Re: [PATCH] perf bench: add benchmark for evlist open/close operations

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

Em Tue, Aug 10, 2021 at 12:31:55PM +0200, Riccardo Mancini escreveu:
> Hi Arnaldo,
> 
> On Mon, 2021-08-09 at 17:28 -0300, Arnaldo Carvalho de Melo wrote:
> > Em Mon, Aug 09, 2021 at 05:23:31PM -0300, Arnaldo Carvalho de Melo escreveu:
> > > Em Mon, Aug 09, 2021 at 10:11:02PM +0200, Riccardo Mancini escreveu:
> > > > +static struct evlist *bench__create_evlist(char *evstr)
> > > > +{
> > > > +       struct evlist *evlist;
> > > > +       struct parse_events_error err;
> > > > +       int ret;
> > 
> > > > +       evlist = evlist__new();
> > > > +       if (!evlist) {
> > > > +               pr_err("Not enough memory to create evlist\n");
> > > > +               return NULL;
> > > > +       }
> > 
> > > > +       bzero(&err, sizeof(err));
> > 
> > > man bzero
> > > 
> > >        The bzero() function is deprecated (marked as LEGACY in POSIX.1-2001);
> > > use memset(3) in new programs.  POSIX.1-2008 removes the specification of
> > > bzero().  The bzero() function first appeared in 4.3BSD.
> 
> Oops, I didn't know, but I saw it is being used in some parts in perf, maybe we
> should get rid of them:
> $ rg -c bzero
> builtin-lock.c:1
> arch/powerpc/util/kvm-stat.c:1
> builtin-stat.c:1
> builtin-trace.c:2
> bench/evlist-open-close.c:1
> bench/numa.c:5
> tests/parse-events.c:1
> tests/backward-ring-buffer.c:1
> tests/bpf.c:2
> util/metricgroup.c:1
> util/parse-events.c:1

Yeah, patches are welcome, but at least lets not add new ones :-)
 
> > > I'm replacing it with a memset().
> > 
> > This one is also equivalent:
> > 
> > tools/perf/tests/pmu-events.c:  struct parse_events_error error = { .idx = 0, };
> > 
> > https://gcc.gnu.org/onlinedocs/gcc/Designated-Inits.html
> > 
> > That text is a bit roundabout, as it says that the members that are not
> > explicitely initialized will be initialized as variables with static
> > storage duration, i.e. zeroed.
> 
> Would it be the same doing the shorter {0}. It would be a general solution for
> these init-to-zero cases.

I'd have to do some extra research to remember why that is also not
optimal, IIRC the '= { .a = 0, };' is the optimal one.
 
> Unrelated to this small issue, I noticed I forgot to check the return of
> bench__create_evlist. Would you like me to send a v2 fixing both issues or are
> you able to apply this other small change yourself?

Nah, as this is the HEAD right now in my local branch, I'll apply it
myself, thanks!

- Arnaldo
 
> diff --git a/tools/perf/bench/evlist-open-close.c b/tools/perf/bench/evlist-open-close.c
> index 40bce06f5ca7bef3..f0b9c330f34f2984 100644
> --- a/tools/perf/bench/evlist-open-close.c
> +++ b/tools/perf/bench/evlist-open-close.c
> @@ -168,7 +168,11 @@ static int bench_evlist_open_close__run(char *evstr)
>  
>         for (i = 0; i < iterations; i++) {
>                 pr_debug("Started iteration %d\n", i);
> +
>                 evlist = bench__create_evlist(evstr);
> +               if (!evlist)
> +                       return -ENOMEM;
> +
>                 gettimeofday(&start, NULL);
>                 err = bench__do_evlist_open_close(evlist);
>                 if (err) {
> 
> Thanks,
> Riccardo
> 
> > 
> > - Arnaldo
> 
> 

-- 

- Arnaldo