[libsanitizer merge from upstream r218156]

git-svn-id: https://gcc.gnu.org/svn/gcc/trunk@215527 138bc75d-0d04-0410-961f-82ee72b054a4

163 files changed, 6355 insertions, 3751 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 62369c7500b..5f50f383272 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,12 @@
+2014-09-23  Kostya Serebryany  <kcc@google.com>
+
+	Update to match the changed asan API.
+	* asan.c (asan_global_struct): Update the __asan_global definition
+	to match the new API.
+	(asan_add_global): Ditto.
+	* sanitizer.def (BUILT_IN_ASAN_INIT): Rename __asan_init_v3
+	to __asan_init_v4.
+
 2014-09-23  Michael Meissner  <meissner@linux.vnet.ibm.com>
 
 	* config/rs6000/rs6000.md (f32_vsx): New mode attributes to
diff --git a/gcc/asan.c b/gcc/asan.c
index 2a90d863d07..6917950477d 100644
--- a/gcc/asan.c
+++ b/gcc/asan.c
@@ -230,6 +230,9 @@ along with GCC; see the file COPYING3.  If not see
 
        // 1 if it has dynamic initialization, 0 otherwise.
        uptr __has_dynamic_init;
+
+       // A pointer to struct that contains source location, could be NULL.
+       __asan_global_source_location *__location;
      }
 
    A destructor function that calls the runtime asan library function
@@ -2136,19 +2139,20 @@ asan_dynamic_init_call (bool after_p)
      const void *__name;
      const void *__module_name;
      uptr __has_dynamic_init;
+     __asan_global_source_location *__location;
    } type.  */
 
 static tree
 asan_global_struct (void)
 {
-  static const char *field_names[6]
+  static const char *field_names[7]
     = { "__beg", "__size", "__size_with_redzone",
-	"__name", "__module_name", "__has_dynamic_init" };
-  tree fields[6], ret;
+	"__name", "__module_name", "__has_dynamic_init", "__location"};
+  tree fields[7], ret;
   int i;
 
   ret = make_node (RECORD_TYPE);
-  for (i = 0; i < 6; i++)
+  for (i = 0; i < 7; i++)
     {
       fields[i]
 	= build_decl (UNKNOWN_LOCATION, FIELD_DECL,
@@ -2220,6 +2224,8 @@ asan_add_global (tree decl, tree type, vec<constructor_elt, va_gc> *v)
   int has_dynamic_init = vnode ? vnode->dynamically_initialized : 0;
   CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
 			  build_int_cst (uptr, has_dynamic_init));
+  CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
+			  build_int_cst (uptr, 0));
   init = build_constructor (type, vinner);
   CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, init);
 }
diff --git a/gcc/sanitizer.def b/gcc/sanitizer.def
index bba28bde334..722311a7868 100644
--- a/gcc/sanitizer.def
+++ b/gcc/sanitizer.def
@@ -27,7 +27,7 @@ along with GCC; see the file COPYING3.  If not see
    for other FEs by asan.c.  */
 
 /* Address Sanitizer */
-DEF_SANITIZER_BUILTIN(BUILT_IN_ASAN_INIT, "__asan_init_v3",
+DEF_SANITIZER_BUILTIN(BUILT_IN_ASAN_INIT, "__asan_init_v4",
 		      BT_FN_VOID, ATTR_NOTHROW_LEAF_LIST)
 /* Do not reorder the BUILT_IN_ASAN_{REPORT,CHECK}* builtins, e.g. cfgcleanup.c
    relies on this order.  */
diff --git a/libsanitizer/ChangeLog b/libsanitizer/ChangeLog
index 7b9d84c86ad..75c24b4756b 100644
--- a/libsanitizer/ChangeLog
+++ b/libsanitizer/ChangeLog
@@ -1,3 +1,17 @@
+2014-09-19  Kostya Serebryany  <kcc@google.com>
+
+	* All source files: Merge from upstream r218156.
+	* asan/Makefile.am (asan_files): Added new files.
+	* asan/Makefile.in: Regenerate.
+	* ubsan/Makefile.am (ubsan_files): Added new files.
+	* ubsan/Makefile.in: Regenerate.
+	* tsan/Makefile.am (tsan_files): Added new files.
+	* tsan/Makefile.in: Regenerate.
+	* sanitizer_common/Makefile.am (sanitizer_common_files): Added new
+	  files.
+	* sanitizer_common/Makefile.in: Regenerate.
+	* asan/libtool-version: Bump the libasan SONAME.
+
 2014-09-10  Jakub Jelinek  <jakub@redhat.com>
 
 	* ubsan/ubsan_handlers.cc, ubsan/ubsan_handlers.h: Cherry pick
diff --git a/libsanitizer/MERGE b/libsanitizer/MERGE
index 84e88e10b77..48a7f26d32d 100644
--- a/libsanitizer/MERGE
+++ b/libsanitizer/MERGE
@@ -1,4 +1,4 @@
-209283
+218156
 
 The first line of this file holds the svn revision number of the
 last merge done from the master library sources.
diff --git a/libsanitizer/asan/Makefile.am b/libsanitizer/asan/Makefile.am
index fdc2b4513d9..12f20ae6d27 100644
--- a/libsanitizer/asan/Makefile.am
+++ b/libsanitizer/asan/Makefile.am
@@ -17,7 +17,7 @@ nodist_toolexeclib_HEADERS = libasan_preinit.o
 asan_files = \
 	asan_activation.cc \
 	asan_allocator2.cc \
-	asan_dll_thunk.cc \
+	asan_debugging.cc \
 	asan_fake_stack.cc \
 	asan_globals.cc \
 	asan_interceptors.cc \
@@ -34,7 +34,9 @@ asan_files = \
 	asan_stack.cc \
 	asan_stats.cc \
 	asan_thread.cc \
-	asan_win.cc
+	asan_win.cc \
+	asan_win_dll_thunk.cc \
+	asan_win_dynamic_runtime_thunk.cc
 
 libasan_la_SOURCES = $(asan_files)
 libasan_la_LIBADD = $(top_builddir)/sanitizer_common/libsanitizer_common.la $(top_builddir)/lsan/libsanitizer_lsan.la
diff --git a/libsanitizer/asan/Makefile.in b/libsanitizer/asan/Makefile.in
index cae6493fc9f..862eec4cea1 100644
--- a/libsanitizer/asan/Makefile.in
+++ b/libsanitizer/asan/Makefile.in
@@ -89,12 +89,13 @@ libasan_la_DEPENDENCIES =  \
 	$(top_builddir)/lsan/libsanitizer_lsan.la $(am__append_2) \
 	$(am__append_3) $(am__DEPENDENCIES_1)
 am__objects_1 = asan_activation.lo asan_allocator2.lo \
-	asan_dll_thunk.lo asan_fake_stack.lo asan_globals.lo \
+	asan_debugging.lo asan_fake_stack.lo asan_globals.lo \
 	asan_interceptors.lo asan_linux.lo asan_mac.lo \
 	asan_malloc_linux.lo asan_malloc_mac.lo asan_malloc_win.lo \
 	asan_new_delete.lo asan_poisoning.lo asan_posix.lo \
 	asan_report.lo asan_rtl.lo asan_stack.lo asan_stats.lo \
-	asan_thread.lo asan_win.lo
+	asan_thread.lo asan_win.lo asan_win_dll_thunk.lo \
+	asan_win_dynamic_runtime_thunk.lo
 am_libasan_la_OBJECTS = $(am__objects_1)
 libasan_la_OBJECTS = $(am_libasan_la_OBJECTS)
 libasan_la_LINK = $(LIBTOOL) --tag=CXX $(AM_LIBTOOLFLAGS) \
@@ -275,7 +276,7 @@ nodist_toolexeclib_HEADERS = libasan_preinit.o
 asan_files = \
 	asan_activation.cc \
 	asan_allocator2.cc \
-	asan_dll_thunk.cc \
+	asan_debugging.cc \
 	asan_fake_stack.cc \
 	asan_globals.cc \
 	asan_interceptors.cc \
@@ -292,7 +293,9 @@ asan_files = \
 	asan_stack.cc \
 	asan_stats.cc \
 	asan_thread.cc \
-	asan_win.cc
+	asan_win.cc \
+	asan_win_dll_thunk.cc \
+	asan_win_dynamic_runtime_thunk.cc
 
 libasan_la_SOURCES = $(asan_files)
 libasan_la_LIBADD =  \
@@ -416,7 +419,7 @@ distclean-compile:
 
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_activation.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_allocator2.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_dll_thunk.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_debugging.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_fake_stack.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_globals.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_interceptors.Plo@am__quote@
@@ -434,6 +437,8 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_stats.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_thread.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_win.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_win_dll_thunk.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/asan_win_dynamic_runtime_thunk.Plo@am__quote@
 
 .cc.o:
 @am__fastdepCXX_TRUE@	$(CXXCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
diff --git a/libsanitizer/asan/asan_allocator.h b/libsanitizer/asan/asan_allocator.h
index 174a5997d4b..567b36867ab 100644
--- a/libsanitizer/asan/asan_allocator.h
+++ b/libsanitizer/asan/asan_allocator.h
@@ -140,6 +140,8 @@ struct AsanThreadLocalMallocStorage {
 void *asan_memalign(uptr alignment, uptr size, StackTrace *stack,
                     AllocType alloc_type);
 void asan_free(void *ptr, StackTrace *stack, AllocType alloc_type);
+void asan_sized_free(void *ptr, uptr size, StackTrace *stack,
+                     AllocType alloc_type);
 
 void *asan_malloc(uptr size, StackTrace *stack);
 void *asan_calloc(uptr nmemb, uptr size, StackTrace *stack);
diff --git a/libsanitizer/asan/asan_allocator2.cc b/libsanitizer/asan/asan_allocator2.cc
index bbc1ff723a4..78c1ec113a6 100644
--- a/libsanitizer/asan/asan_allocator2.cc
+++ b/libsanitizer/asan/asan_allocator2.cc
@@ -19,6 +19,7 @@
 #include "asan_report.h"
 #include "asan_stack.h"
 #include "asan_thread.h"
+#include "sanitizer_common/sanitizer_allocator_interface.h"
 #include "sanitizer_common/sanitizer_flags.h"
 #include "sanitizer_common/sanitizer_internal_defs.h"
 #include "sanitizer_common/sanitizer_list.h"
@@ -165,23 +166,6 @@ struct AsanChunk: ChunkBase {
     }
     return reinterpret_cast<void*>(Beg() - RZLog2Size(rz_log));
   }
-  // If we don't use stack depot, we store the alloc/free stack traces
-  // in the chunk itself.
-  u32 *AllocStackBeg() {
-    return (u32*)(Beg() - RZLog2Size(rz_log));
-  }
-  uptr AllocStackSize() {
-    CHECK_LE(RZLog2Size(rz_log), kChunkHeaderSize);
-    return (RZLog2Size(rz_log) - kChunkHeaderSize) / sizeof(u32);
-  }
-  u32 *FreeStackBeg() {
-    return (u32*)(Beg() + kChunkHeader2Size);
-  }
-  uptr FreeStackSize() {
-    if (user_requested_size < kChunkHeader2Size) return 0;
-    uptr available = RoundUpTo(user_requested_size, SHADOW_GRANULARITY);
-    return (available - kChunkHeader2Size) / sizeof(u32);
-  }
   bool AddrIsInside(uptr addr, bool locked_version = false) {
     return (addr >= Beg()) && (addr < Beg() + UsedSize(locked_version));
   }
@@ -461,12 +445,17 @@ static void QuarantineChunk(AsanChunk *m, void *ptr,
   }
 }
 
-static void Deallocate(void *ptr, StackTrace *stack, AllocType alloc_type) {
+static void Deallocate(void *ptr, uptr delete_size, StackTrace *stack,
+                       AllocType alloc_type) {
   uptr p = reinterpret_cast<uptr>(ptr);
   if (p == 0) return;
 
   uptr chunk_beg = p - kChunkHeaderSize;
   AsanChunk *m = reinterpret_cast<AsanChunk *>(chunk_beg);
+  if (delete_size && flags()->new_delete_type_mismatch &&
+      delete_size != m->UsedSize()) {
+    ReportNewDeleteSizeMismatch(p, delete_size, stack);
+  }
   ASAN_FREE_HOOK(ptr);
   // Must mark the chunk as quarantined before any changes to its metadata.
   AtomicallySetQuarantineFlag(m, ptr, stack);
@@ -493,7 +482,7 @@ static void *Reallocate(void *old_ptr, uptr new_size, StackTrace *stack) {
     // If realloc() races with free(), we may start copying freed memory.
     // However, we will report racy double-free later anyway.
     REAL(memcpy)(new_ptr, old_ptr, memcpy_size);
-    Deallocate(old_ptr, stack, FROM_MALLOC);
+    Deallocate(old_ptr, 0, stack, FROM_MALLOC);
   }
   return new_ptr;
 }
@@ -592,7 +581,12 @@ void *asan_memalign(uptr alignment, uptr size, StackTrace *stack,
 }
 
 void asan_free(void *ptr, StackTrace *stack, AllocType alloc_type) {
-  Deallocate(ptr, stack, alloc_type);
+  Deallocate(ptr, 0, stack, alloc_type);
+}
+
+void asan_sized_free(void *ptr, uptr size, StackTrace *stack,
+                     AllocType alloc_type) {
+  Deallocate(ptr, size, stack, alloc_type);
 }
 
 void *asan_malloc(uptr size, StackTrace *stack) {
@@ -614,7 +608,7 @@ void *asan_realloc(void *p, uptr size, StackTrace *stack) {
   if (p == 0)
     return Allocate(size, 8, stack, FROM_MALLOC, true);
   if (size == 0) {
-    Deallocate(p, stack, FROM_MALLOC);
+    Deallocate(p, 0, stack, FROM_MALLOC);
     return 0;
   }
   return Reallocate(p, size, stack);
@@ -758,23 +752,23 @@ using namespace __asan;  // NOLINT
 
 // ASan allocator doesn't reserve extra bytes, so normally we would
 // just return "size". We don't want to expose our redzone sizes, etc here.
-uptr __asan_get_estimated_allocated_size(uptr size) {
+uptr __sanitizer_get_estimated_allocated_size(uptr size) {
   return size;
 }
 
-int __asan_get_ownership(const void *p) {
+int __sanitizer_get_ownership(const void *p) {
   uptr ptr = reinterpret_cast<uptr>(p);
   return (AllocationSize(ptr) > 0);
 }
 
-uptr __asan_get_allocated_size(const void *p) {
+uptr __sanitizer_get_allocated_size(const void *p) {
   if (p == 0) return 0;
   uptr ptr = reinterpret_cast<uptr>(p);
   uptr allocated_size = AllocationSize(ptr);
   // Die if p is not malloced or if it is already freed.
   if (allocated_size == 0) {
     GET_STACK_TRACE_FATAL_HERE;
-    ReportAsanGetAllocatedSizeNotOwned(ptr, &stack);
+    ReportSanitizerGetAllocatedSizeNotOwned(ptr, &stack);
   }
   return allocated_size;
 }
@@ -783,12 +777,12 @@ uptr __asan_get_allocated_size(const void *p) {
 // Provide default (no-op) implementation of malloc hooks.
 extern "C" {
 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
-void __asan_malloc_hook(void *ptr, uptr size) {
+void __sanitizer_malloc_hook(void *ptr, uptr size) {
   (void)ptr;
   (void)size;
 }
 SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
-void __asan_free_hook(void *ptr) {
+void __sanitizer_free_hook(void *ptr) {
   (void)ptr;
 }
 }  // extern "C"
diff --git a/libsanitizer/asan/asan_asm_instrumentation.S b/libsanitizer/asan/asan_asm_instrumentation.S
deleted file mode 100644
index 36a9d0b5e97..00000000000
--- a/libsanitizer/asan/asan_asm_instrumentation.S
+++ /dev/null
@@ -1,599 +0,0 @@
-// This file was generated by gen_asm_instrumentation.sh. Please, do not edit
-.section .text
-#if defined(__x86_64__) || defined(__i386__)
-.globl __asan_report_store1
-.globl __asan_report_load1
-.globl __asan_report_store2
-.globl __asan_report_load2
-.globl __asan_report_store4
-.globl __asan_report_load4
-.globl __asan_report_store8
-.globl __asan_report_load8
-.globl __asan_report_store16
-.globl __asan_report_load16
-#endif //  defined(__x86_64__) || defined(__i386__)
-#if defined(__i386__)
-// Sanitize 1-byte store. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_store1
-.type __sanitizer_sanitize_store1, @function
-__sanitizer_sanitize_store1:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_store1_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_store1_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_store1@PLT
-.sanitize_store1_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 1-byte load. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_load1
-.type __sanitizer_sanitize_load1, @function
-__sanitizer_sanitize_load1:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_load1_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_load1_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_load1@PLT
-.sanitize_load1_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 2-byte store. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_store2
-.type __sanitizer_sanitize_store2, @function
-__sanitizer_sanitize_store2:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_store2_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  incl %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_store2_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_store2@PLT
-.sanitize_store2_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 2-byte load. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_load2
-.type __sanitizer_sanitize_load2, @function
-__sanitizer_sanitize_load2:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_load2_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  incl %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_load2_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_load2@PLT
-.sanitize_load2_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 4-byte store. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_store4
-.type __sanitizer_sanitize_store4, @function
-__sanitizer_sanitize_store4:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_store4_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  addl $0x3, %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_store4_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_store4@PLT
-.sanitize_store4_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 4-byte load. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_load4
-.type __sanitizer_sanitize_load4, @function
-__sanitizer_sanitize_load4:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushl %edx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  movb 0x20000000(%ecx), %cl
-  testb %cl, %cl
-  je .sanitize_load4_done
-  movl %eax, %edx
-  andl $0x7, %edx
-  addl $0x3, %edx
-  movsbl %cl, %ecx
-  cmpl %ecx, %edx
-  jl .sanitize_load4_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_load4@PLT
-.sanitize_load4_done:
-  popfl
-  popl %edx
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 8-byte store. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_store8
-.type __sanitizer_sanitize_store8, @function
-__sanitizer_sanitize_store8:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  cmpb $0x0, 0x20000000(%ecx)
-  je .sanitize_store8_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_store8@PLT
-.sanitize_store8_done:
-  popfl
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 8-byte load. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_load8
-.type __sanitizer_sanitize_load8, @function
-__sanitizer_sanitize_load8:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  cmpb $0x0, 0x20000000(%ecx)
-  je .sanitize_load8_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_load8@PLT
-.sanitize_load8_done:
-  popfl
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 16-byte store. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_store16
-.type __sanitizer_sanitize_store16, @function
-__sanitizer_sanitize_store16:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  cmpw $0x0, 0x20000000(%ecx)
-  je .sanitize_store16_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_store16@PLT
-.sanitize_store16_done:
-  popfl
-  popl %ecx
-  popl %eax
-  leave
-  ret
-// Sanitize 16-byte load. Takes one 4-byte address as an argument on
-// stack, nothing is returned.
-.globl __sanitizer_sanitize_load16
-.type __sanitizer_sanitize_load16, @function
-__sanitizer_sanitize_load16:
-  pushl %ebp
-  movl %esp, %ebp
-  pushl %eax
-  pushl %ecx
-  pushfl
-  movl 8(%ebp), %eax
-  movl %eax, %ecx
-  shrl $0x3, %ecx
-  cmpw $0x0, 0x20000000(%ecx)
-  je .sanitize_load16_done
-  pushl %eax
-  cld
-  emms
-  call __asan_report_load16@PLT
-.sanitize_load16_done:
-  popfl
-  popl %ecx
-  popl %eax
-  leave
-  ret
-#endif // defined(__i386__)
-#if defined(__x86_64__)
-// Sanitize 1-byte store. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_store1
-.type __sanitizer_sanitize_store1, @function
-__sanitizer_sanitize_store1:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_store1_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_store1_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_store1@PLT
-.sanitize_store1_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 1-byte load. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_load1
-.type __sanitizer_sanitize_load1, @function
-__sanitizer_sanitize_load1:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_load1_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_load1_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_load1@PLT
-.sanitize_load1_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 2-byte store. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_store2
-.type __sanitizer_sanitize_store2, @function
-__sanitizer_sanitize_store2:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_store2_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  incl %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_store2_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_store2@PLT
-.sanitize_store2_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 2-byte load. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_load2
-.type __sanitizer_sanitize_load2, @function
-__sanitizer_sanitize_load2:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_load2_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  incl %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_load2_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_load2@PLT
-.sanitize_load2_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 4-byte store. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_store4
-.type __sanitizer_sanitize_store4, @function
-__sanitizer_sanitize_store4:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_store4_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  addl $0x3, %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_store4_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_store4@PLT
-.sanitize_store4_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 4-byte load. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_load4
-.type __sanitizer_sanitize_load4, @function
-__sanitizer_sanitize_load4:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushq %rcx
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  movb 0x7fff8000(%rax), %al
-  test %al, %al
-  je .sanitize_load4_done
-  movl %edi, %ecx
-  andl $0x7, %ecx
-  addl $0x3, %ecx
-  movsbl %al, %eax
-  cmpl %eax, %ecx
-  jl .sanitize_load4_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_load4@PLT
-.sanitize_load4_done:
-  popfq
-  popq %rcx
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 8-byte store. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_store8
-.type __sanitizer_sanitize_store8, @function
-__sanitizer_sanitize_store8:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  cmpb $0x0, 0x7fff8000(%rax)
-  je .sanitize_store8_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_store8@PLT
-.sanitize_store8_done:
-  popfq
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 8-byte load. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_load8
-.type __sanitizer_sanitize_load8, @function
-__sanitizer_sanitize_load8:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  cmpb $0x0, 0x7fff8000(%rax)
-  je .sanitize_load8_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_load8@PLT
-.sanitize_load8_done:
-  popfq
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 16-byte store. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_store16
-.type __sanitizer_sanitize_store16, @function
-__sanitizer_sanitize_store16:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  cmpw $0x0, 0x7fff8000(%rax)
-  je .sanitize_store16_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_store16@PLT
-.sanitize_store16_done:
-  popfq
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-// Sanitize 16-byte load. Takes one 8-byte address as an argument in %rdi,
-// nothing is returned.
-.globl __sanitizer_sanitize_load16
-.type __sanitizer_sanitize_load16, @function
-__sanitizer_sanitize_load16:
-  leaq -128(%rsp), %rsp
-  pushq %rax
-  pushfq
-  movq %rdi, %rax
-  shrq $0x3, %rax
-  cmpw $0x0, 0x7fff8000(%rax)
-  je .sanitize_load16_done
-  subq $8, %rsp
-  andq $-16, %rsp
-  cld
-  emms
-  call __asan_report_load16@PLT
-.sanitize_load16_done:
-  popfq
-  popq %rax
-  leaq 128(%rsp), %rsp
-  ret
-#endif // defined(__x86_64__)
-/* We do not need executable stack. */
-#if defined(__arm__)
-  .section .note.GNU-stack,"",%progbits
-#else
-  .section .note.GNU-stack,"",@progbits
-#endif // defined(__arm__)
-#endif // __linux__
diff --git a/libsanitizer/asan/asan_debugging.cc b/libsanitizer/asan/asan_debugging.cc
new file mode 100644
index 00000000000..302574d9dd8
--- /dev/null
+++ b/libsanitizer/asan/asan_debugging.cc
@@ -0,0 +1,72 @@
+//===-- asan_debugging.cc -------------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of AddressSanitizer, an address sanity checker.
+//
+// This file contains various functions that are generally useful to call when
+// using a debugger (LLDB, GDB).
+//===----------------------------------------------------------------------===//
+
+#include "asan_allocator.h"
+#include "asan_flags.h"
+#include "asan_internal.h"
+#include "asan_mapping.h"
+#include "asan_thread.h"
+
+namespace __asan {
+
+uptr AsanGetStack(uptr addr, uptr *trace, uptr size, u32 *thread_id,
+                  bool alloc_stack) {
+  AsanChunkView chunk = FindHeapChunkByAddress(addr);
+  if (!chunk.IsValid()) return 0;
+
+  StackTrace stack;
+  if (alloc_stack) {
+    if (chunk.AllocTid() == kInvalidTid) return 0;
+    chunk.GetAllocStack(&stack);
+    if (thread_id) *thread_id = chunk.AllocTid();
+  } else {
+    if (chunk.FreeTid() == kInvalidTid) return 0;
+    chunk.GetFreeStack(&stack);
+    if (thread_id) *thread_id = chunk.FreeTid();
+  }
+
+  if (trace && size) {
+    if (size > kStackTraceMax)
+      size = kStackTraceMax;
+    if (size > stack.size)
+      size = stack.size;
+    for (uptr i = 0; i < size; i++)
+      trace[i] = StackTrace::GetPreviousInstructionPc(stack.trace[i]);
+
+    return size;
+  }
+
+  return 0;
+}
+
+}  // namespace __asan
+
+using namespace __asan;
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __asan_get_alloc_stack(uptr addr, uptr *trace, uptr size, u32 *thread_id) {
+  return AsanGetStack(addr, trace, size, thread_id, /* alloc_stack */ true);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __asan_get_free_stack(uptr addr, uptr *trace, uptr size, u32 *thread_id) {
+  return AsanGetStack(addr, trace, size, thread_id, /* alloc_stack */ false);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __asan_get_shadow_mapping(uptr *shadow_scale, uptr *shadow_offset) {
+  if (shadow_scale)
+    *shadow_scale = SHADOW_SCALE;
+  if (shadow_offset)
+    *shadow_offset = SHADOW_OFFSET;
+}
diff --git a/libsanitizer/asan/asan_flags.h b/libsanitizer/asan/asan_flags.h
index 42463a69b99..2f155eb5cbc 100644
--- a/libsanitizer/asan/asan_flags.h
+++ b/libsanitizer/asan/asan_flags.h
@@ -50,12 +50,13 @@ struct Flags {
   bool print_stats;
   bool print_legend;
   bool atexit;
-  bool disable_core;
   bool allow_reexec;
   bool print_full_thread_history;
   bool poison_heap;
   bool poison_partial;
+  bool poison_array_cookie;
   bool alloc_dealloc_mismatch;
+  bool new_delete_type_mismatch;
   bool strict_memcmp;
   bool strict_init_order;
   bool start_deactivated;
diff --git a/libsanitizer/asan/asan_globals.cc b/libsanitizer/asan/asan_globals.cc
index 132a564f4fe..15c1886af0e 100644
--- a/libsanitizer/asan/asan_globals.cc
+++ b/libsanitizer/asan/asan_globals.cc
@@ -20,6 +20,7 @@
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_mutex.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
+#include "sanitizer_common/sanitizer_stackdepot.h"
 
 namespace __asan {
 
@@ -43,6 +44,14 @@ typedef InternalMmapVector<DynInitGlobal> VectorOfGlobals;
 // Lazy-initialized and never deleted.
 static VectorOfGlobals *dynamic_init_globals;
 
+// We want to remember where a certain range of globals was registered.
+struct GlobalRegistrationSite {
+  u32 stack_id;
+  Global *g_first, *g_last;
+};
+typedef InternalMmapVector<GlobalRegistrationSite> GlobalRegistrationSiteVector;
+static GlobalRegistrationSiteVector *global_registration_site_vector;
+
 ALWAYS_INLINE void PoisonShadowForGlobal(const Global *g, u8 value) {
   FastPoisonShadow(g->beg, g->size_with_redzone, value);
 }
@@ -61,9 +70,14 @@ ALWAYS_INLINE void PoisonRedZones(const Global &g) {
 }
 
 static void ReportGlobal(const Global &g, const char *prefix) {
-  Report("%s Global: beg=%p size=%zu/%zu name=%s module=%s dyn_init=%zu\n",
-         prefix, (void*)g.beg, g.size, g.size_with_redzone, g.name,
+  Report("%s Global[%p]: beg=%p size=%zu/%zu name=%s module=%s dyn_init=%zu\n",
+         prefix, &g, (void *)g.beg, g.size, g.size_with_redzone, g.name,
          g.module_name, g.has_dynamic_init);
+  if (g.location) {
+    Report("  location (%p): name=%s[%p], %d %d\n", g.location,
+           g.location->filename, g.location->filename, g.location->line_no,
+           g.location->column_no);
+  }
 }
 
 bool DescribeAddressIfGlobal(uptr addr, uptr size) {
@@ -79,6 +93,16 @@ bool DescribeAddressIfGlobal(uptr addr, uptr size) {
   return res;
 }
 
+u32 FindRegistrationSite(const Global *g) {
+  CHECK(global_registration_site_vector);
+  for (uptr i = 0, n = global_registration_site_vector->size(); i < n; i++) {
+    GlobalRegistrationSite &grs = (*global_registration_site_vector)[i];
+    if (g >= grs.g_first && g <= grs.g_last)
+      return grs.stack_id;
+  }
+  return 0;
+}
+
 // Register a global variable.
 // This function may be called more than once for every global
 // so we store the globals in a map.
@@ -99,7 +123,8 @@ static void RegisterGlobal(const Global *g) {
       for (ListOfGlobals *l = list_of_all_globals; l; l = l->next) {
         if (g->beg == l->g->beg &&
             (flags()->detect_odr_violation >= 2 || g->size != l->g->size))
-          ReportODRViolation(g, l->g);
+          ReportODRViolation(g, FindRegistrationSite(g),
+                             l->g, FindRegistrationSite(l->g));
       }
     }
   }
@@ -155,7 +180,18 @@ using namespace __asan;  // NOLINT
 // Register an array of globals.
 void __asan_register_globals(__asan_global *globals, uptr n) {
   if (!flags()->report_globals) return;
+  GET_STACK_TRACE_FATAL_HERE;
+  u32 stack_id = StackDepotPut(stack.trace, stack.size);
   BlockingMutexLock lock(&mu_for_globals);
+  if (!global_registration_site_vector)
+    global_registration_site_vector =
+        new(allocator_for_globals) GlobalRegistrationSiteVector(128);
+  GlobalRegistrationSite site = {stack_id, &globals[0], &globals[n - 1]};
+  global_registration_site_vector->push_back(site);
+  if (flags()->report_globals >= 2) {
+    PRINT_CURRENT_STACK();
+    Printf("=== ID %d; %p %p\n", stack_id, &globals[0], &globals[n - 1]);
+  }
   for (uptr i = 0; i < n; i++) {
     RegisterGlobal(&globals[i]);
   }
diff --git a/libsanitizer/asan/asan_init_version.h b/libsanitizer/asan/asan_init_version.h
new file mode 100644
index 00000000000..da232513a09
--- /dev/null
+++ b/libsanitizer/asan/asan_init_version.h
@@ -0,0 +1,30 @@
+//===-- asan_init_version.h -------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of AddressSanitizer, an address sanity checker.
+//
+// This header defines a versioned __asan_init function to be called at the
+// startup of the instrumented program.
+//===----------------------------------------------------------------------===//
+#ifndef ASAN_INIT_VERSION_H
+#define ASAN_INIT_VERSION_H
+
+extern "C" {
+  // Every time the ASan ABI changes we also change the version number in the
+  // __asan_init function name.  Objects built with incompatible ASan ABI
+  // versions will not link with run-time.
+  // Changes between ABI versions:
+  // v1=>v2: added 'module_name' to __asan_global
+  // v2=>v3: stack frame description (created by the compiler)
+  //         contains the function PC as the 3-rd field (see
+  //         DescribeAddressIfStack).
+  // v3=>v4: added '__asan_global_source_location' to __asan_global.
+  #define __asan_init __asan_init_v4
+  #define __asan_init_name "__asan_init_v4"
+}
+
+#endif  // ASAN_INIT_VERSION_H
diff --git a/libsanitizer/asan/asan_interceptors.cc b/libsanitizer/asan/asan_interceptors.cc
index 13deab5766e..182b7842ed9 100644
--- a/libsanitizer/asan/asan_interceptors.cc
+++ b/libsanitizer/asan/asan_interceptors.cc
@@ -144,6 +144,9 @@ DECLARE_REAL_AND_INTERCEPTOR(void, free, void *)
   } while (false)
 #define COMMON_INTERCEPTOR_BLOCK_REAL(name) REAL(name)
 #define COMMON_INTERCEPTOR_ON_EXIT(ctx) OnExit()
+#define COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, res) CovUpdateMapping()
+#define COMMON_INTERCEPTOR_LIBRARY_UNLOADED() CovUpdateMapping()
+#define COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED (!asan_inited)
 #include "sanitizer_common/sanitizer_common_interceptors.inc"
 
 #define COMMON_SYSCALL_PRE_READ_RANGE(p, s) ASAN_READ_RANGE(p, s)
@@ -291,37 +294,29 @@ INTERCEPTOR(void, __cxa_throw, void *a, void *b, void *c) {
 }
 #endif
 
-// intercept mlock and friends.
-// Since asan maps 16T of RAM, mlock is completely unfriendly to asan.
-// All functions return 0 (success).
-static void MlockIsUnsupported() {
-  static bool printed = false;
-  if (printed) return;
-  printed = true;
-  VPrintf(1,
-          "INFO: AddressSanitizer ignores "
-          "mlock/mlockall/munlock/munlockall\n");
-}
-
-INTERCEPTOR(int, mlock, const void *addr, uptr len) {
-  MlockIsUnsupported();
-  return 0;
-}
-
-INTERCEPTOR(int, munlock, const void *addr, uptr len) {
-  MlockIsUnsupported();
-  return 0;
+#if SANITIZER_WINDOWS
+INTERCEPTOR_WINAPI(void, RaiseException, void *a, void *b, void *c, void *d) {
+  CHECK(REAL(RaiseException));
+  __asan_handle_no_return();
+  REAL(RaiseException)(a, b, c, d);
 }
 
-INTERCEPTOR(int, mlockall, int flags) {
-  MlockIsUnsupported();
-  return 0;
+INTERCEPTOR(int, _except_handler3, void *a, void *b, void *c, void *d) {
+  CHECK(REAL(_except_handler3));
+  __asan_handle_no_return();
+  return REAL(_except_handler3)(a, b, c, d);
 }
 
-INTERCEPTOR(int, munlockall, void) {
-  MlockIsUnsupported();
-  return 0;
+#if ASAN_DYNAMIC
+// This handler is named differently in -MT and -MD CRTs.
+#define _except_handler4 _except_handler4_common
+#endif
+INTERCEPTOR(int, _except_handler4, void *a, void *b, void *c, void *d) {
+  CHECK(REAL(_except_handler4));
+  __asan_handle_no_return();
+  return REAL(_except_handler4)(a, b, c, d);
 }
+#endif
 
 static inline int CharCmp(unsigned char c1, unsigned char c2) {
   return (c1 == c2) ? 0 : (c1 < c2) ? -1 : 1;
@@ -523,7 +518,7 @@ INTERCEPTOR(char*, strdup, const char *s) {
 }
 #endif
 
-INTERCEPTOR(uptr, strlen, const char *s) {
+INTERCEPTOR(SIZE_T, strlen, const char *s) {
   if (UNLIKELY(!asan_inited)) return internal_strlen(s);
   // strlen is called from malloc_default_purgeable_zone()
   // in __asan::ReplaceSystemAlloc() on Mac.
@@ -531,15 +526,15 @@ INTERCEPTOR(uptr, strlen, const char *s) {
     return REAL(strlen)(s);
   }
   ENSURE_ASAN_INITED();
-  uptr length = REAL(strlen)(s);
+  SIZE_T length = REAL(strlen)(s);
   if (flags()->replace_str) {
     ASAN_READ_RANGE(s, length + 1);
   }
   return length;
 }
 
-INTERCEPTOR(uptr, wcslen, const wchar_t *s) {
-  uptr length = REAL(wcslen)(s);
+INTERCEPTOR(SIZE_T, wcslen, const wchar_t *s) {
+  SIZE_T length = REAL(wcslen)(s);
   if (!asan_init_is_running) {
     ENSURE_ASAN_INITED();
     ASAN_READ_RANGE(s, (length + 1) * sizeof(wchar_t));
@@ -691,6 +686,16 @@ INTERCEPTOR(int, __cxa_atexit, void (*func)(void *), void *arg,
 }
 #endif  // ASAN_INTERCEPT___CXA_ATEXIT
 
+#if ASAN_INTERCEPT_FORK
+INTERCEPTOR(int, fork, void) {
+  ENSURE_ASAN_INITED();
+  if (common_flags()->coverage) CovBeforeFork();
+  int pid = REAL(fork)();
+  if (common_flags()->coverage) CovAfterFork(pid);
+  return pid;
+}
+#endif  // ASAN_INTERCEPT_FORK
+
 #if SANITIZER_WINDOWS
 INTERCEPTOR_WINAPI(DWORD, CreateThread,
                    void* security, uptr stack_size,
@@ -712,6 +717,9 @@ INTERCEPTOR_WINAPI(DWORD, CreateThread,
 namespace __asan {
 void InitializeWindowsInterceptors() {
   ASAN_INTERCEPT_FUNC(CreateThread);
+  ASAN_INTERCEPT_FUNC(RaiseException);
+  ASAN_INTERCEPT_FUNC(_except_handler3);
+  ASAN_INTERCEPT_FUNC(_except_handler4);
 }
 
 }  // namespace __asan
@@ -759,14 +767,6 @@ void InitializeAsanInterceptors() {
   ASAN_INTERCEPT_FUNC(strtoll);
 #endif
 
-#if ASAN_INTERCEPT_MLOCKX
-  // Intercept mlock/munlock.
-  ASAN_INTERCEPT_FUNC(mlock);
-  ASAN_INTERCEPT_FUNC(munlock);
-  ASAN_INTERCEPT_FUNC(mlockall);
-  ASAN_INTERCEPT_FUNC(munlockall);
-#endif
-
   // Intecept signal- and jump-related functions.
   ASAN_INTERCEPT_FUNC(longjmp);
 #if ASAN_INTERCEPT_SIGNAL_AND_SIGACTION
@@ -789,7 +789,7 @@ void InitializeAsanInterceptors() {
 
   // Intercept exception handling functions.
 #if ASAN_INTERCEPT___CXA_THROW
-  INTERCEPT_FUNCTION(__cxa_throw);
+  ASAN_INTERCEPT_FUNC(__cxa_throw);
 #endif
 
   // Intercept threading-related functions
@@ -802,6 +802,10 @@ void InitializeAsanInterceptors() {
   ASAN_INTERCEPT_FUNC(__cxa_atexit);
 #endif
 
+#if ASAN_INTERCEPT_FORK
+  ASAN_INTERCEPT_FUNC(fork);
+#endif
+
   // Some Windows-specific interceptors.
 #if SANITIZER_WINDOWS
   InitializeWindowsInterceptors();
diff --git a/libsanitizer/asan/asan_interceptors.h b/libsanitizer/asan/asan_interceptors.h
index af7cdc8a916..95a75db4e02 100644
--- a/libsanitizer/asan/asan_interceptors.h
+++ b/libsanitizer/asan/asan_interceptors.h
@@ -24,14 +24,14 @@
 # define ASAN_INTERCEPT_STRDUP 1
 # define ASAN_INTERCEPT_INDEX 1
 # define ASAN_INTERCEPT_PTHREAD_CREATE 1
-# define ASAN_INTERCEPT_MLOCKX 1
+# define ASAN_INTERCEPT_FORK 1
 #else
 # define ASAN_INTERCEPT_ATOLL_AND_STRTOLL 0
 # define ASAN_INTERCEPT__LONGJMP 0
 # define ASAN_INTERCEPT_STRDUP 0
 # define ASAN_INTERCEPT_INDEX 0
 # define ASAN_INTERCEPT_PTHREAD_CREATE 0
-# define ASAN_INTERCEPT_MLOCKX 0
+# define ASAN_INTERCEPT_FORK 0
 #endif
 
 #if SANITIZER_FREEBSD || SANITIZER_LINUX
@@ -64,7 +64,9 @@
 # define ASAN_INTERCEPT_SIGLONGJMP 0
 #endif
 
-#if ASAN_HAS_EXCEPTIONS && !SANITIZER_WINDOWS
+// Android bug: https://code.google.com/p/android/issues/detail?id=61799
+#if ASAN_HAS_EXCEPTIONS && !SANITIZER_WINDOWS && \
+    !(SANITIZER_ANDROID && defined(__i386))
 # define ASAN_INTERCEPT___CXA_THROW 1
 #else
 # define ASAN_INTERCEPT___CXA_THROW 0
@@ -80,7 +82,7 @@ DECLARE_REAL(int, memcmp, const void *a1, const void *a2, uptr size)
 DECLARE_REAL(void*, memcpy, void *to, const void *from, uptr size)
 DECLARE_REAL(void*, memset, void *block, int c, uptr size)
 DECLARE_REAL(char*, strchr, const char *str, int c)
-DECLARE_REAL(uptr, strlen, const char *s)
+DECLARE_REAL(SIZE_T, strlen, const char *s)
 DECLARE_REAL(char*, strncpy, char *to, const char *from, uptr size)
 DECLARE_REAL(uptr, strnlen, const char *s, uptr maxlen)
 DECLARE_REAL(char*, strstr, const char *s1, const char *s2)
diff --git a/libsanitizer/asan/asan_interface_internal.h b/libsanitizer/asan/asan_interface_internal.h
index 1940477f247..1a3b33fed0b 100644
--- a/libsanitizer/asan/asan_interface_internal.h
+++ b/libsanitizer/asan/asan_interface_internal.h
@@ -15,21 +15,24 @@
 
 #include "sanitizer_common/sanitizer_internal_defs.h"
 
+#include "asan_init_version.h"
+
 using __sanitizer::uptr;
 
 extern "C" {
   // This function should be called at the very beginning of the process,
   // before any instrumented code is executed and before any call to malloc.
-  // Every time the asan ABI changes we also change the version number in this
-  // name. Objects build with incompatible asan ABI version
-  // will not link with run-time.
-  // Changes between ABI versions:
-  // v1=>v2: added 'module_name' to __asan_global
-  // v2=>v3: stack frame description (created by the compiler)
-  //         contains the function PC as the 3-rd field (see
-  //         DescribeAddressIfStack).
-  SANITIZER_INTERFACE_ATTRIBUTE void __asan_init_v3();
-  #define __asan_init __asan_init_v3
+  // Please note that __asan_init is a macro that is replaced with
+  // __asan_init_vXXX at compile-time.
+  SANITIZER_INTERFACE_ATTRIBUTE void __asan_init();
+
+  // This structure is used to describe the source location of a place where
+  // global was defined.
+  struct __asan_global_source_location {
+    const char *filename;
+    int line_no;
+    int column_no;
+  };
 
   // This structure describes an instrumented global variable.
   struct __asan_global {
@@ -40,6 +43,8 @@ extern "C" {
     const char *module_name; // Module name as a C string. This pointer is a
                              // unique identifier of a module.
     uptr has_dynamic_init;   // Non-zero if the global has dynamic initializer.
+    __asan_global_source_location *location;  // Source location of a global,
+                                              // or NULL if it is unknown.
   };
 
   // These two functions should be called by the instrumented code.
@@ -84,6 +89,17 @@ extern "C" {
   void __asan_describe_address(uptr addr);
 
   SANITIZER_INTERFACE_ATTRIBUTE
+  uptr __asan_get_alloc_stack(uptr addr, uptr *trace, uptr size,
+                              u32 *thread_id);
+
+  SANITIZER_INTERFACE_ATTRIBUTE
+  uptr __asan_get_free_stack(uptr addr, uptr *trace, uptr size,
+                             u32 *thread_id);
+
+  SANITIZER_INTERFACE_ATTRIBUTE
+  void __asan_get_shadow_mapping(uptr *shadow_scale, uptr *shadow_offset);
+
+  SANITIZER_INTERFACE_ATTRIBUTE
   void __asan_report_error(uptr pc, uptr bp, uptr sp,
                            uptr addr, int is_write, uptr access_size);
 
@@ -97,25 +113,11 @@ extern "C" {
   SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
   /* OPTIONAL */ void __asan_on_error();
 
-  SANITIZER_INTERFACE_ATTRIBUTE
-  uptr __asan_get_estimated_allocated_size(uptr size);
-
-  SANITIZER_INTERFACE_ATTRIBUTE int __asan_get_ownership(const void *p);
-  SANITIZER_INTERFACE_ATTRIBUTE uptr __asan_get_allocated_size(const void *p);
-  SANITIZER_INTERFACE_ATTRIBUTE uptr __asan_get_current_allocated_bytes();
-  SANITIZER_INTERFACE_ATTRIBUTE uptr __asan_get_heap_size();
-  SANITIZER_INTERFACE_ATTRIBUTE uptr __asan_get_free_bytes();
-  SANITIZER_INTERFACE_ATTRIBUTE uptr __asan_get_unmapped_bytes();
   SANITIZER_INTERFACE_ATTRIBUTE void __asan_print_accumulated_stats();
 
   SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
   /* OPTIONAL */ const char* __asan_default_options();
 
-  SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
-  /* OPTIONAL */ void __asan_malloc_hook(void *ptr, uptr size);
-  SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
-  /* OPTIONAL */ void __asan_free_hook(void *ptr);
-
   // Global flag, copy of ASAN_OPTIONS=detect_stack_use_after_return
   SANITIZER_INTERFACE_ATTRIBUTE
   extern int __asan_option_detect_stack_use_after_return;
@@ -142,6 +144,11 @@ extern "C" {
       void* __asan_memset(void *s, int c, uptr n);
   SANITIZER_INTERFACE_ATTRIBUTE
       void* __asan_memmove(void* dest, const void* src, uptr n);
+
+  SANITIZER_INTERFACE_ATTRIBUTE
+  void __asan_poison_cxx_array_cookie(uptr p);
+  SANITIZER_INTERFACE_ATTRIBUTE
+  uptr __asan_load_cxx_array_cookie(uptr *p);
 }  // extern "C"
 
 #endif  // ASAN_INTERFACE_INTERNAL_H
diff --git a/libsanitizer/asan/asan_internal.h b/libsanitizer/asan/asan_internal.h
index d56943a0838..9473bf6a2ca 100644
--- a/libsanitizer/asan/asan_internal.h
+++ b/libsanitizer/asan/asan_internal.h
@@ -43,10 +43,6 @@
 # endif
 #endif
 
-#ifndef ASAN_USE_PREINIT_ARRAY
-# define ASAN_USE_PREINIT_ARRAY (SANITIZER_LINUX && !SANITIZER_ANDROID)
-#endif
-
 #ifndef ASAN_DYNAMIC
 # ifdef PIC
 #  define ASAN_DYNAMIC 1
@@ -96,6 +92,8 @@ void AppendToErrorMessageBuffer(const char *buffer);
 
 void ParseExtraActivationFlags();
 
+void *AsanDlSymNext(const char *sym);
+
 // Platform-specific options.
 #if SANITIZER_MAC
 bool PlatformHasDifferentMemcpyAndMemmove();
@@ -108,9 +106,9 @@ bool PlatformHasDifferentMemcpyAndMemmove();
 // Add convenient macro for interface functions that may be represented as
 // weak hooks.
 #define ASAN_MALLOC_HOOK(ptr, size) \
-  if (&__asan_malloc_hook) __asan_malloc_hook(ptr, size)
+  if (&__sanitizer_malloc_hook) __sanitizer_malloc_hook(ptr, size)
 #define ASAN_FREE_HOOK(ptr) \
-  if (&__asan_free_hook) __asan_free_hook(ptr)
+  if (&__sanitizer_free_hook) __sanitizer_free_hook(ptr)
 #define ASAN_ON_ERROR() \
   if (&__asan_on_error) __asan_on_error()
 
@@ -134,6 +132,7 @@ const int kAsanContiguousContainerOOBMagic = 0xfc;
 const int kAsanStackUseAfterScopeMagic = 0xf8;
 const int kAsanGlobalRedzoneMagic = 0xf9;
 const int kAsanInternalHeapMagic = 0xfe;
+const int kAsanArrayCookieMagic = 0xac;
 
 static const uptr kCurrentStackFrameMagic = 0x41B58AB3;
 static const uptr kRetiredStackFrameMagic = 0x45E0360E;
diff --git a/libsanitizer/asan/asan_linux.cc b/libsanitizer/asan/asan_linux.cc
index 08d2885e548..c504168b614 100644
--- a/libsanitizer/asan/asan_linux.cc
+++ b/libsanitizer/asan/asan_linux.cc
@@ -17,6 +17,7 @@
 #include "asan_internal.h"
 #include "asan_thread.h"
 #include "sanitizer_common/sanitizer_flags.h"
+#include "sanitizer_common/sanitizer_freebsd.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_procmaps.h"
 
@@ -25,6 +26,7 @@
 #include <sys/mman.h>
 #include <sys/syscall.h>
 #include <sys/types.h>
+#include <dlfcn.h>
 #include <fcntl.h>
 #include <pthread.h>
 #include <stdio.h>
@@ -40,19 +42,14 @@
 extern "C" void* _DYNAMIC;
 #else
 #include <sys/ucontext.h>
-#include <dlfcn.h>
 #include <link.h>
 #endif
 
-// x86_64 FreeBSD 9.2 and older define 64-bit register names in both 64-bit
-// and 32-bit modes.
-#if SANITIZER_FREEBSD
-#include <sys/param.h>
-# if __FreeBSD_version <= 902001  // v9.2
-#  define mc_eip mc_rip
-#  define mc_ebp mc_rbp
-#  define mc_esp mc_rsp
-# endif
+// x86-64 FreeBSD 9.2 and older define 'ucontext_t' incorrectly in
+// 32-bit mode.
+#if SANITIZER_FREEBSD && (SANITIZER_WORDSIZE == 32) && \
+  __FreeBSD_version <= 902001  // v9.2
+#define ucontext_t xucontext_t
 #endif
 
 typedef enum {
@@ -241,6 +238,10 @@ void ReadContextStack(void *context, uptr *stack, uptr *ssize) {
 }
 #endif
 
+void *AsanDlSymNext(const char *sym) {
+  return dlsym(RTLD_NEXT, sym);
+}
+
 }  // namespace __asan
 
 #endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
diff --git a/libsanitizer/asan/asan_mac.cc b/libsanitizer/asan/asan_mac.cc
index 4a295e0e355..e4c71cedd31 100644
--- a/libsanitizer/asan/asan_mac.cc
+++ b/libsanitizer/asan/asan_mac.cc
@@ -372,32 +372,44 @@ void dispatch_source_set_event_handler(dispatch_source_t ds, void(^work)(void));
     work(); \
   }
 
+// Forces the compiler to generate a frame pointer in the function.
+#define ENABLE_FRAME_POINTER                                       \
+  do {                                                             \
+    volatile uptr enable_fp;                                       \
+    enable_fp = GET_CURRENT_FRAME();                               \
+  } while (0)
+
 INTERCEPTOR(void, dispatch_async,
             dispatch_queue_t dq, void(^work)(void)) {
+  ENABLE_FRAME_POINTER;
   GET_ASAN_BLOCK(work);
   REAL(dispatch_async)(dq, asan_block);
 }
 
 INTERCEPTOR(void, dispatch_group_async,
             dispatch_group_t dg, dispatch_queue_t dq, void(^work)(void)) {
+  ENABLE_FRAME_POINTER;
   GET_ASAN_BLOCK(work);
   REAL(dispatch_group_async)(dg, dq, asan_block);
 }
 
 INTERCEPTOR(void, dispatch_after,
             dispatch_time_t when, dispatch_queue_t queue, void(^work)(void)) {
+  ENABLE_FRAME_POINTER;
   GET_ASAN_BLOCK(work);
   REAL(dispatch_after)(when, queue, asan_block);
 }
 
 INTERCEPTOR(void, dispatch_source_set_cancel_handler,
             dispatch_source_t ds, void(^work)(void)) {
+  ENABLE_FRAME_POINTER;
   GET_ASAN_BLOCK(work);
   REAL(dispatch_source_set_cancel_handler)(ds, asan_block);
 }
 
 INTERCEPTOR(void, dispatch_source_set_event_handler,
             dispatch_source_t ds, void(^work)(void)) {
+  ENABLE_FRAME_POINTER;
   GET_ASAN_BLOCK(work);
   REAL(dispatch_source_set_event_handler)(ds, asan_block);
 }
diff --git a/libsanitizer/asan/asan_malloc_linux.cc b/libsanitizer/asan/asan_malloc_linux.cc
index ba908e322d9..d03f1bb89c8 100644
--- a/libsanitizer/asan/asan_malloc_linux.cc
+++ b/libsanitizer/asan/asan_malloc_linux.cc
@@ -21,41 +21,6 @@
 #include "asan_internal.h"
 #include "asan_stack.h"
 
-#if SANITIZER_ANDROID
-DECLARE_REAL_AND_INTERCEPTOR(void*, malloc, uptr size)
-DECLARE_REAL_AND_INTERCEPTOR(void, free, void *ptr)
-DECLARE_REAL_AND_INTERCEPTOR(void*, calloc, uptr nmemb, uptr size)
-DECLARE_REAL_AND_INTERCEPTOR(void*, realloc, void *ptr, uptr size)
-DECLARE_REAL_AND_INTERCEPTOR(void*, memalign, uptr boundary, uptr size)
-
-struct MallocDebug {
-  void* (*malloc)(uptr bytes);
-  void  (*free)(void* mem);
-  void* (*calloc)(uptr n_elements, uptr elem_size);
-  void* (*realloc)(void* oldMem, uptr bytes);
-  void* (*memalign)(uptr alignment, uptr bytes);
-};
-
-const MallocDebug asan_malloc_dispatch ALIGNED(32) = {
-  WRAP(malloc), WRAP(free), WRAP(calloc), WRAP(realloc), WRAP(memalign)
-};
-
-extern "C" const MallocDebug* __libc_malloc_dispatch;
-
-namespace __asan {
-void ReplaceSystemMalloc() {
-  __libc_malloc_dispatch = &asan_malloc_dispatch;
-}
-}  // namespace __asan
-
-#else  // ANDROID
-
-namespace __asan {
-void ReplaceSystemMalloc() {
-}
-}  // namespace __asan
-#endif  // ANDROID
-
 // ---------------------- Replacement functions ---------------- {{{1
 using namespace __asan;  // NOLINT
 
@@ -100,6 +65,11 @@ INTERCEPTOR(void*, memalign, uptr boundary, uptr size) {
   return asan_memalign(boundary, size, &stack, FROM_MALLOC);
 }
 
+INTERCEPTOR(void*, aligned_alloc, uptr boundary, uptr size) {
+  GET_STACK_TRACE_MALLOC;
+  return asan_memalign(boundary, size, &stack, FROM_MALLOC);
+}
+
 INTERCEPTOR(void*, __libc_memalign, uptr boundary, uptr size) {
   GET_STACK_TRACE_MALLOC;
   void *res = asan_memalign(boundary, size, &stack, FROM_MALLOC);
@@ -151,4 +121,64 @@ INTERCEPTOR(void, malloc_stats, void) {
   __asan_print_accumulated_stats();
 }
 
+#if SANITIZER_ANDROID
+// Format of __libc_malloc_dispatch has changed in Android L.
+// While we are moving towards a solution that does not depend on bionic
+// internals, here is something to support both K* and L releases.
+struct MallocDebugK {
+  void *(*malloc)(uptr bytes);
+  void (*free)(void *mem);
+  void *(*calloc)(uptr n_elements, uptr elem_size);
+  void *(*realloc)(void *oldMem, uptr bytes);
+  void *(*memalign)(uptr alignment, uptr bytes);
+  uptr (*malloc_usable_size)(void *mem);
+};
+
+struct MallocDebugL {
+  void *(*calloc)(uptr n_elements, uptr elem_size);
+  void (*free)(void *mem);
+  fake_mallinfo (*mallinfo)(void);
+  void *(*malloc)(uptr bytes);
+  uptr (*malloc_usable_size)(void *mem);
+  void *(*memalign)(uptr alignment, uptr bytes);
+  int (*posix_memalign)(void **memptr, uptr alignment, uptr size);
+  void* (*pvalloc)(uptr size);
+  void *(*realloc)(void *oldMem, uptr bytes);
+  void* (*valloc)(uptr size);
+};
+
+ALIGNED(32) const MallocDebugK asan_malloc_dispatch_k = {
+    WRAP(malloc),  WRAP(free),     WRAP(calloc),
+    WRAP(realloc), WRAP(memalign), WRAP(malloc_usable_size)};
+
+ALIGNED(32) const MallocDebugL asan_malloc_dispatch_l = {
+    WRAP(calloc),         WRAP(free),               WRAP(mallinfo),
+    WRAP(malloc),         WRAP(malloc_usable_size), WRAP(memalign),
+    WRAP(posix_memalign), WRAP(pvalloc),            WRAP(realloc),
+    WRAP(valloc)};
+
+namespace __asan {
+void ReplaceSystemMalloc() {
+  void **__libc_malloc_dispatch_p =
+      (void **)AsanDlSymNext("__libc_malloc_dispatch");
+  if (__libc_malloc_dispatch_p) {
+    // Decide on K vs L dispatch format by the presence of
+    // __libc_malloc_default_dispatch export in libc.
+    void *default_dispatch_p = AsanDlSymNext("__libc_malloc_default_dispatch");
+    if (default_dispatch_p)
+      *__libc_malloc_dispatch_p = (void *)&asan_malloc_dispatch_k;
+    else
+      *__libc_malloc_dispatch_p = (void *)&asan_malloc_dispatch_l;
+  }
+}
+}  // namespace __asan
+
+#else  // SANITIZER_ANDROID
+
+namespace __asan {
+void ReplaceSystemMalloc() {
+}
+}  // namespace __asan
+#endif  // SANITIZER_ANDROID
+
 #endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
diff --git a/libsanitizer/asan/asan_malloc_win.cc b/libsanitizer/asan/asan_malloc_win.cc
index 8463d5ef2e9..bbcf80e96ba 100644
--- a/libsanitizer/asan/asan_malloc_win.cc
+++ b/libsanitizer/asan/asan_malloc_win.cc
@@ -21,71 +21,77 @@
 
 #include <stddef.h>
 
-// ---------------------- Replacement functions ---------------- {{{1
 using namespace __asan;  // NOLINT
 
-// FIXME: Simply defining functions with the same signature in *.obj
-// files overrides the standard functions in *.lib
-// This works well for simple helloworld-like tests but might need to be
-// revisited in the future.
+// MT: Simply defining functions with the same signature in *.obj
+// files overrides the standard functions in the CRT.
+// MD: Memory allocation functions are defined in the CRT .dll,
+// so we have to intercept them before they are called for the first time.
+
+#if ASAN_DYNAMIC
+# define ALLOCATION_FUNCTION_ATTRIBUTE
+#else
+# define ALLOCATION_FUNCTION_ATTRIBUTE SANITIZER_INTERFACE_ATTRIBUTE
+#endif
 
 extern "C" {
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void free(void *ptr) {
   GET_STACK_TRACE_FREE;
   return asan_free(ptr, &stack, FROM_MALLOC);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
-void _free_dbg(void* ptr, int) {
+ALLOCATION_FUNCTION_ATTRIBUTE
+void _free_dbg(void *ptr, int) {
   free(ptr);
 }
 
+ALLOCATION_FUNCTION_ATTRIBUTE
 void cfree(void *ptr) {
-  CHECK(!"cfree() should not be used on Windows?");
+  CHECK(!"cfree() should not be used on Windows");
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *malloc(size_t size) {
   GET_STACK_TRACE_MALLOC;
   return asan_malloc(size, &stack);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
-void* _malloc_dbg(size_t size, int , const char*, int) {
+ALLOCATION_FUNCTION_ATTRIBUTE
+void *_malloc_dbg(size_t size, int, const char *, int) {
   return malloc(size);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *calloc(size_t nmemb, size_t size) {
   GET_STACK_TRACE_MALLOC;
   return asan_calloc(nmemb, size, &stack);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
-void* _calloc_dbg(size_t n, size_t size, int, const char*, int) {
-  return calloc(n, size);
+ALLOCATION_FUNCTION_ATTRIBUTE
+void *_calloc_dbg(size_t nmemb, size_t size, int, const char *, int) {
+  return calloc(nmemb, size);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *_calloc_impl(size_t nmemb, size_t size, int *errno_tmp) {
   return calloc(nmemb, size);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *realloc(void *ptr, size_t size) {
   GET_STACK_TRACE_MALLOC;
   return asan_realloc(ptr, size, &stack);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *_realloc_dbg(void *ptr, size_t size, int) {
   CHECK(!"_realloc_dbg should not exist!");
   return 0;
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
-void* _recalloc(void* p, size_t n, size_t elem_size) {
+ALLOCATION_FUNCTION_ATTRIBUTE
+void *_recalloc(void *p, size_t n, size_t elem_size) {
   if (!p)
     return calloc(n, elem_size);
   const size_t size = n * elem_size;
@@ -94,23 +100,23 @@ void* _recalloc(void* p, size_t n, size_t elem_size) {
   return realloc(p, size);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 size_t _msize(void *ptr) {
   GET_CURRENT_PC_BP_SP;
   (void)sp;
   return asan_malloc_usable_size(ptr, pc, bp);
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *_expand(void *memblock, size_t size) {
   // _expand is used in realloc-like functions to resize the buffer if possible.
   // We don't want memory to stand still while resizing buffers, so return 0.
   return 0;
 }
 
-SANITIZER_INTERFACE_ATTRIBUTE
+ALLOCATION_FUNCTION_ATTRIBUTE
 void *_expand_dbg(void *memblock, size_t size) {
-  return 0;
+  return _expand(memblock, size);
 }
 
 // TODO(timurrrr): Might want to add support for _aligned_* allocation
@@ -131,37 +137,38 @@ int _CrtSetReportMode(int, int) {
 }
 }  // extern "C"
 
-using __interception::GetRealFunctionAddress;
-
-// We don't want to include "windows.h" in this file to avoid extra attributes
-// set on malloc/free etc (e.g. dllimport), so declare a few things manually:
-extern "C" int __stdcall VirtualProtect(void* addr, size_t size,
-                                        DWORD prot, DWORD *old_prot);
-const int PAGE_EXECUTE_READWRITE = 0x40;
-
 namespace __asan {
 void ReplaceSystemMalloc() {
-#if defined(_DLL)
-# ifdef _WIN64
-#  error ReplaceSystemMalloc was not tested on x64
-# endif
-  char *crt_malloc;
-  if (GetRealFunctionAddress("malloc", (void**)&crt_malloc)) {
-    // Replace malloc in the CRT dll with a jump to our malloc.
-    DWORD old_prot, unused;
-    CHECK(VirtualProtect(crt_malloc, 16, PAGE_EXECUTE_READWRITE, &old_prot));
-    REAL(memset)(crt_malloc, 0xCC /* int 3 */, 16);  // just in case.
-
-    ptrdiff_t jmp_offset = (char*)malloc - (char*)crt_malloc - 5;
-    crt_malloc[0] = 0xE9;  // jmp, should be followed by an offset.
-    REAL(memcpy)(crt_malloc + 1, &jmp_offset, sizeof(jmp_offset));
-
-    CHECK(VirtualProtect(crt_malloc, 16, old_prot, &unused));
-
-    // FYI: FlushInstructionCache is needed on Itanium etc but not on x86/x64.
-  }
-
-  // FIXME: investigate whether anything else is needed.
+#if defined(ASAN_DYNAMIC)
+  // We don't check the result because CRT might not be used in the process.
+  __interception::OverrideFunction("free", (uptr)free);
+  __interception::OverrideFunction("malloc", (uptr)malloc);
+  __interception::OverrideFunction("_malloc_crt", (uptr)malloc);
+  __interception::OverrideFunction("calloc", (uptr)calloc);
+  __interception::OverrideFunction("_calloc_crt", (uptr)calloc);
+  __interception::OverrideFunction("realloc", (uptr)realloc);
+  __interception::OverrideFunction("_realloc_crt", (uptr)realloc);
+  __interception::OverrideFunction("_recalloc", (uptr)_recalloc);
+  __interception::OverrideFunction("_recalloc_crt", (uptr)_recalloc);
+  __interception::OverrideFunction("_msize", (uptr)_msize);
+  __interception::OverrideFunction("_expand", (uptr)_expand);
+
+  // Override different versions of 'operator new' and 'operator delete'.
+  // No need to override the nothrow versions as they just wrap the throw
+  // versions.
+  // FIXME: Unfortunately, MSVC miscompiles the statements that take the
+  // addresses of the array versions of these operators,
+  // see https://connect.microsoft.com/VisualStudio/feedbackdetail/view/946992
+  // We might want to try to work around this by [inline] assembly or compiling
+  // parts of the RTL with Clang.
+  void *(*op_new)(size_t sz) = operator new;
+  void (*op_delete)(void *p) = operator delete;
+  void *(*op_array_new)(size_t sz) = operator new[];
+  void (*op_array_delete)(void *p) = operator delete[];
+  __interception::OverrideFunction("??2@YAPAXI@Z", (uptr)op_new);
+  __interception::OverrideFunction("??3@YAXPAX@Z", (uptr)op_delete);
+  __interception::OverrideFunction("??_U@YAPAXI@Z", (uptr)op_array_new);
+  __interception::OverrideFunction("??_V@YAXPAX@Z", (uptr)op_array_delete);
 #endif
 }
 }  // namespace __asan
diff --git a/libsanitizer/asan/asan_new_delete.cc b/libsanitizer/asan/asan_new_delete.cc
index a1ab2cd8c39..9d6660ec7b4 100644
--- a/libsanitizer/asan/asan_new_delete.cc
+++ b/libsanitizer/asan/asan_new_delete.cc
@@ -18,6 +18,13 @@
 
 #include <stddef.h>
 
+// C++ operators can't have visibility attributes on Windows.
+#if SANITIZER_WINDOWS
+# define CXX_OPERATOR_ATTRIBUTE
+#else
+# define CXX_OPERATOR_ATTRIBUTE INTERCEPTOR_ATTRIBUTE
+#endif
+
 using namespace __asan;  // NOLINT
 
 // This code has issues on OSX.
@@ -49,14 +56,14 @@ struct nothrow_t {};
 #endif  // __FreeBSD_version
 #endif  // SANITIZER_FREEBSD && SANITIZER_WORDSIZE == 32
 
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void *operator new(size_t size) { OPERATOR_NEW_BODY(FROM_NEW); }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void *operator new[](size_t size) { OPERATOR_NEW_BODY(FROM_NEW_BR); }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void *operator new(size_t size, std::nothrow_t const&)
 { OPERATOR_NEW_BODY(FROM_NEW); }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void *operator new[](size_t size, std::nothrow_t const&)
 { OPERATOR_NEW_BODY(FROM_NEW_BR); }
 
@@ -80,22 +87,32 @@ INTERCEPTOR(void *, _ZnamRKSt9nothrow_t, size_t size, std::nothrow_t const&) {
   asan_free(ptr, &stack, type);
 
 #if !SANITIZER_MAC
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void operator delete(void *ptr) throw() {
   OPERATOR_DELETE_BODY(FROM_NEW);
 }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void operator delete[](void *ptr) throw() {
   OPERATOR_DELETE_BODY(FROM_NEW_BR);
 }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void operator delete(void *ptr, std::nothrow_t const&) {
   OPERATOR_DELETE_BODY(FROM_NEW);
 }
-INTERCEPTOR_ATTRIBUTE
+CXX_OPERATOR_ATTRIBUTE
 void operator delete[](void *ptr, std::nothrow_t const&) {
   OPERATOR_DELETE_BODY(FROM_NEW_BR);
 }
+CXX_OPERATOR_ATTRIBUTE
+void operator delete(void *ptr, size_t size) throw() {
+  GET_STACK_TRACE_FREE;
+  asan_sized_free(ptr, size, &stack, FROM_NEW);
+}
+CXX_OPERATOR_ATTRIBUTE
+void operator delete[](void *ptr, size_t size) throw() {
+  GET_STACK_TRACE_FREE;
+  asan_sized_free(ptr, size, &stack, FROM_NEW_BR);
+}
 
 #else  // SANITIZER_MAC
 INTERCEPTOR(void, _ZdlPv, void *ptr) {
diff --git a/libsanitizer/asan/asan_poisoning.cc b/libsanitizer/asan/asan_poisoning.cc
index a532c5c4388..65f6cf0046e 100644
--- a/libsanitizer/asan/asan_poisoning.cc
+++ b/libsanitizer/asan/asan_poisoning.cc
@@ -225,6 +225,35 @@ void __sanitizer_unaligned_store64(uu64 *p, u64 x) {
   *p = x;
 }
 
+extern "C" SANITIZER_INTERFACE_ATTRIBUTE
+void __asan_poison_cxx_array_cookie(uptr p) {
+  if (SANITIZER_WORDSIZE != 64) return;
+  if (!flags()->poison_array_cookie) return;
+  uptr s = MEM_TO_SHADOW(p);
+  *reinterpret_cast<u8*>(s) = kAsanArrayCookieMagic;
+}
+
+extern "C" SANITIZER_INTERFACE_ATTRIBUTE
+uptr __asan_load_cxx_array_cookie(uptr *p) {
+  if (SANITIZER_WORDSIZE != 64) return *p;
+  if (!flags()->poison_array_cookie) return *p;
+  uptr s = MEM_TO_SHADOW(reinterpret_cast<uptr>(p));
+  u8 sval = *reinterpret_cast<u8*>(s);
+  if (sval == kAsanArrayCookieMagic) return *p;
+  // If sval is not kAsanArrayCookieMagic it can only be freed memory,
+  // which means that we are going to get double-free. So, return 0 to avoid
+  // infinite loop of destructors. We don't want to report a double-free here
+  // though, so print a warning just in case.
+  // CHECK_EQ(sval, kAsanHeapFreeMagic);
+  if (sval == kAsanHeapFreeMagic) {
+    Report("AddressSanitizer: loaded array cookie from free-d memory; "
+           "expect a double-free report\n");
+    return 0;
+  }
+  // FIXME: apparently it can be something else; need to find a reproducer.
+  return *p;
+}
+
 // This is a simplified version of __asan_(un)poison_memory_region, which
 // assumes that left border of region to be poisoned is properly aligned.
 static void PoisonAlignedStackMemory(uptr addr, uptr size, bool do_poison) {
diff --git a/libsanitizer/asan/asan_poisoning.h b/libsanitizer/asan/asan_poisoning.h
index 326d9ba1b67..9644b7d840e 100644
--- a/libsanitizer/asan/asan_poisoning.h
+++ b/libsanitizer/asan/asan_poisoning.h
@@ -33,7 +33,6 @@ void PoisonShadowPartialRightRedzone(uptr addr,
 ALWAYS_INLINE void FastPoisonShadow(uptr aligned_beg, uptr aligned_size,
                                     u8 value) {
   DCHECK(flags()->poison_heap);
-  uptr PageSize = GetPageSizeCached();
   uptr shadow_beg = MEM_TO_SHADOW(aligned_beg);
   uptr shadow_end = MEM_TO_SHADOW(
       aligned_beg + aligned_size - SHADOW_GRANULARITY) + 1;
@@ -46,8 +45,9 @@ ALWAYS_INLINE void FastPoisonShadow(uptr aligned_beg, uptr aligned_size,
       shadow_end - shadow_beg < common_flags()->clear_shadow_mmap_threshold) {
     REAL(memset)((void*)shadow_beg, value, shadow_end - shadow_beg);
   } else {
-    uptr page_beg = RoundUpTo(shadow_beg, PageSize);
-    uptr page_end = RoundDownTo(shadow_end, PageSize);
+    uptr page_size = GetPageSizeCached();
+    uptr page_beg = RoundUpTo(shadow_beg, page_size);
+    uptr page_end = RoundDownTo(shadow_end, page_size);
 
     if (page_beg >= page_end) {
       REAL(memset)((void *)shadow_beg, 0, shadow_end - shadow_beg);
diff --git a/libsanitizer/asan/asan_posix.cc b/libsanitizer/asan/asan_posix.cc
index 8f3798a2e59..4eabb74ba80 100644
--- a/libsanitizer/asan/asan_posix.cc
+++ b/libsanitizer/asan/asan_posix.cc
@@ -48,7 +48,7 @@ void AsanOnSIGSEGV(int, void *siginfo, void *context) {
       (code == si_SEGV_MAPERR || code == si_SEGV_ACCERR))
     ReportStackOverflow(pc, sp, bp, context, addr);
   else
-    ReportSIGSEGV(pc, sp, bp, context, addr);
+    ReportSIGSEGV("SEGV", pc, sp, bp, context, addr);
 }
 
 // ---------------------- TSD ---------------- {{{1
diff --git a/libsanitizer/asan/asan_preinit.cc b/libsanitizer/asan/asan_preinit.cc
index 31042401536..2ce1fb9b666 100644
--- a/libsanitizer/asan/asan_preinit.cc
+++ b/libsanitizer/asan/asan_preinit.cc
@@ -8,22 +8,12 @@
 // This file is a part of AddressSanitizer, an address sanity checker.
 //
 // Call __asan_init at the very early stage of process startup.
-// On Linux we use .preinit_array section (unless PIC macro is defined).
 //===----------------------------------------------------------------------===//
 #include "asan_internal.h"
 
-#if ASAN_USE_PREINIT_ARRAY && !defined(PIC)
-  // On Linux, we force __asan_init to be called before anyone else
-  // by placing it into .preinit_array section.
-  // FIXME: do we have anything like this on Mac?
+#if SANITIZER_CAN_USE_PREINIT_ARRAY
   // The symbol is called __local_asan_preinit, because it's not intended to be
   // exported.
   __attribute__((section(".preinit_array"), used))
   void (*__local_asan_preinit)(void) = __asan_init;
-#elif SANITIZER_WINDOWS && defined(_DLL)
-  // On Windows, when using dynamic CRT (/MD), we can put a pointer
-  // to __asan_init into the global list of C initializers.
-  // See crt0dat.c in the CRT sources for the details.
-  #pragma section(".CRT$XIB", long, read)  // NOLINT
-  __declspec(allocate(".CRT$XIB")) void (*__asan_preinit)() = __asan_init;
 #endif
diff --git a/libsanitizer/asan/asan_report.cc b/libsanitizer/asan/asan_report.cc
index d0a89b9677e..05622a12518 100644
--- a/libsanitizer/asan/asan_report.cc
+++ b/libsanitizer/asan/asan_report.cc
@@ -57,6 +57,7 @@ class Decorator: public __sanitizer::SanitizerCommonDecorator {
     switch (byte) {
       case kAsanHeapLeftRedzoneMagic:
       case kAsanHeapRightRedzoneMagic:
+      case kAsanArrayCookieMagic:
         return Red();
       case kAsanHeapFreeMagic:
         return Magenta();
@@ -141,6 +142,8 @@ static void PrintLegend(InternalScopedString *str) {
                   kAsanUserPoisonedMemoryMagic);
   PrintShadowByte(str, "  Container overflow:      ",
                   kAsanContiguousContainerOOBMagic);
+  PrintShadowByte(str, "  Array cookie:            ",
+                  kAsanArrayCookieMagic);
   PrintShadowByte(str, "  ASan internal:           ", kAsanInternalHeapMagic);
 }
 
@@ -195,7 +198,7 @@ static const char *MaybeDemangleGlobalName(const char *name) {
   else if (SANITIZER_WINDOWS && name[0] == '\01' && name[1] == '?')
     should_demangle = true;
 
-  return should_demangle ? Symbolizer::Get()->Demangle(name) : name;
+  return should_demangle ? Symbolizer::GetOrInit()->Demangle(name) : name;
 }
 
 // Check if the global is a zero-terminated ASCII string. If so, print it.
@@ -210,6 +213,26 @@ static void PrintGlobalNameIfASCII(InternalScopedString *str,
               (char *)g.beg);
 }
 
+static const char *GlobalFilename(const __asan_global &g) {
+  const char *res = g.module_name;
+  // Prefer the filename from source location, if is available.
+  if (g.location)
+    res = g.location->filename;
+  CHECK(res);
+  return res;
+}
+
+static void PrintGlobalLocation(InternalScopedString *str,
+                                const __asan_global &g) {
+  str->append("%s", GlobalFilename(g));
+  if (!g.location)
+    return;
+  if (g.location->line_no)
+    str->append(":%d", g.location->line_no);
+  if (g.location->column_no)
+    str->append(":%d", g.location->column_no);
+}
+
 bool DescribeAddressRelativeToGlobal(uptr addr, uptr size,
                                      const __asan_global &g) {
   static const uptr kMinimalDistanceFromAnotherGlobal = 64;
@@ -230,8 +253,10 @@ bool DescribeAddressRelativeToGlobal(uptr addr, uptr size,
     // Can it happen?
     str.append("%p is located %zd bytes inside", (void *)addr, addr - g.beg);
   }
-  str.append(" of global variable '%s' from '%s' (0x%zx) of size %zu\n",
-             MaybeDemangleGlobalName(g.name), g.module_name, g.beg, g.size);
+  str.append(" of global variable '%s' defined in '",
+             MaybeDemangleGlobalName(g.name));
+  PrintGlobalLocation(&str, g);
+  str.append("' (0x%zx) of size %zu\n", g.beg, g.size);
   str.append("%s", d.EndLocation());
   PrintGlobalNameIfASCII(&str, g);
   Printf("%s", str.data());
@@ -317,12 +342,27 @@ void PrintAccessAndVarIntersection(const char *var_name,
   Printf("%s", str.data());
 }
 
-struct StackVarDescr {
-  uptr beg;
-  uptr size;
-  const char *name_pos;
-  uptr name_len;
-};
+bool ParseFrameDescription(const char *frame_descr,
+                           InternalMmapVector<StackVarDescr> *vars) {
+  char *p;
+  uptr n_objects = (uptr)internal_simple_strtoll(frame_descr, &p, 10);
+  CHECK_GT(n_objects, 0);
+
+  for (uptr i = 0; i < n_objects; i++) {
+    uptr beg  = (uptr)internal_simple_strtoll(p, &p, 10);
+    uptr size = (uptr)internal_simple_strtoll(p, &p, 10);
+    uptr len  = (uptr)internal_simple_strtoll(p, &p, 10);
+    if (beg == 0 || size == 0 || *p != ' ') {
+      return false;
+    }
+    p++;
+    StackVarDescr var = {beg, size, p, len};
+    vars->push_back(var);
+    p += len;
+  }
+
+  return true;
+}
 
 bool DescribeAddressIfStack(uptr addr, uptr access_size) {
   AsanThread *t = FindThreadByStackAddress(addr);
@@ -364,32 +404,19 @@ bool DescribeAddressIfStack(uptr addr, uptr access_size) {
   alloca_stack.size = 1;
   Printf("%s", d.EndLocation());
   alloca_stack.Print();
+
+  InternalMmapVector<StackVarDescr> vars(16);
+  if (!ParseFrameDescription(frame_descr, &vars)) {
+    Printf("AddressSanitizer can't parse the stack frame "
+           "descriptor: |%s|\n", frame_descr);
+    // 'addr' is a stack address, so return true even if we can't parse frame
+    return true;
+  }
+  uptr n_objects = vars.size();
   // Report the number of stack objects.
-  char *p;
-  uptr n_objects = (uptr)internal_simple_strtoll(frame_descr, &p, 10);
-  CHECK_GT(n_objects, 0);
   Printf("  This frame has %zu object(s):\n", n_objects);
 
   // Report all objects in this frame.
-  InternalScopedBuffer<StackVarDescr> vars(n_objects);
-  for (uptr i = 0; i < n_objects; i++) {
-    uptr beg, size;
-    uptr len;
-    beg  = (uptr)internal_simple_strtoll(p, &p, 10);
-    size = (uptr)internal_simple_strtoll(p, &p, 10);
-    len  = (uptr)internal_simple_strtoll(p, &p, 10);
-    if (beg == 0 || size == 0 || *p != ' ') {
-      Printf("AddressSanitizer can't parse the stack frame "
-                 "descriptor: |%s|\n", frame_descr);
-      break;
-    }
-    p++;
-    vars[i].beg = beg;
-    vars[i].size = size;
-    vars[i].name_pos = p;
-    vars[i].name_len = len;
-    p += len;
-  }
   for (uptr i = 0; i < n_objects; i++) {
     buf[0] = 0;
     internal_strncat(buf, vars[i].name_pos,
@@ -401,8 +428,12 @@ bool DescribeAddressIfStack(uptr addr, uptr access_size) {
                                   prev_var_end, next_var_beg);
   }
   Printf("HINT: this may be a false positive if your program uses "
-             "some custom stack unwind mechanism or swapcontext\n"
-             "      (longjmp and C++ exceptions *are* supported)\n");
+         "some custom stack unwind mechanism or swapcontext\n");
+  if (SANITIZER_WINDOWS)
+    Printf("      (longjmp, SEH and C++ exceptions *are* supported)\n");
+  else
+    Printf("      (longjmp and C++ exceptions *are* supported)\n");
+
   DescribeThread(t);
   return true;
 }
@@ -531,7 +562,7 @@ class ScopedInErrorReport {
       // Do not print more than one report, otherwise they will mix up.
       // Error reporting functions shouldn't return at this situation, as
       // they are defined as no-return.
-      Report("AddressSanitizer: while reporting a bug found another one."
+      Report("AddressSanitizer: while reporting a bug found another one. "
                  "Ignoring.\n");
       u32 current_tid = GetCurrentTidOrInvalid();
       if (current_tid != reporting_thread_tid) {
@@ -578,8 +609,8 @@ void ReportStackOverflow(uptr pc, uptr sp, uptr bp, void *context, uptr addr) {
   Printf("%s", d.Warning());
   Report(
       "ERROR: AddressSanitizer: stack-overflow on address %p"
-      " (pc %p sp %p bp %p T%d)\n",
-      (void *)addr, (void *)pc, (void *)sp, (void *)bp,
+      " (pc %p bp %p sp %p T%d)\n",
+      (void *)addr, (void *)pc, (void *)bp, (void *)sp,
       GetCurrentTidOrInvalid());
   Printf("%s", d.EndWarning());
   GET_STACK_TRACE_SIGNAL(pc, bp, context);
@@ -587,15 +618,19 @@ void ReportStackOverflow(uptr pc, uptr sp, uptr bp, void *context, uptr addr) {
   ReportErrorSummary("stack-overflow", &stack);
 }
 
-void ReportSIGSEGV(uptr pc, uptr sp, uptr bp, void *context, uptr addr) {
+void ReportSIGSEGV(const char *description, uptr pc, uptr sp, uptr bp,
+                   void *context, uptr addr) {
   ScopedInErrorReport in_report;
   Decorator d;
   Printf("%s", d.Warning());
   Report(
-      "ERROR: AddressSanitizer: SEGV on unknown address %p"
-      " (pc %p sp %p bp %p T%d)\n",
-      (void *)addr, (void *)pc, (void *)sp, (void *)bp,
+      "ERROR: AddressSanitizer: %s on unknown address %p"
+      " (pc %p bp %p sp %p T%d)\n",
+      description, (void *)addr, (void *)pc, (void *)bp, (void *)sp,
       GetCurrentTidOrInvalid());
+  if (pc < GetPageSizeCached()) {
+    Report("Hint: pc points to the zero page.\n");
+  }
   Printf("%s", d.EndWarning());
   GET_STACK_TRACE_SIGNAL(pc, bp, context);
   stack.Print();
@@ -621,6 +656,30 @@ void ReportDoubleFree(uptr addr, StackTrace *free_stack) {
   ReportErrorSummary("double-free", &stack);
 }
 
+void ReportNewDeleteSizeMismatch(uptr addr, uptr delete_size,
+                                 StackTrace *free_stack) {
+  ScopedInErrorReport in_report;
+  Decorator d;
+  Printf("%s", d.Warning());
+  char tname[128];
+  u32 curr_tid = GetCurrentTidOrInvalid();
+  Report("ERROR: AddressSanitizer: new-delete-type-mismatch on %p in "
+         "thread T%d%s:\n",
+         addr, curr_tid,
+         ThreadNameWithParenthesis(curr_tid, tname, sizeof(tname)));
+  Printf("%s  object passed to delete has wrong type:\n", d.EndWarning());
+  Printf("  size of the allocated type:   %zd bytes;\n"
+         "  size of the deallocated type: %zd bytes.\n",
+         asan_mz_size(reinterpret_cast<void*>(addr)), delete_size);
+  CHECK_GT(free_stack->size, 0);
+  GET_STACK_TRACE_FATAL(free_stack->trace[0], free_stack->top_frame_bp);
+  stack.Print();
+  DescribeHeapAddress(addr, 1);
+  ReportErrorSummary("new-delete-type-mismatch", &stack);
+  Report("HINT: if you don't care about these warnings you may set "
+         "ASAN_OPTIONS=new_delete_type_mismatch=0\n");
+}
+
 void ReportFreeNotMalloced(uptr addr, StackTrace *free_stack) {
   ScopedInErrorReport in_report;
   Decorator d;
@@ -674,17 +733,17 @@ void ReportMallocUsableSizeNotOwned(uptr addr, StackTrace *stack) {
   ReportErrorSummary("bad-malloc_usable_size", stack);
 }
 
-void ReportAsanGetAllocatedSizeNotOwned(uptr addr, StackTrace *stack) {
+void ReportSanitizerGetAllocatedSizeNotOwned(uptr addr, StackTrace *stack) {
   ScopedInErrorReport in_report;
   Decorator d;
   Printf("%s", d.Warning());
   Report("ERROR: AddressSanitizer: attempting to call "
-             "__asan_get_allocated_size() for pointer which is "
+             "__sanitizer_get_allocated_size() for pointer which is "
              "not owned: %p\n", addr);
   Printf("%s", d.EndWarning());
   stack->Print();
   DescribeHeapAddress(addr, 1);
-  ReportErrorSummary("bad-__asan_get_allocated_size", stack);
+  ReportErrorSummary("bad-__sanitizer_get_allocated_size", stack);
 }
 
 void ReportStringFunctionMemoryRangesOverlap(
@@ -733,17 +792,36 @@ void ReportBadParamsToAnnotateContiguousContainer(uptr beg, uptr end,
   ReportErrorSummary("bad-__sanitizer_annotate_contiguous_container", stack);
 }
 
-void ReportODRViolation(const __asan_global *g1, const __asan_global *g2) {
+void ReportODRViolation(const __asan_global *g1, u32 stack_id1,
+                        const __asan_global *g2, u32 stack_id2) {
   ScopedInErrorReport in_report;
   Decorator d;
   Printf("%s", d.Warning());
   Report("ERROR: AddressSanitizer: odr-violation (%p):\n", g1->beg);
   Printf("%s", d.EndWarning());
-  Printf("  [1] size=%zd %s %s\n", g1->size, g1->name, g1->module_name);
-  Printf("  [2] size=%zd %s %s\n", g2->size, g2->name, g2->module_name);
+  InternalScopedString g1_loc(256), g2_loc(256);
+  PrintGlobalLocation(&g1_loc, *g1);
+  PrintGlobalLocation(&g2_loc, *g2);
+  Printf("  [1] size=%zd '%s' %s\n", g1->size,
+         MaybeDemangleGlobalName(g1->name), g1_loc.data());
+  Printf("  [2] size=%zd '%s' %s\n", g2->size,
+         MaybeDemangleGlobalName(g2->name), g2_loc.data());
+  if (stack_id1 && stack_id2) {
+    Printf("These globals were registered at these points:\n");
+    Printf("  [1]:\n");
+    uptr stack_size;
+    const uptr *stack_trace = StackDepotGet(stack_id1, &stack_size);
+    StackTrace::PrintStack(stack_trace, stack_size);
+    Printf("  [2]:\n");
+    stack_trace = StackDepotGet(stack_id2, &stack_size);
+    StackTrace::PrintStack(stack_trace, stack_size);
+  }
   Report("HINT: if you don't care about these warnings you may set "
          "ASAN_OPTIONS=detect_odr_violation=0\n");
-  ReportErrorSummary("odr-violation", g1->module_name, 0, g1->name);
+  InternalScopedString error_msg(256);
+  error_msg.append("odr-violation: global '%s' at %s",
+                   MaybeDemangleGlobalName(g1->name), g1_loc.data());
+  ReportErrorSummary(error_msg.data());
 }
 
 // ----------------------- CheckForInvalidPointerPair ----------- {{{1
@@ -831,6 +909,7 @@ void __asan_report_error(uptr pc, uptr bp, uptr sp, uptr addr, int is_write,
     switch (*shadow_addr) {
       case kAsanHeapLeftRedzoneMagic:
       case kAsanHeapRightRedzoneMagic:
+      case kAsanArrayCookieMagic:
         bug_descr = "heap-buffer-overflow";
         break;
       case kAsanHeapFreeMagic:
@@ -867,7 +946,7 @@ void __asan_report_error(uptr pc, uptr bp, uptr sp, uptr addr, int is_write,
   Decorator d;
   Printf("%s", d.Warning());
   Report("ERROR: AddressSanitizer: %s on address "
-             "%p at pc 0x%zx bp 0x%zx sp 0x%zx\n",
+             "%p at pc %p bp %p sp %p\n",
              bug_descr, (void*)addr, pc, bp, sp);
   Printf("%s", d.EndWarning());
 
