Re: [PATCH v4] pipe: use __pipe_{lock,unlock} instead of spinlock

[Linus Torvalds]

On Mon, Feb 6, 2023 at 7:58 AM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
>
> As for Linus' point about us needing to avoid sleep under RCU +
> spinlock, curious if we can capture *existing* bad users of that with
> Coccinelle SmPL.

Well, we have it dynamically with the "might_sleep()" debugging. But
that obviously only triggers if that is enabled and when those
particular paths are run.

It would be lovely to have a static checker for "sleep under spinlock
or in RCU" (or any of the other situations: preemption disabled either
explicitly or due to get_cpu() and similar).

But I suspect it would be quite hard to do.

               Linus