- 27 Sep, 2023 3 commits
-
-
Ian Rogers authored
Dummy events are created with an attribute where the period and freq are zero. evsel__config will then see the uninitialized values and initialize them in evsel__default_freq_period. As fequency mode is used by default the dummy event would be set to use frequency mode. However, this has no effect on the dummy event but does cause unnecessary timers/interrupts. Avoid this overhead by setting the period to 1 for dummy events. evlist__add_aux_dummy calls evlist__add_dummy then sets freq=0 and period=1. This isn't necessary after this change and so the setting is removed. From Stephane: The dummy event is not counting anything. It is used to collect mmap records and avoid a race condition during the synthesize mmap phase of perf record. As such, it should not cause any overhead during active profiling. Yet, it did. Because of a bug the dummy event was programmed as a sampling event in frequency mode. Events in that mode incur more kernel overheads because on timer tick, the kernel has to look at the number of samples for each event and potentially adjust the sampling period to achieve the desired frequency. The dummy event was therefore adding a frequency event to task and ctx contexts we may otherwise not have any, e.g., perf record -a -e cpu/event=0x3c,period=10000000/. On each timer tick the perf_adjust_freq_unthr_context() is invoked and if ctx->nr_freq is non-zero, then the kernel will loop over ALL the events of the context looking for frequency mode ones. In doing, so it locks the context, and enable/disable the PMU of each hw event. If all the events of the context are in period mode, the kernel will have to traverse the list for nothing incurring overhead. The overhead is multiplied by a very large factor when this happens in a guest kernel. There is no need for the dummy event to be in frequency mode, it does not count anything and therefore should not cause extra overhead for no reason. Fixes: 5bae0250 ("perf evlist: Introduce perf_evlist__new_dummy constructor") Reported-by:
Stephane Eranian <eranian@google.com> Signed-off-by:
Ian Rogers <irogers@google.com> Acked-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Yang Jihong <yangjihong1@huawei.com> Cc: Kan Liang <kan.liang@linux.intel.com> Link: https://lore.kernel.org/r/20230916035640.1074422-1-irogers@google.comSigned-off-by:
Namhyung Kim <namhyung@kernel.org>
-
Charles Han authored
Remove the repeated word "of" in comments. Signed-off-by:
Charles Han <hanchunchao@inspur.com> Reviewed-by:
-
Ilkka Koskinen authored
This patch addresses review comments that were given for 705ed549 ("perf vendor events arm64: Add AmpereOne metrics") but didn't make it to the original patch [1][2] Changes include: A fix for backend_memory formula, use of standard metrics when possible, using #slots, renaming metrics to avoid spaces in the names, and cleanup. [1] https://lore.kernel.org/linux-perf-users/e9bdacb-a231-36af-6a2e-6918ee7effa@os.amperecomputing.com/ [2] https://lore.kernel.org/linux-perf-users/20230826192352.3043220-1-ilkka@os.amperecomputing.com/ Fixes: 705ed549 ("perf vendor events arm64: Add AmpereOne metrics") Signed-off-by:
Ilkka Koskinen <ilkka@os.amperecomputing.com> Reviewed-by:
-
- 21 Sep, 2023 2 commits
-
-
Veronika Molnarova authored
Machines with less then 4 CPUs weren't consistently triggering lock events required for the test. Skip the test on those machines. The limit of 4 CPUs is set as it generates around 100 lock events for a test. Signed-off-by:
Veronika Molnarova <vmolnaro@redhat.com> Acked-by:
Michael Petlan <mpetlan@redhat.com> Link: https://lore.kernel.org/r/20230919150419.23193-2-vmolnaro@redhat.comSigned-off-by:
