1. Patchew
  2. linux
  3. [PATCH 0/6] KVM: x86/pmu: Add support for AMD HG_ONLY bits
  4. View patch

[PATCH 4/6] KVM: x86/pmu: [De]activate HG_ONLY PMCs at SVME changes and nested transitions

Jim Mattson posted 6 patches 2 weeks, 4 days ago
[PATCH 4/6] KVM: x86/pmu: [De]activate HG_ONLY PMCs at SVME changes and nested transitions
Posted by Jim Mattson 2 weeks, 4 days ago 
Add a new function, kvm_pmu_set_pmc_eventsel_hw_enable(), to set or clear
the enable bit in eventsel_hw for PMCs identified by a bitmap.

Use this function to update Host-Only and Guest-Only counters at the
following transitions:

  - svm_set_efer(): When SVME changes, enable Guest-Only counters if SVME
    is being cleared (HG_ONLY bits become ignored), or disable them if SVME
    is being set (L1 is active).

  - nested_svm_vmrun(): Disable Host-Only counters and enable Guest-Only
    counters.

  - nested_svm_vmexit(): Disable Guest-Only counters and enable Host-Only
    counters.

Signed-off-by: Jim Mattson <jmattson@google.com>
---
 arch/x86/include/asm/kvm-x86-pmu-ops.h |  1 +
 arch/x86/kvm/pmu.c                     |  7 +++++++
 arch/x86/kvm/pmu.h                     |  4 ++++
 arch/x86/kvm/svm/nested.c              | 10 ++++++++++
 arch/x86/kvm/svm/pmu.c                 | 17 +++++++++++++++++
 arch/x86/kvm/svm/svm.c                 |  3 +++
 6 files changed, 42 insertions(+)

diff --git a/arch/x86/include/asm/kvm-x86-pmu-ops.h b/arch/x86/include/asm/kvm-x86-pmu-ops.h
index f0aa6996811f..7b32796213a0 100644
--- a/arch/x86/include/asm/kvm-x86-pmu-ops.h
+++ b/arch/x86/include/asm/kvm-x86-pmu-ops.h
@@ -26,6 +26,7 @@ KVM_X86_PMU_OP_OPTIONAL(cleanup)
 KVM_X86_PMU_OP_OPTIONAL(write_global_ctrl)
 KVM_X86_PMU_OP(mediated_load)
 KVM_X86_PMU_OP(mediated_put)
+KVM_X86_PMU_OP_OPTIONAL(set_pmc_eventsel_hw_enable)
 
 #undef KVM_X86_PMU_OP
 #undef KVM_X86_PMU_OP_OPTIONAL
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
index 833ee2ecd43f..1541c201285b 100644
--- a/arch/x86/kvm/pmu.c
+++ b/arch/x86/kvm/pmu.c
@@ -1142,6 +1142,13 @@ void kvm_pmu_branch_retired(struct kvm_vcpu *vcpu)
 }
 EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_branch_retired);
 
+void kvm_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
+				       unsigned long *bitmap, bool enable)
+{
+	kvm_pmu_call(set_pmc_eventsel_hw_enable)(vcpu, bitmap, enable);
+}
+EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_set_pmc_eventsel_hw_enable);
+
 static bool is_masked_filter_valid(const struct kvm_x86_pmu_event_filter *filter)
 {
 	u64 mask = kvm_pmu_ops.EVENTSEL_EVENT |
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index 0925246731cb..b8be8b6e40d8 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -41,6 +41,8 @@ struct kvm_pmu_ops {
 	void (*mediated_load)(struct kvm_vcpu *vcpu);
 	void (*mediated_put)(struct kvm_vcpu *vcpu);
 	void (*write_global_ctrl)(u64 global_ctrl);
+	void (*set_pmc_eventsel_hw_enable)(struct kvm_vcpu *vcpu,
+					   unsigned long *bitmap, bool enable);
 
 	const u64 EVENTSEL_EVENT;
 	const int MAX_NR_GP_COUNTERS;
@@ -258,6 +260,8 @@ void kvm_pmu_destroy(struct kvm_vcpu *vcpu);
 int kvm_vm_ioctl_set_pmu_event_filter(struct kvm *kvm, void __user *argp);
 void kvm_pmu_instruction_retired(struct kvm_vcpu *vcpu);
 void kvm_pmu_branch_retired(struct kvm_vcpu *vcpu);
+void kvm_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
+				       unsigned long *bitmap, bool enable);
 void kvm_mediated_pmu_load(struct kvm_vcpu *vcpu);
 void kvm_mediated_pmu_put(struct kvm_vcpu *vcpu);
 
diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
index de90b104a0dd..edaa76e38417 100644
--- a/arch/x86/kvm/svm/nested.c
+++ b/arch/x86/kvm/svm/nested.c
@@ -28,6 +28,7 @@
 #include "smm.h"
 #include "cpuid.h"
 #include "lapic.h"
+#include "pmu.h"
 #include "svm.h"
 #include "hyperv.h"
 
