aboutsummaryrefslogtreecommitdiff
path: root/arch
diff options
context:
space:
mode:
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/Kconfig.debug9
-rw-r--r--arch/x86/include/asm/ds.h82
-rw-r--r--arch/x86/include/asm/processor.h35
-rw-r--r--arch/x86/include/asm/ptrace.h9
-rw-r--r--arch/x86/include/asm/tlbflush.h8
-rw-r--r--arch/x86/kernel/Makefile1
-rw-r--r--arch/x86/kernel/ds.c921
-rw-r--r--arch/x86/kernel/ds_selftest.c408
-rw-r--r--arch/x86/kernel/ds_selftest.h15
-rw-r--r--arch/x86/kernel/entry_64.S19
-rw-r--r--arch/x86/kernel/process.c5
-rw-r--r--arch/x86/kernel/process_32.c3
-rw-r--r--arch/x86/kernel/process_64.c3
-rw-r--r--arch/x86/kernel/ptrace.c284
-rw-r--r--arch/x86/kernel/stacktrace.c2
-rw-r--r--arch/x86/mm/kmmio.c104
-rw-r--r--arch/x86/mm/mmio-mod.c2
17 files changed, 1444 insertions, 466 deletions
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 33fac6bbe1c..d105f29bb6b 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -174,6 +174,15 @@ config IOMMU_LEAK
Add a simple leak tracer to the IOMMU code. This is useful when you
are debugging a buggy device driver that leaks IOMMU mappings.
+config X86_DS_SELFTEST
+ bool "DS selftest"
+ default y
+ depends on DEBUG_KERNEL
+ depends on X86_DS
+ ---help---
+ Perform Debug Store selftests at boot time.
+ If in doubt, say "N".
+
config HAVE_MMIOTRACE_SUPPORT
def_bool y
diff --git a/arch/x86/include/asm/ds.h b/arch/x86/include/asm/ds.h
index a8f672ba100..70dac199b09 100644
--- a/arch/x86/include/asm/ds.h
+++ b/arch/x86/include/asm/ds.h
@@ -15,8 +15,8 @@
* - buffer allocation (memory accounting)
*
*
- * Copyright (C) 2007-2008 Intel Corporation.
- * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008
+ * Copyright (C) 2007-2009 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
*/
#ifndef _ASM_X86_DS_H
@@ -83,8 +83,10 @@ enum ds_feature {
* The interrupt threshold is independent from the overflow callback
* to allow users to use their own overflow interrupt handling mechanism.
*
- * task: the task to request recording for;
- * NULL for per-cpu recording on the current cpu
+ * The function might sleep.
+ *
+ * task: the task to request recording for
+ * cpu: the cpu to request recording for
* base: the base pointer for the (non-pageable) buffer;
* size: the size of the provided buffer in bytes
* ovfl: pointer to a function to be called on buffer overflow;
@@ -93,19 +95,28 @@ enum ds_feature {
* -1 if no interrupt threshold is requested.
* flags: a bit-mask of the above flags
*/
-extern struct bts_tracer *ds_request_bts(struct task_struct *task,
- void *base, size_t size,
- bts_ovfl_callback_t ovfl,
- size_t th, unsigned int flags);
-extern struct pebs_tracer *ds_request_pebs(struct task_struct *task,
- void *base, size_t size,
- pebs_ovfl_callback_t ovfl,
- size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+ void *base, size_t size,
+ bts_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags);
+extern struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+ bts_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+ void *base, size_t size,
+ pebs_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags);
+extern struct pebs_tracer *ds_request_pebs_cpu(int cpu,
+ void *base, size_t size,
+ pebs_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags);
/*
* Release BTS or PEBS resources
* Suspend and resume BTS or PEBS tracing
*
+ * Must be called with irq's enabled.
+ *
* tracer: the tracer handle returned from ds_request_~()
*/
extern void ds_release_bts(struct bts_tracer *tracer);
@@ -115,6 +126,28 @@ extern void ds_release_pebs(struct pebs_tracer *tracer);
extern void ds_suspend_pebs(struct pebs_tracer *tracer);
extern void ds_resume_pebs(struct pebs_tracer *tracer);
+/*
+ * Release BTS or PEBS resources
+ * Suspend and resume BTS or PEBS tracing
+ *
+ * Cpu tracers must call this on the traced cpu.
+ * Task tracers must call ds_release_~_noirq() for themselves.
+ *
+ * May be called with irq's disabled.
+ *
+ * Returns 0 if successful;
+ * -EPERM if the cpu tracer does not trace the current cpu.
+ * -EPERM if the task tracer does not trace itself.
+ *
+ * tracer: the tracer handle returned from ds_request_~()
+ */
+extern int ds_release_bts_noirq(struct bts_tracer *tracer);
+extern int ds_suspend_bts_noirq(struct bts_tracer *tracer);
+extern int ds_resume_bts_noirq(struct bts_tracer *tracer);
+extern int ds_release_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_suspend_pebs_noirq(struct pebs_tracer *tracer);
+extern int ds_resume_pebs_noirq(struct pebs_tracer *tracer);
+
/*
* The raw DS buffer state as it is used for BTS and PEBS recording.
@@ -170,9 +203,9 @@ struct bts_struct {
} lbr;
/* BTS_TASK_ARRIVES or BTS_TASK_DEPARTS */
struct {
- __u64 jiffies;
+ __u64 clock;
pid_t pid;
- } timestamp;
+ } event;
} variant;
};
@@ -201,8 +234,12 @@ struct bts_trace {
struct pebs_trace {
struct ds_trace ds;
- /* the PEBS reset value */
- unsigned long long reset_value;
+ /* the number of valid counters in the below array */
+ unsigned int counters;
+
+#define MAX_PEBS_COUNTERS 4
+ /* the counter reset value */
+ unsigned long long counter_reset[MAX_PEBS_COUNTERS];
};
@@ -237,9 +274,11 @@ extern int ds_reset_pebs(struct pebs_tracer *tracer);
* Returns 0 on success; -Eerrno on error
*
* tracer: the tracer handle returned from ds_request_pebs()
+ * counter: the index of the counter
* value: the new counter reset value
*/
-extern int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value);
+extern int ds_set_pebs_reset(struct pebs_tracer *tracer,
+ unsigned int counter, u64 value);
/*
* Initialization
@@ -252,21 +291,12 @@ extern void __cpuinit ds_init_intel(struct cpuinfo_x86 *);
*/
extern void ds_switch_to(struct task_struct *prev, struct task_struct *next);
-/*
- * Task clone/init and cleanup work
- */
-extern void ds_copy_thread(struct task_struct *tsk, struct task_struct *father);
-extern void ds_exit_thread(struct task_struct *tsk);
-
#else /* CONFIG_X86_DS */
struct cpuinfo_x86;
static inline void __cpuinit ds_init_intel(struct cpuinfo_x86 *ignored) {}
static inline void ds_switch_to(struct task_struct *prev,
struct task_struct *next) {}
-static inline void ds_copy_thread(struct task_struct *tsk,
- struct task_struct *father) {}
-static inline void ds_exit_thread(struct task_struct *tsk) {}
#endif /* CONFIG_X86_DS */
#endif /* _ASM_X86_DS_H */
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 87ede2f31bc..c7768269b1c 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -462,14 +462,8 @@ struct thread_struct {
unsigned io_bitmap_max;
/* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
unsigned long debugctlmsr;
-#ifdef CONFIG_X86_DS
-/* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
+ /* Debug Store context; see asm/ds.h */
struct ds_context *ds_ctx;
-#endif /* CONFIG_X86_DS */
-#ifdef CONFIG_X86_PTRACE_BTS
-/* the signal to send on a bts buffer overflow */
- unsigned int bts_ovfl_signal;
-#endif /* CONFIG_X86_PTRACE_BTS */
};
static inline unsigned long native_get_debugreg(int regno)
@@ -797,6 +791,21 @@ static inline unsigned long get_debugctlmsr(void)
return debugctlmsr;
}
+static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
+{
+ u64 debugctlmsr = 0;
+ u32 val1, val2;
+
+#ifndef CONFIG_X86_DEBUGCTLMSR
+ if (boot_cpu_data.x86 < 6)
+ return 0;
+#endif
+ rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
+ debugctlmsr = val1 | ((u64)val2 << 32);
+
+ return debugctlmsr;
+}
+
static inline void update_debugctlmsr(unsigned long debugctlmsr)
{
#ifndef CONFIG_X86_DEBUGCTLMSR
@@ -806,6 +815,18 @@ static inline void update_debugctlmsr(unsigned long debugctlmsr)
wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
}
+static inline void update_debugctlmsr_on_cpu(int cpu,
+ unsigned long debugctlmsr)
+{
+#ifndef CONFIG_X86_DEBUGCTLMSR
+ if (boot_cpu_data.x86 < 6)
+ return;
+#endif
+ wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
+ (u32)((u64)debugctlmsr),
+ (u32)((u64)debugctlmsr >> 32));
+}
+
/*
* from system description table in BIOS. Mostly for MCA use, but
* others may find it useful:
diff --git a/arch/x86/include/asm/ptrace.h b/arch/x86/include/asm/ptrace.h
index 624f133943e..0f0d908349a 100644
--- a/arch/x86/include/asm/ptrace.h
+++ b/arch/x86/include/asm/ptrace.h
@@ -236,12 +236,11 @@ extern int do_get_thread_area(struct task_struct *p, int idx,
extern int do_set_thread_area(struct task_struct *p, int idx,
struct user_desc __user *info, int can_allocate);
-extern void x86_ptrace_untrace(struct task_struct *);
-extern void x86_ptrace_fork(struct task_struct *child,
- unsigned long clone_flags);
+#ifdef CONFIG_X86_PTRACE_BTS
+extern void ptrace_bts_untrace(struct task_struct *tsk);
-#define arch_ptrace_untrace(tsk) x86_ptrace_untrace(tsk)
-#define arch_ptrace_fork(child, flags) x86_ptrace_fork(child, flags)
+#define arch_ptrace_untrace(tsk) ptrace_bts_untrace(tsk)
+#endif /* CONFIG_X86_PTRACE_BTS */
#endif /* __KERNEL__ */
diff --git a/arch/x86/include/asm/tlbflush.h b/arch/x86/include/asm/tlbflush.h
index 16a5c84b032..a5ecc9c33e9 100644
--- a/arch/x86/include/asm/tlbflush.h
+++ b/arch/x86/include/asm/tlbflush.h
@@ -17,7 +17,7 @@
static inline void __native_flush_tlb(void)
{
- write_cr3(read_cr3());
+ native_write_cr3(native_read_cr3());
}
static inline void __native_flush_tlb_global(void)
@@ -32,11 +32,11 @@ static inline void __native_flush_tlb_global(void)
*/
raw_local_irq_save(flags);
- cr4 = read_cr4();
+ cr4 = native_read_cr4();
/* clear PGE */
- write_cr4(cr4 & ~X86_CR4_PGE);
+ native_write_cr4(cr4 & ~X86_CR4_PGE);
/* write old PGE again and flush TLBs */
- write_cr4(cr4);
+ native_write_cr4(cr4);
raw_local_irq_restore(flags);
}
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index 235f5927bb9..4f78bd68212 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -44,6 +44,7 @@ obj-y += process.o
obj-y += i387.o xsave.o
obj-y += ptrace.o
obj-$(CONFIG_X86_DS) += ds.o
+obj-$(CONFIG_X86_DS_SELFTEST) += ds_selftest.o
obj-$(CONFIG_X86_32) += tls.o
obj-$(CONFIG_IA32_EMULATION) += tls.o
obj-y += step.o
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 87b67e3a765..48bfe138603 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -19,45 +19,61 @@
* Markus Metzger <markus.t.metzger@intel.com>, 2007-2009
*/
-
-#include <asm/ds.h>
-
-#include <linux/errno.h>
+#include <linux/kernel.h>
#include <linux/string.h>
-#include <linux/slab.h>
+#include <linux/errno.h>
#include <linux/sched.h>
+#include <linux/slab.h>
#include <linux/mm.h>
-#include <linux/kernel.h>
+#include <linux/trace_clock.h>
+
+#include <asm/ds.h>
+#include "ds_selftest.h"
/*
- * The configuration for a particular DS hardware implementation.
