tcg/i386: Do not display HT warning for TCG
RFC PATCH 1/1 tcg/i386: Do not display HT warning for TCG
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding CPUID_HT bit to the TCG features and explicitly checking this bit in the cpuid.
--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
target/i386/cpu.c +5 -5
@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \
@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs); -/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning.
@@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { -error_report ("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) { +error_report ("This CPU doesn't support hyperthreading. Please configure" " -smp options properly." ); ht_warned = true; }
Signature · 1 line
2.11.0
PATCH 1/1 tcg/i386: Do not display HT warning for TCG
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding CPUID_HT bit to the TCG features and explicitly checking this bit in the cpuid.
--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
target/i386/cpu.c +5 -5
@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \
@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs); -/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning.
@@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { -error_report ("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) { +error_report ("This CPU doesn't support hyperthreading. Please configure" " -smp options properly." ); ht_warned = true; }
Signature · 1 line
2.11.0
Quoted 41 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +3
-3@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs); -/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning.
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 6 lines
target/i386/cpu.c +1
-2@@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {
I missed a '!' here. We need to warn if CPUID_HT is not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
Signature · 1 line
Pranith
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);
Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HT is automatically set on the CPU if (nr_threads * nr_cores) > 1. See the "case 1:" block in cpu_x86_cpuid().
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
Signature · 1 line
Eduardo
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().
Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Signature · 1 line
Pranith
Quoted 65 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
Quoted 3 lines
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
Signature · 1 line
Eduardo
Quoted 66 lines · Eduardo Habkost
Quoted 65 lines · Pranith Kumar
Quoted 63 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
Quoted 7 lines
Quoted 3 lines
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
Quoted 4 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus. So we can either:
* Change the vendor to Intel for TCG, or
* Warn based on the HT bit being set
Thanks,
Signature · 1 line
Pranith
Quoted 71 lines · Pranith Kumar
Quoted 70 lines · Eduardo Habkost
Quoted 68 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Quoted 6 lines
Quoted 3 lines
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
I don't know why the default on TCG is AMD, either.
Quoted 11 lines
Quoted 5 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus.
Is Windows XP able to detect 2 threads * 2 cores even if CPUID vendor is AMD?
Quoted 4 lines
So we can either:
* Change the vendor to Intel for TCG, or
* Warn based on the HT bit being set
Could you clarify what you mean by the second item? The HT bit is always set when nr_cores*nr_threads > 1.
Signature · 1 line
Eduardo
Quoted 87 lines · Eduardo Habkost
Quoted 71 lines · Pranith Kumar
Quoted 70 lines · Eduardo Habkost
Quoted 68 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Quoted 6 lines
Quoted 3 lines
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
I don't know why the default on TCG is AMD, either.
Quoted 6 lines
Quoted 5 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
Quoted 9 lines
Quoted 5 lines
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus.
Is Windows XP able to detect 2 threads * 2 cores even if CPUID vendor is AMD?
Yep. I don't think it cares for vendor id.
Quoted 7 lines
Quoted 4 lines
So we can either:
* Change the vendor to Intel for TCG, or
* Warn based on the HT bit being setCould you clarify what you mean by the second item? The HT bit is always set when nr_cores*nr_threads > 1.
Hmm, I see it now. We just warn the user if the host processor does not support HT.
So how about setting the HT bit only if the host processor supports it? Or since it apparently does not matter anyway, remove the warning entirely?
Signature · 1 line
Pranith
Quoted 94 lines · Pranith Kumar
Quoted 93 lines · Eduardo Habkost
Quoted 71 lines · Pranith Kumar
Quoted 70 lines · Eduardo Habkost
Quoted 68 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Quoted 6 lines
Quoted 3 lines
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
I don't know why the default on TCG is AMD, either.
Quoted 11 lines
Quoted 5 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus.
Quoted 5 lines
Quoted 3 lines
Is Windows XP able to detect 2 threads * 2 cores even if CPUID vendor is AMD?
