[RFC PATCH] arm64: Cache HW update of Access Flag support

[Gabriel Krisman Bertazi]

Will Deacon <will at kernel.org> writes:

> FWIW, I had a hack kicking around to avoid the bsearch. It wasn't too
> fiddly to rebase, so I've included it below if you want to take it for a
> spin. The grotty part is having to maintain the extra enum, which is why
> I didn't merge it in the past.

Hi Will,

I applied your patch, converted my fix to use it and re-ran the
benchmark.  For the record, the test is a mmtests workload that builds
the kernel a few times with allmod config.  If you want to reproduce it,
I used the configuration below:

  https://github.com/gormanm/mmtests/blob/master/configs/config-workload-kernbench-max

>From my observation, the array patch doesn't seem to meaningfully impact
this benchmark.  Using my RFC as a baseline, and comparing
read_sanitised_ftr_reg with bsearch and your patch (array), I got:

                          RFC                  bsearch			array
Amean     syst-20     2886.71 (   0.00%)     2913.26 *  -0.92%*     2911.01 *  -0.84%*

I sent a v2 that uses read_sanitised_ftr_reg on another thread [1].  If you decide to
merge your patch, I'm happy to re-spin it to use this new interface.

  [1] https://lore.kernel.org/linux-arm-kernel/20230109151955.8292-1-krisman@suse.de/

Thanks,

-- 
Gabriel Krisman Bertazi