Re: [PATCH] iomap: Address soft lockup in iomap_finish_ioend()

[Dave Chinner <david@fromorbit.com>]

On Wed, Jan 05, 2022 at 04:01:07PM -0800, Darrick J. Wong wrote:
> On Thu, Jan 06, 2022 at 09:48:29AM +1100, Dave Chinner wrote:
> > +
> > +	might_sleep();
> >  
> >  	list_replace_init(&ioend->io_list, &tmp);
> > +	segments = ioend->io_segments;
> >  	iomap_finish_ioend(ioend, error);
> >  
> >  	while (!list_empty(&tmp)) {
> > +		if (segments > 32768) {
> > +			cond_resched();
> > +			segments = 0;
> > +		}
> >  		ioend = list_first_entry(&tmp, struct iomap_ioend, io_list);
> >  		list_del_init(&ioend->io_list);
> > +		segments += ioend->io_segments;
> >  		iomap_finish_ioend(ioend, error);
> >  	}
> 
> I wonder, should we take one more swing at cond_resched at the end of
> the function so that we can return to the caller having given the system
> at least one chance to reschedule?

That's for the caller of xfs_finish_ioends() to deal with as it
loops over each set of merged ioends and does it's own processing.
i.e. that's what the cond_resched() I added to the XFS endio
processing code here provides:

> >  void
> >  xfs_end_io(
> >  	struct work_struct	*work)
> > @@ -157,6 +170,7 @@ xfs_end_io(
> >  		list_del_init(&ioend->io_list);
> >  		iomap_ioend_try_merge(ioend, &tmp);
> >  		xfs_end_ioend(ioend);
> > +		cond_resched();
> >  	}
> >  }

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com