2023-08-12 08:55:13

by Yang Jihong

[permalink] [raw]
Subject: [RFC v1 01/16] perf kwork: Fix incorrect and missing free atom in work_push_atom()

1. Atoms are managed in page mode and should be released using atom_free()
instead of free().
2. When the event does not match, the atom needs to free.

Fixes: f98919ec4fcc ("perf kwork: Implement 'report' subcommand")
Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/builtin-kwork.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-kwork.c b/tools/perf/builtin-kwork.c
index 14bf7a8429e7..73b5dc099a8a 100644
--- a/tools/perf/builtin-kwork.c
+++ b/tools/perf/builtin-kwork.c
@@ -406,12 +406,14 @@ static int work_push_atom(struct perf_kwork *kwork,

work = work_findnew(&class->work_root, &key, &kwork->cmp_id);
if (work == NULL) {
- free(atom);
+ atom_free(atom);
return -1;
}

- if (!profile_event_match(kwork, work, sample))
+ if (!profile_event_match(kwork, work, sample)) {
+ atom_free(atom);
return 0;
+ }

if (dst_type < KWORK_TRACE_MAX) {
dst_atom = list_last_entry_or_null(&work->atom_list[dst_type],
--
2.30.GIT



2023-09-06 17:50:21

by Arnaldo Carvalho de Melo

[permalink] [raw]
Subject: Re: [RFC v1 01/16] perf kwork: Fix incorrect and missing free atom in work_push_atom()

Em Sun, Sep 03, 2023 at 09:05:06PM -0700, Ian Rogers escreveu:
> On Sat, Aug 12, 2023 at 1:52 AM Yang Jihong <[email protected]> wrote:
> >
> > 1. Atoms are managed in page mode and should be released using atom_free()
> > instead of free().
> > 2. When the event does not match, the atom needs to free.
> >
> > Fixes: f98919ec4fcc ("perf kwork: Implement 'report' subcommand")
> > Signed-off-by: Yang Jihong <[email protected]>
>
> Reviewed-by: Ian Rogers <[email protected]>

Thanks, applied locally, will perform some tests and later today
probably push publicly, great work!

- Arnaldo


> > ---
> > tools/perf/builtin-kwork.c | 6 ++++--
> > 1 file changed, 4 insertions(+), 2 deletions(-)
> >
> > diff --git a/tools/perf/builtin-kwork.c b/tools/perf/builtin-kwork.c
> > index 14bf7a8429e7..73b5dc099a8a 100644
> > --- a/tools/perf/builtin-kwork.c
> > +++ b/tools/perf/builtin-kwork.c
> > @@ -406,12 +406,14 @@ static int work_push_atom(struct perf_kwork *kwork,
> >
> > work = work_findnew(&class->work_root, &key, &kwork->cmp_id);
> > if (work == NULL) {
> > - free(atom);
> > + atom_free(atom);
>
> Presumably this is fixing a memory-leak. It would be nice if kwork had
> perf tests, then our builds with -fsanitize=address would highlight
> this kind of issue. Here is a build command I use for this:
> make -C tools/perf O=/tmp/perf DEBUG=1 EXTRA_CFLAGS="-O0 -g
> -fno-omit-frame-pointer -fsanitize=address" BUILD_BPF_SKEL=1
> NO_LIBTRACEEVENT=1
>
> Thanks,
> Ian
>
> > return -1;
> > }
> >
> > - if (!profile_event_match(kwork, work, sample))
> > + if (!profile_event_match(kwork, work, sample)) {
> > + atom_free(atom);
> > return 0;
> > + }
> >
> > if (dst_type < KWORK_TRACE_MAX) {
> > dst_atom = list_last_entry_or_null(&work->atom_list[dst_type],
> > --
> > 2.30.GIT
> >

--

- Arnaldo