Re: [PATCH 07/34] perf/arm: optimize opencoded atomic find_bit() API

[Yury Norov]

On Tue, Nov 21, 2023 at 03:53:44PM +0000, Will Deacon wrote:
> On Sat, Nov 18, 2023 at 07:50:38AM -0800, Yury Norov wrote:
> > Switch subsystem to use atomic find_bit() or atomic iterators as
> > appropriate.
> > 
> > Signed-off-by: Yury Norov <yury.norov@xxxxxxxxx>
> > ---
> >  drivers/perf/arm-cci.c        | 23 +++++------------------
> >  drivers/perf/arm-ccn.c        | 10 ++--------
> >  drivers/perf/arm_dmc620_pmu.c |  9 ++-------
> >  drivers/perf/arm_pmuv3.c      |  8 ++------
> >  4 files changed, 11 insertions(+), 39 deletions(-)
> > 
> > diff --git a/drivers/perf/arm-cci.c b/drivers/perf/arm-cci.c
> > index 61de861eaf91..70fbf9d09d37 100644
> > --- a/drivers/perf/arm-cci.c
> > +++ b/drivers/perf/arm-cci.c
> > @@ -320,12 +320,8 @@ static int cci400_get_event_idx(struct cci_pmu *cci_pmu,
> >  		return CCI400_PMU_CYCLE_CNTR_IDX;
> >  	}
> >  
> > -	for (idx = CCI400_PMU_CNTR0_IDX; idx <= CCI_PMU_CNTR_LAST(cci_pmu); ++idx)
> > -		if (!test_and_set_bit(idx, hw->used_mask))
> > -			return idx;
> > -
> > -	/* No counters available */
> > -	return -EAGAIN;
> > +	idx = find_and_set_bit(hw->used_mask, CCI_PMU_CNTR_LAST(cci_pmu) + 1);
> 
> CCI400_PMU_CNTR0_IDX is defined as 1, so isn't this wrong?

You're right. Will fix in v2
 
> [...]
> 
> > diff --git a/drivers/perf/arm_dmc620_pmu.c b/drivers/perf/arm_dmc620_pmu.c
> > index 30cea6859574..e41c84dabc3e 100644
> > --- a/drivers/perf/arm_dmc620_pmu.c
> > +++ b/drivers/perf/arm_dmc620_pmu.c
> > @@ -303,13 +303,8 @@ static int dmc620_get_event_idx(struct perf_event *event)
> >  		end_idx = DMC620_PMU_MAX_COUNTERS;
> >  	}
> >  
> > -	for (idx = start_idx; idx < end_idx; ++idx) {
> > -		if (!test_and_set_bit(idx, dmc620_pmu->used_mask))
> > -			return idx;
> > -	}
> > -
> > -	/* The counters are all in use. */
> > -	return -EAGAIN;
> > +	idx = find_and_set_next_bit(dmc620_pmu->used_mask, end_idx, start_idx);
> 
> It might just be me, but I'd find this a tonne easier to read if you swapped
> the last two arguments around so that the offset came before the limit in
> the new function.

I personally agree, but we already have find_next_*_bit(addr, nbits, offset)
functions, and having atomic versions of the same with different order
of arguments will make it even more messy...

Thanks,
        Yury