blob: 06aeaf091f213af43ce9f73d8beffbd754a3bf51 [file] [log] [blame]
Christoph Lameter039363f2012-07-06 15:25:10 -05001/*
2 * Slab allocator functions that are independent of the allocator strategy
3 *
4 * (C) 2012 Christoph Lameter <cl@linux.com>
5 */
6#include <linux/slab.h>
7
8#include <linux/mm.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>
Christoph Lameter20cea962012-07-06 15:25:13 -050014#include <linux/cpu.h>
15#include <linux/uaccess.h>
Glauber Costab7454ad2012-10-19 18:20:25 +040016#include <linux/seq_file.h>
17#include <linux/proc_fs.h>
Christoph Lameter039363f2012-07-06 15:25:10 -050018#include <asm/cacheflush.h>
19#include <asm/tlbflush.h>
20#include <asm/page.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080021#include <linux/memcontrol.h>
Andrey Ryabinin928cec92014-08-06 16:04:44 -070022
23#define CREATE_TRACE_POINTS
Christoph Lameterf1b6eb62013-09-04 16:35:34 +000024#include <trace/events/kmem.h>
Christoph Lameter039363f2012-07-06 15:25:10 -050025
Christoph Lameter97d06602012-07-06 15:25:11 -050026#include "slab.h"
27
28enum slab_state slab_state;
Christoph Lameter18004c52012-07-06 15:25:12 -050029LIST_HEAD(slab_caches);
30DEFINE_MUTEX(slab_mutex);
Christoph Lameter9b030cb2012-09-05 00:20:33 +000031struct kmem_cache *kmem_cache;
Christoph Lameter97d06602012-07-06 15:25:11 -050032
Joonsoo Kim07f361b2014-10-09 15:26:00 -070033/*
Joonsoo Kim423c9292014-10-09 15:26:22 -070034 * Set of flags that will prevent slab merging
35 */
36#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
37 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
38 SLAB_FAILSLAB)
39
40#define SLAB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
41 SLAB_CACHE_DMA | SLAB_NOTRACK)
42
43/*
44 * Merge control. If this is set then no merging of slab caches will occur.
45 * (Could be removed. This was introduced to pacify the merge skeptics.)
46 */
47static int slab_nomerge;
48
49static int __init setup_slab_nomerge(char *str)
50{
51 slab_nomerge = 1;
52 return 1;
53}
54
55#ifdef CONFIG_SLUB
56__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
57#endif
58
59__setup("slab_nomerge", setup_slab_nomerge);
60
61/*
Joonsoo Kim07f361b2014-10-09 15:26:00 -070062 * Determine the size of a slab object
63 */
64unsigned int kmem_cache_size(struct kmem_cache *s)
65{
66 return s->object_size;
67}
68EXPORT_SYMBOL(kmem_cache_size);
69
Shuah Khan77be4b12012-08-16 00:09:46 -070070#ifdef CONFIG_DEBUG_VM
Vladimir Davydov794b1242014-04-07 15:39:26 -070071static int kmem_cache_sanity_check(const char *name, size_t size)
Shuah Khan77be4b12012-08-16 00:09:46 -070072{
73 struct kmem_cache *s = NULL;
74
75 if (!name || in_interrupt() || size < sizeof(void *) ||
76 size > KMALLOC_MAX_SIZE) {
77 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
78 return -EINVAL;
79 }
80
81 list_for_each_entry(s, &slab_caches, list) {
82 char tmp;
83 int res;
84
85 /*
86 * This happens when the module gets unloaded and doesn't
87 * destroy its slab cache and no-one else reuses the vmalloc
88 * area of the module. Print a warning.
