]>
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 | |
657dc2f9 TH |
33 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
34 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
35 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
36 | slab_caches_to_rcu_destroy_workfn); | |
37 | ||
423c929c JK |
38 | /* |
39 | * Set of flags that will prevent slab merging | |
40 | */ | |
41 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
42 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ | |
7ed2f9e6 | 43 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 44 | |
230e9fc2 VD |
45 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
46 | SLAB_NOTRACK | SLAB_ACCOUNT) | |
423c929c JK |
47 | |
48 | /* | |
49 | * Merge control. If this is set then no merging of slab caches will occur. | |
50 | * (Could be removed. This was introduced to pacify the merge skeptics.) | |
51 | */ | |
52 | static int slab_nomerge; | |
53 | ||
54 | static int __init setup_slab_nomerge(char *str) | |
55 | { | |
56 | slab_nomerge = 1; | |
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) |
f7ce3190 | 141 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 142 | { |
9eeadc8b | 143 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 144 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 145 | INIT_LIST_HEAD(&s->memcg_params.children); |
f7ce3190 VD |
146 | } |
147 | ||
148 | static int init_memcg_params(struct kmem_cache *s, | |
149 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
150 | { | |
151 | struct memcg_cache_array *arr; | |
33a690c4 | 152 | |
9eeadc8b | 153 | if (root_cache) { |
f7ce3190 | 154 | s->memcg_params.root_cache = root_cache; |
9eeadc8b TH |
155 | s->memcg_params.memcg = memcg; |
156 | INIT_LIST_HEAD(&s->memcg_params.children_node); | |
bc2791f8 | 157 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 158 | return 0; |
f7ce3190 | 159 | } |
33a690c4 | 160 | |
f7ce3190 | 161 | slab_init_memcg_params(s); |
33a690c4 | 162 | |
f7ce3190 VD |
163 | if (!memcg_nr_cache_ids) |
164 | return 0; | |
33a690c4 | 165 | |
f7ce3190 VD |
166 | arr = kzalloc(sizeof(struct memcg_cache_array) + |
167 | memcg_nr_cache_ids * sizeof(void *), | |
168 | GFP_KERNEL); | |
169 | if (!arr) | |
170 | return -ENOMEM; | |
33a690c4 | 171 | |
f7ce3190 | 172 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
173 | return 0; |
174 | } | |
175 | ||
f7ce3190 | 176 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 177 | { |
f7ce3190 VD |
178 | if (is_root_cache(s)) |
179 | kfree(rcu_access_pointer(s->memcg_params.memcg_caches)); | |
33a690c4 VD |
180 | } |
181 | ||
f7ce3190 | 182 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 183 | { |
f7ce3190 | 184 | struct memcg_cache_array *old, *new; |
6f817f4c | 185 | |
f7ce3190 VD |
186 | if (!is_root_cache(s)) |
187 | return 0; | |
6f817f4c | 188 | |
f7ce3190 VD |
189 | new = kzalloc(sizeof(struct memcg_cache_array) + |
190 | new_array_size * sizeof(void *), GFP_KERNEL); | |
191 | if (!new) | |
6f817f4c VD |
192 | return -ENOMEM; |
193 | ||
f7ce3190 VD |
194 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
195 | lockdep_is_held(&slab_mutex)); | |
196 | if (old) | |
197 | memcpy(new->entries, old->entries, | |
198 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 199 | |
f7ce3190 VD |
200 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
201 | if (old) | |
202 | kfree_rcu(old, rcu); | |
6f817f4c VD |
203 | return 0; |
204 | } | |
205 | ||
55007d84 GC |
206 | int memcg_update_all_caches(int num_memcgs) |
207 | { | |
208 | struct kmem_cache *s; | |
209 | int ret = 0; | |
55007d84 | 210 | |
05257a1a | 211 | mutex_lock(&slab_mutex); |
55007d84 | 212 | list_for_each_entry(s, &slab_caches, list) { |
f7ce3190 | 213 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 214 | /* |
55007d84 GC |
215 | * Instead of freeing the memory, we'll just leave the caches |
216 | * up to this point in an updated state. | |
217 | */ | |
218 | if (ret) | |
05257a1a | 219 | break; |
55007d84 | 220 | } |
55007d84 GC |
221 | mutex_unlock(&slab_mutex); |
222 | return ret; | |
223 | } | |
657dc2f9 TH |
224 | |
225 | static void unlink_memcg_cache(struct kmem_cache *s) | |
226 | { | |
9eeadc8b | 227 | list_del(&s->memcg_params.children_node); |
bc2791f8 | 228 | list_del(&s->memcg_params.kmem_caches_node); |
657dc2f9 | 229 | } |
33a690c4 | 230 | #else |
f7ce3190 VD |
231 | static inline int init_memcg_params(struct kmem_cache *s, |
232 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
33a690c4 VD |
233 | { |
234 | return 0; | |
235 | } | |
236 | ||
f7ce3190 | 237 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
238 | { |
239 | } | |
657dc2f9 TH |
240 | |
241 | static inline void unlink_memcg_cache(struct kmem_cache *s) | |
242 | { | |
243 | } | |
127424c8 | 244 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
55007d84 | 245 | |
423c929c JK |
246 | /* |
247 | * Find a mergeable slab cache | |
248 | */ | |
249 | int slab_unmergeable(struct kmem_cache *s) | |
250 | { | |
251 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
252 | return 1; | |
253 | ||
254 | if (!is_root_cache(s)) | |
255 | return 1; | |
256 | ||
257 | if (s->ctor) | |
258 | return 1; | |
259 | ||
260 | /* | |
261 | * We may have set a slab to be unmergeable during bootstrap. | |
262 | */ | |
263 | if (s->refcount < 0) | |
264 | return 1; | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
270 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
271 | { | |
272 | struct kmem_cache *s; | |
273 | ||
c6e28895 | 274 | if (slab_nomerge) |
423c929c JK |
275 | return NULL; |
276 | ||
277 | if (ctor) | |
278 | return NULL; | |
279 | ||
280 | size = ALIGN(size, sizeof(void *)); | |
281 | align = calculate_alignment(flags, align, size); | |
282 | size = ALIGN(size, align); | |
283 | flags = kmem_cache_flags(size, flags, name, NULL); | |
284 | ||
c6e28895 GM |
285 | if (flags & SLAB_NEVER_MERGE) |
286 | return NULL; | |
287 | ||
54362057 | 288 | list_for_each_entry_reverse(s, &slab_caches, list) { |
423c929c JK |
289 | if (slab_unmergeable(s)) |
290 | continue; | |
291 | ||
292 | if (size > s->size) | |
293 | continue; | |
294 | ||
295 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
296 | continue; | |
297 | /* | |
298 | * Check if alignment is compatible. | |
299 | * Courtesy of Adrian Drzewiecki | |
300 | */ | |
301 | if ((s->size & ~(align - 1)) != s->size) | |
302 | continue; | |
303 | ||
304 | if (s->size - size >= sizeof(void *)) | |
305 | continue; | |
306 | ||
95069ac8 JK |
307 | if (IS_ENABLED(CONFIG_SLAB) && align && |
308 | (align > s->align || s->align % align)) | |
309 | continue; | |
310 | ||
423c929c JK |
311 | return s; |
312 | } | |
313 | return NULL; | |
314 | } | |
315 | ||
45906855 CL |
316 | /* |
317 | * Figure out what the alignment of the objects will be given a set of | |
318 | * flags, a user specified alignment and the size of the objects. | |
319 | */ | |
320 | unsigned long calculate_alignment(unsigned long flags, | |
321 | unsigned long align, unsigned long size) | |
322 | { | |
323 | /* | |
324 | * If the user wants hardware cache aligned objects then follow that | |
325 | * suggestion if the object is sufficiently large. | |
326 | * | |
327 | * The hardware cache alignment cannot override the specified | |
328 | * alignment though. If that is greater then use it. | |
329 | */ | |
330 | if (flags & SLAB_HWCACHE_ALIGN) { | |
331 | unsigned long ralign = cache_line_size(); | |
332 | while (size <= ralign / 2) | |
333 | ralign /= 2; | |
334 | align = max(align, ralign); | |
335 | } | |
336 | ||
337 | if (align < ARCH_SLAB_MINALIGN) | |
338 | align = ARCH_SLAB_MINALIGN; | |
339 | ||
340 | return ALIGN(align, sizeof(void *)); | |
341 | } | |
342 | ||
c9a77a79 VD |
343 | static struct kmem_cache *create_cache(const char *name, |
344 | size_t object_size, size_t size, size_t align, | |
345 | unsigned long flags, void (*ctor)(void *), | |
346 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
794b1248 VD |
347 | { |
348 | struct kmem_cache *s; | |
349 | int err; | |
350 | ||
351 | err = -ENOMEM; | |
352 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
353 | if (!s) | |
354 | goto out; | |
355 | ||
356 | s->name = name; | |
357 | s->object_size = object_size; | |
358 | s->size = size; | |
359 | s->align = align; | |
360 | s->ctor = ctor; | |
361 | ||
f7ce3190 | 362 | err = init_memcg_params(s, memcg, root_cache); |
794b1248 VD |
363 | if (err) |
364 | goto out_free_cache; | |
365 | ||
366 | err = __kmem_cache_create(s, flags); | |
367 | if (err) | |
368 | goto out_free_cache; | |
369 | ||
370 | s->refcount = 1; | |
371 | list_add(&s->list, &slab_caches); | |
794b1248 VD |
372 | out: |
373 | if (err) | |
374 | return ERR_PTR(err); | |
375 | return s; | |
376 | ||
377 | out_free_cache: | |
f7ce3190 | 378 | destroy_memcg_params(s); |
7c4da061 | 379 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
380 | goto out; |
381 | } | |
45906855 | 382 | |
77be4b13 SK |
383 | /* |
384 | * kmem_cache_create - Create a cache. | |
385 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
386 | * @size: The size of objects to be created in this cache. | |
387 | * @align: The required alignment for the objects. | |
388 | * @flags: SLAB flags | |
389 | * @ctor: A constructor for the objects. | |
390 | * | |
391 | * Returns a ptr to the cache on success, NULL on failure. | |
392 | * Cannot be called within a interrupt, but can be interrupted. | |
393 | * The @ctor is run when new pages are allocated by the cache. | |
394 | * | |
395 | * The flags are | |
396 | * | |
397 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
398 | * to catch references to uninitialised memory. | |
399 | * | |
400 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
401 | * for buffer overruns. | |
402 | * | |
403 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
404 | * cacheline. This can be beneficial if you're counting cycles as closely | |
405 | * as davem. | |
406 | */ | |
2633d7a0 | 407 | struct kmem_cache * |
794b1248 VD |
408 | kmem_cache_create(const char *name, size_t size, size_t align, |
409 | unsigned long flags, void (*ctor)(void *)) | |
77be4b13 | 410 | { |
40911a79 | 411 | struct kmem_cache *s = NULL; |
3dec16ea | 412 | const char *cache_name; |
3965fc36 | 413 | int err; |
039363f3 | 414 | |
77be4b13 | 415 | get_online_cpus(); |
03afc0e2 | 416 | get_online_mems(); |
05257a1a | 417 | memcg_get_cache_ids(); |
03afc0e2 | 418 | |
77be4b13 | 419 | mutex_lock(&slab_mutex); |
686d550d | 420 | |
794b1248 | 421 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 422 | if (err) { |
3965fc36 | 423 | goto out_unlock; |
3aa24f51 | 424 | } |
686d550d | 425 | |
e70954fd TG |
426 | /* Refuse requests with allocator specific flags */ |
427 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
428 | err = -EINVAL; | |
429 | goto out_unlock; | |
430 | } | |
431 | ||
d8843922 GC |
432 | /* |
433 | * Some allocators will constraint the set of valid flags to a subset | |
434 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
435 | * case, and we'll just provide them with a sanitized version of the | |
436 | * passed flags. | |
437 | */ | |
438 | flags &= CACHE_CREATE_MASK; | |
686d550d | 439 | |
794b1248 VD |
440 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
441 | if (s) | |
3965fc36 | 442 | goto out_unlock; |
2633d7a0 | 443 | |
3dec16ea | 444 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
445 | if (!cache_name) { |
446 | err = -ENOMEM; | |
447 | goto out_unlock; | |
448 | } | |
7c9adf5a | 449 | |
c9a77a79 VD |
450 | s = create_cache(cache_name, size, size, |
451 | calculate_alignment(flags, align, size), | |
452 | flags, ctor, NULL, NULL); | |
794b1248 VD |
453 | if (IS_ERR(s)) { |
454 | err = PTR_ERR(s); | |
3dec16ea | 455 | kfree_const(cache_name); |
794b1248 | 456 | } |
3965fc36 VD |
457 | |
458 | out_unlock: | |
20cea968 | 459 | mutex_unlock(&slab_mutex); |
03afc0e2 | 460 | |
05257a1a | 461 | memcg_put_cache_ids(); |
03afc0e2 | 462 | put_online_mems(); |
20cea968 CL |
463 | put_online_cpus(); |
464 | ||
ba3253c7 | 465 | if (err) { |
686d550d CL |
466 | if (flags & SLAB_PANIC) |
467 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
468 | name, err); | |
469 | else { | |
1170532b | 470 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
471 | name, err); |
472 | dump_stack(); | |
473 | } | |
686d550d CL |
474 | return NULL; |
475 | } | |
039363f3 CL |
476 | return s; |
477 | } | |
794b1248 | 478 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 479 | |
657dc2f9 | 480 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 481 | { |
657dc2f9 TH |
482 | LIST_HEAD(to_destroy); |
483 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 484 | |
657dc2f9 TH |
485 | /* |
486 | * On destruction, SLAB_DESTROY_BY_RCU kmem_caches are put on the | |
487 | * @slab_caches_to_rcu_destroy list. The slab pages are freed | |
488 | * through RCU and and the associated kmem_cache are dereferenced | |
489 | * while freeing the pages, so the kmem_caches should be freed only | |
490 | * after the pending RCU operations are finished. As rcu_barrier() | |
491 | * is a pretty slow operation, we batch all pending destructions | |
492 | * asynchronously. | |
493 | */ | |
494 | mutex_lock(&slab_mutex); | |
495 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
496 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 497 | |
657dc2f9 TH |
498 | if (list_empty(&to_destroy)) |
499 | return; | |
500 | ||
501 | rcu_barrier(); | |
502 | ||
503 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
504 | #ifdef SLAB_SUPPORTS_SYSFS | |
505 | sysfs_slab_release(s); | |
506 | #else | |
507 | slab_kmem_cache_release(s); | |
508 | #endif | |
509 | } | |
d5b3cf71 VD |
510 | } |
511 | ||
657dc2f9 | 512 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 513 | { |
657dc2f9 TH |
514 | if (__kmem_cache_shutdown(s) != 0) |
515 | return -EBUSY; | |
d5b3cf71 | 516 | |
657dc2f9 TH |
517 | list_del(&s->list); |
518 | if (!is_root_cache(s)) | |
519 | unlink_memcg_cache(s); | |
d5b3cf71 | 520 | |
657dc2f9 TH |
521 | if (s->flags & SLAB_DESTROY_BY_RCU) { |
522 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); | |
523 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
524 | } else { | |
d5b3cf71 | 525 | #ifdef SLAB_SUPPORTS_SYSFS |
bf5eb3de | 526 | sysfs_slab_release(s); |
d5b3cf71 VD |
527 | #else |
528 | slab_kmem_cache_release(s); | |
529 | #endif | |
530 | } | |
657dc2f9 TH |
531 | |
532 | return 0; | |
d5b3cf71 VD |
533 | } |
534 | ||
127424c8 | 535 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
794b1248 | 536 | /* |
776ed0f0 | 537 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
538 | * @memcg: The memory cgroup the new cache is for. |
539 | * @root_cache: The parent of the new cache. | |
540 | * | |
541 | * This function attempts to create a kmem cache that will serve allocation | |
542 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
543 | * from its parent. | |
544 | */ | |
d5b3cf71 VD |
545 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
546 | struct kmem_cache *root_cache) | |
2633d7a0 | 547 | { |
3e0350a3 | 548 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 549 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 550 | struct memcg_cache_array *arr; |
bd673145 | 551 | struct kmem_cache *s = NULL; |
794b1248 | 552 | char *cache_name; |
f7ce3190 | 553 | int idx; |
794b1248 VD |
554 | |
555 | get_online_cpus(); | |
03afc0e2 VD |
556 | get_online_mems(); |
557 | ||
794b1248 VD |
558 | mutex_lock(&slab_mutex); |
559 | ||
2a4db7eb | 560 | /* |
567e9ab2 | 561 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
562 | * creation work was pending. |
563 | */ | |
b6ecd2de | 564 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
565 | goto out_unlock; |
566 | ||
f7ce3190 VD |
567 | idx = memcg_cache_id(memcg); |
568 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
569 | lockdep_is_held(&slab_mutex)); | |
570 | ||
d5b3cf71 VD |
571 | /* |
572 | * Since per-memcg caches are created asynchronously on first | |
573 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
574 | * create the same cache, but only one of them may succeed. | |
575 | */ | |
f7ce3190 | 576 | if (arr->entries[idx]) |
d5b3cf71 VD |
577 | goto out_unlock; |
578 | ||
f1008365 | 579 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
580 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
581 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
582 | if (!cache_name) |
