1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
7 #include <linux/slab.h>
10 #include <linux/poison.h>
11 #include <linux/interrupt.h>
12 #include <linux/memory.h>
13 #include <linux/compiler.h>
14 #include <linux/module.h>
15 #include <linux/cpu.h>
16 #include <linux/uaccess.h>
17 #include <linux/seq_file.h>
18 #include <linux/proc_fs.h>
19 #include <asm/cacheflush.h>
20 #include <asm/tlbflush.h>
22 #include <linux/memcontrol.h>
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/kmem.h>
29 enum slab_state slab_state;
30 LIST_HEAD(slab_caches);
31 DEFINE_MUTEX(slab_mutex);
32 struct kmem_cache *kmem_cache;
34 static LIST_HEAD(slab_caches_to_rcu_destroy);
35 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
36 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
37 slab_caches_to_rcu_destroy_workfn);
40 * Set of flags that will prevent slab merging
42 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
43 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
44 SLAB_FAILSLAB | SLAB_KASAN)
46 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
50 * Merge control. If this is set then no merging of slab caches will occur.
52 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
54 static int __init setup_slab_nomerge(char *str)
61 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
64 __setup("slab_nomerge", setup_slab_nomerge);
67 * Determine the size of a slab object
69 unsigned int kmem_cache_size(struct kmem_cache *s)
71 return s->object_size;
73 EXPORT_SYMBOL(kmem_cache_size);
75 #ifdef CONFIG_DEBUG_VM
76 static int kmem_cache_sanity_check(const char *name, size_t size)
78 struct kmem_cache *s = NULL;
80 if (!name || in_interrupt() || size < sizeof(void *) ||
81 size > KMALLOC_MAX_SIZE) {
82 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
86 list_for_each_entry(s, &slab_caches, list) {
91 * This happens when the module gets unloaded and doesn't
92 * destroy its slab cache and no-one else reuses the vmalloc
93 * area of the module. Print a warning.
95 res = probe_kernel_address(s->name, tmp);
97 pr_err("Slab cache with size %d has lost its name\n",
103 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
107 static inline int kmem_cache_sanity_check(const char *name, size_t size)
113 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
117 for (i = 0; i < nr; i++) {
119 kmem_cache_free(s, p[i]);
125 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
130 for (i = 0; i < nr; i++) {
131 void *x = p[i] = kmem_cache_alloc(s, flags);
133 __kmem_cache_free_bulk(s, i, p);
140 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
142 LIST_HEAD(slab_root_caches);
144 void slab_init_memcg_params(struct kmem_cache *s)
146 s->memcg_params.root_cache = NULL;
147 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
148 INIT_LIST_HEAD(&s->memcg_params.children);
151 static int init_memcg_params(struct kmem_cache *s,
152 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
154 struct memcg_cache_array *arr;
157 s->memcg_params.root_cache = root_cache;
158 s->memcg_params.memcg = memcg;
159 INIT_LIST_HEAD(&s->memcg_params.children_node);
160 INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
164 slab_init_memcg_params(s);
166 if (!memcg_nr_cache_ids)
169 arr = kvzalloc(sizeof(struct memcg_cache_array) +
170 memcg_nr_cache_ids * sizeof(void *),
175 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
179 static void destroy_memcg_params(struct kmem_cache *s)
181 if (is_root_cache(s))
182 kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
185 static void free_memcg_params(struct rcu_head *rcu)
187 struct memcg_cache_array *old;
189 old = container_of(rcu, struct memcg_cache_array, rcu);
193 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
195 struct memcg_cache_array *old, *new;
197 new = kvzalloc(sizeof(struct memcg_cache_array) +
198 new_array_size * sizeof(void *), GFP_KERNEL);
202 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
203 lockdep_is_held(&slab_mutex));
205 memcpy(new->entries, old->entries,
206 memcg_nr_cache_ids * sizeof(void *));
208 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
210 call_rcu(&old->rcu, free_memcg_params);
214 int memcg_update_all_caches(int num_memcgs)
216 struct kmem_cache *s;
219 mutex_lock(&slab_mutex);
220 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
221 ret = update_memcg_params(s, num_memcgs);
223 * Instead of freeing the memory, we'll just leave the caches
224 * up to this point in an updated state.
229 mutex_unlock(&slab_mutex);
233 void memcg_link_cache(struct kmem_cache *s)
235 if (is_root_cache(s)) {
236 list_add(&s->root_caches_node, &slab_root_caches);
238 list_add(&s->memcg_params.children_node,
239 &s->memcg_params.root_cache->memcg_params.children);
240 list_add(&s->memcg_params.kmem_caches_node,
241 &s->memcg_params.memcg->kmem_caches);
245 static void memcg_unlink_cache(struct kmem_cache *s)
247 if (is_root_cache(s)) {
248 list_del(&s->root_caches_node);
250 list_del(&s->memcg_params.children_node);
251 list_del(&s->memcg_params.kmem_caches_node);
255 static inline int init_memcg_params(struct kmem_cache *s,
256 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
261 static inline void destroy_memcg_params(struct kmem_cache *s)
265 static inline void memcg_unlink_cache(struct kmem_cache *s)
268 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
271 * Find a mergeable slab cache
273 int slab_unmergeable(struct kmem_cache *s)
275 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
278 if (!is_root_cache(s))
285 * We may have set a slab to be unmergeable during bootstrap.
293 struct kmem_cache *find_mergeable(size_t size, size_t align,
294 slab_flags_t flags, const char *name, void (*ctor)(void *))
296 struct kmem_cache *s;
304 size = ALIGN(size, sizeof(void *));
305 align = calculate_alignment(flags, align, size);
306 size = ALIGN(size, align);
307 flags = kmem_cache_flags(size, flags, name, NULL);
309 if (flags & SLAB_NEVER_MERGE)
312 list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
313 if (slab_unmergeable(s))
319 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
322 * Check if alignment is compatible.
323 * Courtesy of Adrian Drzewiecki
325 if ((s->size & ~(align - 1)) != s->size)
328 if (s->size - size >= sizeof(void *))
331 if (IS_ENABLED(CONFIG_SLAB) && align &&
332 (align > s->align || s->align % align))
341 * Figure out what the alignment of the objects will be given a set of
342 * flags, a user specified alignment and the size of the objects.
344 unsigned long calculate_alignment(slab_flags_t flags,
345 unsigned long align, unsigned long size)
348 * If the user wants hardware cache aligned objects then follow that
349 * suggestion if the object is sufficiently large.
351 * The hardware cache alignment cannot override the specified
352 * alignment though. If that is greater then use it.
354 if (flags & SLAB_HWCACHE_ALIGN) {
355 unsigned long ralign = cache_line_size();
356 while (size <= ralign / 2)
358 align = max(align, ralign);
361 if (align < ARCH_SLAB_MINALIGN)
362 align = ARCH_SLAB_MINALIGN;
364 return ALIGN(align, sizeof(void *));
367 static struct kmem_cache *create_cache(const char *name,
368 size_t object_size, size_t size, size_t align,
369 slab_flags_t flags, void (*ctor)(void *),
370 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
372 struct kmem_cache *s;
376 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
381 s->object_size = object_size;
386 err = init_memcg_params(s, memcg, root_cache);
390 err = __kmem_cache_create(s, flags);
395 list_add(&s->list, &slab_caches);
403 destroy_memcg_params(s);
404 kmem_cache_free(kmem_cache, s);
409 * kmem_cache_create - Create a cache.
410 * @name: A string which is used in /proc/slabinfo to identify this cache.
411 * @size: The size of objects to be created in this cache.
412 * @align: The required alignment for the objects.
414 * @ctor: A constructor for the objects.
416 * Returns a ptr to the cache on success, NULL on failure.
417 * Cannot be called within a interrupt, but can be interrupted.
418 * The @ctor is run when new pages are allocated by the cache.
422 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
423 * to catch references to uninitialised memory.
425 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
426 * for buffer overruns.
428 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
429 * cacheline. This can be beneficial if you're counting cycles as closely
433 kmem_cache_create(const char *name, size_t size, size_t align,
434 slab_flags_t flags, void (*ctor)(void *))
436 struct kmem_cache *s = NULL;
437 const char *cache_name;
442 memcg_get_cache_ids();
444 mutex_lock(&slab_mutex);
446 err = kmem_cache_sanity_check(name, size);
451 /* Refuse requests with allocator specific flags */
452 if (flags & ~SLAB_FLAGS_PERMITTED) {
458 * Some allocators will constraint the set of valid flags to a subset
459 * of all flags. We expect them to define CACHE_CREATE_MASK in this
460 * case, and we'll just provide them with a sanitized version of the
463 flags &= CACHE_CREATE_MASK;
465 s = __kmem_cache_alias(name, size, align, flags, ctor);
469 cache_name = kstrdup_const(name, GFP_KERNEL);
475 s = create_cache(cache_name, size, size,
476 calculate_alignment(flags, align, size),
477 flags, ctor, NULL, NULL);
480 kfree_const(cache_name);
484 mutex_unlock(&slab_mutex);
486 memcg_put_cache_ids();
491 if (flags & SLAB_PANIC)
492 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
495 pr_warn("kmem_cache_create(%s) failed with error %d\n",
503 EXPORT_SYMBOL(kmem_cache_create);
505 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
507 LIST_HEAD(to_destroy);
508 struct kmem_cache *s, *s2;
511 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
512 * @slab_caches_to_rcu_destroy list. The slab pages are freed
513 * through RCU and and the associated kmem_cache are dereferenced
514 * while freeing the pages, so the kmem_caches should be freed only
515 * after the pending RCU operations are finished. As rcu_barrier()
516 * is a pretty slow operation, we batch all pending destructions
519 mutex_lock(&slab_mutex);
520 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
521 mutex_unlock(&slab_mutex);
523 if (list_empty(&to_destroy))
528 list_for_each_entry_safe(s, s2, &to_destroy, list) {
529 #ifdef SLAB_SUPPORTS_SYSFS
530 sysfs_slab_release(s);
532 slab_kmem_cache_release(s);
537 static int shutdown_cache(struct kmem_cache *s)
539 /* free asan quarantined objects */
540 kasan_cache_shutdown(s);
542 if (__kmem_cache_shutdown(s) != 0)
545 memcg_unlink_cache(s);
548 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
549 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
550 schedule_work(&slab_caches_to_rcu_destroy_work);
552 #ifdef SLAB_SUPPORTS_SYSFS
553 sysfs_slab_release(s);
555 slab_kmem_cache_release(s);
562 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
564 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
565 * @memcg: The memory cgroup the new cache is for.
566 * @root_cache: The parent of the new cache.
568 * This function attempts to create a kmem cache that will serve allocation
569 * requests going from @memcg to @root_cache. The new cache inherits properties
572 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
573 struct kmem_cache *root_cache)
575 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
576 struct cgroup_subsys_state *css = &memcg->css;
577 struct memcg_cache_array *arr;
578 struct kmem_cache *s = NULL;
585 mutex_lock(&slab_mutex);
588 * The memory cgroup could have been offlined while the cache
589 * creation work was pending.
591 if (memcg->kmem_state != KMEM_ONLINE)
594 idx = memcg_cache_id(memcg);
595 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
596 lockdep_is_held(&slab_mutex));
599 * Since per-memcg caches are created asynchronously on first
600 * allocation (see memcg_kmem_get_cache()), several threads can try to
601 * create the same cache, but only one of them may succeed.
603 if (arr->entries[idx])
606 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
607 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
608 css->serial_nr, memcg_name_buf);
612 s = create_cache(cache_name, root_cache->object_size,
613 root_cache->size, root_cache->align,
614 root_cache->flags & CACHE_CREATE_MASK,
615 root_cache->ctor, memcg, root_cache);
617 * If we could not create a memcg cache, do not complain, because
618 * that's not critical at all as we can always proceed with the root
627 * Since readers won't lock (see cache_from_memcg_idx()), we need a
628 * barrier here to ensure nobody will see the kmem_cache partially
632 arr->entries[idx] = s;
635 mutex_unlock(&slab_mutex);
641 static void kmemcg_deactivate_workfn(struct work_struct *work)
643 struct kmem_cache *s = container_of(work, struct kmem_cache,
644 memcg_params.deact_work);
649 mutex_lock(&slab_mutex);
651 s->memcg_params.deact_fn(s);
653 mutex_unlock(&slab_mutex);
658 /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
659 css_put(&s->memcg_params.memcg->css);
662 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
664 struct kmem_cache *s = container_of(head, struct kmem_cache,
665 memcg_params.deact_rcu_head);
668 * We need to grab blocking locks. Bounce to ->deact_work. The
669 * work item shares the space with the RCU head and can't be
670 * initialized eariler.
672 INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
673 queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
677 * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
678 * sched RCU grace period
679 * @s: target kmem_cache
680 * @deact_fn: deactivation function to call
682 * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
683 * held after a sched RCU grace period. The slab is guaranteed to stay
684 * alive until @deact_fn is finished. This is to be used from
685 * __kmemcg_cache_deactivate().
687 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
688 void (*deact_fn)(struct kmem_cache *))
690 if (WARN_ON_ONCE(is_root_cache(s)) ||
691 WARN_ON_ONCE(s->memcg_params.deact_fn))
694 /* pin memcg so that @s doesn't get destroyed in the middle */
695 css_get(&s->memcg_params.memcg->css);
697 s->memcg_params.deact_fn = deact_fn;
698 call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
701 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
704 struct memcg_cache_array *arr;
705 struct kmem_cache *s, *c;
707 idx = memcg_cache_id(memcg);
712 mutex_lock(&slab_mutex);
713 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
714 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
715 lockdep_is_held(&slab_mutex));
716 c = arr->entries[idx];
720 __kmemcg_cache_deactivate(c);
721 arr->entries[idx] = NULL;
723 mutex_unlock(&slab_mutex);
729 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
731 struct kmem_cache *s, *s2;
736 mutex_lock(&slab_mutex);
737 list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
738 memcg_params.kmem_caches_node) {
740 * The cgroup is about to be freed and therefore has no charges
741 * left. Hence, all its caches must be empty by now.
743 BUG_ON(shutdown_cache(s));
745 mutex_unlock(&slab_mutex);
751 static int shutdown_memcg_caches(struct kmem_cache *s)
753 struct memcg_cache_array *arr;
754 struct kmem_cache *c, *c2;
758 BUG_ON(!is_root_cache(s));
761 * First, shutdown active caches, i.e. caches that belong to online
764 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
765 lockdep_is_held(&slab_mutex));
766 for_each_memcg_cache_index(i) {
770 if (shutdown_cache(c))
772 * The cache still has objects. Move it to a temporary
773 * list so as not to try to destroy it for a second
774 * time while iterating over inactive caches below.
776 list_move(&c->memcg_params.children_node, &busy);
779 * The cache is empty and will be destroyed soon. Clear
780 * the pointer to it in the memcg_caches array so that
781 * it will never be accessed even if the root cache
784 arr->entries[i] = NULL;
788 * Second, shutdown all caches left from memory cgroups that are now
791 list_for_each_entry_safe(c, c2, &s->memcg_params.children,
792 memcg_params.children_node)
795 list_splice(&busy, &s->memcg_params.children);
798 * A cache being destroyed must be empty. In particular, this means
799 * that all per memcg caches attached to it must be empty too.
801 if (!list_empty(&s->memcg_params.children))
806 static inline int shutdown_memcg_caches(struct kmem_cache *s)
810 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
812 void slab_kmem_cache_release(struct kmem_cache *s)
814 __kmem_cache_release(s);
815 destroy_memcg_params(s);
816 kfree_const(s->name);
817 kmem_cache_free(kmem_cache, s);
820 void kmem_cache_destroy(struct kmem_cache *s)
830 mutex_lock(&slab_mutex);
836 err = shutdown_memcg_caches(s);
838 err = shutdown_cache(s);
841 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
846 mutex_unlock(&slab_mutex);
851 EXPORT_SYMBOL(kmem_cache_destroy);
854 * kmem_cache_shrink - Shrink a cache.
855 * @cachep: The cache to shrink.
857 * Releases as many slabs as possible for a cache.
858 * To help debugging, a zero exit status indicates all slabs were released.
860 int kmem_cache_shrink(struct kmem_cache *cachep)
866 kasan_cache_shrink(cachep);
867 ret = __kmem_cache_shrink(cachep);
872 EXPORT_SYMBOL(kmem_cache_shrink);
874 bool slab_is_available(void)
876 return slab_state >= UP;
880 /* Create a cache during boot when no slab services are available yet */
881 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
887 s->size = s->object_size = size;
888 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
890 slab_init_memcg_params(s);
892 err = __kmem_cache_create(s, flags);
895 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
898 s->refcount = -1; /* Exempt from merging for now */
901 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
904 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
907 panic("Out of memory when creating slab %s\n", name);
909 create_boot_cache(s, name, size, flags);
910 list_add(&s->list, &slab_caches);
916 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
917 EXPORT_SYMBOL(kmalloc_caches);
919 #ifdef CONFIG_ZONE_DMA
920 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
921 EXPORT_SYMBOL(kmalloc_dma_caches);
925 * Conversion table for small slabs sizes / 8 to the index in the
926 * kmalloc array. This is necessary for slabs < 192 since we have non power
927 * of two cache sizes there. The size of larger slabs can be determined using
930 static s8 size_index[24] = {
957 static inline int size_index_elem(size_t bytes)
959 return (bytes - 1) / 8;
963 * Find the kmem_cache structure that serves a given size of
966 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
970 if (unlikely(size > KMALLOC_MAX_SIZE)) {
971 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
977 return ZERO_SIZE_PTR;
979 index = size_index[size_index_elem(size)];
981 index = fls(size - 1);
983 #ifdef CONFIG_ZONE_DMA
984 if (unlikely((flags & GFP_DMA)))
985 return kmalloc_dma_caches[index];
988 return kmalloc_caches[index];
992 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
993 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
996 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
997 {NULL, 0}, {"kmalloc-96", 96},
998 {"kmalloc-192", 192}, {"kmalloc-8", 8},
999 {"kmalloc-16", 16}, {"kmalloc-32", 32},
1000 {"kmalloc-64", 64}, {"kmalloc-128", 128},
1001 {"kmalloc-256", 256}, {"kmalloc-512", 512},
1002 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
1003 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
1004 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
1005 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
1006 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
1007 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
1008 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
1009 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
1010 {"kmalloc-67108864", 67108864}
1014 * Patch up the size_index table if we have strange large alignment
1015 * requirements for the kmalloc array. This is only the case for
1016 * MIPS it seems. The standard arches will not generate any code here.
1018 * Largest permitted alignment is 256 bytes due to the way we
1019 * handle the index determination for the smaller caches.
1021 * Make sure that nothing crazy happens if someone starts tinkering
1022 * around with ARCH_KMALLOC_MINALIGN
1024 void __init setup_kmalloc_cache_index_table(void)
1028 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1029 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1031 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1032 int elem = size_index_elem(i);
1034 if (elem >= ARRAY_SIZE(size_index))
1036 size_index[elem] = KMALLOC_SHIFT_LOW;
1039 if (KMALLOC_MIN_SIZE >= 64) {
1041 * The 96 byte size cache is not used if the alignment
1044 for (i = 64 + 8; i <= 96; i += 8)
1045 size_index[size_index_elem(i)] = 7;
1049 if (KMALLOC_MIN_SIZE >= 128) {
1051 * The 192 byte sized cache is not used if the alignment
1052 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1055 for (i = 128 + 8; i <= 192; i += 8)
1056 size_index[size_index_elem(i)] = 8;
1060 static void __init new_kmalloc_cache(int idx, slab_flags_t flags)
1062 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
1063 kmalloc_info[idx].size, flags);
1067 * Create the kmalloc array. Some of the regular kmalloc arrays
1068 * may already have been created because they were needed to
1069 * enable allocations for slab creation.
1071 void __init create_kmalloc_caches(slab_flags_t flags)
1075 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1076 if (!kmalloc_caches[i])
1077 new_kmalloc_cache(i, flags);
1080 * Caches that are not of the two-to-the-power-of size.
1081 * These have to be created immediately after the
1082 * earlier power of two caches
1084 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1085 new_kmalloc_cache(1, flags);
1086 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1087 new_kmalloc_cache(2, flags);
1090 /* Kmalloc array is now usable */
1093 #ifdef CONFIG_ZONE_DMA
1094 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1095 struct kmem_cache *s = kmalloc_caches[i];
1098 int size = kmalloc_size(i);
1099 char *n = kasprintf(GFP_NOWAIT,
1100 "dma-kmalloc-%d", size);
1103 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1104 size, SLAB_CACHE_DMA | flags);
1109 #endif /* !CONFIG_SLOB */
1112 * To avoid unnecessary overhead, we pass through large allocation requests
1113 * directly to the page allocator. We use __GFP_COMP, because we will need to
1114 * know the allocation order to free the pages properly in kfree.
1116 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1121 flags |= __GFP_COMP;
1122 page = alloc_pages(flags, order);
1123 ret = page ? page_address(page) : NULL;
1124 kmemleak_alloc(ret, size, 1, flags);
1125 kasan_kmalloc_large(ret, size, flags);
1128 EXPORT_SYMBOL(kmalloc_order);
1130 #ifdef CONFIG_TRACING
1131 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1133 void *ret = kmalloc_order(size, flags, order);
1134 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1137 EXPORT_SYMBOL(kmalloc_order_trace);
1140 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1141 /* Randomize a generic freelist */
1142 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1148 for (i = 0; i < count; i++)
1151 /* Fisher-Yates shuffle */
1152 for (i = count - 1; i > 0; i--) {
1153 rand = prandom_u32_state(state);
1155 swap(list[i], list[rand]);
1159 /* Create a random sequence per cache */
1160 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1163 struct rnd_state state;
1165 if (count < 2 || cachep->random_seq)
1168 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1169 if (!cachep->random_seq)
1172 /* Get best entropy at this stage of boot */
1173 prandom_seed_state(&state, get_random_long());
1175 freelist_randomize(&state, cachep->random_seq, count);
1179 /* Destroy the per-cache random freelist sequence */
1180 void cache_random_seq_destroy(struct kmem_cache *cachep)
1182 kfree(cachep->random_seq);
1183 cachep->random_seq = NULL;
1185 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1187 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1189 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1191 #define SLABINFO_RIGHTS S_IRUSR
1194 static void print_slabinfo_header(struct seq_file *m)
1197 * Output format version, so at least we can change it
1198 * without _too_ many complaints.
1200 #ifdef CONFIG_DEBUG_SLAB
1201 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1203 seq_puts(m, "slabinfo - version: 2.1\n");
1205 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1206 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1207 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1208 #ifdef CONFIG_DEBUG_SLAB
1209 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1210 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1215 void *slab_start(struct seq_file *m, loff_t *pos)
1217 mutex_lock(&slab_mutex);
1218 return seq_list_start(&slab_root_caches, *pos);
1221 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1223 return seq_list_next(p, &slab_root_caches, pos);
1226 void slab_stop(struct seq_file *m, void *p)
1228 mutex_unlock(&slab_mutex);
1232 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1234 struct kmem_cache *c;
1235 struct slabinfo sinfo;
1237 if (!is_root_cache(s))
1240 for_each_memcg_cache(c, s) {
1241 memset(&sinfo, 0, sizeof(sinfo));
1242 get_slabinfo(c, &sinfo);
1244 info->active_slabs += sinfo.active_slabs;
1245 info->num_slabs += sinfo.num_slabs;
1246 info->shared_avail += sinfo.shared_avail;
1247 info->active_objs += sinfo.active_objs;
1248 info->num_objs += sinfo.num_objs;
1252 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1254 struct slabinfo sinfo;
1256 memset(&sinfo, 0, sizeof(sinfo));
1257 get_slabinfo(s, &sinfo);
1259 memcg_accumulate_slabinfo(s, &sinfo);
1261 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1262 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1263 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1265 seq_printf(m, " : tunables %4u %4u %4u",
1266 sinfo.limit, sinfo.batchcount, sinfo.shared);
1267 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1268 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1269 slabinfo_show_stats(m, s);
1273 static int slab_show(struct seq_file *m, void *p)
1275 struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1277 if (p == slab_root_caches.next)
1278 print_slabinfo_header(m);
1283 void dump_unreclaimable_slab(void)
1285 struct kmem_cache *s, *s2;
1286 struct slabinfo sinfo;
1289 * Here acquiring slab_mutex is risky since we don't prefer to get
1290 * sleep in oom path. But, without mutex hold, it may introduce a
1292 * Use mutex_trylock to protect the list traverse, dump nothing
1293 * without acquiring the mutex.
