lkml.org 
[lkml]   [2020]   [Mar]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 2/2] mm: swap: use smp_mb__after_atomic() to order LRU bit set
From
Date


On 3/16/20 10:49 AM, Yang Shi wrote:
>
>
> On 3/16/20 10:40 AM, Vlastimil Babka wrote:
>> On 3/13/20 7:34 PM, Yang Shi wrote:
>>> Memory barrier is needed after setting LRU bit, but smp_mb() is too
>>> strong.  Some architectures, i.e. x86, imply memory barrier with atomic
>>> operations, so replacing it with smp_mb__after_atomic() sounds better,
>>> which is nop on strong ordered machines, and full memory barriers on
>>> others.  With this change the vm-calability cases would perform better
>>> on x86, I saw total 6% improvement with this patch and previous inline
>>> fix.
>>>
>>> The test data (lru-file-readtwice throughput) against v5.6-rc4:
>>>     mainline    w/ inline fix    w/ both (adding this)
>>>     150MB        154MB        159MB
>>>
>>> Fixes: 9c4e6b1a7027 ("mm, mlock, vmscan: no more skipping pagevecs")
>>> Cc: Shakeel Butt <shakeelb@google.com>
>>> Cc: Vlastimil Babka <vbabka@suse.cz>
>>> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
>> According to my understanding of Documentation/memory_barriers.txt
>> this would be
>> correct (but it might not say much :)
>
> This is my understanding too.
>
>>
>> Acked-by: Vlastimil Babka <vbabka@suse.cz>
>>
>> But i have some suggestions...
>>
>>> ---
>>>   mm/swap.c | 6 +++---
>>>   1 file changed, 3 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/mm/swap.c b/mm/swap.c
>>> index cf39d24..118bac4 100644
>>> --- a/mm/swap.c
>>> +++ b/mm/swap.c
>>> @@ -945,20 +945,20 @@ static void __pagevec_lru_add_fn(struct page
>>> *page, struct lruvec *lruvec,
>>>        * #0: __pagevec_lru_add_fn        #1: clear_page_mlock
>>>        *
>>>        * SetPageLRU()                TestClearPageMlocked()
>>> -     * smp_mb() // explicit ordering    // above provides strict
>>> +     * MB()     // explicit ordering    // above provides strict
>> Why MB()? That would be the first appareance of 'MB()' in the whole
>> tree. I
>> think it's fine keeping smp_mb()...
>
> I would like to use a more general name, maybe just use "memory barrier"?

Keeping smp_mb() should be just fine...

>
>>
>>>        *                    // ordering
>>>        * PageMlocked()            PageLRU()
>>>        *
>>>        *
>>>        * if '#1' does not observe setting of PG_lru by '#0' and fails
>>>        * isolation, the explicit barrier will make sure that
>>> page_evictable
>>> -     * check will put the page in correct LRU. Without smp_mb(),
>>> SetPageLRU
>>> +     * check will put the page in correct LRU. Without MB(),
>>> SetPageLRU
>> ... same here ...
>>
>>>        * can be reordered after PageMlocked check and can make '#1'
>>> to fail
>>>        * the isolation of the page whose Mlocked bit is cleared (#0
>>> is also
>>>        * looking at the same page) and the evictable page will be
>>> stranded
>>>        * in an unevictable LRU.
>> Only here I would note that SetPageLRU() is an atomic bitop so we can
>> use the
>> __after_atomic() variant. And I would move the actual SetPageLRU()
>> call from
>> above the comment here right before the barrier.
>
> Sure. Thanks.
>
>>
>>>        */
>>> -    smp_mb();
>>> +    smp_mb__after_atomic();
>> Thanks.
>>
>>>         if (page_evictable(page)) {
>>>           lru = page_lru(page);
>>>
>

\
 
 \ /
  Last update: 2020-03-16 23:20    [W:0.087 / U:0.592 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site