Yep. I don't think it cares for vendor id.
What happens if you use -smp 4,cores=4 ?
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines
So we can either:
* Change the vendor to Intel for TCG, or
* Warn based on the HT bit being setCould you clarify what you mean by the second item? The HT bit is always set when nr_cores*nr_threads > 1.
Hmm, I see it now. We just warn the user if the host processor does not support HT.
We don't look at the host processor at all: the warn depends only on the CPUID data we're showing to the guest.
Quoted 4 lines
So how about setting the HT bit only if the host processor supports it? Or since it apparently does not matter anyway, remove the warning entirely?
The host processor doesn't matter. What matters is the CPUID data seen by the guest, and it doesn't matter if we're using TCG or KVM.
Linux guests seem to ignore the hyperthreading information if vendor ID is AMD, and that seems to be the original reason for the warning. I would like to remove the warning, too, but it looks like this fact hasn't changed.
Signature · 1 line
Eduardo
Quoted 95 lines · Eduardo Habkost
Quoted 94 lines · Pranith Kumar
Quoted 93 lines · Eduardo Habkost
Quoted 71 lines · Pranith Kumar
Quoted 70 lines · Eduardo Habkost
Quoted 68 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Quoted 6 lines
Quoted 3 lines
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
I don't know why the default on TCG is AMD, either.
Quoted 11 lines
Quoted 5 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus.
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
Is Windows XP able to detect 2 threads * 2 cores even if CPUID vendor is AMD?
Yep. I don't think it cares for vendor id.
What happens if you use -smp 4,cores=4 ?
This is interesting. The above shows 4 cores in the guest. I then tried various combinations:
-smp 4 shows 2 cores
-smp 4,cores=4 shows 4 cores
-smp 4,threads=4 shows 4 cores
-smp 4,sockets=1 shows 4 cores
I am a bit confused why smp 4 is showing only 2 cores? and why the rest are showing 4 cores
Quoted 29 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
So we can either:
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
* Change the vendor to Intel for TCG, or
* Warn based on the HT bit being setCould you clarify what you mean by the second item? The HT bit is always set when nr_cores*nr_threads > 1.
Hmm, I see it now. We just warn the user if the host processor does not support HT.
We don't look at the host processor at all: the warn depends only on the CPUID data we're showing to the guest.
Quoted 4 lines
So how about setting the HT bit only if the host processor supports it? Or since it apparently does not matter anyway, remove the warning entirely?
The host processor doesn't matter. What matters is the CPUID data seen by the guest, and it doesn't matter if we're using TCG or KVM.
Linux guests seem to ignore the hyperthreading information if vendor ID is AMD, and that seems to be the original reason for the warning. I would like to remove the warning, too, but it looks like this fact hasn't changed.
Signature · 1 line
Pranith
Quoted 100 lines · Pranith Kumar
Quoted 99 lines · Eduardo Habkost
Quoted 98 lines · Pranith Kumar
Quoted 96 lines · Eduardo Habkost
Quoted 71 lines · Pranith Kumar
Quoted 70 lines · Eduardo Habkost
Quoted 68 lines · Pranith Kumar
Quoted 58 lines · Eduardo Habkost
Quoted 36 lines · Pranith Kumar
Quoted 35 lines · Pranith Kumar
When we enable hyperthreading (using threads smp argument), we warn the user if the cpu is an AMD cpu. This does not make sense on TCG and is also obsolete now that AMD Ryzen support hyperthreading.