89 */
90 res = probe_kernel_address(s->name, tmp);
91 if (res) {
92 pr_err("Slab cache with size %d has lost its name\n",
93 s->object_size);
94 continue;
95 }
Shuah Khan77be4b12012-08-16 00:09:46 -070096 }
97
98 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
99 return 0;
100}
101#else
Vladimir Davydov794b1242014-04-07 15:39:26 -0700102static inline int kmem_cache_sanity_check(const char *name, size_t size)
Shuah Khan77be4b12012-08-16 00:09:46 -0700103{
104 return 0;
105}
106#endif
107
Glauber Costa55007d82012-12-18 14:22:38 -0800108#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700109static int memcg_alloc_cache_params(struct mem_cgroup *memcg,
110 struct kmem_cache *s, struct kmem_cache *root_cache)
111{
112 size_t size;
113
114 if (!memcg_kmem_enabled())
115 return 0;
116
117 if (!memcg) {
118 size = offsetof(struct memcg_cache_params, memcg_caches);
119 size += memcg_limited_groups_array_size * sizeof(void *);
120 } else
121 size = sizeof(struct memcg_cache_params);
122
123 s->memcg_params = kzalloc(size, GFP_KERNEL);
124 if (!s->memcg_params)
125 return -ENOMEM;
126
127 if (memcg) {
128 s->memcg_params->memcg = memcg;
129 s->memcg_params->root_cache = root_cache;
130 } else
131 s->memcg_params->is_root_cache = true;
132
133 return 0;
134}
135
136static void memcg_free_cache_params(struct kmem_cache *s)
137{
138 kfree(s->memcg_params);
139}
140
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700141static int memcg_update_cache_params(struct kmem_cache *s, int num_memcgs)
142{
143 int size;
144 struct memcg_cache_params *new_params, *cur_params;
145
146 BUG_ON(!is_root_cache(s));
147
148 size = offsetof(struct memcg_cache_params, memcg_caches);
149 size += num_memcgs * sizeof(void *);
150
151 new_params = kzalloc(size, GFP_KERNEL);
152 if (!new_params)
153 return -ENOMEM;
154
155 cur_params = s->memcg_params;
156 memcpy(new_params->memcg_caches, cur_params->memcg_caches,
157 memcg_limited_groups_array_size * sizeof(void *));
158
159 new_params->is_root_cache = true;
160
161 rcu_assign_pointer(s->memcg_params, new_params);
162 if (cur_params)
163 kfree_rcu(cur_params, rcu_head);
164
165 return 0;
166}
167
Glauber Costa55007d82012-12-18 14:22:38 -0800168int memcg_update_all_caches(int num_memcgs)
169{
170 struct kmem_cache *s;
171 int ret = 0;
172 mutex_lock(&slab_mutex);
173
174 list_for_each_entry(s, &slab_caches, list) {
175 if (!is_root_cache(s))
176 continue;
177
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700178 ret = memcg_update_cache_params(s, num_memcgs);
Glauber Costa55007d82012-12-18 14:22:38 -0800179 /*
Glauber Costa55007d82012-12-18 14:22:38 -0800180 * Instead of freeing the memory, we'll just leave the caches
181 * up to this point in an updated state.
182 */
183 if (ret)
184 goto out;
185 }
186
187 memcg_update_array_size(num_memcgs);
188out:
189 mutex_unlock(&slab_mutex);
190 return ret;
191}
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700192#else
193static inline int memcg_alloc_cache_params(struct mem_cgroup *memcg,
194 struct kmem_cache *s, struct kmem_cache *root_cache)
195{
196 return 0;
197}
198
199static inline void memcg_free_cache_params(struct kmem_cache *s)
200{
201}
202#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa55007d82012-12-18 14:22:38 -0800203
Christoph Lameter039363f2012-07-06 15:25:10 -0500204/*
Joonsoo Kim423c9292014-10-09 15:26:22 -0700205 * Find a mergeable slab cache
206 */
207int slab_unmergeable(struct kmem_cache *s)
208{
209 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
210 return 1;
211
212 if (!is_root_cache(s))
213 return 1;
214
215 if (s->ctor)
216 return 1;
217
218 /*
219 * We may have set a slab to be unmergeable during bootstrap.
220 */
221 if (s->refcount < 0)
222 return 1;
223
224 return 0;
225}
226
227struct kmem_cache *find_mergeable(size_t size, size_t align,
228 unsigned long flags, const char *name, void (*ctor)(void *))
229{
230 struct kmem_cache *s;
231
232 if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
233 return NULL;
234
235 if (ctor)
236 return NULL;
237
238 size = ALIGN(size, sizeof(void *));
239 align = calculate_alignment(flags, align, size);
240 size = ALIGN(size, align);
241 flags = kmem_cache_flags(size, flags, name, NULL);
242
243 list_for_each_entry(s, &slab_caches, list) {
244 if (slab_unmergeable(s))
245 continue;
246
247 if (size > s->size)
248 continue;
249
250 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
251 continue;
252 /*
253 * Check if alignment is compatible.
254 * Courtesy of Adrian Drzewiecki
255 */
256 if ((s->size & ~(align - 1)) != s->size)
257 continue;
258
259 if (s->size - size >= sizeof(void *))
260 continue;
261
Joonsoo Kim95069ac82014-11-13 15:19:25 -0800262 if (IS_ENABLED(CONFIG_SLAB) && align &&
263 (align > s->align || s->align % align))
264 continue;
265
Joonsoo Kim423c9292014-10-09 15:26:22 -0700266 return s;
267 }
268 return NULL;
269}
270
271/*
Christoph Lameter45906852012-11-28 16:23:16 +0000272 * Figure out what the alignment of the objects will be given a set of
273 * flags, a user specified alignment and the size of the objects.
