Re: [PATCH 03/23] perf: Add ability to attach user level registers dump to sample

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Wed, 2012-06-27 at 17:24 +0200, Frederic Weisbecker wrote:
> On Wed, Jun 27, 2012 at 05:13:44PM +0200, Stephane Eranian wrote:
> > On Wed, Jun 27, 2012 at 5:11 PM, Frederic Weisbecker <fweisbec@gmail.com> wrote:
> > > On Tue, Jun 19, 2012 at 05:47:54PM +0200, Jiri Olsa wrote:
> > >> @@ -4061,6 +4105,19 @@ void perf_prepare_sample(struct perf_event_header *header,
> > >>               }
> > >>               header->size += size;
> > >>       }
> > >> +
> > >> +     if (sample_type & PERF_SAMPLE_REGS_USER) {
> > >> +             /* regs dump available bool */
> > >> +             int size = sizeof(u64);
> > >> +
> > >> +             data->regs_user = perf_sample_regs_user(regs);
> > >> +             if (data->regs_user) {
> > >> +                     u64 mask = event->attr.sample_regs_user;
> > >> +                     size += hweight64(mask) * sizeof(u64);
> > >> +             }
> > >> +
> > >> +             header->size += size;
> > >
> > > We'll need to remove the 64 bits registers for compat tasks but other than
> > > that, it looks ok.
> > You cannot do this. You cannot remove register values from under the hood.
> > The only way the user has to parse the sample is the sample_regs_users bitmask.
> > You have to return 0 for those unexisting regs for compat tasks.
> 
> You mean fill unexisting reg values with 0? Yeah that works.

What does x32 look like? Is that still reported as a compat task? If so,
we should record all registers and not 0 out anything.