Fix this by adding
CPUID_HTbit to the TCG features and explicitly checking this bit in the cpuid.Signed-off-by Pranith Kumar <[email protected]>--- target/i386/cpu.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)target/i386/cpu.c +2
-2@@ -202,12 +202,12 @@ static void x86_cpu_vendor_words2str(char *dst, uint32_t vendor1,#define TCG_FEATURES (CPUID_FP87 | CPUID_PSE | CPUID_TSC | CPUID_MSR | \ CPUID_PAE | CPUID_MCE | CPUID_CX8 | CPUID_APIC | CPUID_SEP | \ CPUID_MTRR | CPUID_PGE | CPUID_MCA | CPUID_CMOV | CPUID_PAT | \ - CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | \ + CPUID_PSE36 | CPUID_CLFLUSH | CPUID_ACPI | CPUID_MMX | CPUID_HT | \ CPUID_FXSR | CPUID_SSE | CPUID_SSE2 | CPUID_SS | CPUID_DE)/* partly implemented: CPUID_MTRR, CPUID_MCA, CPUID_CLFLUSH (needed for Win64) */ /* missing: -CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_HT, CPUID_TM, CPUID_PBE */ +CPUID_VME, CPUID_DTS, CPUID_SS, CPUID_TM, CPUID_PBE */ #define TCG_EXT_FEATURES (CPUID_EXT_SSE3 | CPUID_EXT_PCLMULQDQ | \ CPUID_EXT_MONITOR | CPUID_EXT_SSSE3 | CPUID_EXT_CX16 | \ CPUID_EXT_SSE41 | CPUID_EXT_SSE42 | CPUID_EXT_POPCNT | \@@ -3643,7 +3643,7 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)qemu_init_vcpu (cs);Quoted 16 lines
target/i386/cpu.c static void x86_cpu_realizefn(DeviceState *dev, Error **errp) · line 3654
Quoted 12 lines
target/i386/cpu.c +2
-3-/* Only Intel CPUs support hyperthreading. Even though QEMU fixes this +/* Only some CPUs support hyperthreading. Even though QEMU fixes this * issue by adjusting CPUID_0000_0001_EBX and CPUID_8000_0008_ECX * based on inputs (sockets,cores,threads), it is still better to gives * users a warning. @@ -3651,8 +3651,8 @@ static void x86_cpu_realizefn(DeviceState *dev, Error **errp)* NOTE: the following code has to follow qemu_init_vcpu(). Otherwise * cs->nr_threads hasn't be populated yet and the checking is incorrect. */ -if (!IS_INTEL_CPU (env) && cs->nr_threads >1 && !ht_warned) { - error_report("AMD CPU doesn't support hyperthreading. Please configure" +if ((env->features[FEAT_1_EDX ] &CPUID_HT ) && (cs->nr_threads >1 ) && !ht_warned) {I missed a '!' here. We need to warn if
CPUID_HTis not set. But I see that it is not being set even on HT enabled processors (i7-3770). How do I test for it?
CPUID_HTis automatically set on the CPU if (nr_threads*nr_cores) > 1. See the "case 1:" block incpu_x86_cpuid().Quoted 5 lines
AFAIK, the point of the warning is to let the user know that the guest OS is likely to ignore thread topology information if CPUID vendor is not Intel, and has nothing to do with TCG or KVM capabilities. Maybe the warning is obsolete today and guests won't be confused by a HT AMD CPU, but we need to confirm that.
We assume an AMD cpu for x86 TCG and it prints this warning when an smp guest is run with (cores=2,threads=2) argument. The main point of this patch is to remove that warning when using TCG.
Quoted 6 lines
Quoted 3 lines
What is different under TCG? If guests were confused with threads=2 + vendor=AMD using KVM, what exactly would make them not confused by threads=2 + vendor=AMD when using TCG?
I am not sure why we chose AMD to be the vendor for TCG since, as you note, it does not really matter if it's AMD or Intel in TCG mode.
I don't know why the default on TCG is AMD, either.
Quoted 11 lines
Quoted 5 lines
I just ran a Fedora 25 guest using "-machine accel=tcg -smp 2,threads=2", and it still thinks it's running on a 2-core CPU instead of a 2-thread CPU when CPU vendor is AMD.
