Re: [PATCH v2 3/5] KVM: x86/pmu: Refresh Host-Only/Guest-Only eventsel at nested transitions

[Jim Mattson]

On Fri, Jan 30, 2026 at 3:41 PM Yosry Ahmed <yosry.ahmed@xxxxxxxxx> wrote:
>
> January 30, 2026 at 3:30 PM, "Jim Mattson" <jmattson@xxxxxxxxxx> wrote:
>
>
> >
> > On Fri, Jan 30, 2026 at 7:26 AM Yosry Ahmed <yosry.ahmed@xxxxxxxxx> wrote:
> >
> > >
> > > On Thu, Jan 29, 2026 at 03:28:08PM -0800, Jim Mattson wrote:
> > >  Add amd_pmu_refresh_host_guest_eventsel_hw() to recalculate eventsel_hw for
> > >  all PMCs based on the current vCPU state. This is needed because Host-Only
> > >  and Guest-Only counters must be enabled/disabled at:
> > >
> > >  - SVME changes: When EFER.SVME is modified, counters with Guest-Only bits
> > >  need their hardware enable state updated.
> > >
> > >  - Nested transitions: When entering or leaving guest mode, Host-Only
> > >  counters should be disabled/enabled and Guest-Only counters should be
> > >  enabled/disabled accordingly.
> > >
> > >  Introduce svm_enter_guest_mode() and svm_leave_guest_mode() wrappers that
> > >  call enter_guest_mode()/leave_guest_mode() followed by the PMU refresh,
> > >  ensuring the PMU state stays synchronized with guest mode transitions.
> > >
> > >  Signed-off-by: Jim Mattson <jmattson@xxxxxxxxxx>
> > >  ---
> > >  arch/x86/kvm/svm/nested.c | 6 +++---
> > >  arch/x86/kvm/svm/pmu.c | 12 ++++++++++++
> > >  arch/x86/kvm/svm/svm.c | 2 ++
> > >  arch/x86/kvm/svm/svm.h | 17 +++++++++++++++++
> > >  4 files changed, 34 insertions(+), 3 deletions(-)
> > >
> > >  diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
> > >  index de90b104a0dd..a7d1901f256b 100644
> > >  --- a/arch/x86/kvm/svm/nested.c
> > >  +++ b/arch/x86/kvm/svm/nested.c
> > >  @@ -757,7 +757,7 @@ static void nested_vmcb02_prepare_control(struct vcpu_svm *svm,
> > >  nested_svm_transition_tlb_flush(vcpu);
> > >
> > >  /* Enter Guest-Mode */
> > >  - enter_guest_mode(vcpu);
> > >  + svm_enter_guest_mode(vcpu);
> > >
> > >  FWIW, I think this name is a bit confusing because we also have
> > >  enter_svm_guest_mode(). So we end up with:
> > >
> > >  enter_svm_guest_mode() -> nested_vmcb02_prepare_control() ->
> > >  svm_enter_guest_mode() -> enter_guest_mode()
> > >
> > >  I actually have another proposed change [1] that moves
> > >  enter_guest_mode() directly into enter_svm_guest_mode(), so the sequence
> > >  would end up being:
> > >
> > >  enter_svm_guest_mode() -> svm_enter_guest_mode() -> enter_guest_mode()
> > >
> > Yes, that is confusing. What if I renamed the existing function to
> > something like svm_nested_switch_to_vmcb02()?
> >
> > Alternatively, I could go back to introducing a new PMU_OP, call it
> > from {enter,leave}_guest_mode(), and drop the wrappers.
>
> We could just call amd_pmu_refresh_host_guest_eventsel_hw() every time we call enter_guest_mode() and leave_guest_mode(), which is more error-prone but there's already other things in that category.
>
> We could also call it from svm_switch_vmcb(), which will add some calls to extra places but I assume that would be fine?
>
> I personally prefer the former tbh, as it's otherwise easy to miss.

I'm not a fan of replicating code sequences. Maybe instead of a new
PMU_OP, I could introduce a new KVM_X86_OP, which would be
conceptually more general, even though the current usage would be the
same.

> >
> > >
> > > [1] https://lore.kernel.org/kvm/20260115011312.3675857-9-yosry.ahmed@xxxxxxxxx/
> > >
> >