Re: [PATCH v2 12/12] mm: Clear shrinker bit if there are no objects related to memcg

[Vladimir Davydov <vdavydov.dev@gmail.com>]

On Mon, Apr 23, 2018 at 01:01:08PM +0300, Kirill Tkhai wrote:
> On 22.04.2018 21:21, Vladimir Davydov wrote:
> > On Tue, Apr 17, 2018 at 09:54:51PM +0300, Kirill Tkhai wrote:
> >> To avoid further unneed calls of do_shrink_slab()
> >> for shrinkers, which already do not have any charged
> >> objects in a memcg, their bits have to be cleared.
> >>
> >> This patch introduces a lockless mechanism to do that
> >> without races without parallel list lru add. After
> >> do_shrink_slab() returns SHRINK_EMPTY the first time,
> >> we clear the bit and call it once again. Then we restore
> >> the bit, if the new return value is different.
> >>
> >> Note, that single smp_mb__after_atomic() in shrink_slab_memcg()
> >> covers two situations:
> >>
> >> 1)list_lru_add()     shrink_slab_memcg
> >>     list_add_tail()    for_each_set_bit() <--- read bit
> >>                          do_shrink_slab() <--- missed list update (no barrier)
> >>     <MB>                 <MB>
> >>     set_bit()            do_shrink_slab() <--- seen list update
> >>
> >> This situation, when the first do_shrink_slab() sees set bit,
> >> but it doesn't see list update (i.e., race with the first element
> >> queueing), is rare. So we don't add <MB> before the first call
> >> of do_shrink_slab() instead of this to do not slow down generic
> >> case. Also, it's need the second call as seen in below in (2).
> >>
> >> 2)list_lru_add()      shrink_slab_memcg()
> >>     list_add_tail()     ...
> >>     set_bit()           ...
> >>   ...                   for_each_set_bit()
> >>   do_shrink_slab()        do_shrink_slab()
> >>     clear_bit()           ...
> >>   ...                     ...
> >>   list_lru_add()          ...
> >>     list_add_tail()       clear_bit()
> >>     <MB>                  <MB>
> >>     set_bit()             do_shrink_slab()
> >>
> >> The barriers guarantees, the second do_shrink_slab()
> >> in the right side task sees list update if really
> >> cleared the bit. This case is drawn in the code comment.
> >>
> >> [Results/performance of the patchset]
> >>
> >> After the whole patchset applied the below test shows signify
> >> increase of performance:
> >>
> >> $echo 1 > /sys/fs/cgroup/memory/memory.use_hierarchy
> >> $mkdir /sys/fs/cgroup/memory/ct
> >> $echo 4000M > /sys/fs/cgroup/memory/ct/memory.kmem.limit_in_bytes
> >>     $for i in `seq 0 4000`; do mkdir /sys/fs/cgroup/memory/ct/$i; echo $$ > /sys/fs/cgroup/memory/ct/$i/cgroup.procs; mkdir -p s/$i; mount -t tmpfs $i s/$i; touch s/$i/file; done
> >>
> >> Then, 4 sequential calls of drop caches:
> >> $time echo 3 > /proc/sys/vm/drop_caches
> >>
> >> 1)Before:
> >> 0.00user 8.99system 0:08.99elapsed 99%CPU
> >> 0.00user 5.97system 0:05.97elapsed 100%CPU
> >> 0.00user 5.97system 0:05.97elapsed 100%CPU
> >> 0.00user 5.85system 0:05.85elapsed 100%CPU
> >>
> >> 2)After
> >> 0.00user 1.11system 0:01.12elapsed 99%CPU
> >> 0.00user 0.00system 0:00.00elapsed 100%CPU
> >> 0.00user 0.00system 0:00.00elapsed 100%CPU
> >> 0.00user 0.00system 0:00.00elapsed 100%CPU
> >>
> >> Even if we round 0:00.00 up to 0:00.01, the results shows
> >> the performance increases at least in 585 times.
> >>
> >> Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
> >> ---
> >>  include/linux/memcontrol.h |    2 ++
> >>  mm/vmscan.c                |   19 +++++++++++++++++--
> >>  2 files changed, 19 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> >> index e1c1fa8e417a..1c5c68550e2f 100644
> >> --- a/include/linux/memcontrol.h
> >> +++ b/include/linux/memcontrol.h
> >> @@ -1245,6 +1245,8 @@ static inline void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int nr)
> >>  
> >>  		rcu_read_lock();
> >>  		map = SHRINKERS_MAP(memcg, nid);
> >> +		/* Pairs with smp mb in shrink_slab() */
> >> +		smp_mb__before_atomic();
> >>  		set_bit(nr, map->map);
> >>  		rcu_read_unlock();
> >>  	}
> >> diff --git a/mm/vmscan.c b/mm/vmscan.c
> >> index 3be9b4d81c13..a8733bc5377b 100644
> >> --- a/mm/vmscan.c
> >> +++ b/mm/vmscan.c
> >> @@ -579,8 +579,23 @@ static unsigned long shrink_slab_memcg(gfp_t gfp_mask, int nid,
> >>  		}
> >>  
> >>  		ret = do_shrink_slab(&sc, shrinker, priority);
> >> -		if (ret == SHRINK_EMPTY)
> >> -			ret = 0;
> >> +		if (ret == SHRINK_EMPTY) {
> >> +			clear_bit(i, map->map);
> >> +			/*
> >> +			 * Pairs with mb in set_shrinker_bit():
> >> +			 *
> >> +			 * list_lru_add()     shrink_slab_memcg()
> >> +			 *   list_add_tail()    clear_bit()
> >> +			 *   <MB>               <MB>
> >> +			 *   set_bit()          do_shrink_slab()
> >> +			 */
> >> +			smp_mb__after_atomic();
> >> +			ret = do_shrink_slab(&sc, shrinker, priority);
> >> +			if (ret == SHRINK_EMPTY)
> >> +				ret = 0;
> >> +			else
> >> +				set_shrinker_bit(memcg, nid, i);
> >> +		}
> > 
> > This is mind-boggling. Are there any alternatives? For instance, can't
> > we clear the bit in list_lru_del, when we hold the list lock?
> 
> Since a single shrinker may iterate over several lru lists, we can't do that.
> Otherwise, we would have to probe another shrinker's lru list from a lru list,
> which became empty in list_lru_del().
> 
> The solution I suggested, is generic, and it does not depend on low-level
> structure type, used by shrinker. This even doesn't have to be a lru list.

Fair enough. I guess this is the best way to go after all. Please try to
improve the comment so that it isn't just a pure diagram.

Also, please prefix all memcg-related function names (such as
set_shrinker_bit) with memcg_ (or mem_cgroup_) in this and all
other patches.