Re: [PATCH] word-at-a-time: use the same return type for has_zero regardless of endianness

[Nathan Chancellor]

On Tue, Aug 01, 2023 at 06:07:08PM -0700, Linus Torvalds wrote:
> I think the patch is fine, but I guess I'd like to know that people
> who are affected actually don't see any code generation changes (or
> possibly see improvements from not turning it into a bool until later)

We see this warning with ARCH=arm64 defconfig + CONFIG_CPU_BIG_ENDIAN=y.
With both clang 18.0.0 (tip of tree) and GCC 13.1.0, I don't see any
actual code generation changes in fs/namei.o with this configuration.
I'd be pretty surprised if any of the other uses of has_zero() show any
changes, I at least checked lib/string.o with that configuration and
s390 and did not see anything.

As far as I can tell, arm and arm64 with CONFIG_CPU_BIG_ENDIAN=y are the
only configurations that can hit the particular bit of code with the
generic big endian has_zero() implementation because the version of
hash_name() that uses has_zero() in this manner is only used when
CONFIG_DCACHE_WORD_ACCESS is set, which only arm, arm64, powerpc (little
endian), and x86 select.

arch/arm/Kconfig:49:         select DCACHE_WORD_ACCESS if HAVE_EFFICIENT_UNALIGNED_ACCESS
arch/arm64/Kconfig:121:        select DCACHE_WORD_ACCESS
arch/powerpc/Kconfig:183:        select DCACHE_WORD_ACCESS                if PPC64 && CPU_LITTLE_ENDIAN
arch/x86/Kconfig:140:        select DCACHE_WORD_ACCESS                if !KMSAN
arch/x86/um/Kconfig:12:         select DCACHE_WORD_ACCESS

So seems like a pretty low risk patch to me but I could be missing
something.

Cheers,
Nathan