On 02/17/2016 09:14 AM, Tim Sander wrote: > Hi Sebastian Hi Tim, > I have done a bisect run, its a rather innocent looking on liner which seems > to cause the problems. The numbers where reasonably stable so i am pretty > confident that this is the patch giving ~26µs additional latency on the Altera > SOC plattform: > > eec2bf477ac674583a7d73b9d00f47c528b7266d is the first bad commit > commit eec2bf477ac674583a7d73b9d00f47c528b7266d > Author: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> > Date: Thu Feb 4 16:38:10 2016 +0100 > > kernel/perf: mark perf_cpu_context's timer as irqsafe > > Otherwise we get a WARN_ON() backtrace and some events are reported as > "not counted". > > Cc: stable-rt@xxxxxxxxxxxxxxx > Reported-by: Yang Shi <yang.shi@xxxxxxxxxx> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> Seriously? That patch? I played a little and I doubt seriously that this patch has something to do with it. So before that patch you would have a warn_on spotted and complained if that timer would fire. So that is one reason why I doubt that this patch is in charge of the 25us. If I add a printk() to that timer I don't see it under "normal" circumstances. However I do perf_4.3 stat -e branches,branch-misses,bus-cycles,cache-misses,cache-references,cycles,instructions apt-get update then I see them fire. > Best Regards > Tim Sebastian -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
