Commit | Line | Data |
---|---|---|
039363f3 CL |
1 | /* |
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> | |
20cea968 CL |
14 | #include <linux/cpu.h> |
15 | #include <linux/uaccess.h> | |
b7454ad3 GC |
16 | #include <linux/seq_file.h> |
17 | #include <linux/proc_fs.h> | |
039363f3 CL |
18 | #include <asm/cacheflush.h> |
19 | #include <asm/tlbflush.h> | |
20 | #include <asm/page.h> | |
2633d7a0 | 21 | #include <linux/memcontrol.h> |
928cec9c AR |
22 | |
23 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 24 | #include <trace/events/kmem.h> |
039363f3 | 25 | |
97d06609 CL |
26 | #include "slab.h" |
27 | ||
28 | enum slab_state slab_state; | |
18004c5d CL |
29 | LIST_HEAD(slab_caches); |
30 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 31 | struct kmem_cache *kmem_cache; |
97d06609 | 32 | |
423c929c JK |
33 | /* |
34 | * 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 | */ | |
47 | static int slab_nomerge; | |
48 | ||
49 | static 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 | ||
07f361b2 JK |
61 | /* |
62 | * Determine the size of a slab object | |
63 | */ | |
64 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
65 | { | |
66 | return s->object_size; | |
67 | } | |
68 | EXPORT_SYMBOL(kmem_cache_size); | |
69 | ||
77be4b13 | 70 | #ifdef CONFIG_DEBUG_VM |
794b1248 | 71 | static int kmem_cache_sanity_check(const char *name, size_t size) |
039363f3 CL |
72 | { |
73 | struct kmem_cache *s = NULL; | |
74 | ||
039363f3 CL |
75 | if (!name || in_interrupt() || size < sizeof(void *) || |
76 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
77 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
78 | return -EINVAL; | |
039363f3 | 79 | } |
b920536a | 80 | |
20cea968 CL |
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) { | |
77be4b13 | 92 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
93 | s->object_size); |
94 | continue; | |
95 | } | |
20cea968 CL |
96 | } |
97 | ||
98 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
99 | return 0; |
100 | } | |
101 | #else | |
794b1248 | 102 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
77be4b13 SK |
103 | { |
104 | return 0; | |
105 | } | |
20cea968 CL |
106 | #endif |
107 | ||
55007d84 | 108 | #ifdef CONFIG_MEMCG_KMEM |
33a690c4 VD |
109 | static 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 | ||
136 | static void memcg_free_cache_params(struct kmem_cache *s) | |
137 | { | |
138 | kfree(s->memcg_params); | |
139 | } | |
140 | ||
6f817f4c VD |
141 | static 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 | ||
55007d84 GC |
168 | int 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 | ||
6f817f4c | 178 | ret = memcg_update_cache_params(s, num_memcgs); |
55007d84 | 179 | /* |
55007d84 GC |
180 | * 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); | |
188 | out: | |
189 | mutex_unlock(&slab_mutex); | |
190 | return ret; | |
191 | } | |
33a690c4 VD |
192 | #else |
193 | static 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 | ||
199 | static inline void memcg_free_cache_params(struct kmem_cache *s) | |
200 | { | |
201 | } | |
202 | #endif /* CONFIG_MEMCG_KMEM */ | |
55007d84 | 203 | |
423c929c JK |
204 | /* |
205 | * Find a mergeable slab cache | |
206 | */ | |
207 | int 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 | ||
227 | struct 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 | ||
54362057 | 243 | list_for_each_entry_reverse(s, &slab_caches, list) { |
423c929c JK |
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 | ||
95069ac8 JK |
262 | if (IS_ENABLED(CONFIG_SLAB) && align && |
263 | (align > s->align || s->align % align)) | |
264 | continue; | |
265 | ||
423c929c JK |
266 | return s; |
267 | } | |
268 | return NULL; | |
269 | } | |
270 | ||
45906855 CL |
271 | /* |
272 | * 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 | */ | |
275 | unsigned 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 | ||
