blob: 3e5f8f29c28640e44af5f5f9d1c3553986064588 [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 Davydovf7ce3192015-02-12 14:59:20 -0800109void slab_init_memcg_params(struct kmem_cache *s)
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700110{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800111 s->memcg_params.is_root_cache = true;
Vladimir Davydov426589f2015-02-12 14:59:23 -0800112 INIT_LIST_HEAD(&s->memcg_params.list);
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800113 RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
114}
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700115
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800116static int init_memcg_params(struct kmem_cache *s,
117 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
118{
119 struct memcg_cache_array *arr;
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700120
121 if (memcg) {
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800122 s->memcg_params.is_root_cache = false;
123 s->memcg_params.memcg = memcg;
124 s->memcg_params.root_cache = root_cache;
125 return 0;
126 }
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700127
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800128 slab_init_memcg_params(s);
129
130 if (!memcg_nr_cache_ids)
131 return 0;
132
133 arr = kzalloc(sizeof(struct memcg_cache_array) +
134 memcg_nr_cache_ids * sizeof(void *),
135 GFP_KERNEL);
136 if (!arr)
137 return -ENOMEM;
138
139 RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700140 return 0;
141}
142
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800143static void destroy_memcg_params(struct kmem_cache *s)
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700144{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800145 if (is_root_cache(s))
146 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700147}
148
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800149static int update_memcg_params(struct kmem_cache *s, int new_array_size)
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700150{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800151 struct memcg_cache_array *old, *new;
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700152
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800153 if (!is_root_cache(s))
154 return 0;
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700155
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800156 new = kzalloc(sizeof(struct memcg_cache_array) +
157 new_array_size * sizeof(void *), GFP_KERNEL);
158 if (!new)
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700159 return -ENOMEM;
160
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800161 old = rcu_dereference_protected(s->memcg_params.memcg_caches,
162 lockdep_is_held(&slab_mutex));
163 if (old)
164 memcpy(new->entries, old->entries,
165 memcg_nr_cache_ids * sizeof(void *));
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700166
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800167 rcu_assign_pointer(s->memcg_params.memcg_caches, new);
168 if (old)
169 kfree_rcu(old, rcu);
Vladimir Davydov6f817f42014-10-09 15:28:47 -0700170 return 0;
171}
172
Glauber Costa55007d82012-12-18 14:22:38 -0800173int memcg_update_all_caches(int num_memcgs)
174{
175 struct kmem_cache *s;
176 int ret = 0;
Glauber Costa55007d82012-12-18 14:22:38 -0800177
Vladimir Davydov05257a12015-02-12 14:59:01 -0800178 mutex_lock(&slab_mutex);
Glauber Costa55007d82012-12-18 14:22:38 -0800179 list_for_each_entry(s, &slab_caches, list) {
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800180 ret = update_memcg_params(s, num_memcgs);
Glauber Costa55007d82012-12-18 14:22:38 -0800181 /*
Glauber Costa55007d82012-12-18 14:22:38 -0800182 * Instead of freeing the memory, we'll just leave the caches
183 * up to this point in an updated state.
184 */
185 if (ret)
Vladimir Davydov05257a12015-02-12 14:59:01 -0800186 break;
Glauber Costa55007d82012-12-18 14:22:38 -0800187 }
Glauber Costa55007d82012-12-18 14:22:38 -0800188 mutex_unlock(&slab_mutex);
189 return ret;
190}
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700191#else
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800192static inline int init_memcg_params(struct kmem_cache *s,
193 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700194{
195 return 0;
196}
197
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800198static inline void destroy_memcg_params(struct kmem_cache *s)
Vladimir Davydov33a690c2014-10-09 15:28:43 -0700199{
200}
201#endif /* CONFIG_MEMCG_KMEM */
Glauber Costa55007d82012-12-18 14:22:38 -0800202
Christoph Lameter039363f2012-07-06 15:25:10 -0500203/*
Joonsoo Kim423c9292014-10-09 15:26:22 -0700204 * Find a mergeable slab cache
205 */
206int slab_unmergeable(struct kmem_cache *s)
207{
208 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
209 return 1;
210
211 if (!is_root_cache(s))
212 return 1;
213
214 if (s->ctor)
215 return 1;
216
217 /*
218 * We may have set a slab to be unmergeable during bootstrap.
219 */
220 if (s->refcount < 0)
221 return 1;
222
223 return 0;
224}
225
226struct kmem_cache *find_mergeable(size_t size, size_t align,
227 unsigned long flags, const char *name, void (*ctor)(void *))
228{
229 struct kmem_cache *s;
230
231 if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
232 return NULL;
233
234 if (ctor)
235 return NULL;
236
237 size = ALIGN(size, sizeof(void *));
238 align = calculate_alignment(flags, align, size);
239 size = ALIGN(size, align);
240 flags = kmem_cache_flags(size, flags, name, NULL);
241
Joonsoo Kim54362052014-12-10 15:42:18 -0800242 list_for_each_entry_reverse(s, &slab_caches, list) {
Joonsoo Kim423c9292014-10-09 15:26:22 -0700243 if (slab_unmergeable(s))
244 continue;
245
246 if (size > s->size)
247 continue;
248
249 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
250 continue;
251 /*
252 * Check if alignment is compatible.
