1 // SPDX-License-Identifier: GPL-2.0
3 * Slab allocator functions that are independent of the allocator strategy
5 * (C) 2012 Christoph Lameter <cl@linux.com>
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 unsigned long 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(unsigned long 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 unsigned long 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 unsigned long 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 #ifdef SLAB_SUPPORTS_SYSFS
550 sysfs_slab_unlink(s);
552 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
553 schedule_work(&slab_caches_to_rcu_destroy_work);
555 #ifdef SLAB_SUPPORTS_SYSFS
556 sysfs_slab_unlink(s);
557 sysfs_slab_release(s);
559 slab_kmem_cache_release(s);
566 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
568 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
569 * @memcg: The memory cgroup the new cache is for.
570 * @root_cache: The parent of the new cache.
572 * This function attempts to create a kmem cache that will serve allocation
573 * requests going from @memcg to @root_cache. The new cache inherits properties
576 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
577 struct kmem_cache *root_cache)
579 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
580 struct cgroup_subsys_state *css = &memcg->css;
581 struct memcg_cache_array *arr;
582 struct kmem_cache *s = NULL;
589 mutex_lock(&slab_mutex);
592 * The memory cgroup could have been offlined while the cache
593 * creation work was pending.
595 if (memcg->kmem_state != KMEM_ONLINE)
598 idx = memcg_cache_id(memcg);
599 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
600 lockdep_is_held(&slab_mutex));
603 * Since per-memcg caches are created asynchronously on first
604 * allocation (see memcg_kmem_get_cache()), several threads can try to
605 * create the same cache, but only one of them may succeed.
607 if (arr->entries[idx])
610 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
611 cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
612 css->serial_nr, memcg_name_buf);
616 s = create_cache(cache_name, root_cache->object_size,
617 root_cache->size, root_cache->align,
618 root_cache->flags & CACHE_CREATE_MASK,
619 root_cache->ctor, memcg, root_cache);
621 * If we could not create a memcg cache, do not complain, because
622 * that's not critical at all as we can always proceed with the root
631 * Since readers won't lock (see cache_from_memcg_idx()), we need a
632 * barrier here to ensure nobody will see the kmem_cache partially
636 arr->entries[idx] = s;
639 mutex_unlock(&slab_mutex);
645 static void kmemcg_deactivate_workfn(struct work_struct *work)
647 struct kmem_cache *s = container_of(work, struct kmem_cache,
648 memcg_params.deact_work);
653 mutex_lock(&slab_mutex);
655 s->memcg_params.deact_fn(s);
657 mutex_unlock(&slab_mutex);
662 /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
663 css_put(&s->memcg_params.memcg->css);
666 static void kmemcg_deactivate_rcufn(struct rcu_head *head)
668 struct kmem_cache *s = container_of(head, struct kmem_cache,
669 memcg_params.deact_rcu_head);
672 * We need to grab blocking locks. Bounce to ->deact_work. The
673 * work item shares the space with the RCU head and can't be
674 * initialized eariler.
676 INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
677 queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
681 * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
682 * sched RCU grace period
683 * @s: target kmem_cache
684 * @deact_fn: deactivation function to call
686 * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
687 * held after a sched RCU grace period. The slab is guaranteed to stay
688 * alive until @deact_fn is finished. This is to be used from
689 * __kmemcg_cache_deactivate().
691 void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
692 void (*deact_fn)(struct kmem_cache *))
694 if (WARN_ON_ONCE(is_root_cache(s)) ||
695 WARN_ON_ONCE(s->memcg_params.deact_fn))
698 /* pin memcg so that @s doesn't get destroyed in the middle */
699 css_get(&s->memcg_params.memcg->css);
701 s->memcg_params.deact_fn = deact_fn;
702 call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
705 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
708 struct memcg_cache_array *arr;
709 struct kmem_cache *s, *c;
711 idx = memcg_cache_id(memcg);
716 mutex_lock(&slab_mutex);
717 list_for_each_entry(s, &slab_root_caches, root_caches_node) {
718 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
719 lockdep_is_held(&slab_mutex));
720 c = arr->entries[idx];
724 __kmemcg_cache_deactivate(c);
725 arr->entries[idx] = NULL;
727 mutex_unlock(&slab_mutex);
733 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
735 struct kmem_cache *s, *s2;
740 mutex_lock(&slab_mutex);
741 list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
742 memcg_params.kmem_caches_node) {
744 * The cgroup is about to be freed and therefore has no charges
745 * left. Hence, all its caches must be empty by now.
747 BUG_ON(shutdown_cache(s));
749 mutex_unlock(&slab_mutex);
755 static int shutdown_memcg_caches(struct kmem_cache *s)
757 struct memcg_cache_array *arr;
758 struct kmem_cache *c, *c2;
762 BUG_ON(!is_root_cache(s));
765 * First, shutdown active caches, i.e. caches that belong to online
768 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
769 lockdep_is_held(&slab_mutex));
770 for_each_memcg_cache_index(i) {
774 if (shutdown_cache(c))
776 * The cache still has objects. Move it to a temporary
777 * list so as not to try to destroy it for a second
778 * time while iterating over inactive caches below.
780 list_move(&c->memcg_params.children_node, &busy);
783 * The cache is empty and will be destroyed soon. Clear
784 * the pointer to it in the memcg_caches array so that
785 * it will never be accessed even if the root cache
788 arr->entries[i] = NULL;
792 * Second, shutdown all caches left from memory cgroups that are now
795 list_for_each_entry_safe(c, c2, &s->memcg_params.children,
796 memcg_params.children_node)
799 list_splice(&busy, &s->memcg_params.children);
802 * A cache being destroyed must be empty. In particular, this means
803 * that all per memcg caches attached to it must be empty too.
805 if (!list_empty(&s->memcg_params.children))
810 static inline int shutdown_memcg_caches(struct kmem_cache *s)
814 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
816 void slab_kmem_cache_release(struct kmem_cache *s)
818 __kmem_cache_release(s);
819 destroy_memcg_params(s);
820 kfree_const(s->name);
821 kmem_cache_free(kmem_cache, s);
824 void kmem_cache_destroy(struct kmem_cache *s)
834 mutex_lock(&slab_mutex);
840 err = shutdown_memcg_caches(s);
842 err = shutdown_cache(s);
845 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
850 mutex_unlock(&slab_mutex);
855 EXPORT_SYMBOL(kmem_cache_destroy);
858 * kmem_cache_shrink - Shrink a cache.
859 * @cachep: The cache to shrink.
861 * Releases as many slabs as possible for a cache.
862 * To help debugging, a zero exit status indicates all slabs were released.
864 int kmem_cache_shrink(struct kmem_cache *cachep)
870 kasan_cache_shrink(cachep);
871 ret = __kmem_cache_shrink(cachep);
876 EXPORT_SYMBOL(kmem_cache_shrink);
878 bool slab_is_available(void)
880 return slab_state >= UP;
884 /* Create a cache during boot when no slab services are available yet */
885 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
891 s->size = s->object_size = size;
892 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
894 slab_init_memcg_params(s);
896 err = __kmem_cache_create(s, flags);
899 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
902 s->refcount = -1; /* Exempt from merging for now */
905 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
908 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
911 panic("Out of memory when creating slab %s\n", name);
913 create_boot_cache(s, name, size, flags);
914 list_add(&s->list, &slab_caches);
920 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
921 EXPORT_SYMBOL(kmalloc_caches);
923 #ifdef CONFIG_ZONE_DMA
924 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
925 EXPORT_SYMBOL(kmalloc_dma_caches);
929 * Conversion table for small slabs sizes / 8 to the index in the
930 * kmalloc array. This is necessary for slabs < 192 since we have non power
931 * of two cache sizes there. The size of larger slabs can be determined using
934 static s8 size_index[24] = {
961 static inline int size_index_elem(size_t bytes)
963 return (bytes - 1) / 8;
967 * Find the kmem_cache structure that serves a given size of
970 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
976 return ZERO_SIZE_PTR;
978 index = size_index[size_index_elem(size)];
980 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
984 index = fls(size - 1);
987 #ifdef CONFIG_ZONE_DMA
988 if (unlikely((flags & GFP_DMA)))
989 return kmalloc_dma_caches[index];
992 return kmalloc_caches[index];
996 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
997 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1000 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1001 {NULL, 0}, {"kmalloc-96", 96},
1002 {"kmalloc-192", 192}, {"kmalloc-8", 8},
1003 {"kmalloc-16", 16}, {"kmalloc-32", 32},
1004 {"kmalloc-64", 64}, {"kmalloc-128", 128},
1005 {"kmalloc-256", 256}, {"kmalloc-512", 512},
1006 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
1007 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
1008 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
1009 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
1010 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
1011 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
1012 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
1013 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
1014 {"kmalloc-67108864", 67108864}
1018 * Patch up the size_index table if we have strange large alignment
1019 * requirements for the kmalloc array. This is only the case for
1020 * MIPS it seems. The standard arches will not generate any code here.
1022 * Largest permitted alignment is 256 bytes due to the way we
1023 * handle the index determination for the smaller caches.
1025 * Make sure that nothing crazy happens if someone starts tinkering
1026 * around with ARCH_KMALLOC_MINALIGN
1028 void __init setup_kmalloc_cache_index_table(void)
1032 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1033 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1035 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1036 int elem = size_index_elem(i);
1038 if (elem >= ARRAY_SIZE(size_index))
1040 size_index[elem] = KMALLOC_SHIFT_LOW;
1043 if (KMALLOC_MIN_SIZE >= 64) {
1045 * The 96 byte size cache is not used if the alignment
1048 for (i = 64 + 8; i <= 96; i += 8)
1049 size_index[size_index_elem(i)] = 7;
1053 if (KMALLOC_MIN_SIZE >= 128) {
1055 * The 192 byte sized cache is not used if the alignment
1056 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1059 for (i = 128 + 8; i <= 192; i += 8)
1060 size_index[size_index_elem(i)] = 8;
1064 static void __init new_kmalloc_cache(int idx, unsigned long flags)
1066 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
1067 kmalloc_info[idx].size, flags);
1071 * Create the kmalloc array. Some of the regular kmalloc arrays
1072 * may already have been created because they were needed to
1073 * enable allocations for slab creation.
1075 void __init create_kmalloc_caches(unsigned long flags)
1079 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1080 if (!kmalloc_caches[i])
1081 new_kmalloc_cache(i, flags);
1084 * Caches that are not of the two-to-the-power-of size.
1085 * These have to be created immediately after the
1086 * earlier power of two caches
1088 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1089 new_kmalloc_cache(1, flags);
1090 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1091 new_kmalloc_cache(2, flags);
1094 /* Kmalloc array is now usable */
1097 #ifdef CONFIG_ZONE_DMA
1098 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1099 struct kmem_cache *s = kmalloc_caches[i];
1102 int size = kmalloc_size(i);
1103 char *n = kasprintf(GFP_NOWAIT,
1104 "dma-kmalloc-%d", size);
1107 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1108 size, SLAB_CACHE_DMA | flags);
1113 #endif /* !CONFIG_SLOB */
1116 * To avoid unnecessary overhead, we pass through large allocation requests
1117 * directly to the page allocator. We use __GFP_COMP, because we will need to
1118 * know the allocation order to free the pages properly in kfree.
1120 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1125 flags |= __GFP_COMP;
1126 page = alloc_pages(flags, order);
1127 ret = page ? page_address(page) : NULL;
1128 kmemleak_alloc(ret, size, 1, flags);
1129 kasan_kmalloc_large(ret, size, flags);
1132 EXPORT_SYMBOL(kmalloc_order);
1134 #ifdef CONFIG_TRACING
1135 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1137 void *ret = kmalloc_order(size, flags, order);
1138 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1141 EXPORT_SYMBOL(kmalloc_order_trace);
1144 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1145 /* Randomize a generic freelist */
1146 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1152 for (i = 0; i < count; i++)
1155 /* Fisher-Yates shuffle */
1156 for (i = count - 1; i > 0; i--) {
1157 rand = prandom_u32_state(state);
1159 swap(list[i], list[rand]);
1163 /* Create a random sequence per cache */
1164 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1167 struct rnd_state state;
1169 if (count < 2 || cachep->random_seq)
1172 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1173 if (!cachep->random_seq)
1176 /* Get best entropy at this stage of boot */
1177 prandom_seed_state(&state, get_random_long());
1179 freelist_randomize(&state, cachep->random_seq, count);
1183 /* Destroy the per-cache random freelist sequence */
1184 void cache_random_seq_destroy(struct kmem_cache *cachep)
1186 kfree(cachep->random_seq);
1187 cachep->random_seq = NULL;
1189 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1191 #ifdef CONFIG_SLABINFO
1194 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1196 #define SLABINFO_RIGHTS S_IRUSR
1199 static void print_slabinfo_header(struct seq_file *m)
1202 * Output format version, so at least we can change it
1203 * without _too_ many complaints.
1205 #ifdef CONFIG_DEBUG_SLAB
1206 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1208 seq_puts(m, "slabinfo - version: 2.1\n");
1210 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1211 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1212 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1213 #ifdef CONFIG_DEBUG_SLAB
1214 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1215 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1220 void *slab_start(struct seq_file *m, loff_t *pos)
1222 mutex_lock(&slab_mutex);
1223 return seq_list_start(&slab_root_caches, *pos);
1226 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1228 return seq_list_next(p, &slab_root_caches, pos);
1231 void slab_stop(struct seq_file *m, void *p)
1233 mutex_unlock(&slab_mutex);
1237 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1239 struct kmem_cache *c;
1240 struct slabinfo sinfo;
1242 if (!is_root_cache(s))
1245 for_each_memcg_cache(c, s) {
1246 memset(&sinfo, 0, sizeof(sinfo));
1247 get_slabinfo(c, &sinfo);
1249 info->active_slabs += sinfo.active_slabs;
1250 info->num_slabs += sinfo.num_slabs;
1251 info->shared_avail += sinfo.shared_avail;
1252 info->active_objs += sinfo.active_objs;
1253 info->num_objs += sinfo.num_objs;
1257 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1259 struct slabinfo sinfo;
1261 memset(&sinfo, 0, sizeof(sinfo));
1262 get_slabinfo(s, &sinfo);
1264 memcg_accumulate_slabinfo(s, &sinfo);
1266 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1267 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1268 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1270 seq_printf(m, " : tunables %4u %4u %4u",
1271 sinfo.limit, sinfo.batchcount, sinfo.shared);
1272 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1273 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1274 slabinfo_show_stats(m, s);
1278 static int slab_show(struct seq_file *m, void *p)
1280 struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1282 if (p == slab_root_caches.next)
1283 print_slabinfo_header(m);
1288 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1289 void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1291 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1293 mutex_lock(&slab_mutex);
1294 return seq_list_start(&memcg->kmem_caches, *pos);
1297 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1299 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1301 return seq_list_next(p, &memcg->kmem_caches, pos);
1304 void memcg_slab_stop(struct seq_file *m, void *p)
1306 mutex_unlock(&slab_mutex);
1309 int memcg_slab_show(struct seq_file *m, void *p)
1311 struct kmem_cache *s = list_entry(p, struct kmem_cache,
1312 memcg_params.kmem_caches_node);
1313 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1315 if (p == memcg->kmem_caches.next)
1316 print_slabinfo_header(m);
1323 * slabinfo_op - iterator that generates /proc/slabinfo
1332 * num-pages-per-slab
1333 * + further values on SMP and with statistics enabled
1335 static const struct seq_operations slabinfo_op = {
1336 .start = slab_start,
1342 static int slabinfo_open(struct inode *inode, struct file *file)
1344 return seq_open(file, &slabinfo_op);
1347 static const struct file_operations proc_slabinfo_operations = {
1348 .open = slabinfo_open,
1350 .write = slabinfo_write,
1351 .llseek = seq_lseek,
1352 .release = seq_release,
1355 static int __init slab_proc_init(void)
1357 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1358 &proc_slabinfo_operations);
1361 module_init(slab_proc_init);
1362 #endif /* CONFIG_SLABINFO */
1364 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1373 if (ks >= new_size) {
1374 kasan_krealloc((void *)p, new_size, flags);
1378 ret = kmalloc_track_caller(new_size, flags);
1386 * __krealloc - like krealloc() but don't free @p.
1387 * @p: object to reallocate memory for.
1388 * @new_size: how many bytes of memory are required.
1389 * @flags: the type of memory to allocate.
1391 * This function is like krealloc() except it never frees the originally
1392 * allocated buffer. Use this if you don't want to free the buffer immediately
1393 * like, for example, with RCU.
1395 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1397 if (unlikely(!new_size))
1398 return ZERO_SIZE_PTR;
1400 return __do_krealloc(p, new_size, flags);
1403 EXPORT_SYMBOL(__krealloc);
1406 * krealloc - reallocate memory. The contents will remain unchanged.
1407 * @p: object to reallocate memory for.
1408 * @new_size: how many bytes of memory are required.
1409 * @flags: the type of memory to allocate.
1411 * The contents of the object pointed to are preserved up to the
1412 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1413 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1414 * %NULL pointer, the object pointed to is freed.
1416 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1420 if (unlikely(!new_size)) {
1422 return ZERO_SIZE_PTR;
1425 ret = __do_krealloc(p, new_size, flags);
1426 if (ret && p != ret)
1431 EXPORT_SYMBOL(krealloc);
1434 * kzfree - like kfree but zero memory
1435 * @p: object to free memory of
1437 * The memory of the object @p points to is zeroed before freed.
1438 * If @p is %NULL, kzfree() does nothing.
1440 * Note: this function zeroes the whole allocated buffer which can be a good
1441 * deal bigger than the requested buffer size passed to kmalloc(). So be
1442 * careful when using this function in performance sensitive code.
1444 void kzfree(const void *p)
1447 void *mem = (void *)p;
1449 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1452 memzero_explicit(mem, ks);
1455 EXPORT_SYMBOL(kzfree);
1457 /* Tracepoints definitions. */
1458 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1459 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1460 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1461 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1462 EXPORT_TRACEPOINT_SYMBOL(kfree);
1463 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);