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> |
f1b6eb6e | 22 | #include <trace/events/kmem.h> |
039363f3 | 23 | |
97d06609 CL |
24 | #include "slab.h" |
25 | ||
26 | enum slab_state slab_state; | |
18004c5d CL |
27 | LIST_HEAD(slab_caches); |
28 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 29 | struct kmem_cache *kmem_cache; |
97d06609 | 30 | |
77be4b13 | 31 | #ifdef CONFIG_DEBUG_VM |
794b1248 | 32 | static int kmem_cache_sanity_check(const char *name, size_t size) |
039363f3 CL |
33 | { |
34 | struct kmem_cache *s = NULL; | |
35 | ||
039363f3 CL |
36 | if (!name || in_interrupt() || size < sizeof(void *) || |
37 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
38 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
39 | return -EINVAL; | |
039363f3 | 40 | } |
b920536a | 41 | |
20cea968 CL |
42 | list_for_each_entry(s, &slab_caches, list) { |
43 | char tmp; | |
44 | int res; | |
45 | ||
46 | /* | |
47 | * This happens when the module gets unloaded and doesn't | |
48 | * destroy its slab cache and no-one else reuses the vmalloc | |
49 | * area of the module. Print a warning. | |
50 | */ | |
51 | res = probe_kernel_address(s->name, tmp); | |
52 | if (res) { | |
77be4b13 | 53 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
54 | s->object_size); |
55 | continue; | |
56 | } | |
57 | ||
3e374919 | 58 | #if !defined(CONFIG_SLUB) || !defined(CONFIG_SLUB_DEBUG_ON) |
794b1248 | 59 | if (!strcmp(s->name, name)) { |
77be4b13 SK |
60 | pr_err("%s (%s): Cache name already exists.\n", |
61 | __func__, name); | |
20cea968 CL |
62 | dump_stack(); |
63 | s = NULL; | |
77be4b13 | 64 | return -EINVAL; |
20cea968 | 65 | } |
3e374919 | 66 | #endif |
20cea968 CL |
67 | } |
68 | ||
69 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
70 | return 0; |
71 | } | |
72 | #else | |
794b1248 | 73 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
77be4b13 SK |
74 | { |
75 | return 0; | |
76 | } | |
20cea968 CL |
77 | #endif |
78 | ||
55007d84 GC |
79 | #ifdef CONFIG_MEMCG_KMEM |
80 | int memcg_update_all_caches(int num_memcgs) | |
81 | { | |
82 | struct kmem_cache *s; | |
83 | int ret = 0; | |
84 | mutex_lock(&slab_mutex); | |
85 | ||
86 | list_for_each_entry(s, &slab_caches, list) { | |
87 | if (!is_root_cache(s)) | |
88 | continue; | |
89 | ||
90 | ret = memcg_update_cache_size(s, num_memcgs); | |
91 | /* | |
92 | * See comment in memcontrol.c, memcg_update_cache_size: | |
93 | * Instead of freeing the memory, we'll just leave the caches | |
94 | * up to this point in an updated state. | |
95 | */ | |
96 | if (ret) | |
97 | goto out; | |
98 | } | |
99 | ||
100 | memcg_update_array_size(num_memcgs); | |
101 | out: | |
102 | mutex_unlock(&slab_mutex); | |
103 | return ret; | |
104 | } | |
105 | #endif | |
106 | ||
45906855 CL |
107 | /* |
108 | * Figure out what the alignment of the objects will be given a set of | |
109 | * flags, a user specified alignment and the size of the objects. | |
110 | */ | |
111 | unsigned long calculate_alignment(unsigned long flags, | |
112 | unsigned long align, unsigned long size) | |
113 | { | |
114 | /* | |
115 | * If the user wants hardware cache aligned objects then follow that | |
116 | * suggestion if the object is sufficiently large. | |
117 | * | |
118 | * The hardware cache alignment cannot override the specified | |
119 | * alignment though. If that is greater then use it. | |
120 | */ | |
121 | if (flags & SLAB_HWCACHE_ALIGN) { | |
122 | unsigned long ralign = cache_line_size(); | |
123 | while (size <= ralign / 2) | |
124 | ralign /= 2; | |
125 | align = max(align, ralign); | |
126 | } | |
127 | ||
128 | if (align < ARCH_SLAB_MINALIGN) | |
