[PATCH v7 00/20] ARM64 PMU Partitioning
James Clark
james.clark at linaro.org
Mon May 11 07:57:13 PDT 2026
On 04/05/2026 10:17 pm, Colton Lewis wrote:
> This series creates a new PMU scheme on ARM, a partitioned PMU that
> allows reserving a subset of counters for more direct guest access,
> significantly reducing overhead. More details, including performance
> benchmarks, can be read in the v1 cover letter linked below.
>
> An overview of what this series accomplishes was presented at KVM
> Forum 2025. Slides [1] and video [2] are linked below.
>
> After a few false starts, meeting with Will Deacon and Mark Rutland to
> discuss implementation ideas, and a few more false starts, I finally
> have an implementation of dynamic counter reservation that works
> without disrupting host perf too much. Now the host only loses access
> to the guest counters when a vCPU resides on the CPU.
>
> The key was creating perf_pmu_resched_update, which behaves exactly
> like perf_pmu_resched except it takes a callback to call in between
> when the perf events are scheduled out and when they are scheduled
> back in. That allows us to update the PMU's available counters when we
> know they are not currently in use without needing to expose private
> perf core functions and triple check they are not being called in a
> way that violates existing assumptions.
>
> Because this introduces a possibility of perf reschedule during vCPU
> load, I've optimized to only do that operation if there are host
> events occupying the intended guest counters at the time of the load.
>
> The kernel command line parameter for the driver still exists, but now
> only defines an upper limit of counters the guest might use rather
> than taking those counters from the host permanently.
>
> v7:
>
> * Implement dynamic counter reservation as described above. One side
> effect is the PMUv3 driver now needs much fewer changes to enforce
> the boundary.
>
> * Move register accesses out of fast path for non-FGT hardware. The
> performance impact was negligible and this moves bloat out of the
> fast path and allows a more reliable design with more code sharing.
>
> * Make PMCCNTR a special case in the context swap again because trying
> to access it with PMXEVCNTR is undefined.
>
> * Fix a bug where kvm_pmu_guest_counter_mask was using & instead of |.
>
> * Re-expose the dedicated instruction counter to the host since it was
> decided the guest will not own it.
>
> * Change the global armv8pmu_reserved_host_counters to
> armv8pmu_is_partitoned because it was only used in boolean checks.
>
> * Fix typo in vcpu attribute commit so the spelling of the flag in the
> commit message matches the code.
>
> * Rebase to v7.0-rc7
>
> v6:
> https://lore.kernel.org/kvmarm/20260209221414.2169465-1-coltonlewis@google.com/
>
> v5:
> https://lore.kernel.org/kvmarm/20251209205121.1871534-1-coltonlewis@google.com/
>
> v4:
> https://lore.kernel.org/kvmarm/20250714225917.1396543-1-coltonlewis@google.com/
>
> v3:
> https://lore.kernel.org/kvm/20250626200459.1153955-1-coltonlewis@google.com/
>
> v2:
> https://lore.kernel.org/kvm/20250620221326.1261128-1-coltonlewis@google.com/
>
> v1:
> https://lore.kernel.org/kvm/20250602192702.2125115-1-coltonlewis@google.com/
>
> [1] https://gitlab.com/qemu-project/kvm-forum/-/raw/main/_attachments/2025/Optimizing__itvHkhc.pdf
> [2] https://www.youtube.com/watch?v=YRzZ8jMIA6M&list=PLW3ep1uCIRfxwmllXTOA2txfDWN6vUOHp&index=9
>
> Colton Lewis (19):
> arm64: cpufeature: Add cpucap for HPMN0
> KVM: arm64: Reorganize PMU functions
> perf: arm_pmuv3: Generalize counter bitmasks
> perf: arm_pmuv3: Check cntr_mask before using pmccntr
> perf: arm_pmuv3: Add method to partition the PMU
> KVM: arm64: Set up FGT for Partitioned PMU
> KVM: arm64: Add Partitioned PMU register trap handlers
> KVM: arm64: Set up MDCR_EL2 to handle a Partitioned PMU
> KVM: arm64: Context swap Partitioned PMU guest registers
> KVM: arm64: Enforce PMU event filter at vcpu_load()
> perf: Add perf_pmu_resched_update()
> KVM: arm64: Apply dynamic guest counter reservations
> KVM: arm64: Implement lazy PMU context swaps
> perf: arm_pmuv3: Handle IRQs for Partitioned PMU guest counters
> KVM: arm64: Detect overflows for the Partitioned PMU
> KVM: arm64: Add vCPU device attr to partition the PMU
> KVM: selftests: Add find_bit to KVM library
> KVM: arm64: selftests: Add test case for Partitioned PMU
> KVM: arm64: selftests: Relax testing for exceptions when partitioned
>
> Marc Zyngier (1):
> KVM: arm64: Reorganize PMU includes
>
> arch/arm/include/asm/arm_pmuv3.h | 18 +
> arch/arm64/include/asm/arm_pmuv3.h | 12 +-
> arch/arm64/include/asm/kvm_host.h | 17 +-
> arch/arm64/include/asm/kvm_types.h | 6 +-
> arch/arm64/include/uapi/asm/kvm.h | 2 +
> arch/arm64/kernel/cpufeature.c | 8 +
> arch/arm64/kvm/Makefile | 2 +-
> arch/arm64/kvm/arm.c | 2 +
> arch/arm64/kvm/config.c | 41 +-
> arch/arm64/kvm/debug.c | 31 +-
> arch/arm64/kvm/pmu-direct.c | 494 ++++++++++++
> arch/arm64/kvm/pmu-emul.c | 674 +----------------
> arch/arm64/kvm/pmu.c | 701 ++++++++++++++++++
> arch/arm64/kvm/sys_regs.c | 250 ++++++-
> arch/arm64/tools/cpucaps | 1 +
> arch/arm64/tools/sysreg | 6 +-
> drivers/perf/arm_pmuv3.c | 111 ++-
> include/kvm/arm_pmu.h | 110 +++
> include/linux/perf/arm_pmu.h | 3 +
> include/linux/perf/arm_pmuv3.h | 14 +-
> include/linux/perf_event.h | 3 +
> kernel/events/core.c | 28 +-
> tools/testing/selftests/kvm/Makefile.kvm | 1 +
> .../selftests/kvm/arm64/vpmu_counter_access.c | 112 ++-
> tools/testing/selftests/kvm/lib/find_bit.c | 1 +
> 25 files changed, 1861 insertions(+), 787 deletions(-)
> create mode 100644 arch/arm64/kvm/pmu-direct.c
> create mode 100644 tools/testing/selftests/kvm/lib/find_bit.c
>
>
> base-commit: 591cd656a1bf5ea94a222af5ef2ee76df029c1d2
> --
> 2.54.0.545.g6539524ca2-goog
I tested it a bit and ran the kselftests and it all seems to be working
ok. Some of the critical sashiko comments look like they are worth
looking into though:
https://sashiko.dev/#/patchset/20260504211813.1804997-1-coltonlewis%40google.com
For example writing to PMCR_EL0.P from EL2 resets the host's counters,
even if it's KVM doing it after trapping a write from the guest.
More information about the linux-arm-kernel
mailing list