Re: [PATCH 1/2] mm,mlock: drain pagevecs asynchronously

[KOSAKI Motohiro <kosaki.motohiro@gmail.com>]

>> @@ -704,10 +747,23 @@ static void ____pagevec_lru_add_fn(struct page *page, void *arg)
>>   	VM_BUG_ON(PageLRU(page));
>>
>>   	SetPageLRU(page);
>> -	if (active)
>> -		SetPageActive(page);
>> -	update_page_reclaim_stat(zone, page, file, active);
>> -	add_page_to_lru_list(zone, page, lru);
>> + redo:
>> +	if (page_evictable(page, NULL)) {
>> +		if (active)
>> +			SetPageActive(page);
>> +		update_page_reclaim_stat(zone, page, file, active);
>> +		add_page_to_lru_list(zone, page, lru);
>> +	} else {
>> +		SetPageUnevictable(page);
>> +		add_page_to_lru_list(zone, page, LRU_UNEVICTABLE);
>> +		smp_mb();
>
> Why do we need barrier in here? Please comment it.

To cut-n-paste a comment from putback_lru_page() is good idea? :)

+               /*
+                * When racing with an mlock clearing (page is
+                * unlocked), make sure that if the other thread does
+                * not observe our setting of PG_lru and fails
+                * isolation, we see PG_mlocked cleared below and move
+                * the page back to the evictable list.
+                *
+                * The other side is TestClearPageMlocked().
+                */
+               smp_mb();



>> +		if (page_evictable(page, NULL)) {
>> +			del_page_from_lru_list(zone, page, LRU_UNEVICTABLE);
>> +			ClearPageUnevictable(page);
>> +			goto redo;
>> +		}
>> +	}
>
> I am not sure it's a good idea.
> mlock is very rare event but ____pagevec_lru_add_fn is called frequently.
> We are adding more overhead in ____pagevec_lru_add_fn.
> Is it valuable?

dunno.

Personally, I think tao's case is too artificial and I haven't observed
any real world application do such crazy mlock/munlock repeatness. But
he said he has a such application.

If my remember is correct, ltp or some test suite depend on current 
meminfo synching behavior. then I'm afraid simple removing bring us
new annoying bug report.