@@ -1054,6 +1055,11 @@ int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
 		goto out_exit_err;
 
+	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
+		vcpu_to_pmu(vcpu)->pmc_hostonly, false);
+	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
+		vcpu_to_pmu(vcpu)->pmc_guestonly, true);
+
 	if (nested_svm_merge_msrpm(vcpu))
 		goto out;
 
@@ -1137,6 +1143,10 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 
 	/* Exit Guest-Mode */
 	leave_guest_mode(vcpu);
+	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
+		vcpu_to_pmu(vcpu)->pmc_hostonly, true);
+	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
+		vcpu_to_pmu(vcpu)->pmc_guestonly, false);
 	svm->nested.vmcb12_gpa = 0;
 	WARN_ON_ONCE(svm->nested.nested_run_pending);
 
diff --git a/arch/x86/kvm/svm/pmu.c b/arch/x86/kvm/svm/pmu.c
index c06013e2b4b1..85155d65fa38 100644
--- a/arch/x86/kvm/svm/pmu.c
+++ b/arch/x86/kvm/svm/pmu.c
@@ -316,6 +316,22 @@ static void amd_mediated_pmu_put(struct kvm_vcpu *vcpu)
 		wrmsrq(MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR, pmu->global_status);
 }
 
+static void amd_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
+					       unsigned long *bitmap,
+					       bool enable)
+{
+	struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
+	struct kvm_pmc *pmc;
+	int i;
+
+	kvm_for_each_pmc(pmu, pmc, i, bitmap) {
+		if (enable)
+			pmc->eventsel_hw |= ARCH_PERFMON_EVENTSEL_ENABLE;
+		else
+			pmc->eventsel_hw &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
+	}
+}
+
 struct kvm_pmu_ops amd_pmu_ops __initdata = {
 	.rdpmc_ecx_to_pmc = amd_rdpmc_ecx_to_pmc,
 	.msr_idx_to_pmc = amd_msr_idx_to_pmc,
@@ -329,6 +345,7 @@ struct kvm_pmu_ops amd_pmu_ops __initdata = {
 	.is_mediated_pmu_supported = amd_pmu_is_mediated_pmu_supported,
 	.mediated_load = amd_mediated_pmu_load,
 	.mediated_put = amd_mediated_pmu_put,
+	.set_pmc_eventsel_hw_enable = amd_pmu_set_pmc_eventsel_hw_enable,
 
 	.EVENTSEL_EVENT = AMD64_EVENTSEL_EVENT,
 	.MAX_NR_GP_COUNTERS = KVM_MAX_NR_AMD_GP_COUNTERS,
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 7803d2781144..953089b38921 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -244,6 +244,9 @@ int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 			if (svm_gp_erratum_intercept && !sev_guest(vcpu->kvm))
 				set_exception_intercept(svm, GP_VECTOR);
 		}
+
+		kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
+			vcpu_to_pmu(vcpu)->pmc_guestonly, !(efer & EFER_SVME));
 	}
 
 	svm->vmcb->save.efer = efer | EFER_SVME;
-- 
2.52.0.457.g6b5491de43-goog
Re: [PATCH 4/6] KVM: x86/pmu: [De]activate HG_ONLY PMCs at SVME changes and nested transitions
Posted by Sean Christopherson 2 weeks, 3 days ago 
On Wed, Jan 21, 2026, Jim Mattson wrote:
> diff --git a/arch/x86/include/asm/kvm-x86-pmu-ops.h b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> index f0aa6996811f..7b32796213a0 100644
> --- a/arch/x86/include/asm/kvm-x86-pmu-ops.h
> +++ b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> @@ -26,6 +26,7 @@ KVM_X86_PMU_OP_OPTIONAL(cleanup)
>  KVM_X86_PMU_OP_OPTIONAL(write_global_ctrl)
>  KVM_X86_PMU_OP(mediated_load)
>  KVM_X86_PMU_OP(mediated_put)
> +KVM_X86_PMU_OP_OPTIONAL(set_pmc_eventsel_hw_enable)
>  
>  #undef KVM_X86_PMU_OP
>  #undef KVM_X86_PMU_OP_OPTIONAL
> diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> index 833ee2ecd43f..1541c201285b 100644
> --- a/arch/x86/kvm/pmu.c
> +++ b/arch/x86/kvm/pmu.c
> @@ -1142,6 +1142,13 @@ void kvm_pmu_branch_retired(struct kvm_vcpu *vcpu)
>  }
>  EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_branch_retired);
>  
> +void kvm_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
> +				       unsigned long *bitmap, bool enable)
> +{
> +	kvm_pmu_call(set_pmc_eventsel_hw_enable)(vcpu, bitmap, enable);
> +}
> +EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_set_pmc_eventsel_hw_enable);

