[RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

Changes in RFC v3:
 - Directly use __set_bit() and __clear_bit() for the slab_node_partial
   flag operations to avoid exporting non-atomic "workingset" interfaces.
 - Change get_partial() related functions to return a slab instead of
   returning the freelist or single object.
 - Don't freeze any slab under the node list_lock to further reduce
   list_lock holding times, as suggested by Vlastimil Babka.
 - Introduce freeze_slab() to do the delay freezing and return freelist.
 - Reorder patches.
 - RFC v2: https://lore.kernel.org/all/20231021144317.3400916-1-chengming.zhou@linux.dev/

Changes in RFC v2:
 - Reuse PG_workingset bit to keep track of whether slub is on the
   per-node partial list, as suggested by Matthew Wilcox.
 - Fix OOM problem on kernel without CONFIG_SLUB_CPU_PARTIAL, which
   is caused by leak of partial slabs when get_partial_node().
 - Add a patch to simplify acquire_slab().
 - Reorder patches a little.
 - RFC v1: https://lore.kernel.org/all/20231017154439.3036608-1-chengming.zhou@linux.dev/

1. Problem
==========
Now we have to freeze the slab when get from the node partial list, and
unfreeze the slab when put to the node partial list. Because we need to
rely on the node list_lock to synchronize the "frozen" bit changes.

This implementation has some drawbacks:

 - Alloc path: twice cmpxchg_double.
   It has to get some partial slabs from node when the allocator has used
   up the CPU partial slabs. So it freeze the slab (one cmpxchg_double)
   with node list_lock held, put those frozen slabs on its CPU partial
   list. Later ___slab_alloc() will cmpxchg_double try-loop again if that
   slab is picked to use.

 - Alloc path: amplified contention on node list_lock.
   Since we have to synchronize the "frozen" bit changes under the node
   list_lock, the contention of slab (struct page) can be transferred
   to the node list_lock. On machine with many CPUs in one node, the
   contention of list_lock will be amplified by all CPUs' alloc path.

   The current code has to workaround this problem by avoiding using
   cmpxchg_double try-loop, which will just break and return when
   contention of page encountered and the first cmpxchg_double failed.
   But this workaround has its own problem. For more context, see
   9b1ea29bc0d7 ("Revert "mm, slub: consider rest of partial list if
   acquire_slab() fails"").

 - Free path: redundant unfreeze.
   __slab_free() will freeze and cache some slabs on its partial list,
   and flush them to the node partial list when exceed, which has to
   unfreeze those slabs again under the node list_lock. Actually we
   don't need to freeze slab on CPU partial list, in which case we
   can save the unfreeze cmpxchg_double operations in flush path.

2. Solution
===========
We solve these problems by leaving slabs unfrozen when moving out of
the node partial list and on CPU partial list, so "frozen" bit is 0.

These partial slabs won't be manipulate concurrently by alloc path,
the only racer is free path, which may manipulate its list when !inuse.
So we need to introduce another synchronization way to avoid it, we
reuse PG_workingset to keep track of whether the slab is on node partial
list or not, only in that case we can manipulate the slab list.

The slab will be delay frozen when it's picked to actively use by the
CPU, it becomes full at the same time, in which case we still need to
rely on "frozen" bit to avoid manipulating its list. So the slab will
be frozen only when activate use and be unfrozen only when deactivate.

3. Testing
==========
We just did some simple testing on a server with 128 CPUs (2 nodes) to
compare performance for now.

 - perf bench sched messaging -g 5 -t -l 100000
   baseline	RFC
   7.042s	6.966s
   7.022s	7.045s
   7.054s	6.985s

 - stress-ng --rawpkt 128 --rawpkt-ops 100000000
   baseline	RFC
   2.42s	2.15s
   2.45s	2.16s
   2.44s	2.17s

It shows above there is about 10% improvement on stress-ng rawpkt
testcase, although no much improvement on perf sched bench testcase.

Thanks for any comment and code review!

Chengming Zhou (7):
  slub: Keep track of whether slub is on the per-node partial list
  slub: Prepare __slab_free() for unfrozen partial slab out of node
    partial list
  slub: Reflow ___slab_alloc()
  slub: Change get_partial() interfaces to return slab
  slub: Introduce freeze_slab()
  slub: Delay freezing of partial slabs
  slub: Optimize deactivate_slab()

 mm/slab.h |  19 ++++
 mm/slub.c | 306 +++++++++++++++++++++++-------------------------------
 2 files changed, 149 insertions(+), 176 deletions(-)

-- 
2.40.1

[RFC PATCH v3 1/7] slub: Keep track of whether slub is on the per-node partial list

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

Now we rely on the "frozen" bit to see if we should manipulate the
slab->slab_list, which will be changed in the following patch.

Instead we introduce another way to keep track of whether slub is on
the per-node partial list, here we reuse the PG_workingset bit.

We use __set_bit and __clear_bit directly instead of the atomic version
for better performance and it's safe since it's protected by the slub
node list_lock.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slab.h | 19 +++++++++++++++++++
 mm/slub.c |  3 +++
 2 files changed, 22 insertions(+)

diff --git a/mm/slab.h b/mm/slab.h
index 8cd3294fedf5..50522b688cfb 100644
--- a/mm/slab.h
+++ b/mm/slab.h
@@ -193,6 +193,25 @@ static inline void __slab_clear_pfmemalloc(struct slab *slab)
 	__folio_clear_active(slab_folio(slab));
 }
 
+/*
+ * Slub reuse PG_workingset bit to keep track of whether it's on
+ * the per-node partial list.
+ */
+static inline bool slab_test_node_partial(const struct slab *slab)
+{
+	return folio_test_workingset((struct folio *)slab_folio(slab));
+}
+
+static inline void slab_set_node_partial(struct slab *slab)
+{
+	__set_bit(PG_workingset, folio_flags(slab_folio(slab), 0));
+}
+
+static inline void slab_clear_node_partial(struct slab *slab)
+{
+	__clear_bit(PG_workingset, folio_flags(slab_folio(slab), 0));
+}
+
 static inline void *slab_address(const struct slab *slab)
 {
 	return folio_address(slab_folio(slab));
diff --git a/mm/slub.c b/mm/slub.c
index 63d281dfacdb..3fad4edca34b 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2127,6 +2127,7 @@ __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail)
 		list_add_tail(&slab->slab_list, &n->partial);
 	else
 		list_add(&slab->slab_list, &n->partial);
+	slab_set_node_partial(slab);
 }
 
 static inline void add_partial(struct kmem_cache_node *n,
@@ -2141,6 +2142,7 @@ static inline void remove_partial(struct kmem_cache_node *n,
 {
 	lockdep_assert_held(&n->list_lock);
 	list_del(&slab->slab_list);
+	slab_clear_node_partial(slab);
 	n->nr_partial--;
 }
 
@@ -4831,6 +4833,7 @@ static int __kmem_cache_do_shrink(struct kmem_cache *s)
 
 			if (free == slab->objects) {
 				list_move(&slab->slab_list, &discard);
+				slab_clear_node_partial(slab);
 				n->nr_partial--;
 				dec_slabs_node(s, node, slab->objects);
 			} else if (free <= SHRINK_PROMOTE_MAX)
-- 
2.40.1

[RFC PATCH v3 2/7] slub: Prepare __slab_free() for unfrozen partial slab out of node partial list

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

Now the partial slub will be frozen when taken out of node partial list,
so the __slab_free() will know from "was_frozen" that the partial slab
is not on node partial list and is used by one kmem_cache_cpu.

But we will change this, make partial slabs leave the node partial list
with unfrozen state, so we need to change __slab_free() to use the new
slab_test_node_partial() we just introduced.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/mm/slub.c b/mm/slub.c
index 3fad4edca34b..f568a32d7332 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3610,6 +3610,7 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
 	unsigned long counters;
 	struct kmem_cache_node *n = NULL;
 	unsigned long flags;
+	bool on_node_partial;
 
 	stat(s, FREE_SLOWPATH);
 
@@ -3657,6 +3658,7 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
 				 */
 				spin_lock_irqsave(&n->list_lock, flags);
 
+				on_node_partial = slab_test_node_partial(slab);
 			}
 		}
 
@@ -3685,6 +3687,15 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
 		return;
 	}
 
+	/*
+	 * This slab was partial but not on the per-node partial list,
+	 * in which case we shouldn't manipulate its list, just return.
+	 */
+	if (prior && !on_node_partial) {
+		spin_unlock_irqrestore(&n->list_lock, flags);
+		return;
+	}
+
 	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
 		goto slab_empty;
 
-- 
2.40.1

[RFC PATCH v3 3/7] slub: Reflow ___slab_alloc()

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

The get_partial() interface used in ___slab_alloc() may return a single
object in the "kmem_cache_debug(s)" case, in which we will just return
the "freelist" object.

Move this handling up to prepare for later changes.

And the "pfmemalloc_match()" part is not needed for node partial slab,
since we already check this in the get_partial_node().

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 31 +++++++++++++++----------------
 1 file changed, 15 insertions(+), 16 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index f568a32d7332..cd8aa68c156e 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3218,8 +3218,21 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 	pc.slab = &slab;
 	pc.orig_size = orig_size;
 	freelist = get_partial(s, node, &pc);
-	if (freelist)
-		goto check_new_slab;
+	if (freelist) {
+		if (kmem_cache_debug(s)) {
+			/*
+			 * For debug caches here we had to go through
+			 * alloc_single_from_partial() so just store the
+			 * tracking info and return the object.
+			 */
+			if (s->flags & SLAB_STORE_USER)
+				set_track(s, freelist, TRACK_ALLOC, addr);
+
+			return freelist;
+		}
+
+		goto retry_load_slab;
+	}
 
 	slub_put_cpu_ptr(s->cpu_slab);
 	slab = new_slab(s, gfpflags, node);
@@ -3255,20 +3268,6 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 
 	inc_slabs_node(s, slab_nid(slab), slab->objects);
 
-check_new_slab:
-
-	if (kmem_cache_debug(s)) {
-		/*
-		 * For debug caches here we had to go through
-		 * alloc_single_from_partial() so just store the tracking info
-		 * and return the object
-		 */
-		if (s->flags & SLAB_STORE_USER)
-			set_track(s, freelist, TRACK_ALLOC, addr);
-
-		return freelist;
-	}
-
 	if (unlikely(!pfmemalloc_match(slab, gfpflags))) {
 		/*
 		 * For !pfmemalloc_match() case we don't load freelist so that
-- 
2.40.1

[RFC PATCH v3 4/7] slub: Change get_partial() interfaces to return slab

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

We need all get_partial() related interfaces to return a slab, instead
of returning the freelist (or object).

