On Fri, Sep 13, 2019 at 03:22:42PM +0200, Jiri Olsa wrote:
> hi,
> sending changes for exporting basic sampling interface
> in libperf. It's now possible to use following code in
> applications via libperf:
>
> --- (example is without error checks for simplicity)
>
> struct perf_event_attr attr = {
> .type = PERF_TYPE_TRACEPOINT,
> .sample_period = 1,
> .wakeup_watermark = 1,
> .disabled = 1,
> };
> /* ... setup attr */
>
> cpus = perf_cpu_map__new(NULL);
>
> evlist = perf_evlist__new();
> evsel = perf_evsel__new(&attr);
> perf_evlist__add(evlist, evsel);
>
> perf_evlist__set_maps(evlist, cpus, NULL);
>
> err = perf_evlist__open(evlist);
> err = perf_evlist__mmap(evlist, 4);
>
> err = perf_evlist__enable(evlist);
>
> /* ... monitored area, plus all the other cpus */
>
> err = perf_evlist__disable(evlist);
>
> perf_evlist__for_each_mmap(evlist, map) {
> if (perf_mmap__read_init(map) < 0)
> continue;
>
> while ((event = perf_mmap__read_event(map)) != NULL) {
> perf_mmap__consume(map);
> }
>
> perf_mmap__read_done(map);
> }
>
> perf_evlist__delete(evlist);
> perf_cpu_map__put(cpus);
>
> --- (end)
>
> Nothing is carved in stone so far, the interface is exported
> as is available in perf now and we can change it as we want.
>
> New tests are added in test-evlist.c to do thread and cpu based
> sampling.
>
> All the functionality should not change, however there's considerable
> mmap code rewrite, so would be great if guys could run your usual
> workloads to see if all is fine.. so far so good in my tests ;-)
>
> It's also available in here:
> git://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git
> perf/lib
hi,
I rebased the branch to latest Arnaldo's perf/core branch
jirka
