Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.35 as permitted sender) client-ip=23.128.96.35;
Reply-To: Sean Christopherson <seanjc@google.com>
Date:   Fri,  3 Nov 2023 16:05:35 -0700
Mime-Version: 1.0
Message-ID: <20231103230541.352265-1-seanjc@google.com>
Subject: [PATCH v2 0/6] KVM: x86/pmu: Clean up emulated PMC event handling
From:   Sean Christopherson <seanjc@google.com>
To:     Sean Christopherson <seanjc@google.com>,
        Paolo Bonzini <pbonzini@redhat.com>
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        Dapeng Mi <dapeng1.mi@linux.intel.com>,
        Mingwei Zhang <mizhang@google.com>,
        Roman Kagan <rkagan@amazon.de>,
        Jim Mattson <jmattson@google.com>,
        Like Xu <like.xu.linux@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

The ultimate goal of this series is to track emulated counter events using
a dedicated variable instead of trying to track the previous counter value.
Tracking the previous counter value is flawed as it takes a snapshot at
every emulated event, but only checks for overflow prior to VM-Enter, i.e.
KVM could miss an overflow if KVM ever supports emulating event types that
can occur multiple times in a single VM-Exit.

And as Mingwei root caused, emulating overflow while the perf event is
running can result in duplicate overflow events, e.g. if the perf event
overflows between taking and checking the snapshot.  This bug is largely
masked now that KVM correctly sets LVT_MASK when delivering PMIs, but it's
still a bug, e.g. could cause problems if there are other side effects.

Patches 1-5 are (some loosely, some tightly) related fixes and cleanups to
simplify the emulated counter approach implementation.  The fixes are
tagged for stable as usersepace could cause some weirdness around perf
events, but I doubt any real world VMM is actually affected.

Dapeng, I intentionally omitted your Reviewed-by from the last patch as
the change from v1 isn't trivial.

v2:
 - Collect reviews. [Dapeng]
 - Emulate overflow *after* pausing perf event. [Mingwei]

v1: https://lore.kernel.org/all/20231023234000.2499267-1-seanjc@google.com

Sean Christopherson (6):
  KVM: x86/pmu: Move PMU reset logic to common x86 code
  KVM: x86/pmu: Reset the PMU, i.e. stop counters, before refreshing
  KVM: x86/pmu: Stop calling kvm_pmu_reset() at RESET (it's redundant)
  KVM: x86/pmu: Remove manual clearing of fields in kvm_pmu_init()
  KVM: x86/pmu: Update sample period in pmc_write_counter()
  KVM: x86/pmu: Track emulated counter events instead of previous
    counter

 arch/x86/include/asm/kvm-x86-pmu-ops.h |   2 +-
 arch/x86/include/asm/kvm_host.h        |  17 ++-
 arch/x86/kvm/pmu.c                     | 140 +++++++++++++++++++++----
 arch/x86/kvm/pmu.h                     |  47 +--------
 arch/x86/kvm/svm/pmu.c                 |  17 ---
 arch/x86/kvm/vmx/pmu_intel.c           |  22 ----
 arch/x86/kvm/x86.c                     |   1 -
 7 files changed, 137 insertions(+), 109 deletions(-)


base-commit: 45b890f7689eb0aba454fc5831d2d79763781677
-- 
2.42.0.869.gea05f2083d-goog