- 27 May, 2023 28 commits
-
-
Ian Rogers authored
perf_pmu__is_hybrid implicitly uses the hybrid PMU list. Instead return false if hybrid isn't present, if it is then see if any evsel's PMUs are core. Reviewed-by:
Kan Liang <kan.liang@linux.intel.com> Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ali Saidi <alisaidi@amazon.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Dmitrii Dolgov <9erthalion6@gmail.com> Cc: Huacai Chen <chenhuacai@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jing Zhang <renyu.zj@linux.alibaba.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.g.garry@oracle.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kang Minchul <tegongkang@gmail.com> Cc: Leo Yan <leo.yan@linaro.org> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mike Leach <mike.leach@linaro.org> Cc: Ming Wang <wangming01@loongson.cn> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ravi Bangoria <ravi.bangoria@amd.com> Cc: Rob Herring <robh@kernel.org> Cc: Sandipan Das <sandipan.das@amd.com> Cc: Sean Christopherson <seanjc@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Thomas Richter <tmricht@linux.ibm.com> Cc: Will Deacon <will@kernel.org> Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Link: https://lore.kernel.org/r/20230527072210.2900565-23-irogers@google.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Ian Rogers authored
Switch from perf_pmu__is_hybrid to avoid implicitly using the hybrid PMU list. Reviewed-by:
-
Ian Rogers authored
Avoid perf_pmu__for_each_hybrid_pmu by iterating all PMUs are dumping the core ones. This will eventually allow removal of the hybrid PMU list. Reviewed-by:
-
Ian Rogers authored
Short-cut when has_hybrid is false, otherwise return if the evsel's PMU is core. Add a comment for the some what surprising no PMU cases of hardware and legacy cache events. Reviewed-by:
-
Ian Rogers authored
Avoid perf_pmu__for_each_hybrid_pmu in hybrid_topology__new by scanning all PMUs and processing the is_core ones. Add early exit for non-hybrid. Reviewed-by:
-
Ian Rogers authored
Rather than iterating over a separate hybrid list, iterate all PMUs with the hybrid ones having is_core as true. Reviewed-by:
-
Ian Rogers authored
Rather than list empty on perf_pmu__hybrid_pmus, detect if any core PMUs match the hybrid name. Computed values held in statics to avoid recomputation. Reviewed-by:
-
Ian Rogers authored
perf_pmu__hybrid_mounted is used to detect whether cpu_core or cpu_atom is mounted with a non-empty cpus file by pmu_lookup. Discussion [1] showed the empty cpus file check to be redundant and so pmu_lookup needn't have a call to perf_pmu__hybrid_mounted. Checking hybrid_mounted in pmu_is_uncore is redundant as the next cpumask read will fail returning false. Reduce the scope of perf_pmu__find_hybrid_pmu by making it static. [1] https://lore.kernel.org/lkml/20230524221831.1741381-17-irogers@google.com/Reviewed-by:
-
Ian Rogers authored
Function is only used in printout, reduce scope to stat-display.c. Remove the now empty evlist-hybrid.c and evlist-hybrid.h. Reviewed-by:
-
Ian Rogers authored
__evlist__add_default adds a cycles event to a typically empty evlist and was extended for hybrid with evlist__add_default_hybrid, as more than 1 PMU was necessary. Rather than have dedicated logic for the cycles event, this change switches to parsing 'cycles:P' which will handle wildcarding the PMUs appropriately for hybrid. Reviewed-by:
-
Ian Rogers authored
Parse events now corrects PMU groups in parse_events__sort_events_and_fix_groups and so this warning is no longer possible. Reviewed-by:
-
Ian Rogers authored