@@ -899,7 +978,10 @@ void NOINLINE __asan_set_error_report_callback(void (*callback)(const char*)) {
 }
 
 void __asan_describe_address(uptr addr) {
+  // Thread registry must be locked while we're describing an address.
+  asanThreadRegistry().Lock();
   DescribeAddress(addr, 1);
+  asanThreadRegistry().Unlock();
 }
 
 extern "C" {
diff --git a/libsanitizer/asan/asan_report.h b/libsanitizer/asan/asan_report.h
index d9a0bca6423..4e81b9ca3a8 100644
--- a/libsanitizer/asan/asan_report.h
+++ b/libsanitizer/asan/asan_report.h
@@ -16,6 +16,13 @@
 
 namespace __asan {
 
+struct StackVarDescr {
+  uptr beg;
+  uptr size;
+  const char *name_pos;
+  uptr name_len;
+};
+
 // The following functions prints address description depending
 // on the memory type (shadow/heap/stack/global).
 void DescribeHeapAddress(uptr addr, uptr access_size);
@@ -23,6 +30,8 @@ bool DescribeAddressIfGlobal(uptr addr, uptr access_size);
 bool DescribeAddressRelativeToGlobal(uptr addr, uptr access_size,
                                      const __asan_global &g);
 bool DescribeAddressIfShadow(uptr addr);
+bool ParseFrameDescription(const char *frame_descr,
+                           InternalMmapVector<StackVarDescr> *vars);
 bool DescribeAddressIfStack(uptr addr, uptr access_size);
 // Determines memory type on its own.
 void DescribeAddress(uptr addr, uptr access_size);
@@ -32,8 +41,10 @@ void DescribeThread(AsanThreadContext *context);
 // Different kinds of error reports.
 void NORETURN
     ReportStackOverflow(uptr pc, uptr sp, uptr bp, void *context, uptr addr);
-void NORETURN
-    ReportSIGSEGV(uptr pc, uptr sp, uptr bp, void *context, uptr addr);
+void NORETURN ReportSIGSEGV(const char *description, uptr pc, uptr sp, uptr bp,
+                            void *context, uptr addr);
+void NORETURN ReportNewDeleteSizeMismatch(uptr addr, uptr delete_size,
+                                          StackTrace *free_stack);
 void NORETURN ReportDoubleFree(uptr addr, StackTrace *free_stack);
 void NORETURN ReportFreeNotMalloced(uptr addr, StackTrace *free_stack);
 void NORETURN ReportAllocTypeMismatch(uptr addr, StackTrace *free_stack,
@@ -41,8 +52,8 @@ void NORETURN ReportAllocTypeMismatch(uptr addr, StackTrace *free_stack,
                                       AllocType dealloc_type);
 void NORETURN ReportMallocUsableSizeNotOwned(uptr addr,
                                              StackTrace *stack);
-void NORETURN ReportAsanGetAllocatedSizeNotOwned(uptr addr,
-                                                 StackTrace *stack);
+void NORETURN
+ReportSanitizerGetAllocatedSizeNotOwned(uptr addr, StackTrace *stack);
 void NORETURN ReportStringFunctionMemoryRangesOverlap(
     const char *function, const char *offset1, uptr length1,
     const char *offset2, uptr length2, StackTrace *stack);
@@ -53,7 +64,8 @@ ReportBadParamsToAnnotateContiguousContainer(uptr beg, uptr end, uptr old_mid,
                                              uptr new_mid, StackTrace *stack);
 
 void NORETURN
-ReportODRViolation(const __asan_global *g1, const __asan_global *g2);
+ReportODRViolation(const __asan_global *g1, u32 stack_id1,
+                   const __asan_global *g2, u32 stack_id2);
 
 // Mac-specific errors and warnings.
 void WarnMacFreeUnallocated(
diff --git a/libsanitizer/asan/asan_rtl.cc b/libsanitizer/asan/asan_rtl.cc
index 00b4b95868e..8fccc8da967 100644
--- a/libsanitizer/asan/asan_rtl.cc
+++ b/libsanitizer/asan/asan_rtl.cc
@@ -171,11 +171,6 @@ static void ParseFlagsFromString(Flags *f, const char *str) {
       "If set, prints ASan exit stats even after program terminates "
       "successfully.");
 
-  ParseFlag(str, &f->disable_core, "disable_core",
-      "Disable core dumping. By default, disable_core=1 on 64-bit to avoid "
-      "dumping a 16T+ core file. "
-      "Ignored on OSes that don't dump core by default.");
-
   ParseFlag(str, &f->allow_reexec, "allow_reexec",
       "Allow the tool to re-exec the program. This may interfere badly with "
       "the debugger.");
@@ -189,6 +184,9 @@ static void ParseFlagsFromString(Flags *f, const char *str) {
       "Poison (or not) the heap memory on [de]allocation. Zero value is useful "
       "for benchmarking the allocator or instrumentator.");
 
+  ParseFlag(str, &f->poison_array_cookie, "poison_array_cookie",
+      "Poison (or not) the array cookie after operator new[].");
+
   ParseFlag(str, &f->poison_partial, "poison_partial",
       "If true, poison partially addressable 8-byte aligned words "
       "(default=true). This flag affects heap and global buffers, but not "
@@ -196,6 +194,10 @@ static void ParseFlagsFromString(Flags *f, const char *str) {
 
   ParseFlag(str, &f->alloc_dealloc_mismatch, "alloc_dealloc_mismatch",
       "Report errors on malloc/delete, new/free, new/delete[], etc.");
+
+  ParseFlag(str, &f->new_delete_type_mismatch, "new_delete_type_mismatch",
+      "Report errors on mismatch betwen size of new and delete.");
+
   ParseFlag(str, &f->strict_memcmp, "strict_memcmp",
       "If true, assume that memcmp(p1, p2, n) always reads n bytes before "
       "comparing p1 and p2.");
@@ -262,21 +264,23 @@ void InitializeFlags(Flags *f, const char *env) {
   f->print_stats = false;
   f->print_legend = true;
   f->atexit = false;
-  f->disable_core = (SANITIZER_WORDSIZE == 64);
   f->allow_reexec = true;
   f->print_full_thread_history = true;
   f->poison_heap = true;
+  f->poison_array_cookie = true;
   f->poison_partial = true;
   // Turn off alloc/dealloc mismatch checker on Mac and Windows for now.
   // https://code.google.com/p/address-sanitizer/issues/detail?id=131
   // https://code.google.com/p/address-sanitizer/issues/detail?id=309
   // TODO(glider,timurrrr): Fix known issues and enable this back.
   f->alloc_dealloc_mismatch = (SANITIZER_MAC == 0) && (SANITIZER_WINDOWS == 0);
+  f->new_delete_type_mismatch = true;
   f->strict_memcmp = true;
   f->strict_init_order = false;
   f->start_deactivated = false;
   f->detect_invalid_pointer_pairs = 0;
   f->detect_container_overflow = true;
+  f->detect_odr_violation = 2;
 
   // Override from compile definition.
   ParseFlagsFromString(f, MaybeUseAsanDefaultOptionsCompileDefinition());
@@ -456,13 +460,6 @@ static NOINLINE void force_interface_symbols() {
     case 15: __asan_set_error_report_callback(0); break;
     case 16: __asan_handle_no_return(); break;
     case 17: __asan_address_is_poisoned(0); break;
-    case 18: __asan_get_allocated_size(0); break;
-    case 19: __asan_get_current_allocated_bytes(); break;
-    case 20: __asan_get_estimated_allocated_size(0); break;
-    case 21: __asan_get_free_bytes(); break;
-    case 22: __asan_get_heap_size(); break;
-    case 23: __asan_get_ownership(0); break;
-    case 24: __asan_get_unmapped_bytes(); break;
     case 25: __asan_poison_memory_region(0, 0); break;
     case 26: __asan_unpoison_memory_region(0, 0); break;
     case 27: __asan_set_error_exit_code(0); break;
@@ -593,6 +590,11 @@ static void AsanInitInternal() {
 
   InitializeAsanInterceptors();
 
+  // Enable system log ("adb logcat") on Android.
+  // Doing this before interceptors are initialized crashes in:
+  // AsanInitInternal -> android_log_write -> __interceptor_strcmp
+  AndroidLogInit();
+
   ReplaceSystemMalloc();
 
   uptr shadow_start = kLowShadowBeg;
@@ -601,7 +603,8 @@ static void AsanInitInternal() {
   bool full_shadow_is_available =
       MemoryRangeIsAvailable(shadow_start, kHighShadowEnd);
 
-#if SANITIZER_LINUX && defined(__x86_64__) && !ASAN_FIXED_MAPPING
+#if SANITIZER_LINUX && defined(__x86_64__) && defined(_LP64) &&                \
+    !ASAN_FIXED_MAPPING
   if (!full_shadow_is_available) {
     kMidMemBeg = kLowMemEnd < 0x3000000000ULL ? 0x3000000000ULL : 0;
     kMidMemEnd = kLowMemEnd < 0x3000000000ULL ? 0x4fffffffffULL : 0;
@@ -611,9 +614,7 @@ static void AsanInitInternal() {
   if (common_flags()->verbosity)
     PrintAddressSpaceLayout();
 
-  if (flags()->disable_core) {
-    DisableCoreDumper();
-  }
+  DisableCoreDumperIfNecessary();
 
   if (full_shadow_is_available) {
     // mmap the low shadow plus at least one page at the left.
@@ -648,12 +649,8 @@ static void AsanInitInternal() {
   AsanTSDInit(PlatformTSDDtor);
   InstallDeadlySignalHandlers(AsanOnSIGSEGV);
 
-  // Allocator should be initialized before starting external symbolizer, as
-  // fork() on Mac locks the allocator.
   InitializeAllocator();
 
-  Symbolizer::Init(common_flags()->external_symbolizer_path);
-
   // On Linux AsanThread::ThreadStart() calls malloc() that's why asan_inited
   // should be set to 1 prior to initializing the threads.
   asan_inited = 1;
@@ -682,7 +679,7 @@ static void AsanInitInternal() {
   SanitizerInitializeUnwinder();
 
 #if CAN_SANITIZE_LEAKS
-  __lsan::InitCommonLsan();
+  __lsan::InitCommonLsan(false);
   if (common_flags()->detect_leaks && common_flags()->leak_check_at_exit) {
     Atexit(__lsan::DoLeakCheck);
   }
diff --git a/libsanitizer/asan/asan_stats.cc b/libsanitizer/asan/asan_stats.cc
index 71c8582e81c..fbd636ea558 100644
--- a/libsanitizer/asan/asan_stats.cc
+++ b/libsanitizer/asan/asan_stats.cc
@@ -13,6 +13,7 @@
 #include "asan_internal.h"
 #include "asan_stats.h"
 #include "asan_thread.h"
+#include "sanitizer_common/sanitizer_allocator_interface.h"
 #include "sanitizer_common/sanitizer_mutex.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
 
@@ -127,8 +128,8 @@ static void PrintAccumulatedStats() {
   BlockingMutexLock lock(&print_lock);
   stats.Print();
   StackDepotStats *stack_depot_stats = StackDepotGetStats();
-  Printf("Stats: StackDepot: %zd ids; %zdM mapped\n",
-         stack_depot_stats->n_uniq_ids, stack_depot_stats->mapped >> 20);
+  Printf("Stats: StackDepot: %zd ids; %zdM allocated\n",
+         stack_depot_stats->n_uniq_ids, stack_depot_stats->allocated >> 20);
   PrintInternalAllocatorStats();
 }
 
@@ -137,7 +138,7 @@ static void PrintAccumulatedStats() {
 // ---------------------- Interface ---------------- {{{1
 using namespace __asan;  // NOLINT
 
-uptr __asan_get_current_allocated_bytes() {
+uptr __sanitizer_get_current_allocated_bytes() {
   AsanStats stats;
   GetAccumulatedStats(&stats);
   uptr malloced = stats.malloced;
@@ -147,13 +148,13 @@ uptr __asan_get_current_allocated_bytes() {
   return (malloced > freed) ? malloced - freed : 1;
 }
 
-uptr __asan_get_heap_size() {
+uptr __sanitizer_get_heap_size() {
   AsanStats stats;
   GetAccumulatedStats(&stats);
   return stats.mmaped - stats.munmaped;
 }
 
-uptr __asan_get_free_bytes() {
+uptr __sanitizer_get_free_bytes() {
   AsanStats stats;
   GetAccumulatedStats(&stats);
   uptr total_free = stats.mmaped
@@ -167,7 +168,7 @@ uptr __asan_get_free_bytes() {
   return (total_free > total_used) ? total_free - total_used : 1;
 }
 
-uptr __asan_get_unmapped_bytes() {
+uptr __sanitizer_get_unmapped_bytes() {
   return 0;
 }
 
diff --git a/libsanitizer/asan/asan_thread.cc b/libsanitizer/asan/asan_thread.cc
index df85858a67b..87074065633 100644
--- a/libsanitizer/asan/asan_thread.cc
+++ b/libsanitizer/asan/asan_thread.cc
@@ -139,7 +139,10 @@ FakeStack *AsanThread::AsyncSignalSafeLazyInitFakeStack() {
 }
 
 void AsanThread::Init() {
+  fake_stack_ = 0;  // Will be initialized lazily if needed.
+  CHECK_EQ(this->stack_size(), 0U);
   SetThreadStackAndTls();
+  CHECK_GT(this->stack_size(), 0U);
   CHECK(AddrIsInMem(stack_bottom_));
   CHECK(AddrIsInMem(stack_top_ - 1));
   ClearShadowForThreadStackAndTLS();
@@ -147,7 +150,6 @@ void AsanThread::Init() {
   VReport(1, "T%d: stack [%p,%p) size 0x%zx; local=%p\n", tid(),
           (void *)stack_bottom_, (void *)stack_top_, stack_top_ - stack_bottom_,
           &local);
-  fake_stack_ = 0;  // Will be initialized lazily if needed.
   AsanPlatformThreadInit();
 }
 
diff --git a/libsanitizer/asan/asan_win.cc b/libsanitizer/asan/asan_win.cc
index 03d45e3839b..b0028763b11 100644
--- a/libsanitizer/asan/asan_win.cc
+++ b/libsanitizer/asan/asan_win.cc
@@ -19,6 +19,7 @@
 
 #include "asan_interceptors.h"
 #include "asan_internal.h"
+#include "asan_report.h"
 #include "asan_thread.h"
 #include "sanitizer_common/sanitizer_libc.h"
 #include "sanitizer_common/sanitizer_mutex.h"
@@ -68,7 +69,7 @@ void *AsanDoesNotSupportStaticLinkage() {
   return 0;
 }
 
-void AsanCheckDynamicRTPrereqs() { UNIMPLEMENTED(); }
+void AsanCheckDynamicRTPrereqs() {}
 
 void AsanCheckIncompatibleRT() {}
 
@@ -84,6 +85,67 @@ void AsanOnSIGSEGV(int, void *siginfo, void *context) {
   UNIMPLEMENTED();
 }
 
+static LPTOP_LEVEL_EXCEPTION_FILTER default_seh_handler;
+
+static long WINAPI SEHHandler(EXCEPTION_POINTERS *info) {
+  EXCEPTION_RECORD *exception_record = info->ExceptionRecord;
+  CONTEXT *context = info->ContextRecord;
+  uptr pc = (uptr)exception_record->ExceptionAddress;
+#ifdef _WIN64
+  uptr bp = (uptr)context->Rbp, sp = (uptr)context->Rsp;
+#else
+  uptr bp = (uptr)context->Ebp, sp = (uptr)context->Esp;
+#endif
+
+  if (exception_record->ExceptionCode == EXCEPTION_ACCESS_VIOLATION ||
+      exception_record->ExceptionCode == EXCEPTION_IN_PAGE_ERROR) {
+    const char *description =
+        (exception_record->ExceptionCode == EXCEPTION_ACCESS_VIOLATION)
+            ? "access-violation"
+            : "in-page-error";
+    uptr access_addr = exception_record->ExceptionInformation[1];
+    ReportSIGSEGV(description, pc, sp, bp, context, access_addr);
+  }
+
+  // FIXME: Handle EXCEPTION_STACK_OVERFLOW here.
+
+  return default_seh_handler(info);
+}
+
+// We want to install our own exception handler (EH) to print helpful reports
+// on access violations and whatnot.  Unfortunately, the CRT initializers assume
+// they are run before any user code and drop any previously-installed EHs on
+// the floor, so we can't install our handler inside __asan_init.
+// (See crt0dat.c in the CRT sources for the details)
+//
+// Things get even more complicated with the dynamic runtime, as it finishes its
+// initialization before the .exe module CRT begins to initialize.
+//
+// For the static runtime (-MT), it's enough to put a callback to
+// __asan_set_seh_filter in the last section for C initializers.
+//
+// For the dynamic runtime (-MD), we want link the same
+// asan_dynamic_runtime_thunk.lib to all the modules, thus __asan_set_seh_filter
+// will be called for each instrumented module.  This ensures that at least one
+// __asan_set_seh_filter call happens after the .exe module CRT is initialized.
+extern "C" SANITIZER_INTERFACE_ATTRIBUTE
+int __asan_set_seh_filter() {
+  // We should only store the previous handler if it's not our own handler in
+  // order to avoid loops in the EH chain.
+  auto prev_seh_handler = SetUnhandledExceptionFilter(SEHHandler);
+  if (prev_seh_handler != &SEHHandler)
+    default_seh_handler = prev_seh_handler;
+  return 0;
+}
+
+#if !ASAN_DYNAMIC
+// Put a pointer to __asan_set_seh_filter at the end of the global list
+// of C initializers, after the default EH is set by the CRT.
+#pragma section(".CRT$XIZ", long, read)  // NOLINT
+static __declspec(allocate(".CRT$XIZ"))
+    int (*__intercept_seh)() = __asan_set_seh_filter;
+#endif
+
 }  // namespace __asan
 
 #endif  // _WIN32
diff --git a/libsanitizer/asan/asan_dll_thunk.cc b/libsanitizer/asan/asan_win_dll_thunk.cc
index 5bed39ac066..6adb7d2e942 100644
--- a/libsanitizer/asan/asan_dll_thunk.cc
+++ b/libsanitizer/asan/asan_win_dll_thunk.cc
@@ -1,4 +1,4 @@
-//===-- asan_dll_thunk.cc -------------------------------------------------===//
+//===-- asan_win_dll_thunk.cc ---------------------------------------------===//
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
@@ -18,9 +18,10 @@
 // Using #ifdef rather than relying on Makefiles etc.
 // simplifies the build procedure.
 #ifdef ASAN_DLL_THUNK
+#include "asan_init_version.h"
 #include "sanitizer_common/sanitizer_interception.h"
 
-// ----------------- Helper functions and macros --------------------- {{{1
+// ---------- Function interception helper functions and macros ----------- {{{1
 extern "C" {
 void *__stdcall GetModuleHandleA(const char *module_name);
 void *__stdcall GetProcAddress(void *module, const char *proc_name);
@@ -34,68 +35,143 @@ static void *getRealProcAddressOrDie(const char *name) {
   return ret;
 }
 
+// We need to intercept some functions (e.g. ASan interface, memory allocator --
+// let's call them "hooks") exported by the DLL thunk and forward the hooks to
+// the runtime in the main module.
+// However, we don't want to keep two lists of these hooks.
+// To avoid that, the list of hooks should be defined using the
+// INTERCEPT_WHEN_POSSIBLE macro. Then, all these hooks can be intercepted
+// at once by calling INTERCEPT_HOOKS().
+
+// Use macro+template magic to automatically generate the list of hooks.
+// Each hook at line LINE defines a template class with a static
+// FunctionInterceptor<LINE>::Execute() method intercepting the hook.
+// The default implementation of FunctionInterceptor<LINE> is to call
+// the Execute() method corresponding to the previous line.
+template<int LINE>
+struct FunctionInterceptor {
+  static void Execute() { FunctionInterceptor<LINE-1>::Execute(); }
+};
+
+// There shouldn't be any hooks with negative definition line number.
+template<>
+struct FunctionInterceptor<0> {
+  static void Execute() {}
+};
+
+#define INTERCEPT_WHEN_POSSIBLE(main_function, dll_function)                   \
+  template<> struct FunctionInterceptor<__LINE__> {                            \
+    static void Execute() {                                                    \
+      void *wrapper = getRealProcAddressOrDie(main_function);                  \
+      if (!__interception::OverrideFunction((uptr)dll_function,                \
+                                            (uptr)wrapper, 0))                 \
+        abort();                                                               \
+      FunctionInterceptor<__LINE__-1>::Execute();                              \
+    }                                                                          \
+  };
+
+// Special case of hooks -- ASan own interface functions.  Those are only called
+// after __asan_init, thus an empty implementation is sufficient.
+#define INTERFACE_FUNCTION(name)                                               \
+  extern "C" __declspec(noinline) void name() {                                \
+    volatile int prevent_icf = (__LINE__ << 8); (void)prevent_icf;             \
+    __debugbreak();                                                            \
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name)
+
+// INTERCEPT_HOOKS must be used after the last INTERCEPT_WHEN_POSSIBLE.
+#define INTERCEPT_HOOKS FunctionInterceptor<__LINE__>::Execute
+
+// We can't define our own version of strlen etc. because that would lead to
+// link-time or even type mismatch errors.  Instead, we can declare a function
+// just to be able to get its address.  Me may miss the first few calls to the
+// functions since it can be called before __asan_init, but that would lead to
+// false negatives in the startup code before user's global initializers, which
+// isn't a big deal.
+#define INTERCEPT_LIBRARY_FUNCTION(name)                                       \
+  extern "C" void name();                                                      \
+  INTERCEPT_WHEN_POSSIBLE(WRAPPER_NAME(name), name)
+
+// Disable compiler warnings that show up if we declare our own version
+// of a compiler intrinsic (e.g. strlen).
+#pragma warning(disable: 4391)
+#pragma warning(disable: 4392)
+
+static void InterceptHooks();
+// }}}
+
+// ---------- Function wrapping helpers ----------------------------------- {{{1
 #define WRAP_V_V(name)                                                         \
   extern "C" void name() {                                                     \
     typedef void (*fntype)();                                                  \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     fn();                                                                      \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_V_W(name)                                                         \
   extern "C" void name(void *arg) {                                            \
     typedef void (*fntype)(void *arg);                                         \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     fn(arg);                                                                   \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_V_WW(name)                                                        \
   extern "C" void name(void *arg1, void *arg2) {                               \
     typedef void (*fntype)(void *, void *);                                    \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     fn(arg1, arg2);                                                            \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_V_WWW(name)                                                       \
   extern "C" void name(void *arg1, void *arg2, void *arg3) {                   \
     typedef void *(*fntype)(void *, void *, void *);                           \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     fn(arg1, arg2, arg3);                                                      \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_V(name)                                                         \
   extern "C" void *name() {                                                    \
     typedef void *(*fntype)();                                                 \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn();                                                               \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_W(name)                                                         \
   extern "C" void *name(void *arg) {                                           \
     typedef void *(*fntype)(void *arg);                                        \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg);                                                            \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_WW(name)                                                        \
   extern "C" void *name(void *arg1, void *arg2) {                              \
     typedef void *(*fntype)(void *, void *);                                   \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg1, arg2);                                                     \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_WWW(name)                                                       \
   extern "C" void *name(void *arg1, void *arg2, void *arg3) {                  \
     typedef void *(*fntype)(void *, void *, void *);                           \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg1, arg2, arg3);                                               \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_WWWW(name)                                                      \
   extern "C" void *name(void *arg1, void *arg2, void *arg3, void *arg4) {      \
     typedef void *(*fntype)(void *, void *, void *, void *);                   \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg1, arg2, arg3, arg4);                                         \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_WWWWW(name)                                                     \
   extern "C" void *name(void *arg1, void *arg2, void *arg3, void *arg4,        \
@@ -103,7 +179,8 @@ static void *getRealProcAddressOrDie(const char *name) {
     typedef void *(*fntype)(void *, void *, void *, void *, void *);           \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg1, arg2, arg3, arg4, arg5);                                   \
-  }
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 
 #define WRAP_W_WWWWWW(name)                                                    \
   extern "C" void *name(void *arg1, void *arg2, void *arg3, void *arg4,        \
@@ -111,48 +188,8 @@ static void *getRealProcAddressOrDie(const char *name) {
     typedef void *(*fntype)(void *, void *, void *, void *, void *, void *);   \
     static fntype fn = (fntype)getRealProcAddressOrDie(#name);                 \
     return fn(arg1, arg2, arg3, arg4, arg5, arg6);                             \
-  }
-// }}}
-
-// --------- Interface interception helper functions and macros ----------- {{{1
-// We need to intercept the ASan interface exported by the DLL thunk and forward
-// all the functions to the runtime in the main module.
-// However, we don't want to keep two lists of interface functions.
-// To avoid that, the list of interface functions should be defined using the
-// INTERFACE_FUNCTION macro. Then, all the interface can be intercepted at once
-// by calling INTERCEPT_ASAN_INTERFACE().
-
-// Use macro+template magic to automatically generate the list of interface
-// functions.  Each interface function at line LINE defines a template class
-// with a static InterfaceInteceptor<LINE>::Execute() method intercepting the
-// function.  The default implementation of InterfaceInteceptor<LINE> is to call
-// the Execute() method corresponding to the previous line.
-template<int LINE>
-struct InterfaceInteceptor {
-  static void Execute() { InterfaceInteceptor<LINE-1>::Execute(); }
-};
-
-// There shouldn't be any interface function with negative line number.
-template<>
-struct InterfaceInteceptor<0> {
-  static void Execute() {}
-};
-
-#define INTERFACE_FUNCTION(name)                                               \
-  extern "C" void name() { __debugbreak(); }                                   \
-  template<> struct InterfaceInteceptor<__LINE__> {                            \
-    static void Execute() {                                                    \
-      void *wrapper = getRealProcAddressOrDie(#name);                          \
-      if (!__interception::OverrideFunction((uptr)name, (uptr)wrapper, 0))     \
-        abort();                                                               \
-      InterfaceInteceptor<__LINE__-1>::Execute();                              \
-    }                                                                          \
-  };
-
-// INTERCEPT_ASAN_INTERFACE must be used after the last INTERFACE_FUNCTION.
-#define INTERCEPT_ASAN_INTERFACE InterfaceInteceptor<__LINE__>::Execute
-
-static void InterceptASanInterface();
+  }                                                                            \
+  INTERCEPT_WHEN_POSSIBLE(#name, name);
 // }}}
 
 // ----------------- ASan own interface functions --------------------
@@ -165,17 +202,18 @@ extern "C" {
 
   // Manually wrap __asan_init as we need to initialize
   // __asan_option_detect_stack_use_after_return afterwards.
-  void __asan_init_v3() {
+  void __asan_init() {
     typedef void (*fntype)();
     static fntype fn = 0;
+    // __asan_init is expected to be called by only one thread.
     if (fn) return;
 
-    fn = (fntype)getRealProcAddressOrDie("__asan_init_v3");
+    fn = (fntype)getRealProcAddressOrDie(__asan_init_name);
     fn();
     __asan_option_detect_stack_use_after_return =
         (__asan_should_detect_stack_use_after_return() != 0);
 
-    InterceptASanInterface();
+    InterceptHooks();
   }
 }
 
@@ -195,6 +233,20 @@ INTERFACE_FUNCTION(__asan_report_load8)
 INTERFACE_FUNCTION(__asan_report_load16)
 INTERFACE_FUNCTION(__asan_report_load_n)
 
+INTERFACE_FUNCTION(__asan_store1)
+INTERFACE_FUNCTION(__asan_store2)
+INTERFACE_FUNCTION(__asan_store4)
+INTERFACE_FUNCTION(__asan_store8)
+INTERFACE_FUNCTION(__asan_store16)
+INTERFACE_FUNCTION(__asan_storeN)
+
+INTERFACE_FUNCTION(__asan_load1)
+INTERFACE_FUNCTION(__asan_load2)
+INTERFACE_FUNCTION(__asan_load4)
+INTERFACE_FUNCTION(__asan_load8)
+INTERFACE_FUNCTION(__asan_load16)
+INTERFACE_FUNCTION(__asan_loadN)
+
 INTERFACE_FUNCTION(__asan_memcpy);
 INTERFACE_FUNCTION(__asan_memset);
 INTERFACE_FUNCTION(__asan_memmove);
@@ -211,6 +263,9 @@ INTERFACE_FUNCTION(__asan_unpoison_stack_memory)
 INTERFACE_FUNCTION(__asan_poison_memory_region)
 INTERFACE_FUNCTION(__asan_unpoison_memory_region)
 
+INTERFACE_FUNCTION(__asan_address_is_poisoned)
+INTERFACE_FUNCTION(__asan_region_is_poisoned)
+
 INTERFACE_FUNCTION(__asan_get_current_fake_stack)
 INTERFACE_FUNCTION(__asan_addr_is_in_fake_stack)
 
@@ -237,6 +292,8 @@ INTERFACE_FUNCTION(__asan_stack_free_8)
 INTERFACE_FUNCTION(__asan_stack_free_9)
 INTERFACE_FUNCTION(__asan_stack_free_10)
 
+INTERFACE_FUNCTION(__sanitizer_cov_module_init)
+
 // TODO(timurrrr): Add more interface functions on the as-needed basis.
 
 // ----------------- Memory allocation functions ---------------------
@@ -263,8 +320,55 @@ WRAP_W_W(_expand_dbg)
 
 // TODO(timurrrr): Do we need to add _Crt* stuff here? (see asan_malloc_win.cc).
 
-void InterceptASanInterface() {
-  INTERCEPT_ASAN_INTERFACE();
+INTERCEPT_LIBRARY_FUNCTION(atoi);
+INTERCEPT_LIBRARY_FUNCTION(atol);
+INTERCEPT_LIBRARY_FUNCTION(_except_handler3);
+
+// _except_handler4 checks -GS cookie which is different for each module, so we
+// can't use INTERCEPT_LIBRARY_FUNCTION(_except_handler4).
+INTERCEPTOR(int, _except_handler4, void *a, void *b, void *c, void *d) {
+  __asan_handle_no_return();
+  return REAL(_except_handler4)(a, b, c, d);
+}
+
+INTERCEPT_LIBRARY_FUNCTION(frexp);
+INTERCEPT_LIBRARY_FUNCTION(longjmp);
+INTERCEPT_LIBRARY_FUNCTION(memchr);
+INTERCEPT_LIBRARY_FUNCTION(memcmp);
+INTERCEPT_LIBRARY_FUNCTION(memcpy);
+INTERCEPT_LIBRARY_FUNCTION(memmove);
+INTERCEPT_LIBRARY_FUNCTION(memset);
+INTERCEPT_LIBRARY_FUNCTION(strcat);  // NOLINT
+INTERCEPT_LIBRARY_FUNCTION(strchr);
+INTERCEPT_LIBRARY_FUNCTION(strcmp);
+INTERCEPT_LIBRARY_FUNCTION(strcpy);  // NOLINT
+INTERCEPT_LIBRARY_FUNCTION(strlen);
+INTERCEPT_LIBRARY_FUNCTION(strncat);
+INTERCEPT_LIBRARY_FUNCTION(strncmp);
+INTERCEPT_LIBRARY_FUNCTION(strncpy);
+INTERCEPT_LIBRARY_FUNCTION(strnlen);
+INTERCEPT_LIBRARY_FUNCTION(strtol);
+INTERCEPT_LIBRARY_FUNCTION(wcslen);
+
+// Must be after all the interceptor declarations due to the way INTERCEPT_HOOKS
+// is defined.
+void InterceptHooks() {
+  INTERCEPT_HOOKS();
+  INTERCEPT_FUNCTION(_except_handler4);
+}
+
+// We want to call __asan_init before C/C++ initializers/constructors are
+// executed, otherwise functions like memset might be invoked.
+// For some strange reason, merely linking in asan_preinit.cc doesn't work
+// as the callback is never called...  Is link.exe doing something too smart?
+
+// In DLLs, the callbacks are expected to return 0,
+// otherwise CRT initialization fails.
+static int call_asan_init() {
+  __asan_init();
+  return 0;
 }
+#pragma section(".CRT$XIB", long, read)  // NOLINT
+__declspec(allocate(".CRT$XIB")) int (*__asan_preinit)() = call_asan_init;
 
 #endif // ASAN_DLL_THUNK
diff --git a/libsanitizer/asan/asan_win_dynamic_runtime_thunk.cc b/libsanitizer/asan/asan_win_dynamic_runtime_thunk.cc
new file mode 100644
index 00000000000..1b59677eeff
--- /dev/null
+++ b/libsanitizer/asan/asan_win_dynamic_runtime_thunk.cc
@@ -0,0 +1,50 @@
+//===-- asan_win_uar_thunk.cc ---------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of AddressSanitizer, an address sanity checker.
+//
+// This file defines things that need to be present in the application modules
+// to interact with the ASan DLL runtime correctly and can't be implemented
+// using the default "import library" generated when linking the DLL RTL.
+//
+// This includes:
+//  - forwarding the detect_stack_use_after_return runtime option
+//  - installing a custom SEH handler
+//
+//===----------------------------------------------------------------------===//
+
+// Only compile this code when buidling asan_dynamic_runtime_thunk.lib
+// Using #ifdef rather than relying on Makefiles etc.
+// simplifies the build procedure.
+#ifdef ASAN_DYNAMIC_RUNTIME_THUNK
+extern "C" {
+__declspec(dllimport) int __asan_set_seh_filter();
+__declspec(dllimport) int __asan_should_detect_stack_use_after_return();
+
+// Define a copy of __asan_option_detect_stack_use_after_return that should be
+// used when linking an MD runtime with a set of object files on Windows.
+//
+// The ASan MD runtime dllexports '__asan_option_detect_stack_use_after_return',
+// so normally we would just dllimport it.  Unfortunately, the dllimport
+// attribute adds __imp_ prefix to the symbol name of a variable.
+// Since in general we don't know if a given TU is going to be used
+// with a MT or MD runtime and we don't want to use ugly __imp_ names on Windows
+// just to work around this issue, let's clone the a variable that is
+// constant after initialization anyways.
+int __asan_option_detect_stack_use_after_return =
+    __asan_should_detect_stack_use_after_return();
+
+// Set the ASan-specific SEH handler at the end of CRT initialization of each
+// module (see asan_win.cc for the details).
+//
+// Unfortunately, putting a pointer to __asan_set_seh_filter into
+// __asan_intercept_seh gets optimized out, so we have to use an extra function.
+static int SetSEHFilter() { return __asan_set_seh_filter(); }
+#pragma section(".CRT$XIZ", long, read)  // NOLINT
+__declspec(allocate(".CRT$XIZ")) int (*__asan_seh_interceptor)() = SetSEHFilter;
+}
+#endif // ASAN_DYNAMIC_RUNTIME_THUNK
diff --git a/libsanitizer/asan/libtool-version b/libsanitizer/asan/libtool-version
index 9a16cf57844..f18f40737c7 100644
--- a/libsanitizer/asan/libtool-version
+++ b/libsanitizer/asan/libtool-version
@@ -3,4 +3,4 @@
 # a separate file so that version updates don't involve re-running
 # automake.
 # CURRENT:REVISION:AGE
-1:0:0
+2:0:0
diff --git a/libsanitizer/include/sanitizer/allocator_interface.h b/libsanitizer/include/sanitizer/allocator_interface.h
new file mode 100644
index 00000000000..97a72a258eb
--- /dev/null
+++ b/libsanitizer/include/sanitizer/allocator_interface.h
@@ -0,0 +1,64 @@
+//===-- allocator_interface.h ---------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Public interface header for allocator used in sanitizers (ASan/TSan/MSan).
+//===----------------------------------------------------------------------===//
+#ifndef SANITIZER_ALLOCATOR_INTERFACE_H
+#define SANITIZER_ALLOCATOR_INTERFACE_H
+
+#include <stddef.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+  /* Returns the estimated number of bytes that will be reserved by allocator
+     for request of "size" bytes. If allocator can't allocate that much
+     memory, returns the maximal possible allocation size, otherwise returns
+     "size". */
+  size_t __sanitizer_get_estimated_allocated_size(size_t size);
+
+  /* Returns true if p was returned by the allocator and
+     is not yet freed. */
+  int __sanitizer_get_ownership(const volatile void *p);
+
+  /* Returns the number of bytes reserved for the pointer p.
+     Requires (get_ownership(p) == true) or (p == 0). */
+  size_t __sanitizer_get_allocated_size(const volatile void *p);
+
+  /* Number of bytes, allocated and not yet freed by the application. */
+  size_t __sanitizer_get_current_allocated_bytes();
+
+  /* Number of bytes, mmaped by the allocator to fulfill allocation requests.
+     Generally, for request of X bytes, allocator can reserve and add to free
+     lists a large number of chunks of size X to use them for future requests.
+     All these chunks count toward the heap size. Currently, allocator never
+     releases memory to OS (instead, it just puts freed chunks to free
+     lists). */
+  size_t __sanitizer_get_heap_size();
+
+  /* Number of bytes, mmaped by the allocator, which can be used to fulfill
+     allocation requests. When a user program frees memory chunk, it can first
+     fall into quarantine and will count toward __sanitizer_get_free_bytes()
+     later. */
+  size_t __sanitizer_get_free_bytes();
+
+  /* Number of bytes in unmapped pages, that are released to OS. Currently,
+     always returns 0. */
+  size_t __sanitizer_get_unmapped_bytes();
+
+  /* Malloc hooks that may be optionally provided by user.
+     __sanitizer_malloc_hook(ptr, size) is called immediately after
+       allocation of "size" bytes, which returned "ptr".
+     __sanitizer_free_hook(ptr) is called immediately before
+       deallocation of "ptr". */
+  void __sanitizer_malloc_hook(const volatile void *ptr, size_t size);
+  void __sanitizer_free_hook(const volatile void *ptr);
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/libsanitizer/include/sanitizer/asan_interface.h b/libsanitizer/include/sanitizer/asan_interface.h
index bf4c4789536..d1d452991b0 100644
--- a/libsanitizer/include/sanitizer/asan_interface.h
+++ b/libsanitizer/include/sanitizer/asan_interface.h
@@ -60,6 +60,22 @@ extern "C" {
   // Print the description of addr (useful when debugging in gdb).
   void __asan_describe_address(void *addr);
 
+  // Useful for calling from the debugger to get the allocation stack trace
+  // and thread ID for a heap address. Stores up to 'size' frames into 'trace',
+  // returns the number of stored frames or 0 on error.
+  size_t __asan_get_alloc_stack(void *addr, void **trace, size_t size,
+                                int *thread_id);
+
+  // Useful for calling from the debugger to get the free stack trace
+  // and thread ID for a heap address. Stores up to 'size' frames into 'trace',
+  // returns the number of stored frames or 0 on error.
+  size_t __asan_get_free_stack(void *addr, void **trace, size_t size,
+                               int *thread_id);
+
+  // Useful for calling from the debugger to get the current shadow memory
+  // mapping.
+  void __asan_get_shadow_mapping(size_t *shadow_scale, size_t *shadow_offset);
+
   // This is an internal function that is called to report an error.
   // However it is still a part of the interface because users may want to
   // set a breakpoint on this function in a debugger.
@@ -81,32 +97,6 @@ extern "C" {
   // the program crashes before ASan report is printed.
   void __asan_on_error();
 
-  // Returns the estimated number of bytes that will be reserved by allocator
-  // for request of "size" bytes. If ASan allocator can't allocate that much
-  // memory, returns the maximal possible allocation size, otherwise returns
-  // "size".
-  size_t __asan_get_estimated_allocated_size(size_t size);
-  // Returns 1 if p was returned by the ASan allocator and is not yet freed.
-  // Otherwise returns 0.
-  int __asan_get_ownership(const void *p);
-  // Returns the number of bytes reserved for the pointer p.
-  // Requires (get_ownership(p) == true) or (p == 0).
-  size_t __asan_get_allocated_size(const void *p);
-  // Number of bytes, allocated and not yet freed by the application.
-  size_t __asan_get_current_allocated_bytes();
-  // Number of bytes, mmaped by asan allocator to fulfill allocation requests.
-  // Generally, for request of X bytes, allocator can reserve and add to free
-  // lists a large number of chunks of size X to use them for future requests.
-  // All these chunks count toward the heap size. Currently, allocator never
-  // releases memory to OS (instead, it just puts freed chunks to free lists).
-  size_t __asan_get_heap_size();
-  // Number of bytes, mmaped by asan allocator, which can be used to fulfill
-  // allocation requests. When a user program frees memory chunk, it can first
-  // fall into quarantine and will count toward __asan_get_free_bytes() later.
-  size_t __asan_get_free_bytes();
-  // Number of bytes in unmapped pages, that are released to OS. Currently,
-  // always returns 0.
-  size_t __asan_get_unmapped_bytes();
   // Prints accumulated stats to stderr. Used for debugging.
   void __asan_print_accumulated_stats();
 
@@ -114,14 +104,6 @@ extern "C" {
   // a string containing ASan runtime options. See asan_flags.h for details.
   const char* __asan_default_options();
 
-  // Malloc hooks that may be optionally provided by user.
-  // __asan_malloc_hook(ptr, size) is called immediately after
-  //   allocation of "size" bytes, which returned "ptr".
-  // __asan_free_hook(ptr) is called immediately before
-  //   deallocation of "ptr".
-  void __asan_malloc_hook(void *ptr, size_t size);
-  void __asan_free_hook(void *ptr);
-
   // The following 2 functions facilitate garbage collection in presence of
   // asan's fake stack.
 
diff --git a/libsanitizer/include/sanitizer/common_interface_defs.h b/libsanitizer/include/sanitizer/common_interface_defs.h
index 44870a039b4..43e0c30ee8a 100644
--- a/libsanitizer/include/sanitizer/common_interface_defs.h
+++ b/libsanitizer/include/sanitizer/common_interface_defs.h
@@ -64,6 +64,10 @@ extern "C" {
   void __sanitizer_cov_init();
   // Record and dump coverage info.
   void __sanitizer_cov_dump();
+  // Open <name>.sancov.packed in the coverage directory and return the file
+  // descriptor. Returns -1 on failure, or if coverage dumping is disabled.
+  // This is intended for use by sandboxing code.
+  intptr_t __sanitizer_maybe_open_cov_file(const char *name);
 
   // Annotate the current state of a contiguous container, such as
   // std::vector, std::string or similar.
diff --git a/libsanitizer/include/sanitizer/msan_interface.h b/libsanitizer/include/sanitizer/msan_interface.h
index 68e510f19bd..ea7ad1d7c9b 100644
--- a/libsanitizer/include/sanitizer/msan_interface.h
+++ b/libsanitizer/include/sanitizer/msan_interface.h
@@ -17,10 +17,6 @@
 #ifdef __cplusplus
 extern "C" {
 #endif
-  /* Returns a string describing a stack origin.
-     Return NULL if the origin is invalid, or is not a stack origin. */
-  const char *__msan_get_origin_descr_if_stack(uint32_t id);
-
   /* Set raw origin for the memory range. */
   void __msan_set_origin(const volatile void *a, size_t size, uint32_t origin);
 
@@ -91,55 +87,10 @@ extern "C" {
      a string containing Msan runtime options. See msan_flags.h for details. */
   const char* __msan_default_options();
 
-  // Sets the callback to be called right before death on error.
-  // Passing 0 will unset the callback.
+  /* Sets the callback to be called right before death on error.
+     Passing 0 will unset the callback. */
   void __msan_set_death_callback(void (*callback)(void));
 
-  /***********************************/
-  /* Allocator statistics interface. */
-
-  /* Returns the estimated number of bytes that will be reserved by allocator
-     for request of "size" bytes. If Msan allocator can't allocate that much
-     memory, returns the maximal possible allocation size, otherwise returns
-     "size". */
-  size_t __msan_get_estimated_allocated_size(size_t size);
-
-  /* Returns true if p was returned by the Msan allocator and
-     is not yet freed. */
-  int __msan_get_ownership(const volatile void *p);
-
-  /* Returns the number of bytes reserved for the pointer p.
-     Requires (get_ownership(p) == true) or (p == 0). */
-  size_t __msan_get_allocated_size(const volatile void *p);
-
-  /* Number of bytes, allocated and not yet freed by the application. */
-  size_t __msan_get_current_allocated_bytes();
-
-  /* Number of bytes, mmaped by msan allocator to fulfill allocation requests.
-     Generally, for request of X bytes, allocator can reserve and add to free
-     lists a large number of chunks of size X to use them for future requests.
-     All these chunks count toward the heap size. Currently, allocator never
-     releases memory to OS (instead, it just puts freed chunks to free
-     lists). */
-  size_t __msan_get_heap_size();
-
-  /* Number of bytes, mmaped by msan allocator, which can be used to fulfill
-     allocation requests. When a user program frees memory chunk, it can first
-     fall into quarantine and will count toward __msan_get_free_bytes()
-     later. */
-  size_t __msan_get_free_bytes();
-
-  /* Number of bytes in unmapped pages, that are released to OS. Currently,
-     always returns 0. */
-  size_t __msan_get_unmapped_bytes();
-
-  /* Malloc hooks that may be optionally provided by user.
-     __msan_malloc_hook(ptr, size) is called immediately after
-       allocation of "size" bytes, which returned "ptr".
-     __msan_free_hook(ptr) is called immediately before
-       deallocation of "ptr". */
-  void __msan_malloc_hook(const volatile void *ptr, size_t size);
-  void __msan_free_hook(const volatile void *ptr);
 #ifdef __cplusplus
 }  // extern "C"
 #endif
diff --git a/libsanitizer/interception/interception.h b/libsanitizer/interception/interception.h
index 51505e1baa2..458928f14b1 100644
--- a/libsanitizer/interception/interception.h
+++ b/libsanitizer/interception/interception.h
@@ -120,19 +120,23 @@ const interpose_substitution substitution_##func_name[] \
 # define DECLARE_WRAPPER(ret_type, func, ...)
 
 #elif defined(_WIN32)
-# if defined(_DLL)  // DLL CRT
-#  define WRAP(x) x
-#  define WRAPPER_NAME(x) #x
-#  define INTERCEPTOR_ATTRIBUTE
-# else  // Static CRT
-#  define WRAP(x) wrap_##x
-#  define WRAPPER_NAME(x) "wrap_"#x
-#  define INTERCEPTOR_ATTRIBUTE
-# endif
+# define WRAP(x) __asan_wrap_##x
+# define WRAPPER_NAME(x) "__asan_wrap_"#x
+# define INTERCEPTOR_ATTRIBUTE __declspec(dllexport)
 # define DECLARE_WRAPPER(ret_type, func, ...) \
     extern "C" ret_type func(__VA_ARGS__);
 # define DECLARE_WRAPPER_WINAPI(ret_type, func, ...) \
     extern "C" __declspec(dllimport) ret_type __stdcall func(__VA_ARGS__);
+#elif defined(__FreeBSD__)
+# define WRAP(x) __interceptor_ ## x
+# define WRAPPER_NAME(x) "__interceptor_" #x
+# define INTERCEPTOR_ATTRIBUTE __attribute__((visibility("default")))
+// FreeBSD's dynamic linker (incompliantly) gives non-weak symbols higher
+// priority than weak ones so weak aliases won't work for indirect calls
+// in position-independent (-fPIC / -fPIE) mode.
+# define DECLARE_WRAPPER(ret_type, func, ...) \
+     extern "C" ret_type func(__VA_ARGS__) \
+     __attribute__((alias("__interceptor_" #func), visibility("default")));
 #else
 # define WRAP(x) __interceptor_ ## x
 # define WRAPPER_NAME(x) "__interceptor_" #x
diff --git a/libsanitizer/interception/interception_win.cc b/libsanitizer/interception/interception_win.cc
index 9cd71751839..1642bd116cf 100644
--- a/libsanitizer/interception/interception_win.cc
+++ b/libsanitizer/interception/interception_win.cc
@@ -17,20 +17,6 @@
 
 namespace __interception {
 
-bool GetRealFunctionAddress(const char *func_name, uptr *func_addr) {
-  const char *DLLS[] = {
-    "msvcr80.dll",
-    "msvcr90.dll",
-    "kernel32.dll",
-    NULL
-  };
-  *func_addr = 0;
-  for (size_t i = 0; *func_addr == 0 && DLLS[i]; ++i) {
-    *func_addr = (uptr)GetProcAddress(GetModuleHandleA(DLLS[i]), func_name);
-  }
-  return (*func_addr != 0);
-}
-
 // FIXME: internal_str* and internal_mem* functions should be moved from the
 // ASan sources into interception/.
 
@@ -108,9 +94,11 @@ static size_t RoundUpToInstrBoundary(size_t size, char *code) {
       case 0x458B:  // 8B 45 XX = mov eax, dword ptr [ebp+XXh]
       case 0x5D8B:  // 8B 5D XX = mov ebx, dword ptr [ebp+XXh]
       case 0xEC83:  // 83 EC XX = sub esp, XX
+      case 0x75FF:  // FF 75 XX = push dword ptr [ebp+XXh]
         cursor += 3;
         continue;
       case 0xC1F7:  // F7 C1 XX YY ZZ WW = test ecx, WWZZYYXX
+      case 0x25FF:  // FF 25 XX YY ZZ WW = jmp dword ptr ds:[WWZZYYXX]
         cursor += 6;
         continue;
       case 0x3D83:  // 83 3D XX YY ZZ WW TT = cmp TT, WWZZYYXX
@@ -119,6 +107,7 @@ static size_t RoundUpToInstrBoundary(size_t size, char *code) {
     }
     switch (0x00FFFFFF & *(unsigned int*)(code + cursor)) {
       case 0x24448A:  // 8A 44 24 XX = mov eal, dword ptr [esp+XXh]
+      case 0x24448B:  // 8B 44 24 XX = mov eax, dword ptr [esp+XXh]
       case 0x244C8B:  // 8B 4C 24 XX = mov ecx, dword ptr [esp+XXh]
       case 0x24548B:  // 8B 54 24 XX = mov edx, dword ptr [esp+XXh]
       case 0x24748B:  // 8B 74 24 XX = mov esi, dword ptr [esp+XXh]
@@ -131,8 +120,9 @@ static size_t RoundUpToInstrBoundary(size_t size, char *code) {
     // FIXME: Unknown instruction failures might happen when we add a new
     // interceptor or a new compiler version. In either case, they should result
     // in visible and readable error messages. However, merely calling abort()
-    // or __debugbreak() leads to an infinite recursion in CheckFailed.
+    // leads to an infinite recursion in CheckFailed.
     // Do we have a good way to abort with an error message here?
+    __debugbreak();
     return 0;
   }
 
@@ -189,6 +179,33 @@ bool OverrideFunction(uptr old_func, uptr new_func, uptr *orig_old_func) {
   return true;
 }
 
+static const void **InterestingDLLsAvailable() {
+  const char *InterestingDLLs[] = { "kernel32.dll", "msvcr120.dll", NULL };
+  static void *result[ARRAY_SIZE(InterestingDLLs)] = { 0 };
+  if (!result[0]) {
+    for (size_t i = 0, j = 0; InterestingDLLs[i]; ++i) {
+      if (HMODULE h = GetModuleHandleA(InterestingDLLs[i]))
+        result[j++] = (void *)h;
+    }
+  }
+  return (const void **)&result[0];
+}
+
+static bool GetFunctionAddressInDLLs(const char *func_name, uptr *func_addr) {
+  *func_addr = 0;
+  const void **DLLs = InterestingDLLsAvailable();
+  for (size_t i = 0; *func_addr == 0 && DLLs[i]; ++i)
+    *func_addr = (uptr)GetProcAddress((HMODULE)DLLs[i], func_name);
+  return (*func_addr != 0);
+}
+
+bool OverrideFunction(const char *name, uptr new_func, uptr *orig_old_func) {
+  uptr orig_func;
+  if (!GetFunctionAddressInDLLs(name, &orig_func))
+    return false;
+  return OverrideFunction(orig_func, new_func, orig_old_func);
+}
+
 }  // namespace __interception
 
 #endif  // _WIN32
diff --git a/libsanitizer/interception/interception_win.h b/libsanitizer/interception/interception_win.h
index b46ad0dc66f..e9c6200d1b2 100644
--- a/libsanitizer/interception/interception_win.h
+++ b/libsanitizer/interception/interception_win.h
@@ -20,27 +20,29 @@
 #define INTERCEPTION_WIN_H
 
 namespace __interception {
-// returns true if a function with the given name was found.
-bool GetRealFunctionAddress(const char *func_name, uptr *func_addr);
+// All the functions in the OverrideFunction() family return true on success,
+// false on failure (including "couldn't find the function").
 
-// returns true if the old function existed, false on failure.
-bool OverrideFunction(uptr old_func, uptr new_func, uptr *orig_old_func);
+// Overrides a function by its address.
+bool OverrideFunction(uptr old_func, uptr new_func, uptr *orig_old_func = 0);
+
+// Overrides a function in a system DLL or DLL CRT by its exported name.
+bool OverrideFunction(const char *name, uptr new_func, uptr *orig_old_func = 0);
 }  // namespace __interception
 
-#if defined(_DLL)
-# define INTERCEPT_FUNCTION_WIN(func) \
-    ::__interception::GetRealFunctionAddress( \
-        #func, (::__interception::uptr*)&REAL(func))
+#if defined(INTERCEPTION_DYNAMIC_CRT)
+#define INTERCEPT_FUNCTION_WIN(func)                                           \
+  ::__interception::OverrideFunction(#func,                                    \
+                                     (::__interception::uptr)WRAP(func),       \
+                                     (::__interception::uptr *)&REAL(func))
 #else
-# define INTERCEPT_FUNCTION_WIN(func) \
-    ::__interception::OverrideFunction( \
-        (::__interception::uptr)func, \
-        (::__interception::uptr)WRAP(func), \
-        (::__interception::uptr*)&REAL(func))
+#define INTERCEPT_FUNCTION_WIN(func)                                           \
+  ::__interception::OverrideFunction((::__interception::uptr)func,             \
+                                     (::__interception::uptr)WRAP(func),       \
+                                     (::__interception::uptr *)&REAL(func))
 #endif
 
-#define INTERCEPT_FUNCTION_VER_WIN(func, symver) \
-    INTERCEPT_FUNCTION_WIN(func)
+#define INTERCEPT_FUNCTION_VER_WIN(func, symver) INTERCEPT_FUNCTION_WIN(func)
 
 #endif  // INTERCEPTION_WIN_H
 #endif  // _WIN32
diff --git a/libsanitizer/lsan/lsan.cc b/libsanitizer/lsan/lsan.cc
index c1481f5fb97..61792d9c9ea 100644
--- a/libsanitizer/lsan/lsan.cc
+++ b/libsanitizer/lsan/lsan.cc
@@ -23,16 +23,6 @@ bool lsan_init_is_running;
 
 namespace __lsan {
 
-static void InitializeCommonFlags() {
-  CommonFlags *cf = common_flags();
-  SetCommonFlagsDefaults(cf);
-  cf->external_symbolizer_path = GetEnv("LSAN_SYMBOLIZER_PATH");
-  cf->malloc_context_size = 30;
-  cf->detect_leaks = true;
-
-  ParseCommonFlagsFromString(cf, GetEnv("LSAN_OPTIONS"));
-}
-
 ///// Interface to the common LSan module. /////
 bool WordIsPoisoned(uptr addr) {
   return false;
@@ -48,7 +38,7 @@ extern "C" void __lsan_init() {
     return;
   lsan_init_is_running = true;
   SanitizerToolName = "LeakSanitizer";
-  InitializeCommonFlags();
+  InitCommonLsan(true);
   InitializeAllocator();
   InitTlsSize();
   InitializeInterceptors();
@@ -58,11 +48,14 @@ extern "C" void __lsan_init() {
   ThreadStart(tid, GetTid());
   SetCurrentThread(tid);
 
-  Symbolizer::Init(common_flags()->external_symbolizer_path);
-
-  InitCommonLsan();
   if (common_flags()->detect_leaks && common_flags()->leak_check_at_exit)
     Atexit(DoLeakCheck);
   lsan_inited = true;
   lsan_init_is_running = false;
 }
+
+extern "C" SANITIZER_INTERFACE_ATTRIBUTE
+void __sanitizer_print_stack_trace() {
+  GET_STACK_TRACE_FATAL;
+  stack.Print();
+}
diff --git a/libsanitizer/lsan/lsan.h b/libsanitizer/lsan/lsan.h
index 8a5030ce878..57888e3421d 100644
--- a/libsanitizer/lsan/lsan.h
+++ b/libsanitizer/lsan/lsan.h
@@ -13,6 +13,26 @@
 #include "sanitizer_common/sanitizer_flags.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
 
+#define GET_STACK_TRACE(max_size, fast)                                     \
+  StackTrace stack;                                                         \
+  {                                                                         \
+    uptr stack_top = 0, stack_bottom = 0;                                   \
+    ThreadContext *t;                                                       \
+    if (fast && (t = CurrentThreadContext())) {                             \
+      stack_top = t->stack_end();                                           \
+      stack_bottom = t->stack_begin();                                      \
+    }                                                                       \
+    stack.Unwind(max_size, StackTrace::GetCurrentPc(), GET_CURRENT_FRAME(), \
+                 /* context */ 0, stack_top, stack_bottom, fast);           \
+  }
+
+#define GET_STACK_TRACE_FATAL \
+  GET_STACK_TRACE(kStackTraceMax, common_flags()->fast_unwind_on_fatal)
+
+#define GET_STACK_TRACE_MALLOC                                      \
+  GET_STACK_TRACE(__sanitizer::common_flags()->malloc_context_size, \
+                  common_flags()->fast_unwind_on_malloc)
+
 namespace __lsan {
 
 void InitializeInterceptors();
diff --git a/libsanitizer/lsan/lsan_allocator.cc b/libsanitizer/lsan/lsan_allocator.cc
index 3e81ebef2a7..cda2b86ef67 100644
--- a/libsanitizer/lsan/lsan_allocator.cc
+++ b/libsanitizer/lsan/lsan_allocator.cc
@@ -13,6 +13,7 @@
 #include "lsan_allocator.h"
 
 #include "sanitizer_common/sanitizer_allocator.h"
+#include "sanitizer_common/sanitizer_allocator_interface.h"
 #include "sanitizer_common/sanitizer_internal_defs.h"
 #include "sanitizer_common/sanitizer_stackdepot.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
@@ -51,7 +52,7 @@ void AllocatorThreadFinish() {
   allocator.SwallowCache(&cache);
 }
 
-static ChunkMetadata *Metadata(void *p) {
+static ChunkMetadata *Metadata(const void *p) {
   return reinterpret_cast<ChunkMetadata *>(allocator.GetMetaData(p));
 }
 
@@ -85,10 +86,12 @@ void *Allocate(const StackTrace &stack, uptr size, uptr alignment,
   if (cleared && allocator.FromPrimary(p))
     memset(p, 0, size);
   RegisterAllocation(stack, p, size);
+  if (&__sanitizer_malloc_hook) __sanitizer_malloc_hook(p, size);
   return p;
 }
 
 void Deallocate(void *p) {
+  if (&__sanitizer_free_hook) __sanitizer_free_hook(p);
   RegisterDeallocation(p);
   allocator.Deallocate(&cache, p);
 }
@@ -111,7 +114,7 @@ void GetAllocatorCacheRange(uptr *begin, uptr *end) {
   *end = *begin + sizeof(cache);
 }
 
-uptr GetMallocUsableSize(void *p) {
+uptr GetMallocUsableSize(const void *p) {
   ChunkMetadata *m = Metadata(p);
   if (!m) return 0;
   return m->requested_size;
@@ -198,3 +201,38 @@ IgnoreObjectResult IgnoreObjectLocked(const void *p) {
   }
 }
 }  // namespace __lsan
+
+using namespace __lsan;
+
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_current_allocated_bytes() {
+  uptr stats[AllocatorStatCount];
+  allocator.GetStats(stats);
+  return stats[AllocatorStatAllocated];
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_heap_size() {
+  uptr stats[AllocatorStatCount];
+  allocator.GetStats(stats);
+  return stats[AllocatorStatMapped];
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_free_bytes() { return 0; }
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_unmapped_bytes() { return 0; }
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_estimated_allocated_size(uptr size) { return size; }
+
+SANITIZER_INTERFACE_ATTRIBUTE
+int __sanitizer_get_ownership(const void *p) { return Metadata(p) != 0; }
+
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_allocated_size(const void *p) {
+  return GetMallocUsableSize(p);
+}
+}  // extern "C"
diff --git a/libsanitizer/lsan/lsan_allocator.h b/libsanitizer/lsan/lsan_allocator.h
index 61ea86572ef..aae0d28dc4a 100644
--- a/libsanitizer/lsan/lsan_allocator.h
+++ b/libsanitizer/lsan/lsan_allocator.h
@@ -23,7 +23,7 @@ void *Allocate(const StackTrace &stack, uptr size, uptr alignment,
 void Deallocate(void *p);
 void *Reallocate(const StackTrace &stack, void *p, uptr new_size,
                  uptr alignment);
-uptr GetMallocUsableSize(void *p);
+uptr GetMallocUsableSize(const void *p);
 
 template<typename Callable>
 void ForEachChunk(const Callable &callback);
diff --git a/libsanitizer/lsan/lsan_common.cc b/libsanitizer/lsan/lsan_common.cc
index 78afa7706d6..e340b8953e1 100644
--- a/libsanitizer/lsan/lsan_common.cc
+++ b/libsanitizer/lsan/lsan_common.cc
@@ -34,15 +34,13 @@ bool DisabledInThisThread() { return disable_counter > 0; }
 
 Flags lsan_flags;
 
-static void InitializeFlags() {
+static void InitializeFlags(bool standalone) {
   Flags *f = flags();
   // Default values.
   f->report_objects = false;
   f->resolution = 0;
   f->max_leaks = 0;
   f->exitcode = 23;
-  f->print_suppressions = true;
-  f->suppressions="";
   f->use_registers = true;
   f->use_globals = true;
   f->use_stacks = true;
@@ -70,9 +68,18 @@ static void InitializeFlags() {
     ParseFlag(options, &f->log_pointers, "log_pointers", "");
     ParseFlag(options, &f->log_threads, "log_threads", "");
     ParseFlag(options, &f->exitcode, "exitcode", "");
-    ParseFlag(options, &f->print_suppressions, "print_suppressions", "");
-    ParseFlag(options, &f->suppressions, "suppressions", "");
   }
+
+  // Set defaults for common flags (only in standalone mode) and parse
+  // them from LSAN_OPTIONS.
+  CommonFlags *cf = common_flags();
+  if (standalone) {
+    SetCommonFlagsDefaults(cf);
+    cf->external_symbolizer_path = GetEnv("LSAN_SYMBOLIZER_PATH");
+    cf->malloc_context_size = 30;
+    cf->detect_leaks = true;
+  }
+  ParseCommonFlagsFromString(cf, options);
 }
 
 #define LOG_POINTERS(...)                           \
@@ -85,24 +92,14 @@ static void InitializeFlags() {
     if (flags()->log_threads) Report(__VA_ARGS__); \
   } while (0);
 
-SuppressionContext *suppression_ctx;
+static bool suppressions_inited = false;
 
 void InitializeSuppressions() {
-  CHECK(!suppression_ctx);
-  ALIGNED(64) static char placeholder[sizeof(SuppressionContext)];
-  suppression_ctx = new(placeholder) SuppressionContext;
-  char *suppressions_from_file;
-  uptr buffer_size;
-  if (ReadFileToBuffer(flags()->suppressions, &suppressions_from_file,
-                       &buffer_size, 1 << 26 /* max_len */))
-    suppression_ctx->Parse(suppressions_from_file);
-  if (flags()->suppressions[0] && !buffer_size) {
-    Printf("LeakSanitizer: failed to read suppressions file '%s'\n",
-           flags()->suppressions);
-    Die();
-  }
+  CHECK(!suppressions_inited);
+  SuppressionContext::InitIfNecessary();
   if (&__lsan_default_suppressions)
-    suppression_ctx->Parse(__lsan_default_suppressions());
+    SuppressionContext::Get()->Parse(__lsan_default_suppressions());
+  suppressions_inited = true;
 }
 
 struct RootRegion {
@@ -118,8 +115,8 @@ void InitializeRootRegions() {
   root_regions = new(placeholder) InternalMmapVector<RootRegion>(1);
 }
 
-void InitCommonLsan() {
-  InitializeFlags();
+void InitCommonLsan(bool standalone) {
+  InitializeFlags(standalone);
   InitializeRootRegions();
   if (common_flags()->detect_leaks) {
     // Initialization which can fail or print warnings should only be done if
@@ -129,9 +126,9 @@ void InitCommonLsan() {
   }
 }
 
-class Decorator: private __sanitizer::AnsiColorDecorator {
+class Decorator: public __sanitizer::SanitizerCommonDecorator {
  public:
-  Decorator() : __sanitizer::AnsiColorDecorator(PrintsToTtyCached()) { }
+  Decorator() : SanitizerCommonDecorator() { }
   const char *Error() { return Red(); }
   const char *Leak() { return Blue(); }
   const char *End() { return Default(); }
@@ -387,7 +384,7 @@ static void CollectLeaksCb(uptr chunk, void *arg) {
 
 static void PrintMatchedSuppressions() {
   InternalMmapVector<Suppression *> matched(1);
-  suppression_ctx->GetMatched(&matched);
+  SuppressionContext::Get()->GetMatched(&matched);
   if (!matched.size())
     return;
   const char *line = "-----------------------------------------------------";
@@ -448,7 +445,7 @@ void DoLeakCheck() {
     Printf("%s", d.End());
     param.leak_report.ReportTopLeaks(flags()->max_leaks);
   }
-  if (flags()->print_suppressions)
+  if (common_flags()->print_suppressions)
     PrintMatchedSuppressions();
   if (unsuppressed_count > 0) {
     param.leak_report.PrintSummary();
@@ -463,20 +460,22 @@ static Suppression *GetSuppressionForAddr(uptr addr) {
   // Suppress by module name.
   const char *module_name;
   uptr module_offset;
-  if (Symbolizer::Get()->GetModuleNameAndOffsetForPC(addr, &module_name,
-                                                     &module_offset) &&
-      suppression_ctx->Match(module_name, SuppressionLeak, &s))
+  if (Symbolizer::GetOrInit()
+          ->GetModuleNameAndOffsetForPC(addr, &module_name, &module_offset) &&
+      SuppressionContext::Get()->Match(module_name, SuppressionLeak, &s))
     return s;
 
   // Suppress by file or function name.
   static const uptr kMaxAddrFrames = 16;
   InternalScopedBuffer<AddressInfo> addr_frames(kMaxAddrFrames);
   for (uptr i = 0; i < kMaxAddrFrames; i++) new (&addr_frames[i]) AddressInfo();
-  uptr addr_frames_num = Symbolizer::Get()->SymbolizePC(
+  uptr addr_frames_num = Symbolizer::GetOrInit()->SymbolizePC(
       addr, addr_frames.data(), kMaxAddrFrames);
   for (uptr i = 0; i < addr_frames_num; i++) {
-    if (suppression_ctx->Match(addr_frames[i].function, SuppressionLeak, &s) ||
-        suppression_ctx->Match(addr_frames[i].file, SuppressionLeak, &s))
+    if (SuppressionContext::Get()->Match(addr_frames[i].function,
+                                         SuppressionLeak, &s) ||
+        SuppressionContext::Get()->Match(addr_frames[i].file, SuppressionLeak,
+                                         &s))
       return s;
   }
   return 0;
diff --git a/libsanitizer/lsan/lsan_common.h b/libsanitizer/lsan/lsan_common.h
index 0c84d414be2..72523d9ff5c 100644
--- a/libsanitizer/lsan/lsan_common.h
+++ b/libsanitizer/lsan/lsan_common.h
@@ -49,10 +49,6 @@ struct Flags {
   int max_leaks;
   // If nonzero kill the process with this exit code upon finding leaks.
   int exitcode;
-  // Print matched suppressions after leak checking.
-  bool print_suppressions;
-  // Suppressions file name.
-  const char* suppressions;
 
   // Flags controlling the root set of reachable memory.
   // Global variables (.data and .bss).
@@ -133,7 +129,7 @@ enum IgnoreObjectResult {
 };
 
 // Functions called from the parent tool.
-void InitCommonLsan();
+void InitCommonLsan(bool standalone);
 void DoLeakCheck();
 bool DisabledInThisThread();
 
diff --git a/libsanitizer/lsan/lsan_interceptors.cc b/libsanitizer/lsan/lsan_interceptors.cc
index dfaad325672..ac05dfaa54f 100644
--- a/libsanitizer/lsan/lsan_interceptors.cc
+++ b/libsanitizer/lsan/lsan_interceptors.cc
@@ -32,21 +32,6 @@ int pthread_key_create(unsigned *key, void (*destructor)(void* v));
 int pthread_setspecific(unsigned key, const void *v);
 }
 
-#define GET_STACK_TRACE                                              \
-  StackTrace stack;                                                  \
-  {                                                                  \
-    uptr stack_top = 0, stack_bottom = 0;                            \
-    ThreadContext *t;                                                \
-    bool fast = common_flags()->fast_unwind_on_malloc;               \
-    if (fast && (t = CurrentThreadContext())) {                      \
-      stack_top = t->stack_end();                                    \
-      stack_bottom = t->stack_begin();                               \
-    }                                                                \
-    stack.Unwind(__sanitizer::common_flags()->malloc_context_size,   \
-                 StackTrace::GetCurrentPc(), GET_CURRENT_FRAME(), 0, \
-                 stack_top, stack_bottom, fast);                     \
-  }
-
 #define ENSURE_LSAN_INITED do {   \
   CHECK(!lsan_init_is_running);   \
   if (!lsan_inited)               \
@@ -63,7 +48,7 @@ namespace std {
 
 INTERCEPTOR(void*, malloc, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   return Allocate(stack, size, 1, kAlwaysClearMemory);
 }
 
@@ -86,26 +71,32 @@ INTERCEPTOR(void*, calloc, uptr nmemb, uptr size) {
   }
   if (CallocShouldReturnNullDueToOverflow(size, nmemb)) return 0;
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   size *= nmemb;
   return Allocate(stack, size, 1, true);
 }
 
 INTERCEPTOR(void*, realloc, void *q, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   return Reallocate(stack, q, size, 1);
 }
 
 INTERCEPTOR(void*, memalign, uptr alignment, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
+  return Allocate(stack, size, alignment, kAlwaysClearMemory);
+}
+
+INTERCEPTOR(void*, aligned_alloc, uptr alignment, uptr size) {
+  ENSURE_LSAN_INITED;
+  GET_STACK_TRACE_MALLOC;
   return Allocate(stack, size, alignment, kAlwaysClearMemory);
 }
 
 INTERCEPTOR(int, posix_memalign, void **memptr, uptr alignment, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   *memptr = Allocate(stack, size, alignment, kAlwaysClearMemory);
   // FIXME: Return ENOMEM if user requested more than max alloc size.
   return 0;
@@ -113,7 +104,7 @@ INTERCEPTOR(int, posix_memalign, void **memptr, uptr alignment, uptr size) {
 
 INTERCEPTOR(void*, valloc, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   if (size == 0)
     size = GetPageSizeCached();
   return Allocate(stack, size, GetPageSizeCached(), kAlwaysClearMemory);
@@ -140,7 +131,7 @@ INTERCEPTOR(int, mallopt, int cmd, int value) {
 
 INTERCEPTOR(void*, pvalloc, uptr size) {
   ENSURE_LSAN_INITED;
-  GET_STACK_TRACE;
+  GET_STACK_TRACE_MALLOC;
   uptr PageSize = GetPageSizeCached();
   size = RoundUpTo(size, PageSize);
   if (size == 0) {
@@ -154,7 +145,7 @@ INTERCEPTOR(void, cfree, void *p) ALIAS(WRAPPER_NAME(free));
 
 #define OPERATOR_NEW_BODY                              \
   ENSURE_LSAN_INITED;                                  \
-  GET_STACK_TRACE;                                     \
+  GET_STACK_TRACE_MALLOC;                              \
   return Allocate(stack, size, 1, kAlwaysClearMemory);
 
 INTERCEPTOR_ATTRIBUTE
diff --git a/libsanitizer/lsan/lsan_preinit.cc b/libsanitizer/lsan/lsan_preinit.cc
index 856f9f78787..d1efd31406b 100644
--- a/libsanitizer/lsan/lsan_preinit.cc
+++ b/libsanitizer/lsan/lsan_preinit.cc
@@ -12,11 +12,7 @@
 
 #include "lsan.h"
 
-#ifndef LSAN_USE_PREINIT_ARRAY
-#define LSAN_USE_PREINIT_ARRAY 1
-#endif
-
-#if LSAN_USE_PREINIT_ARRAY && !defined(PIC)
+#if SANITIZER_CAN_USE_PREINIT_ARRAY
   // We force __lsan_init to be called before anyone else by placing it into
   // .preinit_array section.
   __attribute__((section(".preinit_array"), used))
diff --git a/libsanitizer/sanitizer_common/Makefile.am b/libsanitizer/sanitizer_common/Makefile.am
index 089c6936e2a..bc1f18c2a18 100644
--- a/libsanitizer/sanitizer_common/Makefile.am
+++ b/libsanitizer/sanitizer_common/Makefile.am
@@ -21,7 +21,8 @@ sanitizer_common_files = \
 	sanitizer_allocator.cc \
 	sanitizer_common.cc \
 	sanitizer_common_libcdep.cc \
-	sanitizer_coverage.cc \
+	sanitizer_coverage_libcdep.cc \
+	sanitizer_coverage_mapping_libcdep.cc \
 	sanitizer_deadlock_detector1.cc \
 	sanitizer_deadlock_detector2.cc \
 	sanitizer_flags.cc \
@@ -30,11 +31,14 @@ sanitizer_common_files = \
 	sanitizer_linux.cc \
 	sanitizer_linux_libcdep.cc \
 	sanitizer_mac.cc \
+	sanitizer_persistent_allocator.cc \
 	sanitizer_platform_limits_linux.cc \
 	sanitizer_platform_limits_posix.cc \
 	sanitizer_posix.cc \
 	sanitizer_posix_libcdep.cc \
 	sanitizer_printf.cc \
+	sanitizer_procmaps_common.cc \
+	sanitizer_procmaps_freebsd.cc \
 	sanitizer_procmaps_linux.cc \
 	sanitizer_procmaps_mac.cc \
 	sanitizer_stackdepot.cc \
@@ -49,8 +53,10 @@ sanitizer_common_files = \
 	sanitizer_symbolizer_win.cc \
 	sanitizer_thread_registry.cc \
 	sanitizer_tls_get_addr.cc \
+	sanitizer_unwind_posix_libcdep.cc \
 	sanitizer_win.cc
 
+
 libsanitizer_common_la_SOURCES = $(sanitizer_common_files) 
 
 # Work around what appears to be a GNU make bug handling MAKEFLAGS
diff --git a/libsanitizer/sanitizer_common/Makefile.in b/libsanitizer/sanitizer_common/Makefile.in
index 1ea631590c0..d82b4834efc 100644
--- a/libsanitizer/sanitizer_common/Makefile.in
+++ b/libsanitizer/sanitizer_common/Makefile.in
@@ -64,14 +64,17 @@ CONFIG_CLEAN_VPATH_FILES =
 LTLIBRARIES = $(noinst_LTLIBRARIES)
 libsanitizer_common_la_LIBADD =
 am__objects_1 = sanitizer_allocator.lo sanitizer_common.lo \
-	sanitizer_common_libcdep.lo sanitizer_coverage.lo \
+	sanitizer_common_libcdep.lo sanitizer_coverage_libcdep.lo \
+	sanitizer_coverage_mapping_libcdep.lo \
 	sanitizer_deadlock_detector1.lo \
 	sanitizer_deadlock_detector2.lo sanitizer_flags.lo \
 	sanitizer_libc.lo sanitizer_libignore.lo sanitizer_linux.lo \
 	sanitizer_linux_libcdep.lo sanitizer_mac.lo \
+	sanitizer_persistent_allocator.lo \
 	sanitizer_platform_limits_linux.lo \
 	sanitizer_platform_limits_posix.lo sanitizer_posix.lo \
 	sanitizer_posix_libcdep.lo sanitizer_printf.lo \
+	sanitizer_procmaps_common.lo sanitizer_procmaps_freebsd.lo \
 	sanitizer_procmaps_linux.lo sanitizer_procmaps_mac.lo \
 	sanitizer_stackdepot.lo sanitizer_stacktrace.lo \
 	sanitizer_stacktrace_libcdep.lo \
@@ -81,7 +84,8 @@ am__objects_1 = sanitizer_allocator.lo sanitizer_common.lo \
 	sanitizer_symbolizer_libcdep.lo \
 	sanitizer_symbolizer_posix_libcdep.lo \
 	sanitizer_symbolizer_win.lo sanitizer_thread_registry.lo \
-	sanitizer_tls_get_addr.lo sanitizer_win.lo
+	sanitizer_tls_get_addr.lo sanitizer_unwind_posix_libcdep.lo \
+	sanitizer_win.lo
 am_libsanitizer_common_la_OBJECTS = $(am__objects_1)
 libsanitizer_common_la_OBJECTS = $(am_libsanitizer_common_la_OBJECTS)
 DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir)
@@ -256,7 +260,8 @@ sanitizer_common_files = \
 	sanitizer_allocator.cc \
 	sanitizer_common.cc \
 	sanitizer_common_libcdep.cc \
-	sanitizer_coverage.cc \
+	sanitizer_coverage_libcdep.cc \
+	sanitizer_coverage_mapping_libcdep.cc \
 	sanitizer_deadlock_detector1.cc \
 	sanitizer_deadlock_detector2.cc \
 	sanitizer_flags.cc \
@@ -265,11 +270,14 @@ sanitizer_common_files = \
 	sanitizer_linux.cc \
 	sanitizer_linux_libcdep.cc \
 	sanitizer_mac.cc \
+	sanitizer_persistent_allocator.cc \
 	sanitizer_platform_limits_linux.cc \
 	sanitizer_platform_limits_posix.cc \
 	sanitizer_posix.cc \
 	sanitizer_posix_libcdep.cc \
 	sanitizer_printf.cc \
+	sanitizer_procmaps_common.cc \
+	sanitizer_procmaps_freebsd.cc \
 	sanitizer_procmaps_linux.cc \
 	sanitizer_procmaps_mac.cc \
 	sanitizer_stackdepot.cc \
@@ -284,6 +292,7 @@ sanitizer_common_files = \
 	sanitizer_symbolizer_win.cc \
 	sanitizer_thread_registry.cc \
 	sanitizer_tls_get_addr.cc \
+	sanitizer_unwind_posix_libcdep.cc \
 	sanitizer_win.cc
 
 libsanitizer_common_la_SOURCES = $(sanitizer_common_files) 
@@ -382,7 +391,8 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_allocator.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_common.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_common_libcdep.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_coverage.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_coverage_libcdep.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_coverage_mapping_libcdep.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_deadlock_detector1.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_deadlock_detector2.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_flags.Plo@am__quote@
@@ -391,11 +401,14 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_linux.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_linux_libcdep.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_mac.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_persistent_allocator.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_platform_limits_linux.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_platform_limits_posix.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_posix.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_posix_libcdep.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_printf.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_procmaps_common.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_procmaps_freebsd.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_procmaps_linux.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_procmaps_mac.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_stackdepot.Plo@am__quote@
@@ -410,6 +423,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_symbolizer_win.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_thread_registry.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_tls_get_addr.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_unwind_posix_libcdep.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/sanitizer_win.Plo@am__quote@
 
 .cc.o:
diff --git a/libsanitizer/sanitizer_common/sanitizer_allocator.h b/libsanitizer/sanitizer_common/sanitizer_allocator.h
index 99be09ba9ba..fb394129202 100644
--- a/libsanitizer/sanitizer_common/sanitizer_allocator.h
+++ b/libsanitizer/sanitizer_common/sanitizer_allocator.h
@@ -271,9 +271,9 @@ class AllocatorGlobalStats : public AllocatorStats {
       if (stats == this)
         break;
     }
-    // All stats must be positive.
+    // All stats must be non-negative.
     for (int i = 0; i < AllocatorStatCount; i++)
-      s[i] = ((sptr)s[i]) > 0 ? s[i] : 1;
+      s[i] = ((sptr)s[i]) >= 0 ? s[i] : 0;
   }
 
  private:
diff --git a/libsanitizer/sanitizer_common/sanitizer_allocator_interface.h b/libsanitizer/sanitizer_common/sanitizer_allocator_interface.h
new file mode 100644
index 00000000000..fd5bed30c4b
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_allocator_interface.h
@@ -0,0 +1,36 @@
+//===-- sanitizer_allocator_interface.h ------------------------- C++ -----===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Re-declaration of functions from public sanitizer allocator interface.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef SANITIZER_ALLOCATOR_INTERFACE_H
+#define SANITIZER_ALLOCATOR_INTERFACE_H
+
+#include "sanitizer_internal_defs.h"
+
+using __sanitizer::uptr;
+
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE
+uptr __sanitizer_get_estimated_allocated_size(uptr size);
+SANITIZER_INTERFACE_ATTRIBUTE int __sanitizer_get_ownership(const void *p);
+SANITIZER_INTERFACE_ATTRIBUTE uptr
+__sanitizer_get_allocated_size(const void *p);
+SANITIZER_INTERFACE_ATTRIBUTE uptr __sanitizer_get_current_allocated_bytes();
+SANITIZER_INTERFACE_ATTRIBUTE uptr __sanitizer_get_heap_size();
+SANITIZER_INTERFACE_ATTRIBUTE uptr __sanitizer_get_free_bytes();
+SANITIZER_INTERFACE_ATTRIBUTE uptr __sanitizer_get_unmapped_bytes();
+
+SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
+    /* OPTIONAL */ void __sanitizer_malloc_hook(void *ptr, uptr size);
+SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
+    /* OPTIONAL */ void __sanitizer_free_hook(void *ptr);
+}  // extern "C"
+
+#endif  // SANITIZER_ALLOCATOR_INTERFACE_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_allocator_internal.h b/libsanitizer/sanitizer_common/sanitizer_allocator_internal.h
index 74e4402f736..5e8cc107706 100644
--- a/libsanitizer/sanitizer_common/sanitizer_allocator_internal.h
+++ b/libsanitizer/sanitizer_common/sanitizer_allocator_internal.h
@@ -23,38 +23,25 @@ typedef CompactSizeClassMap InternalSizeClassMap;
 
 static const uptr kInternalAllocatorSpace = 0;
 static const u64 kInternalAllocatorSize = SANITIZER_MMAP_RANGE_SIZE;
-#if SANITIZER_WORDSIZE == 32
 static const uptr kInternalAllocatorRegionSizeLog = 20;
+#if SANITIZER_WORDSIZE == 32
 static const uptr kInternalAllocatorNumRegions =
     kInternalAllocatorSize >> kInternalAllocatorRegionSizeLog;
 typedef FlatByteMap<kInternalAllocatorNumRegions> ByteMap;
 #else
-static const uptr kInternalAllocatorRegionSizeLog = 24;
 static const uptr kInternalAllocatorNumRegions =
     kInternalAllocatorSize >> kInternalAllocatorRegionSizeLog;
 typedef TwoLevelByteMap<(kInternalAllocatorNumRegions >> 12), 1 << 12> ByteMap;
 #endif
 typedef SizeClassAllocator32<
-    kInternalAllocatorSpace, kInternalAllocatorSize, 16, InternalSizeClassMap,
+    kInternalAllocatorSpace, kInternalAllocatorSize, 0, InternalSizeClassMap,
     kInternalAllocatorRegionSizeLog, ByteMap> PrimaryInternalAllocator;
 
 typedef SizeClassAllocatorLocalCache<PrimaryInternalAllocator>
     InternalAllocatorCache;
 
-// We don't want our internal allocator to do any map/unmap operations from
-// LargeMmapAllocator.
-struct CrashOnMapUnmap {
-  void OnMap(uptr p, uptr size) const {
-    RAW_CHECK_MSG(0, "Unexpected mmap in InternalAllocator!\n");
-  }
-  void OnUnmap(uptr p, uptr size) const {
-    RAW_CHECK_MSG(0, "Unexpected munmap in InternalAllocator!\n");
-  }
-};
-
 typedef CombinedAllocator<PrimaryInternalAllocator, InternalAllocatorCache,
-                          LargeMmapAllocator<CrashOnMapUnmap> >
-    InternalAllocator;
+                          LargeMmapAllocator<> > InternalAllocator;
 
 void *InternalAlloc(uptr size, InternalAllocatorCache *cache = 0);
 void InternalFree(void *p, InternalAllocatorCache *cache = 0);
diff --git a/libsanitizer/sanitizer_common/sanitizer_atomic_msvc.h b/libsanitizer/sanitizer_common/sanitizer_atomic_msvc.h
index 7e18fa38748..1198dec3e97 100644
--- a/libsanitizer/sanitizer_common/sanitizer_atomic_msvc.h
+++ b/libsanitizer/sanitizer_common/sanitizer_atomic_msvc.h
@@ -31,33 +31,20 @@ long long _InterlockedCompareExchange64(  // NOLINT
     long long volatile *Destination,              // NOLINT
     long long Exchange, long long Comparand);     // NOLINT
 #pragma intrinsic(_InterlockedCompareExchange64)
-
-#ifdef _WIN64
-extern "C" long long _InterlockedExchangeAdd64(     // NOLINT
-    long long volatile * Addend, long long Value);  // NOLINT
-#pragma intrinsic(_InterlockedExchangeAdd64)
 extern "C" void *_InterlockedCompareExchangePointer(
     void *volatile *Destination,
     void *Exchange, void *Comparand);
 #pragma intrinsic(_InterlockedCompareExchangePointer)
-#else
-// There's no _InterlockedCompareExchangePointer intrinsic on x86,
-// so call _InterlockedCompareExchange instead.
 extern "C"
 long __cdecl _InterlockedCompareExchange(  // NOLINT
     long volatile *Destination,            // NOLINT
     long Exchange, long Comparand);        // NOLINT
 #pragma intrinsic(_InterlockedCompareExchange)
 
-inline static void *_InterlockedCompareExchangePointer(
-    void *volatile *Destination,
-    void *Exchange, void *Comparand) {
-  return reinterpret_cast<void*>(
-      _InterlockedCompareExchange(
-          reinterpret_cast<long volatile*>(Destination),  // NOLINT
-          reinterpret_cast<long>(Exchange),               // NOLINT
-          reinterpret_cast<long>(Comparand)));            // NOLINT
-}
+#ifdef _WIN64
+extern "C" long long _InterlockedExchangeAdd64(     // NOLINT
+    long long volatile * Addend, long long Value);  // NOLINT
+#pragma intrinsic(_InterlockedExchangeAdd64)
 #endif
 
 namespace __sanitizer {
diff --git a/libsanitizer/sanitizer_common/sanitizer_common.cc b/libsanitizer/sanitizer_common/sanitizer_common.cc
index e76d4d558a1..d4da3206e13 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_common.cc
@@ -12,8 +12,6 @@
 #include "sanitizer_common.h"
 #include "sanitizer_flags.h"
 #include "sanitizer_libc.h"
-#include "sanitizer_stacktrace.h"
-#include "sanitizer_symbolizer.h"
 
 namespace __sanitizer {
 
@@ -195,31 +193,17 @@ void ReportErrorSummary(const char *error_type, const char *file,
   ReportErrorSummary(buff.data());
 }
 
-void ReportErrorSummary(const char *error_type, StackTrace *stack) {
-  if (!common_flags()->print_summary)
-    return;
-  AddressInfo ai;
-#if !SANITIZER_GO
-  if (stack->size > 0 && Symbolizer::Get()->CanReturnFileLineInfo()) {
-    // Currently, we include the first stack frame into the report summary.
-    // Maybe sometimes we need to choose another frame (e.g. skip memcpy/etc).
-    uptr pc = StackTrace::GetPreviousInstructionPc(stack->trace[0]);
-    Symbolizer::Get()->SymbolizePC(pc, &ai, 1);
-  }
-#endif
-  ReportErrorSummary(error_type, ai.file, ai.line, ai.function);
-}
-
 LoadedModule::LoadedModule(const char *module_name, uptr base_address) {
   full_name_ = internal_strdup(module_name);
   base_address_ = base_address;
   n_ranges_ = 0;
 }
 
-void LoadedModule::addAddressRange(uptr beg, uptr end) {
+void LoadedModule::addAddressRange(uptr beg, uptr end, bool executable) {
   CHECK_LT(n_ranges_, kMaxNumberOfAddressRanges);
   ranges_[n_ranges_].beg = beg;
   ranges_[n_ranges_].end = end;
+  exec_[n_ranges_] = executable;
   n_ranges_++;
 }
 
@@ -261,11 +245,6 @@ void DecreaseTotalMmap(uptr size) {
   atomic_fetch_sub(&g_total_mmaped, size, memory_order_relaxed);
 }
 
