* perf: rdpmc and PERF_EVENT_IOC_RESET
@ 2016-08-17 4:53 Vince Weaver
2016-08-17 15:26 ` Peter Zijlstra
0 siblings, 1 reply; 3+ messages in thread
From: Vince Weaver @ 2016-08-17 4:53 UTC (permalink / raw)
To: linux-kernel; +Cc: Ingo Molnar, Peter Zijlstra, Arnaldo Carvalho de Melo
Hello
so using rdpmc() and the mmap page to do fast perf_event reads seems to
interact poorly with the PERF_EVENT_IOC_RESET ioctl.
>From what I can tell, on reset event->count is set to zero, but
event->hw.prev_count is not, so the userpg->offset field ends up negative
and weird things happen.
Shout reset just not be called if you are using the rdpmc() interface?
Vince
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: perf: rdpmc and PERF_EVENT_IOC_RESET
2016-08-17 4:53 perf: rdpmc and PERF_EVENT_IOC_RESET Vince Weaver
@ 2016-08-17 15:26 ` Peter Zijlstra
2016-08-18 4:32 ` Vince Weaver
0 siblings, 1 reply; 3+ messages in thread
From: Peter Zijlstra @ 2016-08-17 15:26 UTC (permalink / raw)
To: Vince Weaver; +Cc: linux-kernel, Ingo Molnar, Arnaldo Carvalho de Melo
On Wed, Aug 17, 2016 at 12:53:01AM -0400, Vince Weaver wrote:
> Hello
>
> so using rdpmc() and the mmap page to do fast perf_event reads seems to
> interact poorly with the PERF_EVENT_IOC_RESET ioctl.
Hurm.. I never considered using RESET with rdpmc(). The typical usage of
rdpmc() I considered is something like:
u64 start = rdpmc();
/* code goes here */
delta = rdpmc() - start;
That said; I don't object to fixing the interaction if it doesn't have
downsides.
> From what I can tell, on reset event->count is set to zero, but
> event->hw.prev_count is not, so the userpg->offset field ends up negative
> and weird things happen.
_perf_event_reset()
perf_event_read() /* updates prev_count */
local64_set(count, 0);
after that, like you say: userpg->offset = perf_event_count() -
prev_count, which does indeed end up negative.
But the rdpmc user function should:
u64 mmap_read_self(void *addr)
{
struct perf_event_mmap_page *pc = addr;
u32 seq, idx, time_mult = 0, time_shift = 0, width = 0;
u64 count, cyc = 0, time_offset = 0, enabled, running, delta;
s64 pmc = 0;
do {
seq = pc->lock;
barrier();
/* enabled/running muck */
idx = pc->index;
count = pc->offset;
if (pc->cap_user_rdpmc && idx) {
width = pc->pmc_width;
pmc = rdpmc(idx - 1);
}
barrier();
} while (pc->lock != seq);
if (idx) {
pmc <<= 64 - width;
pmc >>= 64 - width; /* shift right signed */
count += pmc;
}
/* more enabled/running muck */
return count;
}
Which sign-extends the RDPMC result and adds it to userpg->offset,
resulting in a positive number again.
So I think it should work..
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: perf: rdpmc and PERF_EVENT_IOC_RESET
2016-08-17 15:26 ` Peter Zijlstra
@ 2016-08-18 4:32 ` Vince Weaver
0 siblings, 0 replies; 3+ messages in thread
From: Vince Weaver @ 2016-08-18 4:32 UTC (permalink / raw)
To: Peter Zijlstra; +Cc: linux-kernel, Ingo Molnar, Arnaldo Carvalho de Melo
On Wed, 17 Aug 2016, Peter Zijlstra wrote:
> Hurm.. I never considered using RESET with rdpmc(). The typical usage of
> rdpmc() I considered is something like:
I'm finally trying to hook up PAPI to properly use rdpmc and PAPI likes to
use IOC_RESET in various places. I can just special case to not do so
when in rdpmc mode, but I got curious if RESET was expected to work.
> But the rdpmc user function should:
> ...
>
> Which sign-extends the RDPMC result and adds it to userpg->offset,
> resulting in a positive number again.
I will have to check to see if PAPI is using the proper code.
What happens is the procesed result has bit 49 set for some reason, so the
results are 2^49 too big. I've been trying to track it down, but the PAPI
code is a bit of a mess (to put it mildly).
Vince
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2016-08-18 4:32 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-08-17 4:53 perf: rdpmc and PERF_EVENT_IOC_RESET Vince Weaver
2016-08-17 15:26 ` Peter Zijlstra
2016-08-18 4:32 ` Vince Weaver
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).