+ * The configuration for a particular DS hardware implementation:
*/
struct ds_configuration {
- /* the name of the configuration */
- const char *name;
- /* the size of one pointer-typed field in the DS structure and
- in the BTS and PEBS buffers in bytes;
- this covers the first 8 DS fields related to buffer management. */
- unsigned char sizeof_field;
- /* the size of a BTS/PEBS record in bytes */
- unsigned char sizeof_rec[2];
- /* a series of bit-masks to control various features indexed
- * by enum ds_feature */
- unsigned long ctl[dsf_ctl_max];
+ /* The name of the configuration: */
+ const char *name;
+
+ /* The size of pointer-typed fields in DS, BTS, and PEBS: */
+ unsigned char sizeof_ptr_field;
+
+ /* The size of a BTS/PEBS record in bytes: */
+ unsigned char sizeof_rec[2];
+
+ /* The number of pebs counter reset values in the DS structure. */
+ unsigned char nr_counter_reset;
+
+ /* Control bit-masks indexed by enum ds_feature: */
+ unsigned long ctl[dsf_ctl_max];
};
-static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
+static struct ds_configuration ds_cfg __read_mostly;
+
+
+/* Maximal size of a DS configuration: */
+#define MAX_SIZEOF_DS 0x80
-#define ds_cfg per_cpu(ds_cfg_array, smp_processor_id())
+/* Maximal size of a BTS record: */
+#define MAX_SIZEOF_BTS (3 * 8)
-#define MAX_SIZEOF_DS (12 * 8) /* maximal size of a DS configuration */
-#define MAX_SIZEOF_BTS (3 * 8) /* maximal size of a BTS record */
-#define DS_ALIGNMENT (1 << 3) /* BTS and PEBS buffer alignment */
+/* BTS and PEBS buffer alignment: */
+#define DS_ALIGNMENT (1 << 3)
-#define BTS_CONTROL \
- (ds_cfg.ctl[dsf_bts] | ds_cfg.ctl[dsf_bts_kernel] | ds_cfg.ctl[dsf_bts_user] |\
- ds_cfg.ctl[dsf_bts_overflow])
+/* Number of buffer pointers in DS: */
+#define NUM_DS_PTR_FIELDS 8
+/* Size of a pebs reset value in DS: */
+#define PEBS_RESET_FIELD_SIZE 8
+
+/* Mask of control bits in the DS MSR register: */
+#define BTS_CONTROL \
+ ( ds_cfg.ctl[dsf_bts] | \
+ ds_cfg.ctl[dsf_bts_kernel] | \
+ ds_cfg.ctl[dsf_bts_user] | \
+ ds_cfg.ctl[dsf_bts_overflow] )
/*
* A BTS or PEBS tracer.
@@ -66,29 +82,36 @@ static DEFINE_PER_CPU(struct ds_configuration, ds_cfg_array);
* to identify tracers.
*/
struct ds_tracer {
- /* the DS context (partially) owned by this tracer */
- struct ds_context *context;
- /* the buffer provided on ds_request() and its size in bytes */
- void *buffer;
- size_t size;
+ /* The DS context (partially) owned by this tracer. */
+ struct ds_context *context;
+ /* The buffer provided on ds_request() and its size in bytes. */
+ void *buffer;
+ size_t size;
};
struct bts_tracer {
- /* the common DS part */
- struct ds_tracer ds;
- /* the trace including the DS configuration */
- struct bts_trace trace;
- /* buffer overflow notification function */
- bts_ovfl_callback_t ovfl;
+ /* The common DS part: */
+ struct ds_tracer ds;
+
+ /* The trace including the DS configuration: */
+ struct bts_trace trace;
+
+ /* Buffer overflow notification function: */
+ bts_ovfl_callback_t ovfl;
+
+ /* Active flags affecting trace collection. */
+ unsigned int flags;
};
struct pebs_tracer {
- /* the common DS part */
- struct ds_tracer ds;
- /* the trace including the DS configuration */
- struct pebs_trace trace;
- /* buffer overflow notification function */
- pebs_ovfl_callback_t ovfl;
+ /* The common DS part: */
+ struct ds_tracer ds;
+
+ /* The trace including the DS configuration: */
+ struct pebs_trace trace;
+
+ /* Buffer overflow notification function: */
+ pebs_ovfl_callback_t ovfl;
};
/*
@@ -97,6 +120,7 @@ struct pebs_tracer {
*
* The DS configuration consists of the following fields; different
* architetures vary in the size of those fields.
+ *
* - double-word aligned base linear address of the BTS buffer
* - write pointer into the BTS buffer
* - end linear address of the BTS buffer (one byte beyond the end of
@@ -135,21 +159,22 @@ enum ds_field {
};
enum ds_qualifier {
- ds_bts = 0,
+ ds_bts = 0,
ds_pebs
};
-static inline unsigned long ds_get(const unsigned char *base,
- enum ds_qualifier qual, enum ds_field field)
+static inline unsigned long
+ds_get(const unsigned char *base, enum ds_qualifier qual, enum ds_field field)
{
- base += (ds_cfg.sizeof_field * (field + (4 * qual)));
+ base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
return *(unsigned long *)base;
}
-static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
- enum ds_field field, unsigned long value)
+static inline void
+ds_set(unsigned char *base, enum ds_qualifier qual, enum ds_field field,
+ unsigned long value)
{
- base += (ds_cfg.sizeof_field * (field + (4 * qual)));
+ base += (ds_cfg.sizeof_ptr_field * (field + (4 * qual)));
(*(unsigned long *)base) = value;
}
@@ -159,7 +184,6 @@ static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
*/
static DEFINE_SPINLOCK(ds_lock);
-
/*
* We either support (system-wide) per-cpu or per-thread allocation.
* We distinguish the two based on the task_struct pointer, where a
@@ -178,12 +202,28 @@ static DEFINE_SPINLOCK(ds_lock);
*/
static atomic_t tracers = ATOMIC_INIT(0);
-static inline void get_tracer(struct task_struct *task)
+static inline int get_tracer(struct task_struct *task)
{
- if (task)
+ int error;
+
+ spin_lock_irq(&ds_lock);
+
+ if (task) {
+ error = -EPERM;
+ if (atomic_read(&tracers) < 0)
+ goto out;
atomic_inc(&tracers);
- else
+ } else {
+ error = -EPERM;
+ if (atomic_read(&tracers) > 0)
+ goto out;
atomic_dec(&tracers);
+ }
+
+ error = 0;
+out:
+ spin_unlock_irq(&ds_lock);
+ return error;
}
static inline void put_tracer(struct task_struct *task)
@@ -194,14 +234,6 @@ static inline void put_tracer(struct task_struct *task)
atomic_inc(&tracers);
}
-static inline int check_tracer(struct task_struct *task)
-{
- return task ?
- (atomic_read(&tracers) >= 0) :
- (atomic_read(&tracers) <= 0);
-}
-
-
/*
* The DS context is either attached to a thread or to a cpu:
* - in the former case, the thread_struct contains a pointer to the
@@ -213,61 +245,58 @@ static inline int check_tracer(struct task_struct *task)
* deallocated when the last user puts the context.
*/
struct ds_context {
- /* pointer to the DS configuration; goes into MSR_IA32_DS_AREA */
- unsigned char ds[MAX_SIZEOF_DS];
- /* the owner of the BTS and PEBS configuration, respectively */
- struct bts_tracer *bts_master;
- struct pebs_tracer *pebs_master;
- /* use count */
- unsigned long count;
- /* a pointer to the context location inside the thread_struct
- * or the per_cpu context array */
- struct ds_context **this;
- /* a pointer to the task owning this context, or NULL, if the
- * context is owned by a cpu */
- struct task_struct *task;
-};
+ /* The DS configuration; goes into MSR_IA32_DS_AREA: */
+ unsigned char ds[MAX_SIZEOF_DS];
+
+ /* The owner of the BTS and PEBS configuration, respectively: */
+ struct bts_tracer *bts_master;
+ struct pebs_tracer *pebs_master;
-static DEFINE_PER_CPU(struct ds_context *, system_context_array);
+ /* Use count: */
+ unsigned long count;
-#define system_context per_cpu(system_context_array, smp_processor_id())
+ /* Pointer to the context pointer field: */
+ struct ds_context **this;
+
+ /* The traced task; NULL for cpu tracing: */
+ struct task_struct *task;
+
+ /* The traced cpu; only valid if task is NULL: */
+ int cpu;
+};
+static DEFINE_PER_CPU(struct ds_context *, cpu_context);
-static inline struct ds_context *ds_get_context(struct task_struct *task)
+
+static struct ds_context *ds_get_context(struct task_struct *task, int cpu)
{
struct ds_context **p_context =
- (task ? &task->thread.ds_ctx : &system_context);
+ (task ? &task->thread.ds_ctx : &per_cpu(cpu_context, cpu));
struct ds_context *context = NULL;
struct ds_context *new_context = NULL;
- unsigned long irq;
/* Chances are small that we already have a context. */
new_context = kzalloc(sizeof(*new_context), GFP_KERNEL);
if (!new_context)
return NULL;
- spin_lock_irqsave(&ds_lock, irq);
+ spin_lock_irq(&ds_lock);
context = *p_context;
- if (!context) {
+ if (likely(!context)) {
context = new_context;
context->this = p_context;
context->task = task;
+ context->cpu = cpu;
context->count = 0;
- if (task)
- set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
-
- if (!task || (task == current))
- wrmsrl(MSR_IA32_DS_AREA, (unsigned long)context->ds);
-
*p_context = context;
}
context->count++;
- spin_unlock_irqrestore(&ds_lock, irq);
+ spin_unlock_irq(&ds_lock);
if (context != new_context)
kfree(new_context);
@@ -275,8 +304,9 @@ static inline struct ds_context *ds_get_context(struct task_struct *task)
return context;
}
-static inline void ds_put_context(struct ds_context *context)
+static void ds_put_context(struct ds_context *context)
{
+ struct task_struct *task;
unsigned long irq;
if (!context)
@@ -291,17 +321,55 @@ static inline void ds_put_context(struct ds_context *context)
*(context->this) = NULL;
- if (context->task)
- clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+ task = context->task;
+
+ if (task)
+ clear_tsk_thread_flag(task, TIF_DS_AREA_MSR);
- if (!context->task || (context->task == current))
- wrmsrl(MSR_IA32_DS_AREA, 0);
+ /*
+ * We leave the (now dangling) pointer to the DS configuration in
+ * the DS_AREA msr. This is as good or as bad as replacing it with
+ * NULL - the hardware would crash if we enabled tracing.
+ *
+ * This saves us some problems with having to write an msr on a
+ * different cpu while preventing others from doing the same for the
+ * next context for that same cpu.
+ */
spin_unlock_irqrestore(&ds_lock, irq);
+ /* The context might still be in use for context switching. */
+ if (task && (task != current))
+ wait_task_context_switch(task);
+
kfree(context);
}
+static void ds_install_ds_area(struct ds_context *context)
+{
+ unsigned long ds;
+
+ ds = (unsigned long)context->ds;
+
+ /*
+ * There is a race between the bts master and the pebs master.
+ *
+ * The thread/cpu access is synchronized via get/put_cpu() for
+ * task tracing and via wrmsr_on_cpu for cpu tracing.
+ *
+ * If bts and pebs are collected for the same task or same cpu,
+ * the same confiuration is written twice.
+ */
+ if (context->task) {
+ get_cpu();
+ if (context->task == current)
+ wrmsrl(MSR_IA32_DS_AREA, ds);
+ set_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+ put_cpu();
+ } else
+ wrmsr_on_cpu(context->cpu, MSR_IA32_DS_AREA,
+ (u32)((u64)ds), (u32)((u64)ds >> 32));
+}
/*
* Call the tracer's callback on a buffer overflow.
@@ -332,9 +400,9 @@ static void ds_overflow(struct ds_context *context, enum ds_qualifier qual)
* The remainder of any partially written record is zeroed out.
*
* context: the DS context
- * qual: the buffer type
- * record: the data to write
- * size: the size of the data
+ * qual: the buffer type
+ * record: the data to write
+ * size: the size of the data
*/
static int ds_write(struct ds_context *context, enum ds_qualifier qual,
const void *record, size_t size)
@@ -349,14 +417,14 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
unsigned long write_size, adj_write_size;
/*
- * write as much as possible without producing an
+ * Write as much as possible without producing an
* overflow interrupt.
*
- * interrupt_threshold must either be
+ * Interrupt_threshold must either be
* - bigger than absolute_maximum or
* - point to a record between buffer_base and absolute_maximum
*
- * index points to a valid record.
+ * Index points to a valid record.
*/
base = ds_get(context->ds, qual, ds_buffer_base);
index = ds_get(context->ds, qual, ds_index);
@@ -365,8 +433,10 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
write_end = min(end, int_th);
- /* if we are already beyond the interrupt threshold,
- * we fill the entire buffer */
+ /*
+ * If we are already beyond the interrupt threshold,
+ * we fill the entire buffer.
+ */
if (write_end <= index)
write_end = end;
@@ -383,7 +453,7 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
adj_write_size *= ds_cfg.sizeof_rec[qual];
- /* zero out trailing bytes */
+ /* Zero out trailing bytes. */
memset((char *)index + write_size, 0,
adj_write_size - write_size);
index += adj_write_size;
@@ -410,7 +480,7 @@ static int ds_write(struct ds_context *context, enum ds_qualifier qual,
* Later architectures use 64bit pointers throughout, whereas earlier
* architectures use 32bit pointers in 32bit mode.
*
- * We compute the base address for the first 8 fields based on:
+ * We compute the base address for the fields based on:
* - the field size stored in the DS configuration
* - the relative field position
*
@@ -431,23 +501,23 @@ enum bts_field {
bts_to,
bts_flags,
- bts_qual = bts_from,
- bts_jiffies = bts_to,
- bts_pid = bts_flags,
+ bts_qual = bts_from,
+ bts_clock = bts_to,
+ bts_pid = bts_flags,
- bts_qual_mask = (bts_qual_max - 1),
- bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
+ bts_qual_mask = (bts_qual_max - 1),
+ bts_escape = ((unsigned long)-1 & ~bts_qual_mask)
};
static inline unsigned long bts_get(const char *base, enum bts_field field)
{
- base += (ds_cfg.sizeof_field * field);
+ base += (ds_cfg.sizeof_ptr_field * field);
return *(unsigned long *)base;
}
static inline void bts_set(char *base, enum bts_field field, unsigned long val)
{
- base += (ds_cfg.sizeof_field * field);;
+ base += (ds_cfg.sizeof_ptr_field * field);;
(*(unsigned long *)base) = val;
}
@@ -463,8 +533,8 @@ static inline void bts_set(char *base, enum bts_field field, unsigned long val)
*
* return: bytes read/written on success; -Eerrno, otherwise
*/
-static int bts_read(struct bts_tracer *tracer, const void *at,
- struct bts_struct *out)
+static int
+bts_read(struct bts_tracer *tracer, const void *at, struct bts_struct *out)
{
if (!tracer)
return -EINVAL;
@@ -478,8 +548,8 @@ static int bts_read(struct bts_tracer *tracer, const void *at,
memset(out, 0, sizeof(*out));
if ((bts_get(at, bts_qual) & ~bts_qual_mask) == bts_escape) {
out->qualifier = (bts_get(at, bts_qual) & bts_qual_mask);
- out->variant.timestamp.jiffies = bts_get(at, bts_jiffies);
- out->variant.timestamp.pid = bts_get(at, bts_pid);
+ out->variant.event.clock = bts_get(at, bts_clock);
+ out->variant.event.pid = bts_get(at, bts_pid);
} else {
out->qualifier = bts_branch;
out->variant.lbr.from = bts_get(at, bts_from);
@@ -516,8 +586,8 @@ static int bts_write(struct bts_tracer *tracer, const struct bts_struct *in)
case bts_task_arrives:
case bts_task_departs:
bts_set(raw, bts_qual, (bts_escape | in->qualifier));
- bts_set(raw, bts_jiffies, in->variant.timestamp.jiffies);
- bts_set(raw, bts_pid, in->variant.timestamp.pid);
+ bts_set(raw, bts_clock, in->variant.event.clock);
+ bts_set(raw, bts_pid, in->variant.event.pid);
break;
default:
return -EINVAL;
@@ -555,7 +625,8 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
unsigned int flags) {
unsigned long buffer, adj;
- /* adjust the buffer address and size to meet alignment
+ /*
+ * Adjust the buffer address and size to meet alignment
* constraints:
* - buffer is double-word aligned
* - size is multiple of record size
@@ -577,9 +648,11 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
trace->begin = (void *)buffer;
trace->top = trace->begin;
trace->end = (void *)(buffer + size);
- /* The value for 'no threshold' is -1, which will set the
+ /*
+ * The value for 'no threshold' is -1, which will set the
* threshold outside of the buffer, just like we want it.
*/
+ ith *= ds_cfg.sizeof_rec[qual];
trace->ith = (void *)(buffer + size - ith);
trace->flags = flags;
@@ -588,18 +661,27 @@ static void ds_init_ds_trace(struct ds_trace *trace, enum ds_qualifier qual,
static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
enum ds_qualifier qual, struct task_struct *task,
- void *base, size_t size, size_t th, unsigned int flags)
+ int cpu, void *base, size_t size, size_t th)
{
struct ds_context *context;
int error;
+ size_t req_size;
+
+ error = -EOPNOTSUPP;
+ if (!ds_cfg.sizeof_rec[qual])
+ goto out;
error = -EINVAL;
if (!base)
goto out;
- /* we require some space to do alignment adjustments below */
+ req_size = ds_cfg.sizeof_rec[qual];
+ /* We might need space for alignment adjustments. */
+ if (!IS_ALIGNED((unsigned long)base, DS_ALIGNMENT))
+ req_size += DS_ALIGNMENT;
+
error = -EINVAL;
- if (size < (DS_ALIGNMENT + ds_cfg.sizeof_rec[qual]))
+ if (size < req_size)
goto out;
if (th != (size_t)-1) {
@@ -614,182 +696,318 @@ static int ds_request(struct ds_tracer *tracer, struct ds_trace *trace,
tracer->size = size;
error = -ENOMEM;
- context = ds_get_context(task);
+ context = ds_get_context(task, cpu);
if (!context)
goto out;
tracer->context = context;
- ds_init_ds_trace(trace, qual, base, size, th, flags);
+ /*
+ * Defer any tracer-specific initialization work for the context until
+ * context ownership has been clarified.
+ */
error = 0;
out:
return error;
}
-struct bts_tracer *ds_request_bts(struct task_struct *task,
- void *base, size_t size,
- bts_ovfl_callback_t ovfl, size_t th,
- unsigned int flags)
+static struct bts_tracer *ds_request_bts(struct task_struct *task, int cpu,
+ void *base, size_t size,
+ bts_ovfl_callback_t ovfl, size_t th,
+ unsigned int flags)
{
struct bts_tracer *tracer;
- unsigned long irq;
int error;
+ /* Buffer overflow notification is not yet implemented. */
error = -EOPNOTSUPP;
- if (!ds_cfg.ctl[dsf_bts])
+ if (ovfl)
goto out;
- /* buffer overflow notification is not yet implemented */
- error = -EOPNOTSUPP;
- if (ovfl)
+ error = get_tracer(task);
+ if (error < 0)
goto out;
error = -ENOMEM;
tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
if (!tracer)
- goto out;
+ goto out_put_tracer;
tracer->ovfl = ovfl;
+ /* Do some more error checking and acquire a tracing context. */
error = ds_request(&tracer->ds, &tracer->trace.ds,
- ds_bts, task, base, size, th, flags);
+ ds_bts, task, cpu, base, size, th);
if (error < 0)
goto out_tracer;
-
- spin_lock_irqsave(&ds_lock, irq);
-
- error = -EPERM;
- if (!check_tracer(task))
- goto out_unlock;
- get_tracer(task);
+ /* Claim the bts part of the tracing context we acquired above. */
+ spin_lock_irq(&ds_lock);
error = -EPERM;
if (tracer->ds.context->bts_master)
- goto out_put_tracer;
+ goto out_unlock;
tracer->ds.context->bts_master = tracer;
- spin_unlock_irqrestore(&ds_lock, irq);
+ spin_unlock_irq(&ds_lock);
+ /*
+ * Now that we own the bts part of the context, let's complete the
+ * initialization for that part.
+ */
+ ds_init_ds_trace(&tracer->trace.ds, ds_bts, base, size, th, flags);
+ ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+ ds_install_ds_area(tracer->ds.context);
tracer->trace.read = bts_read;
tracer->trace.write = bts_write;
- ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_bts);
+ /* Start tracing. */
ds_resume_bts(tracer);
return tracer;
- out_put_tracer:
- put_tracer(task);
out_unlock:
- spin_unlock_irqrestore(&ds_lock, irq);
+ spin_unlock_irq(&ds_lock);
ds_put_context(tracer->ds.context);
out_tracer:
kfree(tracer);
+ out_put_tracer:
+ put_tracer(task);
out:
return ERR_PTR(error);
}
-struct pebs_tracer *ds_request_pebs(struct task_struct *task,
- void *base, size_t size,
- pebs_ovfl_callback_t ovfl, size_t th,
- unsigned int flags)
+struct bts_tracer *ds_request_bts_task(struct task_struct *task,
+ void *base, size_t size,
+ bts_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags)
+{
+ return ds_request_bts(task, 0, base, size, ovfl, th, flags);
+}
+
+struct bts_tracer *ds_request_bts_cpu(int cpu, void *base, size_t size,
+ bts_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags)
+{
+ return ds_request_bts(NULL, cpu, base, size, ovfl, th, flags);
+}
+
+static struct pebs_tracer *ds_request_pebs(struct task_struct *task, int cpu,
+ void *base, size_t size,
+ pebs_ovfl_callback_t ovfl, size_t th,
+ unsigned int flags)
{
struct pebs_tracer *tracer;
- unsigned long irq;
int error;
- /* buffer overflow notification is not yet implemented */
+ /* Buffer overflow notification is not yet implemented. */
error = -EOPNOTSUPP;
if (ovfl)
goto out;
+ error = get_tracer(task);
+ if (error < 0)
+ goto out;
+
error = -ENOMEM;
tracer = kzalloc(sizeof(*tracer), GFP_KERNEL);
if (!tracer)
- goto out;
+ goto out_put_tracer;
tracer->ovfl = ovfl;
+ /* Do some more error checking and acquire a tracing context. */
error = ds_request(&tracer->ds, &tracer->trace.ds,
- ds_pebs, task, base, size, th, flags);
+ ds_pebs, task, cpu, base, size, th);
if (error < 0)
goto out_tracer;
- spin_lock_irqsave(&ds_lock, irq);
-
- error = -EPERM;
- if (!check_tracer(task))
- goto out_unlock;
- get_tracer(task);
+ /* Claim the pebs part of the tracing context we acquired above. */
+ spin_lock_irq(&ds_lock);
error = -EPERM;
if (tracer->ds.context->pebs_master)
- goto out_put_tracer;
+ goto out_unlock;
tracer->ds.context->pebs_master = tracer;
- spin_unlock_irqrestore(&ds_lock, irq);
+ spin_unlock_irq(&ds_lock);
+ /*
+ * Now that we own the pebs part of the context, let's complete the
+ * initialization for that part.
