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 |
2633d7a0 GC |
32 | static int kmem_cache_sanity_check(struct mem_cgroup *memcg, const char *name, |
33 | size_t size) | |
039363f3 CL |
34 | { |
35 | struct kmem_cache *s = NULL; | |
36 | ||
039363f3 CL |
37 | if (!name || in_interrupt() || size < sizeof(void *) || |
38 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
39 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
40 | return -EINVAL; | |
039363f3 | 41 | } |
b920536a | 42 | |
20cea968 CL |
43 | list_for_each_entry(s, &slab_caches, list) { |
44 | char tmp; | |
45 | int res; | |
46 | ||
47 | /* | |
48 | * This happens when the module gets unloaded and doesn't | |
49 | * destroy its slab cache and no-one else reuses the vmalloc | |
50 | * area of the module. Print a warning. | |
51 | */ | |
52 | res = probe_kernel_address(s->name, tmp); | |
53 | if (res) { | |
77be4b13 | 54 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
55 | s->object_size); |
56 | continue; | |
57 | } | |
58 | ||
3e374919 | 59 | #if !defined(CONFIG_SLUB) || !defined(CONFIG_SLUB_DEBUG_ON) |
2633d7a0 GC |
60 | /* |
61 | * For simplicity, we won't check this in the list of memcg | |
62 | * caches. We have control over memcg naming, and if there | |
63 | * aren't duplicates in the global list, there won't be any | |
64 | * duplicates in the memcg lists as well. | |
65 | */ | |
66 | if (!memcg && !strcmp(s->name, name)) { | |
77be4b13 SK |
67 | pr_err("%s (%s): Cache name already exists.\n", |
68 | __func__, name); | |
20cea968 CL |
69 | dump_stack(); |
70 | s = NULL; | |
77be4b13 | 71 | return -EINVAL; |
20cea968 | 72 | } |
3e374919 | 73 | #endif |
20cea968 CL |
74 | } |
75 | ||
76 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
77 | return 0; |
78 | } | |
79 | #else | |
2633d7a0 GC |
80 | static inline int kmem_cache_sanity_check(struct mem_cgroup *memcg, |
81 | const char *name, size_t size) | |
77be4b13 SK |
82 | { |
83 | return 0; | |
84 | } | |
20cea968 CL |
85 | #endif |
86 | ||
55007d84 GC |
87 | #ifdef CONFIG_MEMCG_KMEM |
88 | int memcg_update_all_caches(int num_memcgs) | |
89 | { | |
90 | struct kmem_cache *s; | |
91 | int ret = 0; | |
92 | mutex_lock(&slab_mutex); | |
93 | ||
94 | list_for_each_entry(s, &slab_caches, list) { | |
95 | if (!is_root_cache(s)) | |
96 | continue; | |
97 | ||
98 | ret = memcg_update_cache_size(s, num_memcgs); | |
99 | /* | |
100 | * See comment in memcontrol.c, memcg_update_cache_size: | |
101 | * Instead of freeing the memory, we'll just leave the caches | |
102 | * up to this point in an updated state. | |
103 | */ | |
104 | if (ret) | |
105 | goto out; | |
106 | } | |
107 | ||
108 | memcg_update_array_size(num_memcgs); | |
109 | out: | |
110 | mutex_unlock(&slab_mutex); | |
111 | return ret; | |
112 | } | |
113 | #endif | |
114 | ||
45906855 CL |
115 | /* |
116 | * Figure out what the alignment of the objects will be given a set of | |
117 | * flags, a user specified alignment and the size of the objects. | |
118 | */ | |
119 | unsigned long calculate_alignment(unsigned long flags, | |
120 | unsigned long align, unsigned long size) | |
121 | { | |
122 | /* | |
123 | * If the user wants hardware cache aligned objects then follow that | |
124 | * suggestion if the object is sufficiently large. | |
125 | * | |
126 | * The hardware cache alignment cannot override the specified | |
127 | * alignment though. If that is greater then use it. | |
128 | */ | |
129 | if (flags & SLAB_HWCACHE_ALIGN) { | |
130 | unsigned long ralign = cache_line_size(); | |
131 | while (size <= ralign / 2) | |
132 | ralign /= 2; | |
133 | align = max(align, ralign); | |
134 | } | |
135 | ||
136 | if (align < ARCH_SLAB_MINALIGN) | |
137 | align = ARCH_SLAB_MINALIGN; | |
138 | ||
139 | return ALIGN(align, sizeof(void *)); | |
140 | } | |
141 | ||
142 | ||
77be4b13 SK |
143 | /* |
144 | * kmem_cache_create - Create a cache. | |
145 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
146 | * @size: The size of objects to be created in this cache. | |
147 | * @align: The required alignment for the objects. | |
148 | * @flags: SLAB flags | |
149 | * @ctor: A constructor for the objects. | |
150 | * | |
151 | * Returns a ptr to the cache on success, NULL on failure. | |
152 | * Cannot be called within a interrupt, but can be interrupted. | |
153 | * The @ctor is run when new pages are allocated by the cache. | |
154 | * | |
155 | * The flags are | |
156 | * | |
157 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
158 | * to catch references to uninitialised memory. | |
159 | * | |
160 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
161 | * for buffer overruns. | |
162 | * | |
163 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
164 | * cacheline. This can be beneficial if you're counting cycles as closely | |
165 | * as davem. | |
166 | */ | |
167 | ||
2633d7a0 GC |
168 | struct kmem_cache * |
169 | kmem_cache_create_memcg(struct mem_cgroup *memcg, const char *name, size_t size, | |
943a451a GC |
170 | size_t align, unsigned long flags, void (*ctor)(void *), |
171 | struct kmem_cache *parent_cache) | |
77be4b13 SK |
172 | { |
173 | struct kmem_cache *s = NULL; | |
3965fc36 | 174 | int err; |
039363f3 | 175 | |
77be4b13 SK |
176 | get_online_cpus(); |
177 | mutex_lock(&slab_mutex); | |
686d550d | 178 | |
3965fc36 VD |
179 | err = kmem_cache_sanity_check(memcg, name, size); |
180 | if (err) | |
181 | goto out_unlock; | |
686d550d | 182 | |
2edefe11 VD |
183 | if (memcg) { |
184 | /* | |
185 | * Since per-memcg caches are created asynchronously on first | |
186 | * allocation (see memcg_kmem_get_cache()), several threads can | |
187 | * try to create the same cache, but only one of them may | |
188 | * succeed. Therefore if we get here and see the cache has | |
189 | * already been created, we silently return NULL. | |
190 | */ | |
191 | if (cache_from_memcg_idx(parent_cache, memcg_cache_id(memcg))) | |
192 | goto out_unlock; | |
193 | } | |
194 | ||
d8843922 GC |
195 | /* |
196 | * Some allocators will constraint the set of valid flags to a subset | |
197 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
198 | * case, and we'll just provide them with a sanitized version of the | |
199 | * passed flags. | |
200 | */ | |
201 | flags &= CACHE_CREATE_MASK; | |
686d550d | 202 | |
a44cb944 VD |
203 | if (!memcg) { |
204 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
205 | if (s) | |
206 | goto out_unlock; | |
207 | } | |
cbb79694 | 208 | |
3965fc36 | 209 | err = -ENOMEM; |
278b1bb1 | 210 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
3965fc36 VD |
211 | if (!s) |
212 | goto out_unlock; | |
2633d7a0 | 213 | |
3965fc36 VD |
214 | s->object_size = s->size = size; |
215 | s->align = calculate_alignment(flags, align, size); | |
216 | s->ctor = ctor; | |
8a13a4cc | 217 | |
5722d094 VD |
218 | if (memcg) |
219 | s->name = memcg_create_cache_name(memcg, parent_cache); | |
220 | else | |
221 | s->name = kstrdup(name, GFP_KERNEL); | |
3965fc36 VD |
222 | if (!s->name) |
223 | goto out_free_cache; | |
224 | ||
363a044f | 225 | err = memcg_alloc_cache_params(memcg, s, parent_cache); |
3965fc36 VD |
226 | if (err) |
227 | goto out_free_cache; | |
228 | ||
229 | err = __kmem_cache_create(s, flags); | |
230 | if (err) | |
231 | goto out_free_cache; | |
7c9adf5a | 232 | |
3965fc36 VD |
233 | s->refcount = 1; |
234 | list_add(&s->list, &slab_caches); | |
1aa13254 | 235 | memcg_register_cache(s); |
3965fc36 VD |
