Hi David,
On Tue, 24 Feb 2026 at 23:06, David Laight <[email protected]> wrote:
>
> On Tue, 24 Feb 2026 17:54:07 +0000
> Fuad Tabba <[email protected]> wrote:
>
> > Hi Andy,
> >
> > On Tue, 24 Feb 2026 at 17:21, Andy Shevchenko
> > <[email protected]> wrote:
> > >
> > > On Tue, Feb 24, 2026 at 05:04:27PM +0000, Fuad Tabba wrote:
> > > > sized_strscpy() performs word-at-a-time writes to the destination
> > > > buffer. If the destination buffer is not aligned to unsigned long,
> > > > direct assignment causes UBSAN misaligned-access errors.
> > > >
> > > > Use put_unaligned() to safely write the words to the destination.
> > >
> > > Have you measured the performance impact?
> >
> > Not directly. I verified the disassembly for both x86_64 and aarch64.
> > On x86_64, both the raw pointer cast and put_unaligned() compile down
> > to mov %rdi,(%rsi). On aarch64, both compile to str x0, [x1].
>
> What happens on cpu that trap misaligned accesses (eg sparc64)?
> put_unaligned() exists because it can be horrid.
To be honest, I hadn't considered this until now. But looking at it, I
believe that the existing guards in sized_strscpy() already protect
architectures like sparc from the unaligned paths you are concerned
about. Looking at the code and configs, these do not select
CONFIG_DCACHE_WORD_ACCESS nor CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS.
Because of this, they fall into the #else block early in
sized_strscpy():
#ifndef CONFIG_DCACHE_WORD_ACCESS
#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
...
#else
/* If src or dest is unaligned, don't do word-at-a-time. */
if (((long) dest | (long) src) & (sizeof(long) - 1))
max = 0;
#endif
#endif
If either dest or src is unaligned, max is set to 0. This bypasses the
loop with put_unaligned(). I checked by compiling this for sparc: if
aligned, the compiler sees that, and optimizes it into an 8-byte store
(stx %i0, [%i1]), identical to the raw pointer cast.
So this patch shouldn't introduce memcpy fallback penalties on sparc,
but it still fixes the UB on architectures like x86 and arm64.
Cheers,
/fuad
> David
>
> >
> > > Have you read the comment near to
> > >
> > > if (IS_ENABLED(CONFIG_KMSAN))
> >
> > Not until now to be honest. However, are you asking whether
> > put_unaligned() breaks KMSAN? I don't think it does, max is set to 0
> > when KMSAN is enabled, this entire while loop is bypassed.
> >
> > Thanks,
> > /fuad
> >
> > > ?
> > >
> > > --
> > > With Best Regards,
> > > Andy Shevchenko
> > >
> > >
> >
>