-static void (*sandboxing_callback)();
-void SetSandboxingCallback(void (*f)()) {
-  sandboxing_callback = f;
-}
-
 }  // namespace __sanitizer
 
 using namespace __sanitizer;  // NOLINT
@@ -298,13 +277,6 @@ void __sanitizer_set_report_path(const char *path) {
   }
 }
 
-void NOINLINE
-__sanitizer_sandbox_on_notify(__sanitizer_sandbox_arguments *args) {
-  PrepareForSandboxing(args);
-  if (sandboxing_callback)
-    sandboxing_callback();
-}
-
 void __sanitizer_report_error_summary(const char *error_summary) {
   Printf("%s\n", error_summary);
 }
diff --git a/libsanitizer/sanitizer_common/sanitizer_common.h b/libsanitizer/sanitizer_common/sanitizer_common.h
index 93317132c49..a9db1082f48 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common.h
+++ b/libsanitizer/sanitizer_common/sanitizer_common.h
@@ -163,6 +163,9 @@ uptr ReadFileToBuffer(const char *file_name, char **buff,
 // (or NULL if the mapping failes). Stores the size of mmaped region
 // in '*buff_size'.
 void *MapFileToMemory(const char *file_name, uptr *buff_size);
+void *MapWritableFileToMemory(void *addr, uptr size, uptr fd, uptr offset);
+
+bool IsAccessibleMemoryRange(uptr beg, uptr size);
 
 // Error report formatting.
 const char *StripPathPrefix(const char *filepath,
@@ -173,7 +176,7 @@ void PrintModuleAndOffset(InternalScopedString *buffer,
                           const char *module, uptr offset);
 
 // OS
-void DisableCoreDumper();
+void DisableCoreDumperIfNecessary();
 void DumpProcessMap();
 bool FileExists(const char *filename);
 const char *GetEnv(const char *name);
@@ -184,11 +187,17 @@ u32 GetUid();
 void ReExec();
 bool StackSizeIsUnlimited();
 void SetStackSizeLimitInBytes(uptr limit);
+bool AddressSpaceIsUnlimited();
+void SetAddressSpaceUnlimited();
 void AdjustStackSize(void *attr);
 void PrepareForSandboxing(__sanitizer_sandbox_arguments *args);
 void CovPrepareForSandboxing(__sanitizer_sandbox_arguments *args);
 void SetSandboxingCallback(void (*f)());
 
+void CovUpdateMapping(uptr caller_pc = 0);
+void CovBeforeFork();
+void CovAfterFork(int child_pid);
+
 void InitTlsSize();
 uptr GetTlsSize();
 
@@ -239,7 +248,7 @@ const int kMaxSummaryLength = 1024;
 // and pass it to __sanitizer_report_error_summary.
 void ReportErrorSummary(const char *error_message);
 // Same as above, but construct error_message as:
-//   error_type: file:line function
+//   error_type file:line function
 void ReportErrorSummary(const char *error_type, const char *file,
                         int line, const char *function);
 void ReportErrorSummary(const char *error_type, StackTrace *trace);
@@ -475,12 +484,17 @@ uptr InternalBinarySearch(const Container &v, uptr first, uptr last,
 class LoadedModule {
  public:
   LoadedModule(const char *module_name, uptr base_address);
-  void addAddressRange(uptr beg, uptr end);
+  void addAddressRange(uptr beg, uptr end, bool executable);
   bool containsAddress(uptr address) const;
 
   const char *full_name() const { return full_name_; }
   uptr base_address() const { return base_address_; }
 
+  uptr n_ranges() const { return n_ranges_; }
+  uptr address_range_start(int i) const { return ranges_[i].beg; }
+  uptr address_range_end(int i) const { return ranges_[i].end; }
+  bool address_range_executable(int i) const { return exec_[i]; }
+
  private:
   struct AddressRange {
     uptr beg;
@@ -490,6 +504,7 @@ class LoadedModule {
   uptr base_address_;
   static const uptr kMaxNumberOfAddressRanges = 6;
   AddressRange ranges_[kMaxNumberOfAddressRanges];
+  bool exec_[kMaxNumberOfAddressRanges];
   uptr n_ranges_;
 };
 
@@ -529,10 +544,13 @@ F IndirectExternCall(F f) {
 #endif
 
 #if SANITIZER_ANDROID
+// Initialize Android logging. Any writes before this are silently lost.
+void AndroidLogInit();
 void AndroidLogWrite(const char *buffer);
 void GetExtraActivationFlags(char *buf, uptr size);
 void SanitizerInitializeUnwinder();
 #else
+INLINE void AndroidLogInit() {}
 INLINE void AndroidLogWrite(const char *buffer_unused) {}
 INLINE void GetExtraActivationFlags(char *buf, uptr size) { *buf = '\0'; }
 INLINE void SanitizerInitializeUnwinder() {}
@@ -544,4 +562,9 @@ inline void *operator new(__sanitizer::operator_new_size_type size,
   return alloc.Allocate(size);
 }
 
+struct StackDepotStats {
+  uptr n_uniq_ids;
+  uptr allocated;
+};
+
 #endif  // SANITIZER_COMMON_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc b/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc
index 6cbe03ae34a..f55a31de77e 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc
+++ b/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc
@@ -11,6 +11,7 @@
 // This file should be included into the tool's interceptor file,
 // which has to define it's own macros:
 //   COMMON_INTERCEPTOR_ENTER
+//   COMMON_INTERCEPTOR_ENTER_NOIGNORE
 //   COMMON_INTERCEPTOR_READ_RANGE
 //   COMMON_INTERCEPTOR_WRITE_RANGE
 //   COMMON_INTERCEPTOR_INITIALIZE_RANGE
@@ -24,6 +25,7 @@
 //   COMMON_INTERCEPTOR_MUTEX_REPAIR
 //   COMMON_INTERCEPTOR_SET_PTHREAD_NAME
 //   COMMON_INTERCEPTOR_HANDLE_RECVMSG
+//   COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED
 //===----------------------------------------------------------------------===//
 #include "interception/interception.h"
 #include "sanitizer_addrhashmap.h"
@@ -73,6 +75,23 @@
 #define COMMON_INTERCEPTOR_FILE_CLOSE(ctx, file) {}
 #endif
 
+#ifndef COMMON_INTERCEPTOR_LIBRARY_LOADED
+#define COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, map) {}
+#endif
+
+#ifndef COMMON_INTERCEPTOR_LIBRARY_UNLOADED
+#define COMMON_INTERCEPTOR_LIBRARY_UNLOADED() {}
+#endif
+
+#ifndef COMMON_INTERCEPTOR_ENTER_NOIGNORE
+#define COMMON_INTERCEPTOR_ENTER_NOIGNORE(ctx, ...) \
+  COMMON_INTERCEPTOR_ENTER(ctx, __VA_ARGS__)
+#endif
+
+#ifndef COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED
+#define COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED (0)
+#endif
+
 struct FileMetadata {
   // For open_memstream().
   char **addr;
@@ -158,6 +177,8 @@ INTERCEPTOR(int, strcmp, const char *s1, const char *s2) {
 }
 
 INTERCEPTOR(int, strncmp, const char *s1, const char *s2, uptr size) {
+  if (COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED)
+    return internal_strncmp(s1, s2, size);
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, strncmp, s1, s2, size);
   unsigned char c1 = 0, c2 = 0;
@@ -255,8 +276,9 @@ INTERCEPTOR(void*, memrchr, const void *s, int c, SIZE_T n) {
 INTERCEPTOR(double, frexp, double x, int *exp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, frexp, x, exp);
-  double res = REAL(frexp)(x, exp);
+  // Assuming frexp() always writes to |exp|.
   COMMON_INTERCEPTOR_WRITE_RANGE(ctx, exp, sizeof(*exp));
+  double res = REAL(frexp)(x, exp);
   return res;
 }
 
@@ -269,6 +291,9 @@ INTERCEPTOR(double, frexp, double x, int *exp) {
 INTERCEPTOR(float, frexpf, float x, int *exp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, frexpf, x, exp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   float res = REAL(frexpf)(x, exp);
   COMMON_INTERCEPTOR_WRITE_RANGE(ctx, exp, sizeof(*exp));
   return res;
@@ -277,6 +302,9 @@ INTERCEPTOR(float, frexpf, float x, int *exp) {
 INTERCEPTOR(long double, frexpl, long double x, int *exp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, frexpl, x, exp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   long double res = REAL(frexpl)(x, exp);
   COMMON_INTERCEPTOR_WRITE_RANGE(ctx, exp, sizeof(*exp));
   return res;
@@ -315,6 +343,9 @@ INTERCEPTOR(SSIZE_T, read, int fd, void *ptr, SIZE_T count) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, read, fd, ptr, count);
   COMMON_INTERCEPTOR_FD_ACCESS(ctx, fd);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(read)(fd, ptr, count);
   if (res > 0) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ptr, res);
   if (res >= 0 && fd >= 0) COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd);
@@ -330,6 +361,9 @@ INTERCEPTOR(SSIZE_T, pread, int fd, void *ptr, SIZE_T count, OFF_T offset) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, pread, fd, ptr, count, offset);
   COMMON_INTERCEPTOR_FD_ACCESS(ctx, fd);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(pread)(fd, ptr, count, offset);
   if (res > 0) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ptr, res);
   if (res >= 0 && fd >= 0) COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd);
@@ -345,6 +379,9 @@ INTERCEPTOR(SSIZE_T, pread64, int fd, void *ptr, SIZE_T count, OFF64_T offset) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, pread64, fd, ptr, count, offset);
   COMMON_INTERCEPTOR_FD_ACCESS(ctx, fd);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(pread64)(fd, ptr, count, offset);
   if (res > 0) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ptr, res);
   if (res >= 0 && fd >= 0) COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd);
@@ -523,9 +560,11 @@ INTERCEPTOR(int, prctl, int option, unsigned long arg2,
 INTERCEPTOR(unsigned long, time, unsigned long *t) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, time, t);
-  unsigned long res = REAL(time)(t);
+  unsigned long local_t;
+  unsigned long res = REAL(time)(&local_t);
   if (t && res != (unsigned long)-1) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, t, sizeof(*t));
+    *t = local_t;
   }
   return res;
 }
@@ -587,6 +626,9 @@ INTERCEPTOR(__sanitizer_tm *, gmtime_r, unsigned long *timep, void *result) {
 INTERCEPTOR(char *, ctime, unsigned long *timep) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ctime, timep);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(ctime)(timep);
   if (res) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, timep, sizeof(*timep));
@@ -597,6 +639,9 @@ INTERCEPTOR(char *, ctime, unsigned long *timep) {
 INTERCEPTOR(char *, ctime_r, unsigned long *timep, char *result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ctime_r, timep, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(ctime_r)(timep, result);
   if (res) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, timep, sizeof(*timep));
@@ -607,6 +652,9 @@ INTERCEPTOR(char *, ctime_r, unsigned long *timep, char *result) {
 INTERCEPTOR(char *, asctime, __sanitizer_tm *tm) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, asctime, tm);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(asctime)(tm);
   if (res) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, tm, sizeof(*tm));
@@ -617,6 +665,9 @@ INTERCEPTOR(char *, asctime, __sanitizer_tm *tm) {
 INTERCEPTOR(char *, asctime_r, __sanitizer_tm *tm, char *result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, asctime_r, tm, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(asctime_r)(tm, result);
   if (res) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, tm, sizeof(*tm));
@@ -658,6 +709,9 @@ INTERCEPTOR(char *, strptime, char *s, char *format, __sanitizer_tm *tm) {
   COMMON_INTERCEPTOR_ENTER(ctx, strptime, s, format, tm);
   if (format)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, format, REAL(strlen)(format) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(strptime)(s, format, tm);
   if (res) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, s, res - s);
@@ -792,6 +846,9 @@ FORMAT_INTERCEPTOR_IMPL(__isoc99_sscanf, __isoc99_vsscanf, str, format)
     return res;                                                                \
   }
 
+// FIXME: under ASan the REAL() call below may write to freed memory and
+// corrupt its metadata. See
+// https://code.google.com/p/address-sanitizer/issues/detail?id=321.
 #define VSPRINTF_INTERCEPTOR_IMPL(vname, str, ...)                             \
   {                                                                            \
     VPRINTF_INTERCEPTOR_ENTER(vname, str, __VA_ARGS__)                         \
@@ -806,6 +863,9 @@ FORMAT_INTERCEPTOR_IMPL(__isoc99_sscanf, __isoc99_vsscanf, str, format)
     return res;                                                                \
   }
 
+// FIXME: under ASan the REAL() call below may write to freed memory and
+// corrupt its metadata. See
+// https://code.google.com/p/address-sanitizer/issues/detail?id=321.
 #define VSNPRINTF_INTERCEPTOR_IMPL(vname, str, size, ...)                      \
   {                                                                            \
     VPRINTF_INTERCEPTOR_ENTER(vname, str, size, __VA_ARGS__)                   \
@@ -820,6 +880,9 @@ FORMAT_INTERCEPTOR_IMPL(__isoc99_sscanf, __isoc99_vsscanf, str, format)
     return res;                                                                \
   }
 
+// FIXME: under ASan the REAL() call below may write to freed memory and
+// corrupt its metadata. See
+// https://code.google.com/p/address-sanitizer/issues/detail?id=321.
 #define VASPRINTF_INTERCEPTOR_IMPL(vname, strp, ...)                           \
   {                                                                            \
     VPRINTF_INTERCEPTOR_ENTER(vname, strp, __VA_ARGS__)                        \
@@ -972,7 +1035,8 @@ INTERCEPTOR(int, ioctl, int d, unsigned request, void *arg) {
 #endif
 
 #if SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS || \
-    SANITIZER_INTERCEPT_GETPWENT || SANITIZER_INTERCEPT_FGETPWENT
+    SANITIZER_INTERCEPT_GETPWENT || SANITIZER_INTERCEPT_FGETPWENT || \
+    SANITIZER_INTERCEPT_GETPWENT_R || SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS
 static void unpoison_passwd(void *ctx, __sanitizer_passwd *pwd) {
   if (pwd) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, pwd, sizeof(*pwd));
@@ -1019,7 +1083,9 @@ static void unpoison_group(void *ctx, __sanitizer_group *grp) {
   }
 }
 #endif  // SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS ||
-        // SANITIZER_INTERCEPT_GETPWENT || SANITIZER_INTERCEPT_FGETPWENT
+        // SANITIZER_INTERCEPT_GETPWENT || SANITIZER_INTERCEPT_FGETPWENT ||
+        // SANITIZER_INTERCEPT_GETPWENT_R ||
+        // SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS
 
 #if SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS
 INTERCEPTOR(__sanitizer_passwd *, getpwnam, const char *name) {
@@ -1067,6 +1133,9 @@ INTERCEPTOR(int, getpwnam_r, const char *name, __sanitizer_passwd *pwd,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getpwnam_r, name, pwd, buf, buflen, result);
   COMMON_INTERCEPTOR_READ_RANGE(ctx, name, REAL(strlen)(name) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getpwnam_r)(name, pwd, buf, buflen, result);
   if (!res) {
     if (result && *result) unpoison_passwd(ctx, *result);
@@ -1079,6 +1148,9 @@ INTERCEPTOR(int, getpwuid_r, u32 uid, __sanitizer_passwd *pwd, char *buf,
             SIZE_T buflen, __sanitizer_passwd **result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getpwuid_r, uid, pwd, buf, buflen, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getpwuid_r)(uid, pwd, buf, buflen, result);
   if (!res) {
     if (result && *result) unpoison_passwd(ctx, *result);
@@ -1092,6 +1164,9 @@ INTERCEPTOR(int, getgrnam_r, const char *name, __sanitizer_group *grp,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getgrnam_r, name, grp, buf, buflen, result);
   COMMON_INTERCEPTOR_READ_RANGE(ctx, name, REAL(strlen)(name) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getgrnam_r)(name, grp, buf, buflen, result);
   if (!res) {
     if (result && *result) unpoison_group(ctx, *result);
@@ -1104,6 +1179,9 @@ INTERCEPTOR(int, getgrgid_r, u32 gid, __sanitizer_group *grp, char *buf,
             SIZE_T buflen, __sanitizer_group **result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getgrgid_r, gid, grp, buf, buflen, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getgrgid_r)(gid, grp, buf, buflen, result);
   if (!res) {
     if (result && *result) unpoison_group(ctx, *result);
@@ -1170,6 +1248,9 @@ INTERCEPTOR(int, getpwent_r, __sanitizer_passwd *pwbuf, char *buf,
             SIZE_T buflen, __sanitizer_passwd **pwbufp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getpwent_r, pwbuf, buf, buflen, pwbufp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getpwent_r)(pwbuf, buf, buflen, pwbufp);
   if (!res) {
     if (pwbufp && *pwbufp) unpoison_passwd(ctx, *pwbufp);
@@ -1182,6 +1263,9 @@ INTERCEPTOR(int, fgetpwent_r, void *fp, __sanitizer_passwd *pwbuf, char *buf,
             SIZE_T buflen, __sanitizer_passwd **pwbufp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fgetpwent_r, fp, pwbuf, buf, buflen, pwbufp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fgetpwent_r)(fp, pwbuf, buf, buflen, pwbufp);
   if (!res) {
     if (pwbufp && *pwbufp) unpoison_passwd(ctx, *pwbufp);
@@ -1194,6 +1278,9 @@ INTERCEPTOR(int, getgrent_r, __sanitizer_group *pwbuf, char *buf, SIZE_T buflen,
             __sanitizer_group **pwbufp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getgrent_r, pwbuf, buf, buflen, pwbufp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getgrent_r)(pwbuf, buf, buflen, pwbufp);
   if (!res) {
     if (pwbufp && *pwbufp) unpoison_group(ctx, *pwbufp);
@@ -1206,6 +1293,9 @@ INTERCEPTOR(int, fgetgrent_r, void *fp, __sanitizer_group *pwbuf, char *buf,
             SIZE_T buflen, __sanitizer_group **pwbufp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fgetgrent_r, fp, pwbuf, buf, buflen, pwbufp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fgetgrent_r)(fp, pwbuf, buf, buflen, pwbufp);
   if (!res) {
     if (pwbufp && *pwbufp) unpoison_group(ctx, *pwbufp);
@@ -1260,6 +1350,9 @@ INTERCEPTOR(void, endgrent, int dummy) {
 INTERCEPTOR(int, clock_getres, u32 clk_id, void *tp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, clock_getres, clk_id, tp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(clock_getres)(clk_id, tp);
   if (!res && tp) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, tp, struct_timespec_sz);
@@ -1269,6 +1362,9 @@ INTERCEPTOR(int, clock_getres, u32 clk_id, void *tp) {
 INTERCEPTOR(int, clock_gettime, u32 clk_id, void *tp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, clock_gettime, clk_id, tp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(clock_gettime)(clk_id, tp);
   if (!res) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, tp, struct_timespec_sz);
@@ -1293,6 +1389,9 @@ INTERCEPTOR(int, clock_settime, u32 clk_id, const void *tp) {
 INTERCEPTOR(int, getitimer, int which, void *curr_value) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getitimer, which, curr_value);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getitimer)(which, curr_value);
   if (!res && curr_value) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, curr_value, struct_itimerval_sz);
@@ -1304,6 +1403,9 @@ INTERCEPTOR(int, setitimer, int which, const void *new_value, void *old_value) {
   COMMON_INTERCEPTOR_ENTER(ctx, setitimer, which, new_value, old_value);
   if (new_value)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, new_value, struct_itimerval_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(setitimer)(which, new_value, old_value);
   if (!res && old_value) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, old_value, struct_itimerval_sz);
@@ -1433,15 +1535,27 @@ INTERCEPTOR(int, glob64, const char *pattern, int flags,
 INTERCEPTOR_WITH_SUFFIX(int, wait, int *status) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, wait, status);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(wait)(status);
   if (res != -1 && status)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, status, sizeof(*status));
   return res;
 }
+// On FreeBSD id_t is always 64-bit wide.
+#if SANITIZER_FREEBSD && (SANITIZER_WORDSIZE == 32)
+INTERCEPTOR_WITH_SUFFIX(int, waitid, int idtype, long long id, void *infop,
+                        int options) {
+#else
 INTERCEPTOR_WITH_SUFFIX(int, waitid, int idtype, int id, void *infop,
                         int options) {
+#endif
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, waitid, idtype, id, infop, options);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(waitid)(idtype, id, infop, options);
   if (res != -1 && infop)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, infop, siginfo_t_sz);
@@ -1450,6 +1564,9 @@ INTERCEPTOR_WITH_SUFFIX(int, waitid, int idtype, int id, void *infop,
 INTERCEPTOR_WITH_SUFFIX(int, waitpid, int pid, int *status, int options) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, waitpid, pid, status, options);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(waitpid)(pid, status, options);
   if (res != -1 && status)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, status, sizeof(*status));
@@ -1458,6 +1575,9 @@ INTERCEPTOR_WITH_SUFFIX(int, waitpid, int pid, int *status, int options) {
 INTERCEPTOR(int, wait3, int *status, int options, void *rusage) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, wait3, status, options, rusage);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(wait3)(status, options, rusage);
   if (res != -1) {
     if (status) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, status, sizeof(*status));
@@ -1469,6 +1589,9 @@ INTERCEPTOR(int, wait3, int *status, int options, void *rusage) {
 INTERCEPTOR(int, __wait4, int pid, int *status, int options, void *rusage) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, __wait4, pid, status, options, rusage);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(__wait4)(pid, status, options, rusage);
   if (res != -1) {
     if (status) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, status, sizeof(*status));
@@ -1481,6 +1604,9 @@ INTERCEPTOR(int, __wait4, int pid, int *status, int options, void *rusage) {
 INTERCEPTOR(int, wait4, int pid, int *status, int options, void *rusage) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, wait4, pid, status, options, rusage);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(wait4)(pid, status, options, rusage);
   if (res != -1) {
     if (status) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, status, sizeof(*status));
@@ -1507,6 +1633,9 @@ INTERCEPTOR(char *, inet_ntop, int af, const void *src, char *dst, u32 size) {
   uptr sz = __sanitizer_in_addr_sz(af);
   if (sz) COMMON_INTERCEPTOR_READ_RANGE(ctx, src, sz);
   // FIXME: figure out read size based on the address family.
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(inet_ntop)(af, src, dst, size);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, REAL(strlen)(res) + 1);
   return res;
@@ -1515,6 +1644,9 @@ INTERCEPTOR(int, inet_pton, int af, const char *src, void *dst) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, inet_pton, af, src, dst);
   // FIXME: figure out read size based on the address family.
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(inet_pton)(af, src, dst);
   if (res == 1) {
     uptr sz = __sanitizer_in_addr_sz(af);
@@ -1534,6 +1666,9 @@ INTERCEPTOR(int, inet_aton, const char *cp, void *dst) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, inet_aton, cp, dst);
   if (cp) COMMON_INTERCEPTOR_READ_RANGE(ctx, cp, REAL(strlen)(cp) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(inet_aton)(cp, dst);
   if (res != 0) {
     uptr sz = __sanitizer_in_addr_sz(af_inet);
@@ -1550,6 +1685,9 @@ INTERCEPTOR(int, inet_aton, const char *cp, void *dst) {
 INTERCEPTOR(int, pthread_getschedparam, uptr thread, int *policy, int *param) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, pthread_getschedparam, thread, policy, param);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(pthread_getschedparam)(thread, policy, param);
   if (res == 0) {
     if (policy) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, policy, sizeof(*policy));
@@ -1574,6 +1712,9 @@ INTERCEPTOR(int, getaddrinfo, char *node, char *service,
     COMMON_INTERCEPTOR_READ_RANGE(ctx, service, REAL(strlen)(service) + 1);
   if (hints)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, hints, sizeof(__sanitizer_addrinfo));
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getaddrinfo)(node, service, hints, out);
   if (res == 0 && out) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, out, sizeof(*out));
@@ -1603,6 +1744,9 @@ INTERCEPTOR(int, getnameinfo, void *sockaddr, unsigned salen, char *host,
                            serv, servlen, flags);
   // FIXME: consider adding READ_RANGE(sockaddr, salen)
   // There is padding in in_addr that may make this too noisy
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res =
       REAL(getnameinfo)(sockaddr, salen, host, hostlen, serv, servlen, flags);
   if (res == 0) {
@@ -1624,6 +1768,9 @@ INTERCEPTOR(int, getsockname, int sock_fd, void *addr, int *addrlen) {
   COMMON_INTERCEPTOR_ENTER(ctx, getsockname, sock_fd, addr, addrlen);
   COMMON_INTERCEPTOR_READ_RANGE(ctx, addrlen, sizeof(*addrlen));
   int addrlen_in = *addrlen;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getsockname)(sock_fd, addr, addrlen);
   if (res == 0) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, addr, Min(addrlen_in, *addrlen));
@@ -1701,11 +1848,38 @@ INTERCEPTOR(struct __sanitizer_hostent *, gethostbyname2, char *name, int af) {
 #endif
 
 #if SANITIZER_INTERCEPT_GETHOSTBYNAME_R
+INTERCEPTOR(int, gethostbyname_r, char *name, struct __sanitizer_hostent *ret,
+            char *buf, SIZE_T buflen, __sanitizer_hostent **result,
+            int *h_errnop) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER(ctx, gethostbyname_r, name, ret, buf, buflen, result,
+                           h_errnop);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
+  int res = REAL(gethostbyname_r)(name, ret, buf, buflen, result, h_errnop);
+  if (result) {
+    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
+    if (res == 0 && *result) write_hostent(ctx, *result);
+  }
+  if (h_errnop)
+    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, h_errnop, sizeof(*h_errnop));
+  return res;
+}
+#define INIT_GETHOSTBYNAME_R COMMON_INTERCEPT_FUNCTION(gethostbyname_r);
+#else
+#define INIT_GETHOSTBYNAME_R
+#endif
+
+#if SANITIZER_INTERCEPT_GETHOSTENT_R
 INTERCEPTOR(int, gethostent_r, struct __sanitizer_hostent *ret, char *buf,
             SIZE_T buflen, __sanitizer_hostent **result, int *h_errnop) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, gethostent_r, ret, buf, buflen, result,
                            h_errnop);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(gethostent_r)(ret, buf, buflen, result, h_errnop);
   if (result) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
@@ -1715,7 +1889,13 @@ INTERCEPTOR(int, gethostent_r, struct __sanitizer_hostent *ret, char *buf,
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, h_errnop, sizeof(*h_errnop));
   return res;
 }
+#define INIT_GETHOSTENT_R                  \
+  COMMON_INTERCEPT_FUNCTION(gethostent_r);
+#else
+#define INIT_GETHOSTENT_R
+#endif
 
+#if SANITIZER_INTERCEPT_GETHOSTBYADDR_R
 INTERCEPTOR(int, gethostbyaddr_r, void *addr, int len, int type,
             struct __sanitizer_hostent *ret, char *buf, SIZE_T buflen,
             __sanitizer_hostent **result, int *h_errnop) {
@@ -1723,6 +1903,9 @@ INTERCEPTOR(int, gethostbyaddr_r, void *addr, int len, int type,
   COMMON_INTERCEPTOR_ENTER(ctx, gethostbyaddr_r, addr, len, type, ret, buf,
                            buflen, result, h_errnop);
   COMMON_INTERCEPTOR_READ_RANGE(ctx, addr, len);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(gethostbyaddr_r)(addr, len, type, ret, buf, buflen, result,
                                   h_errnop);
   if (result) {
@@ -1733,29 +1916,22 @@ INTERCEPTOR(int, gethostbyaddr_r, void *addr, int len, int type,
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, h_errnop, sizeof(*h_errnop));
   return res;
 }
+#define INIT_GETHOSTBYADDR_R                  \
+  COMMON_INTERCEPT_FUNCTION(gethostbyaddr_r);
+#else
+#define INIT_GETHOSTBYADDR_R
+#endif
 
-INTERCEPTOR(int, gethostbyname_r, char *name, struct __sanitizer_hostent *ret,
-            char *buf, SIZE_T buflen, __sanitizer_hostent **result,
-            int *h_errnop) {
-  void *ctx;
-  COMMON_INTERCEPTOR_ENTER(ctx, gethostbyname_r, name, ret, buf, buflen, result,
-                           h_errnop);
-  int res = REAL(gethostbyname_r)(name, ret, buf, buflen, result, h_errnop);
-  if (result) {
-    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
-    if (res == 0 && *result) write_hostent(ctx, *result);
-  }
-  if (h_errnop)
-    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, h_errnop, sizeof(*h_errnop));
-  return res;
-}
-
+#if SANITIZER_INTERCEPT_GETHOSTBYNAME2_R
 INTERCEPTOR(int, gethostbyname2_r, char *name, int af,
             struct __sanitizer_hostent *ret, char *buf, SIZE_T buflen,
             __sanitizer_hostent **result, int *h_errnop) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, gethostbyname2_r, name, af, ret, buf, buflen,
                            result, h_errnop);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res =
       REAL(gethostbyname2_r)(name, af, ret, buf, buflen, result, h_errnop);
   if (result) {
@@ -1766,13 +1942,10 @@ INTERCEPTOR(int, gethostbyname2_r, char *name, int af,
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, h_errnop, sizeof(*h_errnop));
   return res;
 }
-#define INIT_GETHOSTBYNAME_R                  \
-  COMMON_INTERCEPT_FUNCTION(gethostent_r);    \
-  COMMON_INTERCEPT_FUNCTION(gethostbyaddr_r); \
-  COMMON_INTERCEPT_FUNCTION(gethostbyname_r); \
+#define INIT_GETHOSTBYNAME2_R                  \
   COMMON_INTERCEPT_FUNCTION(gethostbyname2_r);
 #else
-#define INIT_GETHOSTBYNAME_R
+#define INIT_GETHOSTBYNAME2_R
 #endif
 
 #if SANITIZER_INTERCEPT_GETSOCKOPT
@@ -1782,6 +1955,9 @@ INTERCEPTOR(int, getsockopt, int sockfd, int level, int optname, void *optval,
   COMMON_INTERCEPTOR_ENTER(ctx, getsockopt, sockfd, level, optname, optval,
                            optlen);
   if (optlen) COMMON_INTERCEPTOR_READ_RANGE(ctx, optlen, sizeof(*optlen));
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getsockopt)(sockfd, level, optname, optval, optlen);
   if (res == 0)
     if (optval && optlen) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, optval, *optlen);
@@ -1823,6 +1999,9 @@ INTERCEPTOR(int, accept4, int fd, void *addr, unsigned *addrlen, int f) {
     COMMON_INTERCEPTOR_READ_RANGE(ctx, addrlen, sizeof(*addrlen));
     addrlen0 = *addrlen;
   }
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int fd2 = REAL(accept4)(fd, addr, addrlen, f);
   if (fd2 >= 0) {
     if (fd >= 0) COMMON_INTERCEPTOR_FD_SOCKET_ACCEPT(ctx, fd, fd2);
@@ -1840,6 +2019,9 @@ INTERCEPTOR(int, accept4, int fd, void *addr, unsigned *addrlen, int f) {
 INTERCEPTOR(double, modf, double x, double *iptr) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, modf, x, iptr);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   double res = REAL(modf)(x, iptr);
   if (iptr) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, iptr, sizeof(*iptr));
@@ -1849,6 +2031,9 @@ INTERCEPTOR(double, modf, double x, double *iptr) {
 INTERCEPTOR(float, modff, float x, float *iptr) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, modff, x, iptr);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   float res = REAL(modff)(x, iptr);
   if (iptr) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, iptr, sizeof(*iptr));
@@ -1858,6 +2043,9 @@ INTERCEPTOR(float, modff, float x, float *iptr) {
 INTERCEPTOR(long double, modfl, long double x, long double *iptr) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, modfl, x, iptr);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   long double res = REAL(modfl)(x, iptr);
   if (iptr) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, iptr, sizeof(*iptr));
@@ -1890,6 +2078,9 @@ INTERCEPTOR(SSIZE_T, recvmsg, int fd, struct __sanitizer_msghdr *msg,
             int flags) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, recvmsg, fd, msg, flags);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(recvmsg)(fd, msg, flags);
   if (res >= 0) {
     if (fd >= 0) COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd);
@@ -1911,6 +2102,9 @@ INTERCEPTOR(int, getpeername, int sockfd, void *addr, unsigned *addrlen) {
   COMMON_INTERCEPTOR_ENTER(ctx, getpeername, sockfd, addr, addrlen);
   unsigned addr_sz;
   if (addrlen) addr_sz = *addrlen;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getpeername)(sockfd, addr, addrlen);
   if (!res && addr && addrlen)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, addr, Min(addr_sz, *addrlen));
@@ -1924,6 +2118,9 @@ INTERCEPTOR(int, getpeername, int sockfd, void *addr, unsigned *addrlen) {
 #if SANITIZER_INTERCEPT_SYSINFO
 INTERCEPTOR(int, sysinfo, void *info) {
   void *ctx;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   COMMON_INTERCEPTOR_ENTER(ctx, sysinfo, info);
   int res = REAL(sysinfo)(info);
   if (!res && info)
@@ -1939,6 +2136,9 @@ INTERCEPTOR(int, sysinfo, void *info) {
 INTERCEPTOR(__sanitizer_dirent *, readdir, void *dirp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, readdir, dirp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_dirent *res = REAL(readdir)(dirp);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, res->d_reclen);
   return res;
@@ -1948,6 +2148,9 @@ INTERCEPTOR(int, readdir_r, void *dirp, __sanitizer_dirent *entry,
             __sanitizer_dirent **result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, readdir_r, dirp, entry, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(readdir_r)(dirp, entry, result);
   if (!res) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
@@ -1968,6 +2171,9 @@ INTERCEPTOR(int, readdir_r, void *dirp, __sanitizer_dirent *entry,
 INTERCEPTOR(__sanitizer_dirent64 *, readdir64, void *dirp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, readdir64, dirp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_dirent64 *res = REAL(readdir64)(dirp);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, res->d_reclen);
   return res;
@@ -1977,6 +2183,9 @@ INTERCEPTOR(int, readdir64_r, void *dirp, __sanitizer_dirent64 *entry,
             __sanitizer_dirent64 **result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, readdir64_r, dirp, entry, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(readdir64_r)(dirp, entry, result);
   if (!res) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
@@ -2012,10 +2221,13 @@ INTERCEPTOR(uptr, ptrace, int request, int pid, void *addr, void *data) {
     }
   }
 
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   uptr res = REAL(ptrace)(request, pid, addr, data);
 
   if (!res && data) {
-    // Note that PEEK* requests assing different meaning to the return value.
+    // Note that PEEK* requests assign different meaning to the return value.
     // This function does not handle them (nor does it need to).
     if (request == ptrace_getregs)
       COMMON_INTERCEPTOR_WRITE_RANGE(ctx, data, struct_user_regs_struct_sz);
@@ -2025,6 +2237,8 @@ INTERCEPTOR(uptr, ptrace, int request, int pid, void *addr, void *data) {
       COMMON_INTERCEPTOR_WRITE_RANGE(ctx, data, struct_user_fpxregs_struct_sz);
     else if (request == ptrace_getsiginfo)
       COMMON_INTERCEPTOR_WRITE_RANGE(ctx, data, siginfo_t_sz);
+    else if (request == ptrace_geteventmsg)
+      COMMON_INTERCEPTOR_WRITE_RANGE(ctx, data, sizeof(unsigned long));
     else if (request == ptrace_getregset) {
       __sanitizer_iovec *iov = (__sanitizer_iovec *)data;
       COMMON_INTERCEPTOR_WRITE_RANGE(ctx, iov->iov_base, iov->iov_len);
@@ -2058,6 +2272,9 @@ INTERCEPTOR(char *, setlocale, int category, char *locale) {
 INTERCEPTOR(char *, getcwd, char *buf, SIZE_T size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getcwd, buf, size);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(getcwd)(buf, size);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, REAL(strlen)(res) + 1);
   return res;
@@ -2071,6 +2288,9 @@ INTERCEPTOR(char *, getcwd, char *buf, SIZE_T size) {
 INTERCEPTOR(char *, get_current_dir_name, int fake) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, get_current_dir_name, fake);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(get_current_dir_name)(fake);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, REAL(strlen)(res) + 1);
   return res;
@@ -2086,6 +2306,9 @@ INTERCEPTOR(char *, get_current_dir_name, int fake) {
 INTERCEPTOR(INTMAX_T, strtoimax, const char *nptr, char **endptr, int base) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, strtoimax, nptr, endptr, base);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   INTMAX_T res = REAL(strtoimax)(nptr, endptr, base);
   if (endptr) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, endptr, sizeof(*endptr));
   return res;
@@ -2094,6 +2317,9 @@ INTERCEPTOR(INTMAX_T, strtoimax, const char *nptr, char **endptr, int base) {
 INTERCEPTOR(INTMAX_T, strtoumax, const char *nptr, char **endptr, int base) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, strtoumax, nptr, endptr, base);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   INTMAX_T res = REAL(strtoumax)(nptr, endptr, base);
   if (endptr) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, endptr, sizeof(*endptr));
   return res;
@@ -2110,6 +2336,9 @@ INTERCEPTOR(INTMAX_T, strtoumax, const char *nptr, char **endptr, int base) {
 INTERCEPTOR(SIZE_T, mbstowcs, wchar_t *dest, const char *src, SIZE_T len) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, mbstowcs, dest, src, len);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(mbstowcs)(dest, src, len);
   if (res != (SIZE_T) - 1 && dest) {
     SIZE_T write_cnt = res + (res < len);
@@ -2124,6 +2353,9 @@ INTERCEPTOR(SIZE_T, mbsrtowcs, wchar_t *dest, const char **src, SIZE_T len,
   COMMON_INTERCEPTOR_ENTER(ctx, mbsrtowcs, dest, src, len, ps);
   if (src) COMMON_INTERCEPTOR_READ_RANGE(ctx, src, sizeof(*src));
   if (ps) COMMON_INTERCEPTOR_READ_RANGE(ctx, ps, mbstate_t_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(mbsrtowcs)(dest, src, len, ps);
   if (res != (SIZE_T)(-1) && dest && src) {
     // This function, and several others, may or may not write the terminating
@@ -2151,6 +2383,9 @@ INTERCEPTOR(SIZE_T, mbsnrtowcs, wchar_t *dest, const char **src, SIZE_T nms,
     if (nms) COMMON_INTERCEPTOR_READ_RANGE(ctx, *src, nms);
   }
   if (ps) COMMON_INTERCEPTOR_READ_RANGE(ctx, ps, mbstate_t_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(mbsnrtowcs)(dest, src, nms, len, ps);
   if (res != (SIZE_T)(-1) && dest && src) {
     SIZE_T write_cnt = res + !*src;
@@ -2168,6 +2403,9 @@ INTERCEPTOR(SIZE_T, mbsnrtowcs, wchar_t *dest, const char **src, SIZE_T nms,
 INTERCEPTOR(SIZE_T, wcstombs, char *dest, const wchar_t *src, SIZE_T len) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, wcstombs, dest, src, len);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(wcstombs)(dest, src, len);
   if (res != (SIZE_T) - 1 && dest) {
     SIZE_T write_cnt = res + (res < len);
@@ -2182,6 +2420,9 @@ INTERCEPTOR(SIZE_T, wcsrtombs, char *dest, const wchar_t **src, SIZE_T len,
   COMMON_INTERCEPTOR_ENTER(ctx, wcsrtombs, dest, src, len, ps);
   if (src) COMMON_INTERCEPTOR_READ_RANGE(ctx, src, sizeof(*src));
   if (ps) COMMON_INTERCEPTOR_READ_RANGE(ctx, ps, mbstate_t_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(wcsrtombs)(dest, src, len, ps);
   if (res != (SIZE_T) - 1 && dest && src) {
     SIZE_T write_cnt = res + !*src;
@@ -2207,6 +2448,9 @@ INTERCEPTOR(SIZE_T, wcsnrtombs, char *dest, const wchar_t **src, SIZE_T nms,
     if (nms) COMMON_INTERCEPTOR_READ_RANGE(ctx, *src, nms);
   }
   if (ps) COMMON_INTERCEPTOR_READ_RANGE(ctx, ps, mbstate_t_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(wcsnrtombs)(dest, src, nms, len, ps);
   if (res != (SIZE_T) - 1 && dest && src) {
     SIZE_T write_cnt = res + !*src;
@@ -2224,6 +2468,9 @@ INTERCEPTOR(SIZE_T, wcsnrtombs, char *dest, const wchar_t **src, SIZE_T nms,
 INTERCEPTOR(int, tcgetattr, int fd, void *termios_p) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, tcgetattr, fd, termios_p);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(tcgetattr)(fd, termios_p);
   if (!res && termios_p)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, termios_p, struct_termios_sz);
@@ -2278,6 +2525,9 @@ INTERCEPTOR(char *, canonicalize_file_name, const char *path) {
 INTERCEPTOR(SIZE_T, confstr, int name, char *buf, SIZE_T len) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, confstr, name, buf, len);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(confstr)(name, buf, len);
   if (buf && res)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, res < len ? res : len);
@@ -2292,6 +2542,9 @@ INTERCEPTOR(SIZE_T, confstr, int name, char *buf, SIZE_T len) {
 INTERCEPTOR(int, sched_getaffinity, int pid, SIZE_T cpusetsize, void *mask) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sched_getaffinity, pid, cpusetsize, mask);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sched_getaffinity)(pid, cpusetsize, mask);
   if (mask && !res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, mask, cpusetsize);
   return res;
@@ -2318,6 +2571,9 @@ INTERCEPTOR(char *, strerror, int errnum) {
 INTERCEPTOR(char *, strerror_r, int errnum, char *buf, SIZE_T buflen) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, strerror_r, errnum, buf, buflen);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(strerror_r)(errnum, buf, buflen);
   // There are 2 versions of strerror_r:
   //  * POSIX version returns 0 on success, negative error code on failure,
@@ -2346,6 +2602,9 @@ INTERCEPTOR(char *, strerror_r, int errnum, char *buf, SIZE_T buflen) {
 INTERCEPTOR(int, __xpg_strerror_r, int errnum, char *buf, SIZE_T buflen) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, __xpg_strerror_r, errnum, buf, buflen);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(__xpg_strerror_r)(errnum, buf, buflen);
   // This version always returns a null-terminated string.
   if (buf && buflen)
@@ -2388,6 +2647,9 @@ INTERCEPTOR(int, scandir, char *dirp, __sanitizer_dirent ***namelist,
   if (dirp) COMMON_INTERCEPTOR_READ_RANGE(ctx, dirp, REAL(strlen)(dirp) + 1);
   scandir_filter = filter;
   scandir_compar = compar;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(scandir)(dirp, namelist, filter ? wrapped_scandir_filter : 0,
                           compar ? wrapped_scandir_compar : 0);
   scandir_filter = 0;
@@ -2437,6 +2699,9 @@ INTERCEPTOR(int, scandir64, char *dirp, __sanitizer_dirent64 ***namelist,
   if (dirp) COMMON_INTERCEPTOR_READ_RANGE(ctx, dirp, REAL(strlen)(dirp) + 1);
   scandir64_filter = filter;
   scandir64_compar = compar;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res =
       REAL(scandir64)(dirp, namelist, filter ? wrapped_scandir64_filter : 0,
                       compar ? wrapped_scandir64_compar : 0);
@@ -2460,6 +2725,9 @@ INTERCEPTOR(int, scandir64, char *dirp, __sanitizer_dirent64 ***namelist,
 INTERCEPTOR(int, getgroups, int size, u32 *lst) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getgroups, size, lst);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getgroups)(size, lst);
   if (res && lst) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, lst, res * sizeof(*lst));
   return res;
@@ -2523,6 +2791,9 @@ INTERCEPTOR(int, wordexp, char *s, __sanitizer_wordexp_t *p, int flags) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, wordexp, s, p, flags);
   if (s) COMMON_INTERCEPTOR_READ_RANGE(ctx, s, REAL(strlen)(s) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(wordexp)(s, p, flags);
   if (!res && p) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, p, sizeof(*p));
@@ -2546,6 +2817,9 @@ INTERCEPTOR(int, sigwait, __sanitizer_sigset_t *set, int *sig) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigwait, set, sig);
   // FIXME: read sigset_t when all of sigemptyset, etc are intercepted
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigwait)(set, sig);
   if (!res && sig) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, sig, sizeof(*sig));
   return res;
@@ -2560,6 +2834,9 @@ INTERCEPTOR(int, sigwaitinfo, __sanitizer_sigset_t *set, void *info) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigwaitinfo, set, info);
   // FIXME: read sigset_t when all of sigemptyset, etc are intercepted
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigwaitinfo)(set, info);
   if (res > 0 && info) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, info, siginfo_t_sz);
   return res;
@@ -2576,6 +2853,9 @@ INTERCEPTOR(int, sigtimedwait, __sanitizer_sigset_t *set, void *info,
   COMMON_INTERCEPTOR_ENTER(ctx, sigtimedwait, set, info, timeout);
   if (timeout) COMMON_INTERCEPTOR_READ_RANGE(ctx, timeout, struct_timespec_sz);
   // FIXME: read sigset_t when all of sigemptyset, etc are intercepted
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigtimedwait)(set, info, timeout);
   if (res > 0 && info) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, info, siginfo_t_sz);
   return res;
@@ -2589,6 +2869,9 @@ INTERCEPTOR(int, sigtimedwait, __sanitizer_sigset_t *set, void *info,
 INTERCEPTOR(int, sigemptyset, __sanitizer_sigset_t *set) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigemptyset, set);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigemptyset)(set);
   if (!res && set) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, set, sizeof(*set));
   return res;
@@ -2597,6 +2880,9 @@ INTERCEPTOR(int, sigemptyset, __sanitizer_sigset_t *set) {
 INTERCEPTOR(int, sigfillset, __sanitizer_sigset_t *set) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigfillset, set);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigfillset)(set);
   if (!res && set) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, set, sizeof(*set));
   return res;
@@ -2612,6 +2898,9 @@ INTERCEPTOR(int, sigfillset, __sanitizer_sigset_t *set) {
 INTERCEPTOR(int, sigpending, __sanitizer_sigset_t *set) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigpending, set);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigpending)(set);
   if (!res && set) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, set, sizeof(*set));
   return res;
@@ -2627,6 +2916,9 @@ INTERCEPTOR(int, sigprocmask, int how, __sanitizer_sigset_t *set,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sigprocmask, how, set, oldset);
   // FIXME: read sigset_t when all of sigemptyset, etc are intercepted
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(sigprocmask)(how, set, oldset);
   if (!res && oldset)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, oldset, sizeof(*oldset));
@@ -2641,6 +2933,9 @@ INTERCEPTOR(int, sigprocmask, int how, __sanitizer_sigset_t *set,
 INTERCEPTOR(int, backtrace, void **buffer, int size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, backtrace, buffer, size);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(backtrace)(buffer, size);
   if (res && buffer)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buffer, res * sizeof(*buffer));
@@ -2652,6 +2947,9 @@ INTERCEPTOR(char **, backtrace_symbols, void **buffer, int size) {
   COMMON_INTERCEPTOR_ENTER(ctx, backtrace_symbols, buffer, size);
   if (buffer && size)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, buffer, size * sizeof(*buffer));
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char **res = REAL(backtrace_symbols)(buffer, size);
   if (res && size) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, size * sizeof(*res));
@@ -2757,6 +3055,9 @@ INTERCEPTOR(int, statfs, char *path, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, statfs, path, buf);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(statfs)(path, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statfs_sz);
   return res;
@@ -2764,6 +3065,9 @@ INTERCEPTOR(int, statfs, char *path, void *buf) {
 INTERCEPTOR(int, fstatfs, int fd, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fstatfs, fd, buf);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fstatfs)(fd, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statfs_sz);
   return res;
@@ -2780,6 +3084,9 @@ INTERCEPTOR(int, statfs64, char *path, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, statfs64, path, buf);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(statfs64)(path, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statfs64_sz);
   return res;
@@ -2787,6 +3094,9 @@ INTERCEPTOR(int, statfs64, char *path, void *buf) {
 INTERCEPTOR(int, fstatfs64, int fd, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fstatfs64, fd, buf);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fstatfs64)(fd, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statfs64_sz);
   return res;
@@ -2803,6 +3113,9 @@ INTERCEPTOR(int, statvfs, char *path, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, statvfs, path, buf);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(statvfs)(path, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statvfs_sz);
   return res;
@@ -2810,6 +3123,9 @@ INTERCEPTOR(int, statvfs, char *path, void *buf) {
 INTERCEPTOR(int, fstatvfs, int fd, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fstatvfs, fd, buf);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fstatvfs)(fd, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statvfs_sz);
   return res;
@@ -2826,6 +3142,9 @@ INTERCEPTOR(int, statvfs64, char *path, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, statvfs64, path, buf);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(statvfs64)(path, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statvfs64_sz);
   return res;
@@ -2833,6 +3152,9 @@ INTERCEPTOR(int, statvfs64, char *path, void *buf) {
 INTERCEPTOR(int, fstatvfs64, int fd, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fstatvfs64, fd, buf);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(fstatvfs64)(fd, buf);
   if (!res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, buf, struct_statvfs64_sz);
   return res;
@@ -2857,7 +3179,7 @@ INTERCEPTOR(int, initgroups, char *user, u32 group) {
 #define INIT_INITGROUPS
 #endif
 
-#if SANITIZER_INTERCEPT_ETHER
+#if SANITIZER_INTERCEPT_ETHER_NTOA_ATON
 INTERCEPTOR(char *, ether_ntoa, __sanitizer_ether_addr *addr) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ether_ntoa, addr);
@@ -2874,10 +3196,21 @@ INTERCEPTOR(__sanitizer_ether_addr *, ether_aton, char *buf) {
   if (res) COMMON_INTERCEPTOR_INITIALIZE_RANGE(res, sizeof(*res));
   return res;
 }
+#define INIT_ETHER_NTOA_ATON             \
+  COMMON_INTERCEPT_FUNCTION(ether_ntoa); \
+  COMMON_INTERCEPT_FUNCTION(ether_aton);
+#else
+#define INIT_ETHER_NTOA_ATON
+#endif
+
+#if SANITIZER_INTERCEPT_ETHER_HOST
 INTERCEPTOR(int, ether_ntohost, char *hostname, __sanitizer_ether_addr *addr) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ether_ntohost, hostname, addr);
   if (addr) COMMON_INTERCEPTOR_READ_RANGE(ctx, addr, sizeof(*addr));
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(ether_ntohost)(hostname, addr);
   if (!res && hostname)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, hostname, REAL(strlen)(hostname) + 1);
@@ -2888,6 +3221,9 @@ INTERCEPTOR(int, ether_hostton, char *hostname, __sanitizer_ether_addr *addr) {
   COMMON_INTERCEPTOR_ENTER(ctx, ether_hostton, hostname, addr);
   if (hostname)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, hostname, REAL(strlen)(hostname) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(ether_hostton)(hostname, addr);
   if (!res && addr) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, addr, sizeof(*addr));
   return res;
@@ -2897,6 +3233,9 @@ INTERCEPTOR(int, ether_line, char *line, __sanitizer_ether_addr *addr,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ether_line, line, addr, hostname);
   if (line) COMMON_INTERCEPTOR_READ_RANGE(ctx, line, REAL(strlen)(line) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(ether_line)(line, addr, hostname);
   if (!res) {
     if (addr) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, addr, sizeof(*addr));
@@ -2905,14 +3244,12 @@ INTERCEPTOR(int, ether_line, char *line, __sanitizer_ether_addr *addr,
   }
   return res;
 }
-#define INIT_ETHER                          \
-  COMMON_INTERCEPT_FUNCTION(ether_ntoa);    \
-  COMMON_INTERCEPT_FUNCTION(ether_aton);    \
+#define INIT_ETHER_HOST                     \
   COMMON_INTERCEPT_FUNCTION(ether_ntohost); \
   COMMON_INTERCEPT_FUNCTION(ether_hostton); \
   COMMON_INTERCEPT_FUNCTION(ether_line);
 #else
-#define INIT_ETHER
+#define INIT_ETHER_HOST
 #endif
 
 #if SANITIZER_INTERCEPT_ETHER_R
@@ -2920,6 +3257,9 @@ INTERCEPTOR(char *, ether_ntoa_r, __sanitizer_ether_addr *addr, char *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ether_ntoa_r, addr, buf);
   if (addr) COMMON_INTERCEPTOR_READ_RANGE(ctx, addr, sizeof(*addr));
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(ether_ntoa_r)(addr, buf);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, REAL(strlen)(res) + 1);
   return res;
@@ -2929,6 +3269,9 @@ INTERCEPTOR(__sanitizer_ether_addr *, ether_aton_r, char *buf,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ether_aton_r, buf, addr);
   if (buf) COMMON_INTERCEPTOR_READ_RANGE(ctx, buf, REAL(strlen)(buf) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_ether_addr *res = REAL(ether_aton_r)(buf, addr);
   if (res) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, sizeof(*res));
   return res;
@@ -2944,6 +3287,9 @@ INTERCEPTOR(__sanitizer_ether_addr *, ether_aton_r, char *buf,
 INTERCEPTOR(int, shmctl, int shmid, int cmd, void *buf) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, shmctl, shmid, cmd, buf);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(shmctl)(shmid, cmd, buf);
   if (res >= 0) {
     unsigned sz = 0;
@@ -2966,6 +3312,9 @@ INTERCEPTOR(int, shmctl, int shmid, int cmd, void *buf) {
 INTERCEPTOR(int, random_r, void *buf, u32 *result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, random_r, buf, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(random_r)(buf, result);
   if (!res && result)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
@@ -2976,16 +3325,33 @@ INTERCEPTOR(int, random_r, void *buf, u32 *result) {
 #define INIT_RANDOM_R
 #endif
 
-#if SANITIZER_INTERCEPT_PTHREAD_ATTR_GET || \
-    SANITIZER_INTERCEPT_PTHREAD_ATTR_GETINHERITSSCHED
-#define INTERCEPTOR_PTHREAD_ATTR_GET(what, sz)                      \
-  INTERCEPTOR(int, pthread_attr_get##what, void *attr, void *r) {   \
-    void *ctx;                                                      \
-    COMMON_INTERCEPTOR_ENTER(ctx, pthread_attr_get##what, attr, r); \
-    int res = REAL(pthread_attr_get##what)(attr, r);                \
-    if (!res && r) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, r, sz);      \
-    return res;                                                     \
+// FIXME: under ASan the REAL() call below may write to freed memory and corrupt
+// its metadata. See
+// https://code.google.com/p/address-sanitizer/issues/detail?id=321.
+#if SANITIZER_INTERCEPT_PTHREAD_ATTR_GET ||              \
+    SANITIZER_INTERCEPT_PTHREAD_ATTR_GETINHERITSSCHED || \
+    SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GET ||         \
+    SANITIZER_INTERCEPT_PTHREAD_RWLOCKATTR_GET ||        \
+    SANITIZER_INTERCEPT_PTHREAD_CONDATTR_GET ||          \
+    SANITIZER_INTERCEPT_PTHREAD_BARRIERATTR_GET
+#define INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(fn, sz)            \
+  INTERCEPTOR(int, fn, void *attr, void *r) {                  \
+    void *ctx;                                                 \
+    COMMON_INTERCEPTOR_ENTER(ctx, fn, attr, r);                \
+    int res = REAL(fn)(attr, r);                               \
+    if (!res && r) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, r, sz); \
+    return res;                                                \
   }
+#define INTERCEPTOR_PTHREAD_ATTR_GET(what, sz) \
+  INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(pthread_attr_get##what, sz)
+#define INTERCEPTOR_PTHREAD_MUTEXATTR_GET(what, sz) \
+  INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(pthread_mutexattr_get##what, sz)
+#define INTERCEPTOR_PTHREAD_RWLOCKATTR_GET(what, sz) \
+  INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(pthread_rwlockattr_get##what, sz)
+#define INTERCEPTOR_PTHREAD_CONDATTR_GET(what, sz) \
+  INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(pthread_condattr_get##what, sz)
+#define INTERCEPTOR_PTHREAD_BARRIERATTR_GET(what, sz) \
+  INTERCEPTOR_PTHREAD_OBJECT_ATTR_GET(pthread_barrierattr_get##what, sz)
 #endif
 
 #if SANITIZER_INTERCEPT_PTHREAD_ATTR_GET
@@ -2998,6 +3364,9 @@ INTERCEPTOR_PTHREAD_ATTR_GET(stacksize, sizeof(SIZE_T))
 INTERCEPTOR(int, pthread_attr_getstack, void *attr, void **addr, SIZE_T *size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, pthread_attr_getstack, attr, addr, size);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(pthread_attr_getstack)(attr, addr, size);
   if (!res) {
     if (addr) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, addr, sizeof(*addr));
@@ -3044,6 +3413,9 @@ INTERCEPTOR(int, pthread_attr_getaffinity_np, void *attr, SIZE_T cpusetsize,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, pthread_attr_getaffinity_np, attr, cpusetsize,
                            cpuset);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(pthread_attr_getaffinity_np)(attr, cpusetsize, cpuset);
   if (!res && cpusetsize && cpuset)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, cpuset, cpusetsize);
@@ -3056,6 +3428,94 @@ INTERCEPTOR(int, pthread_attr_getaffinity_np, void *attr, SIZE_T cpusetsize,
 #define INIT_PTHREAD_ATTR_GETAFFINITY_NP
 #endif
 
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPSHARED
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(pshared, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETPSHARED \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_getpshared);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETPSHARED
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETTYPE
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(type, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETTYPE \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_gettype);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETTYPE
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPROTOCOL
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(protocol, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETPROTOCOL \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_getprotocol);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETPROTOCOL
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPRIOCEILING
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(prioceiling, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETPRIOCEILING \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_getprioceiling);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETPRIOCEILING
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETROBUST
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(robust, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETROBUST \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_getrobust);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETROBUST
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETROBUST_NP
+INTERCEPTOR_PTHREAD_MUTEXATTR_GET(robust_np, sizeof(int))
+#define INIT_PTHREAD_MUTEXATTR_GETROBUST_NP \
+  COMMON_INTERCEPT_FUNCTION(pthread_mutexattr_getrobust_np);
+#else
+#define INIT_PTHREAD_MUTEXATTR_GETROBUST_NP
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_RWLOCKATTR_GETPSHARED
+INTERCEPTOR_PTHREAD_RWLOCKATTR_GET(pshared, sizeof(int))
+#define INIT_PTHREAD_RWLOCKATTR_GETPSHARED \
+  COMMON_INTERCEPT_FUNCTION(pthread_rwlockattr_getpshared);
+#else
+#define INIT_PTHREAD_RWLOCKATTR_GETPSHARED
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_RWLOCKATTR_GETKIND_NP
+INTERCEPTOR_PTHREAD_RWLOCKATTR_GET(kind_np, sizeof(int))
+#define INIT_PTHREAD_RWLOCKATTR_GETKIND_NP \
+  COMMON_INTERCEPT_FUNCTION(pthread_rwlockattr_getkind_np);
+#else
+#define INIT_PTHREAD_RWLOCKATTR_GETKIND_NP
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_CONDATTR_GETPSHARED
+INTERCEPTOR_PTHREAD_CONDATTR_GET(pshared, sizeof(int))
+#define INIT_PTHREAD_CONDATTR_GETPSHARED \
+  COMMON_INTERCEPT_FUNCTION(pthread_condattr_getpshared);
+#else
+#define INIT_PTHREAD_CONDATTR_GETPSHARED
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_CONDATTR_GETCLOCK
+INTERCEPTOR_PTHREAD_CONDATTR_GET(clock, sizeof(int))
+#define INIT_PTHREAD_CONDATTR_GETCLOCK \
+  COMMON_INTERCEPT_FUNCTION(pthread_condattr_getclock);
+#else
+#define INIT_PTHREAD_CONDATTR_GETCLOCK
+#endif
+
+#if SANITIZER_INTERCEPT_PTHREAD_BARRIERATTR_GETPSHARED
+INTERCEPTOR_PTHREAD_BARRIERATTR_GET(pshared, sizeof(int)) // !mac !android
+#define INIT_PTHREAD_BARRIERATTR_GETPSHARED \
+  COMMON_INTERCEPT_FUNCTION(pthread_barrierattr_getpshared);
+#else
+#define INIT_PTHREAD_BARRIERATTR_GETPSHARED
+#endif
+
 #if SANITIZER_INTERCEPT_TMPNAM
 INTERCEPTOR(char *, tmpnam, char *s) {
   void *ctx;
@@ -3063,6 +3523,9 @@ INTERCEPTOR(char *, tmpnam, char *s) {
   char *res = REAL(tmpnam)(s);
   if (res) {
     if (s)
+      // FIXME: under ASan the call below may write to freed memory and corrupt
+      // its metadata. See
+      // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
       COMMON_INTERCEPTOR_WRITE_RANGE(ctx, s, REAL(strlen)(s) + 1);
     else
       COMMON_INTERCEPTOR_INITIALIZE_RANGE(res, REAL(strlen)(res) + 1);
@@ -3078,6 +3541,9 @@ INTERCEPTOR(char *, tmpnam, char *s) {
 INTERCEPTOR(char *, tmpnam_r, char *s) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, tmpnam_r, s);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(tmpnam_r)(s);
   if (res && s) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, s, REAL(strlen)(s) + 1);
   return res;
@@ -3118,6 +3584,9 @@ INTERCEPTOR(int, pthread_setname_np, uptr thread, const char *name) {
 INTERCEPTOR(void, sincos, double x, double *sin, double *cos) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sincos, x, sin, cos);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   REAL(sincos)(x, sin, cos);
   if (sin) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, sin, sizeof(*sin));
   if (cos) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, cos, sizeof(*cos));
@@ -3125,6 +3594,9 @@ INTERCEPTOR(void, sincos, double x, double *sin, double *cos) {
 INTERCEPTOR(void, sincosf, float x, float *sin, float *cos) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sincosf, x, sin, cos);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   REAL(sincosf)(x, sin, cos);
   if (sin) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, sin, sizeof(*sin));
   if (cos) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, cos, sizeof(*cos));
@@ -3132,6 +3604,9 @@ INTERCEPTOR(void, sincosf, float x, float *sin, float *cos) {
 INTERCEPTOR(void, sincosl, long double x, long double *sin, long double *cos) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, sincosl, x, sin, cos);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   REAL(sincosl)(x, sin, cos);
   if (sin) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, sin, sizeof(*sin));
   if (cos) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, cos, sizeof(*cos));
@@ -3148,6 +3623,9 @@ INTERCEPTOR(void, sincosl, long double x, long double *sin, long double *cos) {
 INTERCEPTOR(double, remquo, double x, double y, int *quo) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, remquo, x, y, quo);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   double res = REAL(remquo)(x, y, quo);
   if (quo) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, quo, sizeof(*quo));
   return res;
@@ -3155,6 +3633,9 @@ INTERCEPTOR(double, remquo, double x, double y, int *quo) {
 INTERCEPTOR(float, remquof, float x, float y, int *quo) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, remquof, x, y, quo);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   float res = REAL(remquof)(x, y, quo);
   if (quo) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, quo, sizeof(*quo));
   return res;
@@ -3162,6 +3643,9 @@ INTERCEPTOR(float, remquof, float x, float y, int *quo) {
 INTERCEPTOR(long double, remquol, long double x, long double y, int *quo) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, remquol, x, y, quo);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   long double res = REAL(remquol)(x, y, quo);
   if (quo) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, quo, sizeof(*quo));
   return res;
@@ -3209,6 +3693,9 @@ INTERCEPTOR(long double, lgammal, long double x) {
 INTERCEPTOR(double, lgamma_r, double x, int *signp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, lgamma_r, x, signp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   double res = REAL(lgamma_r)(x, signp);
   if (signp) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, signp, sizeof(*signp));
   return res;
@@ -3216,29 +3703,43 @@ INTERCEPTOR(double, lgamma_r, double x, int *signp) {
 INTERCEPTOR(float, lgammaf_r, float x, int *signp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, lgammaf_r, x, signp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   float res = REAL(lgammaf_r)(x, signp);
   if (signp) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, signp, sizeof(*signp));
   return res;
 }
+#define INIT_LGAMMA_R                   \
+  COMMON_INTERCEPT_FUNCTION(lgamma_r);  \
+  COMMON_INTERCEPT_FUNCTION(lgammaf_r);
+#else
+#define INIT_LGAMMA_R
+#endif
+
+#if SANITIZER_INTERCEPT_LGAMMAL_R
 INTERCEPTOR(long double, lgammal_r, long double x, int *signp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, lgammal_r, x, signp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   long double res = REAL(lgammal_r)(x, signp);
   if (signp) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, signp, sizeof(*signp));
   return res;
 }
-#define INIT_LGAMMA_R                   \
-  COMMON_INTERCEPT_FUNCTION(lgamma_r);  \
-  COMMON_INTERCEPT_FUNCTION(lgammaf_r); \
-  COMMON_INTERCEPT_FUNCTION(lgammal_r);
+#define INIT_LGAMMAL_R COMMON_INTERCEPT_FUNCTION(lgammal_r);
 #else
-#define INIT_LGAMMA_R
+#define INIT_LGAMMAL_R
 #endif
 
 #if SANITIZER_INTERCEPT_DRAND48_R
 INTERCEPTOR(int, drand48_r, void *buffer, double *result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, drand48_r, buffer, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(drand48_r)(buffer, result);
   if (result) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
   return res;
@@ -3246,6 +3747,9 @@ INTERCEPTOR(int, drand48_r, void *buffer, double *result) {
 INTERCEPTOR(int, lrand48_r, void *buffer, long *result) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, lrand48_r, buffer, result);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(lrand48_r)(buffer, result);
   if (result) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, result, sizeof(*result));
   return res;
@@ -3273,6 +3777,9 @@ INTERCEPTOR(int, rand_r, unsigned *seedp) {
 INTERCEPTOR(SSIZE_T, getline, char **lineptr, SIZE_T *n, void *stream) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getline, lineptr, n, stream);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(getline)(lineptr, n, stream);
   if (res > 0) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, lineptr, sizeof(*lineptr));
@@ -3281,11 +3788,14 @@ INTERCEPTOR(SSIZE_T, getline, char **lineptr, SIZE_T *n, void *stream) {
   }
   return res;
 }
-INTERCEPTOR(SSIZE_T, getdelim, char **lineptr, SIZE_T *n, int delim,
+INTERCEPTOR(SSIZE_T, __getdelim, char **lineptr, SIZE_T *n, int delim,
             void *stream) {
   void *ctx;
-  COMMON_INTERCEPTOR_ENTER(ctx, getdelim, lineptr, n, delim, stream);
-  SSIZE_T res = REAL(getdelim)(lineptr, n, delim, stream);
+  COMMON_INTERCEPTOR_ENTER(ctx, __getdelim, lineptr, n, delim, stream);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
+  SSIZE_T res = REAL(__getdelim)(lineptr, n, delim, stream);
   if (res > 0) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, lineptr, sizeof(*lineptr));
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, n, sizeof(*n));
@@ -3293,8 +3803,13 @@ INTERCEPTOR(SSIZE_T, getdelim, char **lineptr, SIZE_T *n, int delim,
   }
   return res;
 }
-#define INIT_GETLINE                  \
-  COMMON_INTERCEPT_FUNCTION(getline); \
+INTERCEPTOR(SSIZE_T, getdelim, char **lineptr, SIZE_T *n, int delim,
+            void *stream) {
+  return __getdelim(lineptr, n, delim, stream);
+}
+#define INIT_GETLINE                     \
+  COMMON_INTERCEPT_FUNCTION(getline);    \
+  COMMON_INTERCEPT_FUNCTION(__getdelim); \
   COMMON_INTERCEPT_FUNCTION(getdelim);
 #else
 #define INIT_GETLINE
@@ -3313,6 +3828,9 @@ INTERCEPTOR(SIZE_T, iconv, void *cd, char **inbuf, SIZE_T *inbytesleft,
   if (outbytesleft)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, outbytesleft, sizeof(*outbytesleft));
   void *outbuf_orig = outbuf ? *outbuf : 0;
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SIZE_T res = REAL(iconv)(cd, inbuf, inbytesleft, outbuf, outbytesleft);
   if (res != (SIZE_T) - 1 && outbuf && *outbuf > outbuf_orig) {
     SIZE_T sz = (char *)*outbuf - (char *)outbuf_orig;
@@ -3329,6 +3847,9 @@ INTERCEPTOR(SIZE_T, iconv, void *cd, char **inbuf, SIZE_T *inbytesleft,
 INTERCEPTOR(__sanitizer_clock_t, times, void *tms) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, times, tms);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_clock_t res = REAL(times)(tms);
   if (res != (__sanitizer_clock_t)-1 && tms)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, tms, struct_tms_sz);
@@ -3345,7 +3866,11 @@ INTERCEPTOR(void *, __tls_get_addr, void *arg) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, __tls_get_addr, arg);
   void *res = REAL(__tls_get_addr)(arg);
-  DTLS_on_tls_get_addr(arg, res);
+  DTLS::DTV *dtv = DTLS_on_tls_get_addr(arg, res);
+  if (dtv) {
+    // New DTLS block has been allocated.
+    COMMON_INTERCEPTOR_INITIALIZE_RANGE((void *)dtv->beg, dtv->size);
+  }
   return res;
 }
 #else
@@ -3357,6 +3882,9 @@ INTERCEPTOR(SSIZE_T, listxattr, const char *path, char *list, SIZE_T size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, listxattr, path, list, size);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(listxattr)(path, list, size);
   // Here and below, size == 0 is a special case where nothing is written to the
   // buffer, and res contains the desired buffer size.
@@ -3367,6 +3895,9 @@ INTERCEPTOR(SSIZE_T, llistxattr, const char *path, char *list, SIZE_T size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, llistxattr, path, list, size);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(llistxattr)(path, list, size);
   if (size && res > 0 && list) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, list, res);
   return res;
@@ -3374,6 +3905,9 @@ INTERCEPTOR(SSIZE_T, llistxattr, const char *path, char *list, SIZE_T size) {
 INTERCEPTOR(SSIZE_T, flistxattr, int fd, char *list, SIZE_T size) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, flistxattr, fd, list, size);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(flistxattr)(fd, list, size);
   if (size && res > 0 && list) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, list, res);
   return res;
@@ -3393,6 +3927,9 @@ INTERCEPTOR(SSIZE_T, getxattr, const char *path, const char *name, char *value,
   COMMON_INTERCEPTOR_ENTER(ctx, getxattr, path, name, value, size);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
   if (name) COMMON_INTERCEPTOR_READ_RANGE(ctx, name, REAL(strlen)(name) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(getxattr)(path, name, value, size);
   if (size && res > 0 && value) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, value, res);
   return res;
@@ -3403,6 +3940,9 @@ INTERCEPTOR(SSIZE_T, lgetxattr, const char *path, const char *name, char *value,
   COMMON_INTERCEPTOR_ENTER(ctx, lgetxattr, path, name, value, size);
   if (path) COMMON_INTERCEPTOR_READ_RANGE(ctx, path, REAL(strlen)(path) + 1);
   if (name) COMMON_INTERCEPTOR_READ_RANGE(ctx, name, REAL(strlen)(name) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(lgetxattr)(path, name, value, size);
   if (size && res > 0 && value) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, value, res);
   return res;
@@ -3412,6 +3952,9 @@ INTERCEPTOR(SSIZE_T, fgetxattr, int fd, const char *name, char *value,
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fgetxattr, fd, name, value, size);
   if (name) COMMON_INTERCEPTOR_READ_RANGE(ctx, name, REAL(strlen)(name) + 1);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   SSIZE_T res = REAL(fgetxattr)(fd, name, value, size);
   if (size && res > 0 && value) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, value, res);
   return res;
@@ -3428,6 +3971,9 @@ INTERCEPTOR(SSIZE_T, fgetxattr, int fd, const char *name, char *value,
 INTERCEPTOR(int, getresuid, void *ruid, void *euid, void *suid) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getresuid, ruid, euid, suid);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getresuid)(ruid, euid, suid);
   if (res >= 0) {
     if (ruid) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ruid, uid_t_sz);
@@ -3439,6 +3985,9 @@ INTERCEPTOR(int, getresuid, void *ruid, void *euid, void *suid) {
 INTERCEPTOR(int, getresgid, void *rgid, void *egid, void *sgid) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getresgid, rgid, egid, sgid);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getresgid)(rgid, egid, sgid);
   if (res >= 0) {
     if (rgid) COMMON_INTERCEPTOR_WRITE_RANGE(ctx, rgid, gid_t_sz);
@@ -3461,6 +4010,9 @@ INTERCEPTOR(int, getresgid, void *rgid, void *egid, void *sgid) {
 INTERCEPTOR(int, getifaddrs, __sanitizer_ifaddrs **ifap) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, getifaddrs, ifap);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(getifaddrs)(ifap);
   if (res == 0 && ifap) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ifap, sizeof(void *));
@@ -3494,6 +4046,9 @@ INTERCEPTOR(int, getifaddrs, __sanitizer_ifaddrs **ifap) {
 INTERCEPTOR(char *, if_indextoname, unsigned int ifindex, char* ifname) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, if_indextoname, ifindex, ifname);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   char *res = REAL(if_indextoname)(ifindex, ifname);
   if (res && ifname)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ifname, REAL(strlen)(ifname) + 1);
@@ -3519,6 +4074,9 @@ INTERCEPTOR(int, capget, void *hdrp, void *datap) {
   COMMON_INTERCEPTOR_ENTER(ctx, capget, hdrp, datap);
   if (hdrp)
     COMMON_INTERCEPTOR_READ_RANGE(ctx, hdrp, __user_cap_header_struct_sz);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(capget)(hdrp, datap);
   if (res == 0 && datap)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, datap, __user_cap_data_struct_sz);
@@ -3617,6 +4175,9 @@ INTERCEPTOR(void *, __bzero, void *block, uptr size) {
 INTERCEPTOR(int, ftime, __sanitizer_timeb *tp) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, ftime, tp);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(ftime)(tp);
   if (tp)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, tp, sizeof(*tp));
@@ -3632,6 +4193,9 @@ INTERCEPTOR(void, xdrmem_create, __sanitizer_XDR *xdrs, uptr addr,
             unsigned size, int op) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, xdrmem_create, xdrs, addr, size, op);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   REAL(xdrmem_create)(xdrs, addr, size, op);
   COMMON_INTERCEPTOR_WRITE_RANGE(ctx, xdrs, sizeof(*xdrs));
   if (op == __sanitizer_XDR_ENCODE) {
@@ -3644,10 +4208,16 @@ INTERCEPTOR(void, xdrmem_create, __sanitizer_XDR *xdrs, uptr addr,
 INTERCEPTOR(void, xdrstdio_create, __sanitizer_XDR *xdrs, void *file, int op) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, xdrstdio_create, xdrs, file, op);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   REAL(xdrstdio_create)(xdrs, file, op);
   COMMON_INTERCEPTOR_WRITE_RANGE(ctx, xdrs, sizeof(*xdrs));
 }
 
+// FIXME: under ASan the call below may write to freed memory and corrupt
+// its metadata. See
+// https://code.google.com/p/address-sanitizer/issues/detail?id=321.
 #define XDR_INTERCEPTOR(F, T)                             \
   INTERCEPTOR(int, F, __sanitizer_XDR *xdrs, T *p) {      \
     void *ctx;                                            \
@@ -3699,6 +4269,9 @@ INTERCEPTOR(int, xdr_bytes, __sanitizer_XDR *xdrs, char **p, unsigned *sizep,
     COMMON_INTERCEPTOR_READ_RANGE(ctx, sizep, sizeof(*sizep));
     COMMON_INTERCEPTOR_READ_RANGE(ctx, *p, *sizep);
   }
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(xdr_bytes)(xdrs, p, sizep, maxsize);
   if (p && sizep && xdrs->x_op == __sanitizer_XDR_DECODE) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, p, sizeof(*p));
@@ -3716,6 +4289,9 @@ INTERCEPTOR(int, xdr_string, __sanitizer_XDR *xdrs, char **p,
     COMMON_INTERCEPTOR_READ_RANGE(ctx, p, sizeof(*p));
     COMMON_INTERCEPTOR_READ_RANGE(ctx, *p, REAL(strlen)(*p) + 1);
   }
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   int res = REAL(xdr_string)(xdrs, p, maxsize);
   if (p && xdrs->x_op == __sanitizer_XDR_DECODE) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, p, sizeof(*p));
@@ -3765,6 +4341,9 @@ INTERCEPTOR(void *, tsearch, void *key, void **rootp,
             int (*compar)(const void *, const void *)) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, tsearch, key, rootp, compar);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   void *res = REAL(tsearch)(key, rootp, compar);
   if (res && *(void **)res == key)
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, res, sizeof(void *));
@@ -3915,6 +4494,9 @@ INTERCEPTOR(__sanitizer_FILE *, freopen64, const char *path, const char *mode,
 INTERCEPTOR(__sanitizer_FILE *, open_memstream, char **ptr, SIZE_T *sizeloc) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, open_memstream, ptr, sizeloc);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_FILE *res = REAL(open_memstream)(ptr, sizeloc);
   if (res) {
     COMMON_INTERCEPTOR_WRITE_RANGE(ctx, ptr, sizeof(*ptr));
@@ -3943,6 +4525,9 @@ INTERCEPTOR(__sanitizer_FILE *, fmemopen, void *buf, SIZE_T size,
             const char *mode) {
   void *ctx;
   COMMON_INTERCEPTOR_ENTER(ctx, fmemopen, buf, size, mode);
+  // FIXME: under ASan the call below may write to freed memory and corrupt
+  // its metadata. See
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=321.
   __sanitizer_FILE *res = REAL(fmemopen)(buf, size, mode);
   if (res) unpoison_file(res);
   return res;
@@ -4034,6 +4619,114 @@ INTERCEPTOR(int, fclose, __sanitizer_FILE *fp) {
 #define INIT_FCLOSE
 #endif
 
+#if SANITIZER_INTERCEPT_DLOPEN_DLCLOSE
+INTERCEPTOR(void*, dlopen, const char *filename, int flag) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER_NOIGNORE(ctx, dlopen, filename, flag);
+  void *res = REAL(dlopen)(filename, flag);
+  COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, res);
+  return res;
+}
+
+INTERCEPTOR(int, dlclose, void *handle) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER_NOIGNORE(ctx, dlclose, handle);
+  int res = REAL(dlclose)(handle);
+  COMMON_INTERCEPTOR_LIBRARY_UNLOADED();
+  return res;
+}
+#define INIT_DLOPEN_DLCLOSE          \
+  COMMON_INTERCEPT_FUNCTION(dlopen); \
+  COMMON_INTERCEPT_FUNCTION(dlclose);
+#else
+#define INIT_DLOPEN_DLCLOSE
+#endif
+
+#if SANITIZER_INTERCEPT_GETPASS
+INTERCEPTOR(char *, getpass, const char *prompt) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER(ctx, getpass, prompt);
+  if (prompt)
+    COMMON_INTERCEPTOR_READ_RANGE(ctx, prompt, REAL(strlen)(prompt)+1);
+  char *res = REAL(getpass)(prompt);
+  if (res) COMMON_INTERCEPTOR_INITIALIZE_RANGE(res, REAL(strlen)(res)+1);
+  return res;
+}
+
+#define INIT_GETPASS COMMON_INTERCEPT_FUNCTION(getpass);
+#else
+#define INIT_GETPASS
+#endif
+
+#if SANITIZER_INTERCEPT_TIMERFD
+INTERCEPTOR(int, timerfd_settime, int fd, int flags, void *new_value,
+            void *old_value) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER(ctx, timerfd_settime, fd, flags, new_value,
+                           old_value);
+  COMMON_INTERCEPTOR_READ_RANGE(ctx, new_value, struct_itimerspec_sz);
+  int res = REAL(timerfd_settime)(fd, flags, new_value, old_value);
+  if (res != -1 && old_value)
+    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, old_value, struct_itimerspec_sz);
+  return res;
+}
+
+INTERCEPTOR(int, timerfd_gettime, int fd, void *curr_value) {
+  void *ctx;
+  COMMON_INTERCEPTOR_ENTER(ctx, timerfd_gettime, fd, curr_value);
+  int res = REAL(timerfd_gettime)(fd, curr_value);
+  if (res != -1 && curr_value)
+    COMMON_INTERCEPTOR_WRITE_RANGE(ctx, curr_value, struct_itimerspec_sz);
+  return res;
+}
+#define INIT_TIMERFD                          \
+  COMMON_INTERCEPT_FUNCTION(timerfd_settime); \
+  COMMON_INTERCEPT_FUNCTION(timerfd_gettime);
+#else
+#define INIT_TIMERFD
+#endif
+
+#if SANITIZER_INTERCEPT_MLOCKX
+// Linux kernel has a bug that leads to kernel deadlock if a process
+// maps TBs of memory and then calls mlock().
+static void MlockIsUnsupported() {
+  static atomic_uint8_t printed;
+  if (atomic_exchange(&printed, 1, memory_order_relaxed))
+    return;
+  VPrintf(1, "INFO: %s ignores mlock/mlockall/munlock/munlockall\n",
+          SanitizerToolName);
+}
+
+INTERCEPTOR(int, mlock, const void *addr, uptr len) {
+  MlockIsUnsupported();
+  return 0;
+}
+
+INTERCEPTOR(int, munlock, const void *addr, uptr len) {
+  MlockIsUnsupported();
+  return 0;
+}
+
+INTERCEPTOR(int, mlockall, int flags) {
+  MlockIsUnsupported();
+  return 0;
+}
+
+INTERCEPTOR(int, munlockall, void) {
+  MlockIsUnsupported();
+  return 0;
+}
+
+#define INIT_MLOCKX                                                            \
+  COMMON_INTERCEPT_FUNCTION(mlock);                                            \
+  COMMON_INTERCEPT_FUNCTION(munlock);                                          \
+  COMMON_INTERCEPT_FUNCTION(mlockall);                                         \
+  COMMON_INTERCEPT_FUNCTION(munlockall);
+
+#else
+#define INIT_MLOCKX
+#endif  // SANITIZER_INTERCEPT_MLOCKX
+
 static void InitializeCommonInterceptors() {
   static u64 metadata_mem[sizeof(MetadataHashMap) / sizeof(u64) + 1];
   interceptor_metadata_map = new((void *)&metadata_mem) MetadataHashMap();
@@ -4085,6 +4778,9 @@ static void InitializeCommonInterceptors() {
   INIT_GETSOCKNAME;
   INIT_GETHOSTBYNAME;
   INIT_GETHOSTBYNAME_R;
+  INIT_GETHOSTBYNAME2_R;
+  INIT_GETHOSTBYADDR_R;
+  INIT_GETHOSTENT_R;
   INIT_GETSOCKOPT;
   INIT_ACCEPT;
   INIT_ACCEPT4;
@@ -4136,13 +4832,25 @@ static void InitializeCommonInterceptors() {
   INIT_STATVFS;
   INIT_STATVFS64;
   INIT_INITGROUPS;
-  INIT_ETHER;
+  INIT_ETHER_NTOA_ATON;
+  INIT_ETHER_HOST;
   INIT_ETHER_R;
   INIT_SHMCTL;
   INIT_RANDOM_R;
   INIT_PTHREAD_ATTR_GET;
   INIT_PTHREAD_ATTR_GETINHERITSCHED;
   INIT_PTHREAD_ATTR_GETAFFINITY_NP;
+  INIT_PTHREAD_MUTEXATTR_GETPSHARED;
+  INIT_PTHREAD_MUTEXATTR_GETTYPE;
+  INIT_PTHREAD_MUTEXATTR_GETPROTOCOL;
+  INIT_PTHREAD_MUTEXATTR_GETPRIOCEILING;
+  INIT_PTHREAD_MUTEXATTR_GETROBUST;
+  INIT_PTHREAD_MUTEXATTR_GETROBUST_NP;
+  INIT_PTHREAD_RWLOCKATTR_GETPSHARED;
+  INIT_PTHREAD_RWLOCKATTR_GETKIND_NP;
+  INIT_PTHREAD_CONDATTR_GETPSHARED;
+  INIT_PTHREAD_CONDATTR_GETCLOCK;
+  INIT_PTHREAD_BARRIERATTR_GETPSHARED;
   INIT_TMPNAM;
   INIT_TMPNAM_R;
   INIT_TEMPNAM;
@@ -4151,6 +4859,7 @@ static void InitializeCommonInterceptors() {
   INIT_REMQUO;
   INIT_LGAMMA;
   INIT_LGAMMA_R;
+  INIT_LGAMMAL_R;
   INIT_DRAND48_R;
   INIT_RAND_R;
   INIT_GETLINE;
@@ -4175,4 +4884,8 @@ static void InitializeCommonInterceptors() {
   INIT_OBSTACK;
   INIT_FFLUSH;
   INIT_FCLOSE;
+  INIT_DLOPEN_DLCLOSE;
+  INIT_GETPASS;
+  INIT_TIMERFD;
+  INIT_MLOCKX;
 }
diff --git a/libsanitizer/sanitizer_common/sanitizer_common_interceptors_format.inc b/libsanitizer/sanitizer_common/sanitizer_common_interceptors_format.inc
index dfc4ac6b556..26cbe68dc99 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common_interceptors_format.inc
+++ b/libsanitizer/sanitizer_common/sanitizer_common_interceptors_format.inc
@@ -457,6 +457,9 @@ static int printf_get_value_size(PrintfDirective *dir) {
       case 8:                                                      \
         va_arg(*aq, double);                                       \
         break;                                                     \
+      case 12:                                                     \
+        va_arg(*aq, long double);                                  \
+        break;                                                     \
       case 16:                                                     \
         va_arg(*aq, long double);                                  \
         break;                                                     \
diff --git a/libsanitizer/sanitizer_common/sanitizer_common_interceptors_ioctl.inc b/libsanitizer/sanitizer_common/sanitizer_common_interceptors_ioctl.inc
index 96d171a92f5..ca264d899a7 100755
--- a/libsanitizer/sanitizer_common/sanitizer_common_interceptors_ioctl.inc
+++ b/libsanitizer/sanitizer_common/sanitizer_common_interceptors_ioctl.inc
@@ -527,7 +527,7 @@ static bool ioctl_decode(unsigned req, ioctl_desc *desc) {
   desc->name = "<DECODED_IOCTL>";
   desc->size = IOC_SIZE(req);
   // Sanity check.
-  if (desc->size > 1024) return false;
+  if (desc->size > 0xFFFF) return false;
   unsigned dir = IOC_DIR(req);
   switch (dir) {
     case IOC_NONE:
@@ -545,10 +545,10 @@ static bool ioctl_decode(unsigned req, ioctl_desc *desc) {
     default:
       return false;
   }
-  if (desc->type != IOC_NONE && desc->size == 0) return false;
-  char id = IOC_TYPE(req);
+  // Size can be 0 iff type is NONE.
+  if ((desc->type == IOC_NONE) != (desc->size == 0)) return false;
   // Sanity check.
-  if (!(id >= 'a' && id <= 'z') && !(id >= 'A' && id <= 'Z')) return false;
+  if (IOC_TYPE(req) == 0) return false;
   return true;
 }
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_common_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_common_libcdep.cc
index d8330630b5f..b5251444f0f 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_common_libcdep.cc
@@ -11,6 +11,8 @@
 
 #include "sanitizer_common.h"
 #include "sanitizer_flags.h"
+#include "sanitizer_stacktrace.h"
+#include "sanitizer_symbolizer.h"
 
 namespace __sanitizer {
 
@@ -39,4 +41,32 @@ bool ColorizeReports() {
   return internal_strcmp(flag, "always") == 0 ||
          (internal_strcmp(flag, "auto") == 0 && PrintsToTtyCached());
 }
+
+static void (*sandboxing_callback)();
+void SetSandboxingCallback(void (*f)()) {
+  sandboxing_callback = f;
+}
+
+void ReportErrorSummary(const char *error_type, StackTrace *stack) {
+  if (!common_flags()->print_summary)
+    return;
+  AddressInfo ai;
+#if !SANITIZER_GO
+  if (stack->size > 0 && Symbolizer::GetOrInit()->CanReturnFileLineInfo()) {
+    // Currently, we include the first stack frame into the report summary.
+    // Maybe sometimes we need to choose another frame (e.g. skip memcpy/etc).
+    uptr pc = StackTrace::GetPreviousInstructionPc(stack->trace[0]);
+    Symbolizer::GetOrInit()->SymbolizePC(pc, &ai, 1);
+  }
+#endif
+  ReportErrorSummary(error_type, ai.file, ai.line, ai.function);
+}
+
 }  // namespace __sanitizer
+
+void NOINLINE
+__sanitizer_sandbox_on_notify(__sanitizer_sandbox_arguments *args) {
+  PrepareForSandboxing(args);
+  if (sandboxing_callback)
+    sandboxing_callback();
+}
diff --git a/libsanitizer/sanitizer_common/sanitizer_common_syscalls.inc b/libsanitizer/sanitizer_common/sanitizer_common_syscalls.inc
index 6be2f51b80a..24e1b7ec452 100644
--- a/libsanitizer/sanitizer_common/sanitizer_common_syscalls.inc
+++ b/libsanitizer/sanitizer_common/sanitizer_common_syscalls.inc
@@ -829,6 +829,7 @@ POST_SYSCALL(stat)(long res, const void *filename, void *statbuf) {
   }
 }
 
