On Fri, 25 Jul 2025 17:26:54 +0100, Nathan Chancellor nathan@kernel.org wrote:
On Fri, Jul 25, 2025 at 08:30:21AM +0100, Marc Zyngier wrote:
The correct fix would be to backport the series described in e8789ab7047a8, which should be easy enough to apply. it would also make 6.1 less of a terrible kernel.
If doing that is reasonable to clear this up, I think that would be fine to do. This is the only stable-only instance of that warning that I have seen in the build logs, I have sent patches to deal with all the other instances upstream. We would need this in 5.15 to avoid failures from -Werror as well but if it is too hard to backport that series there, we could just disable this warning for this file since we know it is a false positive.
5.15 would be rather challenging, I'm afraid, and I wouldn't want to review such a thing.
The whole reason the warning occurs is due to the constness of the sys_reg_desc parameter in the function created by FUNCTION_INVARIANT(), which I am guessing cannot be removed because it is present in ->access() and it proliferates out from there?
Exactly. Which was a rather bad move when it was introduced over a decade ago (in v3.11), and we only got 'round to killing it entirely in v6.15.
Thanks,
M.