mbox series

[0/4] perf lock contention: Support task/addr aggregation mode (v1)

Message ID 20221209190727.759804-1-namhyung@kernel.org (mailing list archive)
Headers show
Series perf lock contention: Support task/addr aggregation mode (v1) | expand

Message

Namhyung Kim Dec. 9, 2022, 7:07 p.m. UTC
Hello,

This patchset adds two more aggregation modes for perf lock contention.

The first one is the per-task mode with -t/--threads option.  The option
was there already but it remained broken with BPF for a while.  Now it
supports the output with and without BPF.

  # perf lock contention -a -b -t -- sleep 1
   contended   total wait     max wait     avg wait          pid   comm

          11     11.85 us      2.23 us      1.08 us            0   swapper
           2     11.13 us     10.22 us      5.56 us       749053   ThreadPoolForeg
           1      8.15 us      8.15 us      8.15 us       321353   Chrome_ChildIOT
           2      2.73 us      1.77 us      1.37 us       320761   Chrome_ChildIOT
           1      1.40 us      1.40 us      1.40 us       320502   chrome
           1       379 ns       379 ns       379 ns       321227   chrome

The other one is the per-lock-instance mode with -l/--lock-addr option.
If the lock has a symbol, it will be displayed as well.

  # perf lock contention -a -b -l -- sleep 1
   contended   total wait     max wait     avg wait            address   symbol

           3      4.79 us      2.33 us      1.60 us   ffffffffbaed50c0   rcu_state
           4      4.19 us      1.62 us      1.05 us   ffffffffbae07a40   jiffies_lock
           1      1.94 us      1.94 us      1.94 us   ffff9262277861e0
           1       387 ns       387 ns       387 ns   ffff9260bfda4f60

It's based on the current acme/tmp.perf/core branch.
You can find the code in the 'perf/lock-con-aggr-v1' branch in

  git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git

Thanks,
Namhyung


Namhyung Kim (4):
  perf lock contention: Add lock_data.h for common data
  perf lock contention: Implement -t/--threads option for BPF
  perf lock contention: Add -l/--lock-addr option
  perf test: Update perf lock contention test

 tools/perf/Documentation/perf-lock.txt        |  4 +
 tools/perf/builtin-lock.c                     | 95 ++++++++++++++-----
 tools/perf/tests/shell/lock_contention.sh     | 48 ++++++++++
 tools/perf/util/bpf_lock_contention.c         | 72 ++++++++++----
 .../perf/util/bpf_skel/lock_contention.bpf.c  | 67 +++++++++----
 tools/perf/util/bpf_skel/lock_data.h          | 30 ++++++
 tools/perf/util/lock-contention.h             |  1 +
 7 files changed, 255 insertions(+), 62 deletions(-)
 create mode 100644 tools/perf/util/bpf_skel/lock_data.h


base-commit: b22802e295a80ec16e355d7208d2fbbd7bbc1b7a

Comments

Arnaldo Carvalho de Melo Dec. 12, 2022, 7:58 p.m. UTC | #1
Em Fri, Dec 09, 2022 at 11:07:23AM -0800, Namhyung Kim escreveu:
> Hello,
> 
> This patchset adds two more aggregation modes for perf lock contention.


Thanks, applied.

- Arnaldo

 
> The first one is the per-task mode with -t/--threads option.  The option
> was there already but it remained broken with BPF for a while.  Now it
> supports the output with and without BPF.
> 
>   # perf lock contention -a -b -t -- sleep 1
>    contended   total wait     max wait     avg wait          pid   comm
> 
>           11     11.85 us      2.23 us      1.08 us            0   swapper
>            2     11.13 us     10.22 us      5.56 us       749053   ThreadPoolForeg
>            1      8.15 us      8.15 us      8.15 us       321353   Chrome_ChildIOT
>            2      2.73 us      1.77 us      1.37 us       320761   Chrome_ChildIOT
>            1      1.40 us      1.40 us      1.40 us       320502   chrome
>            1       379 ns       379 ns       379 ns       321227   chrome
> 
> The other one is the per-lock-instance mode with -l/--lock-addr option.
> If the lock has a symbol, it will be displayed as well.
> 
>   # perf lock contention -a -b -l -- sleep 1
>    contended   total wait     max wait     avg wait            address   symbol
> 
>            3      4.79 us      2.33 us      1.60 us   ffffffffbaed50c0   rcu_state
>            4      4.19 us      1.62 us      1.05 us   ffffffffbae07a40   jiffies_lock
>            1      1.94 us      1.94 us      1.94 us   ffff9262277861e0
>            1       387 ns       387 ns       387 ns   ffff9260bfda4f60
> 
> It's based on the current acme/tmp.perf/core branch.
> You can find the code in the 'perf/lock-con-aggr-v1' branch in
> 
>   git://git.kernel.org/pub/scm/linux/kernel/git/namhyung/linux-perf.git
> 
> Thanks,
> Namhyung
> 
> 
> Namhyung Kim (4):
>   perf lock contention: Add lock_data.h for common data
>   perf lock contention: Implement -t/--threads option for BPF
>   perf lock contention: Add -l/--lock-addr option
>   perf test: Update perf lock contention test
> 
>  tools/perf/Documentation/perf-lock.txt        |  4 +
>  tools/perf/builtin-lock.c                     | 95 ++++++++++++++-----
>  tools/perf/tests/shell/lock_contention.sh     | 48 ++++++++++
>  tools/perf/util/bpf_lock_contention.c         | 72 ++++++++++----
>  .../perf/util/bpf_skel/lock_contention.bpf.c  | 67 +++++++++----
>  tools/perf/util/bpf_skel/lock_data.h          | 30 ++++++
>  tools/perf/util/lock-contention.h             |  1 +
>  7 files changed, 255 insertions(+), 62 deletions(-)
>  create mode 100644 tools/perf/util/bpf_skel/lock_data.h
> 
> 
> base-commit: b22802e295a80ec16e355d7208d2fbbd7bbc1b7a
> -- 
> 2.39.0.rc1.256.g54fd8350bd-goog