274 */
275unsigned long calculate_alignment(unsigned long flags,
276 unsigned long align, unsigned long size)
277{
278 /*
279 * If the user wants hardware cache aligned objects then follow that
280 * suggestion if the object is sufficiently large.
281 *
282 * The hardware cache alignment cannot override the specified
283 * alignment though. If that is greater then use it.
284 */
285 if (flags & SLAB_HWCACHE_ALIGN) {
286 unsigned long ralign = cache_line_size();
287 while (size <= ralign / 2)
288 ralign /= 2;
289 align = max(align, ralign);
290 }
291
292 if (align < ARCH_SLAB_MINALIGN)
293 align = ARCH_SLAB_MINALIGN;
294
295 return ALIGN(align, sizeof(void *));
296}
297
Vladimir Davydov794b1242014-04-07 15:39:26 -0700298static struct kmem_cache *
299do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align,
300 unsigned long flags, void (*ctor)(void *),
301 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
302{
303 struct kmem_cache *s;
304 int err;
305
306 err = -ENOMEM;
307 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
308 if (!s)
309 goto out;
310
311 s->name = name;
312 s->object_size = object_size;
313 s->size = size;
314 s->align = align;
315 s->ctor = ctor;
316
317 err = memcg_alloc_cache_params(memcg, s, root_cache);
318 if (err)
319 goto out_free_cache;
320
321 err = __kmem_cache_create(s, flags);
322 if (err)
323 goto out_free_cache;
324
325 s->refcount = 1;
326 list_add(&s->list, &slab_caches);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700327out:
328 if (err)
329 return ERR_PTR(err);
330 return s;
331
332out_free_cache:
333 memcg_free_cache_params(s);
334 kfree(s);
335 goto out;
336}
Christoph Lameter45906852012-11-28 16:23:16 +0000337
338/*
Christoph Lameter039363f2012-07-06 15:25:10 -0500339 * kmem_cache_create - Create a cache.
340 * @name: A string which is used in /proc/slabinfo to identify this cache.
341 * @size: The size of objects to be created in this cache.
342 * @align: The required alignment for the objects.
343 * @flags: SLAB flags
344 * @ctor: A constructor for the objects.
345 *
346 * Returns a ptr to the cache on success, NULL on failure.
347 * Cannot be called within a interrupt, but can be interrupted.
348 * The @ctor is run when new pages are allocated by the cache.
349 *
350 * The flags are
351 *
352 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
353 * to catch references to uninitialised memory.
354 *
355 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
356 * for buffer overruns.
357 *
358 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
359 * cacheline. This can be beneficial if you're counting cycles as closely
360 * as davem.
361 */
Glauber Costa2633d7a2012-12-18 14:22:34 -0800362struct kmem_cache *
Vladimir Davydov794b1242014-04-07 15:39:26 -0700363kmem_cache_create(const char *name, size_t size, size_t align,
364 unsigned long flags, void (*ctor)(void *))
Christoph Lameter039363f2012-07-06 15:25:10 -0500365{
Vladimir Davydov794b1242014-04-07 15:39:26 -0700366 struct kmem_cache *s;
367 char *cache_name;
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800368 int err;
Christoph Lameter039363f2012-07-06 15:25:10 -0500369
Pekka Enbergb9205362012-08-16 10:12:18 +0300370 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700371 get_online_mems();
372
Pekka Enbergb9205362012-08-16 10:12:18 +0300373 mutex_lock(&slab_mutex);
Christoph Lameter686d5502012-09-05 00:20:33 +0000374
Vladimir Davydov794b1242014-04-07 15:39:26 -0700375 err = kmem_cache_sanity_check(name, size);
Andrew Morton3aa24f52014-10-09 15:25:58 -0700376 if (err) {
377 s = NULL; /* suppress uninit var warning */
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800378 goto out_unlock;
Andrew Morton3aa24f52014-10-09 15:25:58 -0700379 }
Christoph Lameter686d5502012-09-05 00:20:33 +0000380
Glauber Costad8843922012-10-17 15:36:51 +0400381 /*
382 * Some allocators will constraint the set of valid flags to a subset
383 * of all flags. We expect them to define CACHE_CREATE_MASK in this
384 * case, and we'll just provide them with a sanitized version of the
385 * passed flags.