253 * Courtesy of Adrian Drzewiecki
254 */
255 if ((s->size & ~(align - 1)) != s->size)
256 continue;
257
258 if (s->size - size >= sizeof(void *))
259 continue;
260
Joonsoo Kim95069ac82014-11-13 15:19:25 -0800261 if (IS_ENABLED(CONFIG_SLAB) && align &&
262 (align > s->align || s->align % align))
263 continue;
264
Joonsoo Kim423c9292014-10-09 15:26:22 -0700265 return s;
266 }
267 return NULL;
268}
269
270/*
Christoph Lameter45906852012-11-28 16:23:16 +0000271 * Figure out what the alignment of the objects will be given a set of
272 * flags, a user specified alignment and the size of the objects.
273 */
274unsigned long calculate_alignment(unsigned long flags,
275 unsigned long align, unsigned long size)
276{
277 /*
278 * If the user wants hardware cache aligned objects then follow that
279 * suggestion if the object is sufficiently large.
280 *
281 * The hardware cache alignment cannot override the specified
282 * alignment though. If that is greater then use it.
283 */
284 if (flags & SLAB_HWCACHE_ALIGN) {
285 unsigned long ralign = cache_line_size();
286 while (size <= ralign / 2)
287 ralign /= 2;
288 align = max(align, ralign);
289 }
290
291 if (align < ARCH_SLAB_MINALIGN)
292 align = ARCH_SLAB_MINALIGN;
293
294 return ALIGN(align, sizeof(void *));
295}
296
Vladimir Davydov794b1242014-04-07 15:39:26 -0700297static struct kmem_cache *
Andrzej Hajda3dec16e2015-02-13 14:36:38 -0800298do_kmem_cache_create(const char *name, size_t object_size, size_t size,
299 size_t align, unsigned long flags, void (*ctor)(void *),
Vladimir Davydov794b1242014-04-07 15:39:26 -0700300 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
301{
302 struct kmem_cache *s;
303 int err;
304
305 err = -ENOMEM;
306 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
307 if (!s)
308 goto out;
309
310 s->name = name;
311 s->object_size = object_size;
312 s->size = size;
313 s->align = align;
314 s->ctor = ctor;
315
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800316 err = init_memcg_params(s, memcg, root_cache);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700317 if (err)
318 goto out_free_cache;
319
320 err = __kmem_cache_create(s, flags);
321 if (err)
322 goto out_free_cache;
323
324 s->refcount = 1;
325 list_add(&s->list, &slab_caches);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700326out:
327 if (err)
328 return ERR_PTR(err);
329 return s;
330
331out_free_cache:
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800332 destroy_memcg_params(s);
Vaishali Thakkar7c4da062015-02-10 14:09:40 -0800333 kmem_cache_free(kmem_cache, s);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700334 goto out;
335}
Christoph Lameter45906852012-11-28 16:23:16 +0000336
337/*
Christoph Lameter039363f2012-07-06 15:25:10 -0500338 * kmem_cache_create - Create a cache.
339 * @name: A string which is used in /proc/slabinfo to identify this cache.
340 * @size: The size of objects to be created in this cache.
341 * @align: The required alignment for the objects.
342 * @flags: SLAB flags
343 * @ctor: A constructor for the objects.
344 *
345 * Returns a ptr to the cache on success, NULL on failure.
346 * Cannot be called within a interrupt, but can be interrupted.
347 * The @ctor is run when new pages are allocated by the cache.
348 *
349 * The flags are
350 *
351 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
352 * to catch references to uninitialised memory.
353 *
354 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
355 * for buffer overruns.
356 *
357 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
358 * cacheline. This can be beneficial if you're counting cycles as closely
359 * as davem.
