- 02 Sep, 2023 1 commit
-
-
Ian Rogers authored
A term may have no value in which case it is assumed to have a value of 1. It doesn't just apply to alias/event terms so change the parse_events_term__to_strbuf assert. Commit 99e7138e ("perf tools: Fail on using multiple bits long terms without value") made it so that no_value terms could only be for a single bit. Prior to commit 64199ae4 ("perf parse-events: Fix propagation of term's no_value when cloning") this missed a test case where config1 had no_value. Fixes: 64199ae4 ("perf parse-events: Fix propagation of term's no_value when cloning") Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rob Herring <robh@kernel.org> Link: https://lore.kernel.org/r/20230901233949.2930562-1-irogers@google.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
- 31 Aug, 2023 10 commits
-
-
Ian Rogers authored
The no_value field in 'struct parse_events_term' indicates that the val variable isn't used, the case for an event name. Cloning wasn't propagating this, making cloned event name terms appearing to have a constant assinged to them. Working around the bug would check for a value of 1 assigned to value, but then this meant a user value of 1 couldn't be differentiated causing the value to be lost in debug printing and perf list. The change fixes the cloning and updates the "val.num ==/!= 1" tests to use no_value instead. To better check the no_value is set appropriately parameter comments are added for constant values. This found that no_value wasn't set correctly in parse_events_multi_pmu_add, which matters now that no_value is used to indicate an event name. Fixes: 7a6e9164 ("perf parse-events: Make common term list to strbuf helper") Fixes: 99e7138e ("perf tools: Fail on using multiple bits long terms without value") Signed-off-by:
Kan Liang <kan.liang@linux.intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rob Herring <robh@kernel.org> Link: https://lore.kernel.org/r/20230831071421.2201358-4-irogers@google.comSigned-off-by:
-
Ian Rogers authored
Name the enums used by 'struct parse_events_term' to parse_events__term_val_type and parse_events__term_type. This allows greater compile time error checking. Fix -Wswitch related issues by explicitly listing all enum values prior to default. Add config_term_name to safely look up a parse_events__term_type name, bounds checking the array access first. Add documentation to 'struct parse_events_terms' and reorder to save space. Signed-off-by:
-
Ian Rogers authored
"default_core" was added as a way to demark JSON events whose PMU should be whatever the default core PMU is, previously this had been assumed to be "cpu" but that fails on s390 and ARM. 'perf list' displays the PMU in the event description to save storing it in JSON, but was still comparing against "cpu" and not "default_core", so update this. Fixes: d2045f87 ("perf jevents: Use "default_core" for events with no Unit") Signed-off-by:
-
Ian Rogers authored
The metric is using the wrong format encoding. This fix is in the converter script PR: https://github.com/intel/perfmon/pull/101 Committer testing: Tested on a Lenovo t480s, before 'perf test 100' was failing with: # perf test 100 100: perf all metrics test : FAILED! With 'perf test -vv 100' we can see: <SNIP> Testing MemoryBW Not grouping metric tma_fb_full's events. Try disabling the NMI watchdog to comply NO_NMI_WATCHDOG metric constraint: echo 0 > /proc/sys/kernel/nmi_watchdog perf stat ... echo 1 > /proc/sys/kernel/nmi_watchdog event syntax error: '...DATA_READ/thresh=1,metric-id=UNC_ARB_TRK_OCCUPANCY.DATA_READ!3thresh!21!3/,UNC_ARB_TRK_OCCUPANCY.DATA_READ/metric-id=UNC_ARB_TRK_OCCUPANCY.DATA_READ/}:W,duration_time' \___ Bad event or PMU Unable to find PMU or event on a PMU of 'UNC_ARB_TRK_OCCUPANCY.DATA_READ' <SNIP> With the patch this problem is gone. Reported-by:
-
Adrian Hunter authored
The introduction of reference counting causes the v0 API perf_dlfilter_fns.resolve_address() to leak. v2 API introduced perf_dlfilter_fns.al_cleanup() to prevent that. For the v0 API, avoid the leak by exiting the addr_location immediately, since the documentation makes it clear that pointers obtained via perf_dlfilter_fns are not necessarily valid (dereferenceable) after 'filter_event' and 'filter_event_early' return. Reported-by:
kernel test robot <oliver.sang@intel.com> Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Closes: https://lore.kernel.org/oe-lkp/202308232146.94d82cb4-oliver.sang@intel.com Link: http://lore.kernel.org/lkml/20230830090539.68206-1-adrian.hunter@intel.comSigned-off-by:
-
Ian Rogers authored
Returns the number of CPUs online, unlike #num_cpus that returns the number present. Add a test of the property. This will be used in future Intel metrics. Reviewed-by:
-
Ian Rogers authored
Currently the value is only used in perf list. Compute the value just when needed to avoid unnecessary overhead. Recycle the strbuf to avoid memory allocation overhead. Signed-off-by:
-
Ian Rogers authored
A term list is turned into a string for debug output and for the str value in the alias. Add a helper to do this based on existing code, but then fix for situations like events being identified. Use strbuf to manage the dynamic memory allocation and remove the 256 byte limit. Use in various places the string of the term list is required. Before: $ sudo perf stat -vv -e inst_retired.any true Using CPUID GenuineIntel-6-8D-1 intel_pt default config: tsc,mtc,mtc_period=3,psb_period=3,pt,branch Attempting to add event pmu 'cpu' with 'inst_retired.any,' that may result in non-fatal errors After aliases, add event pmu 'cpu' with 'event,period,' that may result in non-fatal errors inst_retired.any -> cpu/inst_retired.any/ ... After: $ sudo perf stat -vv -e inst_retired.any true Using CPUID GenuineIntel-6-8D-1 intel_pt default config: tsc,mtc,mtc_period=3,psb_period=3,pt,branch Attempt to add: cpu/inst_retired.any/ ..after resolving event: cpu/event=0xc0,period=0x1e8483/ inst_retired.any -> cpu/event=0xc0,period=0x1e8483/ ... Signed-off-by:
-
Ian Rogers authored
Be more specific and fix a typo. Signed-off-by:
-
Ian Rogers authored
alias is allocated with malloc allowing uninitialized memory to be accessed. The initialization of str was moved late after it could have been updated by a JSON event, however, this create a potential for an uninitialized use. Fix this by assigning str to NULL early. Testing on ARM (Raspberry Pi) showed a memory leak in the same code so add a zfree. Fixes: f63a536f ("perf pmu: Merge JSON events with sysfs at load time") Reported-by:
Namhyung Kim <namhyung@kernel.org> Tested-by:
-
- 29 Aug, 2023 13 commits
-
-
Ian Rogers authored
The JSON Unit field encodes the name of the PMU to match the events to. When no name is given it has meant the "cpu" core PMU except for tests. On ARM, Intel hybrid and s390 the core PMU is named differently which means that using "cpu" for this case causes the events not to get matched to the PMU. Introduce a new "default_core" string for this case and in the pmu__name_match force all core PMUs to match this name. Fixes: 2e255b4f ("perf jevents: Group events by PMU") Reported-by:
Arnaldo Carvalho de Melo <acme@kernel.org> Reported-by:
Thomas Richter <tmricht@linux.ibm.com> Signed-off-by:
-
Namhyung Kim authored
It has system-wide test and cpu-list test but the cpu-list test fails sometimes. It runs sleep command on CPU1 and measure both user.slice and system.slice cgroups by default (on systemd-based systems). But if the system was idle enough, sometime the system.slice gets no count and it makes the test failing. Maybe that's because it only looks at the CPU1, let's add CPU0 to increase the chance it finds some tasks. Fixes: 79010860 ("perf test: Add a new test for perf stat cgroup BPF counter") Reported-by:
-
Namhyung Kim authored
As of now, bpf counters (bperf) don't support event groups. But the default perf stat includes topdown metrics if supported (on recent Intel machines) which require groups. That makes perf stat exiting. $ sudo perf stat --bpf-counter true bpf managed perf events do not yet support groups. Actually the test explicitly uses cycles event only, but it missed to pass the option when it checks the availability of the command. Fixes: 2c0cb9f5 ("perf test: Add a shell test for 'perf stat --bpf-counters' new option") Reviewed-by:
Song Liu <song@kernel.org> Signed-off-by:
-
Namhyung Kim authored
The BPF sample filtering requires two kernel changes below: * bpf_cast_to_kernel_ctx() kfunc (added in v6.2) * setting perf_sample_data->sample_flags (finished in v6.3) The perf tools can check bpf_cast_to_kernel_ctx() easily so it can refuse BPF filters on those old kernels (v6.1 and earlier). But checking sample_flags appears to be difficult so current code won't work on v6.2 kernel. That's unfortunate but I don't know what's the correct way to handle it. For now, let's skip v6.2 kernels explicitly (if failed) in the test. Fixes: 9575ecdd ("perf test: Add perf record sample filtering test") Reported-by:
-
Namhyung Kim authored
Now there's no in-tree user of the field. To remove the possible bug later, let's get rid of the 'id' field and add a comment for that. Signed-off-by:
-
Namhyung Kim authored
Instead of accessing the attr.id directly, use the perf_record_header_attr_id() helper to handle old versions. Signed-off-by:
-
Namhyung Kim authored
The HEADER_ATTR record has an event attr followed by the id array. But perf data from a different version could have different size of attr. So it cannot just use event->attr.id to access the array. Let's add the perf_record_header_attr_id() macro to calculate the start of the array. Signed-off-by:
-
Namhyung Kim authored
The PERF_RECORD_ATTR is used for a pipe mode to describe an event with attribute and IDs. The ID table comes after the attr and it calculate size of the table using the total record size and the attr size. n_ids = (total_record_size - end_of_the_attr_field) / sizeof(u64) This is fine for most use cases, but sometimes it saves the pipe output in a file and then process it later. And it becomes a problem if there is a change in attr size between the record and report. $ perf record -o- > perf-pipe.data # old version $ perf report -i- < perf-pipe.data # new version For example, if the attr size is 128 and it has 4 IDs, then it would save them in 168 byte like below: 8 byte: perf event header { .type = PERF_RECORD_ATTR, .size = 168 }, 128 byte: perf event attr { .size = 128, ... }, 32 byte: event IDs [] = { 1234, 1235, 1236, 1237 }, But when report later, it thinks the attr size is 136 then it only read the last 3 entries as ID. 8 byte: perf event header { .type = PERF_RECORD_ATTR, .size = 168 }, 136 byte: perf event attr { .size = 136, ... }, 24 byte: event IDs [] = { 1235, 1236, 1237 }, // 1234 is missing So it should use the recorded version of the attr. The attr has the size field already then it should honor the size when reading data. Fixes: 2c46dbb5 ("perf: Convert perf header attrs into attr events") Signed-off-by: -
Ian Rogers authored
Add a PMUs scan that ignores duplicates. When there are multiple PMUs that differ only by suffix, by default just list the first one and skip all others. The scan routine checks that the PMU names match but doesn't enforce that the numbers are consecutive as for some PMUs there are gaps. If "-v" is passed to "perf list" then list all PMUs. With the previous change duplicate PMUs are no longer printed but the suffix of the first is printed. When duplicate PMUs are being skipped avoid printing the suffix. Before: $ perf list ... uncore_imc_free_running_0/data_read/ [Kernel PMU event] uncore_imc_free_running_0/data_total/ [Kernel PMU event] uncore_imc_free_running_0/data_write/ [Kernel PMU event] uncore_imc_free_running_1/data_read/ [Kernel PMU event] uncore_imc_free_running_1/data_total/ [Kernel PMU event] uncore_imc_free_running_1/data_write/ [Kernel PMU event] After: $ perf list ... uncore_imc_free_running/data_read/ [Kernel PMU event] uncore_imc_free_running/data_total/ [Kernel PMU event] uncore_imc_free_running/data_write/ [Kernel PMU event] ... $ perf list -v uncore_imc_free_running_0/data_read/ [Kernel PMU event] uncore_imc_free_running_0/data_total/ [Kernel PMU event] uncore_imc_free_running_0/data_write/ [Kernel PMU event] uncore_imc_free_running_1/data_read/ [Kernel PMU event] uncore_imc_free_running_1/data_total/ [Kernel PMU event] uncore_imc_free_running_1/data_write/ [Kernel PMU event] ... Reviewed-by: -
Ian Rogers authored
Sort PMUs by name. If two PMUs have the same name but differ by suffix, sort the suffixes numerically. For example, "breakpoint" comes before "cpu", "uncore_imc_free_running_0" comes before "uncore_imc_free_running_1". Suffixes need to be treated specially as otherwise they will be ordered like 0, 1, 10, 11, .., 2, 20, 21, .., etc. Only PMUs starting 'uncore_' are considered to have a potential suffix. Sorting of PMUs is done so that later patches can skip duplicate uncore PMUs that differ only by there suffix. Committer notes: Used the more compact, intention revealing strstarts() function we got from the kernel sources: - if (strncmp(str, "uncore_", 7)) + if (!strstarts(str, "uncore_")) Also in pmus_cmp() the lhs_num and rhs_num variables may end up not being set for non "uncore_" prefixed PMUs in pmu_name_len_no_suffix(), or at least gcc 7.5 in some distros (opensuse 15.5, to be EOLed in Dec/2024) thins so, so initialize both to zero. Reviewed-by:
-
Yanteng Si authored
"[" is part of the shell builtin test (and a synonym for it), not a link to the external command /usr/bin/test. Using the "test" is simpler because it avoids a lot of "[]". Suggested-by:
Yanteng Si <siyanteng@loongson.cn> Acked-by:
Huacai Chen <chenhuacai@kernel.org> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: loongarch@lists.linux.dev Cc: loongson-kernel@lists.loongnix.cn Link: https://lore.kernel.org/r/c50bc0a92dce0ff0fa6504c1a52fb53e2ac007bf.1692962043.git.siyanteng@loongson.cnSigned-off-by:
-
Yanteng Si authored
To address this error: grep: /root/linux-next/tools/arch/xxxxx/include/uapi/asm//mman.h: No such file or directory Signed-off-by:
-
Yanteng Si authored
Define these macros so that the CPU name can be displayed when running 'perf report' and 'perf timechart'. Committer notes: No need to have: if (strcasestr(buf, "Model Name")) { strlcpy(cpu_m, &buf[13], 255); break; } else if (strcasestr(buf, "model name")) { strlcpy(cpu_m, &buf[13], 255); break; } As the point of strcasestr() is to be case insensitive to both the haystack and the needle, so simplify the above to just: if (strcasestr(buf, "model name")) { strlcpy(cpu_m, &buf[13], 255); break; } Signed-off-by:Huacai Chen <chenhuacai@loongson.cn> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: loongarch@lists.linux.dev Cc: loongson-kernel@lists.loongnix.cn Link: https://lore.kernel.org/r/db968a186a10e4629fe10c26a1210f7126ad41ec.1692962043.git.siyanteng@loongson.cnSigned-off-by:
-
- 25 Aug, 2023 7 commits
-
-
Kajol Jain authored
Fix typo in max-stack option description by changing lopck contention to lock contention. Signed-off-by:
Kajol Jain <kjain@linux.ibm.com> Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com> Cc: Disha Goel <disgoel@linux.ibm.com> Cc: Madhavan Srinivasan <maddy@linux.ibm.com> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lore.kernel.org/r/20230825104700.440809-1-kjain@linux.ibm.comSigned-off-by:
-
Ian Rogers authored
Casts were necessary for older versions of libslang, however, these are now 15 years old and so we no longer need to care about supporting them. Tidy the casts and remove unnecessary logic. Move the ENABLE_SLFUTURE_CONST to the libslang.h common include file, and also enable ENABLE_SLFUTURE_VOID. Signed-off-by:
-
Ian Rogers authored
Initialize realname to NULL, rather than name. This avoids a cast and as realpath is either NULL or an allocated string, free can be called unconditionally. Signed-off-by:
-
Ian Rogers authored
The struct pmu id is initialized from pmu_id that is read into allocated memory from a file, as such it needs free-ing in pmu__delete(). Make the id value const so that we can remove casts in tests. Signed-off-by:
-
Ian Rogers authored
This avoids casts in tests. Use zfree in a few places to avoid warnings about a freeing a const pointer. Signed-off-by:
-
Ian Rogers authored
The PMU name could be NULL in the case of the fake_pmu. Initialize the name for the fake_pmu to "fake" so that all other logic can assume it is initialized. Add a const to the type of name so that a literal can be used to avoid additional initialization code. Propagate the cost through related routines and remove now unnecessary "(char *)" casts. Doing this located a bug in builtin-list for the pmu_glob that was missing a strdup. Signed-off-by:
-
Ian Rogers authored
PMU caps are written as HEADER_PMU_CAPS or for the special case of the PMU "cpu" as HEADER_CPU_PMU_CAPS. As the PMU "cpu" is special, and not any "core" PMU, the logic had become broken and core PMUs not called "cpu" were not having their caps written. This affects ARM and s390 non-hybrid PMUs. Simplify the PMU caps writing logic to scan one fewer time and to be more explicit in its behavior. Fixes: 178ddf3b ("perf header: Avoid hybrid PMU list in write_pmu_caps") Reported-by:
Wei Li <liwei391@huawei.com> Signed-off-by:
-
- 24 Aug, 2023 9 commits
-
-
Ian Rogers authored
Unit with the PMU name is appended to desc in jevents.py, but on hybrid platforms it causes the desc to differ from the regular non-hybrid system with a PMU of 'cpu'. Having differing descs means the events don't deduplicate. To make the perf list output not differ, append the Unit on again in the perf list printing code. On x86 reduces the binary size by 409,600 bytes or about 4%. Update pmu-events test expectations to match the differently generated pmu-events.c code. Signed-off-by:
-
Anup Sharma authored
All required libraries have been imported and make sure that none of them are external dependencies. To achieve this, created a virt env and verified. Modified usage information and added combined command. Modified the main() function to read the --save-only command-line option and set the output_file variable accordingly. Modified the trace_end() function to check for the output_file variable. If it is set, the profiler data is saved to a local file in Gecko Profile format, or the profiler.firefox.com is opened on the default browser. Included trace_begin() to initialize the Firefox Profiler and launch the default browser to display the profiler.firefox.com. Added a new function launchFirefox() to start a local server and launch the profiler UI on the default browser with the appropriate URL. Created the "CORSRequestHandler" class to enable Cross-Origin Resource Sharing. Summary: This integration now includes a exiting feature to conveniently host the Gecko Profile data on a local server and open it directly in the default web browser. This means that users can now effortlessly visualize and analyze the profiler results with just a single click. The addition of the --save-only command-line option allows users to save the profiler output to a local file in Gecko Profile format, but the real highlight lies in the capability to seamlessly launch a local server, making the data accessible to Firefox Profiler via a web browser. In addition, it's important to highlight that all data are hosted locally, eliminating any concerns about data privacy rules and regulations. Signed-off-by:
Anup Sharma <anupnewsmail@gmail.com> Tested-by:
-
Anup Sharma authored
Refines the argument handling mechanism in the "gecko-report" script to enable better compatibility and improved user experience. The script now differentiates between scenarios where arguments are provided for record and report cases where gecko.py arguments are passed. Signed-off-by:
-
Ian Rogers authored
Sort the strings within the big C string based on whether they were for a metric and then by when they were added. This helps group related strings and reduce minor faults by approximately 10 in 1740, about 0.57%. Signed-off-by:
-
Ian Rogers authored
Don't load sysfs aliases for a PMU when the PMU is first created, defer until an alias needs to be found. For the pmu-scan benchmark, average core PMU scanning is reduced by 30.8%, and average PMU scanning by 12.6%. Signed-off-by:
-
Ian Rogers authored
Event info is only needed when an event is parsed or when merging data from an JSON and sysfs event. Be lazy in its loading to reduce file accesses. Signed-off-by:
-
Ian Rogers authored
Scan sysfs PMU's type early so that format and aliases aren't attempted to be loaded if the PMU name is invalid. This is the case for event_pmu tokens in parse-events.y where a wildcard name is first assumed to be a PMU name. Signed-off-by:
-
Ian Rogers authored
Rather than scanning all JSON events and adding them when a PMU is created, add the alias when the JSON event is needed. Average core PMU scanning run time reduced by 60.2%. Average PMU scanning run time reduced by 15%. Page faults with no events reduced by 74 page faults, 4% of total. Signed-off-by:
-
Ian Rogers authored
Cache the JSON events table so that finding it isn't done per event/alias. Change the events table find so that when the PMU is given, if the PMU has no JSON events return null. Update usage to always use the PMU variable. Signed-off-by:
-
