Re: [PATCH 2/2] x86/speculation: Simplify indirect_branch_prediction_barrier()

[Josh Poimboeuf]

On Tue, Feb 06, 2018 at 07:44:52PM +0000, David Woodhouse wrote:
> On Fri, 2018-01-26 at 21:08 +0100, Borislav Petkov wrote:
> > Make it all a function which does the WRMSR instead of having a hairy
> > inline asm.
> 
> ...
> 
> > +	alternative_input("",
> > +			Â"call __ibp_barrier",
> > +			ÂX86_FEATURE_IBPB,
> > +			ÂASM_NO_INPUT_CLOBBER("eax", "ecx", "edx", "memory"));
> > Â}
> 
> Dammit. I know the best time to comment is *before* I add my own sign-
> off to it and before Linus has merged it but... I think this is broken.
> 
> If you're calling a C function then you have to mark *all* the call-
> clobbered registers as, well, clobbered.
> 
> If you really really really want to *call* something out of line, then
> it would need to be implemented in asm.

Hm.  In theory I agree this seems like a bug.  On x86_64 I believe we
would need to mark the following registers as clobbered: r8-r11, ax, cx,
dx, si, di, plus "memory" and "cc".

But I'm scratching my head a bit, because we seem to have this bug all
over the kernel.  (Grep for ASM_CALL_CONSTRAINT to see them.)

Many of those inline asm calls have been around a long time.  So why
hasn't it ever bitten us?

-- 
Josh