[v10,0/5] KVM: arm/arm64: add support for chained counters

Message

Andrew Murray June 17, 2019, 7:01 p.m. UTC

ARMv8 provides support for chained PMU counters, where an event type
of 0x001E is set for odd-numbered counters, the event counter will
increment by one for each overflow of the preceding even-numbered
counter. Let's emulate this in KVM by creating a 64 bit perf counter
when a user chains two emulated counters together.

Testing has been performed by hard-coding hwc->sample_period in
__hw_perf_event_init (arm_pmu.c) to a small value, this results in
regular overflows (for non sampling events). The following command
was then used to measure chained and non-chained instruction cycles:

perf stat -e armv8_pmuv3/long=1,inst_retired/u \
          -e armv8_pmuv3/long=0,inst_retired/u dd if=/dev/zero bs=1M \
          count=10 | gzip > /dev/null

The reported values were identical (and for non-chained was in the
same ballpark when running on a kernel without this patchset). Debug
was added to verify that the guest received overflow interrupts for
the chain counter.

The test was also repeated using the cycle counter (cycle:u).

For chained events we only support generating an overflow interrupt
on the high counter. We use the attributes of the low counter to
determine the attributes of the perf event.

Changes since v9:

 - Ensure only 32 bits of cycle counter is returned when !PMCR_LC

 - Add a helper to test for 64 bit counters (e.g. long cycle counter)

 - Rename kvm_pmu_pmc_is_high_counter to kvm_pmu_idx_is_high_counter to
   reflect arguments passed to it

Changes since v8:

 - Correctly calculate the sample_period for the cycle counter

 - Drop "arm64: perf: extract chain helper into header" patch

Changes since v7:

 - Remove pmc->bitmask

 - Remove a couple of instances of using kvm_pmu_get_canonical_pmc
   when not needed

 - Remove unused perf_event variable

Changes since v6:

 - Drop kvm_pmu_{get,set}_perf_event

 - Avoid duplicate work by using kvm_pmu_get_pair_counter_value inside
   kvm_pmu_stop_counter

 - Use GENMASK for 64bit mask

Changes since v5:

 - Use kvm_pmu_pmc_is_high_counter instead of open coding

 - Rename kvm_pmu_event_is_chained to kvm_pmu_idx_has_chain_evtype

 - Use kvm_pmu_get_canonical_pmc only where needed and reintroduce
   the kvm_pmu_{set, get}_perf_event functions

 - Drop masking of counter in kvm_pmu_get_pair_counter_value

 - Only initialise pmc once in kvm_pmu_create_perf_event and other
   minor changes.

Changes since v4:

 - Track pairs of chained counters with a bitmap instead of using
   a struct kvm_pmc_pair.

 - Rebase onto kvmarm/queue

Changes since v3:

 - Simplify approach by not creating events lazily and by introducing
   a struct kvm_pmc_pair to represent the relationship between
   adjacent counters.

 - Rebase onto v5.1-rc2

Changes since v2:

 - Rebased onto v5.0-rc7

 - Add check for cycle counter in correct patch

 - Minor style, naming and comment changes

 - Extract armv8pmu_evtype_is_chain from arch/arm64/kernel/perf_event.c
   into a common header that KVM can use

Changes since v1:

 - Rename kvm_pmu_{enable,disable}_counter to reflect that they can
   operate on multiple counters at once and use these functions where
   possible

 - Fix bugs with overflow handing, kvm_pmu_get_counter_value did not
   take into consideration the perf counter value overflowing the low
   counter

 - Ensure PMCCFILTR_EL0 is used when operating on the cycle counter

 - Rename kvm_pmu_reenable_enabled_{pair, single} and similar

 - Always create perf event disabled to simplify logic elsewhere

 - Move PMCNTENSET_EL0 test to kvm_pmu_enable_counter_mask


Andrew Murray (5):
  KVM: arm/arm64: rename kvm_pmu_{enable/disable}_counter functions
  KVM: arm/arm64: extract duplicated code to own function
  KVM: arm/arm64: re-create event when setting counter value
  KVM: arm/arm64: remove pmc->bitmask
  KVM: arm/arm64: support chained PMU counters

 arch/arm64/kvm/sys_regs.c |   4 +-
 include/kvm/arm_pmu.h     |  11 +-
 virt/kvm/arm/pmu.c        | 350 ++++++++++++++++++++++++++++++--------
 3 files changed, 291 insertions(+), 74 deletions(-)