+#if !SANITIZER_ANDROID
 PRE_SYSCALL(statfs)(const void *path, void *buf) {
   if (path)
     PRE_READ(path, __sanitizer::internal_strlen((const char *)path) + 1);
@@ -866,6 +867,7 @@ POST_SYSCALL(fstatfs64)(long res, long fd, long sz, void *buf) {
     if (buf) POST_WRITE(buf, struct_statfs64_sz);
   }
 }
+#endif // !SANITIZER_ANDROID
 
 PRE_SYSCALL(lstat)(const void *filename, void *statbuf) {
   if (filename)
@@ -1322,13 +1324,13 @@ PRE_SYSCALL(io_submit)(long ctx_id, long nr, __sanitizer_iocb **iocbpp) {
     } else if (op == iocb_cmd_pread && buf && len) {
       POST_WRITE(buf, len);
     } else if (op == iocb_cmd_pwritev) {
-      __sanitizer_iovec *iovec = (__sanitizer_iovec*)iocbpp[i]->aio_buf;
+      __sanitizer_iovec *iovec = (__sanitizer_iovec*)buf;
       for (uptr v = 0; v < len; v++)
-        PRE_READ(iovec[i].iov_base, iovec[i].iov_len);
+        PRE_READ(iovec[v].iov_base, iovec[v].iov_len);
     } else if (op == iocb_cmd_preadv) {
-      __sanitizer_iovec *iovec = (__sanitizer_iovec*)iocbpp[i]->aio_buf;
+      __sanitizer_iovec *iovec = (__sanitizer_iovec*)buf;
       for (uptr v = 0; v < len; v++)
-        POST_WRITE(iovec[i].iov_base, iovec[i].iov_len);
+        POST_WRITE(iovec[v].iov_base, iovec[v].iov_len);
     }
     // See comment in io_getevents.
     COMMON_SYSCALL_RELEASE(data);
@@ -2293,7 +2295,7 @@ PRE_SYSCALL(ni_syscall)() {}
 POST_SYSCALL(ni_syscall)(long res) {}
 
 PRE_SYSCALL(ptrace)(long request, long pid, long addr, long data) {
-#if defined(__i386) || defined (__x86_64)
+#if !SANITIZER_ANDROID && (defined(__i386) || defined (__x86_64))
   if (data) {
     if (request == ptrace_setregs) {
       PRE_READ((void *)data, struct_user_regs_struct_sz);
@@ -2312,7 +2314,7 @@ PRE_SYSCALL(ptrace)(long request, long pid, long addr, long data) {
 }
 
 POST_SYSCALL(ptrace)(long res, long request, long pid, long addr, long data) {
-#if defined(__i386) || defined (__x86_64)
+#if !SANITIZER_ANDROID && (defined(__i386) || defined (__x86_64))
   if (res >= 0 && data) {
     // Note that this is different from the interceptor in
     // sanitizer_common_interceptors.inc.
diff --git a/libsanitizer/sanitizer_common/sanitizer_coverage.cc b/libsanitizer/sanitizer_common/sanitizer_coverage.cc
deleted file mode 100644
index b88e9e7ed0a..00000000000
--- a/libsanitizer/sanitizer_common/sanitizer_coverage.cc
+++ /dev/null
@@ -1,214 +0,0 @@
-//===-- sanitizer_coverage.cc ---------------------------------------------===//
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// Sanitizer Coverage.
-// This file implements run-time support for a poor man's coverage tool.
-//
-// Compiler instrumentation:
-// For every interesting basic block the compiler injects the following code:
-// if (*Guard) {
-//    __sanitizer_cov();
-//    *Guard = 1;
-// }
-// It's fine to call __sanitizer_cov more than once for a given block.
-//
-// Run-time:
-//  - __sanitizer_cov(): record that we've executed the PC (GET_CALLER_PC).
-//  - __sanitizer_cov_dump: dump the coverage data to disk.
-//  For every module of the current process that has coverage data
-//  this will create a file module_name.PID.sancov. The file format is simple:
-//  it's just a sorted sequence of 4-byte offsets in the module.
-//
-// Eventually, this coverage implementation should be obsoleted by a more
-// powerful general purpose Clang/LLVM coverage instrumentation.
-// Consider this implementation as prototype.
-//
-// FIXME: support (or at least test with) dlclose.
-//===----------------------------------------------------------------------===//
-
-#include "sanitizer_allocator_internal.h"
-#include "sanitizer_common.h"
-#include "sanitizer_libc.h"
-#include "sanitizer_mutex.h"
-#include "sanitizer_procmaps.h"
-#include "sanitizer_stacktrace.h"
-#include "sanitizer_flags.h"
-
-atomic_uint32_t dump_once_guard;  // Ensure that CovDump runs only once.
-
-// pc_array is the array containing the covered PCs.
-// To make the pc_array thread- and async-signal-safe it has to be large enough.
-// 128M counters "ought to be enough for anybody" (4M on 32-bit).
-// pc_array is allocated with MmapNoReserveOrDie and so it uses only as
-// much RAM as it really needs.
-static const uptr kPcArraySize = FIRST_32_SECOND_64(1 << 22, 1 << 27);
-static uptr *pc_array;
-static atomic_uintptr_t pc_array_index;
-
-static bool cov_sandboxed = false;
-static int cov_fd = kInvalidFd;
-static unsigned int cov_max_block_size = 0;
-
-namespace __sanitizer {
-
-// Simply add the pc into the vector under lock. If the function is called more
-// than once for a given PC it will be inserted multiple times, which is fine.
-static void CovAdd(uptr pc) {
-  if (!pc_array) return;
-  uptr idx = atomic_fetch_add(&pc_array_index, 1, memory_order_relaxed);
-  CHECK_LT(idx, kPcArraySize);
-  pc_array[idx] = pc;
-}
-
-void CovInit() {
-  pc_array = reinterpret_cast<uptr *>(
-      MmapNoReserveOrDie(sizeof(uptr) * kPcArraySize, "CovInit"));
-}
-
-static inline bool CompareLess(const uptr &a, const uptr &b) {
-  return a < b;
-}
-
-// Block layout for packed file format: header, followed by module name (no
-// trailing zero), followed by data blob.
-struct CovHeader {
-  int pid;
-  unsigned int module_name_length;
-  unsigned int data_length;
-};
-
-static void CovWritePacked(int pid, const char *module, const void *blob,
-                           unsigned int blob_size) {
-  CHECK_GE(cov_fd, 0);
-  unsigned module_name_length = internal_strlen(module);
-  CovHeader header = {pid, module_name_length, blob_size};
-
-  if (cov_max_block_size == 0) {
-    // Writing to a file. Just go ahead.
-    internal_write(cov_fd, &header, sizeof(header));
-    internal_write(cov_fd, module, module_name_length);
-    internal_write(cov_fd, blob, blob_size);
-  } else {
-    // Writing to a socket. We want to split the data into appropriately sized
-    // blocks.
-    InternalScopedBuffer<char> block(cov_max_block_size);
-    CHECK_EQ((uptr)block.data(), (uptr)(CovHeader *)block.data());
-    uptr header_size_with_module = sizeof(header) + module_name_length;
-    CHECK_LT(header_size_with_module, cov_max_block_size);
-    unsigned int max_payload_size =
-        cov_max_block_size - header_size_with_module;
-    char *block_pos = block.data();
-    internal_memcpy(block_pos, &header, sizeof(header));
-    block_pos += sizeof(header);
-    internal_memcpy(block_pos, module, module_name_length);
-    block_pos += module_name_length;
-    char *block_data_begin = block_pos;
-    char *blob_pos = (char *)blob;
-    while (blob_size > 0) {
-      unsigned int payload_size = Min(blob_size, max_payload_size);
-      blob_size -= payload_size;
-      internal_memcpy(block_data_begin, blob_pos, payload_size);
-      blob_pos += payload_size;
-      ((CovHeader *)block.data())->data_length = payload_size;
-      internal_write(cov_fd, block.data(),
-                     header_size_with_module + payload_size);
-    }
-  }
-}
-
-// Dump the coverage on disk.
-static void CovDump() {
-  if (!common_flags()->coverage) return;
-#if !SANITIZER_WINDOWS
-  if (atomic_fetch_add(&dump_once_guard, 1, memory_order_relaxed))
-    return;
-  uptr size = atomic_load(&pc_array_index, memory_order_relaxed);
-  InternalSort(&pc_array, size, CompareLess);
-  InternalMmapVector<u32> offsets(size);
-  const uptr *vb = pc_array;
-  const uptr *ve = vb + size;
-  MemoryMappingLayout proc_maps(/*cache_enabled*/true);
-  uptr mb, me, off, prot;
-  InternalScopedBuffer<char> module(4096);
-  InternalScopedBuffer<char> path(4096 * 2);
-  for (int i = 0;
-       proc_maps.Next(&mb, &me, &off, module.data(), module.size(), &prot);
-       i++) {
-    if ((prot & MemoryMappingLayout::kProtectionExecute) == 0)
-      continue;
-    while (vb < ve && *vb < mb) vb++;
-    if (vb >= ve) break;
-    if (*vb < me) {
-      offsets.clear();
-      const uptr *old_vb = vb;
-      CHECK_LE(off, *vb);
-      for (; vb < ve && *vb < me; vb++) {
-        uptr diff = *vb - (i ? mb : 0) + off;
-        CHECK_LE(diff, 0xffffffffU);
-        offsets.push_back(static_cast<u32>(diff));
-      }
-      char *module_name = StripModuleName(module.data());
-      if (cov_sandboxed) {
-        CovWritePacked(internal_getpid(), module_name, offsets.data(),
-                       offsets.size() * sizeof(u32));
-        VReport(1, " CovDump: %zd PCs written to packed file\n", vb - old_vb);
-      } else {
-        // One file per module per process.
-        internal_snprintf((char *)path.data(), path.size(), "%s.%zd.sancov",
-                          module_name, internal_getpid());
-        uptr fd = OpenFile(path.data(), true);
-        if (internal_iserror(fd)) {
-          Report(" CovDump: failed to open %s for writing\n", path.data());
-        } else {
-          internal_write(fd, offsets.data(), offsets.size() * sizeof(u32));
-          internal_close(fd);
-          VReport(1, " CovDump: %s: %zd PCs written\n", path.data(),
-                  vb - old_vb);
-        }
-      }
-      InternalFree(module_name);
-    }
-  }
-  if (cov_fd >= 0)
-    internal_close(cov_fd);
-#endif  // !SANITIZER_WINDOWS
-}
-
-static void OpenPackedFileForWriting() {
-  CHECK(cov_fd == kInvalidFd);
-  InternalScopedBuffer<char> path(1024);
-  internal_snprintf((char *)path.data(), path.size(), "%zd.sancov.packed",
-                    internal_getpid());
-  uptr fd = OpenFile(path.data(), true);
-  if (internal_iserror(fd)) {
-    Report(" Coverage: failed to open %s for writing\n", path.data());
-    Die();
-  }
-  cov_fd = fd;
-}
-
-void CovPrepareForSandboxing(__sanitizer_sandbox_arguments *args) {
-  if (!args) return;
-  if (!common_flags()->coverage) return;
-  cov_sandboxed = args->coverage_sandboxed;
-  if (!cov_sandboxed) return;
-  cov_fd = args->coverage_fd;
-  cov_max_block_size = args->coverage_max_block_size;
-  if (cov_fd < 0)
-    // Pre-open the file now. The sandbox won't allow us to do it later.
-    OpenPackedFileForWriting();
-}
-
-}  // namespace __sanitizer
-
-extern "C" {
-SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov() {
-  CovAdd(StackTrace::GetPreviousInstructionPc(GET_CALLER_PC()));
-}
-SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov_dump() { CovDump(); }
-SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov_init() { CovInit(); }
-}  // extern "C"
diff --git a/libsanitizer/sanitizer_common/sanitizer_coverage_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_coverage_libcdep.cc
new file mode 100644
index 00000000000..9b02cd75017
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_coverage_libcdep.cc
@@ -0,0 +1,377 @@
+//===-- sanitizer_coverage.cc ---------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Sanitizer Coverage.
+// This file implements run-time support for a poor man's coverage tool.
+//
+// Compiler instrumentation:
+// For every interesting basic block the compiler injects the following code:
+// if (*Guard) {
+//    __sanitizer_cov();
+//    *Guard = 1;
+// }
+// It's fine to call __sanitizer_cov more than once for a given block.
+//
+// Run-time:
+//  - __sanitizer_cov(): record that we've executed the PC (GET_CALLER_PC).
+//  - __sanitizer_cov_dump: dump the coverage data to disk.
+//  For every module of the current process that has coverage data
+//  this will create a file module_name.PID.sancov. The file format is simple:
+//  it's just a sorted sequence of 4-byte offsets in the module.
+//
+// Eventually, this coverage implementation should be obsoleted by a more
+// powerful general purpose Clang/LLVM coverage instrumentation.
+// Consider this implementation as prototype.
+//
+// FIXME: support (or at least test with) dlclose.
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_allocator_internal.h"
+#include "sanitizer_common.h"
+#include "sanitizer_libc.h"
+#include "sanitizer_mutex.h"
+#include "sanitizer_procmaps.h"
+#include "sanitizer_stacktrace.h"
+#include "sanitizer_flags.h"
+
+atomic_uint32_t dump_once_guard;  // Ensure that CovDump runs only once.
+
+// pc_array is the array containing the covered PCs.
+// To make the pc_array thread- and async-signal-safe it has to be large enough.
+// 128M counters "ought to be enough for anybody" (4M on 32-bit).
+
+// With coverage_direct=1 in ASAN_OPTIONS, pc_array memory is mapped to a file.
+// In this mode, __sanitizer_cov_dump does nothing, and CovUpdateMapping()
+// dump current memory layout to another file.
+
+static bool cov_sandboxed = false;
+static int cov_fd = kInvalidFd;
+static unsigned int cov_max_block_size = 0;
+
+namespace __sanitizer {
+
+class CoverageData {
+ public:
+  void Init();
+  void BeforeFork();
+  void AfterFork(int child_pid);
+  void Extend(uptr npcs);
+  void Add(uptr pc);
+
+  uptr *data();
+  uptr size();
+
+ private:
+  // Maximal size pc array may ever grow.
+  // We MmapNoReserve this space to ensure that the array is contiguous.
+  static const uptr kPcArrayMaxSize = FIRST_32_SECOND_64(1 << 22, 1 << 27);
+  // The amount file mapping for the pc array is grown by.
+  static const uptr kPcArrayMmapSize = 64 * 1024;
+
+  // pc_array is allocated with MmapNoReserveOrDie and so it uses only as
+  // much RAM as it really needs.
+  uptr *pc_array;
+  // Index of the first available pc_array slot.
+  atomic_uintptr_t pc_array_index;
+  // Array size.
+  atomic_uintptr_t pc_array_size;
+  // Current file mapped size of the pc array.
+  uptr pc_array_mapped_size;
+  // Descriptor of the file mapped pc array.
+  int pc_fd;
+  StaticSpinMutex mu;
+
+  void DirectOpen();
+  void ReInit();
+};
+
+static CoverageData coverage_data;
+
+void CoverageData::DirectOpen() {
+  InternalScopedString path(1024);
+  internal_snprintf((char *)path.data(), path.size(), "%s/%zd.sancov.raw",
+                    common_flags()->coverage_dir, internal_getpid());
+  pc_fd = OpenFile(path.data(), true);
+  if (internal_iserror(pc_fd)) {
+    Report(" Coverage: failed to open %s for writing\n", path.data());
+    Die();
+  }
+
+  pc_array_mapped_size = 0;
+  CovUpdateMapping();
+}
+
+void CoverageData::Init() {
+  pc_array = reinterpret_cast<uptr *>(
+      MmapNoReserveOrDie(sizeof(uptr) * kPcArrayMaxSize, "CovInit"));
+  pc_fd = kInvalidFd;
+  if (common_flags()->coverage_direct) {
+    atomic_store(&pc_array_size, 0, memory_order_relaxed);
+    atomic_store(&pc_array_index, 0, memory_order_relaxed);
+  } else {
+    atomic_store(&pc_array_size, kPcArrayMaxSize, memory_order_relaxed);
+    atomic_store(&pc_array_index, 0, memory_order_relaxed);
+  }
+}
+
+void CoverageData::ReInit() {
+  internal_munmap(pc_array, sizeof(uptr) * kPcArrayMaxSize);
+  if (pc_fd != kInvalidFd) internal_close(pc_fd);
+  if (common_flags()->coverage_direct) {
+    // In memory-mapped mode we must extend the new file to the known array
+    // size.
+    uptr size = atomic_load(&pc_array_size, memory_order_relaxed);
+    Init();
+    if (size) Extend(size);
+  } else {
+    Init();
+  }
+}
+
+void CoverageData::BeforeFork() {
+  mu.Lock();
+}
+
+void CoverageData::AfterFork(int child_pid) {
+  // We are single-threaded so it's OK to release the lock early.
+  mu.Unlock();
+  if (child_pid == 0) ReInit();
+}
+
+// Extend coverage PC array to fit additional npcs elements.
+void CoverageData::Extend(uptr npcs) {
+  if (!common_flags()->coverage_direct) return;
+  SpinMutexLock l(&mu);
+
+  if (pc_fd == kInvalidFd) DirectOpen();
+  CHECK_NE(pc_fd, kInvalidFd);
+
+  uptr size = atomic_load(&pc_array_size, memory_order_relaxed);
+  size += npcs * sizeof(uptr);
+
+  if (size > pc_array_mapped_size) {
+    uptr new_mapped_size = pc_array_mapped_size;
+    while (size > new_mapped_size) new_mapped_size += kPcArrayMmapSize;
+
+    // Extend the file and map the new space at the end of pc_array.
+    uptr res = internal_ftruncate(pc_fd, new_mapped_size);
+    int err;
+    if (internal_iserror(res, &err)) {
+      Printf("failed to extend raw coverage file: %d\n", err);
+      Die();
+    }
+    void *p = MapWritableFileToMemory(pc_array + pc_array_mapped_size,
+                                      new_mapped_size - pc_array_mapped_size,
+                                      pc_fd, pc_array_mapped_size);
+    CHECK_EQ(p, pc_array + pc_array_mapped_size);
+    pc_array_mapped_size = new_mapped_size;
+  }
+
+  atomic_store(&pc_array_size, size, memory_order_release);
+}
+
+// Simply add the pc into the vector under lock. If the function is called more
+// than once for a given PC it will be inserted multiple times, which is fine.
+void CoverageData::Add(uptr pc) {
+  if (!pc_array) return;
+  uptr idx = atomic_fetch_add(&pc_array_index, 1, memory_order_relaxed);
+  CHECK_LT(idx * sizeof(uptr),
+           atomic_load(&pc_array_size, memory_order_acquire));
+  pc_array[idx] = pc;
+}
+
+uptr *CoverageData::data() {
+  return pc_array;
+}
+
+uptr CoverageData::size() {
+  return atomic_load(&pc_array_index, memory_order_relaxed);
+}
+
+// Block layout for packed file format: header, followed by module name (no
+// trailing zero), followed by data blob.
+struct CovHeader {
+  int pid;
+  unsigned int module_name_length;
+  unsigned int data_length;
+};
+
+static void CovWritePacked(int pid, const char *module, const void *blob,
+                           unsigned int blob_size) {
+  if (cov_fd < 0) return;
+  unsigned module_name_length = internal_strlen(module);
+  CovHeader header = {pid, module_name_length, blob_size};
+
+  if (cov_max_block_size == 0) {
+    // Writing to a file. Just go ahead.
+    internal_write(cov_fd, &header, sizeof(header));
+    internal_write(cov_fd, module, module_name_length);
+    internal_write(cov_fd, blob, blob_size);
+  } else {
+    // Writing to a socket. We want to split the data into appropriately sized
+    // blocks.
+    InternalScopedBuffer<char> block(cov_max_block_size);
+    CHECK_EQ((uptr)block.data(), (uptr)(CovHeader *)block.data());
+    uptr header_size_with_module = sizeof(header) + module_name_length;
+    CHECK_LT(header_size_with_module, cov_max_block_size);
+    unsigned int max_payload_size =
+        cov_max_block_size - header_size_with_module;
+    char *block_pos = block.data();
+    internal_memcpy(block_pos, &header, sizeof(header));
+    block_pos += sizeof(header);
+    internal_memcpy(block_pos, module, module_name_length);
+    block_pos += module_name_length;
+    char *block_data_begin = block_pos;
+    char *blob_pos = (char *)blob;
+    while (blob_size > 0) {
+      unsigned int payload_size = Min(blob_size, max_payload_size);
+      blob_size -= payload_size;
+      internal_memcpy(block_data_begin, blob_pos, payload_size);
+      blob_pos += payload_size;
+      ((CovHeader *)block.data())->data_length = payload_size;
+      internal_write(cov_fd, block.data(),
+                     header_size_with_module + payload_size);
+    }
+  }
+}
+
+// If packed = false: <name>.<pid>.<sancov> (name = module name).
+// If packed = true and name == 0: <pid>.<sancov>.<packed>.
+// If packed = true and name != 0: <name>.<sancov>.<packed> (name is
+// user-supplied).
+static int CovOpenFile(bool packed, const char* name) {
+  InternalScopedBuffer<char> path(1024);
+  if (!packed) {
+    CHECK(name);
+    internal_snprintf((char *)path.data(), path.size(), "%s/%s.%zd.sancov",
+                      common_flags()->coverage_dir, name, internal_getpid());
+  } else {
+    if (!name)
+      internal_snprintf((char *)path.data(), path.size(),
+                        "%s/%zd.sancov.packed", common_flags()->coverage_dir,
+                        internal_getpid());
+    else
+      internal_snprintf((char *)path.data(), path.size(), "%s/%s.sancov.packed",
+                        common_flags()->coverage_dir, name);
+  }
+  uptr fd = OpenFile(path.data(), true);
+  if (internal_iserror(fd)) {
+    Report(" SanitizerCoverage: failed to open %s for writing\n", path.data());
+    return -1;
+  }
+  return fd;
+}
+
+// Dump the coverage on disk.
+static void CovDump() {
+  if (!common_flags()->coverage || common_flags()->coverage_direct) return;
+#if !SANITIZER_WINDOWS
+  if (atomic_fetch_add(&dump_once_guard, 1, memory_order_relaxed))
+    return;
+  uptr size = coverage_data.size();
+  InternalMmapVector<u32> offsets(size);
+  uptr *vb = coverage_data.data();
+  uptr *ve = vb + size;
+  SortArray(vb, size);
+  MemoryMappingLayout proc_maps(/*cache_enabled*/true);
+  uptr mb, me, off, prot;
+  InternalScopedBuffer<char> module(4096);
+  InternalScopedBuffer<char> path(4096 * 2);
+  for (int i = 0;
+       proc_maps.Next(&mb, &me, &off, module.data(), module.size(), &prot);
+       i++) {
+    if ((prot & MemoryMappingLayout::kProtectionExecute) == 0)
+      continue;
+    while (vb < ve && *vb < mb) vb++;
+    if (vb >= ve) break;
+    if (*vb < me) {
+      offsets.clear();
+      const uptr *old_vb = vb;
+      CHECK_LE(off, *vb);
+      for (; vb < ve && *vb < me; vb++) {
+        uptr diff = *vb - (i ? mb : 0) + off;
+        CHECK_LE(diff, 0xffffffffU);
+        offsets.push_back(static_cast<u32>(diff));
+      }
+      char *module_name = StripModuleName(module.data());
+      if (cov_sandboxed) {
+        if (cov_fd >= 0) {
+          CovWritePacked(internal_getpid(), module_name, offsets.data(),
+                         offsets.size() * sizeof(u32));
+          VReport(1, " CovDump: %zd PCs written to packed file\n", vb - old_vb);
+        }
+      } else {
+        // One file per module per process.
+        internal_snprintf((char *)path.data(), path.size(), "%s/%s.%zd.sancov",
+                          common_flags()->coverage_dir, module_name,
+                          internal_getpid());
+        int fd = CovOpenFile(false /* packed */, module_name);
+        if (fd > 0) {
+          internal_write(fd, offsets.data(), offsets.size() * sizeof(u32));
+          internal_close(fd);
+          VReport(1, " CovDump: %s: %zd PCs written\n", path.data(),
+                  vb - old_vb);
+        }
+      }
+      InternalFree(module_name);
+    }
+  }
+  if (cov_fd >= 0)
+    internal_close(cov_fd);
+#endif  // !SANITIZER_WINDOWS
+}
+
+void CovPrepareForSandboxing(__sanitizer_sandbox_arguments *args) {
+  if (!args) return;
+  if (!common_flags()->coverage) return;
+  cov_sandboxed = args->coverage_sandboxed;
+  if (!cov_sandboxed) return;
+  cov_fd = args->coverage_fd;
+  cov_max_block_size = args->coverage_max_block_size;
+  if (cov_fd < 0)
+    // Pre-open the file now. The sandbox won't allow us to do it later.
+    cov_fd = CovOpenFile(true /* packed */, 0);
+}
+
+int MaybeOpenCovFile(const char *name) {
+  CHECK(name);
+  if (!common_flags()->coverage) return -1;
+  return CovOpenFile(true /* packed */, name);
+}
+
+void CovBeforeFork() {
+  coverage_data.BeforeFork();
+}
+
+void CovAfterFork(int child_pid) {
+  coverage_data.AfterFork(child_pid);
+}
+
+}  // namespace __sanitizer
+
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov() {
+  coverage_data.Add(StackTrace::GetPreviousInstructionPc(GET_CALLER_PC()));
+}
+SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov_dump() { CovDump(); }
+SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov_init() {
+  coverage_data.Init();
+}
+SANITIZER_INTERFACE_ATTRIBUTE void __sanitizer_cov_module_init(uptr npcs) {
+  if (!common_flags()->coverage || !common_flags()->coverage_direct) return;
+  if (SANITIZER_ANDROID) {
+    // dlopen/dlclose interceptors do not work on Android, so we rely on
+    // Extend() calls to update .sancov.map.
+    CovUpdateMapping(GET_CALLER_PC());
+  }
+  coverage_data.Extend(npcs);
+}
+SANITIZER_INTERFACE_ATTRIBUTE
+sptr __sanitizer_maybe_open_cov_file(const char *name) {
+  return MaybeOpenCovFile(name);
+}
+}  // extern "C"
diff --git a/libsanitizer/sanitizer_common/sanitizer_coverage_mapping_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_coverage_mapping_libcdep.cc
new file mode 100644
index 00000000000..a134053a719
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_coverage_mapping_libcdep.cc
@@ -0,0 +1,126 @@
+//===-- sanitizer_coverage_mapping.cc -------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Mmap-based implementation of sanitizer coverage.
+//
+// This is part of the implementation of code coverage that does not require
+// __sanitizer_cov_dump() call. Data is stored in 2 files per process.
+//
+// $pid.sancov.map describes process memory layout in the following text-based
+// format:
+// <pointer size in bits>  // 1 line, 32 or 64
+// <mapping start> <mapping end> <base address> <dso name> // repeated
+// ...
+// Mapping lines are NOT sorted. This file is updated every time memory layout
+// is changed (i.e. in dlopen() and dlclose() interceptors).
+//
+// $pid.sancov.raw is a binary dump of PC values, sizeof(uptr) each. Again, not
+// sorted. This file is extended by 64Kb at a time and mapped into memory. It
+// contains one or more 0 words at the end, up to the next 64Kb aligned offset.
+//
+// To convert these 2 files to the usual .sancov format, run sancov.py rawunpack
+// $pid.sancov.raw.
+//
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_allocator_internal.h"
+#include "sanitizer_libc.h"
+#include "sanitizer_procmaps.h"
+
+namespace __sanitizer {
+
+static const uptr kMaxNumberOfModules = 1 << 14;
+static const uptr kMaxTextSize = 64 * 1024;
+
+struct CachedMapping {
+ public:
+  bool NeedsUpdate(uptr pc) {
+    int new_pid = internal_getpid();
+    if (last_pid == new_pid && pc && pc >= last_range_start &&
+        pc < last_range_end)
+      return false;
+    last_pid = new_pid;
+    return true;
+  }
+
+  void SetModuleRange(uptr start, uptr end) {
+    last_range_start = start;
+    last_range_end = end;
+  }
+
+ private:
+  uptr last_range_start, last_range_end;
+  int last_pid;
+};
+
+static CachedMapping cached_mapping;
+static StaticSpinMutex mapping_mu;
+
+void CovUpdateMapping(uptr caller_pc) {
+  if (!common_flags()->coverage || !common_flags()->coverage_direct) return;
+
+  SpinMutexLock l(&mapping_mu);
+
+  if (!cached_mapping.NeedsUpdate(caller_pc))
+    return;
+
+  InternalScopedString text(kMaxTextSize);
+  InternalScopedBuffer<char> modules_data(kMaxNumberOfModules *
+                                          sizeof(LoadedModule));
+  LoadedModule *modules = (LoadedModule *)modules_data.data();
+  CHECK(modules);
+  int n_modules = GetListOfModules(modules, kMaxNumberOfModules,
+                                   /* filter */ 0);
+
+  text.append("%d\n", sizeof(uptr) * 8);
+  for (int i = 0; i < n_modules; ++i) {
+    char *module_name = StripModuleName(modules[i].full_name());
+    for (unsigned j = 0; j < modules[i].n_ranges(); ++j) {
+      if (modules[i].address_range_executable(j)) {
+        uptr start = modules[i].address_range_start(j);
+        uptr end = modules[i].address_range_end(j);
+        uptr base = modules[i].base_address();
+        text.append("%zx %zx %zx %s\n", start, end, base, module_name);
+        if (caller_pc && caller_pc >= start && caller_pc < end)
+          cached_mapping.SetModuleRange(start, end);
+      }
+    }
+    InternalFree(module_name);
+  }
+
+  int err;
+  InternalScopedString tmp_path(64 +
+                                internal_strlen(common_flags()->coverage_dir));
+  uptr res = internal_snprintf((char *)tmp_path.data(), tmp_path.size(),
+                    "%s/%zd.sancov.map.tmp", common_flags()->coverage_dir,
+                    internal_getpid());
+  CHECK_LE(res, tmp_path.size());
+  uptr map_fd = OpenFile(tmp_path.data(), true);
+  if (internal_iserror(map_fd)) {
+    Report(" Coverage: failed to open %s for writing\n", tmp_path.data());
+    Die();
+  }
+
+  res = internal_write(map_fd, text.data(), text.length());
+  if (internal_iserror(res, &err)) {
+    Printf("sancov.map write failed: %d\n", err);
+    Die();
+  }
+  internal_close(map_fd);
+
+  InternalScopedString path(64 + internal_strlen(common_flags()->coverage_dir));
+  res = internal_snprintf((char *)path.data(), path.size(), "%s/%zd.sancov.map",
+                    common_flags()->coverage_dir, internal_getpid());
+  CHECK_LE(res, path.size());
+  res = internal_rename(tmp_path.data(), path.data());
+  if (internal_iserror(res, &err)) {
+    Printf("sancov.map rename failed: %d\n", err);
+    Die();
+  }
+}
+
+}  // namespace __sanitizer
diff --git a/libsanitizer/sanitizer_common/sanitizer_deadlock_detector2.cc b/libsanitizer/sanitizer_common/sanitizer_deadlock_detector2.cc
index f4d46d99736..bc96ad230db 100644
--- a/libsanitizer/sanitizer_common/sanitizer_deadlock_detector2.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_deadlock_detector2.cc
@@ -185,8 +185,7 @@ u32 DD::allocateId(DDCallback *cb) {
     id = id_gen++;
   }
   CHECK_LE(id, kMaxMutex);
-  VPrintf(3, "#%llu: DD::allocateId assign id %d\n",
-      cb->lt->ctx, id);
+  VPrintf(3, "#%llu: DD::allocateId assign id %d\n", cb->lt->ctx, id);
   return id;
 }
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_flags.cc b/libsanitizer/sanitizer_common/sanitizer_flags.cc
index e90d79e4360..476f793c27d 100644
--- a/libsanitizer/sanitizer_common/sanitizer_flags.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_flags.cc
@@ -27,6 +27,11 @@ struct FlagDescription {
 
 IntrusiveList<FlagDescription> flag_descriptions;
 
+// If set, the tool will install its own SEGV signal handler by default.
+#ifndef SANITIZER_NEEDS_SEGV
+# define SANITIZER_NEEDS_SEGV 1
+#endif
+
 void SetCommonFlagsDefaults(CommonFlags *f) {
   f->symbolize = true;
   f->external_symbolizer_path = 0;
@@ -53,7 +58,12 @@ void SetCommonFlagsDefaults(CommonFlags *f) {
   f->legacy_pthread_cond = false;
   f->intercept_tls_get_addr = false;
   f->coverage = false;
+  f->coverage_direct = SANITIZER_ANDROID;
+  f->coverage_dir = ".";
   f->full_address_space = false;
+  f->suppressions = "";
+  f->print_suppressions = true;
+  f->disable_coredump = (SANITIZER_WORDSIZE == 64);
 }
 
 void ParseCommonFlagsFromString(CommonFlags *f, const char *str) {
@@ -125,9 +135,23 @@ void ParseCommonFlagsFromString(CommonFlags *f, const char *str) {
   ParseFlag(str, &f->coverage, "coverage",
       "If set, coverage information will be dumped at program shutdown (if the "
       "coverage instrumentation was enabled at compile time).");
+  ParseFlag(str, &f->coverage_direct, "coverage_direct",
+            "If set, coverage information will be dumped directly to a memory "
+            "mapped file. This way data is not lost even if the process is "
+            "suddenly killed.");
+  ParseFlag(str, &f->coverage_dir, "coverage_dir",
+            "Target directory for coverage dumps. Defaults to the current "
+            "directory.");
   ParseFlag(str, &f->full_address_space, "full_address_space",
             "Sanitize complete address space; "
             "by default kernel area on 32-bit platforms will not be sanitized");
+  ParseFlag(str, &f->suppressions, "suppressions", "Suppressions file name.");
+  ParseFlag(str, &f->print_suppressions, "print_suppressions",
+            "Print matched suppressions at exit.");
+  ParseFlag(str, &f->disable_coredump, "disable_coredump",
+      "Disable core dumping. By default, disable_core=1 on 64-bit to avoid "
+      "dumping a 16T+ core file. Ignored on OSes that don't dump core by"
+      "default and for sanitizers that don't reserve lots of virtual memory.");
 
   // Do a sanity check for certain flags.
   if (f->malloc_context_size < 1)
@@ -143,14 +167,17 @@ static bool GetFlagValue(const char *env, const char *name,
     pos = internal_strstr(env, name);
     if (pos == 0)
       return false;
-    if (pos != env && ((pos[-1] >= 'a' && pos[-1] <= 'z') || pos[-1] == '_')) {
+    const char *name_end = pos + internal_strlen(name);
+    if ((pos != env &&
+         ((pos[-1] >= 'a' && pos[-1] <= 'z') || pos[-1] == '_')) ||
+        *name_end != '=') {
       // Seems to be middle of another flag name or value.
       env = pos + 1;
       continue;
     }
+    pos = name_end;
     break;
   }
-  pos += internal_strlen(name);
   const char *end;
   if (pos[0] != '=') {
     end = pos;
diff --git a/libsanitizer/sanitizer_common/sanitizer_flags.h b/libsanitizer/sanitizer_common/sanitizer_flags.h
index 107b6776b3a..a906c9e4535 100644
--- a/libsanitizer/sanitizer_common/sanitizer_flags.h
+++ b/libsanitizer/sanitizer_common/sanitizer_flags.h
@@ -52,7 +52,12 @@ struct CommonFlags {
   bool help;
   uptr mmap_limit_mb;
   bool coverage;
+  bool coverage_direct;
+  const char *coverage_dir;
   bool full_address_space;
+  const char *suppressions;
+  bool print_suppressions;
+  bool disable_coredump;
 };
 
 inline CommonFlags *common_flags() {
diff --git a/libsanitizer/sanitizer_common/sanitizer_freebsd.h b/libsanitizer/sanitizer_common/sanitizer_freebsd.h
new file mode 100644
index 00000000000..47bb1313e6f
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_freebsd.h
@@ -0,0 +1,135 @@
+//===-- sanitizer_freebsd.h -------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of Sanitizer runtime. It contains FreeBSD-specific
+// definitions.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef SANITIZER_FREEBSD_H
+#define SANITIZER_FREEBSD_H
+
+#include "sanitizer_internal_defs.h"
+
+// x86-64 FreeBSD 9.2 and older define 'ucontext_t' incorrectly in
+// 32-bit mode.
+#if SANITIZER_FREEBSD && (SANITIZER_WORDSIZE == 32)
+# include <osreldate.h>
+# if __FreeBSD_version <= 902001  // v9.2
+#  include <link.h>
+#  include <sys/param.h>
+#  include <ucontext.h>
+
+namespace __sanitizer {
+
+typedef unsigned long long __xuint64_t;
+
+typedef __int32_t __xregister_t;
+
+typedef struct __xmcontext {
+  __xregister_t mc_onstack;
+  __xregister_t mc_gs;
+  __xregister_t mc_fs;
+  __xregister_t mc_es;
+  __xregister_t mc_ds;
+  __xregister_t mc_edi;
+  __xregister_t mc_esi;
+  __xregister_t mc_ebp;
+  __xregister_t mc_isp;
+  __xregister_t mc_ebx;
+  __xregister_t mc_edx;
+  __xregister_t mc_ecx;
+  __xregister_t mc_eax;
+  __xregister_t mc_trapno;
+  __xregister_t mc_err;
+  __xregister_t mc_eip;
+  __xregister_t mc_cs;
+  __xregister_t mc_eflags;
+  __xregister_t mc_esp;
+  __xregister_t mc_ss;
+
+  int mc_len;
+  int mc_fpformat;
+  int mc_ownedfp;
+  __xregister_t mc_flags;
+
+  int mc_fpstate[128] __aligned(16);
+  __xregister_t mc_fsbase;
+  __xregister_t mc_gsbase;
+  __xregister_t mc_xfpustate;
+  __xregister_t mc_xfpustate_len;
+
+  int mc_spare2[4];
+} xmcontext_t;
+
+typedef struct __xucontext {
+  sigset_t  uc_sigmask;
+  xmcontext_t  uc_mcontext;
+
+  struct __ucontext *uc_link;
+  stack_t uc_stack;
+  int uc_flags;
+  int __spare__[4];
+} xucontext_t;
+
+struct xkinfo_vmentry {
+  int kve_structsize;
+  int kve_type;
+  __xuint64_t kve_start;
+  __xuint64_t kve_end;
+  __xuint64_t kve_offset;
+  __xuint64_t kve_vn_fileid;
+  __uint32_t kve_vn_fsid;
+  int kve_flags;
+  int kve_resident;
+  int kve_private_resident;
+  int kve_protection;
+  int kve_ref_count;
+  int kve_shadow_count;
+  int kve_vn_type;
+  __xuint64_t kve_vn_size;
+  __uint32_t kve_vn_rdev;
+  __uint16_t kve_vn_mode;
+  __uint16_t kve_status;
+  int _kve_ispare[12];
+  char kve_path[PATH_MAX];
+};
+
+typedef struct {
+  __uint32_t p_type;
+  __uint32_t p_offset;
+  __uint32_t p_vaddr;
+  __uint32_t p_paddr;
+  __uint32_t p_filesz;
+  __uint32_t p_memsz;
+  __uint32_t p_flags;
+  __uint32_t p_align;
+} XElf32_Phdr;
+
+struct xdl_phdr_info {
+  Elf_Addr dlpi_addr;
+  const char *dlpi_name;
+  const XElf32_Phdr *dlpi_phdr;
+  Elf_Half dlpi_phnum;
+  unsigned long long int dlpi_adds;
+  unsigned long long int dlpi_subs;
+  size_t dlpi_tls_modid;
+  void *dlpi_tls_data;
+};
+
+typedef int (*__xdl_iterate_hdr_callback)(struct xdl_phdr_info*, size_t, void*);
+typedef int xdl_iterate_phdr_t(__xdl_iterate_hdr_callback, void*);
+
+#define xdl_iterate_phdr(callback, param) \
+  (((xdl_iterate_phdr_t*) dl_iterate_phdr)((callback), (param)))
+
+}  // namespace __sanitizer
+
+# endif  // __FreeBSD_version <= 902001
+#endif  // SANITIZER_FREEBSD && (SANITIZER_WORDSIZE == 32)
+
+#endif  // SANITIZER_FREEBSD_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_internal_defs.h b/libsanitizer/sanitizer_common/sanitizer_internal_defs.h
index a925d306b61..f1f243c17e2 100644
--- a/libsanitizer/sanitizer_common/sanitizer_internal_defs.h
+++ b/libsanitizer/sanitizer_common/sanitizer_internal_defs.h
@@ -32,9 +32,13 @@
 # define SANITIZER_SUPPORTS_WEAK_HOOKS 0
 #endif
 
-// If set, the tool will install its own SEGV signal handler.
-#ifndef SANITIZER_NEEDS_SEGV
-# define SANITIZER_NEEDS_SEGV 1
+// We can use .preinit_array section on Linux to call sanitizer initialization
+// functions very early in the process startup (unless PIC macro is defined).
+// FIXME: do we have anything like this on Mac?
+#if SANITIZER_LINUX && !SANITIZER_ANDROID && !defined(PIC)
+# define SANITIZER_CAN_USE_PREINIT_ARRAY 1
+#else
+# define SANITIZER_CAN_USE_PREINIT_ARRAY 0
 #endif
 
 // GCC does not understand __has_feature
diff --git a/libsanitizer/sanitizer_common/sanitizer_libc.h b/libsanitizer/sanitizer_common/sanitizer_libc.h
index fd03b791c58..680b888e219 100644
--- a/libsanitizer/sanitizer_common/sanitizer_libc.h
+++ b/libsanitizer/sanitizer_common/sanitizer_libc.h
@@ -72,6 +72,7 @@ uptr internal_open(const char *filename, int flags, u32 mode);
 
 uptr internal_read(fd_t fd, void *buf, uptr count);
 uptr internal_write(fd_t fd, const void *buf, uptr count);
+uptr internal_ftruncate(fd_t fd, uptr size);
 
 // OS
 uptr internal_filesize(fd_t fd);  // -1 on error.
@@ -81,6 +82,7 @@ uptr internal_fstat(fd_t fd, void *buf);
 uptr internal_dup2(int oldfd, int newfd);
 uptr internal_readlink(const char *path, char *buf, uptr bufsize);
 uptr internal_unlink(const char *path);
+uptr internal_rename(const char *oldpath, const char *newpath);
 void NORETURN internal__exit(int exitcode);
 uptr internal_lseek(fd_t fd, OFF_T offset, int whence);
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_libignore.cc b/libsanitizer/sanitizer_common/sanitizer_libignore.cc
index c88550dec8d..9877d668658 100644
--- a/libsanitizer/sanitizer_common/sanitizer_libignore.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_libignore.cc
@@ -6,7 +6,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_platform.h"
-#if SANITIZER_LINUX
+#if SANITIZER_FREEBSD || SANITIZER_LINUX
 
 #include "sanitizer_libignore.h"
 #include "sanitizer_flags.h"
@@ -101,4 +101,4 @@ void LibIgnore::OnLibraryUnloaded() {
 
 }  // namespace __sanitizer
 
-#endif  // #if SANITIZER_LINUX
+#endif  // #if SANITIZER_FREEBSD || SANITIZER_LINUX
diff --git a/libsanitizer/sanitizer_common/sanitizer_linux.cc b/libsanitizer/sanitizer_common/sanitizer_linux.cc
index faa85acd696..8c7979fc6b5 100644
--- a/libsanitizer/sanitizer_common/sanitizer_linux.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_linux.cc
@@ -44,15 +44,16 @@
 #include <sys/time.h>
 #include <sys/types.h>
 #include <unistd.h>
-#include <unwind.h>
 
 #if SANITIZER_FREEBSD
+#include <sys/sysctl.h>
 #include <machine/atomic.h>
 extern "C" {
 // <sys/umtx.h> must be included after <errno.h> and <sys/types.h> on
 // FreeBSD 9.2 and 10.0.
 #include <sys/umtx.h>
 }
+extern char **environ;  // provided by crt1
 #endif  // SANITIZER_FREEBSD
 
 #if !SANITIZER_ANDROID
@@ -132,7 +133,7 @@ uptr internal_open(const char *filename, int flags, u32 mode) {
 
 uptr OpenFile(const char *filename, bool write) {
   return internal_open(filename,
-      write ? O_WRONLY | O_CREAT /*| O_CLOEXEC*/ : O_RDONLY, 0660);
+      write ? O_RDWR | O_CREAT /*| O_CLOEXEC*/ : O_RDONLY, 0660);
 }
 
 uptr internal_read(fd_t fd, void *buf, uptr count) {
@@ -149,6 +150,12 @@ uptr internal_write(fd_t fd, const void *buf, uptr count) {
   return res;
 }
 
+uptr internal_ftruncate(fd_t fd, uptr size) {
+  sptr res;
+  HANDLE_EINTR(res, (sptr)internal_syscall(SYSCALL(ftruncate), fd, size));
+  return res;
+}
+
 #if !SANITIZER_LINUX_USES_64BIT_SYSCALLS && !SANITIZER_FREEBSD
 static void stat64_to_stat(struct stat64 *in, struct stat *out) {
   internal_memset(out, 0, sizeof(*out));
@@ -244,6 +251,15 @@ uptr internal_unlink(const char *path) {
 #endif
 }
 
+uptr internal_rename(const char *oldpath, const char *newpath) {
+#if SANITIZER_USES_CANONICAL_LINUX_SYSCALLS
+  return internal_syscall(SYSCALL(renameat), AT_FDCWD, (uptr)oldpath, AT_FDCWD,
+                          (uptr)newpath);
+#else
+  return internal_syscall(SYSCALL(rename), (uptr)oldpath, (uptr)newpath);
+#endif
+}
+
 uptr internal_sched_yield() {
   return internal_syscall(SYSCALL(sched_yield));
 }
@@ -297,9 +313,20 @@ u64 NanoTime() {
   return (u64)tv.tv_sec * 1000*1000*1000 + tv.tv_usec * 1000;
 }
 
-// Like getenv, but reads env directly from /proc and does not use libc.
-// This function should be called first inside __asan_init.
+// Like getenv, but reads env directly from /proc (on Linux) or parses the
+// 'environ' array (on FreeBSD) and does not use libc. This function should be
+// called first inside __asan_init.
 const char *GetEnv(const char *name) {
+#if SANITIZER_FREEBSD
+  if (::environ != 0) {
+    uptr NameLen = internal_strlen(name);
+    for (char **Env = ::environ; *Env != 0; Env++) {
+      if (internal_strncmp(*Env, name, NameLen) == 0 && (*Env)[NameLen] == '=')
+        return (*Env) + NameLen + 1;
+    }
+  }
+  return 0;  // Not found.
+#elif SANITIZER_LINUX
   static char *environ;
   static uptr len;
   static bool inited;
@@ -323,6 +350,9 @@ const char *GetEnv(const char *name) {
     p = endp + 1;
   }
   return 0;  // Not found.
+#else
+#error "Unsupported platform"
+#endif
 }
 
 extern "C" {
@@ -388,20 +418,6 @@ void GetThreadStackAndTls(bool main, uptr *stk_addr, uptr *stk_size,
 }
 #endif  // SANITIZER_GO
 
-void PrepareForSandboxing(__sanitizer_sandbox_arguments *args) {
-  // Some kinds of sandboxes may forbid filesystem access, so we won't be able
-  // to read the file mappings from /proc/self/maps. Luckily, neither the
-  // process will be able to load additional libraries, so it's fine to use the
-  // cached mappings.
-  MemoryMappingLayout::CacheMemoryMappings();
-  // Same for /proc/self/exe in the symbolizer.
-#if !SANITIZER_GO
-  if (Symbolizer *sym = Symbolizer::GetOrNull())
-    sym->PrepareForSandboxing();
-  CovPrepareForSandboxing(args);
-#endif
-}
-
 enum MutexState {
   MtxUnlocked = 0,
   MtxLocked = 1,
@@ -506,7 +522,11 @@ uptr internal_sigaltstack(const struct sigaltstack *ss,
 }
 
 int internal_fork() {
+#if SANITIZER_USES_CANONICAL_LINUX_SYSCALLS
+  return internal_syscall(SYSCALL(clone), SIGCHLD, 0);
+#else
   return internal_syscall(SYSCALL(fork));
+#endif
 }
 
 #if SANITIZER_LINUX
@@ -660,24 +680,32 @@ static char proc_self_exe_cache_str[kMaxPathLength];
 static uptr proc_self_exe_cache_len = 0;
 
 uptr ReadBinaryName(/*out*/char *buf, uptr buf_len) {
+  if (proc_self_exe_cache_len > 0) {
+    // If available, use the cached module name.
+    uptr module_name_len =
+        internal_snprintf(buf, buf_len, "%s", proc_self_exe_cache_str);
+    CHECK_LT(module_name_len, buf_len);
+    return module_name_len;
+  }
+#if SANITIZER_FREEBSD
+  const int Mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1 };
+  size_t Size = buf_len;
+  bool IsErr = (sysctl(Mib, 4, buf, &Size, NULL, 0) != 0);
+  int readlink_error = IsErr ? errno : 0;
+  uptr module_name_len = Size;
+#else
   uptr module_name_len = internal_readlink(
       "/proc/self/exe", buf, buf_len);
   int readlink_error;
-  if (internal_iserror(module_name_len, &readlink_error)) {
-    if (proc_self_exe_cache_len) {
-      // If available, use the cached module name.
-      CHECK_LE(proc_self_exe_cache_len, buf_len);
-      internal_strncpy(buf, proc_self_exe_cache_str, buf_len);
-      module_name_len = internal_strlen(proc_self_exe_cache_str);
-    } else {
-      // We can't read /proc/self/exe for some reason, assume the name of the
-      // binary is unknown.
-      Report("WARNING: readlink(\"/proc/self/exe\") failed with errno %d, "
-             "some stack frames may not be symbolized\n", readlink_error);
-      module_name_len = internal_snprintf(buf, buf_len, "/proc/self/exe");
-    }
+  bool IsErr = internal_iserror(module_name_len, &readlink_error);
+#endif
+  if (IsErr) {
+    // We can't read /proc/self/exe for some reason, assume the name of the
+    // binary is unknown.
+    Report("WARNING: readlink(\"/proc/self/exe\") failed with errno %d, "
+           "some stack frames may not be symbolized\n", readlink_error);
+    module_name_len = internal_snprintf(buf, buf_len, "/proc/self/exe");
     CHECK_LT(module_name_len, buf_len);
-    buf[module_name_len] = '\0';
   }
   return module_name_len;
 }
@@ -806,11 +834,19 @@ uptr internal_clone(int (*fn)(void *), void *child_stack, int flags, void *arg,
 #endif  // defined(__x86_64__) && SANITIZER_LINUX
 
 #if SANITIZER_ANDROID
+static atomic_uint8_t android_log_initialized;
+
+void AndroidLogInit() {
+  atomic_store(&android_log_initialized, 1, memory_order_release);
+}
 // This thing is not, strictly speaking, async signal safe, but it does not seem
 // to cause any issues. Alternative is writing to log devices directly, but
 // their location and message format might change in the future, so we'd really
 // like to avoid that.
 void AndroidLogWrite(const char *buffer) {
+  if (!atomic_load(&android_log_initialized, memory_order_acquire))
+    return;
+
   char *copy = internal_strdup(buffer);
   char *p = copy;
   char *q;
diff --git a/libsanitizer/sanitizer_common/sanitizer_linux_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_linux_libcdep.cc
index e754b26e693..36aaafdcc4b 100644
--- a/libsanitizer/sanitizer_common/sanitizer_linux_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_linux_libcdep.cc
@@ -15,23 +15,25 @@
 
 #include "sanitizer_common.h"
 #include "sanitizer_flags.h"
+#include "sanitizer_freebsd.h"
 #include "sanitizer_linux.h"
 #include "sanitizer_placement_new.h"
 #include "sanitizer_procmaps.h"
 #include "sanitizer_stacktrace.h"
 #include "sanitizer_atomic.h"
+#include "sanitizer_symbolizer.h"
+
+#if SANITIZER_ANDROID || SANITIZER_FREEBSD
+#include <dlfcn.h>  // for dlsym()
+#endif
 
-#include <dlfcn.h>
 #include <pthread.h>
 #include <signal.h>
 #include <sys/resource.h>
-#if SANITIZER_FREEBSD
-#define _GNU_SOURCE  // to declare _Unwind_Backtrace() from <unwind.h>
-#endif
-#include <unwind.h>
 
 #if SANITIZER_FREEBSD
 #include <pthread_np.h>
+#include <osreldate.h>
 #define pthread_getattr_np pthread_attr_get_np
 #endif
 
@@ -147,127 +149,6 @@ bool SanitizerGetThreadName(char *name, int max_len) {
 #endif
 }
 
-//------------------------- SlowUnwindStack -----------------------------------
-
-typedef struct {
-  uptr absolute_pc;
-  uptr stack_top;
-  uptr stack_size;
-} backtrace_frame_t;
-
-extern "C" {
-typedef void *(*acquire_my_map_info_list_func)();
-typedef void (*release_my_map_info_list_func)(void *map);
-typedef sptr (*unwind_backtrace_signal_arch_func)(
-    void *siginfo, void *sigcontext, void *map_info_list,
-    backtrace_frame_t *backtrace, uptr ignore_depth, uptr max_depth);
-acquire_my_map_info_list_func acquire_my_map_info_list;
-release_my_map_info_list_func release_my_map_info_list;
-unwind_backtrace_signal_arch_func unwind_backtrace_signal_arch;
-} // extern "C"
-
-#if SANITIZER_ANDROID
-void SanitizerInitializeUnwinder() {
-  void *p = dlopen("libcorkscrew.so", RTLD_LAZY);
-  if (!p) {
-    VReport(1,
-            "Failed to open libcorkscrew.so. You may see broken stack traces "
-            "in SEGV reports.");
-    return;
-  }
-  acquire_my_map_info_list =
-      (acquire_my_map_info_list_func)(uptr)dlsym(p, "acquire_my_map_info_list");
-  release_my_map_info_list =
-      (release_my_map_info_list_func)(uptr)dlsym(p, "release_my_map_info_list");
-  unwind_backtrace_signal_arch = (unwind_backtrace_signal_arch_func)(uptr)dlsym(
-      p, "unwind_backtrace_signal_arch");
-  if (!acquire_my_map_info_list || !release_my_map_info_list ||
-      !unwind_backtrace_signal_arch) {
-    VReport(1,
-            "Failed to find one of the required symbols in libcorkscrew.so. "
-            "You may see broken stack traces in SEGV reports.");
-    acquire_my_map_info_list = NULL;
-    unwind_backtrace_signal_arch = NULL;
-    release_my_map_info_list = NULL;
-  }
-}
-#endif
-
-#ifdef __arm__
-#define UNWIND_STOP _URC_END_OF_STACK
-#define UNWIND_CONTINUE _URC_NO_REASON
-#else
-#define UNWIND_STOP _URC_NORMAL_STOP
-#define UNWIND_CONTINUE _URC_NO_REASON
-#endif
-
-uptr Unwind_GetIP(struct _Unwind_Context *ctx) {
-#ifdef __arm__
-  uptr val;
-  _Unwind_VRS_Result res = _Unwind_VRS_Get(ctx, _UVRSC_CORE,
-      15 /* r15 = PC */, _UVRSD_UINT32, &val);
-  CHECK(res == _UVRSR_OK && "_Unwind_VRS_Get failed");
-  // Clear the Thumb bit.
-  return val & ~(uptr)1;
-#else
-  return _Unwind_GetIP(ctx);
-#endif
-}
-
-struct UnwindTraceArg {
-  StackTrace *stack;
-  uptr max_depth;
-};
-
-_Unwind_Reason_Code Unwind_Trace(struct _Unwind_Context *ctx, void *param) {
-  UnwindTraceArg *arg = (UnwindTraceArg*)param;
-  CHECK_LT(arg->stack->size, arg->max_depth);
-  uptr pc = Unwind_GetIP(ctx);
-  arg->stack->trace[arg->stack->size++] = pc;
-  if (arg->stack->size == arg->max_depth) return UNWIND_STOP;
-  return UNWIND_CONTINUE;
-}
-
-void StackTrace::SlowUnwindStack(uptr pc, uptr max_depth) {
-  CHECK_GE(max_depth, 2);
-  size = 0;
-  UnwindTraceArg arg = {this, Min(max_depth + 1, kStackTraceMax)};
-  _Unwind_Backtrace(Unwind_Trace, &arg);
-  // We need to pop a few frames so that pc is on top.
-  uptr to_pop = LocatePcInTrace(pc);
-  // trace[0] belongs to the current function so we always pop it.
-  if (to_pop == 0)
-    to_pop = 1;
-  PopStackFrames(to_pop);
-  trace[0] = pc;
-}
-
-void StackTrace::SlowUnwindStackWithContext(uptr pc, void *context,
-                                            uptr max_depth) {
-  CHECK_GE(max_depth, 2);
-  if (!unwind_backtrace_signal_arch) {
-    SlowUnwindStack(pc, max_depth);
-    return;
-  }
-
-  void *map = acquire_my_map_info_list();
-  CHECK(map);
-  InternalScopedBuffer<backtrace_frame_t> frames(kStackTraceMax);
-  // siginfo argument appears to be unused.
-  sptr res = unwind_backtrace_signal_arch(/* siginfo */ NULL, context, map,
-                                          frames.data(),
-                                          /* ignore_depth */ 0, max_depth);
-  release_my_map_info_list(map);
-  if (res < 0) return;
-  CHECK_LE((uptr)res, kStackTraceMax);
-
-  size = 0;
-  // +2 compensate for libcorkscrew unwinder returning addresses of call
-  // instructions instead of raw return addresses.
-  for (sptr i = 0; i < res; ++i)
-    trace[size++] = frames[i].absolute_pc + 2;
-}
-
 #if !SANITIZER_FREEBSD
 static uptr g_tls_size;
 #endif
@@ -299,11 +180,11 @@ void InitTlsSize() {
 static atomic_uintptr_t kThreadDescriptorSize;
 
 uptr ThreadDescriptorSize() {
-  char buf[64];
   uptr val = atomic_load(&kThreadDescriptorSize, memory_order_relaxed);
   if (val)
     return val;
 #ifdef _CS_GNU_LIBC_VERSION
+  char buf[64];
   uptr len = confstr(_CS_GNU_LIBC_VERSION, buf, sizeof(buf));
   if (len < sizeof(buf) && internal_strncmp(buf, "glibc 2.", 8) == 0) {
     char *end;
@@ -468,6 +349,10 @@ uptr GetListOfModules(LoadedModule *modules, uptr max_modules,
 #else  // SANITIZER_ANDROID
 # if !SANITIZER_FREEBSD
 typedef ElfW(Phdr) Elf_Phdr;
+# elif SANITIZER_WORDSIZE == 32 && __FreeBSD_version <= 902001  // v9.2
+#  define Elf_Phdr XElf32_Phdr
+#  define dl_phdr_info xdl_phdr_info
+#  define dl_iterate_phdr(c, b) xdl_iterate_phdr((c), (b))
 # endif
 
 struct DlIteratePhdrData {
@@ -504,7 +389,8 @@ static int dl_iterate_phdr_cb(dl_phdr_info *info, size_t size, void *arg) {
     if (phdr->p_type == PT_LOAD) {
       uptr cur_beg = info->dlpi_addr + phdr->p_vaddr;
       uptr cur_end = cur_beg + phdr->p_memsz;
-      cur_module->addAddressRange(cur_beg, cur_end);
+      bool executable = phdr->p_flags & PF_X;
+      cur_module->addAddressRange(cur_beg, cur_end, executable);
     }
   }
   return 0;
@@ -527,6 +413,19 @@ void SetIndirectCallWrapper(uptr wrapper) {
   indirect_call_wrapper = wrapper;
 }
 
+void PrepareForSandboxing(__sanitizer_sandbox_arguments *args) {
+  // Some kinds of sandboxes may forbid filesystem access, so we won't be able
+  // to read the file mappings from /proc/self/maps. Luckily, neither the
+  // process will be able to load additional libraries, so it's fine to use the
+  // cached mappings.
+  MemoryMappingLayout::CacheMemoryMappings();
+  // Same for /proc/self/exe in the symbolizer.
+#if !SANITIZER_GO
+  Symbolizer::GetOrInit()->PrepareForSandboxing();
+  CovPrepareForSandboxing(args);
+#endif
+}
+
 }  // namespace __sanitizer
 
 #endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
diff --git a/libsanitizer/sanitizer_common/sanitizer_mac.cc b/libsanitizer/sanitizer_common/sanitizer_mac.cc
index 6deba53d3e8..2c2a1d74be7 100644
--- a/libsanitizer/sanitizer_common/sanitizer_mac.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_mac.cc
@@ -129,6 +129,14 @@ int internal_fork() {
   return fork();
 }
 
+uptr internal_rename(const char *oldpath, const char *newpath) {
+  return rename(oldpath, newpath);
+}
+
+uptr internal_ftruncate(fd_t fd, uptr size) {
+  return ftruncate(fd, size);
+}
+
 // ----------------- sanitizer_common.h
 bool FileExists(const char *filename) {
   struct stat st;
diff --git a/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.cc b/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.cc
new file mode 100644
index 00000000000..b989ed0c90f
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.cc
@@ -0,0 +1,17 @@
+//===-- sanitizer_persistent_allocator.cc -----------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is shared between AddressSanitizer and ThreadSanitizer
+// run-time libraries.
+//===----------------------------------------------------------------------===//
+#include "sanitizer_persistent_allocator.h"
+
+namespace __sanitizer {
+
+PersistentAllocator thePersistentAllocator;
+
+}  // namespace __sanitizer
diff --git a/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.h b/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.h
new file mode 100644
index 00000000000..e29b7bd57aa
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_persistent_allocator.h
@@ -0,0 +1,69 @@
+//===-- sanitizer_persistent_allocator.h ------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// A fast memory allocator that does not support free() nor realloc().
+// All allocations are forever.
+//===----------------------------------------------------------------------===//
+#ifndef SANITIZER_PERSISTENT_ALLOCATOR_H
+#define SANITIZER_PERSISTENT_ALLOCATOR_H
+
+#include "sanitizer_internal_defs.h"
+#include "sanitizer_mutex.h"
+#include "sanitizer_atomic.h"
+#include "sanitizer_common.h"
+
+namespace __sanitizer {
+
+class PersistentAllocator {
+ public:
+  void *alloc(uptr size);
+
+ private:
+  void *tryAlloc(uptr size);
+  StaticSpinMutex mtx;  // Protects alloc of new blocks for region allocator.
+  atomic_uintptr_t region_pos;  // Region allocator for Node's.
+  atomic_uintptr_t region_end;
+};
+
+inline void *PersistentAllocator::tryAlloc(uptr size) {
+  // Optimisic lock-free allocation, essentially try to bump the region ptr.
+  for (;;) {
+    uptr cmp = atomic_load(&region_pos, memory_order_acquire);
+    uptr end = atomic_load(&region_end, memory_order_acquire);
+    if (cmp == 0 || cmp + size > end) return 0;
+    if (atomic_compare_exchange_weak(&region_pos, &cmp, cmp + size,
+                                     memory_order_acquire))
+      return (void *)cmp;
+  }
+}
+
+inline void *PersistentAllocator::alloc(uptr size) {
+  // First, try to allocate optimisitically.
+  void *s = tryAlloc(size);
+  if (s) return s;
+  // If failed, lock, retry and alloc new superblock.
+  SpinMutexLock l(&mtx);
+  for (;;) {
+    s = tryAlloc(size);
+    if (s) return s;
+    atomic_store(&region_pos, 0, memory_order_relaxed);
+    uptr allocsz = 64 * 1024;
+    if (allocsz < size) allocsz = size;
+    uptr mem = (uptr)MmapOrDie(allocsz, "stack depot");
+    atomic_store(&region_end, mem + allocsz, memory_order_release);
+    atomic_store(&region_pos, mem, memory_order_release);
+  }
+}
+
+extern PersistentAllocator thePersistentAllocator;
+inline void *PersistentAlloc(uptr sz) {
+  return thePersistentAllocator.alloc(sz);
+}
+
+} // namespace __sanitizer
+
+#endif  // SANITIZER_PERSISTENT_ALLOCATOR_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_platform_interceptors.h b/libsanitizer/sanitizer_common/sanitizer_platform_interceptors.h
index 92bbc005106..81a09927d79 100644
--- a/libsanitizer/sanitizer_common/sanitizer_platform_interceptors.h
+++ b/libsanitizer/sanitizer_common/sanitizer_platform_interceptors.h
@@ -27,6 +27,12 @@
 # define SI_LINUX_NOT_ANDROID 0
 #endif
 
+#if SANITIZER_FREEBSD
+# define SI_FREEBSD 1
+#else
+# define SI_FREEBSD 0
+#endif
+
 #if SANITIZER_LINUX
 # define SI_LINUX 1
 #else
@@ -73,11 +79,11 @@
 #define SANITIZER_INTERCEPT_STRPTIME SI_NOT_WINDOWS
 
 #define SANITIZER_INTERCEPT_SCANF SI_NOT_WINDOWS
-#define SANITIZER_INTERCEPT_ISOC99_SCANF SI_LINUX
+#define SANITIZER_INTERCEPT_ISOC99_SCANF SI_LINUX_NOT_ANDROID
 
 #ifndef SANITIZER_INTERCEPT_PRINTF
 # define SANITIZER_INTERCEPT_PRINTF SI_NOT_WINDOWS
-# define SANITIZER_INTERCEPT_ISOC99_PRINTF SI_LINUX
+# define SANITIZER_INTERCEPT_ISOC99_PRINTF SI_LINUX_NOT_ANDROID
 #endif
 
 #define SANITIZER_INTERCEPT_FREXP 1
@@ -86,10 +92,10 @@
 #define SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS \
   SI_MAC || SI_LINUX_NOT_ANDROID
-#define SANITIZER_INTERCEPT_GETPWENT SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_GETPWENT SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_FGETPWENT SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_GETPWENT_R SI_LINUX_NOT_ANDROID
-#define SANITIZER_INTERCEPT_SETPWENT SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_SETPWENT SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_CLOCK_GETTIME SI_LINUX
 #define SANITIZER_INTERCEPT_GETITIMER SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_TIME SI_NOT_WINDOWS
@@ -102,9 +108,12 @@
 #define SANITIZER_INTERCEPT_GETSOCKNAME SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_GETHOSTBYNAME SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_GETHOSTBYNAME_R SI_LINUX
+#define SANITIZER_INTERCEPT_GETHOSTBYNAME2_R SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_GETHOSTBYADDR_R SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_GETHOSTENT_R SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_GETSOCKOPT SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_ACCEPT SI_NOT_WINDOWS
-#define SANITIZER_INTERCEPT_ACCEPT4 SI_LINUX
+#define SANITIZER_INTERCEPT_ACCEPT4 SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_MODF SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_RECVMSG SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_GETPEERNAME SI_NOT_WINDOWS
@@ -117,13 +126,13 @@
    (defined(__i386) || defined (__x86_64))  // NOLINT
 #define SANITIZER_INTERCEPT_SETLOCALE SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_GETCWD SI_NOT_WINDOWS
-#define SANITIZER_INTERCEPT_GET_CURRENT_DIR_NAME SI_LINUX
+#define SANITIZER_INTERCEPT_GET_CURRENT_DIR_NAME SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_STRTOIMAX SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_MBSTOWCS SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_MBSNRTOWCS SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_WCSTOMBS SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_WCSNRTOMBS SI_MAC || SI_LINUX_NOT_ANDROID
-#define SANITIZER_INTERCEPT_TCGETATTR SI_LINUX
+#define SANITIZER_INTERCEPT_TCGETATTR SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_REALPATH SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_CANONICALIZE_FILE_NAME SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_CONFSTR SI_MAC || SI_LINUX_NOT_ANDROID
@@ -140,19 +149,21 @@
 #define SANITIZER_INTERCEPT_SIGWAIT SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_SIGWAITINFO SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_SIGTIMEDWAIT SI_LINUX_NOT_ANDROID
-#define SANITIZER_INTERCEPT_SIGSETOPS SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_SIGSETOPS \
+  SI_FREEBSD || SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_SIGPENDING SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_SIGPROCMASK SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_BACKTRACE SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_GETMNTENT SI_LINUX
 #define SANITIZER_INTERCEPT_GETMNTENT_R SI_LINUX_NOT_ANDROID
-#define SANITIZER_INTERCEPT_STATFS SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_STATFS SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_STATFS64 \
   (SI_MAC && !SI_IOS) || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_STATVFS SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_STATVFS64 SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_INITGROUPS SI_NOT_WINDOWS
-#define SANITIZER_INTERCEPT_ETHER SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_ETHER_NTOA_ATON SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_ETHER_HOST SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_ETHER_R SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_SHMCTL \
   (SI_LINUX_NOT_ANDROID && SANITIZER_WORDSIZE == 64)
@@ -161,6 +172,19 @@
 #define SANITIZER_INTERCEPT_PTHREAD_ATTR_GETINHERITSCHED \
   SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_PTHREAD_ATTR_GETAFFINITY_NP SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPSHARED SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETTYPE SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPROTOCOL \
+  SI_MAC || SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETPRIOCEILING \
+  SI_MAC || SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETROBUST SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_MUTEXATTR_GETROBUST_NP SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_RWLOCKATTR_GETPSHARED SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_PTHREAD_RWLOCKATTR_GETKIND_NP SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_CONDATTR_GETPSHARED SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_PTHREAD_CONDATTR_GETCLOCK SI_LINUX_NOT_ANDROID
+#define SANITIZER_INTERCEPT_PTHREAD_BARRIERATTR_GETPSHARED SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_TMPNAM SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_TMPNAM_R SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_TEMPNAM SI_NOT_WINDOWS
@@ -168,6 +192,7 @@
 #define SANITIZER_INTERCEPT_REMQUO SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_LGAMMA SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_LGAMMA_R SI_LINUX
+#define SANITIZER_INTERCEPT_LGAMMAL_R SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_DRAND48_R SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_RAND_R SI_MAC || SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_ICONV SI_LINUX_NOT_ANDROID
@@ -176,7 +201,7 @@
 // FIXME: getline seems to be available on OSX 10.7
 #define SANITIZER_INTERCEPT_GETLINE SI_LINUX_NOT_ANDROID
 
-#define SANITIZER_INTERCEPT__EXIT SI_LINUX
+#define SANITIZER_INTERCEPT__EXIT SI_LINUX || SI_FREEBSD
 
 #define SANITIZER_INTERCEPT_PHTREAD_MUTEX SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_PTHREAD_SETNAME_NP SI_LINUX_NOT_ANDROID
@@ -201,5 +226,10 @@
 #define SANITIZER_INTERCEPT_OBSTACK SI_LINUX_NOT_ANDROID
 #define SANITIZER_INTERCEPT_FFLUSH SI_NOT_WINDOWS
 #define SANITIZER_INTERCEPT_FCLOSE SI_NOT_WINDOWS
+#define SANITIZER_INTERCEPT_DLOPEN_DLCLOSE SI_LINUX_NOT_ANDROID || SI_MAC
+#define SANITIZER_INTERCEPT_GETPASS SI_LINUX_NOT_ANDROID || SI_MAC
+#define SANITIZER_INTERCEPT_TIMERFD SI_LINUX_NOT_ANDROID
+
+#define SANITIZER_INTERCEPT_MLOCKX SI_NOT_WINDOWS
 
 #endif  // #ifndef SANITIZER_PLATFORM_INTERCEPTORS_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cc b/libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cc
index 76ee9001b7d..8779d8adf72 100644
--- a/libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_platform_limits_linux.cc
@@ -27,7 +27,7 @@
 // are not defined anywhere in userspace headers. Fake them. This seems to work
 // fine with newer headers, too.
 #include <asm/posix_types.h>
-#if defined(__x86_64__)
+#if defined(__x86_64__) ||  defined(__mips__)
 #include <sys/stat.h>
 #else
 #define ino_t __kernel_ino_t
@@ -48,21 +48,19 @@
 
 #include <linux/aio_abi.h>
 
-#if SANITIZER_ANDROID
-#include <asm/statfs.h>
-#else
-#include <sys/statfs.h>
-#endif
-
 #if !SANITIZER_ANDROID
+#include <sys/statfs.h>
 #include <linux/perf_event.h>
 #endif
 
 namespace __sanitizer {
+#if !SANITIZER_ANDROID
   unsigned struct_statfs64_sz = sizeof(struct statfs64);
+#endif
 }  // namespace __sanitizer
 
-#if !defined(__powerpc64__) && !defined(__x86_64__) && !defined(__aarch64__)
+#if !defined(__powerpc64__) && !defined(__x86_64__) && !defined(__aarch64__)\
+                            && !defined(__mips__)
 COMPILER_CHECK(struct___old_kernel_stat_sz == sizeof(struct __old_kernel_stat));
 #endif
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.cc b/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.cc
index a93d38d8aac..f5678dceccc 100644
--- a/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.cc
@@ -33,7 +33,6 @@
 #include <sys/socket.h>
 #include <sys/stat.h>
 #include <sys/time.h>
-#include <sys/timeb.h>
 #include <sys/times.h>
 #include <sys/types.h>
 #include <sys/utsname.h>
@@ -43,6 +42,7 @@
 
 #if !SANITIZER_ANDROID
 #include <sys/mount.h>
+#include <sys/timeb.h>
 #endif
 
 #if SANITIZER_LINUX
@@ -189,13 +189,14 @@ namespace __sanitizer {
   unsigned struct_tms_sz = sizeof(struct tms);
   unsigned struct_sigevent_sz = sizeof(struct sigevent);
   unsigned struct_sched_param_sz = sizeof(struct sched_param);
-  unsigned struct_statfs_sz = sizeof(struct statfs);
+
 
 #if SANITIZER_MAC && !SANITIZER_IOS
   unsigned struct_statfs64_sz = sizeof(struct statfs64);
 #endif // SANITIZER_MAC && !SANITIZER_IOS
 
 #if !SANITIZER_ANDROID
+  unsigned struct_statfs_sz = sizeof(struct statfs);
   unsigned struct_sockaddr_sz = sizeof(struct sockaddr);
   unsigned ucontext_t_sz = sizeof(ucontext_t);
 #endif // !SANITIZER_ANDROID
@@ -287,6 +288,7 @@ namespace __sanitizer {
   int ptrace_setfpregs = PTRACE_SETFPREGS;
   int ptrace_getfpxregs = PTRACE_GETFPXREGS;
   int ptrace_setfpxregs = PTRACE_SETFPXREGS;
+  int ptrace_geteventmsg = PTRACE_GETEVENTMSG;
 #if (defined(PTRACE_GETSIGINFO) && defined(PTRACE_SETSIGINFO)) ||              \
     (defined(PT_GETSIGINFO) && defined(PT_SETSIGINFO))
   int ptrace_getsiginfo = PTRACE_GETSIGINFO;
@@ -396,7 +398,7 @@ namespace __sanitizer {
   unsigned struct_sioc_vif_req_sz = sizeof(struct sioc_vif_req);
 #endif
 
-  unsigned IOCTL_NOT_PRESENT = 0;
+  const unsigned IOCTL_NOT_PRESENT = 0;
 
   unsigned IOCTL_FIOASYNC = FIOASYNC;
   unsigned IOCTL_FIOCLEX = FIOCLEX;
@@ -1056,6 +1058,10 @@ CHECK_SIZE_AND_OFFSET(shmid_ds, shm_nattch);
 
 CHECK_TYPE_SIZE(clock_t);
 
+#if SANITIZER_LINUX
+CHECK_TYPE_SIZE(clockid_t);
+#endif
+
 #if !SANITIZER_ANDROID
 CHECK_TYPE_SIZE(ifaddrs);
 CHECK_SIZE_AND_OFFSET(ifaddrs, ifa_next);
@@ -1086,11 +1092,13 @@ CHECK_SIZE_AND_OFFSET(ifaddrs, ifa_data);
 COMPILER_CHECK(sizeof(__sanitizer_mallinfo) == sizeof(struct mallinfo));
 #endif
 
+#if !SANITIZER_ANDROID
 CHECK_TYPE_SIZE(timeb);
 CHECK_SIZE_AND_OFFSET(timeb, time);
 CHECK_SIZE_AND_OFFSET(timeb, millitm);
 CHECK_SIZE_AND_OFFSET(timeb, timezone);
 CHECK_SIZE_AND_OFFSET(timeb, dstflag);
+#endif
 
 CHECK_TYPE_SIZE(passwd);
 CHECK_SIZE_AND_OFFSET(passwd, pw_name);
diff --git a/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.h b/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.h
index dece2d3cbdd..caa36a4071b 100644
--- a/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.h
+++ b/libsanitizer/sanitizer_common/sanitizer_platform_limits_posix.h
@@ -37,11 +37,11 @@ namespace __sanitizer {
   extern unsigned struct_itimerspec_sz;
   extern unsigned struct_sigevent_sz;
   extern unsigned struct_sched_param_sz;
-  extern unsigned struct_statfs_sz;
   extern unsigned struct_statfs64_sz;
-  extern unsigned struct_sockaddr_sz;
 
 #if !SANITIZER_ANDROID
+  extern unsigned struct_statfs_sz;
+  extern unsigned struct_sockaddr_sz;
   extern unsigned ucontext_t_sz;
 #endif // !SANITIZER_ANDROID
 
@@ -65,6 +65,13 @@ namespace __sanitizer {
 #elif defined(__powerpc64__)
   const unsigned struct_kernel_stat_sz = 144;
   const unsigned struct_kernel_stat64_sz = 104;
+#elif defined(__mips__)
+  #if SANITIZER_WORDSIZE == 64
+  const unsigned struct_kernel_stat_sz = 216;
+  #else
+  const unsigned struct_kernel_stat_sz = 144;
+  #endif
+  const unsigned struct_kernel_stat64_sz = 104;
 #endif
   struct __sanitizer_perf_event_attr {
     unsigned type;
@@ -160,6 +167,12 @@ namespace __sanitizer {
     unsigned __seq;
     u64 __unused1;
     u64 __unused2;
+#elif defined(__mips__)
+    unsigned int mode;
+    unsigned short __seq;
+    unsigned short __pad1;
+    unsigned long __unused1;
+    unsigned long __unused2;
 #else
     unsigned short mode;
     unsigned short __pad1;
@@ -188,15 +201,15 @@ namespace __sanitizer {
     u64 shm_ctime;
   #else
     uptr shm_atime;
-  #ifndef _LP64
+  #if !defined(_LP64) && !defined(__mips__)
     uptr __unused1;
   #endif
     uptr shm_dtime;
-  #ifndef _LP64
+  #if !defined(_LP64) && !defined(__mips__)
     uptr __unused2;
   #endif
     uptr shm_ctime;
-  #ifndef _LP64
+  #if !defined(_LP64) && !defined(__mips__)
     uptr __unused3;
   #endif
   #endif
@@ -438,8 +451,13 @@ namespace __sanitizer {
   typedef long __sanitizer_clock_t;
 #endif
 
+#if SANITIZER_LINUX
+  typedef int __sanitizer_clockid_t;
+#endif
+
 #if SANITIZER_LINUX || SANITIZER_FREEBSD
-#if defined(_LP64) || defined(__x86_64__) || defined(__powerpc__)
+#if defined(_LP64) || defined(__x86_64__) || defined(__powerpc__)\
+                   || defined(__mips__)
   typedef unsigned __sanitizer___kernel_uid_t;
   typedef unsigned __sanitizer___kernel_gid_t;
 #else
@@ -452,7 +470,7 @@ namespace __sanitizer {
   typedef long __sanitizer___kernel_off_t;
 #endif
 
-#if defined(__powerpc__) || defined(__aarch64__)
+#if defined(__powerpc__) || defined(__aarch64__) || defined(__mips__)
   typedef unsigned int __sanitizer___kernel_old_uid_t;
   typedef unsigned int __sanitizer___kernel_old_gid_t;
 #else
@@ -492,6 +510,9 @@ namespace __sanitizer {
 
   // Linux system headers define the 'sa_handler' and 'sa_sigaction' macros.
   struct __sanitizer_sigaction {
+#if defined(__mips__) && !SANITIZER_FREEBSD
+    unsigned int sa_flags;
+#endif
     union {
       void (*sigaction)(int sig, void *siginfo, void *uctx);
       void (*handler)(int sig);
@@ -501,11 +522,16 @@ namespace __sanitizer {
     __sanitizer_sigset_t sa_mask;
 #else
     __sanitizer_sigset_t sa_mask;
+#ifndef __mips__
     int sa_flags;
 #endif
+#endif
 #if SANITIZER_LINUX
     void (*sa_restorer)();
 #endif
+#if defined(__mips__) && (SANITIZER_WORDSIZE == 32)
+    int sa_resv[1];
+#endif
   };
 
 #if SANITIZER_FREEBSD
@@ -676,6 +702,7 @@ namespace __sanitizer {
   extern int ptrace_setsiginfo;
   extern int ptrace_getregset;
   extern int ptrace_setregset;
+  extern int ptrace_geteventmsg;
 #endif
 
 #if (SANITIZER_LINUX || SANITIZER_FREEBSD) && !SANITIZER_ANDROID
@@ -718,7 +745,7 @@ struct __sanitizer_obstack {
 
 #define IOC_NRBITS 8
 #define IOC_TYPEBITS 8
-#if defined(__powerpc__) || defined(__powerpc64__)
+#if defined(__powerpc__) || defined(__powerpc64__) || defined(__mips__)
 #define IOC_SIZEBITS 13
 #define IOC_DIRBITS 3
 #define IOC_NONE 1U
@@ -832,7 +859,7 @@ struct __sanitizer_obstack {
 
   // A special value to mark ioctls that are not present on the target platform,
   // when it can not be determined without including any system headers.
-  extern unsigned IOCTL_NOT_PRESENT;
+  extern const unsigned IOCTL_NOT_PRESENT;
 
   extern unsigned IOCTL_FIOASYNC;
   extern unsigned IOCTL_FIOCLEX;
diff --git a/libsanitizer/sanitizer_common/sanitizer_posix.cc b/libsanitizer/sanitizer_common/sanitizer_posix.cc
index e24d5ed5031..24e99f9472d 100644
--- a/libsanitizer/sanitizer_common/sanitizer_posix.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_posix.cc
@@ -204,6 +204,17 @@ void *MapFileToMemory(const char *file_name, uptr *buff_size) {
   return internal_iserror(map) ? 0 : (void *)map;
 }
 
+void *MapWritableFileToMemory(void *addr, uptr size, uptr fd, uptr offset) {
+  uptr flags = MAP_SHARED;
+  if (addr) flags |= MAP_FIXED;
+  uptr p = internal_mmap(addr, size, PROT_READ | PROT_WRITE, flags, fd, offset);
+  if (internal_iserror(p)) {
+    Printf("could not map writable file (%zd, %zu, %zu): %zd\n", fd, offset,
+           size, p);
+    return 0;
+  }
+  return (void *)p;
+}
 
 static inline bool IntervalsAreSeparate(uptr start1, uptr end1,
                                         uptr start2, uptr end2) {
diff --git a/libsanitizer/sanitizer_common/sanitizer_posix_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_posix_libcdep.cc
index 8e3a96f01e4..b4e42c72462 100644
--- a/libsanitizer/sanitizer_common/sanitizer_posix_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_posix_libcdep.cc
@@ -42,30 +42,49 @@ void FlushUnneededShadowMemory(uptr addr, uptr size) {
   madvise((void*)addr, size, MADV_DONTNEED);
 }
 
-void DisableCoreDumper() {
-  struct rlimit nocore;
-  nocore.rlim_cur = 0;
-  nocore.rlim_max = 0;
-  setrlimit(RLIMIT_CORE, &nocore);
+static rlim_t getlim(int res) {
+  rlimit rlim;
+  CHECK_EQ(0, getrlimit(res, &rlim));
+  return rlim.rlim_cur;
+}
+
+static void setlim(int res, rlim_t lim) {
+  // The following magic is to prevent clang from replacing it with memset.
+  volatile struct rlimit rlim;
+  rlim.rlim_cur = lim;
+  rlim.rlim_max = lim;
+  if (setrlimit(res, (struct rlimit*)&rlim)) {
+    Report("ERROR: %s setrlimit() failed %d\n", SanitizerToolName, errno);
+    Die();
+  }
+}
+
+void DisableCoreDumperIfNecessary() {
+  if (common_flags()->disable_coredump) {
+    setlim(RLIMIT_CORE, 0);
+  }
 }
 
 bool StackSizeIsUnlimited() {
-  struct rlimit rlim;
-  CHECK_EQ(0, getrlimit(RLIMIT_STACK, &rlim));
-  return ((uptr)rlim.rlim_cur == (uptr)-1);
+  rlim_t stack_size = getlim(RLIMIT_STACK);
+  return (stack_size == RLIM_INFINITY);
 }
 
 void SetStackSizeLimitInBytes(uptr limit) {
-  struct rlimit rlim;
-  rlim.rlim_cur = limit;
-  rlim.rlim_max = limit;
-  if (setrlimit(RLIMIT_STACK, &rlim)) {
-    Report("ERROR: %s setrlimit() failed %d\n", SanitizerToolName, errno);
-    Die();
-  }
+  setlim(RLIMIT_STACK, (rlim_t)limit);
   CHECK(!StackSizeIsUnlimited());
 }
 
+bool AddressSpaceIsUnlimited() {
+  rlim_t as_size = getlim(RLIMIT_AS);
+  return (as_size == RLIM_INFINITY);
+}
+
+void SetAddressSpaceUnlimited() {
+  setlim(RLIMIT_AS, RLIM_INFINITY);
+  CHECK(AddressSpaceIsUnlimited());
+}
+
 void SleepForSeconds(int seconds) {
   sleep(seconds);
 }
@@ -127,7 +146,9 @@ static void MaybeInstallSigaction(int signum,
   struct sigaction sigact;
   internal_memset(&sigact, 0, sizeof(sigact));
   sigact.sa_sigaction = (sa_sigaction_t)handler;
-  sigact.sa_flags = SA_SIGINFO;
+  // Do not block the signal from being received in that signal's handler.
+  // Clients are responsible for handling this correctly.
+  sigact.sa_flags = SA_SIGINFO | SA_NODEFER;
   if (common_flags()->use_sigaltstack) sigact.sa_flags |= SA_ONSTACK;
   CHECK_EQ(0, internal_sigaction(signum, &sigact, 0));
   VReport(1, "Installed the sigaction for signal %d\n", signum);
@@ -143,6 +164,28 @@ void InstallDeadlySignalHandlers(SignalHandlerType handler) {
 }
 #endif  // SANITIZER_GO
 
+bool IsAccessibleMemoryRange(uptr beg, uptr size) {
+  uptr page_size = GetPageSizeCached();
+  // Checking too large memory ranges is slow.
+  CHECK_LT(size, page_size * 10);
+  int sock_pair[2];
+  if (pipe(sock_pair))
+    return false;
+  uptr bytes_written =
+      internal_write(sock_pair[1], reinterpret_cast<void *>(beg), size);
+  int write_errno;
+  bool result;
+  if (internal_iserror(bytes_written, &write_errno)) {
+    CHECK_EQ(EFAULT, write_errno);
+    result = false;
+  } else {
+    result = (bytes_written == size);
+  }
+  internal_close(sock_pair[0]);
+  internal_close(sock_pair[1]);
+  return result;
+}
+
 }  // namespace __sanitizer
 
 #endif  // SANITIZER_POSIX
diff --git a/libsanitizer/sanitizer_common/sanitizer_printf.cc b/libsanitizer/sanitizer_common/sanitizer_printf.cc
index 4fc26308ee4..fc0c357ba86 100644
--- a/libsanitizer/sanitizer_common/sanitizer_printf.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_printf.cc
@@ -20,7 +20,8 @@
 #include <stdio.h>
 #include <stdarg.h>
 
