Subject: Re: mmotm woes, mainly compaction
To: Hugh Dickins <hughd@google.com>, Michal Hocko <mhocko@kernel.org>
References: <alpine.LSU.2.11.1604120005350.1832@eggly.anvils>
 <20160412121020.GC10771@dhcp22.suse.cz>
 <alpine.LSU.2.11.1604141114290.1086@eggly.anvils> <571026CA.6000708@suse.cz>
Cc: Andrew Morton <akpm@linux-foundation.org>, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org
From: Vlastimil Babka <vbabka@suse.cz>
Message-ID: <57153205.4050406@suse.cz>
Date: Mon, 18 Apr 2016 15:14:13 -0400
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.7.0
MIME-Version: 1.0
In-Reply-To: <571026CA.6000708@suse.cz>
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 769
Lines: 18

On 04/14/2016 07:24 PM, Vlastimil Babka wrote:
>> > @@ -1459,8 +1459,8 @@ static enum compact_result compact_zone(
>> >   		zone->compact_cached_migrate_pfn[1] = cc->migrate_pfn;
>> >   	}
>> >
>> > -	if (cc->migrate_pfn == start_pfn)
>> > -		cc->whole_zone = true;
>> > +	cc->whole_zone = cc->migrate_pfn == start_pfn &&
>> > +			cc->free_pfn == pageblock_start_pfn(end_pfn - 1);
>> >
>> >   	cc->last_migrated_pfn = 0;
> This would be for Michal, but I agree.

So there's an alternative here that wouldn't have the danger of missing
cc->whole_zone multiple time due to races. When resetting either one
scanner to zone boundary, reset the other as well (and set
cc->whole_zone). I think the situations, where not doing that have any
(performance) advantage, are rare.