In commit 1d3351e6 ("perf tools: Enable on a list of CPUs for hybrid") perf on hybrid will warn if a user requested CPU doesn't match the PMU of the given event but only for hybrid PMUs. Make the logic generic for all PMUs and remove the hybrid logic. Warn if a CPU is requested that isn't present/offline for events not on the core. Warn if a CPU is requested for a core PMU, but the CPU isn't within the cpu map of that PMU. For example on a 16 (0-15) CPU system: ``` $ perf stat -e imc_free_running/data_read/,cycles -C 16 true WARNING: A requested CPU in '16' is not supported by PMU 'uncore_imc_free_running_1' (CPUs 0-15) for event 'imc_free_running/data_read/' WARNING: A requested CPU in '16' is not supported by PMU 'uncore_imc_free_running_0' (CPUs 0-15) for event 'imc_free_running/data_read/' WARNING: A requested CPU in '16' is not supported by PMU 'cpu' (CPUs 0-15) for event 'cycles' Performance counter stats for 'CPU(s) 16': <not supported> MiB imc_free_running/data_read/ <not supported> cycles 0.000575312 seconds time elapsed ``` Remove evlist__fix_hybrid_cpus that previously produced the warnings and also perf_pmu__cpus_match that worked with evlist__fix_hybrid_cpus to change CPU maps for hybrid CPUs, something that is no longer necessary as CPU map propagation properly intersects user requested CPUs with the core PMU's CPU map. Reviewed-by:
-
Ian Rogers authored
Previously this was used to modify CPU map propagation, but it is now unnecessary as map propagation ensure core PMUs only have valid PMUs in the CPU map from user requested CPUs. Reviewed-by:
-
Ian Rogers authored
Now that CPU map propagation only sets valid CPUs for core PMUs, there is no reason to disable "has_user_cpus" for hybrid. Reviewed-by:
-
Ian Rogers authored
The CPU map for a non-core PMU gives a default CPU value for perf_event_open. For core PMUs the CPU map lists all CPUs the evsel may be opened on. If there are >1 core PMU, the CPU maps will list the CPUs for that core PMU, but the user_requested_cpus may contain CPUs that are invalid for the PMU and cause perf_event_open to fail. To avoid this, when propagating the CPU map for core PMUs intersect it with the CPU map of the PMU (the evsel's "own_cpus"). Add comments to __perf_evlist__propagate_maps to explain its somewhat complex behavior. Fix the related comments for system_wide in struct perf_evsel. Reviewed-by:
-
Ian Rogers authored
A typical "cpu" PMU has no "cpus" or "cpumask" file meaning the CPU map is set to NULL, which also encodes an empty CPU map. Update pmu_cpumask so that if the "cpu" PMU fails to load a CPU map, use a default of all online PMUs. Remove const from cpu_map__online for the sake of reference counting. Reviewed-by:
-
Ian Rogers authored
The behaviour of handling cpu maps varies for core and other PMUs. For core PMUs the cpu map lists all valid CPUs, whereas for other PMUs the map is the default CPU. Add a flag in the evsel to indicate if a PMU is core to help with later interpreting of the cpu maps and populate it when the evsel is created during parsing. When propagating cpu maps, core PMUs should intersect the cpu map of the PMU with the user requested one. Reviewed-by:
-
Ian Rogers authored
Cache is_pmu_core in the pmu to avoid recomputation. Reviewed-by:
-
Ian Rogers authored
is_arm_pmu_core detects a core PMU via the presence of a "cpus" file rather than a "cpumask" file. This pattern holds for hybrid PMUs so rename the function and remove redundant perf_pmu__is_hybrid tests. Add a new helper is_pmu_hybrid similar to is_pmu_core. Reviewed-by:
-
Ian Rogers authored
It is common in the code currently to test a map for "empty" when in fact the "any CPU"/dummy value of -1 is being sought. Add a new function to enable this and document the behavior of two other functions. The term "any CPU" comes from perf_event_open, where the value is consumed, but it is more typical in the code to see this value/map referred to as the dummy value. This could be misleading due to the dummy event and also dummy not being intention revealing, so it is hoped to migrate the code to referring to this as "any CPU". Reviewed-by:
-
Ian Rogers authored
Equality is a useful property to compare after merging and intersecting maps. Reviewed-by:
-
Ian Rogers authored
These accessors assume the map is non-null. Rewrite functions to use rather than direct accesses. This also fixes a build regression for REFCNT_CHECKING in the intersect function. Suggested-by:
-
Ian Rogers authored
This avoids picking up a system installed version of the perf python module. Signed-off-by:
-
Namhyung Kim authored
