Excerpts from Madhavan Srinivasan's message of August 17, 2021 11:06 pm: > > On 8/16/21 12:59 PM, Nicholas Piggin wrote: >> Interrupt handling code would like to know whether perf is enabled, to >> know whether it should enable MSR[EE] to improve PMI coverage. >> >> Cc: Madhavan Srinivasan <[email protected]> >> Cc: Athira Rajeev <[email protected]> >> Signed-off-by: Nicholas Piggin <[email protected]> >> --- >> arch/powerpc/include/asm/hw_irq.h | 2 ++ >> arch/powerpc/perf/core-book3s.c | 13 +++++++++++++ >> 2 files changed, 15 insertions(+) >> >> diff --git a/arch/powerpc/include/asm/hw_irq.h >> b/arch/powerpc/include/asm/hw_irq.h >> index 21cc571ea9c2..2d5c0d3ccbb6 100644 >> --- a/arch/powerpc/include/asm/hw_irq.h >> +++ b/arch/powerpc/include/asm/hw_irq.h >> @@ -306,6 +306,8 @@ static inline bool lazy_irq_pending_nocheck(void) >> return __lazy_irq_pending(local_paca->irq_happened); >> } >> >> +bool power_pmu_running(void); >> + >> /* >> * This is called by asynchronous interrupts to conditionally >> * re-enable hard interrupts after having cleared the source >> diff --git a/arch/powerpc/perf/core-book3s.c >> b/arch/powerpc/perf/core-book3s.c >> index bb0ee716de91..76114a9afb2b 100644 >> --- a/arch/powerpc/perf/core-book3s.c >> +++ b/arch/powerpc/perf/core-book3s.c >> @@ -2380,6 +2380,19 @@ static void perf_event_interrupt(struct pt_regs *regs) >> perf_sample_event_took(sched_clock() - start_clock); >> } >> >> +bool power_pmu_running(void) >> +{ >> + struct cpu_hw_events *cpuhw; >> + >> + /* Could this simply test local_paca->pmcregs_in_use? */ >> + >> + if (!ppmu) >> + return false; > > > This covers only when perf platform driver is not registered, > but we should also check for MMCR0[32], since pmu sprs can be > accessed via sysfs.
In that case do they actually do anything with the PMI? I don't think it should matter hopefully. But I do think a lot of this stuff could be cleaned up. We have pmcs_enabled and ppc_enable_pmcs() in sysfs.c, ppc_set_pmu_inuse(), ppc_md.enable_pmcs(), reserve_pmc_hardware(), etc and different users call different things. We don't consistently disable either, e.g., we never disable the H_PERFMON facility after we stop using perf even though it says that slows down partition switch. I started to have a look at sorting it out but it looks like a big job so would take a bit of time if we want to do it. Thanks, Nick
