Re: [perf/x86] 81ec3f3c4c: will-it-scale.per_process_ops -5.5% regression

[Jiri Olsa <jolsa@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 3068 bytes --]

On Fri, Feb 21, 2020 at 04:03:25PM +0800, Feng Tang wrote:
> 
> On Wed, Feb 05, 2020 at 01:58:04PM +0100, Peter Zijlstra wrote:
> > On Wed, Feb 05, 2020 at 08:32:16PM +0800, kernel test robot wrote:
> > > Greeting,
> > > 
> > > FYI, we noticed a -5.5% regression of will-it-scale.per_process_ops due to commit:
> > > 
> > > 
> > > commit: 81ec3f3c4c4d78f2d3b6689c9816bfbdf7417dbb ("perf/x86: Add check_period PMU callback")
> > > https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
> > > 
> > 
> > I'm fairly sure this bisect/result is bogus.
> 
> 
> Hi Peter,
> 
> Some updates:
> 
> We checked more on this. We run 14 times test for it, and the
> results are consistent about the 5.5% degradation, and we
> run the same test on several other platforms, whose test results
> are also consistent, though there are no such -5.5% seen.
> 
> We are also curious that the commit seems to be completely not
> relative to this scalability test of signal, which starts a task
> for each online CPU, and keeps calling raise(), and calculating
> the run numbers.
> 
> One experiment we did is checking which part of the commit
> really affects the test, and it turned out to be the change of
> "struct pmu". Effectively, applying this patch upon 5.0-rc6 
> which triggers the same regression.
> 
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 1d5c551..e1a0517 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -447,6 +447,11 @@ struct pmu {
>  	 * Filter events for PMU-specific reasons.
>  	 */
>  	int (*filter_match)		(struct perf_event *event); /* optional */
> +
> +	/*
> +	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
> +	 */
> +	int (*check_period)		(struct perf_event *event, u64 value); /* optional */
>  };
> 
> So likely, this commit changes the layout of the kernel text
> and data, which may trigger some cacheline level change. From
> the system map of the 2 kernels, a big trunk of symbol's address
> changes which follow the global "pmu",

nice, I wonder we could see that in perf c2c output ;-)
I'll try to run and check

thanks,
jirka

> 
> 5.0-rc6-systemap:
> 
> ffffffff8221d000 d pmu
> ffffffff8221d100 d pmc_reserve_mutex
> ffffffff8221d120 d amd_f15_PMC53
> ffffffff8221d160 d amd_f15_PMC50
> 
> 5.0-rc6+pmu-change-systemap:
> 
> ffffffff8221d000 d pmu
> ffffffff8221d120 d pmc_reserve_mutex
> ffffffff8221d140 d amd_f15_PMC53
> ffffffff8221d180 d amd_f15_PMC50
> 
> But we can hardly identify which exact symbol is responsible
> for the change, as too many symbols are offseted. 
> 
> btw, we've seen similar case that an irrelevant commit changes
> the benchmark, like a hugetlb patch improves pagefault test on
> a platform that never uses hugetlb https://lkml.org/lkml/2020/1/14/150  
> 
> Thanks,
> Feng
> 
> > _______________________________________________
> > LKP mailing list -- lkp(a)lists.01.org
> > To unsubscribe send an email to lkp-leave(a)lists.01.org
>