-#if SANITIZER_WINDOWS && !defined(va_copy)
+#if SANITIZER_WINDOWS && defined(_MSC_VER) && _MSC_VER < 1800 &&               \
+      !defined(va_copy)
 # define va_copy(dst, src) ((dst) = (src))
 #endif
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_procmaps.h b/libsanitizer/sanitizer_common/sanitizer_procmaps.h
index d140c47fda9..7477abf30b6 100644
--- a/libsanitizer/sanitizer_common/sanitizer_procmaps.h
+++ b/libsanitizer/sanitizer_common/sanitizer_procmaps.h
@@ -24,6 +24,9 @@ struct ProcSelfMapsBuff {
   uptr mmaped_size;
   uptr len;
 };
+
+// Reads process memory map in an OS-specific way.
+void ReadProcMaps(ProcSelfMapsBuff *proc_maps);
 #endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
 
 class MemoryMappingLayout {
@@ -55,7 +58,7 @@ class MemoryMappingLayout {
   // platform-specific files.
 # if SANITIZER_FREEBSD || SANITIZER_LINUX
   ProcSelfMapsBuff proc_self_maps_;
-  char *current_;
+  const char *current_;
 
   // Static mappings cache.
   static ProcSelfMapsBuff cached_proc_self_maps_;
@@ -84,6 +87,11 @@ void GetMemoryProfile(fill_profile_f cb, uptr *stats, uptr stats_size);
 // Returns code range for the specified module.
 bool GetCodeRangeForFile(const char *module, uptr *start, uptr *end);
 
+bool IsDecimal(char c);
+uptr ParseDecimal(const char **p);
+bool IsHex(char c);
+uptr ParseHex(const char **p);
+
 }  // namespace __sanitizer
 
 #endif  // SANITIZER_PROCMAPS_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_procmaps_common.cc b/libsanitizer/sanitizer_common/sanitizer_procmaps_common.cc
new file mode 100644
index 00000000000..ca9900fe9ee
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_procmaps_common.cc
@@ -0,0 +1,176 @@
+//===-- sanitizer_procmaps_common.cc --------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Information about the process mappings (common parts).
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_platform.h"
+#if SANITIZER_FREEBSD || SANITIZER_LINUX
+#include "sanitizer_common.h"
+#include "sanitizer_placement_new.h"
+#include "sanitizer_procmaps.h"
+
+namespace __sanitizer {
+
+// Linker initialized.
+ProcSelfMapsBuff MemoryMappingLayout::cached_proc_self_maps_;
+StaticSpinMutex MemoryMappingLayout::cache_lock_;  // Linker initialized.
+
+static int TranslateDigit(char c) {
+  if (c >= '0' && c <= '9')
+    return c - '0';
+  if (c >= 'a' && c <= 'f')
+    return c - 'a' + 10;
+  if (c >= 'A' && c <= 'F')
+    return c - 'A' + 10;
+  return -1;
+}
+
+// Parse a number and promote 'p' up to the first non-digit character.
+static uptr ParseNumber(const char **p, int base) {
+  uptr n = 0;
+  int d;
+  CHECK(base >= 2 && base <= 16);
+  while ((d = TranslateDigit(**p)) >= 0 && d < base) {
+    n = n * base + d;
+    (*p)++;
+  }
+  return n;
+}
+
+bool IsDecimal(char c) {
+  int d = TranslateDigit(c);
+  return d >= 0 && d < 10;
+}
+
+uptr ParseDecimal(const char **p) {
+  return ParseNumber(p, 10);
+}
+
+bool IsHex(char c) {
+  int d = TranslateDigit(c);
+  return d >= 0 && d < 16;
+}
+
+uptr ParseHex(const char **p) {
+  return ParseNumber(p, 16);
+}
+
+MemoryMappingLayout::MemoryMappingLayout(bool cache_enabled) {
+  ReadProcMaps(&proc_self_maps_);
+  if (cache_enabled) {
+    if (proc_self_maps_.mmaped_size == 0) {
+      LoadFromCache();
+      CHECK_GT(proc_self_maps_.len, 0);
+    }
+  } else {
+    CHECK_GT(proc_self_maps_.mmaped_size, 0);
+  }
+  Reset();
+  // FIXME: in the future we may want to cache the mappings on demand only.
+  if (cache_enabled)
+    CacheMemoryMappings();
+}
+
+MemoryMappingLayout::~MemoryMappingLayout() {
+  // Only unmap the buffer if it is different from the cached one. Otherwise
+  // it will be unmapped when the cache is refreshed.
+  if (proc_self_maps_.data != cached_proc_self_maps_.data) {
+    UnmapOrDie(proc_self_maps_.data, proc_self_maps_.mmaped_size);
+  }
+}
+
+void MemoryMappingLayout::Reset() {
+  current_ = proc_self_maps_.data;
+}
+
+// static
+void MemoryMappingLayout::CacheMemoryMappings() {
+  SpinMutexLock l(&cache_lock_);
+  // Don't invalidate the cache if the mappings are unavailable.
+  ProcSelfMapsBuff old_proc_self_maps;
+  old_proc_self_maps = cached_proc_self_maps_;
+  ReadProcMaps(&cached_proc_self_maps_);
+  if (cached_proc_self_maps_.mmaped_size == 0) {
+    cached_proc_self_maps_ = old_proc_self_maps;
+  } else {
+    if (old_proc_self_maps.mmaped_size) {
+      UnmapOrDie(old_proc_self_maps.data,
+                 old_proc_self_maps.mmaped_size);
+    }
+  }
+}
+
+void MemoryMappingLayout::LoadFromCache() {
+  SpinMutexLock l(&cache_lock_);
+  if (cached_proc_self_maps_.data) {
+    proc_self_maps_ = cached_proc_self_maps_;
+  }
+}
+
+uptr MemoryMappingLayout::DumpListOfModules(LoadedModule *modules,
+                                            uptr max_modules,
+                                            string_predicate_t filter) {
+  Reset();
+  uptr cur_beg, cur_end, cur_offset, prot;
+  InternalScopedBuffer<char> module_name(kMaxPathLength);
+  uptr n_modules = 0;
+  for (uptr i = 0; n_modules < max_modules &&
+                       Next(&cur_beg, &cur_end, &cur_offset, module_name.data(),
+                            module_name.size(), &prot);
+       i++) {
+    const char *cur_name = module_name.data();
+    if (cur_name[0] == '\0')
+      continue;
+    if (filter && !filter(cur_name))
+      continue;
+    void *mem = &modules[n_modules];
+    // Don't subtract 'cur_beg' from the first entry:
+    // * If a binary is compiled w/o -pie, then the first entry in
+    //   process maps is likely the binary itself (all dynamic libs
+    //   are mapped higher in address space). For such a binary,
+    //   instruction offset in binary coincides with the actual
+    //   instruction address in virtual memory (as code section
+    //   is mapped to a fixed memory range).
+    // * If a binary is compiled with -pie, all the modules are
+    //   mapped high at address space (in particular, higher than
+    //   shadow memory of the tool), so the module can't be the
+    //   first entry.
+    uptr base_address = (i ? cur_beg : 0) - cur_offset;
+    LoadedModule *cur_module = new(mem) LoadedModule(cur_name, base_address);
+    cur_module->addAddressRange(cur_beg, cur_end, prot & kProtectionExecute);
+    n_modules++;
+  }
+  return n_modules;
+}
+
+void GetMemoryProfile(fill_profile_f cb, uptr *stats, uptr stats_size) {
+  char *smaps = 0;
+  uptr smaps_cap = 0;
+  uptr smaps_len = ReadFileToBuffer("/proc/self/smaps",
+      &smaps, &smaps_cap, 64<<20);
+  uptr start = 0;
+  bool file = false;
+  const char *pos = smaps;
+  while (pos < smaps + smaps_len) {
+    if (IsHex(pos[0])) {
+      start = ParseHex(&pos);
+      for (; *pos != '/' && *pos > '\n'; pos++) {}
+      file = *pos == '/';
+    } else if (internal_strncmp(pos, "Rss:", 4) == 0) {
+      while (!IsDecimal(*pos)) pos++;
+      uptr rss = ParseDecimal(&pos) * 1024;
+      cb(start, rss, file, stats, stats_size);
+    }
+    while (*pos++ != '\n') {}
+  }
+  UnmapOrDie(smaps, smaps_cap);
+}
+
+}  // namespace __sanitizer
+
+#endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
diff --git a/libsanitizer/sanitizer_common/sanitizer_procmaps_freebsd.cc b/libsanitizer/sanitizer_common/sanitizer_procmaps_freebsd.cc
new file mode 100644
index 00000000000..fbc55203ab4
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_procmaps_freebsd.cc
@@ -0,0 +1,86 @@
+//===-- sanitizer_procmaps_freebsd.cc -------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Information about the process mappings (FreeBSD-specific parts).
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_platform.h"
+#if SANITIZER_FREEBSD
+#include "sanitizer_common.h"
+#include "sanitizer_freebsd.h"
+#include "sanitizer_procmaps.h"
+
+#include <unistd.h>
+#include <sys/sysctl.h>
+#include <sys/user.h>
+
+// Fix 'kinfo_vmentry' definition on FreeBSD prior v9.2 in 32-bit mode.
+#if SANITIZER_FREEBSD && (SANITIZER_WORDSIZE == 32)
+# include <osreldate.h>
+# if __FreeBSD_version <= 902001  // v9.2
+#  define kinfo_vmentry xkinfo_vmentry
+# endif
+#endif
+
+namespace __sanitizer {
+
+void ReadProcMaps(ProcSelfMapsBuff *proc_maps) {
+  const int Mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_VMMAP, getpid() };
+  size_t Size = 0;
+  int Err = sysctl(Mib, 4, NULL, &Size, NULL, 0);
+  CHECK_EQ(Err, 0);
+  CHECK_GT(Size, 0);
+
+  size_t MmapedSize = Size * 4 / 3;
+  void *VmMap = MmapOrDie(MmapedSize, "ReadProcMaps()");
+  Size = MmapedSize;
+  Err = sysctl(Mib, 4, VmMap, &Size, NULL, 0);
+  CHECK_EQ(Err, 0);
+
+  proc_maps->data = (char*)VmMap;
+  proc_maps->mmaped_size = MmapedSize;
+  proc_maps->len = Size;
+}
+
+bool MemoryMappingLayout::Next(uptr *start, uptr *end, uptr *offset,
+                               char filename[], uptr filename_size,
+                               uptr *protection) {
+  char *last = proc_self_maps_.data + proc_self_maps_.len;
+  if (current_ >= last) return false;
+  uptr dummy;
+  if (!start) start = &dummy;
+  if (!end) end = &dummy;
+  if (!offset) offset = &dummy;
+  if (!protection) protection = &dummy;
+  struct kinfo_vmentry *VmEntry = (struct kinfo_vmentry*)current_;
+
+  *start = (uptr)VmEntry->kve_start;
+  *end = (uptr)VmEntry->kve_end;
+  *offset = (uptr)VmEntry->kve_offset;
+
+  *protection = 0;
+  if ((VmEntry->kve_protection & KVME_PROT_READ) != 0)
+    *protection |= kProtectionRead;
+  if ((VmEntry->kve_protection & KVME_PROT_WRITE) != 0)
+    *protection |= kProtectionWrite;
+  if ((VmEntry->kve_protection & KVME_PROT_EXEC) != 0)
+    *protection |= kProtectionExecute;
+
+  if (filename != NULL && filename_size > 0) {
+    internal_snprintf(filename,
+                      Min(filename_size, (uptr)PATH_MAX),
+                      "%s", VmEntry->kve_path);
+  }
+
+  current_ += VmEntry->kve_structsize;
+
+  return true;
+}
+
+}  // namespace __sanitizer
+
+#endif  // SANITIZER_FREEBSD
diff --git a/libsanitizer/sanitizer_common/sanitizer_procmaps_linux.cc b/libsanitizer/sanitizer_common/sanitizer_procmaps_linux.cc
index 20a074a799f..43babf89319 100644
--- a/libsanitizer/sanitizer_common/sanitizer_procmaps_linux.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_procmaps_linux.cc
@@ -9,151 +9,20 @@
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_platform.h"
-#if SANITIZER_FREEBSD || SANITIZER_LINUX
+#if SANITIZER_LINUX
 #include "sanitizer_common.h"
-#include "sanitizer_placement_new.h"
 #include "sanitizer_procmaps.h"
 
-#if SANITIZER_FREEBSD
-#include <unistd.h>
-#include <sys/sysctl.h>
-#include <sys/user.h>
-#endif
-
 namespace __sanitizer {
 
-// Linker initialized.
-ProcSelfMapsBuff MemoryMappingLayout::cached_proc_self_maps_;
-StaticSpinMutex MemoryMappingLayout::cache_lock_;  // Linker initialized.
-
-static void ReadProcMaps(ProcSelfMapsBuff *proc_maps) {
-#if SANITIZER_FREEBSD
-  const int Mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_VMMAP, getpid() };
-  size_t Size = 0;
-  int Err = sysctl(Mib, 4, NULL, &Size, NULL, 0);
-  CHECK_EQ(Err, 0);
-  CHECK_GT(Size, 0);
-
-  size_t MmapedSize = Size * 4 / 3;
-  void *VmMap = MmapOrDie(MmapedSize, "ReadProcMaps()");
-  Size = MmapedSize;
-  Err = sysctl(Mib, 4, VmMap, &Size, NULL, 0);
-  CHECK_EQ(Err, 0);
-
-  proc_maps->data = (char*)VmMap;
-  proc_maps->mmaped_size = MmapedSize;
-  proc_maps->len = Size;
-#else
+void ReadProcMaps(ProcSelfMapsBuff *proc_maps) {
   proc_maps->len = ReadFileToBuffer("/proc/self/maps", &proc_maps->data,
                                     &proc_maps->mmaped_size, 1 << 26);
-#endif
-}
-
-MemoryMappingLayout::MemoryMappingLayout(bool cache_enabled) {
-  ReadProcMaps(&proc_self_maps_);
-  if (cache_enabled) {
-    if (proc_self_maps_.mmaped_size == 0) {
-      LoadFromCache();
-      CHECK_GT(proc_self_maps_.len, 0);
-    }
-  } else {
-    CHECK_GT(proc_self_maps_.mmaped_size, 0);
-  }
-  Reset();
-  // FIXME: in the future we may want to cache the mappings on demand only.
-  if (cache_enabled)
-    CacheMemoryMappings();
-}
-
-MemoryMappingLayout::~MemoryMappingLayout() {
-  // Only unmap the buffer if it is different from the cached one. Otherwise
-  // it will be unmapped when the cache is refreshed.
-  if (proc_self_maps_.data != cached_proc_self_maps_.data) {
-    UnmapOrDie(proc_self_maps_.data, proc_self_maps_.mmaped_size);
-  }
-}
-
-void MemoryMappingLayout::Reset() {
-  current_ = proc_self_maps_.data;
-}
-
-// static
-void MemoryMappingLayout::CacheMemoryMappings() {
-  SpinMutexLock l(&cache_lock_);
-  // Don't invalidate the cache if the mappings are unavailable.
-  ProcSelfMapsBuff old_proc_self_maps;
-  old_proc_self_maps = cached_proc_self_maps_;
-  ReadProcMaps(&cached_proc_self_maps_);
-  if (cached_proc_self_maps_.mmaped_size == 0) {
-    cached_proc_self_maps_ = old_proc_self_maps;
-  } else {
-    if (old_proc_self_maps.mmaped_size) {
-      UnmapOrDie(old_proc_self_maps.data,
-                 old_proc_self_maps.mmaped_size);
-    }
-  }
-}
-
-void MemoryMappingLayout::LoadFromCache() {
-  SpinMutexLock l(&cache_lock_);
-  if (cached_proc_self_maps_.data) {
-    proc_self_maps_ = cached_proc_self_maps_;
-  }
-}
-
-#if !SANITIZER_FREEBSD
-// Parse a hex value in str and update str.
-static uptr ParseHex(char **str) {
-  uptr x = 0;
-  char *s;
-  for (s = *str; ; s++) {
-    char c = *s;
-    uptr v = 0;
-    if (c >= '0' && c <= '9')
-      v = c - '0';
-    else if (c >= 'a' && c <= 'f')
-      v = c - 'a' + 10;
-    else if (c >= 'A' && c <= 'F')
-      v = c - 'A' + 10;
-    else
-      break;
-    x = x * 16 + v;
-  }
-  *str = s;
-  return x;
 }
 
 static bool IsOneOf(char c, char c1, char c2) {
   return c == c1 || c == c2;
 }
-#endif
-
-static bool IsDecimal(char c) {
-  return c >= '0' && c <= '9';
-}
-
-static bool IsHex(char c) {
-  return (c >= '0' && c <= '9')
-      || (c >= 'a' && c <= 'f');
-}
-
-static uptr ReadHex(const char *p) {
-  uptr v = 0;
-  for (; IsHex(p[0]); p++) {
-    if (p[0] >= '0' && p[0] <= '9')
-      v = v * 16 + p[0] - '0';
-    else
-      v = v * 16 + p[0] - 'a' + 10;
-  }
-  return v;
-}
-
-static uptr ReadDecimal(const char *p) {
-  uptr v = 0;
-  for (; IsDecimal(p[0]); p++)
-    v = v * 10 + p[0] - '0';
-  return v;
-}
 
 bool MemoryMappingLayout::Next(uptr *start, uptr *end, uptr *offset,
                                char filename[], uptr filename_size,
@@ -165,29 +34,6 @@ bool MemoryMappingLayout::Next(uptr *start, uptr *end, uptr *offset,
   if (!end) end = &dummy;
   if (!offset) offset = &dummy;
   if (!protection) protection = &dummy;
-#if SANITIZER_FREEBSD
-  struct kinfo_vmentry *VmEntry = (struct kinfo_vmentry*)current_;
-
-  *start = (uptr)VmEntry->kve_start;
-  *end = (uptr)VmEntry->kve_end;
-  *offset = (uptr)VmEntry->kve_offset;
-
-  *protection = 0;
-  if ((VmEntry->kve_protection & KVME_PROT_READ) != 0)
-    *protection |= kProtectionRead;
-  if ((VmEntry->kve_protection & KVME_PROT_WRITE) != 0)
-    *protection |= kProtectionWrite;
-  if ((VmEntry->kve_protection & KVME_PROT_EXEC) != 0)
-    *protection |= kProtectionExecute;
-
-  if (filename != NULL && filename_size > 0) {
-    internal_snprintf(filename,
-                      Min(filename_size, (uptr)PATH_MAX),
-                      "%s", VmEntry->kve_path);
-  }
-
-  current_ += VmEntry->kve_structsize;
-#else  // !SANITIZER_FREEBSD
   char *next_line = (char*)internal_memchr(current_, '\n', last - current_);
   if (next_line == 0)
     next_line = last;
@@ -234,69 +80,9 @@ bool MemoryMappingLayout::Next(uptr *start, uptr *end, uptr *offset,
   if (filename && i < filename_size)
     filename[i] = 0;
   current_ = next_line + 1;
-#endif  // !SANITIZER_FREEBSD
   return true;
 }
 
-uptr MemoryMappingLayout::DumpListOfModules(LoadedModule *modules,
-                                            uptr max_modules,
-                                            string_predicate_t filter) {
-  Reset();
-  uptr cur_beg, cur_end, cur_offset;
-  InternalScopedBuffer<char> module_name(kMaxPathLength);
-  uptr n_modules = 0;
-  for (uptr i = 0; n_modules < max_modules &&
-                       Next(&cur_beg, &cur_end, &cur_offset, module_name.data(),
-                            module_name.size(), 0);
-       i++) {
-    const char *cur_name = module_name.data();
-    if (cur_name[0] == '\0')
-      continue;
-    if (filter && !filter(cur_name))
-      continue;
-    void *mem = &modules[n_modules];
-    // Don't subtract 'cur_beg' from the first entry:
-    // * If a binary is compiled w/o -pie, then the first entry in
-    //   process maps is likely the binary itself (all dynamic libs
-    //   are mapped higher in address space). For such a binary,
-    //   instruction offset in binary coincides with the actual
-    //   instruction address in virtual memory (as code section
-    //   is mapped to a fixed memory range).
-    // * If a binary is compiled with -pie, all the modules are
-    //   mapped high at address space (in particular, higher than
-    //   shadow memory of the tool), so the module can't be the
-    //   first entry.
-    uptr base_address = (i ? cur_beg : 0) - cur_offset;
-    LoadedModule *cur_module = new(mem) LoadedModule(cur_name, base_address);
-    cur_module->addAddressRange(cur_beg, cur_end);
-    n_modules++;
-  }
-  return n_modules;
-}
-
-void GetMemoryProfile(fill_profile_f cb, uptr *stats, uptr stats_size) {
-  char *smaps = 0;
-  uptr smaps_cap = 0;
-  uptr smaps_len = ReadFileToBuffer("/proc/self/smaps",
-      &smaps, &smaps_cap, 64<<20);
-  uptr start = 0;
-  bool file = false;
-  const char *pos = smaps;
-  while (pos < smaps + smaps_len) {
-    if (IsHex(pos[0])) {
-      start = ReadHex(pos);
-      for (; *pos != '/' && *pos > '\n'; pos++) {}
-      file = *pos == '/';
-    } else if (internal_strncmp(pos, "Rss:", 4) == 0) {
-      for (; *pos < '0' || *pos > '9'; pos++) {}
-      uptr rss = ReadDecimal(pos) * 1024;
-      cb(start, rss, file, stats, stats_size);
-    }
-    while (*pos++ != '\n') {}
-  }
-  UnmapOrDie(smaps, smaps_cap);
-}
-
 }  // namespace __sanitizer
 
-#endif  // SANITIZER_FREEBSD || SANITIZER_LINUX
+#endif  // SANITIZER_LINUX
diff --git a/libsanitizer/sanitizer_common/sanitizer_procmaps_mac.cc b/libsanitizer/sanitizer_common/sanitizer_procmaps_mac.cc
index c6853068346..81874c21b1f 100644
--- a/libsanitizer/sanitizer_common/sanitizer_procmaps_mac.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_procmaps_mac.cc
@@ -73,14 +73,16 @@ template<u32 kLCSegment, typename SegmentCommand>
 bool MemoryMappingLayout::NextSegmentLoad(
     uptr *start, uptr *end, uptr *offset,
     char filename[], uptr filename_size, uptr *protection) {
-  if (protection)
-    UNIMPLEMENTED();
   const char* lc = current_load_cmd_addr_;
   current_load_cmd_addr_ += ((const load_command *)lc)->cmdsize;
   if (((const load_command *)lc)->cmd == kLCSegment) {
     const sptr dlloff = _dyld_get_image_vmaddr_slide(current_image_);
     const SegmentCommand* sc = (const SegmentCommand *)lc;
     if (start) *start = sc->vmaddr + dlloff;
+    if (protection) {
+      // Return the initial protection.
+      *protection = sc->initprot;
+    }
     if (end) *end = sc->vmaddr + sc->vmsize + dlloff;
     if (offset) {
       if (current_filetype_ == /*MH_EXECUTE*/ 0x2) {
@@ -155,12 +157,12 @@ uptr MemoryMappingLayout::DumpListOfModules(LoadedModule *modules,
                                             uptr max_modules,
                                             string_predicate_t filter) {
   Reset();
-  uptr cur_beg, cur_end;
+  uptr cur_beg, cur_end, prot;
   InternalScopedBuffer<char> module_name(kMaxPathLength);
   uptr n_modules = 0;
   for (uptr i = 0; n_modules < max_modules &&
                        Next(&cur_beg, &cur_end, 0, module_name.data(),
-                            module_name.size(), 0);
+                            module_name.size(), &prot);
        i++) {
     const char *cur_name = module_name.data();
     if (cur_name[0] == '\0')
@@ -176,7 +178,7 @@ uptr MemoryMappingLayout::DumpListOfModules(LoadedModule *modules,
       cur_module = new(mem) LoadedModule(cur_name, cur_beg);
       n_modules++;
     }
-    cur_module->addAddressRange(cur_beg, cur_end);
+    cur_module->addAddressRange(cur_beg, cur_end, prot & kProtectionExecute);
   }
   return n_modules;
 }
diff --git a/libsanitizer/sanitizer_common/sanitizer_report_decorator.h b/libsanitizer/sanitizer_common/sanitizer_report_decorator.h
index c8510585111..e9be29fb3d5 100644
--- a/libsanitizer/sanitizer_common/sanitizer_report_decorator.h
+++ b/libsanitizer/sanitizer_common/sanitizer_report_decorator.h
@@ -18,12 +18,16 @@
 #include "sanitizer_common.h"
 
 namespace __sanitizer {
-class AnsiColorDecorator {
+class SanitizerCommonDecorator {
   // FIXME: This is not portable. It assumes the special strings are printed to
   // stdout, which is not the case on Windows (see SetConsoleTextAttribute()).
  public:
-  explicit AnsiColorDecorator(bool use_ansi_colors) : ansi_(use_ansi_colors) { }
+  SanitizerCommonDecorator() : ansi_(ColorizeReports()) {}
   const char *Bold()    const { return ansi_ ? "\033[1m" : ""; }
+  const char *Default() const { return ansi_ ? "\033[1m\033[0m"  : ""; }
+  const char *Warning()    { return Red(); }
+  const char *EndWarning() { return Default(); }
+ protected:
   const char *Black()   const { return ansi_ ? "\033[1m\033[30m" : ""; }
   const char *Red()     const { return ansi_ ? "\033[1m\033[31m" : ""; }
   const char *Green()   const { return ansi_ ? "\033[1m\033[32m" : ""; }
@@ -32,19 +36,10 @@ class AnsiColorDecorator {
   const char *Magenta() const { return ansi_ ? "\033[1m\033[35m" : ""; }
   const char *Cyan()    const { return ansi_ ? "\033[1m\033[36m" : ""; }
   const char *White()   const { return ansi_ ? "\033[1m\033[37m" : ""; }
-  const char *Default() const { return ansi_ ? "\033[1m\033[0m"  : ""; }
  private:
   bool ansi_;
 };
 
-class SanitizerCommonDecorator: protected AnsiColorDecorator {
- public:
-  SanitizerCommonDecorator()
-      : __sanitizer::AnsiColorDecorator(ColorizeReports()) { }
-  const char *Warning()    { return Red(); }
-  const char *EndWarning() { return Default(); }
-};
-
 }  // namespace __sanitizer
 
 #endif  // SANITIZER_REPORT_DECORATOR_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_stackdepot.cc b/libsanitizer/sanitizer_common/sanitizer_stackdepot.cc
index e8d9f01e7f9..e1915cb808f 100644
--- a/libsanitizer/sanitizer_common/sanitizer_stackdepot.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_stackdepot.cc
@@ -10,193 +10,128 @@
 //===----------------------------------------------------------------------===//
 
 #include "sanitizer_stackdepot.h"
+
 #include "sanitizer_common.h"
-#include "sanitizer_internal_defs.h"
-#include "sanitizer_mutex.h"
-#include "sanitizer_atomic.h"
+#include "sanitizer_stackdepotbase.h"
 
 namespace __sanitizer {
 
-const int kTabSize = 1024 * 1024;  // Hash table size.
-const int kPartBits = 8;
-const int kPartShift = sizeof(u32) * 8 - kPartBits - 1;
-const int kPartCount = 1 << kPartBits;  // Number of subparts in the table.
-const int kPartSize = kTabSize / kPartCount;
-const int kMaxId = 1 << kPartShift;
+struct StackDepotDesc {
+  const uptr *stack;
+  uptr size;
+  u32 hash() const {
+    // murmur2
+    const u32 m = 0x5bd1e995;
+    const u32 seed = 0x9747b28c;
+    const u32 r = 24;
+    u32 h = seed ^ (size * sizeof(uptr));
+    for (uptr i = 0; i < size; i++) {
+      u32 k = stack[i];
+      k *= m;
+      k ^= k >> r;
+      k *= m;
+      h *= m;
+      h ^= k;
+    }
+    h ^= h >> 13;
+    h *= m;
+    h ^= h >> 15;
+    return h;
+  }
+  bool is_valid() { return size > 0 && stack; }
+};
 
-struct StackDesc {
-  StackDesc *link;
+struct StackDepotNode {
+  StackDepotNode *link;
   u32 id;
-  u32 hash;
+  atomic_uint32_t hash_and_use_count; // hash_bits : 12; use_count : 20;
   uptr size;
   uptr stack[1];  // [size]
-};
 
-static struct {
-  StaticSpinMutex mtx;  // Protects alloc of new blocks for region allocator.
-  atomic_uintptr_t region_pos;  // Region allocator for StackDesc's.
-  atomic_uintptr_t region_end;
-  atomic_uintptr_t tab[kTabSize];  // Hash table of StackDesc's.
-  atomic_uint32_t seq[kPartCount];  // Unique id generators.
-} depot;
+  static const u32 kTabSizeLog = 20;
+  // Lower kTabSizeLog bits are equal for all items in one bucket.
+  // We use these bits to store the per-stack use counter.
+  static const u32 kUseCountBits = kTabSizeLog;
+  static const u32 kMaxUseCount = 1 << kUseCountBits;
+  static const u32 kUseCountMask = (1 << kUseCountBits) - 1;
+  static const u32 kHashMask = ~kUseCountMask;
+
+  typedef StackDepotDesc args_type;
+  bool eq(u32 hash, const args_type &args) const {
+    u32 hash_bits =
+        atomic_load(&hash_and_use_count, memory_order_relaxed) & kHashMask;
+    if ((hash & kHashMask) != hash_bits || args.size != size) return false;
+    uptr i = 0;
+    for (; i < size; i++) {
+      if (stack[i] != args.stack[i]) return false;
+    }
+    return true;
+  }
+  static uptr storage_size(const args_type &args) {
+    return sizeof(StackDepotNode) + (args.size - 1) * sizeof(uptr);
+  }
+  void store(const args_type &args, u32 hash) {
+    atomic_store(&hash_and_use_count, hash & kHashMask, memory_order_relaxed);
+    size = args.size;
+    internal_memcpy(stack, args.stack, size * sizeof(uptr));
+  }
+  args_type load() const {
+    args_type ret = {&stack[0], size};
+    return ret;
+  }
+  StackDepotHandle get_handle() { return StackDepotHandle(this); }
 
-static StackDepotStats stats;
+  typedef StackDepotHandle handle_type;
+};
 
-StackDepotStats *StackDepotGetStats() {
-  return &stats;
-}
+COMPILER_CHECK(StackDepotNode::kMaxUseCount == (u32)kStackDepotMaxUseCount);
 
-static u32 hash(const uptr *stack, uptr size) {
-  // murmur2
-  const u32 m = 0x5bd1e995;
-  const u32 seed = 0x9747b28c;
-  const u32 r = 24;
-  u32 h = seed ^ (size * sizeof(uptr));
-  for (uptr i = 0; i < size; i++) {
-    u32 k = stack[i];
-    k *= m;
-    k ^= k >> r;
-    k *= m;
-    h *= m;
-    h ^= k;
-  }
-  h ^= h >> 13;
-  h *= m;
-  h ^= h >> 15;
-  return h;
+u32 StackDepotHandle::id() { return node_->id; }
+int StackDepotHandle::use_count() {
+  return atomic_load(&node_->hash_and_use_count, memory_order_relaxed) &
+         StackDepotNode::kUseCountMask;
 }
-
-static StackDesc *tryallocDesc(uptr memsz) {
-  // Optimisic lock-free allocation, essentially try to bump the region ptr.
-  for (;;) {
-    uptr cmp = atomic_load(&depot.region_pos, memory_order_acquire);
-    uptr end = atomic_load(&depot.region_end, memory_order_acquire);
-    if (cmp == 0 || cmp + memsz > end)
-      return 0;
-    if (atomic_compare_exchange_weak(
-        &depot.region_pos, &cmp, cmp + memsz,
-        memory_order_acquire))
-      return (StackDesc*)cmp;
-  }
+void StackDepotHandle::inc_use_count_unsafe() {
+  u32 prev =
+      atomic_fetch_add(&node_->hash_and_use_count, 1, memory_order_relaxed) &
+      StackDepotNode::kUseCountMask;
+  CHECK_LT(prev + 1, StackDepotNode::kMaxUseCount);
 }
+uptr StackDepotHandle::size() { return node_->size; }
+uptr *StackDepotHandle::stack() { return &node_->stack[0]; }
 
-static StackDesc *allocDesc(uptr size) {
-  // First, try to allocate optimisitically.
-  uptr memsz = sizeof(StackDesc) + (size - 1) * sizeof(uptr);
-  StackDesc *s = tryallocDesc(memsz);
-  if (s)
-    return s;
-  // If failed, lock, retry and alloc new superblock.
-  SpinMutexLock l(&depot.mtx);
-  for (;;) {
-    s = tryallocDesc(memsz);
-    if (s)
-      return s;
-    atomic_store(&depot.region_pos, 0, memory_order_relaxed);
-    uptr allocsz = 64 * 1024;
-    if (allocsz < memsz)
-      allocsz = memsz;
-    uptr mem = (uptr)MmapOrDie(allocsz, "stack depot");
-    stats.mapped += allocsz;
-    atomic_store(&depot.region_end, mem + allocsz, memory_order_release);
-    atomic_store(&depot.region_pos, mem, memory_order_release);
-  }
+// FIXME(dvyukov): this single reserved bit is used in TSan.
+typedef StackDepotBase<StackDepotNode, 1, StackDepotNode::kTabSizeLog>
+    StackDepot;
+static StackDepot theDepot;
+
+StackDepotStats *StackDepotGetStats() {
+  return theDepot.GetStats();
 }
 
-static u32 find(StackDesc *s, const uptr *stack, uptr size, u32 hash) {
-  // Searches linked list s for the stack, returns its id.
-  for (; s; s = s->link) {
-    if (s->hash == hash && s->size == size) {
-      uptr i = 0;
-      for (; i < size; i++) {
-        if (stack[i] != s->stack[i])
-          break;
-      }
-      if (i == size)
-        return s->id;
-    }
-  }
-  return 0;
+u32 StackDepotPut(const uptr *stack, uptr size) {
+  StackDepotDesc desc = {stack, size};
+  StackDepotHandle h = theDepot.Put(desc);
+  return h.valid() ? h.id() : 0;
 }
 
-static StackDesc *lock(atomic_uintptr_t *p) {
-  // Uses the pointer lsb as mutex.
-  for (int i = 0;; i++) {
-    uptr cmp = atomic_load(p, memory_order_relaxed);
-    if ((cmp & 1) == 0
-        && atomic_compare_exchange_weak(p, &cmp, cmp | 1,
-                                        memory_order_acquire))
-      return (StackDesc*)cmp;
-    if (i < 10)
-      proc_yield(10);
-    else
-      internal_sched_yield();
-  }
+StackDepotHandle StackDepotPut_WithHandle(const uptr *stack, uptr size) {
+  StackDepotDesc desc = {stack, size};
+  return theDepot.Put(desc);
 }
 
-static void unlock(atomic_uintptr_t *p, StackDesc *s) {
-  DCHECK_EQ((uptr)s & 1, 0);
-  atomic_store(p, (uptr)s, memory_order_release);
+const uptr *StackDepotGet(u32 id, uptr *size) {
+  StackDepotDesc desc = theDepot.Get(id);
+  *size = desc.size;
+  return desc.stack;
 }
 
-u32 StackDepotPut(const uptr *stack, uptr size) {
-  if (stack == 0 || size == 0)
-    return 0;
-  uptr h = hash(stack, size);
-  atomic_uintptr_t *p = &depot.tab[h % kTabSize];
-  uptr v = atomic_load(p, memory_order_consume);
-  StackDesc *s = (StackDesc*)(v & ~1);
-  // First, try to find the existing stack.
-  u32 id = find(s, stack, size, h);
-  if (id)
-    return id;
-  // If failed, lock, retry and insert new.
-  StackDesc *s2 = lock(p);
-  if (s2 != s) {
-    id = find(s2, stack, size, h);
-    if (id) {
-      unlock(p, s2);
-      return id;
-    }
-  }
-  uptr part = (h % kTabSize) / kPartSize;
-  id = atomic_fetch_add(&depot.seq[part], 1, memory_order_relaxed) + 1;
-  stats.n_uniq_ids++;
-  CHECK_LT(id, kMaxId);
-  id |= part << kPartShift;
-  CHECK_NE(id, 0);
-  CHECK_EQ(id & (1u << 31), 0);
-  s = allocDesc(size);
-  s->id = id;
-  s->hash = h;
-  s->size = size;
-  internal_memcpy(s->stack, stack, size * sizeof(uptr));
-  s->link = s2;
-  unlock(p, s);
-  return id;
+void StackDepotLockAll() {
+  theDepot.LockAll();
 }
 
-const uptr *StackDepotGet(u32 id, uptr *size) {
-  if (id == 0)
-    return 0;
-  CHECK_EQ(id & (1u << 31), 0);
-  // High kPartBits contain part id, so we need to scan at most kPartSize lists.
-  uptr part = id >> kPartShift;
-  for (int i = 0; i != kPartSize; i++) {
-    uptr idx = part * kPartSize + i;
-    CHECK_LT(idx, kTabSize);
-    atomic_uintptr_t *p = &depot.tab[idx];
-    uptr v = atomic_load(p, memory_order_consume);
-    StackDesc *s = (StackDesc*)(v & ~1);
-    for (; s; s = s->link) {
-      if (s->id == id) {
-        *size = s->size;
-        return s->stack;
-      }
-    }
-  }
-  *size = 0;
-  return 0;
+void StackDepotUnlockAll() {
+  theDepot.UnlockAll();
 }
 
 bool StackDepotReverseMap::IdDescPair::IdComparator(
@@ -207,10 +142,10 @@ bool StackDepotReverseMap::IdDescPair::IdComparator(
 
 StackDepotReverseMap::StackDepotReverseMap()
     : map_(StackDepotGetStats()->n_uniq_ids + 100) {
-  for (int idx = 0; idx < kTabSize; idx++) {
-    atomic_uintptr_t *p = &depot.tab[idx];
+  for (int idx = 0; idx < StackDepot::kTabSize; idx++) {
+    atomic_uintptr_t *p = &theDepot.tab[idx];
     uptr v = atomic_load(p, memory_order_consume);
-    StackDesc *s = (StackDesc*)(v & ~1);
+    StackDepotNode *s = (StackDepotNode*)(v & ~1);
     for (; s; s = s->link) {
       IdDescPair pair = {s->id, s};
       map_.push_back(pair);
@@ -228,7 +163,7 @@ const uptr *StackDepotReverseMap::Get(u32 id, uptr *size) {
     *size = 0;
     return 0;
   }
-  StackDesc *desc = map_[idx].desc;
+  StackDepotNode *desc = map_[idx].desc;
   *size = desc->size;
   return desc->stack;
 }
diff --git a/libsanitizer/sanitizer_common/sanitizer_stackdepot.h b/libsanitizer/sanitizer_common/sanitizer_stackdepot.h
index c2c04ef9d64..2b1da4ee14f 100644
--- a/libsanitizer/sanitizer_common/sanitizer_stackdepot.h
+++ b/libsanitizer/sanitizer_common/sanitizer_stackdepot.h
@@ -17,20 +17,29 @@
 namespace __sanitizer {
 
 // StackDepot efficiently stores huge amounts of stack traces.
+struct StackDepotNode;
+struct StackDepotHandle {
+  StackDepotNode *node_;
+  StackDepotHandle() : node_(0) {}
+  explicit StackDepotHandle(StackDepotNode *node) : node_(node) {}
+  bool valid() { return node_; }
+  u32 id();
+  int use_count();
+  void inc_use_count_unsafe();
+  uptr size();
+  uptr *stack();
+};
+
+const int kStackDepotMaxUseCount = 1U << 20;
 
-// Maps stack trace to an unique id.
+StackDepotStats *StackDepotGetStats();
 u32 StackDepotPut(const uptr *stack, uptr size);
+StackDepotHandle StackDepotPut_WithHandle(const uptr *stack, uptr size);
 // Retrieves a stored stack trace by the id.
 const uptr *StackDepotGet(u32 id, uptr *size);
 
-struct StackDepotStats {
-  uptr n_uniq_ids;
-  uptr mapped;
-};
-
-StackDepotStats *StackDepotGetStats();
-
-struct StackDesc;
+void StackDepotLockAll();
+void StackDepotUnlockAll();
 
 // Instantiating this class creates a snapshot of StackDepot which can be
 // efficiently queried with StackDepotGet(). You can use it concurrently with
@@ -44,7 +53,7 @@ class StackDepotReverseMap {
  private:
   struct IdDescPair {
     u32 id;
-    StackDesc *desc;
+    StackDepotNode *desc;
 
     static bool IdComparator(const IdDescPair &a, const IdDescPair &b);
   };
@@ -55,6 +64,7 @@ class StackDepotReverseMap {
   StackDepotReverseMap(const StackDepotReverseMap&);
   void operator=(const StackDepotReverseMap&);
 };
+
 }  // namespace __sanitizer
 
 #endif  // SANITIZER_STACKDEPOT_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_stackdepotbase.h b/libsanitizer/sanitizer_common/sanitizer_stackdepotbase.h
new file mode 100644
index 00000000000..b9dedec2621
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_stackdepotbase.h
@@ -0,0 +1,174 @@
+//===-- sanitizer_stackdepotbase.h ------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Implementation of a mapping from arbitrary values to unique 32-bit
+// identifiers.
+//===----------------------------------------------------------------------===//
+#ifndef SANITIZER_STACKDEPOTBASE_H
+#define SANITIZER_STACKDEPOTBASE_H
+
+#include "sanitizer_internal_defs.h"
+#include "sanitizer_mutex.h"
+#include "sanitizer_atomic.h"
+#include "sanitizer_persistent_allocator.h"
+
+namespace __sanitizer {
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+class StackDepotBase {
+ public:
+  typedef typename Node::args_type args_type;
+  typedef typename Node::handle_type handle_type;
+  // Maps stack trace to an unique id.
+  handle_type Put(args_type args, bool *inserted = 0);
+  // Retrieves a stored stack trace by the id.
+  args_type Get(u32 id);
+
+  StackDepotStats *GetStats() { return &stats; }
+
+  void LockAll();
+  void UnlockAll();
+
+ private:
+  static Node *find(Node *s, args_type args, u32 hash);
+  static Node *lock(atomic_uintptr_t *p);
+  static void unlock(atomic_uintptr_t *p, Node *s);
+
+  static const int kTabSize = 1 << kTabSizeLog;  // Hash table size.
+  static const int kPartBits = 8;
+  static const int kPartShift = sizeof(u32) * 8 - kPartBits - kReservedBits;
+  static const int kPartCount =
+      1 << kPartBits;  // Number of subparts in the table.
+  static const int kPartSize = kTabSize / kPartCount;
+  static const int kMaxId = 1 << kPartShift;
+
+  atomic_uintptr_t tab[kTabSize];   // Hash table of Node's.
+  atomic_uint32_t seq[kPartCount];  // Unique id generators.
+
+  StackDepotStats stats;
+
+  friend class StackDepotReverseMap;
+};
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+Node *StackDepotBase<Node, kReservedBits, kTabSizeLog>::find(Node *s,
+                                                             args_type args,
+                                                             u32 hash) {
+  // Searches linked list s for the stack, returns its id.
+  for (; s; s = s->link) {
+    if (s->eq(hash, args)) {
+      return s;
+    }
+  }
+  return 0;
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+Node *StackDepotBase<Node, kReservedBits, kTabSizeLog>::lock(
+    atomic_uintptr_t *p) {
+  // Uses the pointer lsb as mutex.
+  for (int i = 0;; i++) {
+    uptr cmp = atomic_load(p, memory_order_relaxed);
+    if ((cmp & 1) == 0 &&
+        atomic_compare_exchange_weak(p, &cmp, cmp | 1, memory_order_acquire))
+      return (Node *)cmp;
+    if (i < 10)
+      proc_yield(10);
+    else
+      internal_sched_yield();
+  }
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+void StackDepotBase<Node, kReservedBits, kTabSizeLog>::unlock(
+    atomic_uintptr_t *p, Node *s) {
+  DCHECK_EQ((uptr)s & 1, 0);
+  atomic_store(p, (uptr)s, memory_order_release);
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+typename StackDepotBase<Node, kReservedBits, kTabSizeLog>::handle_type
+StackDepotBase<Node, kReservedBits, kTabSizeLog>::Put(args_type args,
+                                                      bool *inserted) {
+  if (inserted) *inserted = false;
+  if (!args.is_valid()) return handle_type();
+  uptr h = args.hash();
+  atomic_uintptr_t *p = &tab[h % kTabSize];
+  uptr v = atomic_load(p, memory_order_consume);
+  Node *s = (Node *)(v & ~1);
+  // First, try to find the existing stack.
+  Node *node = find(s, args, h);
+  if (node) return node->get_handle();
+  // If failed, lock, retry and insert new.
+  Node *s2 = lock(p);
+  if (s2 != s) {
+    node = find(s2, args, h);
+    if (node) {
+      unlock(p, s2);
+      return node->get_handle();
+    }
+  }
+  uptr part = (h % kTabSize) / kPartSize;
+  u32 id = atomic_fetch_add(&seq[part], 1, memory_order_relaxed) + 1;
+  stats.n_uniq_ids++;
+  CHECK_LT(id, kMaxId);
+  id |= part << kPartShift;
+  CHECK_NE(id, 0);
+  CHECK_EQ(id & (((u32)-1) >> kReservedBits), id);
+  uptr memsz = Node::storage_size(args);
+  s = (Node *)PersistentAlloc(memsz);
+  stats.allocated += memsz;
+  s->id = id;
+  s->store(args, h);
+  s->link = s2;
+  unlock(p, s);
+  if (inserted) *inserted = true;
+  return s->get_handle();
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+typename StackDepotBase<Node, kReservedBits, kTabSizeLog>::args_type
+StackDepotBase<Node, kReservedBits, kTabSizeLog>::Get(u32 id) {
+  if (id == 0) {
+    return args_type();
+  }
+  CHECK_EQ(id & (((u32)-1) >> kReservedBits), id);
+  // High kPartBits contain part id, so we need to scan at most kPartSize lists.
+  uptr part = id >> kPartShift;
+  for (int i = 0; i != kPartSize; i++) {
+    uptr idx = part * kPartSize + i;
+    CHECK_LT(idx, kTabSize);
+    atomic_uintptr_t *p = &tab[idx];
+    uptr v = atomic_load(p, memory_order_consume);
+    Node *s = (Node *)(v & ~1);
+    for (; s; s = s->link) {
+      if (s->id == id) {
+        return s->load();
+      }
+    }
+  }
+  return args_type();
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+void StackDepotBase<Node, kReservedBits, kTabSizeLog>::LockAll() {
+  for (int i = 0; i < kTabSize; ++i) {
+    lock(&tab[i]);
+  }
+}
+
+template <class Node, int kReservedBits, int kTabSizeLog>
+void StackDepotBase<Node, kReservedBits, kTabSizeLog>::UnlockAll() {
+  for (int i = 0; i < kTabSize; ++i) {
+    atomic_uintptr_t *p = &tab[i];
+    uptr s = atomic_load(p, memory_order_relaxed);
+    unlock(p, (Node *)(s & ~1UL));
+  }
+}
+
+}  // namespace __sanitizer
+#endif  // SANITIZER_STACKDEPOTBASE_H
diff --git a/libsanitizer/sanitizer_common/sanitizer_stacktrace.cc b/libsanitizer/sanitizer_common/sanitizer_stacktrace.cc
index 244ac36f5f8..0ce5ae475f6 100644
--- a/libsanitizer/sanitizer_common/sanitizer_stacktrace.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_stacktrace.cc
@@ -19,7 +19,8 @@ uptr StackTrace::GetPreviousInstructionPc(uptr pc) {
 #if defined(__arm__)
   // Cancel Thumb bit.
   pc = pc & (~1);
-#elif defined(__powerpc__) || defined(__powerpc64__)
+#endif
+#if defined(__powerpc__) || defined(__powerpc64__)
   // PCs are always 4 byte aligned.
   return pc - 4;
 #elif defined(__sparc__)
@@ -33,19 +34,39 @@ uptr StackTrace::GetCurrentPc() {
   return GET_CALLER_PC();
 }
 
+// Check if given pointer points into allocated stack area.
+static inline bool IsValidFrame(uptr frame, uptr stack_top, uptr stack_bottom) {
+  return frame > stack_bottom && frame < stack_top - 2 * sizeof (uhwptr);
+}
+
+// In GCC on ARM bp points to saved lr, not fp, so we should check the next
+// cell in stack to be a saved frame pointer. GetCanonicFrame returns the
+// pointer to saved frame pointer in any case.
+static inline uhwptr *GetCanonicFrame(uptr bp,
+                                      uptr stack_top,
+                                      uptr stack_bottom) {
+#ifdef __arm__
+  if (!IsValidFrame(bp, stack_top, stack_bottom)) return 0;
+  uhwptr *bp_prev = (uhwptr *)bp;
+  if (IsValidFrame((uptr)bp_prev[0], stack_top, stack_bottom)) return bp_prev;
+  return bp_prev - 1;
+#else
+  return (uhwptr*)bp;
+#endif
+}
+
 void StackTrace::FastUnwindStack(uptr pc, uptr bp,
                                  uptr stack_top, uptr stack_bottom,
                                  uptr max_depth) {
   CHECK_GE(max_depth, 2);
   trace[0] = pc;
   size = 1;
-  uhwptr *frame = (uhwptr *)bp;
-  uhwptr *prev_frame = frame - 1;
   if (stack_top < 4096) return;  // Sanity check for stack top.
+  uhwptr *frame = GetCanonicFrame(bp, stack_top, stack_bottom);
+  uhwptr *prev_frame = 0;
   // Avoid infinite loop when frame == frame[0] by using frame > prev_frame.
   while (frame > prev_frame &&
-         frame < (uhwptr *)stack_top - 2 &&
-         frame > (uhwptr *)stack_bottom &&
+         IsValidFrame((uptr)frame, stack_top, stack_bottom) &&
          IsAligned((uptr)frame, sizeof(*frame)) &&
          size < max_depth) {
     uhwptr pc1 = frame[1];
@@ -53,7 +74,7 @@ void StackTrace::FastUnwindStack(uptr pc, uptr bp,
       trace[size++] = (uptr) pc1;
     }
     prev_frame = frame;
-    frame = (uhwptr *)frame[0];
+    frame = GetCanonicFrame((uptr)frame[0], stack_top, stack_bottom);
   }
 }
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_stacktrace.h b/libsanitizer/sanitizer_common/sanitizer_stacktrace.h
index f66857ed1fb..0e0f1702228 100644
--- a/libsanitizer/sanitizer_common/sanitizer_stacktrace.h
+++ b/libsanitizer/sanitizer_common/sanitizer_stacktrace.h
@@ -50,9 +50,12 @@ struct StackTrace {
 
   static bool WillUseFastUnwind(bool request_fast_unwind) {
     // Check if fast unwind is available. Fast unwind is the only option on Mac.
+    // It is also the only option on FreeBSD as the slow unwinding that
+    // leverages _Unwind_Backtrace() yields the call stack of the signal's
+    // handler and not of the code that raised the signal (as it does on Linux).
     if (!SANITIZER_CAN_FAST_UNWIND)
       return false;
-    else if (SANITIZER_MAC)
+    else if (SANITIZER_MAC != 0 || SANITIZER_FREEBSD != 0)
       return true;
     return request_fast_unwind;
   }
@@ -83,6 +86,10 @@ struct StackTrace {
   uptr local_stack;                           \
   uptr sp = (uptr)&local_stack
 
+#define GET_CALLER_PC_BP \
+  uptr bp = GET_CURRENT_FRAME();              \
+  uptr pc = GET_CALLER_PC();
+
 // Use this macro if you want to print stack trace with the current
 // function in the top frame.
 #define GET_CURRENT_PC_BP_SP \
diff --git a/libsanitizer/sanitizer_common/sanitizer_stacktrace_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_stacktrace_libcdep.cc
index c3c1045ee04..5dcc0e9ed15 100644
--- a/libsanitizer/sanitizer_common/sanitizer_stacktrace_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_stacktrace_libcdep.cc
@@ -35,6 +35,14 @@ void StackTrace::PrintStack(const uptr *addr, uptr size) {
     uptr pc = GetPreviousInstructionPc(addr[i]);
     uptr addr_frames_num = Symbolizer::GetOrInit()->SymbolizePC(
         pc, addr_frames.data(), addr_frames.size());
+    if (addr_frames_num == 0) {
+      frame_desc.clear();
+      PrintStackFramePrefix(&frame_desc, frame_num, pc);
+      frame_desc.append(" (<unknown module>)");
+      Printf("%s\n", frame_desc.data());
+      frame_num++;
+      continue;
+    }
     for (uptr j = 0; j < addr_frames_num; j++) {
       AddressInfo &info = addr_frames[j];
       frame_desc.clear();
diff --git a/libsanitizer/sanitizer_common/sanitizer_suppressions.cc b/libsanitizer/sanitizer_common/sanitizer_suppressions.cc
index 1766fb5a319..ab40598ae8e 100644
--- a/libsanitizer/sanitizer_common/sanitizer_suppressions.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_suppressions.cc
@@ -13,13 +13,15 @@
 
 #include "sanitizer_allocator_internal.h"
 #include "sanitizer_common.h"
+#include "sanitizer_flags.h"
 #include "sanitizer_libc.h"
+#include "sanitizer_placement_new.h"
 
 namespace __sanitizer {
 
 static const char *const kTypeStrings[SuppressionTypeCount] = {
-    "none",   "race", "mutex",           "thread",
-    "signal", "leak", "called_from_lib", "deadlock"};
+    "none", "race",            "mutex",    "thread",    "signal",
+    "leak", "called_from_lib", "deadlock", "vptr_check"};
 
 bool TemplateMatch(char *templ, const char *str) {
   if (str == 0 || str[0] == 0)
@@ -63,6 +65,33 @@ bool TemplateMatch(char *templ, const char *str) {
   return true;
 }
 
+ALIGNED(64) static char placeholder[sizeof(SuppressionContext)];
+static SuppressionContext *suppression_ctx = 0;
+
+SuppressionContext *SuppressionContext::Get() {
+  CHECK(suppression_ctx);
+  return suppression_ctx;
+}
+
+void SuppressionContext::InitIfNecessary() {
+  if (suppression_ctx)
+    return;
+  suppression_ctx = new(placeholder) SuppressionContext;
+  if (common_flags()->suppressions[0] == '\0')
+    return;
+  char *suppressions_from_file;
+  uptr buffer_size;
+  uptr contents_size =
+      ReadFileToBuffer(common_flags()->suppressions, &suppressions_from_file,
+                       &buffer_size, 1 << 26 /* max_len */);
+  if (contents_size == 0) {
+    Printf("%s: failed to read suppressions file '%s'\n", SanitizerToolName,
+           common_flags()->suppressions);
+    Die();
+  }
+  suppression_ctx->Parse(suppressions_from_file);
+}
+
 bool SuppressionContext::Match(const char *str, SuppressionType type,
                                Suppression **s) {
   can_parse_ = false;
diff --git a/libsanitizer/sanitizer_common/sanitizer_suppressions.h b/libsanitizer/sanitizer_common/sanitizer_suppressions.h
index 033ddc5f52a..84ee834a9fd 100644
--- a/libsanitizer/sanitizer_common/sanitizer_suppressions.h
+++ b/libsanitizer/sanitizer_common/sanitizer_suppressions.h
@@ -25,6 +25,7 @@ enum SuppressionType {
   SuppressionLeak,
   SuppressionLib,
   SuppressionDeadlock,
+  SuppressionVptrCheck,
   SuppressionTypeCount
 };
 
@@ -37,14 +38,21 @@ struct Suppression {
 
 class SuppressionContext {
  public:
-  SuppressionContext() : suppressions_(1), can_parse_(true) {}
   void Parse(const char *str);
   bool Match(const char* str, SuppressionType type, Suppression **s);
   uptr SuppressionCount() const;
   const Suppression *SuppressionAt(uptr i) const;
   void GetMatched(InternalMmapVector<Suppression *> *matched);
 
+  // Create a SuppressionContext singleton if it hasn't been created earlier.
+  // Not thread safe. Must be called early during initialization (but after
+  // runtime flags are parsed).
+  static void InitIfNecessary();
+  // Returns a SuppressionContext singleton.
+  static SuppressionContext *Get();
+
  private:
+  SuppressionContext() : suppressions_(1), can_parse_(true) {}
   InternalMmapVector<Suppression> suppressions_;
   bool can_parse_;
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_symbolizer.cc b/libsanitizer/sanitizer_common/sanitizer_symbolizer.cc
index f417b087ae2..6821e5a3262 100644
--- a/libsanitizer/sanitizer_common/sanitizer_symbolizer.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_symbolizer.cc
@@ -20,17 +20,6 @@ Symbolizer *Symbolizer::symbolizer_;
 StaticSpinMutex Symbolizer::init_mu_;
 LowLevelAllocator Symbolizer::symbolizer_allocator_;
 
-Symbolizer *Symbolizer::GetOrNull() {
-  SpinMutexLock l(&init_mu_);
-  return symbolizer_;
-}
-
-Symbolizer *Symbolizer::Get() {
-  SpinMutexLock l(&init_mu_);
-  RAW_CHECK_MSG(symbolizer_ != 0, "Using uninitialized symbolizer!");
-  return symbolizer_;
-}
-
 Symbolizer *Symbolizer::Disable() {
   CHECK_EQ(0, symbolizer_);
   // Initialize a dummy symbolizer.
diff --git a/libsanitizer/sanitizer_common/sanitizer_symbolizer.h b/libsanitizer/sanitizer_common/sanitizer_symbolizer.h
index 05fc6a7cbb9..73a68b2ee11 100644
--- a/libsanitizer/sanitizer_common/sanitizer_symbolizer.h
+++ b/libsanitizer/sanitizer_common/sanitizer_symbolizer.h
@@ -70,20 +70,9 @@ struct DataInfo {
 
 class Symbolizer {
  public:
-  /// Returns platform-specific implementation of Symbolizer. The symbolizer
-  /// must be initialized (with init or disable) before calling this function.
-  static Symbolizer *Get();
-  /// Returns platform-specific implementation of Symbolizer, or null if not
-  /// initialized.
-  static Symbolizer *GetOrNull();
-  /// Returns platform-specific implementation of Symbolizer.  Will
-  /// automatically initialize symbolizer as if by calling Init(0) if needed.
+  /// Initialize and return platform-specific implementation of symbolizer
+  /// (if it wasn't already initialized).
   static Symbolizer *GetOrInit();
-  /// Initialize and return the symbolizer, given an optional path to an
-  /// external symbolizer.  The path argument is only required for legacy
-  /// reasons as this function will check $PATH for an external symbolizer.  Not
-  /// thread safe.
-  static Symbolizer *Init(const char* path_to_external = 0);
   // Fills at most "max_frames" elements of "frames" with descriptions
   // for a given address (in all inlined functions). Returns the number
   // of descriptions actually filled.
@@ -120,10 +109,7 @@ class Symbolizer {
 
  private:
   /// Platform-specific function for creating a Symbolizer object.
-  static Symbolizer *PlatformInit(const char *path_to_external);
-  /// Create a symbolizer and store it to symbolizer_ without checking if one
-  /// already exists.  Not thread safe.
-  static Symbolizer *CreateAndStore(const char *path_to_external);
+  static Symbolizer *PlatformInit();
   /// Initialize the symbolizer in a disabled state.  Not thread safe.
   static Symbolizer *Disable();
 
diff --git a/libsanitizer/sanitizer_common/sanitizer_symbolizer_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_symbolizer_libcdep.cc
index 2d9caaf4e85..9afd805a82c 100644
--- a/libsanitizer/sanitizer_common/sanitizer_symbolizer_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_symbolizer_libcdep.cc
@@ -14,24 +14,13 @@
 
 namespace __sanitizer {
 
-Symbolizer *Symbolizer::CreateAndStore(const char *path_to_external) {
-  Symbolizer *platform_symbolizer = PlatformInit(path_to_external);
-  if (!platform_symbolizer)
-    return Disable();
-  symbolizer_ = platform_symbolizer;
-  return platform_symbolizer;
-}
-
-Symbolizer *Symbolizer::Init(const char *path_to_external) {
-  CHECK_EQ(0, symbolizer_);
-  return CreateAndStore(path_to_external);
-}
-
 Symbolizer *Symbolizer::GetOrInit() {
   SpinMutexLock l(&init_mu_);
-  if (symbolizer_ == 0)
-    return CreateAndStore(0);
-  return symbolizer_;
+  if (symbolizer_)
+    return symbolizer_;
+  if ((symbolizer_ = PlatformInit()))
+    return symbolizer_;
+  return Disable();
 }
 
 }  // namespace __sanitizer
diff --git a/libsanitizer/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cc
index 161c21b3aad..4f30225a4c8 100644
--- a/libsanitizer/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cc
@@ -253,7 +253,7 @@ class SymbolizerProcess : public ExternalSymbolizerInterface {
       internal_close(outfd[1]);
       internal_close(infd[0]);
       internal_close(infd[1]);
-      for (int fd = getdtablesize(); fd > 2; fd--)
+      for (int fd = sysconf(_SC_OPEN_MAX); fd > 2; fd--)
         internal_close(fd);
       ExecuteWithDefaultArgs(path_);
       internal__exit(1);
@@ -714,7 +714,7 @@ class POSIXSymbolizer : public Symbolizer {
   LibbacktraceSymbolizer *libbacktrace_symbolizer_;   // Leaked.
 };
 
-Symbolizer *Symbolizer::PlatformInit(const char *path_to_external) {
+Symbolizer *Symbolizer::PlatformInit() {
   if (!common_flags()->symbolize) {
     return new(symbolizer_allocator_) POSIXSymbolizer(0, 0, 0);
   }
@@ -727,6 +727,7 @@ Symbolizer *Symbolizer::PlatformInit(const char *path_to_external) {
     libbacktrace_symbolizer =
         LibbacktraceSymbolizer::get(&symbolizer_allocator_);
     if (!libbacktrace_symbolizer) {
+      const char *path_to_external = common_flags()->external_symbolizer_path;
       if (path_to_external && path_to_external[0] == '\0') {
         // External symbolizer is explicitly disabled. Do nothing.
       } else {
diff --git a/libsanitizer/sanitizer_common/sanitizer_symbolizer_win.cc b/libsanitizer/sanitizer_common/sanitizer_symbolizer_win.cc
index c014c6af42b..a1ed4e9a7b7 100644
--- a/libsanitizer/sanitizer_common/sanitizer_symbolizer_win.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_symbolizer_win.cc
@@ -30,11 +30,22 @@ class WinSymbolizer : public Symbolizer {
 
     BlockingMutexLock l(&dbghelp_mu_);
     if (!initialized_) {
-      SymSetOptions(SYMOPT_DEFERRED_LOADS |
-                    SYMOPT_UNDNAME |
-                    SYMOPT_LOAD_LINES);
-      CHECK(SymInitialize(GetCurrentProcess(), 0, TRUE));
-      // FIXME: We don't call SymCleanup() on exit yet - should we?
+      if (!TrySymInitialize()) {
+        // OK, maybe the client app has called SymInitialize already.
+        // That's a bit unfortunate for us as all the DbgHelp functions are
+        // single-threaded and we can't coordinate with the app.
+        // FIXME: Can we stop the other threads at this point?
+        // Anyways, we have to reconfigure stuff to make sure that SymInitialize
+        // has all the appropriate options set.
+        // Cross our fingers and reinitialize DbgHelp.
+        Report("*** WARNING: Failed to initialize DbgHelp!              ***\n");
+        Report("*** Most likely this means that the app is already      ***\n");
+        Report("*** using DbgHelp, possibly with incompatible flags.    ***\n");
+        Report("*** Due to technical reasons, symbolization might crash ***\n");
+        Report("*** or produce wrong results.                           ***\n");
+        SymCleanup(GetCurrentProcess());
+        TrySymInitialize();
+      }
       initialized_ = true;
     }
 
@@ -90,13 +101,19 @@ class WinSymbolizer : public Symbolizer {
   // FIXME: Implement GetModuleNameAndOffsetForPC().
 
  private:
+  bool TrySymInitialize() {
+    SymSetOptions(SYMOPT_DEFERRED_LOADS | SYMOPT_UNDNAME | SYMOPT_LOAD_LINES);
+    return SymInitialize(GetCurrentProcess(), 0, TRUE);
+    // FIXME: We don't call SymCleanup() on exit yet - should we?
+  }
+
   // All DbgHelp functions are single threaded, so we should use a mutex to
   // serialize accesses.
   BlockingMutex dbghelp_mu_;
   bool initialized_;
 };
 
-Symbolizer *Symbolizer::PlatformInit(const char *path_to_external) {
+Symbolizer *Symbolizer::PlatformInit() {
   static bool called_once = false;
   CHECK(!called_once && "Shouldn't create more than one symbolizer");
   called_once = true;
diff --git a/libsanitizer/sanitizer_common/sanitizer_thread_registry.cc b/libsanitizer/sanitizer_common/sanitizer_thread_registry.cc
index 0ad47561b8b..bfa610443c1 100644
--- a/libsanitizer/sanitizer_common/sanitizer_thread_registry.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_thread_registry.cc
@@ -216,7 +216,7 @@ void ThreadRegistry::SetThreadNameByUserId(uptr user_id, const char *name) {
   }
 }
 
-void ThreadRegistry::DetachThread(u32 tid) {
+void ThreadRegistry::DetachThread(u32 tid, void *arg) {
   BlockingMutexLock l(&mtx_);
   CHECK_LT(tid, n_contexts_);
   ThreadContextBase *tctx = threads_[tid];
@@ -225,6 +225,7 @@ void ThreadRegistry::DetachThread(u32 tid) {
     Report("%s: Detach of non-existent thread\n", SanitizerToolName);
     return;
   }
+  tctx->OnDetached(arg);
   if (tctx->status == ThreadStatusFinished) {
     tctx->SetDead();
     QuarantinePush(tctx);
diff --git a/libsanitizer/sanitizer_common/sanitizer_thread_registry.h b/libsanitizer/sanitizer_common/sanitizer_thread_registry.h
index 2d7f9e90e0d..d5a741bbfc4 100644
--- a/libsanitizer/sanitizer_common/sanitizer_thread_registry.h
+++ b/libsanitizer/sanitizer_common/sanitizer_thread_registry.h
@@ -66,6 +66,7 @@ class ThreadContextBase {
   virtual void OnStarted(void *arg) {}
   virtual void OnCreated(void *arg) {}
   virtual void OnReset() {}
+  virtual void OnDetached(void *arg) {}
 };
 
 typedef ThreadContextBase* (*ThreadContextFactory)(u32 tid);
@@ -108,7 +109,7 @@ class ThreadRegistry {
 
   void SetThreadName(u32 tid, const char *name);
   void SetThreadNameByUserId(uptr user_id, const char *name);
-  void DetachThread(u32 tid);
+  void DetachThread(u32 tid, void *arg);
   void JoinThread(u32 tid, void *arg);
   void FinishThread(u32 tid);
   void StartThread(u32 tid, uptr os_id, void *arg);
diff --git a/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.cc b/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.cc
index 1d6170f9157..af828045b59 100644
--- a/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.cc
@@ -76,14 +76,13 @@ void DTLS_Destroy() {
   DTLS_Deallocate(dtls.dtv, s);
 }
 
-void DTLS_on_tls_get_addr(void *arg_void, void *res) {
-  if (!common_flags()->intercept_tls_get_addr) return;
+DTLS::DTV *DTLS_on_tls_get_addr(void *arg_void, void *res) {
+  if (!common_flags()->intercept_tls_get_addr) return 0;
   TlsGetAddrParam *arg = reinterpret_cast<TlsGetAddrParam *>(arg_void);
   uptr dso_id = arg->dso_id;
-  if (dtls.dtv_size == kDestroyedThread) return;
+  if (dtls.dtv_size == kDestroyedThread) return 0;
   DTLS_Resize(dso_id + 1);
-  if (dtls.dtv[dso_id].beg)
-    return;
+  if (dtls.dtv[dso_id].beg) return 0;
   uptr tls_size = 0;
   uptr tls_beg = reinterpret_cast<uptr>(res) - arg->offset;
   VPrintf(2, "__tls_get_addr: %p {%p,%p} => %p; tls_beg: %p; sp: %p "
@@ -108,6 +107,7 @@ void DTLS_on_tls_get_addr(void *arg_void, void *res) {
   }
   dtls.dtv[dso_id].beg = tls_beg;
   dtls.dtv[dso_id].size = tls_size;
+  return dtls.dtv + dso_id;
 }
 
 void DTLS_on_libc_memalign(void *ptr, uptr size) {
@@ -121,7 +121,7 @@ DTLS *DTLS_Get() { return &dtls; }
 
 #else
 void DTLS_on_libc_memalign(void *ptr, uptr size) {}
-void DTLS_on_tls_get_addr(void *arg, void *res) {}
+DTLS::DTV *DTLS_on_tls_get_addr(void *arg, void *res) { return 0; }
 DTLS *DTLS_Get() { return 0; }
 void DTLS_Destroy() {}
 #endif  // SANITIZER_INTERCEPT_TLS_GET_ADDR
diff --git a/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.h b/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.h
index 22a5e1c41c2..8dc629f630e 100644
--- a/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.h
+++ b/libsanitizer/sanitizer_common/sanitizer_tls_get_addr.h
@@ -46,7 +46,9 @@ struct DTLS {
   uptr last_memalign_ptr;
 };
 
-void DTLS_on_tls_get_addr(void *arg, void *res);
+// Returns pointer and size of a linker-allocated TLS block.
+// Each block is returned exactly once.
+DTLS::DTV *DTLS_on_tls_get_addr(void *arg, void *res);
 void DTLS_on_libc_memalign(void *ptr, uptr size);
 DTLS *DTLS_Get();
 void DTLS_Destroy();  // Make sure to call this before the thread is destroyed.
diff --git a/libsanitizer/sanitizer_common/sanitizer_unwind_posix_libcdep.cc b/libsanitizer/sanitizer_common/sanitizer_unwind_posix_libcdep.cc
new file mode 100644
index 00000000000..c49113e1d0b
--- /dev/null
+++ b/libsanitizer/sanitizer_common/sanitizer_unwind_posix_libcdep.cc
@@ -0,0 +1,151 @@
+//===-- sanitizer_unwind_posix.cc ----------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the unwind.h-based (aka "slow") stack unwinding routines
+// available to the tools on Linux, Android, FreeBSD and OS X.
+//===----------------------------------------------------------------------===//
+
+#include "sanitizer_platform.h"
+#if SANITIZER_POSIX
+#include "sanitizer_common.h"
+#include "sanitizer_stacktrace.h"
+
+#if SANITIZER_ANDROID
+#include <dlfcn.h>  // for dlopen()
+#endif
+
+#if SANITIZER_FREEBSD
+#define _GNU_SOURCE  // to declare _Unwind_Backtrace() from <unwind.h>
+#endif
+#include <unwind.h>
+
+namespace __sanitizer {
+
+//------------------------- SlowUnwindStack -----------------------------------
+
+typedef struct {
+  uptr absolute_pc;
+  uptr stack_top;
+  uptr stack_size;
+} backtrace_frame_t;
+
+extern "C" {
+typedef void *(*acquire_my_map_info_list_func)();
+typedef void (*release_my_map_info_list_func)(void *map);
+typedef sptr (*unwind_backtrace_signal_arch_func)(
+    void *siginfo, void *sigcontext, void *map_info_list,
+    backtrace_frame_t *backtrace, uptr ignore_depth, uptr max_depth);
+acquire_my_map_info_list_func acquire_my_map_info_list;
+release_my_map_info_list_func release_my_map_info_list;
+unwind_backtrace_signal_arch_func unwind_backtrace_signal_arch;
+} // extern "C"
+
+#if SANITIZER_ANDROID
+void SanitizerInitializeUnwinder() {
+  void *p = dlopen("libcorkscrew.so", RTLD_LAZY);
+  if (!p) {
+    VReport(1,
+            "Failed to open libcorkscrew.so. You may see broken stack traces "
+            "in SEGV reports.");
+    return;
+  }
+  acquire_my_map_info_list =
+      (acquire_my_map_info_list_func)(uptr)dlsym(p, "acquire_my_map_info_list");
+  release_my_map_info_list =
+      (release_my_map_info_list_func)(uptr)dlsym(p, "release_my_map_info_list");
+  unwind_backtrace_signal_arch = (unwind_backtrace_signal_arch_func)(uptr)dlsym(
+      p, "unwind_backtrace_signal_arch");
+  if (!acquire_my_map_info_list || !release_my_map_info_list ||
+      !unwind_backtrace_signal_arch) {
+    VReport(1,
+            "Failed to find one of the required symbols in libcorkscrew.so. "
+            "You may see broken stack traces in SEGV reports.");
+    acquire_my_map_info_list = 0;
+    unwind_backtrace_signal_arch = 0;
+    release_my_map_info_list = 0;
+  }
+}
+#endif
+
+#ifdef __arm__
+#define UNWIND_STOP _URC_END_OF_STACK
+#define UNWIND_CONTINUE _URC_NO_REASON
+#else
+#define UNWIND_STOP _URC_NORMAL_STOP
+#define UNWIND_CONTINUE _URC_NO_REASON
+#endif
+
+uptr Unwind_GetIP(struct _Unwind_Context *ctx) {
+#ifdef __arm__
+  uptr val;
+  _Unwind_VRS_Result res = _Unwind_VRS_Get(ctx, _UVRSC_CORE,
+      15 /* r15 = PC */, _UVRSD_UINT32, &val);
+  CHECK(res == _UVRSR_OK && "_Unwind_VRS_Get failed");
+  // Clear the Thumb bit.
+  return val & ~(uptr)1;
+#else
+  return _Unwind_GetIP(ctx);
+#endif
+}
+
+struct UnwindTraceArg {
+  StackTrace *stack;
+  uptr max_depth;
+};
+
+_Unwind_Reason_Code Unwind_Trace(struct _Unwind_Context *ctx, void *param) {
+  UnwindTraceArg *arg = (UnwindTraceArg*)param;
+  CHECK_LT(arg->stack->size, arg->max_depth);
+  uptr pc = Unwind_GetIP(ctx);
+  arg->stack->trace[arg->stack->size++] = pc;
+  if (arg->stack->size == arg->max_depth) return UNWIND_STOP;
+  return UNWIND_CONTINUE;
+}
+
+void StackTrace::SlowUnwindStack(uptr pc, uptr max_depth) {
+  CHECK_GE(max_depth, 2);
+  size = 0;
+  UnwindTraceArg arg = {this, Min(max_depth + 1, kStackTraceMax)};
+  _Unwind_Backtrace(Unwind_Trace, &arg);
+  // We need to pop a few frames so that pc is on top.
+  uptr to_pop = LocatePcInTrace(pc);
+  // trace[0] belongs to the current function so we always pop it.
+  if (to_pop == 0)
+    to_pop = 1;
+  PopStackFrames(to_pop);
+  trace[0] = pc;
+}
+
+void StackTrace::SlowUnwindStackWithContext(uptr pc, void *context,
+                                            uptr max_depth) {
+  CHECK_GE(max_depth, 2);
+  if (!unwind_backtrace_signal_arch) {
+    SlowUnwindStack(pc, max_depth);
+    return;
+  }
+
+  void *map = acquire_my_map_info_list();
+  CHECK(map);
+  InternalScopedBuffer<backtrace_frame_t> frames(kStackTraceMax);
+  // siginfo argument appears to be unused.
+  sptr res = unwind_backtrace_signal_arch(/* siginfo */ 0, context, map,
+                                          frames.data(),
+                                          /* ignore_depth */ 0, max_depth);
+  release_my_map_info_list(map);
+  if (res < 0) return;
+  CHECK_LE((uptr)res, kStackTraceMax);
+
+  size = 0;
+  // +2 compensate for libcorkscrew unwinder returning addresses of call
+  // instructions instead of raw return addresses.
+  for (sptr i = 0; i < res; ++i)
+    trace[size++] = frames[i].absolute_pc + 2;
+}
+
+}  // namespace __sanitizer
+
+#endif  // SANITIZER_POSIX
diff --git a/libsanitizer/sanitizer_common/sanitizer_win.cc b/libsanitizer/sanitizer_common/sanitizer_win.cc
index 6065838cefe..9f24510dbd0 100644
--- a/libsanitizer/sanitizer_common/sanitizer_win.cc
+++ b/libsanitizer/sanitizer_common/sanitizer_win.cc
@@ -15,9 +15,10 @@
 
 #define WIN32_LEAN_AND_MEAN
 #define NOGDI
-#include <stdlib.h>
-#include <io.h>
 #include <windows.h>
+#include <dbghelp.h>
+#include <io.h>
+#include <stdlib.h>
 
 #include "sanitizer_common.h"
 #include "sanitizer_libc.h"
@@ -62,6 +63,7 @@ uptr GetThreadSelf() {
   return GetTid();
 }
 
+#if !SANITIZER_GO
 void GetThreadStackTopAndBottom(bool at_initialization, uptr *stack_top,
                                 uptr *stack_bottom) {
   CHECK(stack_top);
@@ -74,6 +76,7 @@ void GetThreadStackTopAndBottom(bool at_initialization, uptr *stack_top,
   *stack_top = (uptr)mbi.BaseAddress + mbi.RegionSize;
   *stack_bottom = (uptr)mbi.AllocationBase;
 }
+#endif  // #if !SANITIZER_GO
 
 void *MmapOrDie(uptr size, const char *mem_type) {
   void *rv = VirtualAlloc(0, size, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
@@ -135,6 +138,10 @@ void *MapFileToMemory(const char *file_name, uptr *buff_size) {
   UNIMPLEMENTED();
 }
 
+void *MapWritableFileToMemory(void *addr, uptr size, uptr fd, uptr offset) {
+  UNIMPLEMENTED();
+}
+
 static const int kMaxEnvNameLength = 128;
 static const DWORD kMaxEnvValueLength = 32767;
 
@@ -182,7 +189,7 @@ void DumpProcessMap() {
   UNIMPLEMENTED();
 }
 
-void DisableCoreDumper() {
+void DisableCoreDumperIfNecessary() {
   // Do nothing.
 }
 
@@ -203,6 +210,14 @@ void SetStackSizeLimitInBytes(uptr limit) {
   UNIMPLEMENTED();
 }
 
+bool AddressSpaceIsUnlimited() {
+  UNIMPLEMENTED();
+}
+
+void SetAddressSpaceUnlimited() {
+  UNIMPLEMENTED();
+}
+
 char *FindPathToBinary(const char *name) {
   // Nothing here for now.
   return 0;
@@ -350,6 +365,14 @@ void internal__exit(int exitcode) {
   ExitProcess(exitcode);
 }
 
+uptr internal_ftruncate(fd_t fd, uptr size) {
+  UNIMPLEMENTED();
+}
+
+uptr internal_rename(const char *oldpath, const char *newpath) {
+  UNIMPLEMENTED();
+}
+
 // ---------------------- BlockingMutex ---------------- {{{1
 const uptr LOCK_UNINITIALIZED = 0;
 const uptr LOCK_READY = (uptr)-1;
@@ -418,6 +441,7 @@ void GetThreadStackAndTls(bool main, uptr *stk_addr, uptr *stk_size,
 #endif
 }
 
+#if !SANITIZER_GO
 void StackTrace::SlowUnwindStack(uptr pc, uptr max_depth) {
   CHECK_GE(max_depth, 2);
   // FIXME: CaptureStackBackTrace might be too slow for us.
@@ -435,8 +459,32 @@ void StackTrace::SlowUnwindStack(uptr pc, uptr max_depth) {
 
 void StackTrace::SlowUnwindStackWithContext(uptr pc, void *context,
                                             uptr max_depth) {
-  UNREACHABLE("no signal context on windows");
+  CONTEXT ctx = *(CONTEXT *)context;
+  STACKFRAME64 stack_frame;
+  memset(&stack_frame, 0, sizeof(stack_frame));
+  size = 0;
+#if defined(_WIN64)
+  int machine_type = IMAGE_FILE_MACHINE_AMD64;
+  stack_frame.AddrPC.Offset = ctx.Rip;
+  stack_frame.AddrFrame.Offset = ctx.Rbp;
+  stack_frame.AddrStack.Offset = ctx.Rsp;
+#else
+  int machine_type = IMAGE_FILE_MACHINE_I386;
+  stack_frame.AddrPC.Offset = ctx.Eip;
+  stack_frame.AddrFrame.Offset = ctx.Ebp;
+  stack_frame.AddrStack.Offset = ctx.Esp;
+#endif
+  stack_frame.AddrPC.Mode = AddrModeFlat;
+  stack_frame.AddrFrame.Mode = AddrModeFlat;
+  stack_frame.AddrStack.Mode = AddrModeFlat;
+  while (StackWalk64(machine_type, GetCurrentProcess(), GetCurrentThread(),
+                     &stack_frame, &ctx, NULL, &SymFunctionTableAccess64,
+                     &SymGetModuleBase64, NULL) &&
+         size < Min(max_depth, kStackTraceMax)) {
+    trace[size++] = (uptr)stack_frame.AddrPC.Offset;
+  }
 }
+#endif  // #if !SANITIZER_GO
 
 void MaybeOpenReportFile() {
   // Windows doesn't have native fork, and we don't support Cygwin or other
@@ -472,6 +520,11 @@ bool IsDeadlySignal(int signum) {
   return false;
 }
 
+bool IsAccessibleMemoryRange(uptr beg, uptr size) {
+  // FIXME: Actually implement this function.
+  return true;
+}
+
 }  // namespace __sanitizer
 
 #endif  // _WIN32
diff --git a/libsanitizer/tsan/Makefile.am b/libsanitizer/tsan/Makefile.am
index 5d98e215e05..ab4748e0878 100644
--- a/libsanitizer/tsan/Makefile.am
+++ b/libsanitizer/tsan/Makefile.am
@@ -32,6 +32,7 @@ tsan_files = \
 	tsan_rtl_mutex.cc \
 	tsan_rtl_report.cc \
 	tsan_rtl_thread.cc \
+	tsan_stack_trace.cc \
 	tsan_stat.cc \
 	tsan_suppressions.cc \
 	tsan_symbolize.cc \
diff --git a/libsanitizer/tsan/Makefile.in b/libsanitizer/tsan/Makefile.in
index 068aaa87fe8..c6f1ecffa75 100644
--- a/libsanitizer/tsan/Makefile.in
+++ b/libsanitizer/tsan/Makefile.in
@@ -91,8 +91,8 @@ am__objects_1 = tsan_clock.lo tsan_fd.lo tsan_flags.lo \
 	tsan_mutexset.lo tsan_platform_linux.lo tsan_platform_mac.lo \
 	tsan_platform_windows.lo tsan_report.lo tsan_rtl.lo \
 	tsan_rtl_mutex.lo tsan_rtl_report.lo tsan_rtl_thread.lo \
-	tsan_stat.lo tsan_suppressions.lo tsan_symbolize.lo \
-	tsan_sync.lo tsan_rtl_amd64.lo
+	tsan_stack_trace.lo tsan_stat.lo tsan_suppressions.lo \
+	tsan_symbolize.lo tsan_sync.lo tsan_rtl_amd64.lo
 am_libtsan_la_OBJECTS = $(am__objects_1)
 libtsan_la_OBJECTS = $(am_libtsan_la_OBJECTS)
 libtsan_la_LINK = $(LIBTOOL) --tag=CXX $(AM_LIBTOOLFLAGS) \
@@ -301,6 +301,7 @@ tsan_files = \
 	tsan_rtl_mutex.cc \
 	tsan_rtl_report.cc \
 	tsan_rtl_thread.cc \
+	tsan_stack_trace.cc \
 	tsan_stat.cc \
 	tsan_suppressions.cc \
 	tsan_symbolize.cc \
@@ -449,6 +450,7 @@ distclean-compile:
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_rtl_mutex.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_rtl_report.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_rtl_thread.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_stack_trace.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_stat.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_suppressions.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/tsan_symbolize.Plo@am__quote@
diff --git a/libsanitizer/tsan/tsan_clock.cc b/libsanitizer/tsan/tsan_clock.cc
index b944cc50d54..a84caa952a6 100644
--- a/libsanitizer/tsan/tsan_clock.cc
+++ b/libsanitizer/tsan/tsan_clock.cc
@@ -10,6 +10,7 @@
 //===----------------------------------------------------------------------===//
 #include "tsan_clock.h"
 #include "tsan_rtl.h"
+#include "sanitizer_common/sanitizer_placement_new.h"
 
 // SyncClock and ThreadClock implement vector clocks for sync variables
 // (mutexes, atomic variables, file descriptors, etc) and threads, respectively.
@@ -100,13 +101,13 @@ ThreadClock::ThreadClock(unsigned tid, unsigned reused)
   clk_[tid_].reused = reused_;
 }
 
-void ThreadClock::acquire(const SyncClock *src) {
+void ThreadClock::acquire(ClockCache *c, const SyncClock *src) {
   DCHECK(nclk_ <= kMaxTid);
-  DCHECK(src->clk_.Size() <= kMaxTid);
+  DCHECK(src->size_ <= kMaxTid);
   CPP_STAT_INC(StatClockAcquire);
 
   // Check if it's empty -> no need to do anything.
-  const uptr nclk = src->clk_.Size();
+  const uptr nclk = src->size_;
   if (nclk == 0) {
     CPP_STAT_INC(StatClockAcquireEmpty);
     return;
@@ -116,12 +117,12 @@ void ThreadClock::acquire(const SyncClock *src) {
   bool acquired = false;
   if (nclk > tid_) {
     CPP_STAT_INC(StatClockAcquireLarge);
-    if (src->clk_[tid_].reused == reused_) {
+    if (src->elem(tid_).reused == reused_) {
       CPP_STAT_INC(StatClockAcquireRepeat);
       for (unsigned i = 0; i < kDirtyTids; i++) {
         unsigned tid = src->dirty_tids_[i];
         if (tid != kInvalidTid) {
-          u64 epoch = src->clk_[tid].epoch;
+          u64 epoch = src->elem(tid).epoch;
           if (clk_[tid].epoch < epoch) {
             clk_[tid].epoch = epoch;
             acquired = true;
@@ -140,7 +141,7 @@ void ThreadClock::acquire(const SyncClock *src) {
   CPP_STAT_INC(StatClockAcquireFull);
   nclk_ = max(nclk_, nclk);
   for (uptr i = 0; i < nclk; i++) {
-    u64 epoch = src->clk_[i].epoch;
+    u64 epoch = src->elem(i).epoch;
     if (clk_[i].epoch < epoch) {
       clk_[i].epoch = epoch;
       acquired = true;
@@ -149,7 +150,7 @@ void ThreadClock::acquire(const SyncClock *src) {
 
   // Remember that this thread has acquired this clock.
   if (nclk > tid_)
-    src->clk_[tid_].reused = reused_;
+    src->elem(tid_).reused = reused_;
 
   if (acquired) {
     CPP_STAT_INC(StatClockAcquiredSomething);
@@ -157,28 +158,26 @@ void ThreadClock::acquire(const SyncClock *src) {
   }
 }
 
-void ThreadClock::release(SyncClock *dst) const {
+void ThreadClock::release(ClockCache *c, SyncClock *dst) const {
   DCHECK_LE(nclk_, kMaxTid);
-  DCHECK_LE(dst->clk_.Size(), kMaxTid);
+  DCHECK_LE(dst->size_, kMaxTid);
 
-  if (dst->clk_.Size() == 0) {
+  if (dst->size_ == 0) {
     // ReleaseStore will correctly set release_store_tid_,
     // which can be important for future operations.
-    ReleaseStore(dst);
+    ReleaseStore(c, dst);
     return;
   }
 
   CPP_STAT_INC(StatClockRelease);
   // Check if we need to resize dst.
-  if (dst->clk_.Size() < nclk_) {
-    CPP_STAT_INC(StatClockReleaseResize);
-    dst->clk_.Resize(nclk_);
-  }
+  if (dst->size_ < nclk_)
+    dst->Resize(c, nclk_);
 
   // Check if we had not acquired anything from other threads
   // since the last release on dst. If so, we need to update
-  // only dst->clk_[tid_].
-  if (dst->clk_[tid_].epoch > last_acquire_) {
+  // only dst->elem(tid_).
+  if (dst->elem(tid_).epoch > last_acquire_) {
     UpdateCurrentThread(dst);
     if (dst->release_store_tid_ != tid_ ||
         dst->release_store_reused_ != reused_)
@@ -194,14 +193,15 @@ void ThreadClock::release(SyncClock *dst) const {
     CPP_STAT_INC(StatClockReleaseAcquired);
   // Update dst->clk_.
   for (uptr i = 0; i < nclk_; i++) {
-    dst->clk_[i].epoch = max(dst->clk_[i].epoch, clk_[i].epoch);
-    dst->clk_[i].reused = 0;
+    ClockElem &ce = dst->elem(i);
+    ce.epoch = max(ce.epoch, clk_[i].epoch);
+    ce.reused = 0;
   }
   // Clear 'acquired' flag in the remaining elements.
-  if (nclk_ < dst->clk_.Size())
+  if (nclk_ < dst->size_)
     CPP_STAT_INC(StatClockReleaseClearTail);
-  for (uptr i = nclk_; i < dst->clk_.Size(); i++)
-    dst->clk_[i].reused = 0;
+  for (uptr i = nclk_; i < dst->size_; i++)
+    dst->elem(i).reused = 0;
   for (unsigned i = 0; i < kDirtyTids; i++)
     dst->dirty_tids_[i] = kInvalidTid;
   dst->release_store_tid_ = kInvalidTid;
@@ -209,23 +209,21 @@ void ThreadClock::release(SyncClock *dst) const {
   // If we've acquired dst, remember this fact,
   // so that we don't need to acquire it on next acquire.
   if (acquired)
-    dst->clk_[tid_].reused = reused_;
+    dst->elem(tid_).reused = reused_;
 }
 