129 | align = ARCH_SLAB_MINALIGN; | |
130 | ||
131 | return ALIGN(align, sizeof(void *)); | |
132 | } | |
133 | ||
794b1248 VD |
134 | static struct kmem_cache * |
135 | do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align, | |
136 | unsigned long flags, void (*ctor)(void *), | |
137 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
138 | { | |
139 | struct kmem_cache *s; | |
140 | int err; | |
141 | ||
142 | err = -ENOMEM; | |
143 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
144 | if (!s) | |
145 | goto out; | |
146 | ||
147 | s->name = name; | |
148 | s->object_size = object_size; | |
149 | s->size = size; | |
150 | s->align = align; | |
151 | s->ctor = ctor; | |
152 | ||
153 | err = memcg_alloc_cache_params(memcg, s, root_cache); | |
154 | if (err) | |
155 | goto out_free_cache; | |
156 | ||
157 | err = __kmem_cache_create(s, flags); | |
158 | if (err) | |
159 | goto out_free_cache; | |
160 | ||
161 | s->refcount = 1; | |
162 | list_add(&s->list, &slab_caches); | |
794b1248 VD |
163 | out: |
164 | if (err) | |
165 | return ERR_PTR(err); | |
166 | return s; | |
167 | ||
168 | out_free_cache: | |
169 | memcg_free_cache_params(s); | |
170 | kfree(s); | |
171 | goto out; | |
172 | } | |
45906855 | 173 | |
77be4b13 SK |
174 | /* |
175 | * kmem_cache_create - Create a cache. | |
176 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
177 | * @size: The size of objects to be created in this cache. | |
178 | * @align: The required alignment for the objects. | |
179 | * @flags: SLAB flags | |
180 | * @ctor: A constructor for the objects. | |
181 | * | |
182 | * Returns a ptr to the cache on success, NULL on failure. | |
183 | * Cannot be called within a interrupt, but can be interrupted. | |
184 | * The @ctor is run when new pages are allocated by the cache. | |
185 | * | |
186 | * The flags are | |
187 | * | |
188 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
189 | * to catch references to uninitialised memory. | |
190 | * | |
191 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
192 | * for buffer overruns. | |
193 | * | |
194 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
195 | * cacheline. This can be beneficial if you're counting cycles as closely | |
196 | * as davem. | |
197 | */ | |
2633d7a0 | 198 | struct kmem_cache * |
794b1248 VD |
199 | kmem_cache_create(const char *name, size_t size, size_t align, |
200 | unsigned long flags, void (*ctor)(void *)) | |
77be4b13 | 201 | { |
794b1248 VD |
202 | struct kmem_cache *s; |
203 | char *cache_name; | |
3965fc36 | 204 | int err; |
039363f3 | 205 | |
77be4b13 | 206 | get_online_cpus(); |
03afc0e2 VD |
207 | get_online_mems(); |
208 | ||
77be4b13 | 209 | mutex_lock(&slab_mutex); |
686d550d | 210 | |
794b1248 | 211 | err = kmem_cache_sanity_check(name, size); |
3965fc36 VD |
212 | if (err) |
213 | goto out_unlock; | |
686d550d | 214 | |
d8843922 GC |
215 | /* |
216 | * Some allocators will constraint the set of valid flags to a subset | |
217 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
218 | * case, and we'll just provide them with a sanitized version of the | |
219 | * passed flags. | |
220 | */ | |
221 | flags &= CACHE_CREATE_MASK; | |
686d550d | 222 | |
794b1248 VD |
223 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
224 | if (s) | |
3965fc36 | 225 | goto out_unlock; |
2633d7a0 | 226 | |
794b1248 VD |
227 | cache_name = kstrdup(name, GFP_KERNEL); |
228 | if (!cache_name) { | |
229 | err = -ENOMEM; | |
230 | goto out_unlock; | |
231 | } | |
7c9adf5a | 232 | |
794b1248 VD |
233 | s = do_kmem_cache_create(cache_name, size, size, |
234 | calculate_alignment(flags, align, size), | |
235 | flags, ctor, NULL, NULL); | |
236 | if (IS_ERR(s)) { | |
237 | err = PTR_ERR(s); | |
238 | kfree(cache_name); | |
239 | } | |
3965fc36 VD |
240 | |
241 | out_unlock: | |
20cea968 | 242 | mutex_unlock(&slab_mutex); |
03afc0e2 VD |
243 | |
244 | put_online_mems(); | |
20cea968 CL |
245 | put_online_cpus(); |
246 | ||
ba3253c7 | 247 | if (err) { |
686d550d CL |
248 | if (flags & SLAB_PANIC) |
249 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
250 | name, err); | |
251 | else { | |
252 | printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d", | |
253 | name, err); | |
254 | dump_stack(); | |
255 | } | |
686d550d CL |
256 | return NULL; |
257 | } | |
039363f3 CL |
258 | return s; |
259 | } | |
794b1248 | 260 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 261 | |
794b1248 VD |
262 | #ifdef CONFIG_MEMCG_KMEM |
263 | /* | |
776ed0f0 | 264 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
265 | * @memcg: The memory cgroup the new cache is for. |
266 | * @root_cache: The parent of the new cache. | |
073ee1c6 | 267 | * @memcg_name: The name of the memory cgroup (used for naming the new cache). |
794b1248 VD |
268 | * |
269 | * This function attempts to create a kmem cache that will serve allocation | |
270 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
271 | * from its parent. | |
272 | */ | |
776ed0f0 | 273 | struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
073ee1c6 VD |
274 | struct kmem_cache *root_cache, |
275 | const char *memcg_name) | |
2633d7a0 | 276 | { |
bd673145 | 277 | struct kmem_cache *s = NULL; |
794b1248 VD |
278 | char *cache_name; |
279 | ||
280 | get_online_cpus(); | |
03afc0e2 VD |
281 | get_online_mems(); |
282 | ||
794b1248 VD |
283 | mutex_lock(&slab_mutex); |
284 | ||
073ee1c6 VD |
285 | cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name, |
286 | memcg_cache_id(memcg), memcg_name); | |
794b1248 VD |
287 | if (!cache_name) |
288 | goto out_unlock; | |
289 | ||
290 | s = do_kmem_cache_create(cache_name, root_cache->object_size, | |
291 | root_cache->size, root_cache->align, | |
292 | root_cache->flags, root_cache->ctor, | |
293 | memcg, root_cache); | |
bd673145 | 294 | if (IS_ERR(s)) { |
794b1248 | 295 | kfree(cache_name); |
bd673145 VD |
296 | s = NULL; |
297 | } | |
794b1248 VD |
298 | |
299 | out_unlock: | |
300 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
301 | |
302 | put_online_mems(); | |
794b1248 | 303 | put_online_cpus(); |
bd673145 VD |
304 | |
305 | return s; | |
2633d7a0 | 306 | } |
b8529907 | 307 | |
776ed0f0 | 308 | static int memcg_cleanup_cache_params(struct kmem_cache *s) |
b8529907 VD |
309 | { |
310 | int rc; | |
311 | ||
312 | if (!s->memcg_params || | |
313 | !s->memcg_params->is_root_cache) | |
314 | return 0; | |
315 | ||
316 | mutex_unlock(&slab_mutex); | |
776ed0f0 | 317 | rc = __memcg_cleanup_cache_params(s); |
b8529907 VD |
318 | mutex_lock(&slab_mutex); |
319 | ||
320 | return rc; | |
321 | } | |
322 | #else | |
776ed0f0 | 323 | static int memcg_cleanup_cache_params(struct kmem_cache *s) |
b8529907 VD |
324 | { |
325 | return 0; | |
326 | } | |
794b1248 | 327 | #endif /* CONFIG_MEMCG_KMEM */ |
97d06609 | 328 | |
41a21285 CL |
329 | void slab_kmem_cache_release(struct kmem_cache *s) |
330 | { | |
331 | kfree(s->name); | |
332 | kmem_cache_free(kmem_cache, s); | |
333 | } | |
334 | ||
945cf2b6 CL |
335 | void kmem_cache_destroy(struct kmem_cache *s) |
336 | { | |
337 | get_online_cpus(); | |
03afc0e2 VD |
338 | get_online_mems(); |
339 | ||
945cf2b6 | 340 | mutex_lock(&slab_mutex); |
b8529907 | 341 | |
945cf2b6 | 342 | s->refcount--; |
b8529907 VD |
