Re: [PATCH 1/4] perf lock contention: Add lock_data.h for common data

[Namhyung Kim]

Hi Arnaldo,

On Mon, Dec 12, 2022 at 11:45 AM Arnaldo Carvalho de Melo
<acme@xxxxxxxxxx> wrote:
>
> Em Mon, Dec 12, 2022 at 04:43:43PM -0300, Arnaldo Carvalho de Melo escreveu:
> > Em Mon, Dec 12, 2022 at 04:42:30PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > Em Fri, Dec 09, 2022 at 11:07:24AM -0800, Namhyung Kim escreveu:
> > > > Accessing BPF maps should use the same data types.  Add bpf_skel/lock_data.h
> > > > to define the common data structures.  No functional changes.
> > >
> > > You forgot to update one of the stack_id users, that field got renamed:
> > >
> > > util/bpf_skel/lock_contention.bpf.c:144:6: error: no member named 'stack_id' in 'struct contention_key'
> > >         key.stack_id = pelem->stack_id;
> > >         ~~~ ^
> > > 1 error generated.
> > > make[2]: *** [Makefile.perf:1075: /tmp/build/perf/util/bpf_skel/.tmp/lock_contention.bpf.o] Error 1
> > > make[1]: *** [Makefile.perf:236: sub-make] Error 2
> > > make: *** [Makefile:113: install-bin] Error 2
> > > make: Leaving directory '/var/home/acme/git/perf/tools/perf'

Oops, right.

> > >
> > >  Performance counter stats for 'make -k NO_LIBTRACEEVENT=1 BUILD_BPF_SKEL=1 CORESIGHT=1 O=/tmp/build/perf -C tools/perf install-bin':
> > >
> > >      7,005,216,342      cycles:u
> > >     11,851,225,594      instructions:u                   #    1.69  insn per cycle
> > >
> > >        3.168945139 seconds time elapsed
> > >
> > >        1.730964000 seconds user
> > >        1.578932000 seconds sys
> > >
> > >
> > > ⬢[acme@toolbox perf]$ git log --oneline -4
> > > f6e7a5f1db49dc8e (HEAD) perf lock contention: Add lock_data.h for common data
> > > 5d9b55713c5c037f perf python: Account for multiple words in CC
> > > d9078bf3f3320457 perf off_cpu: Fix a typo in BTF tracepoint name, it should be 'btf_trace_sched_switch'
> > > 3b7ea76f0f7844f5 perf test: Update event group check for support of uncore event
> > > ⬢[acme@toolbox perf]$
> > >
> > > After some point it builds.
> > >
> > > I'm fixing this to keep it bisectable.
> >
> > I folded this:
> >
> > diff --git a/tools/perf/util/bpf_skel/lock_contention.bpf.c b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> > index 0f63cc28ccbabd21..64fd1e040ac86e58 100644
> > --- a/tools/perf/util/bpf_skel/lock_contention.bpf.c
> > +++ b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> > @@ -141,7 +141,7 @@ int contention_end(u64 *ctx)
> >
> >       duration = bpf_ktime_get_ns() - pelem->timestamp;
> >
> > -     key.stack_id = pelem->stack_id;
> > +     key.stack_or_task_id = pelem->stack_id;
> >       data = bpf_map_lookup_elem(&lock_stat, &key);
> >       if (!data) {
> >               struct contention_data first = {

Thanks for fixing this.

>
>
> And then fixed up this:
>
> Could not apply 3d4947c7bd10beba... perf lock contention: Implement -t/--threads option for BPF
> ⬢[acme@toolbox perf]$
> ⬢[acme@toolbox perf]$
> ⬢[acme@toolbox perf]$ git diff
> diff --cc tools/perf/util/bpf_skel/lock_contention.bpf.c
> index 64fd1e040ac86e58,cd405adcd252b82d..0000000000000000
> --- a/tools/perf/util/bpf_skel/lock_contention.bpf.c
> +++ b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> @@@ -141,7 -168,13 +168,17 @@@ int contention_end(u64 *ctx
>
>         duration = bpf_ktime_get_ns() - pelem->timestamp;
>
> ++<<<<<<< HEAD
>  +      key.stack_or_task_id = pelem->stack_id;
> ++=======
> +       if (aggr_mode == LOCK_AGGR_CALLER) {
> +               key.stack_or_task_id = pelem->stack_id;
> +       } else {
> +               key.stack_or_task_id = pid;
> +               update_task_data(pid);
> +       }
> +
> ++>>>>>>> 3d4947c7bd10beba (perf lock contention: Implement -t/--threads option for BPF)

Sure, it looks good to me.

Thanks,
Namhyung