Skip to contentSkip to conversation

Add support for async page fault to qemu

PATCH Superseded Gleb Natapov 4 participants 6 messages Oct 18, 2010 – Oct 21, 2010 KVM

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Gleb Natapov

author
#0 lore raw

PATCH 1/1 Add support for async page fault to qemu

Add save/restore of MSR for migration and cpuid bit.

Signed-off-by Gleb Natapov <[email protected]>
qemu-kvm-x86.c +5 -0
index bb09fd8..5d8c428 100644
@@ -678,6 +678,9 @@ static int get_msr_entry(struct kvm_msr_entry *entry, CPUState *env)
         env->mcg_ctl = entry->data;
         break;
 #endif
+    case MSR_KVM_ASYNC_PF_EN:
+        env->async_pf_en_msr = entry->data;
+        break;
     default:
 #ifdef KVM_CAP_MCE
         if (entry->index >= MSR_MC0_CTL &&

@@ -967,6 +970,7 @@ void kvm_arch_load_regs(CPUState *env, int level)
         }
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME, env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {

@@ -1186,6 +1190,7 @@ void kvm_arch_save_regs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {
target-i386/cpu.h +2 -1
index 8b6efed..154b76b 100644
@@ -669,6 +669,7 @@ typedef struct CPUX86State {
 #endif
     uint64_t system_time_msr;
     uint64_t wall_clock_msr;
+    uint64_t async_pf_en_msr;
 
     uint64_t tsc;
 

@@ -923,7 +924,7 @@ CPUState *pc_new_cpu(const char *cpu_model);
 #define cpu_list_id x86_cpu_list
 #define cpudef_setup	x86_cpudef_setup
 
-#define CPU_SAVE_VERSION 12
+#define CPU_SAVE_VERSION 13
 
 /* MMU modes definitions */
 #define MMU_MODE0_SUFFIX _kernel
target-i386/cpuid.c +1 -1
index d63fdcb..0ee1f88 100644
@@ -73,7 +73,7 @@ static const char *ext3_feature_name[] = {
 };
 
 static const char *kvm_feature_name[] = {
-    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, NULL, NULL, NULL, NULL,
+    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, "kvm_asyncpf", NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
target-i386/kvm.c +8 -0
index f4fc063..0eb1e90 100644
@@ -151,6 +151,9 @@ struct kvm_para_features {
 #ifdef KVM_CAP_PV_MMU
         { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 #endif
+#ifdef KVM_CAP_ASYNC_PF
+        { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
+#endif
         { -1, -1 }
 };
 

@@ -672,6 +675,7 @@ static int kvm_put_msrs(CPUState *env, int level)
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
                           env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 
     msr_data.info.nmsrs = n;

@@ -880,6 +884,7 @@ static int kvm_get_msrs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
     msr_data.info.nmsrs = n;
     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);

@@ -926,6 +931,9 @@ static int kvm_get_msrs(CPUState *env)
         case MSR_VM_HSAVE_PA:
             env->vm_hsave = msrs[i].data;
             break;
+	case MSR_KVM_ASYNC_PF_EN:
+            env->async_pf_en_msr = msrs[i].data;
+            break;
         }
     }
target-i386/machine.c +3 -0
index 4398801..092c901 100644
@@ -474,6 +474,9 @@ static const VMStateDescription vmstate_cpu = {
         VMSTATE_UINT64_V(xcr0, CPUState, 12),
         VMSTATE_UINT64_V(xstate_bv, CPUState, 12),
         VMSTATE_YMMH_REGS_VARS(ymmh_regs, CPUState, CPU_NB_REGS, 12),
+
+	/* KVM async pf msr */
+        VMSTATE_UINT64_V(async_pf_en_msr, CPUState, 13),
         VMSTATE_END_OF_LIST()
         /* The above list is not sorted /wrt version numbers, watch out! */
     }

--
Gleb.

#1 lore raw
Quoted 1 line · Gleb Natapov

Add save/restore of MSR for migration and cpuid bit.

It is there a way to test if async page faults are in use? if so, we can add a subsection instead of changing the cpuversion.

I think that at some point we are going to need a bitmap that indicates what MSR's have been used or something like that.

What do you think?

Later, Juan.

#2 lore raw
Quoted 4 lines · Juan Quintela
Quoted 1 line · Gleb Natapov

Add save/restore of MSR for migration and cpuid bit.

