Re: [PATCH][RT] xfs: Disable preemption when grabbing all icsb counter locks

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 30 Apr 2015 20:07:21 +0200
Peter Zijlstra <peterz@infradead.org> wrote:

> On Thu, Apr 30, 2015 at 12:33:03PM -0400, Steven Rostedt wrote:
> > diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
> > index 51435dbce9c4..dbaa1ce3f308 100644
> > --- a/fs/xfs/xfs_mount.c
> > +++ b/fs/xfs/xfs_mount.c
> > @@ -1660,6 +1660,12 @@ xfs_icsb_lock_all_counters(
> >  	xfs_icsb_cnts_t *cntp;
> >  	int		i;
> >  
> > +	/*
> > +	 * In PREEMPT_RT, preemption is not disabled here, and it
> > +	 * must be to take the xfs_icsb_lock_cntr.
> > +	 */
> > +	preempt_disable_rt();
> > +
> >  	for_each_online_cpu(i) {
> >  		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
> >  		xfs_icsb_lock_cntr(cntp);
> > @@ -1677,6 +1683,8 @@ xfs_icsb_unlock_all_counters(
> >  		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
> >  		xfs_icsb_unlock_cntr(cntp);
> >  	}
> > +
> > +	preempt_enable_rt();
> >  }
> 
> The irony, this is distinctly non deterministic code you're putting
> under a RT specific preempt_disable ;-)

I know :-(

Unfortunately, a RT behaving fix would be much more invasive and would
probably require the help of the xfs folks. For now, this just prevents
a live lock that can happen and halt the system, where it becomes
deterministic catastrophe.

-- Steve