linux-kernel - Re: [PATCH 6/7] perf expr: Add metric literals for topology.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 10 Nov 2021 06:19:04 -0800
From:   Ian Rogers <irogers@...gle.com>
To:     Jiri Olsa <jolsa@...hat.com>
Cc:     Andi Kleen <ak@...ux.intel.com>, Jin Yao <yao.jin@...ux.intel.com>,
        Namhyung Kim <namhyung@...nel.org>,
        John Garry <john.garry@...wei.com>,
        Kajol Jain <kjain@...ux.ibm.com>,
        "Paul A . Clarke" <pc@...ibm.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Riccardo Mancini <rickyman7@...il.com>,
        Kan Liang <kan.liang@...ux.intel.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Ingo Molnar <mingo@...hat.com>,
        Mark Rutland <mark.rutland@....com>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Madhavan Srinivasan <maddy@...ux.ibm.com>,
        Song Liu <song@...nel.org>, Wan Jiabing <wanjiabing@...o.com>,
        Yury Norov <yury.norov@...il.com>,
        Changbin Du <changbin.du@...el.com>,
        linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 6/7] perf expr: Add metric literals for topology.

On Wed, Nov 10, 2021 at 4:56 AM Jiri Olsa <jolsa@...hat.com> wrote:
>
> On Fri, Nov 05, 2021 at 10:09:42AM -0700, Ian Rogers wrote:
> > Allow the number of cpus, cores, dies and packages to be queried by a
> > metric expression.
> >
> > Signed-off-by: Ian Rogers <irogers@...gle.com>
> > ---
> >  tools/perf/tests/expr.c | 12 +++++++++++-
> >  tools/perf/util/expr.c  | 27 +++++++++++++++++++++++++++
> >  2 files changed, 38 insertions(+), 1 deletion(-)
> >
> > diff --git a/tools/perf/tests/expr.c b/tools/perf/tests/expr.c
> > index 9ee2dc91c27b..0c09ccc76665 100644
> > --- a/tools/perf/tests/expr.c
> > +++ b/tools/perf/tests/expr.c
> > @@ -66,7 +66,7 @@ int test__expr(struct test *t __maybe_unused, int subtest __maybe_unused)
> >  {
> >       struct expr_id_data *val_ptr;
> >       const char *p;
> > -     double val;
> > +     double val, num_cpus, num_cores, num_dies, num_packages;
> >       int ret;
> >       struct expr_parse_ctx *ctx;
> >
> > @@ -161,6 +161,16 @@ int test__expr(struct test *t __maybe_unused, int subtest __maybe_unused)
> >                       NULL, ctx) == 0);
> >       TEST_ASSERT_VAL("find ids", hashmap__size(ctx->ids) == 0);
> >
> > +     /* Test toplogy constants appear well ordered. */
> > +     expr__ctx_clear(ctx);
> > +     TEST_ASSERT_VAL("#num_cpus", expr__parse(&num_cpus, ctx, "#num_cpus") == 0);
> > +     TEST_ASSERT_VAL("#num_cores", expr__parse(&num_cores, ctx, "#num_cores") == 0);
> > +     TEST_ASSERT_VAL("#num_cpus >= #num_cores", num_cpus >= num_cores);
> > +     TEST_ASSERT_VAL("#num_dies", expr__parse(&num_dies, ctx, "#num_dies") == 0);
> > +     TEST_ASSERT_VAL("#num_cores >= #num_dies", num_cores >= num_dies);
> > +     TEST_ASSERT_VAL("#num_packages", expr__parse(&num_packages, ctx, "#num_packages") == 0);
> > +     TEST_ASSERT_VAL("#num_dies >= #num_packages", num_dies >= num_packages);
> > +
> >       expr__ctx_free(ctx);
> >
> >       return 0;
> > diff --git a/tools/perf/util/expr.c b/tools/perf/util/expr.c
> > index 7464739c2890..15af8b8ef5e7 100644
> > --- a/tools/perf/util/expr.c
> > +++ b/tools/perf/util/expr.c
> > @@ -5,6 +5,8 @@
> >  #include <stdlib.h>
> >  #include <string.h>
> >  #include "metricgroup.h"
> > +#include "cpumap.h"
> > +#include "cputopo.h"
> >  #include "debug.h"
> >  #include "expr.h"
> >  #include "expr-bison.h"
> > @@ -375,9 +377,34 @@ double expr_id_data__value(const struct expr_id_data *data)
> >
> >  double expr__get_literal(const char *literal)
> >  {
> > +     static struct cpu_topology *topology;
> > +
> >       if (!strcmp("#smt_on", literal))
> >               return smt_on() > 0 ? 1.0 : 0.0;
> >
> > +     if (!strcmp("#num_cpus", literal))
> > +             return cpu__max_present_cpu();
> > +
> > +     /*
> > +      * Assume that topology strings are consistent, such as CPUs "0-1"
> > +      * wouldn't be listed as "0,1", and so after deduplication the number of
> > +      * these strings gives an indication of the number of packages, dies,
> > +      * etc.
> > +      */
> > +     if (!topology) {
> > +             topology = cpu_topology__new();
>
> any chance we could propagate expr_scanner_ctx in here and store topology
> to it and release it at the end? I think we have several places like this,
> so it'd be nice not to make more if it's possible ;-)

The topology here is static and so will only get computed once per
execution rather than once pre expression parse. I was worried about
the cost of recomputing the topology for something like 'perf stat -I
1000 -M ...' in which case the static will do less recomputation.

Thanks,
Ian

> thanks,
> jirka
>
> > +             if (!topology) {
> > +                     pr_err("Error creating CPU topology");
> > +                     return NAN;
> > +             }
> > +     }
> > +     if (!strcmp("#num_packages", literal))
> > +             return topology->package_cpus_lists;
> > +     if (!strcmp("#num_dies", literal))
> > +             return topology->die_cpus_lists;
> > +     if (!strcmp("#num_cores", literal))
> > +             return topology->core_cpus_lists;
> > +
> >       pr_err("Unrecognized literal '%s'", literal);
> >       return NAN;
> >  }
> > --
> > 2.34.0.rc0.344.g81b53c2807-goog
> >
>