343 | if (s->refcount) |
344 | goto out_unlock; | |
345 | ||
776ed0f0 | 346 | if (memcg_cleanup_cache_params(s) != 0) |
b8529907 VD |
347 | goto out_unlock; |
348 | ||
b8529907 | 349 | if (__kmem_cache_shutdown(s) != 0) { |
b8529907 VD |
350 | printk(KERN_ERR "kmem_cache_destroy %s: " |
351 | "Slab cache still has objects\n", s->name); | |
352 | dump_stack(); | |
353 | goto out_unlock; | |
945cf2b6 | 354 | } |
b8529907 | 355 | |
0bd62b11 VD |
356 | list_del(&s->list); |
357 | ||
b8529907 VD |
358 | mutex_unlock(&slab_mutex); |
359 | if (s->flags & SLAB_DESTROY_BY_RCU) | |
360 | rcu_barrier(); | |
361 | ||
362 | memcg_free_cache_params(s); | |
41a21285 CL |
363 | #ifdef SLAB_SUPPORTS_SYSFS |
364 | sysfs_slab_remove(s); | |
365 | #else | |
366 | slab_kmem_cache_release(s); | |
367 | #endif | |
03afc0e2 | 368 | goto out; |
b8529907 VD |
369 | |
370 | out_unlock: | |
371 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
372 | out: |
373 | put_online_mems(); | |
945cf2b6 CL |
374 | put_online_cpus(); |
375 | } | |
376 | EXPORT_SYMBOL(kmem_cache_destroy); | |
377 | ||
03afc0e2 VD |
378 | /** |
379 | * kmem_cache_shrink - Shrink a cache. | |
380 | * @cachep: The cache to shrink. | |
381 | * | |
382 | * Releases as many slabs as possible for a cache. | |
383 | * To help debugging, a zero exit status indicates all slabs were released. | |
384 | */ | |
385 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
386 | { | |
387 | int ret; | |
388 | ||
389 | get_online_cpus(); | |
390 | get_online_mems(); | |
391 | ret = __kmem_cache_shrink(cachep); | |
392 | put_online_mems(); | |
393 | put_online_cpus(); | |
394 | return ret; | |
395 | } | |
396 | EXPORT_SYMBOL(kmem_cache_shrink); | |
397 | ||
97d06609 CL |
398 | int slab_is_available(void) |
399 | { | |
400 | return slab_state >= UP; | |
401 | } | |
b7454ad3 | 402 | |
45530c44 CL |
403 | #ifndef CONFIG_SLOB |
404 | /* Create a cache during boot when no slab services are available yet */ | |
405 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
406 | unsigned long flags) | |
407 | { | |
408 | int err; | |
409 | ||
410 | s->name = name; | |
411 | s->size = s->object_size = size; | |
45906855 | 412 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
45530c44 CL |
413 | err = __kmem_cache_create(s, flags); |
414 | ||
415 | if (err) | |
31ba7346 | 416 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
417 | name, size, err); |
418 | ||
419 | s->refcount = -1; /* Exempt from merging for now */ | |
420 | } | |
421 | ||
422 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
423 | unsigned long flags) | |
424 | { | |
425 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
426 | ||
427 | if (!s) | |
428 | panic("Out of memory when creating slab %s\n", name); | |
429 | ||
430 | create_boot_cache(s, name, size, flags); | |
431 | list_add(&s->list, &slab_caches); | |
432 | s->refcount = 1; | |
433 | return s; | |
434 | } | |
435 | ||
9425c58e CL |
436 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
437 | EXPORT_SYMBOL(kmalloc_caches); | |
438 | ||
439 | #ifdef CONFIG_ZONE_DMA | |
440 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
441 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
442 | #endif | |
443 | ||
2c59dd65 CL |
444 | /* |
445 | * Conversion table for small slabs sizes / 8 to the index in the | |
446 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
447 | * of two cache sizes there. The size of larger slabs can be determined using | |
448 | * fls. | |
449 | */ | |
450 | static s8 size_index[24] = { | |
451 | 3, /* 8 */ | |
452 | 4, /* 16 */ | |
453 | 5, /* 24 */ | |
454 | 5, /* 32 */ | |
455 | 6, /* 40 */ | |
456 | 6, /* 48 */ | |
457 | 6, /* 56 */ | |
458 | 6, /* 64 */ | |
459 | 1, /* 72 */ | |
460 | 1, /* 80 */ | |
461 | 1, /* 88 */ | |
462 | 1, /* 96 */ | |
463 | 7, /* 104 */ | |
464 | 7, /* 112 */ | |
465 | 7, /* 120 */ | |
466 | 7, /* 128 */ | |
467 | 2, /* 136 */ | |
468 | 2, /* 144 */ | |
469 | 2, /* 152 */ | |
470 | 2, /* 160 */ | |
471 | 2, /* 168 */ | |
472 | 2, /* 176 */ | |
473 | 2, /* 184 */ | |
474 | 2 /* 192 */ | |
475 | }; | |
476 | ||
477 | static inline int size_index_elem(size_t bytes) | |
478 | { | |
479 | return (bytes - 1) / 8; | |
480 | } | |
481 | ||
482 | /* | |
483 | * Find the kmem_cache structure that serves a given size of | |
484 | * allocation | |
485 | */ | |
486 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
487 | { | |
488 | int index; | |
489 | ||
9de1bc87 | 490 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 491 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 492 | return NULL; |
907985f4 | 493 | } |
6286ae97 | 494 | |
2c59dd65 CL |
495 | if (size <= 192) { |
496 | if (!size) | |
497 | return ZERO_SIZE_PTR; | |
498 | ||
499 | index = size_index[size_index_elem(size)]; | |
500 | } else | |
501 | index = fls(size - 1); | |
502 | ||
503 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 504 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
505 | return kmalloc_dma_caches[index]; |
506 | ||
507 | #endif | |
508 | return kmalloc_caches[index]; | |
509 | } | |
510 | ||
f97d5f63 CL |
511 | /* |
512 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
513 | * may already have been created because they were needed to | |
514 | * enable allocations for slab creation. | |
515 | */ | |
516 | void __init create_kmalloc_caches(unsigned long flags) | |
517 | { | |
518 | int i; | |
519 | ||
2c59dd65 CL |
520 | /* |
521 | * Patch up the size_index table if we have strange large alignment | |
522 | * requirements for the kmalloc array. This is only the case for | |
523 | * MIPS it seems. The standard arches will not generate any code here. | |
524 | * | |
525 | * Largest permitted alignment is 256 bytes due to the way we | |
526 | * handle the index determination for the smaller caches. | |
527 | * | |
528 | * Make sure that nothing crazy happens if someone starts tinkering | |
529 | * around with ARCH_KMALLOC_MINALIGN | |
530 | */ | |
531 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
532 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
533 | ||
534 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
535 | int elem = size_index_elem(i); | |
536 | ||
537 | if (elem >= ARRAY_SIZE(size_index)) | |
538 | break; | |
539 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
540 | } | |
541 | ||
542 | if (KMALLOC_MIN_SIZE >= 64) { | |
543 | /* | |
544 | * The 96 byte size cache is not used if the alignment | |
545 | * is 64 byte. | |
546 | */ | |
547 | for (i = 64 + 8; i <= 96; i += 8) | |
548 | size_index[size_index_elem(i)] = 7; | |
549 | ||
550 | } | |
551 | ||
552 | if (KMALLOC_MIN_SIZE >= 128) { | |
553 | /* | |
554 | * The 192 byte sized cache is not used if the alignment | |
555 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
556 | * instead. | |
557 | */ | |
558 | for (i = 128 + 8; i <= 192; i += 8) | |
559 | size_index[size_index_elem(i)] = 8; | |
560 | } | |
8a965b3b CL |
561 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
562 | if (!kmalloc_caches[i]) { | |
f97d5f63 CL |
563 | kmalloc_caches[i] = create_kmalloc_cache(NULL, |
564 | 1 << i, flags); | |
956e46ef | 565 | } |
f97d5f63 | 566 | |
956e46ef CM |
567 | /* |
568 | * Caches that are not of the two-to-the-power-of size. | |
569 | * These have to be created immediately after the | |
570 | * earlier power of two caches | |
571 | */ | |
572 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) | |
573 | kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags); | |
8a965b3b | 574 | |
956e46ef CM |
575 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) |
576 | kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags); | |
8a965b3b CL |
577 | } |
578 | ||
f97d5f63 CL |
579 | /* Kmalloc array is now usable */ |
580 | slab_state = UP; | |
581 | ||
582 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
583 | struct kmem_cache *s = kmalloc_caches[i]; | |
584 | char *n; | |
585 | ||
586 | if (s) { | |
587 | n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i)); | |
588 | ||
589 | BUG_ON(!n); | |
590 | s->name = n; | |
591 | } | |
592 | } | |
593 | ||
594 | #ifdef CONFIG_ZONE_DMA | |
595 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
596 | struct kmem_cache *s = kmalloc_caches[i]; | |
597 | ||
598 | if (s) { | |
599 | int size = kmalloc_size(i); | |
600 | char *n = kasprintf(GFP_NOWAIT, | |
601 | "dma-kmalloc-%d", size); | |
602 | ||
603 | BUG_ON(!n); | |
604 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
605 | size, SLAB_CACHE_DMA | flags); | |
606 | } | |
607 | } | |
608 | #endif | |
609 | } | |
45530c44 CL |
610 | #endif /* !CONFIG_SLOB */ |
611 | ||
cea371f4 VD |
612 | /* |
613 | * To avoid unnecessary overhead, we pass through large allocation requests | |
614 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
615 | * know the allocation order to free the pages properly in kfree. | |
616 | */ | |
52383431 VD |
617 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
618 | { | |
619 | void *ret; | |
620 | struct page *page; | |
621 | ||
622 | flags |= __GFP_COMP; | |
623 | page = alloc_kmem_pages(flags, order); | |
624 | ret = page ? page_address(page) : NULL; | |
625 | kmemleak_alloc(ret, size, 1, flags); | |
626 | return ret; | |
627 | } | |
628 | EXPORT_SYMBOL(kmalloc_order); | |
629 | ||
f1b6eb6e CL |
630 | #ifdef CONFIG_TRACING |
631 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
632 | { | |
633 | void *ret = kmalloc_order(size, flags, order); | |
634 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
635 | return ret; | |
636 | } | |
637 | EXPORT_SYMBOL(kmalloc_order_trace); | |
638 | #endif | |
45530c44 | 639 | |
b7454ad3 | 640 | #ifdef CONFIG_SLABINFO |
e9b4db2b WL |
641 | |
642 | #ifdef CONFIG_SLAB | |
643 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
644 | #else | |
645 | #define SLABINFO_RIGHTS S_IRUSR | |
646 | #endif | |
647 | ||
749c5415 | 648 | void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
649 | { |
650 | /* | |
651 | * Output format version, so at least we can change it | |
652 | * without _too_ many complaints. | |
653 | */ | |
654 | #ifdef CONFIG_DEBUG_SLAB | |
655 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
656 | #else | |
657 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
658 | #endif | |
659 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
660 | "<objperslab> <pagesperslab>"); | |
661 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
662 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
663 | #ifdef CONFIG_DEBUG_SLAB | |
664 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " | |
665 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); | |
666 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
667 | #endif | |
668 | seq_putc(m, '\n'); | |
669 | } | |
670 | ||
b7454ad3 GC |
671 | static void *s_start(struct seq_file *m, loff_t *pos) |
672 | { | |
673 | loff_t n = *pos; | |
674 | ||
675 | mutex_lock(&slab_mutex); | |
676 | if (!n) | |
677 | print_slabinfo_header(m); | |
678 | ||
679 | return seq_list_start(&slab_caches, *pos); | |
680 | } | |
681 | ||
276a2439 | 682 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 GC |
683 | { |
684 | return seq_list_next(p, &slab_caches, pos); | |
685 | } | |
686 | ||
276a2439 | 687 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
688 | { |
689 | mutex_unlock(&slab_mutex); | |
690 | } | |
691 | ||
749c5415 GC |
692 | static void |
693 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
694 | { | |
695 | struct kmem_cache *c; | |
696 | struct slabinfo sinfo; | |
697 | int i; | |
698 | ||
699 | if (!is_root_cache(s)) | |
700 | return; | |
701 | ||
702 | for_each_memcg_cache_index(i) { | |
2ade4de8 | 703 | c = cache_from_memcg_idx(s, i); |
749c5415 GC |
704 | if (!c) |
705 | continue; | |
706 | ||
707 | memset(&sinfo, 0, sizeof(sinfo)); | |
708 | get_slabinfo(c, &sinfo); | |
709 | ||
710 | info->active_slabs += sinfo.active_slabs; | |
711 | info->num_slabs += sinfo.num_slabs; | |
712 | info->shared_avail += sinfo.shared_avail; | |
713 | info->active_objs += sinfo.active_objs; | |
714 | info->num_objs += sinfo.num_objs; | |
715 | } | |
716 | } | |
717 | ||
718 | int cache_show(struct kmem_cache *s, struct seq_file *m) | |
b7454ad3 | 719 | { |
0d7561c6 GC |
720 | struct slabinfo sinfo; |
721 | ||
722 | memset(&sinfo, 0, sizeof(sinfo)); | |
723 | get_slabinfo(s, &sinfo); | |
724 | ||
749c5415 GC |
725 | memcg_accumulate_slabinfo(s, &sinfo); |
726 | ||
0d7561c6 | 727 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 728 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
729 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
730 | ||
731 | seq_printf(m, " : tunables %4u %4u %4u", | |
732 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
733 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
734 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
735 | slabinfo_show_stats(m, s); | |
736 | seq_putc(m, '\n'); | |
737 | return 0; | |
b7454ad3 GC |
738 | } |
739 | ||
749c5415 GC |
740 | static int s_show(struct seq_file *m, void *p) |
741 | { | |
742 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
743 | ||
744 | if (!is_root_cache(s)) | |
745 | return 0; | |
746 | return cache_show(s, m); | |
747 | } | |
748 | ||
b7454ad3 GC |
749 | /* |
750 | * slabinfo_op - iterator that generates /proc/slabinfo | |
751 | * | |
752 | * Output layout: | |
753 | * cache-name | |
754 | * num-active-objs | |
755 | * total-objs | |
756 | * object size | |
757 | * num-active-slabs | |
758 | * total-slabs | |
759 | * num-pages-per-slab | |
760 | * + further values on SMP and with statistics enabled | |
761 | */ | |
762 | static const struct seq_operations slabinfo_op = { | |
763 | .start = s_start, | |
276a2439 WL |
764 | .next = slab_next, |
765 | .stop = slab_stop, | |
b7454ad3 GC |
766 | .show = s_show, |
767 | }; | |
768 | ||
769 | static int slabinfo_open(struct inode *inode, struct file *file) | |
770 | { | |
771 | return seq_open(file, &slabinfo_op); | |
772 | } | |
773 | ||
774 | static const struct file_operations proc_slabinfo_operations = { | |
775 | .open = slabinfo_open, | |
776 | .read = seq_read, | |
777 | .write = slabinfo_write, | |
778 | .llseek = seq_lseek, | |
779 | .release = seq_release, | |
780 | }; | |
781 | ||
782 | static int __init slab_proc_init(void) | |
783 | { | |
e9b4db2b WL |
784 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
785 | &proc_slabinfo_operations); | |
b7454ad3 GC |
786 | return 0; |
787 | } | |
788 | module_init(slab_proc_init); | |
789 | #endif /* CONFIG_SLABINFO */ |