Messages in this thread | | | Date | Sun, 5 Jul 2009 03:11:23 +0200 | From | Ingo Molnar <> | Subject | Re: [PATCH 4/6 -tip] perf_counter: Add Generalized Hardware interrupt support for AMD |
| |
* Jaswinder Singh Rajput <jaswinder@kernel.org> wrote:
> On Sat, 2009-07-04 at 12:22 +0200, Ingo Molnar wrote: > > * Jaswinder Singh Rajput <jaswinder@kernel.org> wrote: > > > > > On Wed, 2009-07-01 at 13:24 +0200, Ingo Molnar wrote: > > > > * Jaswinder Singh Rajput <jaswinder@kernel.org> wrote: > > > > > > > > > > > > > > $ ./perf stat -e interrupts -e masked -e int-pending-mask-cycles -- ls -lR /usr/include/ > /dev/null > > > > > > > > > > Performance counter stats for 'ls -lR /usr/include/': > > > > > > > > > > 377 interrupts > > > > > 53429936 int-mask-cycles > > > > > 1119 int-pending-mask-cycles > > > > > > > > > > 0.371457539 seconds time elapsed > > > > > > > > Agreed, this is another useful generalization - and the 'cycles > > > > pending' metrics are not retrievable via any software means. > > > > > > > > We could and should probably add a software counter for hardirqs > > > > as wel. That would allow the vector/irqnr information to be > > > > passed in, and it would allow architectures without irq metrics > > > > in the PMU to have this counter too. > > > > > > > > > > Please let me know that addition of software counter will be > > > in this patch or we can do it incrementally after this patch. > > > > It should be in this series. That way we can cross-check whether > > the soft counts and the hard counts match up and find potential > > bugs that way, etc. > > > > You want to cross check performance counter events ?
Yes. The events are also more complete if we add per IRQ source counts as well, not just summary counts.
Ingo
| |