-
Veronika Molnarova authored
The test was failing in specific scenarios due to imperfection of FP arithmetics. The `bc` command wasn't correctly rounding the result of division causing the failure. Replace the `bc` with `awk` which should work with more decimal places and add a threshold to catch any possible rounding errors. The acceptable rounding error is set to 0.01 when the test passes with a warning message. Signed-off-by:
-
- 20 Sep, 2023 1 commit
-
-
Xu Yang authored
The struct "pmu_events_table" has been changed after commit 2e255b4f (perf jevents: Group events by PMU, 2023-08-23). So there doesn't exist 'entries' in pmu_events_table anymore. This will align the members with that commit. Othewise, below errors will be printed when run jevent.py: pmu-events/pmu-events.c:5485:26: error: ‘struct pmu_metrics_table’ has no member named ‘entries’ 5485 | .entries = pmu_metrics__freescale_imx8dxl_sys, Signed-off-by:
Xu Yang <xu.yang_2@nxp.com> Reviewed-by:
-
- 18 Sep, 2023 7 commits
-
-
Ian Rogers authored
Fuzzing found that an invalid tracepoint name would create a memory leak with an address sanitizer build: ``` $ perf stat -e '*:o/' true event syntax error: '*:o/' \___ parser error Run 'perf list' for a list of valid events Usage: perf stat [<options>] [<command>] -e, --event <event> event selector. use 'perf list' to list available events ================================================================= ==59380==ERROR: LeakSanitizer: detected memory leaks Direct leak of 4 byte(s) in 2 object(s) allocated from: #0 0x7f38ac07077b in __interceptor_strdup ../../../../src/libsanitizer/asan/asan_interceptors.cpp:439 #1 0x55f2f41be73b in str util/parse-events.l:49 #2 0x55f2f41d08e8 in parse_events_lex util/parse-events.l:338 #3 0x55f2f41dc3b1 in parse_events_parse util/parse-events-bison.c:1464 #4 0x55f2f410b8b3 in parse_events__scanner util/parse-events.c:1822 #5 0x55f2f410d1b9 in __parse_events util/parse-events.c:2094 #6 0x55f2f410e57f in parse_events_option util/parse-events.c:2279 #7 0x55f2f4427b56 in get_value tools/lib/subcmd/parse-options.c:251 #8 0x55f2f4428d98 in parse_short_opt tools/lib/subcmd/parse-options.c:351 #9 0x55f2f4429d80 in parse_options_step tools/lib/subcmd/parse-options.c:539 #10 0x55f2f442acb9 in parse_options_subcommand tools/lib/subcmd/parse-options.c:654 #11 0x55f2f3ec99fc in cmd_stat tools/perf/builtin-stat.c:2501 #12 0x55f2f4093289 in run_builtin tools/perf/perf.c:322 #13 0x55f2f40937f5 in handle_internal_command tools/perf/perf.c:375 #14 0x55f2f4093bbd in run_argv tools/perf/perf.c:419 #15 0x55f2f409412b in main tools/perf/perf.c:535 SUMMARY: AddressSanitizer: 4 byte(s) leaked in 2 allocation(s). ``` Fix by adding the missing destructor. Fixes: 865582c3 ("perf tools: Adds the tracepoint name parsing support") Signed-off-by: -
Ian Rogers authored
Use perf version to detect whether BPF skeletons were enabled in a build rather than a failing perf record. Signed-off-by:
-
Ian Rogers authored
Add to run variable. Signed-off-by:
-
Ian Rogers authored
Fix a target name and set BUILD_BPF_SKEL to 0 rather than 1. Signed-off-by:
-
Ian Rogers authored
LIBBPF is dependent on zlib so move the NO_ZLIB and feature check early to avoid statically building when zlib is disabled. This avoids a linkage failure with perf and static libbpf when zlib isn't specified. Move BUILD_BPF_SKEL logic to one place and if not defined set BUILD_BPF_SKEL to 1. Detect dependencies of building with BPF skeletons and warn/disable if the dependencies aren't present. Change Makefile.perf to contain BPF skeleton logic dependent on the Makefile.config result and refresh the comment about BUILD_BPF_SKEL. Signed-off-by:
-
Ian Rogers authored
Add status for BPF skeletons, to see if a build has them enabled: ``` $ perf version --build-options perf version 6.6.rc1.g0381ae36d1a6 dwarf: [ OFF ] # HAVE_DWARF_SUPPORT dwarf_getlocations: [ OFF ] # HAVE_DWARF_GETLOCATIONS_SUPPORT syscall_table: [ on ] # HAVE_SYSCALL_TABLE_SUPPORT libbfd: [ OFF ] # HAVE_LIBBFD_SUPPORT debuginfod: [ OFF ] # HAVE_DEBUGINFOD_SUPPORT libelf: [ OFF ] # HAVE_LIBELF_SUPPORT libnuma: [ OFF ] # HAVE_LIBNUMA_SUPPORT numa_num_possible_cpus: [ OFF ] # HAVE_LIBNUMA_SUPPORT libperl: [ on ] # HAVE_LIBPERL_SUPPORT libpython: [ on ] # HAVE_LIBPYTHON_SUPPORT libslang: [ on ] # HAVE_SLANG_SUPPORT libcrypto: [ on ] # HAVE_LIBCRYPTO_SUPPORT libunwind: [ OFF ] # HAVE_LIBUNWIND_SUPPORT libdw-dwarf-unwind: [ OFF ] # HAVE_DWARF_SUPPORT zlib: [ on ] # HAVE_ZLIB_SUPPORT lzma: [ on ] # HAVE_LZMA_SUPPORT get_cpuid: [ on ] # HAVE_AUXTRACE_SUPPORT bpf: [ OFF ] # HAVE_LIBBPF_SUPPORT aio: [ on ] # HAVE_AIO_SUPPORT zstd: [ on ] # HAVE_ZSTD_SUPPORT libpfm4: [ on ] # HAVE_LIBPFM libtraceevent: [ on ] # HAVE_LIBTRACEEVENT bpf_skeletons: [ OFF ] # HAVE_BPF_SKEL ``` Signed-off-by: -
Yang Li authored
./tools/perf/util/bpf_kwork_top.c:120:53-58: WARNING: conversion to bool not needed here Signed-off-by:
Yang Li <yang.lee@linux.alibaba.com> Cc: bpf@vger.kernel.org Link: https://lore.kernel.org/r/20230915063832.120274-1-yang.lee@linux.alibaba.comSigned-off-by:
-
- 17 Sep, 2023 1 commit
-
-
Yang Jihong authored
For kernel that supports PERF_FORMAT_LOST, attr->read_format has PERF_FORMAT_LOST bit. Update expected value of attr->read_format of test-record-dummy-C0 for this scenario. Before: # ./perf test 17 -vv 17: Setup struct perf_event_attr : --- start --- test child forked, pid 1609441 <SNIP> running './tests/attr/test-record-dummy-C0' 'PERF_TEST_ATTR=/tmp/tmpm3s60aji ./perf record -o /tmp/tmpm3s60aji/perf.data --no-bpf-event -e dummy -C 0 kill >/dev/null 2>&1' ret '1', expected '1' expected read_format=4, got 20 FAILED './tests/attr/test-record-dummy-C0' - match failure test child finished with -1 ---- end ---- Setup struct perf_event_attr: FAILED! After: # ./perf test 17 -vv 17: Setup struct perf_event_attr : --- start --- test child forked, pid 1609441 <SNIP> running './tests/attr/test-record-dummy-C0' 'PERF_TEST_ATTR=/tmp/tmppa9vxcb7 ./perf record -o /tmp/tmppa9vxcb7/perf.data --no-bpf-event -e dummy -C 0 kill >/dev/null 2>&1' ret '1', expected '1' <SNIP> test child finished with 0 ---- end ---- Setup struct perf_event_attr: Ok Reported-and-Tested-by:Yang Jihong <yangjihong1@huawei.com> Link: https://lore.kernel.org/r/20230916091641.776031-1-yangjihong1@huawei.comSigned-off-by:
-
- 15 Sep, 2023 6 commits
-
-
Colin Ian King authored
There is a spelling mistake in a literal string. Fix it. Signed-off-by:
Colin Ian King <colin.i.king@gmail.com> Reviewed-by:
-
Namhyung Kim authored
Instructions with sign- and zero- extention like movsbl and movzwq were not handled properly. As it can check different size suffix (-b, -w, -l or -q) we can omit that and add the common parts even though some combinations are not possible. Reviewed-by:
-
James Clark authored
pmu_events_table__find() is no longer used so remove it and its Arm specific version. Signed-off-by:
James Clark <james.clark@arm.com> Reviewed-by:
John Garry <john.g.garry@oracle.com> Cc: Ravi Bangoria <ravi.bangoria@amd.com> Cc: Eduard Zingerman <eddyz87@gmail.com> Cc: Will Deacon <will@kernel.org> Cc: Leo Yan <leo.yan@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Jing Zhang <renyu.zj@linux.alibaba.com> Cc: Haixin Yu <yuhaixin.yhx@linux.alibaba.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: linux-arm-kernel@lists.infradead.org Link: https://lore.kernel.org/r/20230913153355.138331-4-james.clark@arm.comSigned-off-by:
-
James Clark authored
Currently the while loop always either exits on the first iteration with a core PMU, or exits with NULL on heterogeneous systems or when not all CPUs are online. Both of the latter behaviors are undesirable for platforms other than Arm so simplify it to always return the first core PMU, or NULL if none exist. This behavior was depended on by the Arm version of pmu_metrics_table__find(), so the logic has been moved there instead. Signed-off-by:
-
James Clark authored
pmu__find_core_pmu() more logically belongs in pmus.c because it iterates over all PMUs, so move it to pmus.c At the same time rename it to perf_pmus__find_core_pmu() to match the naming convention in this file. list_prepare_entry() can't be used in perf_pmus__scan_core() anymore now that it's called from the same compilation unit. This is with -O2 (specifically -O1 -ftree-vrp -finline-functions -finline-small-functions) which allow the bounds of the array access to be determined at compile time. list_prepare_entry() subtracts the offset of the 'list' member in struct perf_pmu from &core_pmus, which isn't a struct perf_pmu. The compiler sees that pmu results in &core_pmus - 8 and refuses to compile. At runtime this works because list_for_each_entry_continue() always adds the offset back again before dereferencing ->next, but it's technically undefined behavior. With -fsanitize=undefined an additional warning is generated. Using list_first_entry_or_null() to get the first entry here avoids doing &core_pmus - 8 but has the same result and fixes both the compile warning and the undefined behavior warning. There are other uses of list_prepare_entry() in pmus.c, but the compiler doesn't seem to be able to see that they can also be called with &core_pmus, so I won't change any at this time. Signed-off-by:
-
Ian Rogers authored
The node (nd) may be NULL and pointer arithmetic on NULL is undefined behavior. Move the computation of next below the NULL check on the node. Signed-off-by:
-
- 12 Sep, 2023 20 commits
-
-
Ian Rogers authored
YYDEBUG enables line numbers and other error helpers in the generated bpf-filter-bison.c. Conditionally enabled only for debug builds. Signed-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Ian Rogers authored
YYDEBUG enables line numbers and other error helpers in the generated pmu-bison.c. Conditionally enabled only for debug builds. Signed-off-by:
-
Ian Rogers authored
YYDEBUG enables line numbers and other error helpers in the generated expr-bison.c. These shouldn't be generated when debugging isn't enabled. Signed-off-by:
-
Ian Rogers authored
YYDEBUG enables line numbers and other error helpers in the generated parse-events-bison.c. These shouldn't be generated when debugging isn't enabled. Signed-off-by:
-
Ian Rogers authored
The fnmatch header is now used in the PMU matching logic in pmu.c. Signed-off-by:
-
Athira Rajeev authored
Makefile.perf uses "CONFIG_*" checks in the code. Example the config for libtraceevent is used to set PYTHON_EXT_SRCS ifeq ($(CONFIG_LIBTRACEEVENT),y) PYTHON_EXT_SRCS := $(shell grep -v ^\# util/python-ext-sources) else PYTHON_EXT_SRCS := $(shell grep -v '^\#\|util/trace-event.c' util/python-ext-sources) endif But this is not picking the value for CONFIG_LIBTRACEEVENT that is set using the settings in Makefile.config. Include the file ".config-detected" so that make will use the system detected configuration in the CONFIG checks. This will fix isues that could arise when other "CONFIG_*" checks are added to Makefile.perf in future as well. Reviewed-by:
Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Disha Goel <disgoel@linux.vnet.ibm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: linuxppc-dev@lists.ozlabs.org Link: https://lore.kernel.org/r/20230912063807.74250-1-atrajeev@linux.vnet.ibm.comSigned-off-by:
-
Ruidong Tian authored
Add ETE as one of the supported device types in perf cs_etm testcase. Reviewed-by:
Ruidong Tian <tianruidong@linux.alibaba.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Mike Leach <mike.leach@linaro.org> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Link: https://lore.kernel.org/r/20230911065541.91293-1-tianruidong@linux.alibaba.comSigned-off-by:
-
James Clark authored
Metrics for V1 weren't previously included in the Perf Jsons, so add them using the telemetry source [1]. After generation any parts identical to the default metrics in sbsa.json were manually removed. [1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-v1.jsonSigned-off-by:
-
James Clark authored
The new data [1] includes descriptions that may have product specific details and new groupings that will be consistent with other products. The following command was used to generate the jsons: $ telemetry-solution/tools/perf_json_generator/generate.py \ linux/tools/perf/ --telemetry-files \ telemetry-solution/data/pmu/cpu/neoverse/neoverse-v1.json [1]: https://gitlab.arm.com/telemetry-solution/telemetry-solution/-/blob/main/data/pmu/cpu/neoverse/neoverse-v1.jsonSigned-off-by:
-
James Clark authored
Test that the new expression builtin returns a match when the current escaped CPU ID is given, and that it doesn't match when "0x0" is given. The CPU ID in test__expr() has to be changed to perf_pmu__getcpuid() which returns the CPU ID string, rather than the raw CPU ID that get_cpuid() returns because that can't be used with strcmp_cpuid_str(). It doesn't affect the is_intel test because both versions contain "Intel". Reviewed-by:
-
James Clark authored
It finds all occurrences of a single character and replaces them with a multi character string. This will be used in a test in a following commit. Reviewed-by:
-
James Clark authored
'cpuid_not_more_than' was the working title of the new 'strcmp_cpuid_str' keyword and was accidentally left in. It was never used so tidying it up has no effect. Reviewed-by:
-
James Clark authored
Currently the function always returns 0, so even when the has_event() test fails, the test still passes. Fix it by returning ret instead. Reviewed-by:
-
Ian Rogers authored
With paranoia set at 2 evsel__open will fail with EACCES for non-root users. To avoid this stopping libpfm4 events from being printed, retry with exclude_kernel enabled - copying the regular is_event_supported test. Signed-off-by:
-
Ian Rogers authored
Use the first core PMU instead. On a Raspberry Pi, before: $ perf list ... cpu/t1=v1[,t2=v2,t3 ...]/modifier [Raw hardware event descriptor] [(see 'man perf-list' on how to encode it)] ... After: $ perf list ... armv8_cortex_a72/t1=v1[,t2=v2,t3 ...]/modifier [Raw hardware event descriptor] [(see 'man perf-list' on how to encode it)] ... ``` Signed-off-by: -
Namhyung Kim authored
Add cgroup aggregation and filter tests. $ sudo ./perf test -v contention 84: kernel lock contention analysis test : --- start --- test child forked, pid 222423 Testing perf lock record and perf lock contention Testing perf lock contention --use-bpf Testing perf lock record and perf lock contention at the same time Testing perf lock contention --threads Testing perf lock contention --lock-addr Testing perf lock contention --lock-cgroup Testing perf lock contention --type-filter (w/ spinlock) Testing perf lock contention --lock-filter (w/ tasklist_lock) Testing perf lock contention --callstack-filter (w/ unix_stream) Testing perf lock contention --callstack-filter with task aggregation Testing perf lock contention --cgroup-filter Testing perf lock contention CSV output test child finished with 0 ---- end ---- kernel lock contention analysis test: Ok Committer testing: [root@quaco ~]# uname -a Linux quaco 6.4.10-200.fc38.x86_64 #1 SMP PREEMPT_DYNAMIC Fri Aug 11 12:20:29 UTC 2023 x86_64 GNU/Linux [root@quaco ~]# perf test -v contention 84: kernel lock contention analysis test : --- start --- test child forked, pid 452625 Testing perf lock record and perf lock contention Testing perf lock contention --use-bpf Testing perf lock record and perf lock contention at the same time Testing perf lock contention --threads Testing perf lock contention --lock-addr Testing perf lock contention --lock-cgroup Testing perf lock contention --type-filter (w/ spinlock) Testing perf lock contention --lock-filter (w/ tasklist_lock) Testing perf lock contention --callstack-filter (w/ unix_stream) Testing perf lock contention --callstack-filter with task aggregation Testing perf lock contention --cgroup-filter Testing perf lock contention CSV output test child finished with 0 ---- end ---- kernel lock contention analysis test: Ok [root@quaco ~]# Reviewed-by:
-
Namhyung Kim authored
The -G/--cgroup-filter is to limit lock contention collection on the tasks in the specific cgroups only. $ sudo ./perf lock con -abt -G /user.slice/.../vte-spawn-52221fb8-b33f-4a52-b5c3-e35d1e6fc0e0.scope \ ./perf bench sched messaging # Running 'sched/messaging' benchmark: # 20 sender and receiver processes per group # 10 groups == 400 processes run Total time: 0.174 [sec] contended total wait max wait avg wait pid comm 4 114.45 us 60.06 us 28.61 us 214847 sched-messaging 2 111.40 us 60.84 us 55.70 us 214848 sched-messaging 2 106.09 us 59.42 us 53.04 us 214837 sched-messaging 1 81.70 us 81.70 us 81.70 us 214709 sched-messaging 68 78.44 us 6.83 us 1.15 us 214633 sched-messaging 69 73.71 us 2.69 us 1.07 us 214632 sched-messaging 4 72.62 us 60.83 us 18.15 us 214850 sched-messaging 2 71.75 us 67.60 us 35.88 us 214840 sched-messaging 2 69.29 us 67.53 us 34.65 us 214804 sched-messaging 2 69.00 us 68.23 us 34.50 us 214826 sched-messaging ... Export cgroup__new() function as it's needed from outside. Reviewed-by: -
Namhyung Kim authored
The --lock-cgroup option shows lock contention stats break down by cgroups. Add LOCK_AGGR_CGROUP mode and use it instead of use_cgroup field. $ sudo ./perf lock con -ab --lock-cgroup sleep 1 contended total wait max wait avg wait cgroup 8 15.70 us 6.34 us 1.96 us / 2 1.48 us 747 ns 738 ns /user.slice/.../app.slice/app-gnome-google\x2dchrome-6442.scope 1 848 ns 848 ns 848 ns /user.slice/.../session.slice/org.gnome.Shell@x11.service 1 220 ns 220 ns 220 ns /user.slice/.../session.slice/pipewire-pulse.service For now, the cgroup mode only works with BPF (-b). Committer notes: Remove -g as it is used in the other tools with a clear meaning of collect/show callchains. As agreed with Namhyung off list. Reviewed-by: -
Namhyung Kim authored
Save cgroup info and display cgroup names if requested. This is a preparation for the next patch. Reviewed-by:
-
Namhyung Kim authored
The read_all_cgroups() is to build a tree of cgroups in the system and users can look up a cgroup using __cgroup_find(). Committer notes: Had to do this to cover that #else block: -static inline u64 __read_cgroup_id(const char *path) { return -1ULL; } +static inline u64 __read_cgroup_id(const char *path __maybe_unused) { return -1ULL; } Reviewed-by:
-
