Re: [PATCH 1/1] kbuild: Only enable -Wtautological-constant-out-of-range-compare for W=2

["Arnd Bergmann" <arnd@arndb.de>]

On Fri, Dec 19, 2025, at 23:18, David Laight wrote:
> On Fri, 19 Dec 2025 13:12:31 -0700 Nathan Chancellor <nathan@kernel.org> wrote:
>
> Somewhere I got confused and must have looked at the wrong email (or just
> failed to separate two very long warning names).
> The actual warning was:
>
>>> drivers/gpu/drm/xe/xe_guc.c:639:19: error: converting the result of '<<' to a boolean always evaluates to true [-Werror,-Wtautological-constant-compare]  
>      639 |                 klvs[count++] = 
> PREP_GUC_KLV_TAG(OPT_IN_FEATURE_EXT_CAT_ERR_TYPE);

This does seem like a completely sensible warning to me, and it's
always been enabled by default. I see three patches in the git history
(all from Nathan), which all make sense as well.

> Inside FIELD_PREP_CONST(mask, val) there is (with the patch, and if I've
> typed it correctly):
> 	BUILD_BUG_ON_ZERO(!(mask) || (mask) & ((mask) + ((mask) & -(mask)))))
> to check the mask is non-zero and contiguous bits.

I think the problem is (as so often) the linux/bitfield.h headers
making things way too complicated. That condition makes no sense to
me, and neither would I expect a compiler to make sense of it either.

If there is no way to express those conditions more clearly, I would
prefer removing the BUILD_BUG_ON stuff from the bitfield.h header,
it keeps causing way more false positives than finding actual bugs
with the input.

     Arnd