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 | |
894b8788 CL |
93 | * lockless_freelist that allows lockless access to |
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 | * | |
143 | * - The per cpu array is updated for each new slab and and is a remote | |
144 | * cacheline for most nodes. This could become a bouncing cacheline given | |
672bba3a CL |
145 | * enough frequent updates. There are 16 pointers in a cacheline, so at |
146 | * max 16 cpus could compete for the cacheline which may be okay. | |
81819f0f CL |
147 | * |
148 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. | |
149 | * | |
81819f0f CL |
150 | * - Variable sizing of the per node arrays |
151 | */ | |
152 | ||
153 | /* Enable to test recovery from slab corruption on boot */ | |
154 | #undef SLUB_RESILIENCY_TEST | |
155 | ||
156 | #if PAGE_SHIFT <= 12 | |
157 | ||
158 | /* | |
159 | * Small page size. Make sure that we do not fragment memory | |
160 | */ | |
161 | #define DEFAULT_MAX_ORDER 1 | |
162 | #define DEFAULT_MIN_OBJECTS 4 | |
163 | ||
164 | #else | |
165 | ||
166 | /* | |
167 | * Large page machines are customarily able to handle larger | |
168 | * page orders. | |
169 | */ | |
170 | #define DEFAULT_MAX_ORDER 2 | |
171 | #define DEFAULT_MIN_OBJECTS 8 | |
172 | ||
173 | #endif | |
174 | ||
2086d26a CL |
175 | /* |
176 | * Mininum number of partial slabs. These will be left on the partial | |
177 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
178 | */ | |
e95eed57 CL |
179 | #define MIN_PARTIAL 2 |
180 | ||
2086d26a CL |
181 | /* |
182 | * Maximum number of desirable partial slabs. | |
183 | * The existence of more partial slabs makes kmem_cache_shrink | |
184 | * sort the partial list by the number of objects in the. | |
185 | */ | |
186 | #define MAX_PARTIAL 10 | |
187 | ||
81819f0f CL |
188 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
189 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 190 | |
81819f0f CL |
191 | /* |
192 | * Set of flags that will prevent slab merging | |
193 | */ | |
194 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
195 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
196 | ||
197 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
198 | SLAB_CACHE_DMA) | |
199 | ||
200 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 201 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
202 | #endif |
203 | ||
204 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 205 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
206 | #endif |
207 | ||
208 | /* Internal SLUB flags */ | |
209 | #define __OBJECT_POISON 0x80000000 /* Poison object */ | |
210 | ||
65c02d4c CL |
211 | /* Not all arches define cache_line_size */ |
212 | #ifndef cache_line_size | |
213 | #define cache_line_size() L1_CACHE_BYTES | |
214 | #endif | |
215 | ||
81819f0f CL |
216 | static int kmem_size = sizeof(struct kmem_cache); |
217 | ||
218 | #ifdef CONFIG_SMP | |
219 | static struct notifier_block slab_notifier; | |
220 | #endif | |
221 | ||
222 | static enum { | |
223 | DOWN, /* No slab functionality available */ | |
224 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 225 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
226 | SYSFS /* Sysfs up */ |
227 | } slab_state = DOWN; | |
228 | ||
229 | /* A list of all slab caches on the system */ | |
230 | static DECLARE_RWSEM(slub_lock); | |
231 | LIST_HEAD(slab_caches); | |
232 | ||
02cbc874 CL |
233 | /* |
234 | * Tracking user of a slab. | |
235 | */ | |
236 | struct track { | |
237 | void *addr; /* Called from address */ | |
238 | int cpu; /* Was running on cpu */ | |
239 | int pid; /* Pid context */ | |
240 | unsigned long when; /* When did the operation occur */ | |
241 | }; | |
242 | ||
243 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
244 | ||
41ecc55b | 245 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
246 | static int sysfs_slab_add(struct kmem_cache *); |
247 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
248 | static void sysfs_slab_remove(struct kmem_cache *); | |
249 | #else | |
250 | static int sysfs_slab_add(struct kmem_cache *s) { return 0; } | |
251 | static int sysfs_slab_alias(struct kmem_cache *s, const char *p) { return 0; } | |
252 | static void sysfs_slab_remove(struct kmem_cache *s) {} | |
253 | #endif | |
254 | ||
255 | /******************************************************************** | |
256 | * Core slab cache functions | |
257 | *******************************************************************/ | |
258 | ||
259 | int slab_is_available(void) | |
260 | { | |
261 | return slab_state >= UP; | |
262 | } | |
263 | ||
264 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
265 | { | |
266 | #ifdef CONFIG_NUMA | |
267 | return s->node[node]; | |
268 | #else | |
269 | return &s->local_node; | |
270 | #endif | |
271 | } | |
272 | ||
02cbc874 CL |
273 | static inline int check_valid_pointer(struct kmem_cache *s, |
274 | struct page *page, const void *object) | |
275 | { | |
276 | void *base; | |
277 | ||
278 | if (!object) | |
279 | return 1; | |
280 | ||
281 | base = page_address(page); | |
282 | if (object < base || object >= base + s->objects * s->size || | |
283 | (object - base) % s->size) { | |
284 | return 0; | |
285 | } | |
286 | ||
287 | return 1; | |
288 | } | |
289 | ||
7656c72b CL |
290 | /* |
291 | * Slow version of get and set free pointer. | |
292 | * | |
293 | * This version requires touching the cache lines of kmem_cache which | |
294 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
295 | * from the page struct. | |
296 | */ | |
297 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
298 | { | |
299 | return *(void **)(object + s->offset); | |
300 | } | |
301 | ||
302 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
303 | { | |
304 | *(void **)(object + s->offset) = fp; | |
305 | } | |
306 | ||
307 | /* Loop over all objects in a slab */ | |
308 | #define for_each_object(__p, __s, __addr) \ | |
309 | for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\ | |
310 | __p += (__s)->size) | |
311 | ||
312 | /* Scan freelist */ | |
313 | #define for_each_free_object(__p, __s, __free) \ | |
314 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) | |
315 | ||
316 | /* Determine object index from a given position */ | |
317 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
318 | { | |
319 | return (p - addr) / s->size; | |
320 | } | |
321 | ||
41ecc55b CL |
322 | #ifdef CONFIG_SLUB_DEBUG |
323 | /* | |
324 | * Debug settings: | |
325 | */ | |
326 | static int slub_debug; | |
327 | ||
328 | static char *slub_debug_slabs; | |
329 | ||
81819f0f CL |
330 | /* |
331 | * Object debugging | |
332 | */ | |
333 | static void print_section(char *text, u8 *addr, unsigned int length) | |
334 | { | |
335 | int i, offset; | |
336 | int newline = 1; | |
337 | char ascii[17]; | |
338 | ||
339 | ascii[16] = 0; | |
340 | ||
341 | for (i = 0; i < length; i++) { | |
342 | if (newline) { | |
343 | printk(KERN_ERR "%10s 0x%p: ", text, addr + i); | |
344 | newline = 0; | |
345 | } | |
346 | printk(" %02x", addr[i]); | |
347 | offset = i % 16; | |
348 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
349 | if (offset == 15) { | |
350 | printk(" %s\n",ascii); | |
351 | newline = 1; | |
352 | } | |
353 | } | |
354 | if (!newline) { | |
355 | i %= 16; | |
356 | while (i < 16) { | |
357 | printk(" "); | |
358 | ascii[i] = ' '; | |
359 | i++; | |
360 | } | |
361 | printk(" %s\n", ascii); | |
362 | } | |
363 | } | |
364 | ||
81819f0f CL |
365 | static struct track *get_track(struct kmem_cache *s, void *object, |
366 | enum track_item alloc) | |
367 | { | |
368 | struct track *p; | |
369 | ||
370 | if (s->offset) | |
371 | p = object + s->offset + sizeof(void *); | |
372 | else | |
373 | p = object + s->inuse; | |
374 | ||
375 | return p + alloc; | |
376 | } | |
377 | ||
378 | static void set_track(struct kmem_cache *s, void *object, | |
379 | enum track_item alloc, void *addr) | |
380 | { | |
381 | struct track *p; | |
382 | ||
383 | if (s->offset) | |
384 | p = object + s->offset + sizeof(void *); | |
385 | else | |
386 | p = object + s->inuse; | |
387 | ||
388 | p += alloc; | |
389 | if (addr) { | |
390 | p->addr = addr; | |
391 | p->cpu = smp_processor_id(); | |
392 | p->pid = current ? current->pid : -1; | |
393 | p->when = jiffies; | |
394 | } else | |
395 | memset(p, 0, sizeof(struct track)); | |
396 | } | |
397 | ||
81819f0f CL |
398 | static void init_tracking(struct kmem_cache *s, void *object) |
399 | { | |
400 | if (s->flags & SLAB_STORE_USER) { | |
401 | set_track(s, object, TRACK_FREE, NULL); | |
402 | set_track(s, object, TRACK_ALLOC, NULL); | |
403 | } | |
404 | } | |
405 | ||
406 | static void print_track(const char *s, struct track *t) | |
407 | { | |
408 | if (!t->addr) | |
409 | return; | |
410 | ||
411 | printk(KERN_ERR "%s: ", s); | |
412 | __print_symbol("%s", (unsigned long)t->addr); | |
413 | printk(" jiffies_ago=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); | |
414 | } | |
415 | ||
416 | static void print_trailer(struct kmem_cache *s, u8 *p) | |
417 | { | |
418 | unsigned int off; /* Offset of last byte */ | |
419 | ||
420 | if (s->flags & SLAB_RED_ZONE) | |
421 | print_section("Redzone", p + s->objsize, | |
422 | s->inuse - s->objsize); | |
423 | ||
424 | printk(KERN_ERR "FreePointer 0x%p -> 0x%p\n", | |
425 | p + s->offset, | |
426 | get_freepointer(s, p)); | |
427 | ||
428 | if (s->offset) | |
429 | off = s->offset + sizeof(void *); | |
430 | else | |
431 | off = s->inuse; | |
432 | ||
433 | if (s->flags & SLAB_STORE_USER) { | |
434 | print_track("Last alloc", get_track(s, p, TRACK_ALLOC)); | |
435 | print_track("Last free ", get_track(s, p, TRACK_FREE)); | |
436 | off += 2 * sizeof(struct track); | |
437 | } | |
438 | ||