360 */
Glauber Costa2633d7a2012-12-18 14:22:34 -0800361struct kmem_cache *
Vladimir Davydov794b1242014-04-07 15:39:26 -0700362kmem_cache_create(const char *name, size_t size, size_t align,
363 unsigned long flags, void (*ctor)(void *))
Christoph Lameter039363f2012-07-06 15:25:10 -0500364{
Vladimir Davydov794b1242014-04-07 15:39:26 -0700365 struct kmem_cache *s;
Andrzej Hajda3dec16e2015-02-13 14:36:38 -0800366 const char *cache_name;
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800367 int err;
Christoph Lameter039363f2012-07-06 15:25:10 -0500368
Pekka Enbergb9205362012-08-16 10:12:18 +0300369 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700370 get_online_mems();
Vladimir Davydov05257a12015-02-12 14:59:01 -0800371 memcg_get_cache_ids();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700372
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
Andrzej Hajda3dec16e2015-02-13 14:36:38 -0800393 cache_name = kstrdup_const(name, GFP_KERNEL);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700394 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);
Andrzej Hajda3dec16e2015-02-13 14:36:38 -0800404 kfree_const(cache_name);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700405 }
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
Vladimir Davydov05257a12015-02-12 14:59:01 -0800410 memcg_put_cache_ids();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700411 put_online_mems();
Christoph Lameter20cea962012-07-06 15:25:13 -0500412 put_online_cpus();
413
Dave Jonesba3253c72014-01-29 14:05:48 -0800414 if (err) {
Christoph Lameter686d5502012-09-05 00:20:33 +0000415 if (flags & SLAB_PANIC)
416 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
417 name, err);
418 else {
419 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
420 name, err);
421 dump_stack();
422 }
Christoph Lameter686d5502012-09-05 00:20:33 +0000423 return NULL;
424 }
Christoph Lameter039363f2012-07-06 15:25:10 -0500425 return s;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800426}
Christoph Lameter039363f2012-07-06 15:25:10 -0500427EXPORT_SYMBOL(kmem_cache_create);
Christoph Lameter97d06602012-07-06 15:25:11 -0500428
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800429static int do_kmem_cache_shutdown(struct kmem_cache *s,
430 struct list_head *release, bool *need_rcu_barrier)
431{
432 if (__kmem_cache_shutdown(s) != 0) {
433 printk(KERN_ERR "kmem_cache_destroy %s: "
434 "Slab cache still has objects\n", s->name);
435 dump_stack();
436 return -EBUSY;
437 }
438
439 if (s->flags & SLAB_DESTROY_BY_RCU)
440 *need_rcu_barrier = true;
441
442#ifdef CONFIG_MEMCG_KMEM
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800443 if (!is_root_cache(s))
Vladimir Davydov426589f2015-02-12 14:59:23 -0800444 list_del(&s->memcg_params.list);
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800445#endif
446 list_move(&s->list, release);
447 return 0;
448}
449
450static void do_kmem_cache_release(struct list_head *release,
451 bool need_rcu_barrier)
452{
453 struct kmem_cache *s, *s2;
454
455 if (need_rcu_barrier)
456 rcu_barrier();
457
458 list_for_each_entry_safe(s, s2, release, list) {
459#ifdef SLAB_SUPPORTS_SYSFS
460 sysfs_slab_remove(s);
461#else
462 slab_kmem_cache_release(s);
463#endif
464 }
465}
466
Vladimir Davydov794b1242014-04-07 15:39:26 -0700467#ifdef CONFIG_MEMCG_KMEM
468/*
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700469 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
Vladimir Davydov794b1242014-04-07 15:39:26 -0700470 * @memcg: The memory cgroup the new cache is for.
471 * @root_cache: The parent of the new cache.
472 *
473 * This function attempts to create a kmem cache that will serve allocation
474 * requests going from @memcg to @root_cache. The new cache inherits properties
475 * from its parent.
476 */
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800477void memcg_create_kmem_cache(struct mem_cgroup *memcg,
478 struct kmem_cache *root_cache)
Vladimir Davydov794b1242014-04-07 15:39:26 -0700479{
Vladimir Davydov3e0350a2015-02-10 14:11:44 -0800480 static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
Vladimir Davydovf1008362015-02-12 14:59:29 -0800481 struct cgroup_subsys_state *css = mem_cgroup_css(memcg);
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800482 struct memcg_cache_array *arr;
Vladimir Davydovbd673142014-06-04 16:07:40 -0700483 struct kmem_cache *s = NULL;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700484 char *cache_name;
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800485 int idx;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700486
487 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700488 get_online_mems();
489
Vladimir Davydov794b1242014-04-07 15:39:26 -0700490 mutex_lock(&slab_mutex);
491
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800492 /*
493 * The memory cgroup could have been deactivated while the cache
494 * creation work was pending.
495 */
496 if (!memcg_kmem_is_active(memcg))
497 goto out_unlock;
498
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800499 idx = memcg_cache_id(memcg);
500 arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
501 lockdep_is_held(&slab_mutex));
502
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800503 /*
504 * Since per-memcg caches are created asynchronously on first
505 * allocation (see memcg_kmem_get_cache()), several threads can try to
506 * create the same cache, but only one of them may succeed.
507 */
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800508 if (arr->entries[idx])
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800509 goto out_unlock;
510
Vladimir Davydovf1008362015-02-12 14:59:29 -0800511 cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700512 cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
Vladimir Davydovf1008362015-02-12 14:59:29 -0800513 css->id, memcg_name_buf);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700514 if (!cache_name)
515 goto out_unlock;
516
517 s = do_kmem_cache_create(cache_name, root_cache->object_size,
518 root_cache->size, root_cache->align,
519 root_cache->flags, root_cache->ctor,
520 memcg, root_cache);
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800521 /*
522 * If we could not create a memcg cache, do not complain, because
523 * that's not critical at all as we can always proceed with the root
524 * cache.
525 */
Vladimir Davydovbd673142014-06-04 16:07:40 -0700526 if (IS_ERR(s)) {
Vladimir Davydov794b1242014-04-07 15:39:26 -0700527 kfree(cache_name);
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800528 goto out_unlock;
Vladimir Davydovbd673142014-06-04 16:07:40 -0700529 }
Vladimir Davydov794b1242014-04-07 15:39:26 -0700530
Vladimir Davydov426589f2015-02-12 14:59:23 -0800531 list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
532
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800533 /*
534 * Since readers won't lock (see cache_from_memcg_idx()), we need a
535 * barrier here to ensure nobody will see the kmem_cache partially
536 * initialized.
537 */
538 smp_wmb();
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800539 arr->entries[idx] = s;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800540
Vladimir Davydov794b1242014-04-07 15:39:26 -0700541out_unlock:
542 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700543
544 put_online_mems();
Vladimir Davydov794b1242014-04-07 15:39:26 -0700545 put_online_cpus();
546}
Vladimir Davydovb8529902014-04-07 15:39:28 -0700547
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800548void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
549{
550 int idx;
551 struct memcg_cache_array *arr;
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800552 struct kmem_cache *s, *c;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800553
554 idx = memcg_cache_id(memcg);
555
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800556 get_online_cpus();
557 get_online_mems();
558
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800559 mutex_lock(&slab_mutex);
560 list_for_each_entry(s, &slab_caches, list) {
561 if (!is_root_cache(s))
562 continue;
563
564 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
565 lockdep_is_held(&slab_mutex));
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800566 c = arr->entries[idx];
567 if (!c)
568 continue;
569
570 __kmem_cache_shrink(c, true);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800571 arr->entries[idx] = NULL;
572 }
573 mutex_unlock(&slab_mutex);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800574
575 put_online_mems();
576 put_online_cpus();
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800577}
578
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800579void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700580{
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800581 LIST_HEAD(release);
582 bool need_rcu_barrier = false;
583 struct kmem_cache *s, *s2;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700584
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800585 get_online_cpus();
586 get_online_mems();
Vladimir Davydovb8529902014-04-07 15:39:28 -0700587
Vladimir Davydovb8529902014-04-07 15:39:28 -0700588 mutex_lock(&slab_mutex);
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800589 list_for_each_entry_safe(s, s2, &slab_caches, list) {
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800590 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800591 continue;
592 /*
593 * The cgroup is about to be freed and therefore has no charges
594 * left. Hence, all its caches must be empty by now.
595 */
596 BUG_ON(do_kmem_cache_shutdown(s, &release, &need_rcu_barrier));
597 }
598 mutex_unlock(&slab_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700599
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800600 put_online_mems();
601 put_online_cpus();
602
603 do_kmem_cache_release(&release, need_rcu_barrier);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700604}
Vladimir Davydov794b1242014-04-07 15:39:26 -0700605#endif /* CONFIG_MEMCG_KMEM */
606
Christoph Lameter41a21282014-05-06 12:50:08 -0700607void slab_kmem_cache_release(struct kmem_cache *s)
608{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800609 destroy_memcg_params(s);
Andrzej Hajda3dec16e2015-02-13 14:36:38 -0800610 kfree_const(s->name);
Christoph Lameter41a21282014-05-06 12:50:08 -0700611 kmem_cache_free(kmem_cache, s);
612}
613
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000614void kmem_cache_destroy(struct kmem_cache *s)
615{
Vladimir Davydov426589f2015-02-12 14:59:23 -0800616 struct kmem_cache *c, *c2;
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800617 LIST_HEAD(release);
618 bool need_rcu_barrier = false;
619 bool busy = false;
620
Vladimir Davydov426589f2015-02-12 14:59:23 -0800621 BUG_ON(!is_root_cache(s));
622
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000623 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700624 get_online_mems();
625
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000626 mutex_lock(&slab_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700627
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000628 s->refcount--;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700629 if (s->refcount)
630 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000631
Vladimir Davydov426589f2015-02-12 14:59:23 -0800632 for_each_memcg_cache_safe(c, c2, s) {
633 if (do_kmem_cache_shutdown(c, &release, &need_rcu_barrier))
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800634 busy = true;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000635 }
Vladimir Davydovb8529902014-04-07 15:39:28 -0700636
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800637 if (!busy)
638 do_kmem_cache_shutdown(s, &release, &need_rcu_barrier);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700639
640out_unlock:
641 mutex_unlock(&slab_mutex);
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800642
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700643 put_online_mems();
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000644 put_online_cpus();
Vladimir Davydovd5b3cf72015-02-10 14:11:47 -0800645
646 do_kmem_cache_release(&release, need_rcu_barrier);
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000647}
648EXPORT_SYMBOL(kmem_cache_destroy);
649
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700650/**
651 * kmem_cache_shrink - Shrink a cache.
652 * @cachep: The cache to shrink.
653 *
654 * Releases as many slabs as possible for a cache.
655 * To help debugging, a zero exit status indicates all slabs were released.
656 */
657int kmem_cache_shrink(struct kmem_cache *cachep)
658{
659 int ret;
660
661 get_online_cpus();
662 get_online_mems();
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800663 ret = __kmem_cache_shrink(cachep, false);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700664 put_online_mems();
665 put_online_cpus();
666 return ret;
667}
668EXPORT_SYMBOL(kmem_cache_shrink);
669
Christoph Lameter97d06602012-07-06 15:25:11 -0500670int slab_is_available(void)
671{
672 return slab_state >= UP;
673}
Glauber Costab7454ad2012-10-19 18:20:25 +0400674
Christoph Lameter45530c42012-11-28 16:23:07 +0000675#ifndef CONFIG_SLOB
676/* Create a cache during boot when no slab services are available yet */
677void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
678 unsigned long flags)
679{
680 int err;
681
682 s->name = name;
683 s->size = s->object_size = size;
Christoph Lameter45906852012-11-28 16:23:16 +0000684 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800685
686 slab_init_memcg_params(s);
687
Christoph Lameter45530c42012-11-28 16:23:07 +0000688 err = __kmem_cache_create(s, flags);
689
690 if (err)
Christoph Lameter31ba7342013-01-10 19:00:53 +0000691 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
Christoph Lameter45530c42012-11-28 16:23:07 +0000692 name, size, err);
693
694 s->refcount = -1; /* Exempt from merging for now */
695}
696
697struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
698 unsigned long flags)
699{
700 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
701
702 if (!s)
703 panic("Out of memory when creating slab %s\n", name);
704
705 create_boot_cache(s, name, size, flags);
706 list_add(&s->list, &slab_caches);
707 s->refcount = 1;
708 return s;
709}
710
Christoph Lameter9425c582013-01-10 19:12:17 +0000711struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
712EXPORT_SYMBOL(kmalloc_caches);
713
714#ifdef CONFIG_ZONE_DMA
715struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
716EXPORT_SYMBOL(kmalloc_dma_caches);
717#endif
718
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000719/*
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000720 * Conversion table for small slabs sizes / 8 to the index in the
721 * kmalloc array. This is necessary for slabs < 192 since we have non power
722 * of two cache sizes there. The size of larger slabs can be determined using
723 * fls.
724 */
725static s8 size_index[24] = {
726 3, /* 8 */
727 4, /* 16 */
728 5, /* 24 */
729 5, /* 32 */
730 6, /* 40 */
731 6, /* 48 */
732 6, /* 56 */
733 6, /* 64 */
734 1, /* 72 */
735 1, /* 80 */
736 1, /* 88 */
737 1, /* 96 */
738 7, /* 104 */
739 7, /* 112 */
740 7, /* 120 */
741 7, /* 128 */
742 2, /* 136 */
743 2, /* 144 */
744 2, /* 152 */
745 2, /* 160 */
746 2, /* 168 */
747 2, /* 176 */
748 2, /* 184 */
749 2 /* 192 */
750};
751
752static inline int size_index_elem(size_t bytes)
753{
754 return (bytes - 1) / 8;
755}
756
757/*
758 * Find the kmem_cache structure that serves a given size of
759 * allocation
760 */
761struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
762{
763 int index;
764
Joonsoo Kim9de1bc82013-08-02 11:02:42 +0900765 if (unlikely(size > KMALLOC_MAX_SIZE)) {
Sasha Levin907985f2013-06-10 15:18:00 -0400766 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
Christoph Lameter6286ae92013-05-03 15:43:18 +0000767 return NULL;
Sasha Levin907985f2013-06-10 15:18:00 -0400768 }
Christoph Lameter6286ae92013-05-03 15:43:18 +0000769
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000770 if (size <= 192) {
771 if (!size)
772 return ZERO_SIZE_PTR;
773
774 index = size_index[size_index_elem(size)];
775 } else
776 index = fls(size - 1);
777
778#ifdef CONFIG_ZONE_DMA
Joonsoo Kimb1e05412013-02-04 23:46:46 +0900779 if (unlikely((flags & GFP_DMA)))
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000780 return kmalloc_dma_caches[index];
781
782#endif
783 return kmalloc_caches[index];
784}
785
786/*
Gavin Guo4066c332015-06-24 16:55:54 -0700787 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
788 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
789 * kmalloc-67108864.
790 */
791static struct {
792 const char *name;
793 unsigned long size;
794} const kmalloc_info[] __initconst = {
795 {NULL, 0}, {"kmalloc-96", 96},
796 {"kmalloc-192", 192}, {"kmalloc-8", 8},
797 {"kmalloc-16", 16}, {"kmalloc-32", 32},
798 {"kmalloc-64", 64}, {"kmalloc-128", 128},
799 {"kmalloc-256", 256}, {"kmalloc-512", 512},
800 {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048},
801 {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192},
802 {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768},
803 {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072},
804 {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288},
805 {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152},
806 {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608},
807 {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432},
808 {"kmalloc-67108864", 67108864}
809};
810
811/*
Daniel Sanders34cc6992015-06-24 16:55:57 -0700812 * Patch up the size_index table if we have strange large alignment
813 * requirements for the kmalloc array. This is only the case for
814 * MIPS it seems. The standard arches will not generate any code here.
815 *
816 * Largest permitted alignment is 256 bytes due to the way we
817 * handle the index determination for the smaller caches.
818 *
819 * Make sure that nothing crazy happens if someone starts tinkering
820 * around with ARCH_KMALLOC_MINALIGN
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000821 */
Daniel Sanders34cc6992015-06-24 16:55:57 -0700822void __init setup_kmalloc_cache_index_table(void)
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000823{
824 int i;
825
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000826 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
827 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
828
829 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
830 int elem = size_index_elem(i);
831
832 if (elem >= ARRAY_SIZE(size_index))
833 break;
834 size_index[elem] = KMALLOC_SHIFT_LOW;
835 }
836
837 if (KMALLOC_MIN_SIZE >= 64) {
838 /*
839 * The 96 byte size cache is not used if the alignment
840 * is 64 byte.
841 */
842 for (i = 64 + 8; i <= 96; i += 8)
843 size_index[size_index_elem(i)] = 7;
844
845 }
846
847 if (KMALLOC_MIN_SIZE >= 128) {
848 /*
849 * The 192 byte sized cache is not used if the alignment
850 * is 128 byte. Redirect kmalloc to use the 256 byte cache
851 * instead.
852 */
853 for (i = 128 + 8; i <= 192; i += 8)
854 size_index[size_index_elem(i)] = 8;
855 }
Daniel Sanders34cc6992015-06-24 16:55:57 -0700856}
857
Christoph Lameterae6f2462015-06-30 09:01:11 -0500858static void __init new_kmalloc_cache(int idx, unsigned long flags)
Christoph Lametera9730fc2015-06-29 09:28:08 -0500859{
860 kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
861 kmalloc_info[idx].size, flags);
862}
863
Daniel Sanders34cc6992015-06-24 16:55:57 -0700864/*
865 * Create the kmalloc array. Some of the regular kmalloc arrays
866 * may already have been created because they were needed to
867 * enable allocations for slab creation.
868 */
869void __init create_kmalloc_caches(unsigned long flags)
870{
871 int i;
872
Christoph Lametera9730fc2015-06-29 09:28:08 -0500873 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
874 if (!kmalloc_caches[i])
875 new_kmalloc_cache(i, flags);
Chris Mason956e46e2013-05-08 15:56:28 -0400876
877 /*
Christoph Lametera9730fc2015-06-29 09:28:08 -0500878 * Caches that are not of the two-to-the-power-of size.
879 * These have to be created immediately after the
880 * earlier power of two caches
Chris Mason956e46e2013-05-08 15:56:28 -0400881 */
Christoph Lametera9730fc2015-06-29 09:28:08 -0500882 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
883 new_kmalloc_cache(1, flags);
884 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
885 new_kmalloc_cache(2, flags);
Christoph Lameter8a965b32013-05-03 18:04:18 +0000886 }
887
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000888 /* Kmalloc array is now usable */
889 slab_state = UP;
890
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000891#ifdef CONFIG_ZONE_DMA
892 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
893 struct kmem_cache *s = kmalloc_caches[i];
894
895 if (s) {
896 int size = kmalloc_size(i);
897 char *n = kasprintf(GFP_NOWAIT,
898 "dma-kmalloc-%d", size);
899
900 BUG_ON(!n);
901 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
902 size, SLAB_CACHE_DMA | flags);
903 }
904 }
905#endif
906}
Christoph Lameter45530c42012-11-28 16:23:07 +0000907#endif /* !CONFIG_SLOB */
908
Vladimir Davydovcea371f2014-06-04 16:07:04 -0700909/*
910 * To avoid unnecessary overhead, we pass through large allocation requests
911 * directly to the page allocator. We use __GFP_COMP, because we will need to
912 * know the allocation order to free the pages properly in kfree.
913 */
Vladimir Davydov52383432014-06-04 16:06:39 -0700914void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
915{
916 void *ret;
917 struct page *page;
918
919 flags |= __GFP_COMP;
920 page = alloc_kmem_pages(flags, order);
921 ret = page ? page_address(page) : NULL;
922 kmemleak_alloc(ret, size, 1, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800923 kasan_kmalloc_large(ret, size);
Vladimir Davydov52383432014-06-04 16:06:39 -0700924 return ret;
925}
926EXPORT_SYMBOL(kmalloc_order);
927
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000928#ifdef CONFIG_TRACING
929void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
930{
931 void *ret = kmalloc_order(size, flags, order);
932 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
933 return ret;
934}
935EXPORT_SYMBOL(kmalloc_order_trace);
936#endif
Christoph Lameter45530c42012-11-28 16:23:07 +0000937
Glauber Costab7454ad2012-10-19 18:20:25 +0400938#ifdef CONFIG_SLABINFO
Wanpeng Lie9b4db22013-07-04 08:33:24 +0800939
940#ifdef CONFIG_SLAB
941#define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
942#else
943#define SLABINFO_RIGHTS S_IRUSR
944#endif
945
Vladimir Davydovb0475012014-12-10 15:44:19 -0800946static void print_slabinfo_header(struct seq_file *m)
Glauber Costabcee6e22012-10-19 18:20:26 +0400947{
948 /*
949 * Output format version, so at least we can change it
950 * without _too_ many complaints.
951 */
952#ifdef CONFIG_DEBUG_SLAB
953 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
954#else
955 seq_puts(m, "slabinfo - version: 2.1\n");
956#endif
957 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
958 "<objperslab> <pagesperslab>");
959 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
960 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
961#ifdef CONFIG_DEBUG_SLAB
962 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
963 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
964 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
965#endif
966 seq_putc(m, '\n');
967}
968
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800969void *slab_start(struct seq_file *m, loff_t *pos)
Glauber Costab7454ad2012-10-19 18:20:25 +0400970{
Glauber Costab7454ad2012-10-19 18:20:25 +0400971 mutex_lock(&slab_mutex);
Glauber Costab7454ad2012-10-19 18:20:25 +0400972 return seq_list_start(&slab_caches, *pos);
973}
974
Wanpeng Li276a2432013-07-08 08:08:28 +0800975void *slab_next(struct seq_file *m, void *p, loff_t *pos)
Glauber Costab7454ad2012-10-19 18:20:25 +0400976{
977 return seq_list_next(p, &slab_caches, pos);
978}
979
Wanpeng Li276a2432013-07-08 08:08:28 +0800980void slab_stop(struct seq_file *m, void *p)
Glauber Costab7454ad2012-10-19 18:20:25 +0400981{
982 mutex_unlock(&slab_mutex);
983}
984
Glauber Costa749c5412012-12-18 14:23:01 -0800985static void
986memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
Glauber Costab7454ad2012-10-19 18:20:25 +0400987{
Glauber Costa749c5412012-12-18 14:23:01 -0800988 struct kmem_cache *c;
989 struct slabinfo sinfo;
Glauber Costa749c5412012-12-18 14:23:01 -0800990
991 if (!is_root_cache(s))
992 return;
993
Vladimir Davydov426589f2015-02-12 14:59:23 -0800994 for_each_memcg_cache(c, s) {
Glauber Costa749c5412012-12-18 14:23:01 -0800995 memset(&sinfo, 0, sizeof(sinfo));
996 get_slabinfo(c, &sinfo);
997
998 info->active_slabs += sinfo.active_slabs;
999 info->num_slabs += sinfo.num_slabs;
1000 info->shared_avail += sinfo.shared_avail;
1001 info->active_objs += sinfo.active_objs;
1002 info->num_objs += sinfo.num_objs;
1003 }
1004}
1005
Vladimir Davydovb0475012014-12-10 15:44:19 -08001006static void cache_show(struct kmem_cache *s, struct seq_file *m)
Glauber Costa749c5412012-12-18 14:23:01 -08001007{
Glauber Costa0d7561c2012-10-19 18:20:27 +04001008 struct slabinfo sinfo;
1009
1010 memset(&sinfo, 0, sizeof(sinfo));
1011 get_slabinfo(s, &sinfo);
1012
Glauber Costa749c5412012-12-18 14:23:01 -08001013 memcg_accumulate_slabinfo(s, &sinfo);
1014
Glauber Costa0d7561c2012-10-19 18:20:27 +04001015 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
Glauber Costa749c5412012-12-18 14:23:01 -08001016 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
Glauber Costa0d7561c2012-10-19 18:20:27 +04001017 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1018
1019 seq_printf(m, " : tunables %4u %4u %4u",
1020 sinfo.limit, sinfo.batchcount, sinfo.shared);
1021 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1022 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1023 slabinfo_show_stats(m, s);
1024 seq_putc(m, '\n');
Glauber Costab7454ad2012-10-19 18:20:25 +04001025}
1026
Vladimir Davydov1df3b262014-12-10 15:42:16 -08001027static int slab_show(struct seq_file *m, void *p)
Glauber Costa749c5412012-12-18 14:23:01 -08001028{
1029 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1030
Vladimir Davydov1df3b262014-12-10 15:42:16 -08001031 if (p == slab_caches.next)
1032 print_slabinfo_header(m);
Vladimir Davydovb0475012014-12-10 15:44:19 -08001033 if (is_root_cache(s))
1034 cache_show(s, m);
1035 return 0;
Glauber Costa749c5412012-12-18 14:23:01 -08001036}
1037
Vladimir Davydovb0475012014-12-10 15:44:19 -08001038#ifdef CONFIG_MEMCG_KMEM
1039int memcg_slab_show(struct seq_file *m, void *p)
1040{
1041 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1042 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1043
1044 if (p == slab_caches.next)
1045 print_slabinfo_header(m);
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08001046 if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
Vladimir Davydovb0475012014-12-10 15:44:19 -08001047 cache_show(s, m);
1048 return 0;
1049}
1050#endif
1051
Glauber Costab7454ad2012-10-19 18:20:25 +04001052/*
1053 * slabinfo_op - iterator that generates /proc/slabinfo
1054 *
1055 * Output layout:
1056 * cache-name
