Re: [PATCH 04/16] perf: Add ability to attach user stack dump tosample

[Jiri Olsa]

On Thu, Jun 07, 2012 at 12:07:56PM +0200, Frederic Weisbecker wrote:
> On Thu, May 24, 2012 at 12:51:20PM +0200, Peter Zijlstra wrote:
> > On Wed, 2012-05-23 at 21:32 +0200, Jiri Olsa wrote:
> > > +static void
> > > +perf_output_sample_ustack(struct perf_output_handle *handle, u64 dump_size,
> > > +                         struct pt_regs *regs)
> > > +{
> > > +       u64 size;
> > > +
> > > +       /* Case of a kernel thread, nothing to dump */
> > > +       if (!regs) {
> > > +               size = 0;
> > > +               perf_output_put(handle, size);
> > > +       } else {
> > > +               unsigned long sp;
> > > +               unsigned int rem;
> > > +               u64 dyn_size;
> > > +
> > > +               /*
> > > +                * Static size: we always dump the size
> > > +                * requested by the user because most of the
> > > +                * time, the top of the user stack is not
> > > +                * paged out.
> > > +                */
> > > +               size = round_up(dump_size, sizeof(u64));
> > 
> > You also do this in the prepare thing..
> > 
> > > +               perf_output_put(handle, size);
> > > +
> > > +               sp = user_stack_pointer(regs);
> > > +               rem = __output_copy_user(handle, (void *)sp, size);
> > > +               dyn_size = size - rem;
> > > +
> > > +               /* What couldn't be dumped is zero padded */
> > > +               while (rem--) {
> > > +                       char zero = 0;
> > > +                       perf_output_put(handle, zero);
> > > +               }
> > 
> > Does this matter? If we don't write it the worst that can happen is that
> > we leave previous ring-bugger content around, but since we already are
> > privileged to read that (and very likely already have) there's no
> > problem with that..
> > 
> > I know not zero-ing is ugly, but its also faster.. and do we care about
> > them silly zeros?
> 
> The problem there is that the unwinder may then deal with random values
> from previous records in the buffer and think it is some real stack values.
> This may give strange results. At least zeroing was limiting this effect.
> 
> I think the best would be to do like you say: if we can't dump everything
> then just leave the state of the buffer as is. But also output in the end
> the effective size of the stack that has been dumped. So that we know what
> we can give to the unwinder and what we can not.

yep, thats the case right now.. just putting away the zeros

jirka
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/