439 | if (off != s->size) | |
440 | /* Beginning of the filler is the free pointer */ | |
441 | print_section("Filler", p + off, s->size - off); | |
442 | } | |
443 | ||
444 | static void object_err(struct kmem_cache *s, struct page *page, | |
445 | u8 *object, char *reason) | |
446 | { | |
447 | u8 *addr = page_address(page); | |
448 | ||
449 | printk(KERN_ERR "*** SLUB %s: %s@0x%p slab 0x%p\n", | |
450 | s->name, reason, object, page); | |
451 | printk(KERN_ERR " offset=%tu flags=0x%04lx inuse=%u freelist=0x%p\n", | |
452 | object - addr, page->flags, page->inuse, page->freelist); | |
453 | if (object > addr + 16) | |
454 | print_section("Bytes b4", object - 16, 16); | |
455 | print_section("Object", object, min(s->objsize, 128)); | |
456 | print_trailer(s, object); | |
457 | dump_stack(); | |
458 | } | |
459 | ||
460 | static void slab_err(struct kmem_cache *s, struct page *page, char *reason, ...) | |
461 | { | |
462 | va_list args; | |
463 | char buf[100]; | |
464 | ||
465 | va_start(args, reason); | |
466 | vsnprintf(buf, sizeof(buf), reason, args); | |
467 | va_end(args); | |
468 | printk(KERN_ERR "*** SLUB %s: %s in slab @0x%p\n", s->name, buf, | |
469 | page); | |
470 | dump_stack(); | |
471 | } | |
472 | ||
473 | static void init_object(struct kmem_cache *s, void *object, int active) | |
474 | { | |
475 | u8 *p = object; | |
476 | ||
477 | if (s->flags & __OBJECT_POISON) { | |
478 | memset(p, POISON_FREE, s->objsize - 1); | |
479 | p[s->objsize -1] = POISON_END; | |
480 | } | |
481 | ||
482 | if (s->flags & SLAB_RED_ZONE) | |
483 | memset(p + s->objsize, | |
484 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
485 | s->inuse - s->objsize); | |
486 | } | |
487 | ||
488 | static int check_bytes(u8 *start, unsigned int value, unsigned int bytes) | |
489 | { | |
490 | while (bytes) { | |
491 | if (*start != (u8)value) | |
492 | return 0; | |
493 | start++; | |
494 | bytes--; | |
495 | } | |
496 | return 1; | |
497 | } | |
498 | ||
81819f0f CL |
499 | /* |
500 | * Object layout: | |
501 | * | |
502 | * object address | |
503 | * Bytes of the object to be managed. | |
504 | * If the freepointer may overlay the object then the free | |
505 | * pointer is the first word of the object. | |
672bba3a | 506 | * |
81819f0f CL |
507 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
508 | * 0xa5 (POISON_END) | |
509 | * | |
510 | * object + s->objsize | |
511 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
512 | * Padding is extended by another word if Redzoning is enabled and |
513 | * objsize == inuse. | |
514 | * | |
81819f0f CL |
515 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
516 | * 0xcc (RED_ACTIVE) for objects in use. | |
517 | * | |
518 | * object + s->inuse | |
672bba3a CL |
519 | * Meta data starts here. |
520 | * | |
81819f0f CL |
521 | * A. Free pointer (if we cannot overwrite object on free) |
522 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a CL |
523 | * C. Padding to reach required alignment boundary or at mininum |
524 | * one word if debuggin is on to be able to detect writes | |
525 | * before the word boundary. | |
526 | * | |
527 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
528 | * |
529 | * object + s->size | |
672bba3a | 530 | * Nothing is used beyond s->size. |
81819f0f | 531 | * |
672bba3a CL |
532 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
533 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
534 | * may be used with merged slabcaches. |
535 | */ | |
536 | ||
537 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
538 | void *from, void *to) | |
539 | { | |
70d71228 | 540 | printk(KERN_ERR "@@@ SLUB %s: Restoring %s (0x%x) from 0x%p-0x%p\n", |
81819f0f CL |
541 | s->name, message, data, from, to - 1); |
542 | memset(from, data, to - from); | |
543 | } | |
544 | ||
545 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) | |
546 | { | |
547 | unsigned long off = s->inuse; /* The end of info */ | |
548 | ||
549 | if (s->offset) | |
550 | /* Freepointer is placed after the object. */ | |
551 | off += sizeof(void *); | |
552 | ||
553 | if (s->flags & SLAB_STORE_USER) | |
554 | /* We also have user information there */ | |
555 | off += 2 * sizeof(struct track); | |
556 | ||
557 | if (s->size == off) | |
558 | return 1; | |
559 | ||
560 | if (check_bytes(p + off, POISON_INUSE, s->size - off)) | |
561 | return 1; | |
562 | ||
563 | object_err(s, page, p, "Object padding check fails"); | |
564 | ||
565 | /* | |
566 | * Restore padding | |
567 | */ | |
568 | restore_bytes(s, "object padding", POISON_INUSE, p + off, p + s->size); | |
569 | return 0; | |
570 | } | |
571 | ||
572 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
573 | { | |
574 | u8 *p; | |
575 | int length, remainder; | |
576 | ||
577 | if (!(s->flags & SLAB_POISON)) | |
578 | return 1; | |
579 | ||
580 | p = page_address(page); | |
581 | length = s->objects * s->size; | |
582 | remainder = (PAGE_SIZE << s->order) - length; | |
583 | if (!remainder) | |
584 | return 1; | |
585 | ||
586 | if (!check_bytes(p + length, POISON_INUSE, remainder)) { | |
70d71228 | 587 | slab_err(s, page, "Padding check failed"); |
81819f0f CL |
588 | restore_bytes(s, "slab padding", POISON_INUSE, p + length, |
589 | p + length + remainder); | |
590 | return 0; | |
591 | } | |
592 | return 1; | |
593 | } | |
594 | ||
595 | static int check_object(struct kmem_cache *s, struct page *page, | |
596 | void *object, int active) | |
597 | { | |
598 | u8 *p = object; | |
599 | u8 *endobject = object + s->objsize; | |
600 | ||
601 | if (s->flags & SLAB_RED_ZONE) { | |
602 | unsigned int red = | |
603 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
604 | ||
605 | if (!check_bytes(endobject, red, s->inuse - s->objsize)) { | |
606 | object_err(s, page, object, | |
607 | active ? "Redzone Active" : "Redzone Inactive"); | |
608 | restore_bytes(s, "redzone", red, | |
609 | endobject, object + s->inuse); | |
610 | return 0; | |
611 | } | |
612 | } else { | |
613 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse && | |
614 | !check_bytes(endobject, POISON_INUSE, | |
615 | s->inuse - s->objsize)) { | |
616 | object_err(s, page, p, "Alignment padding check fails"); | |
617 | /* | |
618 | * Fix it so that there will not be another report. | |
619 | * | |
620 | * Hmmm... We may be corrupting an object that now expects | |
621 | * to be longer than allowed. | |
622 | */ | |
623 | restore_bytes(s, "alignment padding", POISON_INUSE, | |
624 | endobject, object + s->inuse); | |
625 | } | |
626 | } | |
627 | ||
628 | if (s->flags & SLAB_POISON) { | |
629 | if (!active && (s->flags & __OBJECT_POISON) && | |
630 | (!check_bytes(p, POISON_FREE, s->objsize - 1) || | |
631 | p[s->objsize - 1] != POISON_END)) { | |
632 | ||
633 | object_err(s, page, p, "Poison check failed"); | |
634 | restore_bytes(s, "Poison", POISON_FREE, | |
635 | p, p + s->objsize -1); | |
636 | restore_bytes(s, "Poison", POISON_END, | |
637 | p + s->objsize - 1, p + s->objsize); | |
638 | return 0; | |
639 | } | |
640 | /* | |
641 | * check_pad_bytes cleans up on its own. | |
642 | */ | |
643 | check_pad_bytes(s, page, p); | |
644 | } | |
645 | ||
646 | if (!s->offset && active) | |
647 | /* | |
648 | * Object and freepointer overlap. Cannot check | |
649 | * freepointer while object is allocated. | |
650 | */ | |
651 | return 1; | |
652 | ||
653 | /* Check free pointer validity */ | |
654 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
655 | object_err(s, page, p, "Freepointer corrupt"); | |
656 | /* | |
657 | * No choice but to zap it and thus loose the remainder | |
658 | * of the free objects in this slab. May cause | |
672bba3a | 659 | * another error because the object count is now wrong. |
81819f0f CL |
660 | */ |
661 | set_freepointer(s, p, NULL); | |
662 | return 0; | |
663 | } | |
664 | return 1; | |
665 | } | |
666 | ||
667 | static int check_slab(struct kmem_cache *s, struct page *page) | |
668 | { | |
669 | VM_BUG_ON(!irqs_disabled()); | |
670 | ||
671 | if (!PageSlab(page)) { | |
70d71228 CL |
672 | slab_err(s, page, "Not a valid slab page flags=%lx " |
673 | "mapping=0x%p count=%d", page->flags, page->mapping, | |
81819f0f CL |
674 | page_count(page)); |
675 | return 0; | |
676 | } | |
677 | if (page->offset * sizeof(void *) != s->offset) { | |
70d71228 CL |
678 | slab_err(s, page, "Corrupted offset %lu flags=0x%lx " |
679 | "mapping=0x%p count=%d", | |
81819f0f | 680 | (unsigned long)(page->offset * sizeof(void *)), |
81819f0f CL |
681 | page->flags, |
682 | page->mapping, | |
683 | page_count(page)); | |
81819f0f CL |
684 | return 0; |
685 | } | |
686 | if (page->inuse > s->objects) { | |
70d71228 CL |
687 | slab_err(s, page, "inuse %u > max %u @0x%p flags=%lx " |
688 | "mapping=0x%p count=%d", | |
689 | s->name, page->inuse, s->objects, page->flags, | |
81819f0f | 690 | page->mapping, page_count(page)); |
81819f0f CL |
691 | return 0; |
692 | } | |
693 | /* Slab_pad_check fixes things up after itself */ | |
694 | slab_pad_check(s, page); | |
695 | return 1; | |
696 | } | |
697 | ||
698 | /* | |
672bba3a CL |
699 | * Determine if a certain object on a page is on the freelist. Must hold the |
700 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
701 | */ |
702 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
703 | { | |
704 | int nr = 0; | |
705 | void *fp = page->freelist; | |
706 | void *object = NULL; | |
707 | ||
708 | while (fp && nr <= s->objects) { | |
709 | if (fp == search) | |
710 | return 1; | |
711 | if (!check_valid_pointer(s, page, fp)) { | |
712 | if (object) { | |
713 | object_err(s, page, object, | |
714 | "Freechain corrupt"); | |
715 | set_freepointer(s, object, NULL); | |
716 | break; | |
717 | } else { | |
70d71228 CL |
718 | slab_err(s, page, "Freepointer 0x%p corrupt", |
719 | fp); | |
81819f0f CL |
720 | page->freelist = NULL; |
721 | page->inuse = s->objects; | |
70d71228 CL |
722 | printk(KERN_ERR "@@@ SLUB %s: Freelist " |
723 | "cleared. Slab 0x%p\n", | |
724 | s->name, page); | |
81819f0f CL |
725 | return 0; |
726 | } | |
727 | break; | |
728 | } | |
729 | object = fp; | |
730 | fp = get_freepointer(s, object); | |
731 | nr++; | |
732 | } | |
733 | ||
734 | if (page->inuse != s->objects - nr) { | |
70d71228 CL |
735 | slab_err(s, page, "Wrong object count. Counter is %d but " |
736 | "counted were %d", s, page, page->inuse, | |
737 | s->objects - nr); | |
81819f0f | 738 | page->inuse = s->objects - nr; |
70d71228 CL |
739 | printk(KERN_ERR "@@@ SLUB %s: Object count adjusted. " |
740 | "Slab @0x%p\n", s->name, page); | |
81819f0f CL |
741 | } |
742 | return search == NULL; | |
743 | } | |
744 | ||
3ec09742 CL |
745 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
746 | { | |
747 | if (s->flags & SLAB_TRACE) { | |
748 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
749 | s->name, | |
750 | alloc ? "alloc" : "free", | |
751 | object, page->inuse, | |
752 | page->freelist); | |
753 | ||
754 | if (!alloc) | |
755 | print_section("Object", (void *)object, s->objsize); | |
756 | ||
757 | dump_stack(); | |
758 | } | |
759 | } | |
760 | ||
643b1138 | 761 | /* |
672bba3a | 762 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 763 | */ |
e95eed57 | 764 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 765 | { |
643b1138 CL |
766 | spin_lock(&n->list_lock); |
767 | list_add(&page->lru, &n->full); | |
768 | spin_unlock(&n->list_lock); | |
769 | } | |
770 | ||
771 | static void remove_full(struct kmem_cache *s, struct page *page) | |
772 | { | |
773 | struct kmem_cache_node *n; | |
774 | ||
775 | if (!(s->flags & SLAB_STORE_USER)) | |
776 | return; | |
777 | ||
778 | n = get_node(s, page_to_nid(page)); | |
779 | ||
780 | spin_lock(&n->list_lock); | |
781 | list_del(&page->lru); | |
782 | spin_unlock(&n->list_lock); | |
783 | } | |
784 | ||
3ec09742 CL |
785 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
786 | void *object) | |
787 | { | |
788 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
789 | return; | |
790 | ||
791 | init_object(s, object, 0); | |
792 | init_tracking(s, object); | |
793 | } | |
794 | ||
795 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
796 | void *object, void *addr) | |
81819f0f CL |
797 | { |
798 | if (!check_slab(s, page)) | |
799 | goto bad; | |
800 | ||
801 | if (object && !on_freelist(s, page, object)) { | |
70d71228 CL |
802 | slab_err(s, page, "Object 0x%p already allocated", object); |
803 | goto bad; | |
81819f0f CL |
804 | } |
805 | ||
806 | if (!check_valid_pointer(s, page, object)) { | |
807 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 808 | goto bad; |
81819f0f CL |
809 | } |
810 | ||
3ec09742 | 811 | if (object && !check_object(s, page, object, 0)) |
81819f0f | 812 | goto bad; |
81819f0f | 813 | |
3ec09742 CL |
814 | /* Success perform special debug activities for allocs */ |
815 | if (s->flags & SLAB_STORE_USER) | |
816 | set_track(s, object, TRACK_ALLOC, addr); | |
817 | trace(s, page, object, 1); | |
818 | init_object(s, object, 1); | |
81819f0f | 819 | return 1; |
3ec09742 | 820 | |
81819f0f CL |
821 | bad: |
822 | if (PageSlab(page)) { | |
823 | /* | |
824 | * If this is a slab page then lets do the best we can | |
825 | * to avoid issues in the future. Marking all objects | |
672bba3a | 826 | * as used avoids touching the remaining objects. |
81819f0f CL |
827 | */ |
828 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p. Marking all objects used.\n", | |
829 | s->name, page); | |
830 | page->inuse = s->objects; | |
831 | page->freelist = NULL; | |
832 | /* Fix up fields that may be corrupted */ | |
833 | page->offset = s->offset / sizeof(void *); | |
834 | } | |
835 | return 0; | |
836 | } | |
837 | ||
3ec09742 CL |
838 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
839 | void *object, void *addr) | |
81819f0f CL |
840 | { |
841 | if (!check_slab(s, page)) | |
842 | goto fail; | |
843 | ||
844 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 845 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
846 | goto fail; |
847 | } | |
848 | ||
849 | if (on_freelist(s, page, object)) { | |
70d71228 | 850 | slab_err(s, page, "Object 0x%p already free", object); |
81819f0f CL |
851 | goto fail; |
852 | } | |
853 | ||
854 | if (!check_object(s, page, object, 1)) | |
855 | return 0; | |
856 | ||
857 | if (unlikely(s != page->slab)) { | |
858 | if (!PageSlab(page)) | |
70d71228 CL |
859 | slab_err(s, page, "Attempt to free object(0x%p) " |
860 | "outside of slab", object); | |
81819f0f | 861 | else |
70d71228 | 862 | if (!page->slab) { |
81819f0f | 863 | printk(KERN_ERR |
70d71228 | 864 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 865 | object); |
70d71228 CL |
866 | dump_stack(); |
867 | } | |
81819f0f | 868 | else |
70d71228 CL |
869 | slab_err(s, page, "object at 0x%p belongs " |
870 | "to slab %s", object, page->slab->name); | |
81819f0f CL |
871 | goto fail; |
872 | } | |
3ec09742 CL |
873 | |
874 | /* Special debug activities for freeing objects */ | |
875 | if (!SlabFrozen(page) && !page->freelist) | |
876 | remove_full(s, page); | |
877 | if (s->flags & SLAB_STORE_USER) | |
878 | set_track(s, object, TRACK_FREE, addr); | |
879 | trace(s, page, object, 0); | |
880 | init_object(s, object, 0); | |
81819f0f | 881 | return 1; |
3ec09742 | 882 | |
81819f0f | 883 | fail: |
81819f0f CL |
884 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p object at 0x%p not freed.\n", |
885 | s->name, page, object); | |
886 | return 0; | |
887 | } | |
888 | ||
41ecc55b CL |
889 | static int __init setup_slub_debug(char *str) |
890 | { | |
891 | if (!str || *str != '=') | |
892 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
893 | else { | |
894 | str++; | |
895 | if (*str == 0 || *str == ',') | |
896 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
897 | else | |
898 | for( ;*str && *str != ','; str++) | |
899 | switch (*str) { | |
900 | case 'f' : case 'F' : | |
901 | slub_debug |= SLAB_DEBUG_FREE; | |
902 | break; | |
903 | case 'z' : case 'Z' : | |
904 | slub_debug |= SLAB_RED_ZONE; | |
905 | break; | |
906 | case 'p' : case 'P' : | |
907 | slub_debug |= SLAB_POISON; | |
908 | break; | |
909 | case 'u' : case 'U' : | |
910 | slub_debug |= SLAB_STORE_USER; | |
911 | break; | |
912 | case 't' : case 'T' : | |
913 | slub_debug |= SLAB_TRACE; | |
914 | break; | |
915 | default: | |
916 | printk(KERN_ERR "slub_debug option '%c' " | |
917 | "unknown. skipped\n",*str); | |
918 | } | |
919 | } | |
920 | ||
921 | if (*str == ',') | |
922 | slub_debug_slabs = str + 1; | |
923 | return 1; | |
924 | } | |
925 | ||
926 | __setup("slub_debug", setup_slub_debug); | |
927 | ||
928 | static void kmem_cache_open_debug_check(struct kmem_cache *s) | |
929 | { | |
930 | /* | |
931 | * The page->offset field is only 16 bit wide. This is an offset | |
932 | * in units of words from the beginning of an object. If the slab | |
933 | * size is bigger then we cannot move the free pointer behind the | |
934 | * object anymore. | |
935 | * | |
936 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
937 | * the limit is 512k. | |
938 | * | |
c59def9f | 939 | * Debugging or ctor may create a need to move the free |
41ecc55b CL |
940 | * pointer. Fail if this happens. |
941 | */ | |
33e9e241 | 942 | if (s->objsize >= 65535 * sizeof(void *)) { |
41ecc55b CL |
943 | BUG_ON(s->flags & (SLAB_RED_ZONE | SLAB_POISON | |
944 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); | |
c59def9f | 945 | BUG_ON(s->ctor); |
41ecc55b CL |
946 | } |
947 | else | |
948 | /* | |
949 | * Enable debugging if selected on the kernel commandline. | |
950 | */ | |
951 | if (slub_debug && (!slub_debug_slabs || | |
952 | strncmp(slub_debug_slabs, s->name, | |
953 | strlen(slub_debug_slabs)) == 0)) | |
954 | s->flags |= slub_debug; | |
955 | } | |
956 | #else | |
3ec09742 CL |
957 | static inline void setup_object_debug(struct kmem_cache *s, |
958 | struct page *page, void *object) {} | |
41ecc55b | 959 | |
3ec09742 CL |
960 | static inline int alloc_debug_processing(struct kmem_cache *s, |
961 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 962 | |
3ec09742 CL |
963 | static inline int free_debug_processing(struct kmem_cache *s, |
964 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 965 | |
41ecc55b CL |
966 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
967 | { return 1; } | |
968 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
969 | void *object, int active) { return 1; } | |
3ec09742 | 970 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
41ecc55b CL |
971 | static inline void kmem_cache_open_debug_check(struct kmem_cache *s) {} |
972 | #define slub_debug 0 | |
973 | #endif | |
81819f0f CL |
974 | /* |
975 | * Slab allocation and freeing | |
976 | */ | |
977 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
978 | { | |
979 | struct page * page; | |
980 | int pages = 1 << s->order; | |
981 | ||
982 | if (s->order) | |
983 | flags |= __GFP_COMP; | |
984 | ||
985 | if (s->flags & SLAB_CACHE_DMA) | |
986 | flags |= SLUB_DMA; | |
987 | ||
988 | if (node == -1) | |
989 | page = alloc_pages(flags, s->order); | |
990 | else | |
991 | page = alloc_pages_node(node, flags, s->order); | |
992 | ||
993 | if (!page) | |
994 | return NULL; | |
995 | ||
996 | mod_zone_page_state(page_zone(page), | |
997 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
998 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
999 | pages); | |
1000 | ||
1001 | return page; | |
1002 | } | |
1003 | ||
1004 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1005 | void *object) | |
1006 | { | |
3ec09742 | 1007 | setup_object_debug(s, page, object); |
4f104934 | 1008 | if (unlikely(s->ctor)) |
a35afb83 | 1009 | s->ctor(object, s, 0); |
81819f0f CL |
1010 | } |
1011 | ||
1012 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1013 | { | |
1014 | struct page *page; | |
1015 | struct kmem_cache_node *n; | |
1016 | void *start; | |
1017 | void *end; | |
1018 | void *last; | |
1019 | void *p; | |
1020 | ||
81819f0f CL |
1021 | BUG_ON(flags & ~(GFP_DMA | GFP_LEVEL_MASK)); |
1022 | ||
1023 | if (flags & __GFP_WAIT) | |
1024 | local_irq_enable(); | |
1025 | ||
1026 | page = allocate_slab(s, flags & GFP_LEVEL_MASK, node); | |
1027 | if (!page) | |
1028 | goto out; | |
1029 | ||
1030 | n = get_node(s, page_to_nid(page)); | |
1031 | if (n) | |
1032 | atomic_long_inc(&n->nr_slabs); | |
1033 | page->offset = s->offset / sizeof(void *); | |
1034 | page->slab = s; | |
1035 | page->flags |= 1 << PG_slab; | |
1036 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1037 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1038 | SetSlabDebug(page); |
81819f0f CL |
1039 | |
1040 | start = page_address(page); | |
1041 | end = start + s->objects * s->size; | |
1042 | ||
1043 | if (unlikely(s->flags & SLAB_POISON)) | |
1044 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
1045 | ||
1046 | last = start; | |
7656c72b | 1047 | for_each_object(p, s, start) { |
81819f0f CL |
1048 | setup_object(s, page, last); |
1049 | set_freepointer(s, last, p); | |
1050 | last = p; | |
1051 | } | |
1052 | setup_object(s, page, last); | |
1053 | set_freepointer(s, last, NULL); | |
1054 | ||
1055 | page->freelist = start; | |
894b8788 | 1056 | page->lockless_freelist = NULL; |
81819f0f CL |
1057 | page->inuse = 0; |
1058 | out: | |
1059 | if (flags & __GFP_WAIT) | |
1060 | local_irq_disable(); | |
1061 | return page; | |
1062 | } | |
1063 | ||
1064 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1065 | { | |
1066 | int pages = 1 << s->order; | |
1067 | ||
c59def9f | 1068 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1069 | void *p; |
1070 | ||
1071 | slab_pad_check(s, page); | |
c59def9f | 1072 | for_each_object(p, s, page_address(page)) |
81819f0f | 1073 | check_object(s, page, p, 0); |
81819f0f CL |
1074 | } |
1075 | ||
1076 | mod_zone_page_state(page_zone(page), | |
1077 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1078 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1079 | - pages); | |
1080 | ||
1081 | page->mapping = NULL; | |
1082 | __free_pages(page, s->order); | |
1083 | } | |
1084 | ||
1085 | static void rcu_free_slab(struct rcu_head *h) | |
1086 | { | |
1087 | struct page *page; | |
1088 | ||
1089 | page = container_of((struct list_head *)h, struct page, lru); | |
1090 | __free_slab(page->slab, page); | |
1091 | } | |
1092 | ||
1093 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1094 | { | |
1095 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1096 | /* | |
1097 | * RCU free overloads the RCU head over the LRU | |
1098 | */ | |
1099 | struct rcu_head *head = (void *)&page->lru; | |
1100 | ||
1101 | call_rcu(head, rcu_free_slab); | |
1102 | } else | |
1103 | __free_slab(s, page); | |
1104 | } | |
1105 | ||
1106 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1107 | { | |
1108 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1109 | ||
1110 | atomic_long_dec(&n->nr_slabs); | |
1111 | reset_page_mapcount(page); | |
35e5d7ee CL |
1112 | ClearSlabDebug(page); |
1113 | __ClearPageSlab(page); | |
81819f0f CL |
1114 | free_slab(s, page); |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * Per slab locking using the pagelock | |
1119 | */ | |
1120 | static __always_inline void slab_lock(struct page *page) | |
1121 | { | |
1122 | bit_spin_lock(PG_locked, &page->flags); | |
1123 | } | |
1124 | ||
1125 | static __always_inline void slab_unlock(struct page *page) | |
1126 | { | |
1127 | bit_spin_unlock(PG_locked, &page->flags); | |
1128 | } | |
1129 | ||
1130 | static __always_inline int slab_trylock(struct page *page) | |
1131 | { | |
1132 | int rc = 1; | |
1133 | ||
1134 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1135 | return rc; | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Management of partially allocated slabs | |
1140 | */ | |
e95eed57 | 1141 | static void add_partial_tail(struct kmem_cache_node *n, struct page *page) |
81819f0f | 1142 | { |
e95eed57 CL |
1143 | spin_lock(&n->list_lock); |
1144 | n->nr_partial++; | |
1145 | list_add_tail(&page->lru, &n->partial); | |
1146 | spin_unlock(&n->list_lock); | |
1147 | } | |
81819f0f | 1148 | |
e95eed57 CL |
1149 | static void add_partial(struct kmem_cache_node *n, struct page *page) |
1150 | { | |
81819f0f CL |
1151 | spin_lock(&n->list_lock); |
1152 | n->nr_partial++; | |
1153 | list_add(&page->lru, &n->partial); | |
1154 | spin_unlock(&n->list_lock); | |
1155 | } | |
1156 | ||
1157 | static void remove_partial(struct kmem_cache *s, | |
1158 | struct page *page) | |
1159 | { | |
1160 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1161 | ||
1162 | spin_lock(&n->list_lock); | |
1163 | list_del(&page->lru); | |
1164 | n->nr_partial--; | |
1165 | spin_unlock(&n->list_lock); | |
1166 | } | |
1167 | ||
1168 | /* | |
672bba3a | 1169 | * Lock slab and remove from the partial list. |
81819f0f | 1170 | * |
672bba3a | 1171 | * Must hold list_lock. |
81819f0f | 1172 | */ |
4b6f0750 | 1173 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1174 | { |
1175 | if (slab_trylock(page)) { | |
1176 | list_del(&page->lru); | |
1177 | n->nr_partial--; | |
4b6f0750 | 1178 | SetSlabFrozen(page); |
81819f0f CL |
1179 | return 1; |
1180 | } | |
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | /* | |
672bba3a | 1185 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1186 | */ |
1187 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1188 | { | |
1189 | struct page *page; | |
1190 | ||
1191 | /* | |
1192 | * Racy check. If we mistakenly see no partial slabs then we | |
1193 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1194 | * partial slab and there is none available then get_partials() |
1195 | * will return NULL. | |
81819f0f CL |
1196 | */ |
1197 | if (!n || !n->nr_partial) | |
1198 | return NULL; | |
1199 | ||
1200 | spin_lock(&n->list_lock); | |
1201 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1202 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1203 | goto out; |
1204 | page = NULL; | |
1205 | out: | |
1206 | spin_unlock(&n->list_lock); | |
1207 | return page; | |
1208 | } | |
1209 | ||
1210 | /* | |
672bba3a | 1211 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1212 | */ |
1213 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1214 | { | |
1215 | #ifdef CONFIG_NUMA | |
1216 | struct zonelist *zonelist; | |
1217 | struct zone **z; | |
1218 | struct page *page; | |
1219 | ||
1220 | /* | |
672bba3a CL |
1221 | * The defrag ratio allows a configuration of the tradeoffs between |
1222 | * inter node defragmentation and node local allocations. A lower | |
1223 | * defrag_ratio increases the tendency to do local allocations | |
1224 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1225 | * |
672bba3a CL |
1226 | * If the defrag_ratio is set to 0 then kmalloc() always |
1227 | * returns node local objects. If the ratio is higher then kmalloc() | |
1228 | * may return off node objects because partial slabs are obtained | |
1229 | * from other nodes and filled up. | |
81819f0f CL |
1230 | * |
1231 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
672bba3a CL |
1232 | * defrag_ratio = 1000) then every (well almost) allocation will |
1233 | * first attempt to defrag slab caches on other nodes. This means | |
1234 | * scanning over all nodes to look for partial slabs which may be | |
1235 | * expensive if we do it every time we are trying to find a slab | |
1236 | * with available objects. | |
81819f0f CL |
1237 | */ |
1238 | if (!s->defrag_ratio || get_cycles() % 1024 > s->defrag_ratio) | |
1239 | return NULL; | |
1240 | ||
1241 | zonelist = &NODE_DATA(slab_node(current->mempolicy)) | |
1242 | ->node_zonelists[gfp_zone(flags)]; | |
1243 | for (z = zonelist->zones; *z; z++) { | |
1244 | struct kmem_cache_node *n; | |
1245 | ||
1246 | n = get_node(s, zone_to_nid(*z)); | |
1247 | ||
1248 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1249 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1250 | page = get_partial_node(n); |
1251 | if (page) | |
1252 | return page; | |
1253 | } | |
1254 | } | |
1255 | #endif | |
1256 | return NULL; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Get a partial page, lock it and return it. | |
1261 | */ | |
1262 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1263 | { | |
1264 | struct page *page; | |
1265 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1266 | ||
1267 | page = get_partial_node(get_node(s, searchnode)); | |
1268 | if (page || (flags & __GFP_THISNODE)) | |
1269 | return page; | |
1270 | ||
1271 | return get_any_partial(s, flags); | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * Move a page back to the lists. | |
1276 | * | |
1277 | * Must be called with the slab lock held. | |
1278 | * | |
1279 | * On exit the slab lock will have been dropped. | |
1280 | */ | |
4b6f0750 | 1281 | static void unfreeze_slab(struct kmem_cache *s, struct page *page) |
81819f0f | 1282 | { |
e95eed57 CL |
1283 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1284 | ||
4b6f0750 | 1285 | ClearSlabFrozen(page); |
81819f0f | 1286 | if (page->inuse) { |
e95eed57 | 1287 | |
81819f0f | 1288 | if (page->freelist) |
e95eed57 | 1289 | add_partial(n, page); |
35e5d7ee | 1290 | else if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) |
e95eed57 | 1291 | add_full(n, page); |
81819f0f | 1292 | slab_unlock(page); |
e95eed57 | 1293 | |
81819f0f | 1294 | } else { |
e95eed57 CL |
1295 | if (n->nr_partial < MIN_PARTIAL) { |
1296 | /* | |
672bba3a CL |
1297 | * Adding an empty slab to the partial slabs in order |
1298 | * to avoid page allocator overhead. This slab needs | |
1299 | * to come after the other slabs with objects in | |
1300 | * order to fill them up. That way the size of the | |
1301 | * partial list stays small. kmem_cache_shrink can | |
1302 | * reclaim empty slabs from the partial list. | |
e95eed57 CL |
1303 | */ |
1304 | add_partial_tail(n, page); | |
1305 | slab_unlock(page); | |
1306 | } else { | |
1307 | slab_unlock(page); | |
1308 | discard_slab(s, page); | |
1309 | } | |
81819f0f CL |
1310 | } |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Remove the cpu slab | |
1315 | */ | |
1316 | static void deactivate_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1317 | { | |
894b8788 CL |
1318 | /* |
1319 | * Merge cpu freelist into freelist. Typically we get here | |
1320 | * because both freelists are empty. So this is unlikely | |
1321 | * to occur. | |
1322 | */ | |
1323 | while (unlikely(page->lockless_freelist)) { | |
1324 | void **object; | |
1325 | ||
1326 | /* Retrieve object from cpu_freelist */ | |
1327 | object = page->lockless_freelist; | |
1328 | page->lockless_freelist = page->lockless_freelist[page->offset]; | |
1329 | ||
1330 | /* And put onto the regular freelist */ | |
1331 | object[page->offset] = page->freelist; | |
1332 | page->freelist = object; | |
1333 | page->inuse--; | |
1334 | } | |
81819f0f | 1335 | s->cpu_slab[cpu] = NULL; |
4b6f0750 | 1336 | unfreeze_slab(s, page); |
81819f0f CL |
1337 | } |
1338 | ||
1339 | static void flush_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1340 | { | |
1341 | slab_lock(page); | |
1342 | deactivate_slab(s, page, cpu); | |
1343 | } | |
1344 | ||
1345 | /* | |
1346 | * Flush cpu slab. | |
1347 | * Called from IPI handler with interrupts disabled. | |
1348 | */ | |
1349 | static void __flush_cpu_slab(struct kmem_cache *s, int cpu) | |
1350 | { | |
1351 | struct page *page = s->cpu_slab[cpu]; | |
1352 | ||
1353 | if (likely(page)) | |
1354 | flush_slab(s, page, cpu); | |
1355 | } | |
1356 | ||
1357 | static void flush_cpu_slab(void *d) | |
1358 | { | |
1359 | struct kmem_cache *s = d; | |
1360 | int cpu = smp_processor_id(); | |
1361 | ||
1362 | __flush_cpu_slab(s, cpu); | |
1363 | } | |
1364 | ||
1365 | static void flush_all(struct kmem_cache *s) | |
1366 | { | |
1367 | #ifdef CONFIG_SMP | |
1368 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1369 | #else | |
1370 | unsigned long flags; | |
1371 | ||
1372 | local_irq_save(flags); | |
1373 | flush_cpu_slab(s); | |
1374 | local_irq_restore(flags); | |
1375 | #endif | |
1376 | } | |
1377 | ||
1378 | /* | |
894b8788 CL |
1379 | * Slow path. The lockless freelist is empty or we need to perform |
1380 | * debugging duties. | |
1381 | * | |
1382 | * Interrupts are disabled. | |
81819f0f | 1383 | * |
894b8788 CL |
1384 | * Processing is still very fast if new objects have been freed to the |
1385 | * regular freelist. In that case we simply take over the regular freelist | |
1386 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1387 | * |
894b8788 CL |
1388 | * If that is not working then we fall back to the partial lists. We take the |
1389 | * first element of the freelist as the object to allocate now and move the | |
1390 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1391 | * |
894b8788 CL |
1392 | * And if we were unable to get a new slab from the partial slab lists then |
1393 | * we need to allocate a new slab. This is slowest path since we may sleep. | |
81819f0f | 1394 | */ |
894b8788 CL |
1395 | static void *__slab_alloc(struct kmem_cache *s, |
1396 | gfp_t gfpflags, int node, void *addr, struct page *page) | |
81819f0f | 1397 | { |
81819f0f | 1398 | void **object; |
894b8788 | 1399 | int cpu = smp_processor_id(); |
81819f0f | 1400 | |
81819f0f CL |
1401 | if (!page) |
1402 | goto new_slab; | |
1403 | ||
1404 | slab_lock(page); | |
1405 | if (unlikely(node != -1 && page_to_nid(page) != node)) | |
1406 | goto another_slab; | |
894b8788 | 1407 | load_freelist: |
81819f0f CL |
1408 | object = page->freelist; |
1409 | if (unlikely(!object)) | |
1410 | goto another_slab; | |
35e5d7ee | 1411 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1412 | goto debug; |
1413 | ||
894b8788 CL |
1414 | object = page->freelist; |
1415 | page->lockless_freelist = object[page->offset]; | |
1416 | page->inuse = s->objects; | |
1417 | page->freelist = NULL; | |
81819f0f | 1418 | slab_unlock(page); |
81819f0f CL |
1419 | return object; |
1420 | ||
1421 | another_slab: | |
1422 | deactivate_slab(s, page, cpu); | |
1423 | ||
1424 | new_slab: | |
1425 | page = get_partial(s, gfpflags, node); | |
894b8788 | 1426 | if (page) { |
81819f0f | 1427 | s->cpu_slab[cpu] = page; |
894b8788 | 1428 | goto load_freelist; |
81819f0f CL |
1429 | } |
1430 | ||
1431 | page = new_slab(s, gfpflags, node); | |
1432 | if (page) { | |
1433 | cpu = smp_processor_id(); | |
1434 | if (s->cpu_slab[cpu]) { | |
1435 | /* | |
672bba3a CL |
1436 | * Someone else populated the cpu_slab while we |
1437 | * enabled interrupts, or we have gotten scheduled | |
1438 | * on another cpu. The page may not be on the | |
1439 | * requested node even if __GFP_THISNODE was | |
1440 | * specified. So we need to recheck. | |
81819f0f CL |
1441 | */ |
1442 | if (node == -1 || | |
1443 | page_to_nid(s->cpu_slab[cpu]) == node) { | |
1444 | /* | |
1445 | * Current cpuslab is acceptable and we | |
1446 | * want the current one since its cache hot | |
1447 | */ | |
1448 | discard_slab(s, page); | |
1449 | page = s->cpu_slab[cpu]; | |
1450 | slab_lock(page); | |
894b8788 | 1451 | goto load_freelist; |
81819f0f | 1452 | } |
672bba3a | 1453 | /* New slab does not fit our expectations */ |
81819f0f CL |
1454 | flush_slab(s, s->cpu_slab[cpu], cpu); |
1455 | } | |
1456 | slab_lock(page); | |
4b6f0750 CL |
1457 | SetSlabFrozen(page); |
1458 | s->cpu_slab[cpu] = page; | |
1459 | goto load_freelist; | |
81819f0f | 1460 | } |
81819f0f CL |
1461 | return NULL; |
1462 | debug: | |
894b8788 | 1463 | object = page->freelist; |
3ec09742 | 1464 | if (!alloc_debug_processing(s, page, object, addr)) |
81819f0f | 1465 | goto another_slab; |
894b8788 CL |
1466 | |
1467 | page->inuse++; | |
1468 | page->freelist = object[page->offset]; | |
1469 | slab_unlock(page); | |
1470 | return object; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1475 | * have the fastpath folded into their functions. So no function call | |
1476 | * overhead for requests that can be satisfied on the fastpath. | |
1477 | * | |
1478 | * The fastpath works by first checking if the lockless freelist can be used. | |
1479 | * If not then __slab_alloc is called for slow processing. | |
1480 | * | |
1481 | * Otherwise we can simply pick the next object from the lockless free list. | |
1482 | */ | |
1483 | static void __always_inline *slab_alloc(struct kmem_cache *s, | |
1484 | gfp_t gfpflags, int node, void *addr) | |
1485 | { | |
1486 | struct page *page; | |
1487 | void **object; | |
1488 | unsigned long flags; | |
1489 | ||
1490 | local_irq_save(flags); | |
1491 | page = s->cpu_slab[smp_processor_id()]; | |
1492 | if (unlikely(!page || !page->lockless_freelist || | |
1493 | (node != -1 && page_to_nid(page) != node))) | |
1494 | ||
1495 | object = __slab_alloc(s, gfpflags, node, addr, page); | |
1496 | ||
1497 | else { | |
1498 | object = page->lockless_freelist; | |
1499 | page->lockless_freelist = object[page->offset]; | |
1500 | } | |
1501 | local_irq_restore(flags); | |
1502 | return object; | |
81819f0f CL |
1503 | } |
1504 | ||
1505 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1506 | { | |
77c5e2d0 | 1507 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1508 | } |
1509 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1510 | ||
1511 | #ifdef CONFIG_NUMA | |
1512 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1513 | { | |
77c5e2d0 | 1514 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1515 | } |
1516 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1517 | #endif | |
1518 | ||
1519 | /* | |
894b8788 CL |
1520 | * Slow patch handling. This may still be called frequently since objects |
1521 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1522 | * |
894b8788 CL |
1523 | * So we still attempt to reduce cache line usage. Just take the slab |
1524 | * lock and free the item. If there is no additional partial page | |
1525 | * handling required then we can return immediately. | |
81819f0f | 1526 | */ |
894b8788 | 1527 | static void __slab_free(struct kmem_cache *s, struct page *page, |
77c5e2d0 | 1528 | void *x, void *addr) |
81819f0f CL |
1529 | { |
1530 | void *prior; | |
1531 | void **object = (void *)x; | |
81819f0f | 1532 | |
81819f0f CL |
1533 | slab_lock(page); |
1534 | ||
35e5d7ee | 1535 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1536 | goto debug; |
1537 | checks_ok: | |
1538 | prior = object[page->offset] = page->freelist; | |
1539 | page->freelist = object; | |
1540 | page->inuse--; | |
1541 | ||
4b6f0750 | 1542 | if (unlikely(SlabFrozen(page))) |
81819f0f CL |
1543 | goto out_unlock; |
1544 | ||
1545 | if (unlikely(!page->inuse)) | |
1546 | goto slab_empty; | |
1547 | ||
1548 | /* | |
1549 | * Objects left in the slab. If it | |
1550 | * was not on the partial list before | |
1551 | * then add it. | |
1552 | */ | |
1553 | if (unlikely(!prior)) | |
e95eed57 | 1554 | add_partial(get_node(s, page_to_nid(page)), page); |
81819f0f CL |
1555 | |
1556 | out_unlock: | |
1557 | slab_unlock(page); | |
81819f0f CL |
1558 | return; |
1559 | ||
1560 | slab_empty: | |
1561 | if (prior) | |
1562 | /* | |
672bba3a | 1563 | * Slab still on the partial list. |
81819f0f CL |
1564 | */ |
1565 | remove_partial(s, page); | |
1566 | ||
1567 | slab_unlock(page); | |
1568 | discard_slab(s, page); | |
81819f0f CL |
1569 | return; |
1570 | ||
1571 | debug: | |
3ec09742 | 1572 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1573 | goto out_unlock; |
77c5e2d0 | 1574 | goto checks_ok; |
81819f0f CL |
1575 | } |
1576 | ||
894b8788 CL |
1577 | /* |
1578 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1579 | * can perform fastpath freeing without additional function calls. | |
1580 | * | |
1581 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1582 | * of this processor. This typically the case if we have just allocated | |
1583 | * the item before. | |
1584 | * | |
1585 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1586 | * with all sorts of special processing. | |
1587 | */ | |
1588 | static void __always_inline slab_free(struct kmem_cache *s, | |
1589 | struct page *page, void *x, void *addr) | |
1590 | { | |
1591 | void **object = (void *)x; | |
1592 | unsigned long flags; | |
1593 | ||
1594 | local_irq_save(flags); | |
1595 | if (likely(page == s->cpu_slab[smp_processor_id()] && | |
1596 | !SlabDebug(page))) { | |
1597 | object[page->offset] = page->lockless_freelist; | |
1598 | page->lockless_freelist = object; | |
1599 | } else | |
1600 | __slab_free(s, page, x, addr); | |
1601 | ||
1602 | local_irq_restore(flags); | |
1603 | } | |
1604 | ||
81819f0f CL |
1605 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1606 | { | |
77c5e2d0 | 1607 | struct page *page; |
81819f0f | 1608 | |
b49af68f | 1609 | page = virt_to_head_page(x); |
81819f0f | 1610 | |
77c5e2d0 | 1611 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1612 | } |
1613 | EXPORT_SYMBOL(kmem_cache_free); | |
1614 | ||
1615 | /* Figure out on which slab object the object resides */ | |
1616 | static struct page *get_object_page(const void *x) | |
1617 | { | |
b49af68f | 1618 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1619 | |
1620 | if (!PageSlab(page)) | |
1621 | return NULL; | |
1622 | ||
1623 | return page; | |
1624 | } | |
1625 | ||
1626 | /* | |
672bba3a CL |
1627 | * Object placement in a slab is made very easy because we always start at |
1628 | * offset 0. If we tune the size of the object to the alignment then we can | |
1629 | * get the required alignment by putting one properly sized object after | |
1630 | * another. | |
81819f0f CL |
1631 | * |
1632 | * Notice that the allocation order determines the sizes of the per cpu | |
1633 | * caches. Each processor has always one slab available for allocations. | |
1634 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1635 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1636 | * locking overhead. |
81819f0f CL |
1637 | */ |
1638 | ||
1639 | /* | |
1640 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1641 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1642 | * and increases the number of allocations possible without having to | |
1643 | * take the list_lock. | |
1644 | */ | |
1645 | static int slub_min_order; | |
1646 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1647 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1648 | ||
1649 | /* | |
1650 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1651 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1652 | */ |
1653 | static int slub_nomerge; | |
1654 | ||
81819f0f CL |
1655 | /* |
1656 | * Calculate the order of allocation given an slab object size. | |
1657 | * | |
672bba3a CL |
1658 | * The order of allocation has significant impact on performance and other |
1659 | * system components. Generally order 0 allocations should be preferred since | |
1660 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1661 | * be problematic to put into order 0 slabs because there may be too much | |
1662 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1663 | * would be wasted. | |
1664 | * | |
1665 | * In order to reach satisfactory performance we must ensure that a minimum | |
1666 | * number of objects is in one slab. Otherwise we may generate too much | |
1667 | * activity on the partial lists which requires taking the list_lock. This is | |
1668 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1669 | * |
672bba3a CL |
1670 | * slub_max_order specifies the order where we begin to stop considering the |
1671 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1672 | * we try to keep the page order as low as possible. So we accept more waste | |
1673 | * of space in favor of a small page order. | |
81819f0f | 1674 | * |
672bba3a CL |
1675 | * Higher order allocations also allow the placement of more objects in a |
1676 | * slab and thereby reduce object handling overhead. If the user has | |
1677 | * requested a higher mininum order then we start with that one instead of | |
1678 | * the smallest order which will fit the object. | |
81819f0f | 1679 | */ |
5e6d444e CL |
1680 | static inline int slab_order(int size, int min_objects, |
1681 | int max_order, int fract_leftover) | |
81819f0f CL |
1682 | { |
1683 | int order; | |
1684 | int rem; | |
1685 | ||
5e6d444e CL |
1686 | for (order = max(slub_min_order, |
1687 | fls(min_objects * size - 1) - PAGE_SHIFT); | |
1688 | order <= max_order; order++) { | |
81819f0f | 1689 | |
5e6d444e | 1690 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1691 | |
5e6d444e | 1692 | if (slab_size < min_objects * size) |
81819f0f CL |
1693 | continue; |
1694 | ||
1695 | rem = slab_size % size; | |
1696 | ||
5e6d444e | 1697 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1698 | break; |
1699 | ||
1700 | } | |
672bba3a | 1701 | |
81819f0f CL |
1702 | return order; |
1703 | } | |
1704 | ||
5e6d444e CL |
1705 | static inline int calculate_order(int size) |
1706 | { | |
1707 | int order; | |
1708 | int min_objects; | |
1709 | int fraction; | |
1710 | ||
1711 | /* | |
1712 | * Attempt to find best configuration for a slab. This | |
1713 | * works by first attempting to generate a layout with | |
1714 | * the best configuration and backing off gradually. | |
1715 | * | |
1716 | * First we reduce the acceptable waste in a slab. Then | |
1717 | * we reduce the minimum objects required in a slab. | |
1718 | */ | |
1719 | min_objects = slub_min_objects; | |
1720 | while (min_objects > 1) { | |
1721 | fraction = 8; | |
1722 | while (fraction >= 4) { | |
1723 | order = slab_order(size, min_objects, | |
1724 | slub_max_order, fraction); | |
1725 | if (order <= slub_max_order) | |
1726 | return order; | |
1727 | fraction /= 2; | |
1728 | } | |
1729 | min_objects /= 2; | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * We were unable to place multiple objects in a slab. Now | |
1734 | * lets see if we can place a single object there. | |
1735 | */ | |
1736 | order = slab_order(size, 1, slub_max_order, 1); | |
1737 | if (order <= slub_max_order) | |
1738 | return order; | |
1739 | ||
1740 | /* | |
1741 | * Doh this slab cannot be placed using slub_max_order. | |
1742 | */ | |
1743 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1744 | if (order <= MAX_ORDER) | |
1745 | return order; | |
1746 | return -ENOSYS; | |
1747 | } | |
1748 | ||
81819f0f | 1749 | /* |
672bba3a | 1750 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1751 | */ |
1752 | static unsigned long calculate_alignment(unsigned long flags, | |
1753 | unsigned long align, unsigned long size) | |
1754 | { | |
1755 | /* | |
1756 | * If the user wants hardware cache aligned objects then | |
1757 | * follow that suggestion if the object is sufficiently | |
1758 | * large. | |
1759 | * | |
1760 | * The hardware cache alignment cannot override the | |
1761 | * specified alignment though. If that is greater | |
1762 | * then use it. | |
1763 | */ | |
5af60839 | 1764 | if ((flags & SLAB_HWCACHE_ALIGN) && |
65c02d4c CL |
1765 | size > cache_line_size() / 2) |
1766 | return max_t(unsigned long, align, cache_line_size()); | |
81819f0f CL |
1767 | |
1768 | if (align < ARCH_SLAB_MINALIGN) | |
1769 | return ARCH_SLAB_MINALIGN; | |
1770 | ||
1771 | return ALIGN(align, sizeof(void *)); | |
1772 | } | |
1773 | ||
1774 | static void init_kmem_cache_node(struct kmem_cache_node *n) | |
1775 | { | |
1776 | n->nr_partial = 0; | |
1777 | atomic_long_set(&n->nr_slabs, 0); | |
1778 | spin_lock_init(&n->list_lock); | |
1779 | INIT_LIST_HEAD(&n->partial); | |
643b1138 | 1780 | INIT_LIST_HEAD(&n->full); |
81819f0f CL |
1781 | } |
1782 | ||
1783 | #ifdef CONFIG_NUMA | |
1784 | /* | |
1785 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
1786 | * slab on the node for this slabcache. There are no concurrent accesses | |
1787 | * possible. | |
1788 | * | |
1789 | * Note that this function only works on the kmalloc_node_cache | |
1790 | * when allocating for the kmalloc_node_cache. | |
1791 | */ | |
1792 | static struct kmem_cache_node * __init early_kmem_cache_node_alloc(gfp_t gfpflags, | |
1793 | int node) | |
1794 | { | |
1795 | struct page *page; | |
1796 | struct kmem_cache_node *n; | |
1797 | ||
1798 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
1799 | ||
1800 | page = new_slab(kmalloc_caches, gfpflags | GFP_THISNODE, node); | |
1801 | /* new_slab() disables interupts */ | |
1802 | local_irq_enable(); | |
1803 | ||
1804 | BUG_ON(!page); | |
1805 | n = page->freelist; | |
1806 | BUG_ON(!n); | |
1807 | page->freelist = get_freepointer(kmalloc_caches, n); | |
1808 | page->inuse++; | |
1809 | kmalloc_caches->node[node] = n; | |
3ec09742 | 1810 | setup_object_debug(kmalloc_caches, page, n); |
81819f0f CL |
1811 | init_kmem_cache_node(n); |
1812 | atomic_long_inc(&n->nr_slabs); | |
e95eed57 | 1813 | add_partial(n, page); |
81819f0f CL |
1814 | return n; |
1815 | } | |
1816 | ||
1817 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1818 | { | |
1819 | int node; | |
1820 | ||
1821 | for_each_online_node(node) { | |
1822 | struct kmem_cache_node *n = s->node[node]; | |
1823 | if (n && n != &s->local_node) | |
1824 | kmem_cache_free(kmalloc_caches, n); | |
1825 | s->node[node] = NULL; | |
1826 | } | |
1827 | } | |
1828 | ||
1829 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1830 | { | |
1831 | int node; | |
1832 | int local_node; | |
1833 | ||
1834 | if (slab_state >= UP) | |
1835 | local_node = page_to_nid(virt_to_page(s)); | |
1836 | else | |
1837 | local_node = 0; | |
1838 | ||
1839 | for_each_online_node(node) { | |
1840 | struct kmem_cache_node *n; | |
1841 | ||
1842 | if (local_node == node) | |
1843 | n = &s->local_node; | |
1844 | else { | |
1845 | if (slab_state == DOWN) { | |
1846 | n = early_kmem_cache_node_alloc(gfpflags, | |
1847 | node); | |
1848 | continue; | |
1849 | } | |
1850 | n = kmem_cache_alloc_node(kmalloc_caches, | |
1851 | gfpflags, node); | |
1852 | ||
1853 | if (!n) { | |
1854 | free_kmem_cache_nodes(s); | |
1855 | return 0; | |
1856 | } | |
1857 | ||
1858 | } | |
1859 | s->node[node] = n; | |
1860 | init_kmem_cache_node(n); | |
1861 | } | |
1862 | return 1; | |
1863 | } | |
1864 | #else | |
1865 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1866 | { | |
1867 | } | |
1868 | ||
1869 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1870 | { | |
1871 | init_kmem_cache_node(&s->local_node); | |
1872 | return 1; | |
1873 | } | |
1874 | #endif | |
1875 | ||
1876 | /* | |
1877 | * calculate_sizes() determines the order and the distribution of data within | |
1878 | * a slab object. | |
1879 | */ | |
1880 | static int calculate_sizes(struct kmem_cache *s) | |
1881 | { | |
1882 | unsigned long flags = s->flags; | |
1883 | unsigned long size = s->objsize; | |
1884 | unsigned long align = s->align; | |
1885 | ||
1886 | /* | |
1887 | * Determine if we can poison the object itself. If the user of | |
1888 | * the slab may touch the object after free or before allocation | |
1889 | * then we should never poison the object itself. | |
1890 | */ | |
1891 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 1892 | !s->ctor) |
81819f0f CL |
1893 | s->flags |= __OBJECT_POISON; |
1894 | else | |
1895 | s->flags &= ~__OBJECT_POISON; | |
1896 | ||
1897 | /* | |
1898 | * Round up object size to the next word boundary. We can only | |
1899 | * place the free pointer at word boundaries and this determines | |
1900 | * the possible location of the free pointer. | |
1901 | */ | |
1902 | size = ALIGN(size, sizeof(void *)); | |
1903 | ||
41ecc55b | 1904 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f | 1905 | /* |
672bba3a | 1906 | * If we are Redzoning then check if there is some space between the |
81819f0f | 1907 | * end of the object and the free pointer. If not then add an |
672bba3a | 1908 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
1909 | */ |
1910 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
1911 | size += sizeof(void *); | |
41ecc55b | 1912 | #endif |
81819f0f CL |
1913 | |
1914 | /* | |
672bba3a CL |
1915 | * With that we have determined the number of bytes in actual use |
1916 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
1917 | */ |
1918 | s->inuse = size; | |
1919 | ||
1920 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 1921 | s->ctor)) { |
81819f0f CL |
1922 | /* |
1923 | * Relocate free pointer after the object if it is not | |
1924 | * permitted to overwrite the first word of the object on | |
1925 | * kmem_cache_free. | |
1926 | * | |
1927 | * This is the case if we do RCU, have a constructor or | |
1928 | * destructor or are poisoning the objects. | |
1929 | */ | |
1930 | s->offset = size; | |
1931 | size += sizeof(void *); | |
1932 | } | |
1933 | ||
c12b3c62 | 1934 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
1935 | if (flags & SLAB_STORE_USER) |
1936 | /* | |
1937 | * Need to store information about allocs and frees after | |
1938 | * the object. | |
1939 | */ | |
1940 | size += 2 * sizeof(struct track); | |
1941 | ||
be7b3fbc | 1942 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
1943 | /* |
1944 | * Add some empty padding so that we can catch | |
1945 | * overwrites from earlier objects rather than let | |
1946 | * tracking information or the free pointer be | |
1947 | * corrupted if an user writes before the start | |
1948 | * of the object. | |
1949 | */ | |
1950 | size += sizeof(void *); | |
41ecc55b | 1951 | #endif |
672bba3a | 1952 | |
81819f0f CL |
1953 | /* |
1954 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
1955 | * user specified and the dynamic determination of cache line size |
1956 | * on bootup. | |
81819f0f CL |
1957 | */ |
1958 | align = calculate_alignment(flags, align, s->objsize); | |
1959 | ||
1960 | /* | |
1961 | * SLUB stores one object immediately after another beginning from | |
1962 | * offset 0. In order to align the objects we have to simply size | |
1963 | * each object to conform to the alignment. | |
1964 | */ | |
1965 | size = ALIGN(size, align); | |
1966 | s->size = size; | |
1967 | ||
1968 | s->order = calculate_order(size); | |
1969 | if (s->order < 0) | |
1970 | return 0; | |
1971 | ||
1972 | /* | |
1973 | * Determine the number of objects per slab | |
1974 | */ | |
1975 | s->objects = (PAGE_SIZE << s->order) / size; | |
1976 | ||
1977 | /* | |
1978 | * Verify that the number of objects is within permitted limits. | |
1979 | * The page->inuse field is only 16 bit wide! So we cannot have | |
1980 | * more than 64k objects per slab. | |
1981 | */ | |
1982 | if (!s->objects || s->objects > 65535) | |
1983 | return 0; | |
1984 | return 1; | |
1985 | ||
1986 | } | |
1987 | ||
81819f0f CL |
1988 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
1989 | const char *name, size_t size, | |
1990 | size_t align, unsigned long flags, | |
c59def9f | 1991 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) |
81819f0f CL |
1992 | { |
1993 | memset(s, 0, kmem_size); | |
1994 | s->name = name; | |
1995 | s->ctor = ctor; | |
81819f0f CL |
1996 | s->objsize = size; |
1997 | s->flags = flags; | |
1998 | s->align = align; | |
41ecc55b | 1999 | kmem_cache_open_debug_check(s); |
81819f0f CL |
2000 | |
2001 | if (!calculate_sizes(s)) | |
2002 | goto error; | |
2003 | ||
2004 | s->refcount = 1; | |
2005 | #ifdef CONFIG_NUMA | |
2006 | s->defrag_ratio = 100; | |
2007 | #endif | |
2008 | ||
2009 | if (init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) | |
2010 | return 1; | |
2011 | error: | |
2012 | if (flags & SLAB_PANIC) | |
2013 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2014 | "order=%u offset=%u flags=%lx\n", | |
2015 | s->name, (unsigned long)size, s->size, s->order, | |
2016 | s->offset, flags); | |
2017 | return 0; | |
2018 | } | |
2019 | EXPORT_SYMBOL(kmem_cache_open); | |
2020 | ||
2021 | /* | |
2022 | * Check if a given pointer is valid | |
2023 | */ | |
2024 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2025 | { | |
2026 | struct page * page; | |
81819f0f CL |
2027 | |
2028 | page = get_object_page(object); | |
2029 | ||
2030 | if (!page || s != page->slab) | |
2031 | /* No slab or wrong slab */ | |
2032 | return 0; | |
2033 | ||
abcd08a6 | 2034 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2035 | return 0; |
2036 | ||
2037 | /* | |
2038 | * We could also check if the object is on the slabs freelist. | |
2039 | * But this would be too expensive and it seems that the main | |
2040 | * purpose of kmem_ptr_valid is to check if the object belongs | |
2041 | * to a certain slab. | |
2042 | */ | |
2043 | return 1; | |
2044 | } | |
2045 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2046 | ||
2047 | /* | |
2048 | * Determine the size of a slab object | |
2049 | */ | |
2050 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2051 | { | |
2052 | return s->objsize; | |
2053 | } | |
2054 | EXPORT_SYMBOL(kmem_cache_size); | |
2055 | ||
2056 | const char *kmem_cache_name(struct kmem_cache *s) | |
2057 | { | |
2058 | return s->name; | |
2059 | } | |
2060 | EXPORT_SYMBOL(kmem_cache_name); | |
2061 | ||
2062 | /* | |
672bba3a CL |
2063 | * Attempt to free all slabs on a node. Return the number of slabs we |
2064 | * were unable to free. | |
81819f0f CL |
2065 | */ |
2066 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
2067 | struct list_head *list) | |
2068 | { | |
2069 | int slabs_inuse = 0; | |
2070 | unsigned long flags; | |
2071 | struct page *page, *h; | |
2072 | ||
2073 | spin_lock_irqsave(&n->list_lock, flags); | |
2074 | list_for_each_entry_safe(page, h, list, lru) | |
2075 | if (!page->inuse) { | |
2076 | list_del(&page->lru); | |
2077 | discard_slab(s, page); | |
2078 | } else | |
2079 | slabs_inuse++; | |
2080 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2081 | return slabs_inuse; | |
2082 | } | |
2083 | ||
2084 | /* | |
672bba3a | 2085 | * Release all resources used by a slab cache. |
81819f0f CL |
2086 | */ |
2087 | static int kmem_cache_close(struct kmem_cache *s) | |
2088 | { | |
2089 | int node; | |
2090 | ||
2091 | flush_all(s); | |
2092 | ||
2093 | /* Attempt to free all objects */ | |
2094 | for_each_online_node(node) { | |
2095 | struct kmem_cache_node *n = get_node(s, node); | |
2096 | ||
2086d26a | 2097 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
2098 | if (atomic_long_read(&n->nr_slabs)) |
2099 | return 1; | |
2100 | } | |
2101 | free_kmem_cache_nodes(s); | |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | /* | |
2106 | * Close a cache and release the kmem_cache structure | |
2107 | * (must be used for caches created using kmem_cache_create) | |
2108 | */ | |
2109 | void kmem_cache_destroy(struct kmem_cache *s) | |
2110 | { | |
2111 | down_write(&slub_lock); | |
2112 | s->refcount--; | |
2113 | if (!s->refcount) { | |
2114 | list_del(&s->list); | |