1057 * num-active-objs
1058 * total-objs
1059 * object size
1060 * num-active-slabs
1061 * total-slabs
1062 * num-pages-per-slab
1063 * + further values on SMP and with statistics enabled
1064 */
1065static const struct seq_operations slabinfo_op = {
Vladimir Davydov1df3b262014-12-10 15:42:16 -08001066 .start = slab_start,
Wanpeng Li276a2432013-07-08 08:08:28 +08001067 .next = slab_next,
1068 .stop = slab_stop,
Vladimir Davydov1df3b262014-12-10 15:42:16 -08001069 .show = slab_show,
Glauber Costab7454ad2012-10-19 18:20:25 +04001070};
1071
1072static int slabinfo_open(struct inode *inode, struct file *file)
1073{
1074 return seq_open(file, &slabinfo_op);
1075}
1076
1077static const struct file_operations proc_slabinfo_operations = {
1078 .open = slabinfo_open,
1079 .read = seq_read,
1080 .write = slabinfo_write,
1081 .llseek = seq_lseek,
1082 .release = seq_release,
1083};
1084
1085static int __init slab_proc_init(void)
1086{
Wanpeng Lie9b4db22013-07-04 08:33:24 +08001087 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1088 &proc_slabinfo_operations);
Glauber Costab7454ad2012-10-19 18:20:25 +04001089 return 0;
1090}
1091module_init(slab_proc_init);
1092#endif /* CONFIG_SLABINFO */
Andrey Ryabinin928cec92014-08-06 16:04:44 -07001093
1094static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1095 gfp_t flags)
1096{
1097 void *ret;
1098 size_t ks = 0;
1099
1100 if (p)
1101 ks = ksize(p);
1102
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001103 if (ks >= new_size) {
1104 kasan_krealloc((void *)p, new_size);
Andrey Ryabinin928cec92014-08-06 16:04:44 -07001105 return (void *)p;
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001106 }
Andrey Ryabinin928cec92014-08-06 16:04:44 -07001107
1108 ret = kmalloc_track_caller(new_size, flags);
1109 if (ret && p)
1110 memcpy(ret, p, ks);
1111
1112 return ret;
1113}
1114
1115/**
1116 * __krealloc - like krealloc() but don't free @p.
1117 * @p: object to reallocate memory for.
1118 * @new_size: how many bytes of memory are required.
1119 * @flags: the type of memory to allocate.
1120 *
1121 * This function is like krealloc() except it never frees the originally
1122 * allocated buffer. Use this if you don't want to free the buffer immediately
1123 * like, for example, with RCU.
1124 */
1125void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1126{
1127 if (unlikely(!new_size))
1128 return ZERO_SIZE_PTR;
1129
1130 return __do_krealloc(p, new_size, flags);
1131
1132}
1133EXPORT_SYMBOL(__krealloc);
1134
1135/**
1136 * krealloc - reallocate memory. The contents will remain unchanged.
1137 * @p: object to reallocate memory for.
1138 * @new_size: how many bytes of memory are required.
1139 * @flags: the type of memory to allocate.
1140 *
1141 * The contents of the object pointed to are preserved up to the
1142 * lesser of the new and old sizes. If @p is %NULL, krealloc()
1143 * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a
1144 * %NULL pointer, the object pointed to is freed.
1145 */
1146void *krealloc(const void *p, size_t new_size, gfp_t flags)
1147{
1148 void *ret;
1149
1150 if (unlikely(!new_size)) {
1151 kfree(p);
1152 return ZERO_SIZE_PTR;
1153 }
1154
1155 ret = __do_krealloc(p, new_size, flags);
1156 if (ret && p != ret)
1157 kfree(p);
1158
1159 return ret;
1160}
1161EXPORT_SYMBOL(krealloc);
1162
1163/**
1164 * kzfree - like kfree but zero memory
1165 * @p: object to free memory of
1166 *
1167 * The memory of the object @p points to is zeroed before freed.
1168 * If @p is %NULL, kzfree() does nothing.
1169 *
1170 * Note: this function zeroes the whole allocated buffer which can be a good
1171 * deal bigger than the requested buffer size passed to kmalloc(). So be
1172 * careful when using this function in performance sensitive code.
1173 */
1174void kzfree(const void *p)
1175{
1176 size_t ks;
1177 void *mem = (void *)p;
1178
1179 if (unlikely(ZERO_OR_NULL_PTR(mem)))
1180 return;
1181 ks = ksize(mem);
1182 memset(mem, 0, ks);
1183 kfree(mem);
1184}
1185EXPORT_SYMBOL(kzfree);
1186
1187/* Tracepoints definitions. */
1188EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1189EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1190EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1191EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1192EXPORT_TRACEPOINT_SYMBOL(kfree);
1193EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);