Re: [PATCH 2/3] perf: collect BPF metadata from existing BPF programs

[Namhyung Kim]

On Tue, Jun 03, 2025 at 02:54:50PM -0700, Blake Jones wrote:
> Hi Namhyung,
> 
> On Tue, Jun 3, 2025 at 2:44 PM Namhyung Kim <namhyung@xxxxxxxxxx> wrote:
> > > > IIUC the metadata is collected for each BPF program which may have
> > > > multiple subprograms.  Then this patch creates multiple PERF_RECORD_
> > > > BPF_METADATA for each subprogram, right?
> > > >
> > > > Can it be shared using the BPF program ID?
> > >
> > > In theory, yes, it could be shared. But I want to be able to correlate them
> > > with the corresponding PERF_RECORD_KSYMBOL events, and KSYMBOL events for
> > > subprograms don't have the full-program ID, so I wouldn't be able to do that.
> >
> > It's unfortunate that KSYMBOL doesn't have the program ID, but IIRC the
> > following BPF_EVENT should have it.  I think it's safe to think KSYMBOLs
> > belong to the BPF_EVENT when they are from the same thread.
> 
> Hmmm. Is that documented and tested anywhere? Offhand it sounds like an
> implementation detail that I wouldn't feel great about depending on -
> certainly not without a strong guarantee that it wouldn't change.

Good point.  Maybe BPF folks have some idea?

Anyway the current code generates them together in a function.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/events/core.c?h=v6.15#n9825

> 
> Can you say more about why the duplicated records concern you?

More data means more chance to lost something.  I don't expect this is
gonna be a practical concern but in general we should pursue less data.

Thanks,
Namhyung