Re: [RFC PATCH v3 1/4] mm/page_alloc: fix incorrect isolation behavior by rechecking migratetype

[Vlastimil Babka <vbabka@suse.cz>]

On 09/15/2014 04:31 AM, Joonsoo Kim wrote:
> On Mon, Sep 08, 2014 at 10:31:29AM +0200, Vlastimil Babka wrote:
>> On 08/26/2014 10:08 AM, Joonsoo Kim wrote:
>>
>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>> index f86023b..51e0d13 100644
>>> --- a/mm/page_alloc.c
>>> +++ b/mm/page_alloc.c
>>> @@ -740,9 +740,15 @@ static void free_one_page(struct zone *zone,
>>>   	if (nr_scanned)
>>>   		__mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
>>>
>>> +	if (unlikely(has_isolate_pageblock(zone))) {
>>> +		migratetype = get_pfnblock_migratetype(page, pfn);
>>> +		if (is_migrate_isolate(migratetype))
>>> +			goto skip_counting;
>>> +	}
>>> +	__mod_zone_freepage_state(zone, 1 << order, migratetype);
>>> +
>>> +skip_counting:
>>
>> Here, wouldn't a simple 'else __mod_zone_freepage_state...' look
>> better than goto + label? (same for the following 2 patches). Or
>> does that generate worse code?
>
> To remove goto label, we need two __mod_zone_freepage_state() like
> as below. On my system, it doesn't generate worse code, but, I am not
> sure that this is true if more logic would be added. I think that
> goto + label is better.

Oh right, I missed that. It's a bit subtle, but I don't see a nicer 
solution right now.

> +	if (unlikely(has_isolate_pageblock(zone))) {
> +		migratetype = get_pfnblock_migratetype(page, pfn);
> +               if (!is_migrate_isolate(migratetype))
> +                       __mod_zone_freepage_state(zone, 1 << order, migratetype);
> +       } else {
> +               __mod_zone_freepage_state(zone, 1 << order, migratetype);
>          }
>

Yeah that would be uglier I guess.

> Anyway, What do you think which one is better, either v2 or v3? Still, v3? :)

Yeah v3 is much better than v1 was, and better for backporting than v2. 
The changelogs also look quite clear. The overhead shouldn't be bad with 
the per-zone flag guarding get_pfnblock_migratetype.

I'm just not sure about patch 4 and potentially leaving unmerged budies 
behind. How would it look if instead we made sure isolation works on 
whole MAX_ORDER blocks instead?

Vlastimil

> Thanks.
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>