Why bounce through a PMU op just to go from nested.c to pmu.c?  AFAICT, common
x86 code never calls kvm_pmu_set_pmc_eventsel_hw_enable(), just wire up calls
directly to amd_pmu_refresh_host_guest_eventsels().

> @@ -1054,6 +1055,11 @@ int nested_svm_vmrun(struct kvm_vcpu *vcpu)
>  	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
>  		goto out_exit_err;
>  
> +	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> +		vcpu_to_pmu(vcpu)->pmc_hostonly, false);
> +	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> +		vcpu_to_pmu(vcpu)->pmc_guestonly, true);
> +
>  	if (nested_svm_merge_msrpm(vcpu))
>  		goto out;
>  
> @@ -1137,6 +1143,10 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
>  
>  	/* Exit Guest-Mode */
>  	leave_guest_mode(vcpu);
> +	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> +		vcpu_to_pmu(vcpu)->pmc_hostonly, true);
> +	kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> +		vcpu_to_pmu(vcpu)->pmc_guestonly, false);
>  	svm->nested.vmcb12_gpa = 0;
>  	WARN_ON_ONCE(svm->nested.nested_run_pending);

I don't think these are the right places to hook.  Shouldn't KVM update the
event selectors on _all_ transitions, whether they're architectural or not?  E.g.
by wrapping {enter,leave}_guest_mode()?

static void svm_enter_guest_mode(struct kvm_vcpu *vcpu)
{
	enter_guest_mode(vcpu);
	amd_pmu_refresh_host_guest_eventsels(vcpu);
}

static void svm_leave_guest_mode(struct kvm_vcpu *vcpu)
{
	leave_guest_mode(vcpu);
	amd_pmu_refresh_host_guest_eventsels(vcpu);
}
Re: [PATCH 4/6] KVM: x86/pmu: [De]activate HG_ONLY PMCs at SVME changes and nested transitions
Posted by Jim Mattson 1 week, 4 days ago 
On Thu, Jan 22, 2026 at 8:55 AM Sean Christopherson <seanjc@google.com> wrote:
>
> On Wed, Jan 21, 2026, Jim Mattson wrote:
> > diff --git a/arch/x86/include/asm/kvm-x86-pmu-ops.h b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > index f0aa6996811f..7b32796213a0 100644
> > --- a/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > +++ b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > @@ -26,6 +26,7 @@ KVM_X86_PMU_OP_OPTIONAL(cleanup)
> >  KVM_X86_PMU_OP_OPTIONAL(write_global_ctrl)
> >  KVM_X86_PMU_OP(mediated_load)
> >  KVM_X86_PMU_OP(mediated_put)
> > +KVM_X86_PMU_OP_OPTIONAL(set_pmc_eventsel_hw_enable)
> >
> >  #undef KVM_X86_PMU_OP
> >  #undef KVM_X86_PMU_OP_OPTIONAL
> > diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> > index 833ee2ecd43f..1541c201285b 100644
> > --- a/arch/x86/kvm/pmu.c
> > +++ b/arch/x86/kvm/pmu.c
> > @@ -1142,6 +1142,13 @@ void kvm_pmu_branch_retired(struct kvm_vcpu *vcpu)
> >  }
> >  EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_branch_retired);
> >
> > +void kvm_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
> > +                                    unsigned long *bitmap, bool enable)
> > +{
> > +     kvm_pmu_call(set_pmc_eventsel_hw_enable)(vcpu, bitmap, enable);
> > +}
> > +EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_set_pmc_eventsel_hw_enable);
>
> Why bounce through a PMU op just to go from nested.c to pmu.c?  AFAICT, common
> x86 code never calls kvm_pmu_set_pmc_eventsel_hw_enable(), just wire up calls
> directly to amd_pmu_refresh_host_guest_eventsels().

It seemed that pmu.c deliberately didn't export anything. All accesses
were via virtual function table. But maybe that was just happenstance.
Should I create a separate pmu.h, or just throw the prototype into
svm.h?

