Re: [PATCH 2/7] mm, vmscan: add active list aging tracepoint

From: Vlastimil Babka
Date: Wed Jan 04 2017 - 08:41:18 EST


On 01/04/2017 02:16 PM, Michal Hocko wrote:
> On Wed 04-01-17 13:52:24, Vlastimil Babka wrote:
>> On 01/04/2017 11:19 AM, Michal Hocko wrote:
>>> From: Michal Hocko <mhocko@xxxxxxxx>
>>>
>>> Our reclaim process has several tracepoints to tell us more about how
>>> things are progressing. We are, however, missing a tracepoint to track
>>> active list aging. Introduce mm_vmscan_lru_shrink_active which reports
>>> the number of
>>> - nr_scanned, nr_taken pages to tell us the LRU isolation
>>> effectiveness.
>>
>> Well, this point is no longer true, is it...
>
> ups, leftover
> - nr_take - the number of isolated pages

nr_taken

>
>>> - nr_referenced pages which tells us that we are hitting referenced
>>> pages which are deactivated. If this is a large part of the
>>> reported nr_deactivated pages then we might be hitting into
>>> the active list too early because they might be still part of
>>> the working set. This might help to debug performance issues.
>>> - nr_activated pages which tells us how many pages are kept on the
>>
>> "nr_activated" is slightly misleading? They remain active, they are not
>> being activated (that's why the pgactivate vmstat is also not increased
>> on them, right?). I guess rename to "nr_active" ? Or something like
>> "nr_remain_active" although that's longer.
>
> will go with nr_active

OK.

>
>>
>> [...]
>>
>>> @@ -1857,6 +1859,7 @@ static void move_active_pages_to_lru(struct lruvec *lruvec,
>>> unsigned long pgmoved = 0;
>>> struct page *page;
>>> int nr_pages;
>>> + int nr_moved = 0;
>>>
>>> while (!list_empty(list)) {
>>> page = lru_to_page(list);
>>> @@ -1882,11 +1885,15 @@ static void move_active_pages_to_lru(struct lruvec *lruvec,
>>> spin_lock_irq(&pgdat->lru_lock);
>>> } else
>>> list_add(&page->lru, pages_to_free);
>>> + } else {
>>> + nr_moved += nr_pages;
>>> }
>>> }
>>>
>>> if (!is_active_lru(lru))
>>> __count_vm_events(PGDEACTIVATE, pgmoved);
>>
>> So we now have pgmoved and nr_moved. One is used for vmstat, other for
>> tracepoint, and the only difference is that vmstat includes pages where
>> we raced with page being unmapped from all pte's (IIUC?) and thus
>> removed from lru, which should be rather rare? I guess those are being
>> counted into vmstat only due to how the code evolved from using pagevec.
>> If we don't consider them in the tracepoint, then I'd suggest we don't
>> count them into vmstat either, and simplify this.
>
> OK, but I would prefer to have this in a separate patch, OK?

Sure, thanks!