The recent --per-cache option test caused a problem. According to the option name, I think it should check args.per_cache instead of args.per_cache_instance. $ sudo ./perf test -v 99 99: perf stat JSON output linter : --- start --- test child forked, pid 3086101 Checking json output: no args [Success] Checking json output: system wide [Success] Checking json output: interval [Success] Checking json output: event [Success] Checking json output: per thread [Success] Checking json output: per node [Success] Checking json output: system wide no aggregation [Success] Checking json output: per core [Success] Checking json output: per cache_instance Test failed for input: ... Traceback (most recent call last): File "linux/tools/perf/tests/shell/lib/perf_json_output_lint.py", line 88, in <module> elif args.per_core or args.per_socket or args.per_node or args.per_die or args.per_cache_instance: AttributeError: 'Namespace' object has no attribute 'per_cache_instance' test child finished with -1 ---- end ---- perf stat JSON output linter: FAILED! Fixes: bfce728d ("pert tests: Add tests for new "perf stat --per-cache" aggregation option") Signed-off-by:Namhyung Kim <namhyung@kernel.org> Tested-by:
K Prateek Nayak <kprateek.nayak@amd.com> Acked-by:
-
Ian Rogers authored
Go from 4 suites to a single suite with 4 test cases. Reviewed-by:
-
Ian Rogers authored
The merge function gives the union of two cpu maps. Add an intersect function which is necessary, for example, when intersecting a PMUs supported CPUs with user requested. Reviewed-by:
-
Ian Rogers authored
Add metric group descriptions created by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.py The descriptions add some additional detail in perf list. Committer notes: Removed unrelated changes to tools/perf/pmu-events/arch/x86/mapfile.csv that removed AMD mappings and ended up breaking the build with things like: CC /tmp/build/perf-tools-next/pmu-events/pmu-events.o /tmp/build/perf-tools-next/pmu-events/pmu-events.c:23808:39: error: ‘pmu_metrics__amdzen4’ defined but not used [-Werror=unused-const-variable=] 23808 | static const struct compact_pmu_event pmu_metrics__amdzen4[] = { | ^~~~~~~~~~~~~~~~~~~~ /tmp/build/perf-tools-next/pmu-events/pmu-events.c:23316:39: error: ‘pmu_events__amdzen4’ defined but not used [-Werror=unused-const-variable=] 23316 | static const struct compact_pmu_event pmu_events__amdzen4[] = { | ^~~~~~~~~~~~~~~~~~~ Reviewed-by:
-
Ian Rogers authored
Metrics have a field where the groups they belong to are listed like the following from tools/perf/pmu-events/arch/x86/skylakex/skx-metrics.json: "MetricGroup": "PGO;TmaL1;TopdownL1;tma_L1_group", "MetricName": "tma_frontend_bound", The metric groups are shown in 'perf list' like the following where TopdownL1 is a metric group: TopdownL1: tma_backend_bound [This category represents fraction of slots where no uops are being delivered due to a lack of required resources for accepting new uops in the Backend] tma_bad_speculation [This category represents fraction of slots wasted due to incorrect speculations] tma_frontend_bound [This category represents fraction of slots where the processor's Frontend undersupplies its Backend] tma_retiring [This category represents fraction of slots utilized by useful work i.e. issued uops that eventually get retired] This patch adds support for a new json file in each model directory called metricgroups.json that comprises a dictionary containing entries that map from a metric group to a description: { ... "TopdownL1": "Metrics for top-down breakdown at level 1", ... } perf list is then updated to support this changing the above output to: TopdownL1: [Metrics for top-down breakdown at level 1] Committer notes: Added a (int) cast to the ARRAY_SIZE() introduced in this patch to address: /tmp/build/perf-tools-next/pmu-events/pmu-events.c: In function ‘describe_metricgroup’: /var/home/acme/git/perf-tools-next/tools/include/linux/kernel.h:102:25: error: overflow in conversion from ‘long unsigned int’ to ‘int’ changes value from ‘18446744073709551615’ to ‘-1’ [-Werror=overflow] 102 | #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) + __must_be_array(arr)) | ^ /tmp/build/perf-tools-next/pmu-events/pmu-events.c:61603:29: note: in expansion of macro ‘ARRAY_SIZE’ 61603 | int low = 0, high = ARRAY_SIZE(metricgroups) - 1; | ^~~~~~~~~~ cc1: all warnings being treated as errors Reviewed-by:
-
- 23 May, 2023 12 commits
-
-
K Prateek Nayak authored
Add tests for the new "--per-cache" option in 'perf stat' for CSV and JSON generation as well as for the JSON linting. Suggested-by:
Gautham Shenoy <gautham.shenoy@amd.com> Signed-off-by:
-
K Prateek Nayak authored
This patch adds support for "--per-cache" option for aggregation at a particular cache level and documents the same. Following is the output of 'perf stat' with aggregation at L3 for the event "ls_dmnd_fills_from_sys.ext_cache_remote" on a dual socket 3rd Generation EPYC Processor (2 x 64C/128T - 16 LLCs) when running hackbench pinned to 4 LLCs: $ sudo perf stat --per-cache=L3 -a -e ls_dmnd_fills_from_sys.ext_cache_remote -- \ taskset -c 0-15,64-79,128-143,192-207 \ perf bench sched messaging -p -t -l 100000 -g 8 ... Performance counter stats for 'system wide': S0-D0-L3-ID0 16 9,500,803 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID8 16 6,338,099 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID16 16 355,005 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID24 16 22,067 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID32 16 16,321 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID40 16 11,619 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID48 16 4,238 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID56 16 31,158 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID64 16 28,242,452 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID72 16 22,906,973 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID80 16 72,898 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID88 16 56,907 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID96 16 20,456 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID104 16 40,913 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID112 16 78,113 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID120 16 37,897 ls_dmnd_fills_from_sys.ext_cache_remote Also support 'perf stat record' and 'perf stat report' with the ability to specify a different cache level to aggregate data at when running 'perf stat report'. $ sudo perf stat record --per-cache=L2 -a -e ls_dmnd_fills_from_sys.ext_cache_remote -- \ taskset -c 0-15,64-79,128-143,192-207 \ perf bench sched messaging -p -t -l 100000 -g 8 ... Performance counter stats for 'system wide': S0-D0-L2-ID0 2 1,442,061 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID1 2 1,548,994 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID2 2 1,553,557 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID3 2 1,420,122 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID4 2 1,465,461 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID5 2 1,455,153 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID6 2 1,595,237 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID7 2 1,499,321 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L2-ID8 2 1,919,025 ls_dmnd_fills_from_sys.ext_cache_remote ... S1-D1-L2-ID127 2 21,295 ls_dmnd_fills_from_sys.ext_cache_remote $ sudo perf stat report --per-cache=L3 Performance counter stats for 'perf stat record --per-cache=L2 -a -e ls_dmnd_fills_from_sys.ext_cache_remote --\ taskset -c 0-15,64-79,128-143,192-207 \ perf bench sched messaging -p -t -l 100000 -g 8': S0-D0-L3-ID0 16 11,979,906 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID8 16 14,257,202 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID16 16 377,484 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID24 16 27,224 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID32 16 26,816 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID40 16 14,461 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID48 16 10,499 ls_dmnd_fills_from_sys.ext_cache_remote S0-D0-L3-ID56 16 53,817 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID64 16 27,361,987 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID72 16 37,299,024 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID80 16 84,125 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID88 16 64,561 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID96 16 13,403 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID104 16 20,138 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID112 16 93,220 ls_dmnd_fills_from_sys.ext_cache_remote S1-D1-L3-ID120 16 35,465 ls_dmnd_fills_from_sys.ext_cache_remote On the above system, the domain covered by S0-D0-L3-ID0 contains S0-D0-L2-ID0 to S0-D0-L2-ID7, the corresponding count for L3-ID0 is equal to the sum of counts for L2-ID0 to L2-ID7. Add documentation for the newly introduced "--per-cache" option. Suggested-by: -