It is there a way to test if async page faults are in use?

Yes, msr != 0 -> need a subsection. Good idea.

Quoted 6 lines

if so, we can add a subsection instead of changing the cpuversion.

I think that at some point we are going to need a bitmap that indicates what MSR's have been used or something like that.

What do you think?

We just need to check if an msr is different from its default value (which we can get by reading msrs immediately after the initial reset).

Currently the reset code assumes msr reset value is zero, that's wrong.

Signature · 1 line
error compiling committee.c: too many arguments to function
author
#3 lore raw
In reply to Juan Quintela
Quoted 6 lines · Juan Quintela
Quoted 1 line · Gleb Natapov

Add save/restore of MSR for migration and cpuid bit.

It is there a way to test if async page faults are in use? if so, we can add a subsection instead of changing the cpuversion.

Yeah. Good idea. Forgot about our cool new subsection feature.

--
Gleb.

author
#4 lore raw

PATCH v2 1/1 Add support for async page fault to qemu

In reply to Gleb Natapov

Add save/restore of MSR for migration and cpuid bit.

Signed-off-by Gleb Natapov <[email protected]>

---
v1->v2
- use vmstate subsection to migrate new msr.

qemu-kvm-x86.c +5 -0
index 59aacd0..145c863 100644
@@ -678,6 +678,9 @@ static int get_msr_entry(struct kvm_msr_entry *entry, CPUState *env)
         env->mcg_ctl = entry->data;
         break;
 #endif
+    case MSR_KVM_ASYNC_PF_EN:
+        env->async_pf_en_msr = entry->data;
+        break;
     default:
 #ifdef KVM_CAP_MCE
         if (entry->index >= MSR_MC0_CTL &&

@@ -967,6 +970,7 @@ void kvm_arch_load_regs(CPUState *env, int level)
         }
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME, env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {

@@ -1186,6 +1190,7 @@ void kvm_arch_save_regs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {
target-i386/cpu.h +1 -0
index 8b6efed..6d1d6a0 100644
@@ -669,6 +669,7 @@ typedef struct CPUX86State {
 #endif
     uint64_t system_time_msr;
     uint64_t wall_clock_msr;
+    uint64_t async_pf_en_msr;
 
     uint64_t tsc;
target-i386/cpuid.c +1 -1
index d63fdcb..0ee1f88 100644
@@ -73,7 +73,7 @@ static const char *ext3_feature_name[] = {
 };
 
 static const char *kvm_feature_name[] = {
-    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, NULL, NULL, NULL, NULL,
+    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, "kvm_asyncpf", NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
target-i386/kvm.c +8 -0
index f4fc063..0eb1e90 100644
@@ -151,6 +151,9 @@ struct kvm_para_features {
 #ifdef KVM_CAP_PV_MMU
         { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 #endif
+#ifdef KVM_CAP_ASYNC_PF
+        { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
+#endif
         { -1, -1 }
 };
 

@@ -672,6 +675,7 @@ static int kvm_put_msrs(CPUState *env, int level)
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
                           env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 
     msr_data.info.nmsrs = n;

@@ -880,6 +884,7 @@ static int kvm_get_msrs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
     msr_data.info.nmsrs = n;
     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);

@@ -926,6 +931,9 @@ static int kvm_get_msrs(CPUState *env)
         case MSR_VM_HSAVE_PA:
             env->vm_hsave = msrs[i].data;
             break;
+	case MSR_KVM_ASYNC_PF_EN:
+            env->async_pf_en_msr = msrs[i].data;
+            break;
         }
     }
target-i386/machine.c +26 -0
index 4398801..bf14067 100644
@@ -373,6 +373,24 @@ static int cpu_post_load(void *opaque, int version_id)
     return 0;
 }
 
+static bool async_pf_msr_needed(void *opaque)
+{
+    CPUState *cpu = opaque;
+
+    return cpu->async_pf_en_msr != 0;
+}
+
+static const VMStateDescription vmstate_async_pf_msr = {
+    .name = "cpu/async_pf_msr",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .minimum_version_id_old = 1,
+    .fields      = (VMStateField []) {
+        VMSTATE_UINT64(async_pf_en_msr, CPUState),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 static const VMStateDescription vmstate_cpu = {
     .name = "cpu",
     .version_id = CPU_SAVE_VERSION,

@@ -476,6 +494,14 @@ static const VMStateDescription vmstate_cpu = {
         VMSTATE_YMMH_REGS_VARS(ymmh_regs, CPUState, CPU_NB_REGS, 12),
         VMSTATE_END_OF_LIST()
         /* The above list is not sorted /wrt version numbers, watch out! */
+    },
+    .subsections = (VMStateSubsection []) {
+        {
+            .vmsd = &vmstate_async_pf_msr,
+            .needed = async_pf_msr_needed,
+        } , {
+            /* empty */
+        }
     }
 };

--
Gleb.

#5 lore raw
Quoted 93 lines · Gleb Natapov

Add save/restore of MSR for migration and cpuid bit.

Signed-off-by Gleb Natapov <[email protected]>

---
v1->v2
- use vmstate subsection to migrate new msr.

qemu-kvm-x86.c +5 -0
index 59aacd0..145c863 100644
@@ -678,6 +678,9 @@ static int get_msr_entry(struct kvm_msr_entry *entry, CPUState *env)
         env->mcg_ctl = entry->data;
         break;
 #endif
+    case MSR_KVM_ASYNC_PF_EN:
+        env->async_pf_en_msr = entry->data;
+        break;
     default:
 #ifdef KVM_CAP_MCE
         if (entry->index >= MSR_MC0_CTL &&

@@ -967,6 +970,7 @@ void kvm_arch_load_regs(CPUState *env, int level)
         }
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME, env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {

@@ -1186,6 +1190,7 @@ void kvm_arch_save_regs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
 #ifdef KVM_CAP_MCE
     if (env->mcg_cap) {
target-i386/cpu.h +1 -0
index 8b6efed..6d1d6a0 100644
@@ -669,6 +669,7 @@ typedef struct CPUX86State {
 #endif
     uint64_t system_time_msr;
     uint64_t wall_clock_msr;
+    uint64_t async_pf_en_msr;
 
     uint64_t tsc;
target-i386/cpuid.c +1 -1
index d63fdcb..0ee1f88 100644
@@ -73,7 +73,7 @@ static const char *ext3_feature_name[] = {
 };
 
 static const char *kvm_feature_name[] = {
-    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, NULL, NULL, NULL, NULL,
+    "kvmclock", "kvm_nopiodelay", "kvm_mmu", NULL, "kvm_asyncpf", NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
target-i386/kvm.c +5 -0
index f4fc063..0eb1e90 100644
@@ -151,6 +151,9 @@ struct kvm_para_features {
 #ifdef KVM_CAP_PV_MMU
         { KVM_CAP_PV_MMU, KVM_FEATURE_MMU_OP },
 #endif
+#ifdef KVM_CAP_ASYNC_PF
+        { KVM_CAP_ASYNC_PF, KVM_FEATURE_ASYNC_PF },
+#endif
         { -1, -1 }
 };
 

@@ -672,6 +675,7 @@ static int kvm_put_msrs(CPUState *env, int level)
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_SYSTEM_TIME,
                           env->system_time_msr);
         kvm_msr_entry_set(&msrs[n++], MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
+        kvm_msr_entry_set(&msrs[n++], MSR_KVM_ASYNC_PF_EN, env->async_pf_en_msr);
     }
 
     msr_data.info.nmsrs = n;

@@ -880,6 +884,7 @@ static int kvm_get_msrs(CPUState *env)
 #endif
     msrs[n++].index = MSR_KVM_SYSTEM_TIME;
     msrs[n++].index = MSR_KVM_WALL_CLOCK;
+    msrs[n++].index = MSR_KVM_ASYNC_PF_EN;
 
     msr_data.info.nmsrs = n;
     ret = kvm_vcpu_ioctl(env, KVM_GET_MSRS, &msr_data);

target-i386/kvm.c static int kvm_get_msrs(CPUState *env) · line 937

Quoted 7 lines
target-i386/kvm.c +3 -0
@@ -926,6 +931,9 @@ static int kvm_get_msrs(CPUState *env)
         case MSR_VM_HSAVE_PA:
             env->vm_hsave = msrs[i].data;
             break;
+	case MSR_KVM_ASYNC_PF_EN:
+            env->async_pf_en_msr = msrs[i].data;
+            break;
         }

I think this is going to break the build if MSR_KVM_ASYNC_PF_EN is not defined.

Please regenerate against uq/master.