386 */
387 flags &= CACHE_CREATE_MASK;
Christoph Lameter686d5502012-09-05 00:20:33 +0000388
Vladimir Davydov794b1242014-04-07 15:39:26 -0700389 s = __kmem_cache_alias(name, size, align, flags, ctor);
390 if (s)
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800391 goto out_unlock;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800392
Vladimir Davydov794b1242014-04-07 15:39:26 -0700393 cache_name = kstrdup(name, GFP_KERNEL);
394 if (!cache_name) {
395 err = -ENOMEM;
396 goto out_unlock;
397 }
Glauber Costa2633d7a2012-12-18 14:22:34 -0800398
Vladimir Davydov794b1242014-04-07 15:39:26 -0700399 s = do_kmem_cache_create(cache_name, size, size,
400 calculate_alignment(flags, align, size),
401 flags, ctor, NULL, NULL);
402 if (IS_ERR(s)) {
403 err = PTR_ERR(s);
404 kfree(cache_name);
405 }
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800406
407out_unlock:
Christoph Lameter20cea962012-07-06 15:25:13 -0500408 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700409
410 put_online_mems();
Christoph Lameter20cea962012-07-06 15:25:13 -0500411 put_online_cpus();
412
Dave Jonesba3253c72014-01-29 14:05:48 -0800413 if (err) {
Christoph Lameter686d5502012-09-05 00:20:33 +0000414 if (flags & SLAB_PANIC)
415 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
416 name, err);
417 else {
418 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
419 name, err);
420 dump_stack();
421 }
Christoph Lameter686d5502012-09-05 00:20:33 +0000422 return NULL;
423 }
Christoph Lameter039363f2012-07-06 15:25:10 -0500424 return s;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800425}
Christoph Lameter039363f2012-07-06 15:25:10 -0500426EXPORT_SYMBOL(kmem_cache_create);
Christoph Lameter97d06602012-07-06 15:25:11 -0500427
Vladimir Davydov794b1242014-04-07 15:39:26 -0700428#ifdef CONFIG_MEMCG_KMEM
429/*
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700430 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
Vladimir Davydov794b1242014-04-07 15:39:26 -0700431 * @memcg: The memory cgroup the new cache is for.
432 * @root_cache: The parent of the new cache.
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700433 * @memcg_name: The name of the memory cgroup (used for naming the new cache).
Vladimir Davydov794b1242014-04-07 15:39:26 -0700434 *
435 * This function attempts to create a kmem cache that will serve allocation
436 * requests going from @memcg to @root_cache. The new cache inherits properties
437 * from its parent.
438 */
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700439struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700440 struct kmem_cache *root_cache,
441 const char *memcg_name)
Vladimir Davydov794b1242014-04-07 15:39:26 -0700442{
Vladimir Davydovbd673142014-06-04 16:07:40 -0700443 struct kmem_cache *s = NULL;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700444 char *cache_name;
445
446 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700447 get_online_mems();
448
Vladimir Davydov794b1242014-04-07 15:39:26 -0700449 mutex_lock(&slab_mutex);
450
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700451 cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
452 memcg_cache_id(memcg), memcg_name);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700453 if (!cache_name)
454 goto out_unlock;
455
456 s = do_kmem_cache_create(cache_name, root_cache->object_size,
457 root_cache->size, root_cache->align,
458 root_cache->flags, root_cache->ctor,
459 memcg, root_cache);
Vladimir Davydovbd673142014-06-04 16:07:40 -0700460 if (IS_ERR(s)) {
Vladimir Davydov794b1242014-04-07 15:39:26 -0700461 kfree(cache_name);
Vladimir Davydovbd673142014-06-04 16:07:40 -0700462 s = NULL;
463 }
Vladimir Davydov794b1242014-04-07 15:39:26 -0700464
465out_unlock:
466 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700467
468 put_online_mems();
Vladimir Davydov794b1242014-04-07 15:39:26 -0700469 put_online_cpus();
Vladimir Davydovbd673142014-06-04 16:07:40 -0700470
471 return s;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700472}
Vladimir Davydovb8529902014-04-07 15:39:28 -0700473
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700474static int memcg_cleanup_cache_params(struct kmem_cache *s)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700475{
476 int rc;
477
478 if (!s->memcg_params ||
479 !s->memcg_params->is_root_cache)
480 return 0;
481
482 mutex_unlock(&slab_mutex);
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700483 rc = __memcg_cleanup_cache_params(s);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700484 mutex_lock(&slab_mutex);
485
486 return rc;
487}
488#else
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700489static int memcg_cleanup_cache_params(struct kmem_cache *s)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700490{
491 return 0;
492}
Vladimir Davydov794b1242014-04-07 15:39:26 -0700493#endif /* CONFIG_MEMCG_KMEM */
494
Christoph Lameter41a21282014-05-06 12:50:08 -0700495void slab_kmem_cache_release(struct kmem_cache *s)
496{
497 kfree(s->name);
498 kmem_cache_free(kmem_cache, s);
499}
500
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000501void kmem_cache_destroy(struct kmem_cache *s)
502{
503 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700504 get_online_mems();
505
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000506 mutex_lock(&slab_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700507
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000508 s->refcount--;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700509 if (s->refcount)
510 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000511
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700512 if (memcg_cleanup_cache_params(s) != 0)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700513 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000514
Vladimir Davydovb8529902014-04-07 15:39:28 -0700515 if (__kmem_cache_shutdown(s) != 0) {
Vladimir Davydovb8529902014-04-07 15:39:28 -0700516 printk(KERN_ERR "kmem_cache_destroy %s: "
517 "Slab cache still has objects\n", s->name);
518 dump_stack();
519 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000520 }
Vladimir Davydovb8529902014-04-07 15:39:28 -0700521
Vladimir Davydov0bd62b12014-06-04 16:10:03 -0700522 list_del(&s->list);
523
Vladimir Davydovb8529902014-04-07 15:39:28 -0700524 mutex_unlock(&slab_mutex);
525 if (s->flags & SLAB_DESTROY_BY_RCU)
526 rcu_barrier();
527
528 memcg_free_cache_params(s);
Christoph Lameter41a21282014-05-06 12:50:08 -0700529#ifdef SLAB_SUPPORTS_SYSFS
530 sysfs_slab_remove(s);
531#else
532 slab_kmem_cache_release(s);
533#endif
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700534 goto out;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700535
536out_unlock:
537 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700538out:
539 put_online_mems();
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000540 put_online_cpus();
541}
542EXPORT_SYMBOL(kmem_cache_destroy);
543
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700544/**
545 * kmem_cache_shrink - Shrink a cache.
546 * @cachep: The cache to shrink.
547 *
548 * Releases as many slabs as possible for a cache.
549 * To help debugging, a zero exit status indicates all slabs were released.
550 */
551int kmem_cache_shrink(struct kmem_cache *cachep)
552{
553 int ret;
554
555 get_online_cpus();
556 get_online_mems();
557 ret = __kmem_cache_shrink(cachep);
558 put_online_mems();
559 put_online_cpus();
560 return ret;
561}
562EXPORT_SYMBOL(kmem_cache_shrink);
563
Christoph Lameter97d06602012-07-06 15:25:11 -0500564int slab_is_available(void)
565{
566 return slab_state >= UP;
567}
Glauber Costab7454ad2012-10-19 18:20:25 +0400568
Christoph Lameter45530c42012-11-28 16:23:07 +0000569#ifndef CONFIG_SLOB
570/* Create a cache during boot when no slab services are available yet */
571void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
572 unsigned long flags)
573{
574 int err;
575
576 s->name = name;
577 s->size = s->object_size = size;
Christoph Lameter45906852012-11-28 16:23:16 +0000578 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
Christoph Lameter45530c42012-11-28 16:23:07 +0000579 err = __kmem_cache_create(s, flags);
580
581 if (err)
Christoph Lameter31ba7342013-01-10 19:00:53 +0000582 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
Christoph Lameter45530c42012-11-28 16:23:07 +0000583 name, size, err);
584
585 s->refcount = -1; /* Exempt from merging for now */
586}
587
588struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
589 unsigned long flags)
590{
591 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
592
593 if (!s)
594 panic("Out of memory when creating slab %s\n", name);
595
596 create_boot_cache(s, name, size, flags);
597 list_add(&s->list, &slab_caches);
598 s->refcount = 1;
599 return s;
600}
601
Christoph Lameter9425c582013-01-10 19:12:17 +0000602struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
603EXPORT_SYMBOL(kmalloc_caches);
604
605#ifdef CONFIG_ZONE_DMA
606struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
607EXPORT_SYMBOL(kmalloc_dma_caches);
608#endif
609
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000610/*
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000611 * Conversion table for small slabs sizes / 8 to the index in the
612 * kmalloc array. This is necessary for slabs < 192 since we have non power
613 * of two cache sizes there. The size of larger slabs can be determined using
614 * fls.
615 */
616static s8 size_index[24] = {
617 3, /* 8 */
618 4, /* 16 */
619 5, /* 24 */
620 5, /* 32 */
621 6, /* 40 */
622 6, /* 48 */
623 6, /* 56 */
624 6, /* 64 */
625 1, /* 72 */
626 1, /* 80 */
627 1, /* 88 */
628 1, /* 96 */
629 7, /* 104 */
630 7, /* 112 */
631 7, /* 120 */
632 7, /* 128 */
633 2, /* 136 */
634 2, /* 144 */
635 2, /* 152 */
636 2, /* 160 */
637 2, /* 168 */
638 2, /* 176 */
639 2, /* 184 */
640 2 /* 192 */
641};
642
643static inline int size_index_elem(size_t bytes)
644{
645 return (bytes - 1) / 8;
646}
647
648/*
649 * Find the kmem_cache structure that serves a given size of
650 * allocation
651 */
652struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
653{
654 int index;
655
Joonsoo Kim9de1bc82013-08-02 11:02:42 +0900656 if (unlikely(size > KMALLOC_MAX_SIZE)) {
Sasha Levin907985f2013-06-10 15:18:00 -0400657 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
Christoph Lameter6286ae92013-05-03 15:43:18 +0000658 return NULL;
Sasha Levin907985f2013-06-10 15:18:00 -0400659 }
Christoph Lameter6286ae92013-05-03 15:43:18 +0000660
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000661 if (size <= 192) {
662 if (!size)
663 return ZERO_SIZE_PTR;
664
665 index = size_index[size_index_elem(size)];
666 } else
667 index = fls(size - 1);
668
669#ifdef CONFIG_ZONE_DMA
Joonsoo Kimb1e05412013-02-04 23:46:46 +0900670 if (unlikely((flags & GFP_DMA)))
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000671 return kmalloc_dma_caches[index];
672
673#endif
674 return kmalloc_caches[index];
675}
676
677/*
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000678 * Create the kmalloc array. Some of the regular kmalloc arrays
679 * may already have been created because they were needed to
680 * enable allocations for slab creation.
681 */
682void __init create_kmalloc_caches(unsigned long flags)
683{
684 int i;
685
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000686 /*
687 * Patch up the size_index table if we have strange large alignment
688 * requirements for the kmalloc array. This is only the case for
689 * MIPS it seems. The standard arches will not generate any code here.
690 *
691 * Largest permitted alignment is 256 bytes due to the way we
692 * handle the index determination for the smaller caches.
693 *
694 * Make sure that nothing crazy happens if someone starts tinkering
695 * around with ARCH_KMALLOC_MINALIGN
696 */
697 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
698 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
699
700 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
701 int elem = size_index_elem(i);
702
703 if (elem >= ARRAY_SIZE(size_index))
704 break;
705 size_index[elem] = KMALLOC_SHIFT_LOW;
706 }
707
708 if (KMALLOC_MIN_SIZE >= 64) {
709 /*
710 * The 96 byte size cache is not used if the alignment
711 * is 64 byte.
712 */
713 for (i = 64 + 8; i <= 96; i += 8)
714 size_index[size_index_elem(i)] = 7;
715
716 }
717
718 if (KMALLOC_MIN_SIZE >= 128) {
719 /*
720 * The 192 byte sized cache is not used if the alignment
721 * is 128 byte. Redirect kmalloc to use the 256 byte cache
722 * instead.
723 */
724 for (i = 128 + 8; i <= 192; i += 8)
725 size_index[size_index_elem(i)] = 8;
726 }
Christoph Lameter8a965b32013-05-03 18:04:18 +0000727 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
728 if (!kmalloc_caches[i]) {
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000729 kmalloc_caches[i] = create_kmalloc_cache(NULL,
730 1 << i, flags);
Christoph Lameter8a965b32013-05-03 18:04:18 +0000731 }
Chris Mason956e46e2013-05-08 15:56:28 -0400732
733 /*
734 * Caches that are not of the two-to-the-power-of size.
735 * These have to be created immediately after the
736 * earlier power of two caches
737 */
738 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
739 kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
740
741 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
742 kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
Christoph Lameter8a965b32013-05-03 18:04:18 +0000743 }
744
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000745 /* Kmalloc array is now usable */
746 slab_state = UP;
747
748 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
749 struct kmem_cache *s = kmalloc_caches[i];
750 char *n;
751
752 if (s) {
753 n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
754
755 BUG_ON(!n);
756 s->name = n;
757 }
758 }
759
760#ifdef CONFIG_ZONE_DMA
761 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
762 struct kmem_cache *s = kmalloc_caches[i];
763
764 if (s) {
765 int size = kmalloc_size(i);
766 char *n = kasprintf(GFP_NOWAIT,
767 "dma-kmalloc-%d", size);
768
769 BUG_ON(!n);
770 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
771 size, SLAB_CACHE_DMA | flags);
772 }
773 }
774#endif
775}
Christoph Lameter45530c42012-11-28 16:23:07 +0000776#endif /* !CONFIG_SLOB */
777
Vladimir Davydovcea371f2014-06-04 16:07:04 -0700778/*
779 * To avoid unnecessary overhead, we pass through large allocation requests
780 * directly to the page allocator. We use __GFP_COMP, because we will need to
781 * know the allocation order to free the pages properly in kfree.
782 */
Vladimir Davydov52383432014-06-04 16:06:39 -0700783void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
784{
785 void *ret;
786 struct page *page;
787
788 flags |= __GFP_COMP;
789 page = alloc_kmem_pages(flags, order);
790 ret = page ? page_address(page) : NULL;
791 kmemleak_alloc(ret, size, 1, flags);
792 return ret;
793}
794EXPORT_SYMBOL(kmalloc_order);
795
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000796#ifdef CONFIG_TRACING
797void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
798{
799 void *ret = kmalloc_order(size, flags, order);
800 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
801 return ret;
802}
803EXPORT_SYMBOL(kmalloc_order_trace);
804#endif
Christoph Lameter45530c42012-11-28 16:23:07 +0000805
Glauber Costab7454ad2012-10-19 18:20:25 +0400806#ifdef CONFIG_SLABINFO
Wanpeng Lie9b4db22013-07-04 08:33:24 +0800807
808#ifdef CONFIG_SLAB
809#define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
810#else
811#define SLABINFO_RIGHTS S_IRUSR
812#endif
813
Glauber Costa749c5412012-12-18 14:23:01 -0800814void print_slabinfo_header(struct seq_file *m)
Glauber Costabcee6e22012-10-19 18:20:26 +0400815{
816 /*
817 * Output format version, so at least we can change it
818 * without _too_ many complaints.
819 */
820#ifdef CONFIG_DEBUG_SLAB
821 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
822#else
823 seq_puts(m, "slabinfo - version: 2.1\n");
824#endif
825 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
826 "<objperslab> <pagesperslab>");
827 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
828 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
829#ifdef CONFIG_DEBUG_SLAB
830 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
831 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
832 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
833#endif
834 seq_putc(m, '\n');
835}
836
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800837void *slab_start(struct seq_file *m, loff_t *pos)
Glauber Costab7454ad2012-10-19 18:20:25 +0400838{
Glauber Costab7454ad2012-10-19 18:20:25 +0400839 mutex_lock(&slab_mutex);
Glauber Costab7454ad2012-10-19 18:20:25 +0400840 return seq_list_start(&slab_caches, *pos);
841}
842
Wanpeng Li276a2432013-07-08 08:08:28 +0800843void *slab_next(struct seq_file *m, void *p, loff_t *pos)
Glauber Costab7454ad2012-10-19 18:20:25 +0400844{
845 return seq_list_next(p, &slab_caches, pos);
846}
847
Wanpeng Li276a2432013-07-08 08:08:28 +0800848void slab_stop(struct seq_file *m, void *p)
Glauber Costab7454ad2012-10-19 18:20:25 +0400849{
850 mutex_unlock(&slab_mutex);
851}
852
Glauber Costa749c5412012-12-18 14:23:01 -0800853static void
854memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
Glauber Costab7454ad2012-10-19 18:20:25 +0400855{
Glauber Costa749c5412012-12-18 14:23:01 -0800856 struct kmem_cache *c;
857 struct slabinfo sinfo;
858 int i;
859
860 if (!is_root_cache(s))
861 return;
862
863 for_each_memcg_cache_index(i) {
Qiang Huang2ade4de2013-11-12 15:08:23 -0800864 c = cache_from_memcg_idx(s, i);
Glauber Costa749c5412012-12-18 14:23:01 -0800865 if (!c)
866 continue;
867
868 memset(&sinfo, 0, sizeof(sinfo));
869 get_slabinfo(c, &sinfo);
870
871 info->active_slabs += sinfo.active_slabs;
872 info->num_slabs += sinfo.num_slabs;
873 info->shared_avail += sinfo.shared_avail;
874 info->active_objs += sinfo.active_objs;
875 info->num_objs += sinfo.num_objs;
876 }
877}
878
879int cache_show(struct kmem_cache *s, struct seq_file *m)
880{
Glauber Costa0d7561c2012-10-19 18:20:27 +0400881 struct slabinfo sinfo;
882
883 memset(&sinfo, 0, sizeof(sinfo));
884 get_slabinfo(s, &sinfo);
885
Glauber Costa749c5412012-12-18 14:23:01 -0800886 memcg_accumulate_slabinfo(s, &sinfo);
887
Glauber Costa0d7561c2012-10-19 18:20:27 +0400888 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
Glauber Costa749c5412012-12-18 14:23:01 -0800889 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
Glauber Costa0d7561c2012-10-19 18:20:27 +0400890 sinfo.objects_per_slab, (1 << sinfo.cache_order));
891
892 seq_printf(m, " : tunables %4u %4u %4u",
893 sinfo.limit, sinfo.batchcount, sinfo.shared);
894 seq_printf(m, " : slabdata %6lu %6lu %6lu",
895 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
896 slabinfo_show_stats(m, s);
897 seq_putc(m, '\n');
898 return 0;
Glauber Costab7454ad2012-10-19 18:20:25 +0400899}
900
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800901static int slab_show(struct seq_file *m, void *p)
Glauber Costa749c5412012-12-18 14:23:01 -0800902{
