2 * Slab allocator functions that are independent of the allocator strategy
4 * (C) 2012 Christoph Lameter <cl@linux.com>
6 #include <linux/slab.h>
9 #include <linux/poison.h>
10 #include <linux/interrupt.h>
11 #include <linux/memory.h>
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/cpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/seq_file.h>
17 #include <linux/proc_fs.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
21 #include <linux/memcontrol.h>
22 #include <trace/events/kmem.h>
26 enum slab_state slab_state;
27 LIST_HEAD(slab_caches);
28 DEFINE_MUTEX(slab_mutex);
29 struct kmem_cache *kmem_cache;
31 #ifdef CONFIG_DEBUG_VM
32 static int kmem_cache_sanity_check(const char *name, size_t size)
34 struct kmem_cache *s = NULL;
36 if (!name || in_interrupt() || size < sizeof(void *) ||
37 size > KMALLOC_MAX_SIZE) {
38 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
42 list_for_each_entry(s, &slab_caches, list) {
47 * This happens when the module gets unloaded and doesn't
48 * destroy its slab cache and no-one else reuses the vmalloc
49 * area of the module. Print a warning.
51 res = probe_kernel_address(s->name, tmp);
53 pr_err("Slab cache with size %d has lost its name\n",
58 #if !defined(CONFIG_SLUB) || !defined(CONFIG_SLUB_DEBUG_ON)
59 if (!strcmp(s->name, name)) {
60 pr_err("%s (%s): Cache name already exists.\n",
69 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
73 static inline int kmem_cache_sanity_check(const char *name, size_t size)
79 #ifdef CONFIG_MEMCG_KMEM
80 int memcg_update_all_caches(int num_memcgs)
84 mutex_lock(&slab_mutex);
86 list_for_each_entry(s, &slab_caches, list) {
87 if (!is_root_cache(s))
90 ret = memcg_update_cache_size(s, num_memcgs);
92 * See comment in memcontrol.c, memcg_update_cache_size:
93 * Instead of freeing the memory, we'll just leave the caches
94 * up to this point in an updated state.
100 memcg_update_array_size(num_memcgs);
102 mutex_unlock(&slab_mutex);
108 * Figure out what the alignment of the objects will be given a set of
109 * flags, a user specified alignment and the size of the objects.
111 unsigned long calculate_alignment(unsigned long flags,
112 unsigned long align, unsigned long size)
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
121 if (flags & SLAB_HWCACHE_ALIGN) {
122 unsigned long ralign = cache_line_size();
123 while (size <= ralign / 2)
125 align = max(align, ralign);
128 if (align < ARCH_SLAB_MINALIGN)
129 align = ARCH_SLAB_MINALIGN;
131 return ALIGN(align, sizeof(void *));
134 static struct kmem_cache *
135 do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align,
136 unsigned long flags, void (*ctor)(void *),
137 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
139 struct kmem_cache *s;
143 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
148 s->object_size = object_size;
153 err = memcg_alloc_cache_params(memcg, s, root_cache);
157 err = __kmem_cache_create(s, flags);
162 list_add(&s->list, &slab_caches);
169 memcg_free_cache_params(s);
175 * kmem_cache_create - Create a cache.
176 * @name: A string which is used in /proc/slabinfo to identify this cache.
177 * @size: The size of objects to be created in this cache.
178 * @align: The required alignment for the objects.
180 * @ctor: A constructor for the objects.
182 * Returns a ptr to the cache on success, NULL on failure.
183 * Cannot be called within a interrupt, but can be interrupted.
184 * The @ctor is run when new pages are allocated by the cache.
188 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
189 * to catch references to uninitialised memory.
191 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
192 * for buffer overruns.
194 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
195 * cacheline. This can be beneficial if you're counting cycles as closely
199 kmem_cache_create(const char *name, size_t size, size_t align,
200 unsigned long flags, void (*ctor)(void *))
202 struct kmem_cache *s;
209 mutex_lock(&slab_mutex);
211 err = kmem_cache_sanity_check(name, size);
216 * Some allocators will constraint the set of valid flags to a subset
217 * of all flags. We expect them to define CACHE_CREATE_MASK in this
218 * case, and we'll just provide them with a sanitized version of the
221 flags &= CACHE_CREATE_MASK;
223 s = __kmem_cache_alias(name, size, align, flags, ctor);
227 cache_name = kstrdup(name, GFP_KERNEL);
233 s = do_kmem_cache_create(cache_name, size, size,
234 calculate_alignment(flags, align, size),
235 flags, ctor, NULL, NULL);
242 mutex_unlock(&slab_mutex);
248 if (flags & SLAB_PANIC)
249 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
252 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
260 EXPORT_SYMBOL(kmem_cache_create);
262 #ifdef CONFIG_MEMCG_KMEM
264 * kmem_cache_create_memcg - Create a cache for a memory cgroup.
265 * @memcg: The memory cgroup the new cache is for.
266 * @root_cache: The parent of the new cache.
268 * This function attempts to create a kmem cache that will serve allocation
269 * requests going from @memcg to @root_cache. The new cache inherits properties
272 struct kmem_cache *kmem_cache_create_memcg(struct mem_cgroup *memcg,
273 struct kmem_cache *root_cache)
275 struct kmem_cache *s = NULL;
281 mutex_lock(&slab_mutex);
283 cache_name = memcg_create_cache_name(memcg, root_cache);
287 s = do_kmem_cache_create(cache_name, root_cache->object_size,
288 root_cache->size, root_cache->align,
289 root_cache->flags, root_cache->ctor,
297 mutex_unlock(&slab_mutex);
305 static int kmem_cache_destroy_memcg_children(struct kmem_cache *s)
309 if (!s->memcg_params ||
310 !s->memcg_params->is_root_cache)
313 mutex_unlock(&slab_mutex);
314 rc = __kmem_cache_destroy_memcg_children(s);
315 mutex_lock(&slab_mutex);
320 static int kmem_cache_destroy_memcg_children(struct kmem_cache *s)
324 #endif /* CONFIG_MEMCG_KMEM */
326 void slab_kmem_cache_release(struct kmem_cache *s)
329 kmem_cache_free(kmem_cache, s);
332 void kmem_cache_destroy(struct kmem_cache *s)
337 mutex_lock(&slab_mutex);
343 if (kmem_cache_destroy_memcg_children(s) != 0)
347 if (__kmem_cache_shutdown(s) != 0) {
348 list_add(&s->list, &slab_caches);
349 printk(KERN_ERR "kmem_cache_destroy %s: "
350 "Slab cache still has objects\n", s->name);
355 mutex_unlock(&slab_mutex);
356 if (s->flags & SLAB_DESTROY_BY_RCU)
359 memcg_free_cache_params(s);
360 #ifdef SLAB_SUPPORTS_SYSFS
361 sysfs_slab_remove(s);
363 slab_kmem_cache_release(s);
368 mutex_unlock(&slab_mutex);
373 EXPORT_SYMBOL(kmem_cache_destroy);
376 * kmem_cache_shrink - Shrink a cache.
377 * @cachep: The cache to shrink.
379 * Releases as many slabs as possible for a cache.
380 * To help debugging, a zero exit status indicates all slabs were released.
382 int kmem_cache_shrink(struct kmem_cache *cachep)
388 ret = __kmem_cache_shrink(cachep);
393 EXPORT_SYMBOL(kmem_cache_shrink);
395 int slab_is_available(void)
397 return slab_state >= UP;
401 /* Create a cache during boot when no slab services are available yet */
402 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
408 s->size = s->object_size = size;
409 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
410 err = __kmem_cache_create(s, flags);
413 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
416 s->refcount = -1; /* Exempt from merging for now */
419 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
422 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
425 panic("Out of memory when creating slab %s\n", name);
427 create_boot_cache(s, name, size, flags);
428 list_add(&s->list, &slab_caches);
433 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
434 EXPORT_SYMBOL(kmalloc_caches);
436 #ifdef CONFIG_ZONE_DMA
437 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
438 EXPORT_SYMBOL(kmalloc_dma_caches);
442 * Conversion table for small slabs sizes / 8 to the index in the
443 * kmalloc array. This is necessary for slabs < 192 since we have non power
444 * of two cache sizes there. The size of larger slabs can be determined using
447 static s8 size_index[24] = {
474 static inline int size_index_elem(size_t bytes)
476 return (bytes - 1) / 8;
480 * Find the kmem_cache structure that serves a given size of
483 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
487 if (unlikely(size > KMALLOC_MAX_SIZE)) {
488 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
494 return ZERO_SIZE_PTR;
496 index = size_index[size_index_elem(size)];
498 index = fls(size - 1);
500 #ifdef CONFIG_ZONE_DMA
501 if (unlikely((flags & GFP_DMA)))
502 return kmalloc_dma_caches[index];
505 return kmalloc_caches[index];
509 * Create the kmalloc array. Some of the regular kmalloc arrays
510 * may already have been created because they were needed to
511 * enable allocations for slab creation.
513 void __init create_kmalloc_caches(unsigned long flags)
518 * Patch up the size_index table if we have strange large alignment
519 * requirements for the kmalloc array. This is only the case for
520 * MIPS it seems. The standard arches will not generate any code here.
522 * Largest permitted alignment is 256 bytes due to the way we
523 * handle the index determination for the smaller caches.
525 * Make sure that nothing crazy happens if someone starts tinkering
526 * around with ARCH_KMALLOC_MINALIGN
528 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
529 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
531 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
532 int elem = size_index_elem(i);
534 if (elem >= ARRAY_SIZE(size_index))
536 size_index[elem] = KMALLOC_SHIFT_LOW;
539 if (KMALLOC_MIN_SIZE >= 64) {
541 * The 96 byte size cache is not used if the alignment
544 for (i = 64 + 8; i <= 96; i += 8)
545 size_index[size_index_elem(i)] = 7;
549 if (KMALLOC_MIN_SIZE >= 128) {
551 * The 192 byte sized cache is not used if the alignment
552 * is 128 byte. Redirect kmalloc to use the 256 byte cache
555 for (i = 128 + 8; i <= 192; i += 8)
556 size_index[size_index_elem(i)] = 8;
558 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
559 if (!kmalloc_caches[i]) {
560 kmalloc_caches[i] = create_kmalloc_cache(NULL,
565 * Caches that are not of the two-to-the-power-of size.
566 * These have to be created immediately after the
567 * earlier power of two caches
569 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
570 kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
572 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
573 kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
576 /* Kmalloc array is now usable */
579 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
580 struct kmem_cache *s = kmalloc_caches[i];
584 n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
591 #ifdef CONFIG_ZONE_DMA
592 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
593 struct kmem_cache *s = kmalloc_caches[i];
596 int size = kmalloc_size(i);
597 char *n = kasprintf(GFP_NOWAIT,
598 "dma-kmalloc-%d", size);
601 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
602 size, SLAB_CACHE_DMA | flags);
607 #endif /* !CONFIG_SLOB */
610 * To avoid unnecessary overhead, we pass through large allocation requests
611 * directly to the page allocator. We use __GFP_COMP, because we will need to
612 * know the allocation order to free the pages properly in kfree.
614 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
620 page = alloc_kmem_pages(flags, order);
621 ret = page ? page_address(page) : NULL;
622 kmemleak_alloc(ret, size, 1, flags);
625 EXPORT_SYMBOL(kmalloc_order);
627 #ifdef CONFIG_TRACING
628 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
630 void *ret = kmalloc_order(size, flags, order);
631 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
634 EXPORT_SYMBOL(kmalloc_order_trace);
637 #ifdef CONFIG_SLABINFO
640 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
642 #define SLABINFO_RIGHTS S_IRUSR
645 void print_slabinfo_header(struct seq_file *m)
648 * Output format version, so at least we can change it
649 * without _too_ many complaints.
651 #ifdef CONFIG_DEBUG_SLAB
652 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
654 seq_puts(m, "slabinfo - version: 2.1\n");
656 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
657 "<objperslab> <pagesperslab>");
658 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
659 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
660 #ifdef CONFIG_DEBUG_SLAB
661 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
662 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
663 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
668 static void *s_start(struct seq_file *m, loff_t *pos)
672 mutex_lock(&slab_mutex);
674 print_slabinfo_header(m);
676 return seq_list_start(&slab_caches, *pos);
679 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
681 return seq_list_next(p, &slab_caches, pos);
684 void slab_stop(struct seq_file *m, void *p)
686 mutex_unlock(&slab_mutex);
690 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
692 struct kmem_cache *c;
693 struct slabinfo sinfo;
696 if (!is_root_cache(s))
699 for_each_memcg_cache_index(i) {
700 c = cache_from_memcg_idx(s, i);
704 memset(&sinfo, 0, sizeof(sinfo));
705 get_slabinfo(c, &sinfo);
707 info->active_slabs += sinfo.active_slabs;
708 info->num_slabs += sinfo.num_slabs;
709 info->shared_avail += sinfo.shared_avail;
710 info->active_objs += sinfo.active_objs;
711 info->num_objs += sinfo.num_objs;
715 int cache_show(struct kmem_cache *s, struct seq_file *m)
717 struct slabinfo sinfo;
719 memset(&sinfo, 0, sizeof(sinfo));
720 get_slabinfo(s, &sinfo);
722 memcg_accumulate_slabinfo(s, &sinfo);
724 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
725 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
726 sinfo.objects_per_slab, (1 << sinfo.cache_order));
728 seq_printf(m, " : tunables %4u %4u %4u",
729 sinfo.limit, sinfo.batchcount, sinfo.shared);
730 seq_printf(m, " : slabdata %6lu %6lu %6lu",
731 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
732 slabinfo_show_stats(m, s);
737 static int s_show(struct seq_file *m, void *p)
739 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
741 if (!is_root_cache(s))
743 return cache_show(s, m);
747 * slabinfo_op - iterator that generates /proc/slabinfo
757 * + further values on SMP and with statistics enabled
759 static const struct seq_operations slabinfo_op = {
766 static int slabinfo_open(struct inode *inode, struct file *file)
768 return seq_open(file, &slabinfo_op);
771 static const struct file_operations proc_slabinfo_operations = {
772 .open = slabinfo_open,
774 .write = slabinfo_write,
776 .release = seq_release,
779 static int __init slab_proc_init(void)
781 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
782 &proc_slabinfo_operations);
785 module_init(slab_proc_init);
786 #endif /* CONFIG_SLABINFO */