Re: perf_event: rdpmc self-monitoring overhead issue

[Andi Kleen]

> I assume he means MAP_POPULATE

Yes.

> 
> which does improve things, from ~3000 cycles to ~219 cycles but that's 
> still more overhead than the ~130 or so you get by manually touching the 
> page first.

That seems odd. It should be the same.

Can you do a trace-cmd function trace and compare the two cases?

trace-cmd record -p function_graph ...
trace-cmd report

(as usual for tracing perf remove the useless -pg removal for perf in
kernel/events/Makefile and arch/x86/kernel/cpu/Makefile first)

-Andi
--
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/