Re: [PATCH v3] perf parse: Copy string to perf_evsel_config_term

[Jiri Olsa]

On Tue, Jan 07, 2020 at 02:45:27PM -0700, Mathieu Poirier wrote:
> Hi Leo,
> 
> On Tue, 7 Jan 2020 at 05:03, Leo Yan <leo.yan@xxxxxxxxxx> wrote:
> >
> > perf with CoreSight fails to record trace data with command:
> >
> >   perf record -e cs_etm/@tmc_etr0/u --per-thread ls
> >   failed to set sink "" on event cs_etm/@tmc_etr0/u with 21 (Is a
> >   directory)/perf/
> >
> > This failure is root caused with the commit 1dc925568f01 ("perf
> > parse: Add a deep delete for parse event terms").
> >
> > The log shows, cs_etm fails to parse the sink attribution; cs_etm event
> > relies on the event configuration to pass sink name, but the event
> > specific configuration data cannot be passed properly with flow:
> >
> >   get_config_terms()
> >     ADD_CONFIG_TERM(DRV_CFG, drv_cfg, term->val.str);
> >       __t->val.drv_cfg = term->val.str;
> >         `> __t->val.drv_cfg is assigned to term->val.str;
> >
> >   parse_events_terms__purge()
> >     parse_events_term__delete()
> >       zfree(&term->val.str);
> >         `> term->val.str is freed and assigned to NULL pointer;
> >
> >   cs_etm_set_sink_attr()
> >     sink = __t->val.drv_cfg;
> >       `> sink string has been freed.
> >
> > To fix this issue, in the function get_config_terms(), this patch
> > changes from directly assignment pointer value for the strings to
> > use strdup() for allocation a new duplicate string for the cases:
> >
> >   perf_evsel_config_term::val.callgraph
> >   perf_evsel_config_term::val.branch
> >   perf_evsel_config_term::val.drv_cfg.
> >
> > In the data structure perf_evsel_config_term, this patch adds
> > 'char *str' pointer in the val union and new field 'free_str'.  When the
> > union is used as a string pointer, 'free_str' will be set to true;
> > finally it's flag to tell perf_evsel__free_config_terms() to free the
> > string with perf_evsel_config_term::val.str.
> 
> Many thanks for digging into this and stepping forward to provide a
> solution - it is much appreciated.
> 
> >
> > Fixes: 1dc925568f01 ("perf parse: Add a deep delete for parse event terms")
> > Suggested-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> > Signed-off-by: Leo Yan <leo.yan@xxxxxxxxxx>
> > ---
> >  tools/perf/util/evsel.c        |  2 ++
> >  tools/perf/util/evsel_config.h |  2 ++
> >  tools/perf/util/parse-events.c | 56 +++++++++++++++++++++-------------
> >  3 files changed, 39 insertions(+), 21 deletions(-)
> >
> > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> > index a69e64236120..ab9925cc1aa7 100644
> > --- a/tools/perf/util/evsel.c
> > +++ b/tools/perf/util/evsel.c
> > @@ -1265,6 +1265,8 @@ static void perf_evsel__free_config_terms(struct evsel *evsel)
> >
> >         list_for_each_entry_safe(term, h, &evsel->config_terms, list) {
> >                 list_del_init(&term->list);
> > +               if (term->free_str)
> > +                       free(term->val.str);
> 
> This will do the trick but we can definitely do better.
> 
> Part of his comments on V2, Jiri hinted that we should move to a
> common perf_evsel_config_term::str to replace {callgraph, drv_cfg,
> branch}, something that will work because we have
> perf_evsel_config_term::type.  That means  functions
> apply_config_terms() and cs_etm_set_sink_attr() need to be modified
> but the changes are quite small and well worth for the benefit they'll
> carry.
> 
> With that the above becomes neat and clean.

I wonder if there was some reason for keeping the variables
like that for every type and not just one per type as we did
'struct parse_events_term'

if the change is possible, the code would be cleaner, let's see ;-)

thanks,
jirka