236 | |
237 | out_unlock: | |
20cea968 CL |
238 | mutex_unlock(&slab_mutex); |
239 | put_online_cpus(); | |
240 | ||
ba3253c7 DJ |
241 | if (err) { |
242 | /* | |
243 | * There is no point in flooding logs with warnings or | |
244 | * especially crashing the system if we fail to create a cache | |
245 | * for a memcg. In this case we will be accounting the memcg | |
246 | * allocation to the root cgroup until we succeed to create its | |
247 | * own cache, but it isn't that critical. | |
248 | */ | |
249 | if (!memcg) | |
250 | return NULL; | |
251 | ||
686d550d CL |
252 | if (flags & SLAB_PANIC) |
253 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
254 | name, err); | |
255 | else { | |
256 | printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d", | |
257 | name, err); | |
258 | dump_stack(); | |
259 | } | |
686d550d CL |
260 | return NULL; |
261 | } | |
039363f3 | 262 | return s; |
3965fc36 VD |
263 | |
264 | out_free_cache: | |
363a044f | 265 | memcg_free_cache_params(s); |
3965fc36 VD |
266 | kfree(s->name); |
267 | kmem_cache_free(kmem_cache, s); | |
268 | goto out_unlock; | |
039363f3 | 269 | } |
2633d7a0 GC |
270 | |
271 | struct kmem_cache * | |
272 | kmem_cache_create(const char *name, size_t size, size_t align, | |
273 | unsigned long flags, void (*ctor)(void *)) | |
274 | { | |
943a451a | 275 | return kmem_cache_create_memcg(NULL, name, size, align, flags, ctor, NULL); |
2633d7a0 | 276 | } |
039363f3 | 277 | EXPORT_SYMBOL(kmem_cache_create); |
97d06609 | 278 | |
945cf2b6 CL |
279 | void kmem_cache_destroy(struct kmem_cache *s) |
280 | { | |
7cf27982 GC |
281 | /* Destroy all the children caches if we aren't a memcg cache */ |
282 | kmem_cache_destroy_memcg_children(s); | |
283 | ||
945cf2b6 CL |
284 | get_online_cpus(); |
285 | mutex_lock(&slab_mutex); | |
286 | s->refcount--; | |
287 | if (!s->refcount) { | |
288 | list_del(&s->list); | |
289 | ||
290 | if (!__kmem_cache_shutdown(s)) { | |
2edefe11 | 291 | memcg_unregister_cache(s); |
210ed9de | 292 | mutex_unlock(&slab_mutex); |
945cf2b6 CL |
293 | if (s->flags & SLAB_DESTROY_BY_RCU) |
294 | rcu_barrier(); | |
295 | ||
1aa13254 | 296 | memcg_free_cache_params(s); |
db265eca | 297 | kfree(s->name); |
8f4c765c | 298 | kmem_cache_free(kmem_cache, s); |
945cf2b6 CL |
299 | } else { |
300 | list_add(&s->list, &slab_caches); | |
210ed9de | 301 | mutex_unlock(&slab_mutex); |
945cf2b6 CL |
302 | printk(KERN_ERR "kmem_cache_destroy %s: Slab cache still has objects\n", |
303 | s->name); | |
304 | dump_stack(); | |
305 | } | |
210ed9de JK |
306 | } else { |
307 | mutex_unlock(&slab_mutex); | |
945cf2b6 | 308 | } |
945cf2b6 CL |
309 | put_online_cpus(); |
310 | } | |
311 | EXPORT_SYMBOL(kmem_cache_destroy); | |
312 | ||
97d06609 CL |
313 | int slab_is_available(void) |
314 | { | |
315 | return slab_state >= UP; | |
316 | } | |
b7454ad3 | 317 | |
45530c44 CL |
318 | #ifndef CONFIG_SLOB |
319 | /* Create a cache during boot when no slab services are available yet */ | |
320 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
321 | unsigned long flags) | |
322 | { | |
323 | int err; | |
324 | ||
325 | s->name = name; | |
326 | s->size = s->object_size = size; | |
45906855 | 327 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
45530c44 CL |
328 | err = __kmem_cache_create(s, flags); |
329 | ||
330 | if (err) | |
31ba7346 | 331 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
332 | name, size, err); |
333 | ||
334 | s->refcount = -1; /* Exempt from merging for now */ | |
335 | } | |
336 | ||
337 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
338 | unsigned long flags) | |
339 | { | |
340 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
341 | ||
342 | if (!s) | |
343 | panic("Out of memory when creating slab %s\n", name); | |
344 | ||
345 | create_boot_cache(s, name, size, flags); | |
346 | list_add(&s->list, &slab_caches); | |
347 | s->refcount = 1; | |
348 | return s; | |
349 | } | |
350 | ||
9425c58e CL |
351 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
352 | EXPORT_SYMBOL(kmalloc_caches); | |
353 | ||
354 | #ifdef CONFIG_ZONE_DMA | |
355 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
356 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
357 | #endif | |
358 | ||
2c59dd65 CL |
359 | /* |
360 | * Conversion table for small slabs sizes / 8 to the index in the | |
361 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
362 | * of two cache sizes there. The size of larger slabs can be determined using | |
363 | * fls. | |
364 | */ | |
365 | static s8 size_index[24] = { | |
366 | 3, /* 8 */ | |
367 | 4, /* 16 */ | |
368 | 5, /* 24 */ | |
369 | 5, /* 32 */ | |
370 | 6, /* 40 */ | |
371 | 6, /* 48 */ | |
372 | 6, /* 56 */ | |
373 | 6, /* 64 */ | |
374 | 1, /* 72 */ | |
375 | 1, /* 80 */ | |
376 | 1, /* 88 */ | |
377 | 1, /* 96 */ | |
378 | 7, /* 104 */ | |
379 | 7, /* 112 */ | |
380 | 7, /* 120 */ | |
381 | 7, /* 128 */ | |
382 | 2, /* 136 */ | |
383 | 2, /* 144 */ | |
384 | 2, /* 152 */ | |
385 | 2, /* 160 */ | |
386 | 2, /* 168 */ | |
387 | 2, /* 176 */ | |
388 | 2, /* 184 */ | |
389 | 2 /* 192 */ | |
390 | }; | |
391 | ||
392 | static inline int size_index_elem(size_t bytes) | |
393 | { | |
394 | return (bytes - 1) / 8; | |
395 | } | |
396 | ||
397 | /* | |
398 | * Find the kmem_cache structure that serves a given size of | |
399 | * allocation | |
400 | */ | |
401 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
402 | { | |
403 | int index; | |
404 | ||
9de1bc87 | 405 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 406 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 407 | return NULL; |
907985f4 | 408 | } |
6286ae97 | 409 | |
2c59dd65 CL |
410 | if (size <= 192) { |
411 | if (!size) | |
412 | return ZERO_SIZE_PTR; | |
413 | ||
414 | index = size_index[size_index_elem(size)]; | |
415 | } else | |
416 | index = fls(size - 1); | |
417 | ||
418 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 419 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
420 | return kmalloc_dma_caches[index]; |
421 | ||
422 | #endif | |
423 | return kmalloc_caches[index]; | |
424 | } | |
425 | ||
f97d5f63 CL |
426 | /* |
427 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
428 | * may already have been created because they were needed to | |
429 | * enable allocations for slab creation. | |
430 | */ | |
431 | void __init create_kmalloc_caches(unsigned long flags) | |
432 | { | |
433 | int i; | |
434 | ||
2c59dd65 CL |
435 | /* |
436 | * Patch up the size_index table if we have strange large alignment | |
437 | * requirements for the kmalloc array. This is only the case for | |
438 | * MIPS it seems. The standard arches will not generate any code here. | |
439 | * | |
440 | * Largest permitted alignment is 256 bytes due to the way we | |
441 | * handle the index determination for the smaller caches. | |
442 | * | |
443 | * Make sure that nothing crazy happens if someone starts tinkering | |
444 | * around with ARCH_KMALLOC_MINALIGN | |
445 | */ | |
446 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
447 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
448 | ||
449 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
450 | int elem = size_index_elem(i); | |
451 | ||
452 | if (elem >= ARRAY_SIZE(size_index)) | |
453 | break; | |
454 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
455 | } | |
456 | ||
457 | if (KMALLOC_MIN_SIZE >= 64) { | |
458 | /* | |
459 | * The 96 byte size cache is not used if the alignment | |
460 | * is 64 byte. | |
461 | */ | |
462 | for (i = 64 + 8; i <= 96; i += 8) | |
463 | size_index[size_index_elem(i)] = 7; | |
464 | ||
465 | } | |
466 | ||
467 | if (KMALLOC_MIN_SIZE >= 128) { | |
468 | /* | |
469 | * The 192 byte sized cache is not used if the alignment | |
470 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
471 | * instead. | |
472 | */ | |
473 | for (i = 128 + 8; i <= 192; i += 8) | |
474 | size_index[size_index_elem(i)] = 8; | |
475 | } | |
8a965b3b CL |
476 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
477 | if (!kmalloc_caches[i]) { | |
f97d5f63 CL |
478 | kmalloc_caches[i] = create_kmalloc_cache(NULL, |
479 | 1 << i, flags); | |
956e46ef | 480 | } |
f97d5f63 | 481 | |
956e46ef CM |
482 | /* |
483 | * Caches that are not of the two-to-the-power-of size. | |
484 | * These have to be created immediately after the | |
485 | * earlier power of two caches | |
486 | */ | |
487 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) | |
488 | kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags); | |
8a965b3b | 489 | |
956e46ef CM |
490 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) |
491 | kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags); | |
8a965b3b CL |
492 | } |
493 | ||
f97d5f63 CL |
494 | /* Kmalloc array is now usable */ |
495 | slab_state = UP; | |
496 | ||
497 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
498 | struct kmem_cache *s = kmalloc_caches[i]; | |
499 | char *n; | |
500 | ||
501 | if (s) { | |
502 | n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i)); | |
503 | ||
504 | BUG_ON(!n); | |
505 | s->name = n; | |
506 | } | |
507 | } | |
508 | ||
509 | #ifdef CONFIG_ZONE_DMA | |
510 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
511 | struct kmem_cache *s = kmalloc_caches[i]; | |
512 | ||
513 | if (s) { | |
514 | int size = kmalloc_size(i); | |
515 | char *n = kasprintf(GFP_NOWAIT, | |
516 | "dma-kmalloc-%d", size); | |
517 | ||
518 | BUG_ON(!n); | |
519 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
520 | size, SLAB_CACHE_DMA | flags); | |
521 | } | |
522 | } | |
523 | #endif | |
524 | } | |
45530c44 CL |
525 | #endif /* !CONFIG_SLOB */ |
526 | ||
f1b6eb6e CL |
527 | #ifdef CONFIG_TRACING |
528 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
529 | { | |
530 | void *ret = kmalloc_order(size, flags, order); | |
531 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
532 | return ret; | |
533 | } | |
534 | EXPORT_SYMBOL(kmalloc_order_trace); | |
535 | #endif | |
45530c44 | 536 | |
b7454ad3 | 537 | #ifdef CONFIG_SLABINFO |
e9b4db2b WL |
538 | |
539 | #ifdef CONFIG_SLAB | |
540 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
541 | #else | |
542 | #define SLABINFO_RIGHTS S_IRUSR | |
543 | #endif | |
544 | ||
749c5415 | 545 | void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
546 | { |
547 | /* | |
548 | * Output format version, so at least we can change it | |
549 | * without _too_ many complaints. | |
550 | */ | |
551 | #ifdef CONFIG_DEBUG_SLAB | |
552 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
553 | #else | |
554 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
555 | #endif | |
556 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
557 | "<objperslab> <pagesperslab>"); | |
558 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
559 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
560 | #ifdef CONFIG_DEBUG_SLAB | |
561 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " | |
562 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); | |
563 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
564 | #endif | |
565 | seq_putc(m, '\n'); | |
566 | } | |
567 | ||
b7454ad3 GC |
568 | static void *s_start(struct seq_file *m, loff_t *pos) |
569 | { | |
570 | loff_t n = *pos; | |
571 | ||
572 | mutex_lock(&slab_mutex); | |
573 | if (!n) | |
574 | print_slabinfo_header(m); | |
575 | ||
576 | return seq_list_start(&slab_caches, *pos); | |
577 | } | |
578 | ||
276a2439 | 579 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 GC |
580 | { |
581 | return seq_list_next(p, &slab_caches, pos); | |
582 | } | |
583 | ||
276a2439 | 584 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
585 | { |
586 | mutex_unlock(&slab_mutex); | |
587 | } | |
588 | ||
749c5415 GC |
589 | static void |
590 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
591 | { | |
592 | struct kmem_cache *c; | |
593 | struct slabinfo sinfo; | |
594 | int i; | |
595 | ||
596 | if (!is_root_cache(s)) | |
597 | return; | |
598 | ||
599 | for_each_memcg_cache_index(i) { | |
2ade4de8 | 600 | c = cache_from_memcg_idx(s, i); |
749c5415 GC |
601 | if (!c) |
602 | continue; | |
603 | ||
604 | memset(&sinfo, 0, sizeof(sinfo)); | |
605 | get_slabinfo(c, &sinfo); | |
606 | ||
607 | info->active_slabs += sinfo.active_slabs; | |
608 | info->num_slabs += sinfo.num_slabs; | |
609 | info->shared_avail += sinfo.shared_avail; | |
610 | info->active_objs += sinfo.active_objs; | |
611 | info->num_objs += sinfo.num_objs; | |
612 | } | |
613 | } | |
614 | ||
615 | int cache_show(struct kmem_cache *s, struct seq_file *m) | |
b7454ad3 | 616 | { |
0d7561c6 GC |
617 | struct slabinfo sinfo; |
618 | ||
619 | memset(&sinfo, 0, sizeof(sinfo)); | |
620 | get_slabinfo(s, &sinfo); | |
621 | ||
749c5415 GC |
622 | memcg_accumulate_slabinfo(s, &sinfo); |
623 | ||
0d7561c6 | 624 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 625 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
626 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
627 | ||
628 | seq_printf(m, " : tunables %4u %4u %4u", | |
629 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
630 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
631 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
632 | slabinfo_show_stats(m, s); | |
633 | seq_putc(m, '\n'); | |
634 | return 0; | |
b7454ad3 GC |
635 | } |
636 | ||
749c5415 GC |
637 | static int s_show(struct seq_file *m, void *p) |
638 | { | |
639 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
640 | ||
641 | if (!is_root_cache(s)) | |
642 | return 0; | |
643 | return cache_show(s, m); | |
644 | } | |
645 | ||
b7454ad3 GC |
646 | /* |
647 | * slabinfo_op - iterator that generates /proc/slabinfo | |
648 | * | |
649 | * Output layout: | |
650 | * cache-name | |
651 | * num-active-objs | |
652 | * total-objs | |
653 | * object size | |
654 | * num-active-slabs | |
655 | * total-slabs | |
656 | * num-pages-per-slab | |
657 | * + further values on SMP and with statistics enabled | |
658 | */ | |
659 | static const struct seq_operations slabinfo_op = { | |
660 | .start = s_start, | |
276a2439 WL |
661 | .next = slab_next, |
662 | .stop = slab_stop, | |
b7454ad3 GC |
663 | .show = s_show, |
664 | }; | |
665 | ||
666 | static int slabinfo_open(struct inode *inode, struct file *file) | |
667 | { | |
668 | return seq_open(file, &slabinfo_op); | |
669 | } | |
670 | ||
671 | static const struct file_operations proc_slabinfo_operations = { | |
672 | .open = slabinfo_open, | |
673 | .read = seq_read, | |
674 | .write = slabinfo_write, | |
675 | .llseek = seq_lseek, | |
676 | .release = seq_release, | |
677 | }; | |
678 | ||
679 | static int __init slab_proc_init(void) | |
680 | { | |
e9b4db2b WL |
681 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
682 | &proc_slabinfo_operations); | |
b7454ad3 GC |
683 | return 0; |
684 | } | |
685 | module_init(slab_proc_init); | |
686 | #endif /* CONFIG_SLABINFO */ |