- 11 Oct, 2013 11 commits
-
-
David Ahern authored
The "perf stat" tool displays the command run in its summary output which is misleading when using a cpu list or system wide collection. Before: perf stat -a -- sleep 1 Performance counter stats for 'sleep 1': 16152.670249 task-clock # 16.132 CPUs utilized 417 context-switches # 0.002 M/sec 7 cpu-migrations # 0.030 K/sec ... After: perf stat -a -- sleep 1 Performance counter stats for 'system wide': 16206.931120 task-clock # 16.144 CPUs utilized 395 context-switches # 0.002 M/sec 5 cpu-migrations # 0.030 K/sec ... or perf stat -C1 -- sleep 1 Performance counter stats for 'CPU(s) 1': 1001.669257 task-clock # 1.000 CPUs utilized 4,264 context-switches # 0.004 M/sec 3 cpu-migrations # 0.003 K/sec ... Signed-off-by:David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1380400080-9211-2-git-send-email-dsahern@gmail.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Jiri Olsa authored
The perf_evlist__mmap_read used 'union perf_event' as a placeholder for event crossing the mmap boundary. This is ok for sample shorter than ~PATH_MAX. However we could grow up to the maximum sample size which is 16 bits max. I hit this overflow issue when using 'perf top -G dwarf' which produces sample with the size around 8192 bytes. We could configure any valid sample size here using: '-G dwarf,size'. Using array with sample max size instead for the event placeholder. Also adding another safe check for the dynamic size of the user stack. TODO: The 'struct perf_mmap' is quite big now, maybe we could use some lazy allocation for event_copy size. Signed-off-by:
Jiri Olsa <jolsa@redhat.com> Acked-by:
-
Petr Holasek authored
Patch adds more subtle handling of -C and -N parameters in parse_{cpu,node}_setup_list() functions when there isn't enough NUMA nodes or CPUs present. Instead of assertion and terminating benchmark, partial test is skipped with error message and perf will continue to the next one. Fixed problem can be easily reproduced on machine with only one NUMA node: # Running numa/mem benchmark... # Running main, "perf bench numa mem -a" ... # Running RAM-bw-remote, "perf bench numa mem -p 1 -t 1 -P 1024 -C 0 -M 1 -s perf: bench/numa.c:622: parse_setup_node_list: Assertion `!(bind_node_0 < 0 || bind_node_0 >= g->p.nr_nodes)' failed. Aborted Signed-off-by:Petr Holasek <pholasek@redhat.com> Acked-by:
Ingo Molnar <mingo@kernel.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Petr Benas <pbenas@redhat.com> Link: http://lkml.kernel.org/r/1380821325-4017-1-git-send-email-pholasek@redhat.comSigned-off-by:
Petr Benas <pbenas@redhat.com> Signed-off-by:
-
Ramkumar Ramachandra authored
The default output file produced by the 'perf timechart' tool is called output.svg, add it to .gitignore. Signed-off-by:
Ramkumar Ramachandra <artagnon@gmail.com> Cc: Ingo Molnar <mingo@kernel.org> Link: http://lkml.kernel.org/r/1380789636-4512-1-git-send-email-artagnon@gmail.comSigned-off-by:
-
Ramkumar Ramachandra authored
When only the instructions event is requested: $ perf stat -e instructions git s M builtin-stat.c Performance counter stats for 'git s': 917,453,420 instructions # 0.00 insns per cycle 0.213002926 seconds time elapsed The 0.00 insns per cycle comment in the output is totally bogus and misleading. It happens because update_shadow_stats() doesn't touch runtime_cycles_stats when only the instructions event is requested. So, omit printing the bogus data altogether. Signed-off-by: -
Ramkumar Ramachandra authored
When only the cycles event is requested: $ perf stat -e cycles dd if=/dev/zero of=/dev/null count=1000000 1000000+0 records in 1000000+0 records out 512000000 bytes (512 MB) copied, 0.26123 s, 2.0 GB/s Performance counter stats for 'dd if=/dev/zero of=/dev/null count=1000000': 911,626,453 cycles # 0.000 GHz 0.262113350 seconds time elapsed The 0.000 GHz comment in the output is totally bogus and misleading. It happens because update_shadow_stats() doesn't touch runtime_nsecs_stats; it is only written when a requested counter matches a SW_TASK_CLOCK. In our case, since we have only requested HW_CPU_CYCLES, runtime_nsecs_stats is unavailable. So, omit printing the comment altogether. Signed-off-by: -
Jiri Olsa authored
Moving start conditions to start of the flex file so it's clear what the INITIAL condition rules are. Plus adding default rule for INITIAL condition. This prevents default space to be printed for events like: $ ./perf stat -e "cycles " kill 2>/dev/null $ ^^^^^^^^ Signed-off-by: -
Jiri Olsa authored
If we build perf with NO_LIBPYTHON=1 NO_LIBPERL=1 the '-ldl' is not added to libs build fails if we have gtk2 code in, because it depends on it. Signed-off-by:
Namhyung Kim <namhyung@kernel.org> Cc: Andi Kleen <ak@linux.intel.com> Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1380221754-29865-1-git-send-email-jolsa@redhat.comSigned-off-by:
-
Adrian Hunter authored
To avoid buffer overruns. Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/r/1379845338-29637-2-git-send-email-adrian.hunter@intel.com [ Split from aa7fe3b0 ] Signed-off-by:
-
Ingo Molnar authored
Allow the measurement of thread versus process context switch performance. The default stays at 'process' based measurement, like lmbench's lat_ctx benchmark. Sample output: comet:~/tip/tools/perf> taskset 1 ./perf bench sched pipe # Running sched/pipe benchmark... # Executed 1000000 pipe operations between two processes Total time: 4.138 [sec] 4.138729 usecs/op 241620 ops/sec comet:~/tip/tools/perf> taskset 1 ./perf bench sched pipe --threaded # Running sched/pipe benchmark... # Executed 1000000 pipe operations between two threads Total time: 3.667 [sec] 3.667667 usecs/op 272652 ops/sec Signed-off-by: -
Arnaldo Carvalho de Melo authored
Make 'perf trace' more accessible by aliasing it to just 'trace': [root@zoo linux]# trace --duration 15 -a -e futex sleep 1 110.092 (16.188 ms): libvirtd/1166 futex(uaddr: 0x185b344, op: WAIT|PRIV, val: 174293 ) = 0 110.101 (15.903 ms): libvirtd/1171 futex(uaddr: 0x185b3dc, op: WAIT|PRIV, val: 139265 ) = 0 111.594 (15.776 ms): libvirtd/1165 futex(uaddr: 0x185b344, op: WAIT|PRIV, val: 174295 ) = 0 111.610 (15.969 ms): libvirtd/1169 futex(uaddr: 0x185b3dc, op: WAIT|PRIV, val: 139267 ) = 0 113.556 (16.216 ms): libvirtd/1168 futex(uaddr: 0x185b3dc, op: WAIT|PRIV, val: 139269 ) = 0 291.265 (199.508 ms): chromium-brows/15830 futex(uaddr: 0x7fff2986bcb4, op: WAIT_BITSET|PRIV|CLKRT, val: 1, utime: 0x7fff2986bab0, val3: 4294967295) = -1 ETIMEDOUT Connection timed out 360.354 (69.053 ms): chromium-brows/15830 futex(uaddr: 0x7fff2986bcb4, op: WAIT_BITSET|PRIV|CLKRT, val: 1, utime: 0x7fff2986bab0, val3: 4294967295) = -1 ETIMEDOUT Connection timed out [root@zoo linux]# I.e. looking for futex calls that take at least 15ms, system wide, during a one second window. Now to get callchains into 'trace' to figure out what are those locks :-) Requested-by:
-
- 09 Oct, 2013 29 commits
-
-
Arnaldo Carvalho de Melo authored
For live sessions we can just access /proc to map an fd to its path, on a best effort way, i.e. sometimes the fd will have gone away when we try to do the mapping, as it is done in a lazy way, only when a reference to such fd is made then the path will be looked up in /proc. This is disabled when processing perf.data files, where we will have to have a way to get getname events, be it via an on-the-fly 'perf probe' event or after a vfs_getname tracepoint is added to the kernel. A first step will be to synthesize such event for the use cases where the threads in the monitored workload exist already. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-1r1ti33ye1666jezu2d8q1c3@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Printing it as an hex number. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-gd68zmnwbbofsv5m6w18intw@git.kernel.orgSigned-off-by:
-
David Ahern authored
Needed for compiles on Fedora 12 for example. Signed-off-by:
-
David Ahern authored
Before: 0.030 ( 0.002 ms): 2571 clock_gettime(which_clock: 1, tp: 0x7f3b45729cd0 ) = 0 After: 0.030 ( 0.002 ms): 2571 clock_gettime(which_clock: MONOTONIC, tp: 0x7f3b45729cd0 ) = 0 v2: Update to use the STRARRAY option Signed-off-by:
-
Arnaldo Carvalho de Melo authored
4.234 (0.005 ms): fetchmail/3224 pipe2(fildes: 0x7fffc72bcee0, flags: CLOEXEC) = 0 Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-9e1jz78i6q6e0xr9fsitqbpe@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
For the address family and socket type. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-3a6cwwskobvan823pau76cm4@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Case in hand: 9.682 ( 0.001 ms): Xorg/13079 setitimer(which: REAL, value: 0x7fffede42470) = 0 ITIMER_REAL is zero. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-6hnoqsjh99t4hxi3xu2nlwep@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
In such cases just stating the (arg, name, array) is enough, reducing the size of the syscall formatters table. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-3k53p6dv2sh4ydsc5k5otoia@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
4735.638 ( 0.003 ms): man/19881 flock(fd: 3, cmd: SH|NB) = 0 4735.832 ( 0.002 ms): man/19881 flock(fd: 3, cmd: UN ) = 0 Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-amh3y88kh1nmclpwezqlarl8@git.kernel.orgSigned-off-by:
-
Arnaldo Carvalho de Melo authored
[root@sandy ~]# perf trace -e epoll_ctl 2.490 (0.003 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: ADD, fd: 24, event: 0x7fff22314ef0) = 0 2.621 (0.003 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: DEL, fd: 24 ) = 0 2.833 (0.010 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: ADD, fd: 24, event: 0x7fff22314cd0) = 0 2.953 (0.002 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: DEL, fd: 24 ) = 0 3.118 (0.002 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: ADD, fd: 24, event: 0x7fff22314d20) = 0 4.762 (0.002 ms): systemd-logind/586 epoll_ctl(epfd: 10, op: DEL, fd: 24 ) = 0 ^C[root@sandy ~]# Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: David Ahern <dsahern@gmail.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Link: http://lkml.kernel.org/n/tip-88xz9phc8cbicnxonud6if8h@git.kernel.orgSigned-off-by:
-
Namhyung Kim authored
Currently the srcline sort key compares ip rather than srcline info. I guess this was due to a performance reason to run external addr2line utility. Now we have implemented the functionality inside, use the srcline info when comparing hist entries. Also constantly print "??:0" string for unknown srcline rather than printing ip. Signed-off-by:
-
Roberto Vitillo authored
When the srcline sort key is used , the external addr2line utility needs to be run for each hist entry to get the srcline info. This can consume quite a time if one has a huge perf.data file. So rather than executing the external utility, implement it internally and just call it. We can do it since we've linked with libbfd already. Signed-off-by:
Roberto Agostino Vitillo <ravitillo@lbl.gov> Reviewed-by:
-
Namhyung Kim authored
Some dso's lack srcline info, so there's no point to keep trying on them. Just save failture status and skip them. Signed-off-by:
-
Namhyung Kim authored
This is a preparation of next change. No functional changes are intended. Signed-off-by:
-
Namhyung Kim authored
No need to call addr2line since they don't have such information. Signed-off-by:
-
Namhyung Kim authored
Currently external addr2line tool is used for srcline sort key and annotate with srcline info. Separate the common code to prepare upcoming enhancements. Signed-off-by:
-
Namhyung Kim authored
We've been leaked srcline of hist_entry, it should be freed also. Signed-off-by:
-
Namhyung Kim authored
In the symbol__get_source_line(), path and src_line->path will have same value, but they were allocated separately, and leaks one. Just share path to src_line->path. Signed-off-by:
-
Namhyung Kim authored
In the hist_entry__srcline_snprintf(), path and self->srcline are pointing the same memory region, but they are doubly allocated. Signed-off-by:
-
Namhyung Kim authored
Separate out GTK codes to a shared object called libperf-gtk.so. This time only GTK codes are built with -fPIC and libperf remains as is. Now run GTK hist and annotation browser using libdl. Signed-off-by:
Pekka Enberg <penberg@kernel.org> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Pekka Enberg <penberg@kernel.org> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1379053663-13706-1-git-send-email-namhyung@kernel.org [ Fix it up wrt Ingo's tools/perf build speedups ] Signed-off-by:
-
Willy Tarreau authored
Running "perf top" on a machine with possibly invalid or non-matching vmlinux at the various places results in no symbol resolving despite /proc/kallsyms being present and valid. Add a new option --ignore-vmlinux to explicitly indicate that we do not want to use these kernels and just use what we have (kallsyms). Signed-off-by:
Willy Tarreau <w@1wt.eu> Cc: Ingo Molnar <mingo@redhat.com> Link: http://lkml.kernel.org/r/20130914083259.GA3418@1wt.euSigned-off-by:
-
Jiri Olsa authored
Moving 'struct throttle_event' out of python code and making it global as any other event. There's no usage of throttling events in any perf commands so far (besides python support), but we'll need this event data backup for upcoming test. Signed-off-by:
-
Jiri Olsa authored
Adding new common function to create evlist with default event. It spares some code lines in automated tests. Signed-off-by:
-
Jiri Olsa authored
Adding possibility to specify mmap size via -m/--mmap-pages by appending unit size character (B/K/M/G) to the number, like: $ perf record -m 8K ls $ perf record -m 2M ls The size is rounded up appropriately to follow perf mmap restrictions. If no unit is specified the number provides pages as of now, like: $ perf record -m 8 ls Signed-off-by:
-
Jiri Olsa authored
Move the check of the mmap_pages value to the options parsing time, so we could rely on this value on other parts of code. Related changes come in the next patches. Also changes perf_evlist::mmap_len to proper size_t type. Signed-off-by:
-
Davidlohr Bueso authored
While perf-lock currently reports both the total wait time and the number of contentions, it doesn't explicitly show the average wait time. Having this value immediately in the report can be quite useful when looking into performance issues. Furthermore, allowing report to sort by averages is another handy feature to have - and thus do not only print the value, but add it to the lock_stat structure. Signed-off-by:
Davidlohr Bueso <davidlohr@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp> Cc: Ingo Molnar <mingo@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1378693159-8747-8-git-send-email-davidlohr@hp.comSigned-off-by:
-
Davidlohr Bueso authored
Two decimal precision should be enough for this. Signed-off-by:
-
Davidlohr Bueso authored
This function should be straightforward, and we can remove some trivial logic by moving the functionality of read_events() into __cmd_report() - thus allowing a new session to be properly deleted. Since the 'info' subcommand also needs to process the recorded events, add a 'display_info' flag to differentiate between report and info commands. Furthermore, this patch also calls perf_session__has_traces(), making sure that we don't compare apples and oranges, fixing a segfault when using an perf.data file generated by a different subcommand. ie: ./perf mem record sleep 1 [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.017 MB perf.data (~724 samples) ] ./perf lock report Segmentation fault (core dumped) Signed-off-by:
-
Davidlohr Bueso authored
Address some trivial leaks. Signed-off-by:
-
