Re: [PATCH 2/5] fs/procfs: implement efficient VMA querying API for /proc/<pid>/maps

[Arnaldo Carvalho de Melo]

On Mon, May 06, 2024 at 03:53:40PM -0300, Arnaldo Carvalho de Melo wrote:
> On Mon, May 06, 2024 at 11:05:17AM -0700, Namhyung Kim wrote:
> > On Mon, May 6, 2024 at 6:58 AM Arnaldo Carvalho de Melo <acme@xxxxxxxxxx> wrote:
> > > On Sat, May 04, 2024 at 02:50:31PM -0700, Andrii Nakryiko wrote:
> > > > On Sat, May 4, 2024 at 8:28 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > > > On Fri, May 03, 2024 at 05:30:03PM -0700, Andrii Nakryiko wrote:
> > > > > > Note also, that fetching VMA name (e.g., backing file path, or special
> > > > > > hard-coded or user-provided names) is optional just like build ID. If
> > > > > > user sets vma_name_size to zero, kernel code won't attempt to retrieve
> > > > > > it, saving resources.
> 
> > > > > > Signed-off-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> 
> > > > > Where is the userspace code that uses this new api you have created?
> 
> > > > So I added a faithful comparison of existing /proc/<pid>/maps vs new
> > > > ioctl() API to solve a common problem (as described above) in patch
> > > > #5. The plan is to put it in mentioned blazesym library at the very
> > > > least.
> > > >
> > > > I'm sure perf would benefit from this as well (cc'ed Arnaldo and
> > > > linux-perf-user), as they need to do stack symbolization as well.
>  
> > I think the general use case in perf is different.  This ioctl API is great
> > for live tracing of a single (or a small number of) process(es).  And
> > yes, perf tools have those tracing use cases too.  But I think the
> > major use case of perf tools is system-wide profiling.
>  
> > For system-wide profiling, you need to process samples of many
> > different processes at a high frequency.  Now perf record doesn't
> > process them and just save it for offline processing (well, it does
> > at the end to find out build-ID but it can be omitted).
> 
> Since:
> 
>   Author: Jiri Olsa <jolsa@xxxxxxxxxx>
>   Date:   Mon Dec 14 11:54:49 2020 +0100
>   1ca6e80254141d26 ("perf tools: Store build id when available in PERF_RECORD_MMAP2 metadata events")
> 
> We don't need to to process the events to find the build ids. I haven't
> checked if we still do it to find out which DSOs had hits, but we
> shouldn't need to do it for build-ids (unless they were not in memory
> when the kernel tried to stash them in the PERF_RECORD_MMAP2, which I
> haven't checked but IIRC is a possibility if that ELF part isn't in
> memory at the time we want to copy it).

> If we're still traversing it like that I guess we can have a knob and
> make it the default to not do that and instead create the perf.data
> build ID header table with all the build-ids we got from
> PERF_RECORD_MMAP2, a (slightly) bigger perf.data file but no event
> processing at the end of a 'perf record' session.

But then we don't process the PERF_RECORD_MMAP2 in 'perf record', it
just goes on directly to the perf.data file :-\

Humm, perhaps the sideband thread...

- Arnaldo