| Message ID | 20240604000222.75065-3-zide.chen@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | improve -overcommit cpu-pm=on|off | expand |
On Mon, Jun 03, 2024 at 05:02:22PM -0700, Zide Chen wrote: > Date: Mon, 3 Jun 2024 17:02:22 -0700 > From: Zide Chen <zide.chen@intel.com> > Subject: [PATCH V3 2/2] target/i386: Advertise MWAIT iff host supports > X-Mailer: git-send-email 2.34.1 > > host_cpu_realizefn() sets CPUID_EXT_MONITOR without consulting host/KVM > capabilities. This may cause problems: > > - If MWAIT/MONITOR is not available on the host, advertising this > feature to the guest and executing MWAIT/MONITOR from the guest > triggers #UD and the guest doesn't boot. This is because typically > #UD takes priority over VM-Exit interception checks and KVM doesn't > emulate MONITOR/MWAIT on #UD. > > - If KVM doesn't support KVM_X86_DISABLE_EXITS_MWAIT, MWAIT/MONITOR > from the guest are intercepted by KVM, which is not what cpu-pm=on > intends to do. > > In these cases, MWAIT/MONITOR should not be exposed to the guest. > > The logic in kvm_arch_get_supported_cpuid() to handle CPUID_EXT_MONITOR > is correct and sufficient, and we can't set CPUID_EXT_MONITOR after > x86_cpu_filter_features(). > > This was not an issue before commit 662175b91ff ("i386: reorder call to > cpu_exec_realizefn") because the feature added in the accel-specific > realizefn could be checked against host availability and filtered out. > > Additionally, it seems not a good idea to handle guest CPUID leaves in > host_cpu_realizefn(), and this patch merges host_cpu_enable_cpu_pm() > into kvm_cpu_realizefn(). > > Fixes: f5cc5a5c1686 ("i386: split cpu accelerators from cpu.c, using AccelCPUClass") > Fixes: 662175b91ff2 ("i386: reorder call to cpu_exec_realizefn") > Signed-off-by: Zide Chen <zide.chen@intel.com> > --- LGTM, Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
On 6/4/2024 8:02 AM, Zide Chen wrote: > host_cpu_realizefn() sets CPUID_EXT_MONITOR without consulting host/KVM > capabilities. This may cause problems: > > - If MWAIT/MONITOR is not available on the host, advertising this > feature to the guest and executing MWAIT/MONITOR from the guest > triggers #UD and the guest doesn't boot. This is because typically > #UD takes priority over VM-Exit interception checks and KVM doesn't > emulate MONITOR/MWAIT on #UD. > > - If KVM doesn't support KVM_X86_DISABLE_EXITS_MWAIT, MWAIT/MONITOR > from the guest are intercepted by KVM, which is not what cpu-pm=on > intends to do. > > In these cases, MWAIT/MONITOR should not be exposed to the guest. > > The logic in kvm_arch_get_supported_cpuid() to handle CPUID_EXT_MONITOR > is correct and sufficient, and we can't set CPUID_EXT_MONITOR after > x86_cpu_filter_features(). > > This was not an issue before commit 662175b91ff ("i386: reorder call to > cpu_exec_realizefn") because the feature added in the accel-specific > realizefn could be checked against host availability and filtered out. > > Additionally, it seems not a good idea to handle guest CPUID leaves in > host_cpu_realizefn(), and this patch merges host_cpu_enable_cpu_pm() > into kvm_cpu_realizefn(). > > Fixes: f5cc5a5c1686 ("i386: split cpu accelerators from cpu.c, using AccelCPUClass") > Fixes: 662175b91ff2 ("i386: reorder call to cpu_exec_realizefn") > Signed-off-by: Zide Chen <zide.chen@intel.com> Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com> > --- > > V3: > - don't set CPUID_EXT_MONITOR in kvm_cpu_realizefn(). > - Change title to reflect the main purpose of this patch. > > target/i386/host-cpu.c | 12 ------------ > target/i386/kvm/kvm-cpu.c | 11 +++++++++-- > 2 files changed, 9 insertions(+), 14 deletions(-) > > diff --git a/target/i386/host-cpu.c b/target/i386/host-cpu.c > index 280e427c017c..8b8bf5afeccf 100644 > --- a/target/i386/host-cpu.c > +++ b/target/i386/host-cpu.c > @@ -42,15 +42,6 @@ static uint32_t host_cpu_phys_bits(void) > return host_phys_bits; > } > > -static void host_cpu_enable_cpu_pm(X86CPU *cpu) > -{ > - CPUX86State *env = &cpu->env; > - > - host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, > - &cpu->mwait.ecx, &cpu->mwait.edx); > - env->features[FEAT_1_ECX] |= CPUID_EXT_MONITOR; > -} > - > static uint32_t host_cpu_adjust_phys_bits(X86CPU *cpu) > { > uint32_t host_phys_bits = host_cpu_phys_bits(); > @@ -83,9 +74,6 @@ bool host_cpu_realizefn(CPUState *cs, Error **errp) > X86CPU *cpu = X86_CPU(cs); > CPUX86State *env = &cpu->env; > > - if (cpu->max_features && enable_cpu_pm) { > - host_cpu_enable_cpu_pm(cpu); > - } > if (env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_LM) { > uint32_t phys_bits = host_cpu_adjust_phys_bits(cpu); > > diff --git a/target/i386/kvm/kvm-cpu.c b/target/i386/kvm/kvm-cpu.c > index f76972e47e61..148d10ce3711 100644 > --- a/target/i386/kvm/kvm-cpu.c > +++ b/target/i386/kvm/kvm-cpu.c > @@ -65,8 +65,15 @@ static bool kvm_cpu_realizefn(CPUState *cs, Error **errp) > * cpu_common_realizefn() (via xcc->parent_realize) > */ > if (cpu->max_features) { > - if (enable_cpu_pm && kvm_has_waitpkg()) { > - env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; > + if (enable_cpu_pm) { > + if (kvm_has_waitpkg()) { > + env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; > + } > + > + if (env->features[FEAT_1_ECX] & CPUID_EXT_MONITOR) { > + host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, > + &cpu->mwait.ecx, &cpu->mwait.edx); > + } > } > if (cpu->ucode_rev == 0) { > cpu->ucode_rev =
On Mon, 3 Jun 2024 17:02:22 -0700 Zide Chen <zide.chen@intel.com> wrote: > host_cpu_realizefn() sets CPUID_EXT_MONITOR without consulting host/KVM > capabilities. This may cause problems: > > - If MWAIT/MONITOR is not available on the host, advertising this > feature to the guest and executing MWAIT/MONITOR from the guest > triggers #UD and the guest doesn't boot. This is because typically > #UD takes priority over VM-Exit interception checks and KVM doesn't > emulate MONITOR/MWAIT on #UD. > > - If KVM doesn't support KVM_X86_DISABLE_EXITS_MWAIT, MWAIT/MONITOR > from the guest are intercepted by KVM, which is not what cpu-pm=on > intends to do. > > In these cases, MWAIT/MONITOR should not be exposed to the guest. > > The logic in kvm_arch_get_supported_cpuid() to handle CPUID_EXT_MONITOR > is correct and sufficient, and we can't set CPUID_EXT_MONITOR after > x86_cpu_filter_features(). > > This was not an issue before commit 662175b91ff ("i386: reorder call to > cpu_exec_realizefn") because the feature added in the accel-specific > realizefn could be checked against host availability and filtered out. > > Additionally, it seems not a good idea to handle guest CPUID leaves in > host_cpu_realizefn(), and this patch merges host_cpu_enable_cpu_pm() > into kvm_cpu_realizefn(). > > Fixes: f5cc5a5c1686 ("i386: split cpu accelerators from cpu.c, using AccelCPUClass") > Fixes: 662175b91ff2 ("i386: reorder call to cpu_exec_realizefn") > Signed-off-by: Zide Chen <zide.chen@intel.com> Reviewed-by: Igor Mammedov <imammedo@redhat.com> > --- > > V3: > - don't set CPUID_EXT_MONITOR in kvm_cpu_realizefn(). > - Change title to reflect the main purpose of this patch. > > target/i386/host-cpu.c | 12 ------------ > target/i386/kvm/kvm-cpu.c | 11 +++++++++-- > 2 files changed, 9 insertions(+), 14 deletions(-) > > diff --git a/target/i386/host-cpu.c b/target/i386/host-cpu.c > index 280e427c017c..8b8bf5afeccf 100644 > --- a/target/i386/host-cpu.c > +++ b/target/i386/host-cpu.c > @@ -42,15 +42,6 @@ static uint32_t host_cpu_phys_bits(void) > return host_phys_bits; > } > > -static void host_cpu_enable_cpu_pm(X86CPU *cpu) > -{ > - CPUX86State *env = &cpu->env; > - > - host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, > - &cpu->mwait.ecx, &cpu->mwait.edx); > - env->features[FEAT_1_ECX] |= CPUID_EXT_MONITOR; > -} > - > static uint32_t host_cpu_adjust_phys_bits(X86CPU *cpu) > { > uint32_t host_phys_bits = host_cpu_phys_bits(); > @@ -83,9 +74,6 @@ bool host_cpu_realizefn(CPUState *cs, Error **errp) > X86CPU *cpu = X86_CPU(cs); > CPUX86State *env = &cpu->env; > > - if (cpu->max_features && enable_cpu_pm) { > - host_cpu_enable_cpu_pm(cpu); > - } > if (env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_LM) { > uint32_t phys_bits = host_cpu_adjust_phys_bits(cpu); > > diff --git a/target/i386/kvm/kvm-cpu.c b/target/i386/kvm/kvm-cpu.c > index f76972e47e61..148d10ce3711 100644 > --- a/target/i386/kvm/kvm-cpu.c > +++ b/target/i386/kvm/kvm-cpu.c > @@ -65,8 +65,15 @@ static bool kvm_cpu_realizefn(CPUState *cs, Error **errp) > * cpu_common_realizefn() (via xcc->parent_realize) > */ > if (cpu->max_features) { > - if (enable_cpu_pm && kvm_has_waitpkg()) { > - env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; > + if (enable_cpu_pm) { > + if (kvm_has_waitpkg()) { > + env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; > + } > + > + if (env->features[FEAT_1_ECX] & CPUID_EXT_MONITOR) { > + host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, > + &cpu->mwait.ecx, &cpu->mwait.edx); > + } > } > if (cpu->ucode_rev == 0) { > cpu->ucode_rev =
diff --git a/target/i386/host-cpu.c b/target/i386/host-cpu.c index 280e427c017c..8b8bf5afeccf 100644 --- a/target/i386/host-cpu.c +++ b/target/i386/host-cpu.c @@ -42,15 +42,6 @@ static uint32_t host_cpu_phys_bits(void) return host_phys_bits; } -static void host_cpu_enable_cpu_pm(X86CPU *cpu) -{ - CPUX86State *env = &cpu->env; - - host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, - &cpu->mwait.ecx, &cpu->mwait.edx); - env->features[FEAT_1_ECX] |= CPUID_EXT_MONITOR; -} - static uint32_t host_cpu_adjust_phys_bits(X86CPU *cpu) { uint32_t host_phys_bits = host_cpu_phys_bits(); @@ -83,9 +74,6 @@ bool host_cpu_realizefn(CPUState *cs, Error **errp) X86CPU *cpu = X86_CPU(cs); CPUX86State *env = &cpu->env; - if (cpu->max_features && enable_cpu_pm) { - host_cpu_enable_cpu_pm(cpu); - } if (env->features[FEAT_8000_0001_EDX] & CPUID_EXT2_LM) { uint32_t phys_bits = host_cpu_adjust_phys_bits(cpu); diff --git a/target/i386/kvm/kvm-cpu.c b/target/i386/kvm/kvm-cpu.c index f76972e47e61..148d10ce3711 100644 --- a/target/i386/kvm/kvm-cpu.c +++ b/target/i386/kvm/kvm-cpu.c @@ -65,8 +65,15 @@ static bool kvm_cpu_realizefn(CPUState *cs, Error **errp) * cpu_common_realizefn() (via xcc->parent_realize) */ if (cpu->max_features) { - if (enable_cpu_pm && kvm_has_waitpkg()) { - env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; + if (enable_cpu_pm) { + if (kvm_has_waitpkg()) { + env->features[FEAT_7_0_ECX] |= CPUID_7_0_ECX_WAITPKG; + } + + if (env->features[FEAT_1_ECX] & CPUID_EXT_MONITOR) { + host_cpuid(5, 0, &cpu->mwait.eax, &cpu->mwait.ebx, + &cpu->mwait.ecx, &cpu->mwait.edx); + } } if (cpu->ucode_rev == 0) { cpu->ucode_rev =
host_cpu_realizefn() sets CPUID_EXT_MONITOR without consulting host/KVM capabilities. This may cause problems: - If MWAIT/MONITOR is not available on the host, advertising this feature to the guest and executing MWAIT/MONITOR from the guest triggers #UD and the guest doesn't boot. This is because typically #UD takes priority over VM-Exit interception checks and KVM doesn't emulate MONITOR/MWAIT on #UD. - If KVM doesn't support KVM_X86_DISABLE_EXITS_MWAIT, MWAIT/MONITOR from the guest are intercepted by KVM, which is not what cpu-pm=on intends to do. In these cases, MWAIT/MONITOR should not be exposed to the guest. The logic in kvm_arch_get_supported_cpuid() to handle CPUID_EXT_MONITOR is correct and sufficient, and we can't set CPUID_EXT_MONITOR after x86_cpu_filter_features(). This was not an issue before commit 662175b91ff ("i386: reorder call to cpu_exec_realizefn") because the feature added in the accel-specific realizefn could be checked against host availability and filtered out. Additionally, it seems not a good idea to handle guest CPUID leaves in host_cpu_realizefn(), and this patch merges host_cpu_enable_cpu_pm() into kvm_cpu_realizefn(). Fixes: f5cc5a5c1686 ("i386: split cpu accelerators from cpu.c, using AccelCPUClass") Fixes: 662175b91ff2 ("i386: reorder call to cpu_exec_realizefn") Signed-off-by: Zide Chen <zide.chen@intel.com> --- V3: - don't set CPUID_EXT_MONITOR in kvm_cpu_realizefn(). - Change title to reflect the main purpose of this patch. target/i386/host-cpu.c | 12 ------------ target/i386/kvm/kvm-cpu.c | 11 +++++++++-- 2 files changed, 9 insertions(+), 14 deletions(-)