On Wed, Mar 04, 2020 at 11:11:04AM -0800, Yonghong Song wrote:
>
> - if (in_nmi()) {
> + /* Delay sending signal if irq is disabled. Otherwise,
> + * we risk deadlock with rq_lock.
> + */
This comment read in isolation is confusing. It's not clear how irq
has anything to do with rq_lock and why deadlock is possible.
But commit log has very nice summary,
so I've just deleted that comment and applied to bpf tree.
The patch 2 did reproduce the lockdep splat for me. Which was great.
Running test_progs few times before and after I noticed that older
send_signal tests are flaky and fail from time to time.
It's unrelated to this patch, but please take a look.
