]> git.ipfire.org Git - thirdparty/linux.git/commit
KVM: x86/pmu: Load/put mediated PMU context when entering/exiting guest
authorDapeng Mi <dapeng1.mi@linux.intel.com>
Sat, 6 Dec 2025 00:17:03 +0000 (16:17 -0800)
committerSean Christopherson <seanjc@google.com>
Thu, 8 Jan 2026 19:52:11 +0000 (11:52 -0800)
commit56bb2736975068cc03648718bb8e50a456ce7173
treecf4973ef59813cecc52d47b290684751c85cd5c2
parenta2f4ba534cc5d681a2d017c82e282bb32d8447df
KVM: x86/pmu: Load/put mediated PMU context when entering/exiting guest

Implement the PMU "world switch" between host perf and guest mediated PMU.
When loading guest state, call into perf to switch from host to guest, and
then load guest state into hardware, and then reverse those actions when
putting guest state.

On the KVM side, when loading guest state, zero PERF_GLOBAL_CTRL to ensure
all counters are disabled, then load selectors and counters, and finally
call into vendor code to load control/status information.  While VMX and
SVM use different mechanisms to avoid counting host activity while guest
controls are loaded, both implementations require PERF_GLOBAL_CTRL to be
zeroed when the event selectors are in flux.

When putting guest state, reverse the order, and save and zero controls
and status prior to saving+zeroing selectors and counters.  Defer clearing
PERF_GLOBAL_CTRL to vendor code, as only SVM needs to manually clear the
MSR; VMX configures PERF_GLOBAL_CTRL to be atomically cleared by the CPU
on VM-Exit.

Handle the difference in MSR layouts between Intel and AMD by communicating
the bases and stride via kvm_pmu_ops.  Because KVM requires Intel v4 (and
full-width writes) and AMD v2, the MSRs to load/save are constant for a
given vendor, i.e. do not vary based on the guest PMU, and do not vary
based on host PMU (because KVM will simply disable mediated PMU support if
the necessary MSRs are unsupported).

Except for retrieving the guest's PERF_GLOBAL_CTRL, which needs to be read
before invoking any fastpath handler (spoiler alert), perform the context
switch around KVM's inner run loop.  State only needs to be synchronized
from hardware before KVM can access the software "caches".

Note, VMX already grabs the guest's PERF_GLOBAL_CTRL immediately after
VM-Exit, as hardware saves value into the VMCS.

Co-developed-by: Mingwei Zhang <mizhang@google.com>
Signed-off-by: Mingwei Zhang <mizhang@google.com>
Co-developed-by: Sandipan Das <sandipan.das@amd.com>
Signed-off-by: Sandipan Das <sandipan.das@amd.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Tested-by: Xudong Hao <xudong.hao@intel.com>
Co-developed-by: Sean Christopherson <seanjc@google.com>
Tested-by: Manali Shukla <manali.shukla@amd.com>
Link: https://patch.msgid.link/20251206001720.468579-28-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>
arch/x86/include/asm/kvm-x86-pmu-ops.h
arch/x86/include/asm/msr-index.h
arch/x86/kvm/pmu.c
arch/x86/kvm/pmu.h
arch/x86/kvm/svm/pmu.c
arch/x86/kvm/svm/svm.c
arch/x86/kvm/vmx/pmu_intel.c
arch/x86/kvm/x86.c