This all started with using Windows XP. It has an arbitrary limit on the number of cores it can support to 2 cores. So using -smp 4 will still only show 2 cores. However, we can overcome this by using -smp cores=2,threads=2 and 4 cores show up in the guest. But we see the ugly warning that HT is not supported on AMD cpus.
Is Windows XP able to detect 2 threads * 2 cores even if CPUID vendor is AMD?
Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Yep. I don't think it cares for vendor id.
What happens if you use -smp 4,cores=4 ?
This is interesting. The above shows 4 cores in the guest. I then tried various combinations:
-smp 4 shows 2 cores
This is a shortcut for 4 sockets, 1 core per socket, 1 thread per core.
Quoted 1 line
-smp 4,cores=4 shows 4 cores
This is a shortcut for 1 socket with 4 cores, 1 thread per core.
Quoted 1 line
-smp 4,threads=4 shows 4 cores
This means 1 socket with 1 core, 4 threads per core. But if vendor ID is AMD, guests will probably ignore the hyperthreading info and treat each thread as a separate core. That's why we have a warning.
Quoted 1 line
-smp 4,sockets=1 shows 4 cores
This is a shortcut for 1 socket with 4 cores, 1 thread per core.
Quoted 3 lines
I am a bit confused why smp 4 is showing only 2 cores? and why the rest are showing 4 cores
Probably because Windows XP is limited to 2 sockets.
Signature · 1 line
Eduardo
Quoted 5 lines · Eduardo Habkost
Quoted 1 line
-smp 4,threads=4 shows 4 cores
This means 1 socket with 1 core, 4 threads per core. But if vendor ID is AMD, guests will probably ignore the hyperthreading info and treat each thread as a separate core. That's why we have a warning.
I'm not sure about the "probably". As long as the HT bit is 1 and the number of threads is > 1 in CPUID[EAX=1].EBX[bits 23:16], I am not sure why the OS would care about the vendor.
x86info does care (it just does not try to detect the topology for !Intel processors), but Linux seems not to. detect_ht is called unconditionally.
Quoted 3 lines
Quoted 2 lines
I am a bit confused why smp 4 is showing only 2 cores? and why the rest are showing 4 cores
Probably because Windows XP is limited to 2 sockets.
This is correct.
Paolo
Quoted 4 lines · Paolo Bonzini
Quoted 1 line · Eduardo Habkost
Quoted 1 line
-smp 4,threads=4 shows 4 cores
Quoted 9 lines
Quoted 4 lines
This means 1 socket with 1 core, 4 threads per core. But if vendor ID is AMD, guests will probably ignore the hyperthreading info and treat each thread as a separate core. That's why we have a warning.
I'm not sure about the "probably". As long as the HT bit is 1 and the number of threads is > 1 in CPUID[EAX=1].EBX[bits 23:16], I am not sure why the OS would care about the vendor.
On my testing, "-cpu qemu64,vendor=AuthenticAMD" and "-cpu qemu64,vendor=GenuineIntel" gave different results on lscpu and /proc/cpuinfo in a Fedora 25 guest (Linux 4.8.6-300.fc25).
Quoted 3 lines
x86info does care (it just does not try to detect the topology for !Intel processors), but Linux seems not to.
detect_htis called unconditionally.
I see detect_ht() inside a "#ifdef CONFIG_X86_32" in amd_cpu_dev.c_init (init_amd()), and intel_cpudev.c_init (init_intel()). It is inside a "#ifdef CONFIG_X86_64" in identify_cpu(), though, so it looks like it is called at different places, but always called.
But the logic becomes very confusing after that: the core/sibling CPU masks are initialized based on phys_proc_id and cpu_core_id, and detect_ht() isn't the only place where those fields are initialized (see amd_get_topology(), amd_detect_cmp(), detect_extended_topology()). It looks like amd_detect_cmp() and/or amd_get_topology() will undo what detect_ht() did.
Signature · 1 line
Eduardo