Sample the latency of memory accesses in dirty_log_perf_test and report summary stats to give a picture of the latency distribution. Specifically, focus on the right tail with the 50th, 90th, and 99th percentile reported in ns. v3: * Move timing functions to lib/$ARCH/processor.c * Add memory barriers to prevent reordering measured memory accesses * Remove floating point from conversion functions * Change command line flag to -l for latency and provide help summary * Do not show latency measurements unless -l is given with argument * Change output formatting to seconds for consistency with other output v2: https://lore.kernel.org/kvm/20230316222752.1911001-1-coltonlewis@xxxxxxxxxx/ v1: https://lore.kernel.org/kvm/20221115173258.2530923-1-coltonlewis@xxxxxxxxxx/ Colton Lewis (2): KVM: selftests: Provide generic way to read system counter KVM: selftests: Print summary stats of memory latency distribution .../selftests/kvm/access_tracking_perf_test.c | 3 +- .../selftests/kvm/demand_paging_test.c | 2 +- .../selftests/kvm/dirty_log_perf_test.c | 21 +++++- .../selftests/kvm/include/aarch64/processor.h | 13 ++++ .../testing/selftests/kvm/include/memstress.h | 10 ++- .../selftests/kvm/include/x86_64/processor.h | 13 ++++ .../selftests/kvm/lib/aarch64/processor.c | 12 ++++ tools/testing/selftests/kvm/lib/memstress.c | 68 ++++++++++++++++--- .../selftests/kvm/lib/x86_64/processor.c | 13 ++++ .../kvm/memslot_modification_stress_test.c | 2 +- .../kvm/system_counter_offset_test.c | 10 +-- 11 files changed, 143 insertions(+), 24 deletions(-) -- 2.40.0.348.gf938b09366-goog
