> On 1/9/2023 3:28 pm, Xiong Zhang wrote: > > vLBR event will be released at vcpu sched-in time if LBR_EN bit is not > > set in GUEST_IA32_DEBUGCTL VMCS field, this bit is cleared in two cases: > > 1. guest disable LBR through WRMSR > > 2. KVM disable LBR at PMI injection to emulate guest FREEZE_LBR_ON_PMI. > > > > The first case is guest LBR won't be used anymore and vLBR event can > > be released, but guest LBR is still be used in the second case, vLBR > > event can not be released. > > > > Considering this serial: > > 1. vPMC overflow, KVM injects vPMI and clears guest LBR_EN 2. guest > > handles PMI, and reads LBR records. > > 3. vCPU is sched-out, later sched-in, vLBR event is released. > > This has nothing to do with vPMI. If guest lbr is disabled and the guest LBR driver > doesn't read it before the KVM vLBR event is released (typically after two sched > slices), that part of the LBR records are lost in terms of design. What is needed > here is a generic KVM mechanism to close this gap. PMI handler will read LBR records, so it is easier to find this issue with vPMI. Actually I found this issue with this test https://lore.kernel.org/kvm/20230901074052.640296-3-xiong.y.zhang@xxxxxxxxx/T/#u. Indeed it can be extended to generic case as you said, in order to fix this gap, my rough idea is: 1. find a point to save LBR snapshot, where ???? it is too heavy when guest disable LBR, it is too late when vcpu sched-out. 2. Before guest enable LBR, kvm return LBR snapshot when guest read LBR records 3. When guest enable LBR, kvm create vLBR and restore snapshot (maybe this restore isn't needed). As LBR msrs number is big, this will make vLBR overhead bigger. Anyway, as vPMI may be frequent, this commit could reduce the number of vLBR release and creation, so it is still needed. > > > 4. Guest continue reading LBR records, KVM creates vLBR event again, > > the vLBR event is the only LBR user on host now, host PMU driver will > > reset HW LBR facility at vLBR creataion. > > 5. Guest gets the remain LBR records with reset state. > > This is conflict with FREEZE_LBR_ON_PMI meaning, so vLBR event can not > > be release on PMI. > > > > This commit adds a freeze_on_pmi flag, this flag is set at pmi > > injection and is cleared when guest operates guest DEBUGCTL_MSR. If > > this flag is true, vLBR event will not be released. > > > > Signed-off-by: Xiong Zhang <xiong.y.zhang@xxxxxxxxx> > > --- > > arch/x86/kvm/vmx/pmu_intel.c | 5 ++++- > > arch/x86/kvm/vmx/vmx.c | 12 +++++++++--- > > arch/x86/kvm/vmx/vmx.h | 3 +++ > > 3 files changed, 16 insertions(+), 4 deletions(-) > > > > diff --git a/arch/x86/kvm/vmx/pmu_intel.c > > b/arch/x86/kvm/vmx/pmu_intel.c index f2efa0bf7ae8..3a36a91638c6 100644 > > --- a/arch/x86/kvm/vmx/pmu_intel.c > > +++ b/arch/x86/kvm/vmx/pmu_intel.c > > @@ -628,6 +628,7 @@ static void intel_pmu_init(struct kvm_vcpu *vcpu) > > lbr_desc->records.nr = 0; > > lbr_desc->event = NULL; > > lbr_desc->msr_passthrough = false; > > + lbr_desc->freeze_on_pmi = false; > > } > > > > static void intel_pmu_reset(struct kvm_vcpu *vcpu) @@ -670,6 +671,7 > > @@ static void intel_pmu_legacy_freezing_lbrs_on_pmi(struct kvm_vcpu > *vcpu) > > if (data & DEBUGCTLMSR_FREEZE_LBRS_ON_PMI) { > > data &= ~DEBUGCTLMSR_LBR; > > vmcs_write64(GUEST_IA32_DEBUGCTL, data); > > + vcpu_to_lbr_desc(vcpu)->freeze_on_pmi = true; > > } > > } > > > > @@ -761,7 +763,8 @@ void vmx_passthrough_lbr_msrs(struct kvm_vcpu > > *vcpu) > > > > static void intel_pmu_cleanup(struct kvm_vcpu *vcpu) > > { > > - if (!(vmcs_read64(GUEST_IA32_DEBUGCTL) & DEBUGCTLMSR_LBR)) > > + if (!(vmcs_read64(GUEST_IA32_DEBUGCTL) & DEBUGCTLMSR_LBR) && > > + !vcpu_to_lbr_desc(vcpu)->freeze_on_pmi) > > intel_pmu_release_guest_lbr_event(vcpu); > > } > > > > diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c index > > e6849f780dba..199d0da1dbee 100644 > > --- a/arch/x86/kvm/vmx/vmx.c > > +++ b/arch/x86/kvm/vmx/vmx.c > > @@ -2223,9 +2223,15 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, > struct msr_data *msr_info) > > get_vmcs12(vcpu)->guest_ia32_debugctl = data; > > > > vmcs_write64(GUEST_IA32_DEBUGCTL, data); > > - if (intel_pmu_lbr_is_enabled(vcpu) && !to_vmx(vcpu)- > >lbr_desc.event && > > - (data & DEBUGCTLMSR_LBR)) > > - intel_pmu_create_guest_lbr_event(vcpu); > > + > > + if (intel_pmu_lbr_is_enabled(vcpu)) { > > + struct lbr_desc *lbr_desc = vcpu_to_lbr_desc(vcpu); > > + > > + lbr_desc->freeze_on_pmi = false; > > + if (!lbr_desc->event && (data & DEBUGCTLMSR_LBR)) > > + intel_pmu_create_guest_lbr_event(vcpu); > > + } > > + > > return 0; > > } > > case MSR_IA32_BNDCFGS: > > diff --git a/arch/x86/kvm/vmx/vmx.h b/arch/x86/kvm/vmx/vmx.h index > > c2130d2c8e24..9729ccfa75ae 100644 > > --- a/arch/x86/kvm/vmx/vmx.h > > +++ b/arch/x86/kvm/vmx/vmx.h > > @@ -107,6 +107,9 @@ struct lbr_desc { > > > > /* True if LBRs are marked as not intercepted in the MSR bitmap */ > > bool msr_passthrough; > > + > > + /* True if LBR is frozen on PMI */ > > + bool freeze_on_pmi; > > }; > > > > /*
