]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
039363f3 CL |
2 | /* |
3 | * Slab allocator functions that are independent of the allocator strategy | |
4 | * | |
5 | * (C) 2012 Christoph Lameter <cl@linux.com> | |
6 | */ | |
7 | #include <linux/slab.h> | |
8 | ||
9 | #include <linux/mm.h> | |
10 | #include <linux/poison.h> | |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/memory.h> | |
13 | #include <linux/compiler.h> | |
14 | #include <linux/module.h> | |
20cea968 CL |
15 | #include <linux/cpu.h> |
16 | #include <linux/uaccess.h> | |
b7454ad3 GC |
17 | #include <linux/seq_file.h> |
18 | #include <linux/proc_fs.h> | |
039363f3 CL |
19 | #include <asm/cacheflush.h> |
20 | #include <asm/tlbflush.h> | |
21 | #include <asm/page.h> | |
2633d7a0 | 22 | #include <linux/memcontrol.h> |
928cec9c AR |
23 | |
24 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 25 | #include <trace/events/kmem.h> |
039363f3 | 26 | |
97d06609 CL |
27 | #include "slab.h" |
28 | ||
29 | enum slab_state slab_state; | |
18004c5d CL |
30 | LIST_HEAD(slab_caches); |
31 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 32 | struct kmem_cache *kmem_cache; |
97d06609 | 33 | |
657dc2f9 TH |
34 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
35 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
36 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
37 | slab_caches_to_rcu_destroy_workfn); | |
38 | ||
423c929c JK |
39 | /* |
40 | * Set of flags that will prevent slab merging | |
41 | */ | |
42 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 43 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
7ed2f9e6 | 44 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 45 | |
230e9fc2 VD |
46 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
47 | SLAB_NOTRACK | SLAB_ACCOUNT) | |
423c929c JK |
48 | |
49 | /* | |
50 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 51 | */ |
7660a6fd | 52 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
53 | |
54 | static int __init setup_slab_nomerge(char *str) | |
55 | { | |
7660a6fd | 56 | slab_nomerge = true; |
423c929c JK |
57 | return 1; |
58 | } | |
59 | ||
60 | #ifdef CONFIG_SLUB | |
61 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
62 | #endif | |
63 | ||
64 | __setup("slab_nomerge", setup_slab_nomerge); | |
65 | ||
07f361b2 JK |
66 | /* |
67 | * Determine the size of a slab object | |
68 | */ | |
69 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
70 | { | |
71 | return s->object_size; | |
72 | } | |
73 | EXPORT_SYMBOL(kmem_cache_size); | |
74 | ||
77be4b13 | 75 | #ifdef CONFIG_DEBUG_VM |
794b1248 | 76 | static int kmem_cache_sanity_check(const char *name, size_t size) |
039363f3 CL |
77 | { |
78 | struct kmem_cache *s = NULL; | |
79 | ||
039363f3 CL |
80 | if (!name || in_interrupt() || size < sizeof(void *) || |
81 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
82 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
83 | return -EINVAL; | |
039363f3 | 84 | } |
b920536a | 85 | |
20cea968 CL |
86 | list_for_each_entry(s, &slab_caches, list) { |
87 | char tmp; | |
88 | int res; | |
89 | ||
90 | /* | |
91 | * This happens when the module gets unloaded and doesn't | |
92 | * destroy its slab cache and no-one else reuses the vmalloc | |
93 | * area of the module. Print a warning. | |
94 | */ | |
95 | res = probe_kernel_address(s->name, tmp); | |
96 | if (res) { | |
77be4b13 | 97 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
98 | s->object_size); |
99 | continue; | |
100 | } | |
20cea968 CL |
101 | } |
102 | ||
103 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
104 | return 0; |
105 | } | |
106 | #else | |
794b1248 | 107 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
77be4b13 SK |
108 | { |
109 | return 0; | |
110 | } | |
20cea968 CL |
111 | #endif |
112 | ||
484748f0 CL |
113 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
114 | { | |
115 | size_t i; | |
116 | ||
ca257195 JDB |
117 | for (i = 0; i < nr; i++) { |
118 | if (s) | |
119 | kmem_cache_free(s, p[i]); | |
120 | else | |
121 | kfree(p[i]); | |
122 | } | |
484748f0 CL |
123 | } |
124 | ||
865762a8 | 125 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
126 | void **p) |
127 | { | |
128 | size_t i; | |
129 | ||
130 | for (i = 0; i < nr; i++) { | |
131 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
132 | if (!x) { | |
133 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 134 | return 0; |
484748f0 CL |
135 | } |
136 | } | |
865762a8 | 137 | return i; |
484748f0 CL |
138 | } |
139 | ||
127424c8 | 140 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
510ded33 TH |
141 | |
142 | LIST_HEAD(slab_root_caches); | |
143 | ||
f7ce3190 | 144 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 145 | { |
9eeadc8b | 146 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 147 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 148 | INIT_LIST_HEAD(&s->memcg_params.children); |
f7ce3190 VD |
149 | } |
150 | ||
151 | static int init_memcg_params(struct kmem_cache *s, | |
152 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
153 | { | |
154 | struct memcg_cache_array *arr; | |
33a690c4 | 155 | |
9eeadc8b | 156 | if (root_cache) { |
f7ce3190 | 157 | s->memcg_params.root_cache = root_cache; |
9eeadc8b TH |
158 | s->memcg_params.memcg = memcg; |
159 | INIT_LIST_HEAD(&s->memcg_params.children_node); | |
bc2791f8 | 160 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 161 | return 0; |
f7ce3190 | 162 | } |
33a690c4 | 163 | |
f7ce3190 | 164 | slab_init_memcg_params(s); |
33a690c4 | 165 | |
f7ce3190 VD |
166 | if (!memcg_nr_cache_ids) |
167 | return 0; | |
33a690c4 | 168 | |
f80c7dab JW |
169 | arr = kvzalloc(sizeof(struct memcg_cache_array) + |
170 | memcg_nr_cache_ids * sizeof(void *), | |
171 | GFP_KERNEL); | |
f7ce3190 VD |
172 | if (!arr) |
173 | return -ENOMEM; | |
33a690c4 | 174 | |
f7ce3190 | 175 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
176 | return 0; |
177 | } | |
178 | ||
f7ce3190 | 179 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 180 | { |
f7ce3190 | 181 | if (is_root_cache(s)) |
f80c7dab JW |
182 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
183 | } | |
184 | ||
185 | static void free_memcg_params(struct rcu_head *rcu) | |
186 | { | |
187 | struct memcg_cache_array *old; | |
188 | ||
189 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
190 | kvfree(old); | |
33a690c4 VD |
191 | } |
192 | ||
f7ce3190 | 193 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 194 | { |
f7ce3190 | 195 | struct memcg_cache_array *old, *new; |
6f817f4c | 196 | |
f80c7dab JW |
197 | new = kvzalloc(sizeof(struct memcg_cache_array) + |
198 | new_array_size * sizeof(void *), GFP_KERNEL); | |
f7ce3190 | 199 | if (!new) |
6f817f4c VD |
200 | return -ENOMEM; |
201 | ||
f7ce3190 VD |
202 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
203 | lockdep_is_held(&slab_mutex)); | |
204 | if (old) | |
205 | memcpy(new->entries, old->entries, | |
206 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 207 | |
f7ce3190 VD |
208 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
209 | if (old) | |
f80c7dab | 210 | call_rcu(&old->rcu, free_memcg_params); |
6f817f4c VD |
211 | return 0; |
212 | } | |
213 | ||
55007d84 GC |
214 | int memcg_update_all_caches(int num_memcgs) |
215 | { | |
216 | struct kmem_cache *s; | |
217 | int ret = 0; | |
55007d84 | 218 | |
05257a1a | 219 | mutex_lock(&slab_mutex); |
510ded33 | 220 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
f7ce3190 | 221 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 222 | /* |
55007d84 GC |
223 | * Instead of freeing the memory, we'll just leave the caches |
224 | * up to this point in an updated state. | |
225 | */ | |
226 | if (ret) | |
05257a1a | 227 | break; |
55007d84 | 228 | } |
55007d84 GC |
229 | mutex_unlock(&slab_mutex); |
230 | return ret; | |
231 | } | |
657dc2f9 | 232 | |
510ded33 | 233 | void memcg_link_cache(struct kmem_cache *s) |
657dc2f9 | 234 | { |
510ded33 TH |
235 | if (is_root_cache(s)) { |
236 | list_add(&s->root_caches_node, &slab_root_caches); | |
237 | } else { | |
238 | list_add(&s->memcg_params.children_node, | |
239 | &s->memcg_params.root_cache->memcg_params.children); | |
240 | list_add(&s->memcg_params.kmem_caches_node, | |
241 | &s->memcg_params.memcg->kmem_caches); | |
242 | } | |
243 | } | |
244 | ||
245 | static void memcg_unlink_cache(struct kmem_cache *s) | |
246 | { | |
247 | if (is_root_cache(s)) { | |
248 | list_del(&s->root_caches_node); | |
249 | } else { | |
250 | list_del(&s->memcg_params.children_node); | |
251 | list_del(&s->memcg_params.kmem_caches_node); | |
252 | } | |
657dc2f9 | 253 | } |
33a690c4 | 254 | #else |
f7ce3190 VD |
255 | static inline int init_memcg_params(struct kmem_cache *s, |
256 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
33a690c4 VD |
257 | { |
258 | return 0; | |
259 | } | |
260 | ||
f7ce3190 | 261 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
262 | { |
263 | } | |
657dc2f9 | 264 | |
510ded33 | 265 | static inline void memcg_unlink_cache(struct kmem_cache *s) |
657dc2f9 TH |
266 | { |
267 | } | |
127424c8 | 268 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
55007d84 | 269 | |
423c929c JK |
270 | /* |
271 | * Find a mergeable slab cache | |
272 | */ | |
273 | int slab_unmergeable(struct kmem_cache *s) | |
274 | { | |
275 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
276 | return 1; | |
277 | ||
278 | if (!is_root_cache(s)) | |
279 | return 1; | |
280 | ||
281 | if (s->ctor) | |
282 | return 1; | |
283 | ||
284 | /* | |
285 | * We may have set a slab to be unmergeable during bootstrap. | |
286 | */ | |
287 | if (s->refcount < 0) | |
288 | return 1; | |
289 | ||
290 | return 0; | |
291 | } | |
292 | ||
293 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
294 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
295 | { | |
296 | struct kmem_cache *s; | |
297 | ||
c6e28895 | 298 | if (slab_nomerge) |
423c929c JK |
299 | return NULL; |
300 | ||
301 | if (ctor) | |
302 | return NULL; | |
303 | ||
304 | size = ALIGN(size, sizeof(void *)); | |
305 | align = calculate_alignment(flags, align, size); | |
306 | size = ALIGN(size, align); | |
307 | flags = kmem_cache_flags(size, flags, name, NULL); | |
308 | ||
c6e28895 GM |
309 | if (flags & SLAB_NEVER_MERGE) |
310 | return NULL; | |
311 | ||
510ded33 | 312 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { |
423c929c JK |
313 | if (slab_unmergeable(s)) |
314 | continue; | |
315 | ||
316 | if (size > s->size) | |
317 | continue; | |
318 | ||
319 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
320 | continue; | |
321 | /* | |
322 | * Check if alignment is compatible. | |
323 | * Courtesy of Adrian Drzewiecki | |
324 | */ | |
325 | if ((s->size & ~(align - 1)) != s->size) | |
326 | continue; | |
327 | ||
328 | if (s->size - size >= sizeof(void *)) | |
329 | continue; | |
330 | ||
95069ac8 JK |
331 | if (IS_ENABLED(CONFIG_SLAB) && align && |
332 | (align > s->align || s->align % align)) | |
333 | continue; | |
334 | ||
423c929c JK |
335 | return s; |
336 | } | |
337 | return NULL; | |
338 | } | |
339 | ||
45906855 CL |
340 | /* |
341 | * Figure out what the alignment of the objects will be given a set of | |
342 | * flags, a user specified alignment and the size of the objects. | |
343 | */ | |
344 | unsigned long calculate_alignment(unsigned long flags, | |
345 | unsigned long align, unsigned long size) | |
346 | { | |
347 | /* | |
348 | * If the user wants hardware cache aligned objects then follow that | |
349 | * suggestion if the object is sufficiently large. | |
350 | * | |
351 | * The hardware cache alignment cannot override the specified | |
352 | * alignment though. If that is greater then use it. | |
353 | */ | |
354 | if (flags & SLAB_HWCACHE_ALIGN) { | |
355 | unsigned long ralign = cache_line_size(); | |
356 | while (size <= ralign / 2) | |
357 | ralign /= 2; | |
358 | align = max(align, ralign); | |
359 | } | |
360 | ||
361 | if (align < ARCH_SLAB_MINALIGN) | |
362 | align = ARCH_SLAB_MINALIGN; | |
363 | ||
364 | return ALIGN(align, sizeof(void *)); | |
365 | } | |
366 | ||
c9a77a79 VD |
367 | static struct kmem_cache *create_cache(const char *name, |
368 | size_t object_size, size_t size, size_t align, | |
369 | unsigned long flags, void (*ctor)(void *), | |
370 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
794b1248 VD |
371 | { |
372 | struct kmem_cache *s; | |
373 | int err; | |
374 | ||
375 | err = -ENOMEM; | |
376 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
377 | if (!s) | |
378 | goto out; | |
379 | ||
380 | s->name = name; | |
381 | s->object_size = object_size; | |
382 | s->size = size; | |
383 | s->align = align; | |
384 | s->ctor = ctor; | |
385 | ||
f7ce3190 | 386 | err = init_memcg_params(s, memcg, root_cache); |
794b1248 VD |
387 | if (err) |
388 | goto out_free_cache; | |
389 | ||
390 | err = __kmem_cache_create(s, flags); | |
391 | if (err) | |
392 | goto out_free_cache; | |
393 | ||
394 | s->refcount = 1; | |
395 | list_add(&s->list, &slab_caches); | |
510ded33 | 396 | memcg_link_cache(s); |
794b1248 VD |
397 | out: |
398 | if (err) | |
399 | return ERR_PTR(err); | |
400 | return s; | |
401 | ||
402 | out_free_cache: | |
f7ce3190 | 403 | destroy_memcg_params(s); |
7c4da061 | 404 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
405 | goto out; |
406 | } | |
45906855 | 407 | |
77be4b13 SK |
408 | /* |
409 | * kmem_cache_create - Create a cache. | |
410 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
411 | * @size: The size of objects to be created in this cache. | |
412 | * @align: The required alignment for the objects. | |
413 | * @flags: SLAB flags | |
414 | * @ctor: A constructor for the objects. | |
415 | * | |
416 | * Returns a ptr to the cache on success, NULL on failure. | |
417 | * Cannot be called within a interrupt, but can be interrupted. | |
418 | * The @ctor is run when new pages are allocated by the cache. | |
419 | * | |
420 | * The flags are | |
421 | * | |
422 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
423 | * to catch references to uninitialised memory. | |
424 | * | |
425 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
426 | * for buffer overruns. | |
427 | * | |
428 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
429 | * cacheline. This can be beneficial if you're counting cycles as closely | |
430 | * as davem. | |
431 | */ | |
2633d7a0 | 432 | struct kmem_cache * |
794b1248 VD |
433 | kmem_cache_create(const char *name, size_t size, size_t align, |
434 | unsigned long flags, void (*ctor)(void *)) | |
77be4b13 | 435 | { |
40911a79 | 436 | struct kmem_cache *s = NULL; |
3dec16ea | 437 | const char *cache_name; |
3965fc36 | 438 | int err; |
039363f3 | 439 | |
77be4b13 | 440 | get_online_cpus(); |
03afc0e2 | 441 | get_online_mems(); |
05257a1a | 442 | memcg_get_cache_ids(); |
03afc0e2 | 443 | |
77be4b13 | 444 | mutex_lock(&slab_mutex); |
686d550d | 445 | |
794b1248 | 446 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 447 | if (err) { |
3965fc36 | 448 | goto out_unlock; |
3aa24f51 | 449 | } |
686d550d | 450 | |
e70954fd TG |
451 | /* Refuse requests with allocator specific flags */ |
452 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
453 | err = -EINVAL; | |
454 | goto out_unlock; | |
455 | } | |
456 | ||
d8843922 GC |
457 | /* |
458 | * Some allocators will constraint the set of valid flags to a subset | |
459 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
460 | * case, and we'll just provide them with a sanitized version of the | |
461 | * passed flags. | |
462 | */ | |
463 | flags &= CACHE_CREATE_MASK; | |
686d550d | 464 | |
794b1248 VD |
465 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
466 | if (s) | |
3965fc36 | 467 | goto out_unlock; |
2633d7a0 | 468 | |
3dec16ea | 469 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
470 | if (!cache_name) { |
471 | err = -ENOMEM; | |
472 | goto out_unlock; | |
473 | } | |
7c9adf5a | 474 | |
c9a77a79 VD |
475 | s = create_cache(cache_name, size, size, |
476 | calculate_alignment(flags, align, size), | |
477 | flags, ctor, NULL, NULL); | |
794b1248 VD |
478 | if (IS_ERR(s)) { |
479 | err = PTR_ERR(s); | |
3dec16ea | 480 | kfree_const(cache_name); |
794b1248 | 481 | } |
3965fc36 VD |
482 | |
483 | out_unlock: | |
20cea968 | 484 | mutex_unlock(&slab_mutex); |
03afc0e2 | 485 | |
05257a1a | 486 | memcg_put_cache_ids(); |
03afc0e2 | 487 | put_online_mems(); |
20cea968 CL |
488 | put_online_cpus(); |
489 | ||
ba3253c7 | 490 | if (err) { |
686d550d CL |
491 | if (flags & SLAB_PANIC) |
492 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
493 | name, err); | |
494 | else { | |
1170532b | 495 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
496 | name, err); |
497 | dump_stack(); | |
498 | } | |
686d550d CL |
499 | return NULL; |
500 | } | |
039363f3 CL |
501 | return s; |
502 | } | |
794b1248 | 503 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 504 | |
657dc2f9 | 505 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 506 | { |
657dc2f9 TH |
507 | LIST_HEAD(to_destroy); |
508 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 509 | |
657dc2f9 | 510 | /* |
5f0d5a3a | 511 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 TH |
512 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
513 | * through RCU and and the associated kmem_cache are dereferenced | |
514 | * while freeing the pages, so the kmem_caches should be freed only | |
515 | * after the pending RCU operations are finished. As rcu_barrier() | |
516 | * is a pretty slow operation, we batch all pending destructions | |
517 | * asynchronously. | |
518 | */ | |
519 | mutex_lock(&slab_mutex); | |
520 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
521 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 522 | |
657dc2f9 TH |
523 | if (list_empty(&to_destroy)) |
524 | return; | |
525 | ||
526 | rcu_barrier(); | |
527 | ||
528 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
529 | #ifdef SLAB_SUPPORTS_SYSFS | |
530 | sysfs_slab_release(s); | |
531 | #else | |
532 | slab_kmem_cache_release(s); | |
533 | #endif | |
534 | } | |
d5b3cf71 VD |
535 | } |
536 | ||
657dc2f9 | 537 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 538 | { |
f9fa1d91 GT |
539 | /* free asan quarantined objects */ |
540 | kasan_cache_shutdown(s); | |
541 | ||
657dc2f9 TH |
542 | if (__kmem_cache_shutdown(s) != 0) |
543 | return -EBUSY; | |
d5b3cf71 | 544 | |
510ded33 | 545 | memcg_unlink_cache(s); |
657dc2f9 | 546 | list_del(&s->list); |
d5b3cf71 | 547 | |
5f0d5a3a | 548 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
657dc2f9 TH |
549 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
550 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
551 | } else { | |
d5b3cf71 | 552 | #ifdef SLAB_SUPPORTS_SYSFS |
bf5eb3de | 553 | sysfs_slab_release(s); |
d5b3cf71 VD |
554 | #else |
555 | slab_kmem_cache_release(s); | |
556 | #endif | |
557 | } | |
657dc2f9 TH |
558 | |
559 | return 0; | |
d5b3cf71 VD |
560 | } |
561 | ||
127424c8 | 562 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
794b1248 | 563 | /* |
776ed0f0 | 564 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
565 | * @memcg: The memory cgroup the new cache is for. |
566 | * @root_cache: The parent of the new cache. | |
567 | * | |
568 | * This function attempts to create a kmem cache that will serve allocation | |
569 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
570 | * from its parent. | |
571 | */ | |
d5b3cf71 VD |
572 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
573 | struct kmem_cache *root_cache) | |
2633d7a0 | 574 | { |
3e0350a3 | 575 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 576 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 577 | struct memcg_cache_array *arr; |
bd673145 | 578 | struct kmem_cache *s = NULL; |
794b1248 | 579 | char *cache_name; |
f7ce3190 | 580 | int idx; |
794b1248 VD |
581 | |
582 | get_online_cpus(); | |
03afc0e2 VD |
583 | get_online_mems(); |
584 | ||
794b1248 VD |
585 | mutex_lock(&slab_mutex); |
586 | ||
2a4db7eb | 587 | /* |
567e9ab2 | 588 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
589 | * creation work was pending. |
590 | */ | |
b6ecd2de | 591 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
592 | goto out_unlock; |
593 | ||
f7ce3190 VD |
594 | idx = memcg_cache_id(memcg); |
595 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
596 | lockdep_is_held(&slab_mutex)); | |
597 | ||
d5b3cf71 VD |
598 | /* |
599 | * Since per-memcg caches are created asynchronously on first | |
600 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
601 | * create the same cache, but only one of them may succeed. | |
602 | */ | |
f7ce3190 | 603 | if (arr->entries[idx]) |
d5b3cf71 VD |
604 | goto out_unlock; |
605 | ||
f1008365 | 606 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
607 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
608 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
609 | if (!cache_name) |
610 | goto out_unlock; | |
611 | ||
c9a77a79 VD |
612 | s = create_cache(cache_name, root_cache->object_size, |
613 | root_cache->size, root_cache->align, | |
f773e36d GT |
614 | root_cache->flags & CACHE_CREATE_MASK, |
615 | root_cache->ctor, memcg, root_cache); | |
d5b3cf71 VD |
616 | /* |
617 | * If we could not create a memcg cache, do not complain, because | |
618 | * that's not critical at all as we can always proceed with the root | |
619 | * cache. | |
620 | */ | |
bd673145 | 621 | if (IS_ERR(s)) { |
794b1248 | 622 | kfree(cache_name); |
d5b3cf71 | 623 | goto out_unlock; |
bd673145 | 624 | } |
794b1248 | 625 | |
d5b3cf71 VD |
626 | /* |
627 | * Since readers won't lock (see cache_from_memcg_idx()), we need a | |
628 | * barrier here to ensure nobody will see the kmem_cache partially | |
629 | * initialized. | |
630 | */ | |
631 | smp_wmb(); | |
f7ce3190 | 632 | arr->entries[idx] = s; |
d5b3cf71 | 633 | |
794b1248 VD |
634 | out_unlock: |
635 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
636 | |
637 | put_online_mems(); | |
794b1248 | 638 | put_online_cpus(); |
2633d7a0 | 639 | } |
b8529907 | 640 | |
01fb58bc TH |
641 | static void kmemcg_deactivate_workfn(struct work_struct *work) |
642 | { | |
643 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
644 | memcg_params.deact_work); | |
645 | ||
646 | get_online_cpus(); | |
647 | get_online_mems(); | |
648 | ||
649 | mutex_lock(&slab_mutex); | |
650 | ||
651 | s->memcg_params.deact_fn(s); | |
652 | ||
653 | mutex_unlock(&slab_mutex); | |
654 | ||
655 | put_online_mems(); | |
656 | put_online_cpus(); | |
657 | ||
658 | /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */ | |
659 | css_put(&s->memcg_params.memcg->css); | |
660 | } | |
661 | ||
662 | static void kmemcg_deactivate_rcufn(struct rcu_head *head) | |
663 | { | |
664 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
665 | memcg_params.deact_rcu_head); | |
666 | ||
667 | /* | |
668 | * We need to grab blocking locks. Bounce to ->deact_work. The | |
669 | * work item shares the space with the RCU head and can't be | |
670 | * initialized eariler. | |
671 | */ | |
672 | INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn); | |
17cc4dfe | 673 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work); |
01fb58bc TH |
674 | } |
675 | ||
676 | /** | |
677 | * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a | |
678 | * sched RCU grace period | |
679 | * @s: target kmem_cache | |
680 | * @deact_fn: deactivation function to call | |
681 | * | |
682 | * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex | |
683 | * held after a sched RCU grace period. The slab is guaranteed to stay | |
684 | * alive until @deact_fn is finished. This is to be used from | |
685 | * __kmemcg_cache_deactivate(). | |
686 | */ | |
687 | void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s, | |
688 | void (*deact_fn)(struct kmem_cache *)) | |
689 | { | |
690 | if (WARN_ON_ONCE(is_root_cache(s)) || | |
691 | WARN_ON_ONCE(s->memcg_params.deact_fn)) | |
692 | return; | |
693 | ||
694 | /* pin memcg so that @s doesn't get destroyed in the middle */ | |
695 | css_get(&s->memcg_params.memcg->css); | |
696 | ||
697 | s->memcg_params.deact_fn = deact_fn; | |
698 | call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn); | |
699 | } | |
700 | ||
2a4db7eb VD |
701 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
702 | { | |
703 | int idx; | |
704 | struct memcg_cache_array *arr; | |
d6e0b7fa | 705 | struct kmem_cache *s, *c; |
2a4db7eb VD |
706 | |
707 | idx = memcg_cache_id(memcg); | |
708 | ||
d6e0b7fa VD |
709 | get_online_cpus(); |
710 | get_online_mems(); | |
711 | ||
2a4db7eb | 712 | mutex_lock(&slab_mutex); |
510ded33 | 713 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
2a4db7eb VD |
714 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
715 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
716 | c = arr->entries[idx]; |
717 | if (!c) | |
718 | continue; | |
719 | ||
c9fc5864 | 720 | __kmemcg_cache_deactivate(c); |
2a4db7eb VD |
721 | arr->entries[idx] = NULL; |
722 | } | |
723 | mutex_unlock(&slab_mutex); | |
d6e0b7fa VD |
724 | |
725 | put_online_mems(); | |
726 | put_online_cpus(); | |
2a4db7eb VD |
727 | } |
728 | ||
d5b3cf71 | 729 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
b8529907 | 730 | { |
d5b3cf71 | 731 | struct kmem_cache *s, *s2; |
b8529907 | 732 | |
d5b3cf71 VD |
733 | get_online_cpus(); |
734 | get_online_mems(); | |
b8529907 | 735 | |
b8529907 | 736 | mutex_lock(&slab_mutex); |
bc2791f8 TH |
737 | list_for_each_entry_safe(s, s2, &memcg->kmem_caches, |
738 | memcg_params.kmem_caches_node) { | |
d5b3cf71 VD |
739 | /* |
740 | * The cgroup is about to be freed and therefore has no charges | |
741 | * left. Hence, all its caches must be empty by now. | |
742 | */ | |
657dc2f9 | 743 | BUG_ON(shutdown_cache(s)); |
d5b3cf71 VD |
744 | } |
745 | mutex_unlock(&slab_mutex); | |
b8529907 | 746 | |
d5b3cf71 VD |
747 | put_online_mems(); |
748 | put_online_cpus(); | |
b8529907 | 749 | } |
d60fdcc9 | 750 | |
657dc2f9 | 751 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
752 | { |
753 | struct memcg_cache_array *arr; | |
754 | struct kmem_cache *c, *c2; | |
755 | LIST_HEAD(busy); | |
756 | int i; | |
757 | ||
758 | BUG_ON(!is_root_cache(s)); | |
759 | ||
760 | /* | |
761 | * First, shutdown active caches, i.e. caches that belong to online | |
762 | * memory cgroups. | |
763 | */ | |
764 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
765 | lockdep_is_held(&slab_mutex)); | |
766 | for_each_memcg_cache_index(i) { | |
767 | c = arr->entries[i]; | |
768 | if (!c) | |
769 | continue; | |
657dc2f9 | 770 | if (shutdown_cache(c)) |
d60fdcc9 VD |
771 | /* |
772 | * The cache still has objects. Move it to a temporary | |
773 | * list so as not to try to destroy it for a second | |
774 | * time while iterating over inactive caches below. | |
775 | */ | |
9eeadc8b | 776 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
777 | else |
778 | /* | |
779 | * The cache is empty and will be destroyed soon. Clear | |
780 | * the pointer to it in the memcg_caches array so that | |
781 | * it will never be accessed even if the root cache | |
782 | * stays alive. | |
783 | */ | |
784 | arr->entries[i] = NULL; | |
785 | } | |
786 | ||
787 | /* | |
788 | * Second, shutdown all caches left from memory cgroups that are now | |
789 | * offline. | |
790 | */ | |
9eeadc8b TH |
791 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
792 | memcg_params.children_node) | |
657dc2f9 | 793 | shutdown_cache(c); |
d60fdcc9 | 794 | |
9eeadc8b | 795 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
796 | |
797 | /* | |
798 | * A cache being destroyed must be empty. In particular, this means | |
799 | * that all per memcg caches attached to it must be empty too. | |
800 | */ | |
9eeadc8b | 801 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
802 | return -EBUSY; |
803 | return 0; | |
804 | } | |
805 | #else | |
657dc2f9 | 806 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
807 | { |
808 | return 0; | |
809 | } | |
127424c8 | 810 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
97d06609 | 811 | |
41a21285 CL |
812 | void slab_kmem_cache_release(struct kmem_cache *s) |
813 | { | |
52b4b950 | 814 | __kmem_cache_release(s); |
f7ce3190 | 815 | destroy_memcg_params(s); |
3dec16ea | 816 | kfree_const(s->name); |
41a21285 CL |
817 | kmem_cache_free(kmem_cache, s); |
818 | } | |
819 | ||
945cf2b6 CL |
820 | void kmem_cache_destroy(struct kmem_cache *s) |
821 | { | |
d60fdcc9 | 822 | int err; |
d5b3cf71 | 823 | |
3942d299 SS |
824 | if (unlikely(!s)) |
825 | return; | |
826 | ||
945cf2b6 | 827 | get_online_cpus(); |
03afc0e2 VD |
828 | get_online_mems(); |
829 | ||
945cf2b6 | 830 | mutex_lock(&slab_mutex); |
b8529907 | 831 | |
945cf2b6 | 832 | s->refcount--; |
b8529907 VD |
833 | if (s->refcount) |
834 | goto out_unlock; | |
835 | ||
657dc2f9 | 836 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 837 | if (!err) |
657dc2f9 | 838 | err = shutdown_cache(s); |
b8529907 | 839 | |
cd918c55 | 840 | if (err) { |
756a025f JP |
841 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
842 | s->name); | |
cd918c55 VD |
843 | dump_stack(); |
844 | } | |
b8529907 VD |
845 | out_unlock: |
846 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 847 | |
03afc0e2 | 848 | put_online_mems(); |
945cf2b6 CL |
849 | put_online_cpus(); |
850 | } | |
851 | EXPORT_SYMBOL(kmem_cache_destroy); | |
852 | ||
03afc0e2 VD |
853 | /** |
854 | * kmem_cache_shrink - Shrink a cache. | |
855 | * @cachep: The cache to shrink. | |
856 | * | |
857 | * Releases as many slabs as possible for a cache. | |
858 | * To help debugging, a zero exit status indicates all slabs were released. | |
859 | */ | |
860 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
861 | { | |
862 | int ret; | |
863 | ||
864 | get_online_cpus(); | |
865 | get_online_mems(); | |
55834c59 | 866 | kasan_cache_shrink(cachep); |
c9fc5864 | 867 | ret = __kmem_cache_shrink(cachep); |
03afc0e2 VD |
868 | put_online_mems(); |
869 | put_online_cpus(); | |
870 | return ret; | |
871 | } | |
872 | EXPORT_SYMBOL(kmem_cache_shrink); | |
873 | ||
fda90124 | 874 | bool slab_is_available(void) |
97d06609 CL |
875 | { |
876 | return slab_state >= UP; | |
877 | } | |
b7454ad3 | 878 | |
45530c44 CL |
879 | #ifndef CONFIG_SLOB |
880 | /* Create a cache during boot when no slab services are available yet */ | |
881 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
882 | unsigned long flags) | |
883 | { | |
884 | int err; | |
885 | ||
886 | s->name = name; | |
887 | s->size = s->object_size = size; | |
45906855 | 888 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
f7ce3190 VD |
889 | |
890 | slab_init_memcg_params(s); | |
891 | ||
45530c44 CL |
892 | err = __kmem_cache_create(s, flags); |
893 | ||
894 | if (err) | |
31ba7346 | 895 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
896 | name, size, err); |
897 | ||
898 | s->refcount = -1; /* Exempt from merging for now */ | |
899 | } | |
900 | ||
901 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
902 | unsigned long flags) | |
903 | { | |
904 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
905 | ||
906 | if (!s) | |
907 | panic("Out of memory when creating slab %s\n", name); | |
908 | ||
909 | create_boot_cache(s, name, size, flags); | |
910 | list_add(&s->list, &slab_caches); | |
510ded33 | 911 | memcg_link_cache(s); |
45530c44 CL |
912 | s->refcount = 1; |
913 | return s; | |
914 | } | |
915 | ||
9425c58e CL |
916 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
917 | EXPORT_SYMBOL(kmalloc_caches); | |
918 | ||
919 | #ifdef CONFIG_ZONE_DMA | |
920 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
921 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
922 | #endif | |
923 | ||
2c59dd65 CL |
924 | /* |
925 | * Conversion table for small slabs sizes / 8 to the index in the | |
926 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
927 | * of two cache sizes there. The size of larger slabs can be determined using | |
928 | * fls. | |
929 | */ | |
930 | static s8 size_index[24] = { | |
931 | 3, /* 8 */ | |
932 | 4, /* 16 */ | |
933 | 5, /* 24 */ | |
934 | 5, /* 32 */ | |
935 | 6, /* 40 */ | |
936 | 6, /* 48 */ | |
937 | 6, /* 56 */ | |
938 | 6, /* 64 */ | |
939 | 1, /* 72 */ | |
940 | 1, /* 80 */ | |
941 | 1, /* 88 */ | |
942 | 1, /* 96 */ | |
943 | 7, /* 104 */ | |
944 | 7, /* 112 */ | |
945 | 7, /* 120 */ | |
946 | 7, /* 128 */ | |
947 | 2, /* 136 */ | |
948 | 2, /* 144 */ | |
949 | 2, /* 152 */ | |
950 | 2, /* 160 */ | |
951 | 2, /* 168 */ | |
952 | 2, /* 176 */ | |
953 | 2, /* 184 */ | |
954 | 2 /* 192 */ | |
955 | }; | |
956 | ||
957 | static inline int size_index_elem(size_t bytes) | |
958 | { | |
959 | return (bytes - 1) / 8; | |
960 | } | |
961 | ||
962 | /* | |
963 | * Find the kmem_cache structure that serves a given size of | |
964 | * allocation | |
965 | */ | |
966 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
967 | { | |
968 | int index; | |
969 | ||
9de1bc87 | 970 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 971 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 972 | return NULL; |
907985f4 | 973 | } |
6286ae97 | 974 | |
2c59dd65 CL |
975 | if (size <= 192) { |
976 | if (!size) | |
977 | return ZERO_SIZE_PTR; | |
978 | ||
979 | index = size_index[size_index_elem(size)]; | |
980 | } else | |
981 | index = fls(size - 1); | |
982 | ||
983 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 984 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
985 | return kmalloc_dma_caches[index]; |
986 | ||
987 | #endif | |
988 | return kmalloc_caches[index]; | |
989 | } | |
990 | ||
4066c33d GG |
991 | /* |
992 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
993 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
994 | * kmalloc-67108864. | |
995 | */ | |
af3b5f87 | 996 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
4066c33d GG |
997 | {NULL, 0}, {"kmalloc-96", 96}, |
998 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
999 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
1000 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
1001 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
1002 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, | |
1003 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, | |
1004 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, | |
1005 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, | |
1006 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, | |
1007 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, | |
1008 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, | |
1009 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, | |
1010 | {"kmalloc-67108864", 67108864} | |
1011 | }; | |
1012 | ||
f97d5f63 | 1013 | /* |
34cc6990 DS |
1014 | * Patch up the size_index table if we have strange large alignment |
1015 | * requirements for the kmalloc array. This is only the case for | |
1016 | * MIPS it seems. The standard arches will not generate any code here. | |
1017 | * | |
1018 | * Largest permitted alignment is 256 bytes due to the way we | |
1019 | * handle the index determination for the smaller caches. | |
1020 | * | |
1021 | * Make sure that nothing crazy happens if someone starts tinkering | |
1022 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 1023 | */ |
34cc6990 | 1024 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 CL |
1025 | { |
1026 | int i; | |
1027 | ||
2c59dd65 CL |
1028 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
1029 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1030 | ||
1031 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
1032 | int elem = size_index_elem(i); | |
1033 | ||
1034 | if (elem >= ARRAY_SIZE(size_index)) | |
1035 | break; | |
1036 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1037 | } | |
1038 | ||
1039 | if (KMALLOC_MIN_SIZE >= 64) { | |
1040 | /* | |
1041 | * The 96 byte size cache is not used if the alignment | |
1042 | * is 64 byte. | |
1043 | */ | |
1044 | for (i = 64 + 8; i <= 96; i += 8) | |
1045 | size_index[size_index_elem(i)] = 7; | |
1046 | ||
1047 | } | |
1048 | ||
1049 | if (KMALLOC_MIN_SIZE >= 128) { | |
1050 | /* | |
1051 | * The 192 byte sized cache is not used if the alignment | |
1052 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1053 | * instead. | |
1054 | */ | |
1055 | for (i = 128 + 8; i <= 192; i += 8) | |
1056 | size_index[size_index_elem(i)] = 8; | |
1057 | } | |
34cc6990 DS |
1058 | } |
1059 | ||
ae6f2462 | 1060 | static void __init new_kmalloc_cache(int idx, unsigned long flags) |
a9730fca CL |
1061 | { |
1062 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, | |
1063 | kmalloc_info[idx].size, flags); | |
1064 | } | |
1065 | ||
34cc6990 DS |
1066 | /* |
1067 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1068 | * may already have been created because they were needed to | |
1069 | * enable allocations for slab creation. | |
1070 | */ | |
1071 | void __init create_kmalloc_caches(unsigned long flags) | |
1072 | { | |
1073 | int i; | |
1074 | ||
a9730fca CL |
1075 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
1076 | if (!kmalloc_caches[i]) | |
1077 | new_kmalloc_cache(i, flags); | |
f97d5f63 | 1078 | |
956e46ef | 1079 | /* |
a9730fca CL |
1080 | * Caches that are not of the two-to-the-power-of size. |
1081 | * These have to be created immediately after the | |
1082 | * earlier power of two caches | |
956e46ef | 1083 | */ |
a9730fca CL |
1084 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
1085 | new_kmalloc_cache(1, flags); | |
1086 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) | |
1087 | new_kmalloc_cache(2, flags); | |
8a965b3b CL |
1088 | } |
1089 | ||
f97d5f63 CL |
1090 | /* Kmalloc array is now usable */ |
1091 | slab_state = UP; | |
1092 | ||
f97d5f63 CL |
1093 | #ifdef CONFIG_ZONE_DMA |
1094 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1095 | struct kmem_cache *s = kmalloc_caches[i]; | |
1096 | ||
1097 | if (s) { | |
1098 | int size = kmalloc_size(i); | |
1099 | char *n = kasprintf(GFP_NOWAIT, | |
1100 | "dma-kmalloc-%d", size); | |
1101 | ||
1102 | BUG_ON(!n); | |
1103 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
1104 | size, SLAB_CACHE_DMA | flags); | |
1105 | } | |
1106 | } | |
1107 | #endif | |
1108 | } | |
45530c44 CL |
1109 | #endif /* !CONFIG_SLOB */ |
1110 | ||
cea371f4 VD |
1111 | /* |
1112 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1113 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1114 | * know the allocation order to free the pages properly in kfree. | |
1115 | */ | |
52383431 VD |
1116 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1117 | { | |
1118 | void *ret; | |
1119 | struct page *page; | |
1120 | ||
1121 | flags |= __GFP_COMP; | |
4949148a | 1122 | page = alloc_pages(flags, order); |
52383431 VD |
1123 | ret = page ? page_address(page) : NULL; |
1124 | kmemleak_alloc(ret, size, 1, flags); | |
505f5dcb | 1125 | kasan_kmalloc_large(ret, size, flags); |
52383431 VD |
1126 | return ret; |
1127 | } | |
1128 | EXPORT_SYMBOL(kmalloc_order); | |
1129 | ||
f1b6eb6e CL |
1130 | #ifdef CONFIG_TRACING |
1131 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1132 | { | |
1133 | void *ret = kmalloc_order(size, flags, order); | |
1134 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1135 | return ret; | |
1136 | } | |
1137 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1138 | #endif | |
45530c44 | 1139 | |
7c00fce9 TG |
1140 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1141 | /* Randomize a generic freelist */ | |
1142 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
1143 | size_t count) | |
1144 | { | |
1145 | size_t i; | |
1146 | unsigned int rand; | |
1147 | ||
1148 | for (i = 0; i < count; i++) | |
1149 | list[i] = i; | |
1150 | ||
1151 | /* Fisher-Yates shuffle */ | |
1152 | for (i = count - 1; i > 0; i--) { | |
1153 | rand = prandom_u32_state(state); | |
1154 | rand %= (i + 1); | |
1155 | swap(list[i], list[rand]); | |
1156 | } | |
1157 | } | |
1158 | ||
1159 | /* Create a random sequence per cache */ | |
1160 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1161 | gfp_t gfp) | |
1162 | { | |
1163 | struct rnd_state state; | |
1164 | ||
1165 | if (count < 2 || cachep->random_seq) | |
1166 | return 0; | |
1167 | ||
1168 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1169 | if (!cachep->random_seq) | |
1170 | return -ENOMEM; | |
1171 | ||
1172 | /* Get best entropy at this stage of boot */ | |
1173 | prandom_seed_state(&state, get_random_long()); | |
1174 | ||
1175 | freelist_randomize(&state, cachep->random_seq, count); | |
1176 | return 0; | |
1177 | } | |
1178 | ||
1179 | /* Destroy the per-cache random freelist sequence */ | |
1180 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1181 | { | |
1182 | kfree(cachep->random_seq); | |
1183 | cachep->random_seq = NULL; | |
1184 | } | |
1185 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1186 | ||
5b365771 | 1187 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b WL |
1188 | #ifdef CONFIG_SLAB |
1189 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
1190 | #else | |
1191 | #define SLABINFO_RIGHTS S_IRUSR | |
1192 | #endif | |
1193 | ||
b047501c | 1194 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1195 | { |
1196 | /* | |
1197 | * Output format version, so at least we can change it | |
1198 | * without _too_ many complaints. | |
1199 | */ | |
1200 | #ifdef CONFIG_DEBUG_SLAB | |
1201 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1202 | #else | |
1203 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1204 | #endif | |
756a025f | 1205 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1206 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1207 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1208 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1209 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1210 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1211 | #endif | |
1212 | seq_putc(m, '\n'); | |
1213 | } | |
1214 | ||
1df3b26f | 1215 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1216 | { |
b7454ad3 | 1217 | mutex_lock(&slab_mutex); |
510ded33 | 1218 | return seq_list_start(&slab_root_caches, *pos); |
b7454ad3 GC |
1219 | } |
1220 | ||
276a2439 | 1221 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1222 | { |
510ded33 | 1223 | return seq_list_next(p, &slab_root_caches, pos); |
b7454ad3 GC |
1224 | } |
1225 | ||
276a2439 | 1226 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1227 | { |
1228 | mutex_unlock(&slab_mutex); | |
1229 | } | |
1230 | ||
749c5415 GC |
1231 | static void |
1232 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1233 | { | |
1234 | struct kmem_cache *c; | |
1235 | struct slabinfo sinfo; | |
749c5415 GC |
1236 | |
1237 | if (!is_root_cache(s)) | |
1238 | return; | |
1239 | ||
426589f5 | 1240 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1241 | memset(&sinfo, 0, sizeof(sinfo)); |
1242 | get_slabinfo(c, &sinfo); | |
1243 | ||
1244 | info->active_slabs += sinfo.active_slabs; | |
1245 | info->num_slabs += sinfo.num_slabs; | |
1246 | info->shared_avail += sinfo.shared_avail; | |
1247 | info->active_objs += sinfo.active_objs; | |
1248 | info->num_objs += sinfo.num_objs; | |
1249 | } | |
1250 | } | |
1251 | ||
b047501c | 1252 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1253 | { |
0d7561c6 GC |
1254 | struct slabinfo sinfo; |
1255 | ||
1256 | memset(&sinfo, 0, sizeof(sinfo)); | |
1257 | get_slabinfo(s, &sinfo); | |
1258 | ||
749c5415 GC |
1259 | memcg_accumulate_slabinfo(s, &sinfo); |
1260 | ||
0d7561c6 | 1261 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1262 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1263 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1264 | ||
1265 | seq_printf(m, " : tunables %4u %4u %4u", | |
1266 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1267 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1268 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1269 | slabinfo_show_stats(m, s); | |
1270 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1271 | } |
1272 | ||
1df3b26f | 1273 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1274 | { |
510ded33 | 1275 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); |
749c5415 | 1276 | |
510ded33 | 1277 | if (p == slab_root_caches.next) |
1df3b26f | 1278 | print_slabinfo_header(m); |
510ded33 | 1279 | cache_show(s, m); |
b047501c VD |
1280 | return 0; |
1281 | } | |
1282 | ||
852d8be0 YS |
1283 | void dump_unreclaimable_slab(void) |
1284 | { | |
1285 | struct kmem_cache *s, *s2; | |
1286 | struct slabinfo sinfo; | |
1287 | ||
1288 | /* | |
1289 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1290 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1291 | * risk of crash. | |
1292 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1293 | * without acquiring the mutex. | |
1294 | */ | |
1295 | if (!mutex_trylock(&slab_mutex)) { | |
1296 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1297 | return; | |
1298 | } | |
1299 | ||
1300 | pr_info("Unreclaimable slab info:\n"); | |
1301 | pr_info("Name Used Total\n"); | |
1302 | ||
1303 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1304 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1305 | continue; | |
1306 | ||
1307 | get_slabinfo(s, &sinfo); | |
1308 | ||
1309 | if (sinfo.num_objs > 0) | |
1310 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1311 | (sinfo.active_objs * s->size) / 1024, | |
1312 | (sinfo.num_objs * s->size) / 1024); | |
1313 | } | |
1314 | mutex_unlock(&slab_mutex); | |
1315 | } | |
1316 | ||
5b365771 | 1317 | #if defined(CONFIG_MEMCG) |
bc2791f8 TH |
1318 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1319 | { | |
1320 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1321 | ||
1322 | mutex_lock(&slab_mutex); | |
1323 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1324 | } | |
1325 | ||
1326 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1327 | { | |
1328 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1329 | ||
1330 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1331 | } | |
1332 | ||
1333 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1334 | { | |
1335 | mutex_unlock(&slab_mutex); | |
1336 | } | |
1337 | ||
b047501c VD |
1338 | int memcg_slab_show(struct seq_file *m, void *p) |
1339 | { | |
bc2791f8 TH |
1340 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1341 | memcg_params.kmem_caches_node); | |
b047501c VD |
1342 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
1343 | ||
bc2791f8 | 1344 | if (p == memcg->kmem_caches.next) |
b047501c | 1345 | print_slabinfo_header(m); |
bc2791f8 | 1346 | cache_show(s, m); |
b047501c | 1347 | return 0; |
749c5415 | 1348 | } |
b047501c | 1349 | #endif |
749c5415 | 1350 | |
b7454ad3 GC |
1351 | /* |
1352 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1353 | * | |
1354 | * Output layout: | |
1355 | * cache-name | |
1356 | * num-active-objs | |
1357 | * total-objs | |
1358 | * object size | |
1359 | * num-active-slabs | |
1360 | * total-slabs | |
1361 | * num-pages-per-slab | |
1362 | * + further values on SMP and with statistics enabled | |
1363 | */ | |
1364 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1365 | .start = slab_start, |
276a2439 WL |
1366 | .next = slab_next, |
1367 | .stop = slab_stop, | |
1df3b26f | 1368 | .show = slab_show, |
b7454ad3 GC |
1369 | }; |
1370 | ||
1371 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1372 | { | |
1373 | return seq_open(file, &slabinfo_op); | |
1374 | } | |
1375 | ||
1376 | static const struct file_operations proc_slabinfo_operations = { | |
1377 | .open = slabinfo_open, | |
1378 | .read = seq_read, | |
1379 | .write = slabinfo_write, | |
1380 | .llseek = seq_lseek, | |
1381 | .release = seq_release, | |
1382 | }; | |
1383 | ||
1384 | static int __init slab_proc_init(void) | |
1385 | { | |
e9b4db2b WL |
1386 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
1387 | &proc_slabinfo_operations); | |
b7454ad3 GC |
1388 | return 0; |
1389 | } | |
1390 | module_init(slab_proc_init); | |
5b365771 | 1391 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1392 | |
1393 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1394 | gfp_t flags) | |
1395 | { | |
1396 | void *ret; | |
1397 | size_t ks = 0; | |
1398 | ||
1399 | if (p) | |
1400 | ks = ksize(p); | |
1401 | ||
0316bec2 | 1402 | if (ks >= new_size) { |
505f5dcb | 1403 | kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1404 | return (void *)p; |
0316bec2 | 1405 | } |
928cec9c AR |
1406 | |
1407 | ret = kmalloc_track_caller(new_size, flags); | |
1408 | if (ret && p) | |
1409 | memcpy(ret, p, ks); | |
1410 | ||
1411 | return ret; | |
1412 | } | |
1413 | ||
1414 | /** | |
1415 | * __krealloc - like krealloc() but don't free @p. | |
1416 | * @p: object to reallocate memory for. | |
1417 | * @new_size: how many bytes of memory are required. | |
1418 | * @flags: the type of memory to allocate. | |
1419 | * | |
1420 | * This function is like krealloc() except it never frees the originally | |
1421 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1422 | * like, for example, with RCU. | |
1423 | */ | |
1424 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1425 | { | |
1426 | if (unlikely(!new_size)) | |
1427 | return ZERO_SIZE_PTR; | |
1428 | ||
1429 | return __do_krealloc(p, new_size, flags); | |
1430 | ||
1431 | } | |
1432 | EXPORT_SYMBOL(__krealloc); | |
1433 | ||
1434 | /** | |
1435 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1436 | * @p: object to reallocate memory for. | |
1437 | * @new_size: how many bytes of memory are required. | |
1438 | * @flags: the type of memory to allocate. | |
1439 | * | |
1440 | * The contents of the object pointed to are preserved up to the | |
1441 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1442 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1443 | * %NULL pointer, the object pointed to is freed. | |
1444 | */ | |
1445 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1446 | { | |
1447 | void *ret; | |
1448 | ||
1449 | if (unlikely(!new_size)) { | |
1450 | kfree(p); | |
1451 | return ZERO_SIZE_PTR; | |
1452 | } | |
1453 | ||
1454 | ret = __do_krealloc(p, new_size, flags); | |
1455 | if (ret && p != ret) | |
1456 | kfree(p); | |
1457 | ||
1458 | return ret; | |
1459 | } | |
1460 | EXPORT_SYMBOL(krealloc); | |
1461 | ||
1462 | /** | |
1463 | * kzfree - like kfree but zero memory | |
1464 | * @p: object to free memory of | |
1465 | * | |
1466 | * The memory of the object @p points to is zeroed before freed. | |
1467 | * If @p is %NULL, kzfree() does nothing. | |
1468 | * | |
1469 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1470 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1471 | * careful when using this function in performance sensitive code. | |
1472 | */ | |
1473 | void kzfree(const void *p) | |
1474 | { | |
1475 | size_t ks; | |
1476 | void *mem = (void *)p; | |
1477 | ||
1478 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1479 | return; | |
1480 | ks = ksize(mem); | |
1481 | memset(mem, 0, ks); | |
1482 | kfree(mem); | |
1483 | } | |
1484 | EXPORT_SYMBOL(kzfree); | |
1485 | ||
1486 | /* Tracepoints definitions. */ | |
1487 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1488 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1489 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1490 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1491 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1492 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |