On Tue, Nov 15, 2022 at 05:32:57PM +0000, Colton Lewis wrote: > Collect memory access latency measured in clock cycles. > > This introduces a dependency on the timers for ARM and x86. No other > architectures are implemented and their samples will all be 0. > > Because keeping all samples is impractical due to the space required > in some cases (pooled memory w/ 64 vcpus would be 64 GB/vcpu * 64 > vcpus * 250,000 samples/GB * 8 bytes/sample ~ 8 Gb extra memory just > for samples), resevior sampling is used to only keep a small number of nit: reservoir > samples per vcpu (1000 samples in this patch). Didn't see this before my previous comment. But, I guess it still applies: isn't it possible to know the number of events to store? to avoid the "100" obtained via trial and error. > > Resevoir sampling means despite keeping only a small number of > samples, each sample has an equal chance of making it to the > resevoir. Simple proofs of this can be found online. This makes the > resevoir a good representation of the distribution of samples and reservoir > enables calculation of reasonably accurate percentiles. > > All samples are stored in a statically allocated flat array for ease > of combining them later. Samples are stored at an offset in this array > calculated by the vcpu index (so vcpu 5 sample 10 would be stored at > address sample_times + 5 * vcpu_idx + 10). > > Signed-off-by: Colton Lewis <coltonlewis@xxxxxxxxxx> > --- > .../selftests/kvm/lib/perf_test_util.c | 34 +++++++++++++++++-- > 1 file changed, 32 insertions(+), 2 deletions(-) > > diff --git a/tools/testing/selftests/kvm/lib/perf_test_util.c b/tools/testing/selftests/kvm/lib/perf_test_util.c > index a48904b64e19..0311da76bae0 100644 > --- a/tools/testing/selftests/kvm/lib/perf_test_util.c > +++ b/tools/testing/selftests/kvm/lib/perf_test_util.c > @@ -4,6 +4,9 @@ > */ > #include <inttypes.h> > > +#if defined(__aarch64__) > +#include "aarch64/arch_timer.h" > +#endif > #include "kvm_util.h" > #include "perf_test_util.h" > #include "processor.h" > @@ -44,6 +47,18 @@ static struct kvm_vcpu *vcpus[KVM_MAX_VCPUS]; > /* Store all samples in a flat array so they can be easily sorted later. */ > uint64_t latency_samples[SAMPLE_CAPACITY]; > > +static uint64_t perf_test_timer_read(void) > +{ > +#if defined(__aarch64__) > + return timer_get_cntct(VIRTUAL); > +#elif defined(__x86_64__) > + return rdtsc(); > +#else > +#warn __func__ " is not implemented for this architecture, will return 0" > + return 0; > +#endif > +} > + > /* > * Continuously write to the first 8 bytes of each page in the > * specified region. > @@ -59,6 +74,10 @@ void perf_test_guest_code(uint32_t vcpu_idx) > int i; > struct guest_random_state rand_state = > new_guest_random_state(pta->random_seed + vcpu_idx); > + uint64_t *latency_samples_offset = latency_samples + SAMPLES_PER_VCPU * vcpu_idx; > + uint64_t count_before; > + uint64_t count_after; > + uint32_t maybe_sample; > > gva = vcpu_args->gva; > pages = vcpu_args->pages; > @@ -75,10 +94,21 @@ void perf_test_guest_code(uint32_t vcpu_idx) > > addr = gva + (page * pta->guest_page_size); > > - if (guest_random_u32(&rand_state) % 100 < pta->write_percent) > + if (guest_random_u32(&rand_state) % 100 < pta->write_percent) { > + count_before = perf_test_timer_read(); > *(uint64_t *)addr = 0x0123456789ABCDEF; > - else > + count_after = perf_test_timer_read(); > + } else { > + count_before = perf_test_timer_read(); > READ_ONCE(*(uint64_t *)addr); > + count_after = perf_test_timer_read(); "count_before ... ACCESS count_after" could be moved to some macro, e.g.,: t = MEASURE(READ_ONCE(*(uint64_t *)addr)); > + } > + > + maybe_sample = guest_random_u32(&rand_state) % (i + 1); > + if (i < SAMPLES_PER_VCPU) > + latency_samples_offset[i] = count_after - count_before; > + else if (maybe_sample < SAMPLES_PER_VCPU) > + latency_samples_offset[maybe_sample] = count_after - count_before; I would prefer these reservoir sampling details to be in a helper, e.g.,: reservoir_sample_record(t, i); > } > > GUEST_SYNC(1); > -- > 2.38.1.431.g37b22c650d-goog >