Re: [PATCH 12/13] mm: Throttle direct reclaimers if PF_MEMALLOC reserves are low and swap is backed by network storage

[Mel Gorman <mgorman@suse.de>]

On Thu, Apr 28, 2011 at 10:22:44AM +1000, NeilBrown wrote:
> On Wed, 27 Apr 2011 17:08:10 +0100 Mel Gorman <mgorman@suse.de> wrote:
> 
> 
> > +/*
> > + * Throttle direct reclaimers if backing storage is backed by the network
> > + * and the PFMEMALLOC reserve for the preferred node is getting dangerously
> > + * depleted. kswapd will continue to make progress and wake the processes
> > + * when the low watermark is reached
> > + */
> > +static void throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist,
> > +					nodemask_t *nodemask)
> > +{
> > +	struct zone *zone;
> > +	int high_zoneidx = gfp_zone(gfp_mask);
> > +	DEFINE_WAIT(wait);
> > +
> > +	/* Check if the pfmemalloc reserves are ok */
> > +	first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone);
> > +	if (pfmemalloc_watermark_ok(zone->zone_pgdat, high_zoneidx))
> > +		return;
> 
> As the first thing that 'wait_event_interruptible" does is test the condition
> and return if it is true, this "if () return;" is pointless.
>  

In patch 13, we count the number of times we got throttled. In this
patch, the check is pointless but it makes sense in the context of
the following patch.

> > +
> > +	/* Throttle */
> > +	wait_event_interruptible(zone->zone_pgdat->pfmemalloc_wait,
> > +		pfmemalloc_watermark_ok(zone->zone_pgdat, high_zoneidx));
> > +}
> 
> I was surprised that you chose wait_event_interruptible as your previous code
> was almost exactly "wait_event_killable".
> 
> Is there some justification for not throttling processes which happen to have
> a (non-fatal) signal pending?
> 

No justification, wait_event_killable() is indeed a better fit.

> > +
> >  unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
> >  				gfp_t gfp_mask, nodemask_t *nodemask)
> >  {
> > @@ -2133,6 +2172,15 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
> >  		.nodemask = nodemask,
> >  	};
> >  
> > +	throttle_direct_reclaim(gfp_mask, zonelist, nodemask);
> > +
> > +	/*
> > +	 * Do not enter reclaim if fatal signal is pending. 1 is returned so
> > +	 * that the page allocator does not consider triggering OOM
> > +	 */
> > +	if (fatal_signal_pending(current))
> > +		return 1;
> > +
> >  	trace_mm_vmscan_direct_reclaim_begin(order,
> >  				sc.may_writepage,
> >  				gfp_mask);
> > @@ -2488,6 +2536,12 @@ loop_again:
> >  			}
> >  
> >  		}
> > +
> > +		/* Wake throttled direct reclaimers if low watermark is met */
> > +		if (waitqueue_active(&pgdat->pfmemalloc_wait) &&
> > +				pfmemalloc_watermark_ok(pgdat, MAX_NR_ZONES - 1))
> > +			wake_up_interruptible(&pgdat->pfmemalloc_wait);
> > +
> >  		if (all_zones_ok || (order && pgdat_balanced(pgdat, balanced, *classzone_idx)))
> >  			break;		/* kswapd: all done */
> >  		/*
> 

-- 
Mel Gorman
SUSE Labs