Use the partial_context.object to return back freelist or object for
now. This patch shouldn't have any functional changes.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 63 +++++++++++++++++++++++++++++--------------------------
 1 file changed, 33 insertions(+), 30 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index cd8aa68c156e..7d0234bffad3 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -204,9 +204,9 @@ DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
 
 /* Structure holding parameters for get_partial() call chain */
 struct partial_context {
-	struct slab **slab;
 	gfp_t flags;
 	unsigned int orig_size;
+	void *object;
 };
 
 static inline bool kmem_cache_debug(struct kmem_cache *s)
@@ -2271,10 +2271,11 @@ static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags);
 /*
  * Try to allocate a partial slab from a specific node.
  */
-static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
-			      struct partial_context *pc)
+static struct slab *get_partial_node(struct kmem_cache *s,
+				     struct kmem_cache_node *n,
+				     struct partial_context *pc)
 {
-	struct slab *slab, *slab2;
+	struct slab *slab, *slab2, *partial = NULL;
 	void *object = NULL;
 	unsigned long flags;
 	unsigned int partial_slabs = 0;
@@ -2290,27 +2291,28 @@ static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
 
 	spin_lock_irqsave(&n->list_lock, flags);
 	list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) {
-		void *t;
-
 		if (!pfmemalloc_match(slab, pc->flags))
 			continue;
 
 		if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
 			object = alloc_single_from_partial(s, n, slab,
 							pc->orig_size);
-			if (object)
+			if (object) {
+				partial = slab;
+				pc->object = object;
 				break;
+			}
 			continue;
 		}
 
-		t = acquire_slab(s, n, slab, object == NULL);
-		if (!t)
+		object = acquire_slab(s, n, slab, object == NULL);
+		if (!object)
 			break;
 
-		if (!object) {
-			*pc->slab = slab;
+		if (!partial) {
+			partial = slab;
+			pc->object = object;
 			stat(s, ALLOC_FROM_PARTIAL);
-			object = t;
 		} else {
 			put_cpu_partial(s, slab, 0);
 			stat(s, CPU_PARTIAL_NODE);
@@ -2326,20 +2328,21 @@ static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
 
 	}
 	spin_unlock_irqrestore(&n->list_lock, flags);
-	return object;
+	return partial;
 }
 
 /*
  * Get a slab from somewhere. Search in increasing NUMA distances.
  */
-static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
+static struct slab *get_any_partial(struct kmem_cache *s,
+				    struct partial_context *pc)
 {
 #ifdef CONFIG_NUMA
 	struct zonelist *zonelist;
 	struct zoneref *z;
 	struct zone *zone;
 	enum zone_type highest_zoneidx = gfp_zone(pc->flags);
-	void *object;
+	struct slab *slab;
 	unsigned int cpuset_mems_cookie;
 
 	/*
@@ -2374,8 +2377,8 @@ static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
 
 			if (n && cpuset_zone_allowed(zone, pc->flags) &&
 					n->nr_partial > s->min_partial) {
-				object = get_partial_node(s, n, pc);
-				if (object) {
+				slab = get_partial_node(s, n, pc);
+				if (slab) {
 					/*
 					 * Don't check read_mems_allowed_retry()
 					 * here - if mems_allowed was updated in
@@ -2383,7 +2386,7 @@ static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
 					 * between allocation and the cpuset
 					 * update
 					 */
-					return object;
+					return slab;
 				}
 			}
 		}
@@ -2395,17 +2398,18 @@ static void *get_any_partial(struct kmem_cache *s, struct partial_context *pc)
 /*
  * Get a partial slab, lock it and return it.
  */
-static void *get_partial(struct kmem_cache *s, int node, struct partial_context *pc)
+static struct slab *get_partial(struct kmem_cache *s, int node,
+				struct partial_context *pc)
 {
-	void *object;
+	struct slab *slab;
 	int searchnode = node;
 
 	if (node == NUMA_NO_NODE)
 		searchnode = numa_mem_id();
 
-	object = get_partial_node(s, get_node(s, searchnode), pc);
-	if (object || node != NUMA_NO_NODE)
-		return object;
+	slab = get_partial_node(s, get_node(s, searchnode), pc);
+	if (slab || node != NUMA_NO_NODE)
+		return slab;
 
 	return get_any_partial(s, pc);
 }
@@ -3215,10 +3219,10 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 new_objects:
 
 	pc.flags = gfpflags;
-	pc.slab = &slab;
 	pc.orig_size = orig_size;
-	freelist = get_partial(s, node, &pc);
-	if (freelist) {
+	slab = get_partial(s, node, &pc);
+	if (slab) {
+		freelist = pc.object;
 		if (kmem_cache_debug(s)) {
 			/*
 			 * For debug caches here we had to go through
@@ -3410,12 +3414,11 @@ static void *__slab_alloc_node(struct kmem_cache *s,
 	void *object;
 
 	pc.flags = gfpflags;
-	pc.slab = &slab;
 	pc.orig_size = orig_size;
-	object = get_partial(s, node, &pc);
+	slab = get_partial(s, node, &pc);
 
-	if (object)
-		return object;
+	if (slab)
+		return pc.object;
 
 	slab = new_slab(s, gfpflags, node);
 	if (unlikely(!slab)) {
-- 
2.40.1

[RFC PATCH v3 5/7] slub: Introduce freeze_slab()

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

We will have unfrozen slabs out of the node partial list later, so we
need a freeze_slab() function to freeze the partial slab and get its
freelist.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/mm/slub.c b/mm/slub.c
index 7d0234bffad3..5b428648021f 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3079,6 +3079,33 @@ static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
 	return freelist;
 }
 
+/*
+ * Freeze the partial slab and return the pointer to the freelist.
+ */
+static inline void *freeze_slab(struct kmem_cache *s, struct slab *slab)
+{
+	struct slab new;
+	unsigned long counters;
+	void *freelist;
+
+	do {
+		freelist = slab->freelist;
+		counters = slab->counters;
+
+		new.counters = counters;
+		VM_BUG_ON(new.frozen);
+
+		new.inuse = slab->objects;
+		new.frozen = 1;
+
+	} while (!__slab_update_freelist(s, slab,
+		freelist, counters,
+		NULL, new.counters,
+		"freeze_slab"));
+
+	return freelist;
+}
+
 /*
  * Slow path. The lockless freelist is empty or we need to perform
  * debugging duties.
-- 
2.40.1

[RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

Now we will freeze slabs when moving them out of node partial list to
cpu partial list, this method needs two cmpxchg_double operations:

1. freeze slab (acquire_slab()) under the node list_lock
2. get_freelist() when pick used in ___slab_alloc()

Actually we don't need to freeze when moving slabs out of node partial
list, we can delay freezing to when use slab freelist in ___slab_alloc(),
so we can save one cmpxchg_double().

And there are other good points:
 - The moving of slabs between node partial list and cpu partial list
   becomes simpler, since we don't need to freeze or unfreeze at all.

 - The node list_lock contention would be less, since we don't need to
   freeze any slab under the node list_lock.

We can achieve this because there is no concurrent path would manipulate
the partial slab list except the __slab_free() path, which is serialized
now.

Since the slab returned by get_partial() interfaces is not frozen anymore
and no freelist in the partial_context, so we need to use the introduced
freeze_slab() to freeze it and get its freelist.

Similarly, the slabs on the CPU partial list are not frozen anymore,
we need to freeze_slab() on it before use.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 111 +++++++++++-------------------------------------------
 1 file changed, 21 insertions(+), 90 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index 5b428648021f..486d44421432 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2215,51 +2215,6 @@ static void *alloc_single_from_new_slab(struct kmem_cache *s,
 	return object;
 }
 
-/*
- * Remove slab from the partial list, freeze it and
- * return the pointer to the freelist.
- *
- * Returns a list of objects or NULL if it fails.
- */
-static inline void *acquire_slab(struct kmem_cache *s,
-		struct kmem_cache_node *n, struct slab *slab,
-		int mode)
-{
-	void *freelist;
-	unsigned long counters;
-	struct slab new;
-
-	lockdep_assert_held(&n->list_lock);
-
-	/*
-	 * Zap the freelist and set the frozen bit.
-	 * The old freelist is the list of objects for the
-	 * per cpu allocation list.
-	 */
-	freelist = slab->freelist;
-	counters = slab->counters;
-	new.counters = counters;
-	if (mode) {
-		new.inuse = slab->objects;
-		new.freelist = NULL;
-	} else {
-		new.freelist = freelist;
-	}
-
-	VM_BUG_ON(new.frozen);
-	new.frozen = 1;
-
-	if (!__slab_update_freelist(s, slab,
-			freelist, counters,
-			new.freelist, new.counters,
-			"acquire_slab"))
-		return NULL;
-
-	remove_partial(n, slab);
-	WARN_ON(!freelist);
-	return freelist;
-}
-
 #ifdef CONFIG_SLUB_CPU_PARTIAL
 static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
 #else
@@ -2276,7 +2231,6 @@ static struct slab *get_partial_node(struct kmem_cache *s,
 				     struct partial_context *pc)
 {
 	struct slab *slab, *slab2, *partial = NULL;
-	void *object = NULL;
 	unsigned long flags;
 	unsigned int partial_slabs = 0;
 
@@ -2295,7 +2249,7 @@ static struct slab *get_partial_node(struct kmem_cache *s,
 			continue;
 
 		if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
-			object = alloc_single_from_partial(s, n, slab,
+			void *object = alloc_single_from_partial(s, n, slab,
 							pc->orig_size);
 			if (object) {
 				partial = slab;
@@ -2305,13 +2259,10 @@ static struct slab *get_partial_node(struct kmem_cache *s,
 			continue;
 		}
 
-		object = acquire_slab(s, n, slab, object == NULL);
-		if (!object)
-			break;
+		remove_partial(n, slab);
 
 		if (!partial) {
 			partial = slab;
-			pc->object = object;
 			stat(s, ALLOC_FROM_PARTIAL);
 		} else {
 			put_cpu_partial(s, slab, 0);
@@ -2610,9 +2561,6 @@ static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
 	unsigned long flags = 0;
 
 	while (partial_slab) {
-		struct slab new;
-		struct slab old;
-
 		slab = partial_slab;
 		partial_slab = slab->next;
 
