On Thu, Nov 03, 2022, Jim Mattson wrote: > On Thu, Nov 3, 2022 at 12:18 PM Aaron Lewis <aaronlewis@xxxxxxxxxx> wrote: > > > > When the PMU is disabled, don't bother sharing the PMU MSRs with > > userspace through KVM_GET_MSR_INDEX_LIST. Instead, filter them out so > > userspace doesn't have to keep track of them. > > > > Note that 'enable_pmu' is read-only, so userspace has no control over > > whether the PMU MSRs are included in the list or not. > > > > Suggested-by: Sean Christopherson <seanjc@xxxxxxxxxx> > > Signed-off-by: Aaron Lewis <aaronlewis@xxxxxxxxxx> > > --- > > arch/x86/kvm/x86.c | 15 +++++++++++---- > > 1 file changed, 11 insertions(+), 4 deletions(-) > > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c > > index 521b433f978c..19bc42a6946d 100644 > > --- a/arch/x86/kvm/x86.c > > +++ b/arch/x86/kvm/x86.c > > @@ -7042,13 +7042,20 @@ static void kvm_init_msr_list(void) > > continue; > > break; > > case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17: > > - if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= > > - min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp)) > > + if ((msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= > > + min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp)) || > > + !enable_pmu) > > continue; > > break; > > case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17: > > - if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= > > - min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp)) > > + if ((msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= > > + min(INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp)) || > > + !enable_pmu) > > + continue; > > + break; > > + case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: > > + case MSR_K7_EVNTSEL0 ... MSR_K7_PERFCTR3: > > + if (!enable_pmu) > > continue; > > break; > > case MSR_IA32_XFD: > > I think you've missed a bunch: > > MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1, > MSR_ARCH_PERFMON_FIXED_CTR0 + 2, > MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS, > MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL, > MSR_IA32_PEBS_ENABLE, MSR_PEBS_DATA_CFG Ooh, better idea! Assuming this isn't a pressing concern, what about organizing the MSR lists into sublists, kinda like what I proposed for VMX MSRs[*], but do it in a generic way. E.g. have all the PMU MSRs in a sub-list so that they can be skipped as a group instead of needing to add a bunch of enable_pmu checks. [*] https://lore.kernel.org/all/20220805172945.35412-3-seanjc@xxxxxxxxxx
