[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5088c94c-3c80-0799-08a2-f1d53b95380b@c-s.fr>
Date: Tue, 29 Oct 2019 17:10:28 +0100
From: Christophe Leroy <christophe.leroy@....fr>
To: Michael Ellerman <mpe@...erman.id.au>,
Benjamin Herrenschmidt <benh@...nel.crashing.org>,
Paul Mackerras <paulus@...ba.org>
Cc: linux-kernel@...r.kernel.org, linuxppc-dev@...ts.ozlabs.org
Subject: Re: [PATCH v2 1/8] powerpc/32: Add VDSO version of getcpu
Le 18/09/2019 à 07:51, Michael Ellerman a écrit :
>
> We are still in the middle of the years long process of removing the
> "magic" syscall on 64-bit:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/kernel/exceptions-64s.S?commit=4d856f72c10ecb060868ed10ff1b1453943fc6c8#n1578
>
>
> Can we not add another one on 32-bit?
>
> Is it really such a fast path that it's worth putting a wart in the
> syscall entry like that?
>
> Is there some other method? On s390 they have a per-cpu VDSO page, that
> would be a nice option. How we do that would be specific to a particular
> MMU, and maybe not even possible with some MMUs. So maybe that's not
> feasible.
Ok, for now I remove the fast syscall and only keep the VDSO getcpu()
for non SMP.
I think we may in the future implement a per-cpu VDSO page that will be
mapped to the process based on the CPU it is running on. I need to look
at that in more details.
Christophe
Powered by blists - more mailing lists