@@ -2625,23 +2573,7 @@ static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
 			spin_lock_irqsave(&n->list_lock, flags);
 		}
 
-		do {
-
-			old.freelist = slab->freelist;
-			old.counters = slab->counters;
-			VM_BUG_ON(!old.frozen);
-
-			new.counters = old.counters;
-			new.freelist = old.freelist;
-
-			new.frozen = 0;
-
-		} while (!__slab_update_freelist(s, slab,
-				old.freelist, old.counters,
-				new.freelist, new.counters,
-				"unfreezing slab"));
-
-		if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
+		if (unlikely(!slab->inuse && n->nr_partial >= s->min_partial)) {
 			slab->next = slab_to_discard;
 			slab_to_discard = slab;
 		} else {
@@ -3148,7 +3080,6 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 			node = NUMA_NO_NODE;
 		goto new_slab;
 	}
-redo:
 
 	if (unlikely(!node_match(slab, node))) {
 		/*
@@ -3224,7 +3155,7 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 
 new_slab:
 
-	if (slub_percpu_partial(c)) {
+	while (slub_percpu_partial(c)) {
 		local_lock_irqsave(&s->cpu_slab->lock, flags);
 		if (unlikely(c->slab)) {
 			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
@@ -3236,11 +3167,20 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 			goto new_objects;
 		}
 
-		slab = c->slab = slub_percpu_partial(c);
+		slab = slub_percpu_partial(c);
 		slub_set_percpu_partial(c, slab);
 		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
 		stat(s, CPU_PARTIAL_ALLOC);
-		goto redo;
+
+		if (unlikely(!node_match(slab, node) ||
+			     !pfmemalloc_match(slab, gfpflags))) {
+			slab->next = NULL;
+			__unfreeze_partials(s, slab);
+			continue;
+		}
+
+		freelist = freeze_slab(s, slab);
+		goto retry_load_slab;
 	}
 
 new_objects:
@@ -3249,8 +3189,8 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 	pc.orig_size = orig_size;
 	slab = get_partial(s, node, &pc);
 	if (slab) {
-		freelist = pc.object;
 		if (kmem_cache_debug(s)) {
+			freelist = pc.object;
 			/*
 			 * For debug caches here we had to go through
 			 * alloc_single_from_partial() so just store the
@@ -3262,6 +3202,7 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
 			return freelist;
 		}
 
+		freelist = freeze_slab(s, slab);
 		goto retry_load_slab;
 	}
 
@@ -3663,18 +3604,8 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
 		was_frozen = new.frozen;
 		new.inuse -= cnt;
 		if ((!new.inuse || !prior) && !was_frozen) {
-
-			if (kmem_cache_has_cpu_partial(s) && !prior) {
-
-				/*
-				 * Slab was on no list before and will be
-				 * partially empty
-				 * We can defer the list move and instead
-				 * freeze it.
-				 */
-				new.frozen = 1;
-
-			} else { /* Needs to be taken off a list */
+			/* Needs to be taken off a list */
+			if (!kmem_cache_has_cpu_partial(s) || prior) {
 
 				n = get_node(s, slab_nid(slab));
 				/*
@@ -3704,9 +3635,9 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
 			 * activity can be necessary.
 			 */
 			stat(s, FREE_FROZEN);
-		} else if (new.frozen) {
+		} else if (kmem_cache_has_cpu_partial(s) && !prior) {
 			/*
-			 * If we just froze the slab then put it onto the
+			 * If we started with a full slab then put it onto the
 			 * per cpu partial list.
 			 */
 			put_cpu_partial(s, slab, 1);
-- 
2.40.1

[RFC PATCH v3 7/7] slub: Optimize deactivate_slab()

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

Since the introduce of unfrozen slabs on cpu partial list, we don't
need to synchronize the slab frozen state under the node list_lock.

The caller of deactivate_slab() and the caller of __slab_free() won't
manipulate the slab list concurrently.

So we can get node list_lock in the last stage if we really need to
manipulate the slab list in this path.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/slub.c | 70 ++++++++++++++++++++-----------------------------------
 1 file changed, 25 insertions(+), 45 deletions(-)

diff --git a/mm/slub.c b/mm/slub.c
index 486d44421432..64d550e415eb 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2449,10 +2449,8 @@ static void init_kmem_cache_cpus(struct kmem_cache *s)
 static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
 			    void *freelist)
 {
-	enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
 	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
 	int free_delta = 0;
-	enum slab_modes mode = M_NONE;
 	void *nextfree, *freelist_iter, *freelist_tail;
 	int tail = DEACTIVATE_TO_HEAD;
 	unsigned long flags = 0;
@@ -2499,58 +2497,40 @@ static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
 	 * unfrozen and number of objects in the slab may have changed.
 	 * Then release lock and retry cmpxchg again.
 	 */
-redo:
-
-	old.freelist = READ_ONCE(slab->freelist);
-	old.counters = READ_ONCE(slab->counters);
-	VM_BUG_ON(!old.frozen);
-
-	/* Determine target state of the slab */
-	new.counters = old.counters;
-	if (freelist_tail) {
-		new.inuse -= free_delta;
-		set_freepointer(s, freelist_tail, old.freelist);
-		new.freelist = freelist;
-	} else
-		new.freelist = old.freelist;
+	do {
+		old.freelist = READ_ONCE(slab->freelist);
+		old.counters = READ_ONCE(slab->counters);
+		VM_BUG_ON(!old.frozen);
+
+		/* Determine target state of the slab */
+		new.counters = old.counters;
+		new.frozen = 0;
+		if (freelist_tail) {
+			new.inuse -= free_delta;
+			set_freepointer(s, freelist_tail, old.freelist);
+			new.freelist = freelist;
+		} else
+			new.freelist = old.freelist;
 
-	new.frozen = 0;
+	} while (!slab_update_freelist(s, slab,
+		old.freelist, old.counters,
+		new.freelist, new.counters,
+		"unfreezing slab"));
 
+	/*
+	 * Stage three: Manipulate the slab list based on the updated state.
+	 */
 	if (!new.inuse && n->nr_partial >= s->min_partial) {
-		mode = M_FREE;
+		stat(s, DEACTIVATE_EMPTY);
+		discard_slab(s, slab);
+		stat(s, FREE_SLAB);
 	} else if (new.freelist) {
-		mode = M_PARTIAL;
-		/*
-		 * Taking the spinlock removes the possibility that
-		 * acquire_slab() will see a slab that is frozen
-		 */
 		spin_lock_irqsave(&n->list_lock, flags);
-	} else {
-		mode = M_FULL_NOLIST;
-	}
-
-
-	if (!slab_update_freelist(s, slab,
-				old.freelist, old.counters,
-				new.freelist, new.counters,
-				"unfreezing slab")) {
-		if (mode == M_PARTIAL)
-			spin_unlock_irqrestore(&n->list_lock, flags);
-		goto redo;
-	}
-
-
-	if (mode == M_PARTIAL) {
 		add_partial(n, slab, tail);
 		spin_unlock_irqrestore(&n->list_lock, flags);
 		stat(s, tail);
-	} else if (mode == M_FREE) {
-		stat(s, DEACTIVATE_EMPTY);
-		discard_slab(s, slab);
-		stat(s, FREE_SLAB);
-	} else if (mode == M_FULL_NOLIST) {
+	} else
 		stat(s, DEACTIVATE_FULL);
-	}
 }
 
 #ifdef CONFIG_SLUB_CPU_PARTIAL
-- 
2.40.1

Re: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

[Chengming Zhou]

