Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
8 | * (C) 2007 SGI, Christoph Lameter <clameter@sgi.com> | |
9 | */ | |
10 | ||
11 | #include <linux/mm.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/bit_spinlock.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/bitops.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/seq_file.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/cpuset.h> | |
20 | #include <linux/mempolicy.h> | |
21 | #include <linux/ctype.h> | |
22 | #include <linux/kallsyms.h> | |
23 | ||
24 | /* | |
25 | * Lock order: | |
26 | * 1. slab_lock(page) | |
27 | * 2. slab->list_lock | |
28 | * | |
29 | * The slab_lock protects operations on the object of a particular | |
30 | * slab and its metadata in the page struct. If the slab lock | |
31 | * has been taken then no allocations nor frees can be performed | |
32 | * on the objects in the slab nor can the slab be added or removed | |
33 | * from the partial or full lists since this would mean modifying | |
34 | * the page_struct of the slab. | |
35 | * | |
36 | * The list_lock protects the partial and full list on each node and | |
37 | * the partial slab counter. If taken then no new slabs may be added or | |
38 | * removed from the lists nor make the number of partial slabs be modified. | |
39 | * (Note that the total number of slabs is an atomic value that may be | |
40 | * modified without taking the list lock). | |
41 | * | |
42 | * The list_lock is a centralized lock and thus we avoid taking it as | |
43 | * much as possible. As long as SLUB does not have to handle partial | |
44 | * slabs, operations can continue without any centralized lock. F.e. | |
45 | * allocating a long series of objects that fill up slabs does not require | |
46 | * the list lock. | |
47 | * | |
48 | * The lock order is sometimes inverted when we are trying to get a slab | |
49 | * off a list. We take the list_lock and then look for a page on the list | |
50 | * to use. While we do that objects in the slabs may be freed. We can | |
51 | * only operate on the slab if we have also taken the slab_lock. So we use | |
52 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
53 | * can occur anymore and we can use the slab for allocations etc. If the | |
54 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
55 | * we must stay away from it for a while since we may cause a bouncing | |
56 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
57 | * If all pages are busy then we may allocate a new slab instead of reusing | |
58 | * a partial slab. A new slab has noone operating on it and thus there is | |
59 | * no danger of cacheline contention. | |
60 | * | |
61 | * Interrupts are disabled during allocation and deallocation in order to | |
62 | * make the slab allocator safe to use in the context of an irq. In addition | |
63 | * interrupts are disabled to ensure that the processor does not change | |
64 | * while handling per_cpu slabs, due to kernel preemption. | |
65 | * | |
66 | * SLUB assigns one slab for allocation to each processor. | |
67 | * Allocations only occur from these slabs called cpu slabs. | |
68 | * | |
672bba3a CL |
69 | * Slabs with free elements are kept on a partial list and during regular |
70 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 71 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
72 | * We track full slabs for debugging purposes though because otherwise we |
73 | * cannot scan all objects. | |
81819f0f CL |
74 | * |
75 | * Slabs are freed when they become empty. Teardown and setup is | |
76 | * minimal so we rely on the page allocators per cpu caches for | |
77 | * fast frees and allocs. | |
78 | * | |
79 | * Overloading of page flags that are otherwise used for LRU management. | |
80 | * | |
4b6f0750 CL |
81 | * PageActive The slab is frozen and exempt from list processing. |
82 | * This means that the slab is dedicated to a purpose | |
83 | * such as satisfying allocations for a specific | |
84 | * processor. Objects may be freed in the slab while | |
85 | * it is frozen but slab_free will then skip the usual | |
86 | * list operations. It is up to the processor holding | |
87 | * the slab to integrate the slab into the slab lists | |
88 | * when the slab is no longer needed. | |
89 | * | |
90 | * One use of this flag is to mark slabs that are | |
91 | * used for allocations. Then such a slab becomes a cpu | |
92 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 93 | * freelist that allows lockless access to |
894b8788 CL |
94 | * free objects in addition to the regular freelist |
95 | * that requires the slab lock. | |
81819f0f CL |
96 | * |
97 | * PageError Slab requires special handling due to debug | |
98 | * options set. This moves slab handling out of | |
894b8788 | 99 | * the fast path and disables lockless freelists. |
81819f0f CL |
100 | */ |
101 | ||
5577bd8a CL |
102 | #define FROZEN (1 << PG_active) |
103 | ||
104 | #ifdef CONFIG_SLUB_DEBUG | |
105 | #define SLABDEBUG (1 << PG_error) | |
106 | #else | |
107 | #define SLABDEBUG 0 | |
108 | #endif | |
109 | ||
4b6f0750 CL |
110 | static inline int SlabFrozen(struct page *page) |
111 | { | |
5577bd8a | 112 | return page->flags & FROZEN; |
4b6f0750 CL |
113 | } |
114 | ||
115 | static inline void SetSlabFrozen(struct page *page) | |
116 | { | |
5577bd8a | 117 | page->flags |= FROZEN; |
4b6f0750 CL |
118 | } |
119 | ||
120 | static inline void ClearSlabFrozen(struct page *page) | |
121 | { | |
5577bd8a | 122 | page->flags &= ~FROZEN; |
4b6f0750 CL |
123 | } |
124 | ||
35e5d7ee CL |
125 | static inline int SlabDebug(struct page *page) |
126 | { | |
5577bd8a | 127 | return page->flags & SLABDEBUG; |
35e5d7ee CL |
128 | } |
129 | ||
130 | static inline void SetSlabDebug(struct page *page) | |
131 | { | |
5577bd8a | 132 | page->flags |= SLABDEBUG; |
35e5d7ee CL |
133 | } |
134 | ||
135 | static inline void ClearSlabDebug(struct page *page) | |
136 | { | |
5577bd8a | 137 | page->flags &= ~SLABDEBUG; |
35e5d7ee CL |
138 | } |
139 | ||
81819f0f CL |
140 | /* |
141 | * Issues still to be resolved: | |
142 | * | |
81819f0f CL |
143 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
144 | * | |
81819f0f CL |
145 | * - Variable sizing of the per node arrays |
146 | */ | |
147 | ||
148 | /* Enable to test recovery from slab corruption on boot */ | |
149 | #undef SLUB_RESILIENCY_TEST | |
150 | ||
151 | #if PAGE_SHIFT <= 12 | |
152 | ||
153 | /* | |
154 | * Small page size. Make sure that we do not fragment memory | |
155 | */ | |
156 | #define DEFAULT_MAX_ORDER 1 | |
157 | #define DEFAULT_MIN_OBJECTS 4 | |
158 | ||
159 | #else | |
160 | ||
161 | /* | |
162 | * Large page machines are customarily able to handle larger | |
163 | * page orders. | |
164 | */ | |
165 | #define DEFAULT_MAX_ORDER 2 | |
166 | #define DEFAULT_MIN_OBJECTS 8 | |
167 | ||
168 | #endif | |
169 | ||
2086d26a CL |
170 | /* |
171 | * Mininum number of partial slabs. These will be left on the partial | |
172 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
173 | */ | |
e95eed57 CL |
174 | #define MIN_PARTIAL 2 |
175 | ||
2086d26a CL |
176 | /* |
177 | * Maximum number of desirable partial slabs. | |
178 | * The existence of more partial slabs makes kmem_cache_shrink | |
179 | * sort the partial list by the number of objects in the. | |
180 | */ | |
181 | #define MAX_PARTIAL 10 | |
182 | ||
81819f0f CL |
183 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
184 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 185 | |
81819f0f CL |
186 | /* |
187 | * Set of flags that will prevent slab merging | |
188 | */ | |
189 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
190 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
191 | ||
192 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
193 | SLAB_CACHE_DMA) | |
194 | ||
195 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 196 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
197 | #endif |
198 | ||
199 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 200 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
201 | #endif |
202 | ||
203 | /* Internal SLUB flags */ | |
1ceef402 CL |
204 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
205 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f | 206 | |
65c02d4c CL |
207 | /* Not all arches define cache_line_size */ |
208 | #ifndef cache_line_size | |
209 | #define cache_line_size() L1_CACHE_BYTES | |
210 | #endif | |
211 | ||
81819f0f CL |
212 | static int kmem_size = sizeof(struct kmem_cache); |
213 | ||
214 | #ifdef CONFIG_SMP | |
215 | static struct notifier_block slab_notifier; | |
216 | #endif | |
217 | ||
218 | static enum { | |
219 | DOWN, /* No slab functionality available */ | |
220 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 221 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
222 | SYSFS /* Sysfs up */ |
223 | } slab_state = DOWN; | |
224 | ||
225 | /* A list of all slab caches on the system */ | |
226 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 227 | static LIST_HEAD(slab_caches); |
81819f0f | 228 | |
02cbc874 CL |
229 | /* |
230 | * Tracking user of a slab. | |
231 | */ | |
232 | struct track { | |
233 | void *addr; /* Called from address */ | |
234 | int cpu; /* Was running on cpu */ | |
235 | int pid; /* Pid context */ | |
236 | unsigned long when; /* When did the operation occur */ | |
237 | }; | |
238 | ||
239 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
240 | ||
41ecc55b | 241 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
242 | static int sysfs_slab_add(struct kmem_cache *); |
243 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
244 | static void sysfs_slab_remove(struct kmem_cache *); | |
245 | #else | |
0c710013 CL |
246 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
247 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
248 | { return 0; } | |
249 | static inline void sysfs_slab_remove(struct kmem_cache *s) {} | |
81819f0f CL |
250 | #endif |
251 | ||
252 | /******************************************************************** | |
253 | * Core slab cache functions | |
254 | *******************************************************************/ | |
255 | ||
256 | int slab_is_available(void) | |
257 | { | |
258 | return slab_state >= UP; | |
259 | } | |
260 | ||
261 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
262 | { | |
263 | #ifdef CONFIG_NUMA | |
264 | return s->node[node]; | |
265 | #else | |
266 | return &s->local_node; | |
267 | #endif | |
268 | } | |
269 | ||
dfb4f096 CL |
270 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
271 | { | |
4c93c355 CL |
272 | #ifdef CONFIG_SMP |
273 | return s->cpu_slab[cpu]; | |
274 | #else | |
275 | return &s->cpu_slab; | |
276 | #endif | |
dfb4f096 CL |
277 | } |
278 | ||
02cbc874 CL |
279 | static inline int check_valid_pointer(struct kmem_cache *s, |
280 | struct page *page, const void *object) | |
281 | { | |
282 | void *base; | |
283 | ||
284 | if (!object) | |
285 | return 1; | |
286 | ||
287 | base = page_address(page); | |
288 | if (object < base || object >= base + s->objects * s->size || | |
289 | (object - base) % s->size) { | |
290 | return 0; | |
291 | } | |
292 | ||
293 | return 1; | |
294 | } | |
295 | ||
7656c72b CL |
296 | /* |
297 | * Slow version of get and set free pointer. | |
298 | * | |
299 | * This version requires touching the cache lines of kmem_cache which | |
300 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
301 | * from the page struct. | |
302 | */ | |
303 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
304 | { | |
305 | return *(void **)(object + s->offset); | |
306 | } | |
307 | ||
308 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
309 | { | |
310 | *(void **)(object + s->offset) = fp; | |
311 | } | |
312 | ||
313 | /* Loop over all objects in a slab */ | |
314 | #define for_each_object(__p, __s, __addr) \ | |
315 | for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\ | |
316 | __p += (__s)->size) | |
317 | ||
318 | /* Scan freelist */ | |
319 | #define for_each_free_object(__p, __s, __free) \ | |
320 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) | |
321 | ||
322 | /* Determine object index from a given position */ | |
323 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
324 | { | |
325 | return (p - addr) / s->size; | |
326 | } | |
327 | ||
41ecc55b CL |
328 | #ifdef CONFIG_SLUB_DEBUG |
329 | /* | |
330 | * Debug settings: | |
331 | */ | |
f0630fff CL |
332 | #ifdef CONFIG_SLUB_DEBUG_ON |
333 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
334 | #else | |
41ecc55b | 335 | static int slub_debug; |
f0630fff | 336 | #endif |
41ecc55b CL |
337 | |
338 | static char *slub_debug_slabs; | |
339 | ||
81819f0f CL |
340 | /* |
341 | * Object debugging | |
342 | */ | |
343 | static void print_section(char *text, u8 *addr, unsigned int length) | |
344 | { | |
345 | int i, offset; | |
346 | int newline = 1; | |
347 | char ascii[17]; | |
348 | ||
349 | ascii[16] = 0; | |
350 | ||
351 | for (i = 0; i < length; i++) { | |
352 | if (newline) { | |
24922684 | 353 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
354 | newline = 0; |
355 | } | |
356 | printk(" %02x", addr[i]); | |
357 | offset = i % 16; | |
358 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
359 | if (offset == 15) { | |
360 | printk(" %s\n",ascii); | |
361 | newline = 1; | |
362 | } | |
363 | } | |
364 | if (!newline) { | |
365 | i %= 16; | |
366 | while (i < 16) { | |
367 | printk(" "); | |
368 | ascii[i] = ' '; | |
369 | i++; | |
370 | } | |
371 | printk(" %s\n", ascii); | |
372 | } | |
373 | } | |
374 | ||
81819f0f CL |
375 | static struct track *get_track(struct kmem_cache *s, void *object, |
376 | enum track_item alloc) | |
377 | { | |
378 | struct track *p; | |
379 | ||
380 | if (s->offset) | |
381 | p = object + s->offset + sizeof(void *); | |
382 | else | |
383 | p = object + s->inuse; | |
384 | ||
385 | return p + alloc; | |
386 | } | |
387 | ||
388 | static void set_track(struct kmem_cache *s, void *object, | |
389 | enum track_item alloc, void *addr) | |
390 | { | |
391 | struct track *p; | |
392 | ||
393 | if (s->offset) | |
394 | p = object + s->offset + sizeof(void *); | |
395 | else | |
396 | p = object + s->inuse; | |
397 | ||
398 | p += alloc; | |
399 | if (addr) { | |
400 | p->addr = addr; | |
401 | p->cpu = smp_processor_id(); | |
402 | p->pid = current ? current->pid : -1; | |
403 | p->when = jiffies; | |
404 | } else | |
405 | memset(p, 0, sizeof(struct track)); | |
406 | } | |
407 | ||
81819f0f CL |
408 | static void init_tracking(struct kmem_cache *s, void *object) |
409 | { | |
24922684 CL |
410 | if (!(s->flags & SLAB_STORE_USER)) |
411 | return; | |
412 | ||
413 | set_track(s, object, TRACK_FREE, NULL); | |
414 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
415 | } |
416 | ||
417 | static void print_track(const char *s, struct track *t) | |
418 | { | |
419 | if (!t->addr) | |
420 | return; | |
421 | ||
24922684 | 422 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 423 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
424 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
425 | } | |
426 | ||
427 | static void print_tracking(struct kmem_cache *s, void *object) | |
428 | { | |
429 | if (!(s->flags & SLAB_STORE_USER)) | |
430 | return; | |
431 | ||
432 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
433 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
434 | } | |
435 | ||
436 | static void print_page_info(struct page *page) | |
437 | { | |
438 | printk(KERN_ERR "INFO: Slab 0x%p used=%u fp=0x%p flags=0x%04lx\n", | |
439 | page, page->inuse, page->freelist, page->flags); | |
440 | ||
441 | } | |
442 | ||
443 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
444 | { | |
445 | va_list args; | |
446 | char buf[100]; | |
447 | ||
448 | va_start(args, fmt); | |
449 | vsnprintf(buf, sizeof(buf), fmt, args); | |
450 | va_end(args); | |
451 | printk(KERN_ERR "========================================" | |
452 | "=====================================\n"); | |
453 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
454 | printk(KERN_ERR "----------------------------------------" | |
455 | "-------------------------------------\n\n"); | |
81819f0f CL |
456 | } |
457 | ||
24922684 CL |
458 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
459 | { | |
460 | va_list args; | |
461 | char buf[100]; | |
462 | ||
463 | va_start(args, fmt); | |
464 | vsnprintf(buf, sizeof(buf), fmt, args); | |
465 | va_end(args); | |
466 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
467 | } | |
468 | ||
469 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
470 | { |
471 | unsigned int off; /* Offset of last byte */ | |
24922684 CL |
472 | u8 *addr = page_address(page); |
473 | ||
474 | print_tracking(s, p); | |
475 | ||
476 | print_page_info(page); | |
477 | ||
478 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
479 | p, p - addr, get_freepointer(s, p)); | |
480 | ||
481 | if (p > addr + 16) | |
482 | print_section("Bytes b4", p - 16, 16); | |
483 | ||
484 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
485 | |
486 | if (s->flags & SLAB_RED_ZONE) | |
487 | print_section("Redzone", p + s->objsize, | |
488 | s->inuse - s->objsize); | |
489 | ||
81819f0f CL |
490 | if (s->offset) |
491 | off = s->offset + sizeof(void *); | |
492 | else | |
493 | off = s->inuse; | |
494 | ||
24922684 | 495 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 496 | off += 2 * sizeof(struct track); |
81819f0f CL |
497 | |
498 | if (off != s->size) | |
499 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
500 | print_section("Padding", p + off, s->size - off); |
501 | ||
502 | dump_stack(); | |
81819f0f CL |
503 | } |
504 | ||
505 | static void object_err(struct kmem_cache *s, struct page *page, | |
506 | u8 *object, char *reason) | |
507 | { | |
24922684 CL |
508 | slab_bug(s, reason); |
509 | print_trailer(s, page, object); | |
81819f0f CL |
510 | } |
511 | ||
24922684 | 512 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
513 | { |
514 | va_list args; | |
515 | char buf[100]; | |
516 | ||
24922684 CL |
517 | va_start(args, fmt); |
518 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 519 | va_end(args); |
24922684 CL |
520 | slab_bug(s, fmt); |
521 | print_page_info(page); | |
81819f0f CL |
522 | dump_stack(); |
523 | } | |
524 | ||
525 | static void init_object(struct kmem_cache *s, void *object, int active) | |
526 | { | |
527 | u8 *p = object; | |
528 | ||
529 | if (s->flags & __OBJECT_POISON) { | |
530 | memset(p, POISON_FREE, s->objsize - 1); | |
531 | p[s->objsize -1] = POISON_END; | |
532 | } | |
533 | ||
534 | if (s->flags & SLAB_RED_ZONE) | |
535 | memset(p + s->objsize, | |
536 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
537 | s->inuse - s->objsize); | |
538 | } | |
539 | ||
24922684 | 540 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
541 | { |
542 | while (bytes) { | |
543 | if (*start != (u8)value) | |
24922684 | 544 | return start; |
81819f0f CL |
545 | start++; |
546 | bytes--; | |
547 | } | |
24922684 CL |
548 | return NULL; |
549 | } | |
550 | ||
551 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
552 | void *from, void *to) | |
553 | { | |
554 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
555 | memset(from, data, to - from); | |
556 | } | |
557 | ||
558 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
559 | u8 *object, char *what, | |
560 | u8* start, unsigned int value, unsigned int bytes) | |
561 | { | |
562 | u8 *fault; | |
563 | u8 *end; | |
564 | ||
565 | fault = check_bytes(start, value, bytes); | |
566 | if (!fault) | |
567 | return 1; | |
568 | ||
569 | end = start + bytes; | |
570 | while (end > fault && end[-1] == value) | |
571 | end--; | |
572 | ||
573 | slab_bug(s, "%s overwritten", what); | |
574 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
575 | fault, end - 1, fault[0], value); | |
576 | print_trailer(s, page, object); | |
577 | ||
578 | restore_bytes(s, what, value, fault, end); | |
579 | return 0; | |
81819f0f CL |
580 | } |
581 | ||
81819f0f CL |
582 | /* |
583 | * Object layout: | |
584 | * | |
585 | * object address | |
586 | * Bytes of the object to be managed. | |
587 | * If the freepointer may overlay the object then the free | |
588 | * pointer is the first word of the object. | |
672bba3a | 589 | * |
81819f0f CL |
590 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
591 | * 0xa5 (POISON_END) | |
592 | * | |
593 | * object + s->objsize | |
594 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
595 | * Padding is extended by another word if Redzoning is enabled and |
596 | * objsize == inuse. | |
597 | * | |
81819f0f CL |
598 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
599 | * 0xcc (RED_ACTIVE) for objects in use. | |
600 | * | |
601 | * object + s->inuse | |
672bba3a CL |
602 | * Meta data starts here. |
603 | * | |
81819f0f CL |
604 | * A. Free pointer (if we cannot overwrite object on free) |
605 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a CL |
606 | * C. Padding to reach required alignment boundary or at mininum |
607 | * one word if debuggin is on to be able to detect writes | |
608 | * before the word boundary. | |
609 | * | |
610 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
611 | * |
612 | * object + s->size | |
672bba3a | 613 | * Nothing is used beyond s->size. |
81819f0f | 614 | * |
672bba3a CL |
615 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
616 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
617 | * may be used with merged slabcaches. |
618 | */ | |
619 | ||
81819f0f CL |
620 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
621 | { | |
622 | unsigned long off = s->inuse; /* The end of info */ | |
623 | ||
624 | if (s->offset) | |
625 | /* Freepointer is placed after the object. */ | |
626 | off += sizeof(void *); | |
627 | ||
628 | if (s->flags & SLAB_STORE_USER) | |
629 | /* We also have user information there */ | |
630 | off += 2 * sizeof(struct track); | |
631 | ||
632 | if (s->size == off) | |
633 | return 1; | |
634 | ||
24922684 CL |
635 | return check_bytes_and_report(s, page, p, "Object padding", |
636 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
637 | } |
638 | ||
639 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
640 | { | |
24922684 CL |
641 | u8 *start; |
642 | u8 *fault; | |
643 | u8 *end; | |
644 | int length; | |
645 | int remainder; | |
81819f0f CL |
646 | |
647 | if (!(s->flags & SLAB_POISON)) | |
648 | return 1; | |
649 | ||
24922684 CL |
650 | start = page_address(page); |
651 | end = start + (PAGE_SIZE << s->order); | |
81819f0f | 652 | length = s->objects * s->size; |
24922684 | 653 | remainder = end - (start + length); |
81819f0f CL |
654 | if (!remainder) |
655 | return 1; | |
656 | ||
24922684 CL |
657 | fault = check_bytes(start + length, POISON_INUSE, remainder); |
658 | if (!fault) | |
659 | return 1; | |
660 | while (end > fault && end[-1] == POISON_INUSE) | |
661 | end--; | |
662 | ||
663 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
664 | print_section("Padding", start, length); | |
665 | ||
666 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
667 | return 0; | |
81819f0f CL |
668 | } |
669 | ||
670 | static int check_object(struct kmem_cache *s, struct page *page, | |
671 | void *object, int active) | |
672 | { | |
673 | u8 *p = object; | |
674 | u8 *endobject = object + s->objsize; | |
675 | ||
676 | if (s->flags & SLAB_RED_ZONE) { | |
677 | unsigned int red = | |
678 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
679 | ||
24922684 CL |
680 | if (!check_bytes_and_report(s, page, object, "Redzone", |
681 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 682 | return 0; |
81819f0f | 683 | } else { |
24922684 CL |
684 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) |
685 | check_bytes_and_report(s, page, p, "Alignment padding", endobject, | |
686 | POISON_INUSE, s->inuse - s->objsize); | |
81819f0f CL |
687 | } |
688 | ||
689 | if (s->flags & SLAB_POISON) { | |
690 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
691 | (!check_bytes_and_report(s, page, p, "Poison", p, |
692 | POISON_FREE, s->objsize - 1) || | |
693 | !check_bytes_and_report(s, page, p, "Poison", | |
694 | p + s->objsize -1, POISON_END, 1))) | |
81819f0f | 695 | return 0; |
81819f0f CL |
696 | /* |
697 | * check_pad_bytes cleans up on its own. | |
698 | */ | |
699 | check_pad_bytes(s, page, p); | |
700 | } | |
701 | ||
702 | if (!s->offset && active) | |
703 | /* | |
704 | * Object and freepointer overlap. Cannot check | |
705 | * freepointer while object is allocated. | |
706 | */ | |
707 | return 1; | |
708 | ||
709 | /* Check free pointer validity */ | |
710 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
711 | object_err(s, page, p, "Freepointer corrupt"); | |
712 | /* | |
713 | * No choice but to zap it and thus loose the remainder | |
714 | * of the free objects in this slab. May cause | |
672bba3a | 715 | * another error because the object count is now wrong. |
81819f0f CL |
716 | */ |
717 | set_freepointer(s, p, NULL); | |
718 | return 0; | |
719 | } | |
720 | return 1; | |
721 | } | |
722 | ||
723 | static int check_slab(struct kmem_cache *s, struct page *page) | |
724 | { | |
725 | VM_BUG_ON(!irqs_disabled()); | |
726 | ||
727 | if (!PageSlab(page)) { | |
24922684 | 728 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
729 | return 0; |
730 | } | |
81819f0f | 731 | if (page->inuse > s->objects) { |
24922684 CL |
732 | slab_err(s, page, "inuse %u > max %u", |
733 | s->name, page->inuse, s->objects); | |
81819f0f CL |
734 | return 0; |
735 | } | |
736 | /* Slab_pad_check fixes things up after itself */ | |
737 | slab_pad_check(s, page); | |
738 | return 1; | |
739 | } | |
740 | ||
741 | /* | |
672bba3a CL |
742 | * Determine if a certain object on a page is on the freelist. Must hold the |
743 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
744 | */ |
745 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
746 | { | |
747 | int nr = 0; | |
748 | void *fp = page->freelist; | |
749 | void *object = NULL; | |
750 | ||
751 | while (fp && nr <= s->objects) { | |
752 | if (fp == search) | |
753 | return 1; | |
754 | if (!check_valid_pointer(s, page, fp)) { | |
755 | if (object) { | |
756 | object_err(s, page, object, | |
757 | "Freechain corrupt"); | |
758 | set_freepointer(s, object, NULL); | |
759 | break; | |
760 | } else { | |
24922684 | 761 | slab_err(s, page, "Freepointer corrupt"); |
81819f0f CL |
762 | page->freelist = NULL; |
763 | page->inuse = s->objects; | |
24922684 | 764 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
765 | return 0; |
766 | } | |
767 | break; | |
768 | } | |
769 | object = fp; | |
770 | fp = get_freepointer(s, object); | |
771 | nr++; | |
772 | } | |
773 | ||
774 | if (page->inuse != s->objects - nr) { | |
70d71228 | 775 | slab_err(s, page, "Wrong object count. Counter is %d but " |
24922684 | 776 | "counted were %d", page->inuse, s->objects - nr); |
81819f0f | 777 | page->inuse = s->objects - nr; |
24922684 | 778 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
779 | } |
780 | return search == NULL; | |
781 | } | |
782 | ||
3ec09742 CL |
783 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
784 | { | |
785 | if (s->flags & SLAB_TRACE) { | |
786 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
787 | s->name, | |
788 | alloc ? "alloc" : "free", | |
789 | object, page->inuse, | |
790 | page->freelist); | |
791 | ||
792 | if (!alloc) | |
793 | print_section("Object", (void *)object, s->objsize); | |
794 | ||
795 | dump_stack(); | |
796 | } | |
797 | } | |
798 | ||
643b1138 | 799 | /* |
672bba3a | 800 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 801 | */ |
e95eed57 | 802 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 803 | { |
643b1138 CL |
804 | spin_lock(&n->list_lock); |
805 | list_add(&page->lru, &n->full); | |
806 | spin_unlock(&n->list_lock); | |
807 | } | |
808 | ||
809 | static void remove_full(struct kmem_cache *s, struct page *page) | |
810 | { | |
811 | struct kmem_cache_node *n; | |
812 | ||
813 | if (!(s->flags & SLAB_STORE_USER)) | |
814 | return; | |
815 | ||
816 | n = get_node(s, page_to_nid(page)); | |
817 | ||
818 | spin_lock(&n->list_lock); | |
819 | list_del(&page->lru); | |
820 | spin_unlock(&n->list_lock); | |
821 | } | |
822 | ||
3ec09742 CL |
823 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
824 | void *object) | |
825 | { | |
826 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
827 | return; | |
828 | ||
829 | init_object(s, object, 0); | |
830 | init_tracking(s, object); | |
831 | } | |
832 | ||
833 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
834 | void *object, void *addr) | |
81819f0f CL |
835 | { |
836 | if (!check_slab(s, page)) | |
837 | goto bad; | |
838 | ||
839 | if (object && !on_freelist(s, page, object)) { | |
24922684 | 840 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 841 | goto bad; |
81819f0f CL |
842 | } |
843 | ||
844 | if (!check_valid_pointer(s, page, object)) { | |
845 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 846 | goto bad; |
81819f0f CL |
847 | } |
848 | ||
3ec09742 | 849 | if (object && !check_object(s, page, object, 0)) |
81819f0f | 850 | goto bad; |
81819f0f | 851 | |
3ec09742 CL |
852 | /* Success perform special debug activities for allocs */ |
853 | if (s->flags & SLAB_STORE_USER) | |
854 | set_track(s, object, TRACK_ALLOC, addr); | |
855 | trace(s, page, object, 1); | |
856 | init_object(s, object, 1); | |
81819f0f | 857 | return 1; |
3ec09742 | 858 | |
81819f0f CL |
859 | bad: |
860 | if (PageSlab(page)) { | |
861 | /* | |
862 | * If this is a slab page then lets do the best we can | |
863 | * to avoid issues in the future. Marking all objects | |
672bba3a | 864 | * as used avoids touching the remaining objects. |
81819f0f | 865 | */ |
24922684 | 866 | slab_fix(s, "Marking all objects used"); |
81819f0f CL |
867 | page->inuse = s->objects; |
868 | page->freelist = NULL; | |
81819f0f CL |
869 | } |
870 | return 0; | |
871 | } | |
872 | ||
3ec09742 CL |
873 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
874 | void *object, void *addr) | |
81819f0f CL |
875 | { |
876 | if (!check_slab(s, page)) | |
877 | goto fail; | |
878 | ||
879 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 880 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
881 | goto fail; |
882 | } | |
883 | ||
884 | if (on_freelist(s, page, object)) { | |
24922684 | 885 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
886 | goto fail; |
887 | } | |
888 | ||
889 | if (!check_object(s, page, object, 1)) | |
890 | return 0; | |
891 | ||
892 | if (unlikely(s != page->slab)) { | |
893 | if (!PageSlab(page)) | |
70d71228 CL |
894 | slab_err(s, page, "Attempt to free object(0x%p) " |
895 | "outside of slab", object); | |
81819f0f | 896 | else |
70d71228 | 897 | if (!page->slab) { |
81819f0f | 898 | printk(KERN_ERR |
70d71228 | 899 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 900 | object); |
70d71228 CL |
901 | dump_stack(); |
902 | } | |
81819f0f | 903 | else |
24922684 CL |
904 | object_err(s, page, object, |
905 | "page slab pointer corrupt."); | |
81819f0f CL |
906 | goto fail; |
907 | } | |
3ec09742 CL |
908 | |
909 | /* Special debug activities for freeing objects */ | |
910 | if (!SlabFrozen(page) && !page->freelist) | |
911 | remove_full(s, page); | |
912 | if (s->flags & SLAB_STORE_USER) | |
913 | set_track(s, object, TRACK_FREE, addr); | |
914 | trace(s, page, object, 0); | |
915 | init_object(s, object, 0); | |
81819f0f | 916 | return 1; |
3ec09742 | 917 | |
81819f0f | 918 | fail: |
24922684 | 919 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
920 | return 0; |
921 | } | |
922 | ||
41ecc55b CL |
923 | static int __init setup_slub_debug(char *str) |
924 | { | |
f0630fff CL |
925 | slub_debug = DEBUG_DEFAULT_FLAGS; |
926 | if (*str++ != '=' || !*str) | |
927 | /* | |
928 | * No options specified. Switch on full debugging. | |
929 | */ | |
930 | goto out; | |
931 | ||
932 | if (*str == ',') | |
933 | /* | |
934 | * No options but restriction on slabs. This means full | |
935 | * debugging for slabs matching a pattern. | |
936 | */ | |
937 | goto check_slabs; | |
938 | ||
939 | slub_debug = 0; | |
940 | if (*str == '-') | |
941 | /* | |
942 | * Switch off all debugging measures. | |
943 | */ | |
944 | goto out; | |
945 | ||
946 | /* | |
947 | * Determine which debug features should be switched on | |
948 | */ | |
949 | for ( ;*str && *str != ','; str++) { | |
950 | switch (tolower(*str)) { | |
951 | case 'f': | |
952 | slub_debug |= SLAB_DEBUG_FREE; | |
953 | break; | |
954 | case 'z': | |
955 | slub_debug |= SLAB_RED_ZONE; | |
956 | break; | |
957 | case 'p': | |
958 | slub_debug |= SLAB_POISON; | |
959 | break; | |
960 | case 'u': | |
961 | slub_debug |= SLAB_STORE_USER; | |
962 | break; | |
963 | case 't': | |
964 | slub_debug |= SLAB_TRACE; | |
965 | break; | |
966 | default: | |
967 | printk(KERN_ERR "slub_debug option '%c' " | |
968 | "unknown. skipped\n",*str); | |
969 | } | |
41ecc55b CL |
970 | } |
971 | ||
f0630fff | 972 | check_slabs: |
41ecc55b CL |
973 | if (*str == ',') |
974 | slub_debug_slabs = str + 1; | |
f0630fff | 975 | out: |
41ecc55b CL |
976 | return 1; |
977 | } | |
978 | ||
979 | __setup("slub_debug", setup_slub_debug); | |
980 | ||
ba0268a8 CL |
981 | static unsigned long kmem_cache_flags(unsigned long objsize, |
982 | unsigned long flags, const char *name, | |
983 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) | |
41ecc55b CL |
984 | { |
985 | /* | |
986 | * The page->offset field is only 16 bit wide. This is an offset | |
987 | * in units of words from the beginning of an object. If the slab | |
988 | * size is bigger then we cannot move the free pointer behind the | |
989 | * object anymore. | |
990 | * | |
991 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
992 | * the limit is 512k. | |
993 | * | |
c59def9f | 994 | * Debugging or ctor may create a need to move the free |
41ecc55b CL |
995 | * pointer. Fail if this happens. |
996 | */ | |
ba0268a8 CL |
997 | if (objsize >= 65535 * sizeof(void *)) { |
998 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
41ecc55b | 999 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); |
ba0268a8 CL |
1000 | BUG_ON(ctor); |
1001 | } else { | |
41ecc55b CL |
1002 | /* |
1003 | * Enable debugging if selected on the kernel commandline. | |
1004 | */ | |
1005 | if (slub_debug && (!slub_debug_slabs || | |
ba0268a8 | 1006 | strncmp(slub_debug_slabs, name, |
41ecc55b | 1007 | strlen(slub_debug_slabs)) == 0)) |
ba0268a8 CL |
1008 | flags |= slub_debug; |
1009 | } | |
1010 | ||
1011 | return flags; | |
41ecc55b CL |
1012 | } |
1013 | #else | |
3ec09742 CL |
1014 | static inline void setup_object_debug(struct kmem_cache *s, |
1015 | struct page *page, void *object) {} | |
41ecc55b | 1016 | |
3ec09742 CL |
1017 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1018 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1019 | |
3ec09742 CL |
1020 | static inline int free_debug_processing(struct kmem_cache *s, |
1021 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1022 | |
41ecc55b CL |
1023 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1024 | { return 1; } | |
1025 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1026 | void *object, int active) { return 1; } | |
3ec09742 | 1027 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1028 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1029 | unsigned long flags, const char *name, | |
1030 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) | |
1031 | { | |
1032 | return flags; | |
1033 | } | |
41ecc55b CL |
1034 | #define slub_debug 0 |
1035 | #endif | |
81819f0f CL |
1036 | /* |
1037 | * Slab allocation and freeing | |
1038 | */ | |
1039 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1040 | { | |
1041 | struct page * page; | |
1042 | int pages = 1 << s->order; | |
1043 | ||
1044 | if (s->order) | |
1045 | flags |= __GFP_COMP; | |
1046 | ||
1047 | if (s->flags & SLAB_CACHE_DMA) | |
1048 | flags |= SLUB_DMA; | |
1049 | ||
e12ba74d MG |
1050 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
1051 | flags |= __GFP_RECLAIMABLE; | |
1052 | ||
81819f0f CL |
1053 | if (node == -1) |
1054 | page = alloc_pages(flags, s->order); | |
1055 | else | |
1056 | page = alloc_pages_node(node, flags, s->order); | |
1057 | ||
1058 | if (!page) | |
1059 | return NULL; | |
1060 | ||
1061 | mod_zone_page_state(page_zone(page), | |
1062 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1063 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1064 | pages); | |
1065 | ||
1066 | return page; | |
1067 | } | |
1068 | ||
1069 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1070 | void *object) | |
1071 | { | |
3ec09742 | 1072 | setup_object_debug(s, page, object); |
4f104934 | 1073 | if (unlikely(s->ctor)) |
a35afb83 | 1074 | s->ctor(object, s, 0); |
81819f0f CL |
1075 | } |
1076 | ||
1077 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1078 | { | |
1079 | struct page *page; | |
1080 | struct kmem_cache_node *n; | |
1081 | void *start; | |
1082 | void *end; | |
1083 | void *last; | |
1084 | void *p; | |
1085 | ||
6cb06229 | 1086 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f CL |
1087 | |
1088 | if (flags & __GFP_WAIT) | |
1089 | local_irq_enable(); | |
1090 | ||
6cb06229 CL |
1091 | page = allocate_slab(s, |
1092 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1093 | if (!page) |
1094 | goto out; | |
1095 | ||
1096 | n = get_node(s, page_to_nid(page)); | |
1097 | if (n) | |
1098 | atomic_long_inc(&n->nr_slabs); | |
81819f0f CL |
1099 | page->slab = s; |
1100 | page->flags |= 1 << PG_slab; | |
1101 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1102 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1103 | SetSlabDebug(page); |
81819f0f CL |
1104 | |
1105 | start = page_address(page); | |
1106 | end = start + s->objects * s->size; | |
1107 | ||
1108 | if (unlikely(s->flags & SLAB_POISON)) | |
1109 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
1110 | ||
1111 | last = start; | |
7656c72b | 1112 | for_each_object(p, s, start) { |
81819f0f CL |
1113 | setup_object(s, page, last); |
1114 | set_freepointer(s, last, p); | |
1115 | last = p; | |
1116 | } | |
1117 | setup_object(s, page, last); | |
1118 | set_freepointer(s, last, NULL); | |
1119 | ||
1120 | page->freelist = start; | |
1121 | page->inuse = 0; | |
1122 | out: | |
1123 | if (flags & __GFP_WAIT) | |
1124 | local_irq_disable(); | |
1125 | return page; | |
1126 | } | |
1127 | ||
1128 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1129 | { | |
1130 | int pages = 1 << s->order; | |
1131 | ||
c59def9f | 1132 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1133 | void *p; |
1134 | ||
1135 | slab_pad_check(s, page); | |
c59def9f | 1136 | for_each_object(p, s, page_address(page)) |
81819f0f | 1137 | check_object(s, page, p, 0); |
2208b764 | 1138 | ClearSlabDebug(page); |
81819f0f CL |
1139 | } |
1140 | ||
1141 | mod_zone_page_state(page_zone(page), | |
1142 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1143 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1144 | - pages); | |
1145 | ||
81819f0f CL |
1146 | __free_pages(page, s->order); |
1147 | } | |
1148 | ||
1149 | static void rcu_free_slab(struct rcu_head *h) | |
1150 | { | |
1151 | struct page *page; | |
1152 | ||
1153 | page = container_of((struct list_head *)h, struct page, lru); | |
1154 | __free_slab(page->slab, page); | |
1155 | } | |
1156 | ||
1157 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1158 | { | |
1159 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1160 | /* | |
1161 | * RCU free overloads the RCU head over the LRU | |
1162 | */ | |
1163 | struct rcu_head *head = (void *)&page->lru; | |
1164 | ||
1165 | call_rcu(head, rcu_free_slab); | |
1166 | } else | |
1167 | __free_slab(s, page); | |
1168 | } | |
1169 | ||
1170 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1171 | { | |
1172 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1173 | ||
1174 | atomic_long_dec(&n->nr_slabs); | |
1175 | reset_page_mapcount(page); | |
35e5d7ee | 1176 | __ClearPageSlab(page); |
81819f0f CL |
1177 | free_slab(s, page); |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * Per slab locking using the pagelock | |
1182 | */ | |
1183 | static __always_inline void slab_lock(struct page *page) | |
1184 | { | |
1185 | bit_spin_lock(PG_locked, &page->flags); | |
1186 | } | |
1187 | ||
1188 | static __always_inline void slab_unlock(struct page *page) | |
1189 | { | |
1190 | bit_spin_unlock(PG_locked, &page->flags); | |
1191 | } | |
1192 | ||
1193 | static __always_inline int slab_trylock(struct page *page) | |
1194 | { | |
1195 | int rc = 1; | |
1196 | ||
1197 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1198 | return rc; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Management of partially allocated slabs | |
1203 | */ | |
e95eed57 | 1204 | static void add_partial_tail(struct kmem_cache_node *n, struct page *page) |
81819f0f | 1205 | { |
e95eed57 CL |
1206 | spin_lock(&n->list_lock); |
1207 | n->nr_partial++; | |
1208 | list_add_tail(&page->lru, &n->partial); | |
1209 | spin_unlock(&n->list_lock); | |
1210 | } | |
81819f0f | 1211 | |
e95eed57 CL |
1212 | static void add_partial(struct kmem_cache_node *n, struct page *page) |
1213 | { | |
81819f0f CL |
1214 | spin_lock(&n->list_lock); |
1215 | n->nr_partial++; | |
1216 | list_add(&page->lru, &n->partial); | |
1217 | spin_unlock(&n->list_lock); | |
1218 | } | |
1219 | ||
1220 | static void remove_partial(struct kmem_cache *s, | |
1221 | struct page *page) | |
1222 | { | |
1223 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1224 | ||
1225 | spin_lock(&n->list_lock); | |
1226 | list_del(&page->lru); | |
1227 | n->nr_partial--; | |
1228 | spin_unlock(&n->list_lock); | |
1229 | } | |
1230 | ||
1231 | /* | |
672bba3a | 1232 | * Lock slab and remove from the partial list. |
81819f0f | 1233 | * |
672bba3a | 1234 | * Must hold list_lock. |
81819f0f | 1235 | */ |
4b6f0750 | 1236 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1237 | { |
1238 | if (slab_trylock(page)) { | |
1239 | list_del(&page->lru); | |
1240 | n->nr_partial--; | |
4b6f0750 | 1241 | SetSlabFrozen(page); |
81819f0f CL |
1242 | return 1; |
1243 | } | |
1244 | return 0; | |
1245 | } | |
1246 | ||
1247 | /* | |
672bba3a | 1248 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1249 | */ |
1250 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1251 | { | |
1252 | struct page *page; | |
1253 | ||
1254 | /* | |
1255 | * Racy check. If we mistakenly see no partial slabs then we | |
1256 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1257 | * partial slab and there is none available then get_partials() |
1258 | * will return NULL. | |
81819f0f CL |
1259 | */ |
1260 | if (!n || !n->nr_partial) | |
1261 | return NULL; | |
1262 | ||
1263 | spin_lock(&n->list_lock); | |
1264 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1265 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1266 | goto out; |
1267 | page = NULL; | |
1268 | out: | |
1269 | spin_unlock(&n->list_lock); | |
1270 | return page; | |
1271 | } | |
1272 | ||
1273 | /* | |
672bba3a | 1274 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1275 | */ |
1276 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1277 | { | |
1278 | #ifdef CONFIG_NUMA | |
1279 | struct zonelist *zonelist; | |
1280 | struct zone **z; | |
1281 | struct page *page; | |
1282 | ||
1283 | /* | |
672bba3a CL |
1284 | * The defrag ratio allows a configuration of the tradeoffs between |
1285 | * inter node defragmentation and node local allocations. A lower | |
1286 | * defrag_ratio increases the tendency to do local allocations | |
1287 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1288 | * |
672bba3a CL |
1289 | * If the defrag_ratio is set to 0 then kmalloc() always |
1290 | * returns node local objects. If the ratio is higher then kmalloc() | |
1291 | * may return off node objects because partial slabs are obtained | |
1292 | * from other nodes and filled up. | |
81819f0f CL |
1293 | * |
1294 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
672bba3a CL |
1295 | * defrag_ratio = 1000) then every (well almost) allocation will |
1296 | * first attempt to defrag slab caches on other nodes. This means | |
1297 | * scanning over all nodes to look for partial slabs which may be | |
1298 | * expensive if we do it every time we are trying to find a slab | |
1299 | * with available objects. | |
81819f0f CL |
1300 | */ |
1301 | if (!s->defrag_ratio || get_cycles() % 1024 > s->defrag_ratio) | |
1302 | return NULL; | |
1303 | ||
1304 | zonelist = &NODE_DATA(slab_node(current->mempolicy)) | |
1305 | ->node_zonelists[gfp_zone(flags)]; | |
1306 | for (z = zonelist->zones; *z; z++) { | |
1307 | struct kmem_cache_node *n; | |
1308 | ||
1309 | n = get_node(s, zone_to_nid(*z)); | |
1310 | ||
1311 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1312 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1313 | page = get_partial_node(n); |
1314 | if (page) | |
1315 | return page; | |
1316 | } | |
1317 | } | |
1318 | #endif | |
1319 | return NULL; | |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * Get a partial page, lock it and return it. | |
1324 | */ | |
1325 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1326 | { | |
1327 | struct page *page; | |
1328 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1329 | ||
1330 | page = get_partial_node(get_node(s, searchnode)); | |
1331 | if (page || (flags & __GFP_THISNODE)) | |
1332 | return page; | |
1333 | ||
1334 | return get_any_partial(s, flags); | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * Move a page back to the lists. | |
1339 | * | |
1340 | * Must be called with the slab lock held. | |
1341 | * | |
1342 | * On exit the slab lock will have been dropped. | |
1343 | */ | |
4b6f0750 | 1344 | static void unfreeze_slab(struct kmem_cache *s, struct page *page) |
81819f0f | 1345 | { |
e95eed57 CL |
1346 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1347 | ||
4b6f0750 | 1348 | ClearSlabFrozen(page); |
81819f0f | 1349 | if (page->inuse) { |
e95eed57 | 1350 | |
81819f0f | 1351 | if (page->freelist) |
e95eed57 | 1352 | add_partial(n, page); |
35e5d7ee | 1353 | else if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) |
e95eed57 | 1354 | add_full(n, page); |
81819f0f | 1355 | slab_unlock(page); |
e95eed57 | 1356 | |
81819f0f | 1357 | } else { |
e95eed57 CL |
1358 | if (n->nr_partial < MIN_PARTIAL) { |
1359 | /* | |
672bba3a CL |
1360 | * Adding an empty slab to the partial slabs in order |
1361 | * to avoid page allocator overhead. This slab needs | |
1362 | * to come after the other slabs with objects in | |
1363 | * order to fill them up. That way the size of the | |
1364 | * partial list stays small. kmem_cache_shrink can | |
1365 | * reclaim empty slabs from the partial list. | |
e95eed57 CL |
1366 | */ |
1367 | add_partial_tail(n, page); | |
1368 | slab_unlock(page); | |
1369 | } else { | |
1370 | slab_unlock(page); | |
1371 | discard_slab(s, page); | |
1372 | } | |
81819f0f CL |
1373 | } |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Remove the cpu slab | |
1378 | */ | |
dfb4f096 | 1379 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1380 | { |
dfb4f096 | 1381 | struct page *page = c->page; |
894b8788 CL |
1382 | /* |
1383 | * Merge cpu freelist into freelist. Typically we get here | |
1384 | * because both freelists are empty. So this is unlikely | |
1385 | * to occur. | |
1386 | */ | |
dfb4f096 | 1387 | while (unlikely(c->freelist)) { |
894b8788 CL |
1388 | void **object; |
1389 | ||
1390 | /* Retrieve object from cpu_freelist */ | |
dfb4f096 | 1391 | object = c->freelist; |
b3fba8da | 1392 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1393 | |
1394 | /* And put onto the regular freelist */ | |
b3fba8da | 1395 | object[c->offset] = page->freelist; |
894b8788 CL |
1396 | page->freelist = object; |
1397 | page->inuse--; | |
1398 | } | |
dfb4f096 | 1399 | c->page = NULL; |
4b6f0750 | 1400 | unfreeze_slab(s, page); |
81819f0f CL |
1401 | } |
1402 | ||
dfb4f096 | 1403 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1404 | { |
dfb4f096 CL |
1405 | slab_lock(c->page); |
1406 | deactivate_slab(s, c); | |
81819f0f CL |
1407 | } |
1408 | ||
1409 | /* | |
1410 | * Flush cpu slab. | |
1411 | * Called from IPI handler with interrupts disabled. | |
1412 | */ | |
0c710013 | 1413 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1414 | { |
dfb4f096 | 1415 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1416 | |
dfb4f096 CL |
1417 | if (likely(c && c->page)) |
1418 | flush_slab(s, c); | |
81819f0f CL |
1419 | } |
1420 | ||
1421 | static void flush_cpu_slab(void *d) | |
1422 | { | |
1423 | struct kmem_cache *s = d; | |
81819f0f | 1424 | |
dfb4f096 | 1425 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1426 | } |
1427 | ||
1428 | static void flush_all(struct kmem_cache *s) | |
1429 | { | |
1430 | #ifdef CONFIG_SMP | |
1431 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1432 | #else | |
1433 | unsigned long flags; | |
1434 | ||
1435 | local_irq_save(flags); | |
1436 | flush_cpu_slab(s); | |
1437 | local_irq_restore(flags); | |
1438 | #endif | |
1439 | } | |
1440 | ||
dfb4f096 CL |
1441 | /* |
1442 | * Check if the objects in a per cpu structure fit numa | |
1443 | * locality expectations. | |
1444 | */ | |
1445 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1446 | { | |
1447 | #ifdef CONFIG_NUMA | |
1448 | if (node != -1 && c->node != node) | |
1449 | return 0; | |
1450 | #endif | |
1451 | return 1; | |
1452 | } | |
1453 | ||
81819f0f | 1454 | /* |
894b8788 CL |
1455 | * Slow path. The lockless freelist is empty or we need to perform |
1456 | * debugging duties. | |
1457 | * | |
1458 | * Interrupts are disabled. | |
81819f0f | 1459 | * |
894b8788 CL |
1460 | * Processing is still very fast if new objects have been freed to the |
1461 | * regular freelist. In that case we simply take over the regular freelist | |
1462 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1463 | * |
894b8788 CL |
1464 | * If that is not working then we fall back to the partial lists. We take the |
1465 | * first element of the freelist as the object to allocate now and move the | |
1466 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1467 | * |
894b8788 CL |
1468 | * And if we were unable to get a new slab from the partial slab lists then |
1469 | * we need to allocate a new slab. This is slowest path since we may sleep. | |
81819f0f | 1470 | */ |
894b8788 | 1471 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1472 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1473 | { |
81819f0f | 1474 | void **object; |
dfb4f096 | 1475 | struct page *new; |
81819f0f | 1476 | |
dfb4f096 | 1477 | if (!c->page) |
81819f0f CL |
1478 | goto new_slab; |
1479 | ||
dfb4f096 CL |
1480 | slab_lock(c->page); |
1481 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1482 | goto another_slab; |
894b8788 | 1483 | load_freelist: |
dfb4f096 | 1484 | object = c->page->freelist; |
81819f0f CL |
1485 | if (unlikely(!object)) |
1486 | goto another_slab; | |
dfb4f096 | 1487 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1488 | goto debug; |
1489 | ||
dfb4f096 | 1490 | object = c->page->freelist; |
b3fba8da | 1491 | c->freelist = object[c->offset]; |
dfb4f096 CL |
1492 | c->page->inuse = s->objects; |
1493 | c->page->freelist = NULL; | |
1494 | c->node = page_to_nid(c->page); | |
1495 | slab_unlock(c->page); | |
81819f0f CL |
1496 | return object; |
1497 | ||
1498 | another_slab: | |
dfb4f096 | 1499 | deactivate_slab(s, c); |
81819f0f CL |
1500 | |
1501 | new_slab: | |
dfb4f096 CL |
1502 | new = get_partial(s, gfpflags, node); |
1503 | if (new) { | |
1504 | c->page = new; | |
894b8788 | 1505 | goto load_freelist; |
81819f0f CL |
1506 | } |
1507 | ||
dfb4f096 CL |
1508 | new = new_slab(s, gfpflags, node); |
1509 | if (new) { | |
1510 | c = get_cpu_slab(s, smp_processor_id()); | |
1511 | if (c->page) { | |
81819f0f | 1512 | /* |
672bba3a CL |
1513 | * Someone else populated the cpu_slab while we |
1514 | * enabled interrupts, or we have gotten scheduled | |
1515 | * on another cpu. The page may not be on the | |
1516 | * requested node even if __GFP_THISNODE was | |
1517 | * specified. So we need to recheck. | |
81819f0f | 1518 | */ |
dfb4f096 | 1519 | if (node_match(c, node)) { |
81819f0f CL |
1520 | /* |
1521 | * Current cpuslab is acceptable and we | |
1522 | * want the current one since its cache hot | |
1523 | */ | |
dfb4f096 CL |
1524 | discard_slab(s, new); |
1525 | slab_lock(c->page); | |
894b8788 | 1526 | goto load_freelist; |
81819f0f | 1527 | } |
672bba3a | 1528 | /* New slab does not fit our expectations */ |
dfb4f096 | 1529 | flush_slab(s, c); |
81819f0f | 1530 | } |
dfb4f096 CL |
1531 | slab_lock(new); |
1532 | SetSlabFrozen(new); | |
1533 | c->page = new; | |
4b6f0750 | 1534 | goto load_freelist; |
81819f0f | 1535 | } |
81819f0f CL |
1536 | return NULL; |
1537 | debug: | |
dfb4f096 CL |
1538 | object = c->page->freelist; |
1539 | if (!alloc_debug_processing(s, c->page, object, addr)) | |
81819f0f | 1540 | goto another_slab; |
894b8788 | 1541 | |
dfb4f096 | 1542 | c->page->inuse++; |
b3fba8da | 1543 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1544 | c->node = -1; |
dfb4f096 | 1545 | slab_unlock(c->page); |
894b8788 CL |
1546 | return object; |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1551 | * have the fastpath folded into their functions. So no function call | |
1552 | * overhead for requests that can be satisfied on the fastpath. | |
1553 | * | |
1554 | * The fastpath works by first checking if the lockless freelist can be used. | |
1555 | * If not then __slab_alloc is called for slow processing. | |
1556 | * | |
1557 | * Otherwise we can simply pick the next object from the lockless free list. | |
1558 | */ | |
1559 | static void __always_inline *slab_alloc(struct kmem_cache *s, | |
ce15fea8 | 1560 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1561 | { |
894b8788 CL |
1562 | void **object; |
1563 | unsigned long flags; | |
dfb4f096 | 1564 | struct kmem_cache_cpu *c; |
894b8788 CL |
1565 | |
1566 | local_irq_save(flags); | |
dfb4f096 | 1567 | c = get_cpu_slab(s, smp_processor_id()); |
ee3c72a1 | 1568 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1569 | |
dfb4f096 | 1570 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1571 | |
1572 | else { | |
dfb4f096 | 1573 | object = c->freelist; |
b3fba8da | 1574 | c->freelist = object[c->offset]; |
894b8788 CL |
1575 | } |
1576 | local_irq_restore(flags); | |
d07dbea4 CL |
1577 | |
1578 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
ce15fea8 | 1579 | memset(object, 0, s->objsize); |
d07dbea4 | 1580 | |
894b8788 | 1581 | return object; |
81819f0f CL |
1582 | } |
1583 | ||
1584 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1585 | { | |
ce15fea8 | 1586 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1587 | } |
1588 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1589 | ||
1590 | #ifdef CONFIG_NUMA | |
1591 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1592 | { | |
ce15fea8 | 1593 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1594 | } |
1595 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1596 | #endif | |
1597 | ||
1598 | /* | |
894b8788 CL |
1599 | * Slow patch handling. This may still be called frequently since objects |
1600 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1601 | * |
894b8788 CL |
1602 | * So we still attempt to reduce cache line usage. Just take the slab |
1603 | * lock and free the item. If there is no additional partial page | |
1604 | * handling required then we can return immediately. | |
81819f0f | 1605 | */ |
894b8788 | 1606 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1607 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1608 | { |
1609 | void *prior; | |
1610 | void **object = (void *)x; | |
81819f0f | 1611 | |
81819f0f CL |
1612 | slab_lock(page); |
1613 | ||
35e5d7ee | 1614 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1615 | goto debug; |
1616 | checks_ok: | |
b3fba8da | 1617 | prior = object[offset] = page->freelist; |
81819f0f CL |
1618 | page->freelist = object; |
1619 | page->inuse--; | |
1620 | ||
4b6f0750 | 1621 | if (unlikely(SlabFrozen(page))) |
81819f0f CL |
1622 | goto out_unlock; |
1623 | ||
1624 | if (unlikely(!page->inuse)) | |
1625 | goto slab_empty; | |
1626 | ||
1627 | /* | |
1628 | * Objects left in the slab. If it | |
1629 | * was not on the partial list before | |
1630 | * then add it. | |
1631 | */ | |
1632 | if (unlikely(!prior)) | |
e95eed57 | 1633 | add_partial(get_node(s, page_to_nid(page)), page); |
81819f0f CL |
1634 | |
1635 | out_unlock: | |
1636 | slab_unlock(page); | |
81819f0f CL |
1637 | return; |
1638 | ||
1639 | slab_empty: | |
1640 | if (prior) | |
1641 | /* | |
672bba3a | 1642 | * Slab still on the partial list. |
81819f0f CL |
1643 | */ |
1644 | remove_partial(s, page); | |
1645 | ||
1646 | slab_unlock(page); | |
1647 | discard_slab(s, page); | |
81819f0f CL |
1648 | return; |
1649 | ||
1650 | debug: | |
3ec09742 | 1651 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1652 | goto out_unlock; |
77c5e2d0 | 1653 | goto checks_ok; |
81819f0f CL |
1654 | } |
1655 | ||
894b8788 CL |
1656 | /* |
1657 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1658 | * can perform fastpath freeing without additional function calls. | |
1659 | * | |
1660 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1661 | * of this processor. This typically the case if we have just allocated | |
1662 | * the item before. | |
1663 | * | |
1664 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1665 | * with all sorts of special processing. | |
1666 | */ | |
1667 | static void __always_inline slab_free(struct kmem_cache *s, | |
1668 | struct page *page, void *x, void *addr) | |
1669 | { | |
1670 | void **object = (void *)x; | |
1671 | unsigned long flags; | |
dfb4f096 | 1672 | struct kmem_cache_cpu *c; |
894b8788 CL |
1673 | |
1674 | local_irq_save(flags); | |
02febdf7 | 1675 | debug_check_no_locks_freed(object, s->objsize); |
dfb4f096 | 1676 | c = get_cpu_slab(s, smp_processor_id()); |
ee3c72a1 | 1677 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1678 | object[c->offset] = c->freelist; |
dfb4f096 | 1679 | c->freelist = object; |
894b8788 | 1680 | } else |
b3fba8da | 1681 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1682 | |
1683 | local_irq_restore(flags); | |
1684 | } | |
1685 | ||
81819f0f CL |
1686 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1687 | { | |
77c5e2d0 | 1688 | struct page *page; |
81819f0f | 1689 | |
b49af68f | 1690 | page = virt_to_head_page(x); |
81819f0f | 1691 | |
77c5e2d0 | 1692 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1693 | } |
1694 | EXPORT_SYMBOL(kmem_cache_free); | |
1695 | ||
1696 | /* Figure out on which slab object the object resides */ | |
1697 | static struct page *get_object_page(const void *x) | |
1698 | { | |
b49af68f | 1699 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1700 | |
1701 | if (!PageSlab(page)) | |
1702 | return NULL; | |
1703 | ||
1704 | return page; | |
1705 | } | |
1706 | ||
1707 | /* | |
672bba3a CL |
1708 | * Object placement in a slab is made very easy because we always start at |
1709 | * offset 0. If we tune the size of the object to the alignment then we can | |
1710 | * get the required alignment by putting one properly sized object after | |
1711 | * another. | |
81819f0f CL |
1712 | * |
1713 | * Notice that the allocation order determines the sizes of the per cpu | |
1714 | * caches. Each processor has always one slab available for allocations. | |
1715 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1716 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1717 | * locking overhead. |
81819f0f CL |
1718 | */ |
1719 | ||
1720 | /* | |
1721 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1722 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1723 | * and increases the number of allocations possible without having to | |
1724 | * take the list_lock. | |
1725 | */ | |
1726 | static int slub_min_order; | |
1727 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1728 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1729 | ||
1730 | /* | |
1731 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1732 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1733 | */ |
1734 | static int slub_nomerge; | |
1735 | ||
81819f0f CL |
1736 | /* |
1737 | * Calculate the order of allocation given an slab object size. | |
1738 | * | |
672bba3a CL |
1739 | * The order of allocation has significant impact on performance and other |
1740 | * system components. Generally order 0 allocations should be preferred since | |
1741 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1742 | * be problematic to put into order 0 slabs because there may be too much | |
1743 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1744 | * would be wasted. | |
1745 | * | |
1746 | * In order to reach satisfactory performance we must ensure that a minimum | |
1747 | * number of objects is in one slab. Otherwise we may generate too much | |
1748 | * activity on the partial lists which requires taking the list_lock. This is | |
1749 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1750 | * |
672bba3a CL |
1751 | * slub_max_order specifies the order where we begin to stop considering the |
1752 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1753 | * we try to keep the page order as low as possible. So we accept more waste | |
1754 | * of space in favor of a small page order. | |
81819f0f | 1755 | * |
672bba3a CL |
1756 | * Higher order allocations also allow the placement of more objects in a |
1757 | * slab and thereby reduce object handling overhead. If the user has | |
1758 | * requested a higher mininum order then we start with that one instead of | |
1759 | * the smallest order which will fit the object. | |
81819f0f | 1760 | */ |
5e6d444e CL |
1761 | static inline int slab_order(int size, int min_objects, |
1762 | int max_order, int fract_leftover) | |
81819f0f CL |
1763 | { |
1764 | int order; | |
1765 | int rem; | |
6300ea75 | 1766 | int min_order = slub_min_order; |
81819f0f | 1767 | |
6300ea75 | 1768 | for (order = max(min_order, |
5e6d444e CL |
1769 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1770 | order <= max_order; order++) { | |
81819f0f | 1771 | |
5e6d444e | 1772 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1773 | |
5e6d444e | 1774 | if (slab_size < min_objects * size) |
81819f0f CL |
1775 | continue; |
1776 | ||
1777 | rem = slab_size % size; | |
1778 | ||
5e6d444e | 1779 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1780 | break; |
1781 | ||
1782 | } | |
672bba3a | 1783 | |
81819f0f CL |
1784 | return order; |
1785 | } | |
1786 | ||
5e6d444e CL |
1787 | static inline int calculate_order(int size) |
1788 | { | |
1789 | int order; | |
1790 | int min_objects; | |
1791 | int fraction; | |
1792 | ||
1793 | /* | |
1794 | * Attempt to find best configuration for a slab. This | |
1795 | * works by first attempting to generate a layout with | |
1796 | * the best configuration and backing off gradually. | |
1797 | * | |
1798 | * First we reduce the acceptable waste in a slab. Then | |
1799 | * we reduce the minimum objects required in a slab. | |
1800 | */ | |
1801 | min_objects = slub_min_objects; | |
1802 | while (min_objects > 1) { | |
1803 | fraction = 8; | |
1804 | while (fraction >= 4) { | |
1805 | order = slab_order(size, min_objects, | |
1806 | slub_max_order, fraction); | |
1807 | if (order <= slub_max_order) | |
1808 | return order; | |
1809 | fraction /= 2; | |
1810 | } | |
1811 | min_objects /= 2; | |
1812 | } | |
1813 | ||
1814 | /* | |
1815 | * We were unable to place multiple objects in a slab. Now | |
1816 | * lets see if we can place a single object there. | |
1817 | */ | |
1818 | order = slab_order(size, 1, slub_max_order, 1); | |
1819 | if (order <= slub_max_order) | |
1820 | return order; | |
1821 | ||
1822 | /* | |
1823 | * Doh this slab cannot be placed using slub_max_order. | |
1824 | */ | |
1825 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1826 | if (order <= MAX_ORDER) | |
1827 | return order; | |
1828 | return -ENOSYS; | |
1829 | } | |
1830 | ||
81819f0f | 1831 | /* |
672bba3a | 1832 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1833 | */ |
1834 | static unsigned long calculate_alignment(unsigned long flags, | |
1835 | unsigned long align, unsigned long size) | |
1836 | { | |
1837 | /* | |
1838 | * If the user wants hardware cache aligned objects then | |
1839 | * follow that suggestion if the object is sufficiently | |
1840 | * large. | |
1841 | * | |
1842 | * The hardware cache alignment cannot override the | |
1843 | * specified alignment though. If that is greater | |
1844 | * then use it. | |
1845 | */ | |
5af60839 | 1846 | if ((flags & SLAB_HWCACHE_ALIGN) && |
65c02d4c CL |
1847 | size > cache_line_size() / 2) |
1848 | return max_t(unsigned long, align, cache_line_size()); | |
81819f0f CL |
1849 | |
1850 | if (align < ARCH_SLAB_MINALIGN) | |
1851 | return ARCH_SLAB_MINALIGN; | |
1852 | ||
1853 | return ALIGN(align, sizeof(void *)); | |
1854 | } | |
1855 | ||
dfb4f096 CL |
1856 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1857 | struct kmem_cache_cpu *c) | |
1858 | { | |
1859 | c->page = NULL; | |
1860 | c->freelist = NULL; | |
b3fba8da | 1861 | c->offset = s->offset / sizeof(void *); |
dfb4f096 CL |
1862 | c->node = 0; |
1863 | } | |
1864 | ||
81819f0f CL |
1865 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
1866 | { | |
1867 | n->nr_partial = 0; | |
1868 | atomic_long_set(&n->nr_slabs, 0); | |
1869 | spin_lock_init(&n->list_lock); | |
1870 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1871 | #ifdef CONFIG_SLUB_DEBUG |
643b1138 | 1872 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1873 | #endif |
81819f0f CL |
1874 | } |
1875 | ||
4c93c355 CL |
1876 | #ifdef CONFIG_SMP |
1877 | /* | |
1878 | * Per cpu array for per cpu structures. | |
1879 | * | |
1880 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1881 | * close together meaning that it becomes possible that multiple per cpu | |
1882 | * structures are contained in one cacheline. This may be particularly | |
1883 | * beneficial for the kmalloc caches. | |
1884 | * | |
1885 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1886 | * likely able to get per cpu structures for all caches from the array defined | |
1887 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
1888 | * | |
1889 | * If the per cpu array is exhausted then fall back to kmalloc | |
1890 | * of individual cachelines. No sharing is possible then. | |
1891 | */ | |
1892 | #define NR_KMEM_CACHE_CPU 100 | |
1893 | ||
1894 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
1895 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
1896 | ||
1897 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
1898 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
1899 | ||
1900 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
1901 | int cpu, gfp_t flags) | |
1902 | { | |
1903 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
1904 | ||
1905 | if (c) | |
1906 | per_cpu(kmem_cache_cpu_free, cpu) = | |
1907 | (void *)c->freelist; | |
1908 | else { | |
1909 | /* Table overflow: So allocate ourselves */ | |
1910 | c = kmalloc_node( | |
1911 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
1912 | flags, cpu_to_node(cpu)); | |
1913 | if (!c) | |
1914 | return NULL; | |
1915 | } | |
1916 | ||
1917 | init_kmem_cache_cpu(s, c); | |
1918 | return c; | |
1919 | } | |
1920 | ||
1921 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
1922 | { | |
1923 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
1924 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
1925 | kfree(c); | |
1926 | return; | |
1927 | } | |
1928 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
1929 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
1930 | } | |
1931 | ||
1932 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
1933 | { | |
1934 | int cpu; | |
1935 | ||
1936 | for_each_online_cpu(cpu) { | |
1937 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
1938 | ||
1939 | if (c) { | |
1940 | s->cpu_slab[cpu] = NULL; | |
1941 | free_kmem_cache_cpu(c, cpu); | |
1942 | } | |
1943 | } | |
1944 | } | |
1945 | ||
1946 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
1947 | { | |
1948 | int cpu; | |
1949 | ||
1950 | for_each_online_cpu(cpu) { | |
1951 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
1952 | ||
1953 | if (c) | |
1954 | continue; | |
1955 | ||
1956 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
1957 | if (!c) { | |
1958 | free_kmem_cache_cpus(s); | |
1959 | return 0; | |
1960 | } | |
1961 | s->cpu_slab[cpu] = c; | |
1962 | } | |
1963 | return 1; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Initialize the per cpu array. | |
1968 | */ | |
1969 | static void init_alloc_cpu_cpu(int cpu) | |
1970 | { | |
1971 | int i; | |
1972 | ||
1973 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
1974 | return; | |
1975 | ||
1976 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
1977 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
1978 | ||
1979 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
1980 | } | |
1981 | ||
1982 | static void __init init_alloc_cpu(void) | |
1983 | { | |
1984 | int cpu; | |
1985 | ||
1986 | for_each_online_cpu(cpu) | |
1987 | init_alloc_cpu_cpu(cpu); | |
1988 | } | |
1989 | ||
1990 | #else | |
1991 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
1992 | static inline void init_alloc_cpu(void) {} | |
1993 | ||
1994 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
1995 | { | |
1996 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
1997 | return 1; | |
1998 | } | |
1999 | #endif | |
2000 | ||
81819f0f CL |
2001 | #ifdef CONFIG_NUMA |
2002 | /* | |
2003 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2004 | * slab on the node for this slabcache. There are no concurrent accesses | |
2005 | * possible. | |
2006 | * | |
2007 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2008 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2009 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2010 | */ |
1cd7daa5 AB |
2011 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2012 | int node) | |
81819f0f CL |
2013 | { |
2014 | struct page *page; | |
2015 | struct kmem_cache_node *n; | |
2016 | ||
2017 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2018 | ||
a2f92ee7 | 2019 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2020 | |
2021 | BUG_ON(!page); | |
a2f92ee7 CL |
2022 | if (page_to_nid(page) != node) { |
2023 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2024 | "node %d\n", node); | |
2025 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2026 | "in order to be able to continue\n"); | |
2027 | } | |
2028 | ||
81819f0f CL |
2029 | n = page->freelist; |
2030 | BUG_ON(!n); | |
2031 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2032 | page->inuse++; | |
2033 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2034 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2035 | init_object(kmalloc_caches, n, 1); |
2036 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2037 | #endif |
81819f0f CL |
2038 | init_kmem_cache_node(n); |
2039 | atomic_long_inc(&n->nr_slabs); | |
e95eed57 | 2040 | add_partial(n, page); |
dbc55faa CL |
2041 | |
2042 | /* | |
2043 | * new_slab() disables interupts. If we do not reenable interrupts here | |
2044 | * then bootup would continue with interrupts disabled. | |
2045 | */ | |
2046 | local_irq_enable(); | |
81819f0f CL |
2047 | return n; |
2048 | } | |
2049 | ||
2050 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2051 | { | |
2052 | int node; | |
2053 | ||
f64dc58c | 2054 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2055 | struct kmem_cache_node *n = s->node[node]; |
2056 | if (n && n != &s->local_node) | |
2057 | kmem_cache_free(kmalloc_caches, n); | |
2058 | s->node[node] = NULL; | |
2059 | } | |
2060 | } | |
2061 | ||
2062 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2063 | { | |
2064 | int node; | |
2065 | int local_node; | |
2066 | ||
2067 | if (slab_state >= UP) | |
2068 | local_node = page_to_nid(virt_to_page(s)); | |
2069 | else | |
2070 | local_node = 0; | |
2071 | ||
f64dc58c | 2072 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2073 | struct kmem_cache_node *n; |
2074 | ||
2075 | if (local_node == node) | |
2076 | n = &s->local_node; | |
2077 | else { | |
2078 | if (slab_state == DOWN) { | |
2079 | n = early_kmem_cache_node_alloc(gfpflags, | |
2080 | node); | |
2081 | continue; | |
2082 | } | |
2083 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2084 | gfpflags, node); | |
2085 | ||
2086 | if (!n) { | |
2087 | free_kmem_cache_nodes(s); | |
2088 | return 0; | |
2089 | } | |
2090 | ||
2091 | } | |
2092 | s->node[node] = n; | |
2093 | init_kmem_cache_node(n); | |
2094 | } | |
2095 | return 1; | |
2096 | } | |
2097 | #else | |
2098 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2099 | { | |
2100 | } | |
2101 | ||
2102 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2103 | { | |
2104 | init_kmem_cache_node(&s->local_node); | |
2105 | return 1; | |
2106 | } | |
2107 | #endif | |
2108 | ||
2109 | /* | |
2110 | * calculate_sizes() determines the order and the distribution of data within | |
2111 | * a slab object. | |
2112 | */ | |
2113 | static int calculate_sizes(struct kmem_cache *s) | |
2114 | { | |
2115 | unsigned long flags = s->flags; | |
2116 | unsigned long size = s->objsize; | |
2117 | unsigned long align = s->align; | |
2118 | ||
2119 | /* | |
2120 | * Determine if we can poison the object itself. If the user of | |
2121 | * the slab may touch the object after free or before allocation | |
2122 | * then we should never poison the object itself. | |
2123 | */ | |
2124 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2125 | !s->ctor) |
81819f0f CL |
2126 | s->flags |= __OBJECT_POISON; |
2127 | else | |
2128 | s->flags &= ~__OBJECT_POISON; | |
2129 | ||
2130 | /* | |
2131 | * Round up object size to the next word boundary. We can only | |
2132 | * place the free pointer at word boundaries and this determines | |
2133 | * the possible location of the free pointer. | |
2134 | */ | |
2135 | size = ALIGN(size, sizeof(void *)); | |
2136 | ||
41ecc55b | 2137 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f | 2138 | /* |
672bba3a | 2139 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2140 | * end of the object and the free pointer. If not then add an |
672bba3a | 2141 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2142 | */ |
2143 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2144 | size += sizeof(void *); | |
41ecc55b | 2145 | #endif |
81819f0f CL |
2146 | |
2147 | /* | |
672bba3a CL |
2148 | * With that we have determined the number of bytes in actual use |
2149 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2150 | */ |
2151 | s->inuse = size; | |
2152 | ||
2153 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2154 | s->ctor)) { |
81819f0f CL |
2155 | /* |
2156 | * Relocate free pointer after the object if it is not | |
2157 | * permitted to overwrite the first word of the object on | |
2158 | * kmem_cache_free. | |
2159 | * | |
2160 | * This is the case if we do RCU, have a constructor or | |
2161 | * destructor or are poisoning the objects. | |
2162 | */ | |
2163 | s->offset = size; | |
2164 | size += sizeof(void *); | |
2165 | } | |
2166 | ||
c12b3c62 | 2167 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2168 | if (flags & SLAB_STORE_USER) |
2169 | /* | |
2170 | * Need to store information about allocs and frees after | |
2171 | * the object. | |
2172 | */ | |
2173 | size += 2 * sizeof(struct track); | |
2174 | ||
be7b3fbc | 2175 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2176 | /* |
2177 | * Add some empty padding so that we can catch | |
2178 | * overwrites from earlier objects rather than let | |
2179 | * tracking information or the free pointer be | |
2180 | * corrupted if an user writes before the start | |
2181 | * of the object. | |
2182 | */ | |
2183 | size += sizeof(void *); | |
41ecc55b | 2184 | #endif |
672bba3a | 2185 | |
81819f0f CL |
2186 | /* |
2187 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2188 | * user specified and the dynamic determination of cache line size |
2189 | * on bootup. | |
81819f0f CL |
2190 | */ |
2191 | align = calculate_alignment(flags, align, s->objsize); | |
2192 | ||
2193 | /* | |
2194 | * SLUB stores one object immediately after another beginning from | |
2195 | * offset 0. In order to align the objects we have to simply size | |
2196 | * each object to conform to the alignment. | |
2197 | */ | |
2198 | size = ALIGN(size, align); | |
2199 | s->size = size; | |
2200 | ||
2201 | s->order = calculate_order(size); | |
2202 | if (s->order < 0) | |
2203 | return 0; | |
2204 | ||
2205 | /* | |
2206 | * Determine the number of objects per slab | |
2207 | */ | |
2208 | s->objects = (PAGE_SIZE << s->order) / size; | |
2209 | ||
b3fba8da | 2210 | return !!s->objects; |
81819f0f CL |
2211 | |
2212 | } | |
2213 | ||
81819f0f CL |
2214 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2215 | const char *name, size_t size, | |
2216 | size_t align, unsigned long flags, | |
c59def9f | 2217 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) |
81819f0f CL |
2218 | { |
2219 | memset(s, 0, kmem_size); | |
2220 | s->name = name; | |
2221 | s->ctor = ctor; | |
81819f0f | 2222 | s->objsize = size; |
81819f0f | 2223 | s->align = align; |
ba0268a8 | 2224 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f CL |
2225 | |
2226 | if (!calculate_sizes(s)) | |
2227 | goto error; | |
2228 | ||
2229 | s->refcount = 1; | |
2230 | #ifdef CONFIG_NUMA | |
2231 | s->defrag_ratio = 100; | |
2232 | #endif | |
dfb4f096 CL |
2233 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2234 | goto error; | |
81819f0f | 2235 | |
dfb4f096 | 2236 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2237 | return 1; |
4c93c355 | 2238 | free_kmem_cache_nodes(s); |
81819f0f CL |
2239 | error: |
2240 | if (flags & SLAB_PANIC) | |
2241 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2242 | "order=%u offset=%u flags=%lx\n", | |
2243 | s->name, (unsigned long)size, s->size, s->order, | |
2244 | s->offset, flags); | |
2245 | return 0; | |
2246 | } | |
81819f0f CL |
2247 | |
2248 | /* | |
2249 | * Check if a given pointer is valid | |
2250 | */ | |
2251 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2252 | { | |
2253 | struct page * page; | |
81819f0f CL |
2254 | |
2255 | page = get_object_page(object); | |
2256 | ||
2257 | if (!page || s != page->slab) | |
2258 | /* No slab or wrong slab */ | |
2259 | return 0; | |
2260 | ||
abcd08a6 | 2261 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2262 | return 0; |
2263 | ||
2264 | /* | |
2265 | * We could also check if the object is on the slabs freelist. | |
2266 | * But this would be too expensive and it seems that the main | |
2267 | * purpose of kmem_ptr_valid is to check if the object belongs | |
2268 | * to a certain slab. | |
2269 | */ | |
2270 | return 1; | |
2271 | } | |
2272 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2273 | ||
2274 | /* | |
2275 | * Determine the size of a slab object | |
2276 | */ | |
2277 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2278 | { | |
2279 | return s->objsize; | |
2280 | } | |
2281 | EXPORT_SYMBOL(kmem_cache_size); | |
2282 | ||
2283 | const char *kmem_cache_name(struct kmem_cache *s) | |
2284 | { | |
2285 | return s->name; | |
2286 | } | |
2287 | EXPORT_SYMBOL(kmem_cache_name); | |
2288 | ||
2289 | /* | |
672bba3a CL |
2290 | * Attempt to free all slabs on a node. Return the number of slabs we |
2291 | * were unable to free. | |
81819f0f CL |
2292 | */ |
2293 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
2294 | struct list_head *list) | |
2295 | { | |
2296 | int slabs_inuse = 0; | |
2297 | unsigned long flags; | |
2298 | struct page *page, *h; | |
2299 | ||
2300 | spin_lock_irqsave(&n->list_lock, flags); | |
2301 | list_for_each_entry_safe(page, h, list, lru) | |
2302 | if (!page->inuse) { | |
2303 | list_del(&page->lru); | |
2304 | discard_slab(s, page); | |
2305 | } else | |
2306 | slabs_inuse++; | |
2307 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2308 | return slabs_inuse; | |
2309 | } | |
2310 | ||
2311 | /* | |
672bba3a | 2312 | * Release all resources used by a slab cache. |
81819f0f | 2313 | */ |
0c710013 | 2314 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2315 | { |
2316 | int node; | |
2317 | ||
2318 | flush_all(s); | |
2319 | ||
2320 | /* Attempt to free all objects */ | |
4c93c355 | 2321 | free_kmem_cache_cpus(s); |
f64dc58c | 2322 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2323 | struct kmem_cache_node *n = get_node(s, node); |
2324 | ||
2086d26a | 2325 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
2326 | if (atomic_long_read(&n->nr_slabs)) |
2327 | return 1; | |
2328 | } | |
2329 | free_kmem_cache_nodes(s); | |
2330 | return 0; | |
2331 | } | |
2332 | ||
2333 | /* | |
2334 | * Close a cache and release the kmem_cache structure | |
2335 | * (must be used for caches created using kmem_cache_create) | |
2336 | */ | |
2337 | void kmem_cache_destroy(struct kmem_cache *s) | |
2338 | { | |
2339 | down_write(&slub_lock); | |
2340 | s->refcount--; | |
2341 | if (!s->refcount) { | |
2342 | list_del(&s->list); | |
a0e1d1be | 2343 | up_write(&slub_lock); |
81819f0f CL |
2344 | if (kmem_cache_close(s)) |
2345 | WARN_ON(1); | |
2346 | sysfs_slab_remove(s); | |
2347 | kfree(s); | |
a0e1d1be CL |
2348 | } else |
2349 | up_write(&slub_lock); | |
81819f0f CL |
2350 | } |
2351 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2352 | ||
2353 | /******************************************************************** | |
2354 | * Kmalloc subsystem | |
2355 | *******************************************************************/ | |
2356 | ||
aadb4bc4 | 2357 | struct kmem_cache kmalloc_caches[PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2358 | EXPORT_SYMBOL(kmalloc_caches); |
2359 | ||
2360 | #ifdef CONFIG_ZONE_DMA | |
aadb4bc4 | 2361 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT]; |
81819f0f CL |
2362 | #endif |
2363 | ||
2364 | static int __init setup_slub_min_order(char *str) | |
2365 | { | |
2366 | get_option (&str, &slub_min_order); | |
2367 | ||
2368 | return 1; | |
2369 | } | |
2370 | ||
2371 | __setup("slub_min_order=", setup_slub_min_order); | |
2372 | ||
2373 | static int __init setup_slub_max_order(char *str) | |
2374 | { | |
2375 | get_option (&str, &slub_max_order); | |
2376 | ||
2377 | return 1; | |
2378 | } | |
2379 | ||
2380 | __setup("slub_max_order=", setup_slub_max_order); | |
2381 | ||
2382 | static int __init setup_slub_min_objects(char *str) | |
2383 | { | |
2384 | get_option (&str, &slub_min_objects); | |
2385 | ||
2386 | return 1; | |
2387 | } | |
2388 | ||
2389 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2390 | ||
2391 | static int __init setup_slub_nomerge(char *str) | |
2392 | { | |
2393 | slub_nomerge = 1; | |
2394 | return 1; | |
2395 | } | |
2396 | ||
2397 | __setup("slub_nomerge", setup_slub_nomerge); | |
2398 | ||
81819f0f CL |
2399 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2400 | const char *name, int size, gfp_t gfp_flags) | |
2401 | { | |
2402 | unsigned int flags = 0; | |
2403 | ||
2404 | if (gfp_flags & SLUB_DMA) | |
2405 | flags = SLAB_CACHE_DMA; | |
2406 | ||
2407 | down_write(&slub_lock); | |
2408 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
c59def9f | 2409 | flags, NULL)) |
81819f0f CL |
2410 | goto panic; |
2411 | ||
2412 | list_add(&s->list, &slab_caches); | |
2413 | up_write(&slub_lock); | |
2414 | if (sysfs_slab_add(s)) | |
2415 | goto panic; | |
2416 | return s; | |
2417 | ||
2418 | panic: | |
2419 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2420 | } | |
2421 | ||
2e443fd0 | 2422 | #ifdef CONFIG_ZONE_DMA |
1ceef402 CL |
2423 | |
2424 | static void sysfs_add_func(struct work_struct *w) | |
2425 | { | |
2426 | struct kmem_cache *s; | |
2427 | ||
2428 | down_write(&slub_lock); | |
2429 | list_for_each_entry(s, &slab_caches, list) { | |
2430 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2431 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2432 | sysfs_slab_add(s); | |
2433 | } | |
2434 | } | |
2435 | up_write(&slub_lock); | |
2436 | } | |
2437 | ||
2438 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2439 | ||
2e443fd0 CL |
2440 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2441 | { | |
2442 | struct kmem_cache *s; | |
2e443fd0 CL |
2443 | char *text; |
2444 | size_t realsize; | |
2445 | ||
2446 | s = kmalloc_caches_dma[index]; | |
2447 | if (s) | |
2448 | return s; | |
2449 | ||
2450 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2451 | if (flags & __GFP_WAIT) |
2452 | down_write(&slub_lock); | |
2453 | else { | |
2454 | if (!down_write_trylock(&slub_lock)) | |
2455 | goto out; | |
2456 | } | |
2457 | ||
2458 | if (kmalloc_caches_dma[index]) | |
2459 | goto unlock_out; | |
2e443fd0 | 2460 | |
7b55f620 | 2461 | realsize = kmalloc_caches[index].objsize; |
1ceef402 CL |
2462 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", (unsigned int)realsize), |
2463 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); | |
2464 | ||
2465 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2466 | realsize, ARCH_KMALLOC_MINALIGN, | |
2467 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2468 | kfree(s); | |
2469 | kfree(text); | |
2470 | goto unlock_out; | |
dfce8648 | 2471 | } |
1ceef402 CL |
2472 | |
2473 | list_add(&s->list, &slab_caches); | |
2474 | kmalloc_caches_dma[index] = s; | |
2475 | ||
2476 | schedule_work(&sysfs_add_work); | |
2477 | ||
2478 | unlock_out: | |
dfce8648 | 2479 | up_write(&slub_lock); |
1ceef402 | 2480 | out: |
dfce8648 | 2481 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2482 | } |
2483 | #endif | |
2484 | ||
f1b26339 CL |
2485 | /* |
2486 | * Conversion table for small slabs sizes / 8 to the index in the | |
2487 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2488 | * of two cache sizes there. The size of larger slabs can be determined using | |
2489 | * fls. | |
2490 | */ | |
2491 | static s8 size_index[24] = { | |
2492 | 3, /* 8 */ | |
2493 | 4, /* 16 */ | |
2494 | 5, /* 24 */ | |
2495 | 5, /* 32 */ | |
2496 | 6, /* 40 */ | |
2497 | 6, /* 48 */ | |
2498 | 6, /* 56 */ | |
2499 | 6, /* 64 */ | |
2500 | 1, /* 72 */ | |
2501 | 1, /* 80 */ | |
2502 | 1, /* 88 */ | |
2503 | 1, /* 96 */ | |
2504 | 7, /* 104 */ | |
2505 | 7, /* 112 */ | |
2506 | 7, /* 120 */ | |
2507 | 7, /* 128 */ | |
2508 | 2, /* 136 */ | |
2509 | 2, /* 144 */ | |
2510 | 2, /* 152 */ | |
2511 | 2, /* 160 */ | |
2512 | 2, /* 168 */ | |
2513 | 2, /* 176 */ | |
2514 | 2, /* 184 */ | |
2515 | 2 /* 192 */ | |
2516 | }; | |
2517 | ||
81819f0f CL |
2518 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2519 | { | |
f1b26339 | 2520 | int index; |
81819f0f | 2521 | |
f1b26339 CL |
2522 | if (size <= 192) { |
2523 | if (!size) | |
2524 | return ZERO_SIZE_PTR; | |
81819f0f | 2525 | |
f1b26339 | 2526 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2527 | } else |
f1b26339 | 2528 | index = fls(size - 1); |
81819f0f CL |
2529 | |
2530 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2531 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2532 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2533 | |
81819f0f CL |
2534 | #endif |
2535 | return &kmalloc_caches[index]; | |
2536 | } | |
2537 | ||
2538 | void *__kmalloc(size_t size, gfp_t flags) | |
2539 | { | |
aadb4bc4 | 2540 | struct kmem_cache *s; |
81819f0f | 2541 | |
aadb4bc4 CL |
2542 | if (unlikely(size > PAGE_SIZE / 2)) |
2543 | return (void *)__get_free_pages(flags | __GFP_COMP, | |
2544 | get_order(size)); | |
2545 | ||
2546 | s = get_slab(size, flags); | |
2547 | ||
2548 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2549 | return s; |
2550 | ||
ce15fea8 | 2551 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2552 | } |
2553 | EXPORT_SYMBOL(__kmalloc); | |
2554 | ||
2555 | #ifdef CONFIG_NUMA | |
2556 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2557 | { | |
aadb4bc4 | 2558 | struct kmem_cache *s; |
81819f0f | 2559 | |
aadb4bc4 CL |
2560 | if (unlikely(size > PAGE_SIZE / 2)) |
2561 | return (void *)__get_free_pages(flags | __GFP_COMP, | |
2562 | get_order(size)); | |
2563 | ||
2564 | s = get_slab(size, flags); | |
2565 | ||
2566 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2567 | return s; |
2568 | ||
ce15fea8 | 2569 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2570 | } |
2571 | EXPORT_SYMBOL(__kmalloc_node); | |
2572 | #endif | |
2573 | ||
2574 | size_t ksize(const void *object) | |
2575 | { | |
272c1d21 | 2576 | struct page *page; |
81819f0f CL |
2577 | struct kmem_cache *s; |
2578 | ||
ef8b4520 CL |
2579 | BUG_ON(!object); |
2580 | if (unlikely(object == ZERO_SIZE_PTR)) | |
272c1d21 CL |
2581 | return 0; |
2582 | ||
2583 | page = get_object_page(object); | |
81819f0f CL |
2584 | BUG_ON(!page); |
2585 | s = page->slab; | |
2586 | BUG_ON(!s); | |
2587 | ||
2588 | /* | |
2589 | * Debugging requires use of the padding between object | |
2590 | * and whatever may come after it. | |
2591 | */ | |
2592 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2593 | return s->objsize; | |
2594 | ||
2595 | /* | |
2596 | * If we have the need to store the freelist pointer | |
2597 | * back there or track user information then we can | |
2598 | * only use the space before that information. | |
2599 | */ | |
2600 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2601 | return s->inuse; | |
2602 | ||
2603 | /* | |
2604 | * Else we can use all the padding etc for the allocation | |
2605 | */ | |
2606 | return s->size; | |
2607 | } | |
2608 | EXPORT_SYMBOL(ksize); | |
2609 | ||
2610 | void kfree(const void *x) | |
2611 | { | |
81819f0f CL |
2612 | struct page *page; |
2613 | ||
2408c550 | 2614 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2615 | return; |
2616 | ||
b49af68f | 2617 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2618 | if (unlikely(!PageSlab(page))) { |
2619 | put_page(page); | |
2620 | return; | |
2621 | } | |
2622 | slab_free(page->slab, page, (void *)x, __builtin_return_address(0)); | |
81819f0f CL |
2623 | } |
2624 | EXPORT_SYMBOL(kfree); | |
2625 | ||
2086d26a | 2626 | /* |
672bba3a CL |
2627 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2628 | * the remaining slabs by the number of items in use. The slabs with the | |
2629 | * most items in use come first. New allocations will then fill those up | |
2630 | * and thus they can be removed from the partial lists. | |
2631 | * | |
2632 | * The slabs with the least items are placed last. This results in them | |
2633 | * being allocated from last increasing the chance that the last objects | |
2634 | * are freed in them. | |
2086d26a CL |
2635 | */ |
2636 | int kmem_cache_shrink(struct kmem_cache *s) | |
2637 | { | |
2638 | int node; | |
2639 | int i; | |
2640 | struct kmem_cache_node *n; | |
2641 | struct page *page; | |
2642 | struct page *t; | |
2643 | struct list_head *slabs_by_inuse = | |
2644 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2645 | unsigned long flags; | |
2646 | ||
2647 | if (!slabs_by_inuse) | |
2648 | return -ENOMEM; | |
2649 | ||
2650 | flush_all(s); | |
f64dc58c | 2651 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2652 | n = get_node(s, node); |
2653 | ||
2654 | if (!n->nr_partial) | |
2655 | continue; | |
2656 | ||
2657 | for (i = 0; i < s->objects; i++) | |
2658 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2659 | ||
2660 | spin_lock_irqsave(&n->list_lock, flags); | |
2661 | ||
2662 | /* | |
672bba3a | 2663 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2664 | * |
672bba3a CL |
2665 | * Note that concurrent frees may occur while we hold the |
2666 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2667 | */ |
2668 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2669 | if (!page->inuse && slab_trylock(page)) { | |
2670 | /* | |
2671 | * Must hold slab lock here because slab_free | |
2672 | * may have freed the last object and be | |
2673 | * waiting to release the slab. | |
2674 | */ | |
2675 | list_del(&page->lru); | |
2676 | n->nr_partial--; | |
2677 | slab_unlock(page); | |
2678 | discard_slab(s, page); | |
2679 | } else { | |
fcda3d89 CL |
2680 | list_move(&page->lru, |
2681 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2682 | } |
2683 | } | |
2684 | ||
2086d26a | 2685 | /* |
672bba3a CL |
2686 | * Rebuild the partial list with the slabs filled up most |
2687 | * first and the least used slabs at the end. | |
2086d26a CL |
2688 | */ |
2689 | for (i = s->objects - 1; i >= 0; i--) | |
2690 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2691 | ||
2086d26a CL |
2692 | spin_unlock_irqrestore(&n->list_lock, flags); |
2693 | } | |
2694 | ||
2695 | kfree(slabs_by_inuse); | |
2696 | return 0; | |
2697 | } | |
2698 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2699 | ||
81819f0f CL |
2700 | /******************************************************************** |
2701 | * Basic setup of slabs | |
2702 | *******************************************************************/ | |
2703 | ||
2704 | void __init kmem_cache_init(void) | |
2705 | { | |
2706 | int i; | |
4b356be0 | 2707 | int caches = 0; |
81819f0f | 2708 | |
4c93c355 CL |
2709 | init_alloc_cpu(); |
2710 | ||
81819f0f CL |
2711 | #ifdef CONFIG_NUMA |
2712 | /* | |
2713 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2714 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2715 | * kmem_cache_open for slab_state == DOWN. |
2716 | */ | |
2717 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2718 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2719 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 2720 | caches++; |
81819f0f CL |
2721 | #endif |
2722 | ||
2723 | /* Able to allocate the per node structures */ | |
2724 | slab_state = PARTIAL; | |
2725 | ||
2726 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
2727 | if (KMALLOC_MIN_SIZE <= 64) { |
2728 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 2729 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
2730 | caches++; |
2731 | } | |
2732 | if (KMALLOC_MIN_SIZE <= 128) { | |
2733 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 2734 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
2735 | caches++; |
2736 | } | |
81819f0f | 2737 | |
aadb4bc4 | 2738 | for (i = KMALLOC_SHIFT_LOW; i < PAGE_SHIFT; i++) { |
81819f0f CL |
2739 | create_kmalloc_cache(&kmalloc_caches[i], |
2740 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
2741 | caches++; |
2742 | } | |
81819f0f | 2743 | |
f1b26339 CL |
2744 | |
2745 | /* | |
2746 | * Patch up the size_index table if we have strange large alignment | |
2747 | * requirements for the kmalloc array. This is only the case for | |
2748 | * mips it seems. The standard arches will not generate any code here. | |
2749 | * | |
2750 | * Largest permitted alignment is 256 bytes due to the way we | |
2751 | * handle the index determination for the smaller caches. | |
2752 | * | |
2753 | * Make sure that nothing crazy happens if someone starts tinkering | |
2754 | * around with ARCH_KMALLOC_MINALIGN | |
2755 | */ | |
2756 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
2757 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
2758 | ||
12ad6843 | 2759 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
2760 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
2761 | ||
81819f0f CL |
2762 | slab_state = UP; |
2763 | ||
2764 | /* Provide the correct kmalloc names now that the caches are up */ | |
aadb4bc4 | 2765 | for (i = KMALLOC_SHIFT_LOW; i < PAGE_SHIFT; i++) |
81819f0f CL |
2766 | kmalloc_caches[i]. name = |
2767 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
2768 | ||
2769 | #ifdef CONFIG_SMP | |
2770 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
2771 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
2772 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
2773 | #else | |
2774 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
2775 | #endif |
2776 | ||
81819f0f CL |
2777 | |
2778 | printk(KERN_INFO "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
2779 | " CPUs=%d, Nodes=%d\n", |
2780 | caches, cache_line_size(), | |
81819f0f CL |
2781 | slub_min_order, slub_max_order, slub_min_objects, |
2782 | nr_cpu_ids, nr_node_ids); | |
2783 | } | |
2784 | ||
2785 | /* | |
2786 | * Find a mergeable slab cache | |
2787 | */ | |
2788 | static int slab_unmergeable(struct kmem_cache *s) | |
2789 | { | |
2790 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
2791 | return 1; | |
2792 | ||
c59def9f | 2793 | if (s->ctor) |
81819f0f CL |
2794 | return 1; |
2795 | ||
8ffa6875 CL |
2796 | /* |
2797 | * We may have set a slab to be unmergeable during bootstrap. | |
2798 | */ | |
2799 | if (s->refcount < 0) | |
2800 | return 1; | |
2801 | ||
81819f0f CL |
2802 | return 0; |
2803 | } | |
2804 | ||
2805 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 2806 | size_t align, unsigned long flags, const char *name, |
c59def9f | 2807 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) |
81819f0f | 2808 | { |
5b95a4ac | 2809 | struct kmem_cache *s; |
81819f0f CL |
2810 | |
2811 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
2812 | return NULL; | |
2813 | ||
c59def9f | 2814 | if (ctor) |
81819f0f CL |
2815 | return NULL; |
2816 | ||
2817 | size = ALIGN(size, sizeof(void *)); | |
2818 | align = calculate_alignment(flags, align, size); | |
2819 | size = ALIGN(size, align); | |
ba0268a8 | 2820 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 2821 | |
5b95a4ac | 2822 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
2823 | if (slab_unmergeable(s)) |
2824 | continue; | |
2825 | ||
2826 | if (size > s->size) | |
2827 | continue; | |
2828 | ||
ba0268a8 | 2829 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
2830 | continue; |
2831 | /* | |
2832 | * Check if alignment is compatible. | |
2833 | * Courtesy of Adrian Drzewiecki | |
2834 | */ | |
2835 | if ((s->size & ~(align -1)) != s->size) | |
2836 | continue; | |
2837 | ||
2838 | if (s->size - size >= sizeof(void *)) | |
2839 | continue; | |
2840 | ||
2841 | return s; | |
2842 | } | |
2843 | return NULL; | |
2844 | } | |
2845 | ||
2846 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
2847 | size_t align, unsigned long flags, | |
20c2df83 | 2848 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) |
81819f0f CL |
2849 | { |
2850 | struct kmem_cache *s; | |
2851 | ||
2852 | down_write(&slub_lock); | |
ba0268a8 | 2853 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
2854 | if (s) { |
2855 | s->refcount++; | |
2856 | /* | |
2857 | * Adjust the object sizes so that we clear | |
2858 | * the complete object on kzalloc. | |
2859 | */ | |
2860 | s->objsize = max(s->objsize, (int)size); | |
2861 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
a0e1d1be | 2862 | up_write(&slub_lock); |
81819f0f CL |
2863 | if (sysfs_slab_alias(s, name)) |
2864 | goto err; | |
a0e1d1be CL |
2865 | return s; |
2866 | } | |
2867 | s = kmalloc(kmem_size, GFP_KERNEL); | |
2868 | if (s) { | |
2869 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 2870 | size, align, flags, ctor)) { |
81819f0f | 2871 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
2872 | up_write(&slub_lock); |
2873 | if (sysfs_slab_add(s)) | |
2874 | goto err; | |
2875 | return s; | |
2876 | } | |
2877 | kfree(s); | |
81819f0f CL |
2878 | } |
2879 | up_write(&slub_lock); | |
81819f0f CL |
2880 | |
2881 | err: | |
81819f0f CL |
2882 | if (flags & SLAB_PANIC) |
2883 | panic("Cannot create slabcache %s\n", name); | |
2884 | else | |
2885 | s = NULL; | |
2886 | return s; | |
2887 | } | |
2888 | EXPORT_SYMBOL(kmem_cache_create); | |
2889 | ||
81819f0f | 2890 | #ifdef CONFIG_SMP |
81819f0f | 2891 | /* |
672bba3a CL |
2892 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
2893 | * necessary. | |
81819f0f CL |
2894 | */ |
2895 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
2896 | unsigned long action, void *hcpu) | |
2897 | { | |
2898 | long cpu = (long)hcpu; | |
5b95a4ac CL |
2899 | struct kmem_cache *s; |
2900 | unsigned long flags; | |
81819f0f CL |
2901 | |
2902 | switch (action) { | |
4c93c355 CL |
2903 | case CPU_UP_PREPARE: |
2904 | case CPU_UP_PREPARE_FROZEN: | |
2905 | init_alloc_cpu_cpu(cpu); | |
2906 | down_read(&slub_lock); | |
2907 | list_for_each_entry(s, &slab_caches, list) | |
2908 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
2909 | GFP_KERNEL); | |
2910 | up_read(&slub_lock); | |
2911 | break; | |
2912 | ||
81819f0f | 2913 | case CPU_UP_CANCELED: |
8bb78442 | 2914 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 2915 | case CPU_DEAD: |
8bb78442 | 2916 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
2917 | down_read(&slub_lock); |
2918 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
2919 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
2920 | ||
5b95a4ac CL |
2921 | local_irq_save(flags); |
2922 | __flush_cpu_slab(s, cpu); | |
2923 | local_irq_restore(flags); | |
4c93c355 CL |
2924 | free_kmem_cache_cpu(c, cpu); |
2925 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
2926 | } |
2927 | up_read(&slub_lock); | |
81819f0f CL |
2928 | break; |
2929 | default: | |
2930 | break; | |
2931 | } | |
2932 | return NOTIFY_OK; | |
2933 | } | |
2934 | ||
2935 | static struct notifier_block __cpuinitdata slab_notifier = | |
2936 | { &slab_cpuup_callback, NULL, 0 }; | |
2937 | ||
2938 | #endif | |
2939 | ||
81819f0f CL |
2940 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
2941 | { | |
aadb4bc4 CL |
2942 | struct kmem_cache *s; |
2943 | ||
2944 | if (unlikely(size > PAGE_SIZE / 2)) | |
2945 | return (void *)__get_free_pages(gfpflags | __GFP_COMP, | |
2946 | get_order(size)); | |
2947 | s = get_slab(size, gfpflags); | |
81819f0f | 2948 | |
2408c550 | 2949 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 2950 | return s; |
81819f0f | 2951 | |
ce15fea8 | 2952 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
2953 | } |
2954 | ||
2955 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
2956 | int node, void *caller) | |
2957 | { | |
aadb4bc4 CL |
2958 | struct kmem_cache *s; |
2959 | ||
2960 | if (unlikely(size > PAGE_SIZE / 2)) | |
2961 | return (void *)__get_free_pages(gfpflags | __GFP_COMP, | |
2962 | get_order(size)); | |
2963 | s = get_slab(size, gfpflags); | |
81819f0f | 2964 | |
2408c550 | 2965 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 2966 | return s; |
81819f0f | 2967 | |
ce15fea8 | 2968 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
2969 | } |
2970 | ||
41ecc55b | 2971 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
2972 | static int validate_slab(struct kmem_cache *s, struct page *page, |
2973 | unsigned long *map) | |
53e15af0 CL |
2974 | { |
2975 | void *p; | |
2976 | void *addr = page_address(page); | |
53e15af0 CL |
2977 | |
2978 | if (!check_slab(s, page) || | |
2979 | !on_freelist(s, page, NULL)) | |
2980 | return 0; | |
2981 | ||
2982 | /* Now we know that a valid freelist exists */ | |
2983 | bitmap_zero(map, s->objects); | |
2984 | ||
7656c72b CL |
2985 | for_each_free_object(p, s, page->freelist) { |
2986 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
2987 | if (!check_object(s, page, p, 0)) |
2988 | return 0; | |
2989 | } | |
2990 | ||
7656c72b CL |
2991 | for_each_object(p, s, addr) |
2992 | if (!test_bit(slab_index(p, s, addr), map)) | |
53e15af0 CL |
2993 | if (!check_object(s, page, p, 1)) |
2994 | return 0; | |
2995 | return 1; | |
2996 | } | |
2997 | ||
434e245d CL |
2998 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
2999 | unsigned long *map) | |
53e15af0 CL |
3000 | { |
3001 | if (slab_trylock(page)) { | |
434e245d | 3002 | validate_slab(s, page, map); |
53e15af0 CL |
3003 | slab_unlock(page); |
3004 | } else | |
3005 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3006 | s->name, page); | |
3007 | ||
3008 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3009 | if (!SlabDebug(page)) |
3010 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3011 | "on slab 0x%p\n", s->name, page); |
3012 | } else { | |
35e5d7ee CL |
3013 | if (SlabDebug(page)) |
3014 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3015 | "slab 0x%p\n", s->name, page); |
3016 | } | |
3017 | } | |
3018 | ||
434e245d CL |
3019 | static int validate_slab_node(struct kmem_cache *s, |
3020 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3021 | { |
3022 | unsigned long count = 0; | |
3023 | struct page *page; | |
3024 | unsigned long flags; | |
3025 | ||
3026 | spin_lock_irqsave(&n->list_lock, flags); | |
3027 | ||
3028 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3029 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3030 | count++; |
3031 | } | |
3032 | if (count != n->nr_partial) | |
3033 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3034 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3035 | ||
3036 | if (!(s->flags & SLAB_STORE_USER)) | |
3037 | goto out; | |
3038 | ||
3039 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3040 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3041 | count++; |
3042 | } | |
3043 | if (count != atomic_long_read(&n->nr_slabs)) | |
3044 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3045 | "counter=%ld\n", s->name, count, | |
3046 | atomic_long_read(&n->nr_slabs)); | |
3047 | ||
3048 | out: | |
3049 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3050 | return count; | |
3051 | } | |
3052 | ||
434e245d | 3053 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3054 | { |
3055 | int node; | |
3056 | unsigned long count = 0; | |
434e245d CL |
3057 | unsigned long *map = kmalloc(BITS_TO_LONGS(s->objects) * |
3058 | sizeof(unsigned long), GFP_KERNEL); | |
3059 | ||
3060 | if (!map) | |
3061 | return -ENOMEM; | |
53e15af0 CL |
3062 | |
3063 | flush_all(s); | |
f64dc58c | 3064 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3065 | struct kmem_cache_node *n = get_node(s, node); |
3066 | ||
434e245d | 3067 | count += validate_slab_node(s, n, map); |
53e15af0 | 3068 | } |
434e245d | 3069 | kfree(map); |
53e15af0 CL |
3070 | return count; |
3071 | } | |
3072 | ||
b3459709 CL |
3073 | #ifdef SLUB_RESILIENCY_TEST |
3074 | static void resiliency_test(void) | |
3075 | { | |
3076 | u8 *p; | |
3077 | ||
3078 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3079 | printk(KERN_ERR "-----------------------\n"); | |
3080 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3081 | ||
3082 | p = kzalloc(16, GFP_KERNEL); | |
3083 | p[16] = 0x12; | |
3084 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3085 | " 0x12->0x%p\n\n", p + 16); | |
3086 | ||
3087 | validate_slab_cache(kmalloc_caches + 4); | |
3088 | ||
3089 | /* Hmmm... The next two are dangerous */ | |
3090 | p = kzalloc(32, GFP_KERNEL); | |
3091 | p[32 + sizeof(void *)] = 0x34; | |
3092 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3093 | " 0x34 -> -0x%p\n", p); | |
3094 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
3095 | ||
3096 | validate_slab_cache(kmalloc_caches + 5); | |
3097 | p = kzalloc(64, GFP_KERNEL); | |
3098 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3099 | *p = 0x56; | |
3100 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3101 | p); | |
3102 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
3103 | validate_slab_cache(kmalloc_caches + 6); | |
3104 | ||
3105 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3106 | p = kzalloc(128, GFP_KERNEL); | |
3107 | kfree(p); | |
3108 | *p = 0x78; | |
3109 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3110 | validate_slab_cache(kmalloc_caches + 7); | |
3111 | ||
3112 | p = kzalloc(256, GFP_KERNEL); | |
3113 | kfree(p); | |
3114 | p[50] = 0x9a; | |
3115 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); | |
3116 | validate_slab_cache(kmalloc_caches + 8); | |
3117 | ||
3118 | p = kzalloc(512, GFP_KERNEL); | |
3119 | kfree(p); | |
3120 | p[512] = 0xab; | |
3121 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3122 | validate_slab_cache(kmalloc_caches + 9); | |
3123 | } | |
3124 | #else | |
3125 | static void resiliency_test(void) {}; | |
3126 | #endif | |
3127 | ||
88a420e4 | 3128 | /* |
672bba3a | 3129 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3130 | * and freed. |
3131 | */ | |
3132 | ||
3133 | struct location { | |
3134 | unsigned long count; | |
3135 | void *addr; | |
45edfa58 CL |
3136 | long long sum_time; |
3137 | long min_time; | |
3138 | long max_time; | |
3139 | long min_pid; | |
3140 | long max_pid; | |
3141 | cpumask_t cpus; | |
3142 | nodemask_t nodes; | |
88a420e4 CL |
3143 | }; |
3144 | ||
3145 | struct loc_track { | |
3146 | unsigned long max; | |
3147 | unsigned long count; | |
3148 | struct location *loc; | |
3149 | }; | |
3150 | ||
3151 | static void free_loc_track(struct loc_track *t) | |
3152 | { | |
3153 | if (t->max) | |
3154 | free_pages((unsigned long)t->loc, | |
3155 | get_order(sizeof(struct location) * t->max)); | |
3156 | } | |
3157 | ||
68dff6a9 | 3158 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3159 | { |
3160 | struct location *l; | |
3161 | int order; | |
3162 | ||
88a420e4 CL |
3163 | order = get_order(sizeof(struct location) * max); |
3164 | ||
68dff6a9 | 3165 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3166 | if (!l) |
3167 | return 0; | |
3168 | ||
3169 | if (t->count) { | |
3170 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3171 | free_loc_track(t); | |
3172 | } | |
3173 | t->max = max; | |
3174 | t->loc = l; | |
3175 | return 1; | |
3176 | } | |
3177 | ||
3178 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3179 | const struct track *track) |
88a420e4 CL |
3180 | { |
3181 | long start, end, pos; | |
3182 | struct location *l; | |
3183 | void *caddr; | |
45edfa58 | 3184 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3185 | |
3186 | start = -1; | |
3187 | end = t->count; | |
3188 | ||
3189 | for ( ; ; ) { | |
3190 | pos = start + (end - start + 1) / 2; | |
3191 | ||
3192 | /* | |
3193 | * There is nothing at "end". If we end up there | |
3194 | * we need to add something to before end. | |
3195 | */ | |
3196 | if (pos == end) | |
3197 | break; | |
3198 | ||
3199 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3200 | if (track->addr == caddr) { |
3201 | ||
3202 | l = &t->loc[pos]; | |
3203 | l->count++; | |
3204 | if (track->when) { | |
3205 | l->sum_time += age; | |
3206 | if (age < l->min_time) | |
3207 | l->min_time = age; | |
3208 | if (age > l->max_time) | |
3209 | l->max_time = age; | |
3210 | ||
3211 | if (track->pid < l->min_pid) | |
3212 | l->min_pid = track->pid; | |
3213 | if (track->pid > l->max_pid) | |
3214 | l->max_pid = track->pid; | |
3215 | ||
3216 | cpu_set(track->cpu, l->cpus); | |
3217 | } | |
3218 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3219 | return 1; |
3220 | } | |
3221 | ||
45edfa58 | 3222 | if (track->addr < caddr) |
88a420e4 CL |
3223 | end = pos; |
3224 | else | |
3225 | start = pos; | |
3226 | } | |
3227 | ||
3228 | /* | |
672bba3a | 3229 | * Not found. Insert new tracking element. |
88a420e4 | 3230 | */ |
68dff6a9 | 3231 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3232 | return 0; |
3233 | ||
3234 | l = t->loc + pos; | |
3235 | if (pos < t->count) | |
3236 | memmove(l + 1, l, | |
3237 | (t->count - pos) * sizeof(struct location)); | |
3238 | t->count++; | |
3239 | l->count = 1; | |
45edfa58 CL |
3240 | l->addr = track->addr; |
3241 | l->sum_time = age; | |
3242 | l->min_time = age; | |
3243 | l->max_time = age; | |
3244 | l->min_pid = track->pid; | |
3245 | l->max_pid = track->pid; | |
3246 | cpus_clear(l->cpus); | |
3247 | cpu_set(track->cpu, l->cpus); | |
3248 | nodes_clear(l->nodes); | |
3249 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3250 | return 1; |
3251 | } | |
3252 | ||
3253 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3254 | struct page *page, enum track_item alloc) | |
3255 | { | |
3256 | void *addr = page_address(page); | |
7656c72b | 3257 | DECLARE_BITMAP(map, s->objects); |
88a420e4 CL |
3258 | void *p; |
3259 | ||
3260 | bitmap_zero(map, s->objects); | |
7656c72b CL |
3261 | for_each_free_object(p, s, page->freelist) |
3262 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3263 | |
7656c72b | 3264 | for_each_object(p, s, addr) |
45edfa58 CL |
3265 | if (!test_bit(slab_index(p, s, addr), map)) |
3266 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3267 | } |
3268 | ||
3269 | static int list_locations(struct kmem_cache *s, char *buf, | |
3270 | enum track_item alloc) | |
3271 | { | |
3272 | int n = 0; | |
3273 | unsigned long i; | |
68dff6a9 | 3274 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3275 | int node; |
3276 | ||
68dff6a9 CL |
3277 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3278 | GFP_KERNEL)) | |
3279 | return sprintf(buf, "Out of memory\n"); | |
88a420e4 CL |
3280 | |
3281 | /* Push back cpu slabs */ | |
3282 | flush_all(s); | |
3283 | ||
f64dc58c | 3284 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3285 | struct kmem_cache_node *n = get_node(s, node); |
3286 | unsigned long flags; | |
3287 | struct page *page; | |
3288 | ||
9e86943b | 3289 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3290 | continue; |
3291 | ||
3292 | spin_lock_irqsave(&n->list_lock, flags); | |
3293 | list_for_each_entry(page, &n->partial, lru) | |
3294 | process_slab(&t, s, page, alloc); | |
3295 | list_for_each_entry(page, &n->full, lru) | |
3296 | process_slab(&t, s, page, alloc); | |
3297 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3298 | } | |
3299 | ||
3300 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3301 | struct location *l = &t.loc[i]; |
88a420e4 CL |
3302 | |
3303 | if (n > PAGE_SIZE - 100) | |
3304 | break; | |
45edfa58 CL |
3305 | n += sprintf(buf + n, "%7ld ", l->count); |
3306 | ||
3307 | if (l->addr) | |
3308 | n += sprint_symbol(buf + n, (unsigned long)l->addr); | |
88a420e4 CL |
3309 | else |
3310 | n += sprintf(buf + n, "<not-available>"); | |
45edfa58 CL |
3311 | |
3312 | if (l->sum_time != l->min_time) { | |
3313 | unsigned long remainder; | |
3314 | ||
3315 | n += sprintf(buf + n, " age=%ld/%ld/%ld", | |
3316 | l->min_time, | |
3317 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3318 | l->max_time); | |
3319 | } else | |
3320 | n += sprintf(buf + n, " age=%ld", | |
3321 | l->min_time); | |
3322 | ||
3323 | if (l->min_pid != l->max_pid) | |
3324 | n += sprintf(buf + n, " pid=%ld-%ld", | |
3325 | l->min_pid, l->max_pid); | |
3326 | else | |
3327 | n += sprintf(buf + n, " pid=%ld", | |
3328 | l->min_pid); | |
3329 | ||
84966343 CL |
3330 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
3331 | n < PAGE_SIZE - 60) { | |
45edfa58 CL |
3332 | n += sprintf(buf + n, " cpus="); |
3333 | n += cpulist_scnprintf(buf + n, PAGE_SIZE - n - 50, | |
3334 | l->cpus); | |
3335 | } | |
3336 | ||
84966343 CL |
3337 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
3338 | n < PAGE_SIZE - 60) { | |
45edfa58 CL |
3339 | n += sprintf(buf + n, " nodes="); |
3340 | n += nodelist_scnprintf(buf + n, PAGE_SIZE - n - 50, | |
3341 | l->nodes); | |
3342 | } | |
3343 | ||
88a420e4 CL |
3344 | n += sprintf(buf + n, "\n"); |
3345 | } | |
3346 | ||
3347 | free_loc_track(&t); | |
3348 | if (!t.count) | |
3349 | n += sprintf(buf, "No data\n"); | |
3350 | return n; | |
3351 | } | |
3352 | ||
81819f0f CL |
3353 | static unsigned long count_partial(struct kmem_cache_node *n) |
3354 | { | |
3355 | unsigned long flags; | |
3356 | unsigned long x = 0; | |
3357 | struct page *page; | |
3358 | ||
3359 | spin_lock_irqsave(&n->list_lock, flags); | |
3360 | list_for_each_entry(page, &n->partial, lru) | |
3361 | x += page->inuse; | |
3362 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3363 | return x; | |
3364 | } | |
3365 | ||
3366 | enum slab_stat_type { | |
3367 | SL_FULL, | |
3368 | SL_PARTIAL, | |
3369 | SL_CPU, | |
3370 | SL_OBJECTS | |
3371 | }; | |
3372 | ||
3373 | #define SO_FULL (1 << SL_FULL) | |
3374 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
3375 | #define SO_CPU (1 << SL_CPU) | |
3376 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
3377 | ||
3378 | static unsigned long slab_objects(struct kmem_cache *s, | |
3379 | char *buf, unsigned long flags) | |
3380 | { | |
3381 | unsigned long total = 0; | |
3382 | int cpu; | |
3383 | int node; | |
3384 | int x; | |
3385 | unsigned long *nodes; | |
3386 | unsigned long *per_cpu; | |
3387 | ||
3388 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
3389 | per_cpu = nodes + nr_node_ids; | |
3390 | ||
3391 | for_each_possible_cpu(cpu) { | |
dfb4f096 | 3392 | struct page *page; |
ee3c72a1 | 3393 | int node; |
dfb4f096 | 3394 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 3395 | |
dfb4f096 CL |
3396 | if (!c) |
3397 | continue; | |
3398 | ||
3399 | page = c->page; | |
ee3c72a1 CL |
3400 | node = c->node; |
3401 | if (node < 0) | |
3402 | continue; | |
81819f0f | 3403 | if (page) { |
81819f0f CL |
3404 | if (flags & SO_CPU) { |
3405 | int x = 0; | |
3406 | ||
3407 | if (flags & SO_OBJECTS) | |
3408 | x = page->inuse; | |
3409 | else | |
3410 | x = 1; | |
3411 | total += x; | |
ee3c72a1 | 3412 | nodes[node] += x; |
81819f0f | 3413 | } |
ee3c72a1 | 3414 | per_cpu[node]++; |
81819f0f CL |
3415 | } |
3416 | } | |
3417 | ||
f64dc58c | 3418 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3419 | struct kmem_cache_node *n = get_node(s, node); |
3420 | ||
3421 | if (flags & SO_PARTIAL) { | |
3422 | if (flags & SO_OBJECTS) | |
3423 | x = count_partial(n); | |
3424 | else | |
3425 | x = n->nr_partial; | |
3426 | total += x; | |
3427 | nodes[node] += x; | |
3428 | } | |
3429 | ||
3430 | if (flags & SO_FULL) { | |
9e86943b | 3431 | int full_slabs = atomic_long_read(&n->nr_slabs) |
81819f0f CL |
3432 | - per_cpu[node] |
3433 | - n->nr_partial; | |
3434 | ||
3435 | if (flags & SO_OBJECTS) | |
3436 | x = full_slabs * s->objects; | |
3437 | else | |
3438 | x = full_slabs; | |
3439 | total += x; | |
3440 | nodes[node] += x; | |
3441 | } | |
3442 | } | |
3443 | ||
3444 | x = sprintf(buf, "%lu", total); | |
3445 | #ifdef CONFIG_NUMA | |
f64dc58c | 3446 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3447 | if (nodes[node]) |
3448 | x += sprintf(buf + x, " N%d=%lu", | |
3449 | node, nodes[node]); | |
3450 | #endif | |
3451 | kfree(nodes); | |
3452 | return x + sprintf(buf + x, "\n"); | |
3453 | } | |
3454 | ||
3455 | static int any_slab_objects(struct kmem_cache *s) | |
3456 | { | |
3457 | int node; | |
3458 | int cpu; | |
3459 | ||
dfb4f096 CL |
3460 | for_each_possible_cpu(cpu) { |
3461 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
3462 | ||
3463 | if (c && c->page) | |
81819f0f | 3464 | return 1; |
dfb4f096 | 3465 | } |
81819f0f | 3466 | |
dfb4f096 | 3467 | for_each_online_node(node) { |
81819f0f CL |
3468 | struct kmem_cache_node *n = get_node(s, node); |
3469 | ||
dfb4f096 CL |
3470 | if (!n) |
3471 | continue; | |
3472 | ||
9e86943b | 3473 | if (n->nr_partial || atomic_long_read(&n->nr_slabs)) |
81819f0f CL |
3474 | return 1; |
3475 | } | |
3476 | return 0; | |
3477 | } | |
3478 | ||
3479 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3480 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3481 | ||
3482 | struct slab_attribute { | |
3483 | struct attribute attr; | |
3484 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3485 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3486 | }; | |
3487 | ||
3488 | #define SLAB_ATTR_RO(_name) \ | |
3489 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3490 | ||
3491 | #define SLAB_ATTR(_name) \ | |
3492 | static struct slab_attribute _name##_attr = \ | |
3493 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3494 | ||
81819f0f CL |
3495 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3496 | { | |
3497 | return sprintf(buf, "%d\n", s->size); | |
3498 | } | |
3499 | SLAB_ATTR_RO(slab_size); | |
3500 | ||
3501 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3502 | { | |
3503 | return sprintf(buf, "%d\n", s->align); | |
3504 | } | |
3505 | SLAB_ATTR_RO(align); | |
3506 | ||
3507 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3508 | { | |
3509 | return sprintf(buf, "%d\n", s->objsize); | |
3510 | } | |
3511 | SLAB_ATTR_RO(object_size); | |
3512 | ||
3513 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3514 | { | |
3515 | return sprintf(buf, "%d\n", s->objects); | |
3516 | } | |
3517 | SLAB_ATTR_RO(objs_per_slab); | |
3518 | ||
3519 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3520 | { | |
3521 | return sprintf(buf, "%d\n", s->order); | |
3522 | } | |
3523 | SLAB_ATTR_RO(order); | |
3524 | ||
3525 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3526 | { | |
3527 | if (s->ctor) { | |
3528 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3529 | ||
3530 | return n + sprintf(buf + n, "\n"); | |
3531 | } | |
3532 | return 0; | |
3533 | } | |
3534 | SLAB_ATTR_RO(ctor); | |
3535 | ||
81819f0f CL |
3536 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3537 | { | |
3538 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3539 | } | |
3540 | SLAB_ATTR_RO(aliases); | |
3541 | ||
3542 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3543 | { | |
3544 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3545 | } | |
3546 | SLAB_ATTR_RO(slabs); | |
3547 | ||
3548 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3549 | { | |
3550 | return slab_objects(s, buf, SO_PARTIAL); | |
3551 | } | |
3552 | SLAB_ATTR_RO(partial); | |
3553 | ||
3554 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3555 | { | |
3556 | return slab_objects(s, buf, SO_CPU); | |
3557 | } | |
3558 | SLAB_ATTR_RO(cpu_slabs); | |
3559 | ||
3560 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3561 | { | |
3562 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3563 | } | |
3564 | SLAB_ATTR_RO(objects); | |
3565 | ||
3566 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3567 | { | |
3568 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3569 | } | |
3570 | ||
3571 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3572 | const char *buf, size_t length) | |
3573 | { | |
3574 | s->flags &= ~SLAB_DEBUG_FREE; | |
3575 | if (buf[0] == '1') | |
3576 | s->flags |= SLAB_DEBUG_FREE; | |
3577 | return length; | |
3578 | } | |
3579 | SLAB_ATTR(sanity_checks); | |
3580 | ||
3581 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3582 | { | |
3583 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3584 | } | |
3585 | ||
3586 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3587 | size_t length) | |
3588 | { | |
3589 | s->flags &= ~SLAB_TRACE; | |
3590 | if (buf[0] == '1') | |
3591 | s->flags |= SLAB_TRACE; | |
3592 | return length; | |
3593 | } | |
3594 | SLAB_ATTR(trace); | |
3595 | ||
3596 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3597 | { | |
3598 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3599 | } | |
3600 | ||
3601 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3602 | const char *buf, size_t length) | |
3603 | { | |
3604 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3605 | if (buf[0] == '1') | |
3606 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3607 | return length; | |
3608 | } | |
3609 | SLAB_ATTR(reclaim_account); | |
3610 | ||
3611 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3612 | { | |
5af60839 | 3613 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3614 | } |
3615 | SLAB_ATTR_RO(hwcache_align); | |
3616 | ||
3617 | #ifdef CONFIG_ZONE_DMA | |
3618 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3619 | { | |
3620 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3621 | } | |
3622 | SLAB_ATTR_RO(cache_dma); | |
3623 | #endif | |
3624 | ||
3625 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3626 | { | |
3627 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3628 | } | |
3629 | SLAB_ATTR_RO(destroy_by_rcu); | |
3630 | ||
3631 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3632 | { | |
3633 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3634 | } | |
3635 | ||
3636 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3637 | const char *buf, size_t length) | |
3638 | { | |
3639 | if (any_slab_objects(s)) | |
3640 | return -EBUSY; | |
3641 | ||
3642 | s->flags &= ~SLAB_RED_ZONE; | |
3643 | if (buf[0] == '1') | |
3644 | s->flags |= SLAB_RED_ZONE; | |
3645 | calculate_sizes(s); | |
3646 | return length; | |
3647 | } | |
3648 | SLAB_ATTR(red_zone); | |
3649 | ||
3650 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3651 | { | |
3652 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3653 | } | |
3654 | ||
3655 | static ssize_t poison_store(struct kmem_cache *s, | |
3656 | const char *buf, size_t length) | |
3657 | { | |
3658 | if (any_slab_objects(s)) | |
3659 | return -EBUSY; | |
3660 | ||
3661 | s->flags &= ~SLAB_POISON; | |
3662 | if (buf[0] == '1') | |
3663 | s->flags |= SLAB_POISON; | |
3664 | calculate_sizes(s); | |
3665 | return length; | |
3666 | } | |
3667 | SLAB_ATTR(poison); | |
3668 | ||
3669 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3670 | { | |
3671 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3672 | } | |
3673 | ||
3674 | static ssize_t store_user_store(struct kmem_cache *s, | |
3675 | const char *buf, size_t length) | |
3676 | { | |
3677 | if (any_slab_objects(s)) | |
3678 | return -EBUSY; | |
3679 | ||
3680 | s->flags &= ~SLAB_STORE_USER; | |
3681 | if (buf[0] == '1') | |
3682 | s->flags |= SLAB_STORE_USER; | |
3683 | calculate_sizes(s); | |
3684 | return length; | |
3685 | } | |
3686 | SLAB_ATTR(store_user); | |
3687 | ||
53e15af0 CL |
3688 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3689 | { | |
3690 | return 0; | |
3691 | } | |
3692 | ||
3693 | static ssize_t validate_store(struct kmem_cache *s, | |
3694 | const char *buf, size_t length) | |
3695 | { | |
434e245d CL |
3696 | int ret = -EINVAL; |
3697 | ||
3698 | if (buf[0] == '1') { | |
3699 | ret = validate_slab_cache(s); | |
3700 | if (ret >= 0) | |
3701 | ret = length; | |
3702 | } | |
3703 | return ret; | |
53e15af0 CL |
3704 | } |
3705 | SLAB_ATTR(validate); | |
3706 | ||
2086d26a CL |
3707 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
3708 | { | |
3709 | return 0; | |
3710 | } | |
3711 | ||
3712 | static ssize_t shrink_store(struct kmem_cache *s, | |
3713 | const char *buf, size_t length) | |
3714 | { | |
3715 | if (buf[0] == '1') { | |
3716 | int rc = kmem_cache_shrink(s); | |
3717 | ||
3718 | if (rc) | |
3719 | return rc; | |
3720 | } else | |
3721 | return -EINVAL; | |
3722 | return length; | |
3723 | } | |
3724 | SLAB_ATTR(shrink); | |
3725 | ||
88a420e4 CL |
3726 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
3727 | { | |
3728 | if (!(s->flags & SLAB_STORE_USER)) | |
3729 | return -ENOSYS; | |
3730 | return list_locations(s, buf, TRACK_ALLOC); | |
3731 | } | |
3732 | SLAB_ATTR_RO(alloc_calls); | |
3733 | ||
3734 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
3735 | { | |
3736 | if (!(s->flags & SLAB_STORE_USER)) | |
3737 | return -ENOSYS; | |
3738 | return list_locations(s, buf, TRACK_FREE); | |
3739 | } | |
3740 | SLAB_ATTR_RO(free_calls); | |
3741 | ||
81819f0f CL |
3742 | #ifdef CONFIG_NUMA |
3743 | static ssize_t defrag_ratio_show(struct kmem_cache *s, char *buf) | |
3744 | { | |
3745 | return sprintf(buf, "%d\n", s->defrag_ratio / 10); | |
3746 | } | |
3747 | ||
3748 | static ssize_t defrag_ratio_store(struct kmem_cache *s, | |
3749 | const char *buf, size_t length) | |
3750 | { | |
3751 | int n = simple_strtoul(buf, NULL, 10); | |
3752 | ||
3753 | if (n < 100) | |
3754 | s->defrag_ratio = n * 10; | |
3755 | return length; | |
3756 | } | |
3757 | SLAB_ATTR(defrag_ratio); | |
3758 | #endif | |
3759 | ||
3760 | static struct attribute * slab_attrs[] = { | |
3761 | &slab_size_attr.attr, | |
3762 | &object_size_attr.attr, | |
3763 | &objs_per_slab_attr.attr, | |
3764 | &order_attr.attr, | |
3765 | &objects_attr.attr, | |
3766 | &slabs_attr.attr, | |
3767 | &partial_attr.attr, | |
3768 | &cpu_slabs_attr.attr, | |
3769 | &ctor_attr.attr, | |
81819f0f CL |
3770 | &aliases_attr.attr, |
3771 | &align_attr.attr, | |
3772 | &sanity_checks_attr.attr, | |
3773 | &trace_attr.attr, | |
3774 | &hwcache_align_attr.attr, | |
3775 | &reclaim_account_attr.attr, | |
3776 | &destroy_by_rcu_attr.attr, | |
3777 | &red_zone_attr.attr, | |
3778 | &poison_attr.attr, | |
3779 | &store_user_attr.attr, | |
53e15af0 | 3780 | &validate_attr.attr, |
2086d26a | 3781 | &shrink_attr.attr, |
88a420e4 CL |
3782 | &alloc_calls_attr.attr, |
3783 | &free_calls_attr.attr, | |
81819f0f CL |
3784 | #ifdef CONFIG_ZONE_DMA |
3785 | &cache_dma_attr.attr, | |
3786 | #endif | |
3787 | #ifdef CONFIG_NUMA | |
3788 | &defrag_ratio_attr.attr, | |
3789 | #endif | |
3790 | NULL | |
3791 | }; | |
3792 | ||
3793 | static struct attribute_group slab_attr_group = { | |
3794 | .attrs = slab_attrs, | |
3795 | }; | |
3796 | ||
3797 | static ssize_t slab_attr_show(struct kobject *kobj, | |
3798 | struct attribute *attr, | |
3799 | char *buf) | |
3800 | { | |
3801 | struct slab_attribute *attribute; | |
3802 | struct kmem_cache *s; | |
3803 | int err; | |
3804 | ||
3805 | attribute = to_slab_attr(attr); | |
3806 | s = to_slab(kobj); | |
3807 | ||
3808 | if (!attribute->show) | |
3809 | return -EIO; | |
3810 | ||
3811 | err = attribute->show(s, buf); | |
3812 | ||
3813 | return err; | |
3814 | } | |
3815 | ||
3816 | static ssize_t slab_attr_store(struct kobject *kobj, | |
3817 | struct attribute *attr, | |
3818 | const char *buf, size_t len) | |
3819 | { | |
3820 | struct slab_attribute *attribute; | |
3821 | struct kmem_cache *s; | |
3822 | int err; | |
3823 | ||
3824 | attribute = to_slab_attr(attr); | |
3825 | s = to_slab(kobj); | |
3826 | ||
3827 | if (!attribute->store) | |
3828 | return -EIO; | |
3829 | ||
3830 | err = attribute->store(s, buf, len); | |
3831 | ||
3832 | return err; | |
3833 | } | |
3834 | ||
3835 | static struct sysfs_ops slab_sysfs_ops = { | |
3836 | .show = slab_attr_show, | |
3837 | .store = slab_attr_store, | |
3838 | }; | |
3839 | ||
3840 | static struct kobj_type slab_ktype = { | |
3841 | .sysfs_ops = &slab_sysfs_ops, | |
3842 | }; | |
3843 | ||
3844 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
3845 | { | |
3846 | struct kobj_type *ktype = get_ktype(kobj); | |
3847 | ||
3848 | if (ktype == &slab_ktype) | |
3849 | return 1; | |
3850 | return 0; | |
3851 | } | |
3852 | ||
3853 | static struct kset_uevent_ops slab_uevent_ops = { | |
3854 | .filter = uevent_filter, | |
3855 | }; | |
3856 | ||
5af328a5 | 3857 | static decl_subsys(slab, &slab_ktype, &slab_uevent_ops); |
81819f0f CL |
3858 | |
3859 | #define ID_STR_LENGTH 64 | |
3860 | ||
3861 | /* Create a unique string id for a slab cache: | |
3862 | * format | |
3863 | * :[flags-]size:[memory address of kmemcache] | |
3864 | */ | |
3865 | static char *create_unique_id(struct kmem_cache *s) | |
3866 | { | |
3867 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
3868 | char *p = name; | |
3869 | ||
3870 | BUG_ON(!name); | |
3871 | ||
3872 | *p++ = ':'; | |
3873 | /* | |
3874 | * First flags affecting slabcache operations. We will only | |
3875 | * get here for aliasable slabs so we do not need to support | |
3876 | * too many flags. The flags here must cover all flags that | |
3877 | * are matched during merging to guarantee that the id is | |
3878 | * unique. | |
3879 | */ | |
3880 | if (s->flags & SLAB_CACHE_DMA) | |
3881 | *p++ = 'd'; | |
3882 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3883 | *p++ = 'a'; | |
3884 | if (s->flags & SLAB_DEBUG_FREE) | |
3885 | *p++ = 'F'; | |
3886 | if (p != name + 1) | |
3887 | *p++ = '-'; | |
3888 | p += sprintf(p, "%07d", s->size); | |
3889 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
3890 | return name; | |
3891 | } | |
3892 | ||
3893 | static int sysfs_slab_add(struct kmem_cache *s) | |
3894 | { | |
3895 | int err; | |
3896 | const char *name; | |
3897 | int unmergeable; | |
3898 | ||
3899 | if (slab_state < SYSFS) | |
3900 | /* Defer until later */ | |
3901 | return 0; | |
3902 | ||
3903 | unmergeable = slab_unmergeable(s); | |
3904 | if (unmergeable) { | |
3905 | /* | |
3906 | * Slabcache can never be merged so we can use the name proper. | |
3907 | * This is typically the case for debug situations. In that | |
3908 | * case we can catch duplicate names easily. | |
3909 | */ | |
0f9008ef | 3910 | sysfs_remove_link(&slab_subsys.kobj, s->name); |
81819f0f CL |
3911 | name = s->name; |
3912 | } else { | |
3913 | /* | |
3914 | * Create a unique name for the slab as a target | |
3915 | * for the symlinks. | |
3916 | */ | |
3917 | name = create_unique_id(s); | |
3918 | } | |
3919 | ||
3920 | kobj_set_kset_s(s, slab_subsys); | |
3921 | kobject_set_name(&s->kobj, name); | |
3922 | kobject_init(&s->kobj); | |
3923 | err = kobject_add(&s->kobj); | |
3924 | if (err) | |
3925 | return err; | |
3926 | ||
3927 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
3928 | if (err) | |
3929 | return err; | |
3930 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
3931 | if (!unmergeable) { | |
3932 | /* Setup first alias */ | |
3933 | sysfs_slab_alias(s, s->name); | |
3934 | kfree(name); | |
3935 | } | |
3936 | return 0; | |
3937 | } | |
3938 | ||
3939 | static void sysfs_slab_remove(struct kmem_cache *s) | |
3940 | { | |
3941 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
3942 | kobject_del(&s->kobj); | |
3943 | } | |
3944 | ||
3945 | /* | |
3946 | * Need to buffer aliases during bootup until sysfs becomes | |
3947 | * available lest we loose that information. | |
3948 | */ | |
3949 | struct saved_alias { | |
3950 | struct kmem_cache *s; | |
3951 | const char *name; | |
3952 | struct saved_alias *next; | |
3953 | }; | |
3954 | ||
5af328a5 | 3955 | static struct saved_alias *alias_list; |
81819f0f CL |
3956 | |
3957 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
3958 | { | |
3959 | struct saved_alias *al; | |
3960 | ||
3961 | if (slab_state == SYSFS) { | |
3962 | /* | |
3963 | * If we have a leftover link then remove it. | |
3964 | */ | |
0f9008ef LT |
3965 | sysfs_remove_link(&slab_subsys.kobj, name); |
3966 | return sysfs_create_link(&slab_subsys.kobj, | |
81819f0f CL |
3967 | &s->kobj, name); |
3968 | } | |
3969 | ||
3970 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
3971 | if (!al) | |
3972 | return -ENOMEM; | |
3973 | ||
3974 | al->s = s; | |
3975 | al->name = name; | |
3976 | al->next = alias_list; | |
3977 | alias_list = al; | |
3978 | return 0; | |
3979 | } | |
3980 | ||
3981 | static int __init slab_sysfs_init(void) | |
3982 | { | |
5b95a4ac | 3983 | struct kmem_cache *s; |
81819f0f CL |
3984 | int err; |
3985 | ||
3986 | err = subsystem_register(&slab_subsys); | |
3987 | if (err) { | |
3988 | printk(KERN_ERR "Cannot register slab subsystem.\n"); | |
3989 | return -ENOSYS; | |
3990 | } | |
3991 | ||
26a7bd03 CL |
3992 | slab_state = SYSFS; |
3993 | ||
5b95a4ac | 3994 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 3995 | err = sysfs_slab_add(s); |
5d540fb7 CL |
3996 | if (err) |
3997 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
3998 | " to sysfs\n", s->name); | |
26a7bd03 | 3999 | } |
81819f0f CL |
4000 | |
4001 | while (alias_list) { | |
4002 | struct saved_alias *al = alias_list; | |
4003 | ||
4004 | alias_list = alias_list->next; | |
4005 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4006 | if (err) |
4007 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4008 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4009 | kfree(al); |
4010 | } | |
4011 | ||
4012 | resiliency_test(); | |
4013 | return 0; | |
4014 | } | |
4015 | ||
4016 | __initcall(slab_sysfs_init); | |
81819f0f | 4017 | #endif |