kernel/tools/perf/util/evsel.c, branch linux-4.17.y

perf evsel: Only fall back group read for leader

2018-04-24T19:11:59Z

Perf doesn't support mixed events from different PMUs (except software event) in a group. The perf stat should output / for all events, but it doesn't. For example, perf stat -e '{cycles,uncore_imc_5/umask=0xF,event=0x4/,instructions}' cycles uncore_imc_5/umask=0xF,event=0x4/ 1,024,300 instructions If perf fails to open an event, it doesn't error out directly. It will disable some features and retry, until the event is opened or all features are disabled. The disabled features will not be re-enabled. The group read is one of these features. For the example as above, the IMC event and the leader event "cycles" are from different PMUs. Opening the IMC event must fail. The group read feature must be disabled for IMC event and the followed event "instructions". The "instructions" event has the same PMU as the leader "cycles". It can be opened successfully. Since the group read feature has been disabled, the "instructions" event will be read as a single event, which definitely has a value. The group read fallback is still useful for the case which kernel doesn't support group read. It is good enough to be handled only by the leader. For the fallback request from members, it must be caused by an error. The fallback only breaks the semantics of group. Limit the group read fallback only for the leader. Committer testing: On a broadwell t450s notebook: Before: # perf stat -e '{cycles,unc_cbo_cache_lookup.read_i,instructions}' sleep 1 Performance counter stats for 'sleep 1': cycles unc_cbo_cache_lookup.read_i 818,206 instructions 1.003170887 seconds time elapsed Some events weren't counted. Try disabling the NMI watchdog: echo 0 > /proc/sys/kernel/nmi_watchdog perf stat ... echo 1 > /proc/sys/kernel/nmi_watchdog After: # perf stat -e '{cycles,unc_cbo_cache_lookup.read_i,instructions}' sleep 1 Performance counter stats for 'sleep 1': cycles unc_cbo_cache_lookup.read_i instructions 1.001380511 seconds time elapsed Some events weren't counted. Try disabling the NMI watchdog: echo 0 > /proc/sys/kernel/nmi_watchdog perf stat ... echo 1 > /proc/sys/kernel/nmi_watchdog # Reported-by: Andi Kleen Signed-off-by: Kan Liang Tested-by: Arnaldo Carvalho de Melo Cc: Agustin Vega-Frias Cc: Ganapatrao Kulkarni Cc: Jin Yao Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Shaokun Zhang Cc: Will Deacon Fixes: 82bf311e15d2 ("perf stat: Use group read for event groups") Link: http://lkml.kernel.org/r/1524594014-79243-3-git-send-email-kan.liang@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf evsel: Disable write_backward for leader sampling group events

2018-04-23T14:21:56Z

.. and other related fields that do not need to be enabled for events that have sampling leader. It fixes the perf top usage Ingo reported broken: # perf top -e '{cycles,msr/aperf/}:S' The 'msr/aperf/' event is configured for write_back sampling, which is not allowed by the MSR PMU, so it fails to create the event. Adjusting related attr test. Reported-by: Ingo Molnar Signed-off-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: David Ahern Cc: Kan Liang Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/20180423090823.32309-6-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Keep the / modifier separator in fallback

2018-04-23T14:14:10Z

The 'perf stat' fallback for EACCES error sets the exclude_kernel perf_event_attr and tries perf_event_open() again with it. In addition, it also changes the name of the event to reflect that change by adding the 'u' modifier. But it does not take into account the '/' separator, so the event name can end up mangled, like: (note the '/:' characters) $ perf stat -e cpu/cpu-cycles/ kill ... 386,832 cpu/cpu-cycles/:u Adding the code to check on the '/' separator and set the following correct event name: $ perf stat -e cpu/cpu-cycles/ kill ... 388,548 cpu/cpu-cycles/u Signed-off-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: David Ahern Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/20180423090823.32309-4-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf record: Remove suggestion to enable APIC

2018-04-18T18:35:50Z

'perf record' suggests to enable the APIC on errors. APIC is practically always used today and the problem is usually somewhere else. Just remove the outdated suggestion. Signed-off-by: Andi Kleen Acked-by: Jiri Olsa Link: http://lkml.kernel.org/r/20180406203812.3087-5-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo

perf record: Remove misleading error suggestion

2018-04-18T18:35:49Z

When perf record encounters an error setting up an event it suggests to enable CONFIG_PERF_EVENTS. This is misleading because: - Usually it is enabled (it is really hard to disable on x86) - The problem is usually somewhere else, e.g. the CPU is not supported or an invalid configuration has been used. Remove the misleading suggestion. Signed-off-by: Andi Kleen Acked-by: Jiri Olsa Link: http://lkml.kernel.org/r/20180406203812.3087-4-andi@firstfloor.org Signed-off-by: Arnaldo Carvalho de Melo

perf report: Fix the output for stdio events list

2018-03-08T14:30:36Z

Changing the output header for reporting forced groups via --groups option on non grouped events, like: $ perf record -e 'cycles,instructions' $ perf report --stdio --group Before: # Samples: 24 of event 'anon group { cycles:u, instructions:u }' After: # Samples: 24 of events 'cycles:u, instructions:u' Signed-off-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: David Ahern Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Fixes: ad52b8cb4886 ("perf report: Add support to display group output for non group events") Link: http://lkml.kernel.org/r/20180307155020.32613-2-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf pmu: Display pmu name when printing unmerged events in stat

2018-03-08T13:05:49Z

To simplify creation of events accross multiple instances of the same type of PMU stat supports two methods for creating multiple events from a single event specification: 1. A prefix or glob can be used in the PMU name. 2. Aliases, which are listed immediately after the Kernel PMU events by perf list, are used. When the --no-merge option is passed and these events are displayed individually the PMU name is lost and it's not possible to see which count corresponds to which pmu: $ perf stat -a -e l3cache/read-miss/ --no-merge ls > /dev/null Performance counter stats for 'system wide': 67 l3cache/read-miss/ 67 l3cache/read-miss/ 63 l3cache/read-miss/ 60 l3cache/read-miss/ 0.001675706 seconds time elapsed $ perf stat -a -e l3cache_read_miss --no-merge ls > /dev/null Performance counter stats for 'system wide': 12 l3cache_read_miss 17 l3cache_read_miss 10 l3cache_read_miss 8 l3cache_read_miss 0.001661305 seconds time elapsed This change adds the original pmu name to the event. For dynamic pmu events the pmu name is restored in the event name: $ perf stat -a -e l3cache/read-miss/ --no-merge ls > /dev/null Performance counter stats for 'system wide': 63 l3cache_0_3/read-miss/ 74 l3cache_0_1/read-miss/ 64 l3cache_0_2/read-miss/ 74 l3cache_0_0/read-miss/ 0.001675706 seconds time elapsed For alias events the name is added after the event name: $ perf stat -a -e l3cache_read_miss --no-merge ls > /dev/null Performance counter stats for 'system wide': 10 l3cache_read_miss [l3cache_0_3] 12 l3cache_read_miss [l3cache_0_1] 10 l3cache_read_miss [l3cache_0_2] 17 l3cache_read_miss [l3cache_0_0] 0.001661305 seconds time elapsed Signed-off-by: Agustin Vega-Frias Acked-by: Andi Kleen Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Timur Tabi Cc: linux-arm-kernel@lists.infradead.org Change-Id: I8056b9eda74bda33e95065056167ad96e97cb1fb Link: http://lkml.kernel.org/r/1520345084-42646-3-git-send-email-agustinv@codeaurora.org Signed-off-by: Arnaldo Carvalho de Melo

perf cgroup: Rename close_cgroup() to cgroup__put()

2018-03-07T13:22:25Z

It is not really closing the cgroup, but instead dropping a reference count and if it hits zero, then calling delete, which will, among other cleanup shores, close the cgroup fd. So it is really dropping a reference to that cgroup, and the method name for that is "put", so rename close_cgroup() to cgroup__put() to follow this naming convention. Cc: Adrian Hunter Cc: David Ahern Cc: Jiri Olsa Cc: Namhyung Kim Cc: Stephane Eranian Cc: Wang Nan Link: https://lkml.kernel.org/n/tip-sccxpnd7bgwc1llgokt6fcey@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Ignore error thread when enabling system-wide --per-thread

2018-02-27T14:29:21Z

If we execute 'perf stat --per-thread' with non-root account (even set kernel.perf_event_paranoid = -1 yet), it reports the error: jinyao@skl:~$ perf stat --per-thread Error: You may not have permission to collect system-wide stats. Consider tweaking /proc/sys/kernel/perf_event_paranoid, which controls use of the performance events system by unprivileged users (without CAP_SYS_ADMIN). The current value is 2: -1: Allow use of (almost) all events by all users Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK >= 0: Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN Disallow raw tracepoint access by users without CAP_SYS_ADMIN >= 1: Disallow CPU event access by users without CAP_SYS_ADMIN >= 2: Disallow kernel profiling by users without CAP_SYS_ADMIN To make this setting permanent, edit /etc/sysctl.conf too, e.g.: kernel.perf_event_paranoid = -1 Perhaps the ptrace rule doesn't allow to trace some processes. But anyway the global --per-thread mode had better ignore such errors and continue working on other threads. This patch will record the index of error thread in perf_evsel__open() and remove this thread before retrying. For example (run with non-root, kernel.perf_event_paranoid isn't set): jinyao@skl:~$ perf stat --per-thread ^C Performance counter stats for 'system wide': vmstat-3458 6.171984 cpu-clock:u (msec) # 0.000 CPUs utilized perf-3670 0.515599 cpu-clock:u (msec) # 0.000 CPUs utilized vmstat-3458 1,163,643 cycles:u # 0.189 GHz perf-3670 40,881 cycles:u # 0.079 GHz vmstat-3458 1,410,238 instructions:u # 1.21 insn per cycle perf-3670 3,536 instructions:u # 0.09 insn per cycle vmstat-3458 288,937 branches:u # 46.814 M/sec perf-3670 936 branches:u # 1.815 M/sec vmstat-3458 15,195 branch-misses:u # 5.26% of all branches perf-3670 76 branch-misses:u # 8.12% of all branches 12.651675247 seconds time elapsed Signed-off-by: Jin Yao Acked-by: Jiri Olsa Tested-by: Arnaldo Carvalho de Melo Cc: Alexander Shishkin Cc: Andi Kleen Cc: Kan Liang Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1516117388-10120-1-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf evsel: Expose the perf_missing_features struct

2018-02-15T12:54:53Z

As tools may need to adjust to missing features, as 'perf top' will, in the next csets, to cope with a missing 'write_backward' feature. Cc: Andi Kleen Cc: Jin Yao Cc: Jiri Olsa Cc: Kan Liang Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Wang Nan Link: https://lkml.kernel.org/n/tip-jelngl9q1ooaizvkcput9tic@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo