Re: [RFC] mm/vmscan.c: avoid possible long latency caused by too_many_isolated()

[Huang, Ying]

Yu Zhao <yuzhao@xxxxxxxxxx> writes:
[snip]

> @@ -2966,13 +2938,20 @@ static void shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
>  			/* need some check for avoid more shrink_zone() */
>  		}
>  
> -		/* See comment about same check for global reclaim above */
> -		if (zone->zone_pgdat == last_pgdat)
> -			continue;
> -		last_pgdat = zone->zone_pgdat;
>  		shrink_node(zone->zone_pgdat, sc);
>  	}
>  
> +	if (last_pgdat)
> +		atomic_dec(&last_pgdat->nr_reclaimers);
> +	else if (should_retry) {
> +		/* wait a bit for the reclaimer. */
> +		if (!schedule_timeout_killable(HZ / 10))

Once we reached here, even accidentally, the caller needs to sleep at
least 100ms.  How about use a semaphore for pgdat->nr_reclaimers?  Then
the sleeper can be waken up when the resource is considered enough.

Best Regards,
Huang, Ying

> +			goto retry;
> +
> +		/* We are about to die and free our memory. Return now. */
> +		sc->nr_reclaimed += SWAP_CLUSTER_MAX;
> +	}
> +
>  	/*
>  	 * Restore to original mask to avoid the impact on the caller if we
>  	 * promoted it to __GFP_HIGHMEM.

[snip]