Re: State of RFC PATCH Remove CONFIG_DCACHE_WORD_ACCESS

[Christophe Leroy]

Le 12/09/2022 à 17:22, Christophe Leroy a écrit :
> Le 12/09/2022 à 15:46, Lukas Bulwahn a écrit :
> > Hi Joe, hi Ben,
> >
> > While reviewing some kernel config, I came across
> > CONFIG_DCACHE_WORD_ACCESS and tried to understand its purpose.
> >
> > Then, I discovered this RFC patch from 2014 that seems never to have
> > been integrated:
> >
> > https://lore.kernel.org/all/1393964591.20435.58.camel@joe-AO722/
> > [RFC] Remove CONFIG_DCACHE_WORD_ACCESS
> >
> > The discussion seemed to just not continue and the patch was just not
> > integrated by anyone.
> >
> > In the meantime, the use of CONFIG_DCACHE_WORD_ACCESS has spread into
> > a few more files, but replacing it with
> > CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS still seems feasible.
> >
> > Are you aware of reasons that this patch from 2014 should not be 
> > integrated?
> >
> > I would spend some time to move the integration of this patch further
> > if you consider that the patch is not completely wrong.
>
> As far as I can see, for the time being this is not equivalent on powerpc:
>
> select HAVE_EFFICIENT_UNALIGNED_ACCESS  if !(CPU_LITTLE_ENDIAN && 
> POWER7_CPU)
>
> select DCACHE_WORD_ACCESS               if PPC64 && CPU_LITTLE_ENDIAN
>
> This will need to be investigated I guess.
>
> In the meantime I'll try to see if it makes any difference for ppc32.

Selecting DCACHE_WORD_ACCESS on powerpc32 provides an improvement of 
approx 9% on a powerpc 8xx and about 2% on a powerpc 832x, using the 
benchmark test in commit a75c380c7129 ("powerpc: Enable 
DCACHE_WORD_ACCESS on ppc64le")

Christophe