Re: mmotm 2019-04-19-14-53 uploaded (objtool)

[Linus Torvalds]

On Tue, Apr 23, 2019 at 1:25 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> Now, we could of course allow this symbol, but I found only the below
> was required to make allyesconfig build without issue.
>
> Andy, Linus?

Ack on that patch. Except I think the uaccess.h part should be a
separate commit: I think it goes along with 2a418cf3f5f1
("x86/uaccess: Don't leak the AC flag into __put_user() value
evaluation") we did earlier. I think the logic is the same - it's not
just the _value_ that can have complex calculations, the address can
too (although admittedly that's really not supposed to be common, but
you clearly found one case where a complier misfeature made it happen,
so...).

I also wonder if we should just make "count" be "unsigned long" in
do_{strncpy_from,strnlen}_user() too, since we've already done

        if (unlikely(count <= 0))
                return 0;

in the caller, so it *is* unsigned by then, and we'd not be mixing
signedness when comparing "max/count/res".

                   Linus