903 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
904
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800905 if (p == slab_caches.next)
906 print_slabinfo_header(m);
Glauber Costa749c5412012-12-18 14:23:01 -0800907 if (!is_root_cache(s))
908 return 0;
909 return cache_show(s, m);
910}
911
Glauber Costab7454ad2012-10-19 18:20:25 +0400912/*
913 * slabinfo_op - iterator that generates /proc/slabinfo
914 *
915 * Output layout:
916 * cache-name
917 * num-active-objs
918 * total-objs
919 * object size
920 * num-active-slabs
921 * total-slabs
922 * num-pages-per-slab
923 * + further values on SMP and with statistics enabled
924 */
925static const struct seq_operations slabinfo_op = {
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800926 .start = slab_start,
Wanpeng Li276a2432013-07-08 08:08:28 +0800927 .next = slab_next,
928 .stop = slab_stop,
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800929 .show = slab_show,
Glauber Costab7454ad2012-10-19 18:20:25 +0400930};
931
932static int slabinfo_open(struct inode *inode, struct file *file)
933{
934 return seq_open(file, &slabinfo_op);
935}
936
937static const struct file_operations proc_slabinfo_operations = {
938 .open = slabinfo_open,
939 .read = seq_read,
940 .write = slabinfo_write,
941 .llseek = seq_lseek,
942 .release = seq_release,
943};
944
945static int __init slab_proc_init(void)
946{
Wanpeng Lie9b4db22013-07-04 08:33:24 +0800947 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
948 &proc_slabinfo_operations);
Glauber Costab7454ad2012-10-19 18:20:25 +0400949 return 0;
950}
951module_init(slab_proc_init);
952#endif /* CONFIG_SLABINFO */
Andrey Ryabinin928cec92014-08-06 16:04:44 -0700953
954static __always_inline void *__do_krealloc(const void *p, size_t new_size,
955 gfp_t flags)
956{
957 void *ret;
958 size_t ks = 0;
959
960 if (p)
961 ks = ksize(p);
962
963 if (ks >= new_size)
964 return (void *)p;
965
966 ret = kmalloc_track_caller(new_size, flags);
967 if (ret && p)
968 memcpy(ret, p, ks);
969
970 return ret;
971}
972
973/**
974 * __krealloc - like krealloc() but don't free @p.
975 * @p: object to reallocate memory for.
976 * @new_size: how many bytes of memory are required.
977 * @flags: the type of memory to allocate.
978 *
979 * This function is like krealloc() except it never frees the originally
980 * allocated buffer. Use this if you don't want to free the buffer immediately
981 * like, for example, with RCU.
982 */
983void *__krealloc(const void *p, size_t new_size, gfp_t flags)
984{
985 if (unlikely(!new_size))
986 return ZERO_SIZE_PTR;
987
988 return __do_krealloc(p, new_size, flags);
989
990}
991EXPORT_SYMBOL(__krealloc);
992
993/**
994 * krealloc - reallocate memory. The contents will remain unchanged.
995 * @p: object to reallocate memory for.
996 * @new_size: how many bytes of memory are required.
997 * @flags: the type of memory to allocate.
998 *
999 * The contents of the object pointed to are preserved up to the
1000 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1001 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1002 * %NULL pointer, the object pointed to is freed.
1003 */
1004void *krealloc(const void *p, size_t new_size, gfp_t flags)
1005{
1006 void *ret;
1007
1008 if (unlikely(!new_size)) {
1009 kfree(p);
1010 return ZERO_SIZE_PTR;
1011 }
1012
1013 ret = __do_krealloc(p, new_size, flags);
1014 if (ret && p != ret)
1015 kfree(p);
1016
1017 return ret;
1018}
1019EXPORT_SYMBOL(krealloc);
1020
1021/**
1022 * kzfree - like kfree but zero memory
1023 * @p: object to free memory of
1024 *
1025 * The memory of the object @p points to is zeroed before freed.
1026 * If @p is %NULL, kzfree() does nothing.
1027 *
1028 * Note: this function zeroes the whole allocated buffer which can be a good
1029 * deal bigger than the requested buffer size passed to kmalloc(). So be
1030 * careful when using this function in performance sensitive code.
1031 */
1032void kzfree(const void *p)
1033{
1034 size_t ks;
1035 void *mem = (void *)p;
1036
1037 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1038 return;
1039 ks = ksize(mem);
1040 memset(mem, 0, ks);
1041 kfree(mem);
1042}
1043EXPORT_SYMBOL(kzfree);
1044
1045/* Tracepoints definitions. */
1046EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1047EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1048EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1049EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1050EXPORT_TRACEPOINT_SYMBOL(kfree);
1051EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);