Message ID | 20210223013958.1280444-1-like.xu@linux.intel.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | [1/2] KVM: vmx/pmu: Fix dummy check if lbr_desc->event is created | expand |
On 23/02/21 02:39, Like Xu wrote: > If lbr_desc->event is successfully created, the intel_pmu_create_ > guest_lbr_event() will return 0, otherwise it will return -ENOENT, > and then jump to LBR msrs dummy handling. > > Fixes: 1b5ac3226a1a ("KVM: vmx/pmu: Pass-through LBR msrs when the guest LBR event is ACTIVE") > Signed-off-by: Like Xu <like.xu@linux.intel.com> > --- > arch/x86/kvm/vmx/pmu_intel.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c > index d1df618cb7de..d6a5fe19ff09 100644 > --- a/arch/x86/kvm/vmx/pmu_intel.c > +++ b/arch/x86/kvm/vmx/pmu_intel.c > @@ -320,7 +320,7 @@ static bool intel_pmu_handle_lbr_msrs_access(struct kvm_vcpu *vcpu, > if (!intel_pmu_is_valid_lbr_msr(vcpu, index)) > return false; > > - if (!lbr_desc->event && !intel_pmu_create_guest_lbr_event(vcpu)) > + if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) > goto dummy; Queued, adding a "< 0" to clarify the semantics of the function. Paolo
On Tue, Feb 23, 2021, Like Xu wrote: > If lbr_desc->event is successfully created, the intel_pmu_create_ > guest_lbr_event() will return 0, otherwise it will return -ENOENT, > and then jump to LBR msrs dummy handling. > > Fixes: 1b5ac3226a1a ("KVM: vmx/pmu: Pass-through LBR msrs when the guest LBR event is ACTIVE") > Signed-off-by: Like Xu <like.xu@linux.intel.com> > --- > arch/x86/kvm/vmx/pmu_intel.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c > index d1df618cb7de..d6a5fe19ff09 100644 > --- a/arch/x86/kvm/vmx/pmu_intel.c > +++ b/arch/x86/kvm/vmx/pmu_intel.c > @@ -320,7 +320,7 @@ static bool intel_pmu_handle_lbr_msrs_access(struct kvm_vcpu *vcpu, > if (!intel_pmu_is_valid_lbr_msr(vcpu, index)) > return false; > > - if (!lbr_desc->event && !intel_pmu_create_guest_lbr_event(vcpu)) > + if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) > goto dummy; Wouldn't it be better to create an event only on write? And really, why create the event in this flow in the first place? In normal operation, can't event creation be deferred until GUEST_IA32_DEBUGCTL.DEBUGCTLMSR_LBR=1? If event creation fails in that flow, I would think KVM would do its best to create an event in future runs without waiting for additional actions from the guest. Also, this bug suggests there's a big gaping hole in the test coverage. AFAICT, event contention would lead to a #GP crash in the host due to lbr_desc->event being dereferenced, no? > > /* > -- > 2.29.2 >
On 23/02/21 18:15, Sean Christopherson wrote: > If event > creation fails in that flow, I would think KVM would do its best to create an > event in future runs without waiting for additional actions from the guest. > > Also, this bug suggests there's a big gaping hole in the test coverage. AFAICT, > event contention would lead to a #GP crash in the host due to lbr_desc->event > being dereferenced, no? Yes, testing contention would use the tools/testing/selftests/kvm framework rather than just kvm-unit-tests. Paolo
On 2021/2/24 1:15, Sean Christopherson wrote: > On Tue, Feb 23, 2021, Like Xu wrote: >> If lbr_desc->event is successfully created, the intel_pmu_create_ >> guest_lbr_event() will return 0, otherwise it will return -ENOENT, >> and then jump to LBR msrs dummy handling. >> >> Fixes: 1b5ac3226a1a ("KVM: vmx/pmu: Pass-through LBR msrs when the guest LBR event is ACTIVE") >> Signed-off-by: Like Xu <like.xu@linux.intel.com> >> --- >> arch/x86/kvm/vmx/pmu_intel.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c >> index d1df618cb7de..d6a5fe19ff09 100644 >> --- a/arch/x86/kvm/vmx/pmu_intel.c >> +++ b/arch/x86/kvm/vmx/pmu_intel.c >> @@ -320,7 +320,7 @@ static bool intel_pmu_handle_lbr_msrs_access(struct kvm_vcpu *vcpu, >> if (!intel_pmu_is_valid_lbr_msr(vcpu, index)) >> return false; >> >> - if (!lbr_desc->event && !intel_pmu_create_guest_lbr_event(vcpu)) >> + if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) >> goto dummy; > Wouldn't it be better to create an event only on write? And really, why create > the event in this flow in the first place? In normal operation, can't event > creation be deferred until GUEST_IA32_DEBUGCTL.DEBUGCTLMSR_LBR=1? We need event creation and pass-through for both read and write. The LBR driver would firstly access the MSR_LBR_SELECT to configure branch types and we also create LBR event for GUEST_IA32_DEBUGCTL.DEBUGCTLMSR_LBR=1 trap. A lazy approach requests more cached values and more traps. > If event > creation fails in that flow, I would think KVM would do its best to create an > event in future runs without waiting for additional actions from the guest. We have done this via releasing the LBR event for next creation and pass-through try. > > Also, this bug suggests there's a big gaping hole in the test coverage. Not a big but concern one. To hit that, we need run LBR agent on the host and grab LBR from the guest. And it's not covered in the current test cases since we do not recommend this kind of usage in the comment. > AFAICT, > event contention would lead to a #GP crash in the host due to lbr_desc->event > being dereferenced, no? a #GP crash in the host ?Can you share more understanding about it ? > >> >> /* >> -- >> 2.29.2 >>
On Wed, Feb 24, 2021, Xu, Like wrote: > On 2021/2/24 1:15, Sean Christopherson wrote: > > On Tue, Feb 23, 2021, Like Xu wrote: > > > If lbr_desc->event is successfully created, the intel_pmu_create_ > > > guest_lbr_event() will return 0, otherwise it will return -ENOENT, > > > and then jump to LBR msrs dummy handling. > > > > > > Fixes: 1b5ac3226a1a ("KVM: vmx/pmu: Pass-through LBR msrs when the guest LBR event is ACTIVE") > > > Signed-off-by: Like Xu <like.xu@linux.intel.com> > > > --- > > > arch/x86/kvm/vmx/pmu_intel.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c > > > index d1df618cb7de..d6a5fe19ff09 100644 > > > --- a/arch/x86/kvm/vmx/pmu_intel.c > > > +++ b/arch/x86/kvm/vmx/pmu_intel.c > > > @@ -320,7 +320,7 @@ static bool intel_pmu_handle_lbr_msrs_access(struct kvm_vcpu *vcpu, > > > if (!intel_pmu_is_valid_lbr_msr(vcpu, index)) > > > return false; > > > - if (!lbr_desc->event && !intel_pmu_create_guest_lbr_event(vcpu)) > > > + if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) > > > goto dummy; ... > > AFAICT, event contention would lead to a #GP crash in the host due to > > lbr_desc->event being dereferenced, no? > > a #GP crash in the host ?Can you share more understanding about it ? The original code is will dereference a null lbr_desc->event if intel_pmu_create_guest_lbr_event() fails. if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) <- falls through goto dummy; /* * Disable irq to ensure the LBR feature doesn't get reclaimed by the * host at the time the value is read from the msr, and this avoids the * host LBR value to be leaked to the guest. If LBR has been reclaimed, * return 0 on guest reads. */ local_irq_disable(); if (lbr_desc->event->state == PERF_EVENT_STATE_ACTIVE) { <--------- kaboom if (read) rdmsrl(index, msr_info->data); else wrmsrl(index, msr_info->data); __set_bit(INTEL_PMC_IDX_FIXED_VLBR, vcpu_to_pmu(vcpu)->pmc_in_use); local_irq_enable(); return true; }
diff --git a/arch/x86/kvm/vmx/pmu_intel.c b/arch/x86/kvm/vmx/pmu_intel.c index d1df618cb7de..d6a5fe19ff09 100644 --- a/arch/x86/kvm/vmx/pmu_intel.c +++ b/arch/x86/kvm/vmx/pmu_intel.c @@ -320,7 +320,7 @@ static bool intel_pmu_handle_lbr_msrs_access(struct kvm_vcpu *vcpu, if (!intel_pmu_is_valid_lbr_msr(vcpu, index)) return false; - if (!lbr_desc->event && !intel_pmu_create_guest_lbr_event(vcpu)) + if (!lbr_desc->event && intel_pmu_create_guest_lbr_event(vcpu)) goto dummy; /*
If lbr_desc->event is successfully created, the intel_pmu_create_ guest_lbr_event() will return 0, otherwise it will return -ENOENT, and then jump to LBR msrs dummy handling. Fixes: 1b5ac3226a1a ("KVM: vmx/pmu: Pass-through LBR msrs when the guest LBR event is ACTIVE") Signed-off-by: Like Xu <like.xu@linux.intel.com> --- arch/x86/kvm/vmx/pmu_intel.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)