Revert "mm: vmscan: hold write lock to reparent shrinker nr_deferred"
authorQi Zheng <zhengqi.arch@bytedance.com>
Fri, 9 Jun 2023 08:15:14 +0000 (08:15 +0000)
committerAndrew Morton <akpm@linux-foundation.org>
Mon, 19 Jun 2023 20:19:33 +0000 (13:19 -0700)
This reverts commit b3cabea3c9153fd42fe5cb851ac58b51ea2b32b8.

Kernel test robot reports -88.8% regression in stress-ng.ramfs.ops_per_sec
test case [1], which is caused by commit f95bdb700bc6 ("mm: vmscan: make
global slab shrink lockless"). The root cause is that SRCU has to be careful
to not frequently check for SRCU read-side critical section exits. Therefore,
even if no one is currently in the SRCU read-side critical section,
synchronize_srcu() cannot return quickly. That's why unregister_shrinker()
has become slower.

We will try to use the refcount+RCU method [2] proposed by Dave Chinner
to continue to re-implement the lockless slab shrink. Because there will
be other readers after reverting the shrinker_srcu related changes, so
it is better to restore to hold read lock to reparent shrinker nr_deferred.

[1]. https://lore.kernel.org/lkml/202305230837.db2c233f-yujie.liu@intel.com/
[2]. https://lore.kernel.org/lkml/ZIJhou1d55d4H1s0@dread.disaster.area/

Link: https://lkml.kernel.org/r/20230609081518.3039120-4-qi.zheng@linux.dev
Reported-by: kernel test robot <yujie.liu@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202305230837.db2c233f-yujie.liu@intel.com
Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Kirill Tkhai <tkhai@ya.ru>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/vmscan.c

index 0ba0e11..d1d309f 100644 (file)
@@ -433,7 +433,7 @@ void reparent_shrinker_deferred(struct mem_cgroup *memcg)
                parent = root_mem_cgroup;
 
        /* Prevent from concurrent shrinker_info expand */
-       down_write(&shrinker_rwsem);
+       down_read(&shrinker_rwsem);
        for_each_node(nid) {
                child_info = shrinker_info_protected(memcg, nid);
                parent_info = shrinker_info_protected(parent, nid);
@@ -442,7 +442,7 @@ void reparent_shrinker_deferred(struct mem_cgroup *memcg)
                        atomic_long_add(nr, &parent_info->nr_deferred[i]);
                }
        }
-       up_write(&shrinker_rwsem);
+       up_read(&shrinker_rwsem);
 }
 
 static bool cgroup_reclaim(struct scan_control *sc)