On 2023/10/24 17:33, chengming.zhou@linux.dev wrote:
> From: Chengming Zhou <zhouchengming@bytedance.com>
> 
> Now we will freeze slabs when moving them out of node partial list to
> cpu partial list, this method needs two cmpxchg_double operations:
> 
> 1. freeze slab (acquire_slab()) under the node list_lock
> 2. get_freelist() when pick used in ___slab_alloc()
> 
> Actually we don't need to freeze when moving slabs out of node partial
> list, we can delay freezing to when use slab freelist in ___slab_alloc(),
> so we can save one cmpxchg_double().
> 
> And there are other good points:
>  - The moving of slabs between node partial list and cpu partial list
>    becomes simpler, since we don't need to freeze or unfreeze at all.
> 
>  - The node list_lock contention would be less, since we don't need to
>    freeze any slab under the node list_lock.
> 
> We can achieve this because there is no concurrent path would manipulate
> the partial slab list except the __slab_free() path, which is serialized
> now.
> 
> Since the slab returned by get_partial() interfaces is not frozen anymore
> and no freelist in the partial_context, so we need to use the introduced
> freeze_slab() to freeze it and get its freelist.
> 
> Similarly, the slabs on the CPU partial list are not frozen anymore,
> we need to freeze_slab() on it before use.
> 
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
> ---
>  mm/slub.c | 111 +++++++++++-------------------------------------------
>  1 file changed, 21 insertions(+), 90 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 5b428648021f..486d44421432 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -2215,51 +2215,6 @@ static void *alloc_single_from_new_slab(struct kmem_cache *s,
>  	return object;
>  }
>  
> -/*
> - * Remove slab from the partial list, freeze it and
> - * return the pointer to the freelist.
> - *
> - * Returns a list of objects or NULL if it fails.
> - */
> -static inline void *acquire_slab(struct kmem_cache *s,
> -		struct kmem_cache_node *n, struct slab *slab,
> -		int mode)
> -{
> -	void *freelist;
> -	unsigned long counters;
> -	struct slab new;
> -
> -	lockdep_assert_held(&n->list_lock);
> -
> -	/*
> -	 * Zap the freelist and set the frozen bit.
> -	 * The old freelist is the list of objects for the
> -	 * per cpu allocation list.
> -	 */
> -	freelist = slab->freelist;
> -	counters = slab->counters;
> -	new.counters = counters;
> -	if (mode) {
> -		new.inuse = slab->objects;
> -		new.freelist = NULL;
> -	} else {
> -		new.freelist = freelist;
> -	}
> -
> -	VM_BUG_ON(new.frozen);
> -	new.frozen = 1;
> -
> -	if (!__slab_update_freelist(s, slab,
> -			freelist, counters,
> -			new.freelist, new.counters,
> -			"acquire_slab"))
> -		return NULL;
> -
> -	remove_partial(n, slab);
> -	WARN_ON(!freelist);
> -	return freelist;
> -}
> -
>  #ifdef CONFIG_SLUB_CPU_PARTIAL
>  static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain);
>  #else
> @@ -2276,7 +2231,6 @@ static struct slab *get_partial_node(struct kmem_cache *s,
>  				     struct partial_context *pc)
>  {
>  	struct slab *slab, *slab2, *partial = NULL;
> -	void *object = NULL;
>  	unsigned long flags;
>  	unsigned int partial_slabs = 0;
>  
> @@ -2295,7 +2249,7 @@ static struct slab *get_partial_node(struct kmem_cache *s,
>  			continue;
>  
>  		if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) {
> -			object = alloc_single_from_partial(s, n, slab,
> +			void *object = alloc_single_from_partial(s, n, slab,
>  							pc->orig_size);
>  			if (object) {
>  				partial = slab;
> @@ -2305,13 +2259,10 @@ static struct slab *get_partial_node(struct kmem_cache *s,
>  			continue;
>  		}
>  
> -		object = acquire_slab(s, n, slab, object == NULL);
> -		if (!object)
> -			break;
> +		remove_partial(n, slab);
>  
>  		if (!partial) {
>  			partial = slab;
> -			pc->object = object;
>  			stat(s, ALLOC_FROM_PARTIAL);
>  		} else {
>  			put_cpu_partial(s, slab, 0);
> @@ -2610,9 +2561,6 @@ static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
>  	unsigned long flags = 0;
>  
>  	while (partial_slab) {
> -		struct slab new;
> -		struct slab old;
> -
>  		slab = partial_slab;
>  		partial_slab = slab->next;
>  
> @@ -2625,23 +2573,7 @@ static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab)
>  			spin_lock_irqsave(&n->list_lock, flags);
>  		}
>  
> -		do {
> -
> -			old.freelist = slab->freelist;
> -			old.counters = slab->counters;
> -			VM_BUG_ON(!old.frozen);
> -
> -			new.counters = old.counters;
> -			new.freelist = old.freelist;
> -
> -			new.frozen = 0;
> -
> -		} while (!__slab_update_freelist(s, slab,
> -				old.freelist, old.counters,
> -				new.freelist, new.counters,
> -				"unfreezing slab"));
> -
> -		if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
> +		if (unlikely(!slab->inuse && n->nr_partial >= s->min_partial)) {
>  			slab->next = slab_to_discard;
>  			slab_to_discard = slab;
>  		} else {
> @@ -3148,7 +3080,6 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
>  			node = NUMA_NO_NODE;
>  		goto new_slab;
>  	}
> -redo:
>  
>  	if (unlikely(!node_match(slab, node))) {
>  		/*
> @@ -3224,7 +3155,7 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
>  
>  new_slab:
>  
> -	if (slub_percpu_partial(c)) {
> +	while (slub_percpu_partial(c)) {
>  		local_lock_irqsave(&s->cpu_slab->lock, flags);
>  		if (unlikely(c->slab)) {
>  			local_unlock_irqrestore(&s->cpu_slab->lock, flags);
> @@ -3236,11 +3167,20 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
>  			goto new_objects;
>  		}
>  
> -		slab = c->slab = slub_percpu_partial(c);
> +		slab = slub_percpu_partial(c);
>  		slub_set_percpu_partial(c, slab);
>  		local_unlock_irqrestore(&s->cpu_slab->lock, flags);
>  		stat(s, CPU_PARTIAL_ALLOC);
> -		goto redo;
> +
> +		if (unlikely(!node_match(slab, node) ||
> +			     !pfmemalloc_match(slab, gfpflags))) {
> +			slab->next = NULL;
> +			__unfreeze_partials(s, slab);
> +			continue;
> +		}
> +
> +		freelist = freeze_slab(s, slab);
> +		goto retry_load_slab;
>  	}

Oops, this while(slub_percpu_partial(c)) loop block should be put in #ifdef CONFIG_SLUB_CPU_PARTIAL,
since the slab->next and __unfreeze_partials() only defined when CONFIG_SLUB_CPU_PARTIAL.

And I should append a cleanup patch to rename all *unfreeze_partials* functions to *put_partials*
since there is no "unfreeze" in these functions anymore.

Will do in the next version.

Thanks.

>  
>  new_objects:
> @@ -3249,8 +3189,8 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
>  	pc.orig_size = orig_size;
>  	slab = get_partial(s, node, &pc);
>  	if (slab) {
> -		freelist = pc.object;
>  		if (kmem_cache_debug(s)) {
> +			freelist = pc.object;
>  			/*
>  			 * For debug caches here we had to go through
>  			 * alloc_single_from_partial() so just store the
> @@ -3262,6 +3202,7 @@ static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
>  			return freelist;
>  		}
>  
> +		freelist = freeze_slab(s, slab);
>  		goto retry_load_slab;
>  	}
>  
> @@ -3663,18 +3604,8 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
>  		was_frozen = new.frozen;
>  		new.inuse -= cnt;
>  		if ((!new.inuse || !prior) && !was_frozen) {
> -
> -			if (kmem_cache_has_cpu_partial(s) && !prior) {
> -
> -				/*
> -				 * Slab was on no list before and will be
> -				 * partially empty
> -				 * We can defer the list move and instead
> -				 * freeze it.
> -				 */
> -				new.frozen = 1;
> -
> -			} else { /* Needs to be taken off a list */
> +			/* Needs to be taken off a list */
> +			if (!kmem_cache_has_cpu_partial(s) || prior) {
>  
>  				n = get_node(s, slab_nid(slab));
>  				/*
> @@ -3704,9 +3635,9 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
>  			 * activity can be necessary.
>  			 */
>  			stat(s, FREE_FROZEN);
> -		} else if (new.frozen) {
> +		} else if (kmem_cache_has_cpu_partial(s) && !prior) {
>  			/*
> -			 * If we just froze the slab then put it onto the
> +			 * If we started with a full slab then put it onto the
>  			 * per cpu partial list.
>  			 */
>  			put_cpu_partial(s, slab, 1);

Re: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

[kernel test robot]

Hello,

kernel test robot noticed "WARNING:at_mm/slub.c:#___slab_alloc" on:

commit: b34342e2732b0dc92b29d6807c5314e2e5e0c27c ("[RFC PATCH v3 6/7] slub: Delay freezing of partial slabs")
url: https://github.com/intel-lab-lkp/linux/commits/chengming-zhou-linux-dev/slub-Keep-track-of-whether-slub-is-on-the-per-node-partial-list/20231024-173519
base: git://git.kernel.org/cgit/linux/kernel/git/vbabka/slab.git for-next
patch link: https://lore.kernel.org/all/20231024093345.3676493-7-chengming.zhou@linux.dev/
patch subject: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

in testcase: rcutorture
version: 
with following parameters:

	runtime: 300s
	test: default
	torture_type: rcu



compiler: gcc-12
test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G

(please refer to attached dmesg/kmsg for entire log/backtrace)


+------------------------------------------------+------------+------------+
|                                                | e87d1d9973 | b34342e273 |
+------------------------------------------------+------------+------------+
| boot_successes                                 | 15         | 0          |
| boot_failures                                  | 0          | 15         |
| WARNING:at_mm/slub.c:#___slab_alloc            | 0          | 15         |
| RIP:___slab_alloc                              | 0          | 15         |
+------------------------------------------------+------------+------------+


If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <oliver.sang@intel.com>
| Closes: https://lore.kernel.org/oe-lkp/202310261032.b4590eb0-oliver.sang@intel.com


[    4.907500][    T1] ------------[ cut here ]------------
[ 4.908232][ T1] WARNING: CPU: 0 PID: 1 at mm/slub.c:577 ___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 
[    4.909242][    T1] Modules linked in:
[    4.909739][    T1] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.6.0-rc5-00013-gb34342e2732b #1
[ 4.910746][ T1] RIP: 0010:___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 
[ 4.911433][ T1] Code: 9f 05 66 25 ff 7f 66 89 45 b8 48 8b 4d b8 45 85 e4 74 1a 65 8b 05 dd c4 a3 7e 85 c0 75 0f 65 8b 05 de c1 87 7e 85 c0 74 04 90 <0f> 0b 90 81 e6 00 00 00 40 74 31 4c 89 f8 f0 49 0f c7 48 20 0f 84
All code
========
   0:	9f                   	lahf
   1:	05 66 25 ff 7f       	add    $0x7fff2566,%eax
   6:	66 89 45 b8          	mov    %ax,-0x48(%rbp)
   a:	48 8b 4d b8          	mov    -0x48(%rbp),%rcx
   e:	45 85 e4             	test   %r12d,%r12d
  11:	74 1a                	je     0x2d
  13:	65 8b 05 dd c4 a3 7e 	mov    %gs:0x7ea3c4dd(%rip),%eax        # 0x7ea3c4f7
  1a:	85 c0                	test   %eax,%eax
  1c:	75 0f                	jne    0x2d
  1e:	65 8b 05 de c1 87 7e 	mov    %gs:0x7e87c1de(%rip),%eax        # 0x7e87c203
  25:	85 c0                	test   %eax,%eax
  27:	74 04                	je     0x2d
  29:	90                   	nop
  2a:*	0f 0b                	ud2		<-- trapping instruction
  2c:	90                   	nop
  2d:	81 e6 00 00 00 40    	and    $0x40000000,%esi
  33:	74 31                	je     0x66
  35:	4c 89 f8             	mov    %r15,%rax
  38:	f0 49 0f c7 48 20    	lock cmpxchg16b 0x20(%r8)
  3e:	0f                   	.byte 0xf
  3f:	84                   	.byte 0x84