583 | goto out_unlock; | |
584 | ||
c9a77a79 VD |
585 | s = create_cache(cache_name, root_cache->object_size, |
586 | root_cache->size, root_cache->align, | |
f773e36d GT |
587 | root_cache->flags & CACHE_CREATE_MASK, |
588 | root_cache->ctor, memcg, root_cache); | |
d5b3cf71 VD |
589 | /* |
590 | * If we could not create a memcg cache, do not complain, because | |
591 | * that's not critical at all as we can always proceed with the root | |
592 | * cache. | |
593 | */ | |
bd673145 | 594 | if (IS_ERR(s)) { |
794b1248 | 595 | kfree(cache_name); |
d5b3cf71 | 596 | goto out_unlock; |
bd673145 | 597 | } |
794b1248 | 598 | |
9eeadc8b TH |
599 | list_add(&s->memcg_params.children_node, |
600 | &root_cache->memcg_params.children); | |
bc2791f8 | 601 | list_add(&s->memcg_params.kmem_caches_node, &memcg->kmem_caches); |
426589f5 | 602 | |
d5b3cf71 VD |
603 | /* |
604 | * Since readers won't lock (see cache_from_memcg_idx()), we need a | |
605 | * barrier here to ensure nobody will see the kmem_cache partially | |
606 | * initialized. | |
607 | */ | |
608 | smp_wmb(); | |
f7ce3190 | 609 | arr->entries[idx] = s; |
d5b3cf71 | 610 | |
794b1248 VD |
611 | out_unlock: |
612 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
613 | |
614 | put_online_mems(); | |
794b1248 | 615 | put_online_cpus(); |
2633d7a0 | 616 | } |
b8529907 | 617 | |
2a4db7eb VD |
618 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
619 | { | |
620 | int idx; | |
621 | struct memcg_cache_array *arr; | |
d6e0b7fa | 622 | struct kmem_cache *s, *c; |
2a4db7eb VD |
623 | |
624 | idx = memcg_cache_id(memcg); | |
625 | ||
d6e0b7fa VD |
626 | get_online_cpus(); |
627 | get_online_mems(); | |
628 | ||
2a4db7eb VD |
629 | mutex_lock(&slab_mutex); |
630 | list_for_each_entry(s, &slab_caches, list) { | |
631 | if (!is_root_cache(s)) | |
632 | continue; | |
633 | ||
634 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
635 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
636 | c = arr->entries[idx]; |
637 | if (!c) | |
638 | continue; | |
639 | ||
290b6a58 | 640 | __kmem_cache_shrink(c, true); |
2a4db7eb VD |
641 | arr->entries[idx] = NULL; |
642 | } | |
643 | mutex_unlock(&slab_mutex); | |
d6e0b7fa VD |
644 | |
645 | put_online_mems(); | |
646 | put_online_cpus(); | |
2a4db7eb VD |
647 | } |
648 | ||
d5b3cf71 | 649 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
b8529907 | 650 | { |
d5b3cf71 | 651 | struct kmem_cache *s, *s2; |
b8529907 | 652 | |
d5b3cf71 VD |
653 | get_online_cpus(); |
654 | get_online_mems(); | |
b8529907 | 655 | |
b8529907 | 656 | mutex_lock(&slab_mutex); |
bc2791f8 TH |
657 | list_for_each_entry_safe(s, s2, &memcg->kmem_caches, |
658 | memcg_params.kmem_caches_node) { | |
d5b3cf71 VD |
659 | /* |
660 | * The cgroup is about to be freed and therefore has no charges | |
661 | * left. Hence, all its caches must be empty by now. | |
662 | */ | |
657dc2f9 | 663 | BUG_ON(shutdown_cache(s)); |
d5b3cf71 VD |
664 | } |
665 | mutex_unlock(&slab_mutex); | |
b8529907 | 666 | |
d5b3cf71 VD |
667 | put_online_mems(); |
668 | put_online_cpus(); | |
b8529907 | 669 | } |
d60fdcc9 | 670 | |
657dc2f9 | 671 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
672 | { |
673 | struct memcg_cache_array *arr; | |
674 | struct kmem_cache *c, *c2; | |
675 | LIST_HEAD(busy); | |
676 | int i; | |
677 | ||
678 | BUG_ON(!is_root_cache(s)); | |
679 | ||
680 | /* | |
681 | * First, shutdown active caches, i.e. caches that belong to online | |
682 | * memory cgroups. | |
683 | */ | |
684 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
685 | lockdep_is_held(&slab_mutex)); | |
686 | for_each_memcg_cache_index(i) { | |
687 | c = arr->entries[i]; | |
688 | if (!c) | |
689 | continue; | |
657dc2f9 | 690 | if (shutdown_cache(c)) |
d60fdcc9 VD |
691 | /* |
692 | * The cache still has objects. Move it to a temporary | |
693 | * list so as not to try to destroy it for a second | |
694 | * time while iterating over inactive caches below. | |
695 | */ | |
9eeadc8b | 696 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
697 | else |
698 | /* | |
699 | * The cache is empty and will be destroyed soon. Clear | |
700 | * the pointer to it in the memcg_caches array so that | |
701 | * it will never be accessed even if the root cache | |
702 | * stays alive. | |
703 | */ | |
704 | arr->entries[i] = NULL; | |
705 | } | |
706 | ||
707 | /* | |
708 | * Second, shutdown all caches left from memory cgroups that are now | |
709 | * offline. | |
710 | */ | |
9eeadc8b TH |
711 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
712 | memcg_params.children_node) | |
657dc2f9 | 713 | shutdown_cache(c); |
d60fdcc9 | 714 | |
9eeadc8b | 715 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
716 | |
717 | /* | |
718 | * A cache being destroyed must be empty. In particular, this means | |
719 | * that all per memcg caches attached to it must be empty too. | |
720 | */ | |
9eeadc8b | 721 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
722 | return -EBUSY; |
723 | return 0; | |
724 | } | |
725 | #else | |
657dc2f9 | 726 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
727 | { |
728 | return 0; | |
729 | } | |
127424c8 | 730 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
97d06609 | 731 | |
41a21285 CL |
732 | void slab_kmem_cache_release(struct kmem_cache *s) |
733 | { | |
52b4b950 | 734 | __kmem_cache_release(s); |
f7ce3190 | 735 | destroy_memcg_params(s); |
3dec16ea | 736 | kfree_const(s->name); |
41a21285 CL |
737 | kmem_cache_free(kmem_cache, s); |
738 | } | |
739 | ||
945cf2b6 CL |
740 | void kmem_cache_destroy(struct kmem_cache *s) |
741 | { | |
d60fdcc9 | 742 | int err; |
d5b3cf71 | 743 | |
3942d299 SS |
744 | if (unlikely(!s)) |
745 | return; | |
746 | ||
945cf2b6 | 747 | get_online_cpus(); |
03afc0e2 VD |
748 | get_online_mems(); |
749 | ||
55834c59 | 750 | kasan_cache_destroy(s); |
945cf2b6 | 751 | mutex_lock(&slab_mutex); |
b8529907 | 752 | |
945cf2b6 | 753 | s->refcount--; |
b8529907 VD |
754 | if (s->refcount) |
755 | goto out_unlock; | |
756 | ||
657dc2f9 | 757 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 758 | if (!err) |
657dc2f9 | 759 | err = shutdown_cache(s); |
b8529907 | 760 | |
cd918c55 | 761 | if (err) { |
756a025f JP |
762 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
763 | s->name); | |
cd918c55 VD |
764 | dump_stack(); |
765 | } | |
b8529907 VD |
766 | out_unlock: |
767 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 768 | |
03afc0e2 | 769 | put_online_mems(); |
945cf2b6 CL |
770 | put_online_cpus(); |
771 | } | |
772 | EXPORT_SYMBOL(kmem_cache_destroy); | |
773 | ||
03afc0e2 VD |
774 | /** |
775 | * kmem_cache_shrink - Shrink a cache. | |
776 | * @cachep: The cache to shrink. | |
777 | * | |
778 | * Releases as many slabs as possible for a cache. | |
779 | * To help debugging, a zero exit status indicates all slabs were released. | |
780 | */ | |
781 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
782 | { | |
783 | int ret; | |
784 | ||
785 | get_online_cpus(); | |
786 | get_online_mems(); | |
55834c59 | 787 | kasan_cache_shrink(cachep); |
290b6a58 | 788 | ret = __kmem_cache_shrink(cachep, false); |
03afc0e2 VD |
789 | put_online_mems(); |
790 | put_online_cpus(); | |
791 | return ret; | |
792 | } | |
793 | EXPORT_SYMBOL(kmem_cache_shrink); | |
794 | ||
fda90124 | 795 | bool slab_is_available(void) |
97d06609 CL |
796 | { |
797 | return slab_state >= UP; | |
798 | } | |
b7454ad3 | 799 | |
45530c44 CL |
800 | #ifndef CONFIG_SLOB |
801 | /* Create a cache during boot when no slab services are available yet */ | |
802 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
803 | unsigned long flags) | |
804 | { | |
805 | int err; | |
806 | ||
807 | s->name = name; | |
808 | s->size = s->object_size = size; | |
45906855 | 809 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
f7ce3190 VD |
810 | |
811 | slab_init_memcg_params(s); | |
812 | ||
45530c44 CL |
813 | err = __kmem_cache_create(s, flags); |
814 | ||
815 | if (err) | |
31ba7346 | 816 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
817 | name, size, err); |
818 | ||
819 | s->refcount = -1; /* Exempt from merging for now */ | |
820 | } | |
821 | ||
822 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
823 | unsigned long flags) | |
824 | { | |
825 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
826 | ||
827 | if (!s) | |
828 | panic("Out of memory when creating slab %s\n", name); | |
829 | ||
830 | create_boot_cache(s, name, size, flags); | |
831 | list_add(&s->list, &slab_caches); | |
832 | s->refcount = 1; | |
833 | return s; | |
834 | } | |
835 | ||
9425c58e CL |
836 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
837 | EXPORT_SYMBOL(kmalloc_caches); | |
838 | ||
839 | #ifdef CONFIG_ZONE_DMA | |
840 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
841 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
842 | #endif | |
843 | ||
2c59dd65 CL |
844 | /* |
845 | * Conversion table for small slabs sizes / 8 to the index in the | |
846 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
847 | * of two cache sizes there. The size of larger slabs can be determined using | |
848 | * fls. | |
849 | */ | |
850 | static s8 size_index[24] = { | |
851 | 3, /* 8 */ | |
852 | 4, /* 16 */ | |
853 | 5, /* 24 */ | |
854 | 5, /* 32 */ | |
855 | 6, /* 40 */ | |
856 | 6, /* 48 */ | |
857 | 6, /* 56 */ | |
858 | 6, /* 64 */ | |
859 | 1, /* 72 */ | |
860 | 1, /* 80 */ | |
861 | 1, /* 88 */ | |
862 | 1, /* 96 */ | |
863 | 7, /* 104 */ | |
864 | 7, /* 112 */ | |
865 | 7, /* 120 */ | |
866 | 7, /* 128 */ | |
867 | 2, /* 136 */ | |
868 | 2, /* 144 */ | |
869 | 2, /* 152 */ | |
870 | 2, /* 160 */ | |
871 | 2, /* 168 */ | |
872 | 2, /* 176 */ | |
873 | 2, /* 184 */ | |
874 | 2 /* 192 */ | |
875 | }; | |
876 | ||
877 | static inline int size_index_elem(size_t bytes) | |
878 | { | |
879 | return (bytes - 1) / 8; | |
880 | } | |
881 | ||
882 | /* | |
883 | * Find the kmem_cache structure that serves a given size of | |
884 | * allocation | |
885 | */ | |
886 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
887 | { | |
888 | int index; | |
889 | ||
9de1bc87 | 890 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 891 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 892 | return NULL; |
907985f4 | 893 | } |
6286ae97 | 894 | |
2c59dd65 CL |
895 | if (size <= 192) { |
896 | if (!size) | |
897 | return ZERO_SIZE_PTR; | |
898 | ||
899 | index = size_index[size_index_elem(size)]; | |
900 | } else | |
901 | index = fls(size - 1); | |
902 | ||
903 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 904 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
905 | return kmalloc_dma_caches[index]; |
906 | ||
907 | #endif | |
908 | return kmalloc_caches[index]; | |
909 | } | |
910 | ||
4066c33d GG |
911 | /* |
912 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
913 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
914 | * kmalloc-67108864. | |
915 | */ | |
af3b5f87 | 916 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
4066c33d GG |
917 | {NULL, 0}, {"kmalloc-96", 96}, |
918 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
919 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
920 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
921 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
922 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, | |
923 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, | |
924 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, | |
925 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, | |
926 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, | |
927 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, | |
928 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, | |
929 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, | |
930 | {"kmalloc-67108864", 67108864} | |
931 | }; | |
932 | ||
f97d5f63 | 933 | /* |
34cc6990 DS |
934 | * Patch up the size_index table if we have strange large alignment |
935 | * requirements for the kmalloc array. This is only the case for | |
936 | * MIPS it seems. The standard arches will not generate any code here. | |
937 | * | |
938 | * Largest permitted alignment is 256 bytes due to the way we | |
939 | * handle the index determination for the smaller caches. | |
940 | * | |
941 | * Make sure that nothing crazy happens if someone starts tinkering | |
942 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 943 | */ |
34cc6990 | 944 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 CL |
945 | { |
946 | int i; | |
947 | ||
2c59dd65 CL |
948 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
949 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
950 | ||
951 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
952 | int elem = size_index_elem(i); | |
953 | ||
954 | if (elem >= ARRAY_SIZE(size_index)) | |
955 | break; | |
956 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
957 | } | |
958 | ||
959 | if (KMALLOC_MIN_SIZE >= 64) { | |
960 | /* | |
961 | * The 96 byte size cache is not used if the alignment | |
962 | * is 64 byte. | |
963 | */ | |
964 | for (i = 64 + 8; i <= 96; i += 8) | |
965 | size_index[size_index_elem(i)] = 7; | |
966 | ||
967 | } | |
968 | ||
969 | if (KMALLOC_MIN_SIZE >= 128) { | |
970 | /* | |
971 | * The 192 byte sized cache is not used if the alignment | |
972 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
973 | * instead. | |
974 | */ | |
975 | for (i = 128 + 8; i <= 192; i += 8) | |
976 | size_index[size_index_elem(i)] = 8; | |
977 | } | |
34cc6990 DS |
978 | } |
979 | ||
ae6f2462 | 980 | static void __init new_kmalloc_cache(int idx, unsigned long flags) |
a9730fca CL |
981 | { |
982 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, | |
983 | kmalloc_info[idx].size, flags); | |
984 | } | |
985 | ||
34cc6990 DS |
986 | /* |
987 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
988 | * may already have been created because they were needed to | |
989 | * enable allocations for slab creation. | |
990 | */ | |
991 | void __init create_kmalloc_caches(unsigned long flags) | |
992 | { | |
993 | int i; | |
994 | ||
a9730fca CL |
995 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
996 | if (!kmalloc_caches[i]) | |
997 | new_kmalloc_cache(i, flags); | |
f97d5f63 | 998 | |
956e46ef | 999 | /* |
a9730fca CL |
1000 | * Caches that are not of the two-to-the-power-of size. |
1001 | * These have to be created immediately after the | |
1002 | * earlier power of two caches | |
956e46ef | 1003 | */ |
a9730fca CL |
1004 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
1005 | new_kmalloc_cache(1, flags); | |
1006 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) | |
1007 | new_kmalloc_cache(2, flags); | |
8a965b3b CL |
1008 | } |
1009 | ||
f97d5f63 CL |
1010 | /* Kmalloc array is now usable */ |
1011 | slab_state = UP; | |
1012 | ||
f97d5f63 CL |
1013 | #ifdef CONFIG_ZONE_DMA |
1014 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1015 | struct kmem_cache *s = kmalloc_caches[i]; | |
1016 | ||
1017 | if (s) { | |
1018 | int size = kmalloc_size(i); | |
1019 | char *n = kasprintf(GFP_NOWAIT, | |
1020 | "dma-kmalloc-%d", size); | |
1021 | ||
1022 | BUG_ON(!n); | |
1023 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
1024 | size, SLAB_CACHE_DMA | flags); | |
1025 | } | |
1026 | } | |
1027 | #endif | |
1028 | } | |
45530c44 CL |
1029 | #endif /* !CONFIG_SLOB */ |
1030 | ||
cea371f4 VD |
1031 | /* |
1032 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1033 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1034 | * know the allocation order to free the pages properly in kfree. | |
1035 | */ | |
52383431 VD |
1036 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1037 | { | |
1038 | void *ret; | |
1039 | struct page *page; | |
1040 | ||
1041 | flags |= __GFP_COMP; | |
4949148a | 1042 | page = alloc_pages(flags, order); |
52383431 VD |
1043 | ret = page ? page_address(page) : NULL; |
1044 | kmemleak_alloc(ret, size, 1, flags); | |
505f5dcb | 1045 | kasan_kmalloc_large(ret, size, flags); |
52383431 VD |
1046 | return ret; |
1047 | } | |
1048 | EXPORT_SYMBOL(kmalloc_order); | |
1049 | ||
f1b6eb6e CL |
1050 | #ifdef CONFIG_TRACING |
1051 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1052 | { | |
1053 | void *ret = kmalloc_order(size, flags, order); | |
1054 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1055 | return ret; | |
1056 | } | |
1057 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1058 | #endif | |
45530c44 | 1059 | |
7c00fce9 TG |
1060 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1061 | /* Randomize a generic freelist */ | |
1062 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
1063 | size_t count) | |
1064 | { | |
1065 | size_t i; | |
1066 | unsigned int rand; | |
1067 | ||
1068 | for (i = 0; i < count; i++) | |
1069 | list[i] = i; | |
1070 | ||
1071 | /* Fisher-Yates shuffle */ | |
1072 | for (i = count - 1; i > 0; i--) { | |
1073 | rand = prandom_u32_state(state); | |
1074 | rand %= (i + 1); | |
1075 | swap(list[i], list[rand]); | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | /* Create a random sequence per cache */ | |
1080 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1081 | gfp_t gfp) | |
1082 | { | |
1083 | struct rnd_state state; | |
1084 | ||
1085 | if (count < 2 || cachep->random_seq) | |
1086 | return 0; | |
1087 | ||
1088 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1089 | if (!cachep->random_seq) | |
1090 | return -ENOMEM; | |
1091 | ||
1092 | /* Get best entropy at this stage of boot */ | |
1093 | prandom_seed_state(&state, get_random_long()); | |
1094 | ||
1095 | freelist_randomize(&state, cachep->random_seq, count); | |
1096 | return 0; | |
1097 | } | |
1098 | ||
1099 | /* Destroy the per-cache random freelist sequence */ | |
1100 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1101 | { | |
1102 | kfree(cachep->random_seq); | |
1103 | cachep->random_seq = NULL; | |
1104 | } | |
1105 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1106 | ||
b7454ad3 | 1107 | #ifdef CONFIG_SLABINFO |
e9b4db2b WL |
1108 | |
1109 | #ifdef CONFIG_SLAB | |
1110 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
1111 | #else | |
1112 | #define SLABINFO_RIGHTS S_IRUSR | |
1113 | #endif | |
1114 | ||
b047501c | 1115 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1116 | { |
1117 | /* | |
1118 | * Output format version, so at least we can change it | |
1119 | * without _too_ many complaints. | |
1120 | */ | |
1121 | #ifdef CONFIG_DEBUG_SLAB | |
1122 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1123 | #else | |
1124 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1125 | #endif | |
756a025f | 1126 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1127 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1128 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1129 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1130 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1131 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1132 | #endif | |
1133 | seq_putc(m, '\n'); | |
1134 | } | |
1135 | ||
1df3b26f | 1136 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1137 | { |
b7454ad3 | 1138 | mutex_lock(&slab_mutex); |
b7454ad3 GC |
1139 | return seq_list_start(&slab_caches, *pos); |
1140 | } | |
1141 | ||
276a2439 | 1142 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 GC |
1143 | { |
1144 | return seq_list_next(p, &slab_caches, pos); | |
1145 | } | |
1146 | ||
276a2439 | 1147 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1148 | { |
1149 | mutex_unlock(&slab_mutex); | |
1150 | } | |
1151 | ||
749c5415 GC |
1152 | static void |
1153 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1154 | { | |
1155 | struct kmem_cache *c; | |
1156 | struct slabinfo sinfo; | |
749c5415 GC |
1157 | |
1158 | if (!is_root_cache(s)) | |
1159 | return; | |
1160 | ||
426589f5 | 1161 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1162 | memset(&sinfo, 0, sizeof(sinfo)); |
1163 | get_slabinfo(c, &sinfo); | |
1164 | ||
1165 | info->active_slabs += sinfo.active_slabs; | |
1166 | info->num_slabs += sinfo.num_slabs; | |
1167 | info->shared_avail += sinfo.shared_avail; | |
1168 | info->active_objs += sinfo.active_objs; | |
1169 | info->num_objs += sinfo.num_objs; | |
1170 | } | |
1171 | } | |
1172 | ||
b047501c | 1173 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1174 | { |
0d7561c6 GC |
1175 | struct slabinfo sinfo; |
1176 | ||
1177 | memset(&sinfo, 0, sizeof(sinfo)); | |
1178 | get_slabinfo(s, &sinfo); | |
1179 | ||
749c5415 GC |
1180 | memcg_accumulate_slabinfo(s, &sinfo); |
1181 | ||
0d7561c6 | 1182 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1183 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1184 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1185 | ||
1186 | seq_printf(m, " : tunables %4u %4u %4u", | |
1187 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1188 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1189 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1190 | slabinfo_show_stats(m, s); | |
1191 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1192 | } |
1193 | ||
1df3b26f | 1194 | static int slab_show(struct seq_file *m, void *p) |
749c5415 GC |
1195 | { |
1196 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
1197 | ||
1df3b26f VD |
1198 | if (p == slab_caches.next) |
1199 | print_slabinfo_header(m); | |
b047501c VD |
1200 | if (is_root_cache(s)) |
1201 | cache_show(s, m); | |
1202 | return 0; | |
1203 | } | |
1204 | ||
127424c8 | 1205 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
bc2791f8 TH |
1206 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1207 | { | |
1208 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1209 | ||
1210 | mutex_lock(&slab_mutex); | |
1211 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1212 | } | |
1213 | ||
1214 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1215 | { | |
1216 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1217 | ||
1218 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1219 | } | |
1220 | ||
1221 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1222 | { | |
1223 | mutex_unlock(&slab_mutex); | |
1224 | } | |
1225 | ||
b047501c VD |
1226 | int memcg_slab_show(struct seq_file *m, void *p) |
1227 | { | |
bc2791f8 TH |
1228 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1229 | memcg_params.kmem_caches_node); | |
b047501c VD |
1230 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
1231 | ||
bc2791f8 | 1232 | if (p == memcg->kmem_caches.next) |
b047501c | 1233 | print_slabinfo_header(m); |
bc2791f8 | 1234 | cache_show(s, m); |
b047501c | 1235 | return 0; |
749c5415 | 1236 | } |
b047501c | 1237 | #endif |
749c5415 | 1238 | |
b7454ad3 GC |
1239 | /* |
1240 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1241 | * | |
1242 | * Output layout: | |
1243 | * cache-name | |
1244 | * num-active-objs | |
1245 | * total-objs | |
1246 | * object size | |
1247 | * num-active-slabs | |
1248 | * total-slabs | |
1249 | * num-pages-per-slab | |
1250 | * + further values on SMP and with statistics enabled | |
1251 | */ | |
1252 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1253 | .start = slab_start, |
276a2439 WL |
1254 | .next = slab_next, |
1255 | .stop = slab_stop, | |
1df3b26f | 1256 | .show = slab_show, |
b7454ad3 GC |
1257 | }; |
1258 | ||
1259 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1260 | { | |
1261 | return seq_open(file, &slabinfo_op); | |
1262 | } | |
1263 | ||
1264 | static const struct file_operations proc_slabinfo_operations = { | |
1265 | .open = slabinfo_open, | |
1266 | .read = seq_read, | |
1267 | .write = slabinfo_write, | |
1268 | .llseek = seq_lseek, | |
1269 | .release = seq_release, | |
1270 | }; | |
1271 | ||
1272 | static int __init slab_proc_init(void) | |
1273 | { | |
e9b4db2b WL |
1274 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
1275 | &proc_slabinfo_operations); | |
b7454ad3 GC |
1276 | return 0; |
1277 | } | |
1278 | module_init(slab_proc_init); | |
1279 | #endif /* CONFIG_SLABINFO */ | |
928cec9c AR |
1280 | |
1281 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1282 | gfp_t flags) | |
1283 | { | |
1284 | void *ret; | |
1285 | size_t ks = 0; | |
1286 | ||
1287 | if (p) | |
1288 | ks = ksize(p); | |
1289 | ||
0316bec2 | 1290 | if (ks >= new_size) { |
505f5dcb | 1291 | kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1292 | return (void *)p; |
0316bec2 | 1293 | } |
928cec9c AR |
1294 | |
1295 | ret = kmalloc_track_caller(new_size, flags); | |
1296 | if (ret && p) | |
1297 | memcpy(ret, p, ks); | |
1298 | ||
1299 | return ret; | |
1300 | } | |
1301 | ||
1302 | /** | |
1303 | * __krealloc - like krealloc() but don't free @p. | |
1304 | * @p: object to reallocate memory for. | |
1305 | * @new_size: how many bytes of memory are required. | |
1306 | * @flags: the type of memory to allocate. | |
1307 | * | |
1308 | * This function is like krealloc() except it never frees the originally | |
1309 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1310 | * like, for example, with RCU. | |
1311 | */ | |
1312 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1313 | { | |
1314 | if (unlikely(!new_size)) | |
1315 | return ZERO_SIZE_PTR; | |
1316 | ||
1317 | return __do_krealloc(p, new_size, flags); | |
1318 | ||
1319 | } | |
1320 | EXPORT_SYMBOL(__krealloc); | |
1321 | ||
1322 | /** | |
1323 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1324 | * @p: object to reallocate memory for. | |
1325 | * @new_size: how many bytes of memory are required. | |
1326 | * @flags: the type of memory to allocate. | |
1327 | * | |
1328 | * The contents of the object pointed to are preserved up to the | |
1329 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1330 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1331 | * %NULL pointer, the object pointed to is freed. | |
1332 | */ | |
1333 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1334 | { | |
1335 | void *ret; | |
1336 | ||
1337 | if (unlikely(!new_size)) { | |
1338 | kfree(p); | |
1339 | return ZERO_SIZE_PTR; | |
1340 | } | |
1341 | ||
1342 | ret = __do_krealloc(p, new_size, flags); | |
1343 | if (ret && p != ret) | |
1344 | kfree(p); | |
1345 | ||
1346 | return ret; | |
1347 | } | |
1348 | EXPORT_SYMBOL(krealloc); | |
1349 | ||
1350 | /** | |
1351 | * kzfree - like kfree but zero memory | |
1352 | * @p: object to free memory of | |
1353 | * | |
1354 | * The memory of the object @p points to is zeroed before freed. | |
1355 | * If @p is %NULL, kzfree() does nothing. | |
1356 | * | |
1357 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1358 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1359 | * careful when using this function in performance sensitive code. | |
1360 | */ | |
1361 | void kzfree(const void *p) | |
1362 | { | |
1363 | size_t ks; | |
1364 | void *mem = (void *)p; | |
1365 | ||
1366 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1367 | return; | |
1368 | ks = ksize(mem); | |
1369 | memset(mem, 0, ks); | |
1370 | kfree(mem); | |
1371 | } | |
1372 | EXPORT_SYMBOL(kzfree); | |
1373 | ||
1374 | /* Tracepoints definitions. */ | |
1375 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1376 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1377 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1378 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1379 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1380 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |