mm: slab: optimize memcg_slab_free_hook()

Most callers of memcg_slab_free_hook() already know the slab,  which could
be passed to memcg_slab_free_hook() directly to reduce the overhead of an
another call of virt_to_slab().  For bulk freeing of objects, the call of
slab_objcgs() in the loop in memcg_slab_free_hook() is redundant as well.
Rework memcg_slab_free_hook() and build_detached_freelist() to reduce
those unnecessary overhead and make memcg_slab_free_hook() can handle bulk
freeing in slab_free().

Move the calling site of memcg_slab_free_hook() from do_slab_free() to
slab_free() for slub to make the code clearer since the logic is weird
(e.g. the caller need to judge whether it needs to call
memcg_slab_free_hook()). It is easy to make mistakes like missing calling
of memcg_slab_free_hook() like fixes of:

  commit d1b2cf6 ("mm: memcg/slab: uncharge during kmem_cache_free_bulk()")
  commit ae085d7 ("mm: kfence: fix missing objcg housekeeping for SLAB")

This optimization is mainly for bulk objects freeing.  The following numbers
is shown for 16-object freeing.

                           before      after
  kmem_cache_free_bulk:   ~430 ns     ~400 ns

The overhead is reduced by about 7% for 16-object freeing.

Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
Link: https://lore.kernel.org/r/20220429123044.37885-1-songmuchun@bytedance.com
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>

Author: Muchun Song

mm/slab.c

@@ -3406,9 +3406,10 @@ static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp,
 {
 	bool init;
 
+	memcg_slab_free_hook(cachep, virt_to_slab(objp), &objp, 1);
+
 	if (is_kfence_address(objp)) {
 		kmemleak_free_recursive(objp, cachep->flags);
-		memcg_slab_free_hook(cachep, &objp, 1);
 		__kfence_free(objp);
 		return;
 	}
@@ -3441,7 +3442,6 @@ void ___cache_free(struct kmem_cache *cachep, void *objp,
 	check_irq_off();
 	kmemleak_free_recursive(objp, cachep->flags);
 	objp = cache_free_debugcheck(cachep, objp, caller);
-	memcg_slab_free_hook(cachep, &objp, 1);
 
 	/*
 	 * Skip calling cache_free_alien() when the platform is not numa.

mm/slab.h

@@ -547,36 +547,22 @@ static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 	obj_cgroup_put(objcg);
 }
 
-static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
+static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
 					void **p, int objects)
 {
-	struct kmem_cache *s;
 	struct obj_cgroup **objcgs;
-	struct obj_cgroup *objcg;
-	struct slab *slab;
-	unsigned int off;
 	int i;
 
 	if (!memcg_kmem_enabled())
 		return;
 
-	for (i = 0; i < objects; i++) {
-		if (unlikely(!p[i]))
-			continue;
-
-		slab = virt_to_slab(p[i]);
-		/* we could be given a kmalloc_large() object, skip those */
-		if (!slab)
-			continue;
-
-		objcgs = slab_objcgs(slab);
-		if (!objcgs)
-			continue;
+	objcgs = slab_objcgs(slab);
+	if (!objcgs)
+		return;
 
-		if (!s_orig)
-			s = slab->slab_cache;
-		else
-			s = s_orig;
+	for (i = 0; i < objects; i++) {
+		struct obj_cgroup *objcg;
+		unsigned int off;
 
 		off = obj_to_index(s, slab, p[i]);
 		objcg = objcgs[off];
@@ -628,7 +614,7 @@ static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
 {
 }
 
-static inline void memcg_slab_free_hook(struct kmem_cache *s,
+static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
 					void **p, int objects)
 {
 }

mm/slub.c

@@ -3464,9 +3464,6 @@ static __always_inline void do_slab_free(struct kmem_cache *s,
 	struct kmem_cache_cpu *c;
 	unsigned long tid;
 
-	/* memcg_slab_free_hook() is already called for bulk free. */
-	if (!tail)
-		memcg_slab_free_hook(s, &head, 1);
 redo:
 	/*
 	 * Determine the currently cpus per cpu slab.
@@ -3526,9 +3523,10 @@ static __always_inline void do_slab_free(struct kmem_cache *s,
 }
 
 static __always_inline void slab_free(struct kmem_cache *s, struct slab *slab,
-				      void *head, void *tail, int cnt,
+				      void *head, void *tail, void **p, int cnt,
 				      unsigned long addr)
 {
+	memcg_slab_free_hook(s, slab, p, cnt);
 	/*
 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
 	 * to remove objects, whose reuse must be delayed.
@@ -3550,7 +3548,7 @@ void kmem_cache_free(struct kmem_cache *s, void *x)
 	if (!s)
 		return;
 	trace_kmem_cache_free(_RET_IP_, x, s->name);
-	slab_free(s, virt_to_slab(x), x, NULL, 1, _RET_IP_);
+	slab_free(s, virt_to_slab(x), x, NULL, &x, 1, _RET_IP_);
 }
 EXPORT_SYMBOL(kmem_cache_free);
 
@@ -3591,79 +3589,59 @@ static inline
 int build_detached_freelist(struct kmem_cache *s, size_t size,
 			    void **p, struct detached_freelist *df)
 {
-	size_t first_skipped_index = 0;
 	int lookahead = 3;
 	void *object;
 	struct folio *folio;
-	struct slab *slab;
-
-	/* Always re-init detached_freelist */
-	df->slab = NULL;
-
-	do {
-		object = p[--size];
-		/* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
-	} while (!object && size);
-
-	if (!object)
-		return 0;
+	size_t same;
 
+	object = p[--size];
 	folio = virt_to_folio(object);
 	if (!s) {
 		/* Handle kalloc'ed objects */
 		if (unlikely(!folio_test_slab(folio))) {
 			free_large_kmalloc(folio, object);
-			p[size] = NULL; /* mark object processed */
+			df->slab = NULL;
 			return size;
 		}
 		/* Derive kmem_cache from object */
-		slab = folio_slab(folio);
-		df->s = slab->slab_cache;
+		df->slab = folio_slab(folio);
+		df->s = df->slab->slab_cache;
 	} else {
-		slab = folio_slab(folio);
+		df->slab = folio_slab(folio);
 		df->s = cache_from_obj(s, object); /* Support for memcg */
 	}
 
-	if (is_kfence_address(object)) {
-		slab_free_hook(df->s, object, false);
-		__kfence_free(object);
-		p[size] = NULL; /* mark object processed */
-		return size;
-	}
-
 	/* Start new detached freelist */
-	df->slab = slab;
-	set_freepointer(df->s, object, NULL);
 	df->tail = object;
 	df->freelist = object;
-	p[size] = NULL; /* mark object processed */
 	df->cnt = 1;
 
+	if (is_kfence_address(object))
+		return size;
+
+	set_freepointer(df->s, object, NULL);
+
+	same = size;
 	while (size) {
 		object = p[--size];
-		if (!object)
-			continue; /* Skip processed objects */
-
 		/* df->slab is always set at this point */
 		if (df->slab == virt_to_slab(object)) {
 			/* Opportunity build freelist */
 			set_freepointer(df->s, object, df->freelist);
 			df->freelist = object;
 			df->cnt++;
-			p[size] = NULL; /* mark object processed */
-
+			same--;
+			if (size != same)
+				swap(p[size], p[same]);
 			continue;
 		}
 
 		/* Limit look ahead search */
 		if (!--lookahead)
 			break;
-
-		if (!first_skipped_index)
-			first_skipped_index = size + 1;
 	}
 
-	return first_skipped_index;
+	return same;
 }
 
 /* Note that interrupts must be enabled when calling this function. */
@@ -3672,15 +3650,15 @@ void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
 	if (WARN_ON(!size))
 		return;
 
-	memcg_slab_free_hook(s, p, size);
 	do {
 		struct detached_freelist df;
 
 		size = build_detached_freelist(s, size, p, &df);
 		if (!df.slab)
 			continue;
 
-		slab_free(df.s, df.slab, df.freelist, df.tail, df.cnt, _RET_IP_);
+		slab_free(df.s, df.slab, df.freelist, df.tail, &p[size], df.cnt,
+			  _RET_IP_);
 	} while (likely(size));
 }
 EXPORT_SYMBOL(kmem_cache_free_bulk);
@@ -4581,7 +4559,7 @@ void kfree(const void *x)
 		return;
 	}
 	slab = folio_slab(folio);
-	slab_free(slab->slab_cache, slab, object, NULL, 1, _RET_IP_);
+	slab_free(slab->slab_cache, slab, object, NULL, &object, 1, _RET_IP_);
 }
 EXPORT_SYMBOL(kfree);