s390x/mmu: Use ioctl for reading and writing from/to guest memory

Add code to make use of the new ioctl for reading from / writing to
virtual guest memory. By using the ioctl, the memory accesses are now
protected with the so-called ipte-lock in the kernel.

[CH: moved error message into kvm_s390_mem_op()]
Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>

3 files changed, 54 insertions, 0 deletions

diff --git a/target-s390x/cpu.h b/target-s390x/cpu.h
index 79bc80b58..9c4274325 100644
--- a/target-s390x/cpu.h
+++ b/target-s390x/cpu.h
@@ -401,6 +401,8 @@ void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq);
 void kvm_s390_floating_interrupt(struct kvm_s390_irq *irq);
 int kvm_s390_inject_flic(struct kvm_s390_irq *irq);
 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code);
+int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, void *hostbuf, int len,
+                    bool is_write);
 int kvm_s390_get_clock(uint8_t *tod_high, uint64_t *tod_clock);
 int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_clock);
 #else
@@ -418,6 +420,11 @@ static inline int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
 {
     return -ENOSYS;
 }
+static inline int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, void *hostbuf,
+                                  int len, bool is_write)
+{
+    return -ENOSYS;
+}
 static inline void kvm_s390_access_exception(S390CPU *cpu, uint16_t code,
                                              uint64_t te_code)
 {
diff --git a/target-s390x/kvm.c b/target-s390x/kvm.c
index 619684b9e..1c0e78c64 100644
--- a/target-s390x/kvm.c
+++ b/target-s390x/kvm.c
@@ -123,6 +123,7 @@ const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
 
 static int cap_sync_regs;
 static int cap_async_pf;
+static int cap_mem_op;
 
 static void *legacy_s390_alloc(size_t size, uint64_t *align);
 
@@ -247,6 +248,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
 {
     cap_sync_regs = kvm_check_extension(s, KVM_CAP_SYNC_REGS);
     cap_async_pf = kvm_check_extension(s, KVM_CAP_ASYNC_PF);
+    cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
 
     kvm_s390_enable_cmma(s);
 
@@ -550,6 +552,44 @@ int kvm_s390_set_clock(uint8_t *tod_high, uint64_t *tod_low)
     return kvm_vm_ioctl(kvm_state, KVM_SET_DEVICE_ATTR, &attr);
 }
 
+/**
+ * kvm_s390_mem_op:
+ * @addr:      the logical start address in guest memory
+ * @hostbuf:   buffer in host memory. NULL = do only checks w/o copying
+ * @len:       length that should be transfered
+ * @is_write:  true = write, false = read
+ * Returns:    0 on success, non-zero if an exception or error occured
+ *
+ * Use KVM ioctl to read/write from/to guest memory. An access exception
+ * is injected into the vCPU in case of translation errors.
+ */
+int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, void *hostbuf, int len,
+                    bool is_write)
+{
+    struct kvm_s390_mem_op mem_op = {
+        .gaddr = addr,
+        .flags = KVM_S390_MEMOP_F_INJECT_EXCEPTION,
+        .size = len,
+        .op = is_write ? KVM_S390_MEMOP_LOGICAL_WRITE
+                       : KVM_S390_MEMOP_LOGICAL_READ,
+        .buf = (uint64_t)hostbuf,
+    };
+    int ret;
+
+    if (!cap_mem_op) {
+        return -ENOSYS;
+    }
+    if (!hostbuf) {
+        mem_op.flags |= KVM_S390_MEMOP_F_CHECK_ONLY;
+    }
+
+    ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
+    if (ret < 0) {
+        error_printf("KVM_S390_MEM_OP failed: %s\n", strerror(-ret));
+    }
+    return ret;
+}
+
 /*
  * Legacy layout for s390:
  * Older S390 KVM requires the topmost vma of the RAM to be
diff --git a/target-s390x/mmu_helper.c b/target-s390x/mmu_helper.c
index 9b88498b3..cd2cb5162 100644
--- a/target-s390x/mmu_helper.c
+++ b/target-s390x/mmu_helper.c
@@ -450,6 +450,13 @@ int s390_cpu_virt_mem_rw(S390CPU *cpu, vaddr laddr, void *hostbuf,
     target_ulong *pages;
     int ret;
 
+    if (kvm_enabled()) {
+        ret = kvm_s390_mem_op(cpu, laddr, hostbuf, len, is_write);
+        if (ret >= 0) {
+            return ret;
+        }
+    }
+
     nr_pages = (((laddr & ~TARGET_PAGE_MASK) + len - 1) >> TARGET_PAGE_BITS)
                + 1;
     pages = g_malloc(nr_pages * sizeof(*pages));