[PATCH 0/3] record non-slab shrinkers for non-root memcgs when kmem is disabled

[PATCH 0/3] record non-slab shrinkers for non-root memcgs when kmem is disabled

[Haifeng Xu]

When registering new shrinkers, all memcgs need to expand shrinker info
if new allocated id exceeds shrinker_nr_max. But if kmem is disabled,
only non-slab shrinkers is useful in memcg slab shrink. So in this case,
it is enough to allocate non-slab shrinker info for non-root memcg. This
can save a bit of memory and reduce the holding time of shrinker lock.

With this optimization, the finish time of pod creation in our internal
test is reduced from 150 senconds to 69 senconds. We test it based on
stable kernel 6.6.102.

Haifeng Xu (3):
  mm: shrinker: introduce new function get_shrinker_id()
  mm: shrinker: optimize the allocation of shrinker_info when setting
    cgroup_memory_nokmem
  mm: shrinker: remove unnecessary check in shrink_slab_memcg()

 include/linux/shrinker.h |   3 +
 mm/huge_memory.c         |  21 ++++--
 mm/shrinker.c            | 155 +++++++++++++++++++++++++++++++++------
 3 files changed, 150 insertions(+), 29 deletions(-)

-- 
2.43.0

[PATCH 1/3] mm: shrinker: introduce new function get_shrinker_id()

[Haifeng Xu]

Use get_shrinker_id() to retrieve the id of shrinker. No functional change.

Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
---
 mm/shrinker.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/mm/shrinker.c b/mm/shrinker.c
index 7b61fc0ee78f..f0c6dfa026b0 100644
--- a/mm/shrinker.c
+++ b/mm/shrinker.c
@@ -193,6 +193,11 @@ static inline int calc_shrinker_id(int index, int offset)
 	return index * SHRINKER_UNIT_BITS + offset;
 }
 
+static inline int get_shrinker_id(struct mem_cgroup *memcg, struct shrinker *shrinker)
+{
+	return shrinker->id;
+}
+
 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id)
 {
 	if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) {
@@ -255,11 +260,13 @@ static long xchg_nr_deferred_memcg(int nid, struct shrinker *shrinker,
 	struct shrinker_info *info;
 	struct shrinker_info_unit *unit;
 	long nr_deferred;
+	int id;
 
 	rcu_read_lock();
+	id = get_shrinker_id(memcg, shrinker);
 	info = rcu_dereference(memcg->nodeinfo[nid]->shrinker_info);
-	unit = info->unit[shrinker_id_to_index(shrinker->id)];
-	nr_deferred = atomic_long_xchg(&unit->nr_deferred[shrinker_id_to_offset(shrinker->id)], 0);
+	unit = info->unit[shrinker_id_to_index(id)];
+	nr_deferred = atomic_long_xchg(&unit->nr_deferred[shrinker_id_to_offset(id)], 0);
 	rcu_read_unlock();
 
 	return nr_deferred;
@@ -271,12 +278,14 @@ static long add_nr_deferred_memcg(long nr, int nid, struct shrinker *shrinker,
 	struct shrinker_info *info;
 	struct shrinker_info_unit *unit;
 	long nr_deferred;
+	int id;
 
 	rcu_read_lock();
+	id = get_shrinker_id(memcg, shrinker);
 	info = rcu_dereference(memcg->nodeinfo[nid]->shrinker_info);
-	unit = info->unit[shrinker_id_to_index(shrinker->id)];
+	unit = info->unit[shrinker_id_to_index(id)];
 	nr_deferred =
-		atomic_long_add_return(nr, &unit->nr_deferred[shrinker_id_to_offset(shrinker->id)]);
+		atomic_long_add_return(nr, &unit->nr_deferred[shrinker_id_to_offset(id)]);
 	rcu_read_unlock();
 
 	return nr_deferred;
-- 
2.43.0

[PATCH 2/3] mm: shrinker: optimize the allocation of shrinker_info when setting cgroup_memory_nokmem

[Haifeng Xu]

When kmem is disabled, memcg slab shrink only call non-slab shrinkers,
so just allocates shrinker info for non-slab shrinkers to non-root memcgs.

Therefore, if memcg_kmem_online is true, all things keep same as before.
Otherwise, root memcg allocates id from shrinker_idr to identify each
shrinker and non-root memcgs use nonslab_id to identify non-slab shrinkers.
The size of shrinkers_info in non-root memcgs can be very low because the
number of shrinkers marked as SHRINKER_NONSLAB | SHRINKER_MEMCG_AWARE is
few. Also, the time spending in expand_shrinker_info() can reduce a lot.

When setting shrinker bit or updating nr_deferred, use nonslab_id for
non-root memcgs if the shrinker is marked as SHRINKER_NONSLAB.

Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
---
 include/linux/shrinker.h |   3 +
 mm/huge_memory.c         |  21 ++++--
 mm/shrinker.c            | 135 ++++++++++++++++++++++++++++++++++-----
 3 files changed, 138 insertions(+), 21 deletions(-)

diff --git a/include/linux/shrinker.h b/include/linux/shrinker.h
index 1a00be90d93a..df53008ed8b5 100644
--- a/include/linux/shrinker.h
+++ b/include/linux/shrinker.h
@@ -107,6 +107,9 @@ struct shrinker {
 #ifdef CONFIG_MEMCG
 	/* ID in shrinker_idr */
 	int id;
+
+	/* ID in shrinker_nonslab_idr */
+	int nonslab_id;
 #endif
 #ifdef CONFIG_SHRINKER_DEBUG
 	int debugfs_id;
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 8e2746ea74ad..319349b5da5d 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -4351,9 +4351,14 @@ void deferred_split_folio(struct folio *folio, bool partially_mapped)
 		memcg = folio_split_queue_memcg(folio, ds_queue);
 		list_add_tail(&folio->_deferred_list, &ds_queue->split_queue);
 		ds_queue->split_queue_len++;
-		if (memcg)
-			set_shrinker_bit(memcg, folio_nid(folio),
-					 shrinker_id(deferred_split_shrinker));
+		if (memcg) {
+			int id = deferred_split_shrinker->id;
+
+			if (!memcg_kmem_online() &&  memcg != root_mem_cgroup)
+				id = deferred_split_shrinker->nonslab_id;
+
+			set_shrinker_bit(memcg, folio_nid(folio), id);
+		}
 	}
 	split_queue_unlock_irqrestore(ds_queue, flags);
 }
@@ -4508,8 +4513,14 @@ void reparent_deferred_split_queue(struct mem_cgroup *memcg)
 	parent_ds_queue->split_queue_len += ds_queue->split_queue_len;
 	ds_queue->split_queue_len = 0;
 
-	for_each_node(nid)
-		set_shrinker_bit(parent, nid, shrinker_id(deferred_split_shrinker));
+	for_each_node(nid) {
+		int id = deferred_split_shrinker->id;
+
+		if (!memcg_kmem_online() && parent != root_mem_cgroup)
+			id = deferred_split_shrinker->nonslab_id;
+
+		set_shrinker_bit(parent, nid, id);
+	}
 
 unlock:
 	spin_unlock(&parent_ds_queue->split_queue_lock);
diff --git a/mm/shrinker.c b/mm/shrinker.c
index f0c6dfa026b0..52ea0e6391af 100644
--- a/mm/shrinker.c
+++ b/mm/shrinker.c
@@ -12,6 +12,7 @@ DEFINE_MUTEX(shrinker_mutex);
 
 #ifdef CONFIG_MEMCG
 static int shrinker_nr_max;
+static int shrinker_nonslab_nr_max;
 
 static inline int shrinker_unit_size(int nr_items)
 {
@@ -78,15 +79,25 @@ int alloc_shrinker_info(struct mem_cgroup *memcg)
 {
 	int nid, ret = 0;
 	int array_size = 0;
+	int alloc_nr_max;
+
+	if (memcg_kmem_online()) {
+		alloc_nr_max = shrinker_nr_max;
+	} else {
+		if (memcg == root_mem_cgroup)
+			alloc_nr_max = shrinker_nr_max;
+		else
+			alloc_nr_max = shrinker_nonslab_nr_max;
+	}
 
 	mutex_lock(&shrinker_mutex);
-	array_size = shrinker_unit_size(shrinker_nr_max);
+	array_size = shrinker_unit_size(alloc_nr_max);
 	for_each_node(nid) {
 		struct shrinker_info *info = kvzalloc_node(sizeof(*info) + array_size,
 							   GFP_KERNEL, nid);
 		if (!info)
 			goto err;
-		info->map_nr_max = shrinker_nr_max;
+		info->map_nr_max = alloc_nr_max;
 		if (shrinker_unit_alloc(info, NULL, nid)) {
 			kvfree(info);
 			goto err;
@@ -147,33 +158,47 @@ static int expand_one_shrinker_info(struct mem_cgroup *memcg, int new_size,
 	return 0;
 }
 
-static int expand_shrinker_info(int new_id)
+static int expand_shrinker_info(int new_id, bool full, bool root)
 {
 	int ret = 0;
 	int new_nr_max = round_up(new_id + 1, SHRINKER_UNIT_BITS);
 	int new_size, old_size = 0;
 	struct mem_cgroup *memcg;
+	struct mem_cgroup *start = NULL;
+	int old_nr_max = shrinker_nr_max;
 
 	if (!root_mem_cgroup)
 		goto out;
 
 	lockdep_assert_held(&shrinker_mutex);
 
+	if (!full && !root) {
+		start = root_mem_cgroup;
+		old_nr_max = shrinker_nonslab_nr_max;
+	}
+
 	new_size = shrinker_unit_size(new_nr_max);
-	old_size = shrinker_unit_size(shrinker_nr_max);
+	old_size = shrinker_unit_size(old_nr_max);
+
+	memcg = mem_cgroup_iter(NULL, start, NULL);
+	if (!memcg)
+		goto out;
 
-	memcg = mem_cgroup_iter(NULL, NULL, NULL);
 	do {
 		ret = expand_one_shrinker_info(memcg, new_size, old_size,
 					       new_nr_max);
-		if (ret) {
+		if (ret || (root && memcg == root_mem_cgroup)) {
 			mem_cgroup_iter_break(NULL, memcg);
 			goto out;
 		}
 	} while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL);
 out:
-	if (!ret)
-		shrinker_nr_max = new_nr_max;
+	if (!ret) {
+		if (!full && !root)
+			shrinker_nonslab_nr_max = new_nr_max;
+		else
+			shrinker_nr_max = new_nr_max;
+	}
 
 	return ret;
 }
@@ -195,7 +220,13 @@ static inline int calc_shrinker_id(int index, int offset)
 
 static inline int get_shrinker_id(struct mem_cgroup *memcg, struct shrinker *shrinker)
 {
-	return shrinker->id;
+	int id = shrinker->id;
+
+	if (!memcg_kmem_online() && (shrinker->flags & SHRINKER_NONSLAB) &&
+	    memcg != root_mem_cgroup)
+		id = shrinker->nonslab_id;
+
+	return id;
 }
 
 void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id)
@@ -217,6 +248,8 @@ void set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id)
 }
 
 static DEFINE_IDR(shrinker_idr);
+static DEFINE_IDR(shrinker_nonslab_idr);
+
 
 static int shrinker_memcg_alloc(struct shrinker *shrinker)
 {
@@ -230,10 +263,46 @@ static int shrinker_memcg_alloc(struct shrinker *shrinker)
 	if (id < 0)
 		goto unlock;
 
-	if (id >= shrinker_nr_max) {
-		if (expand_shrinker_info(id)) {
-			idr_remove(&shrinker_idr, id);
-			goto unlock;
+	shrinker->nonslab_id = -1;
+
+	if (!mem_cgroup_kmem_disabled()) {
+		if (id >= shrinker_nr_max) {
+			/* expand shrinker info for all memory cgroups */
+			if (expand_shrinker_info(id, true, false)) {
+				idr_remove(&shrinker_idr, id);
+				goto unlock;
+			}
+		}
+	} else {
+		/*
+		 * If cgroup_memory_nokmem is set, every shrinker needs to be recorded in
+		 * root memory cgroup because gloal slab shrink traverse all shrinkers. For
+		 * non-root memcgs, record shrinkers with SHRINKER_NONSLAB because memcg
+		 * slab shrink only call non-slab shrinkers.
+		 */
+		if (id >= shrinker_nr_max) {
+			/* expand shrinker info for root memory cgroup only */
+			if (expand_shrinker_info(id, false, true)) {
+				idr_remove(&shrinker_idr, id);
+				goto unlock;
+			}
+		}
+
+		if (shrinker->flags & SHRINKER_NONSLAB) {
+			int nonslab_id;
+
+			nonslab_id = idr_alloc(&shrinker_nonslab_idr, shrinker, 0, 0, GFP_KERNEL);
+			if (nonslab_id < 0)
+				goto unlock;
+
+			if (nonslab_id >= shrinker_nonslab_nr_max) {
+				/* expand shrinker info for non-root memory cgroups */
+				if (expand_shrinker_info(nonslab_id, false, false)) {
+					idr_remove(&shrinker_nonslab_idr, id);
+					goto unlock;
+				}
+			}
+			shrinker->nonslab_id = nonslab_id;
 		}
 	}
 	shrinker->id = id;
@@ -252,6 +321,12 @@ static void shrinker_memcg_remove(struct shrinker *shrinker)
 	lockdep_assert_held(&shrinker_mutex);
 
 	idr_remove(&shrinker_idr, id);
+
+	if (shrinker->flags & SHRINKER_NONSLAB) {
+		id = shrinker->nonslab_id;
+		if (id >= 0)
+			idr_remove(&shrinker_nonslab_idr, id);
+	}
 }
 
 static long xchg_nr_deferred_memcg(int nid, struct shrinker *shrinker,
@@ -310,10 +385,33 @@ void reparent_shrinker_deferred(struct mem_cgroup *memcg)
 		parent_info = shrinker_info_protected(parent, nid);
 		for (index = 0; index < shrinker_id_to_index(child_info->map_nr_max); index++) {
 			child_unit = child_info->unit[index];
-			parent_unit = parent_info->unit[index];
 			for (offset = 0; offset < SHRINKER_UNIT_BITS; offset++) {
 				nr = atomic_long_read(&child_unit->nr_deferred[offset]);
-				atomic_long_add(nr, &parent_unit->nr_deferred[offset]);
+
+				/*
+				 * If memcg_kmem_online() is false, the non-root memcgs use
+				 * nonslab_id but root memory cgroup use id. When reparenting
+				 * shrinker info to it, must convert the nonslab_id to id.
+				 */
+				if (!memcg_kmem_online() &&  parent == root_mem_cgroup) {
+					int id, p_index, p_off;
+					struct shrinker *shrinker;
+
+					id = calc_shrinker_id(index, offset);
+					shrinker = idr_find(&shrinker_nonslab_idr, id);
+					if (shrinker) {
+						id = shrinker->id;
+						p_index = shrinker_id_to_index(id);
+						p_off = shrinker_id_to_offset(id);
+
+						parent_unit = parent_info->unit[p_index];
+						atomic_long_add(nr,
+								&parent_unit->nr_deferred[p_off]);
+					}
+				} else {
+					parent_unit = parent_info->unit[index];
+					atomic_long_add(nr, &parent_unit->nr_deferred[offset]);
+				}
 			}
 		}
 	}
@@ -543,7 +641,12 @@ static unsigned long shrink_slab_memcg(gfp_t gfp_mask, int nid,
 			int shrinker_id = calc_shrinker_id(index, offset);
 
 			rcu_read_lock();
-			shrinker = idr_find(&shrinker_idr, shrinker_id);
+
+			if (memcg_kmem_online())
+				shrinker = idr_find(&shrinker_idr, shrinker_id);
+			else
+				shrinker = idr_find(&shrinker_nonslab_idr, shrinker_id);
+
 			if (unlikely(!shrinker || !shrinker_try_get(shrinker))) {
 				clear_bit(offset, unit->map);
 				rcu_read_unlock();
-- 
2.43.0

[PATCH 3/3] mm: shrinker: remove unnecessary check in shrink_slab_memcg()

[Haifeng Xu]

If memcg_kmem_online() is false, only non-slab shrinkers are recorded
in the map, so remove the check.

Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
---
 mm/shrinker.c | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/mm/shrinker.c b/mm/shrinker.c
index 52ea0e6391af..c8b090e8a972 100644
--- a/mm/shrinker.c
+++ b/mm/shrinker.c
@@ -654,11 +654,6 @@ static unsigned long shrink_slab_memcg(gfp_t gfp_mask, int nid,
 			}
 			rcu_read_unlock();
 
-			/* Call non-slab shrinkers even though kmem is disabled */
-			if (!memcg_kmem_online() &&
-			    !(shrinker->flags & SHRINKER_NONSLAB))
-				continue;
-
 			ret = do_shrink_slab(&sc, shrinker, priority);
 			if (ret == SHRINK_EMPTY) {
 				clear_bit(offset, unit->map);
-- 
2.43.0

Re: [PATCH 1/3] mm: shrinker: introduce new function get_shrinker_id()

[Qi Zheng]

On 3/6/26 3:57 PM, Haifeng Xu wrote:
> Use get_shrinker_id() to retrieve the id of shrinker. No functional change.
> 
> Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
> ---
>   mm/shrinker.c | 17 +++++++++++++----
>   1 file changed, 13 insertions(+), 4 deletions(-)
> 

We already have shrinker_id() in memcontrol.h.

Re: [PATCH 1/3] mm: shrinker: introduce new function get_shrinker_id()

[Haifeng Xu]

On 2026/3/6 16:24, Qi Zheng wrote:
> On 3/6/26 3:57 PM, Haifeng Xu wrote:
>> Use get_shrinker_id() to retrieve the id of shrinker. No functional change.
>>
>> Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
>> ---
>>   mm/shrinker.c | 17 +++++++++++++----
>>   1 file changed, 13 insertions(+), 4 deletions(-)
>>
> 
> We already have shrinker_id() in memcontrol.h.

Yes, thanks for your reminding.