Skip to contentSkip to conversation

target-i386: Support migratable=no properly

PATCH Applied Eduardo Habkost 3 participants 5 messages Aug 21, 2014 – Sep 4, 2014

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 target-i386: Support migratable=no properly

When the "migratable" property was implemented, the behavior was tested by changing the default on the code, but actually using the option on the command-line (e.g. "-cpu host,migratable=false") doesn't work as expected. This is a regression for a common use case of "-cpu host", which is to enable features that are supported by the host CPU + kernel before feature-specific code is added to QEMU.

Fix this by initializing the feature words for "-cpu host" on x86_cpu_parse_featurestr(), right after parsing the CPU options.

---
I was considering a more complex (but less hacky) fix, by introducing an enum FeatureBit { OFF, ON, HOST }, making FeatureWord a 32-element FeatureBit array, and initializing the actual feature bits on x86_cpu_realizefn(). But as this is a patch for qemu-stable, I kept the fix as simple as possible.

---
 target-i386/cpu-qom.h |  1 +
 target-i386/cpu.c     | 17 ++++++++++++-----
 2 files changed, 13 insertions(+), 5 deletions(-)
target-i386/cpu-qom.h +1 -0
index 71a1b97..7755466 100644
@@ -92,6 +92,7 @@ typedef struct X86CPU {
     bool enforce_cpuid;
     bool expose_kvm;
     bool migratable;
+    bool host_features;
 
     /* if true the CPUID code directly forward host cache leaves to the guest */
     bool cache_info_passthrough;
target-i386/cpu.c +12 -5
index 6d008ab..c0f8efc 100644
@@ -1318,18 +1318,18 @@ static void host_x86_cpu_initfn(Object *obj)
     X86CPU *cpu = X86_CPU(obj);
     CPUX86State *env = &cpu->env;
     KVMState *s = kvm_state;
-    FeatureWord w;
 
     assert(kvm_enabled());
 
+    /* We can't fill the features array here because we don't know yet if
+     * "migratable" is true or false.
+     */
+    cpu->host_features = true;
+
     env->cpuid_level = kvm_arch_get_supported_cpuid(s, 0x0, 0, R_EAX);
     env->cpuid_xlevel = kvm_arch_get_supported_cpuid(s, 0x80000000, 0, R_EAX);
     env->cpuid_xlevel2 = kvm_arch_get_supported_cpuid(s, 0xC0000000, 0, R_EAX);
 
-    for (w = 0; w < FEATURE_WORDS; w++) {
-        env->features[w] =
-            x86_cpu_get_supported_feature_word(w, cpu->migratable);
-    }
     object_property_set_bool(OBJECT(cpu), true, "pmu", &error_abort);
 }
 
@@ -1828,6 +1828,13 @@ static void x86_cpu_parse_featurestr(CPUState *cs, char *features,
         featurestr = strtok(NULL, ",");
     }
 
+    if (cpu->host_features) {
+        for (w = 0; w < FEATURE_WORDS; w++) {
+            env->features[w] =
+                x86_cpu_get_supported_feature_word(w, cpu->migratable);
+        }
+    }
+
     for (w = 0; w < FEATURE_WORDS; w++) {
         env->features[w] |= plus_features[w];
         env->features[w] &= ~minus_features[w];
Signature · 1 line
1.9.3
#1 lore raw

Forgot to add qemu-devel to "To:". Adding it.

Quoted 79 lines · Eduardo Habkost

When the "migratable" property was implemented, the behavior was tested by changing the default on the code, but actually using the option on the command-line (e.g. "-cpu host,migratable=false") doesn't work as expected. This is a regression for a common use case of "-cpu host", which is to enable features that are supported by the host CPU + kernel before feature-specific code is added to QEMU.

Fix this by initializing the feature words for "-cpu host" on x86_cpu_parse_featurestr(), right after parsing the CPU options.

---
I was considering a more complex (but less hacky) fix, by introducing an enum FeatureBit { OFF, ON, HOST }, making FeatureWord a 32-element FeatureBit array, and initializing the actual feature bits on x86_cpu_realizefn(). But as this is a patch for qemu-stable, I kept the fix as simple as possible.

---
 target-i386/cpu-qom.h |  1 +
 target-i386/cpu.c     | 17 ++++++++++++-----
 2 files changed, 13 insertions(+), 5 deletions(-)
target-i386/cpu-qom.h +1 -0
index 71a1b97..7755466 100644
@@ -92,6 +92,7 @@ typedef struct X86CPU {
     bool enforce_cpuid;
     bool expose_kvm;
     bool migratable;
+    bool host_features;
 
     /* if true the CPUID code directly forward host cache leaves to the guest */
     bool cache_info_passthrough;
target-i386/cpu.c +12 -5
index 6d008ab..c0f8efc 100644
@@ -1318,18 +1318,18 @@ static void host_x86_cpu_initfn(Object *obj)
     X86CPU *cpu = X86_CPU(obj);
     CPUX86State *env = &cpu->env;
     KVMState *s = kvm_state;
-    FeatureWord w;
 
     assert(kvm_enabled());
 
+    /* We can't fill the features array here because we don't know yet if
+     * "migratable" is true or false.
+     */
+    cpu->host_features = true;
+
     env->cpuid_level = kvm_arch_get_supported_cpuid(s, 0x0, 0, R_EAX);
     env->cpuid_xlevel = kvm_arch_get_supported_cpuid(s, 0x80000000, 0, R_EAX);
     env->cpuid_xlevel2 = kvm_arch_get_supported_cpuid(s, 0xC0000000, 0, R_EAX);
 
-    for (w = 0; w < FEATURE_WORDS; w++) {
-        env->features[w] =
-            x86_cpu_get_supported_feature_word(w, cpu->migratable);
-    }
     object_property_set_bool(OBJECT(cpu), true, "pmu", &error_abort);
 }
 
@@ -1828,6 +1828,13 @@ static void x86_cpu_parse_featurestr(CPUState *cs, char *features,
         featurestr = strtok(NULL, ",");
     }
 
+    if (cpu->host_features) {
+        for (w = 0; w < FEATURE_WORDS; w++) {
+            env->features[w] =
+                x86_cpu_get_supported_feature_word(w, cpu->migratable);
+        }
+    }
+
     for (w = 0; w < FEATURE_WORDS; w++) {
         env->features[w] |= plus_features[w];
         env->features[w] &= ~minus_features[w];
Signature · 1 line
1.9.3
Signature · 1 line
Eduardo
Reviewed-by
#2 lore raw

Quoting Eduardo Habkost (2014-08-21 15:14:38)

Quoted 14 lines

Forgot to add qemu-devel to "To:". Adding it.

Quoted 11 lines · Eduardo Habkost

When the "migratable" property was implemented, the behavior was tested by changing the default on the code, but actually using the option on the command-line (e.g. "-cpu host,migratable=false") doesn't work as expected. This is a regression for a common use case of "-cpu host", which is to enable features that are supported by the host CPU + kernel before feature-specific code is added to QEMU.

Fix this by initializing the feature words for "-cpu host" on x86_cpu_parse_featurestr(), right after parsing the CPU options.

Ping for stable 2.1.1, freeze is Wednesday.

Quoted 71 lines
Quoted 68 lines

---
I was considering a more complex (but less hacky) fix, by introducing an enum FeatureBit { OFF, ON, HOST }, making FeatureWord a 32-element FeatureBit array, and initializing the actual feature bits on x86_cpu_realizefn(). But as this is a patch for qemu-stable, I kept the fix as simple as possible.

---
 target-i386/cpu-qom.h |  1 +
 target-i386/cpu.c     | 17 ++++++++++++-----
 2 files changed, 13 insertions(+), 5 deletions(-)
target-i386/cpu-qom.h +1 -0
index 71a1b97..7755466 100644
@@ -92,6 +92,7 @@ typedef struct X86CPU {
     bool enforce_cpuid;
     bool expose_kvm;
     bool migratable;
+    bool host_features;
 
     /* if true the CPUID code directly forward host cache leaves to the guest */
     bool cache_info_passthrough;
target-i386/cpu.c +12 -5
index 6d008ab..c0f8efc 100644
@@ -1318,18 +1318,18 @@ static void host_x86_cpu_initfn(Object *obj)
     X86CPU *cpu = X86_CPU(obj);
     CPUX86State *env = &cpu->env;
     KVMState *s = kvm_state;
-    FeatureWord w;
 
     assert(kvm_enabled());
 
+    /* We can't fill the features array here because we don't know yet if
+     * "migratable" is true or false.
+     */
+    cpu->host_features = true;
+
     env->cpuid_level = kvm_arch_get_supported_cpuid(s, 0x0, 0, R_EAX);
     env->cpuid_xlevel = kvm_arch_get_supported_cpuid(s, 0x80000000, 0, R_EAX);
     env->cpuid_xlevel2 = kvm_arch_get_supported_cpuid(s, 0xC0000000, 0, R_EAX);
 
-    for (w = 0; w < FEATURE_WORDS; w++) {
-        env->features[w] =
-            x86_cpu_get_supported_feature_word(w, cpu->migratable);
-    }
     object_property_set_bool(OBJECT(cpu), true, "pmu", &error_abort);
 }
 
@@ -1828,6 +1828,13 @@ static void x86_cpu_parse_featurestr(CPUState *cs, char *features,
         featurestr = strtok(NULL, ",");
     }
 
+    if (cpu->host_features) {
+        for (w = 0; w < FEATURE_WORDS; w++) {
+            env->features[w] =
+                x86_cpu_get_supported_feature_word(w, cpu->migratable);
+        }
+    }
+
     for (w = 0; w < FEATURE_WORDS; w++) {
         env->features[w] |= plus_features[w];
         env->features[w] &= ~minus_features[w];
Signature · 1 line
1.9.3
Signature · 1 line
Eduardo
#3 lore raw
Quoted 11 lines · Michael Roth

Quoting Eduardo Habkost (2014-08-21 15:14:38)

Quoted 10 lines

Forgot to add qemu-devel to "To:". Adding it.

Quoted 7 lines · Eduardo Habkost

When the "migratable" property was implemented, the behavior was tested by changing the default on the code, but actually using the option on the command-line (e.g. "-cpu host,migratable=false") doesn't work as expected. This is a regression for a common use case of "-cpu host", which is to enable features that are supported by the host CPU + kernel before feature-specific code is added to QEMU.

Quoted 8 lines
Quoted 4 lines
Quoted 4 lines

Fix this by initializing the feature words for "-cpu host" on x86_cpu_parse_featurestr(), right after parsing the CPU options.

Ping for stable 2.1.1, freeze is Wednesday.

Ping? Who can help me get this applied?

I was expecting to need to argue to get this considered as -stable material, but not to have trouble getting it applied to master...

Signature · 1 line
Eduardo
#4 lore raw

Am 04.09.2014 17:19, schrieb Eduardo Habkost:

Quoted 17 lines · Michael Roth
Quoted 16 lines

Quoting Eduardo Habkost (2014-08-21 15:14:38)

Quoted 14 lines

Forgot to add qemu-devel to "To:". Adding it.

Quoted 11 lines · Eduardo Habkost

When the "migratable" property was implemented, the behavior was tested by changing the default on the code, but actually using the option on the command-line (e.g. "-cpu host,migratable=false") doesn't work as expected. This is a regression for a common use case of "-cpu host", which is to enable features that are supported by the host CPU + kernel before feature-specific code is added to QEMU.

Fix this by initializing the feature words for "-cpu host" on x86_cpu_parse_featurestr(), right after parsing the CPU options.

Quoted 8 lines
Quoted 3 lines

Ping for stable 2.1.1, freeze is Wednesday.

Ping? Who can help me get this applied?

I was expecting to need to argue to get this considered as -stable material, but not to have trouble getting it applied to master...

Thanks, applied to qom-cpu:
https://github.com/afaerber/qemu-cpu/commits/qom-cpu

Andreas

Signature · 2 lines
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg