Re: [RFC 8/8] compaction: make compaction use in-order putback

[Minchan Kim]

On Wed, Apr 27, 2011 at 5:39 PM, KAMEZAWA Hiroyuki
<kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> On Wed, 27 Apr 2011 01:25:25 +0900
> Minchan Kim <minchan.kim@xxxxxxxxx> wrote:
>
>> Compaction is good solution to get contiguos page but it makes
>> LRU churing which is not good.
>> This patch makes that compaction code use in-order putback so
>> after compaction completion, migrated pages are keeping LRU ordering.
>>
>> Cc: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>
>> Cc: Mel Gorman <mgorman@xxxxxxx>
>> Cc: Rik van Riel <riel@xxxxxxxxxx>
>> Cc: Andrea Arcangeli <aarcange@xxxxxxxxxx>
>> Signed-off-by: Minchan Kim <minchan.kim@xxxxxxxxx>
>> ---
>> Âmm/compaction.c | Â 22 +++++++++++++++-------
>> Â1 files changed, 15 insertions(+), 7 deletions(-)
>>
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index a2f6e96..480d2ac 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -211,11 +211,11 @@ static void isolate_freepages(struct zone *zone,
>> Â/* Update the number of anon and file isolated pages in the zone */
>> Âstatic void acct_isolated(struct zone *zone, struct compact_control *cc)
>> Â{
>> - Â Â struct page *page;
>> + Â Â struct pages_lru *pages_lru;
>> Â Â Â unsigned int count[NR_LRU_LISTS] = { 0, };
>>
>> - Â Â list_for_each_entry(page, &cc->migratepages, lru) {
>> - Â Â Â Â Â Â int lru = page_lru_base_type(page);
>> + Â Â list_for_each_entry(pages_lru, &cc->migratepages, lru) {
>> + Â Â Â Â Â Â int lru = page_lru_base_type(pages_lru->page);
>> Â Â Â Â Â Â Â count[lru]++;
>> Â Â Â }
>>
>> @@ -281,6 +281,7 @@ static unsigned long isolate_migratepages(struct zone *zone,
>> Â Â Â spin_lock_irq(&zone->lru_lock);
>> Â Â Â for (; low_pfn < end_pfn; low_pfn++) {
>> Â Â Â Â Â Â Â struct page *page;
>> + Â Â Â Â Â Â struct pages_lru *pages_lru;
>> Â Â Â Â Â Â Â bool locked = true;
>>
>> Â Â Â Â Â Â Â /* give a chance to irqs before checking need_resched() */
>> @@ -334,10 +335,16 @@ static unsigned long isolate_migratepages(struct zone *zone,
>> Â Â Â Â Â Â Â Â Â Â Â continue;
>> Â Â Â Â Â Â Â }
>>
>> + Â Â Â Â Â Â pages_lru = kmalloc(sizeof(struct pages_lru), GFP_ATOMIC);
>> + Â Â Â Â Â Â if (pages_lru)
>> + Â Â Â Â Â Â Â Â Â Â continue;
>
> Hmm, can't we use fixed size of statically allocated pages_lru, per-node or
> per-zone ? I think using kmalloc() in memory reclaim path is risky.

Yes. we can enhance it with pagevec-like approach.
It's my TODO list.  :)

In compaction POV, it is used by reclaiming big order pages so most of
time order-0 pages are enough. It's basic assumption of compaction so
it shouldn't be a problem.

Thanks for the review, Kame.


-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/