-void ThreadClock::ReleaseStore(SyncClock *dst) const {
+void ThreadClock::ReleaseStore(ClockCache *c, SyncClock *dst) const {
   DCHECK(nclk_ <= kMaxTid);
-  DCHECK(dst->clk_.Size() <= kMaxTid);
+  DCHECK(dst->size_ <= kMaxTid);
   CPP_STAT_INC(StatClockStore);
 
   // Check if we need to resize dst.
-  if (dst->clk_.Size() < nclk_) {
-    CPP_STAT_INC(StatClockStoreResize);
-    dst->clk_.Resize(nclk_);
-  }
+  if (dst->size_ < nclk_)
+    dst->Resize(c, nclk_);
 
   if (dst->release_store_tid_ == tid_ &&
       dst->release_store_reused_ == reused_ &&
-      dst->clk_[tid_].epoch > last_acquire_) {
+      dst->elem(tid_).epoch > last_acquire_) {
     CPP_STAT_INC(StatClockStoreFast);
     UpdateCurrentThread(dst);
     return;
@@ -234,13 +232,17 @@ void ThreadClock::ReleaseStore(SyncClock *dst) const {
   // O(N) release-store.
   CPP_STAT_INC(StatClockStoreFull);
   for (uptr i = 0; i < nclk_; i++) {
-    dst->clk_[i].epoch = clk_[i].epoch;
-    dst->clk_[i].reused = 0;
+    ClockElem &ce = dst->elem(i);
+    ce.epoch = clk_[i].epoch;
+    ce.reused = 0;
   }
   // Clear the tail of dst->clk_.
-  if (nclk_ < dst->clk_.Size()) {
-    internal_memset(&dst->clk_[nclk_], 0,
-        (dst->clk_.Size() - nclk_) * sizeof(dst->clk_[0]));
+  if (nclk_ < dst->size_) {
+    for (uptr i = nclk_; i < dst->size_; i++) {
+      ClockElem &ce = dst->elem(i);
+      ce.epoch = 0;
+      ce.reused = 0;
+    }
     CPP_STAT_INC(StatClockStoreTail);
   }
   for (unsigned i = 0; i < kDirtyTids; i++)
@@ -248,19 +250,19 @@ void ThreadClock::ReleaseStore(SyncClock *dst) const {
   dst->release_store_tid_ = tid_;
   dst->release_store_reused_ = reused_;
   // Rememeber that we don't need to acquire it in future.
-  dst->clk_[tid_].reused = reused_;
+  dst->elem(tid_).reused = reused_;
 }
 
-void ThreadClock::acq_rel(SyncClock *dst) {
+void ThreadClock::acq_rel(ClockCache *c, SyncClock *dst) {
   CPP_STAT_INC(StatClockAcquireRelease);
-  acquire(dst);
-  ReleaseStore(dst);
+  acquire(c, dst);
+  ReleaseStore(c, dst);
 }
 
 // Updates only single element related to the current thread in dst->clk_.
 void ThreadClock::UpdateCurrentThread(SyncClock *dst) const {
   // Update the threads time, but preserve 'acquired' flag.
-  dst->clk_[tid_].epoch = clk_[tid_].epoch;
+  dst->elem(tid_).epoch = clk_[tid_].epoch;
 
   for (unsigned i = 0; i < kDirtyTids; i++) {
     if (dst->dirty_tids_[i] == tid_) {
@@ -275,27 +277,73 @@ void ThreadClock::UpdateCurrentThread(SyncClock *dst) const {
   }
   // Reset all 'acquired' flags, O(N).
   CPP_STAT_INC(StatClockReleaseSlow);
-  for (uptr i = 0; i < dst->clk_.Size(); i++) {
-    dst->clk_[i].reused = 0;
-  }
+  for (uptr i = 0; i < dst->size_; i++)
+    dst->elem(i).reused = 0;
   for (unsigned i = 0; i < kDirtyTids; i++)
     dst->dirty_tids_[i] = kInvalidTid;
 }
 
 // Checks whether the current threads has already acquired src.
 bool ThreadClock::IsAlreadyAcquired(const SyncClock *src) const {
-  if (src->clk_[tid_].reused != reused_)
+  if (src->elem(tid_).reused != reused_)
     return false;
   for (unsigned i = 0; i < kDirtyTids; i++) {
     unsigned tid = src->dirty_tids_[i];
     if (tid != kInvalidTid) {
-      if (clk_[tid].epoch < src->clk_[tid].epoch)
+      if (clk_[tid].epoch < src->elem(tid).epoch)
         return false;
     }
   }
   return true;
 }
 
+void SyncClock::Resize(ClockCache *c, uptr nclk) {
+  CPP_STAT_INC(StatClockReleaseResize);
+  if (RoundUpTo(nclk, ClockBlock::kClockCount) <=
+      RoundUpTo(size_, ClockBlock::kClockCount)) {
+    // Growing within the same block.
+    // Memory is already allocated, just increase the size.
+    size_ = nclk;
+    return;
+  }
+  if (nclk <= ClockBlock::kClockCount) {
+    // Grow from 0 to one-level table.
+    CHECK_EQ(size_, 0);
+    CHECK_EQ(tab_, 0);
+    CHECK_EQ(tab_idx_, 0);
+    size_ = nclk;
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+    return;
+  }
+  // Growing two-level table.
+  if (size_ == 0) {
+    // Allocate first level table.
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+  } else if (size_ <= ClockBlock::kClockCount) {
+    // Transform one-level table to two-level table.
+    u32 old = tab_idx_;
+    tab_idx_ = ctx->clock_alloc.Alloc(c);
+    tab_ = ctx->clock_alloc.Map(tab_idx_);
+    internal_memset(tab_, 0, sizeof(*tab_));
+    tab_->table[0] = old;
+  }
+  // At this point we have first level table allocated.
+  // Add second level tables as necessary.
+  for (uptr i = RoundUpTo(size_, ClockBlock::kClockCount);
+      i < nclk; i += ClockBlock::kClockCount) {
+    u32 idx = ctx->clock_alloc.Alloc(c);
+    ClockBlock *cb = ctx->clock_alloc.Map(idx);
+    internal_memset(cb, 0, sizeof(*cb));
+    CHECK_EQ(tab_->table[i/ClockBlock::kClockCount], 0);
+    tab_->table[i/ClockBlock::kClockCount] = idx;
+  }
+  size_ = nclk;
+}
+
 // Sets a single element in the vector clock.
 // This function is called only from weird places like AcquireGlobal.
 void ThreadClock::set(unsigned tid, u64 v) {
@@ -319,28 +367,59 @@ void ThreadClock::DebugDump(int(*printf)(const char *s, ...)) {
 }
 
 SyncClock::SyncClock()
-    : clk_(MBlockClock) {
-  release_store_tid_ = kInvalidTid;
-  release_store_reused_ = 0;
+    : release_store_tid_(kInvalidTid)
+    , release_store_reused_()
+    , tab_()
+    , tab_idx_()
+    , size_() {
   for (uptr i = 0; i < kDirtyTids; i++)
     dirty_tids_[i] = kInvalidTid;
 }
 
-void SyncClock::Reset() {
-  clk_.Reset();
+SyncClock::~SyncClock() {
+  // Reset must be called before dtor.
+  CHECK_EQ(size_, 0);
+  CHECK_EQ(tab_, 0);
+  CHECK_EQ(tab_idx_, 0);
+}
+
+void SyncClock::Reset(ClockCache *c) {
+  if (size_ == 0) {
+    // nothing
+  } else if (size_ <= ClockBlock::kClockCount) {
+    // One-level table.
+    ctx->clock_alloc.Free(c, tab_idx_);
+  } else {
+    // Two-level table.
+    for (uptr i = 0; i < size_; i += ClockBlock::kClockCount)
+      ctx->clock_alloc.Free(c, tab_->table[i / ClockBlock::kClockCount]);
+    ctx->clock_alloc.Free(c, tab_idx_);
+  }
+  tab_ = 0;
+  tab_idx_ = 0;
+  size_ = 0;
   release_store_tid_ = kInvalidTid;
   release_store_reused_ = 0;
   for (uptr i = 0; i < kDirtyTids; i++)
     dirty_tids_[i] = kInvalidTid;
 }
 
+ClockElem &SyncClock::elem(unsigned tid) const {
+  DCHECK_LT(tid, size_);
+  if (size_ <= ClockBlock::kClockCount)
+    return tab_->clock[tid];
+  u32 idx = tab_->table[tid / ClockBlock::kClockCount];
+  ClockBlock *cb = ctx->clock_alloc.Map(idx);
+  return cb->clock[tid % ClockBlock::kClockCount];
+}
+
 void SyncClock::DebugDump(int(*printf)(const char *s, ...)) {
   printf("clock=[");
-  for (uptr i = 0; i < clk_.Size(); i++)
-    printf("%s%llu", i == 0 ? "" : ",", clk_[i].epoch);
+  for (uptr i = 0; i < size_; i++)
+    printf("%s%llu", i == 0 ? "" : ",", elem(i).epoch);
   printf("] reused=[");
-  for (uptr i = 0; i < clk_.Size(); i++)
-    printf("%s%llu", i == 0 ? "" : ",", clk_[i].reused);
+  for (uptr i = 0; i < size_; i++)
+    printf("%s%llu", i == 0 ? "" : ",", elem(i).reused);
   printf("] release_store_tid=%d/%d dirty_tids=%d/%d",
       release_store_tid_, release_store_reused_,
       dirty_tids_[0], dirty_tids_[1]);
diff --git a/libsanitizer/tsan/tsan_clock.h b/libsanitizer/tsan/tsan_clock.h
index 2ce480b2d16..3deb7f5198c 100644
--- a/libsanitizer/tsan/tsan_clock.h
+++ b/libsanitizer/tsan/tsan_clock.h
@@ -12,7 +12,7 @@
 #define TSAN_CLOCK_H
 
 #include "tsan_defs.h"
-#include "tsan_vector.h"
+#include "tsan_dense_alloc.h"
 
 namespace __tsan {
 
@@ -21,36 +21,65 @@ struct ClockElem {
   u64 reused : 64 - kClkBits;
 };
 
+struct ClockBlock {
+  static const uptr kSize = 512;
+  static const uptr kTableSize = kSize / sizeof(u32);
+  static const uptr kClockCount = kSize / sizeof(ClockElem);
+
+  union {
+    u32       table[kTableSize];
+    ClockElem clock[kClockCount];
+  };
+
+  ClockBlock() {
+  }
+};
+
+typedef DenseSlabAlloc<ClockBlock, 1<<16, 1<<10> ClockAlloc;
+typedef DenseSlabAllocCache ClockCache;
+
 // The clock that lives in sync variables (mutexes, atomics, etc).
 class SyncClock {
  public:
   SyncClock();
+  ~SyncClock();
 
   uptr size() const {
-    return clk_.Size();
+    return size_;
   }
 
   u64 get(unsigned tid) const {
-    DCHECK_LT(tid, clk_.Size());
-    return clk_[tid].epoch;
+    return elem(tid).epoch;
   }
 
-  void Reset();
+  void Resize(ClockCache *c, uptr nclk);
+  void Reset(ClockCache *c);
 
   void DebugDump(int(*printf)(const char *s, ...));
 
  private:
+  friend struct ThreadClock;
+  static const uptr kDirtyTids = 2;
+
   unsigned release_store_tid_;
   unsigned release_store_reused_;
-  static const uptr kDirtyTids = 2;
   unsigned dirty_tids_[kDirtyTids];
-  mutable Vector<ClockElem> clk_;
-  friend struct ThreadClock;
+  // tab_ contains indirect pointer to a 512b block using DenseSlabAlloc.
+  // If size_ <= 64, then tab_ points to an array with 64 ClockElem's.
+  // Otherwise, tab_ points to an array with 128 u32 elements,
+  // each pointing to the second-level 512b block with 64 ClockElem's.
+  ClockBlock *tab_;
+  u32 tab_idx_;
+  u32 size_;
+
+  ClockElem &elem(unsigned tid) const;
 };
 
 // The clock that lives in threads.
 struct ThreadClock {
  public:
+  typedef DenseSlabAllocCache Cache;
+
   explicit ThreadClock(unsigned tid, unsigned reused = 0);
 
   u64 get(unsigned tid) const {
@@ -73,10 +102,10 @@ struct ThreadClock {
     return nclk_;
   }
 
-  void acquire(const SyncClock *src);
-  void release(SyncClock *dst) const;
-  void acq_rel(SyncClock *dst);
-  void ReleaseStore(SyncClock *dst) const;
+  void acquire(ClockCache *c, const SyncClock *src);
+  void release(ClockCache *c, SyncClock *dst) const;
+  void acq_rel(ClockCache *c, SyncClock *dst);
+  void ReleaseStore(ClockCache *c, SyncClock *dst) const;
 
   void DebugReset();
   void DebugDump(int(*printf)(const char *s, ...));
diff --git a/libsanitizer/tsan/tsan_defs.h b/libsanitizer/tsan/tsan_defs.h
index cc65ae8b4a3..a8528cbb44a 100644
--- a/libsanitizer/tsan/tsan_defs.h
+++ b/libsanitizer/tsan/tsan_defs.h
@@ -52,6 +52,7 @@ const uptr kShadowCnt = TSAN_SHADOW_COUNT;
 # endif
 #else
 // Count of shadow values in a shadow cell.
+#define TSAN_SHADOW_COUNT 4
 const uptr kShadowCnt = 4;
 #endif
 
@@ -64,6 +65,13 @@ const uptr kShadowSize = 8;
 // Shadow memory is kShadowMultiplier times larger than user memory.
 const uptr kShadowMultiplier = kShadowSize * kShadowCnt / kShadowCell;
 
+// That many user bytes are mapped onto a single meta shadow cell.
+// Must be less or equal to minimal memory allocator alignment.
+const uptr kMetaShadowCell = 8;
+
+// Size of a single meta shadow value (u32).
+const uptr kMetaShadowSize = 4;
+
 #if defined(TSAN_NO_HISTORY) && TSAN_NO_HISTORY
 const bool kCollectHistory = false;
 #else
@@ -165,7 +173,15 @@ struct ReportStack;
 class ReportDesc;
 class RegionAlloc;
 class StackTrace;
-struct MBlock;
+
+// Descriptor of user's memory block.
+struct MBlock {
+  u64  siz;
+  u32  stk;
+  u16  tid;
+};
+
+COMPILER_CHECK(sizeof(MBlock) == 16);
 
 }  // namespace __tsan
 
diff --git a/libsanitizer/tsan/tsan_dense_alloc.h b/libsanitizer/tsan/tsan_dense_alloc.h
new file mode 100644
index 00000000000..651c112c78a
--- /dev/null
+++ b/libsanitizer/tsan/tsan_dense_alloc.h
@@ -0,0 +1,135 @@
+//===-- tsan_dense_alloc.h --------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+// A DenseSlabAlloc is a freelist-based allocator of fixed-size objects.
+// DenseSlabAllocCache is a thread-local cache for DenseSlabAlloc.
+// The only difference with traditional slab allocators is that DenseSlabAlloc
+// allocates/free indices of objects and provide a functionality to map
+// the index onto the real pointer. The index is u32, that is, 2 times smaller
+// than uptr (hense the Dense prefix).
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_DENSE_ALLOC_H
+#define TSAN_DENSE_ALLOC_H
+
+#include "sanitizer_common/sanitizer_common.h"
+#include "tsan_defs.h"
+#include "tsan_mutex.h"
+
+namespace __tsan {
+
+class DenseSlabAllocCache {
+  static const uptr kSize = 128;
+  typedef u32 IndexT;
+  uptr pos;
+  IndexT cache[kSize];
+  template<typename T, uptr kL1Size, uptr kL2Size> friend class DenseSlabAlloc;
+};
+
+template<typename T, uptr kL1Size, uptr kL2Size>
+class DenseSlabAlloc {
+ public:
+  typedef DenseSlabAllocCache Cache;
+  typedef typename Cache::IndexT IndexT;
+
+  DenseSlabAlloc() {
+    // Check that kL1Size and kL2Size are sane.
+    CHECK_EQ(kL1Size & (kL1Size - 1), 0);
+    CHECK_EQ(kL2Size & (kL2Size - 1), 0);
+    CHECK_GE(1ull << (sizeof(IndexT) * 8), kL1Size * kL2Size);
+    // Check that it makes sense to use the dense alloc.
+    CHECK_GE(sizeof(T), sizeof(IndexT));
+    internal_memset(map_, 0, sizeof(map_));
+    freelist_ = 0;
+    fillpos_ = 0;
+  }
+
+  ~DenseSlabAlloc() {
+    for (uptr i = 0; i < kL1Size; i++) {
+      if (map_[i] != 0)
+        UnmapOrDie(map_[i], kL2Size * sizeof(T));
+    }
+  }
+
+  IndexT Alloc(Cache *c) {
+    if (c->pos == 0)
+      Refill(c);
+    return c->cache[--c->pos];
+  }
+
+  void Free(Cache *c, IndexT idx) {
+    DCHECK_NE(idx, 0);
+    if (c->pos == Cache::kSize)
+      Drain(c);
+    c->cache[c->pos++] = idx;
+  }
+
+  T *Map(IndexT idx) {
+    DCHECK_NE(idx, 0);
+    DCHECK_LE(idx, kL1Size * kL2Size);
+    return &map_[idx / kL2Size][idx % kL2Size];
+  }
+
+  void FlushCache(Cache *c) {
+    SpinMutexLock lock(&mtx_);
+    while (c->pos) {
+      IndexT idx = c->cache[--c->pos];
+      *(IndexT*)Map(idx) = freelist_;
+      freelist_ = idx;
+    }
+  }
+
+  void InitCache(Cache *c) {
+    c->pos = 0;
+    internal_memset(c->cache, 0, sizeof(c->cache));
+  }
+
+ private:
+  T *map_[kL1Size];
+  SpinMutex mtx_;
+  IndexT freelist_;
+  uptr fillpos_;
+
+  void Refill(Cache *c) {
+    SpinMutexLock lock(&mtx_);
+    if (freelist_ == 0) {
+      if (fillpos_ == kL1Size) {
+        Printf("ThreadSanitizer: DenseSlabAllocator overflow. Dying.\n");
+        Die();
+      }
+      T *batch = (T*)MmapOrDie(kL2Size * sizeof(T), "DenseSlabAllocator");
+      // Reserve 0 as invalid index.
+      IndexT start = fillpos_ == 0 ? 1 : 0;
+      for (IndexT i = start; i < kL2Size; i++) {
+        new(batch + i) T();
+        *(IndexT*)(batch + i) = i + 1 + fillpos_ * kL2Size;
+      }
+      *(IndexT*)(batch + kL2Size - 1) = 0;
+      freelist_ = fillpos_ * kL2Size + start;
+      map_[fillpos_++] = batch;
+    }
+    for (uptr i = 0; i < Cache::kSize / 2 && freelist_ != 0; i++) {
+      IndexT idx = freelist_;
+      c->cache[c->pos++] = idx;
+      freelist_ = *(IndexT*)Map(idx);
+    }
+  }
+
+  void Drain(Cache *c) {
+    SpinMutexLock lock(&mtx_);
+    for (uptr i = 0; i < Cache::kSize / 2; i++) {
+      IndexT idx = c->cache[--c->pos];
+      *(IndexT*)Map(idx) = freelist_;
+      freelist_ = idx;
+    }
+  }
+};
+
+}  // namespace __tsan
+
+#endif  // TSAN_DENSE_ALLOC_H
diff --git a/libsanitizer/tsan/tsan_fd.cc b/libsanitizer/tsan/tsan_fd.cc
index 7d62ae4c5ff..a74a668c13b 100644
--- a/libsanitizer/tsan/tsan_fd.cc
+++ b/libsanitizer/tsan/tsan_fd.cc
@@ -45,8 +45,8 @@ static bool bogusfd(int fd) {
   return fd < 0 || fd >= kTableSize;
 }
 
-static FdSync *allocsync() {
-  FdSync *s = (FdSync*)internal_alloc(MBlockFD, sizeof(FdSync));
+static FdSync *allocsync(ThreadState *thr, uptr pc) {
+  FdSync *s = (FdSync*)user_alloc(thr, pc, sizeof(FdSync));
   atomic_store(&s->rc, 1, memory_order_relaxed);
   return s;
 }
@@ -63,10 +63,7 @@ static void unref(ThreadState *thr, uptr pc, FdSync *s) {
       CHECK_NE(s, &fdctx.globsync);
       CHECK_NE(s, &fdctx.filesync);
       CHECK_NE(s, &fdctx.socksync);
-      SyncVar *v = ctx->synctab.GetAndRemove(thr, pc, (uptr)s);
-      if (v)
-        DestroyAndFree(v);
-      internal_free(s);
+      user_free(thr, pc, s);
     }
   }
 }
@@ -217,7 +214,7 @@ void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd) {
 
 void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd) {
   DPrintf("#%d: FdCreatePipe(%d, %d)\n", thr->tid, rfd, wfd);
-  FdSync *s = allocsync();
+  FdSync *s = allocsync(thr, pc);
   init(thr, pc, rfd, ref(s));
   init(thr, pc, wfd, ref(s));
   unref(thr, pc, s);
@@ -227,7 +224,7 @@ void FdEventCreate(ThreadState *thr, uptr pc, int fd) {
   DPrintf("#%d: FdEventCreate(%d)\n", thr->tid, fd);
   if (bogusfd(fd))
     return;
-  init(thr, pc, fd, allocsync());
+  init(thr, pc, fd, allocsync(thr, pc));
 }
 
 void FdSignalCreate(ThreadState *thr, uptr pc, int fd) {
@@ -248,7 +245,7 @@ void FdPollCreate(ThreadState *thr, uptr pc, int fd) {
   DPrintf("#%d: FdPollCreate(%d)\n", thr->tid, fd);
   if (bogusfd(fd))
     return;
-  init(thr, pc, fd, allocsync());
+  init(thr, pc, fd, allocsync(thr, pc));
 }
 
 void FdSocketCreate(ThreadState *thr, uptr pc, int fd) {
diff --git a/libsanitizer/tsan/tsan_flags.cc b/libsanitizer/tsan/tsan_flags.cc
index e241cb6cda4..02bf6527e6a 100644
--- a/libsanitizer/tsan/tsan_flags.cc
+++ b/libsanitizer/tsan/tsan_flags.cc
@@ -35,7 +35,6 @@ static void ParseFlags(Flags *f, const char *env) {
   ParseFlag(env, &f->enable_annotations, "enable_annotations", "");
   ParseFlag(env, &f->suppress_equal_stacks, "suppress_equal_stacks", "");
   ParseFlag(env, &f->suppress_equal_addresses, "suppress_equal_addresses", "");
-  ParseFlag(env, &f->suppress_java, "suppress_java", "");
   ParseFlag(env, &f->report_bugs, "report_bugs", "");
   ParseFlag(env, &f->report_thread_leaks, "report_thread_leaks", "");
   ParseFlag(env, &f->report_destroy_locked, "report_destroy_locked", "");
@@ -43,8 +42,6 @@ static void ParseFlags(Flags *f, const char *env) {
   ParseFlag(env, &f->report_signal_unsafe, "report_signal_unsafe", "");
   ParseFlag(env, &f->report_atomic_races, "report_atomic_races", "");
   ParseFlag(env, &f->force_seq_cst_atomics, "force_seq_cst_atomics", "");
-  ParseFlag(env, &f->suppressions, "suppressions", "");
-  ParseFlag(env, &f->print_suppressions, "print_suppressions", "");
   ParseFlag(env, &f->print_benign, "print_benign", "");
   ParseFlag(env, &f->exitcode, "exitcode", "");
   ParseFlag(env, &f->halt_on_error, "halt_on_error", "");
@@ -70,7 +67,6 @@ void InitializeFlags(Flags *f, const char *env) {
   f->enable_annotations = true;
   f->suppress_equal_stacks = true;
   f->suppress_equal_addresses = true;
-  f->suppress_java = false;
   f->report_bugs = true;
   f->report_thread_leaks = true;
   f->report_destroy_locked = true;
@@ -78,8 +74,6 @@ void InitializeFlags(Flags *f, const char *env) {
   f->report_signal_unsafe = true;
   f->report_atomic_races = true;
   f->force_seq_cst_atomics = false;
-  f->suppressions = "";
-  f->print_suppressions = false;
   f->print_benign = false;
   f->exitcode = 66;
   f->halt_on_error = false;
@@ -97,19 +91,19 @@ void InitializeFlags(Flags *f, const char *env) {
   // DDFlags
   f->second_deadlock_stack = false;
 
-  SetCommonFlagsDefaults(f);
+  CommonFlags *cf = common_flags();
+  SetCommonFlagsDefaults(cf);
   // Override some common flags defaults.
-  f->allow_addr2line = true;
+  cf->allow_addr2line = true;
+  cf->detect_deadlocks = true;
+  cf->print_suppressions = false;
 
   // Let a frontend override.
   ParseFlags(f, __tsan_default_options());
-  ParseCommonFlagsFromString(f, __tsan_default_options());
+  ParseCommonFlagsFromString(cf, __tsan_default_options());
   // Override from command line.
   ParseFlags(f, env);
-  ParseCommonFlagsFromString(f, env);
-
-  // Copy back to common flags.
-  *common_flags() = *f;
+  ParseCommonFlagsFromString(cf, env);
 
   // Sanity check.
   if (!f->report_bugs) {
@@ -118,7 +112,7 @@ void InitializeFlags(Flags *f, const char *env) {
     f->report_signal_unsafe = false;
   }
 
-  if (f->help) PrintFlagDescriptions();
+  if (cf->help) PrintFlagDescriptions();
 
   if (f->history_size < 0 || f->history_size > 7) {
     Printf("ThreadSanitizer: incorrect value for history_size"
diff --git a/libsanitizer/tsan/tsan_flags.h b/libsanitizer/tsan/tsan_flags.h
index 4bf459d5981..182d9b298af 100644
--- a/libsanitizer/tsan/tsan_flags.h
+++ b/libsanitizer/tsan/tsan_flags.h
@@ -17,7 +17,7 @@
 
 namespace __tsan {
 
-struct Flags : CommonFlags, DDFlags {
+struct Flags : DDFlags {
   // Enable dynamic annotations, otherwise they are no-ops.
   bool enable_annotations;
   // Suppress a race report if we've already output another race report
@@ -26,9 +26,6 @@ struct Flags : CommonFlags, DDFlags {
   // Suppress a race report if we've already output another race report
   // on the same address.
   bool suppress_equal_addresses;
-  // Suppress weird race reports that can be seen if JVM is embed
-  // into the process.
-  bool suppress_java;
   // Turns off bug reporting entirely (useful for benchmarking).
   bool report_bugs;
   // Report thread leaks at exit?
@@ -45,10 +42,6 @@ struct Flags : CommonFlags, DDFlags {
   // If set, all atomics are effectively sequentially consistent (seq_cst),
   // regardless of what user actually specified.
   bool force_seq_cst_atomics;
-  // Suppressions filename.
-  const char *suppressions;
-  // Print matched suppressions at exit.
-  bool print_suppressions;
   // Print matched "benign" races at exit.
   bool print_benign;
   // Override exit status if something was reported.
diff --git a/libsanitizer/tsan/tsan_interceptors.cc b/libsanitizer/tsan/tsan_interceptors.cc
index 19a3b7b0643..b49622b3ad4 100644
--- a/libsanitizer/tsan/tsan_interceptors.cc
+++ b/libsanitizer/tsan/tsan_interceptors.cc
@@ -45,7 +45,7 @@ DECLARE_REAL(int, pthread_attr_getdetachstate, void *, void *)
 extern "C" int pthread_attr_setstacksize(void *attr, uptr stacksize);
 extern "C" int pthread_key_create(unsigned *key, void (*destructor)(void* v));
 extern "C" int pthread_setspecific(unsigned key, const void *v);
-extern "C" int pthread_mutexattr_gettype(void *a, int *type);
+DECLARE_REAL(int, pthread_mutexattr_gettype, void *, void *)
 extern "C" int pthread_yield();
 extern "C" int pthread_sigmask(int how, const __sanitizer_sigset_t *set,
                                __sanitizer_sigset_t *oldset);
@@ -91,6 +91,11 @@ typedef void (*sighandler_t)(int sig);
 
 #define errno (*__errno_location())
 
+// 16K loaded modules should be enough for everyone.
+static const uptr kMaxModules = 1 << 14;
+static LoadedModule *modules;
+static uptr nmodules;
+
 struct sigaction_t {
   union {
     sighandler_t sa_handler;
@@ -122,9 +127,9 @@ struct SignalDesc {
 };
 
 struct SignalContext {
-  int in_blocking_func;
   int int_signal_send;
-  int pending_signal_count;
+  atomic_uintptr_t in_blocking_func;
+  atomic_uintptr_t have_pending_signals;
   SignalDesc pending_signals[kSigCount];
 };
 
@@ -136,7 +141,7 @@ static LibIgnore *libignore() {
 }
 
 void InitializeLibIgnore() {
-  libignore()->Init(*GetSuppressionContext());
+  libignore()->Init(*SuppressionContext::Get());
   libignore()->OnLibraryLoaded(0);
 }
 
@@ -144,7 +149,7 @@ void InitializeLibIgnore() {
 
 static SignalContext *SigCtx(ThreadState *thr) {
   SignalContext *ctx = (SignalContext*)thr->signal_ctx;
-  if (ctx == 0 && thr->is_alive) {
+  if (ctx == 0 && !thr->is_dead) {
     ctx = (SignalContext*)MmapOrDie(sizeof(*ctx), "SignalContext");
     MemoryResetRange(thr, (uptr)&SigCtx, (uptr)ctx, sizeof(*ctx));
     thr->signal_ctx = ctx;
@@ -189,6 +194,7 @@ ScopedInterceptor::~ScopedInterceptor() {
   if (!thr_->ignore_interceptors) {
     ProcessPendingSignals(thr_);
     FuncExit(thr_);
+    CheckNoLocks(thr_);
   }
 }
 
@@ -205,7 +211,7 @@ ScopedInterceptor::~ScopedInterceptor() {
     if (REAL(func) == 0) { \
       Report("FATAL: ThreadSanitizer: failed to intercept %s\n", #func); \
       Die(); \
-    } \
+    }                                                    \
     if (thr->ignore_interceptors || thr->in_ignored_lib) \
       return REAL(func)(__VA_ARGS__); \
 /**/
@@ -218,22 +224,30 @@ ScopedInterceptor::~ScopedInterceptor() {
 
 struct BlockingCall {
   explicit BlockingCall(ThreadState *thr)
-      : ctx(SigCtx(thr)) {
-    ctx->in_blocking_func++;
+      : thr(thr)
+      , ctx(SigCtx(thr)) {
+    for (;;) {
+      atomic_store(&ctx->in_blocking_func, 1, memory_order_relaxed);
+      if (atomic_load(&ctx->have_pending_signals, memory_order_relaxed) == 0)
+        break;
+      atomic_store(&ctx->in_blocking_func, 0, memory_order_relaxed);
+      ProcessPendingSignals(thr);
+    }
+    // When we are in a "blocking call", we process signals asynchronously
+    // (right when they arrive). In this context we do not expect to be
+    // executing any user/runtime code. The known interceptor sequence when
+    // this is not true is: pthread_join -> munmap(stack). It's fine
+    // to ignore munmap in this case -- we handle stack shadow separately.
+    thr->ignore_interceptors++;
   }
 
   ~BlockingCall() {
-    ctx->in_blocking_func--;
+    thr->ignore_interceptors--;
+    atomic_store(&ctx->in_blocking_func, 0, memory_order_relaxed);
   }
 
+  ThreadState *thr;
   SignalContext *ctx;
-
-  // When we are in a "blocking call", we process signals asynchronously
-  // (right when they arrive). In this context we do not expect to be
-  // executing any user/runtime code. The known interceptor sequence when
-  // this is not true is: pthread_join -> munmap(stack). It's fine
-  // to ignore munmap in this case -- we handle stack shadow separately.
-  ScopedIgnoreInterceptors ignore_interceptors;
 };
 
 TSAN_INTERCEPTOR(unsigned, sleep, unsigned sec) {
@@ -257,74 +271,60 @@ TSAN_INTERCEPTOR(int, nanosleep, void *req, void *rem) {
   return res;
 }
 
-TSAN_INTERCEPTOR(void*, dlopen, const char *filename, int flag) {
-  SCOPED_INTERCEPTOR_RAW(dlopen, filename, flag);
-  void *res = REAL(dlopen)(filename, flag);
-  libignore()->OnLibraryLoaded(filename);
-  return res;
-}
-
-TSAN_INTERCEPTOR(int, dlclose, void *handle) {
-  SCOPED_INTERCEPTOR_RAW(dlclose, handle);
-  int res = REAL(dlclose)(handle);
-  libignore()->OnLibraryUnloaded();
-  return res;
-}
-
 class AtExitContext {
  public:
   AtExitContext()
     : mtx_(MutexTypeAtExit, StatMtxAtExit)
-    , pos_() {
+    , stack_(MBlockAtExit) {
   }
 
-  typedef void(*atexit_t)();
+  typedef void(*atexit_cb_t)();
 
   int atexit(ThreadState *thr, uptr pc, bool is_on_exit,
-             atexit_t f, void *arg) {
+             atexit_cb_t f, void *arg, void *dso) {
     Lock l(&mtx_);
-    if (pos_ == kMaxAtExit)
-      return 1;
     Release(thr, pc, (uptr)this);
-    stack_[pos_] = f;
-    args_[pos_] = arg;
-    is_on_exits_[pos_] = is_on_exit;
-    pos_++;
+    atexit_t *a = stack_.PushBack();
+    a->cb = f;
+    a->arg = arg;
+    a->dso = dso;
+    a->is_on_exit = is_on_exit;
     return 0;
   }
 
   void exit(ThreadState *thr, uptr pc) {
     for (;;) {
-      atexit_t f = 0;
-      void *arg = 0;
-      bool is_on_exit = false;
+      atexit_t a = {};
       {
         Lock l(&mtx_);
-        if (pos_) {
-          pos_--;
-          f = stack_[pos_];
-          arg = args_[pos_];
-          is_on_exit = is_on_exits_[pos_];
+        if (stack_.Size() != 0) {
+          a = stack_[stack_.Size() - 1];
+          stack_.PopBack();
           Acquire(thr, pc, (uptr)this);
         }
       }
-      if (f == 0)
+      if (a.cb == 0)
         break;
-      DPrintf("#%d: executing atexit func %p\n", thr->tid, f);
-      if (is_on_exit)
-        ((void(*)(int status, void *arg))f)(0, arg);
+      VPrintf(2, "#%d: executing atexit func %p(%p) dso=%p\n",
+          thr->tid, a.cb, a.arg, a.dso);
+      if (a.is_on_exit)
+        ((void(*)(int status, void *arg))a.cb)(0, a.arg);
       else
-        ((void(*)(void *arg, void *dso))f)(arg, 0);
+        ((void(*)(void *arg, void *dso))a.cb)(a.arg, a.dso);
     }
   }
 
  private:
-  static const int kMaxAtExit = 128;
+  struct atexit_t {
+    atexit_cb_t cb;
+    void *arg;
+    void *dso;
+    bool is_on_exit;
+  };
+
+  static const int kMaxAtExit = 1024;
   Mutex mtx_;
-  atexit_t stack_[kMaxAtExit];
-  void *args_[kMaxAtExit];
-  bool is_on_exits_[kMaxAtExit];
-  int pos_;
+  Vector<atexit_t> stack_;
 };
 
 static AtExitContext *atexit_ctx;
@@ -335,29 +335,42 @@ TSAN_INTERCEPTOR(int, atexit, void (*f)()) {
   // We want to setup the atexit callback even if we are in ignored lib
   // or after fork.
   SCOPED_INTERCEPTOR_RAW(atexit, f);
-  return atexit_ctx->atexit(thr, pc, false, (void(*)())f, 0);
+  return atexit_ctx->atexit(thr, pc, false, (void(*)())f, 0, 0);
 }
 
 TSAN_INTERCEPTOR(int, on_exit, void(*f)(int, void*), void *arg) {
   if (cur_thread()->in_symbolizer)
     return 0;
   SCOPED_TSAN_INTERCEPTOR(on_exit, f, arg);
-  return atexit_ctx->atexit(thr, pc, true, (void(*)())f, arg);
+  return atexit_ctx->atexit(thr, pc, true, (void(*)())f, arg, 0);
+}
+
+bool IsSaticModule(void *dso) {
+  if (modules == 0)
+    return false;
+  for (uptr i = 0; i < nmodules; i++) {
+    if (modules[i].containsAddress((uptr)dso))
+      return true;
+  }
+  return false;
 }
 
 TSAN_INTERCEPTOR(int, __cxa_atexit, void (*f)(void *a), void *arg, void *dso) {
   if (cur_thread()->in_symbolizer)
     return 0;
   SCOPED_TSAN_INTERCEPTOR(__cxa_atexit, f, arg, dso);
-  if (dso) {
-    // Memory allocation in __cxa_atexit will race with free during exit,
-    // because we do not see synchronization around atexit callback list.
-    ThreadIgnoreBegin(thr, pc);
-    int res = REAL(__cxa_atexit)(f, arg, dso);
-    ThreadIgnoreEnd(thr, pc);
-    return res;
-  }
-  return atexit_ctx->atexit(thr, pc, false, (void(*)())f, arg);
+  // If it's the main executable or a statically loaded library,
+  // we will call the callback.
+  if (dso == 0 || IsSaticModule(dso))
+    return atexit_ctx->atexit(thr, pc, false, (void(*)())f, arg, dso);
+
+  // Dynamically load module, don't know when to call the callback for it.
+  // Memory allocation in __cxa_atexit will race with free during exit,
+  // because we do not see synchronization around atexit callback list.
+  ThreadIgnoreBegin(thr, pc);
+  int res = REAL(__cxa_atexit)(f, arg, dso);
+  ThreadIgnoreEnd(thr, pc);
+  return res;
 }
 
 // Cleanup old bufs.
@@ -383,6 +396,13 @@ static void SetJmp(ThreadState *thr, uptr sp, uptr mangled_sp) {
   buf->sp = sp;
   buf->mangled_sp = mangled_sp;
   buf->shadow_stack_pos = thr->shadow_stack_pos;
+  SignalContext *sctx = SigCtx(thr);
+  buf->int_signal_send = sctx ? sctx->int_signal_send : 0;
+  buf->in_blocking_func = sctx ?
+      atomic_load(&sctx->in_blocking_func, memory_order_relaxed) :
+      false;
+  buf->in_signal_handler = atomic_load(&thr->in_signal_handler,
+      memory_order_relaxed);
 }
 
 static void LongJmp(ThreadState *thr, uptr *env) {
@@ -395,6 +415,14 @@ static void LongJmp(ThreadState *thr, uptr *env) {
       // Unwind the stack.
       while (thr->shadow_stack_pos > buf->shadow_stack_pos)
         FuncExit(thr);
+      SignalContext *sctx = SigCtx(thr);
+      if (sctx) {
+        sctx->int_signal_send = buf->int_signal_send;
+        atomic_store(&sctx->in_blocking_func, buf->in_blocking_func,
+            memory_order_relaxed);
+      }
+      atomic_store(&thr->in_signal_handler, buf->in_signal_handler,
+          memory_order_relaxed);
       JmpBufGarbageCollect(thr, buf->sp - 1);  // do not collect buf->sp
       return;
     }
@@ -531,7 +559,7 @@ TSAN_INTERCEPTOR(void, cfree, void *p) {
 
 TSAN_INTERCEPTOR(uptr, malloc_usable_size, void *p) {
   SCOPED_INTERCEPTOR_RAW(malloc_usable_size, p);
-  return user_alloc_usable_size(thr, pc, p);
+  return user_alloc_usable_size(p);
 }
 
 #define OPERATOR_NEW_BODY(mangled_name) \
@@ -749,6 +777,11 @@ TSAN_INTERCEPTOR(void*, memalign, uptr align, uptr sz) {
   return user_alloc(thr, pc, sz, align);
 }
 
+TSAN_INTERCEPTOR(void*, aligned_alloc, uptr align, uptr sz) {
+  SCOPED_INTERCEPTOR_RAW(memalign, align, sz);
+  return user_alloc(thr, pc, sz, align);
+}
+
 TSAN_INTERCEPTOR(void*, valloc, uptr sz) {
   SCOPED_INTERCEPTOR_RAW(valloc, sz);
   return user_alloc(thr, pc, sz, GetPageSizeCached());
@@ -1040,7 +1073,7 @@ TSAN_INTERCEPTOR(int, pthread_mutex_init, void *m, void *a) {
     bool recursive = false;
     if (a) {
       int type = 0;
-      if (pthread_mutexattr_gettype(a, &type) == 0)
+      if (REAL(pthread_mutexattr_gettype)(a, &type) == 0)
         recursive = (type == PTHREAD_MUTEX_RECURSIVE
             || type == PTHREAD_MUTEX_RECURSIVE_NP);
     }
@@ -1151,7 +1184,7 @@ TSAN_INTERCEPTOR(int, pthread_rwlock_tryrdlock, void *m) {
   SCOPED_TSAN_INTERCEPTOR(pthread_rwlock_tryrdlock, m);
   int res = REAL(pthread_rwlock_tryrdlock)(m);
   if (res == 0) {
-    MutexLock(thr, pc, (uptr)m, /*rec=*/1, /*try_lock=*/true);
+    MutexReadLock(thr, pc, (uptr)m, /*try_lock=*/true);
   }
   return res;
 }
@@ -1689,8 +1722,10 @@ TSAN_INTERCEPTOR(int, epoll_wait, int epfd, void *ev, int cnt, int timeout) {
 
 namespace __tsan {
 
-static void CallUserSignalHandler(ThreadState *thr, bool sync, bool sigact,
-    int sig, my_siginfo_t *info, void *uctx) {
+static void CallUserSignalHandler(ThreadState *thr, bool sync, bool acquire,
+    bool sigact, int sig, my_siginfo_t *info, void *uctx) {
+  if (acquire)
+    Acquire(thr, 0, (uptr)&sigactions[sig]);
   // Ensure that the handler does not spoil errno.
   const int saved_errno = errno;
   errno = 99;
@@ -1716,8 +1751,8 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool sigact,
     ThreadRegistryLock l(ctx->thread_registry);
     ScopedReport rep(ReportTypeErrnoInSignal);
     if (!IsFiredSuppression(ctx, rep, stack)) {
-      rep.AddStack(&stack);
-      OutputReport(ctx, rep, rep.GetReport()->stacks[0]);
+      rep.AddStack(&stack, true);
+      OutputReport(thr, rep);
     }
   }
   errno = saved_errno;
@@ -1725,10 +1760,11 @@ static void CallUserSignalHandler(ThreadState *thr, bool sync, bool sigact,
 
 void ProcessPendingSignals(ThreadState *thr) {
   SignalContext *sctx = SigCtx(thr);
-  if (sctx == 0 || sctx->pending_signal_count == 0 || thr->in_signal_handler)
+  if (sctx == 0 ||
+      atomic_load(&sctx->have_pending_signals, memory_order_relaxed) == 0)
     return;
-  thr->in_signal_handler = true;
-  sctx->pending_signal_count = 0;
+  atomic_store(&sctx->have_pending_signals, 0, memory_order_relaxed);
+  atomic_fetch_add(&thr->in_signal_handler, 1, memory_order_relaxed);
   // These are too big for stack.
   static THREADLOCAL __sanitizer_sigset_t emptyset, oldset;
   REAL(sigfillset)(&emptyset);
@@ -1739,14 +1775,13 @@ void ProcessPendingSignals(ThreadState *thr) {
       signal->armed = false;
       if (sigactions[sig].sa_handler != SIG_DFL
           && sigactions[sig].sa_handler != SIG_IGN) {
-        CallUserSignalHandler(thr, false, signal->sigaction,
+        CallUserSignalHandler(thr, false, true, signal->sigaction,
             sig, &signal->siginfo, &signal->ctx);
       }
     }
   }
   pthread_sigmask(SIG_SETMASK, &oldset, 0);
-  CHECK_EQ(thr->in_signal_handler, true);
-  thr->in_signal_handler = false;
+  atomic_fetch_add(&thr->in_signal_handler, -1, memory_order_relaxed);
 }
 
 }  // namespace __tsan
@@ -1772,21 +1807,27 @@ void ALWAYS_INLINE rtl_generic_sighandler(bool sigact, int sig,
       // If we are in blocking function, we can safely process it now
       // (but check if we are in a recursive interceptor,
       // i.e. pthread_join()->munmap()).
-      (sctx && sctx->in_blocking_func == 1)) {
-    CHECK_EQ(thr->in_signal_handler, false);
-    thr->in_signal_handler = true;
-    if (sctx && sctx->in_blocking_func == 1) {
+      (sctx && atomic_load(&sctx->in_blocking_func, memory_order_relaxed))) {
+    atomic_fetch_add(&thr->in_signal_handler, 1, memory_order_relaxed);
+    if (sctx && atomic_load(&sctx->in_blocking_func, memory_order_relaxed)) {
       // We ignore interceptors in blocking functions,
       // temporary enbled them again while we are calling user function.
       int const i = thr->ignore_interceptors;
       thr->ignore_interceptors = 0;
-      CallUserSignalHandler(thr, sync, sigact, sig, info, ctx);
+      atomic_store(&sctx->in_blocking_func, 0, memory_order_relaxed);
+      CallUserSignalHandler(thr, sync, true, sigact, sig, info, ctx);
       thr->ignore_interceptors = i;
+      atomic_store(&sctx->in_blocking_func, 1, memory_order_relaxed);
     } else {
-      CallUserSignalHandler(thr, sync, sigact, sig, info, ctx);
+      // Be very conservative with when we do acquire in this case.
+      // It's unsafe to do acquire in async handlers, because ThreadState
+      // can be in inconsistent state.
+      // SIGSYS looks relatively safe -- it's synchronous and can actually
+      // need some global state.
+      bool acq = (sig == SIGSYS);
+      CallUserSignalHandler(thr, sync, acq, sigact, sig, info, ctx);
     }
-    CHECK_EQ(thr->in_signal_handler, true);
-    thr->in_signal_handler = false;
+    atomic_fetch_add(&thr->in_signal_handler, -1, memory_order_relaxed);
     return;
   }
 
@@ -1800,7 +1841,7 @@ void ALWAYS_INLINE rtl_generic_sighandler(bool sigact, int sig,
       internal_memcpy(&signal->siginfo, info, sizeof(*info));
     if (ctx)
       internal_memcpy(&signal->ctx, ctx, sizeof(signal->ctx));
-    sctx->pending_signal_count++;
+    atomic_store(&sctx->have_pending_signals, 1, memory_order_relaxed);
   }
 }
 
@@ -1828,6 +1869,7 @@ TSAN_INTERCEPTOR(int, sigaction, int sig, sigaction_t *act, sigaction_t *old) {
     else
       newact.sa_handler = rtl_sighandler;
   }
+  ReleaseStore(thr, pc, (uptr)&sigactions[sig]);
   int res = REAL(sigaction)(sig, &newact, 0);
   return res;
 }
@@ -1911,35 +1953,6 @@ TSAN_INTERCEPTOR(int, getaddrinfo, void *node, void *service,
   return res;
 }
 
-// Linux kernel has a bug that leads to kernel deadlock if a process
-// maps TBs of memory and then calls mlock().
-static void MlockIsUnsupported() {
-  static atomic_uint8_t printed;
-  if (atomic_exchange(&printed, 1, memory_order_relaxed))
-    return;
-  VPrintf(1, "INFO: ThreadSanitizer ignores mlock/munlock[all]\n");
-}
-
-TSAN_INTERCEPTOR(int, mlock, const void *addr, uptr len) {
-  MlockIsUnsupported();
-  return 0;
-}
-
-TSAN_INTERCEPTOR(int, munlock, const void *addr, uptr len) {
-  MlockIsUnsupported();
-  return 0;
-}
-
-TSAN_INTERCEPTOR(int, mlockall, int flags) {
-  MlockIsUnsupported();
-  return 0;
-}
-
-TSAN_INTERCEPTOR(int, munlockall, void) {
-  MlockIsUnsupported();
-  return 0;
-}
-
 TSAN_INTERCEPTOR(int, fork, int fake) {
   if (cur_thread()->in_symbolizer)
     return REAL(fork)(fake);
@@ -2000,6 +2013,18 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
 #include "sanitizer_common/sanitizer_platform_interceptors.h"
 // Causes interceptor recursion (getaddrinfo() and fopen())
 #undef SANITIZER_INTERCEPT_GETADDRINFO
+// There interceptors do not seem to be strictly necessary for tsan.
+// But we see cases where the interceptors consume 70% of execution time.
+// Memory blocks passed to fgetgrent_r are "written to" by tsan several times.
+// First, there is some recursion (getgrnam_r calls fgetgrent_r), and each
+// function "writes to" the buffer. Then, the same memory is "written to"
+// twice, first as buf and then as pwbufp (both of them refer to the same
+// addresses).
+#undef SANITIZER_INTERCEPT_GETPWENT
+#undef SANITIZER_INTERCEPT_GETPWENT_R
+#undef SANITIZER_INTERCEPT_FGETPWENT
+#undef SANITIZER_INTERCEPT_GETPWNAM_AND_FRIENDS
+#undef SANITIZER_INTERCEPT_GETPWNAM_R_AND_FRIENDS
 
 #define COMMON_INTERCEPT_FUNCTION(name) INTERCEPT_FUNCTION(name)
 
@@ -2019,6 +2044,12 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
   ctx = (void *)&_ctx;                                \
   (void) ctx;
 
+#define COMMON_INTERCEPTOR_ENTER_NOIGNORE(ctx, func, ...) \
+  SCOPED_INTERCEPTOR_RAW(func, __VA_ARGS__);              \
+  TsanInterceptorContext _ctx = {thr, caller_pc, pc};     \
+  ctx = (void *)&_ctx;                                    \
+  (void) ctx;
+
 #define COMMON_INTERCEPTOR_FILE_OPEN(ctx, file, path) \
   Acquire(thr, pc, File2addr(path));                  \
   if (file) {                                         \
@@ -2032,6 +2063,12 @@ static void HandleRecvmsg(ThreadState *thr, uptr pc,
     if (fd >= 0) FdClose(thr, pc, fd);           \
   }
 
+#define COMMON_INTERCEPTOR_LIBRARY_LOADED(filename, res)  \
+  libignore()->OnLibraryLoaded(filename)
+
+#define COMMON_INTERCEPTOR_LIBRARY_UNLOADED() \
+  libignore()->OnLibraryUnloaded()
+
 #define COMMON_INTERCEPTOR_FD_ACQUIRE(ctx, fd) \
   FdAcquire(((TsanInterceptorContext *) ctx)->thr, pc, fd)
 
@@ -2367,15 +2404,8 @@ void InitializeInterceptors() {
   TSAN_INTERCEPT(gettimeofday);
   TSAN_INTERCEPT(getaddrinfo);
 
-  TSAN_INTERCEPT(mlock);
-  TSAN_INTERCEPT(munlock);
-  TSAN_INTERCEPT(mlockall);
-  TSAN_INTERCEPT(munlockall);
-
   TSAN_INTERCEPT(fork);
   TSAN_INTERCEPT(vfork);
-  TSAN_INTERCEPT(dlopen);
-  TSAN_INTERCEPT(dlclose);
   TSAN_INTERCEPT(on_exit);
   TSAN_INTERCEPT(__cxa_atexit);
   TSAN_INTERCEPT(_exit);
@@ -2397,6 +2427,11 @@ void InitializeInterceptors() {
   }
 
   FdInit();
+
+  // Remember list of loaded libraries for atexit interceptors.
+  modules = (LoadedModule*)MmapOrDie(sizeof(*modules)*kMaxModules,
+      "LoadedModule");
+  nmodules = GetListOfModules(modules, kMaxModules, 0);
 }
 
 void *internal_start_thread(void(*func)(void *arg), void *arg) {
diff --git a/libsanitizer/tsan/tsan_interface_ann.cc b/libsanitizer/tsan/tsan_interface_ann.cc
index 20e3d9a35ac..85de2e6221f 100644
--- a/libsanitizer/tsan/tsan_interface_ann.cc
+++ b/libsanitizer/tsan/tsan_interface_ann.cc
@@ -38,6 +38,7 @@ class ScopedAnnotation {
 
   ~ScopedAnnotation() {
     FuncExit(thr_);
+    CheckNoLocks(thr_);
   }
  private:
   ThreadState *const thr_;
@@ -123,8 +124,6 @@ static ExpectRace *FindRace(ExpectRace *list, uptr addr, uptr size) {
 
 static bool CheckContains(ExpectRace *list, uptr addr, uptr size) {
   ExpectRace *race = FindRace(list, addr, size);
-  if (race == 0 && AlternativeAddress(addr))
-    race = FindRace(list, AlternativeAddress(addr), size);
   if (race == 0)
     return false;
   DPrintf("Hit expected/benign race: %s addr=%zx:%d %s:%d\n",
diff --git a/libsanitizer/tsan/tsan_interface_atomic.cc b/libsanitizer/tsan/tsan_interface_atomic.cc
index 3f5a4ccc9f7..316614dd486 100644
--- a/libsanitizer/tsan/tsan_interface_atomic.cc
+++ b/libsanitizer/tsan/tsan_interface_atomic.cc
@@ -25,33 +25,23 @@
 
 using namespace __tsan;  // NOLINT
 
-#define SCOPED_ATOMIC(func, ...) \
-    const uptr callpc = (uptr)__builtin_return_address(0); \
-    uptr pc = __sanitizer::StackTrace::GetCurrentPc(); \
-    mo = flags()->force_seq_cst_atomics ? (morder)mo_seq_cst : mo; \
-    ThreadState *const thr = cur_thread(); \
-    if (thr->ignore_interceptors) \
-      return NoTsanAtomic##func(__VA_ARGS__); \
-    AtomicStatInc(thr, sizeof(*a), mo, StatAtomic##func); \
-    ScopedAtomic sa(thr, callpc, a, mo, __func__); \
-    return Atomic##func(thr, pc, __VA_ARGS__); \
-/**/
-
 // These should match declarations from public tsan_interface_atomic.h header.
 typedef unsigned char      a8;
 typedef unsigned short     a16;  // NOLINT
 typedef unsigned int       a32;
 typedef unsigned long long a64;  // NOLINT
-#if defined(__SIZEOF_INT128__) \
-    || (__clang_major__ * 100 + __clang_minor__ >= 302)
+#if !defined(TSAN_GO) && (defined(__SIZEOF_INT128__) \
+    || (__clang_major__ * 100 + __clang_minor__ >= 302))
 __extension__ typedef __int128 a128;
 # define __TSAN_HAS_INT128 1
 #else
 # define __TSAN_HAS_INT128 0
 #endif
 
+#ifndef TSAN_GO
 // Protects emulation of 128-bit atomic operations.
 static StaticSpinMutex mutex128;
+#endif
 
 // Part of ABI, do not change.
 // http://llvm.org/viewvc/llvm-project/libcxx/trunk/include/atomic?view=markup
@@ -64,38 +54,6 @@ typedef enum {
   mo_seq_cst
 } morder;
 
-class ScopedAtomic {
- public:
-  ScopedAtomic(ThreadState *thr, uptr pc, const volatile void *a,
-               morder mo, const char *func)
-      : thr_(thr) {
-    FuncEntry(thr_, pc);
-    DPrintf("#%d: %s(%p, %d)\n", thr_->tid, func, a, mo);
-  }
-  ~ScopedAtomic() {
-    ProcessPendingSignals(thr_);
-    FuncExit(thr_);
-  }
- private:
-  ThreadState *thr_;
-};
-
-static void AtomicStatInc(ThreadState *thr, uptr size, morder mo, StatType t) {
-  StatInc(thr, StatAtomic);
-  StatInc(thr, t);
-  StatInc(thr, size == 1 ? StatAtomic1
-             : size == 2 ? StatAtomic2
-             : size == 4 ? StatAtomic4
-             : size == 8 ? StatAtomic8
-             :             StatAtomic16);
-  StatInc(thr, mo == mo_relaxed ? StatAtomicRelaxed
-             : mo == mo_consume ? StatAtomicConsume
-             : mo == mo_acquire ? StatAtomicAcquire
-             : mo == mo_release ? StatAtomicRelease
-             : mo == mo_acq_rel ? StatAtomicAcq_Rel
-             :                    StatAtomicSeq_Cst);
-}
-
 static bool IsLoadOrder(morder mo) {
   return mo == mo_relaxed || mo == mo_consume
       || mo == mo_acquire || mo == mo_seq_cst;
@@ -165,7 +123,7 @@ template<typename T> T func_cas(volatile T *v, T cmp, T xch) {
 // Atomic ops are executed under tsan internal mutex,
 // here we assume that the atomic variables are not accessed
 // from non-instrumented code.
-#ifndef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16
+#if !defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16) && !defined(TSAN_GO)
 a128 func_xchg(volatile a128 *v, a128 op) {
   SpinMutexLock lock(&mutex128);
   a128 cmp = *v;
@@ -238,6 +196,7 @@ static int SizeLog() {
   // this leads to false negatives only in very obscure cases.
 }
 
+#ifndef TSAN_GO
 static atomic_uint8_t *to_atomic(const volatile a8 *a) {
   return (atomic_uint8_t*)a;
 }
@@ -245,6 +204,7 @@ static atomic_uint8_t *to_atomic(const volatile a8 *a) {
 static atomic_uint16_t *to_atomic(const volatile a16 *a) {
   return (atomic_uint16_t*)a;
 }
+#endif
 
 static atomic_uint32_t *to_atomic(const volatile a32 *a) {
   return (atomic_uint32_t*)a;
@@ -272,10 +232,12 @@ static T NoTsanAtomicLoad(const volatile T *a, morder mo) {
   return atomic_load(to_atomic(a), to_mo(mo));
 }
 
+#if __TSAN_HAS_INT128 && !defined(TSAN_GO)
 static a128 NoTsanAtomicLoad(const volatile a128 *a, morder mo) {
   SpinMutexLock lock(&mutex128);
   return *a;
 }
+#endif
 
 template<typename T>
 static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
@@ -287,7 +249,7 @@ static T AtomicLoad(ThreadState *thr, uptr pc, const volatile T *a,
     MemoryReadAtomic(thr, pc, (uptr)a, SizeLog<T>());
     return NoTsanAtomicLoad(a, mo);
   }
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, false);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, (uptr)a, false);
   AcquireImpl(thr, pc, &s->clock);
   T v = NoTsanAtomicLoad(a, mo);
   s->mtx.ReadUnlock();
@@ -300,10 +262,12 @@ static void NoTsanAtomicStore(volatile T *a, T v, morder mo) {
   atomic_store(to_atomic(a), v, to_mo(mo));
 }
 
+#if __TSAN_HAS_INT128 && !defined(TSAN_GO)
 static void NoTsanAtomicStore(volatile a128 *a, a128 v, morder mo) {
   SpinMutexLock lock(&mutex128);
   *a = v;
 }
+#endif
 
 template<typename T>
 static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
@@ -319,7 +283,7 @@ static void AtomicStore(ThreadState *thr, uptr pc, volatile T *a, T v,
     return;
   }
   __sync_synchronize();
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, (uptr)a, true);
   thr->fast_state.IncrementEpoch();
   // Can't increment epoch w/o writing to the trace as well.
   TraceAddEvent(thr, thr->fast_state, EventTypeMop, 0);
@@ -333,7 +297,7 @@ static T AtomicRMW(ThreadState *thr, uptr pc, volatile T *a, T v, morder mo) {
   MemoryWriteAtomic(thr, pc, (uptr)a, SizeLog<T>());
   SyncVar *s = 0;
   if (mo != mo_relaxed) {
-    s = ctx->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, true);
+    s = ctx->metamap.GetOrCreateAndLock(thr, pc, (uptr)a, true);
     thr->fast_state.IncrementEpoch();
     // Can't increment epoch w/o writing to the trace as well.
     TraceAddEvent(thr, thr->fast_state, EventTypeMop, 0);
@@ -432,6 +396,7 @@ static bool NoTsanAtomicCAS(volatile T *a, T *c, T v, morder mo, morder fmo) {
   return atomic_compare_exchange_strong(to_atomic(a), c, v, to_mo(mo));
 }
 
+#if __TSAN_HAS_INT128
 static bool NoTsanAtomicCAS(volatile a128 *a, a128 *c, a128 v,
     morder mo, morder fmo) {
   a128 old = *c;
@@ -441,10 +406,12 @@ static bool NoTsanAtomicCAS(volatile a128 *a, a128 *c, a128 v,
   *c = cur;
   return false;
 }
+#endif
 
 template<typename T>
-static bool NoTsanAtomicCAS(volatile T *a, T c, T v, morder mo, morder fmo) {
-  return NoTsanAtomicCAS(a, &c, v, mo, fmo);
+static T NoTsanAtomicCAS(volatile T *a, T c, T v, morder mo, morder fmo) {
+  NoTsanAtomicCAS(a, &c, v, mo, fmo);
+  return c;
 }
 
 template<typename T>
@@ -455,7 +422,7 @@ static bool AtomicCAS(ThreadState *thr, uptr pc,
   SyncVar *s = 0;
   bool write_lock = mo != mo_acquire && mo != mo_consume;
   if (mo != mo_relaxed) {
-    s = ctx->synctab.GetOrCreateAndLock(thr, pc, (uptr)a, write_lock);
+    s = ctx->metamap.GetOrCreateAndLock(thr, pc, (uptr)a, write_lock);
     thr->fast_state.IncrementEpoch();
     // Can't increment epoch w/o writing to the trace as well.
     TraceAddEvent(thr, thr->fast_state, EventTypeMop, 0);
@@ -487,6 +454,7 @@ static T AtomicCAS(ThreadState *thr, uptr pc,
   return c;
 }
 
+#ifndef TSAN_GO
 static void NoTsanAtomicFence(morder mo) {
   __sync_synchronize();
 }
@@ -495,6 +463,56 @@ static void AtomicFence(ThreadState *thr, uptr pc, morder mo) {
   // FIXME(dvyukov): not implemented.
   __sync_synchronize();
 }
+#endif
+
+// Interface functions follow.
+#ifndef TSAN_GO
+
+// C/C++
+
+#define SCOPED_ATOMIC(func, ...) \
+    const uptr callpc = (uptr)__builtin_return_address(0); \
+    uptr pc = __sanitizer::StackTrace::GetCurrentPc(); \
+    mo = flags()->force_seq_cst_atomics ? (morder)mo_seq_cst : mo; \
+    ThreadState *const thr = cur_thread(); \
+    if (thr->ignore_interceptors) \
+      return NoTsanAtomic##func(__VA_ARGS__); \
+    AtomicStatInc(thr, sizeof(*a), mo, StatAtomic##func); \
+    ScopedAtomic sa(thr, callpc, a, mo, __func__); \
+    return Atomic##func(thr, pc, __VA_ARGS__); \
+/**/
+
+class ScopedAtomic {
+ public:
+  ScopedAtomic(ThreadState *thr, uptr pc, const volatile void *a,
+               morder mo, const char *func)
+      : thr_(thr) {
+    FuncEntry(thr_, pc);
+    DPrintf("#%d: %s(%p, %d)\n", thr_->tid, func, a, mo);
+  }
+  ~ScopedAtomic() {
+    ProcessPendingSignals(thr_);
+    FuncExit(thr_);
+  }
+ private:
+  ThreadState *thr_;
+};
+
+static void AtomicStatInc(ThreadState *thr, uptr size, morder mo, StatType t) {
+  StatInc(thr, StatAtomic);
+  StatInc(thr, t);
+  StatInc(thr, size == 1 ? StatAtomic1
+             : size == 2 ? StatAtomic2
+             : size == 4 ? StatAtomic4
+             : size == 8 ? StatAtomic8
+             :             StatAtomic16);
+  StatInc(thr, mo == mo_relaxed ? StatAtomicRelaxed
+             : mo == mo_consume ? StatAtomicConsume
+             : mo == mo_acquire ? StatAtomicAcquire
+             : mo == mo_release ? StatAtomicRelease
+             : mo == mo_acq_rel ? StatAtomicAcq_Rel
+             :                    StatAtomicSeq_Cst);
+}
 
 extern "C" {
 SANITIZER_INTERFACE_ATTRIBUTE
@@ -846,3 +864,88 @@ SANITIZER_INTERFACE_ATTRIBUTE
 void __tsan_atomic_signal_fence(morder mo) {
 }
 }  // extern "C"
+
+#else  // #ifndef TSAN_GO
+
+// Go
+
+#define ATOMIC(func, ...) \
+    if (thr->ignore_sync) { \
+      NoTsanAtomic##func(__VA_ARGS__); \
+    } else { \
+      FuncEntry(thr, cpc); \
+      Atomic##func(thr, pc, __VA_ARGS__); \
+      FuncExit(thr); \
+    } \
+/**/
+
+#define ATOMIC_RET(func, ret, ...) \
+    if (thr->ignore_sync) { \
+      (ret) = NoTsanAtomic##func(__VA_ARGS__); \
+    } else { \
+      FuncEntry(thr, cpc); \
+      (ret) = Atomic##func(thr, pc, __VA_ARGS__); \
+      FuncExit(thr); \
+    } \
+/**/
+
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_load(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(Load, *(a32*)(a+8), *(a32**)a, mo_acquire);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_load(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(Load, *(a64*)(a+8), *(a64**)a, mo_acquire);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_store(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC(Store, *(a32**)a, *(a32*)(a+8), mo_release);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_store(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC(Store, *(a64**)a, *(a64*)(a+8), mo_release);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_fetch_add(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(FetchAdd, *(a32*)(a+16), *(a32**)a, *(a32*)(a+8), mo_acq_rel);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_fetch_add(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(FetchAdd, *(a64*)(a+16), *(a64**)a, *(a64*)(a+8), mo_acq_rel);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_exchange(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(Exchange, *(a32*)(a+16), *(a32**)a, *(a32*)(a+8), mo_acq_rel);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_exchange(ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  ATOMIC_RET(Exchange, *(a64*)(a+16), *(a64**)a, *(a64*)(a+8), mo_acq_rel);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic32_compare_exchange(
+    ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  a32 cur = 0;
+  a32 cmp = *(a32*)(a+8);
+  ATOMIC_RET(CAS, cur, *(a32**)a, cmp, *(a32*)(a+12), mo_acq_rel, mo_acquire);
+  *(bool*)(a+16) = (cur == cmp);
+}
+
+SANITIZER_INTERFACE_ATTRIBUTE
+void __tsan_go_atomic64_compare_exchange(
+    ThreadState *thr, uptr cpc, uptr pc, u8 *a) {
+  a64 cur = 0;
+  a64 cmp = *(a64*)(a+8);
+  ATOMIC_RET(CAS, cur, *(a64**)a, cmp, *(a64*)(a+16), mo_acq_rel, mo_acquire);
+  *(bool*)(a+24) = (cur == cmp);
+}
+}  // extern "C"
+#endif  // #ifndef TSAN_GO
diff --git a/libsanitizer/tsan/tsan_interface_java.cc b/libsanitizer/tsan/tsan_interface_java.cc
index 7f451690946..03a9e280ddc 100644
--- a/libsanitizer/tsan/tsan_interface_java.cc
+++ b/libsanitizer/tsan/tsan_interface_java.cc
@@ -20,54 +20,17 @@
 
 using namespace __tsan;  // NOLINT
 
-namespace __tsan {
-
-const uptr kHeapShadow = 0x300000000000ull;
-const uptr kHeapAlignment = 8;
+const jptr kHeapAlignment = 8;
 
-struct BlockDesc {
-  bool begin;
-  Mutex mtx;
-  SyncVar *head;
-
-  BlockDesc()
-      : mtx(MutexTypeJavaMBlock, StatMtxJavaMBlock)
-      , head() {
-    CHECK_EQ(begin, false);
-    begin = true;
-  }
-
-  ~BlockDesc() {
-    CHECK_EQ(begin, true);
-    begin = false;
-    ThreadState *thr = cur_thread();
-    SyncVar *s = head;
-    while (s) {
-      SyncVar *s1 = s->next;
-      StatInc(thr, StatSyncDestroyed);
-      s->mtx.Lock();
-      s->mtx.Unlock();
-      thr->mset.Remove(s->GetId());
-      DestroyAndFree(s);
-      s = s1;
-    }
-  }
-};
+namespace __tsan {
 
 struct JavaContext {
   const uptr heap_begin;
   const uptr heap_size;
-  BlockDesc *heap_shadow;
 
   JavaContext(jptr heap_begin, jptr heap_size)
       : heap_begin(heap_begin)
       , heap_size(heap_size) {
-    uptr size = heap_size / kHeapAlignment * sizeof(BlockDesc);
-    heap_shadow = (BlockDesc*)MmapFixedNoReserve(kHeapShadow, size);
-    if ((uptr)heap_shadow != kHeapShadow) {
-      Printf("ThreadSanitizer: failed to mmap Java heap shadow\n");
-      Die();
-    }
   }
 };
 
@@ -91,63 +54,6 @@ class ScopedJavaFunc {
 static u64 jctx_buf[sizeof(JavaContext) / sizeof(u64) + 1];
 static JavaContext *jctx;
 
-static BlockDesc *getblock(uptr addr) {
-  uptr i = (addr - jctx->heap_begin) / kHeapAlignment;
-  return &jctx->heap_shadow[i];
-}
-
-static uptr USED getmem(BlockDesc *b) {
-  uptr i = b - jctx->heap_shadow;
-  uptr p = jctx->heap_begin + i * kHeapAlignment;
-  CHECK_GE(p, jctx->heap_begin);
-  CHECK_LT(p, jctx->heap_begin + jctx->heap_size);
-  return p;
-}
-
-static BlockDesc *getblockbegin(uptr addr) {
-  for (BlockDesc *b = getblock(addr);; b--) {
-    CHECK_GE(b, jctx->heap_shadow);
-    if (b->begin)
-      return b;
-  }
-  return 0;
-}
-
-SyncVar* GetJavaSync(ThreadState *thr, uptr pc, uptr addr,
-                     bool write_lock, bool create) {
-  if (jctx == 0 || addr < jctx->heap_begin
-      || addr >= jctx->heap_begin + jctx->heap_size)
-    return 0;
-  BlockDesc *b = getblockbegin(addr);
-  DPrintf("#%d: GetJavaSync %p->%p\n", thr->tid, addr, b);
-  Lock l(&b->mtx);
-  SyncVar *s = b->head;
-  for (; s; s = s->next) {
-    if (s->addr == addr) {
-      DPrintf("#%d: found existing sync for %p\n", thr->tid, addr);
-      break;
-    }
-  }
-  if (s == 0 && create) {
-    DPrintf("#%d: creating new sync for %p\n", thr->tid, addr);
-    s = ctx->synctab.Create(thr, pc, addr);
-    s->next = b->head;
-    b->head = s;
-  }
-  if (s) {
-    if (write_lock)
-      s->mtx.Lock();
-    else
-      s->mtx.ReadLock();
-  }
-  return s;
-}
-
-SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr) {
-  // We do not destroy Java mutexes other than in __tsan_java_free().
-  return 0;
-}
-
 }  // namespace __tsan
 
 #define SCOPED_JAVA_FUNC(func) \
@@ -190,8 +96,7 @@ void __tsan_java_alloc(jptr ptr, jptr size) {
   CHECK_GE(ptr, jctx->heap_begin);
   CHECK_LE(ptr + size, jctx->heap_begin + jctx->heap_size);
 
-  BlockDesc *b = getblock(ptr);
-  new(b) BlockDesc();
+  OnUserAlloc(thr, pc, ptr, size, false);
 }
 
 void __tsan_java_free(jptr ptr, jptr size) {
@@ -204,12 +109,7 @@ void __tsan_java_free(jptr ptr, jptr size) {
   CHECK_GE(ptr, jctx->heap_begin);
   CHECK_LE(ptr + size, jctx->heap_begin + jctx->heap_size);
 
-  BlockDesc *beg = getblock(ptr);
-  BlockDesc *end = getblock(ptr + size);
-  for (BlockDesc *b = beg; b != end; b++) {
-    if (b->begin)
-      b->~BlockDesc();
-  }
+  ctx->metamap.FreeRange(thr, pc, ptr, size);
 }
 
 void __tsan_java_move(jptr src, jptr dst, jptr size) {
@@ -224,42 +124,36 @@ void __tsan_java_move(jptr src, jptr dst, jptr size) {
   CHECK_LE(src + size, jctx->heap_begin + jctx->heap_size);
   CHECK_GE(dst, jctx->heap_begin);
   CHECK_LE(dst + size, jctx->heap_begin + jctx->heap_size);
-  CHECK(dst >= src + size || src >= dst + size);
+  CHECK_NE(dst, src);
+  CHECK_NE(size, 0);
 
   // Assuming it's not running concurrently with threads that do
   // memory accesses and mutex operations (stop-the-world phase).
-  {  // NOLINT
-    BlockDesc *s = getblock(src);
-    BlockDesc *d = getblock(dst);
-    BlockDesc *send = getblock(src + size);
-    for (; s != send; s++, d++) {
-      CHECK_EQ(d->begin, false);
-      if (s->begin) {
-        DPrintf("#%d: moving block %p->%p\n", thr->tid, getmem(s), getmem(d));
-        new(d) BlockDesc;
-        d->head = s->head;
-        for (SyncVar *sync = d->head; sync; sync = sync->next) {
-          uptr newaddr = sync->addr - src + dst;
-          DPrintf("#%d: moving sync %p->%p\n", thr->tid, sync->addr, newaddr);
-          sync->addr = newaddr;
-        }
-        s->head = 0;
-        s->~BlockDesc();
-      }
-    }
+  ctx->metamap.MoveMemory(src, dst, size);
+
+  // Move shadow.
+  u64 *s = (u64*)MemToShadow(src);
+  u64 *d = (u64*)MemToShadow(dst);
+  u64 *send = (u64*)MemToShadow(src + size);
+  uptr inc = 1;
+  if (dst > src) {
+    s = (u64*)MemToShadow(src + size) - 1;
+    d = (u64*)MemToShadow(dst + size) - 1;
+    send = (u64*)MemToShadow(src) - 1;
+    inc = -1;
   }
-
-  {  // NOLINT
-    u64 *s = (u64*)MemToShadow(src);
-    u64 *d = (u64*)MemToShadow(dst);
-    u64 *send = (u64*)MemToShadow(src + size);
-    for (; s != send; s++, d++) {
-      *d = *s;
-      *s = 0;
-    }
+  for (; s != send; s += inc, d += inc) {
+    *d = *s;
+    *s = 0;
   }
 }
 
+void __tsan_java_finalize() {
+  SCOPED_JAVA_FUNC(__tsan_java_finalize);
+  DPrintf("#%d: java_mutex_finalize()\n", thr->tid);
+  AcquireGlobal(thr, 0);
+}
+
 void __tsan_java_mutex_lock(jptr addr) {
   SCOPED_JAVA_FUNC(__tsan_java_mutex_lock);
   DPrintf("#%d: java_mutex_lock(%p)\n", thr->tid, addr);
diff --git a/libsanitizer/tsan/tsan_interface_java.h b/libsanitizer/tsan/tsan_interface_java.h
index 885ff289751..9af1f3fe21a 100644
--- a/libsanitizer/tsan/tsan_interface_java.h
+++ b/libsanitizer/tsan/tsan_interface_java.h
@@ -48,8 +48,13 @@ void __tsan_java_alloc(jptr ptr, jptr size) INTERFACE_ATTRIBUTE;
 void __tsan_java_free(jptr ptr, jptr size) INTERFACE_ATTRIBUTE;
 // Callback for memory move by GC.
 // Can be aggregated for several objects (preferably).
-// The ranges must not overlap.
+// The ranges can overlap.
 void __tsan_java_move(jptr src, jptr dst, jptr size) INTERFACE_ATTRIBUTE;
+// This function must be called on the finalizer thread
+// before executing a batch of finalizers.
+// It ensures necessary synchronization between
+// java object creation and finalization.
+void __tsan_java_finalize() INTERFACE_ATTRIBUTE;
 
 // Mutex lock.
 // Addr is any unique address associated with the mutex.
diff --git a/libsanitizer/tsan/tsan_mman.cc b/libsanitizer/tsan/tsan_mman.cc
index 3df0531f0c8..2eae60de512 100644
--- a/libsanitizer/tsan/tsan_mman.cc
+++ b/libsanitizer/tsan/tsan_mman.cc
@@ -8,6 +8,7 @@
 // This file is a part of ThreadSanitizer (TSan), a race detector.
 //
 //===----------------------------------------------------------------------===//
+#include "sanitizer_common/sanitizer_allocator_interface.h"
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_placement_new.h"
 #include "tsan_mman.h"
@@ -16,43 +17,17 @@
 #include "tsan_flags.h"
 
 // May be overriden by front-end.
-extern "C" void WEAK __tsan_malloc_hook(void *ptr, uptr size) {
+extern "C" void WEAK __sanitizer_malloc_hook(void *ptr, uptr size) {
   (void)ptr;
   (void)size;
 }
 
-extern "C" void WEAK __tsan_free_hook(void *ptr) {
+extern "C" void WEAK __sanitizer_free_hook(void *ptr) {
   (void)ptr;
 }
 
 namespace __tsan {
 
-COMPILER_CHECK(sizeof(MBlock) == 16);
-
-void MBlock::Lock() {
-  atomic_uintptr_t *a = reinterpret_cast<atomic_uintptr_t*>(this);
-  uptr v = atomic_load(a, memory_order_relaxed);
-  for (int iter = 0;; iter++) {
-    if (v & 1) {
-      if (iter < 10)
-        proc_yield(20);
-      else
-        internal_sched_yield();
-      v = atomic_load(a, memory_order_relaxed);
-      continue;
-    }
-    if (atomic_compare_exchange_weak(a, &v, v | 1, memory_order_acquire))
-      break;
-  }
-}
-
-void MBlock::Unlock() {
-  atomic_uintptr_t *a = reinterpret_cast<atomic_uintptr_t*>(this);
-  uptr v = atomic_load(a, memory_order_relaxed);
-  DCHECK(v & 1);
-  atomic_store(a, v & ~1, memory_order_relaxed);
-}
-
 struct MapUnmapCallback {
   void OnMap(uptr p, uptr size) const { }
   void OnUnmap(uptr p, uptr size) const {
@@ -86,15 +61,16 @@ void AllocatorPrintStats() {
 }
 
 static void SignalUnsafeCall(ThreadState *thr, uptr pc) {
-  if (!thr->in_signal_handler || !flags()->report_signal_unsafe)
+  if (atomic_load(&thr->in_signal_handler, memory_order_relaxed) == 0 ||
+      !flags()->report_signal_unsafe)
     return;
   StackTrace stack;
   stack.ObtainCurrent(thr, pc);
   ThreadRegistryLock l(ctx->thread_registry);
   ScopedReport rep(ReportTypeSignalUnsafe);
   if (!IsFiredSuppression(ctx, rep, stack)) {
-    rep.AddStack(&stack);
-    OutputReport(ctx, rep, rep.GetReport()->stacks[0]);
+    rep.AddStack(&stack, true);
+    OutputReport(thr, rep);
   }
 }
 
@@ -104,43 +80,36 @@ void *user_alloc(ThreadState *thr, uptr pc, uptr sz, uptr align) {
   void *p = allocator()->Allocate(&thr->alloc_cache, sz, align);
   if (p == 0)
     return 0;
-  MBlock *b = new(allocator()->GetMetaData(p)) MBlock;
-  b->Init(sz, thr->tid, CurrentStackId(thr, pc));
-  if (ctx && ctx->initialized) {
-    if (thr->ignore_reads_and_writes == 0)
-      MemoryRangeImitateWrite(thr, pc, (uptr)p, sz);
-    else
-      MemoryResetRange(thr, pc, (uptr)p, sz);
-  }
-  DPrintf("#%d: alloc(%zu) = %p\n", thr->tid, sz, p);
+  if (ctx && ctx->initialized)
+    OnUserAlloc(thr, pc, (uptr)p, sz, true);
   SignalUnsafeCall(thr, pc);
   return p;
 }
 
 void user_free(ThreadState *thr, uptr pc, void *p) {
-  CHECK_NE(p, (void*)0);
-  DPrintf("#%d: free(%p)\n", thr->tid, p);
-  MBlock *b = (MBlock*)allocator()->GetMetaData(p);
-  if (b->ListHead()) {
-    MBlock::ScopedLock l(b);
-    for (SyncVar *s = b->ListHead(); s;) {
-      SyncVar *res = s;
-      s = s->next;
-      StatInc(thr, StatSyncDestroyed);
-      res->mtx.Lock();
-      res->mtx.Unlock();
-      DestroyAndFree(res);
-    }
-    b->ListReset();
-  }
-  if (ctx && ctx->initialized) {
-    if (thr->ignore_reads_and_writes == 0)
-      MemoryRangeFreed(thr, pc, (uptr)p, b->Size());
-  }
+  if (ctx && ctx->initialized)
+    OnUserFree(thr, pc, (uptr)p, true);
   allocator()->Deallocate(&thr->alloc_cache, p);
   SignalUnsafeCall(thr, pc);
 }
 
+void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write) {
+  DPrintf("#%d: alloc(%zu) = %p\n", thr->tid, sz, p);
+  ctx->metamap.AllocBlock(thr, pc, p, sz);
+  if (write && thr->ignore_reads_and_writes == 0)
+    MemoryRangeImitateWrite(thr, pc, (uptr)p, sz);
+  else
+    MemoryResetRange(thr, pc, (uptr)p, sz);
+}
+
+void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write) {
+  CHECK_NE(p, (void*)0);
+  uptr sz = ctx->metamap.FreeBlock(thr, pc, p);
+  DPrintf("#%d: free(%p, %zu)\n", thr->tid, p, sz);
+  if (write && thr->ignore_reads_and_writes == 0)
+    MemoryRangeFreed(thr, pc, (uptr)p, sz);
+}
+
 void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz) {
   void *p2 = 0;
   // FIXME: Handle "shrinking" more efficiently,
@@ -150,9 +119,8 @@ void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz) {
     if (p2 == 0)
       return 0;
     if (p) {
-      MBlock *b = user_mblock(thr, p);
-      CHECK_NE(b, 0);
-      internal_memcpy(p2, p, min(b->Size(), sz));
+      uptr oldsz = user_alloc_usable_size(p);
+      internal_memcpy(p2, p, min(oldsz, sz));
     }
   }
   if (p)
@@ -160,39 +128,29 @@ void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz) {
   return p2;
 }
 
-uptr user_alloc_usable_size(ThreadState *thr, uptr pc, void *p) {
+uptr user_alloc_usable_size(const void *p) {
   if (p == 0)
     return 0;
-  MBlock *b = (MBlock*)allocator()->GetMetaData(p);
-  return b ? b->Size() : 0;
-}
-
-MBlock *user_mblock(ThreadState *thr, void *p) {
-  CHECK_NE(p, 0);
-  Allocator *a = allocator();
-  void *b = a->GetBlockBegin(p);
-  if (b == 0)
-    return 0;
-  return (MBlock*)a->GetMetaData(b);
+  MBlock *b = ctx->metamap.GetBlock((uptr)p);
+  return b ? b->siz : 0;
 }
 
 void invoke_malloc_hook(void *ptr, uptr size) {
   ThreadState *thr = cur_thread();
   if (ctx == 0 || !ctx->initialized || thr->ignore_interceptors)
     return;
-  __tsan_malloc_hook(ptr, size);
+  __sanitizer_malloc_hook(ptr, size);
 }
 
 void invoke_free_hook(void *ptr) {
   ThreadState *thr = cur_thread();
   if (ctx == 0 || !ctx->initialized || thr->ignore_interceptors)
     return;
-  __tsan_free_hook(ptr);
+  __sanitizer_free_hook(ptr);
 }
 
 void *internal_alloc(MBlockType typ, uptr sz) {
   ThreadState *thr = cur_thread();
-  CHECK_LE(sz, InternalSizeClassMap::kMaxSize);
   if (thr->nomalloc) {
     thr->nomalloc = 0;  // CHECK calls internal_malloc().
     CHECK(0);
@@ -214,47 +172,42 @@ void internal_free(void *p) {
 using namespace __tsan;
 
 extern "C" {
-uptr __tsan_get_current_allocated_bytes() {
+uptr __sanitizer_get_current_allocated_bytes() {
   uptr stats[AllocatorStatCount];
   allocator()->GetStats(stats);
   return stats[AllocatorStatAllocated];
 }
 
-uptr __tsan_get_heap_size() {
+uptr __sanitizer_get_heap_size() {
   uptr stats[AllocatorStatCount];
   allocator()->GetStats(stats);
   return stats[AllocatorStatMapped];
 }
 
-uptr __tsan_get_free_bytes() {
+uptr __sanitizer_get_free_bytes() {
   return 1;
 }
 
-uptr __tsan_get_unmapped_bytes() {
+uptr __sanitizer_get_unmapped_bytes() {
   return 1;
 }
 
-uptr __tsan_get_estimated_allocated_size(uptr size) {
+uptr __sanitizer_get_estimated_allocated_size(uptr size) {
   return size;
 }
 
-bool __tsan_get_ownership(void *p) {
+int __sanitizer_get_ownership(const void *p) {
   return allocator()->GetBlockBegin(p) != 0;
 }
 
-uptr __tsan_get_allocated_size(void *p) {
-  if (p == 0)
-    return 0;
-  p = allocator()->GetBlockBegin(p);
-  if (p == 0)
-    return 0;
-  MBlock *b = (MBlock*)allocator()->GetMetaData(p);
-  return b->Size();
+uptr __sanitizer_get_allocated_size(const void *p) {
+  return user_alloc_usable_size(p);
 }
 
 void __tsan_on_thread_idle() {
   ThreadState *thr = cur_thread();
   allocator()->SwallowCache(&thr->alloc_cache);
   internal_allocator()->SwallowCache(&thr->internal_alloc_cache);
+  ctx->metamap.OnThreadIdle(thr);
 }
 }  // extern "C"
diff --git a/libsanitizer/tsan/tsan_mman.h b/libsanitizer/tsan/tsan_mman.h
index 90faaffa1fc..ab8eb83f919 100644
--- a/libsanitizer/tsan/tsan_mman.h
+++ b/libsanitizer/tsan/tsan_mman.h
@@ -29,10 +29,7 @@ void *user_alloc(ThreadState *thr, uptr pc, uptr sz,
 void user_free(ThreadState *thr, uptr pc, void *p);
 void *user_realloc(ThreadState *thr, uptr pc, void *p, uptr sz);
 void *user_alloc_aligned(ThreadState *thr, uptr pc, uptr sz, uptr align);
-uptr user_alloc_usable_size(ThreadState *thr, uptr pc, void *p);
-// Given the pointer p into a valid allocated block,
-// returns the descriptor of the block.
-MBlock *user_mblock(ThreadState *thr, void *p);
+uptr user_alloc_usable_size(const void *p);
 
 // Invoking malloc/free hooks that may be installed by the user.
 void invoke_malloc_hook(void *ptr, uptr size);
@@ -60,7 +57,6 @@ enum MBlockType {
   MBlockSuppression,
   MBlockExpectRace,
   MBlockSignal,
-  MBlockFD,
   MBlockJmpBuf,
 
