On Wed, 2 Jul 2025 18:36:09 +0200
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > +static u64 get_timestamp(struct unwind_task_info *info)
> > +{
> > + lockdep_assert_irqs_disabled();
> > +
> > + if (!info->timestamp)
> > + info->timestamp = local_clock();
> > +
> > + return info->timestamp;
> > +}
>
> I'm very hesitant about this. Modern hardware can do this, but older
> hardware (think Intel Core and AMD Bulldozer etc hardware) might
> struggle with this. They don't have stable TSC and as such will use
> the magic in kernel/sched/clock.c; which can get stuck on a window
> edge for a little bit and re-use timestamps.
Well, the idea of using timestamps came from Microsoft as that's what
they do for a similar feature. It just seemed like an easier approach.
But I could definitely go back to the "cookie" idea that is just a per
cpu counter (with the CPU number as part of the cookie).
As the timestamp is likely not going to be as useful as it is with
Microsoft as there's no guarantee that the timestamp counter used is
the same as the timestamp used by the tracer asking for this, the cookie
approach may indeed be better.
-- Steve
