linux/tools/perf, branch v5.11

perf script: Fix overrun issue for dynamically-allocated PMU type number

2021-01-21T20:25:33Z

When unpacking the event which is from dynamic PMU, the array output[OUTPUT_TYPE_MAX] may be overrun. For example, type number of SKL uncore_imc is 10, but OUTPUT_TYPE_MAX is 7 now (OUTPUT_TYPE_MAX = PERF_TYPE_MAX + 1). /* In builtin-script.c */ process_event() { unsigned int type = output_type(attr->type); if (output[type].fields == 0) return; } output[10] is overrun. Create a type OUTPUT_TYPE_OTHER for dynamic PMU events, then output_type(attr->type) will return OUTPUT_TYPE_OTHER here. Note that if PERF_TYPE_MAX ever changed, then there would be a conflict between old perf.data files that had a dynamicaliy allocated PMU number that would then be the same as a fixed PERF_TYPE. Example: # perf record --switch-events -C 0 -e "{cpu-clock,uncore_imc/data_reads/,uncore_imc/data_writes/}:SD" -a -- sleep 1 # perf script Before: swapper 0 [000] 1479253.987551: 277766 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.987797: 246709 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.988127: 329883 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.988273: 146393 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.988523: 249977 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.988877: 354090 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.989023: 145940 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.989383: 359856 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1479253.989523: 140082 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) After: swapper 0 [000] 1397040.402011: 272384 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1397040.402011: 5396 uncore_imc/data_reads/: swapper 0 [000] 1397040.402011: 967 uncore_imc/data_writes/: swapper 0 [000] 1397040.402259: 249153 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1397040.402259: 7231 uncore_imc/data_reads/: swapper 0 [000] 1397040.402259: 1297 uncore_imc/data_writes/: swapper 0 [000] 1397040.402508: 249108 cpu-clock: ffffffff9d4ddb6f cpuidle_enter_state+0xdf ([kernel.kallsyms]) swapper 0 [000] 1397040.402508: 5333 uncore_imc/data_reads/: swapper 0 [000] 1397040.402508: 1008 uncore_imc/data_writes/: Signed-off-by: Jin Yao Acked-by: Adrian Hunter Acked-by: Jiri Olsa Cc: Peter Zijlstra Cc: Kan Liang Cc: Andi Kleen Cc: Alexander Shishkin Cc: Ingo Molnar Link: https://lore.kernel.org/r/20201209005828.21302-1-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf metricgroup: Fix system PMU metrics

2021-01-21T20:25:33Z

Joakim reports that getting "perf stat" for multiple system PMU metrics segfaults: $ perf stat -a -I 1000 -M imx8mm_ddr_write.all,imx8mm_ddr_write.all Segmentation fault $ While the same works without issue for a single metric. The logic in metricgroup__add_metric_sys_event_iter() is broken, in that add_metric() @m argument should be NULL for each new metric. Fix by not passing a holder for that, and rather make local in metricgroup__add_metric_sys_event_iter(). Fixes: be335ec28efa ("perf metricgroup: Support adding metrics for system PMUs") Reported-by: Joakim Zhang Signed-off-by: John Garry Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Ian Rogers Cc: Ingo Molnar Cc: Kajol Jain Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Cc: linuxarm@openeuler.org Link: https://lore.kernel.org/r/1611050655-44020-1-git-send-email-john.garry@huawei.com Signed-off-by: Arnaldo Carvalho de Melo

perf metricgroup: Fix for metrics containing duration_time

2021-01-21T20:25:33Z

Metrics containing duration_time cause a segfault: $ perf stat -v -M L1D_Cache_Fill_BW sleep 1 Using CPUID GenuineIntel-6-3D-4 metric expr 64 * l1d.replacement / 1000000000 / duration_time for L1D_Cache_Fill_BW found event duration_time found event l1d.replacement adding {l1d.replacement}:W,duration_time l1d.replacement -> cpu/umask=0x1,(null)=0x1e8483,event=0x51/ Segmentation fault $ In commit c2337d67199a1ea1 ("perf metricgroup: Fix metrics using aliases covering multiple PMUs"), the logic in find_evsel_group() when iter'ing events was changed to not only select events in same group, but also for aliased PMUs. Checking whether events were for aliased PMUs was done by comparing the event PMU name. This was not safe for duration_time event, which has no associated PMU (and no PMU name), so fix by checking if the event PMU name is set also. Committer testing: Reproduced the bug, then, on a: $ grep -m1 ^'model name' /proc/cpuinfo model name : Intel(R) Core(TM) i7-8650U CPU @ 1.90GHz $ We now get: $ perf stat -M L1D_Cache_Fill_BW sleep 1 Performance counter stats for 'sleep 1': 4,141 l1d.replacement:u 1,001,285,107 ns duration_time:u 1.001285107 seconds time elapsed 0.000000000 seconds user 0.001119000 seconds sys $ Detais from -v: Using CPUID GenuineIntel-6-8E-A metric expr 64 * l1d.replacement / 1000000000 / duration_time for L1D_Cache_Fill_BW found event duration_time found event l1d.replacement adding {l1d.replacement}:W,duration_time l1d.replacement -> cpu/(null)=0x1e8483,umask=0x1,event=0x51/ Control descriptor is not initialized Warning: kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor samples Warning: kernel.perf_event_paranoid=2, trying to fall back to excluding kernel and hypervisor samples l1d.replacement:u: 4592 612201 612201 duration_time:u: 1001478621 1001478621 1001478621 Fixes: c2337d67199a1ea1 ("perf metricgroup: Fix metrics using aliases covering multiple PMUs") Reported-by: Joakim Zhang Signed-off-by: John Garry Tested-by: Arnaldo Carvalho de Melo Tested-by: Jiri Olsa Acked-by: Ian Rogers Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Ingo Molnar Cc: Kajol Jain Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Cc: linuxarm@openeuler.org Link: https://lore.kernel.org/r/1611159518-226883-1-git-send-email-john.garry@huawei.com Signed-off-by: Arnaldo Carvalho de Melo

perf inject: Correct event attribute sizes

2021-01-15T20:28:28Z

When 'perf inject' reads a perf.data file from an older version of perf, it writes event attributes into the output with the original size field, but lays them out as if they had the size currently used. Readers see a corrupt file. Update the size field to match the layout. Signed-off-by: Al Grant Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Mark Rutland Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lore.kernel.org/lkml/20201124195818.30603-1-al.grant@arm.com Signed-off-by: Denis Nikitin Signed-off-by: Arnaldo Carvalho de Melo

perf intel-pt: Fix 'CPU too large' error

2021-01-15T20:28:27Z

In some cases, the number of cpus (nr_cpus_online) is confused with the maximum cpu number (nr_cpus_avail), which results in the error in the example below: Example on system with 8 cpus: Before: # echo 0 > /sys/devices/system/cpu/cpu2/online # ./perf record --kcore -e intel_pt// taskset --cpu-list 7 uname Linux [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.147 MB perf.data ] # ./perf script --itrace=e Requested CPU 7 too large. Consider raising MAX_NR_CPUS 0x25908 [0x8]: failed to process type: 68 [Invalid argument] After: # ./perf script --itrace=e # Fixes: 8c7274691f0d ("perf machine: Replace MAX_NR_CPUS with perf_env::nr_cpus_online") Fixes: 7df4e36a4785 ("perf session: Replace MAX_NR_CPUS with perf_env::nr_cpus_online") Signed-off-by: Adrian Hunter Tested-by: Kan Liang Cc: Jiri Olsa Cc: stable@vger.kernel.org Link: http://lore.kernel.org/lkml/20210107174159.24897-1-adrian.hunter@intel.com Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Take cgroups into account for shadow stats

2021-01-15T20:28:27Z

As of now it doesn't consider cgroups when collecting shadow stats and metrics so counter values from different cgroups will be saved in a same slot. This resulted in incorrect numbers when those cgroups have different workloads. For example, let's look at the scenario below: cgroups A and C runs same workload which burns a cpu while cgroup B runs a light workload. $ perf stat -a -e cycles,instructions --for-each-cgroup A,B,C sleep 1 Performance counter stats for 'system wide': 3,958,116,522 cycles A 6,722,650,929 instructions A # 2.53 insn per cycle 1,132,741 cycles B 571,743 instructions B # 0.00 insn per cycle 4,007,799,935 cycles C 6,793,181,523 instructions C # 2.56 insn per cycle 1.001050869 seconds time elapsed When I run 'perf stat' with single workload, it usually shows IPC around 1.7. We can verify it (6,722,650,929.0 / 3,958,116,522 = 1.698) for cgroup A. But in this case, since cgroups are ignored, cycles are averaged so it used the lower value for IPC calculation and resulted in around 2.5. avg cycle: (3958116522 + 1132741 + 4007799935) / 3 = 2655683066 IPC (A) : 6722650929 / 2655683066 = 2.531 IPC (B) : 571743 / 2655683066 = 0.0002 IPC (C) : 6793181523 / 2655683066 = 2.557 We can simply compare cgroup pointers in the evsel and it'll be NULL when cgroups are not specified. With this patch, I can see correct numbers like below: $ perf stat -a -e cycles,instructions --for-each-cgroup A,B,C sleep 1 Performance counter stats for 'system wide': 4,171,051,687 cycles A 7,219,793,922 instructions A # 1.73 insn per cycle 1,051,189 cycles B 583,102 instructions B # 0.55 insn per cycle 4,171,124,710 cycles C 7,192,944,580 instructions C # 1.72 insn per cycle 1.007909814 seconds time elapsed Signed-off-by: Namhyung Kim Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Andi Kleen Cc: Ian Rogers Cc: Jin Yao Cc: Mark Rutland Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lore.kernel.org/lkml/20210115071139.257042-2-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Introduce struct runtime_stat_data

2021-01-15T20:28:27Z

To pass more info to the saved_value in the runtime_stat, add a new struct runtime_stat_data. Currently it only has 'ctx' field but later patch will add more. Note that we intentionally pass 0 as ctx to clock-related events for compatibility. It was already there in a few places. So move the code into the saved_value_lookup() explicitly and add a comment. Suggested-by: Andi Kleen Signed-off-by: Namhyung Kim Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Andi Kleen Cc: Ian Rogers Cc: Jin Yao Cc: Mark Rutland Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lore.kernel.org/lkml/20210115071139.257042-1-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf test: Fix shadow stat test for non-bash shells

2021-01-15T19:31:46Z

It was using some bash-specific features and failed to parse when running with a different shell like below: root@kbl-ppc:~/kbl-ws/perf-dev/lck-9077/acme.tmp/tools/perf# ./perf test 83 -vv 83: perf stat metrics (shadow stat) test : --- start --- test child forked, pid 3922 ./tests/shell/stat+shadow_stat.sh: 19: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 24: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 30: ./tests/shell/stat+shadow_stat.sh: [[: not found (standard_in) 2: syntax error ./tests/shell/stat+shadow_stat.sh: 36: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 19: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 24: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 30: ./tests/shell/stat+shadow_stat.sh: [[: not found (standard_in) 2: syntax error ./tests/shell/stat+shadow_stat.sh: 36: ./tests/shell/stat+shadow_stat.sh: [[: not found ./tests/shell/stat+shadow_stat.sh: 45: ./tests/shell/stat+shadow_stat.sh: declare: not found test child finished with -1 ---- end ---- perf stat metrics (shadow stat) test: FAILED! Reported-by: Jin Yao Signed-off-by: Namhyung Kim Acked-by: Jiri Olsa Cc: Alexander Shishkin Cc: Andi Kleen Cc: David Laight Cc: Ian Rogers Cc: Mark Rutland Cc: Peter Zijlstra Cc: Stephane Eranian Link: http://lore.kernel.org/lkml/20210114050609.1258820-1-namhyung@kernel.org Signed-off-by: Arnaldo Carvalho de Melo

perf bpf examples: Fix bpf.h header include directive in 5sec.c example

2021-01-15T19:31:46Z

It was looking at bpf/bpf.h, which caused this problem: # perf trace -e tools/perf/examples/bpf/5sec.c /home/acme/git/perf/tools/perf/examples/bpf/5sec.c:42:10: fatal error: 'bpf/bpf.h' file not found #include ^~~~~~~~~~~ 1 error generated. ERROR: unable to compile tools/perf/examples/bpf/5sec.c Hint: Check error message shown above. Hint: You can also pre-compile it into .o using: clang -target bpf -O2 -c tools/perf/examples/bpf/5sec.c with proper -I and -D options. event syntax error: 'tools/perf/examples/bpf/5sec.c' \___ Failed to load tools/perf/examples/bpf/5sec.c from source: Error when compiling BPF scriptlet # Change that to plain bpf.h, to make it work again: # perf trace -e tools/perf/examples/bpf/5sec.c sleep 5s 0.000 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1776891872, rqtp: 5000000000) # perf trace -e tools/perf/examples/bpf/5sec.c/max-stack=16/ sleep 5s 0.000 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1776891872, rqtp: 5000000000) hrtimer_nanosleep ([kernel.kallsyms]) common_nsleep ([kernel.kallsyms]) __x64_sys_clock_nanosleep ([kernel.kallsyms]) do_syscall_64 ([kernel.kallsyms]) entry_SYSCALL_64_after_hwframe ([kernel.kallsyms]) __clock_nanosleep_2 (/usr/lib64/libc-2.32.so) # perf trace -e tools/perf/examples/bpf/5sec.c sleep 4s # Cc: Adrian Hunter Cc: Ian Rogers Cc: Jiri Olsa Cc: Namhyung Kim Signed-off-by: Arnaldo Carvalho de Melo

perf probe: Fix memory leak when synthesizing SDT probes

2020-12-24T13:52:10Z

The argv_split() function must be paired with argv_free(), else we must keep a reference to the argv array received or do the freeing ourselves, in synthesize_sdt_probe_command() we were simply leaking that argv[] array. Fixes: 3b1f8311f6963cd1 ("perf probe: Add sdt probes arguments into the uprobe cmd string") Cc: Adrian Hunter Cc: Alexander Shishkin Cc: Alexandre Truong Cc: Alexis Berlemont Cc: He Zhe Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: John Garry Cc: Mark Rutland Cc: Masami Hiramatsu Cc: Mathieu Poirier Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Sumanth Korikkar Cc: Thomas Richter Cc: Will Deacon Link: https://lore.kernel.org/r/20201224135139.GF477817@kernel.org Signed-off-by: Arnaldo Carvalho de Melo