1295 if (!mutex_trylock(&slab_mutex)) {
1296 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1300 pr_info("Unreclaimable slab info:\n");
1301 pr_info("Name Used Total\n");
1303 list_for_each_entry_safe(s, s2, &slab_caches, list) {
1304 if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1307 get_slabinfo(s, &sinfo);
1309 if (sinfo.num_objs > 0)
1310 pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1311 (sinfo.active_objs * s->size) / 1024,
1312 (sinfo.num_objs * s->size) / 1024);
1314 mutex_unlock(&slab_mutex);
1317 #if defined(CONFIG_MEMCG)
1318 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1320 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1322 mutex_lock(&slab_mutex);
1323 return seq_list_start(&memcg->kmem_caches, *pos);
1326 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1328 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1330 return seq_list_next(p, &memcg->kmem_caches, pos);
1333 void memcg_slab_stop(struct seq_file *m, void *p)
1335 mutex_unlock(&slab_mutex);
1338 int memcg_slab_show(struct seq_file *m, void *p)
1340 struct kmem_cache *s = list_entry(p, struct kmem_cache,
1341 memcg_params.kmem_caches_node);
1342 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1344 if (p == memcg->kmem_caches.next)
1345 print_slabinfo_header(m);
1352 * slabinfo_op - iterator that generates /proc/slabinfo
1361 * num-pages-per-slab
1362 * + further values on SMP and with statistics enabled
1364 static const struct seq_operations slabinfo_op = {
1365 .start = slab_start,
1371 static int slabinfo_open(struct inode *inode, struct file *file)
1373 return seq_open(file, &slabinfo_op);
1376 static const struct file_operations proc_slabinfo_operations = {
1377 .open = slabinfo_open,
1379 .write = slabinfo_write,
1380 .llseek = seq_lseek,
1381 .release = seq_release,
1384 static int __init slab_proc_init(void)
1386 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1387 &proc_slabinfo_operations);
1390 module_init(slab_proc_init);
1391 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1393 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1402 if (ks >= new_size) {
1403 kasan_krealloc((void *)p, new_size, flags);
1407 ret = kmalloc_track_caller(new_size, flags);
1415 * __krealloc - like krealloc() but don't free @p.
1416 * @p: object to reallocate memory for.
1417 * @new_size: how many bytes of memory are required.
1418 * @flags: the type of memory to allocate.
1420 * This function is like krealloc() except it never frees the originally
1421 * allocated buffer. Use this if you don't want to free the buffer immediately
1422 * like, for example, with RCU.
1424 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1426 if (unlikely(!new_size))
1427 return ZERO_SIZE_PTR;
1429 return __do_krealloc(p, new_size, flags);
1432 EXPORT_SYMBOL(__krealloc);
1435 * krealloc - reallocate memory. The contents will remain unchanged.
1436 * @p: object to reallocate memory for.
1437 * @new_size: how many bytes of memory are required.
1438 * @flags: the type of memory to allocate.
1440 * The contents of the object pointed to are preserved up to the
1441 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1442 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1443 * %NULL pointer, the object pointed to is freed.
1445 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1449 if (unlikely(!new_size)) {
1451 return ZERO_SIZE_PTR;
1454 ret = __do_krealloc(p, new_size, flags);
1455 if (ret && p != ret)
1460 EXPORT_SYMBOL(krealloc);
1463 * kzfree - like kfree but zero memory
1464 * @p: object to free memory of
1466 * The memory of the object @p points to is zeroed before freed.
1467 * If @p is %NULL, kzfree() does nothing.
1469 * Note: this function zeroes the whole allocated buffer which can be a good
1470 * deal bigger than the requested buffer size passed to kmalloc(). So be
1471 * careful when using this function in performance sensitive code.
1473 void kzfree(const void *p)
1476 void *mem = (void *)p;
1478 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1484 EXPORT_SYMBOL(kzfree);
1486 /* Tracepoints definitions. */
1487 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1488 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1489 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1490 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1491 EXPORT_TRACEPOINT_SYMBOL(kfree);
1492 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);