+ */
+ ds_init_ds_trace(&tracer->trace.ds, ds_pebs, base, size, th, flags);
ds_write_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
+ ds_install_ds_area(tracer->ds.context);
+
+ /* Start tracing. */
ds_resume_pebs(tracer);
return tracer;
- out_put_tracer:
- put_tracer(task);
out_unlock:
- spin_unlock_irqrestore(&ds_lock, irq);
+ spin_unlock_irq(&ds_lock);
ds_put_context(tracer->ds.context);
out_tracer:
kfree(tracer);
+ out_put_tracer:
+ put_tracer(task);
out:
return ERR_PTR(error);
}
-void ds_release_bts(struct bts_tracer *tracer)
+struct pebs_tracer *ds_request_pebs_task(struct task_struct *task,
+ void *base, size_t size,
+ pebs_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags)
{
- if (!tracer)
- return;
+ return ds_request_pebs(task, 0, base, size, ovfl, th, flags);
+}
- ds_suspend_bts(tracer);
+struct pebs_tracer *ds_request_pebs_cpu(int cpu, void *base, size_t size,
+ pebs_ovfl_callback_t ovfl,
+ size_t th, unsigned int flags)
+{
+ return ds_request_pebs(NULL, cpu, base, size, ovfl, th, flags);
+}
+
+static void ds_free_bts(struct bts_tracer *tracer)
+{
+ struct task_struct *task;
+
+ task = tracer->ds.context->task;
WARN_ON_ONCE(tracer->ds.context->bts_master != tracer);
tracer->ds.context->bts_master = NULL;
- put_tracer(tracer->ds.context->task);
+ /* Make sure tracing stopped and the tracer is not in use. */
+ if (task && (task != current))
+ wait_task_context_switch(task);
+
ds_put_context(tracer->ds.context);
+ put_tracer(task);
kfree(tracer);
}
+void ds_release_bts(struct bts_tracer *tracer)
+{
+ might_sleep();
+
+ if (!tracer)
+ return;
+
+ ds_suspend_bts(tracer);
+ ds_free_bts(tracer);
+}
+
+int ds_release_bts_noirq(struct bts_tracer *tracer)
+{
+ struct task_struct *task;
+ unsigned long irq;
+ int error;
+
+ if (!tracer)
+ return 0;
+
+ task = tracer->ds.context->task;
+
+ local_irq_save(irq);
+
+ error = -EPERM;
+ if (!task &&
+ (tracer->ds.context->cpu != smp_processor_id()))
+ goto out;
+
+ error = -EPERM;
+ if (task && (task != current))
+ goto out;
+
+ ds_suspend_bts_noirq(tracer);
+ ds_free_bts(tracer);
+
+ error = 0;
+ out:
+ local_irq_restore(irq);
+ return error;
+}
+
+static void update_task_debugctlmsr(struct task_struct *task,
+ unsigned long debugctlmsr)
+{
+ task->thread.debugctlmsr = debugctlmsr;
+
+ get_cpu();
+ if (task == current)
+ update_debugctlmsr(debugctlmsr);
+ put_cpu();
+}
+
void ds_suspend_bts(struct bts_tracer *tracer)
{
struct task_struct *task;
+ unsigned long debugctlmsr;
+ int cpu;
if (!tracer)
return;
+ tracer->flags = 0;
+
task = tracer->ds.context->task;
+ cpu = tracer->ds.context->cpu;
- if (!task || (task == current))
- update_debugctlmsr(get_debugctlmsr() & ~BTS_CONTROL);
+ WARN_ON(!task && irqs_disabled());
- if (task) {
- task->thread.debugctlmsr &= ~BTS_CONTROL;
+ debugctlmsr = (task ?
+ task->thread.debugctlmsr :
+ get_debugctlmsr_on_cpu(cpu));
+ debugctlmsr &= ~BTS_CONTROL;
- if (!task->thread.debugctlmsr)
- clear_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
- }
+ if (task)
+ update_task_debugctlmsr(task, debugctlmsr);
+ else
+ update_debugctlmsr_on_cpu(cpu, debugctlmsr);
}
-void ds_resume_bts(struct bts_tracer *tracer)
+int ds_suspend_bts_noirq(struct bts_tracer *tracer)
{
struct task_struct *task;
- unsigned long control;
+ unsigned long debugctlmsr, irq;
+ int cpu, error = 0;
if (!tracer)
- return;
+ return 0;
+
+ tracer->flags = 0;
task = tracer->ds.context->task;
+ cpu = tracer->ds.context->cpu;
+
+ local_irq_save(irq);
+
+ error = -EPERM;
+ if (!task && (cpu != smp_processor_id()))
+ goto out;
+
+ debugctlmsr = (task ?
+ task->thread.debugctlmsr :
+ get_debugctlmsr());
+ debugctlmsr &= ~BTS_CONTROL;
+
+ if (task)
+ update_task_debugctlmsr(task, debugctlmsr);
+ else
+ update_debugctlmsr(debugctlmsr);
+
+ error = 0;
+ out:
+ local_irq_restore(irq);
+ return error;
+}
+
+static unsigned long ds_bts_control(struct bts_tracer *tracer)
+{
+ unsigned long control;
control = ds_cfg.ctl[dsf_bts];
if (!(tracer->trace.ds.flags & BTS_KERNEL))
@@ -797,41 +1015,149 @@ void ds_resume_bts(struct bts_tracer *tracer)
if (!(tracer->trace.ds.flags & BTS_USER))
control |= ds_cfg.ctl[dsf_bts_user];
- if (task) {
- task->thread.debugctlmsr |= control;
- set_tsk_thread_flag(task, TIF_DEBUGCTLMSR);
- }
-
- if (!task || (task == current))
- update_debugctlmsr(get_debugctlmsr() | control);
+ return control;
}
-void ds_release_pebs(struct pebs_tracer *tracer)
+void ds_resume_bts(struct bts_tracer *tracer)
{
+ struct task_struct *task;
+ unsigned long debugctlmsr;
+ int cpu;
+
if (!tracer)
return;
- ds_suspend_pebs(tracer);
+ tracer->flags = tracer->trace.ds.flags;
+
+ task = tracer->ds.context->task;
+ cpu = tracer->ds.context->cpu;
+
+ WARN_ON(!task && irqs_disabled());
+
+ debugctlmsr = (task ?
+ task->thread.debugctlmsr :
+ get_debugctlmsr_on_cpu(cpu));
+ debugctlmsr |= ds_bts_control(tracer);
+
+ if (task)
+ update_task_debugctlmsr(task, debugctlmsr);
+ else
+ update_debugctlmsr_on_cpu(cpu, debugctlmsr);
+}
+
+int ds_resume_bts_noirq(struct bts_tracer *tracer)
+{
+ struct task_struct *task;
+ unsigned long debugctlmsr, irq;
+ int cpu, error = 0;
+
+ if (!tracer)
+ return 0;
+
+ tracer->flags = tracer->trace.ds.flags;
+
+ task = tracer->ds.context->task;
+ cpu = tracer->ds.context->cpu;
+
+ local_irq_save(irq);
+
+ error = -EPERM;
+ if (!task && (cpu != smp_processor_id()))
+ goto out;
+
+ debugctlmsr = (task ?
+ task->thread.debugctlmsr :
+ get_debugctlmsr());
+ debugctlmsr |= ds_bts_control(tracer);
+
+ if (task)
+ update_task_debugctlmsr(task, debugctlmsr);
+ else
+ update_debugctlmsr(debugctlmsr);
+
+ error = 0;
+ out:
+ local_irq_restore(irq);
+ return error;
+}
+
+static void ds_free_pebs(struct pebs_tracer *tracer)
+{
+ struct task_struct *task;
+
+ task = tracer->ds.context->task;
WARN_ON_ONCE(tracer->ds.context->pebs_master != tracer);
tracer->ds.context->pebs_master = NULL;
- put_tracer(tracer->ds.context->task);
ds_put_context(tracer->ds.context);
+ put_tracer(task);
kfree(tracer);
}
+void ds_release_pebs(struct pebs_tracer *tracer)
+{
+ might_sleep();
+
+ if (!tracer)
+ return;
+
+ ds_suspend_pebs(tracer);
+ ds_free_pebs(tracer);
+}
+
+int ds_release_pebs_noirq(struct pebs_tracer *tracer)
+{
+ struct task_struct *task;
+ unsigned long irq;
+ int error;
+
+ if (!tracer)
+ return 0;
+
+ task = tracer->ds.context->task;
+
+ local_irq_save(irq);
+
+ error = -EPERM;
+ if (!task &&
+ (tracer->ds.context->cpu != smp_processor_id()))
+ goto out;
+
+ error = -EPERM;
+ if (task && (task != current))
+ goto out;
+
+ ds_suspend_pebs_noirq(tracer);
+ ds_free_pebs(tracer);
+
+ error = 0;
+ out:
+ local_irq_restore(irq);
+ return error;
+}
+
void ds_suspend_pebs(struct pebs_tracer *tracer)
{
}
+int ds_suspend_pebs_noirq(struct pebs_tracer *tracer)
+{
+ return 0;
+}
+
void ds_resume_pebs(struct pebs_tracer *tracer)
{
}
+int ds_resume_pebs_noirq(struct pebs_tracer *tracer)
+{
+ return 0;
+}
+
const struct bts_trace *ds_read_bts(struct bts_tracer *tracer)
{
if (!tracer)
@@ -847,8 +1173,12 @@ const struct pebs_trace *ds_read_pebs(struct pebs_tracer *tracer)
return NULL;
ds_read_config(tracer->ds.context, &tracer->trace.ds, ds_pebs);
- tracer->trace.reset_value =
- *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8));
+
+ tracer->trace.counters = ds_cfg.nr_counter_reset;
+ memcpy(tracer->trace.counter_reset,
+ tracer->ds.context->ds +
+ (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field),
+ ds_cfg.nr_counter_reset * PEBS_RESET_FIELD_SIZE);
return &tracer->trace;
}
@@ -873,18 +1203,24 @@ int ds_reset_pebs(struct pebs_tracer *tracer)
tracer->trace.ds.top = tracer->trace.ds.begin;
- ds_set(tracer->ds.context->ds, ds_bts, ds_index,
+ ds_set(tracer->ds.context->ds, ds_pebs, ds_index,
(unsigned long)tracer->trace.ds.top);
return 0;
}
-int ds_set_pebs_reset(struct pebs_tracer *tracer, u64 value)
+int ds_set_pebs_reset(struct pebs_tracer *tracer,
+ unsigned int counter, u64 value)
{
if (!tracer)
return -EINVAL;
- *(u64 *)(tracer->ds.context->ds + (ds_cfg.sizeof_field * 8)) = value;
+ if (ds_cfg.nr_counter_reset < counter)
+ return -EINVAL;
+
+ *(u64 *)(tracer->ds.context->ds +
+ (NUM_DS_PTR_FIELDS * ds_cfg.sizeof_ptr_field) +
+ (counter * PEBS_RESET_FIELD_SIZE)) = value;
return 0;
}
@@ -894,73 +1230,117 @@ static const struct ds_configuration ds_cfg_netburst = {
.ctl[dsf_bts] = (1 << 2) | (1 << 3),
.ctl[dsf_bts_kernel] = (1 << 5),
.ctl[dsf_bts_user] = (1 << 6),
-
- .sizeof_field = sizeof(long),
- .sizeof_rec[ds_bts] = sizeof(long) * 3,
-#ifdef __i386__
- .sizeof_rec[ds_pebs] = sizeof(long) * 10,
-#else
- .sizeof_rec[ds_pebs] = sizeof(long) * 18,
-#endif
+ .nr_counter_reset = 1,
};
static const struct ds_configuration ds_cfg_pentium_m = {
.name = "Pentium M",
.ctl[dsf_bts] = (1 << 6) | (1 << 7),
-
- .sizeof_field = sizeof(long),
- .sizeof_rec[ds_bts] = sizeof(long) * 3,
-#ifdef __i386__
- .sizeof_rec[ds_pebs] = sizeof(long) * 10,
-#else
- .sizeof_rec[ds_pebs] = sizeof(long) * 18,
-#endif
+ .nr_counter_reset = 1,
};
static const struct ds_configuration ds_cfg_core2_atom = {
.name = "Core 2/Atom",
.ctl[dsf_bts] = (1 << 6) | (1 << 7),
.ctl[dsf_bts_kernel] = (1 << 9),
.ctl[dsf_bts_user] = (1 << 10),
-
- .sizeof_field = 8,
- .sizeof_rec[ds_bts] = 8 * 3,
- .sizeof_rec[ds_pebs] = 8 * 18,
+ .nr_counter_reset = 1,
+};
+static const struct ds_configuration ds_cfg_core_i7 = {
+ .name = "Core i7",
+ .ctl[dsf_bts] = (1 << 6) | (1 << 7),
+ .ctl[dsf_bts_kernel] = (1 << 9),
+ .ctl[dsf_bts_user] = (1 << 10),
+ .nr_counter_reset = 4,
};
static void
-ds_configure(const struct ds_configuration *cfg)
+ds_configure(const struct ds_configuration *cfg,
+ struct cpuinfo_x86 *cpu)
{
+ unsigned long nr_pebs_fields = 0;
+
+ printk(KERN_INFO "[ds] using %s configuration\n", cfg->name);
+
+#ifdef __i386__
+ nr_pebs_fields = 10;
+#else
+ nr_pebs_fields = 18;
+#endif
+
+ /*
+ * Starting with version 2, architectural performance
+ * monitoring supports a format specifier.
+ */
+ if ((cpuid_eax(0xa) & 0xff) > 1) {
+ unsigned long perf_capabilities, format;
+
+ rdmsrl(MSR_IA32_PERF_CAPABILITIES, perf_capabilities);
+
+ format = (perf_capabilities >> 8) & 0xf;
+
+ switch (format) {
+ case 0:
+ nr_pebs_fields = 18;
+ break;
+ case 1:
+ nr_pebs_fields = 22;
+ break;
+ default:
+ printk(KERN_INFO
+ "[ds] unknown PEBS format: %lu\n", format);
+ nr_pebs_fields = 0;
+ break;
+ }
+ }
+
memset(&ds_cfg, 0, sizeof(ds_cfg));
ds_cfg = *cfg;
- printk(KERN_INFO "[ds] using %s configuration\n", ds_cfg.name);
+ ds_cfg.sizeof_ptr_field =
+ (cpu_has(cpu, X86_FEATURE_DTES64) ? 8 : 4);
+
+ ds_cfg.sizeof_rec[ds_bts] = ds_cfg.sizeof_ptr_field * 3;
+ ds_cfg.sizeof_rec[ds_pebs] = ds_cfg.sizeof_ptr_field * nr_pebs_fields;
- if (!cpu_has_bts) {
- ds_cfg.ctl[dsf_bts] = 0;
+ if (!cpu_has(cpu, X86_FEATURE_BTS)) {
+ ds_cfg.sizeof_rec[ds_bts] = 0;
printk(KERN_INFO "[ds] bts not available\n");
}
- if (!cpu_has_pebs)
+ if (!cpu_has(cpu, X86_FEATURE_PEBS)) {
+ ds_cfg.sizeof_rec[ds_pebs] = 0;
printk(KERN_INFO "[ds] pebs not available\n");
+ }
+
+ printk(KERN_INFO "[ds] sizes: address: %u bit, ",
+ 8 * ds_cfg.sizeof_ptr_field);
+ printk("bts/pebs record: %u/%u bytes\n",
+ ds_cfg.sizeof_rec[ds_bts], ds_cfg.sizeof_rec[ds_pebs]);
- WARN_ON_ONCE(MAX_SIZEOF_DS < (12 * ds_cfg.sizeof_field));
+ WARN_ON_ONCE(MAX_PEBS_COUNTERS < ds_cfg.nr_counter_reset);
}
void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
{
+ /* Only configure the first cpu. Others are identical. */
+ if (ds_cfg.name)
+ return;
+
switch (c->x86) {
case 0x6:
switch (c->x86_model) {
case 0x9:
case 0xd: /* Pentium M */
- ds_configure(&ds_cfg_pentium_m);
+ ds_configure(&ds_cfg_pentium_m, c);
break;
case 0xf:
case 0x17: /* Core2 */
case 0x1c: /* Atom */
- ds_configure(&ds_cfg_core2_atom);
+ ds_configure(&ds_cfg_core2_atom, c);
+ break;
+ case 0x1a: /* Core i7 */
+ ds_configure(&ds_cfg_core_i7, c);
break;
- case 0x1a: /* i7 */
default:
- /* sorry, don't know about them */
+ /* Sorry, don't know about them. */
break;
}
break;
@@ -969,64 +1349,89 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c)
case 0x0:
case 0x1:
case 0x2: /* Netburst */
- ds_configure(&ds_cfg_netburst);
+ ds_configure(&ds_cfg_netburst, c);
break;
default:
- /* sorry, don't know about them */
+ /* Sorry, don't know about them. */
break;
}
break;
default:
- /* sorry, don't know about them */
+ /* Sorry, don't know about them. */
break;
}
}
+static inline void ds_take_timestamp(struct ds_context *context,
+ enum bts_qualifier qualifier,
+ struct task_struct *task)
+{
+ struct bts_tracer *tracer = context->bts_master;
+ struct bts_struct ts;
+
+ /* Prevent compilers from reading the tracer pointer twice. */
+ barrier();
+
+ if (!tracer || !(tracer->flags & BTS_TIMESTAMPS))
+ return;
+
+ memset(&ts, 0, sizeof(ts));
+ ts.qualifier = qualifier;
+ ts.variant.event.clock = trace_clock_global();
+ ts.variant.event.pid = task->pid;
+
+ bts_write(tracer, &ts);
+}
+
/*
* Change the DS configuration from tracing prev to tracing next.
*/
void ds_switch_to(struct task_struct *prev, struct task_struct *next)
{
- struct ds_context *prev_ctx = prev->thread.ds_ctx;
- struct ds_context *next_ctx = next->thread.ds_ctx;
+ struct ds_context *prev_ctx = prev->thread.ds_ctx;
+ struct ds_context *next_ctx = next->thread.ds_ctx;
+ unsigned long debugctlmsr = next->thread.debugctlmsr;
+
+ /* Make sure all data is read before we start. */
+ barrier();
if (prev_ctx) {
update_debugctlmsr(0);
- if (prev_ctx->bts_master &&
- (prev_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
- struct bts_struct ts = {
- .qualifier = bts_task_departs,
- .variant.timestamp.jiffies = jiffies_64,
- .variant.timestamp.pid = prev->pid
- };
- bts_write(prev_ctx->bts_master, &ts);
- }
+ ds_take_timestamp(prev_ctx, bts_task_departs, prev);
}
if (next_ctx) {
- if (next_ctx->bts_master &&
- (next_ctx->bts_master->trace.ds.flags & BTS_TIMESTAMPS)) {
- struct bts_struct ts = {
- .qualifier = bts_task_arrives,
- .variant.timestamp.jiffies = jiffies_64,
- .variant.timestamp.pid = next->pid
- };
- bts_write(next_ctx->bts_master, &ts);
- }
+ ds_take_timestamp(next_ctx, bts_task_arrives, next);
wrmsrl(MSR_IA32_DS_AREA, (unsigned long)next_ctx->ds);
}
- update_debugctlmsr(next->thread.debugctlmsr);
+ update_debugctlmsr(debugctlmsr);
}
-void ds_copy_thread(struct task_struct *tsk, struct task_struct *father)
+static __init int ds_selftest(void)
{
- clear_tsk_thread_flag(tsk, TIF_DS_AREA_MSR);
- tsk->thread.ds_ctx = NULL;
-}
+ if (ds_cfg.sizeof_rec[ds_bts]) {
+ int error;
-void ds_exit_thread(struct task_struct *tsk)
-{
+ error = ds_selftest_bts();
+ if (error) {
+ WARN(1, "[ds] selftest failed. disabling bts.\n");
+ ds_cfg.sizeof_rec[ds_bts] = 0;
+ }
+ }
+
+ if (ds_cfg.sizeof_rec[ds_pebs]) {
+ int error;
+
+ error = ds_selftest_pebs();
+ if (error) {
+ WARN(1, "[ds] selftest failed. disabling pebs.\n");
+ ds_cfg.sizeof_rec[ds_pebs] = 0;
+ }
+ }
+
+ return 0;
}
+device_initcall(ds_selftest);
diff --git a/arch/x86/kernel/ds_selftest.c b/arch/x86/kernel/ds_selftest.c
new file mode 100644
index 00000000000..6bc7c199ab9
--- /dev/null
+++ b/arch/x86/kernel/ds_selftest.c
@@ -0,0 +1,408 @@
+/*
+ * Debug Store support - selftest
+ *
+ *
+ * Copyright (C) 2009 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2009
+ */
+
+#include "ds_selftest.h"
+
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/smp.h>
+#include <linux/cpu.h>
+
+#include <asm/ds.h>
+
+
+#define BUFFER_SIZE 521 /* Intentionally chose an odd size. */
+#define SMALL_BUFFER_SIZE 24 /* A single bts entry. */
+
+struct ds_selftest_bts_conf {
+ struct bts_tracer *tracer;
+ int error;
+ int (*suspend)(struct bts_tracer *);
+ int (*resume)(struct bts_tracer *);
+};
+
+static int ds_selftest_bts_consistency(const struct bts_trace *trace)
+{
+ int error = 0;
+
+ if (!trace) {
+ printk(KERN_CONT "failed to access trace...");
+ /* Bail out. Other tests are pointless. */
+ return -1;
+ }
+
+ if (!trace->read) {
+ printk(KERN_CONT "bts read not available...");
+ error = -1;
+ }
+
+ /* Do some sanity checks on the trace configuration. */
+ if (!trace->ds.n) {
+ printk(KERN_CONT "empty bts buffer...");
+ error = -1;
+ }
+ if (!trace->ds.size) {
+ printk(KERN_CONT "bad bts trace setup...");
+ error = -1;
+ }
+ if (trace->ds.end !=
+ (char *)trace->ds.begin + (trace->ds.n * trace->ds.size)) {
+ printk(KERN_CONT "bad bts buffer setup...");
+ error = -1;
+ }
+ /*
+ * We allow top in [begin; end], since its not clear when the
+ * overflow adjustment happens: after the increment or before the
+ * write.
+ */
+ if ((trace->ds.top < trace->ds.begin) ||
+ (trace->ds.end < trace->ds.top)) {
+ printk(KERN_CONT "bts top out of bounds...");
+ error = -1;
+ }
+
+ return error;
+}
+
+static int ds_selftest_bts_read(struct bts_tracer *tracer,
+ const struct bts_trace *trace,
+ const void *from, const void *to)
+{
+ const unsigned char *at;
+
+ /*
+ * Check a few things which do not belong to this test.
+ * They should be covered by other tests.
+ */
+ if (!trace)
+ return -1;
+
+ if (!trace->read)
+ return -1;
+
+ if (to < from)
+ return -1;
+
+ if (from < trace->ds.begin)
+ return -1;
+
+ if (trace->ds.end < to)
+ return -1;
+
+ if (!trace->ds.size)
+ return -1;
+
+ /* Now to the test itself. */
+ for (at = from; (void *)at < to; at += trace->ds.size) {
+ struct bts_struct bts;
+ unsigned long index;
+ int error;
+
+ if (((void *)at - trace->ds.begin) % trace->ds.size) {
+ printk(KERN_CONT
+ "read from non-integer index...");
+ return -1;
+ }
+ index = ((void *)at - trace->ds.begin) / trace->ds.size;
+
+ memset(&bts, 0, sizeof(bts));
+ error = trace->read(tracer, at, &bts);
+ if (error < 0) {
+ printk(KERN_CONT
+ "error reading bts trace at [%lu] (0x%p)...",
+ index, at);
+ return error;
+ }
+
+ switch (bts.qualifier) {
+ case BTS_BRANCH:
+ break;
+ default:
+ printk(KERN_CONT
+ "unexpected bts entry %llu at [%lu] (0x%p)...",
+ bts.qualifier, index, at);
+ return -1;
+ }
+ }
+
+ return 0;
+}
+
+static void ds_selftest_bts_cpu(void *arg)
+{
+ struct ds_selftest_bts_conf *conf = arg;
+ const struct bts_trace *trace;
+ void *top;
+
+ if (IS_ERR(conf->tracer)) {
+ conf->error = PTR_ERR(conf->tracer);
+ conf->tracer = NULL;
+
+ printk(KERN_CONT
+ "initialization failed (err: %d)...", conf->error);
+ return;
+ }
+
+ /* We should meanwhile have enough trace. */
+ conf->error = conf->suspend(conf->tracer);
+ if (conf->error < 0)
+ return;
+
+ /* Let's see if we can access the trace. */
+ trace = ds_read_bts(conf->tracer);
+
+ conf->error = ds_selftest_bts_consistency(trace);
+ if (conf->error < 0)
+ return;
+
+ /* If everything went well, we should have a few trace entries. */
+ if (trace->ds.top == trace->ds.begin) {
+ /*
+ * It is possible but highly unlikely that we got a
+ * buffer overflow and end up at exactly the same
+ * position we started from.
+ * Let's issue a warning, but continue.
+ */
+ printk(KERN_CONT "no trace/overflow...");
+ }
+
+ /* Let's try to read the trace we collected. */
+ conf->error =
+ ds_selftest_bts_read(conf->tracer, trace,
+ trace->ds.begin, trace->ds.top);
+ if (conf->error < 0)
+ return;
+
+ /*
+ * Let's read the trace again.
+ * Since we suspended tracing, we should get the same result.
+ */
+ top = trace->ds.top;
+
+ trace = ds_read_bts(conf->tracer);
+ conf->error = ds_selftest_bts_consistency(trace);
+ if (conf->error < 0)
+ return;
+
+ if (top != trace->ds.top) {
+ printk(KERN_CONT "suspend not working...");
+ conf->error = -1;
+ return;
+ }
+
+ /* Let's collect some more trace - see if resume is working. */
+ conf->error = conf->resume(conf->tracer);
+ if (conf->error < 0)
+ return;
+
+ conf->error = conf->suspend(conf->tracer);
+ if (conf->error < 0)
+ return;
+
+ trace = ds_read_bts(conf->tracer);
+
+ conf->error = ds_selftest_bts_consistency(trace);
+ if (conf->error < 0)
+ return;
+
+ if (trace->ds.top == top) {
+ /*
+ * It is possible but highly unlikely that we got a
+ * buffer overflow and end up at exactly the same
+ * position we started from.
+ * Let's issue a warning and check the full trace.
+ */
+ printk(KERN_CONT
+ "no resume progress/overflow...");
+
+ conf->error =
+ ds_selftest_bts_read(conf->tracer, trace,
+ trace->ds.begin, trace->ds.end);
+ } else if (trace->ds.top < top) {
+ /*
+ * We had a buffer overflow - the entire buffer should
+ * contain trace records.
+ */
+ conf->error =
+ ds_selftest_bts_read(conf->tracer, trace,
+ trace->ds.begin, trace->ds.end);
+ } else {
+ /*
+ * It is quite likely that the buffer did not overflow.
+ * Let's just check the delta trace.
+ */
+ conf->error =
+ ds_selftest_bts_read(conf->tracer, trace, top,
+ trace->ds.top);
+ }
+ if (conf->error < 0)
+ return;
+
+ conf->error = 0;
+}
+
+static int ds_suspend_bts_wrap(struct bts_tracer *tracer)
+{
+ ds_suspend_bts(tracer);
+ return 0;
+}
+
+static int ds_resume_bts_wrap(struct bts_tracer *tracer)
+{
+ ds_resume_bts(tracer);
+ return 0;
+}
+
+static void ds_release_bts_noirq_wrap(void *tracer)
+{
+ (void)ds_release_bts_noirq(tracer);
+}
+
+static int ds_selftest_bts_bad_release_noirq(int cpu,
+ struct bts_tracer *tracer)
+{
+ int error = -EPERM;
+
+ /* Try to release the tracer on the wrong cpu. */
+ get_cpu();
+ if (cpu != smp_processor_id()) {
+ error = ds_release_bts_noirq(tracer);
+ if (error != -EPERM)
+ printk(KERN_CONT "release on wrong cpu...");
+ }
+ put_cpu();
+
+ return error ? 0 : -1;
+}
+
+static int ds_selftest_bts_bad_request_cpu(int cpu, void *buffer)
+{
+ struct bts_tracer *tracer;
+ int error;
+
+ /* Try to request cpu tracing while task tracing is active. */
+ tracer = ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE, NULL,
+ (size_t)-1, BTS_KERNEL);
+ error = PTR_ERR(tracer);
+ if (!IS_ERR(tracer)) {
+ ds_release_bts(tracer);
+ error = 0;
+ }
+
+ if (error != -EPERM)
+ printk(KERN_CONT "cpu/task tracing overlap...");
+
+ return error ? 0 : -1;
+}
+
+static int ds_selftest_bts_bad_request_task(void *buffer)
+{
+ struct bts_tracer *tracer;
+ int error;
+
+ /* Try to request cpu tracing while task tracing is active. */
+ tracer = ds_request_bts_task(current, buffer, BUFFER_SIZE, NULL,
+ (size_t)-1, BTS_KERNEL);
+ error = PTR_ERR(tracer);
+ if (!IS_ERR(tracer)) {
+ error = 0;
+ ds_release_bts(tracer);
+ }
+
+ if (error != -EPERM)
+ printk(KERN_CONT "task/cpu tracing overlap...");
+
+ return error ? 0 : -1;
+}
+
+int ds_selftest_bts(void)
+{
+ struct ds_selftest_bts_conf conf;
+ unsigned char buffer[BUFFER_SIZE], *small_buffer;
+ unsigned long irq;
+ int cpu;
+
+ printk(KERN_INFO "[ds] bts selftest...");
+ conf.error = 0;
+
+ small_buffer = (unsigned char *)ALIGN((unsigned long)buffer, 8) + 8;
+
+ get_online_cpus();
+ for_each_online_cpu(cpu) {
+ conf.suspend = ds_suspend_bts_wrap;
+ conf.resume = ds_resume_bts_wrap;
+ conf.tracer =
+ ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+ NULL, (size_t)-1, BTS_KERNEL);
+ ds_selftest_bts_cpu(&conf);
+ if (conf.error >= 0)
+ conf.error = ds_selftest_bts_bad_request_task(buffer);
+ ds_release_bts(conf.tracer);
+ if (conf.error < 0)
+ goto out;
+
+ conf.suspend = ds_suspend_bts_noirq;
+ conf.resume = ds_resume_bts_noirq;
+ conf.tracer =
+ ds_request_bts_cpu(cpu, buffer, BUFFER_SIZE,
+ NULL, (size_t)-1, BTS_KERNEL);
+ smp_call_function_single(cpu, ds_selftest_bts_cpu, &conf, 1);
+ if (conf.error >= 0) {
+ conf.error =
+ ds_selftest_bts_bad_release_noirq(cpu,
+ conf.tracer);
+ /* We must not release the tracer twice. */
+ if (conf.error < 0)
+ conf.tracer = NULL;
+ }
+ if (conf.error >= 0)
+ conf.error = ds_selftest_bts_bad_request_task(buffer);
+ smp_call_function_single(cpu, ds_release_bts_noirq_wrap,
+ conf.tracer, 1);
+ if (conf.error < 0)
+ goto out;
+ }
+
+ conf.suspend = ds_suspend_bts_wrap;
+ conf.resume = ds_resume_bts_wrap;
+ conf.tracer =
+ ds_request_bts_task(current, buffer, BUFFER_SIZE,
+ NULL, (size_t)-1, BTS_KERNEL);
+ ds_selftest_bts_cpu(&conf);
+ if (conf.error >= 0)
+ conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+ ds_release_bts(conf.tracer);
+ if (conf.error < 0)
+ goto out;
+
+ conf.suspend = ds_suspend_bts_noirq;
+ conf.resume = ds_resume_bts_noirq;
+ conf.tracer =
+ ds_request_bts_task(current, small_buffer, SMALL_BUFFER_SIZE,
+ NULL, (size_t)-1, BTS_KERNEL);
+ local_irq_save(irq);
+ ds_selftest_bts_cpu(&conf);
+ if (conf.error >= 0)
+ conf.error = ds_selftest_bts_bad_request_cpu(0, buffer);
+ ds_release_bts_noirq(conf.tracer);
+ local_irq_restore(irq);
+ if (conf.error < 0)
+ goto out;
+
+ conf.error = 0;
+ out:
+ put_online_cpus();
+ printk(KERN_CONT "%s.\n", (conf.error ? "failed" : "passed"));
+
+ return conf.error;
+}
+
+int ds_selftest_pebs(void)
+{
+ return 0;
+}
diff --git a/arch/x86/kernel/ds_selftest.h b/arch/x86/kernel/ds_selftest.h
new file mode 100644
index 00000000000..2ba8745c666
--- /dev/null
+++ b/arch/x86/kernel/ds_selftest.h
@@ -0,0 +1,15 @@
+/*
+ * Debug Store support - selftest
+ *
+ *
+ * Copyright (C) 2009 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2009
+ */
+
+#ifdef CONFIG_X86_DS_SELFTEST
+extern int ds_selftest_bts(void);
+extern int ds_selftest_pebs(void);
+#else
+static inline int ds_selftest_bts(void) { return 0; }
+static inline int ds_selftest_pebs(void) { return 0; }
+#endif
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index bb01ce080b8..1c17d7c751a 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -147,27 +147,14 @@ END(ftrace_graph_caller)
GLOBAL(return_to_handler)
subq $80, %rsp
+ /* Save the return values */
movq %rax, (%rsp)
- movq %rcx, 8(%rsp)
- movq %rdx, 16(%rsp)
- movq %rsi, 24(%rsp)
- movq %rdi, 32(%rsp)
- movq %r8, 40(%rsp)
- movq %r9, 48(%rsp)
- movq %r10, 56(%rsp)
- movq %r11, 64(%rsp)
+ movq %rdx, 8(%rsp)
call ftrace_return_to_handler
movq %rax, 72(%rsp)
- movq 64(%rsp), %r11
- movq 56(%rsp), %r10
- movq 48(%rsp), %r9
- movq 40(%rsp), %r8
- movq 32(%rsp), %rdi
- movq 24(%rsp), %rsi
- movq 16(%rsp), %rdx
- movq 8(%rsp), %rcx
+ movq 8(%rsp), %rdx
movq (%rsp), %rax
addq $72, %rsp
retq
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index e22d63bdc8f..3bb2be1649b 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -16,6 +16,7 @@
#include <asm/idle.h>
#include <asm/uaccess.h>
#include <asm/i387.h>
+#include <asm/ds.h>
unsigned long idle_halt;
EXPORT_SYMBOL(idle_halt);
@@ -47,6 +48,8 @@ void free_thread_xstate(struct task_struct *tsk)
kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
tsk->thread.xstate = NULL;
}
+
+ WARN(tsk->thread.ds_ctx, "leaking DS context\n");
}
void free_thread_info(struct thread_info *ti)
@@ -85,8 +88,6 @@ void exit_thread(void)
put_cpu();
kfree(bp);
}
-
- ds_exit_thread(current);
}
void flush_thread(void)
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index c60924b5d12..59f4524984a 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -287,7 +287,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
p->thread.io_bitmap_max = 0;
}
- ds_copy_thread(p, current);
+ clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+ p->thread.ds_ctx = NULL;
clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
p->thread.debugctlmsr = 0;
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 45f010fb2e2..ebefb5407b9 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -332,7 +332,8 @@ int copy_thread(unsigned long clone_flags, unsigned long sp,
goto out;
}
- ds_copy_thread(p, me);
+ clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
+ p->thread.ds_ctx = NULL;
clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
p->thread.debugctlmsr = 0;
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index 23b7c8f017e..09ecbde91c1 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -21,6 +21,7 @@
#include <linux/audit.h>
#include <linux/seccomp.h>
#include <linux/signal.h>
+#include <linux/workqueue.h>
#include <asm/uaccess.h>
#include <asm/pgtable.h>
@@ -578,17 +579,130 @@ static int ioperm_get(struct task_struct *target,
}
#ifdef CONFIG_X86_PTRACE_BTS
+/*
+ * A branch trace store context.
+ *
+ * Contexts may only be installed by ptrace_bts_config() and only for
+ * ptraced tasks.
+ *
+ * Contexts are destroyed when the tracee is detached from the tracer.
+ * The actual destruction work requires interrupts enabled, so the
+ * work is deferred and will be scheduled during __ptrace_unlink().
+ *
+ * Contexts hold an additional task_struct reference on the traced
+ * task, as well as a reference on the tracer's mm.
+ *
+ * Ptrace already holds a task_struct for the duration of ptrace operations,
+ * but since destruction is deferred, it may be executed after both
+ * tracer and tracee exited.
+ */
+struct bts_context {
+ /* The branch trace handle. */
+ struct bts_tracer *tracer;
+
+ /* The buffer used to store the branch trace and its size. */
+ void *buffer;
+ unsigned int size;
+
+ /* The mm that paid for the above buffer. */
+ struct mm_struct *mm;
+
+ /* The task this context belongs to. */
+ struct task_struct *task;
+
+ /* The signal to send on a bts buffer overflow. */
+ unsigned int bts_ovfl_signal;
+
+ /* The work struct to destroy a context. */
+ struct work_struct work;
+};
+
+static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
+{
+ void *buffer = NULL;
+ int err = -ENOMEM;
+
+ err = account_locked_memory(current->mm, current->signal->rlim, size);
+ if (err < 0)
+ return err;
+
+ buffer = kzalloc(size, GFP_KERNEL);
+ if (!buffer)
+ goto out_refund;
+
+ context->buffer = buffer;
+ context->size = size;
+ context->mm = get_task_mm(current);
+
+ return 0;
+
+ out_refund:
+ refund_locked_memory(current->mm, size);
+ return err;
+}
+
+static inline void free_bts_buffer(struct bts_context *context)
+{
+ if (!context->buffer)
+ return;
+
+ kfree(context->buffer);
+ context->buffer = NULL;
+
+ refund_locked_memory(context->mm, context->size);
+ context->size = 0;
+
+ mmput(context->mm);
+ context->mm = NULL;
+}
+
+static void free_bts_context_work(struct work_struct *w)
+{
+ struct bts_context *context;
+
+ context = container_of(w, struct bts_context, work);
+
+ ds_release_bts(context->tracer);
+ put_task_struct(context->task);
+ free_bts_buffer(context);
+ kfree(context);
+}
+
+static inline void free_bts_context(struct bts_context *context)
+{
+ INIT_WORK(&context->work, free_bts_context_work);
+ schedule_work(&context->work);
+}
+
+static inline struct bts_context *alloc_bts_context(struct task_struct *task)
+{
+ struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
+ if (context) {
+ context->task = task;
+ task->bts = context;
+
+ get_task_struct(task);
+ }
+
+ return context;
+}
+
static int ptrace_bts_read_record(struct task_struct *child, size_t index,
struct bts_struct __user *out)
{
+ struct bts_context *context;
const struct bts_trace *trace;
struct bts_struct bts;
const unsigned char *at;
int error;
- trace = ds_read_bts(child->bts);
+ context = child->bts;
+ if (!context)
+ return -ESRCH;
+
+ trace = ds_read_bts(context->tracer);
if (!trace)
- return -EPERM;
+ return -ESRCH;
at = trace->ds.top - ((index + 1) * trace->ds.size);
if ((void *)at < trace->ds.begin)
@@ -597,7 +711,7 @@ static int ptrace_bts_read_record(struct task_struct *child, size_t index,
if (!trace->read)
return -EOPNOTSUPP;
- error = trace->read(child->bts, at, &bts);
+ error = trace->read(context->tracer, at, &bts);
if (error < 0)
return error;
@@ -611,13 +725,18 @@ static int ptrace_bts_drain(struct task_struct *child,
long size,
struct bts_struct __user *out)
{
+ struct bts_context *context;
const struct bts_trace *trace;
const unsigned char *at;
int error, drained = 0;
- trace = ds_read_bts(child->bts);
+ context = child->bts;
+ if (!context)
+ return -ESRCH;
+
+ trace = ds_read_bts(context->tracer);
if (!trace)
- return -EPERM;
+ return -ESRCH;
if (!trace->read)
return -EOPNOTSUPP;
@@ -628,9 +747,8 @@ static int ptrace_bts_drain(struct task_struct *child,
for (at = trace->ds.begin; (void *)at < trace->ds.top;
out++, drained++, at += trace->ds.size) {
struct bts_struct bts;
- int error;
- error = trace->read(child->bts, at, &bts);
+ error = trace->read(context->tracer, at, &bts);
if (error < 0)
return error;
@@ -640,35 +758,18 @@ static int ptrace_bts_drain(struct task_struct *child,
memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
- error = ds_reset_bts(child->bts);
+ error = ds_reset_bts(context->tracer);
if (error < 0)
return error;
return drained;
}
-static int ptrace_bts_allocate_buffer(struct task_struct *child, size_t size)
-{
- child->bts_buffer = alloc_locked_buffer(size);
- if (!child->bts_buffer)
- return -ENOMEM;
-
- child->bts_size = size;
-
- return 0;
-}
-
-static void ptrace_bts_free_buffer(struct task_struct *child)
-{
- free_locked_buffer(child->bts_buffer, child->bts_size);
- child->bts_buffer = NULL;
- child->bts_size = 0;
-}
-
static int ptrace_bts_config(struct task_struct *child,
long cfg_size,
const struct ptrace_bts_config __user *ucfg)
{
+ struct bts_context *context;
struct ptrace_bts_config cfg;
unsigned int flags = 0;
@@ -678,28 +779,33 @@ static int ptrace_bts_config(struct task_struct *child,
if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
return -EFAULT;
- if (child->bts) {
- ds_release_bts(child->bts);
- child->bts = NULL;
- }
+ context = child->bts;
+ if (!context)
+ context = alloc_bts_context(child);
+ if (!context)
+ return -ENOMEM;
if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
if (!cfg.signal)
return -EINVAL;
- child->thread.bts_ovfl_signal = cfg.signal;
return -EOPNOTSUPP;
+ context->bts_ovfl_signal = cfg.signal;
}
- if ((cfg.flags & PTRACE_BTS_O_ALLOC) &&
- (cfg.size != child->bts_size)) {
- int error;
+ ds_release_bts(context->tracer);
+ context->tracer = NULL;
- ptrace_bts_free_buffer(child);
+ if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
+ int err;
- error = ptrace_bts_allocate_buffer(child, cfg.size);
- if (error < 0)
- return error;
+ free_bts_buffer(context);
+ if (!cfg.size)
+ return 0;
+
+ err = alloc_bts_buffer(context, cfg.size);
+ if (err < 0)
+ return err;
}
if (cfg.flags & PTRACE_BTS_O_TRACE)
@@ -708,15 +814,14 @@ static int ptrace_bts_config(struct task_struct *child,
if (cfg.flags & PTRACE_BTS_O_SCHED)
flags |= BTS_TIMESTAMPS;
- child->bts = ds_request_bts(child, child->bts_buffer, child->bts_size,
- /* ovfl = */ NULL, /* th = */ (size_t)-1,
- flags);
- if (IS_ERR(child->bts)) {
- int error = PTR_ERR(child->bts);
-
- ptrace_bts_free_buffer(child);
- child->bts = NULL;
+ context->tracer =
+ ds_request_bts_task(child, context->buffer, context->size,
+ NULL, (size_t)-1, flags);
+ if (unlikely(IS_ERR(context->tracer))) {
+ int error = PTR_ERR(context->tracer);
+ free_bts_buffer(context);
+ context->tracer = NULL;
return error;
}
@@ -727,20 +832,25 @@ static int ptrace_bts_status(struct task_struct *child,
long cfg_size,
struct ptrace_bts_config __user *ucfg)
{
+ struct bts_context *context;
const struct bts_trace *trace;
struct ptrace_bts_config cfg;
+ context = child->bts;
+ if (!context)
+ return -ESRCH;
+
if (cfg_size < sizeof(cfg))
return -EIO;
- trace = ds_read_bts(child->bts);
+ trace = ds_read_bts(context->tracer);
if (!trace)
- return -EPERM;
+ return -ESRCH;
memset(&cfg, 0, sizeof(cfg));
- cfg.size = trace->ds.end - trace->ds.begin;
- cfg.signal = child->thread.bts_ovfl_signal;
- cfg.bts_size = sizeof(struct bts_struct);
+ cfg.size = trace->ds.end - trace->ds.begin;
+ cfg.signal = context->bts_ovfl_signal;
+ cfg.bts_size = sizeof(struct bts_struct);
if (cfg.signal)
cfg.flags |= PTRACE_BTS_O_SIGNAL;
@@ -759,80 +869,51 @@ static int ptrace_bts_status(struct task_struct *child,
static int ptrace_bts_clear(struct task_struct *child)
{
+ struct bts_context *context;
const struct bts_trace *trace;
- trace = ds_read_bts(child->bts);
+ context = child->bts;
+ if (!context)
+ return -ESRCH;
+
+ trace = ds_read_bts(context->tracer);
if (!trace)
- return -EPERM;
+ return -ESRCH;
memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
- return ds_reset_bts(child->bts);
+ return ds_reset_bts(context->tracer);
}
static int ptrace_bts_size(struct task_struct *child)
{
+ struct bts_context *context;
const struct bts_trace *trace;
- trace = ds_read_bts(child->bts);
+ context = child->bts;
+ if (!context)
+ return -ESRCH;
+
+ trace = ds_read_bts(context->tracer);
if (!trace)
- return -EPERM;
+ return -ESRCH;
return (trace->ds.top - trace->ds.begin) / trace->ds.size;
}
-static void ptrace_bts_fork(struct task_struct *tsk)
-{
- tsk->bts = NULL;
- tsk->bts_buffer = NULL;
- tsk->bts_size = 0;
- tsk->thread.bts_ovfl_signal = 0;
-}
-
-static void ptrace_bts_untrace(struct task_struct *child)
+/*
+ * Called from __ptrace_unlink() after the child has been moved back
+ * to its original parent.
