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);
163 memcg_register_cache(s);
170 memcg_free_cache_params(s);
176 * kmem_cache_create - Create a cache.
177 * @name: A string which is used in /proc/slabinfo to identify this cache.
178 * @size: The size of objects to be created in this cache.
179 * @align: The required alignment for the objects.
181 * @ctor: A constructor for the objects.
183 * Returns a ptr to the cache on success, NULL on failure.
184 * Cannot be called within a interrupt, but can be interrupted.
185 * The @ctor is run when new pages are allocated by the cache.
189 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
190 * to catch references to uninitialised memory.
192 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
193 * for buffer overruns.
195 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
196 * cacheline. This can be beneficial if you're counting cycles as closely
200 kmem_cache_create(const char *name, size_t size, size_t align,
201 unsigned long flags, void (*ctor)(void *))
203 struct kmem_cache *s;
208 mutex_lock(&slab_mutex);
210 err = kmem_cache_sanity_check(name, size);
215 * Some allocators will constraint the set of valid flags to a subset
216 * of all flags. We expect them to define CACHE_CREATE_MASK in this
217 * case, and we'll just provide them with a sanitized version of the
220 flags &= CACHE_CREATE_MASK;
222 s = __kmem_cache_alias(name, size, align, flags, ctor);
226 cache_name = kstrdup(name, GFP_KERNEL);
232 s = do_kmem_cache_create(cache_name, size, size,
233 calculate_alignment(flags, align, size),
234 flags, ctor, NULL, NULL);
241 mutex_unlock(&slab_mutex);
245 if (flags & SLAB_PANIC)
246 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
249 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
257 EXPORT_SYMBOL(kmem_cache_create);
259 #ifdef CONFIG_MEMCG_KMEM
261 * kmem_cache_create_memcg - Create a cache for a memory cgroup.
262 * @memcg: The memory cgroup the new cache is for.
263 * @root_cache: The parent of the new cache.
265 * This function attempts to create a kmem cache that will serve allocation
266 * requests going from @memcg to @root_cache. The new cache inherits properties
269 void kmem_cache_create_memcg(struct mem_cgroup *memcg, struct kmem_cache *root_cache)
271 struct kmem_cache *s;
275 mutex_lock(&slab_mutex);
278 * Since per-memcg caches are created asynchronously on first
279 * allocation (see memcg_kmem_get_cache()), several threads can try to
280 * create the same cache, but only one of them may succeed.
282 if (cache_from_memcg_idx(root_cache, memcg_cache_id(memcg)))
285 cache_name = memcg_create_cache_name(memcg, root_cache);
289 s = do_kmem_cache_create(cache_name, root_cache->object_size,
290 root_cache->size, root_cache->align,
291 root_cache->flags, root_cache->ctor,
297 mutex_unlock(&slab_mutex);
301 static int kmem_cache_destroy_memcg_children(struct kmem_cache *s)
305 if (!s->memcg_params ||
306 !s->memcg_params->is_root_cache)
309 mutex_unlock(&slab_mutex);
310 rc = __kmem_cache_destroy_memcg_children(s);
311 mutex_lock(&slab_mutex);
316 static int kmem_cache_destroy_memcg_children(struct kmem_cache *s)
320 #endif /* CONFIG_MEMCG_KMEM */
322 void slab_kmem_cache_release(struct kmem_cache *s)
325 kmem_cache_free(kmem_cache, s);
328 void kmem_cache_destroy(struct kmem_cache *s)
331 mutex_lock(&slab_mutex);
337 if (kmem_cache_destroy_memcg_children(s) != 0)
341 memcg_unregister_cache(s);
343 if (__kmem_cache_shutdown(s) != 0) {
344 list_add(&s->list, &slab_caches);
345 memcg_register_cache(s);
346 printk(KERN_ERR "kmem_cache_destroy %s: "
347 "Slab cache still has objects\n", s->name);
352 mutex_unlock(&slab_mutex);
353 if (s->flags & SLAB_DESTROY_BY_RCU)
356 memcg_free_cache_params(s);
357 #ifdef SLAB_SUPPORTS_SYSFS
358 sysfs_slab_remove(s);
360 slab_kmem_cache_release(s);
365 mutex_unlock(&slab_mutex);
369 EXPORT_SYMBOL(kmem_cache_destroy);
371 int slab_is_available(void)
373 return slab_state >= UP;
377 /* Create a cache during boot when no slab services are available yet */
378 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
384 s->size = s->object_size = size;
385 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
386 err = __kmem_cache_create(s, flags);
389 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
392 s->refcount = -1; /* Exempt from merging for now */
395 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
398 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
401 panic("Out of memory when creating slab %s\n", name);
403 create_boot_cache(s, name, size, flags);
404 list_add(&s->list, &slab_caches);
409 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
410 EXPORT_SYMBOL(kmalloc_caches);
412 #ifdef CONFIG_ZONE_DMA
413 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
414 EXPORT_SYMBOL(kmalloc_dma_caches);
418 * Conversion table for small slabs sizes / 8 to the index in the
419 * kmalloc array. This is necessary for slabs < 192 since we have non power
420 * of two cache sizes there. The size of larger slabs can be determined using
423 static s8 size_index[24] = {
450 static inline int size_index_elem(size_t bytes)
452 return (bytes - 1) / 8;
456 * Find the kmem_cache structure that serves a given size of
459 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
463 if (unlikely(size > KMALLOC_MAX_SIZE)) {
464 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
470 return ZERO_SIZE_PTR;
472 index = size_index[size_index_elem(size)];
474 index = fls(size - 1);
476 #ifdef CONFIG_ZONE_DMA
477 if (unlikely((flags & GFP_DMA)))
478 return kmalloc_dma_caches[index];
481 return kmalloc_caches[index];
485 * Create the kmalloc array. Some of the regular kmalloc arrays
486 * may already have been created because they were needed to
487 * enable allocations for slab creation.
