Re: [PATCH v1 10/13] perf inject: Fix build ID injection
From: Arnaldo Carvalho de Melo
Date: Tue Sep 03 2024 - 14:34:16 EST
On Mon, Sep 02, 2024 at 11:26:09AM -0700, Namhyung Kim wrote:
> Hi Ian,
>
> On Wed, Aug 28, 2024 at 08:15:10AM -0700, Ian Rogers wrote:
> > On Mon, Aug 19, 2024 at 12:54 PM Ian Rogers <irogers@xxxxxxxxxx> wrote:
> > >
> > > On Mon, Aug 19, 2024 at 11:01 AM Arnaldo Carvalho de Melo
> > > <acme@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, Aug 16, 2024 at 11:44:39PM -0700, Ian Rogers wrote:
> > > > > Build ID injection wasn't inserting a sample ID and aligning events to
> > > > > 64 bytes rather than 8. No sample ID means events are unordered and
> > > > > two different build_id events for the same path, as happens when a
> > > > > file is replaced, can't be differentiated.
> > > > >
> > > > > Add in sample ID insertion for the build_id events alongside some
> > > > > refactoring. The refactoring better aligns the function arguments for
> > > > > different use cases, such as synthesizing build_id events without
> > > > > needing to have a dso. The misc bits are explicitly passed as with
> > > > > callchains the maps/dsos may span user and kernel land, so using
> > > > > sample->cpumode isn't good enough.
> > > > >
> > > > > Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> > > > > ---
> > > > > tools/perf/builtin-inject.c | 170 ++++++++++++++++++++++-------
> > > > > tools/perf/util/build-id.c | 6 +-
> > > > > tools/perf/util/synthetic-events.c | 44 ++++++--
> > > > > tools/perf/util/synthetic-events.h | 10 +-
> > > > > 4 files changed, 175 insertions(+), 55 deletions(-)
> > > > >
> > > > > diff --git a/tools/perf/builtin-inject.c b/tools/perf/builtin-inject.c
> > > > > index a7c859db2e15..84a4bdb5cb0a 100644
> > > > > --- a/tools/perf/builtin-inject.c
> > > > > +++ b/tools/perf/builtin-inject.c
> > > > > @@ -131,6 +131,7 @@ struct perf_inject {
> > > > > struct perf_file_section secs[HEADER_FEAT_BITS];
> > > > > struct guest_session guest_session;
> > > > > struct strlist *known_build_ids;
> > > > > + const struct evsel *mmap_evsel;
> > > > > };
> > > > >
> > > > > struct event_entry {
> > > > > @@ -139,8 +140,13 @@ struct event_entry {
> > > > > union perf_event event[];
> > > > > };
> > > > >
> > > > > -static int dso__inject_build_id(struct dso *dso, const struct perf_tool *tool,
> > > > > - struct machine *machine, u8 cpumode, u32 flags);
> > > > > +static int dso__inject_build_id(const struct perf_tool *tool,
> > > > > + struct perf_sample *sample,
> > > > > + struct machine *machine,
> > > > > + const struct evsel *evsel,
> > > > > + __u16 misc,
> > > > > + const char *filename,
> > > > > + struct dso *dso, u32 flags);
> > > >
> > > > So in the end the dso was needed, the name of the function remains
> > > > dso__something(), so first arg would be a 'struct dso *'
> > > >
> > > > I processed the patches up to 9/13, so that they can get tested now.
> > >
> > > Maybe we should rename the function? We can get the build ID from
> > > mmap2 events now, not just stored away in the dso by build_id events.
> > > Reordering the arguments isn't a problem, I was just aiming for
> > > consistency between the caller, perf_event__synthesize_build_id and
> > > eventually the call to process that all take the arguments in this
> > > order.
> >
> > Arnaldo, did you have any more thoughts on this?
>
> I'm ok to rename it to tool__inject_build_id().
I thought I had this processed, but I stopped at this patch, yeah,
please rebase on top of perf-tools-next with the suggested rename.
- Arnaldo