Code starting with the faulting instruction
===========================================
   0:	0f 0b                	ud2
   2:	90                   	nop
   3:	81 e6 00 00 00 40    	and    $0x40000000,%esi
   9:	74 31                	je     0x3c
   b:	4c 89 f8             	mov    %r15,%rax
   e:	f0 49 0f c7 48 20    	lock cmpxchg16b 0x20(%r8)
  14:	0f                   	.byte 0xf
  15:	84                   	.byte 0x84
[    4.913822][    T1] RSP: 0000:ffffc9000001f830 EFLAGS: 00010202
[    4.914602][    T1] RAX: 0000000000000001 RBX: 0000000000000000 RCX: 0000000080270027
[    4.914743][    T1] RDX: 0000000000270021 RSI: 0000000048000000 RDI: ffffffff842e066b
[    4.915745][    T1] RBP: ffffc9000001f8e0 R08: ffffea0005931d40 R09: fffffbfff0e33a5a
[    4.916743][    T1] R10: ffffffff8719d2d7 R11: 0000000000000000 R12: 0000000000000001
[    4.917696][    T1] R13: ffff8883ae809430 R14: ffff88810033e3c0 R15: ffff888164c75dd0
[    4.918688][    T1] FS:  0000000000000000(0000) GS:ffff8883ae600000(0000) knlGS:0000000000000000
[    4.918754][    T1] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    4.919522][    T1] CR2: ffff88843ffff000 CR3: 0000000005c89000 CR4: 00000000000406f0
[    4.920500][    T1] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    4.921496][    T1] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[    4.922461][    T1] Call Trace:
[    4.922742][    T1]  <TASK>
[ 4.923107][ T1] ? __warn (kernel/panic.c:673) 
[ 4.923632][ T1] ? ___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 
[ 4.924250][ T1] ? report_bug (lib/bug.c:180 lib/bug.c:219) 
[ 4.924831][ T1] ? handle_bug (arch/x86/kernel/traps.c:237) 
[ 4.925376][ T1] ? exc_invalid_op (arch/x86/kernel/traps.c:258 (discriminator 1)) 
[ 4.925962][ T1] ? asm_exc_invalid_op (arch/x86/include/asm/idtentry.h:568) 
[ 4.926595][ T1] ? _raw_spin_unlock_irqrestore (arch/x86/include/asm/irqflags.h:42 arch/x86/include/asm/irqflags.h:77 arch/x86/include/asm/irqflags.h:135 include/linux/spinlock_api_smp.h:151 kernel/locking/spinlock.c:194) 
[ 4.926741][ T1] ? ___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 
[ 4.927370][ T1] ? acpi_ut_allocate_object_desc_dbg (include/linux/slab.h:710 include/acpi/platform/aclinuxex.h:67 drivers/acpi/acpica/utobject.c:359) 
[ 4.928186][ T1] ? kmem_cache_alloc (mm/slub.c:3295 mm/slub.c:3348 mm/slub.c:3440 mm/slub.c:3458 mm/slub.c:3465 mm/slub.c:3474) 
[ 4.928824][ T1] kmem_cache_alloc (mm/slub.c:3295 mm/slub.c:3348 mm/slub.c:3440 mm/slub.c:3458 mm/slub.c:3465 mm/slub.c:3474) 
[ 4.929442][ T1] ? acpi_ut_allocate_object_desc_dbg (include/linux/slab.h:710 include/acpi/platform/aclinuxex.h:67 drivers/acpi/acpica/utobject.c:359) 
[ 4.930221][ T1] acpi_ut_allocate_object_desc_dbg (include/linux/slab.h:710 include/acpi/platform/aclinuxex.h:67 drivers/acpi/acpica/utobject.c:359) 
[ 4.930743][ T1] acpi_ut_create_internal_object_dbg (drivers/acpi/acpica/utobject.c:71) 
[ 4.931576][ T1] acpi_ut_copy_esimple_to_isimple (drivers/acpi/acpica/utcopy.c:434) 
[ 4.932394][ T1] acpi_ut_copy_eobject_to_iobject (drivers/acpi/acpica/utcopy.c:618) 
[ 4.933185][ T1] ? __kasan_kmalloc (mm/kasan/common.c:374 mm/kasan/common.c:383) 
[ 4.933816][ T1] acpi_evaluate_object (drivers/acpi/acpica/nsxfeval.c:259) 
[ 4.934486][ T1] ? acpi_get_data_full (drivers/acpi/acpica/nsxfeval.c:167) 
[ 4.934756][ T1] ? hlock_class (arch/x86/include/asm/bitops.h:228 arch/x86/include/asm/bitops.h:240 include/asm-generic/bitops/instrumented-non-atomic.h:142 kernel/locking/lockdep.c:228) 
[ 4.935345][ T1] ? __uuid_parse+0xd0/0x1b0 
[ 4.936024][ T1] acpi_run_osc (drivers/acpi/bus.c:217) 
[ 4.936609][ T1] ? acpi_bus_detach_private_data (drivers/acpi/bus.c:187) 
[ 4.937363][ T1] ? _raw_spin_unlock_irqrestore (arch/x86/include/asm/irqflags.h:42 arch/x86/include/asm/irqflags.h:77 arch/x86/include/asm/irqflags.h:135 include/linux/spinlock_api_smp.h:151 kernel/locking/spinlock.c:194) 
[ 4.938127][ T1] ? acpi_get_data (drivers/acpi/acpica/nsxfname.c:48) 
[ 4.938710][ T1] acpi_bus_init (drivers/acpi/bus.c:352 drivers/acpi/bus.c:1329) 
[ 4.938742][ T1] ? up (include/linux/list.h:373 kernel/locking/semaphore.c:188) 
[ 4.939205][ T1] ? acpi_sleep_proc_init (drivers/acpi/bus.c:1289) 
[ 4.939862][ T1] ? _raw_spin_unlock_irqrestore (arch/x86/include/asm/irqflags.h:42 arch/x86/include/asm/irqflags.h:77 arch/x86/include/asm/irqflags.h:135 include/linux/spinlock_api_smp.h:151 kernel/locking/spinlock.c:194) 
[ 4.940581][ T1] ? acpi_os_signal_semaphore (drivers/acpi/osl.c:1294 (discriminator 5)) 
[ 4.941313][ T1] ? acpi_ut_release_mutex (drivers/acpi/acpica/utmutex.c:329) 
[ 4.942007][ T1] ? acpi_install_address_space_handler_internal (drivers/acpi/acpica/evxfregn.c:94) 
[ 4.942744][ T1] ? acpi_bus_init (drivers/acpi/bus.c:1390) 
[ 4.943325][ T1] acpi_init (drivers/acpi/bus.c:1404) 
[ 4.943832][ T1] ? acpi_bus_init (drivers/acpi/bus.c:1390) 
[ 4.944420][ T1] ? kb3886_init (drivers/video/fbdev/core/fbmem.c:1111) 
[ 4.944976][ T1] ? acpi_bus_init (drivers/acpi/bus.c:1390) 
[ 4.945579][ T1] do_one_initcall (init/main.c:1232) 
[ 4.946157][ T1] ? trace_event_raw_event_initcall_level (init/main.c:1223) 
[ 4.946747][ T1] ? kasan_set_track (mm/kasan/common.c:52) 
[ 4.947373][ T1] ? __kasan_kmalloc (mm/kasan/common.c:374 mm/kasan/common.c:383) 
[ 4.948010][ T1] do_initcalls (init/main.c:1293 init/main.c:1310) 
[ 4.948614][ T1] kernel_init_freeable (init/main.c:1549) 
[ 4.949315][ T1] ? rest_init (init/main.c:1429) 
[ 4.949894][ T1] kernel_init (init/main.c:1439) 
[ 4.950743][ T1] ? _raw_spin_unlock_irq (arch/x86/include/asm/irqflags.h:42 arch/x86/include/asm/irqflags.h:77 include/linux/spinlock_api_smp.h:159 kernel/locking/spinlock.c:202) 
[ 4.951386][ T1] ret_from_fork (arch/x86/kernel/process.c:153) 
[ 4.951957][ T1] ? rest_init (init/main.c:1429) 
[ 4.952520][ T1] ret_from_fork_asm (arch/x86/entry/entry_64.S:312) 
[    4.953143][    T1]  </TASK>
[    4.953514][    T1] irq event stamp: 85121
[ 4.954040][ T1] hardirqs last enabled at (85129): console_unlock (arch/x86/include/asm/irqflags.h:42 arch/x86/include/asm/irqflags.h:77 arch/x86/include/asm/irqflags.h:135 kernel/printk/printk.c:347 kernel/printk/printk.c:2718 kernel/printk/printk.c:3037) 
[ 4.954736][ T1] hardirqs last disabled at (85138): console_unlock (kernel/printk/printk.c:345 kernel/printk/printk.c:2718 kernel/printk/printk.c:3037) 
[ 4.955884][ T1] softirqs last enabled at (84862): __do_softirq (arch/x86/include/asm/preempt.h:27 kernel/softirq.c:400 kernel/softirq.c:582) 
[ 4.957017][ T1] softirqs last disabled at (84857): irq_exit_rcu (kernel/softirq.c:427 kernel/softirq.c:632 kernel/softirq.c:644) 
[    4.958128][    T1] ---[ end trace 0000000000000000 ]---


The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20231026/202310261032.b4590eb0-oliver.sang@intel.com



-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

[Chengming Zhou]

On 2023/10/26 13:49, kernel test robot wrote:
> 
> 
> Hello,
> 
> kernel test robot noticed "WARNING:at_mm/slub.c:#___slab_alloc" on:
> 
> commit: b34342e2732b0dc92b29d6807c5314e2e5e0c27c ("[RFC PATCH v3 6/7] slub: Delay freezing of partial slabs")
> url: https://github.com/intel-lab-lkp/linux/commits/chengming-zhou-linux-dev/slub-Keep-track-of-whether-slub-is-on-the-per-node-partial-list/20231024-173519
> base: git://git.kernel.org/cgit/linux/kernel/git/vbabka/slab.git for-next
> patch link: https://lore.kernel.org/all/20231024093345.3676493-7-chengming.zhou@linux.dev/
> patch subject: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs
> 
> in testcase: rcutorture
> version: 
> with following parameters:
> 
> 	runtime: 300s
> 	test: default
> 	torture_type: rcu
> 
> 
> 
> compiler: gcc-12
> test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
> 
> (please refer to attached dmesg/kmsg for entire log/backtrace)
> 
> 
> +------------------------------------------------+------------+------------+
> |                                                | e87d1d9973 | b34342e273 |
> +------------------------------------------------+------------+------------+
> | boot_successes                                 | 15         | 0          |
> | boot_failures                                  | 0          | 15         |
> | WARNING:at_mm/slub.c:#___slab_alloc            | 0          | 15         |
> | RIP:___slab_alloc                              | 0          | 15         |
> +------------------------------------------------+------------+------------+
> 
> 
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <oliver.sang@intel.com>
> | Closes: https://lore.kernel.org/oe-lkp/202310261032.b4590eb0-oliver.sang@intel.com
> 
> 
> [    4.907500][    T1] ------------[ cut here ]------------
> [ 4.908232][ T1] WARNING: CPU: 0 PID: 1 at mm/slub.c:577 ___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 
> [    4.909242][    T1] Modules linked in:
> [    4.909739][    T1] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 6.6.0-rc5-00013-gb34342e2732b #1
> [ 4.910746][ T1] RIP: 0010:___slab_alloc (mm/slub.c:577 mm/slub.c:3033 mm/slub.c:3205) 

The warning is triggered by "lockdep_assert_irqs_disabled()" in __slab_update_freelist(),
which is called in the new introduced freeze_slab().