489 void __init create_kmalloc_caches(unsigned long flags)
494 * Patch up the size_index table if we have strange large alignment
495 * requirements for the kmalloc array. This is only the case for
496 * MIPS it seems. The standard arches will not generate any code here.
498 * Largest permitted alignment is 256 bytes due to the way we
499 * handle the index determination for the smaller caches.
501 * Make sure that nothing crazy happens if someone starts tinkering
502 * around with ARCH_KMALLOC_MINALIGN
504 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
505 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
507 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
508 int elem = size_index_elem(i);
510 if (elem >= ARRAY_SIZE(size_index))
512 size_index[elem] = KMALLOC_SHIFT_LOW;
515 if (KMALLOC_MIN_SIZE >= 64) {
517 * The 96 byte size cache is not used if the alignment
520 for (i = 64 + 8; i <= 96; i += 8)
521 size_index[size_index_elem(i)] = 7;
525 if (KMALLOC_MIN_SIZE >= 128) {
527 * The 192 byte sized cache is not used if the alignment
528 * is 128 byte. Redirect kmalloc to use the 256 byte cache
531 for (i = 128 + 8; i <= 192; i += 8)
532 size_index[size_index_elem(i)] = 8;
534 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
535 if (!kmalloc_caches[i]) {
536 kmalloc_caches[i] = create_kmalloc_cache(NULL,
541 * Caches that are not of the two-to-the-power-of size.
542 * These have to be created immediately after the
543 * earlier power of two caches
545 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
546 kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
548 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
549 kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
552 /* Kmalloc array is now usable */
555 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
556 struct kmem_cache *s = kmalloc_caches[i];
560 n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
567 #ifdef CONFIG_ZONE_DMA
568 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
569 struct kmem_cache *s = kmalloc_caches[i];
572 int size = kmalloc_size(i);
573 char *n = kasprintf(GFP_NOWAIT,
574 "dma-kmalloc-%d", size);
577 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
578 size, SLAB_CACHE_DMA | flags);
583 #endif /* !CONFIG_SLOB */
586 * To avoid unnecessary overhead, we pass through large allocation requests
587 * directly to the page allocator. We use __GFP_COMP, because we will need to
588 * know the allocation order to free the pages properly in kfree.
590 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
596 page = alloc_kmem_pages(flags, order);
597 ret = page ? page_address(page) : NULL;
598 kmemleak_alloc(ret, size, 1, flags);
601 EXPORT_SYMBOL(kmalloc_order);
603 #ifdef CONFIG_TRACING
604 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
606 void *ret = kmalloc_order(size, flags, order);
607 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
610 EXPORT_SYMBOL(kmalloc_order_trace);
613 #ifdef CONFIG_SLABINFO
616 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
618 #define SLABINFO_RIGHTS S_IRUSR
621 void print_slabinfo_header(struct seq_file *m)
624 * Output format version, so at least we can change it
625 * without _too_ many complaints.
627 #ifdef CONFIG_DEBUG_SLAB
628 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
630 seq_puts(m, "slabinfo - version: 2.1\n");
632 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
633 "<objperslab> <pagesperslab>");
634 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
635 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
636 #ifdef CONFIG_DEBUG_SLAB
637 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
638 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
639 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
644 static void *s_start(struct seq_file *m, loff_t *pos)
648 mutex_lock(&slab_mutex);
650 print_slabinfo_header(m);
652 return seq_list_start(&slab_caches, *pos);
655 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
657 return seq_list_next(p, &slab_caches, pos);
660 void slab_stop(struct seq_file *m, void *p)
662 mutex_unlock(&slab_mutex);
666 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
668 struct kmem_cache *c;
669 struct slabinfo sinfo;
672 if (!is_root_cache(s))
675 for_each_memcg_cache_index(i) {
676 c = cache_from_memcg_idx(s, i);
680 memset(&sinfo, 0, sizeof(sinfo));
681 get_slabinfo(c, &sinfo);
683 info->active_slabs += sinfo.active_slabs;
684 info->num_slabs += sinfo.num_slabs;
685 info->shared_avail += sinfo.shared_avail;
686 info->active_objs += sinfo.active_objs;
687 info->num_objs += sinfo.num_objs;
691 int cache_show(struct kmem_cache *s, struct seq_file *m)
693 struct slabinfo sinfo;
695 memset(&sinfo, 0, sizeof(sinfo));
696 get_slabinfo(s, &sinfo);
698 memcg_accumulate_slabinfo(s, &sinfo);
700 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
701 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
702 sinfo.objects_per_slab, (1 << sinfo.cache_order));
704 seq_printf(m, " : tunables %4u %4u %4u",
705 sinfo.limit, sinfo.batchcount, sinfo.shared);
706 seq_printf(m, " : slabdata %6lu %6lu %6lu",
707 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
708 slabinfo_show_stats(m, s);
713 static int s_show(struct seq_file *m, void *p)
715 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
717 if (!is_root_cache(s))
719 return cache_show(s, m);
723 * slabinfo_op - iterator that generates /proc/slabinfo
733 * + further values on SMP and with statistics enabled
735 static const struct seq_operations slabinfo_op = {
742 static int slabinfo_open(struct inode *inode, struct file *file)
744 return seq_open(file, &slabinfo_op);
747 static const struct file_operations proc_slabinfo_operations = {
748 .open = slabinfo_open,
750 .write = slabinfo_write,
752 .release = seq_release,
755 static int __init slab_proc_init(void)
757 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
758 &proc_slabinfo_operations);
761 module_init(slab_proc_init);
762 #endif /* CONFIG_SLABINFO */