On Thu, Jan 11, 2024, at 12:16, Dan Carpenter wrote:
> On Thu, Jan 11, 2024 at 04:23:09PM +0530, Naresh Kamboju wrote:
>> Following build failures noticed on i386 and x86 with clang builds on the
>> Linux next-20240111 tag.
>>
>> Reported-by: Linux Kernel Functional Testing <lkft@xxxxxxxxxx>
>>
>> Build error:
>> ----------
>> mm/vmalloc.c:4691:25: error: variable 'addr' is uninitialized when
>> used here [-Werror,-Wuninitialized]
>> 4691 | va = __find_vmap_area(addr, &vn->busy.root);
>> | ^~~~
>> mm/vmalloc.c:4684:20: note: initialize the variable 'addr' to silence
>> this warning
>> 4684 | unsigned long addr;
>> | ^
>> | = 0
>> 1 error generated.
>
> We turned off uninitialized variable warnings for GCC a long time ago...
> :/ I don't know if we'll be able to re-enable it in a -Werror world
> although Clang seems to be managing alright so perhaps there is hope.
The problem with gcc's warning is that it is non-deterministic and
in recent versions actually got more false-positives even without
-Os or -fsanitize=. Clang does not catch all that gcc does because
it doesn't track state across inline functions, but at least its
output is always the same regardless of optimization and other
options.
At least this particular one is an obvious bug and easily gets
caught by lkft and lkp even if gcc's -Wuninitilized doesn't
flag it.
Arnd