> > @@ -1054,6 +1055,11 @@ int nested_svm_vmrun(struct kvm_vcpu *vcpu)
> >       if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
> >               goto out_exit_err;
> >
> > +     kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> > +             vcpu_to_pmu(vcpu)->pmc_hostonly, false);
> > +     kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> > +             vcpu_to_pmu(vcpu)->pmc_guestonly, true);
> > +
> >       if (nested_svm_merge_msrpm(vcpu))
> >               goto out;
> >
> > @@ -1137,6 +1143,10 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
> >
> >       /* Exit Guest-Mode */
> >       leave_guest_mode(vcpu);
> > +     kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> > +             vcpu_to_pmu(vcpu)->pmc_hostonly, true);
> > +     kvm_pmu_set_pmc_eventsel_hw_enable(vcpu,
> > +             vcpu_to_pmu(vcpu)->pmc_guestonly, false);
> >       svm->nested.vmcb12_gpa = 0;
> >       WARN_ON_ONCE(svm->nested.nested_run_pending);
>
> I don't think these are the right places to hook.  Shouldn't KVM update the
> event selectors on _all_ transitions, whether they're architectural or not?  E.g.
> by wrapping {enter,leave}_guest_mode()?

You are so right! I will fix this in the next version.

> static void svm_enter_guest_mode(struct kvm_vcpu *vcpu)
> {
>         enter_guest_mode(vcpu);
>         amd_pmu_refresh_host_guest_eventsels(vcpu);
> }
>
> static void svm_leave_guest_mode(struct kvm_vcpu *vcpu)
> {
>         leave_guest_mode(vcpu);
>         amd_pmu_refresh_host_guest_eventsels(vcpu);
> }
Re: [PATCH 4/6] KVM: x86/pmu: [De]activate HG_ONLY PMCs at SVME changes and nested transitions
Posted by Sean Christopherson 1 week, 3 days ago 
On Wed, Jan 28, 2026, Jim Mattson wrote:
> On Thu, Jan 22, 2026 at 8:55 AM Sean Christopherson <seanjc@google.com> wrote:
> >
> > On Wed, Jan 21, 2026, Jim Mattson wrote:
> > > diff --git a/arch/x86/include/asm/kvm-x86-pmu-ops.h b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > > index f0aa6996811f..7b32796213a0 100644
> > > --- a/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > > +++ b/arch/x86/include/asm/kvm-x86-pmu-ops.h
> > > @@ -26,6 +26,7 @@ KVM_X86_PMU_OP_OPTIONAL(cleanup)
> > >  KVM_X86_PMU_OP_OPTIONAL(write_global_ctrl)
> > >  KVM_X86_PMU_OP(mediated_load)
> > >  KVM_X86_PMU_OP(mediated_put)
> > > +KVM_X86_PMU_OP_OPTIONAL(set_pmc_eventsel_hw_enable)
> > >
> > >  #undef KVM_X86_PMU_OP
> > >  #undef KVM_X86_PMU_OP_OPTIONAL
> > > diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
> > > index 833ee2ecd43f..1541c201285b 100644
> > > --- a/arch/x86/kvm/pmu.c
> > > +++ b/arch/x86/kvm/pmu.c
> > > @@ -1142,6 +1142,13 @@ void kvm_pmu_branch_retired(struct kvm_vcpu *vcpu)
> > >  }
> > >  EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_branch_retired);
> > >
> > > +void kvm_pmu_set_pmc_eventsel_hw_enable(struct kvm_vcpu *vcpu,
> > > +                                    unsigned long *bitmap, bool enable)
> > > +{
> > > +     kvm_pmu_call(set_pmc_eventsel_hw_enable)(vcpu, bitmap, enable);
> > > +}
> > > +EXPORT_SYMBOL_FOR_KVM_INTERNAL(kvm_pmu_set_pmc_eventsel_hw_enable);
> >
> > Why bounce through a PMU op just to go from nested.c to pmu.c?  AFAICT, common
> > x86 code never calls kvm_pmu_set_pmc_eventsel_hw_enable(), just wire up calls
> > directly to amd_pmu_refresh_host_guest_eventsels().
> 
> It seemed that pmu.c deliberately didn't export anything. All accesses
> were via virtual function table. But maybe that was just happenstance.

Probably just happenstance?

> Should I create a separate pmu.h, or just throw the prototype into
> svm.h?

I say just throw it in svm.h.  We've had pmu_intel.h for a long time, and there's
hardly anything in there.  And somewhat surprisingly, only two things in vmx.h
that obviously could go in pmu_intel.h.

  void intel_pmu_cross_mapped_check(struct kvm_pmu *pmu);
  int intel_pmu_create_guest_lbr_event(struct kvm_vcpu *vcpu);

Patchew 2.1-devel-b67e4c2

© 2016 - 2026 Red Hat, Inc.