Re: [syzbot] [net?] general protection fault in dev_prep_valid_name
From: Aleksandr Nogikh
Date: Mon Nov 25 2024 - 06:35:11 EST
On Tue, Nov 19, 2024 at 3:53 AM 'Jakub Kicinski' via syzkaller-bugs
<syzkaller-bugs@xxxxxxxxxxxxxxxx> wrote:
>
> On Mon, 18 Nov 2024 18:19:23 -0800 Caleb Sander wrote:
> > Hmm, it seems very unlikely that "mlx5/core: Schedule EQ comp tasklet
> > only if necessary" could have caused this issue. The commit only
> > touches the mlx5 driver. Does the test machine have ConnectX NICs? The
> > commit itself simply moves where tasklet_schedule() is called for the
> > mlx5_cq_tasklet_cb() tasklet, making it so the tasklet will only be
> > scheduled to process userspace RDMA completions.
> > Is it possible that the failure is not consistently reproducible and
> > the bisection is landing on the wrong commit?
>
> Yes, most likely bad bisection, I looked at the syzbot docs but I don't
> see the command for invalidating the bisection results.
>
Thanks for analyzing the bisection!
It's indeed not possible to cancel the result via an email command.
I've marked it as invalid using other means and left a comment on the
corresponding backlog issue:
https://github.com/google/syzkaller/issues/5414#issuecomment-2497667350
--
Aleksandr