Re: [PATCH] futex: Set USER_DS for the futex_detect_cmpxchg() test

[Andy Lutomirski]

On Fri, Sep 28, 2018 at 7:53 AM Martin Schwidefsky
<schwidefsky@xxxxxxxxxx> wrote:
>
> On Fri, 28 Sep 2018 07:11:44 -0700
> Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>
> > > On Sep 28, 2018, at 1:42 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> > >
> > >> On Fri, 28 Sep 2018, Martin Schwidefsky wrote:
> > >> On Fri, 28 Sep 2018 09:12:10 +0200
> > >> Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > >>> I don't know if that has happened, and whether it would work on s390 now.
> > >>
> > >> commit 03b8c7b623c80af264c4c8d6111e5c6289933666
> > >> Author: Heiko Carstens <heiko.carstens@xxxxxxxxxx>
> > >> Date:   Sun Mar 2 13:09:47 2014 +0100
> > >>
> > >>    futex: Allow architectures to skip futex_atomic_cmpxchg_inatomic() test
> > >>
> > >>    If an architecture has futex_atomic_cmpxchg_inatomic() implemented and there
> > >>    is no runtime check necessary, allow to skip the test within futex_init().
> > >>
> > >>    This allows to get rid of some code which would always give the same result,
> > >>    and also allows the compiler to optimize a couple of if statements away.
> > >>
> > >>    Signed-off-by: Heiko Carstens <heiko.carstens@xxxxxxxxxx>
> > >>    Cc: Finn Thain <fthain@xxxxxxxxxxxxxxxxxxx>
> > >>    Cc: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> > >>    Link: http://lkml.kernel.org/r/20140302120947.GA3641@osiris
> > >>    Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > >>
> > >>
> > >> Heiko created the CONFIG_HAVE_FUTEX_CMPXCHG to get around this issue.
> > >> We just skip the runtime check as well as arc, m68k and sh. Not sure
> > >> about xtensa, the set it config option only for !MMU.
> > >
> > > Duh. grep would have told me. -ENOTENOUGHCOFFEE
> > >
> > >
> >
> > Thereâs another way to skin this cat: keep KERNEL_DS but pass a valid pointer.
> > I donât suppose you remember why you didnât do that?
>
> No, I don't remember. To use a valid kernel pointer with KERNEL_DS and
> then test for == 0 (vs -ENOSYS) imho should work.
>

There may be a much nicer solution.  Unless I missed something, only
mips and xtensa even have the possibility of cmpxchg being missing.
We could just make those arches supply a futex-detecting helper.