blob: d31c4bacc6a203b0bc555bd76c2a97e90e78fa6c [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>
Christoph Lameterf1b6eb62013-09-04 16:35:34 +000022#include <trace/events/kmem.h>
Christoph Lameter039363f2012-07-06 15:25:10 -050023
Christoph Lameter97d06602012-07-06 15:25:11 -050024#include "slab.h"
25
26enum slab_state slab_state;
Christoph Lameter18004c52012-07-06 15:25:12 -050027LIST_HEAD(slab_caches);
28DEFINE_MUTEX(slab_mutex);
Christoph Lameter9b030cb2012-09-05 00:20:33 +000029struct kmem_cache *kmem_cache;
Christoph Lameter97d06602012-07-06 15:25:11 -050030
Shuah Khan77be4b12012-08-16 00:09:46 -070031#ifdef CONFIG_DEBUG_VM
Vladimir Davydov794b1242014-04-07 15:39:26 -070032static int kmem_cache_sanity_check(const char *name, size_t size)
Shuah Khan77be4b12012-08-16 00:09:46 -070033{
34 struct kmem_cache *s = NULL;
35
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);
39 return -EINVAL;
40 }
41
42 list_for_each_entry(s, &slab_caches, list) {
43 char tmp;
44 int res;
45
46 /*
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.
50 */
51 res = probe_kernel_address(s->name, tmp);
52 if (res) {
53 pr_err("Slab cache with size %d has lost its name\n",
54 s->object_size);
55 continue;
56 }
57
Mikulas Patocka69461742014-03-04 17:13:47 -050058#if !defined(CONFIG_SLUB)
Vladimir Davydov794b1242014-04-07 15:39:26 -070059 if (!strcmp(s->name, name)) {
Shuah Khan77be4b12012-08-16 00:09:46 -070060 pr_err("%s (%s): Cache name already exists.\n",
61 __func__, name);
62 dump_stack();
63 s = NULL;
64 return -EINVAL;
65 }
Christoph Lameter3e374912013-09-21 21:56:34 +000066#endif
Shuah Khan77be4b12012-08-16 00:09:46 -070067 }
68
69 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
70 return 0;
71}
72#else
Vladimir Davydov794b1242014-04-07 15:39:26 -070073static inline int kmem_cache_sanity_check(const char *name, size_t size)
Shuah Khan77be4b12012-08-16 00:09:46 -070074{
75 return 0;
76}
77#endif
78
Glauber Costa55007d82012-12-18 14:22:38 -080079#ifdef CONFIG_MEMCG_KMEM
80int memcg_update_all_caches(int num_memcgs)
81{
82 struct kmem_cache *s;
83 int ret = 0;
84 mutex_lock(&slab_mutex);
85
86 list_for_each_entry(s, &slab_caches, list) {
87 if (!is_root_cache(s))
88 continue;
89
90 ret = memcg_update_cache_size(s, num_memcgs);
91 /*
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.
95 */
96 if (ret)
97 goto out;
98 }
99
100 memcg_update_array_size(num_memcgs);
101out:
102 mutex_unlock(&slab_mutex);
103 return ret;
104}
105#endif
106
Christoph Lameter039363f2012-07-06 15:25:10 -0500107/*
Christoph Lameter45906852012-11-28 16:23:16 +0000108 * 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.
110 */
111unsigned long calculate_alignment(unsigned long flags,
112 unsigned long align, unsigned long size)
113{
114 /*
115 * If the user wants hardware cache aligned objects then follow that
116 * suggestion if the object is sufficiently large.
117 *
118 * The hardware cache alignment cannot override the specified
119 * alignment though. If that is greater then use it.
120 */
121 if (flags & SLAB_HWCACHE_ALIGN) {
122 unsigned long ralign = cache_line_size();
123 while (size <= ralign / 2)
124 ralign /= 2;
125 align = max(align, ralign);
126 }
127
128 if (align < ARCH_SLAB_MINALIGN)
129 align = ARCH_SLAB_MINALIGN;
130
131 return ALIGN(align, sizeof(void *));
132}
133
Vladimir Davydov794b1242014-04-07 15:39:26 -0700134static struct kmem_cache *
135do_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)
138{
139 struct kmem_cache *s;
140 int err;
141
142 err = -ENOMEM;
143 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
144 if (!s)
145 goto out;
146
147 s->name = name;
148 s->object_size = object_size;
149 s->size = size;
150 s->align = align;
151 s->ctor = ctor;
152
153 err = memcg_alloc_cache_params(memcg, s, root_cache);
154 if (err)
155 goto out_free_cache;
156
157 err = __kmem_cache_create(s, flags);
158 if (err)
159 goto out_free_cache;
160
161 s->refcount = 1;
162 list_add(&s->list, &slab_caches);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700163out:
164 if (err)
165 return ERR_PTR(err);
166 return s;
167
168out_free_cache:
169 memcg_free_cache_params(s);
170 kfree(s);
171 goto out;
172}
Christoph Lameter45906852012-11-28 16:23:16 +0000173
174/*
Christoph Lameter039363f2012-07-06 15:25:10 -0500175 * kmem_cache_create - Create a cache.
176 * @name: A string which is used in /proc/slabinfo to identify this cache.
177 * @size: The size of objects to be created in this cache.
178 * @align: The required alignment for the objects.
179 * @flags: SLAB flags
180 * @ctor: A constructor for the objects.
181 *
182 * Returns a ptr to the cache on success, NULL on failure.
183 * Cannot be called within a interrupt, but can be interrupted.
184 * The @ctor is run when new pages are allocated by the cache.
185 *
186 * The flags are
187 *
188 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
189 * to catch references to uninitialised memory.
190 *
191 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
192 * for buffer overruns.
193 *
194 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
195 * cacheline. This can be beneficial if you're counting cycles as closely
196 * as davem.
197 */
Glauber Costa2633d7a2012-12-18 14:22:34 -0800198struct kmem_cache *
Vladimir Davydov794b1242014-04-07 15:39:26 -0700199kmem_cache_create(const char *name, size_t size, size_t align,
200 unsigned long flags, void (*ctor)(void *))
Christoph Lameter039363f2012-07-06 15:25:10 -0500201{
Vladimir Davydov794b1242014-04-07 15:39:26 -0700202 struct kmem_cache *s;
203 char *cache_name;
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800204 int err;
Christoph Lameter039363f2012-07-06 15:25:10 -0500205
Pekka Enbergb9205362012-08-16 10:12:18 +0300206 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700207 get_online_mems();
208
Pekka Enbergb9205362012-08-16 10:12:18 +0300209 mutex_lock(&slab_mutex);
Christoph Lameter686d5502012-09-05 00:20:33 +0000210
Vladimir Davydov794b1242014-04-07 15:39:26 -0700211 err = kmem_cache_sanity_check(name, size);
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800212 if (err)
213 goto out_unlock;
Christoph Lameter686d5502012-09-05 00:20:33 +0000214
Glauber Costad8843922012-10-17 15:36:51 +0400215 /*
216 * Some allocators will constraint the set of valid flags to a subset
217 * of all flags. We expect them to define CACHE_CREATE_MASK in this
218 * case, and we'll just provide them with a sanitized version of the
219 * passed flags.
220 */
221 flags &= CACHE_CREATE_MASK;
Christoph Lameter686d5502012-09-05 00:20:33 +0000222
Vladimir Davydov794b1242014-04-07 15:39:26 -0700223 s = __kmem_cache_alias(name, size, align, flags, ctor);
224 if (s)
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800225 goto out_unlock;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800226
Vladimir Davydov794b1242014-04-07 15:39:26 -0700227 cache_name = kstrdup(name, GFP_KERNEL);
228 if (!cache_name) {
229 err = -ENOMEM;
230 goto out_unlock;
231 }
Glauber Costa2633d7a2012-12-18 14:22:34 -0800232
Vladimir Davydov794b1242014-04-07 15:39:26 -0700233 s = do_kmem_cache_create(cache_name, size, size,
234 calculate_alignment(flags, align, size),
235 flags, ctor, NULL, NULL);
236 if (IS_ERR(s)) {
237 err = PTR_ERR(s);
238 kfree(cache_name);
239 }
Vladimir Davydov3965fc32014-01-23 15:52:55 -0800240
241out_unlock:
Christoph Lameter20cea962012-07-06 15:25:13 -0500242 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700243
244 put_online_mems();
Christoph Lameter20cea962012-07-06 15:25:13 -0500245 put_online_cpus();
246
Dave Jonesba3253c72014-01-29 14:05:48 -0800247 if (err) {
Christoph Lameter686d5502012-09-05 00:20:33 +0000248 if (flags & SLAB_PANIC)
249 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
250 name, err);
251 else {
252 printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
253 name, err);
254 dump_stack();
255 }
Christoph Lameter686d5502012-09-05 00:20:33 +0000256 return NULL;
257 }
Christoph Lameter039363f2012-07-06 15:25:10 -0500258 return s;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800259}
Christoph Lameter039363f2012-07-06 15:25:10 -0500260EXPORT_SYMBOL(kmem_cache_create);
Christoph Lameter97d06602012-07-06 15:25:11 -0500261
Vladimir Davydov794b1242014-04-07 15:39:26 -0700262#ifdef CONFIG_MEMCG_KMEM
263/*
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700264 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
Vladimir Davydov794b1242014-04-07 15:39:26 -0700265 * @memcg: The memory cgroup the new cache is for.
266 * @root_cache: The parent of the new cache.
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700267 * @memcg_name: The name of the memory cgroup (used for naming the new cache).
Vladimir Davydov794b1242014-04-07 15:39:26 -0700268 *
269 * This function attempts to create a kmem cache that will serve allocation
270 * requests going from @memcg to @root_cache. The new cache inherits properties
271 * from its parent.
272 */
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700273struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700274 struct kmem_cache *root_cache,
275 const char *memcg_name)
Vladimir Davydov794b1242014-04-07 15:39:26 -0700276{
Vladimir Davydovbd673142014-06-04 16:07:40 -0700277 struct kmem_cache *s = NULL;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700278 char *cache_name;
279
280 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700281 get_online_mems();
282
Vladimir Davydov794b1242014-04-07 15:39:26 -0700283 mutex_lock(&slab_mutex);
284
Vladimir Davydov073ee1c2014-06-04 16:08:23 -0700285 cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
286 memcg_cache_id(memcg), memcg_name);
Vladimir Davydov794b1242014-04-07 15:39:26 -0700287 if (!cache_name)
288 goto out_unlock;
289
290 s = do_kmem_cache_create(cache_name, root_cache->object_size,
291 root_cache->size, root_cache->align,
292 root_cache->flags, root_cache->ctor,
293 memcg, root_cache);
Vladimir Davydovbd673142014-06-04 16:07:40 -0700294 if (IS_ERR(s)) {
Vladimir Davydov794b1242014-04-07 15:39:26 -0700295 kfree(cache_name);
Vladimir Davydovbd673142014-06-04 16:07:40 -0700296 s = NULL;
297 }
Vladimir Davydov794b1242014-04-07 15:39:26 -0700298
299out_unlock:
300 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700301
302 put_online_mems();
Vladimir Davydov794b1242014-04-07 15:39:26 -0700303 put_online_cpus();
Vladimir Davydovbd673142014-06-04 16:07:40 -0700304
305 return s;
Vladimir Davydov794b1242014-04-07 15:39:26 -0700306}
Vladimir Davydovb8529902014-04-07 15:39:28 -0700307
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700308static int memcg_cleanup_cache_params(struct kmem_cache *s)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700309{
310 int rc;
311
312 if (!s->memcg_params ||
313 !s->memcg_params->is_root_cache)
314 return 0;
315
316 mutex_unlock(&slab_mutex);
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700317 rc = __memcg_cleanup_cache_params(s);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700318 mutex_lock(&slab_mutex);
319
320 return rc;
321}
322#else
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700323static int memcg_cleanup_cache_params(struct kmem_cache *s)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700324{
325 return 0;
326}
Vladimir Davydov794b1242014-04-07 15:39:26 -0700327#endif /* CONFIG_MEMCG_KMEM */
328
Christoph Lameter41a21282014-05-06 12:50:08 -0700329void slab_kmem_cache_release(struct kmem_cache *s)
330{
331 kfree(s->name);
332 kmem_cache_free(kmem_cache, s);
333}
334
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000335void kmem_cache_destroy(struct kmem_cache *s)
336{
337 get_online_cpus();
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700338 get_online_mems();
339
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000340 mutex_lock(&slab_mutex);
Vladimir Davydovb8529902014-04-07 15:39:28 -0700341
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000342 s->refcount--;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700343 if (s->refcount)
344 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000345
Vladimir Davydov776ed0f2014-06-04 16:10:02 -0700346 if (memcg_cleanup_cache_params(s) != 0)
Vladimir Davydovb8529902014-04-07 15:39:28 -0700347 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000348
Vladimir Davydovb8529902014-04-07 15:39:28 -0700349 if (__kmem_cache_shutdown(s) != 0) {
Vladimir Davydovb8529902014-04-07 15:39:28 -0700350 printk(KERN_ERR "kmem_cache_destroy %s: "
351 "Slab cache still has objects\n", s->name);
352 dump_stack();
353 goto out_unlock;
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000354 }
Vladimir Davydovb8529902014-04-07 15:39:28 -0700355
Vladimir Davydov0bd62b12014-06-04 16:10:03 -0700356 list_del(&s->list);
357
Vladimir Davydovb8529902014-04-07 15:39:28 -0700358 mutex_unlock(&slab_mutex);
359 if (s->flags & SLAB_DESTROY_BY_RCU)
360 rcu_barrier();
361
362 memcg_free_cache_params(s);
Christoph Lameter41a21282014-05-06 12:50:08 -0700363#ifdef SLAB_SUPPORTS_SYSFS
364 sysfs_slab_remove(s);
365#else
366 slab_kmem_cache_release(s);
367#endif
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700368 goto out;
Vladimir Davydovb8529902014-04-07 15:39:28 -0700369
370out_unlock:
371 mutex_unlock(&slab_mutex);
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700372out:
373 put_online_mems();
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000374 put_online_cpus();
375}
376EXPORT_SYMBOL(kmem_cache_destroy);
377
Vladimir Davydov03afc0e2014-06-04 16:07:20 -0700378/**
379 * kmem_cache_shrink - Shrink a cache.
380 * @cachep: The cache to shrink.
381 *
382 * Releases as many slabs as possible for a cache.
383 * To help debugging, a zero exit status indicates all slabs were released.
384 */
385int kmem_cache_shrink(struct kmem_cache *cachep)
386{
387 int ret;
388
389 get_online_cpus();
390 get_online_mems();
391 ret = __kmem_cache_shrink(cachep);
392 put_online_mems();
393 put_online_cpus();
394 return ret;
395}
396EXPORT_SYMBOL(kmem_cache_shrink);
397
Christoph Lameter97d06602012-07-06 15:25:11 -0500398int slab_is_available(void)
399{
400 return slab_state >= UP;
401}
Glauber Costab7454ad2012-10-19 18:20:25 +0400402
Christoph Lameter45530c42012-11-28 16:23:07 +0000403#ifndef CONFIG_SLOB
404/* Create a cache during boot when no slab services are available yet */
405void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
406 unsigned long flags)
407{
408 int err;
409
410 s->name = name;
411 s->size = s->object_size = size;
Christoph Lameter45906852012-11-28 16:23:16 +0000412 s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
Christoph Lameter45530c42012-11-28 16:23:07 +0000413 err = __kmem_cache_create(s, flags);
414
415 if (err)
Christoph Lameter31ba7342013-01-10 19:00:53 +0000416 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
Christoph Lameter45530c42012-11-28 16:23:07 +0000417 name, size, err);
418
419 s->refcount = -1; /* Exempt from merging for now */
420}
421
422struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
423 unsigned long flags)
424{
425 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
426
427 if (!s)
428 panic("Out of memory when creating slab %s\n", name);
429
430 create_boot_cache(s, name, size, flags);
431 list_add(&s->list, &slab_caches);
432 s->refcount = 1;
433 return s;
434}
435
Christoph Lameter9425c582013-01-10 19:12:17 +0000436struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
437EXPORT_SYMBOL(kmalloc_caches);
438
439#ifdef CONFIG_ZONE_DMA
440struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
441EXPORT_SYMBOL(kmalloc_dma_caches);
442#endif
443
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000444/*
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000445 * Conversion table for small slabs sizes / 8 to the index in the
446 * kmalloc array. This is necessary for slabs < 192 since we have non power
447 * of two cache sizes there. The size of larger slabs can be determined using
448 * fls.
449 */
450static s8 size_index[24] = {
451 3, /* 8 */
452 4, /* 16 */
453 5, /* 24 */
454 5, /* 32 */
455 6, /* 40 */
456 6, /* 48 */
457 6, /* 56 */
458 6, /* 64 */
459 1, /* 72 */
460 1, /* 80 */
461 1, /* 88 */
462 1, /* 96 */
463 7, /* 104 */
464 7, /* 112 */
465 7, /* 120 */
466 7, /* 128 */
467 2, /* 136 */
468 2, /* 144 */
469 2, /* 152 */
470 2, /* 160 */
471 2, /* 168 */
472 2, /* 176 */
473 2, /* 184 */
474 2 /* 192 */
475};
476
477static inline int size_index_elem(size_t bytes)
478{
479 return (bytes - 1) / 8;
480}
481
482/*
483 * Find the kmem_cache structure that serves a given size of
484 * allocation
485 */
486struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
487{
488 int index;
489
Joonsoo Kim9de1bc82013-08-02 11:02:42 +0900490 if (unlikely(size > KMALLOC_MAX_SIZE)) {
Sasha Levin907985f2013-06-10 15:18:00 -0400491 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
Christoph Lameter6286ae92013-05-03 15:43:18 +0000492 return NULL;
Sasha Levin907985f2013-06-10 15:18:00 -0400493 }
Christoph Lameter6286ae92013-05-03 15:43:18 +0000494
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000495 if (size <= 192) {
496 if (!size)
497 return ZERO_SIZE_PTR;
498
499 index = size_index[size_index_elem(size)];
500 } else
501 index = fls(size - 1);
502
503#ifdef CONFIG_ZONE_DMA
Joonsoo Kimb1e05412013-02-04 23:46:46 +0900504 if (unlikely((flags & GFP_DMA)))
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000505 return kmalloc_dma_caches[index];
506
507#endif
508 return kmalloc_caches[index];
509}
510
511/*
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000512 * Create the kmalloc array. Some of the regular kmalloc arrays
513 * may already have been created because they were needed to
514 * enable allocations for slab creation.
515 */
516void __init create_kmalloc_caches(unsigned long flags)
517{
518 int i;
519
Christoph Lameter2c59dd62013-01-10 19:14:19 +0000520 /*
521 * Patch up the size_index table if we have strange large alignment
522 * requirements for the kmalloc array. This is only the case for
523 * MIPS it seems. The standard arches will not generate any code here.
524 *
525 * Largest permitted alignment is 256 bytes due to the way we
526 * handle the index determination for the smaller caches.
527 *
528 * Make sure that nothing crazy happens if someone starts tinkering
529 * around with ARCH_KMALLOC_MINALIGN
530 */
531 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
532 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
533
534 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
535 int elem = size_index_elem(i);
536
537 if (elem >= ARRAY_SIZE(size_index))
538 break;
539 size_index[elem] = KMALLOC_SHIFT_LOW;
540 }
541
542 if (KMALLOC_MIN_SIZE >= 64) {
543 /*
544 * The 96 byte size cache is not used if the alignment
545 * is 64 byte.
546 */
547 for (i = 64 + 8; i <= 96; i += 8)
548 size_index[size_index_elem(i)] = 7;
549
550 }
551
552 if (KMALLOC_MIN_SIZE >= 128) {
553 /*
554 * The 192 byte sized cache is not used if the alignment
555 * is 128 byte. Redirect kmalloc to use the 256 byte cache
556 * instead.
557 */
558 for (i = 128 + 8; i <= 192; i += 8)
559 size_index[size_index_elem(i)] = 8;
560 }
Christoph Lameter8a965b32013-05-03 18:04:18 +0000561 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
562 if (!kmalloc_caches[i]) {
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000563 kmalloc_caches[i] = create_kmalloc_cache(NULL,
564 1 << i, flags);
Christoph Lameter8a965b32013-05-03 18:04:18 +0000565 }
Chris Mason956e46e2013-05-08 15:56:28 -0400566
567 /*
568 * Caches that are not of the two-to-the-power-of size.
569 * These have to be created immediately after the
570 * earlier power of two caches
571 */
572 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
573 kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
574
575 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
576 kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
Christoph Lameter8a965b32013-05-03 18:04:18 +0000577 }
578
Christoph Lameterf97d5f62013-01-10 19:12:17 +0000579 /* Kmalloc array is now usable */
580 slab_state = UP;
581
582 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
583 struct kmem_cache *s = kmalloc_caches[i];
584 char *n;
585
586 if (s) {
587 n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
588
589 BUG_ON(!n);
590 s->name = n;
591 }
592 }
593
594#ifdef CONFIG_ZONE_DMA
595 for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
596 struct kmem_cache *s = kmalloc_caches[i];
597
598 if (s) {
599 int size = kmalloc_size(i);
600 char *n = kasprintf(GFP_NOWAIT,
601 "dma-kmalloc-%d", size);
602
603 BUG_ON(!n);
604 kmalloc_dma_caches[i] = create_kmalloc_cache(n,
605 size, SLAB_CACHE_DMA | flags);
606 }
607 }
608#endif
609}
Christoph Lameter45530c42012-11-28 16:23:07 +0000610#endif /* !CONFIG_SLOB */
611
Vladimir Davydovcea371f2014-06-04 16:07:04 -0700612/*
613 * To avoid unnecessary overhead, we pass through large allocation requests
614 * directly to the page allocator. We use __GFP_COMP, because we will need to
615 * know the allocation order to free the pages properly in kfree.
616 */
Vladimir Davydov52383432014-06-04 16:06:39 -0700617void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
618{
619 void *ret;
620 struct page *page;
621
622 flags |= __GFP_COMP;
623 page = alloc_kmem_pages(flags, order);
624 ret = page ? page_address(page) : NULL;
625 kmemleak_alloc(ret, size, 1, flags);
626 return ret;
627}
628EXPORT_SYMBOL(kmalloc_order);
629
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000630#ifdef CONFIG_TRACING
631void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
632{
633 void *ret = kmalloc_order(size, flags, order);
634 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
635 return ret;
636}
637EXPORT_SYMBOL(kmalloc_order_trace);
638#endif
Christoph Lameter45530c42012-11-28 16:23:07 +0000639
Glauber Costab7454ad2012-10-19 18:20:25 +0400640#ifdef CONFIG_SLABINFO
Wanpeng Lie9b4db22013-07-04 08:33:24 +0800641
642#ifdef CONFIG_SLAB
643#define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
644#else
645#define SLABINFO_RIGHTS S_IRUSR
646#endif
647
Glauber Costa749c5412012-12-18 14:23:01 -0800648void print_slabinfo_header(struct seq_file *m)
Glauber Costabcee6e22012-10-19 18:20:26 +0400649{
650 /*
651 * Output format version, so at least we can change it
652 * without _too_ many complaints.
653 */
654#ifdef CONFIG_DEBUG_SLAB
655 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
656#else
657 seq_puts(m, "slabinfo - version: 2.1\n");
658#endif
659 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
660 "<objperslab> <pagesperslab>");
661 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
662 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
663#ifdef CONFIG_DEBUG_SLAB
664 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
665 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
666 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
667#endif
668 seq_putc(m, '\n');
669}
670
Glauber Costab7454ad2012-10-19 18:20:25 +0400671static void *s_start(struct seq_file *m, loff_t *pos)
672{
673 loff_t n = *pos;
674
675 mutex_lock(&slab_mutex);
676 if (!n)
677 print_slabinfo_header(m);
678
679 return seq_list_start(&slab_caches, *pos);
680}
681
Wanpeng Li276a2432013-07-08 08:08:28 +0800682void *slab_next(struct seq_file *m, void *p, loff_t *pos)
Glauber Costab7454ad2012-10-19 18:20:25 +0400683{
684 return seq_list_next(p, &slab_caches, pos);
685}
686
Wanpeng Li276a2432013-07-08 08:08:28 +0800687void slab_stop(struct seq_file *m, void *p)
Glauber Costab7454ad2012-10-19 18:20:25 +0400688{
689 mutex_unlock(&slab_mutex);
690}
691
Glauber Costa749c5412012-12-18 14:23:01 -0800692static void
693memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
Glauber Costab7454ad2012-10-19 18:20:25 +0400694{
Glauber Costa749c5412012-12-18 14:23:01 -0800695 struct kmem_cache *c;
696 struct slabinfo sinfo;
697 int i;
698
699 if (!is_root_cache(s))
700 return;
701
702 for_each_memcg_cache_index(i) {
Qiang Huang2ade4de2013-11-12 15:08:23 -0800703 c = cache_from_memcg_idx(s, i);
Glauber Costa749c5412012-12-18 14:23:01 -0800704 if (!c)
705 continue;
706
707 memset(&sinfo, 0, sizeof(sinfo));
708 get_slabinfo(c, &sinfo);
709
710 info->active_slabs += sinfo.active_slabs;
711 info->num_slabs += sinfo.num_slabs;
712 info->shared_avail += sinfo.shared_avail;
713 info->active_objs += sinfo.active_objs;
714 info->num_objs += sinfo.num_objs;
715 }
716}
717
718int cache_show(struct kmem_cache *s, struct seq_file *m)
719{
Glauber Costa0d7561c2012-10-19 18:20:27 +0400720 struct slabinfo sinfo;
721
722 memset(&sinfo, 0, sizeof(sinfo));
723 get_slabinfo(s, &sinfo);
724
Glauber Costa749c5412012-12-18 14:23:01 -0800725 memcg_accumulate_slabinfo(s, &sinfo);
726
Glauber Costa0d7561c2012-10-19 18:20:27 +0400727 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
Glauber Costa749c5412012-12-18 14:23:01 -0800728 cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
Glauber Costa0d7561c2012-10-19 18:20:27 +0400729 sinfo.objects_per_slab, (1 << sinfo.cache_order));
730
731 seq_printf(m, " : tunables %4u %4u %4u",
732 sinfo.limit, sinfo.batchcount, sinfo.shared);
733 seq_printf(m, " : slabdata %6lu %6lu %6lu",
734 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
735 slabinfo_show_stats(m, s);
736 seq_putc(m, '\n');
737 return 0;
Glauber Costab7454ad2012-10-19 18:20:25 +0400738}
739
Glauber Costa749c5412012-12-18 14:23:01 -0800740static int s_show(struct seq_file *m, void *p)
741{
742 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
743
744 if (!is_root_cache(s))
745 return 0;
746 return cache_show(s, m);
747}
748
Glauber Costab7454ad2012-10-19 18:20:25 +0400749/*
750 * slabinfo_op - iterator that generates /proc/slabinfo
751 *
752 * Output layout:
753 * cache-name
754 * num-active-objs
755 * total-objs
756 * object size
757 * num-active-slabs
758 * total-slabs
759 * num-pages-per-slab
760 * + further values on SMP and with statistics enabled
761 */
762static const struct seq_operations slabinfo_op = {
763 .start = s_start,
Wanpeng Li276a2432013-07-08 08:08:28 +0800764 .next = slab_next,
765 .stop = slab_stop,
Glauber Costab7454ad2012-10-19 18:20:25 +0400766 .show = s_show,
767};
768
769static int slabinfo_open(struct inode *inode, struct file *file)
770{
771 return seq_open(file, &slabinfo_op);
772}
773
774static const struct file_operations proc_slabinfo_operations = {
775 .open = slabinfo_open,
776 .read = seq_read,
777 .write = slabinfo_write,
778 .llseek = seq_lseek,
779 .release = seq_release,
780};
781
782static int __init slab_proc_init(void)
783{
Wanpeng Lie9b4db22013-07-04 08:33:24 +0800784 proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
785 &proc_slabinfo_operations);
Glauber Costab7454ad2012-10-19 18:20:25 +0400786 return 0;
787}
788module_init(slab_proc_init);
789#endif /* CONFIG_SLABINFO */