794b1248 VD |
298 | static struct kmem_cache * |
299 | do_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); | |
794b1248 VD |
327 | out: |
328 | if (err) | |
329 | return ERR_PTR(err); | |
330 | return s; | |
331 | ||
332 | out_free_cache: | |
333 | memcg_free_cache_params(s); | |
7c4da061 | 334 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
335 | goto out; |
336 | } | |
45906855 | 337 | |
77be4b13 SK |
338 | /* |
339 | * 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 | */ | |
2633d7a0 | 362 | struct kmem_cache * |
794b1248 VD |
363 | kmem_cache_create(const char *name, size_t size, size_t align, |
364 | unsigned long flags, void (*ctor)(void *)) | |
77be4b13 | 365 | { |
794b1248 VD |
366 | struct kmem_cache *s; |
367 | char *cache_name; | |
3965fc36 | 368 | int err; |
039363f3 | 369 | |
77be4b13 | 370 | get_online_cpus(); |
03afc0e2 VD |
371 | get_online_mems(); |
372 | ||
77be4b13 | 373 | mutex_lock(&slab_mutex); |
686d550d | 374 | |
794b1248 | 375 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 AM |
376 | if (err) { |
377 | s = NULL; /* suppress uninit var warning */ | |
3965fc36 | 378 | goto out_unlock; |
3aa24f51 | 379 | } |
686d550d | 380 | |
d8843922 GC |
381 | /* |
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; | |
686d550d | 388 | |
794b1248 VD |
389 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
390 | if (s) | |
3965fc36 | 391 | goto out_unlock; |
2633d7a0 | 392 | |
794b1248 VD |
393 | cache_name = kstrdup(name, GFP_KERNEL); |
394 | if (!cache_name) { | |
395 | err = -ENOMEM; | |
396 | goto out_unlock; | |
397 | } | |
7c9adf5a | 398 | |
794b1248 VD |
399 | 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 | } | |
3965fc36 VD |
406 | |
407 | out_unlock: | |
20cea968 | 408 | mutex_unlock(&slab_mutex); |
03afc0e2 VD |
409 | |
410 | put_online_mems(); | |
20cea968 CL |
411 | put_online_cpus(); |
412 | ||
ba3253c7 | 413 | if (err) { |
686d550d CL |
414 | 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 | } | |
686d550d CL |
422 | return NULL; |
423 | } | |
039363f3 CL |
424 | return s; |
425 | } | |
794b1248 | 426 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 427 | |
794b1248 VD |
428 | #ifdef CONFIG_MEMCG_KMEM |
429 | /* | |
776ed0f0 | 430 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
431 | * @memcg: The memory cgroup the new cache is for. |
432 | * @root_cache: The parent of the new cache. | |
433 | * | |
434 | * This function attempts to create a kmem cache that will serve allocation | |
435 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
436 | * from its parent. | |
437 | */ | |
776ed0f0 | 438 | struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3e0350a3 | 439 | struct kmem_cache *root_cache) |
2633d7a0 | 440 | { |
3e0350a3 | 441 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
bd673145 | 442 | struct kmem_cache *s = NULL; |
794b1248 VD |
443 | char *cache_name; |
444 | ||
445 | get_online_cpus(); | |
03afc0e2 VD |
446 | get_online_mems(); |
447 | ||
794b1248 VD |
448 | mutex_lock(&slab_mutex); |
449 | ||
3e0350a3 VD |
450 | cgroup_name(mem_cgroup_css(memcg)->cgroup, |
451 | memcg_name_buf, sizeof(memcg_name_buf)); | |
073ee1c6 | 452 | cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name, |
3e0350a3 | 453 | memcg_cache_id(memcg), memcg_name_buf); |
794b1248 VD |
454 | if (!cache_name) |
455 | goto out_unlock; | |
456 | ||
457 | s = do_kmem_cache_create(cache_name, root_cache->object_size, | |
458 | root_cache->size, root_cache->align, | |
459 | root_cache->flags, root_cache->ctor, | |
460 | memcg, root_cache); | |
bd673145 | 461 | if (IS_ERR(s)) { |
794b1248 | 462 | kfree(cache_name); |
bd673145 VD |
463 | s = NULL; |
464 | } | |
794b1248 VD |
465 | |
466 | out_unlock: | |
467 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
468 | |
469 | put_online_mems(); | |
794b1248 | 470 | put_online_cpus(); |