+ */
+void ptrace_bts_untrace(struct task_struct *child)
{
if (unlikely(child->bts)) {
- ds_release_bts(child->bts);
+ free_bts_context(child->bts);
child->bts = NULL;
-
- /* We cannot update total_vm and locked_vm since
- child's mm is already gone. But we can reclaim the
- memory. */
- kfree(child->bts_buffer);
- child->bts_buffer = NULL;
- child->bts_size = 0;
}
}
-
-static void ptrace_bts_detach(struct task_struct *child)
-{
- /*
- * Ptrace_detach() races with ptrace_untrace() in case
- * the child dies and is reaped by another thread.
- *
- * We only do the memory accounting at this point and
- * leave the buffer deallocation and the bts tracer
- * release to ptrace_bts_untrace() which will be called
- * later on with tasklist_lock held.
- */
- release_locked_buffer(child->bts_buffer, child->bts_size);
-}
-#else
-static inline void ptrace_bts_fork(struct task_struct *tsk) {}
-static inline void ptrace_bts_detach(struct task_struct *child) {}
-static inline void ptrace_bts_untrace(struct task_struct *child) {}
#endif /* CONFIG_X86_PTRACE_BTS */
-void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
-{
- ptrace_bts_fork(child);
-}
-
-void x86_ptrace_untrace(struct task_struct *child)
-{
- ptrace_bts_untrace(child);
-}
-
/*
* Called by kernel/ptrace.c when detaching..
*
@@ -844,7 +925,6 @@ void ptrace_disable(struct task_struct *child)
#ifdef TIF_SYSCALL_EMU
clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
#endif
- ptrace_bts_detach(child);
}
#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
diff --git a/arch/x86/kernel/stacktrace.c b/arch/x86/kernel/stacktrace.c
index f7bddc2e37d..4aaf7e48394 100644
--- a/arch/x86/kernel/stacktrace.c
+++ b/arch/x86/kernel/stacktrace.c
@@ -20,7 +20,7 @@ save_stack_warning_symbol(void *data, char *msg, unsigned long symbol)
static int save_stack_stack(void *data, char *name)
{
- return -1;
+ return 0;
}
static void save_stack_address(void *data, unsigned long addr, int reliable)
diff --git a/arch/x86/mm/kmmio.c b/arch/x86/mm/kmmio.c
index 50dc802a1c4..16ccbd77917 100644
--- a/arch/x86/mm/kmmio.c
+++ b/arch/x86/mm/kmmio.c
@@ -32,7 +32,7 @@ struct kmmio_fault_page {
struct list_head list;
struct kmmio_fault_page *release_next;
unsigned long page; /* location of the fault page */
- bool old_presence; /* page presence prior to arming */
+ pteval_t old_presence; /* page presence prior to arming */
bool armed;
/*
@@ -97,60 +97,62 @@ static struct kmmio_probe *get_kmmio_probe(unsigned long addr)
static struct kmmio_fault_page *get_kmmio_fault_page(unsigned long page)
{
struct list_head *head;
- struct kmmio_fault_page *p;
+ struct kmmio_fault_page *f;
page &= PAGE_MASK;
head = kmmio_page_list(page);
- list_for_each_entry_rcu(p, head, list) {
- if (p->page == page)
- return p;
+ list_for_each_entry_rcu(f, head, list) {
+ if (f->page == page)
+ return f;
}
return NULL;
}
-static void set_pmd_presence(pmd_t *pmd, bool present, bool *old)
+static void clear_pmd_presence(pmd_t *pmd, bool clear, pmdval_t *old)
{
pmdval_t v = pmd_val(*pmd);
- *old = !!(v & _PAGE_PRESENT);
- v &= ~_PAGE_PRESENT;
- if (present)
- v |= _PAGE_PRESENT;
+ if (clear) {
+ *old = v & _PAGE_PRESENT;
+ v &= ~_PAGE_PRESENT;
+ } else /* presume this has been called with clear==true previously */
+ v |= *old;
set_pmd(pmd, __pmd(v));
}
-static void set_pte_presence(pte_t *pte, bool present, bool *old)
+static void clear_pte_presence(pte_t *pte, bool clear, pteval_t *old)
{
pteval_t v = pte_val(*pte);
- *old = !!(v & _PAGE_PRESENT);
- v &= ~_PAGE_PRESENT;
- if (present)
- v |= _PAGE_PRESENT;
+ if (clear) {
+ *old = v & _PAGE_PRESENT;
+ v &= ~_PAGE_PRESENT;
+ } else /* presume this has been called with clear==true previously */
+ v |= *old;
set_pte_atomic(pte, __pte(v));
}
-static int set_page_presence(unsigned long addr, bool present, bool *old)
+static int clear_page_presence(struct kmmio_fault_page *f, bool clear)
{
unsigned int level;
- pte_t *pte = lookup_address(addr, &level);
+ pte_t *pte = lookup_address(f->page, &level);
if (!pte) {
- pr_err("kmmio: no pte for page 0x%08lx\n", addr);
+ pr_err("kmmio: no pte for page 0x%08lx\n", f->page);
return -1;
}
switch (level) {
case PG_LEVEL_2M:
- set_pmd_presence((pmd_t *)pte, present, old);
+ clear_pmd_presence((pmd_t *)pte, clear, &f->old_presence);
break;
case PG_LEVEL_4K:
- set_pte_presence(pte, present, old);
+ clear_pte_presence(pte, clear, &f->old_presence);
break;
default:
pr_err("kmmio: unexpected page level 0x%x.\n", level);
return -1;
}
- __flush_tlb_one(addr);
+ __flush_tlb_one(f->page);
return 0;
}
@@ -171,9 +173,9 @@ static int arm_kmmio_fault_page(struct kmmio_fault_page *f)
WARN_ONCE(f->armed, KERN_ERR "kmmio page already armed.\n");
if (f->armed) {
pr_warning("kmmio double-arm: page 0x%08lx, ref %d, old %d\n",
- f->page, f->count, f->old_presence);
+ f->page, f->count, !!f->old_presence);
}
- ret = set_page_presence(f->page, false, &f->old_presence);
+ ret = clear_page_presence(f, true);
WARN_ONCE(ret < 0, KERN_ERR "kmmio arming 0x%08lx failed.\n", f->page);
f->armed = true;
return ret;
@@ -182,8 +184,7 @@ static int arm_kmmio_fault_page(struct kmmio_fault_page *f)
/** Restore the given page to saved presence state. */
static void disarm_kmmio_fault_page(struct kmmio_fault_page *f)
{
- bool tmp;
- int ret = set_page_presence(f->page, f->old_presence, &tmp);
+ int ret = clear_page_presence(f, false);
WARN_ONCE(ret < 0,
KERN_ERR "kmmio disarming 0x%08lx failed.\n", f->page);
f->armed = false;
@@ -310,7 +311,12 @@ static int post_kmmio_handler(unsigned long condition, struct pt_regs *regs)
struct kmmio_context *ctx = &get_cpu_var(kmmio_ctx);
if (!ctx->active) {
- pr_debug("kmmio: spurious debug trap on CPU %d.\n",
+ /*
+ * debug traps without an active context are due to either
+ * something external causing them (f.e. using a debugger while
+ * mmio tracing enabled), or erroneous behaviour
+ */
+ pr_warning("kmmio: unexpected debug trap on CPU %d.\n",
smp_processor_id());
goto out;
}
@@ -439,12 +445,12 @@ static void rcu_free_kmmio_fault_pages(struct rcu_head *head)
head,
struct kmmio_delayed_release,
rcu);
- struct kmmio_fault_page *p = dr->release_list;
- while (p) {
- struct kmmio_fault_page *next = p->release_next;
- BUG_ON(p->count);
- kfree(p);
- p = next;
+ struct kmmio_fault_page *f = dr->release_list;
+ while (f) {
+ struct kmmio_fault_page *next = f->release_next;
+ BUG_ON(f->count);
+ kfree(f);
+ f = next;
}
kfree(dr);
}
@@ -453,19 +459,19 @@ static void remove_kmmio_fault_pages(struct rcu_head *head)
{
struct kmmio_delayed_release *dr =
container_of(head, struct kmmio_delayed_release, rcu);
- struct kmmio_fault_page *p = dr->release_list;
+ struct kmmio_fault_page *f = dr->release_list;
struct kmmio_fault_page **prevp = &dr->release_list;
unsigned long flags;
spin_lock_irqsave(&kmmio_lock, flags);
- while (p) {
- if (!p->count) {
- list_del_rcu(&p->list);
- prevp = &p->release_next;
+ while (f) {
+ if (!f->count) {
+ list_del_rcu(&f->list);
+ prevp = &f->release_next;
} else {
- *prevp = p->release_next;
+ *prevp = f->release_next;
}
- p = p->release_next;
+ f = f->release_next;
}
spin_unlock_irqrestore(&kmmio_lock, flags);
@@ -528,8 +534,8 @@ void unregister_kmmio_probe(struct kmmio_probe *p)
}
EXPORT_SYMBOL(unregister_kmmio_probe);
-static int kmmio_die_notifier(struct notifier_block *nb, unsigned long val,
- void *args)
+static int
+kmmio_die_notifier(struct notifier_block *nb, unsigned long val, void *args)
{
struct die_args *arg = args;
@@ -544,11 +550,23 @@ static struct notifier_block nb_die = {
.notifier_call = kmmio_die_notifier
};
-static int __init init_kmmio(void)
+int kmmio_init(void)
{
int i;
+
for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++)
INIT_LIST_HEAD(&kmmio_page_table[i]);
+
return register_die_notifier(&nb_die);
}
-fs_initcall(init_kmmio); /* should be before device_initcall() */
+
+void kmmio_cleanup(void)
+{
+ int i;
+
+ unregister_die_notifier(&nb_die);
+ for (i = 0; i < KMMIO_PAGE_TABLE_SIZE; i++) {
+ WARN_ONCE(!list_empty(&kmmio_page_table[i]),
+ KERN_ERR "kmmio_page_table not empty at cleanup, any further tracing will leak memory.\n");
+ }
+}
diff --git a/arch/x86/mm/mmio-mod.c b/arch/x86/mm/mmio-mod.c
index c9342ed8b40..132772a8ec5 100644
--- a/arch/x86/mm/mmio-mod.c
+++ b/arch/x86/mm/mmio-mod.c
@@ -451,6 +451,7 @@ void enable_mmiotrace(void)
if (nommiotrace)
pr_info(NAME "MMIO tracing disabled.\n");
+ kmmio_init();
enter_uniprocessor();
spin_lock_irq(&trace_lock);
atomic_inc(&mmiotrace_enabled);
@@ -473,6 +474,7 @@ void disable_mmiotrace(void)
clear_trace_list(); /* guarantees: no more kmmio callbacks */
leave_uniprocessor();
+ kmmio_cleanup();
pr_info(NAME "disabled.\n");
out:
mutex_unlock(&mmiotrace_mutex);