K Prateek Nayak authored
When aggregating based on cache-topology, in addition to the aggregation mode, knowing the cache level at which data is aggregated is necessary to ensure consistency when running 'perf stat record' and later 'perf stat report'. Save the cache level for aggregation as a part of the env data that can be later retrieved when running perf stat report. Suggested-by:
-
K Prateek Nayak authored
Processors based on chiplet architecture, such as AMD EPYC and Hygon do not expose the chiplet details in the sysfs CPU topology information. However, this information can be derived from the per CPU cache level information from the sysfs. 'perf stat' has already supported aggregation based on topology information using core ID, socket ID, etc. It'll be useful to aggregate based on the cache topology to detect problems like imbalance and cache-to-cache sharing at various cache levels. This patch lays the foundation for aggregating data in 'perf stat' based on the processor's cache topology. The cmdline option to aggregate data based on the cache topology is added in Patch 4 of the series while this patch sets up all the necessary functions and variables required to support the new aggregation option. The patch also adds support to display per-cache aggregation, or save it as a JSON or CSV, as splitting it into a separate patch would break builds when compiling with "-Werror=switch-enum" where the compiler will complain about the lack of handling for the AGGR_CACHE case in the output functions. Committer notes: Don't use perf_stat_config in tools/perf/util/cpumap.c, this would make code that is in util/, thus not really specific to a single builtin, use a specific builtin config structure. Move the functions introduced in this patch from tools/perf/util/cpumap.c since it needs access to builtin specific and is not strictly needed to live in the util/ directory. With this 'perf test python' is back building. Suggested-by:
-
K Prateek Nayak authored
build_caches() builds the complete cache topology of the system by iterating over all CPU, building and comparing cache levels of each CPU, keeping only the unique ones at the end. Extract the unit that build the cache levels for a single CPU into a separate function. Expose this function, and the MAX_CACHE_LVL value to be used elsewhere in perf too. Signed-off-by:
-
Ian Rogers authored
Update tigerlake events to v1.12 including the new events MEM_LOAD_MISC_RETIRED.UC and SQ_MISC.BUS_LOCK. Metrics are updated to make TMA info metric names synchronized. Events and metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Update snowridgex to v1.21 that marks deprecated a number of events and adds improves descriptions. The events data was generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Update skylake events to v60 and skylakex events to v1.30, adding the events FP_ARITH_INST_RETIRED.4_FLOPS, FP_ARITH_INST_RETIRED.8_FLOPS, FP_ARITH_INST_RETIRED.SCALAR, FP_ARITH_INST_RETIRED.VECTOR and INT_MISC.CLEARS_COUNT. Metrics are updated to make TMA info metric names synchronized. Events and metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Update sapphirerapids events to v1.13 improving event descriptions. Metrics are updated to make TMA info metric names synchronized. Events and metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Metrics are updated to make TMA info metric names synchronized. Metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Metrics are updated to make TMA info metric names synchronized. Metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
Ian Rogers authored
Metrics are updated to make TMA info metric names synchronized. Metrics were generated by: https://github.com/intel/perfmon/blob/main/scripts/create_perf_json.pyReviewed-by:
-
