Re: [PATCH 1/4] sched/wakeup: Strengthen current_save_and_set_rtlock_wait_state()

[Will Deacon]

On Fri, Sep 10, 2021 at 04:01:29PM +0200, Peter Zijlstra wrote:
> On Fri, Sep 10, 2021 at 03:17:04PM +0200, Peter Zijlstra wrote:
> > On Fri, Sep 10, 2021 at 01:57:26PM +0100, Will Deacon wrote:
> > > On Thu, Sep 09, 2021 at 04:27:46PM +0200, Peter Zijlstra wrote:
> > > > Moo yes, so the earlier changelog I wrote was something like:
> > > > 
> > > > 	current_save_and_set_rtlock_wait_state();
> > > > 	for (;;) {
> > > > 		if (try_lock())
> > > > 			break;
> > > > 
> > > > 		raw_spin_unlock_irq(&lock->wait_lock);
> > > > 		if (!cond)
> > > > 			schedule();
> > > > 		raw_spin_lock_irq(&lock->wait_lock);
> > > > 
> > > > 		set_current_state(TASK_RTLOCK_WAIT);
> > > > 	}
> > > > 	current_restore_rtlock_saved_state();
> > > > 
> > > > which is more what the code looks like before these patches, and in that
> > > > case the @cond load can be lifted before __state.
> > > 
> > > Ah, so that makes more sense, thanks. I can't see how the try_lock() could
> > > be reordered though, as it's going to have to do an atomic rmw.
> > 
> > OK, lemme go update the Changelog and make it __flags for bigeasy :-)
> 
> The patch now reads:
> 
> ---
> Subject: sched/wakeup: Strengthen current_save_and_set_rtlock_wait_state()
> From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Date: Thu, 09 Sep 2021 12:59:16 +0200
> 
> While looking at current_save_and_set_rtlock_wait_state() I'm thinking
> it really ought to use smp_store_mb(), because using it for a more
> traditional wait loop like:
> 
> 	current_save_and_set_rtlock_wait_state();
> 	for (;;) {
> 		if (cond)
> 			schedule();
> 
> 		set_current_state(TASK_RTLOCK_WAIT);
> 	}
> 	current_restore_rtlock_saved_state();
> 
> is actually broken, since the cond load could be re-ordered against
> the state store, which could lead to a missed wakeup -> BAD (tm).
> 
> While there, make them consistent with the IRQ usage in
> set_special_state().

Cheers, that's much better:

Acked-by: Will Deacon <will@xxxxxxxxxx>

Will