bd673145 VD |
471 | |
472 | return s; | |
2633d7a0 | 473 | } |
b8529907 | 474 | |
776ed0f0 | 475 | static int memcg_cleanup_cache_params(struct kmem_cache *s) |
b8529907 VD |
476 | { |
477 | int rc; | |
478 | ||
479 | if (!s->memcg_params || | |
480 | !s->memcg_params->is_root_cache) | |
481 | return 0; | |
482 | ||
483 | mutex_unlock(&slab_mutex); | |
776ed0f0 | 484 | rc = __memcg_cleanup_cache_params(s); |
b8529907 VD |
485 | mutex_lock(&slab_mutex); |
486 | ||
487 | return rc; | |
488 | } | |
489 | #else | |
776ed0f0 | 490 | static int memcg_cleanup_cache_params(struct kmem_cache *s) |
b8529907 VD |
491 | { |
492 | return 0; | |
493 | } | |
794b1248 | 494 | #endif /* CONFIG_MEMCG_KMEM */ |
97d06609 | 495 | |
41a21285 CL |
496 | void slab_kmem_cache_release(struct kmem_cache *s) |
497 | { | |
498 | kfree(s->name); | |
499 | kmem_cache_free(kmem_cache, s); | |
500 | } | |
501 | ||
945cf2b6 CL |
502 | void kmem_cache_destroy(struct kmem_cache *s) |
503 | { | |
504 | get_online_cpus(); | |
03afc0e2 VD |
505 | get_online_mems(); |
506 | ||
945cf2b6 | 507 | mutex_lock(&slab_mutex); |
b8529907 | 508 | |
945cf2b6 | 509 | s->refcount--; |
b8529907 VD |
510 | if (s->refcount) |
511 | goto out_unlock; | |
512 | ||
776ed0f0 | 513 | if (memcg_cleanup_cache_params(s) != 0) |
b8529907 VD |
514 | goto out_unlock; |
515 | ||
b8529907 | 516 | if (__kmem_cache_shutdown(s) != 0) { |
b8529907 VD |
517 | printk(KERN_ERR "kmem_cache_destroy %s: " |
518 | "Slab cache still has objects\n", s->name); | |
519 | dump_stack(); | |
520 | goto out_unlock; | |
945cf2b6 | 521 | } |
b8529907 | 522 | |
0bd62b11 VD |
523 | list_del(&s->list); |
524 | ||
b8529907 VD |
525 | mutex_unlock(&slab_mutex); |
526 | if (s->flags & SLAB_DESTROY_BY_RCU) | |
527 | rcu_barrier(); | |
528 | ||
529 | memcg_free_cache_params(s); | |
41a21285 CL |
530 | #ifdef SLAB_SUPPORTS_SYSFS |
531 | sysfs_slab_remove(s); | |
532 | #else | |
533 | slab_kmem_cache_release(s); | |
534 | #endif | |
03afc0e2 | 535 | goto out; |
b8529907 VD |
536 | |
537 | out_unlock: | |
538 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
539 | out: |
540 | put_online_mems(); | |
945cf2b6 CL |
541 | put_online_cpus(); |
542 | } | |
543 | EXPORT_SYMBOL(kmem_cache_destroy); | |
544 | ||
03afc0e2 VD |
545 | /** |
546 | * kmem_cache_shrink - Shrink a cache. | |
547 | * @cachep: The cache to shrink. | |
548 | * | |
549 | * Releases as many slabs as possible for a cache. | |
550 | * To help debugging, a zero exit status indicates all slabs were released. | |
551 | */ | |
552 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
553 | { | |
554 | int ret; | |
555 | ||
556 | get_online_cpus(); | |
557 | get_online_mems(); | |
558 | ret = __kmem_cache_shrink(cachep); | |
559 | put_online_mems(); | |
560 | put_online_cpus(); | |
561 | return ret; | |
562 | } | |
563 | EXPORT_SYMBOL(kmem_cache_shrink); | |
564 | ||
97d06609 CL |
565 | int slab_is_available(void) |
566 | { | |
567 | return slab_state >= UP; | |
568 | } | |
b7454ad3 | 569 | |
45530c44 CL |
570 | #ifndef CONFIG_SLOB |
571 | /* Create a cache during boot when no slab services are available yet */ | |
572 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
573 | unsigned long flags) | |
574 | { | |
575 | int err; | |
576 | ||
577 | s->name = name; | |
578 | s->size = s->object_size = size; | |
45906855 | 579 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
45530c44 CL |
580 | err = __kmem_cache_create(s, flags); |
581 | ||
582 | if (err) | |
31ba7346 | 583 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
584 | name, size, err); |
585 | ||
586 | s->refcount = -1; /* Exempt from merging for now */ | |
587 | } | |
588 | ||
589 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
590 | unsigned long flags) | |
591 | { | |
592 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
593 | ||
594 | if (!s) | |
595 | panic("Out of memory when creating slab %s\n", name); | |
