Re: [PATCH v7 00/10] perf record --off-cpu: Dump off-cpu samples directly

[Arnaldo Carvalho de Melo <acme@kernel.org>]

On Tue, Nov 12, 2024 at 04:17:19PM -0300, Arnaldo Carvalho de Melo wrote:
> I squashed the patch below and I'm trying to apply the other patches to do some
> minimal testing on the feature itself, but the organization of the
> patches needs some work.
 
> +++ b/tools/perf/util/bpf_off_cpu.c
> @@ -61,6 +61,9 @@ static int off_cpu_config(struct evlist *evlist)
>  static void off_cpu_start(void *arg)
>  {
>  	struct evlist *evlist = arg;
> +	struct evsel *evsel;
> +	struct perf_cpu pcpu;
> +	int i;
>  
>  	/* update task filter for the given workload */
>  	if (skel->rodata->has_task && skel->rodata->uses_tgid &&
> @@ -82,6 +85,8 @@ static void off_cpu_start(void *arg)
>  	}
>  
>  	perf_cpu_map__for_each_cpu(pcpu, i, evsel->core.cpus) {
> +		int err;
> +
>  		err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
>  					   xyarray__entry(evsel->core.fd, i, 0),
>  					   sizeof(__u32), BPF_ANY);

This is not enough, as it in the end tries to use that
skel->maps.offcpu_output that is only introduced at a later patch, it
seems, not checked yet, but explains the error below:

  LD      /tmp/build/perf-tools-next/perf-test-in.o
  AR      /tmp/build/perf-tools-next/libperf-test.a
  CC      /tmp/build/perf-tools-next/util/parse-events.o
util/bpf_off_cpu.c: In function ‘off_cpu_start’:
util/bpf_off_cpu.c:90:54: error: ‘struct <anonymous>’ has no member named ‘offcpu_output’
   90 |                 err = bpf_map__update_elem(skel->maps.offcpu_output, &pcpu.cpu, sizeof(__u32),
      |                                                      ^
make[4]: *** [/home/acme/git/perf-tools-next/tools/build/Makefile.build:106: /tmp/build/perf-tools-next/util/bpf_off_cpu.o] Error 1
make[4]: *** Waiting for unfinished jobs....
make[3]: *** [/home/acme/git/perf-tools-next/tools/build/Makefile.build:158: util] Error 2
make[2]: *** [Makefile.perf:789: /tmp/build/perf-tools-next/perf-util-in.o] Error 2
make[2]: *** Waiting for unfinished jobs....
  CC      /tmp/build/perf-tools-next/pmu-events/pmu-events.o
  LD      /tmp/build/perf-tools-next/pmu-events/pmu-events-in.o
make[1]: *** [Makefile.perf:292: sub-make] Error 2
make: *** [Makefile:119: install-bin] Error 2
make: Leaving directory '/home/acme/git/perf-tools-next/tools/perf'
⬢ [acme@toolbox perf-tools-next]$ 


Ok, at the end of the series it builds, and the 'perf test' entry
introduced in this series passes:

root@x1:~# perf test off
121: perf record offcpu profiling tests                              : Ok
root@x1:~# perf test -v off
121: perf record offcpu profiling tests                              : Ok
root@x1:~# perf test -vv off
121: perf record offcpu profiling tests:
--- start ---
test child forked, pid 1303134
Checking off-cpu privilege
Basic off-cpu test
Basic off-cpu test [Success]
Child task off-cpu test
Child task off-cpu test [Success]
Direct off-cpu test
Direct off-cpu test [Success]
---- end(0) ----
121: perf record offcpu profiling tests                              : Ok
root@x1:~#

But the only examples I could find so far for this feature were on the
'perf test' at the end of this series.

I think we need to have some examples in the 'perf-record' man page
showing how to use it, explaining the whole process, etc.

I'll continue testing it and trying to move things around so that it
gets bisectable and testable step by step, documenting the whole
process as I go, probably tomorrow.

The series with my fixes is at:

https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git perf-off-cpu77918
https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/log/?h=perf-off-cpu


- Arnaldo