We should fix it by using "slab_update_freelist()" in freeze_slab() instead, which will
disable the interrupts correctly.

Thanks!

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Christoph Lameter]

On Tue, 24 Oct 2023, chengming.zhou@linux.dev wrote:

> 2. Solution
> ===========
> We solve these problems by leaving slabs unfrozen when moving out of
> the node partial list and on CPU partial list, so "frozen" bit is 0.
>
> These partial slabs won't be manipulate concurrently by alloc path,
> the only racer is free path, which may manipulate its list when !inuse.
> So we need to introduce another synchronization way to avoid it, we
> reuse PG_workingset to keep track of whether the slab is on node partial
> list or not, only in that case we can manipulate the slab list.
>
> The slab will be delay frozen when it's picked to actively use by the
> CPU, it becomes full at the same time, in which case we still need to
> rely on "frozen" bit to avoid manipulating its list. So the slab will
> be frozen only when activate use and be unfrozen only when deactivate.

I think we have to clear our terminology a bit about what a "frozen" slab 
is.

Before this patch a frozen slab is not on the node partial list and 
therefore its state on the list does not have to be considered during 
freeing and other operations. The frozen slab could be actively allocated 
from.

From the source:

*   Frozen slabs
  *
  *   If a slab is frozen then it is exempt from list management. It is not
  *   on any list except per cpu partial list. The processor that froze the
  *   slab is the one who can perform list operations on the slab. Other
  *   processors may put objects onto the freelist but the processor that
  *   froze the slab is the only one that can retrieve the objects from the
  *   slab's freelist.
  *


After this patch the PG_workingset indicates the state of being on 
the partial lists.

What does "frozen slab" then mean? The slab is being allocated from? Is 
that information useful or can we drop the frozen flag?

Update the definition?

Re: [RFC PATCH v3 1/7] slub: Keep track of whether slub is on the per-node partial list

[Chengming Zhou]

On 2023/10/27 23:09, Vlastimil Babka wrote:
> On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
>> From: Chengming Zhou <zhouchengming@bytedance.com>
>>
>> Now we rely on the "frozen" bit to see if we should manipulate the
>> slab->slab_list, which will be changed in the following patch.
>>
>> Instead we introduce another way to keep track of whether slub is on
>> the per-node partial list, here we reuse the PG_workingset bit.
>>
>> We use __set_bit and __clear_bit directly instead of the atomic version
>> for better performance and it's safe since it's protected by the slub
>> node list_lock.
>>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
>> ---
>>  mm/slab.h | 19 +++++++++++++++++++
>>  mm/slub.c |  3 +++
>>  2 files changed, 22 insertions(+)
>>
>> diff --git a/mm/slab.h b/mm/slab.h
>> index 8cd3294fedf5..50522b688cfb 100644
>> --- a/mm/slab.h
>> +++ b/mm/slab.h
> 
> I think these helper might just go to mm/slub.c as nothing else would use them.
> 

Ok.

>> @@ -193,6 +193,25 @@ static inline void __slab_clear_pfmemalloc(struct slab *slab)
>>  	__folio_clear_active(slab_folio(slab));
>>  }
>>  
>> +/*
>> + * Slub reuse PG_workingset bit to keep track of whether it's on
> 
> "SLUB reuses" ...
> 
> Looks fine otherwise!
> 

Will fix it. Thanks!

Re: [RFC PATCH v3 2/7] slub: Prepare __slab_free() for unfrozen partial slab out of node partial list

[Chengming Zhou]

On 2023/10/27 23:18, Vlastimil Babka wrote:
> On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
>> From: Chengming Zhou <zhouchengming@bytedance.com>
>>
>> Now the partial slub will be frozen when taken out of node partial list,
> 
> 	  partially empty slab
> 
>> so the __slab_free() will know from "was_frozen" that the partial slab
>> is not on node partial list and is used by one kmem_cache_cpu.
> 
> 				... is a cpu or cpu partial slab of some cpu.
> 
>> But we will change this, make partial slabs leave the node partial list
>> with unfrozen state, so we need to change __slab_free() to use the new
>> slab_test_node_partial() we just introduced.
>>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
>> ---
>>  mm/slub.c | 11 +++++++++++
>>  1 file changed, 11 insertions(+)
>>
>> diff --git a/mm/slub.c b/mm/slub.c
>> index 3fad4edca34b..f568a32d7332 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -3610,6 +3610,7 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
>>  	unsigned long counters;
>>  	struct kmem_cache_node *n = NULL;
>>  	unsigned long flags;
>> +	bool on_node_partial;
>>  
>>  	stat(s, FREE_SLOWPATH);
>>  
>> @@ -3657,6 +3658,7 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
>>  				 */
>>  				spin_lock_irqsave(&n->list_lock, flags);
>>  
>> +				on_node_partial = slab_test_node_partial(slab);
>>  			}
>>  		}
>>  
>> @@ -3685,6 +3687,15 @@ static void __slab_free(struct kmem_cache *s, struct slab *slab,
>>  		return;
>>  	}
>>  
>> +	/*
>> +	 * This slab was partial but not on the per-node partial list,
> 
> This slab was partially empty ...
> 
> Otherwise LGTM!

Ok, will fix.

Thanks!

> 
>> +	 * in which case we shouldn't manipulate its list, just return.
>> +	 */
>> +	if (prior && !on_node_partial) {
>> +		spin_unlock_irqrestore(&n->list_lock, flags);
>> +		return;
>> +	}
>> +
>>  	if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
>>  		goto slab_empty;
>>  
> 

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Chengming Zhou]

On 2023/10/28 01:57, Christoph Lameter wrote:
> On Tue, 24 Oct 2023, chengming.zhou@linux.dev wrote:
> 
>> 2. Solution
>> ===========
>> We solve these problems by leaving slabs unfrozen when moving out of
>> the node partial list and on CPU partial list, so "frozen" bit is 0.
>>
>> These partial slabs won't be manipulate concurrently by alloc path,
>> the only racer is free path, which may manipulate its list when !inuse.
>> So we need to introduce another synchronization way to avoid it, we
>> reuse PG_workingset to keep track of whether the slab is on node partial
>> list or not, only in that case we can manipulate the slab list.
>>
>> The slab will be delay frozen when it's picked to actively use by the
>> CPU, it becomes full at the same time, in which case we still need to
>> rely on "frozen" bit to avoid manipulating its list. So the slab will
>> be frozen only when activate use and be unfrozen only when deactivate.
> 
> I think we have to clear our terminology a bit about what a "frozen" slab is.

Yes, we need to clean up these inconsistent documentations in the source.

> 
> Before this patch a frozen slab is not on the node partial list and therefore its state on the list does not have to be considered during freeing and other operations. The frozen slab could be actively allocated from.
> 
> From the source:
> 
> *   Frozen slabs
>  *
>  *   If a slab is frozen then it is exempt from list management. It is not
>  *   on any list except per cpu partial list. The processor that froze the

~~ except per cpu partial list ~~

Frozen slab is not on any list, it's actively allocated from by the processor
that froze it. IOW, frozen slab is the cpu slab.

>  *   slab is the one who can perform list operations on the slab. Other
>  *   processors may put objects onto the freelist but the processor that
>  *   froze the slab is the only one that can retrieve the objects from the
>  *   slab's freelist.
>  *

This part I think is unchanged.

> 
> 
> After this patch the PG_workingset indicates the state of being on the partial lists.
> 
> What does "frozen slab" then mean? The slab is being allocated from? Is that information useful or can we drop the frozen flag?

Right, frozen slab is the cpu slab, which is being allocated from by the cpu that froze it.

IMHO, the "frozen" bit is useful because:

1. PG_workingset is only useful on partial slab, which indicates the slab is on the node
   partial list, so we can manipulate its list in the __slab_free() path.

2. But for full slab (slab->freelist == NULL), PG_workingset is not much useful, we don't
   safely know whether it's used as the cpu slab or not just from this flag. So __slab_free()
   still rely on the "frozen" bit to know it.

3. And the maintaining of "frozen" has no extra cost now, since it's changed together with "freelist"
   and other counter using cmpxchg, we already have the cmpxchg when start to use a slab as the cpu slab.

Maybe I missed something, I don't know how to drop the frozen flag.

> 
> Update the definition?
> 

Ok, will add a cleanup patch to update.

Thanks!

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Vlastimil Babka]

On 10/28/23 04:36, Chengming Zhou wrote:
>> 
>> 
>> After this patch the PG_workingset indicates the state of being on the partial lists.
>> 
>> What does "frozen slab" then mean? The slab is being allocated from? Is that information useful or can we drop the frozen flag?
> 
> Right, frozen slab is the cpu slab, which is being allocated from by the cpu that froze it.
> 
> IMHO, the "frozen" bit is useful because:
> 
> 1. PG_workingset is only useful on partial slab, which indicates the slab is on the node
>    partial list, so we can manipulate its list in the __slab_free() path.
> 
> 2. But for full slab (slab->freelist == NULL), PG_workingset is not much useful, we don't
>    safely know whether it's used as the cpu slab or not just from this flag. So __slab_free()
>    still rely on the "frozen" bit to know it.

Well, we could extend the meaning of PG_workingset to mean "not a cpu slab
or pecpu partial slab" i.e. both on node partial list and full. However it
would increase the number of cases where __slab_free() has to lock the
list_lock and check the PG_working set. "slab->freelist == NULL" might
happen often exactly because the freelist became cpu freelist.

> 3. And the maintaining of "frozen" has no extra cost now, since it's changed together with "freelist"
>    and other counter using cmpxchg, we already have the cmpxchg when start to use a slab as the cpu slab.

And together with this point, I don't see a reason to drop the frozen bit.
It's still useful for cpu slabs. It just wasn't the best possible solution
for percpu partial slabs.

> Maybe I missed something, I don't know how to drop the frozen flag.

Should be possible, but not worth it IMHO.

>> 
>> Update the definition?
>> 
> 
> Ok, will add a cleanup patch to update.
> 
> Thanks!

Re: [RFC PATCH v3 4/7] slub: Change get_partial() interfaces to return slab

[Vlastimil Babka]

On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
> From: Chengming Zhou <zhouchengming@bytedance.com>
> 
> We need all get_partial() related interfaces to return a slab, instead
> of returning the freelist (or object).
> 
> Use the partial_context.object to return back freelist or object for
> now. This patch shouldn't have any functional changes.
> 
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>

Reviewed-by: Vlastimil Babka <vbabka@suse.cz>

I think you could even move patches 3/7 and 4/7 to the front of the series,
as cleanups that are useful on their own.

Thanks!

Re: [RFC PATCH v3 5/7] slub: Introduce freeze_slab()

[Vlastimil Babka]

On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
> From: Chengming Zhou <zhouchengming@bytedance.com>
> 
> We will have unfrozen slabs out of the node partial list later, so we
> need a freeze_slab() function to freeze the partial slab and get its
> freelist.
> 
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>

As you noted, we'll need slab_update_freelist().
Otherwise,

Reviewed-by: Vlastimil Babka <vbabka@suse.cz>

> ---
>  mm/slub.c | 27 +++++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 7d0234bffad3..5b428648021f 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -3079,6 +3079,33 @@ static inline void *get_freelist(struct kmem_cache *s, struct slab *slab)
>  	return freelist;
>  }
>  
> +/*
> + * Freeze the partial slab and return the pointer to the freelist.
> + */
> +static inline void *freeze_slab(struct kmem_cache *s, struct slab *slab)
> +{
> +	struct slab new;
> +	unsigned long counters;
> +	void *freelist;
> +
> +	do {
> +		freelist = slab->freelist;
> +		counters = slab->counters;
> +
> +		new.counters = counters;
> +		VM_BUG_ON(new.frozen);
> +
> +		new.inuse = slab->objects;
> +		new.frozen = 1;
> +
> +	} while (!__slab_update_freelist(s, slab,
> +		freelist, counters,
> +		NULL, new.counters,
> +		"freeze_slab"));
> +
> +	return freelist;
> +}
> +
>  /*
>   * Slow path. The lockless freelist is empty or we need to perform
>   * debugging duties.

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Christoph Lameter]

On Sat, 28 Oct 2023, Chengming Zhou wrote:

> 2. But for full slab (slab->freelist == NULL), PG_workingset is not much useful, we don't
>   safely know whether it's used as the cpu slab or not just from this flag. So __slab_free()
>   still rely on the "frozen" bit to know it.
>
> 3. And the maintaining of "frozen" has no extra cost now, since it's changed together with "freelist"
>   and other counter using cmpxchg, we already have the cmpxchg when start to use a slab as the cpu slab.
>
> Maybe I missed something, I don't know how to drop the frozen flag.


Maybe frozen is now = PG_Workingset | cmpxchg-frozen?

Re: [RFC PATCH v3 4/7] slub: Change get_partial() interfaces to return slab

[Chengming Zhou]

On 2023/10/31 00:55, Vlastimil Babka wrote:
> On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
>> From: Chengming Zhou <zhouchengming@bytedance.com>
>>
>> We need all get_partial() related interfaces to return a slab, instead
>> of returning the freelist (or object).
>>
>> Use the partial_context.object to return back freelist or object for
>> now. This patch shouldn't have any functional changes.
>>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
> 
> Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
> 
> I think you could even move patches 3/7 and 4/7 to the front of the series,
> as cleanups that are useful on their own.
> 

Right, I will move these two to the front in the next version.

Thanks!

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Chengming Zhou]

On 2023/10/31 00:19, Vlastimil Babka wrote:
> On 10/28/23 04:36, Chengming Zhou wrote:
>>>
>>>
>>> After this patch the PG_workingset indicates the state of being on the partial lists.
>>>
>>> What does "frozen slab" then mean? The slab is being allocated from? Is that information useful or can we drop the frozen flag?
>>
>> Right, frozen slab is the cpu slab, which is being allocated from by the cpu that froze it.
>>
>> IMHO, the "frozen" bit is useful because:
>>
>> 1. PG_workingset is only useful on partial slab, which indicates the slab is on the node
>>    partial list, so we can manipulate its list in the __slab_free() path.
>>
>> 2. But for full slab (slab->freelist == NULL), PG_workingset is not much useful, we don't
>>    safely know whether it's used as the cpu slab or not just from this flag. So __slab_free()
>>    still rely on the "frozen" bit to know it.
> 
> Well, we could extend the meaning of PG_workingset to mean "not a cpu slab
> or pecpu partial slab" i.e. both on node partial list and full. However it
> would increase the number of cases where __slab_free() has to lock the
> list_lock and check the PG_working set. "slab->freelist == NULL" might
> happen often exactly because the freelist became cpu freelist.

Ah, right, it's possible to do like this.

> 
>> 3. And the maintaining of "frozen" has no extra cost now, since it's changed together with "freelist"
>>    and other counter using cmpxchg, we already have the cmpxchg when start to use a slab as the cpu slab.
> 
> And together with this point, I don't see a reason to drop the frozen bit.
> It's still useful for cpu slabs. It just wasn't the best possible solution
> for percpu partial slabs.
> 
>> Maybe I missed something, I don't know how to drop the frozen flag.
> 
> Should be possible, but not worth it IMHO.

Agree, we'll just keep "frozen" for the cpu slabs.

Thanks!

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Chengming Zhou]

On 2023/10/31 03:25, Christoph Lameter wrote:
> On Sat, 28 Oct 2023, Chengming Zhou wrote:
> 
>> 2. But for full slab (slab->freelist == NULL), PG_workingset is not much useful, we don't
>>   safely know whether it's used as the cpu slab or not just from this flag. So __slab_free()
>>   still rely on the "frozen" bit to know it.
>>
>> 3. And the maintaining of "frozen" has no extra cost now, since it's changed together with "freelist"
>>   and other counter using cmpxchg, we already have the cmpxchg when start to use a slab as the cpu slab.
>>
>> Maybe I missed something, I don't know how to drop the frozen flag.
> 
> 
> Maybe frozen is now = PG_Workingset | cmpxchg-frozen?
> 
> 

The current scheme (which this series implemented) is:

- node partial slabs: PG_Workingset (set or clear with per-node list_lock protection)
- cpu partial slabs: !PG_Workingset
- cpu slabs: !PG_Workingset && frozen (set or clear using cmpxchg together with freelist)
- full slabs: !PG_Workingset

As Vlastimil noted, it's possible to drop "frozen" bit for cpu slabs, but
we keep it for performance, since we don't need to grab node list_lock to
check whether PG_Workingset is set or not if the "frozen" bit is set.

Thanks!

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Christoph Lameter]

On Tue, 31 Oct 2023, Chengming Zhou wrote:

> The current scheme (which this series implemented) is:
>
> - node partial slabs: PG_Workingset (set or clear with per-node list_lock protection)
> - cpu partial slabs: !PG_Workingset

And then the frozen flag needs to be set. Otherwise slab_free() would 
conclude it is on a partial list?

> - cpu slabs: !PG_Workingset && frozen (set or clear using cmpxchg together with freelist)



> - full slabs: !PG_Workingset

And frozen is clear? Otherwise it is the same as a cpu partial slab.

> As Vlastimil noted, it's possible to drop "frozen" bit for cpu slabs, but
> we keep it for performance, since we don't need to grab node list_lock to
> check whether PG_Workingset is set or not if the "frozen" bit is set.
>
> Thanks!
>

Re: [RFC PATCH v3 0/7] slub: Delay freezing of CPU partial slabs

[Chengming Zhou]

On 2023/10/31 11:47, Christoph Lameter wrote:
> On Tue, 31 Oct 2023, Chengming Zhou wrote:
> 
>> The current scheme (which this series implemented) is:
>>
>> - node partial slabs: PG_Workingset (set or clear with per-node list_lock protection)
>> - cpu partial slabs: !PG_Workingset
> 
> And then the frozen flag needs to be set. Otherwise slab_free() would conclude it is on a partial list?
> 

- cpu partial slabs: !PG_Workingset && !frozen

Here comes the optimization that "frozen" is not set for the cpu partial slabs,
slab_free() will grab node list_lock then check by !PG_Workingset that it's not
on a node partial list.

>> - cpu slabs: !PG_Workingset && frozen (set or clear using cmpxchg together with freelist)
> 
> 
> 
>> - full slabs: !PG_Workingset
> 
> And frozen is clear? Otherwise it is the same as a cpu partial slab.
> 

Right, - full slabs: !PG_Workingset && !frozen

Yes, it's the same as a cpu partial slab from only these two flags, but
slab_free() also know whether it was full or not.

>> As Vlastimil noted, it's possible to drop "frozen" bit for cpu slabs, but
>> we keep it for performance, since we don't need to grab node list_lock to
>> check whether PG_Workingset is set or not if the "frozen" bit is set.
>>
>> Thanks!
>>

Re: [RFC PATCH v3 6/7] slub: Delay freezing of partial slabs

[Vlastimil Babka]

On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
> From: Chengming Zhou <zhouchengming@bytedance.com>
> 
> Now we will freeze slabs when moving them out of node partial list to
> cpu partial list, this method needs two cmpxchg_double operations:
> 
> 1. freeze slab (acquire_slab()) under the node list_lock
> 2. get_freelist() when pick used in ___slab_alloc()
> 
> Actually we don't need to freeze when moving slabs out of node partial
> list, we can delay freezing to when use slab freelist in ___slab_alloc(),
> so we can save one cmpxchg_double().
> 
> And there are other good points:
>  - The moving of slabs between node partial list and cpu partial list
>    becomes simpler, since we don't need to freeze or unfreeze at all.
> 
>  - The node list_lock contention would be less, since we don't need to
>    freeze any slab under the node list_lock.
> 
> We can achieve this because there is no concurrent path would manipulate
> the partial slab list except the __slab_free() path, which is serialized
> now.

"which is now serialized by slab_test_node_partial() under the list_lock." ?

> Since the slab returned by get_partial() interfaces is not frozen anymore
> and no freelist in the partial_context, so we need to use the introduced

		 ^ is returned in

> freeze_slab() to freeze it and get its freelist.
> 
> Similarly, the slabs on the CPU partial list are not frozen anymore,
> we need to freeze_slab() on it before use.

We can now delete acquire_slab() as it became unused.

> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>

With the fixup for CONFIG_SLUB_CPU_PARTIAL you mentioned,
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>

Also agreed with followup patch to rename unfreeze_partials().
Thanks!

Re: [RFC PATCH v3 7/7] slub: Optimize deactivate_slab()

[Vlastimil Babka]

On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
> From: Chengming Zhou <zhouchengming@bytedance.com>
> 
> Since the introduce of unfrozen slabs on cpu partial list, we don't
> need to synchronize the slab frozen state under the node list_lock.
> 
> The caller of deactivate_slab() and the caller of __slab_free() won't
> manipulate the slab list concurrently.
> 
> So we can get node list_lock in the last stage if we really need to
> manipulate the slab list in this path.
> 
> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>

Nice simplification!

> ---
>  mm/slub.c | 70 ++++++++++++++++++++-----------------------------------
>  1 file changed, 25 insertions(+), 45 deletions(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 486d44421432..64d550e415eb 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -2449,10 +2449,8 @@ static void init_kmem_cache_cpus(struct kmem_cache *s)
>  static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
>  			    void *freelist)
>  {
> -	enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
>  	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
>  	int free_delta = 0;
> -	enum slab_modes mode = M_NONE;
>  	void *nextfree, *freelist_iter, *freelist_tail;
>  	int tail = DEACTIVATE_TO_HEAD;
>  	unsigned long flags = 0;
> @@ -2499,58 +2497,40 @@ static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
>  	 * unfrozen and number of objects in the slab may have changed.
>  	 * Then release lock and retry cmpxchg again.
>  	 */

This comment above (including parts not visible here) should be updated as
there is no more list manipulation during unfreeze.

> -redo:
> -
> -	old.freelist = READ_ONCE(slab->freelist);
> -	old.counters = READ_ONCE(slab->counters);
> -	VM_BUG_ON(!old.frozen);
> -
> -	/* Determine target state of the slab */
> -	new.counters = old.counters;
> -	if (freelist_tail) {
> -		new.inuse -= free_delta;
> -		set_freepointer(s, freelist_tail, old.freelist);
> -		new.freelist = freelist;
> -	} else
> -		new.freelist = old.freelist;
> +	do {
> +		old.freelist = READ_ONCE(slab->freelist);
> +		old.counters = READ_ONCE(slab->counters);
> +		VM_BUG_ON(!old.frozen);
> +
> +		/* Determine target state of the slab */
> +		new.counters = old.counters;
> +		new.frozen = 0;
> +		if (freelist_tail) {
> +			new.inuse -= free_delta;
> +			set_freepointer(s, freelist_tail, old.freelist);
> +			new.freelist = freelist;
> +		} else
> +			new.freelist = old.freelist;

Per coding style we should have the else with { } even if it's one line, to
match the if branch. Since we touch the code that was previously violating
the style, we can fix up.

>  
> -	new.frozen = 0;
> +	} while (!slab_update_freelist(s, slab,
> +		old.freelist, old.counters,
> +		new.freelist, new.counters,
> +		"unfreezing slab"));
>  
> +	/*
> +	 * Stage three: Manipulate the slab list based on the updated state.
> +	 */
>  	if (!new.inuse && n->nr_partial >= s->min_partial) {
> -		mode = M_FREE;
> +		stat(s, DEACTIVATE_EMPTY);
> +		discard_slab(s, slab);
> +		stat(s, FREE_SLAB);
>  	} else if (new.freelist) {
> -		mode = M_PARTIAL;
> -		/*
> -		 * Taking the spinlock removes the possibility that
> -		 * acquire_slab() will see a slab that is frozen
> -		 */
>  		spin_lock_irqsave(&n->list_lock, flags);
> -	} else {
> -		mode = M_FULL_NOLIST;
> -	}
> -
> -
> -	if (!slab_update_freelist(s, slab,
> -				old.freelist, old.counters,
> -				new.freelist, new.counters,
> -				"unfreezing slab")) {
> -		if (mode == M_PARTIAL)
> -			spin_unlock_irqrestore(&n->list_lock, flags);
> -		goto redo;
> -	}
> -
> -
> -	if (mode == M_PARTIAL) {
>  		add_partial(n, slab, tail);
>  		spin_unlock_irqrestore(&n->list_lock, flags);
>  		stat(s, tail);
> -	} else if (mode == M_FREE) {
> -		stat(s, DEACTIVATE_EMPTY);
> -		discard_slab(s, slab);
> -		stat(s, FREE_SLAB);
> -	} else if (mode == M_FULL_NOLIST) {
> +	} else
>  		stat(s, DEACTIVATE_FULL);
> -	}

Same here.

Thanks!

>  }
>  
>  #ifdef CONFIG_SLUB_CPU_PARTIAL

Re: [RFC PATCH v3 7/7] slub: Optimize deactivate_slab()

[Chengming Zhou]

On 2023/10/31 19:15, Vlastimil Babka wrote:
> On 10/24/23 11:33, chengming.zhou@linux.dev wrote:
>> From: Chengming Zhou <zhouchengming@bytedance.com>
>>
>> Since the introduce of unfrozen slabs on cpu partial list, we don't
>> need to synchronize the slab frozen state under the node list_lock.
>>
>> The caller of deactivate_slab() and the caller of __slab_free() won't
>> manipulate the slab list concurrently.
>>
>> So we can get node list_lock in the last stage if we really need to
>> manipulate the slab list in this path.
>>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
> 
> Nice simplification!
> 
>> ---
>>  mm/slub.c | 70 ++++++++++++++++++++-----------------------------------
>>  1 file changed, 25 insertions(+), 45 deletions(-)
>>
>> diff --git a/mm/slub.c b/mm/slub.c
>> index 486d44421432..64d550e415eb 100644
>> --- a/mm/slub.c
>> +++ b/mm/slub.c
>> @@ -2449,10 +2449,8 @@ static void init_kmem_cache_cpus(struct kmem_cache *s)
>>  static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
>>  			    void *freelist)
>>  {
>> -	enum slab_modes { M_NONE, M_PARTIAL, M_FREE, M_FULL_NOLIST };
>>  	struct kmem_cache_node *n = get_node(s, slab_nid(slab));
>>  	int free_delta = 0;
>> -	enum slab_modes mode = M_NONE;
>>  	void *nextfree, *freelist_iter, *freelist_tail;
>>  	int tail = DEACTIVATE_TO_HEAD;
>>  	unsigned long flags = 0;
>> @@ -2499,58 +2497,40 @@ static void deactivate_slab(struct kmem_cache *s, struct slab *slab,
>>  	 * unfrozen and number of objects in the slab may have changed.
>>  	 * Then release lock and retry cmpxchg again.
>>  	 */
> 
> This comment above (including parts not visible here) should be updated as
> there is no more list manipulation during unfreeze.

Right!

> 
>> -redo:
>> -
>> -	old.freelist = READ_ONCE(slab->freelist);
>> -	old.counters = READ_ONCE(slab->counters);
>> -	VM_BUG_ON(!old.frozen);
>> -
>> -	/* Determine target state of the slab */
>> -	new.counters = old.counters;
>> -	if (freelist_tail) {
>> -		new.inuse -= free_delta;
>> -		set_freepointer(s, freelist_tail, old.freelist);
>> -		new.freelist = freelist;
>> -	} else
>> -		new.freelist = old.freelist;
>> +	do {
>> +		old.freelist = READ_ONCE(slab->freelist);
>> +		old.counters = READ_ONCE(slab->counters);
>> +		VM_BUG_ON(!old.frozen);
>> +
>> +		/* Determine target state of the slab */
>> +		new.counters = old.counters;
>> +		new.frozen = 0;
>> +		if (freelist_tail) {
>> +			new.inuse -= free_delta;
>> +			set_freepointer(s, freelist_tail, old.freelist);
>> +			new.freelist = freelist;
>> +		} else
>> +			new.freelist = old.freelist;
> 
> Per coding style we should have the else with { } even if it's one line, to
> match the if branch. Since we touch the code that was previously violating
> the style, we can fix up.

Ok, I will fix all these.

Big thanks for your review!

> 
>>  
>> -	new.frozen = 0;
>> +	} while (!slab_update_freelist(s, slab,
>> +		old.freelist, old.counters,
>> +		new.freelist, new.counters,
>> +		"unfreezing slab"));
>>  
>> +	/*
>> +	 * Stage three: Manipulate the slab list based on the updated state.
>> +	 */
>>  	if (!new.inuse && n->nr_partial >= s->min_partial) {
>> -		mode = M_FREE;
>> +		stat(s, DEACTIVATE_EMPTY);
>> +		discard_slab(s, slab);
>> +		stat(s, FREE_SLAB);
>>  	} else if (new.freelist) {
>> -		mode = M_PARTIAL;
>> -		/*
>> -		 * Taking the spinlock removes the possibility that
>> -		 * acquire_slab() will see a slab that is frozen
>> -		 */
>>  		spin_lock_irqsave(&n->list_lock, flags);
>> -	} else {
>> -		mode = M_FULL_NOLIST;
>> -	}
>> -
>> -
>> -	if (!slab_update_freelist(s, slab,
>> -				old.freelist, old.counters,
>> -				new.freelist, new.counters,
>> -				"unfreezing slab")) {
>> -		if (mode == M_PARTIAL)
>> -			spin_unlock_irqrestore(&n->list_lock, flags);
>> -		goto redo;
>> -	}
>> -
>> -
>> -	if (mode == M_PARTIAL) {
>>  		add_partial(n, slab, tail);
>>  		spin_unlock_irqrestore(&n->list_lock, flags);
>>  		stat(s, tail);
>> -	} else if (mode == M_FREE) {
>> -		stat(s, DEACTIVATE_EMPTY);
>> -		discard_slab(s, slab);
>> -		stat(s, FREE_SLAB);
>> -	} else if (mode == M_FULL_NOLIST) {
>> +	} else
>>  		stat(s, DEACTIVATE_FULL);
>> -	}
> 
> Same here.
> 
> Thanks!
> 
>>  }
>>  
>>  #ifdef CONFIG_SLUB_CPU_PARTIAL
>