Re: [PATCH v2 1/2] x86/uaccess: Extend CMPXCHG user helpers to 128-bit operands

[Sean Christopherson <seanjc@google.com>]

On Mon, Jul 06, 2026, Sairaj Kodilkar wrote:
> On 7/6/2026 12:22 PM, sashiko-bot@kernel.org wrote:
> > > diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
> > > index 367297b188c39..123755d471099 100644
> > > --- a/arch/x86/include/asm/uaccess.h
> > > +++ b/arch/x86/include/asm/uaccess.h
> > [ ... ]
> > 
> > > @@ -551,11 +594,18 @@ do {                                                                            \
> > > 
> > >   extern void __try_cmpxchg_user_wrong_size(void);
> > > 
> > > -#ifndef CONFIG_X86_32
> > > +#ifdef CONFIG_X86_32
> > > +/* Always fail on 32 bit arch as it do not support 128 cmpxchg (i.e. cmpxchg16b
> > > + * instruction).
> > > + */
> > > +#define __try_cmpxchg128_user_asm(_ptr, _pold, _new, label) (1)
> > [Severity: Medium]
> > Does this 32-bit stub correctly signal failure?
> > 
> > Since this macro evaluates to 1, won't the surrounding unsafe_try_cmpxchg_user
> > macro treat this as success (true) rather than failure (false)?
> > 
> > This could cause the wrapper to negate the value and return 0 (success) to the
> > caller, falsely signaling a successful memory exchange without actually
> > updating memory or the old value. Even though there are no current 32-bit
> > callers using a 16-byte operand, should this evaluate to 0 instead of 1 to
> > ensure theoretical future callers don't encounter a silent atomicity break?
> > 
> > --
> > Sashiko AI review · https://sashiko.dev/#/patchset/20260706063035.1139-1-sarunkod@amd.com?part=1
> Hi,
> 
> Right, It should return 0 instead of 1, will rectify this in the next patch.

Is it possible to fail the build instead?  Attempting to generate CMPXCHG16B on
32-bit should straigt up fail.