Re: [PATCH v3 2/8] mm, compaction: remove redundant watermark check in compact_finished()

[Vlastimil Babka]

On 03/16/2017 02:30 AM, Joonsoo Kim wrote:
> Hello,

Hi, sorry for the late replies.

> On Tue, Mar 07, 2017 at 02:15:39PM +0100, Vlastimil Babka wrote:
>> When detecting whether compaction has succeeded in forming a high-order page,
>> __compact_finished() employs a watermark check, followed by an own search for
>> a suitable page in the freelists. This is not ideal for two reasons:
>> 
>> - The watermark check also searches high-order freelists, but has a less strict
>>   criteria wrt fallback. It's therefore redundant and waste of cycles. This was
>>   different in the past when high-order watermark check attempted to apply
>>   reserves to high-order pages.
> 
> Although it looks redundant now, I don't like removal of the watermark
> check here. Criteria in watermark check would be changed to more strict
> later and we would easily miss to apply it on compaction side if the
> watermark check is removed.

I see, but compaction is already full of various watermark(-like) checks that
have to be considered/updated if watermark checking changes significantly, or
things will go subtly wrong. I doubt this extra check can really help much in
such cases.

>> 
>> - The watermark check might actually fail due to lack of order-0 pages.
>>   Compaction can't help with that, so there's no point in continuing because of
>>   that. It's possible that high-order page still exists and it terminates.
> 
> If lack of order-0 pages is the reason for stopping compaction, we
> need to insert the watermark check for order-0 to break the compaction
> instead of removing it. Am I missing something?

You proposed that once IIRC, but didn't follow up? Currently we learn about
insufficient order-0 watermark in __isolate_free_page() from the free scanner.
We could potentially stop compacting earlier by checking it also in
compact_finished(), but maybe it doesn't happen that often and it's just extra
checking overhead.

So I wouldn't be terribly opposed by converting the current check to an order-0
fail-compaction check (instead of removing it), but I really wouldn't like to
insert the order-0 one and also keep the current one.

> Thanks.
>