Re: [LTP] mmstress[1309]: segfault at 7f3d71a36ee8 ip 00007f3d77132bdf sp 00007f3d71a36ee8 error 4 in libc-2.27.so[7f3d77058000+1aa000]

[Linus Torvalds]

On Fri, Oct 23, 2020 at 8:54 AM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Fri, Oct 23, 2020 at 12:14 AM Rasmus Villemoes
> <linux@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > That's certainly garbage. Now, I don't know if it's a sufficient fix (or
> > could break something else), but the obvious first step of rearranging
> > so that the ptr argument is evaluated before the assignment to __val_pu
>
> Ack. We could do that.
>
> I'm more inclined to just bite the bullet and go back to the ugly
> conditional on the size that I had hoped to avoid, but if that turns
> out too ugly, mind signing off on your patch and I'll have that as a
> fallback?

Actually, looking at that code, and the fact that we've used the
"register asm()" format forever for the get_user() side, I think your
approach is the right one.

I'd rename the internal ptr variable to "__ptr_pu", and make sure the
assignments happen just before the asm call (with the __val_pu
assignment being the final thing).

lso, it needs to be

        void __user *__ptr_pu;

instead of

        __typeof__(ptr) __ptr = (ptr);

because "ptr" may actually be an array, and we need to have the usual
C "array to pointer" conversions happen, rather than try to make
__ptr_pu be an array too.

So the patch would become something like the appended instead, but I'd
still like your sign-off (and I'd put you as author of the fix).

Narest, can you confirm that this patch fixes the issue for you?

                  Linus

Attachment: patch
Description: Binary data