On Mon, Sep 14, 2020 at 8:18 PM Namhyung Kim <namhy...@kernel.org> wrote: > > Hello, > > I've found and fixed a bunch of memory leaks during perf pmu and > metric tests with address sanitizer. Before this, the tests were > mostly failed due to the leaks since ASAN makes it return non-zero. > > Now I'm seeing no error with ASAN like below: > > $ ./perf test pmu metric > 9: Parse perf pmu format : Ok > 10: PMU events : > 10.1: PMU event table sanity : Ok > 10.2: PMU event map aliases : Ok > 10.3: Parsing of PMU event table metrics : Skip (some > metrics failed) > 10.4: Parsing of PMU event table metrics with fake PMUs : Ok > 67: Parse and process metrics : Ok > > The failure in 10.3 seems due to parse errors like below: > > Multiple errors dropping message: unknown term 'filter_opc' for pmu > 'uncore_cbox_0' > (valid terms: > event,edge,inv,umask,cmask,config,config1,config2,name,period,freq, > > branch_type,time,call-graph,stack-size,no-inherit,inherit,max-stack, > > nr,no-overwrite,overwrite,driver-config,percore,aux-output,aux-sample-size) > > > Parse event failed metric 'DRAM_Parallel_Reads' id > 'arb/event=0x80,umask=0x2,thresh=1/' > expr 'arb@event\=0x80\,umask\=0x2@ / > arb@event\=0x80\,umask\=0x2\,thresh\=1@' > Error string 'unknown term 'thresh' for pmu 'uncore_arb'' help > 'valid terms: > event,edge,inv,umask,cmask,config,config1,config2,name,period,freq, > > branch_type,time,call-graph,stack-size,no-inherit,inherit,max-stack, > > nr,no-overwrite,overwrite,driver-config,percore,aux-output,aux-sample-size'
The 10.3 failure seems to be a problem in the skl metric DRAM_Parallel_Reads: https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/perf/pmu-events/arch/x86/skylake/skl-metrics.json?h=perf/core#n319 arb@event\\=0x80\\,umask\\=0x2@ / arb@event\\=0x80\\,umask\\=0x2\\,thresh\\=1@ The test failure message is: Parse event failed metric 'DRAM_Parallel_Reads' id 'arb/event=0x80,umask=0x2,thresh=1/' expr 'arb@event\=0x80\,umask\=0x2@ / arb@event\=0x80\,umask\=0x2\,thresh\=1@' Error string 'unknown term 'thresh' for pmu 'uncore_arb'' help 'valid terms: event,edge,inv,umask,cmask,config,config1,config2,name,period,freq,branch_type,time,call-graph,stack-size,no-inherit,inherit,max-stack,nr,no-overwrite,overwrite,driver-config,percore,aux-output,aux-sample-size' The 01.org version of this from: https://download.01.org/perfmon/TMA_Metrics.xlsx is: UNC_ARB_TRK_OCCUPANCY.DATA_READ / UNC_ARB_TRK_OCCUPANCY.DATA_READ:c1 It seems that :c1 has been translated into thresh=1 but that thresh doesn't exist as a format option (just "cmask edge event inv umask" are present). I wonder if Andi or Jin you could look into this broken metric? Thanks, Ian > * Changes from v1: > - Add 'Acked-by: Jiri' > > > The patches are also available at 'perf/metric-fix-v2' branch on > > git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git > > Thanks > Namhyung > > > Namhyung Kim (11): > perf metric: Fix some memory leaks > perf metric: Fix some memory leaks - part 2 > perf evlist: Fix cpu/thread map leak > perf parse-event: Fix cpu map leaks > perf parse-event: Fix memory leak in evsel->unit > perf test: Fix memory leaks in parse-metric test > perf metric: Release expr_parse_ctx after testing > perf metric: Free metric when it failed to resolve > perf metric: Do not free metric when failed to resolve > perf test: Free aliases for PMU event map aliases test > perf test: Free formats for perf pmu parse test > > tools/perf/tests/parse-metric.c | 14 ++++++++----- > tools/perf/tests/pmu-events.c | 5 +++++ > tools/perf/tests/pmu.c | 1 + > tools/perf/util/evlist.c | 11 ++++++++--- > tools/perf/util/metricgroup.c | 35 +++++++++++++++++++++++---------- > tools/perf/util/parse-events.c | 9 +++++++-- > tools/perf/util/pmu.c | 13 +++++++++++- > tools/perf/util/pmu.h | 2 ++ > tools/perf/util/stat-shadow.c | 8 +++++--- > 9 files changed, 74 insertions(+), 24 deletions(-) > > -- > 2.28.0.618.gf4bc123cb7-goog >
