On 2/7/2026 6:53 AM, Jim Mattson wrote: > Update amd_pmu_set_eventsel_hw() to clear the event selector's hardware > enable bit when the PMC should not count based on the guest's Host-Only and > Guest-Only event selector bits and the current vCPU state. > > Signed-off-by: Jim Mattson <[email protected]> > --- > arch/x86/include/asm/perf_event.h | 2 ++ > arch/x86/kvm/svm/pmu.c | 18 ++++++++++++++++++ > 2 files changed, 20 insertions(+) > > diff --git a/arch/x86/include/asm/perf_event.h > b/arch/x86/include/asm/perf_event.h > index 0d9af4135e0a..4dfe12053c09 100644 > --- a/arch/x86/include/asm/perf_event.h > +++ b/arch/x86/include/asm/perf_event.h > @@ -58,6 +58,8 @@ > #define AMD64_EVENTSEL_INT_CORE_ENABLE (1ULL << 36) > #define AMD64_EVENTSEL_GUESTONLY (1ULL << 40) > #define AMD64_EVENTSEL_HOSTONLY (1ULL << 41) > +#define AMD64_EVENTSEL_HOST_GUEST_MASK \ > + (AMD64_EVENTSEL_HOSTONLY | AMD64_EVENTSEL_GUESTONLY) > > #define AMD64_EVENTSEL_INT_CORE_SEL_SHIFT 37 > #define AMD64_EVENTSEL_INT_CORE_SEL_MASK \ > diff --git a/arch/x86/kvm/svm/pmu.c b/arch/x86/kvm/svm/pmu.c > index d9ca633f9f49..8d451110a94d 100644 > --- a/arch/x86/kvm/svm/pmu.c > +++ b/arch/x86/kvm/svm/pmu.c > @@ -149,8 +149,26 @@ static int amd_pmu_get_msr(struct kvm_vcpu *vcpu, struct > msr_data *msr_info) > > static void amd_pmu_set_eventsel_hw(struct kvm_pmc *pmc) > { > + struct kvm_vcpu *vcpu = pmc->vcpu; > + u64 host_guest_bits; > + > pmc->eventsel_hw = (pmc->eventsel & ~AMD64_EVENTSEL_HOSTONLY) | > AMD64_EVENTSEL_GUESTONLY; > + > + if (!(pmc->eventsel & ARCH_PERFMON_EVENTSEL_ENABLE)) > + return; > + > + if (!(vcpu->arch.efer & EFER_SVME)) > + return; > + > + host_guest_bits = pmc->eventsel & AMD64_EVENTSEL_HOST_GUEST_MASK; > + if (!host_guest_bits || host_guest_bits == > AMD64_EVENTSEL_HOST_GUEST_MASK) > + return; > + > + if (!!(host_guest_bits & AMD64_EVENTSEL_GUESTONLY) == > is_guest_mode(vcpu)) > + return;
This seems to disable the PMCs after exits from an L2 guest to the L0 hypervisor. For such transitions, the corresponding L1 vCPU's PMC has GuestOnly set but is_guest_mode() is false as this function is called at the very end of leave_guest_mode() after vcpu->stat.guest_mode is set to 0. Is this a correct interpretation of the condition above? > + > + pmc->eventsel_hw &= ~ARCH_PERFMON_EVENTSEL_ENABLE; > } > > static int amd_pmu_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
