On Mon, 22 Apr 2024, Peter Zijlstra wrote:
> On Wed, Apr 17, 2024 at 09:57:04PM -0700, Christoph Hellwig wrote:
> > On Wed, Apr 17, 2024 at 08:00:22PM +0200, Mikulas Patocka wrote:
> > > > > +EXPORT_SYMBOL(wait_for_completion_long_io);
> > > >
> > > > Urgh, why is it a sane thing to circumvent the hang check timer?
> > >
> > > The block layer already does it - the bios can have arbitrary size, so
> > > waiting for them takes arbitrary time.
> >
> > And as mentioned the last few times around, I think we want a task
> > state to say that task can sleep long or even forever and not propagate
> > this hack even further.
>
> A bit like TASK_NOLOAD (which is used to make TASK_IDLE work), but
> different I suppose.
>
> TASK_NOHUNG would be trivial to add ofc. But is it worth it?
>
> Anyway, as per the other email, anything like this needs to come with a
> big fat warning. You get to keep the pieces etc..
This seems better than the blk_wait_io hack.
Reviewed-by: Mikulas Patocka <[email protected]>
> ---
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 3c2abbc587b4..83b25327c233 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -112,7 +112,8 @@ struct user_event_mm;
> #define TASK_FREEZABLE 0x00002000
> #define __TASK_FREEZABLE_UNSAFE (0x00004000 *
> IS_ENABLED(CONFIG_LOCKDEP))
> #define TASK_FROZEN 0x00008000
> -#define TASK_STATE_MAX 0x00010000
> +#define TASK_NOHUNG 0x00010000
> +#define TASK_STATE_MAX 0x00020000
>
> #define TASK_ANY (TASK_STATE_MAX-1)
>
> diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> index b2fc2727d654..126fac835e5e 100644
> --- a/kernel/hung_task.c
> +++ b/kernel/hung_task.c
> @@ -210,7 +210,8 @@ static void check_hung_uninterruptible_tasks(unsigned
> long timeout)
> state = READ_ONCE(t->__state);
> if ((state & TASK_UNINTERRUPTIBLE) &&
> !(state & TASK_WAKEKILL) &&
> - !(state & TASK_NOLOAD))
> + !(state & TASK_NOLOAD) &&
> + !(state & TASK_NOHUNG))
> check_hung_task(t, timeout);
> }
> unlock:
>
