https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105727
--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Jan Hubicka from comment #6)
> I don't know what clang does, but GCC keeps builtin_constant_p till late
> optimization and resolves it then. So here we cross module inline (or
> constant propagate) and then it becomes constant.
>
> Outcome of __builtin_constant_p should depend on optimization settings and
> LTO is one of them. So I would say that both clang and gcc are correct, just
> gcc has better QOI. Why this breaks kernel?
Using KASAN, we end up with:
In function ‘memory_is_poisoned’,
inlined from ‘check_region_inline’ at mm/kasan/generic.c:180:6,
inlined from ‘kasan_check_range’ at mm/kasan/generic.c:189:9,
inlined from ‘memset’ at mm/kasan/shadow.c:44:7,
inlined from ‘do_sysinfo.isra’ at kernel/sys.c:2656:2:
mm/kasan/generic.c:155:25: error: call to ‘__compiletime_assert_243’ declared
with attribute error: BUILD_BUG failed
155 | BUILD_BUG();
static __always_inline bool memory_is_poisoned(unsigned long addr, size_t size)
{
if (__builtin_constant_p(size)) {
switch (size) {
case 1:
return memory_is_poisoned_1(addr);
case 2:
case 4:
case 8:
return memory_is_poisoned_2_4_8(addr, size);
case 16:
return memory_is_poisoned_16(addr);
default:
BUILD_BUG();
}
}
return memory_is_poisoned_n(addr, size);
}
where we propagate large some value during LTRANS and that's why we end up with
BUILD_BUG().
