While writing new PMUv3 event counter kvm-unit-tests I found some bugs related to the chained counter implementation: - the enable state of the high counter is not taken into account, - chain counters are not implemented along with SW_INCR type, - SW_INCR does not take into account the global enable state The last patch rather is an optimization that avoids manipulating non supported counters. Best Regards Eric This series can be found at: https://github.com/eauger/linux/tree/v5.5-rc7-pmu-fixes-v1 Test: Tested with kvm-unit-tests [1]: all tests now pass, at the exception of one sub-test in pmu-chain-promotion but this is a bug in test. Other testing at higher level (perf) appreciated. references: [1] KVM: arm64: PMUv3 Event Counter Tests (https://lore.kernel.org/kvmarm/c1831b6c-dc75-1bd3-6657-0375682c30af@xxxxxxxxxx/T/) History: RFC -> v1: - remove [RFC 3/3] KVM: arm64: pmu: Enforce PMEVTYPER evtCount size - add KVM: arm64: pmu: Only handle supported event counters - Take into account the enable state of the CHAIN high counter - revisit kvm_pmu_software_increment() implementation as suggested by Marc Eric Auger (4): KVM: arm64: pmu: Don't increment SW_INCR if PMCR.E is unset KVM: arm64: pmu: Don't mark a counter as chained if the odd one is disabled KVM: arm64: pmu: Fix chained SW_INCR counters KVM: arm64: pmu: Only handle supported event counters virt/kvm/arm/pmu.c | 114 +++++++++++++++++++++++++++------------------ 1 file changed, 69 insertions(+), 45 deletions(-) -- 2.20.1 _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
