Re: [PATCH 1/9] perf/core: Add PERF_RECORD_CGROUP event

[Namhyung Kim]

Hi Tejun,

On Sat, Dec 21, 2019 at 12:16 AM Tejun Heo <tj@xxxxxxxxxx> wrote:
>
> On Fri, Dec 20, 2019 at 10:33:35AM +0100, Peter Zijlstra wrote:
> > On Fri, Dec 20, 2019 at 01:32:45PM +0900, Namhyung Kim wrote:
> > > To support cgroup tracking, add CGROUP event to save a link between
> > > cgroup path and inode number.  The attr.cgroup bit was also added to
> > > enable cgroup tracking from userspace.
> > >
> > > This event will be generated when a new cgroup becomes active.
> > > Userspace might need to synthesize those events for existing cgroups.
> > >
> > > Cc: Tejun Heo <tj@xxxxxxxxxx>
> > > Cc: Li Zefan <lizefan@xxxxxxxxxx>
> > > Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
> > > Cc: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> > > Signed-off-by: Namhyung Kim <namhyung@xxxxxxxxxx>
> >
> > TJ, is this the right thing to do? ISTR you had concerns on this topic
> > on the past.
>
> Yeah, cgroup->id is now the same as ino (on 64bit ino matchines) and
> fhandle and uniquely identifies a cgroup instance in that boot
> instance.  That said, id -> path mapping can be done from userspace by
> passing the cgroup id to open_by_handle_at(2) and then reading the
> symlink in /proc/self/fd, so this event isn't necessary per-se if the
> goal is mapping back ids to paths.

But we should support offline report even on a different machine.
Also cgroups might go away during the record. so I think we need it
anyway.

Thanks
Namhyung