Re: [PATCH 06/27] mm, vmscan: Make kswapd reclaim in terms of nodes

[Vlastimil Babka]

On 04/15/2016 11:13 AM, Mel Gorman wrote:
>   	/*
> -	 * If a zone reaches its high watermark, consider it to be no longer
> -	 * congested. It's possible there are dirty pages backed by congested
> -	 * BDIs but as pressure is relieved, speculatively avoid congestion
> -	 * waits.
> +	 * Fragmentation may mean that the system cannot be rebalanced for
> +	 * high-order allocations. If twice the allocation size has been
> +	 * reclaimed then recheck watermarks only at order-0 to prevent
> +	 * excessive reclaim. Assume that a process requested a high-order
> +	 * can direct reclaim/compact.

Also kcompactd is woken up in this case...

>   	 */
> -	if (pgdat_reclaimable(zone->zone_pgdat) &&
> -	    zone_balanced(zone, sc->order, false, 0, classzone_idx)) {
> -		clear_bit(PGDAT_CONGESTED, &pgdat->flags);
> -		clear_bit(PGDAT_DIRTY, &pgdat->flags);
> -	}
> +	if (sc->order && sc->nr_reclaimed >= 2UL << sc->order)
> +		sc->order = 0;
>
>   	return sc->nr_scanned >= sc->nr_to_reclaim;

This looks indeed simpler than my earlier zone_balanced() modification 
you removed. However I think there's still potential of overreclaim due 
to a stream of kswapd_wakeups where each will have to reclaim 2UL << 
sc->order pages, regardless of watermarks. Could be some high-order 
wakeups from GFP_ATOMIC context that have order-0 fallbacks but will 
cause kswapd to keep reclaiming when kcompactd can't keep up due to 
fragmentation...