   // This must be the last.
diff --git a/libsanitizer/tsan/tsan_mutex.cc b/libsanitizer/tsan/tsan_mutex.cc
index 0c3bb4a6721..2e49b9d2de9 100644
--- a/libsanitizer/tsan/tsan_mutex.cc
+++ b/libsanitizer/tsan/tsan_mutex.cc
@@ -29,13 +29,13 @@ static MutexType CanLockTab[MutexTypeCount][MutexTypeCount] = {
   /*0  MutexTypeInvalid*/     {},
   /*1  MutexTypeTrace*/       {MutexTypeLeaf},
   /*2  MutexTypeThreads*/     {MutexTypeReport},
-  /*3  MutexTypeReport*/      {MutexTypeSyncTab, MutexTypeSyncVar,
+  /*3  MutexTypeReport*/      {MutexTypeSyncVar,
                                MutexTypeMBlock, MutexTypeJavaMBlock},
   /*4  MutexTypeSyncVar*/     {MutexTypeDDetector},
-  /*5  MutexTypeSyncTab*/     {MutexTypeSyncVar},
+  /*5  MutexTypeSyncTab*/     {},  // unused
   /*6  MutexTypeSlab*/        {MutexTypeLeaf},
   /*7  MutexTypeAnnotations*/ {},
-  /*8  MutexTypeAtExit*/      {MutexTypeSyncTab},
+  /*8  MutexTypeAtExit*/      {MutexTypeSyncVar},
   /*9  MutexTypeMBlock*/      {MutexTypeSyncVar},
   /*10 MutexTypeJavaMBlock*/  {MutexTypeSyncVar},
   /*11 MutexTypeDDetector*/   {},
@@ -159,7 +159,19 @@ void InternalDeadlockDetector::Unlock(MutexType t) {
   CHECK(locked_[t]);
   locked_[t] = 0;
 }
+
+void InternalDeadlockDetector::CheckNoLocks() {
+  for (int i = 0; i != MutexTypeCount; i++) {
+    CHECK_EQ(locked_[i], 0);
+  }
+}
+#endif
+
+void CheckNoLocks(ThreadState *thr) {
+#if TSAN_DEBUG && !TSAN_GO
+  thr->internal_deadlock_detector.CheckNoLocks();
 #endif
+}
 
 const uptr kUnlocked = 0;
 const uptr kWriteLock = 1;
@@ -220,7 +232,7 @@ void Mutex::Lock() {
       cmp = kUnlocked;
       if (atomic_compare_exchange_weak(&state_, &cmp, kWriteLock,
                                        memory_order_acquire)) {
-#if TSAN_COLLECT_STATS
+#if TSAN_COLLECT_STATS && !TSAN_GO
         StatInc(cur_thread(), stat_type_, backoff.Contention());
 #endif
         return;
@@ -248,7 +260,7 @@ void Mutex::ReadLock() {
   for (Backoff backoff; backoff.Do();) {
     prev = atomic_load(&state_, memory_order_acquire);
     if ((prev & kWriteLock) == 0) {
-#if TSAN_COLLECT_STATS
+#if TSAN_COLLECT_STATS && !TSAN_GO
       StatInc(cur_thread(), stat_type_, backoff.Contention());
 #endif
       return;
diff --git a/libsanitizer/tsan/tsan_mutex.h b/libsanitizer/tsan/tsan_mutex.h
index f075ce831e6..c27bc2b4459 100644
--- a/libsanitizer/tsan/tsan_mutex.h
+++ b/libsanitizer/tsan/tsan_mutex.h
@@ -69,6 +69,7 @@ class InternalDeadlockDetector {
   InternalDeadlockDetector();
   void Lock(MutexType t);
   void Unlock(MutexType t);
+  void CheckNoLocks();
  private:
   u64 seq_;
   u64 locked_[MutexTypeCount];
@@ -76,6 +77,10 @@ class InternalDeadlockDetector {
 
 void InitializeMutex();
 
+// Checks that the current thread does not hold any runtime locks
+// (e.g. when returning from an interceptor).
+void CheckNoLocks(ThreadState *thr);
+
 }  // namespace __tsan
 
 #endif  // TSAN_MUTEX_H
diff --git a/libsanitizer/tsan/tsan_platform.h b/libsanitizer/tsan/tsan_platform.h
index 60eb1a84995..b7a6376e8ed 100644
--- a/libsanitizer/tsan/tsan_platform.h
+++ b/libsanitizer/tsan/tsan_platform.h
@@ -14,31 +14,23 @@
 C++ linux memory layout:
 0000 0000 0000 - 03c0 0000 0000: protected
 03c0 0000 0000 - 1000 0000 0000: shadow
-1000 0000 0000 - 6000 0000 0000: protected
+1000 0000 0000 - 3000 0000 0000: protected
+3000 0000 0000 - 4000 0000 0000: metainfo (memory blocks and sync objects)
+4000 0000 0000 - 6000 0000 0000: protected
 6000 0000 0000 - 6200 0000 0000: traces
 6200 0000 0000 - 7d00 0000 0000: -
 7d00 0000 0000 - 7e00 0000 0000: heap
 7e00 0000 0000 - 7fff ffff ffff: modules and main thread stack
 
-C++ COMPAT linux memory layout:
-0000 0000 0000 - 0400 0000 0000: protected
-0400 0000 0000 - 1000 0000 0000: shadow
-1000 0000 0000 - 2900 0000 0000: protected
-2900 0000 0000 - 2c00 0000 0000: modules
-2c00 0000 0000 - 6000 0000 0000: -
-6000 0000 0000 - 6200 0000 0000: traces
-6200 0000 0000 - 7d00 0000 0000: -
-7d00 0000 0000 - 7e00 0000 0000: heap
-7e00 0000 0000 - 7f00 0000 0000: -
-7f00 0000 0000 - 7fff ffff ffff: main thread stack
-
 Go linux and darwin memory layout:
 0000 0000 0000 - 0000 1000 0000: executable
 0000 1000 0000 - 00f8 0000 0000: -
 00c0 0000 0000 - 00e0 0000 0000: heap
 00e0 0000 0000 - 1000 0000 0000: -
 1000 0000 0000 - 1380 0000 0000: shadow
-1460 0000 0000 - 6000 0000 0000: -
+1460 0000 0000 - 2000 0000 0000: -
+3000 0000 0000 - 4000 0000 0000: metainfo (memory blocks and sync objects)
+4000 0000 0000 - 6000 0000 0000: -
 6000 0000 0000 - 6200 0000 0000: traces
 6200 0000 0000 - 7fff ffff ffff: -
 
@@ -49,7 +41,8 @@ Go windows memory layout:
 00e0 0000 0000 - 0100 0000 0000: -
 0100 0000 0000 - 0560 0000 0000: shadow
 0560 0000 0000 - 0760 0000 0000: traces
-0760 0000 0000 - 07ff ffff ffff: -
+0760 0000 0000 - 07d0 0000 0000: metainfo (memory blocks and sync objects)
+07d0 0000 0000 - 07ff ffff ffff: -
 */
 
 #ifndef TSAN_PLATFORM_H
@@ -66,18 +59,16 @@ static const uptr kLinuxAppMemBeg = 0x000000000000ULL;
 static const uptr kLinuxAppMemEnd = 0x04dfffffffffULL;
 # if SANITIZER_WINDOWS
 static const uptr kLinuxShadowMsk = 0x010000000000ULL;
-# else
+static const uptr kMetaShadow     = 0x076000000000ULL;
+static const uptr kMetaSize       = 0x007000000000ULL;
+# else  // if SANITIZER_WINDOWS
 static const uptr kLinuxShadowMsk = 0x200000000000ULL;
-# endif
-// TSAN_COMPAT_SHADOW is intended for COMPAT virtual memory layout,
-// when memory addresses are of the 0x2axxxxxxxxxx form.
-// The option is enabled with 'setarch x86_64 -L'.
-#elif defined(TSAN_COMPAT_SHADOW) && TSAN_COMPAT_SHADOW
-static const uptr kLinuxAppMemBeg = 0x290000000000ULL;
-static const uptr kLinuxAppMemEnd = 0x7fffffffffffULL;
-static const uptr kAppMemGapBeg   = 0x2c0000000000ULL;
-static const uptr kAppMemGapEnd   = 0x7d0000000000ULL;
-#else
+static const uptr kMetaShadow     = 0x300000000000ULL;
+static const uptr kMetaSize       = 0x100000000000ULL;
+# endif  // if SANITIZER_WINDOWS
+#else  // defined(TSAN_GO)
+static const uptr kMetaShadow     = 0x300000000000ULL;
+static const uptr kMetaSize       = 0x100000000000ULL;
 static const uptr kLinuxAppMemBeg = 0x7cf000000000ULL;
 static const uptr kLinuxAppMemEnd = 0x7fffffffffffULL;
 #endif
@@ -94,10 +85,16 @@ const uptr kTraceMemSize = 0x020000000000ULL;
 // This has to be a macro to allow constant initialization of constants below.
 #ifndef TSAN_GO
 #define MemToShadow(addr) \
-    (((addr) & ~(kLinuxAppMemMsk | (kShadowCell - 1))) * kShadowCnt)
+    ((((uptr)addr) & ~(kLinuxAppMemMsk | (kShadowCell - 1))) * kShadowCnt)
+#define MemToMeta(addr) \
+    (u32*)(((((uptr)addr) & ~(kLinuxAppMemMsk | (kMetaShadowCell - 1))) \
+    / kMetaShadowCell * kMetaShadowSize) | kMetaShadow)
 #else
 #define MemToShadow(addr) \
-    ((((addr) & ~(kShadowCell - 1)) * kShadowCnt) | kLinuxShadowMsk)
+    (((((uptr)addr) & ~(kShadowCell - 1)) * kShadowCnt) | kLinuxShadowMsk)
+#define MemToMeta(addr) \
+    (u32*)(((((uptr)addr) & ~(kMetaShadowCell - 1)) \
+    / kMetaShadowCell * kMetaShadowSize) | kMetaShadow)
 #endif
 
 static const uptr kLinuxShadowBeg = MemToShadow(kLinuxAppMemBeg);
@@ -105,9 +102,8 @@ static const uptr kLinuxShadowEnd =
     MemToShadow(kLinuxAppMemEnd) | 0xff;
 
 static inline bool IsAppMem(uptr mem) {
-#if defined(TSAN_COMPAT_SHADOW) && TSAN_COMPAT_SHADOW
-  return (mem >= kLinuxAppMemBeg && mem < kAppMemGapBeg) ||
-         (mem >= kAppMemGapEnd   && mem <= kLinuxAppMemEnd);
+#if defined(TSAN_GO)
+  return mem <= kLinuxAppMemEnd;
 #else
   return mem >= kLinuxAppMemBeg && mem <= kLinuxAppMemEnd;
 #endif
@@ -126,22 +122,11 @@ static inline uptr ShadowToMem(uptr shadow) {
 #endif
 }
 
-// For COMPAT mapping returns an alternative address
-// that mapped to the same shadow address.
-// COMPAT mapping is not quite one-to-one.
-static inline uptr AlternativeAddress(uptr addr) {
-#if defined(TSAN_COMPAT_SHADOW) && TSAN_COMPAT_SHADOW
-  return (addr & ~kLinuxAppMemMsk) | 0x280000000000ULL;
-#else
-  return 0;
-#endif
-}
-
 void FlushShadowMemory();
-void WriteMemoryProfile(char *buf, uptr buf_size);
+void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive);
 uptr GetRSS();
 
-const char *InitializePlatform();
+void InitializePlatform();
 void FinalizePlatform();
 
 // The additional page is to catch shadow stack overflow as paging fault.
diff --git a/libsanitizer/tsan/tsan_platform_linux.cc b/libsanitizer/tsan/tsan_platform_linux.cc
index 062e84615cc..ba81fd242e3 100644
--- a/libsanitizer/tsan/tsan_platform_linux.cc
+++ b/libsanitizer/tsan/tsan_platform_linux.cc
@@ -12,7 +12,7 @@
 
 
 #include "sanitizer_common/sanitizer_platform.h"
-#if SANITIZER_LINUX
+#if SANITIZER_LINUX || SANITIZER_FREEBSD
 
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_libc.h"
@@ -30,7 +30,6 @@
 #include <string.h>
 #include <stdarg.h>
 #include <sys/mman.h>
-#include <sys/prctl.h>
 #include <sys/syscall.h>
 #include <sys/socket.h>
 #include <sys/time.h>
@@ -41,9 +40,10 @@
 #include <errno.h>
 #include <sched.h>
 #include <dlfcn.h>
+#if SANITIZER_LINUX
 #define __need_res_state
 #include <resolv.h>
-#include <malloc.h>
+#endif
 
 #ifdef sa_handler
 # undef sa_handler
@@ -53,61 +53,98 @@
 # undef sa_sigaction
 #endif
 
-extern "C" struct mallinfo __libc_mallinfo();
+#if SANITIZER_FREEBSD
+extern "C" void *__libc_stack_end;
+void *__libc_stack_end = 0;
+#endif
 
 namespace __tsan {
 
 const uptr kPageSize = 4096;
 
+enum {
+  MemTotal  = 0,
+  MemShadow = 1,
+  MemMeta   = 2,
+  MemFile   = 3,
+  MemMmap   = 4,
+  MemTrace  = 5,
+  MemHeap   = 6,
+  MemOther  = 7,
+  MemCount  = 8,
+};
+
 void FillProfileCallback(uptr start, uptr rss, bool file,
                          uptr *mem, uptr stats_size) {
-  CHECK_EQ(7, stats_size);
-  mem[6] += rss;  // total
+  mem[MemTotal] += rss;
   start >>= 40;
-  if (start < 0x10)  // shadow
-    mem[0] += rss;
-  else if (start >= 0x20 && start < 0x30)  // compat modules
-    mem[file ? 1 : 2] += rss;
-  else if (start >= 0x7e)  // modules
-    mem[file ? 1 : 2] += rss;
-  else if (start >= 0x60 && start < 0x62)  // traces
-    mem[3] += rss;
-  else if (start >= 0x7d && start < 0x7e)  // heap
-    mem[4] += rss;
-  else  // other
-    mem[5] += rss;
+  if (start < 0x10)
+    mem[MemShadow] += rss;
+  else if (start >= 0x20 && start < 0x30)
+    mem[file ? MemFile : MemMmap] += rss;
+  else if (start >= 0x30 && start < 0x40)
+    mem[MemMeta] += rss;
+  else if (start >= 0x7e)
+    mem[file ? MemFile : MemMmap] += rss;
+  else if (start >= 0x60 && start < 0x62)
+    mem[MemTrace] += rss;
+  else if (start >= 0x7d && start < 0x7e)
+    mem[MemHeap] += rss;
+  else
+    mem[MemOther] += rss;
 }
 
-void WriteMemoryProfile(char *buf, uptr buf_size) {
-  uptr mem[7] = {};
+void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
+  uptr mem[MemCount] = {};
   __sanitizer::GetMemoryProfile(FillProfileCallback, mem, 7);
-  char *buf_pos = buf;
-  char *buf_end = buf + buf_size;
-  buf_pos += internal_snprintf(buf_pos, buf_end - buf_pos,
-      "RSS %zd MB: shadow:%zd file:%zd mmap:%zd trace:%zd heap:%zd other:%zd\n",
-      mem[6] >> 20, mem[0] >> 20, mem[1] >> 20, mem[2] >> 20,
-      mem[3] >> 20, mem[4] >> 20, mem[5] >> 20);
-  struct mallinfo mi = __libc_mallinfo();
-  buf_pos += internal_snprintf(buf_pos, buf_end - buf_pos,
-      "mallinfo: arena=%d mmap=%d fordblks=%d keepcost=%d\n",
-      mi.arena >> 20, mi.hblkhd >> 20, mi.fordblks >> 20, mi.keepcost >> 20);
+  internal_snprintf(buf, buf_size,
+      "RSS %zd MB: shadow:%zd meta:%zd file:%zd mmap:%zd"
+      " trace:%zd heap:%zd other:%zd nthr=%zd/%zd\n",
+      mem[MemTotal] >> 20, mem[MemShadow] >> 20, mem[MemMeta] >> 20,
+      mem[MemFile] >> 20, mem[MemMmap] >> 20, mem[MemTrace] >> 20,
+      mem[MemHeap] >> 20, mem[MemOther] >> 20,
+      nlive, nthread);
 }
 
 uptr GetRSS() {
-  uptr mem[7] = {};
-  __sanitizer::GetMemoryProfile(FillProfileCallback, mem, 7);
-  return mem[6];
+  uptr fd = OpenFile("/proc/self/statm", false);
+  if ((sptr)fd < 0)
+    return 0;
+  char buf[64];
+  uptr len = internal_read(fd, buf, sizeof(buf) - 1);
+  internal_close(fd);
+  if ((sptr)len <= 0)
+    return 0;
+  buf[len] = 0;
+  // The format of the file is:
+  // 1084 89 69 11 0 79 0
+  // We need the second number which is RSS in 4K units.
+  char *pos = buf;
+  // Skip the first number.
+  while (*pos >= '0' && *pos <= '9')
+    pos++;
+  // Skip whitespaces.
+  while (!(*pos >= '0' && *pos <= '9') && *pos != 0)
+    pos++;
+  // Read the number.
+  uptr rss = 0;
+  while (*pos >= '0' && *pos <= '9')
+    rss = rss * 10 + *pos++ - '0';
+  return rss * 4096;
 }
 
-
+#if SANITIZER_LINUX
 void FlushShadowMemoryCallback(
     const SuspendedThreadsList &suspended_threads_list,
     void *argument) {
   FlushUnneededShadowMemory(kLinuxShadowBeg, kLinuxShadowEnd - kLinuxShadowBeg);
 }
+#endif
 
 void FlushShadowMemory() {
+#if SANITIZER_LINUX
   StopTheWorld(FlushShadowMemoryCallback, 0);
+#endif
 }
 
 #ifndef TSAN_GO
@@ -121,9 +158,7 @@ static void ProtectRange(uptr beg, uptr end) {
     Die();
   }
 }
-#endif
 
-#ifndef TSAN_GO
 // Mark shadow for .rodata sections with the special kShadowRodata marker.
 // Accesses to .rodata can't race, so this saves time, memory and trace space.
 static void MapRodata() {
@@ -182,6 +217,7 @@ static void MapRodata() {
 }
 
 void InitializeShadowMemory() {
+  // Map memory shadow.
   uptr shadow = (uptr)MmapFixedNoReserve(kLinuxShadowBeg,
     kLinuxShadowEnd - kLinuxShadowBeg);
   if (shadow != kLinuxShadowBeg) {
@@ -190,23 +226,56 @@ void InitializeShadowMemory() {
                "to link with -pie (%p, %p).\n", shadow, kLinuxShadowBeg);
     Die();
   }
+  // This memory range is used for thread stacks and large user mmaps.
+  // Frequently a thread uses only a small part of stack and similarly
+  // a program uses a small part of large mmap. On some programs
+  // we see 20% memory usage reduction without huge pages for this range.
+#ifdef MADV_NOHUGEPAGE
+  madvise((void*)MemToShadow(0x7f0000000000ULL),
+      0x10000000000ULL * kShadowMultiplier, MADV_NOHUGEPAGE);
+#endif
+  DPrintf("memory shadow: %zx-%zx (%zuGB)\n",
+      kLinuxShadowBeg, kLinuxShadowEnd,
+      (kLinuxShadowEnd - kLinuxShadowBeg) >> 30);
+
+  // Map meta shadow.
+  if (MemToMeta(kLinuxAppMemBeg) < (u32*)kMetaShadow) {
+    Printf("ThreadSanitizer: bad meta shadow (%p -> %p < %p)\n",
+        kLinuxAppMemBeg, MemToMeta(kLinuxAppMemBeg), kMetaShadow);
+    Die();
+  }
+  if (MemToMeta(kLinuxAppMemEnd) >= (u32*)(kMetaShadow + kMetaSize)) {
+    Printf("ThreadSanitizer: bad meta shadow (%p -> %p >= %p)\n",
+        kLinuxAppMemEnd, MemToMeta(kLinuxAppMemEnd), kMetaShadow + kMetaSize);
+    Die();
+  }
+  uptr meta = (uptr)MmapFixedNoReserve(kMetaShadow, kMetaSize);
+  if (meta != kMetaShadow) {
+    Printf("FATAL: ThreadSanitizer can not mmap the shadow memory\n");
+    Printf("FATAL: Make sure to compile with -fPIE and "
+               "to link with -pie (%p, %p).\n", meta, kMetaShadow);
+    Die();
+  }
+  DPrintf("meta shadow: %zx-%zx (%zuGB)\n",
+      kMetaShadow, kMetaShadow + kMetaSize, kMetaSize >> 30);
+
+  // Protect gaps.
   const uptr kClosedLowBeg  = 0x200000;
   const uptr kClosedLowEnd  = kLinuxShadowBeg - 1;
   const uptr kClosedMidBeg = kLinuxShadowEnd + 1;
-  const uptr kClosedMidEnd = min(kLinuxAppMemBeg, kTraceMemBegin);
+  const uptr kClosedMidEnd = min(min(kLinuxAppMemBeg, kTraceMemBegin),
+      kMetaShadow);
+
   ProtectRange(kClosedLowBeg, kClosedLowEnd);
   ProtectRange(kClosedMidBeg, kClosedMidEnd);
-  DPrintf("kClosedLow   %zx-%zx (%zuGB)\n",
+  VPrintf(2, "kClosedLow   %zx-%zx (%zuGB)\n",
       kClosedLowBeg, kClosedLowEnd, (kClosedLowEnd - kClosedLowBeg) >> 30);
-  DPrintf("kLinuxShadow %zx-%zx (%zuGB)\n",
-      kLinuxShadowBeg, kLinuxShadowEnd,
-      (kLinuxShadowEnd - kLinuxShadowBeg) >> 30);
-  DPrintf("kClosedMid   %zx-%zx (%zuGB)\n",
+  VPrintf(2, "kClosedMid   %zx-%zx (%zuGB)\n",
       kClosedMidBeg, kClosedMidEnd, (kClosedMidEnd - kClosedMidBeg) >> 30);
-  DPrintf("kLinuxAppMem %zx-%zx (%zuGB)\n",
+  VPrintf(2, "app mem: %zx-%zx (%zuGB)\n",
       kLinuxAppMemBeg, kLinuxAppMemEnd,
       (kLinuxAppMemEnd - kLinuxAppMemBeg) >> 30);
-  DPrintf("stack        %zx\n", (uptr)&shadow);
+  VPrintf(2, "stack: %zx\n", (uptr)&shadow);
 
   MapRodata();
 }
@@ -261,26 +330,8 @@ static void InitDataSeg() {
 
 #endif  // #ifndef TSAN_GO
 
-static rlim_t getlim(int res) {
-  rlimit rlim;
-  CHECK_EQ(0, getrlimit(res, &rlim));
-  return rlim.rlim_cur;
-}
-
-static void setlim(int res, rlim_t lim) {
-  // The following magic is to prevent clang from replacing it with memset.
-  volatile rlimit rlim;
-  rlim.rlim_cur = lim;
-  rlim.rlim_max = lim;
-  setrlimit(res, (rlimit*)&rlim);
-}
-
-const char *InitializePlatform() {
-  void *p = 0;
-  if (sizeof(p) == 8) {
-    // Disable core dumps, dumping of 16TB usually takes a bit long.
-    setlim(RLIMIT_CORE, 0);
-  }
+void InitializePlatform() {
+  DisableCoreDumperIfNecessary();
 
   // Go maps shadow memory lazily and works fine with limited address space.
   // Unlimited stack is not a problem as well, because the executable
@@ -290,7 +341,7 @@ const char *InitializePlatform() {
     // TSan doesn't play well with unlimited stack size (as stack
     // overlaps with shadow memory). If we detect unlimited stack size,
     // we re-exec the program with limited stack size as a best effort.
-    if (getlim(RLIMIT_STACK) == (rlim_t)-1) {
+    if (StackSizeIsUnlimited()) {
       const uptr kMaxStackSize = 32 * 1024 * 1024;
       VReport(1, "Program is run with unlimited stack size, which wouldn't "
                  "work with ThreadSanitizer.\n"
@@ -300,11 +351,11 @@ const char *InitializePlatform() {
       reexec = true;
     }
 
-    if (getlim(RLIMIT_AS) != (rlim_t)-1) {
+    if (!AddressSpaceIsUnlimited()) {
       Report("WARNING: Program is run with limited virtual address space,"
              " which wouldn't work with ThreadSanitizer.\n");
       Report("Re-execing with unlimited virtual address space.\n");
-      setlim(RLIMIT_AS, -1);
+      SetAddressSpaceUnlimited();
       reexec = true;
     }
     if (reexec)
@@ -316,7 +367,6 @@ const char *InitializePlatform() {
   InitTlsSize();
   InitDataSeg();
 #endif
-  return GetEnv(kTsanOptionsEnv);
 }
 
 bool IsGlobalVar(uptr addr) {
@@ -328,6 +378,7 @@ bool IsGlobalVar(uptr addr) {
 // This is required to properly "close" the fds, because we do not see internal
 // closes within glibc. The code is a pure hack.
 int ExtractResolvFDs(void *state, int *fds, int nfd) {
+#if SANITIZER_LINUX
   int cnt = 0;
   __res_state *statp = (__res_state*)state;
   for (int i = 0; i < MAXNS && cnt < nfd; i++) {
@@ -335,6 +386,9 @@ int ExtractResolvFDs(void *state, int *fds, int nfd) {
       fds[cnt++] = statp->_u._ext.nssocks[i];
   }
   return cnt;
+#else
+  return 0;
+#endif
 }
 
 // Extract file descriptors passed via UNIX domain sockets.
diff --git a/libsanitizer/tsan/tsan_platform_mac.cc b/libsanitizer/tsan/tsan_platform_mac.cc
index c3d4d905219..95527c79431 100644
--- a/libsanitizer/tsan/tsan_platform_mac.cc
+++ b/libsanitizer/tsan/tsan_platform_mac.cc
@@ -45,7 +45,7 @@ uptr GetShadowMemoryConsumption() {
 void FlushShadowMemory() {
 }
 
-void WriteMemoryProfile(char *buf, uptr buf_size) {
+void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
 }
 
 uptr GetRSS() {
@@ -71,18 +71,8 @@ void InitializeShadowMemory() {
 }
 #endif
 
-const char *InitializePlatform() {
-  void *p = 0;
-  if (sizeof(p) == 8) {
-    // Disable core dumps, dumping of 16TB usually takes a bit long.
-    // The following magic is to prevent clang from replacing it with memset.
-    volatile rlimit lim;
-    lim.rlim_cur = 0;
-    lim.rlim_max = 0;
-    setrlimit(RLIMIT_CORE, (rlimit*)&lim);
-  }
-
-  return GetEnv(kTsanOptionsEnv);
+void InitializePlatform() {
+  DisableCoreDumperIfNecessary();
 }
 
 void FinalizePlatform() {
diff --git a/libsanitizer/tsan/tsan_platform_windows.cc b/libsanitizer/tsan/tsan_platform_windows.cc
index f16bebf7f4c..d6e9e6d33f4 100644
--- a/libsanitizer/tsan/tsan_platform_windows.cc
+++ b/libsanitizer/tsan/tsan_platform_windows.cc
@@ -26,15 +26,14 @@ uptr GetShadowMemoryConsumption() {
 void FlushShadowMemory() {
 }
 
-void WriteMemoryProfile(char *buf, uptr buf_size) {
+void WriteMemoryProfile(char *buf, uptr buf_size, uptr nthread, uptr nlive) {
 }
 
 uptr GetRSS() {
   return 0;
 }
 
-const char *InitializePlatform() {
-  return GetEnv(kTsanOptionsEnv);
+void InitializePlatform() {
 }
 
 void FinalizePlatform() {
diff --git a/libsanitizer/tsan/tsan_report.cc b/libsanitizer/tsan/tsan_report.cc
index 00a512e0cf8..b14856cd153 100644
--- a/libsanitizer/tsan/tsan_report.cc
+++ b/libsanitizer/tsan/tsan_report.cc
@@ -15,9 +15,9 @@
 
 namespace __tsan {
 
-class Decorator: private __sanitizer::AnsiColorDecorator {
+class Decorator: public __sanitizer::SanitizerCommonDecorator {
  public:
-  Decorator() : __sanitizer::AnsiColorDecorator(PrintsToTtyCached()) { }
+  Decorator() : SanitizerCommonDecorator() { }
   const char *Warning()    { return Red(); }
   const char *EndWarning() { return Default(); }
   const char *Access()     { return Blue(); }
diff --git a/libsanitizer/tsan/tsan_report.h b/libsanitizer/tsan/tsan_report.h
index d773e057af0..0bde59b1675 100644
--- a/libsanitizer/tsan/tsan_report.h
+++ b/libsanitizer/tsan/tsan_report.h
@@ -40,6 +40,7 @@ struct ReportStack {
   char *file;
   int line;
   int col;
+  bool suppressable;
 };
 
 struct ReportMopMutex {
@@ -78,6 +79,7 @@ struct ReportLocation {
   char *name;
   char *file;
   int line;
+  bool suppressable;
   ReportStack *stack;
 };
 
diff --git a/libsanitizer/tsan/tsan_rtl.cc b/libsanitizer/tsan/tsan_rtl.cc
index 7932a6d9e04..f5942bcaa3e 100644
--- a/libsanitizer/tsan/tsan_rtl.cc
+++ b/libsanitizer/tsan/tsan_rtl.cc
@@ -23,6 +23,16 @@
 #include "tsan_suppressions.h"
 #include "tsan_symbolize.h"
 
+#ifdef __SSE3__
+// <emmintrin.h> transitively includes <stdlib.h>,
+// and it's prohibited to include std headers into tsan runtime.
+// So we do this dirty trick.
+#define _MM_MALLOC_H_INCLUDED
+#define __MM_MALLOC_H
+#include <emmintrin.h>
+typedef __m128i m128;
+#endif
+
 volatile int __tsan_resumed = 0;
 
 extern "C" void __tsan_resume() {
@@ -110,10 +120,7 @@ static void MemoryProfiler(Context *ctx, fd_t fd, int i) {
   uptr n_running_threads;
   ctx->thread_registry->GetNumberOfThreads(&n_threads, &n_running_threads);
   InternalScopedBuffer<char> buf(4096);
-  internal_snprintf(buf.data(), buf.size(), "%d: nthr=%d nlive=%d\n",
-      i, n_threads, n_running_threads);
-  internal_write(fd, buf.data(), internal_strlen(buf.data()));
-  WriteMemoryProfile(buf.data(), buf.size());
+  WriteMemoryProfile(buf.data(), buf.size(), n_threads, n_running_threads);
   internal_write(fd, buf.data(), internal_strlen(buf.data()));
 }
 
@@ -129,15 +136,21 @@ static void BackgroundThread(void *arg) {
 
   fd_t mprof_fd = kInvalidFd;
   if (flags()->profile_memory && flags()->profile_memory[0]) {
-    InternalScopedBuffer<char> filename(4096);
-    internal_snprintf(filename.data(), filename.size(), "%s.%d",
-        flags()->profile_memory, (int)internal_getpid());
-    uptr openrv = OpenFile(filename.data(), true);
-    if (internal_iserror(openrv)) {
-      Printf("ThreadSanitizer: failed to open memory profile file '%s'\n",
-          &filename[0]);
+    if (internal_strcmp(flags()->profile_memory, "stdout") == 0) {
+      mprof_fd = 1;
+    } else if (internal_strcmp(flags()->profile_memory, "stderr") == 0) {
+      mprof_fd = 2;
     } else {
-      mprof_fd = openrv;
+      InternalScopedBuffer<char> filename(4096);
+      internal_snprintf(filename.data(), filename.size(), "%s.%d",
+          flags()->profile_memory, (int)internal_getpid());
+      uptr openrv = OpenFile(filename.data(), true);
+      if (internal_iserror(openrv)) {
+        Printf("ThreadSanitizer: failed to open memory profile file '%s'\n",
+            &filename[0]);
+      } else {
+        mprof_fd = openrv;
+      }
     }
   }
 
@@ -152,27 +165,23 @@ static void BackgroundThread(void *arg) {
     // Flush memory if requested.
     if (flags()->flush_memory_ms > 0) {
       if (last_flush + flags()->flush_memory_ms * kMs2Ns < now) {
-        if (flags()->verbosity > 0)
-          Printf("ThreadSanitizer: periodic memory flush\n");
+        VPrintf(1, "ThreadSanitizer: periodic memory flush\n");
         FlushShadowMemory();
         last_flush = NanoTime();
       }
     }
+    // GetRSS can be expensive on huge programs, so don't do it every 100ms.
     if (flags()->memory_limit_mb > 0) {
       uptr rss = GetRSS();
       uptr limit = uptr(flags()->memory_limit_mb) << 20;
-      if (flags()->verbosity > 0) {
-        Printf("ThreadSanitizer: memory flush check"
-               " RSS=%llu LAST=%llu LIMIT=%llu\n",
-               (u64)rss>>20, (u64)last_rss>>20, (u64)limit>>20);
-      }
+      VPrintf(1, "ThreadSanitizer: memory flush check"
+                 " RSS=%llu LAST=%llu LIMIT=%llu\n",
+              (u64)rss >> 20, (u64)last_rss >> 20, (u64)limit >> 20);
       if (2 * rss > limit + last_rss) {
-        if (flags()->verbosity > 0)
-          Printf("ThreadSanitizer: flushing memory due to RSS\n");
+        VPrintf(1, "ThreadSanitizer: flushing memory due to RSS\n");
         FlushShadowMemory();
         rss = GetRSS();
-        if (flags()->verbosity > 0)
-          Printf("ThreadSanitizer: memory flushed RSS=%llu\n", (u64)rss>>20);
+        VPrintf(1, "ThreadSanitizer: memory flushed RSS=%llu\n", (u64)rss>>20);
       }
       last_rss = rss;
     }
@@ -201,11 +210,13 @@ static void StartBackgroundThread() {
   ctx->background_thread = internal_start_thread(&BackgroundThread, 0);
 }
 
+#ifndef TSAN_GO
 static void StopBackgroundThread() {
   atomic_store(&ctx->stop_background_thread, 1, memory_order_relaxed);
   internal_join_thread(ctx->background_thread);
   ctx->background_thread = 0;
 }
+#endif
 
 void DontNeedShadowFor(uptr addr, uptr size) {
   uptr shadow_beg = MemToShadow(addr);
@@ -218,6 +229,32 @@ void MapShadow(uptr addr, uptr size) {
   // so we can get away with unaligned mapping.
   // CHECK_EQ(addr, addr & ~((64 << 10) - 1));  // windows wants 64K alignment
   MmapFixedNoReserve(MemToShadow(addr), size * kShadowMultiplier);
+
+  // Meta shadow is 2:1, so tread carefully.
+  static bool data_mapped = false;
+  static uptr mapped_meta_end = 0;
+  uptr meta_begin = (uptr)MemToMeta(addr);
+  uptr meta_end = (uptr)MemToMeta(addr + size);
+  meta_begin = RoundDownTo(meta_begin, 64 << 10);
+  meta_end = RoundUpTo(meta_end, 64 << 10);
+  if (!data_mapped) {
+    // First call maps data+bss.
+    data_mapped = true;
+    MmapFixedNoReserve(meta_begin, meta_end - meta_begin);
+  } else {
+    // Mapping continous heap.
+    // Windows wants 64K alignment.
+    meta_begin = RoundDownTo(meta_begin, 64 << 10);
+    meta_end = RoundUpTo(meta_end, 64 << 10);
+    if (meta_end <= mapped_meta_end)
+      return;
+    if (meta_begin < mapped_meta_end)
+      meta_begin = mapped_meta_end;
+    MmapFixedNoReserve(meta_begin, meta_end - meta_begin);
+    mapped_meta_end = meta_end;
+  }
+  VPrintf(2, "mapped meta shadow for (%p-%p) at (%p-%p)\n",
+      addr, addr+size, meta_begin, meta_end);
 }
 
 void MapThreadTrace(uptr addr, uptr size) {
@@ -245,34 +282,35 @@ void Initialize(ThreadState *thr) {
   // Install tool-specific callbacks in sanitizer_common.
   SetCheckFailedCallback(TsanCheckFailed);
 
+  ctx = new(ctx_placeholder) Context;
+  const char *options = GetEnv(kTsanOptionsEnv);
+  InitializeFlags(&ctx->flags, options);
 #ifndef TSAN_GO
   InitializeAllocator();
 #endif
   InitializeInterceptors();
-  const char *env = InitializePlatform();
+  InitializePlatform();
   InitializeMutex();
   InitializeDynamicAnnotations();
-  ctx = new(ctx_placeholder) Context;
 #ifndef TSAN_GO
   InitializeShadowMemory();
 #endif
-  InitializeFlags(&ctx->flags, env);
   // Setup correct file descriptor for error reports.
-  __sanitizer_set_report_path(flags()->log_path);
+  __sanitizer_set_report_path(common_flags()->log_path);
   InitializeSuppressions();
 #ifndef TSAN_GO
   InitializeLibIgnore();
-  Symbolizer::Init(common_flags()->external_symbolizer_path);
-  Symbolizer::Get()->AddHooks(EnterSymbolizer, ExitSymbolizer);
+  Symbolizer::GetOrInit()->AddHooks(EnterSymbolizer, ExitSymbolizer);
 #endif
   StartBackgroundThread();
+#ifndef TSAN_GO
   SetSandboxingCallback(StopBackgroundThread);
-  if (flags()->detect_deadlocks)
+#endif
+  if (common_flags()->detect_deadlocks)
     ctx->dd = DDetector::Create(flags());
 
-  if (ctx->flags.verbosity)
-    Printf("***** Running under ThreadSanitizer v2 (pid %d) *****\n",
-           (int)internal_getpid());
+  VPrintf(1, "***** Running under ThreadSanitizer v2 (pid %d) *****\n",
+          (int)internal_getpid());
 
   // Initialize thread 0.
   int tid = ThreadCreate(thr, 0, 0, true);
@@ -291,7 +329,6 @@ void Initialize(ThreadState *thr) {
 }
 
 int Finalize(ThreadState *thr) {
-  Context *ctx = __tsan::ctx;
   bool failed = false;
 
   if (flags()->atexit_sleep_ms > 0 && ThreadCount(thr) > 1)
@@ -304,7 +341,7 @@ int Finalize(ThreadState *thr) {
   ctx->report_mtx.Unlock();
 
 #ifndef TSAN_GO
-  if (ctx->flags.verbosity)
+  if (common_flags()->verbosity)
     AllocatorPrintStats();
 #endif
 
@@ -325,7 +362,7 @@ int Finalize(ThreadState *thr) {
         ctx->nmissed_expected);
   }
 
-  if (flags()->print_suppressions)
+  if (common_flags()->print_suppressions)
     PrintMatchedSuppressions();
 #ifndef TSAN_GO
   if (flags()->print_benign)
@@ -372,16 +409,37 @@ void ForkChildAfter(ThreadState *thr, uptr pc) {
 }
 #endif
 
+#ifdef TSAN_GO
+NOINLINE
+void GrowShadowStack(ThreadState *thr) {
+  const int sz = thr->shadow_stack_end - thr->shadow_stack;
+  const int newsz = 2 * sz;
+  uptr *newstack = (uptr*)internal_alloc(MBlockShadowStack,
+      newsz * sizeof(uptr));
+  internal_memcpy(newstack, thr->shadow_stack, sz * sizeof(uptr));
+  internal_free(thr->shadow_stack);
+  thr->shadow_stack = newstack;
+  thr->shadow_stack_pos = newstack + sz;
+  thr->shadow_stack_end = newstack + newsz;
+}
+#endif
+
 u32 CurrentStackId(ThreadState *thr, uptr pc) {
   if (thr->shadow_stack_pos == 0)  // May happen during bootstrap.
     return 0;
-  if (pc) {
+  if (pc != 0) {
+#ifndef TSAN_GO
+    DCHECK_LT(thr->shadow_stack_pos, thr->shadow_stack_end);
+#else
+    if (thr->shadow_stack_pos == thr->shadow_stack_end)
+      GrowShadowStack(thr);
+#endif
     thr->shadow_stack_pos[0] = pc;
     thr->shadow_stack_pos++;
   }
   u32 id = StackDepotPut(thr->shadow_stack,
                          thr->shadow_stack_pos - thr->shadow_stack);
-  if (pc)
+  if (pc != 0)
     thr->shadow_stack_pos--;
   return id;
 }
@@ -443,7 +501,8 @@ void StoreIfNotYetStored(u64 *sp, u64 *s) {
   *s = 0;
 }
 
-static inline void HandleRace(ThreadState *thr, u64 *shadow_mem,
+ALWAYS_INLINE
+void HandleRace(ThreadState *thr, u64 *shadow_mem,
                               Shadow cur, Shadow old) {
   thr->racy_state[0] = cur.raw();
   thr->racy_state[1] = old.raw();
@@ -455,16 +514,12 @@ static inline void HandleRace(ThreadState *thr, u64 *shadow_mem,
 #endif
 }
 
-static inline bool OldIsInSameSynchEpoch(Shadow old, ThreadState *thr) {
-  return old.epoch() >= thr->fast_synch_epoch;
-}
-
 static inline bool HappensBefore(Shadow old, ThreadState *thr) {
   return thr->clock.get(old.TidWithIgnore()) >= old.epoch();
 }
 
-ALWAYS_INLINE USED
-void MemoryAccessImpl(ThreadState *thr, uptr addr,
+ALWAYS_INLINE
+void MemoryAccessImpl1(ThreadState *thr, uptr addr,
     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
     u64 *shadow_mem, Shadow cur) {
   StatInc(thr, StatMop);
@@ -542,13 +597,13 @@ void UnalignedMemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   while (size) {
     int size1 = 1;
     int kAccessSizeLog = kSizeLog1;
-    if (size >= 8 && (addr & ~7) == ((addr + 8) & ~7)) {
+    if (size >= 8 && (addr & ~7) == ((addr + 7) & ~7)) {
       size1 = 8;
       kAccessSizeLog = kSizeLog8;
-    } else if (size >= 4 && (addr & ~7) == ((addr + 4) & ~7)) {
+    } else if (size >= 4 && (addr & ~7) == ((addr + 3) & ~7)) {
       size1 = 4;
       kAccessSizeLog = kSizeLog4;
-    } else if (size >= 2 && (addr & ~7) == ((addr + 2) & ~7)) {
+    } else if (size >= 2 && (addr & ~7) == ((addr + 1) & ~7)) {
       size1 = 2;
       kAccessSizeLog = kSizeLog2;
     }
@@ -558,6 +613,90 @@ void UnalignedMemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   }
 }
 
+ALWAYS_INLINE
+bool ContainsSameAccessSlow(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
+  Shadow cur(a);
+  for (uptr i = 0; i < kShadowCnt; i++) {
+    Shadow old(LoadShadow(&s[i]));
+    if (Shadow::Addr0AndSizeAreEqual(cur, old) &&
+        old.TidWithIgnore() == cur.TidWithIgnore() &&
+        old.epoch() > sync_epoch &&
+        old.IsAtomic() == cur.IsAtomic() &&
+        old.IsRead() <= cur.IsRead())
+      return true;
+  }
+  return false;
+}
+
+#if defined(__SSE3__) && TSAN_SHADOW_COUNT == 4
+#define SHUF(v0, v1, i0, i1, i2, i3) _mm_castps_si128(_mm_shuffle_ps( \
+    _mm_castsi128_ps(v0), _mm_castsi128_ps(v1), \
+    (i0)*1 + (i1)*4 + (i2)*16 + (i3)*64))
+ALWAYS_INLINE
+bool ContainsSameAccessFast(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
+  // This is an optimized version of ContainsSameAccessSlow.
+  // load current access into access[0:63]
+  const m128 access     = _mm_cvtsi64_si128(a);
+  // duplicate high part of access in addr0:
+  // addr0[0:31]        = access[32:63]
+  // addr0[32:63]       = access[32:63]
+  // addr0[64:95]       = access[32:63]
+  // addr0[96:127]      = access[32:63]
+  const m128 addr0      = SHUF(access, access, 1, 1, 1, 1);
+  // load 4 shadow slots
+  const m128 shadow0    = _mm_load_si128((__m128i*)s);
+  const m128 shadow1    = _mm_load_si128((__m128i*)s + 1);
+  // load high parts of 4 shadow slots into addr_vect:
+  // addr_vect[0:31]    = shadow0[32:63]
+  // addr_vect[32:63]   = shadow0[96:127]
+  // addr_vect[64:95]   = shadow1[32:63]
+  // addr_vect[96:127]  = shadow1[96:127]
+  m128 addr_vect        = SHUF(shadow0, shadow1, 1, 3, 1, 3);
+  if (!is_write) {
+    // set IsRead bit in addr_vect
+    const m128 rw_mask1 = _mm_cvtsi64_si128(1<<15);
+    const m128 rw_mask  = SHUF(rw_mask1, rw_mask1, 0, 0, 0, 0);
+    addr_vect           = _mm_or_si128(addr_vect, rw_mask);
+  }
+  // addr0 == addr_vect?
+  const m128 addr_res   = _mm_cmpeq_epi32(addr0, addr_vect);
+  // epoch1[0:63]       = sync_epoch
+  const m128 epoch1     = _mm_cvtsi64_si128(sync_epoch);
+  // epoch[0:31]        = sync_epoch[0:31]
+  // epoch[32:63]       = sync_epoch[0:31]
+  // epoch[64:95]       = sync_epoch[0:31]
+  // epoch[96:127]      = sync_epoch[0:31]
+  const m128 epoch      = SHUF(epoch1, epoch1, 0, 0, 0, 0);
+  // load low parts of shadow cell epochs into epoch_vect:
+  // epoch_vect[0:31]   = shadow0[0:31]
+  // epoch_vect[32:63]  = shadow0[64:95]
+  // epoch_vect[64:95]  = shadow1[0:31]
+  // epoch_vect[96:127] = shadow1[64:95]
+  const m128 epoch_vect = SHUF(shadow0, shadow1, 0, 2, 0, 2);
+  // epoch_vect >= sync_epoch?
+  const m128 epoch_res  = _mm_cmpgt_epi32(epoch_vect, epoch);
+  // addr_res & epoch_res
+  const m128 res        = _mm_and_si128(addr_res, epoch_res);
+  // mask[0] = res[7]
+  // mask[1] = res[15]
+  // ...
+  // mask[15] = res[127]
+  const int mask        = _mm_movemask_epi8(res);
+  return mask != 0;
+}
+#endif
+
+ALWAYS_INLINE
+bool ContainsSameAccess(u64 *s, u64 a, u64 sync_epoch, bool is_write) {
+#if defined(__SSE3__) && TSAN_SHADOW_COUNT == 4
+  bool res = ContainsSameAccessFast(s, a, sync_epoch, is_write);
+  DCHECK_EQ(res, ContainsSameAccessSlow(s, a, sync_epoch, is_write));
+  return res;
+#else
+  return ContainsSameAccessSlow(s, a, sync_epoch, is_write);
+#endif
+}
+
 ALWAYS_INLINE USED
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic) {
@@ -579,7 +718,7 @@ void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   }
 #endif
 
-  if (*shadow_mem == kShadowRodata) {
+  if (kCppMode && *shadow_mem == kShadowRodata) {
     // Access to .rodata section, no races here.
     // Measurements show that it can be 10-20% of all memory accesses.
     StatInc(thr, StatMop);
@@ -590,14 +729,12 @@ void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   }
 
   FastState fast_state = thr->fast_state;
-  if (fast_state.GetIgnoreBit())
+  if (fast_state.GetIgnoreBit()) {
+    StatInc(thr, StatMop);
+    StatInc(thr, kAccessIsWrite ? StatMopWrite : StatMopRead);
+    StatInc(thr, (StatType)(StatMop1 + kAccessSizeLog));
+    StatInc(thr, StatMopIgnored);
     return;
-  if (kCollectHistory) {
-    fast_state.IncrementEpoch();
-    thr->fast_state = fast_state;
-    // We must not store to the trace if we do not store to the shadow.
-    // That is, this call must be moved somewhere below.
-    TraceAddEvent(thr, fast_state, EventTypeMop, pc);
   }
 
   Shadow cur(fast_state);
@@ -605,7 +742,41 @@ void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
   cur.SetWrite(kAccessIsWrite);
   cur.SetAtomic(kIsAtomic);
 
-  MemoryAccessImpl(thr, addr, kAccessSizeLog, kAccessIsWrite, kIsAtomic,
+  if (LIKELY(ContainsSameAccess(shadow_mem, cur.raw(),
+      thr->fast_synch_epoch, kAccessIsWrite))) {
+    StatInc(thr, StatMop);
+    StatInc(thr, kAccessIsWrite ? StatMopWrite : StatMopRead);
+    StatInc(thr, (StatType)(StatMop1 + kAccessSizeLog));
+    StatInc(thr, StatMopSame);
+    return;
+  }
+
+  if (kCollectHistory) {
+    fast_state.IncrementEpoch();
+    thr->fast_state = fast_state;
+    TraceAddEvent(thr, fast_state, EventTypeMop, pc);
+    cur.IncrementEpoch();
+  }
+
+  MemoryAccessImpl1(thr, addr, kAccessSizeLog, kAccessIsWrite, kIsAtomic,
+      shadow_mem, cur);
+}
+
+// Called by MemoryAccessRange in tsan_rtl_thread.cc
+ALWAYS_INLINE USED
+void MemoryAccessImpl(ThreadState *thr, uptr addr,
+    int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
+    u64 *shadow_mem, Shadow cur) {
+  if (LIKELY(ContainsSameAccess(shadow_mem, cur.raw(),
+      thr->fast_synch_epoch, kAccessIsWrite))) {
+    StatInc(thr, StatMop);
+    StatInc(thr, kAccessIsWrite ? StatMopWrite : StatMopRead);
+    StatInc(thr, (StatType)(StatMop1 + kAccessSizeLog));
+    StatInc(thr, StatMopSame);
+    return;
+  }
+
+  MemoryAccessImpl1(thr, addr, kAccessSizeLog, kAccessIsWrite, kIsAtomic,
       shadow_mem, cur);
 }
 
@@ -723,17 +894,8 @@ void FuncEntry(ThreadState *thr, uptr pc) {
 #ifndef TSAN_GO
   DCHECK_LT(thr->shadow_stack_pos, thr->shadow_stack_end);
 #else
-  if (thr->shadow_stack_pos == thr->shadow_stack_end) {
-    const int sz = thr->shadow_stack_end - thr->shadow_stack;
-    const int newsz = 2 * sz;
-    uptr *newstack = (uptr*)internal_alloc(MBlockShadowStack,
-        newsz * sizeof(uptr));
-    internal_memcpy(newstack, thr->shadow_stack, sz * sizeof(uptr));
-    internal_free(thr->shadow_stack);
-    thr->shadow_stack = newstack;
-    thr->shadow_stack_pos = newstack + sz;
-    thr->shadow_stack_end = newstack + newsz;
-  }
+  if (thr->shadow_stack_pos == thr->shadow_stack_end)
+    GrowShadowStack(thr);
 #endif
   thr->shadow_stack_pos[0] = pc;
   thr->shadow_stack_pos++;
diff --git a/libsanitizer/tsan/tsan_rtl.h b/libsanitizer/tsan/tsan_rtl.h
index e6bc8b28595..c7ea94dfbde 100644
--- a/libsanitizer/tsan/tsan_rtl.h
+++ b/libsanitizer/tsan/tsan_rtl.h
@@ -42,6 +42,7 @@
 #include "tsan_platform.h"
 #include "tsan_mutexset.h"
 #include "tsan_ignoreset.h"
+#include "tsan_stack_trace.h"
 
 #if SANITIZER_WORDSIZE != 64
 # error "ThreadSanitizer is supported only on 64-bit platforms"
@@ -49,87 +50,12 @@
 
 namespace __tsan {
 
-// Descriptor of user's memory block.
-struct MBlock {
-  /*
-  u64 mtx : 1;  // must be first
-  u64 lst : 44;
-  u64 stk : 31;  // on word boundary
-  u64 tid : kTidBits;
-  u64 siz : 128 - 1 - 31 - 44 - kTidBits;  // 39
-  */
-  u64 raw[2];
-
-  void Init(uptr siz, u32 tid, u32 stk) {
-    raw[0] = raw[1] = 0;
-    raw[1] |= (u64)siz << ((1 + 44 + 31 + kTidBits) % 64);
-    raw[1] |= (u64)tid << ((1 + 44 + 31) % 64);
-    raw[0] |= (u64)stk << (1 + 44);
-    raw[1] |= (u64)stk >> (64 - 44 - 1);
-    DCHECK_EQ(Size(), siz);
-    DCHECK_EQ(Tid(), tid);
-    DCHECK_EQ(StackId(), stk);
-  }
-
-  u32 Tid() const {
-    return GetLsb(raw[1] >> ((1 + 44 + 31) % 64), kTidBits);
-  }
-
-  uptr Size() const {
-    return raw[1] >> ((1 + 31 + 44 + kTidBits) % 64);
-  }
-
-  u32 StackId() const {
-    return (raw[0] >> (1 + 44)) | GetLsb(raw[1] << (64 - 44 - 1), 31);
-  }
-
-  SyncVar *ListHead() const {
-    return (SyncVar*)(GetLsb(raw[0] >> 1, 44) << 3);
-  }
-
-  void ListPush(SyncVar *v) {
-    SyncVar *lst = ListHead();
-    v->next = lst;
-    u64 x = (u64)v ^ (u64)lst;
-    x = (x >> 3) << 1;
-    raw[0] ^= x;
-    DCHECK_EQ(ListHead(), v);
-  }
-
-  SyncVar *ListPop() {
-    SyncVar *lst = ListHead();
-    SyncVar *nxt = lst->next;
-    lst->next = 0;
-    u64 x = (u64)lst ^ (u64)nxt;
-    x = (x >> 3) << 1;
-    raw[0] ^= x;
-    DCHECK_EQ(ListHead(), nxt);
-    return lst;
-  }
-
-  void ListReset() {
-    SyncVar *lst = ListHead();
-    u64 x = (u64)lst;
-    x = (x >> 3) << 1;
-    raw[0] ^= x;
-    DCHECK_EQ(ListHead(), 0);
-  }
-
-  void Lock();
-  void Unlock();
-  typedef GenericScopedLock<MBlock> ScopedLock;
-};
-
 #ifndef TSAN_GO
-#if defined(TSAN_COMPAT_SHADOW) && TSAN_COMPAT_SHADOW
-const uptr kAllocatorSpace = 0x7d0000000000ULL;
-#else
 const uptr kAllocatorSpace = 0x7d0000000000ULL;
-#endif
 const uptr kAllocatorSize  =  0x10000000000ULL;  // 1T.
 
 struct MapUnmapCallback;
-typedef SizeClassAllocator64<kAllocatorSpace, kAllocatorSize, sizeof(MBlock),
+typedef SizeClassAllocator64<kAllocatorSpace, kAllocatorSize, 0,
     DefaultSizeClassMap, MapUnmapCallback> PrimaryAllocator;
 typedef SizeClassAllocatorLocalCache<PrimaryAllocator> AllocatorCache;
 typedef LargeMmapAllocator<MapUnmapCallback> SecondaryAllocator;
@@ -146,14 +72,14 @@ const u64 kShadowRodata = (u64)-1;  // .rodata shadow marker
 // FastState (from most significant bit):
 //   ignore          : 1
 //   tid             : kTidBits
-//   epoch           : kClkBits
 //   unused          : -
 //   history_size    : 3
+//   epoch           : kClkBits
 class FastState {
  public:
   FastState(u64 tid, u64 epoch) {
     x_ = tid << kTidShift;
-    x_ |= epoch << kClkShift;
+    x_ |= epoch;
     DCHECK_EQ(tid, this->tid());
     DCHECK_EQ(epoch, this->epoch());
     DCHECK_EQ(GetIgnoreBit(), false);
@@ -178,13 +104,13 @@ class FastState {
   }
 
   u64 epoch() const {
-    u64 res = (x_ << (kTidBits + 1)) >> (64 - kClkBits);
+    u64 res = x_ & ((1ull << kClkBits) - 1);
     return res;
   }
 
   void IncrementEpoch() {
     u64 old_epoch = epoch();
-    x_ += 1 << kClkShift;
+    x_ += 1;
     DCHECK_EQ(old_epoch + 1, epoch());
     (void)old_epoch;
   }
@@ -196,17 +122,19 @@ class FastState {
   void SetHistorySize(int hs) {
     CHECK_GE(hs, 0);
     CHECK_LE(hs, 7);
-    x_ = (x_ & ~7) | hs;
+    x_ = (x_ & ~(kHistoryMask << kHistoryShift)) | (u64(hs) << kHistoryShift);
   }
 
+  ALWAYS_INLINE
   int GetHistorySize() const {
-    return (int)(x_ & 7);
+    return (int)((x_ >> kHistoryShift) & kHistoryMask);
   }
 
   void ClearHistorySize() {
-    x_ &= ~7;
+    SetHistorySize(0);
   }
 
+  ALWAYS_INLINE
   u64 GetTracePos() const {
     const int hs = GetHistorySize();
     // When hs == 0, the trace consists of 2 parts.
@@ -217,20 +145,21 @@ class FastState {
  private:
   friend class Shadow;
   static const int kTidShift = 64 - kTidBits - 1;
-  static const int kClkShift = kTidShift - kClkBits;
   static const u64 kIgnoreBit = 1ull << 63;
   static const u64 kFreedBit = 1ull << 63;
+  static const u64 kHistoryShift = kClkBits;
+  static const u64 kHistoryMask = 7;
   u64 x_;
 };
 
 // Shadow (from most significant bit):
 //   freed           : 1
 //   tid             : kTidBits
-//   epoch           : kClkBits
 //   is_atomic       : 1
 //   is_read         : 1
 //   size_log        : 2
 //   addr0           : 3
+//   epoch           : kClkBits
 class Shadow : public FastState {
  public:
   explicit Shadow(u64 x)
@@ -243,10 +172,10 @@ class Shadow : public FastState {
   }
 
   void SetAddr0AndSizeLog(u64 addr0, unsigned kAccessSizeLog) {
-    DCHECK_EQ(x_ & 31, 0);
+    DCHECK_EQ((x_ >> kClkBits) & 31, 0);
     DCHECK_LE(addr0, 7);
     DCHECK_LE(kAccessSizeLog, 3);
-    x_ |= (kAccessSizeLog << 3) | addr0;
+    x_ |= ((kAccessSizeLog << 3) | addr0) << kClkBits;
     DCHECK_EQ(kAccessSizeLog, size_log());
     DCHECK_EQ(addr0, this->addr0());
   }
@@ -279,47 +208,34 @@ class Shadow : public FastState {
     return shifted_xor == 0;
   }
 
-  static inline bool Addr0AndSizeAreEqual(const Shadow s1, const Shadow s2) {
-    u64 masked_xor = (s1.x_ ^ s2.x_) & 31;
+  static ALWAYS_INLINE
+  bool Addr0AndSizeAreEqual(const Shadow s1, const Shadow s2) {
+    u64 masked_xor = ((s1.x_ ^ s2.x_) >> kClkBits) & 31;
     return masked_xor == 0;
   }
 
-  static inline bool TwoRangesIntersect(Shadow s1, Shadow s2,
+  static ALWAYS_INLINE bool TwoRangesIntersect(Shadow s1, Shadow s2,
       unsigned kS2AccessSize) {
     bool res = false;
     u64 diff = s1.addr0() - s2.addr0();
     if ((s64)diff < 0) {  // s1.addr0 < s2.addr0  // NOLINT
       // if (s1.addr0() + size1) > s2.addr0()) return true;
-      if (s1.size() > -diff)  res = true;
+      if (s1.size() > -diff)
+        res = true;
     } else {
       // if (s2.addr0() + kS2AccessSize > s1.addr0()) return true;
-      if (kS2AccessSize > diff) res = true;
+      if (kS2AccessSize > diff)
+        res = true;
     }
-    DCHECK_EQ(res, TwoRangesIntersectSLOW(s1, s2));
-    DCHECK_EQ(res, TwoRangesIntersectSLOW(s2, s1));
+    DCHECK_EQ(res, TwoRangesIntersectSlow(s1, s2));
+    DCHECK_EQ(res, TwoRangesIntersectSlow(s2, s1));
     return res;
   }
 
-  // The idea behind the offset is as follows.
-  // Consider that we have 8 bool's contained within a single 8-byte block
-  // (mapped to a single shadow "cell"). Now consider that we write to the bools
-  // from a single thread (which we consider the common case).
-  // W/o offsetting each access will have to scan 4 shadow values at average
-  // to find the corresponding shadow value for the bool.
-  // With offsetting we start scanning shadow with the offset so that
-  // each access hits necessary shadow straight off (at least in an expected
-  // optimistic case).
-  // This logic works seamlessly for any layout of user data. For example,
-  // if user data is {int, short, char, char}, then accesses to the int are
-  // offsetted to 0, short - 4, 1st char - 6, 2nd char - 7. Hopefully, accesses
-  // from a single thread won't need to scan all 8 shadow values.
-  unsigned ComputeSearchOffset() {
-    return x_ & 7;
-  }
-  u64 addr0() const { return x_ & 7; }
-  u64 size() const { return 1ull << size_log(); }
-  bool IsWrite() const { return !IsRead(); }
-  bool IsRead() const { return x_ & kReadBit; }
+  u64 ALWAYS_INLINE addr0() const { return (x_ >> kClkBits) & 7; }
+  u64 ALWAYS_INLINE size() const { return 1ull << size_log(); }
+  bool ALWAYS_INLINE IsWrite() const { return !IsRead(); }
+  bool ALWAYS_INLINE IsRead() const { return x_ & kReadBit; }
 
   // The idea behind the freed bit is as follows.
   // When the memory is freed (or otherwise unaccessible) we write to the shadow
@@ -344,15 +260,14 @@ class Shadow : public FastState {
     return res;
   }
 
-  bool IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
-    // analyzes 5-th bit (is_read) and 6-th bit (is_atomic)
-    bool v = x_ & u64(((kIsWrite ^ 1) << kReadShift)
-        | (kIsAtomic << kAtomicShift));
+  bool ALWAYS_INLINE IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
+    bool v = x_ & ((u64(kIsWrite ^ 1) << kReadShift)
+        | (u64(kIsAtomic) << kAtomicShift));
     DCHECK_EQ(v, (!IsWrite() && !kIsWrite) || (IsAtomic() && kIsAtomic));
     return v;
   }
 
-  bool IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
+  bool ALWAYS_INLINE IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
     bool v = ((x_ >> kReadShift) & 3)
         <= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
     DCHECK_EQ(v, (IsAtomic() < kIsAtomic) ||
@@ -360,7 +275,7 @@ class Shadow : public FastState {
     return v;
   }
 
-  bool IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
+  bool ALWAYS_INLINE IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
     bool v = ((x_ >> kReadShift) & 3)
         >= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
     DCHECK_EQ(v, (IsAtomic() > kIsAtomic) ||
@@ -369,14 +284,14 @@ class Shadow : public FastState {
   }
 
  private:
-  static const u64 kReadShift   = 5;
+  static const u64 kReadShift   = 5 + kClkBits;
   static const u64 kReadBit     = 1ull << kReadShift;
-  static const u64 kAtomicShift = 6;
+  static const u64 kAtomicShift = 6 + kClkBits;
   static const u64 kAtomicBit   = 1ull << kAtomicShift;
 
-  u64 size_log() const { return (x_ >> 3) & 3; }
+  u64 size_log() const { return (x_ >> (3 + kClkBits)) & 3; }
 
-  static bool TwoRangesIntersectSLOW(const Shadow s1, const Shadow s2) {
+  static bool TwoRangesIntersectSlow(const Shadow s1, const Shadow s2) {
     if (s1.addr0() == s2.addr0()) return true;
     if (s1.addr0() < s2.addr0() && s1.addr0() + s1.size() > s2.addr0())
       return true;
@@ -391,6 +306,9 @@ struct SignalContext;
 struct JmpBuf {
   uptr sp;
   uptr mangled_sp;
+  int int_signal_send;
+  bool in_blocking_func;
+  uptr in_signal_handler;
   uptr *shadow_stack_pos;
 };
 
@@ -439,7 +357,7 @@ struct ThreadState {
   const int unique_id;
   bool in_symbolizer;
   bool in_ignored_lib;
-  bool is_alive;
+  bool is_dead;
   bool is_freeing;
   bool is_vptr_access;
   const uptr stk_addr;
@@ -452,9 +370,13 @@ struct ThreadState {
   DDPhysicalThread *dd_pt;
   DDLogicalThread *dd_lt;
 
-  bool in_signal_handler;
+  atomic_uintptr_t in_signal_handler;
   SignalContext *signal_ctx;
 
+  DenseSlabAllocCache block_cache;
+  DenseSlabAllocCache sync_cache;
+  DenseSlabAllocCache clock_cache;
+
 #ifndef TSAN_GO
   u32 last_sleep_stack_id;
   ThreadClock last_sleep_clock;
@@ -498,6 +420,7 @@ class ThreadContext : public ThreadContextBase {
   void OnStarted(void *arg);
   void OnCreated(void *arg);
   void OnReset();
+  void OnDetached(void *arg);
 };
 
 struct RacyStacks {
@@ -528,7 +451,7 @@ struct Context {
   bool initialized;
   bool after_multithreaded_fork;
 
-  SyncTab synctab;
+  MetaMap metamap;
 
   Mutex report_mtx;
   int nreported;
@@ -546,6 +469,8 @@ struct Context {
   InternalMmapVector<FiredSuppression> fired_suppressions;
   DDetector *dd;
 
+  ClockAlloc clock_alloc;
+
   Flags flags;
 
   u64 stat[StatCnt];
@@ -574,11 +499,11 @@ class ScopedReport {
   explicit ScopedReport(ReportType typ);
   ~ScopedReport();
 
-  void AddStack(const StackTrace *stack);
   void AddMemoryAccess(uptr addr, Shadow s, const StackTrace *stack,
                        const MutexSet *mset);
-  void AddThread(const ThreadContext *tctx);
-  void AddThread(int unique_tid);
+  void AddStack(const StackTrace *stack, bool suppressable = false);
+  void AddThread(const ThreadContext *tctx, bool suppressable = false);
+  void AddThread(int unique_tid, bool suppressable = false);
   void AddUniqueTid(int unique_tid);
   void AddMutex(const SyncVar *s);
   u64 AddMutex(u64 id);
@@ -626,11 +551,7 @@ void ForkParentAfter(ThreadState *thr, uptr pc);
 void ForkChildAfter(ThreadState *thr, uptr pc);
 
 void ReportRace(ThreadState *thr);
-bool OutputReport(Context *ctx,
-                  const ScopedReport &srep,
-                  const ReportStack *suppress_stack1,
-                  const ReportStack *suppress_stack2 = 0,
-                  const ReportLocation *suppress_loc = 0);
+bool OutputReport(ThreadState *thr, const ScopedReport &srep);
 bool IsFiredSuppression(Context *ctx,
                         const ScopedReport &srep,
                         const StackTrace &trace);
@@ -659,9 +580,8 @@ void PrintCurrentStackSlow();  // uses libunwind
 void Initialize(ThreadState *thr);
 int Finalize(ThreadState *thr);
 
-SyncVar* GetJavaSync(ThreadState *thr, uptr pc, uptr addr,
-                     bool write_lock, bool create);
-SyncVar* GetAndRemoveJavaSync(ThreadState *thr, uptr pc, uptr addr);
+void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write);
+void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write);
 
 void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
     int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic);
diff --git a/libsanitizer/tsan/tsan_rtl_mutex.cc b/libsanitizer/tsan/tsan_rtl_mutex.cc
index 3724571cfff..cc183138aba 100644
--- a/libsanitizer/tsan/tsan_rtl_mutex.cc
+++ b/libsanitizer/tsan/tsan_rtl_mutex.cc
@@ -50,14 +50,18 @@ void DDMutexInit(ThreadState *thr, uptr pc, SyncVar *s) {
 
 static void ReportMutexMisuse(ThreadState *thr, uptr pc, ReportType typ,
     uptr addr, u64 mid) {
+  // In Go, these misuses are either impossible, or detected by std lib,
+  // or false positives (e.g. unlock in a different thread).
+  if (kGoMode)
+    return;
   ThreadRegistryLock l(ctx->thread_registry);
   ScopedReport rep(typ);
   rep.AddMutex(mid);
   StackTrace trace;
   trace.ObtainCurrent(thr, pc);
-  rep.AddStack(&trace);
+  rep.AddStack(&trace, true);
   rep.AddLocation(addr, 1);
-  OutputReport(ctx, rep, rep.GetReport()->stacks[0]);
+  OutputReport(thr, rep);
 }
 
 void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
@@ -70,10 +74,12 @@ void MutexCreate(ThreadState *thr, uptr pc, uptr addr,
     MemoryWrite(thr, pc, addr, kSizeLog1);
     thr->is_freeing = false;
   }
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   s->is_rw = rw;
   s->is_recursive = recursive;
   s->is_linker_init = linker_init;
+  if (kCppMode && s->creation_stack_id == 0)
+    s->creation_stack_id = CurrentStackId(thr, pc);
   s->mtx.Unlock();
 }
 
@@ -86,37 +92,54 @@ void MutexDestroy(ThreadState *thr, uptr pc, uptr addr) {
   if (IsGlobalVar(addr))
     return;
 #endif
-  SyncVar *s = ctx->synctab.GetAndRemove(thr, pc, addr);
-  if (s == 0)
-    return;
-  if (flags()->detect_deadlocks) {
-    Callback cb(thr, pc);
-    ctx->dd->MutexDestroy(&cb, &s->dd);
-  }
   if (IsAppMem(addr)) {
     CHECK(!thr->is_freeing);
     thr->is_freeing = true;
     MemoryWrite(thr, pc, addr, kSizeLog1);
     thr->is_freeing = false;
   }
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
+  if (s == 0)
+    return;
+  if (common_flags()->detect_deadlocks) {
+    Callback cb(thr, pc);
+    ctx->dd->MutexDestroy(&cb, &s->dd);
+    ctx->dd->MutexInit(&cb, &s->dd);
+  }
+  bool unlock_locked = false;
   if (flags()->report_destroy_locked
       && s->owner_tid != SyncVar::kInvalidTid
       && !s->is_broken) {
     s->is_broken = true;
+    unlock_locked = true;
+  }
+  u64 mid = s->GetId();
+  u32 last_lock = s->last_lock;
+  if (!unlock_locked)
+    s->Reset(thr);  // must not reset it before the report is printed
+  s->mtx.Unlock();
+  if (unlock_locked) {
     ThreadRegistryLock l(ctx->thread_registry);
     ScopedReport rep(ReportTypeMutexDestroyLocked);
-    rep.AddMutex(s);
+    rep.AddMutex(mid);
     StackTrace trace;
     trace.ObtainCurrent(thr, pc);
     rep.AddStack(&trace);
-    FastState last(s->last_lock);
+    FastState last(last_lock);
     RestoreStack(last.tid(), last.epoch(), &trace, 0);
-    rep.AddStack(&trace);
-    rep.AddLocation(s->addr, 1);
-    OutputReport(ctx, rep, rep.GetReport()->stacks[0]);
+    rep.AddStack(&trace, true);
+    rep.AddLocation(addr, 1);
+    OutputReport(thr, rep);
+  }
+  if (unlock_locked) {
+    SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
+    if (s != 0) {
+      s->Reset(thr);
+      s->mtx.Unlock();
+    }
   }
-  thr->mset.Remove(s->GetId());
-  DestroyAndFree(s);
+  thr->mset.Remove(mid);
+  // s will be destroyed and freed in MetaMap::FreeBlock.
 }
 
 void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec, bool try_lock) {
@@ -124,7 +147,7 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec, bool try_lock) {
   CHECK_GT(rec, 0);
   if (IsAppMem(addr))
     MemoryReadAtomic(thr, pc, addr, kSizeLog1);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeLock, s->GetId());
   bool report_double_lock = false;
@@ -147,7 +170,7 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec, bool try_lock) {
   }
   s->recursion += rec;
   thr->mset.Add(s->GetId(), true, thr->fast_state.epoch());
-  if (flags()->detect_deadlocks && s->recursion == 1) {
+  if (common_flags()->detect_deadlocks && (s->recursion - rec) == 0) {
     Callback cb(thr, pc);
     if (!try_lock)
       ctx->dd->MutexBeforeLock(&cb, &s->dd, true);
@@ -158,7 +181,7 @@ void MutexLock(ThreadState *thr, uptr pc, uptr addr, int rec, bool try_lock) {
   // Can't touch s after this point.
   if (report_double_lock)
     ReportMutexMisuse(thr, pc, ReportTypeMutexDoubleLock, addr, mid);
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     Callback cb(thr, pc);
     ReportDeadlock(thr, pc, ctx->dd->GetReport(&cb));
   }
@@ -168,12 +191,12 @@ int MutexUnlock(ThreadState *thr, uptr pc, uptr addr, bool all) {
   DPrintf("#%d: MutexUnlock %zx all=%d\n", thr->tid, addr, all);
   if (IsAppMem(addr))
     MemoryReadAtomic(thr, pc, addr, kSizeLog1);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeUnlock, s->GetId());
   int rec = 0;
   bool report_bad_unlock = false;
-  if (s->recursion == 0 || s->owner_tid != thr->tid) {
+  if (kCppMode && (s->recursion == 0 || s->owner_tid != thr->tid)) {
     if (flags()->report_mutex_bugs && !s->is_broken) {
       s->is_broken = true;
       report_bad_unlock = true;
@@ -190,7 +213,8 @@ int MutexUnlock(ThreadState *thr, uptr pc, uptr addr, bool all) {
     }
   }
   thr->mset.Del(s->GetId(), true);
-  if (flags()->detect_deadlocks && s->recursion == 0) {
+  if (common_flags()->detect_deadlocks && s->recursion == 0 &&
+      !report_bad_unlock) {
     Callback cb(thr, pc);
     ctx->dd->MutexBeforeUnlock(&cb, &s->dd, true);
   }
@@ -199,7 +223,7 @@ int MutexUnlock(ThreadState *thr, uptr pc, uptr addr, bool all) {
   // Can't touch s after this point.
   if (report_bad_unlock)
     ReportMutexMisuse(thr, pc, ReportTypeMutexBadUnlock, addr, mid);
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks && !report_bad_unlock) {
     Callback cb(thr, pc);
     ReportDeadlock(thr, pc, ctx->dd->GetReport(&cb));
   }
@@ -211,7 +235,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool trylock) {
   StatInc(thr, StatMutexReadLock);
   if (IsAppMem(addr))
     MemoryReadAtomic(thr, pc, addr, kSizeLog1);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, false);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, false);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeRLock, s->GetId());
   bool report_bad_lock = false;
@@ -224,7 +248,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool trylock) {
   AcquireImpl(thr, pc, &s->clock);
   s->last_lock = thr->fast_state.raw();
   thr->mset.Add(s->GetId(), false, thr->fast_state.epoch());
-  if (flags()->detect_deadlocks && s->recursion == 0) {
+  if (common_flags()->detect_deadlocks && s->recursion == 0) {
     Callback cb(thr, pc);
     if (!trylock)
       ctx->dd->MutexBeforeLock(&cb, &s->dd, false);
@@ -235,7 +259,7 @@ void MutexReadLock(ThreadState *thr, uptr pc, uptr addr, bool trylock) {
   // Can't touch s after this point.
   if (report_bad_lock)
     ReportMutexMisuse(thr, pc, ReportTypeMutexBadReadLock, addr, mid);
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     Callback cb(thr, pc);
     ReportDeadlock(thr, pc, ctx->dd->GetReport(&cb));
   }
@@ -246,7 +270,7 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
   StatInc(thr, StatMutexReadUnlock);
   if (IsAppMem(addr))
     MemoryReadAtomic(thr, pc, addr, kSizeLog1);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   TraceAddEvent(thr, thr->fast_state, EventTypeRUnlock, s->GetId());
   bool report_bad_unlock = false;
@@ -257,7 +281,7 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
     }
   }
   ReleaseImpl(thr, pc, &s->read_clock);
-  if (flags()->detect_deadlocks && s->recursion == 0) {
+  if (common_flags()->detect_deadlocks && s->recursion == 0) {
     Callback cb(thr, pc);
     ctx->dd->MutexBeforeUnlock(&cb, &s->dd, false);
   }
@@ -267,7 +291,7 @@ void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr) {
   thr->mset.Del(mid, false);
   if (report_bad_unlock)
     ReportMutexMisuse(thr, pc, ReportTypeMutexBadReadUnlock, addr, mid);
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     Callback cb(thr, pc);
     ReportDeadlock(thr, pc, ctx->dd->GetReport(&cb));
   }
@@ -277,7 +301,7 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexReadOrWriteUnlock %zx\n", thr->tid, addr);
   if (IsAppMem(addr))
     MemoryReadAtomic(thr, pc, addr, kSizeLog1);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   bool write = true;
   bool report_bad_unlock = false;
   if (s->owner_tid == SyncVar::kInvalidTid) {
@@ -305,7 +329,7 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
     report_bad_unlock = true;
   }
   thr->mset.Del(s->GetId(), write);
-  if (flags()->detect_deadlocks && s->recursion == 0) {
+  if (common_flags()->detect_deadlocks && s->recursion == 0) {
     Callback cb(thr, pc);
     ctx->dd->MutexBeforeUnlock(&cb, &s->dd, write);
   }
@@ -314,7 +338,7 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
   // Can't touch s after this point.
   if (report_bad_unlock)
     ReportMutexMisuse(thr, pc, ReportTypeMutexBadUnlock, addr, mid);
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     Callback cb(thr, pc);
     ReportDeadlock(thr, pc, ctx->dd->GetReport(&cb));
   }
@@ -322,7 +346,7 @@ void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr) {
 
 void MutexRepair(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: MutexRepair %zx\n", thr->tid, addr);
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   s->owner_tid = SyncVar::kInvalidTid;
   s->recursion = 0;
   s->mtx.Unlock();
@@ -332,7 +356,7 @@ void Acquire(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: Acquire %zx\n", thr->tid, addr);
   if (thr->ignore_sync)
     return;
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, false);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, false);
   AcquireImpl(thr, pc, &s->clock);
   s->mtx.ReadUnlock();
 }
@@ -359,7 +383,7 @@ void Release(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: Release %zx\n", thr->tid, addr);
   if (thr->ignore_sync)
     return;
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   // Can't increment epoch w/o writing to the trace as well.
   TraceAddEvent(thr, thr->fast_state, EventTypeMop, 0);
@@ -371,7 +395,7 @@ void ReleaseStore(ThreadState *thr, uptr pc, uptr addr) {
   DPrintf("#%d: ReleaseStore %zx\n", thr->tid, addr);
   if (thr->ignore_sync)
     return;
-  SyncVar *s = ctx->synctab.GetOrCreateAndLock(thr, pc, addr, true);
+  SyncVar *s = ctx->metamap.GetOrCreateAndLock(thr, pc, addr, true);
   thr->fast_state.IncrementEpoch();
   // Can't increment epoch w/o writing to the trace as well.
   TraceAddEvent(thr, thr->fast_state, EventTypeMop, 0);
@@ -404,7 +428,7 @@ void AcquireImpl(ThreadState *thr, uptr pc, SyncClock *c) {
   if (thr->ignore_sync)
     return;
   thr->clock.set(thr->fast_state.epoch());
-  thr->clock.acquire(c);
+  thr->clock.acquire(&thr->clock_cache, c);
   StatInc(thr, StatSyncAcquire);
 }
 
@@ -413,7 +437,7 @@ void ReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.release(c);
+  thr->clock.release(&thr->clock_cache, c);
   StatInc(thr, StatSyncRelease);
 }
 
@@ -422,7 +446,7 @@ void ReleaseStoreImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.ReleaseStore(c);
+  thr->clock.ReleaseStore(&thr->clock_cache, c);
   StatInc(thr, StatSyncRelease);
 }
 
@@ -431,7 +455,7 @@ void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c) {
     return;
   thr->clock.set(thr->fast_state.epoch());
   thr->fast_synch_epoch = thr->fast_state.epoch();
-  thr->clock.acq_rel(c);
+  thr->clock.acq_rel(&thr->clock_cache, c);
   StatInc(thr, StatSyncAcquire);
   StatInc(thr, StatSyncRelease);
 }
@@ -446,7 +470,7 @@ void ReportDeadlock(ThreadState *thr, uptr pc, DDReport *r) {
     rep.AddUniqueTid((int)r->loop[i].thr_ctx);
     rep.AddThread((int)r->loop[i].thr_ctx);
   }
-  StackTrace stacks[2 * DDReport::kMaxLoopSize];
+  InternalScopedBuffer<StackTrace> stacks(2 * DDReport::kMaxLoopSize);
   uptr dummy_pc = 0x42;
   for (int i = 0; i < r->n; i++) {
     uptr size;
@@ -460,12 +484,10 @@ void ReportDeadlock(ThreadState *thr, uptr pc, DDReport *r) {
         // but we should still produce some stack trace in the report.
         stacks[i].Init(&dummy_pc, 1);
       }
-      rep.AddStack(&stacks[i]);
+      rep.AddStack(&stacks[i], true);
     }
   }
-  // FIXME: use all stacks for suppressions, not just the second stack of the
-  // first edge.
-  OutputReport(ctx, rep, rep.GetReport()->stacks[0]);
+  OutputReport(thr, rep);
 }
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_rtl_report.cc b/libsanitizer/tsan/tsan_rtl_report.cc
index d19deb066df..eafd1f4dfcd 100644
--- a/libsanitizer/tsan/tsan_rtl_report.cc
+++ b/libsanitizer/tsan/tsan_rtl_report.cc
@@ -59,7 +59,7 @@ static void StackStripMain(ReportStack *stack) {
   ReportStack *last_frame2 = 0;
   const char *prefix = "__interceptor_";
   uptr prefix_len = internal_strlen(prefix);
-  const char *path_prefix = flags()->strip_path_prefix;
+  const char *path_prefix = common_flags()->strip_path_prefix;
   uptr path_prefix_len = internal_strlen(path_prefix);
   char *pos;
   for (ReportStack *ent = stack; ent; ent = ent->next) {
@@ -160,9 +160,10 @@ ScopedReport::~ScopedReport() {
   DestroyAndFree(rep_);
 }
 
-void ScopedReport::AddStack(const StackTrace *stack) {
+void ScopedReport::AddStack(const StackTrace *stack, bool suppressable) {
   ReportStack **rs = rep_->stacks.PushBack();
   *rs = SymbolizeStack(*stack);
+  (*rs)->suppressable = suppressable;
 }
 
 void ScopedReport::AddMemoryAccess(uptr addr, Shadow s,
@@ -176,6 +177,8 @@ void ScopedReport::AddMemoryAccess(uptr addr, Shadow s,
   mop->write = s.IsWrite();
   mop->atomic = s.IsAtomic();
   mop->stack = SymbolizeStack(*stack);
+  if (mop->stack)
+    mop->stack->suppressable = true;
   for (uptr i = 0; i < mset->Size(); i++) {
     MutexSet::Desc d = mset->Get(i);
     u64 mid = this->AddMutex(d.id);
@@ -188,7 +191,7 @@ void ScopedReport::AddUniqueTid(int unique_tid) {
   rep_->unique_tids.PushBack(unique_tid);
 }
 
-void ScopedReport::AddThread(const ThreadContext *tctx) {
+void ScopedReport::AddThread(const ThreadContext *tctx, bool suppressable) {
   for (uptr i = 0; i < rep_->threads.Size(); i++) {
     if ((u32)rep_->threads[i]->id == tctx->tid)
       return;
@@ -203,6 +206,8 @@ void ScopedReport::AddThread(const ThreadContext *tctx) {
   rt->parent_tid = tctx->parent_tid;
   rt->stack = 0;
   rt->stack = SymbolizeStackId(tctx->creation_stack_id);
+  if (rt->stack)
+    rt->stack->suppressable = suppressable;
 }
 
 #ifndef TSAN_GO
@@ -249,9 +254,9 @@ ThreadContext *IsThreadStackOrTls(uptr addr, bool *is_stack) {
 }
 #endif
 
-void ScopedReport::AddThread(int unique_tid) {
+void ScopedReport::AddThread(int unique_tid, bool suppressable) {
 #ifndef TSAN_GO
-  AddThread(FindThreadByUidLocked(unique_tid));
+  AddThread(FindThreadByUidLocked(unique_tid), suppressable);
 #endif
 }
 
@@ -273,7 +278,7 @@ u64 ScopedReport::AddMutex(u64 id) {
   u64 uid = 0;
   u64 mid = id;
   uptr addr = SyncVar::SplitId(id, &uid);
-  SyncVar *s = ctx->synctab.GetIfExistsAndLock(addr, false);
+  SyncVar *s = ctx->metamap.GetIfExistsAndLock(addr);
   // Check that the mutex is still alive.
   // Another mutex can be created at the same address,
   // so check uid as well.
@@ -284,7 +289,7 @@ u64 ScopedReport::AddMutex(u64 id) {
     AddDeadMutex(id);
   }
   if (s)
-    s->mtx.ReadUnlock();
+    s->mtx.Unlock();
   return mid;
 }
 
@@ -309,8 +314,7 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
   int fd = -1;
   int creat_tid = -1;
   u32 creat_stack = 0;
-  if (FdLocation(addr, &fd, &creat_tid, &creat_stack)
-      || FdLocation(AlternativeAddress(addr), &fd, &creat_tid, &creat_stack)) {
+  if (FdLocation(addr, &fd, &creat_tid, &creat_stack)) {
     void *mem = internal_alloc(MBlockReportLoc, sizeof(ReportLocation));
     ReportLocation *loc = new(mem) ReportLocation();
     rep_->locs.PushBack(loc);
@@ -324,21 +328,26 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
     return;
   }
   MBlock *b = 0;
-  if (allocator()->PointerIsMine((void*)addr)
-      && (b = user_mblock(0, (void*)addr))) {
-    ThreadContext *tctx = FindThreadByTidLocked(b->Tid());
+  Allocator *a = allocator();
+  if (a->PointerIsMine((void*)addr)) {
+    void *block_begin = a->GetBlockBegin((void*)addr);
+    if (block_begin)
+      b = ctx->metamap.GetBlock((uptr)block_begin);
+  }
+  if (b != 0) {
+    ThreadContext *tctx = FindThreadByTidLocked(b->tid);
     void *mem = internal_alloc(MBlockReportLoc, sizeof(ReportLocation));
     ReportLocation *loc = new(mem) ReportLocation();
     rep_->locs.PushBack(loc);
     loc->type = ReportLocationHeap;
     loc->addr = (uptr)allocator()->GetBlockBegin((void*)addr);
-    loc->size = b->Size();
-    loc->tid = tctx ? tctx->tid : b->Tid();
+    loc->size = b->siz;
+    loc->tid = tctx ? tctx->tid : b->tid;
     loc->name = 0;
     loc->file = 0;
     loc->line = 0;
     loc->stack = 0;
-    loc->stack = SymbolizeStackId(b->StackId());
+    loc->stack = SymbolizeStackId(b->stk);
     if (tctx)
       AddThread(tctx);
     return;
@@ -354,6 +363,7 @@ void ScopedReport::AddLocation(uptr addr, uptr size) {
   }
   ReportLocation *loc = SymbolizeData(addr);
   if (loc) {
+    loc->suppressable = true;
     rep_->locs.PushBack(loc);
     return;
   }
@@ -493,25 +503,31 @@ static void AddRacyStacks(ThreadState *thr, const StackTrace (&traces)[2],
   }
 }
 
-bool OutputReport(Context *ctx,
-                  const ScopedReport &srep,
-                  const ReportStack *suppress_stack1,
-                  const ReportStack *suppress_stack2,
-                  const ReportLocation *suppress_loc) {
+bool OutputReport(ThreadState *thr, const ScopedReport &srep) {
   atomic_store(&ctx->last_symbolize_time_ns, NanoTime(), memory_order_relaxed);
   const ReportDesc *rep = srep.GetReport();
   Suppression *supp = 0;
-  uptr suppress_pc = IsSuppressed(rep->typ, suppress_stack1, &supp);
-  if (suppress_pc == 0)
-    suppress_pc = IsSuppressed(rep->typ, suppress_stack2, &supp);
-  if (suppress_pc == 0)
-    suppress_pc = IsSuppressed(rep->typ, suppress_loc, &supp);
+  uptr suppress_pc = 0;
+  for (uptr i = 0; suppress_pc == 0 && i < rep->mops.Size(); i++)
+    suppress_pc = IsSuppressed(rep->typ, rep->mops[i]->stack, &supp);
+  for (uptr i = 0; suppress_pc == 0 && i < rep->stacks.Size(); i++)
+    suppress_pc = IsSuppressed(rep->typ, rep->stacks[i], &supp);
+  for (uptr i = 0; suppress_pc == 0 && i < rep->threads.Size(); i++)
+    suppress_pc = IsSuppressed(rep->typ, rep->threads[i]->stack, &supp);
+  for (uptr i = 0; suppress_pc == 0 && i < rep->locs.Size(); i++)
+    suppress_pc = IsSuppressed(rep->typ, rep->locs[i], &supp);
   if (suppress_pc != 0) {
     FiredSuppression s = {srep.GetReport()->typ, suppress_pc, supp};
     ctx->fired_suppressions.push_back(s);
   }
-  if (OnReport(rep, suppress_pc != 0))
-    return false;
+  {
+    bool old_is_freeing = thr->is_freeing;
+    thr->is_freeing = false;
+    bool suppressed = OnReport(rep, suppress_pc != 0);
+    thr->is_freeing = old_is_freeing;
+    if (suppressed)
+      return false;
+  }
   PrintReport(rep);
   ctx->nreported++;
   if (flags()->halt_on_error)
@@ -560,41 +576,6 @@ bool FrameIsInternal(const ReportStack *frame) {
           internal_strstr(frame->file, "tsan_interface_"));
 }
 
-// On programs that use Java we see weird reports like:
-// WARNING: ThreadSanitizer: data race (pid=22512)
-//   Read of size 8 at 0x7d2b00084318 by thread 100:
-//     #0 memcpy tsan_interceptors.cc:406 (foo+0x00000d8dfae3)
-//     #1 <null> <null>:0 (0x7f7ad9b40193)
-//   Previous write of size 8 at 0x7d2b00084318 by thread 105:
-//     #0 strncpy tsan_interceptors.cc:501 (foo+0x00000d8e0919)
-//     #1 <null> <null>:0 (0x7f7ad9b42707)
-static bool IsJavaNonsense(const ReportDesc *rep) {
-#ifndef TSAN_GO
-  for (uptr i = 0; i < rep->mops.Size(); i++) {
-    ReportMop *mop = rep->mops[i];
-    ReportStack *frame = mop->stack;
-    if (frame == 0
-        || (frame->func == 0 && frame->file == 0 && frame->line == 0
-          && frame->module == 0)) {
-      return true;
-    }
-    if (FrameIsInternal(frame)) {
-      frame = frame->next;
-      if (frame == 0
-          || (frame->func == 0 && frame->file == 0 && frame->line == 0
-          && frame->module == 0)) {
-        if (frame) {
-          FiredSuppression supp = {rep->typ, frame->pc, 0};
-          ctx->fired_suppressions.push_back(supp);
-        }
-        return true;
-      }
-    }
-  }
-#endif
-  return false;
-}
-
 static bool RaceBetweenAtomicAndFree(ThreadState *thr) {
   Shadow s0(thr->racy_state[0]);
   Shadow s1(thr->racy_state[1]);
@@ -609,6 +590,8 @@ static bool RaceBetweenAtomicAndFree(ThreadState *thr) {
 }
 
 void ReportRace(ThreadState *thr) {
+  CheckNoLocks(thr);
+
   // Symbolizer makes lots of intercepted calls. If we try to process them,
   // at best it will cause deadlocks on internal mutexes.
   ScopedIgnoreInterceptors ignore;
@@ -671,9 +654,6 @@ void ReportRace(ThreadState *thr) {
                         i == 0 ? &thr->mset : mset2.data());
   }
 
-  if (flags()->suppress_java && IsJavaNonsense(rep.GetReport()))
-    return;
-
   for (uptr i = 0; i < kMop; i++) {
     FastState s(thr->racy_state[i]);
     ThreadContext *tctx = static_cast<ThreadContext*>(
@@ -693,11 +673,7 @@ void ReportRace(ThreadState *thr) {
   }
 #endif
 
-  ReportLocation *suppress_loc = rep.GetReport()->locs.Size() ?
-                                 rep.GetReport()->locs[0] : 0;
-  if (!OutputReport(ctx, rep, rep.GetReport()->mops[0]->stack,
-                              rep.GetReport()->mops[1]->stack,
-                              suppress_loc))
+  if (!OutputReport(thr, rep))
     return;
 
   AddRacyStacks(thr, traces, addr_min, addr_max);
diff --git a/libsanitizer/tsan/tsan_rtl_thread.cc b/libsanitizer/tsan/tsan_rtl_thread.cc
index 385af7e1fa6..0c0acc2787f 100644
--- a/libsanitizer/tsan/tsan_rtl_thread.cc
+++ b/libsanitizer/tsan/tsan_rtl_thread.cc
@@ -34,13 +34,13 @@ ThreadContext::~ThreadContext() {
 #endif
 
 void ThreadContext::OnDead() {
-  sync.Reset();
+  CHECK_EQ(sync.size(), 0);
 }
 
 void ThreadContext::OnJoined(void *arg) {
   ThreadState *caller_thr = static_cast<ThreadState *>(arg);
   AcquireImpl(caller_thr, 0, &sync);
-  sync.Reset();
+  sync.Reset(&caller_thr->clock_cache);
 }
 
 struct OnCreatedArgs {
@@ -63,11 +63,16 @@ void ThreadContext::OnCreated(void *arg) {
 }
 
 void ThreadContext::OnReset() {
-  sync.Reset();
+  CHECK_EQ(sync.size(), 0);
   FlushUnneededShadowMemory(GetThreadTrace(tid), TraceSize() * sizeof(Event));
   //!!! FlushUnneededShadowMemory(GetThreadTraceHeader(tid), sizeof(Trace));
 }
 
+void ThreadContext::OnDetached(void *arg) {
+  ThreadState *thr1 = static_cast<ThreadState*>(arg);
+  sync.Reset(&thr1->clock_cache);
+}
+
 struct OnStartedArgs {
   ThreadState *thr;
   uptr stk_addr;
@@ -100,7 +105,7 @@ void ThreadContext::OnStarted(void *arg) {
 #ifndef TSAN_GO
   AllocatorThreadStart(thr);
 #endif
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     thr->dd_pt = ctx->dd->CreatePhysicalThread();
     thr->dd_lt = ctx->dd->CreateLogicalThread(unique_id);
   }
@@ -111,12 +116,11 @@ void ThreadContext::OnStarted(void *arg) {
   Trace *thr_trace = ThreadTrace(thr->tid);
   thr_trace->headers[trace].epoch0 = epoch0;
   StatInc(thr, StatSyncAcquire);
-  sync.Reset();
+  sync.Reset(&thr->clock_cache);
   DPrintf("#%d: ThreadStart epoch=%zu stk_addr=%zx stk_size=%zx "
           "tls_addr=%zx tls_size=%zx\n",
           tid, (uptr)epoch0, args->stk_addr, args->stk_size,
           args->tls_addr, args->tls_size);
-  thr->is_alive = true;
 }
 
 void ThreadContext::OnFinished() {
@@ -128,10 +132,12 @@ void ThreadContext::OnFinished() {
   }
   epoch1 = thr->fast_state.epoch();
 
-  if (flags()->detect_deadlocks) {
+  if (common_flags()->detect_deadlocks) {
     ctx->dd->DestroyPhysicalThread(thr->dd_pt);
     ctx->dd->DestroyLogicalThread(thr->dd_lt);
   }
+  ctx->clock_alloc.FlushCache(&thr->clock_cache);
+  ctx->metamap.OnThreadIdle(thr);
 #ifndef TSAN_GO
   AllocatorThreadFinish(thr);
 #endif
@@ -203,9 +209,9 @@ void ThreadFinalize(ThreadState *thr) {
       MaybeReportThreadLeak, &leaks);
   for (uptr i = 0; i < leaks.Size(); i++) {
     ScopedReport rep(ReportTypeThreadLeak);
-    rep.AddThread(leaks[i].tctx);
+    rep.AddThread(leaks[i].tctx, true);
     rep.SetCount(leaks[i].count);
-    OutputReport(ctx, rep, rep.GetReport()->threads[0]->stack);
+    OutputReport(thr, rep);
   }
 #endif
 }
@@ -275,7 +281,7 @@ void ThreadFinish(ThreadState *thr) {
     DontNeedShadowFor(thr->stk_addr, thr->stk_size);
   if (thr->tls_addr && thr->tls_size)
     DontNeedShadowFor(thr->tls_addr, thr->tls_size);
-  thr->is_alive = false;
+  thr->is_dead = true;
   ctx->thread_registry->FinishThread(thr->tid);
 }
 
@@ -304,7 +310,7 @@ void ThreadJoin(ThreadState *thr, uptr pc, int tid) {
 void ThreadDetach(ThreadState *thr, uptr pc, int tid) {
   CHECK_GT(tid, 0);
   CHECK_LT(tid, kMaxTid);
-  ctx->thread_registry->DetachThread(tid);
+  ctx->thread_registry->DetachThread(tid, thr);
 }
 
 void ThreadSetName(ThreadState *thr, const char *name) {
diff --git a/libsanitizer/tsan/tsan_stack_trace.cc b/libsanitizer/tsan/tsan_stack_trace.cc
new file mode 100644
index 00000000000..45bd2517837
--- /dev/null
+++ b/libsanitizer/tsan/tsan_stack_trace.cc
@@ -0,0 +1,110 @@
+//===-- tsan_stack_trace.cc -----------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+//#include "sanitizer_common/sanitizer_placement_new.h"
+#include "tsan_stack_trace.h"
+#include "tsan_rtl.h"
+#include "tsan_mman.h"
+
+namespace __tsan {
+
+StackTrace::StackTrace()
+    : n_()
+    , s_()
+    , c_() {
+}
+
+StackTrace::StackTrace(uptr *buf, uptr cnt)
+    : n_()
+    , s_(buf)
+    , c_(cnt) {
+  CHECK_NE(buf, 0);
+  CHECK_NE(cnt, 0);
+}
+
+StackTrace::~StackTrace() {
+  Reset();
+}
+
+void StackTrace::Reset() {
+  if (s_ && !c_) {
+    CHECK_NE(n_, 0);
+    internal_free(s_);
+    s_ = 0;
+  }
+  n_ = 0;
+}
+
+void StackTrace::Init(const uptr *pcs, uptr cnt) {
+  Reset();
+  if (cnt == 0)
+    return;
+  if (c_) {
+    CHECK_NE(s_, 0);
+    CHECK_LE(cnt, c_);
+  } else {
+    s_ = (uptr*)internal_alloc(MBlockStackTrace, cnt * sizeof(s_[0]));
+  }
+  n_ = cnt;
+  internal_memcpy(s_, pcs, cnt * sizeof(s_[0]));
+}
+
+void StackTrace::ObtainCurrent(ThreadState *thr, uptr toppc) {
+  Reset();
+  n_ = thr->shadow_stack_pos - thr->shadow_stack;
+  if (n_ + !!toppc == 0)
+    return;
+  uptr start = 0;
+  if (c_) {
+    CHECK_NE(s_, 0);
+    if (n_ + !!toppc > c_) {
+      start = n_ - c_ + !!toppc;
+      n_ = c_ - !!toppc;
+    }
+  } else {
+    // Cap potentially huge stacks.
+    if (n_ + !!toppc > kTraceStackSize) {
+      start = n_ - kTraceStackSize + !!toppc;
+      n_ = kTraceStackSize - !!toppc;
+    }
+    s_ = (uptr*)internal_alloc(MBlockStackTrace,
+                               (n_ + !!toppc) * sizeof(s_[0]));
+  }
+  for (uptr i = 0; i < n_; i++)
+    s_[i] = thr->shadow_stack[start + i];
+  if (toppc) {
+    s_[n_] = toppc;
+    n_++;
+  }
+}
+
+void StackTrace::CopyFrom(const StackTrace& other) {
+  Reset();
+  Init(other.Begin(), other.Size());
+}
+
+bool StackTrace::IsEmpty() const {
+  return n_ == 0;
+}
+
+uptr StackTrace::Size() const {
+  return n_;
+}
+
+uptr StackTrace::Get(uptr i) const {
+  CHECK_LT(i, n_);
+  return s_[i];
+}
+
+const uptr *StackTrace::Begin() const {
+  return s_;
+}
+
+}  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_stack_trace.h b/libsanitizer/tsan/tsan_stack_trace.h
new file mode 100644
index 00000000000..ce0cb8859d2
--- /dev/null
+++ b/libsanitizer/tsan/tsan_stack_trace.h
@@ -0,0 +1,52 @@
+//===-- tsan_stack_trace.h --------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of ThreadSanitizer (TSan), a race detector.
+//
+//===----------------------------------------------------------------------===//
+#ifndef TSAN_STACK_TRACE_H
+#define TSAN_STACK_TRACE_H
+
+//#include "sanitizer_common/sanitizer_atomic.h"
+//#include "sanitizer_common/sanitizer_common.h"
+//#include "sanitizer_common/sanitizer_deadlock_detector_interface.h"
+#include "tsan_defs.h"
+//#include "tsan_clock.h"
+//#include "tsan_mutex.h"
+//#include "tsan_dense_alloc.h"
+
+namespace __tsan {
+
+class StackTrace {
+ public:
+  StackTrace();
+  // Initialized the object in "static mode",
+  // in this mode it never calls malloc/free but uses the provided buffer.
+  StackTrace(uptr *buf, uptr cnt);
+  ~StackTrace();
+  void Reset();
+
+  void Init(const uptr *pcs, uptr cnt);
+  void ObtainCurrent(ThreadState *thr, uptr toppc);
+  bool IsEmpty() const;
+  uptr Size() const;
+  uptr Get(uptr i) const;
+  const uptr *Begin() const;
+  void CopyFrom(const StackTrace& other);
+
+ private:
+  uptr n_;
+  uptr *s_;
+  const uptr c_;
+
+  StackTrace(const StackTrace&);
+  void operator = (const StackTrace&);
+};
+
+}  // namespace __tsan
+
+#endif  // TSAN_STACK_TRACE_H
diff --git a/libsanitizer/tsan/tsan_stat.cc b/libsanitizer/tsan/tsan_stat.cc
index e8d3a790b1c..cdb48358551 100644
--- a/libsanitizer/tsan/tsan_stat.cc
+++ b/libsanitizer/tsan/tsan_stat.cc
@@ -35,6 +35,7 @@ void StatOutput(u64 *stat) {
   name[StatMop4]                         = "            size 4                ";
   name[StatMop8]                         = "            size 8                ";
   name[StatMopSame]                      = "  Including same                  ";
+  name[StatMopIgnored]                   = "  Including ignored               ";
   name[StatMopRange]                     = "  Including range                 ";
   name[StatMopRodata]                    = "  Including .rodata               ";
   name[StatMopRangeRodata]               = "  Including .rodata range         ";
diff --git a/libsanitizer/tsan/tsan_stat.h b/libsanitizer/tsan/tsan_stat.h
index 5bdd9de8213..132656f035f 100644
--- a/libsanitizer/tsan/tsan_stat.h
+++ b/libsanitizer/tsan/tsan_stat.h
@@ -24,6 +24,7 @@ enum StatType {
   StatMop4,
   StatMop8,
   StatMopSame,
+  StatMopIgnored,
   StatMopRange,
   StatMopRodata,
   StatMopRangeRodata,
diff --git a/libsanitizer/tsan/tsan_suppressions.cc b/libsanitizer/tsan/tsan_suppressions.cc
index ce8d5fe8611..6b42d3a67b8 100644
--- a/libsanitizer/tsan/tsan_suppressions.cc
+++ b/libsanitizer/tsan/tsan_suppressions.cc
@@ -39,55 +39,16 @@ extern "C" const char *WEAK __tsan_default_suppressions() {
 
 namespace __tsan {
 
-static SuppressionContext* g_ctx;
-
-static char *ReadFile(const char *filename) {
-  if (filename == 0 || filename[0] == 0)
-    return 0;
-  InternalScopedBuffer<char> tmp(4*1024);
-  if (filename[0] == '/' || GetPwd() == 0)
-    internal_snprintf(tmp.data(), tmp.size(), "%s", filename);
-  else
-    internal_snprintf(tmp.data(), tmp.size(), "%s/%s", GetPwd(), filename);
-  uptr openrv = OpenFile(tmp.data(), false);
-  if (internal_iserror(openrv)) {
-    Printf("ThreadSanitizer: failed to open suppressions file '%s'\n",
-               tmp.data());
-    Die();
-  }
-  fd_t fd = openrv;
-  const uptr fsize = internal_filesize(fd);
-  if (fsize == (uptr)-1) {
-    Printf("ThreadSanitizer: failed to stat suppressions file '%s'\n",
-               tmp.data());
-    Die();
-  }
-  char *buf = (char*)internal_alloc(MBlockSuppression, fsize + 1);
-  if (fsize != internal_read(fd, buf, fsize)) {
-    Printf("ThreadSanitizer: failed to read suppressions file '%s'\n",
-               tmp.data());
-    Die();
-  }
-  internal_close(fd);
-  buf[fsize] = 0;
-  return buf;
-}
+static bool suppressions_inited = false;
 
 void InitializeSuppressions() {
-  ALIGNED(64) static char placeholder_[sizeof(SuppressionContext)];
-  g_ctx = new(placeholder_) SuppressionContext;
-  const char *supp = ReadFile(flags()->suppressions);
-  g_ctx->Parse(supp);
+  CHECK(!suppressions_inited);
+  SuppressionContext::InitIfNecessary();
 #ifndef TSAN_GO
-  supp = __tsan_default_suppressions();
-  g_ctx->Parse(supp);
-  g_ctx->Parse(std_suppressions);
+  SuppressionContext::Get()->Parse(__tsan_default_suppressions());
+  SuppressionContext::Get()->Parse(std_suppressions);
 #endif
-}
-
-SuppressionContext *GetSuppressionContext() {
-  CHECK_NE(g_ctx, 0);
-  return g_ctx;
+  suppressions_inited = true;
 }
 
 SuppressionType conv(ReportType typ) {
@@ -120,16 +81,17 @@ SuppressionType conv(ReportType typ) {
 }
 
 uptr IsSuppressed(ReportType typ, const ReportStack *stack, Suppression **sp) {
-  CHECK(g_ctx);
-  if (!g_ctx->SuppressionCount() || stack == 0) return 0;
+  if (!SuppressionContext::Get()->SuppressionCount() || stack == 0 ||
+      !stack->suppressable)
+    return 0;
   SuppressionType stype = conv(typ);
   if (stype == SuppressionNone)
     return 0;
   Suppression *s;
   for (const ReportStack *frame = stack; frame; frame = frame->next) {
-    if (g_ctx->Match(frame->func, stype, &s) ||
-        g_ctx->Match(frame->file, stype, &s) ||
-        g_ctx->Match(frame->module, stype, &s)) {
+    if (SuppressionContext::Get()->Match(frame->func, stype, &s) ||
+        SuppressionContext::Get()->Match(frame->file, stype, &s) ||
+        SuppressionContext::Get()->Match(frame->module, stype, &s)) {
       DPrintf("ThreadSanitizer: matched suppression '%s'\n", s->templ);
       s->hit_count++;
       *sp = s;
@@ -140,17 +102,16 @@ uptr IsSuppressed(ReportType typ, const ReportStack *stack, Suppression **sp) {
 }
 
 uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp) {
-  CHECK(g_ctx);
-  if (!g_ctx->SuppressionCount() || loc == 0 ||
-      loc->type != ReportLocationGlobal)
+  if (!SuppressionContext::Get()->SuppressionCount() || loc == 0 ||
+      loc->type != ReportLocationGlobal || !loc->suppressable)
     return 0;
   SuppressionType stype = conv(typ);
   if (stype == SuppressionNone)
     return 0;
   Suppression *s;
-  if (g_ctx->Match(loc->name, stype, &s) ||
-      g_ctx->Match(loc->file, stype, &s) ||
-      g_ctx->Match(loc->module, stype, &s)) {
+  if (SuppressionContext::Get()->Match(loc->name, stype, &s) ||
+      SuppressionContext::Get()->Match(loc->file, stype, &s) ||
+      SuppressionContext::Get()->Match(loc->module, stype, &s)) {
       DPrintf("ThreadSanitizer: matched suppression '%s'\n", s->templ);
       s->hit_count++;
       *sp = s;
@@ -160,9 +121,8 @@ uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp) {
 }
 
 void PrintMatchedSuppressions() {
-  CHECK(g_ctx);
   InternalMmapVector<Suppression *> matched(1);
-  g_ctx->GetMatched(&matched);
+  SuppressionContext::Get()->GetMatched(&matched);
   if (!matched.size())
     return;
   int hit_count = 0;
diff --git a/libsanitizer/tsan/tsan_suppressions.h b/libsanitizer/tsan/tsan_suppressions.h
index 2939e9a8b9f..e38d81ece85 100644
--- a/libsanitizer/tsan/tsan_suppressions.h
+++ b/libsanitizer/tsan/tsan_suppressions.h
@@ -20,7 +20,6 @@ void InitializeSuppressions();
 void PrintMatchedSuppressions();
 uptr IsSuppressed(ReportType typ, const ReportStack *stack, Suppression **sp);
 uptr IsSuppressed(ReportType typ, const ReportLocation *loc, Suppression **sp);
-SuppressionContext *GetSuppressionContext();
 
 }  // namespace __tsan
 
diff --git a/libsanitizer/tsan/tsan_symbolize.cc b/libsanitizer/tsan/tsan_symbolize.cc
index fa36017846c..49ae3dffa47 100644
--- a/libsanitizer/tsan/tsan_symbolize.cc
+++ b/libsanitizer/tsan/tsan_symbolize.cc
@@ -109,7 +109,7 @@ ReportStack *SymbolizeCode(uptr addr) {
   InternalScopedBuffer<AddressInfo> addr_frames(kMaxAddrFrames);
   for (uptr i = 0; i < kMaxAddrFrames; i++)
     new(&addr_frames[i]) AddressInfo();
-  uptr addr_frames_num = Symbolizer::Get()->SymbolizePC(
+  uptr addr_frames_num = Symbolizer::GetOrInit()->SymbolizePC(
       addr, addr_frames.data(), kMaxAddrFrames);
   if (addr_frames_num == 0)
     return NewReportStackEntry(addr);
@@ -130,7 +130,7 @@ ReportStack *SymbolizeCode(uptr addr) {
 
 ReportLocation *SymbolizeData(uptr addr) {
   DataInfo info;
-  if (!Symbolizer::Get()->SymbolizeData(addr, &info))
+  if (!Symbolizer::GetOrInit()->SymbolizeData(addr, &info))
     return 0;
   ReportLocation *ent = (ReportLocation*)internal_alloc(MBlockReportStack,
                                                         sizeof(ReportLocation));
@@ -146,7 +146,7 @@ ReportLocation *SymbolizeData(uptr addr) {
 }
 
 void SymbolizeFlush() {
-  Symbolizer::Get()->Flush();
+  Symbolizer::GetOrInit()->Flush();
 }
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_sync.cc b/libsanitizer/tsan/tsan_sync.cc
index f6f2cb731e7..2209199ac48 100644
--- a/libsanitizer/tsan/tsan_sync.cc
+++ b/libsanitizer/tsan/tsan_sync.cc
@@ -17,293 +17,207 @@ namespace __tsan {
 
 void DDMutexInit(ThreadState *thr, uptr pc, SyncVar *s);
 
-SyncVar::SyncVar(uptr addr, u64 uid)
-  : mtx(MutexTypeSyncVar, StatMtxSyncVar)
-  , addr(addr)
-  , uid(uid)
-  , creation_stack_id()
-  , owner_tid(kInvalidTid)
-  , last_lock()
-  , recursion()
-  , is_rw()
-  , is_recursive()
-  , is_broken()
-  , is_linker_init() {
+SyncVar::SyncVar()
+    : mtx(MutexTypeSyncVar, StatMtxSyncVar) {
+  Reset(0);
+}
+
+void SyncVar::Init(ThreadState *thr, uptr pc, uptr addr, u64 uid) {
+  this->addr = addr;
+  this->uid = uid;
+  this->next = 0;
+
+  creation_stack_id = 0;
+  if (kCppMode)  // Go does not use them
+    creation_stack_id = CurrentStackId(thr, pc);
+  if (common_flags()->detect_deadlocks)
+    DDMutexInit(thr, pc, this);
+}
+
+void SyncVar::Reset(ThreadState *thr) {
+  uid = 0;
+  creation_stack_id = 0;
+  owner_tid = kInvalidTid;
+  last_lock = 0;
+  recursion = 0;
+  is_rw = 0;
+  is_recursive = 0;
+  is_broken = 0;
+  is_linker_init = 0;
+
+  if (thr == 0) {
+    CHECK_EQ(clock.size(), 0);
+    CHECK_EQ(read_clock.size(), 0);
+  } else {
+    clock.Reset(&thr->clock_cache);
+    read_clock.Reset(&thr->clock_cache);
+  }
 }
 
-SyncTab::Part::Part()
-  : mtx(MutexTypeSyncTab, StatMtxSyncTab)
-  , val() {
+MetaMap::MetaMap() {
+  atomic_store(&uid_gen_, 0, memory_order_relaxed);
 }
 
-SyncTab::SyncTab() {
+void MetaMap::AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+  u32 idx = block_alloc_.Alloc(&thr->block_cache);
+  MBlock *b = block_alloc_.Map(idx);
+  b->siz = sz;
+  b->tid = thr->tid;
+  b->stk = CurrentStackId(thr, pc);
+  u32 *meta = MemToMeta(p);
+  DCHECK_EQ(*meta, 0);
+  *meta = idx | kFlagBlock;
 }
 
-SyncTab::~SyncTab() {
-  for (int i = 0; i < kPartCount; i++) {
-    while (tab_[i].val) {
-      SyncVar *tmp = tab_[i].val;
-      tab_[i].val = tmp->next;
-      DestroyAndFree(tmp);
+uptr MetaMap::FreeBlock(ThreadState *thr, uptr pc, uptr p) {
+  MBlock* b = GetBlock(p);
+  if (b == 0)
+    return 0;
+  uptr sz = RoundUpTo(b->siz, kMetaShadowCell);
+  FreeRange(thr, pc, p, sz);
+  return sz;
+}
+
+void MetaMap::FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz) {
+  u32 *meta = MemToMeta(p);
+  u32 *end = MemToMeta(p + sz);
+  if (end == meta)
+    end++;
+  for (; meta < end; meta++) {
+    u32 idx = *meta;
+    *meta = 0;
+    for (;;) {
+      if (idx == 0)
+        break;
+      if (idx & kFlagBlock) {
+        block_alloc_.Free(&thr->block_cache, idx & ~kFlagMask);
+        break;
+      } else if (idx & kFlagSync) {
+        DCHECK(idx & kFlagSync);
+        SyncVar *s = sync_alloc_.Map(idx & ~kFlagMask);
+        u32 next = s->next;
+        s->Reset(thr);
+        sync_alloc_.Free(&thr->sync_cache, idx & ~kFlagMask);
+        idx = next;
+      } else {
+        CHECK(0);
+      }
     }
   }
 }
 
-SyncVar* SyncTab::GetOrCreateAndLock(ThreadState *thr, uptr pc,
-                                     uptr addr, bool write_lock) {
-  return GetAndLock(thr, pc, addr, write_lock, true);
+MBlock* MetaMap::GetBlock(uptr p) {
+  u32 *meta = MemToMeta(p);
+  u32 idx = *meta;
+  for (;;) {
+    if (idx == 0)
+      return 0;
+    if (idx & kFlagBlock)
+      return block_alloc_.Map(idx & ~kFlagMask);
+    DCHECK(idx & kFlagSync);
+    SyncVar * s = sync_alloc_.Map(idx & ~kFlagMask);
+    idx = s->next;
+  }
 }
 
-SyncVar* SyncTab::GetIfExistsAndLock(uptr addr, bool write_lock) {
-  return GetAndLock(0, 0, addr, write_lock, false);
+SyncVar* MetaMap::GetOrCreateAndLock(ThreadState *thr, uptr pc,
+                              uptr addr, bool write_lock) {
+  return GetAndLock(thr, pc, addr, write_lock, true);
 }
 
-SyncVar* SyncTab::Create(ThreadState *thr, uptr pc, uptr addr) {
-  StatInc(thr, StatSyncCreated);
-  void *mem = internal_alloc(MBlockSync, sizeof(SyncVar));
-  const u64 uid = atomic_fetch_add(&uid_gen_, 1, memory_order_relaxed);
-  SyncVar *res = new(mem) SyncVar(addr, uid);
-  res->creation_stack_id = 0;
-  if (!kGoMode)  // Go does not use them
-    res->creation_stack_id = CurrentStackId(thr, pc);
-  if (flags()->detect_deadlocks)
-    DDMutexInit(thr, pc, res);
-  return res;
+SyncVar* MetaMap::GetIfExistsAndLock(uptr addr) {
+  return GetAndLock(0, 0, addr, true, false);
 }
 
-SyncVar* SyncTab::GetAndLock(ThreadState *thr, uptr pc,
+SyncVar* MetaMap::GetAndLock(ThreadState *thr, uptr pc,
                              uptr addr, bool write_lock, bool create) {
-#ifndef TSAN_GO
-  {  // NOLINT
-    SyncVar *res = GetJavaSync(thr, pc, addr, write_lock, create);
-    if (res)
-      return res;
-  }
-
-  // Here we ask only PrimaryAllocator, because
-  // SecondaryAllocator::PointerIsMine() is slow and we have fallback on
-  // the hashmap anyway.
-  if (PrimaryAllocator::PointerIsMine((void*)addr)) {
-    MBlock *b = user_mblock(thr, (void*)addr);
-    CHECK_NE(b, 0);
-    MBlock::ScopedLock l(b);
-    SyncVar *res = 0;
-    for (res = b->ListHead(); res; res = res->next) {
-      if (res->addr == addr)
+  u32 *meta = MemToMeta(addr);
+  u32 idx0 = *meta;
+  u32 myidx = 0;
+  SyncVar *mys = 0;
+  for (;;) {
+    u32 idx = idx0;
+    for (;;) {
+      if (idx == 0)
         break;
-    }
-    if (res == 0) {
-      if (!create)
-        return 0;
-      res = Create(thr, pc, addr);
-      b->ListPush(res);
-    }
-    if (write_lock)
-      res->mtx.Lock();
-    else
-      res->mtx.ReadLock();
-    return res;
-  }
-#endif
-
-  Part *p = &tab_[PartIdx(addr)];
-  {
-    ReadLock l(&p->mtx);
-    for (SyncVar *res = p->val; res; res = res->next) {
-      if (res->addr == addr) {
+      if (idx & kFlagBlock)
+        break;
+      DCHECK(idx & kFlagSync);
+      SyncVar * s = sync_alloc_.Map(idx & ~kFlagMask);
+      if (s->addr == addr) {
+        if (myidx != 0) {
+          mys->Reset(thr);
+          sync_alloc_.Free(&thr->sync_cache, myidx);
+        }
         if (write_lock)
-          res->mtx.Lock();
+          s->mtx.Lock();
         else
-          res->mtx.ReadLock();
-        return res;
+          s->mtx.ReadLock();
+        return s;
       }
+      idx = s->next;
     }
-  }
-  if (!create)
-    return 0;
-  {
-    Lock l(&p->mtx);
-    SyncVar *res = p->val;
-    for (; res; res = res->next) {
-      if (res->addr == addr)
-        break;
-    }
-    if (res == 0) {
-      res = Create(thr, pc, addr);
-      res->next = p->val;
-      p->val = res;
+    if (!create)
+      return 0;
+    if (*meta != idx0) {
+      idx0 = *meta;
+      continue;
     }
-    if (write_lock)
-      res->mtx.Lock();
-    else
-      res->mtx.ReadLock();
-    return res;
-  }
-}
 
-SyncVar* SyncTab::GetAndRemove(ThreadState *thr, uptr pc, uptr addr) {
-#ifndef TSAN_GO
-  {  // NOLINT
-    SyncVar *res = GetAndRemoveJavaSync(thr, pc, addr);
-    if (res)
-      return res;
-  }
-  if (PrimaryAllocator::PointerIsMine((void*)addr)) {
-    MBlock *b = user_mblock(thr, (void*)addr);
-    CHECK_NE(b, 0);
-    SyncVar *res = 0;
-    {
-      MBlock::ScopedLock l(b);
-      res = b->ListHead();
-      if (res) {
-        if (res->addr == addr) {
-          if (res->is_linker_init)
-            return 0;
-          b->ListPop();
-        } else {
-          SyncVar **prev = &res->next;
-          res = *prev;
-          while (res) {
-            if (res->addr == addr) {
-              if (res->is_linker_init)
-                return 0;
-              *prev = res->next;
-              break;
-            }
-            prev = &res->next;
-            res = *prev;
-          }
-        }
-        if (res) {
-          StatInc(thr, StatSyncDestroyed);
-          res->mtx.Lock();
-          res->mtx.Unlock();
-        }
-      }
+    if (myidx == 0) {
+      const u64 uid = atomic_fetch_add(&uid_gen_, 1, memory_order_relaxed);
+      myidx = sync_alloc_.Alloc(&thr->sync_cache);
+      mys = sync_alloc_.Map(myidx);
+      mys->Init(thr, pc, addr, uid);
     }
-    return res;
-  }
-#endif
-
-  Part *p = &tab_[PartIdx(addr)];
-  SyncVar *res = 0;
-  {
-    Lock l(&p->mtx);
-    SyncVar **prev = &p->val;
-    res = *prev;
-    while (res) {
-      if (res->addr == addr) {
-        if (res->is_linker_init)
-          return 0;
-        *prev = res->next;
-        break;
-      }
-      prev = &res->next;
-      res = *prev;
+    mys->next = idx0;
+    if (atomic_compare_exchange_strong((atomic_uint32_t*)meta, &idx0,
+        myidx | kFlagSync, memory_order_release)) {
+      if (write_lock)
+        mys->mtx.Lock();
+      else
+        mys->mtx.ReadLock();
+      return mys;
     }
   }
-  if (res) {
-    StatInc(thr, StatSyncDestroyed);
-    res->mtx.Lock();
-    res->mtx.Unlock();
-  }
-  return res;
-}
-
-int SyncTab::PartIdx(uptr addr) {
-  return (addr >> 3) % kPartCount;
-}
-
-StackTrace::StackTrace()
-    : n_()
-    , s_()
-    , c_() {
 }
 
-StackTrace::StackTrace(uptr *buf, uptr cnt)
-    : n_()
-    , s_(buf)
-    , c_(cnt) {
-  CHECK_NE(buf, 0);
-  CHECK_NE(cnt, 0);
-}
-
-StackTrace::~StackTrace() {
-  Reset();
-}
-
-void StackTrace::Reset() {
-  if (s_ && !c_) {
-    CHECK_NE(n_, 0);
-    internal_free(s_);
-    s_ = 0;
-  }
-  n_ = 0;
-}
-
-void StackTrace::Init(const uptr *pcs, uptr cnt) {
-  Reset();
-  if (cnt == 0)
-    return;
-  if (c_) {
-    CHECK_NE(s_, 0);
-    CHECK_LE(cnt, c_);
-  } else {
-    s_ = (uptr*)internal_alloc(MBlockStackTrace, cnt * sizeof(s_[0]));
+void MetaMap::MoveMemory(uptr src, uptr dst, uptr sz) {
+  // src and dst can overlap,
+  // there are no concurrent accesses to the regions (e.g. stop-the-world).
+  CHECK_NE(src, dst);
+  CHECK_NE(sz, 0);
+  uptr diff = dst - src;
+  u32 *src_meta = MemToMeta(src);
+  u32 *dst_meta = MemToMeta(dst);
+  u32 *src_meta_end = MemToMeta(src + sz);
+  uptr inc = 1;
+  if (dst > src) {
+    src_meta = MemToMeta(src + sz) - 1;
+    dst_meta = MemToMeta(dst + sz) - 1;
+    src_meta_end = MemToMeta(src) - 1;
+    inc = -1;
   }
-  n_ = cnt;
-  internal_memcpy(s_, pcs, cnt * sizeof(s_[0]));
-}
-
-void StackTrace::ObtainCurrent(ThreadState *thr, uptr toppc) {
-  Reset();
-  n_ = thr->shadow_stack_pos - thr->shadow_stack;
-  if (n_ + !!toppc == 0)
-    return;
-  uptr start = 0;
-  if (c_) {
-    CHECK_NE(s_, 0);
-    if (n_ + !!toppc > c_) {
-      start = n_ - c_ + !!toppc;
-      n_ = c_ - !!toppc;
-    }
-  } else {
-    // Cap potentially huge stacks.
-    if (n_ + !!toppc > kTraceStackSize) {
-      start = n_ - kTraceStackSize + !!toppc;
-      n_ = kTraceStackSize - !!toppc;
+  for (; src_meta != src_meta_end; src_meta += inc, dst_meta += inc) {
+    CHECK_EQ(*dst_meta, 0);
+    u32 idx = *src_meta;
+    *src_meta = 0;
+    *dst_meta = idx;
+    // Patch the addresses in sync objects.
+    while (idx != 0) {
+      if (idx & kFlagBlock)
+        break;
+      CHECK(idx & kFlagSync);
+      SyncVar *s = sync_alloc_.Map(idx & ~kFlagMask);
+      s->addr += diff;
+      idx = s->next;
     }
-    s_ = (uptr*)internal_alloc(MBlockStackTrace,
-                               (n_ + !!toppc) * sizeof(s_[0]));
-  }
-  for (uptr i = 0; i < n_; i++)
-    s_[i] = thr->shadow_stack[start + i];
-  if (toppc) {
-    s_[n_] = toppc;
-    n_++;
   }
 }
 
-void StackTrace::CopyFrom(const StackTrace& other) {
-  Reset();
-  Init(other.Begin(), other.Size());
-}
-
-bool StackTrace::IsEmpty() const {
-  return n_ == 0;
-}
-
-uptr StackTrace::Size() const {
-  return n_;
-}
-
-uptr StackTrace::Get(uptr i) const {
-  CHECK_LT(i, n_);
-  return s_[i];
-}
-
-const uptr *StackTrace::Begin() const {
-  return s_;
+void MetaMap::OnThreadIdle(ThreadState *thr) {
+  block_alloc_.FlushCache(&thr->block_cache);
+  sync_alloc_.FlushCache(&thr->sync_cache);
 }
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_sync.h b/libsanitizer/tsan/tsan_sync.h
index 3838df91d75..6ed3715ee0a 100644
--- a/libsanitizer/tsan/tsan_sync.h
+++ b/libsanitizer/tsan/tsan_sync.h
@@ -14,46 +14,21 @@
 #include "sanitizer_common/sanitizer_atomic.h"
 #include "sanitizer_common/sanitizer_common.h"
 #include "sanitizer_common/sanitizer_deadlock_detector_interface.h"
-#include "tsan_clock.h"
 #include "tsan_defs.h"
+#include "tsan_clock.h"
 #include "tsan_mutex.h"
+#include "tsan_dense_alloc.h"
 
 namespace __tsan {
 
-class StackTrace {
- public:
-  StackTrace();
-  // Initialized the object in "static mode",
-  // in this mode it never calls malloc/free but uses the provided buffer.
-  StackTrace(uptr *buf, uptr cnt);
-  ~StackTrace();
-  void Reset();
-
-  void Init(const uptr *pcs, uptr cnt);
-  void ObtainCurrent(ThreadState *thr, uptr toppc);
-  bool IsEmpty() const;
-  uptr Size() const;
-  uptr Get(uptr i) const;
-  const uptr *Begin() const;
-  void CopyFrom(const StackTrace& other);
-
- private:
-  uptr n_;
-  uptr *s_;
-  const uptr c_;
-
-  StackTrace(const StackTrace&);
-  void operator = (const StackTrace&);
-};
-
 struct SyncVar {
-  explicit SyncVar(uptr addr, u64 uid);
+  SyncVar();
 
   static const int kInvalidTid = -1;
 
+  uptr addr;  // overwritten by DenseSlabAlloc freelist
   Mutex mtx;
-  uptr addr;
-  const u64 uid;  // Globally unique id.
+  u64 uid;  // Globally unique id.
   u32 creation_stack_id;
   int owner_tid;  // Set only by exclusive owners.
   u64 last_lock;
@@ -62,13 +37,16 @@ struct SyncVar {
   bool is_recursive;
   bool is_broken;
   bool is_linker_init;
-  SyncVar *next;  // In SyncTab hashtable.
+  u32 next;  // in MetaMap
   DDMutex dd;
   SyncClock read_clock;  // Used for rw mutexes only.
   // The clock is placed last, so that it is situated on a different cache line
   // with the mtx. This reduces contention for hot sync objects.
   SyncClock clock;
 
+  void Init(ThreadState *thr, uptr pc, uptr addr, u64 uid);
+  void Reset(ThreadState *thr);
+
   u64 GetId() const {
     // 47 lsb is addr, then 14 bits is low part of uid, then 3 zero bits.
     return GetLsb((u64)addr | (uid << 47), 61);
@@ -83,40 +61,39 @@ struct SyncVar {
   }
 };
 
-class SyncTab {
+/* MetaMap allows to map arbitrary user pointers onto various descriptors.
+   Currently it maps pointers to heap block descriptors and sync var descs.
+   It uses 1/2 direct shadow, see tsan_platform.h.
+*/
+class MetaMap {
  public:
-  SyncTab();
-  ~SyncTab();
+  MetaMap();
+
+  void AllocBlock(ThreadState *thr, uptr pc, uptr p, uptr sz);
+  uptr FreeBlock(ThreadState *thr, uptr pc, uptr p);
+  void FreeRange(ThreadState *thr, uptr pc, uptr p, uptr sz);
+  MBlock* GetBlock(uptr p);
 
   SyncVar* GetOrCreateAndLock(ThreadState *thr, uptr pc,
                               uptr addr, bool write_lock);
-  SyncVar* GetIfExistsAndLock(uptr addr, bool write_lock);
+  SyncVar* GetIfExistsAndLock(uptr addr);
 
-  // If the SyncVar does not exist, returns 0.
-  SyncVar* GetAndRemove(ThreadState *thr, uptr pc, uptr addr);
+  void MoveMemory(uptr src, uptr dst, uptr sz);
 
-  SyncVar* Create(ThreadState *thr, uptr pc, uptr addr);
+  void OnThreadIdle(ThreadState *thr);
 
  private:
-  struct Part {
-    Mutex mtx;
-    SyncVar *val;
-    char pad[kCacheLineSize - sizeof(Mutex) - sizeof(SyncVar*)];  // NOLINT
-    Part();
-  };
-
-  // FIXME: Implement something more sane.
-  static const int kPartCount = 1009;
-  Part tab_[kPartCount];
+  static const u32 kFlagMask  = 3 << 30;
+  static const u32 kFlagBlock = 1 << 30;
+  static const u32 kFlagSync  = 2 << 30;
+  typedef DenseSlabAlloc<MBlock, 1<<16, 1<<12> BlockAlloc;
+  typedef DenseSlabAlloc<SyncVar, 1<<16, 1<<10> SyncAlloc;
+  BlockAlloc block_alloc_;
+  SyncAlloc sync_alloc_;
   atomic_uint64_t uid_gen_;
 
-  int PartIdx(uptr addr);
-
-  SyncVar* GetAndLock(ThreadState *thr, uptr pc,
-                      uptr addr, bool write_lock, bool create);
-
-  SyncTab(const SyncTab&);  // Not implemented.
-  void operator = (const SyncTab&);  // Not implemented.
+  SyncVar* GetAndLock(ThreadState *thr, uptr pc, uptr addr, bool write_lock,
+                      bool create);
 };
 
 }  // namespace __tsan
diff --git a/libsanitizer/tsan/tsan_trace.h b/libsanitizer/tsan/tsan_trace.h
index 93ed8d907e5..af140686f5d 100644
--- a/libsanitizer/tsan/tsan_trace.h
+++ b/libsanitizer/tsan/tsan_trace.h
@@ -13,7 +13,7 @@
 
 #include "tsan_defs.h"
 #include "tsan_mutex.h"
-#include "tsan_sync.h"
+#include "tsan_stack_trace.h"
 #include "tsan_mutexset.h"
 
 namespace __tsan {
diff --git a/libsanitizer/tsan/tsan_update_shadow_word_inl.h b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
index 42caf80d349..91def7bb1e8 100644
--- a/libsanitizer/tsan/tsan_update_shadow_word_inl.h
+++ b/libsanitizer/tsan/tsan_update_shadow_word_inl.h
@@ -14,8 +14,7 @@
 do {
   StatInc(thr, StatShadowProcessed);
   const unsigned kAccessSize = 1 << kAccessSizeLog;
-  unsigned off = cur.ComputeSearchOffset();
-  u64 *sp = &shadow_mem[(idx + off) % kShadowCnt];
+  u64 *sp = &shadow_mem[idx];
   old = LoadShadow(sp);
   if (old.IsZero()) {
     StatInc(thr, StatShadowZero);
@@ -31,16 +30,6 @@ do {
     // same thread?
     if (Shadow::TidsAreEqual(old, cur)) {
       StatInc(thr, StatShadowSameThread);
-      if (OldIsInSameSynchEpoch(old, thr)) {
-        if (old.IsRWNotWeaker(kAccessIsWrite, kIsAtomic)) {
-          // found a slot that holds effectively the same info
-          // (that is, same tid, same sync epoch and same size)
-          StatInc(thr, StatMopSame);
-          return;
-        }
-        StoreIfNotYetStored(sp, &store_word);
-        break;
-      }
       if (old.IsRWWeakerOrEqual(kAccessIsWrite, kIsAtomic))
         StoreIfNotYetStored(sp, &store_word);
       break;
diff --git a/libsanitizer/tsan/tsan_vector.h b/libsanitizer/tsan/tsan_vector.h
index f65ad2b5560..c0485513ee2 100644
--- a/libsanitizer/tsan/tsan_vector.h
+++ b/libsanitizer/tsan/tsan_vector.h
@@ -76,6 +76,10 @@ class Vector {
   }
 
   void Resize(uptr size) {
+    if (size == 0) {
+      end_ = begin_;
+      return;
+    }
     uptr old_size = Size();
     EnsureSize(size);
     if (old_size < size) {
@@ -98,7 +102,7 @@ class Vector {
       return;
     }
     uptr cap0 = last_ - begin_;
-    uptr cap = 2 * cap0;
+    uptr cap = cap0 * 5 / 4;  // 25% growth
     if (cap == 0)
       cap = 16;
     if (cap < size)
diff --git a/libsanitizer/ubsan/Makefile.am b/libsanitizer/ubsan/Makefile.am
index 80b79e1cb82..d010b91681d 100644
--- a/libsanitizer/ubsan/Makefile.am
+++ b/libsanitizer/ubsan/Makefile.am
@@ -12,8 +12,10 @@ toolexeclib_LTLIBRARIES = libubsan.la
 
 ubsan_files = \
 	ubsan_diag.cc \
+	ubsan_flags.cc \
 	ubsan_handlers.cc \
 	ubsan_handlers_cxx.cc \
+	ubsan_init.cc \
 	ubsan_type_hash.cc \
 	ubsan_value.cc
 
diff --git a/libsanitizer/ubsan/Makefile.in b/libsanitizer/ubsan/Makefile.in
index c38740be666..102eff6138e 100644
--- a/libsanitizer/ubsan/Makefile.in
+++ b/libsanitizer/ubsan/Makefile.in
@@ -84,8 +84,9 @@ am__DEPENDENCIES_1 =
 libubsan_la_DEPENDENCIES =  \
 	$(top_builddir)/sanitizer_common/libsanitizer_common.la \
 	$(am__append_1) $(am__append_2) $(am__DEPENDENCIES_1)
-am__objects_1 = ubsan_diag.lo ubsan_handlers.lo ubsan_handlers_cxx.lo \
-	ubsan_type_hash.lo ubsan_value.lo
+am__objects_1 = ubsan_diag.lo ubsan_flags.lo ubsan_handlers.lo \
+	ubsan_handlers_cxx.lo ubsan_init.lo ubsan_type_hash.lo \
+	ubsan_value.lo
 am_libubsan_la_OBJECTS = $(am__objects_1)
 libubsan_la_OBJECTS = $(am_libubsan_la_OBJECTS)
 libubsan_la_LINK = $(LIBTOOL) --tag=CXX $(AM_LIBTOOLFLAGS) \
@@ -260,8 +261,10 @@ ACLOCAL_AMFLAGS = -I m4
 toolexeclib_LTLIBRARIES = libubsan.la
 ubsan_files = \
 	ubsan_diag.cc \
+	ubsan_flags.cc \
 	ubsan_handlers.cc \
 	ubsan_handlers_cxx.cc \
+	ubsan_init.cc \
 	ubsan_type_hash.cc \
 	ubsan_value.cc
 
@@ -385,8 +388,10 @@ distclean-compile:
 	-rm -f *.tab.c
 
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_diag.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_flags.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_handlers.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_handlers_cxx.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_init.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_type_hash.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/ubsan_value.Plo@am__quote@
 
diff --git a/libsanitizer/ubsan/ubsan_diag.cc b/libsanitizer/ubsan/ubsan_diag.cc
index 1dfe7255f68..828127ab84d 100644
--- a/libsanitizer/ubsan/ubsan_diag.cc
+++ b/libsanitizer/ubsan/ubsan_diag.cc
@@ -10,9 +10,8 @@
 //===----------------------------------------------------------------------===//
 
 #include "ubsan_diag.h"
-#include "sanitizer_common/sanitizer_common.h"
-#include "sanitizer_common/sanitizer_flags.h"
-#include "sanitizer_common/sanitizer_libc.h"
+#include "ubsan_init.h"
+#include "ubsan_flags.h"
 #include "sanitizer_common/sanitizer_report_decorator.h"
 #include "sanitizer_common/sanitizer_stacktrace.h"
 #include "sanitizer_common/sanitizer_symbolizer.h"
@@ -20,20 +19,48 @@
 
 using namespace __ubsan;
 
-static void InitializeSanitizerCommon() {
-  static StaticSpinMutex init_mu;
-  SpinMutexLock l(&init_mu);
-  static bool initialized;
-  if (initialized)
-   return;
-  if (0 == internal_strcmp(SanitizerToolName, "SanitizerTool")) {
-    // UBSan is run in a standalone mode. Initialize it now.
-    SanitizerToolName = "UndefinedBehaviorSanitizer";
-    CommonFlags *cf = common_flags();
-    SetCommonFlagsDefaults(cf);
-    cf->print_summary = false;
+static void MaybePrintStackTrace(uptr pc, uptr bp) {
+  // We assume that flags are already parsed: InitIfNecessary
+  // will definitely be called when we print the first diagnostics message.
+  if (!flags()->print_stacktrace)
+    return;
+  // We can only use slow unwind, as we don't have any information about stack
+  // top/bottom.
+  // FIXME: It's better to respect "fast_unwind_on_fatal" runtime flag and
+  // fetch stack top/bottom information if we have it (e.g. if we're running
+  // under ASan).
+  if (StackTrace::WillUseFastUnwind(false))
+    return;
+  StackTrace stack;
+  stack.Unwind(kStackTraceMax, pc, bp, 0, 0, 0, false);
+  stack.Print();
+}
+
+static void MaybeReportErrorSummary(Location Loc) {
+  if (!common_flags()->print_summary)
+    return;
+  // Don't try to unwind the stack trace in UBSan summaries: just use the
+  // provided location.
+  if (Loc.isSourceLocation()) {
+    SourceLocation SLoc = Loc.getSourceLocation();
+    if (!SLoc.isInvalid()) {
+      ReportErrorSummary("runtime-error", SLoc.getFilename(), SLoc.getLine(),
+                         "");
+      return;
+    }
   }
-  initialized = true;
+  ReportErrorSummary("runtime-error");
+}
+
+namespace {
+class Decorator : public SanitizerCommonDecorator {
+ public:
+  Decorator() : SanitizerCommonDecorator() {}
+  const char *Highlight() const { return Green(); }
+  const char *EndHighlight() const { return Default(); }
+  const char *Note() const { return Black(); }
+  const char *EndNote() const { return Default(); }
+};
 }
 
 Location __ubsan::getCallerLocation(uptr CallerLoc) {
@@ -47,12 +74,11 @@ Location __ubsan::getCallerLocation(uptr CallerLoc) {
 Location __ubsan::getFunctionLocation(uptr Loc, const char **FName) {
   if (!Loc)
     return Location();
-  // FIXME: We may need to run initialization earlier.
-  InitializeSanitizerCommon();
+  InitIfNecessary();
 
   AddressInfo Info;
-  if (!Symbolizer::GetOrInit()->SymbolizePC(Loc, &Info, 1) ||
-      !Info.module || !*Info.module)
+  if (!Symbolizer::GetOrInit()->SymbolizePC(Loc, &Info, 1) || !Info.module ||
+      !*Info.module)
     return Location(Loc);
 
   if (FName && Info.function)
@@ -181,36 +207,49 @@ static Range *upperBound(MemoryLocation Loc, Range *Ranges,
   return Best;
 }
 
+static inline uptr subtractNoOverflow(uptr LHS, uptr RHS) {
+  return (LHS < RHS) ? 0 : LHS - RHS;
+}
+
+static inline uptr addNoOverflow(uptr LHS, uptr RHS) {
+  const uptr Limit = (uptr)-1;
+  return (LHS > Limit - RHS) ? Limit : LHS + RHS;
+}
+
 /// Render a snippet of the address space near a location.
-static void renderMemorySnippet(const __sanitizer::AnsiColorDecorator &Decor,
-                                MemoryLocation Loc,
+static void renderMemorySnippet(const Decorator &Decor, MemoryLocation Loc,
                                 Range *Ranges, unsigned NumRanges,
                                 const Diag::Arg *Args) {
-  const unsigned BytesToShow = 32;
-  const unsigned MinBytesNearLoc = 4;
-
   // Show at least the 8 bytes surrounding Loc.
-  MemoryLocation Min = Loc - MinBytesNearLoc, Max = Loc + MinBytesNearLoc;
+  const unsigned MinBytesNearLoc = 4;
+  MemoryLocation Min = subtractNoOverflow(Loc, MinBytesNearLoc);
+  MemoryLocation Max = addNoOverflow(Loc, MinBytesNearLoc);
+  MemoryLocation OrigMin = Min;
   for (unsigned I = 0; I < NumRanges; ++I) {
     Min = __sanitizer::Min(Ranges[I].getStart().getMemoryLocation(), Min);
     Max = __sanitizer::Max(Ranges[I].getEnd().getMemoryLocation(), Max);
   }
 
   // If we have too many interesting bytes, prefer to show bytes after Loc.
+  const unsigned BytesToShow = 32;
   if (Max - Min > BytesToShow)
-    Min = __sanitizer::Min(Max - BytesToShow, Loc - MinBytesNearLoc);
-  Max = Min + BytesToShow;
+    Min = __sanitizer::Min(Max - BytesToShow, OrigMin);
+  Max = addNoOverflow(Min, BytesToShow);
+
+  if (!IsAccessibleMemoryRange(Min, Max - Min)) {
+    Printf("<memory cannot be printed>\n");
+    return;
+  }
 
   // Emit data.
   for (uptr P = Min; P != Max; ++P) {
-    // FIXME: Check that the address is readable before printing it.
     unsigned char C = *reinterpret_cast<const unsigned char*>(P);
     Printf("%s%02x", (P % 8 == 0) ? "  " : " ", C);
   }
   Printf("\n");
 
   // Emit highlights.
-  Printf(Decor.Green());
+  Printf(Decor.Highlight());
   Range *InRange = upperBound(Min, Ranges, NumRanges);
   for (uptr P = Min; P != Max; ++P) {
     char Pad = ' ', Byte = ' ';
@@ -225,7 +264,7 @@ static void renderMemorySnippet(const __sanitizer::AnsiColorDecorator &Decor,
     char Buffer[] = { Pad, Pad, P == Loc ? '^' : Byte, Byte, 0 };
     Printf((P % 8 == 0) ? Buffer : &Buffer[1]);
   }
-  Printf("%s\n", Decor.Default());
+  Printf("%s\n", Decor.EndHighlight());
 
   // Go over the line again, and print names for the ranges.
   InRange = 0;
@@ -263,8 +302,9 @@ static void renderMemorySnippet(const __sanitizer::AnsiColorDecorator &Decor,
 }
 
 Diag::~Diag() {
-  __sanitizer::AnsiColorDecorator Decor(PrintsToTty());
-  SpinMutexLock l(&CommonSanitizerReportMutex);
+  // All diagnostics should be printed under report mutex.
+  CommonSanitizerReportMutex.CheckLocked();
+  Decorator Decor;
   Printf(Decor.Bold());
 
   renderLocation(Loc);
@@ -272,11 +312,11 @@ Diag::~Diag() {
   switch (Level) {
   case DL_Error:
     Printf("%s runtime error: %s%s",
-           Decor.Red(), Decor.Default(), Decor.Bold());
+           Decor.Warning(), Decor.EndWarning(), Decor.Bold());
     break;
 
   case DL_Note:
-    Printf("%s note: %s", Decor.Black(), Decor.Default());
+    Printf("%s note: %s", Decor.Note(), Decor.EndNote());
     break;
   }
 
@@ -288,3 +328,26 @@ Diag::~Diag() {
     renderMemorySnippet(Decor, Loc.getMemoryLocation(), Ranges,
                         NumRanges, Args);
 }
+
+ScopedReport::ScopedReport(ReportOptions Opts, Location SummaryLoc)
+    : Opts(Opts), SummaryLoc(SummaryLoc) {
+  InitIfNecessary();
+  CommonSanitizerReportMutex.Lock();
+}
+
+ScopedReport::~ScopedReport() {
+  MaybePrintStackTrace(Opts.pc, Opts.bp);
+  MaybeReportErrorSummary(SummaryLoc);
+  CommonSanitizerReportMutex.Unlock();
+  if (Opts.DieAfterReport || flags()->halt_on_error)
+    Die();
+}
+
+bool __ubsan::MatchSuppression(const char *Str, SuppressionType Type) {
+  Suppression *s;
+  // If .preinit_array is not used, it is possible that the UBSan runtime is not
+  // initialized.
+  if (!SANITIZER_CAN_USE_PREINIT_ARRAY)
+    InitIfNecessary();
+  return SuppressionContext::Get()->Match(Str, Type, &s);
+}
diff --git a/libsanitizer/ubsan/ubsan_diag.h b/libsanitizer/ubsan/ubsan_diag.h
index 04503687568..fbb3f1ff9d4 100644
--- a/libsanitizer/ubsan/ubsan_diag.h
+++ b/libsanitizer/ubsan/ubsan_diag.h
@@ -12,6 +12,8 @@
 #define UBSAN_DIAG_H
 
 #include "ubsan_value.h"
+#include "sanitizer_common/sanitizer_stacktrace.h"
+#include "sanitizer_common/sanitizer_suppressions.h"
 
 namespace __ubsan {
 
@@ -201,6 +203,33 @@ public:
   Diag &operator<<(const Range &R) { return AddRange(R); }
 };
 
+struct ReportOptions {
+  /// If DieAfterReport is specified, UBSan will terminate the program after the
+  /// report is printed.
+  bool DieAfterReport;
+  /// pc/bp are used to unwind the stack trace.
+  uptr pc;
+  uptr bp;
+};
+
+#define GET_REPORT_OPTIONS(die_after_report) \
+    GET_CALLER_PC_BP; \
+    ReportOptions Opts = {die_after_report, pc, bp}
+
+/// \brief Instantiate this class before printing diagnostics in the error
+/// report. This class ensures that reports from different threads and from
+/// different sanitizers won't be mixed.
+class ScopedReport {
+  ReportOptions Opts;
+  Location SummaryLoc;
+
+public:
+  ScopedReport(ReportOptions Opts, Location SummaryLoc);
+  ~ScopedReport();
+};
+
+bool MatchSuppression(const char *Str, SuppressionType Type);
+
 } // namespace __ubsan
 
 #endif // UBSAN_DIAG_H
diff --git a/libsanitizer/ubsan/ubsan_flags.cc b/libsanitizer/ubsan/ubsan_flags.cc
new file mode 100644
index 00000000000..9645a857260
--- /dev/null
+++ b/libsanitizer/ubsan/ubsan_flags.cc
@@ -0,0 +1,61 @@
+//===-- ubsan_flags.cc ----------------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Runtime flags for UndefinedBehaviorSanitizer.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ubsan_flags.h"
+#include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_flags.h"
+
+namespace __ubsan {
+
+static const char *MaybeCallUbsanDefaultOptions() {
+  return (&__ubsan_default_options) ? __ubsan_default_options() : "";
+}
+
+void InitializeCommonFlags() {
+  CommonFlags *cf = common_flags();
+  SetCommonFlagsDefaults(cf);
+  cf->print_summary = false;
+  // Override from user-specified string.
+  ParseCommonFlagsFromString(cf, MaybeCallUbsanDefaultOptions());
+  // Override from environment variable.
+  ParseCommonFlagsFromString(cf, GetEnv("UBSAN_OPTIONS"));
+}
+
+Flags ubsan_flags;
+
+static void ParseFlagsFromString(Flags *f, const char *str) {
+  if (!str)
+    return;
+  ParseFlag(str, &f->halt_on_error, "halt_on_error",
+            "Crash the program after printing the first error report");
+  ParseFlag(str, &f->print_stacktrace, "print_stacktrace",
+            "Include full stacktrace into an error report");
+}
+
+void InitializeFlags() {
+  Flags *f = flags();
+  // Default values.
+  f->halt_on_error = false;
+  f->print_stacktrace = false;
+  // Override from user-specified string.
+  ParseFlagsFromString(f, MaybeCallUbsanDefaultOptions());
+  // Override from environment variable.
+  ParseFlagsFromString(f, GetEnv("UBSAN_OPTIONS"));
+}
+
+}  // namespace __ubsan
+
+#if !SANITIZER_SUPPORTS_WEAK_HOOKS
+extern "C" {
+SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
+const char *__ubsan_default_options() { return ""; }
+}  // extern "C"
+#endif
diff --git a/libsanitizer/ubsan/ubsan_flags.h b/libsanitizer/ubsan/ubsan_flags.h
new file mode 100644
index 00000000000..00be9b004f4
--- /dev/null
+++ b/libsanitizer/ubsan/ubsan_flags.h
@@ -0,0 +1,38 @@
+//===-- ubsan_flags.h -------------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Runtime flags for UndefinedBehaviorSanitizer.
+//
+//===----------------------------------------------------------------------===//
+#ifndef UBSAN_FLAGS_H
+#define UBSAN_FLAGS_H
+
+#include "sanitizer_common/sanitizer_internal_defs.h"
+
+namespace __ubsan {
+
+struct Flags {
+  bool halt_on_error;
+  bool print_stacktrace;
+};
+
+extern Flags ubsan_flags;
+inline Flags *flags() { return &ubsan_flags; }
+
+void InitializeCommonFlags();
+void InitializeFlags();
+
+}  // namespace __ubsan
+
+extern "C" {
+// Users may provide their own implementation of __ubsan_default_options to
+// override the default flag values.
+SANITIZER_INTERFACE_ATTRIBUTE SANITIZER_WEAK_ATTRIBUTE
+const char *__ubsan_default_options();
+}  // extern "C"
+
+#endif  // UBSAN_FLAGS_H
diff --git a/libsanitizer/ubsan/ubsan_handlers.cc b/libsanitizer/ubsan/ubsan_handlers.cc
index 42f948ae2e6..bb9322f068f 100644
--- a/libsanitizer/ubsan/ubsan_handlers.cc
+++ b/libsanitizer/ubsan/ubsan_handlers.cc
@@ -17,6 +17,16 @@
 using namespace __sanitizer;
 using namespace __ubsan;
 
+static bool ignoreReport(SourceLocation SLoc, ReportOptions Opts) {
+  // If source location is already acquired, we don't need to print an error
+  // report for the second time. However, if we're in an unrecoverable handler,
+  // it's possible that location was required by concurrently running thread.
+  // In this case, we should continue the execution to ensure that any of
+  // threads will grab the report mutex and print the report before
+  // crashing the program.
+  return SLoc.isDisabled() && !Opts.DieAfterReport;
+}
+
 namespace __ubsan {
   const char *TypeCheckKinds[] = {
     "load of", "store to", "reference binding to", "member access within",
@@ -25,15 +35,17 @@ namespace __ubsan {
 }
 
 static void handleTypeMismatchImpl(TypeMismatchData *Data, ValueHandle Pointer,
-                                   Location FallbackLoc) {
+                                   Location FallbackLoc, ReportOptions Opts) {
   Location Loc = Data->Loc.acquire();
-
   // Use the SourceLocation from Data to track deduplication, even if 'invalid'
-  if (Loc.getSourceLocation().isDisabled())
+  if (ignoreReport(Loc.getSourceLocation(), Opts))
     return;
+
   if (Data->Loc.isInvalid())
     Loc = FallbackLoc;
 
+  ScopedReport R(Opts, Loc);
+
   if (!Pointer)
     Diag(Loc, DL_Error, "%0 null pointer of type %1")
       << TypeCheckKinds[Data->TypeCheckKind] << Data->Type;
@@ -49,70 +61,59 @@ static void handleTypeMismatchImpl(TypeMismatchData *Data, ValueHandle Pointer,
   if (Pointer)
     Diag(Pointer, DL_Note, "pointer points here");
 }
+
 void __ubsan::__ubsan_handle_type_mismatch(TypeMismatchData *Data,
                                            ValueHandle Pointer) {
-  handleTypeMismatchImpl(Data, Pointer, getCallerLocation());
+  GET_REPORT_OPTIONS(false);
+  handleTypeMismatchImpl(Data, Pointer, getCallerLocation(), Opts);
 }
 void __ubsan::__ubsan_handle_type_mismatch_abort(TypeMismatchData *Data,
                                                  ValueHandle Pointer) {
-  handleTypeMismatchImpl(Data, Pointer, getCallerLocation());
+  GET_REPORT_OPTIONS(true);
+  handleTypeMismatchImpl(Data, Pointer, getCallerLocation(), Opts);
   Die();
 }
 
 /// \brief Common diagnostic emission for various forms of integer overflow.
-template<typename T> static void HandleIntegerOverflow(OverflowData *Data,
-                                                       ValueHandle LHS,
-                                                       const char *Operator,
-                                                       T RHS) {
+template <typename T>
+static void handleIntegerOverflowImpl(OverflowData *Data, ValueHandle LHS,
+                                      const char *Operator, T RHS,
+                                      ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error, "%0 integer overflow: "
                       "%1 %2 %3 cannot be represented in type %4")
     << (Data->Type.isSignedIntegerTy() ? "signed" : "unsigned")
     << Value(Data->Type, LHS) << Operator << RHS << Data->Type;
 }
 
-void __ubsan::__ubsan_handle_add_overflow(OverflowData *Data,
-                                          ValueHandle LHS, ValueHandle RHS) {
-  HandleIntegerOverflow(Data, LHS, "+", Value(Data->Type, RHS));
-}
-void __ubsan::__ubsan_handle_add_overflow_abort(OverflowData *Data,
-                                                 ValueHandle LHS,
-                                                 ValueHandle RHS) {
-  __ubsan_handle_add_overflow(Data, LHS, RHS);
-  Die();
-}
+#define UBSAN_OVERFLOW_HANDLER(handler_name, op, abort)                        \
+  void __ubsan::handler_name(OverflowData *Data, ValueHandle LHS,              \
+                             ValueHandle RHS) {                                \
+    GET_REPORT_OPTIONS(abort);                                                 \
+    handleIntegerOverflowImpl(Data, LHS, op, Value(Data->Type, RHS), Opts);    \
+    if (abort) Die();                                                          \
+  }
 
-void __ubsan::__ubsan_handle_sub_overflow(OverflowData *Data,
-                                          ValueHandle LHS, ValueHandle RHS) {
-  HandleIntegerOverflow(Data, LHS, "-", Value(Data->Type, RHS));
-}
-void __ubsan::__ubsan_handle_sub_overflow_abort(OverflowData *Data,
-                                                 ValueHandle LHS,
-                                                 ValueHandle RHS) {
-  __ubsan_handle_sub_overflow(Data, LHS, RHS);
-  Die();
-}
-
-void __ubsan::__ubsan_handle_mul_overflow(OverflowData *Data,
-                                          ValueHandle LHS, ValueHandle RHS) {
-  HandleIntegerOverflow(Data, LHS, "*", Value(Data->Type, RHS));
-}
-void __ubsan::__ubsan_handle_mul_overflow_abort(OverflowData *Data,
-                                                 ValueHandle LHS,
-                                                 ValueHandle RHS) {
-  __ubsan_handle_mul_overflow(Data, LHS, RHS);
-  Die();
-}
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_add_overflow, "+", false)
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_add_overflow_abort, "+", true)
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_sub_overflow, "-", false)
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_sub_overflow_abort, "-", true)
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_mul_overflow, "*", false)
+UBSAN_OVERFLOW_HANDLER(__ubsan_handle_mul_overflow_abort, "*", true)
 
-void __ubsan::__ubsan_handle_negate_overflow(OverflowData *Data,
-                                             ValueHandle OldVal) {
+static void handleNegateOverflowImpl(OverflowData *Data, ValueHandle OldVal,
+                                     ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   if (Data->Type.isSignedIntegerTy())
     Diag(Loc, DL_Error,
          "negation of %0 cannot be represented in type %1; "
@@ -123,18 +124,27 @@ void __ubsan::__ubsan_handle_negate_overflow(OverflowData *Data,
          "negation of %0 cannot be represented in type %1")
       << Value(Data->Type, OldVal) << Data->Type;
 }
+
+void __ubsan::__ubsan_handle_negate_overflow(OverflowData *Data,
+                                             ValueHandle OldVal) {
+  GET_REPORT_OPTIONS(false);
+  handleNegateOverflowImpl(Data, OldVal, Opts);
+}
 void __ubsan::__ubsan_handle_negate_overflow_abort(OverflowData *Data,
                                                     ValueHandle OldVal) {
-  __ubsan_handle_negate_overflow(Data, OldVal);
+  GET_REPORT_OPTIONS(true);
+  handleNegateOverflowImpl(Data, OldVal, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_divrem_overflow(OverflowData *Data,
-                                             ValueHandle LHS, ValueHandle RHS) {
+static void handleDivremOverflowImpl(OverflowData *Data, ValueHandle LHS,
+                                     ValueHandle RHS, ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Value LHSVal(Data->Type, LHS);
   Value RHSVal(Data->Type, RHS);
   if (RHSVal.isMinusOne())
@@ -144,20 +154,29 @@ void __ubsan::__ubsan_handle_divrem_overflow(OverflowData *Data,
   else
     Diag(Loc, DL_Error, "division by zero");
 }
+
+void __ubsan::__ubsan_handle_divrem_overflow(OverflowData *Data,
+                                             ValueHandle LHS, ValueHandle RHS) {
+  GET_REPORT_OPTIONS(false);
+  handleDivremOverflowImpl(Data, LHS, RHS, Opts);
+}
 void __ubsan::__ubsan_handle_divrem_overflow_abort(OverflowData *Data,
                                                     ValueHandle LHS,
                                                     ValueHandle RHS) {
-  __ubsan_handle_divrem_overflow(Data, LHS, RHS);
+  GET_REPORT_OPTIONS(true);
+  handleDivremOverflowImpl(Data, LHS, RHS, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_shift_out_of_bounds(ShiftOutOfBoundsData *Data,
-                                                 ValueHandle LHS,
-                                                 ValueHandle RHS) {
+static void handleShiftOutOfBoundsImpl(ShiftOutOfBoundsData *Data,
+                                       ValueHandle LHS, ValueHandle RHS,
+                                       ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Value LHSVal(Data->LHSType, LHS);
   Value RHSVal(Data->RHSType, RHS);
   if (RHSVal.isNegative())
@@ -173,116 +192,182 @@ void __ubsan::__ubsan_handle_shift_out_of_bounds(ShiftOutOfBoundsData *Data,
          "left shift of %0 by %1 places cannot be represented in type %2")
       << LHSVal << RHSVal << Data->LHSType;
 }
+
+void __ubsan::__ubsan_handle_shift_out_of_bounds(ShiftOutOfBoundsData *Data,
+                                                 ValueHandle LHS,
+                                                 ValueHandle RHS) {
+  GET_REPORT_OPTIONS(false);
+  handleShiftOutOfBoundsImpl(Data, LHS, RHS, Opts);
+}
 void __ubsan::__ubsan_handle_shift_out_of_bounds_abort(
                                                      ShiftOutOfBoundsData *Data,
                                                      ValueHandle LHS,
                                                      ValueHandle RHS) {
-  __ubsan_handle_shift_out_of_bounds(Data, LHS, RHS);
+  GET_REPORT_OPTIONS(true);
+  handleShiftOutOfBoundsImpl(Data, LHS, RHS, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_out_of_bounds(OutOfBoundsData *Data,
-                                           ValueHandle Index) {
+static void handleOutOfBoundsImpl(OutOfBoundsData *Data, ValueHandle Index,
+                                  ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Value IndexVal(Data->IndexType, Index);
   Diag(Loc, DL_Error, "index %0 out of bounds for type %1")
     << IndexVal << Data->ArrayType;
 }
+
+void __ubsan::__ubsan_handle_out_of_bounds(OutOfBoundsData *Data,
+                                           ValueHandle Index) {
+  GET_REPORT_OPTIONS(false);
+  handleOutOfBoundsImpl(Data, Index, Opts);
+}
 void __ubsan::__ubsan_handle_out_of_bounds_abort(OutOfBoundsData *Data,
                                                  ValueHandle Index) {
-  __ubsan_handle_out_of_bounds(Data, Index);
+  GET_REPORT_OPTIONS(true);
+  handleOutOfBoundsImpl(Data, Index, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_builtin_unreachable(UnreachableData *Data) {
+static void handleBuiltinUnreachableImpl(UnreachableData *Data,
+                                         ReportOptions Opts) {
+  ScopedReport R(Opts, Data->Loc);
   Diag(Data->Loc, DL_Error, "execution reached a __builtin_unreachable() call");
+}
+
+void __ubsan::__ubsan_handle_builtin_unreachable(UnreachableData *Data) {
+  GET_REPORT_OPTIONS(true);
+  handleBuiltinUnreachableImpl(Data, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_missing_return(UnreachableData *Data) {
+static void handleMissingReturnImpl(UnreachableData *Data, ReportOptions Opts) {
+  ScopedReport R(Opts, Data->Loc);
   Diag(Data->Loc, DL_Error,
        "execution reached the end of a value-returning function "
        "without returning a value");
+}
+
+void __ubsan::__ubsan_handle_missing_return(UnreachableData *Data) {
+  GET_REPORT_OPTIONS(true);
+  handleMissingReturnImpl(Data, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_vla_bound_not_positive(VLABoundData *Data,
-                                                    ValueHandle Bound) {
+static void handleVLABoundNotPositive(VLABoundData *Data, ValueHandle Bound,
+                                      ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error, "variable length array bound evaluates to "
                       "non-positive value %0")
     << Value(Data->Type, Bound);
 }
+
+void __ubsan::__ubsan_handle_vla_bound_not_positive(VLABoundData *Data,
+                                                    ValueHandle Bound) {
+  GET_REPORT_OPTIONS(false);
+  handleVLABoundNotPositive(Data, Bound, Opts);
+}
 void __ubsan::__ubsan_handle_vla_bound_not_positive_abort(VLABoundData *Data,
-                                                           ValueHandle Bound) {
-  __ubsan_handle_vla_bound_not_positive(Data, Bound);
+                                                          ValueHandle Bound) {
+  GET_REPORT_OPTIONS(true);
+  handleVLABoundNotPositive(Data, Bound, Opts);
   Die();
 }
 
-
-void __ubsan::__ubsan_handle_float_cast_overflow(FloatCastOverflowData *Data,
-                                                 ValueHandle From) {
+static void handleFloatCastOverflow(FloatCastOverflowData *Data,
+                                    ValueHandle From, ReportOptions Opts) {
   // TODO: Add deduplication once a SourceLocation is generated for this check.
-  Diag(getCallerLocation(), DL_Error,
+  Location Loc = getCallerLocation();
+  ScopedReport R(Opts, Loc);
+
+  Diag(Loc, DL_Error,
        "value %0 is outside the range of representable values of type %2")
-    << Value(Data->FromType, From) << Data->FromType << Data->ToType;
+      << Value(Data->FromType, From) << Data->FromType << Data->ToType;
 }
-void __ubsan::__ubsan_handle_float_cast_overflow_abort(
-                                                    FloatCastOverflowData *Data,
-                                                    ValueHandle From) {
-  Diag(getCallerLocation(), DL_Error,
-       "value %0 is outside the range of representable values of type %2")
-    << Value(Data->FromType, From) << Data->FromType << Data->ToType;
+
+void __ubsan::__ubsan_handle_float_cast_overflow(FloatCastOverflowData *Data,
+                                                 ValueHandle From) {
+  GET_REPORT_OPTIONS(false);
+  handleFloatCastOverflow(Data, From, Opts);
+}
+void
+__ubsan::__ubsan_handle_float_cast_overflow_abort(FloatCastOverflowData *Data,
+                                                  ValueHandle From) {
+  GET_REPORT_OPTIONS(true);
+  handleFloatCastOverflow(Data, From, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_load_invalid_value(InvalidValueData *Data,
-                                                ValueHandle Val) {
+static void handleLoadInvalidValue(InvalidValueData *Data, ValueHandle Val,
+                                   ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error,
        "load of value %0, which is not a valid value for type %1")
     << Value(Data->Type, Val) << Data->Type;
 }
+
+void __ubsan::__ubsan_handle_load_invalid_value(InvalidValueData *Data,
+                                                ValueHandle Val) {
+  GET_REPORT_OPTIONS(false);
+  handleLoadInvalidValue(Data, Val, Opts);
+}
 void __ubsan::__ubsan_handle_load_invalid_value_abort(InvalidValueData *Data,
                                                       ValueHandle Val) {
-  __ubsan_handle_load_invalid_value(Data, Val);
+  GET_REPORT_OPTIONS(true);
+  handleLoadInvalidValue(Data, Val, Opts);
   Die();
 }
 
-void __ubsan::__ubsan_handle_function_type_mismatch(
-    FunctionTypeMismatchData *Data,
-    ValueHandle Function) {
+static void handleFunctionTypeMismatch(FunctionTypeMismatchData *Data,
+                                       ValueHandle Function,
+                                       ReportOptions Opts) {
   const char *FName = "(unknown)";
 
   Location Loc = getFunctionLocation(Function, &FName);
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Data->Loc, DL_Error,
        "call to function %0 through pointer to incorrect function type %1")
     << FName << Data->Type;
   Diag(Loc, DL_Note, "%0 defined here") << FName;
 }
 
+void
+__ubsan::__ubsan_handle_function_type_mismatch(FunctionTypeMismatchData *Data,
+                                               ValueHandle Function) {
+  GET_REPORT_OPTIONS(false);
+  handleFunctionTypeMismatch(Data, Function, Opts);
+}
+
 void __ubsan::__ubsan_handle_function_type_mismatch_abort(
-    FunctionTypeMismatchData *Data,
-    ValueHandle Function) {
-  __ubsan_handle_function_type_mismatch(Data, Function);
+    FunctionTypeMismatchData *Data, ValueHandle Function) {
+  GET_REPORT_OPTIONS(true);
+  handleFunctionTypeMismatch(Data, Function, Opts);
   Die();
 }
 
-static void handleNonnullReturn(NonNullReturnData *Data) {
+static void handleNonNullReturn(NonNullReturnData *Data, ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error, "null pointer returned from function declared to never "
                       "return null");
   if (!Data->AttrLoc.isInvalid())
@@ -290,19 +375,23 @@ static void handleNonnullReturn(NonNullReturnData *Data) {
 }
 
 void __ubsan::__ubsan_handle_nonnull_return(NonNullReturnData *Data) {
-  handleNonnullReturn(Data);
+  GET_REPORT_OPTIONS(false);
+  handleNonNullReturn(Data, Opts);
 }
 
 void __ubsan::__ubsan_handle_nonnull_return_abort(NonNullReturnData *Data) {
-  handleNonnullReturn(Data);
+  GET_REPORT_OPTIONS(true);
+  handleNonNullReturn(Data, Opts);
   Die();
 }
 
-static void handleNonNullArg(NonNullArgData *Data) {
+static void handleNonNullArg(NonNullArgData *Data, ReportOptions Opts) {
   SourceLocation Loc = Data->Loc.acquire();
-  if (Loc.isDisabled())
+  if (ignoreReport(Loc, Opts))
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error, "null pointer passed as argument %0, which is declared to "
        "never be null") << Data->ArgIndex;
   if (!Data->AttrLoc.isInvalid())
@@ -310,10 +399,12 @@ static void handleNonNullArg(NonNullArgData *Data) {
 }
 
 void __ubsan::__ubsan_handle_nonnull_arg(NonNullArgData *Data) {
-  handleNonNullArg(Data);
+  GET_REPORT_OPTIONS(false);
+  handleNonNullArg(Data, Opts);
 }
 
 void __ubsan::__ubsan_handle_nonnull_arg_abort(NonNullArgData *Data) {
-  handleNonNullArg(Data);
+  GET_REPORT_OPTIONS(true);
+  handleNonNullArg(Data, Opts);
   Die();
 }
diff --git a/libsanitizer/ubsan/ubsan_handlers.h b/libsanitizer/ubsan/ubsan_handlers.h
index 641fbfe993f..92365d81893 100644
--- a/libsanitizer/ubsan/ubsan_handlers.h
+++ b/libsanitizer/ubsan/ubsan_handlers.h
@@ -22,10 +22,14 @@ struct TypeMismatchData {
   unsigned char TypeCheckKind;
 };
 
+#define UNRECOVERABLE(checkname, ...) \
+  extern "C" SANITIZER_INTERFACE_ATTRIBUTE NORETURN \
+    void __ubsan_handle_ ## checkname( __VA_ARGS__ );
+
 #define RECOVERABLE(checkname, ...) \
   extern "C" SANITIZER_INTERFACE_ATTRIBUTE \
     void __ubsan_handle_ ## checkname( __VA_ARGS__ ); \
-  extern "C" SANITIZER_INTERFACE_ATTRIBUTE \
+  extern "C" SANITIZER_INTERFACE_ATTRIBUTE NORETURN \
     void __ubsan_handle_ ## checkname ## _abort( __VA_ARGS__ );
 
 /// \brief Handle a runtime type check failure, caused by either a misaligned
@@ -79,11 +83,9 @@ struct UnreachableData {
 };
 
 /// \brief Handle a __builtin_unreachable which is reached.
-extern "C" SANITIZER_INTERFACE_ATTRIBUTE
-void __ubsan_handle_builtin_unreachable(UnreachableData *Data);
+UNRECOVERABLE(builtin_unreachable, UnreachableData *Data)
 /// \brief Handle reaching the end of a value-returning function.
-extern "C" SANITIZER_INTERFACE_ATTRIBUTE
-void __ubsan_handle_missing_return(UnreachableData *Data);
+UNRECOVERABLE(missing_return, UnreachableData *Data)
 
 struct VLABoundData {
   SourceLocation Loc;
diff --git a/libsanitizer/ubsan/ubsan_handlers_cxx.cc b/libsanitizer/ubsan/ubsan_handlers_cxx.cc
index bb43cc75cfc..86f3e57a8bc 100644
--- a/libsanitizer/ubsan/ubsan_handlers_cxx.cc
+++ b/libsanitizer/ubsan/ubsan_handlers_cxx.cc
@@ -16,6 +16,7 @@
 #include "ubsan_type_hash.h"
 
 #include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_suppressions.h"
 
 using namespace __sanitizer;
 using namespace __ubsan;
@@ -26,47 +27,54 @@ namespace __ubsan {
 
 static void HandleDynamicTypeCacheMiss(
     DynamicTypeCacheMissData *Data, ValueHandle Pointer, ValueHandle Hash,
-    bool Abort) {
+    ReportOptions Opts) {
   if (checkDynamicType((void*)Pointer, Data->TypeInfo, Hash))
     // Just a cache miss. The type matches after all.
     return;
 
+  // Check if error report should be suppressed.
+  DynamicTypeInfo DTI = getDynamicTypeInfo((void*)Pointer);
+  if (DTI.isValid() &&
+      MatchSuppression(DTI.getMostDerivedTypeName(), SuppressionVptrCheck))
+    return;
+
   SourceLocation Loc = Data->Loc.acquire();
   if (Loc.isDisabled())
     return;
 
+  ScopedReport R(Opts, Loc);
+
   Diag(Loc, DL_Error,
        "%0 address %1 which does not point to an object of type %2")
     << TypeCheckKinds[Data->TypeCheckKind] << (void*)Pointer << Data->Type;
 
   // If possible, say what type it actually points to.
-  DynamicTypeInfo DTI = getDynamicTypeInfo((void*)Pointer);
   if (!DTI.isValid())
     Diag(Pointer, DL_Note, "object has invalid vptr")
-      << MangledName(DTI.getMostDerivedTypeName())
-      << Range(Pointer, Pointer + sizeof(uptr), "invalid vptr");
+        << MangledName(DTI.getMostDerivedTypeName())
+        << Range(Pointer, Pointer + sizeof(uptr), "invalid vptr");
   else if (!DTI.getOffset())
     Diag(Pointer, DL_Note, "object is of type %0")
-      << MangledName(DTI.getMostDerivedTypeName())
-      << Range(Pointer, Pointer + sizeof(uptr), "vptr for %0");
+        << MangledName(DTI.getMostDerivedTypeName())
+        << Range(Pointer, Pointer + sizeof(uptr), "vptr for %0");
   else
     // FIXME: Find the type at the specified offset, and include that
     //        in the note.
     Diag(Pointer - DTI.getOffset(), DL_Note,
          "object is base class subobject at offset %0 within object of type %1")
-      << DTI.getOffset() << MangledName(DTI.getMostDerivedTypeName())
-      << MangledName(DTI.getSubobjectTypeName())
-      << Range(Pointer, Pointer + sizeof(uptr), "vptr for %2 base class of %1");
-
-  if (Abort)
-    Die();
+        << DTI.getOffset() << MangledName(DTI.getMostDerivedTypeName())
+        << MangledName(DTI.getSubobjectTypeName())
+        << Range(Pointer, Pointer + sizeof(uptr),
+                 "vptr for %2 base class of %1");
 }
 
 void __ubsan::__ubsan_handle_dynamic_type_cache_miss(
     DynamicTypeCacheMissData *Data, ValueHandle Pointer, ValueHandle Hash) {
-  HandleDynamicTypeCacheMiss(Data, Pointer, Hash, false);
+  GET_REPORT_OPTIONS(false);
+  HandleDynamicTypeCacheMiss(Data, Pointer, Hash, Opts);
 }
 void __ubsan::__ubsan_handle_dynamic_type_cache_miss_abort(
     DynamicTypeCacheMissData *Data, ValueHandle Pointer, ValueHandle Hash) {
-  HandleDynamicTypeCacheMiss(Data, Pointer, Hash, true);
+  GET_REPORT_OPTIONS(true);
+  HandleDynamicTypeCacheMiss(Data, Pointer, Hash, Opts);
 }
diff --git a/libsanitizer/ubsan/ubsan_init.cc b/libsanitizer/ubsan/ubsan_init.cc
new file mode 100644
index 00000000000..f28fc811f50
--- /dev/null
+++ b/libsanitizer/ubsan/ubsan_init.cc
@@ -0,0 +1,59 @@
+//===-- ubsan_init.cc -----------------------------------------------------===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Initialization of UBSan runtime.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ubsan_init.h"
+#include "ubsan_flags.h"
+#include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_libc.h"
+#include "sanitizer_common/sanitizer_mutex.h"
+#include "sanitizer_common/sanitizer_suppressions.h"
+#include "sanitizer_common/sanitizer_symbolizer.h"
+
+using namespace __ubsan;
+
+static bool ubsan_inited;
+
+void __ubsan::InitIfNecessary() {
+#if !SANITIZER_CAN_USE_PREINIT_ARRAY
+  // No need to lock mutex if we're initializing from preinit array.
+  static StaticSpinMutex init_mu;
+  SpinMutexLock l(&init_mu);
+#endif
+  if (LIKELY(ubsan_inited))
+   return;
+  if (0 == internal_strcmp(SanitizerToolName, "SanitizerTool")) {
+    // WARNING: If this condition holds, then either UBSan runs in a standalone
+    // mode, or initializer for another sanitizer hasn't run yet. In a latter
+    // case, another sanitizer will overwrite "SanitizerToolName" and reparse
+    // common flags. It means, that we are not allowed to *use* common flags
+    // in this function.
+    SanitizerToolName = "UndefinedBehaviorSanitizer";
+    InitializeCommonFlags();
+  }
+  // Initialize UBSan-specific flags.
+  InitializeFlags();
+  SuppressionContext::InitIfNecessary();
+  ubsan_inited = true;
+}
+
+#if SANITIZER_CAN_USE_PREINIT_ARRAY
+__attribute__((section(".preinit_array"), used))
+void (*__local_ubsan_preinit)(void) = __ubsan::InitIfNecessary;
+#else
+// Use a dynamic initializer.
+class UbsanInitializer {
+ public:
+  UbsanInitializer() {
+    InitIfNecessary();
+  }
+};
+static UbsanInitializer ubsan_initializer;
+#endif  // SANITIZER_CAN_USE_PREINIT_ARRAY
diff --git a/libsanitizer/ubsan/ubsan_init.h b/libsanitizer/ubsan/ubsan_init.h
new file mode 100644
index 00000000000..b76bbfe7bb7
--- /dev/null
+++ b/libsanitizer/ubsan/ubsan_init.h
@@ -0,0 +1,22 @@
+//===-- ubsan_init.h --------------------------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Initialization function for UBSan runtime.
+//
+//===----------------------------------------------------------------------===//
+#ifndef UBSAN_INIT_H
+#define UBSAN_INIT_H
+
+namespace __ubsan {
+
+// NOTE: This function might take a lock (if .preinit_array initialization is
+// not used). It's generally a bad idea to call it on a fast path.
+void InitIfNecessary();
+
+}  // namespace __ubsan
+
+#endif  // UBSAN_INIT_H