Re: [v3 03/12] x86/fsgsbase/64: Add intrinsics/macros for FSGSBASE instructions

From: Andy Lutomirski
Date: Thu Oct 25 2018 - 19:14:32 EST


On Wed, Oct 24, 2018 at 12:21 PM Andi Kleen <ak@xxxxxxxxxxxxxxx> wrote:
>
> On Wed, Oct 24, 2018 at 11:53:54AM -0700, Andy Lutomirski wrote:
> > On Tue, Oct 23, 2018 at 11:43 AM Chang S. Bae <chang.seok.bae@xxxxxxxxx> wrote:
> > >
> > > From: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> > >
> > > Add C intrinsics and assembler macros for the new FSBASE and GSBASE
> > > instructions.
> > >
> > > Very straight forward. Used in followon patches.
> > >
> > > [ luto: Rename the variables from FS and GS to FSBASE and GSBASE and
> > > make <asm/fsgsbase.h> safe to include on 32-bit kernels. ]
> > >
> > > v2: Use __always_inline
> > >
> > > [ chang: Revise the changelog. Place them in <asm/fsgsbase.h>. Replace
> > > the macros with GAS-compatible ones. ]
> > >
> > > If GCC supports it, we can add -mfsgsbase to CFLAGS and use the builtins
> > > here for extra performance.
> >
> > Reviewed-by: Andy Lutomirski <luto@xxxxxxxxxx> # C parts only
> >
> > With the caveat that I'm not convinced that the memory clobbers are
> > needed. The __force_order trick in special_insns.h would probably be
> > more appropriate.
> >
> > I don't feel qualified to review the asm part without some research.
> > Whereas hpa or Boris could probably review it with their eyes closed.
>
> BTW the other option would be to update the min-binutils requirement
> to 2.21 (currently it is 2.20) and then write it directly without .byte.
> I believe 2.21 added support for these instructions.
>
> (It's only a binutils requirement, don't need gcc support)
>

I'd personally be fine with this. Linus? Thomas? Ingo?