On Wed, 2 Jul 2025 at 09:42, Steven Rostedt [off-list ref] wrote:
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.
I think having just a percpu counter is probably the safest thing to
do, since the main reason just seems to be "correlate with the user
event". Using some kind of "real time" for correlation purposes seems
like a bad idea from any portability standpoint, considering just how
many broken timers we've seen across pretty much every architecture
out there.
Also, does it actually have to be entirely unique? IOW, a 32-bit
counter (or even less) might be sufficient if there's some guarantee
that processing happens before the counter wraps around? Again - for
correlation purposes, just *how* many outstanding events can you have
that aren't ordered by other things too?
I'm sure people want to also get some kind of rough time idea, but
don't most perf events have them simply because people want time
information for _informatioal_ reasons, rather than to correlate two
events?
Linus