596 | ||
597 | create_boot_cache(s, name, size, flags); | |
598 | list_add(&s->list, &slab_caches); | |
599 | s->refcount = 1; | |
600 | return s; | |
601 | } | |
602 | ||
9425c58e CL |
603 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
604 | EXPORT_SYMBOL(kmalloc_caches); | |
605 | ||
606 | #ifdef CONFIG_ZONE_DMA | |
607 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
608 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
609 | #endif | |
610 | ||
2c59dd65 CL |
611 | /* |
612 | * Conversion table for small slabs sizes / 8 to the index in the | |
613 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
614 | * of two cache sizes there. The size of larger slabs can be determined using | |
615 | * fls. | |
616 | */ | |
617 | static s8 size_index[24] = { | |
618 | 3, /* 8 */ | |
619 | 4, /* 16 */ | |
620 | 5, /* 24 */ | |
621 | 5, /* 32 */ | |
622 | 6, /* 40 */ | |
623 | 6, /* 48 */ | |
624 | 6, /* 56 */ | |
625 | 6, /* 64 */ | |
626 | 1, /* 72 */ | |
627 | 1, /* 80 */ | |
628 | 1, /* 88 */ | |
629 | 1, /* 96 */ | |
630 | 7, /* 104 */ | |
631 | 7, /* 112 */ | |
632 | 7, /* 120 */ | |
633 | 7, /* 128 */ | |
634 | 2, /* 136 */ | |
635 | 2, /* 144 */ | |
636 | 2, /* 152 */ | |
637 | 2, /* 160 */ | |
638 | 2, /* 168 */ | |
639 | 2, /* 176 */ | |
640 | 2, /* 184 */ | |
641 | 2 /* 192 */ | |
642 | }; | |
643 | ||
644 | static inline int size_index_elem(size_t bytes) | |
645 | { | |
646 | return (bytes - 1) / 8; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Find the kmem_cache structure that serves a given size of | |
651 | * allocation | |
652 | */ | |
653 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
654 | { | |
655 | int index; | |
656 | ||
9de1bc87 | 657 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 658 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 659 | return NULL; |
907985f4 | 660 | } |
6286ae97 | 661 | |
2c59dd65 CL |
662 | if (size <= 192) { |
663 | if (!size) | |
664 | return ZERO_SIZE_PTR; | |
665 | ||
666 | index = size_index[size_index_elem(size)]; | |
667 | } else | |
668 | index = fls(size - 1); | |
669 | ||
670 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 671 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
672 | return kmalloc_dma_caches[index]; |
673 | ||
674 | #endif | |
675 | return kmalloc_caches[index]; | |
676 | } | |
677 | ||
f97d5f63 CL |
678 | /* |
679 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
680 | * may already have been created because they were needed to | |
681 | * enable allocations for slab creation. | |
682 | */ | |
683 | void __init create_kmalloc_caches(unsigned long flags) | |
684 | { | |
685 | int i; | |
686 | ||
2c59dd65 CL |
687 | /* |
688 | * Patch up the size_index table if we have strange large alignment | |
689 | * requirements for the kmalloc array. This is only the case for | |
690 | * MIPS it seems. The standard arches will not generate any code here. | |
691 | * | |
692 | * Largest permitted alignment is 256 bytes due to the way we | |
693 | * handle the index determination for the smaller caches. | |
694 | * | |
695 | * Make sure that nothing crazy happens if someone starts tinkering | |
696 | * around with ARCH_KMALLOC_MINALIGN | |
697 | */ | |
698 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
699 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
700 | ||
701 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
702 | int elem = size_index_elem(i); | |
703 | ||
704 | if (elem >= ARRAY_SIZE(size_index)) | |
705 | break; | |
706 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
707 | } | |
708 | ||
709 | if (KMALLOC_MIN_SIZE >= 64) { | |
710 | /* | |
711 | * The 96 byte size cache is not used if the alignment | |
712 | * is 64 byte. | |
713 | */ | |
714 | for (i = 64 + 8; i <= 96; i += 8) | |
715 | size_index[size_index_elem(i)] = 7; | |
716 | ||
717 | } | |
718 | ||
719 | if (KMALLOC_MIN_SIZE >= 128) { | |
720 | /* | |
721 | * The 192 byte sized cache is not used if the alignment | |
722 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
723 | * instead. | |
724 | */ | |
725 | for (i = 128 + 8; i <= 192; i += 8) | |
726 | size_index[size_index_elem(i)] = 8; | |
727 | } | |
8a965b3b CL |
728 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
729 | if (!kmalloc_caches[i]) { | |
f97d5f63 CL |
730 | kmalloc_caches[i] = create_kmalloc_cache(NULL, |
731 | 1 << i, flags); | |
956e46ef | 732 | } |
f97d5f63 | 733 | |
956e46ef CM |
734 | /* |
735 | * Caches that are not of the two-to-the-power-of size. | |
736 | * These have to be created immediately after the | |
737 | * earlier power of two caches | |
738 | */ | |
739 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) | |
740 | kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags); | |
8a965b3b | 741 | |
956e46ef CM |
742 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) |
743 | kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags); | |
8a965b3b CL |
744 | } |
745 | ||
f97d5f63 CL |
746 | /* Kmalloc array is now usable */ |
747 | slab_state = UP; | |
748 | ||
749 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
750 | struct kmem_cache *s = kmalloc_caches[i]; | |
751 | char *n; | |
752 | ||
753 | if (s) { | |
754 | n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i)); | |
755 | ||
756 | BUG_ON(!n); | |
757 | s->name = n; | |
758 | } | |
759 | } | |
760 | ||
761 | #ifdef CONFIG_ZONE_DMA | |
762 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
763 | struct kmem_cache *s = kmalloc_caches[i]; | |
764 | ||
765 | if (s) { | |
766 | int size = kmalloc_size(i); | |
767 | char *n = kasprintf(GFP_NOWAIT, | |
768 | "dma-kmalloc-%d", size); | |
769 | ||
770 | BUG_ON(!n); | |
771 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
772 | size, SLAB_CACHE_DMA | flags); | |
773 | } | |
774 | } | |
775 | #endif | |
776 | } | |
45530c44 CL |
777 | #endif /* !CONFIG_SLOB */ |
778 | ||
cea371f4 VD |
779 | /* |
780 | * To avoid unnecessary overhead, we pass through large allocation requests | |
781 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
782 | * know the allocation order to free the pages properly in kfree. | |
783 | */ | |
52383431 VD |
784 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
785 | { | |
786 | void *ret; | |
787 | struct page *page; | |
788 | ||
789 | flags |= __GFP_COMP; | |
790 | page = alloc_kmem_pages(flags, order); | |
791 | ret = page ? page_address(page) : NULL; | |
792 | kmemleak_alloc(ret, size, 1, flags); | |
793 | return ret; | |
794 | } | |
795 | EXPORT_SYMBOL(kmalloc_order); | |
796 | ||
f1b6eb6e CL |
797 | #ifdef CONFIG_TRACING |
798 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
799 | { | |
800 | void *ret = kmalloc_order(size, flags, order); | |
801 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
802 | return ret; | |
803 | } | |
804 | EXPORT_SYMBOL(kmalloc_order_trace); | |
805 | #endif | |
45530c44 | 806 | |
b7454ad3 | 807 | #ifdef CONFIG_SLABINFO |
e9b4db2b WL |
808 | |
809 | #ifdef CONFIG_SLAB | |
810 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
811 | #else | |
812 | #define SLABINFO_RIGHTS S_IRUSR | |
813 | #endif | |
814 | ||
b047501c | 815 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
816 | { |
817 | /* | |
818 | * Output format version, so at least we can change it | |
819 | * without _too_ many complaints. | |
820 | */ | |
821 | #ifdef CONFIG_DEBUG_SLAB | |
822 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
823 | #else | |
824 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
825 | #endif | |
826 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
827 | "<objperslab> <pagesperslab>"); | |
828 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
829 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
830 | #ifdef CONFIG_DEBUG_SLAB | |
831 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " | |
832 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); | |
833 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
834 | #endif | |
835 | seq_putc(m, '\n'); | |
836 | } | |
837 | ||
1df3b26f | 838 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 839 | { |
b7454ad3 | 840 | mutex_lock(&slab_mutex); |
b7454ad3 GC |
841 | return seq_list_start(&slab_caches, *pos); |
842 | } | |
843 | ||
276a2439 | 844 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 GC |
845 | { |
846 | return seq_list_next(p, &slab_caches, pos); | |
847 | } | |
848 | ||
276a2439 | 849 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
850 | { |
851 | mutex_unlock(&slab_mutex); | |
852 | } | |
853 | ||
749c5415 GC |
854 | static void |
855 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
856 | { | |
857 | struct kmem_cache *c; | |
858 | struct slabinfo sinfo; | |
859 | int i; | |
860 | ||
861 | if (!is_root_cache(s)) | |
862 | return; | |
863 | ||
864 | for_each_memcg_cache_index(i) { | |
2ade4de8 | 865 | c = cache_from_memcg_idx(s, i); |
749c5415 GC |
866 | if (!c) |
867 | continue; | |
868 | ||
869 | memset(&sinfo, 0, sizeof(sinfo)); | |
870 | get_slabinfo(c, &sinfo); | |
871 | ||
872 | info->active_slabs += sinfo.active_slabs; | |
873 | info->num_slabs += sinfo.num_slabs; | |
874 | info->shared_avail += sinfo.shared_avail; | |
875 | info->active_objs += sinfo.active_objs; | |
876 | info->num_objs += sinfo.num_objs; | |
877 | } | |
878 | } | |
879 | ||
b047501c | 880 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 881 | { |
0d7561c6 GC |
882 | struct slabinfo sinfo; |
883 | ||
884 | memset(&sinfo, 0, sizeof(sinfo)); | |
885 | get_slabinfo(s, &sinfo); | |
886 | ||
749c5415 GC |
887 | memcg_accumulate_slabinfo(s, &sinfo); |
888 | ||
0d7561c6 | 889 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 890 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
891 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
892 | ||
893 | seq_printf(m, " : tunables %4u %4u %4u", | |
894 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
895 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
896 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
897 | slabinfo_show_stats(m, s); | |
898 | seq_putc(m, '\n'); | |
b7454ad3 GC |
899 | } |
900 | ||
1df3b26f | 901 | static int slab_show(struct seq_file *m, void *p) |
749c5415 GC |
902 | { |
903 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
904 | ||
1df3b26f VD |
905 | if (p == slab_caches.next) |
906 | print_slabinfo_header(m); | |
b047501c VD |
907 | if (is_root_cache(s)) |
908 | cache_show(s, m); | |
909 | return 0; | |
910 | } | |
911 | ||
912 | #ifdef CONFIG_MEMCG_KMEM | |
913 | int memcg_slab_show(struct seq_file *m, void *p) | |
914 | { | |
915 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
916 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
917 | ||
918 | if (p == slab_caches.next) | |
919 | print_slabinfo_header(m); | |
920 | if (!is_root_cache(s) && s->memcg_params->memcg == memcg) | |
921 | cache_show(s, m); | |
922 | return 0; | |
749c5415 | 923 | } |
b047501c | 924 | #endif |
749c5415 | 925 | |
b7454ad3 GC |
926 | /* |
927 | * slabinfo_op - iterator that generates /proc/slabinfo | |
928 | * | |
929 | * Output layout: | |
930 | * cache-name | |
931 | * num-active-objs | |
932 | * total-objs | |
933 | * object size | |
934 | * num-active-slabs | |
935 | * total-slabs | |
936 | * num-pages-per-slab | |
937 | * + further values on SMP and with statistics enabled | |
938 | */ | |
939 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 940 | .start = slab_start, |
276a2439 WL |
941 | .next = slab_next, |
942 | .stop = slab_stop, | |
1df3b26f | 943 | .show = slab_show, |
b7454ad3 GC |
944 | }; |
945 | ||
946 | static int slabinfo_open(struct inode *inode, struct file *file) | |
947 | { | |
948 | return seq_open(file, &slabinfo_op); | |
949 | } | |
950 | ||
951 | static const struct file_operations proc_slabinfo_operations = { | |
952 | .open = slabinfo_open, | |
953 | .read = seq_read, | |
954 | .write = slabinfo_write, | |
955 | .llseek = seq_lseek, | |
956 | .release = seq_release, | |
957 | }; | |
958 | ||
959 | static int __init slab_proc_init(void) | |
960 | { | |
e9b4db2b WL |
961 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
962 | &proc_slabinfo_operations); | |
b7454ad3 GC |
963 | return 0; |
964 | } | |
965 | module_init(slab_proc_init); | |
966 | #endif /* CONFIG_SLABINFO */ | |
928cec9c AR |
967 | |
968 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
969 | gfp_t flags) | |
970 | { | |
971 | void *ret; | |
972 | size_t ks = 0; | |
973 | ||
974 | if (p) | |
975 | ks = ksize(p); | |
976 | ||
977 | if (ks >= new_size) | |
978 | return (void *)p; | |
979 | ||
980 | ret = kmalloc_track_caller(new_size, flags); | |
981 | if (ret && p) | |
982 | memcpy(ret, p, ks); | |
983 | ||
984 | return ret; | |
985 | } | |
986 | ||
987 | /** | |
988 | * __krealloc - like krealloc() but don't free @p. | |
989 | * @p: object to reallocate memory for. | |
990 | * @new_size: how many bytes of memory are required. | |
991 | * @flags: the type of memory to allocate. | |
992 | * | |
993 | * This function is like krealloc() except it never frees the originally | |
994 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
995 | * like, for example, with RCU. | |
996 | */ | |
997 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
998 | { | |
999 | if (unlikely(!new_size)) | |
1000 | return ZERO_SIZE_PTR; | |
1001 | ||
1002 | return __do_krealloc(p, new_size, flags); | |
1003 | ||
1004 | } | |
1005 | EXPORT_SYMBOL(__krealloc); | |
1006 | ||
1007 | /** | |
1008 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1009 | * @p: object to reallocate memory for. | |
1010 | * @new_size: how many bytes of memory are required. | |
1011 | * @flags: the type of memory to allocate. | |
1012 | * | |
1013 | * The contents of the object pointed to are preserved up to the | |
1014 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1015 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1016 | * %NULL pointer, the object pointed to is freed. | |
1017 | */ | |
1018 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1019 | { | |
1020 | void *ret; | |
1021 | ||
1022 | if (unlikely(!new_size)) { | |
1023 | kfree(p); | |
1024 | return ZERO_SIZE_PTR; | |
1025 | } | |
1026 | ||
1027 | ret = __do_krealloc(p, new_size, flags); | |
1028 | if (ret && p != ret) | |
1029 | kfree(p); | |
1030 | ||
1031 | return ret; | |
1032 | } | |
1033 | EXPORT_SYMBOL(krealloc); | |
1034 | ||
1035 | /** | |
1036 | * kzfree - like kfree but zero memory | |
1037 | * @p: object to free memory of | |
1038 | * | |
1039 | * The memory of the object @p points to is zeroed before freed. | |
1040 | * If @p is %NULL, kzfree() does nothing. | |
1041 | * | |
1042 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1043 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1044 | * careful when using this function in performance sensitive code. | |
1045 | */ | |
1046 | void kzfree(const void *p) | |
1047 | { | |
1048 | size_t ks; | |
1049 | void *mem = (void *)p; | |
1050 | ||
1051 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1052 | return; | |
1053 | ks = ksize(mem); | |
1054 | memset(mem, 0, ks); | |
1055 | kfree(mem); | |
1056 | } | |
1057 | EXPORT_SYMBOL(kzfree); | |
1058 | ||
1059 | /* Tracepoints definitions. */ | |
1060 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1061 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1062 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1063 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1064 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1065 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |