perf: register pmu implementations
in [Kernel]
|
Prev: [PATCH] procfs: Do not release pid_ns->proc_mnt too early
Next: perf/perf_events: misleading number of samples due to mmap()
From: Paul Mackerras on 8 Jul 2010 23:10 On Thu, Jun 24, 2010 at 04:28:09PM +0200, Peter Zijlstra wrote: > Simple registration interface for struct pmu, this provides the > infrastructure for removing all the weak functions. Seems to work on powerpc, but the sh bits don't seem quite right: > Index: linux-2.6/arch/sh/kernel/perf_event.c > =================================================================== > --- linux-2.6.orig/arch/sh/kernel/perf_event.c > +++ linux-2.6/arch/sh/kernel/perf_event.c > @@ -257,24 +257,24 @@ static void sh_pmu_read(struct perf_even > sh_perf_event_update(event, &event->hw, event->hw.idx); > } > > -static struct pmu pmu = { > - .enable = sh_pmu_enable, > - .disable = sh_pmu_disable, > - .read = sh_pmu_read, > -}; > - > -struct pmu *hw_perf_event_init(struct perf_event *event) > +static in sh_pmu_event_init(struct perf_event *event) int? > { > int err = __hw_perf_event_init(event); We need a switch on event->attr.type so we return -ENOENT if it's not PERF_TYPE_{HARDWARE,HW_CACHE,RAW}. As it is we don't ever return -ENOENT, which might stop software and tracepoint events from working. Paul. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Peter Zijlstra on 9 Jul 2010 04:20
On Fri, 2010-07-09 at 13:08 +1000, Paul Mackerras wrote: > > -struct pmu *hw_perf_event_init(struct perf_event *event) > > +static in sh_pmu_event_init(struct perf_event *event) > > int? Argh, I fixed that a few times, but the hunk keeps slipping into different patches.. cured. > > { > > int err = __hw_perf_event_init(event); > > We need a switch on event->attr.type so we return -ENOENT if it's > not PERF_TYPE_{HARDWARE,HW_CACHE,RAW}. As it is we don't ever return > -ENOENT, which might stop software and tracepoint events from working. Aaah, indeed! That is why Matt's perf record broke, perf record defaults to -e cycles which automagically falls back to a software timer, which then doesn't work. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |