Hi Andrew,
On 07/04/2012 06:48 AM, Andrew Morton wrote:
> On Tue, 3 Jul 2012 11:10:24 +0100
> Mel Gorman <mel@xxxxxxxxx> wrote:
>
>>>>>>>> + if (cc->order> 0)
>>>>>>>> + zone->compact_cached_free_pfn = high_pfn;
>>>>>>>
>>>>>>> Is high_pfn guaranteed to be aligned to pageblock_nr_pages here? I
>>>>>>> assume so, if lots of code in other places is correct but it's
>>>>>>> unobvious from reading this function.
>>>>>>
>>>>>> Reading the code a few more times, I believe that it is
>>>>>> indeed aligned to pageblock size.
>>>>>
>>>>> I'll slip this into -next for a while.
>>>>>
>>>>> --- a/mm/compaction.c~isolate_freepages-check-that-high_pfn-is-aligned-as-expected
>>>>> +++ a/mm/compaction.c
>>>>> @@ -456,6 +456,7 @@ static void isolate_freepages(struct zon
>>>>> }
>>>>> spin_unlock_irqrestore(&zone->lock, flags);
>>>>>
>>>>> + WARN_ON_ONCE(high_pfn& (pageblock_nr_pages - 1));
>>>>> /*
>>>>> * Record the highest PFN we isolated pages from. When next
>>>>> * looking for free pages, the search will restart here as
>>>>
>>>> I've triggered the following with today's -next:
>>>
>>> I've been staring at the migrate code for most of the afternoon,
>>> and am not sure how this is triggered.
>>>
>>
>> That warning is placed in isolate_freepages(). When the migration
>> scanner and free scanner have almost met it is possible for high_pfn to
>> be
>>
>> cc->migrate_pfn + pageblock_nr_pages
>>
>> and that is not necessarily pageblock aligned. Forcing it to be aligned
>> raises the possibility that the free scanner moves to another zone. This
>> is very unlikely but could happen if a high zone was very small.
>>
>> I should have caught this when the warning was proposed :( IMO it's
>> safe to just drop the warning.
>
> The rest of this patch takes care to ensure that
> ->compact_cached_free_pfn is aligned to pageblock_nr_pages. But it now
> appears that this particular site will violate that.
>
> What's up? Do we need to fix this site, or do we remove all that
> make-compact_cached_free_pfn-aligned code?
I vote removing the warning because it doesn't related to Rik's incremental compaction.
Let's see.
high_pfn = min(low_pfn, pfn) = cc->migrate_pfn + pageblock_nr_pages.
In here, cc->migrate_pfn isn't necessarily pageblock aligined.
So if we don't consider compact_cached_free_pfn, it can hit.
static void isolate_freepages()
{
high_pfn = min(low_pfn, pfn) = cc->migrate_pfn + pageblock_nr_pages;
for (..) {
...
WARN_ON_ONCE(high_pfn & (pageblock_nr_pages - 1));
}
}
--
Kind regards,
Minchan Kim
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>