2115 | if (kmem_cache_close(s)) | |
2116 | WARN_ON(1); | |
2117 | sysfs_slab_remove(s); | |
2118 | kfree(s); | |
2119 | } | |
2120 | up_write(&slub_lock); | |
2121 | } | |
2122 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2123 | ||
2124 | /******************************************************************** | |
2125 | * Kmalloc subsystem | |
2126 | *******************************************************************/ | |
2127 | ||
2128 | struct kmem_cache kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __cacheline_aligned; | |
2129 | EXPORT_SYMBOL(kmalloc_caches); | |
2130 | ||
2131 | #ifdef CONFIG_ZONE_DMA | |
2132 | static struct kmem_cache *kmalloc_caches_dma[KMALLOC_SHIFT_HIGH + 1]; | |
2133 | #endif | |
2134 | ||
2135 | static int __init setup_slub_min_order(char *str) | |
2136 | { | |
2137 | get_option (&str, &slub_min_order); | |
2138 | ||
2139 | return 1; | |
2140 | } | |
2141 | ||
2142 | __setup("slub_min_order=", setup_slub_min_order); | |
2143 | ||
2144 | static int __init setup_slub_max_order(char *str) | |
2145 | { | |
2146 | get_option (&str, &slub_max_order); | |
2147 | ||
2148 | return 1; | |
2149 | } | |
2150 | ||
2151 | __setup("slub_max_order=", setup_slub_max_order); | |
2152 | ||
2153 | static int __init setup_slub_min_objects(char *str) | |
2154 | { | |
2155 | get_option (&str, &slub_min_objects); | |
2156 | ||
2157 | return 1; | |
2158 | } | |
2159 | ||
2160 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2161 | ||
2162 | static int __init setup_slub_nomerge(char *str) | |
2163 | { | |
2164 | slub_nomerge = 1; | |
2165 | return 1; | |
2166 | } | |
2167 | ||
2168 | __setup("slub_nomerge", setup_slub_nomerge); | |
2169 | ||
81819f0f CL |
2170 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2171 | const char *name, int size, gfp_t gfp_flags) | |
2172 | { | |
2173 | unsigned int flags = 0; | |
2174 | ||
2175 | if (gfp_flags & SLUB_DMA) | |
2176 | flags = SLAB_CACHE_DMA; | |
2177 | ||
2178 | down_write(&slub_lock); | |
2179 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
c59def9f | 2180 | flags, NULL)) |
81819f0f CL |
2181 | goto panic; |
2182 | ||
2183 | list_add(&s->list, &slab_caches); | |
2184 | up_write(&slub_lock); | |
2185 | if (sysfs_slab_add(s)) | |
2186 | goto panic; | |
2187 | return s; | |
2188 | ||
2189 | panic: | |
2190 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2191 | } | |
2192 | ||
2193 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) | |
2194 | { | |
2195 | int index = kmalloc_index(size); | |
2196 | ||
614410d5 | 2197 | if (!index) |
81819f0f CL |
2198 | return NULL; |
2199 | ||
2200 | /* Allocation too large? */ | |
2201 | BUG_ON(index < 0); | |
2202 | ||
2203 | #ifdef CONFIG_ZONE_DMA | |
2204 | if ((flags & SLUB_DMA)) { | |
2205 | struct kmem_cache *s; | |
2206 | struct kmem_cache *x; | |
2207 | char *text; | |
2208 | size_t realsize; | |
2209 | ||
2210 | s = kmalloc_caches_dma[index]; | |
2211 | if (s) | |
2212 | return s; | |
2213 | ||
2214 | /* Dynamically create dma cache */ | |
2215 | x = kmalloc(kmem_size, flags & ~SLUB_DMA); | |
2216 | if (!x) | |
2217 | panic("Unable to allocate memory for dma cache\n"); | |
2218 | ||
2219 | if (index <= KMALLOC_SHIFT_HIGH) | |
2220 | realsize = 1 << index; | |
2221 | else { | |
2222 | if (index == 1) | |
2223 | realsize = 96; | |
2224 | else | |
2225 | realsize = 192; | |
2226 | } | |
2227 | ||
2228 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", | |
2229 | (unsigned int)realsize); | |
2230 | s = create_kmalloc_cache(x, text, realsize, flags); | |
2231 | kmalloc_caches_dma[index] = s; | |
2232 | return s; | |
2233 | } | |
2234 | #endif | |
2235 | return &kmalloc_caches[index]; | |
2236 | } | |
2237 | ||
2238 | void *__kmalloc(size_t size, gfp_t flags) | |
2239 | { | |
2240 | struct kmem_cache *s = get_slab(size, flags); | |
2241 | ||
2242 | if (s) | |
77c5e2d0 | 2243 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
272c1d21 | 2244 | return ZERO_SIZE_PTR; |
81819f0f CL |
2245 | } |
2246 | EXPORT_SYMBOL(__kmalloc); | |
2247 | ||
2248 | #ifdef CONFIG_NUMA | |
2249 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2250 | { | |
2251 | struct kmem_cache *s = get_slab(size, flags); | |
2252 | ||
2253 | if (s) | |
77c5e2d0 | 2254 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
272c1d21 | 2255 | return ZERO_SIZE_PTR; |
81819f0f CL |
2256 | } |
2257 | EXPORT_SYMBOL(__kmalloc_node); | |
2258 | #endif | |
2259 | ||
2260 | size_t ksize(const void *object) | |
2261 | { | |
272c1d21 | 2262 | struct page *page; |
81819f0f CL |
2263 | struct kmem_cache *s; |
2264 | ||
272c1d21 CL |
2265 | if (object == ZERO_SIZE_PTR) |
2266 | return 0; | |
2267 | ||
2268 | page = get_object_page(object); | |
81819f0f CL |
2269 | BUG_ON(!page); |
2270 | s = page->slab; | |
2271 | BUG_ON(!s); | |
2272 | ||
2273 | /* | |
2274 | * Debugging requires use of the padding between object | |
2275 | * and whatever may come after it. | |
2276 | */ | |
2277 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2278 | return s->objsize; | |
2279 | ||
2280 | /* | |
2281 | * If we have the need to store the freelist pointer | |
2282 | * back there or track user information then we can | |
2283 | * only use the space before that information. | |
2284 | */ | |
2285 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2286 | return s->inuse; | |
2287 | ||
2288 | /* | |
2289 | * Else we can use all the padding etc for the allocation | |
2290 | */ | |
2291 | return s->size; | |
2292 | } | |
2293 | EXPORT_SYMBOL(ksize); | |
2294 | ||
2295 | void kfree(const void *x) | |
2296 | { | |
2297 | struct kmem_cache *s; | |
2298 | struct page *page; | |
2299 | ||
272c1d21 CL |
2300 | /* |
2301 | * This has to be an unsigned comparison. According to Linus | |
2302 | * some gcc version treat a pointer as a signed entity. Then | |
2303 | * this comparison would be true for all "negative" pointers | |
2304 | * (which would cover the whole upper half of the address space). | |
2305 | */ | |
2306 | if ((unsigned long)x <= (unsigned long)ZERO_SIZE_PTR) | |
81819f0f CL |
2307 | return; |
2308 | ||
b49af68f | 2309 | page = virt_to_head_page(x); |
81819f0f CL |
2310 | s = page->slab; |
2311 | ||
77c5e2d0 | 2312 | slab_free(s, page, (void *)x, __builtin_return_address(0)); |
81819f0f CL |
2313 | } |
2314 | EXPORT_SYMBOL(kfree); | |
2315 | ||
2086d26a | 2316 | /* |
672bba3a CL |
2317 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2318 | * the remaining slabs by the number of items in use. The slabs with the | |
2319 | * most items in use come first. New allocations will then fill those up | |
2320 | * and thus they can be removed from the partial lists. | |
2321 | * | |
2322 | * The slabs with the least items are placed last. This results in them | |
2323 | * being allocated from last increasing the chance that the last objects | |
2324 | * are freed in them. | |
2086d26a CL |
2325 | */ |
2326 | int kmem_cache_shrink(struct kmem_cache *s) | |
2327 | { | |
2328 | int node; | |
2329 | int i; | |
2330 | struct kmem_cache_node *n; | |
2331 | struct page *page; | |
2332 | struct page *t; | |
2333 | struct list_head *slabs_by_inuse = | |
2334 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2335 | unsigned long flags; | |
2336 | ||
2337 | if (!slabs_by_inuse) | |
2338 | return -ENOMEM; | |
2339 | ||
2340 | flush_all(s); | |
2341 | for_each_online_node(node) { | |
2342 | n = get_node(s, node); | |
2343 | ||
2344 | if (!n->nr_partial) | |
2345 | continue; | |
2346 | ||
2347 | for (i = 0; i < s->objects; i++) | |
2348 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2349 | ||
2350 | spin_lock_irqsave(&n->list_lock, flags); | |
2351 | ||
2352 | /* | |
672bba3a | 2353 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2354 | * |
672bba3a CL |
2355 | * Note that concurrent frees may occur while we hold the |
2356 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2357 | */ |
2358 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2359 | if (!page->inuse && slab_trylock(page)) { | |
2360 | /* | |
2361 | * Must hold slab lock here because slab_free | |
2362 | * may have freed the last object and be | |
2363 | * waiting to release the slab. | |
2364 | */ | |
2365 | list_del(&page->lru); | |
2366 | n->nr_partial--; | |
2367 | slab_unlock(page); | |
2368 | discard_slab(s, page); | |
2369 | } else { | |
2370 | if (n->nr_partial > MAX_PARTIAL) | |
2371 | list_move(&page->lru, | |
2372 | slabs_by_inuse + page->inuse); | |
2373 | } | |
2374 | } | |
2375 | ||
2376 | if (n->nr_partial <= MAX_PARTIAL) | |
2377 | goto out; | |
2378 | ||
2379 | /* | |
672bba3a CL |
2380 | * Rebuild the partial list with the slabs filled up most |
2381 | * first and the least used slabs at the end. | |
2086d26a CL |
2382 | */ |
2383 | for (i = s->objects - 1; i >= 0; i--) | |
2384 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2385 | ||
2386 | out: | |
2387 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2388 | } | |
2389 | ||
2390 | kfree(slabs_by_inuse); | |
2391 | return 0; | |
2392 | } | |
2393 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2394 | ||
81819f0f CL |
2395 | /** |
2396 | * krealloc - reallocate memory. The contents will remain unchanged. | |
81819f0f CL |
2397 | * @p: object to reallocate memory for. |
2398 | * @new_size: how many bytes of memory are required. | |
2399 | * @flags: the type of memory to allocate. | |
2400 | * | |
2401 | * The contents of the object pointed to are preserved up to the | |
2402 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
2403 | * behaves exactly like kmalloc(). If @size is 0 and @p is not a | |
2404 | * %NULL pointer, the object pointed to is freed. | |
2405 | */ | |
2406 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
2407 | { | |
81819f0f | 2408 | void *ret; |
1f99a283 | 2409 | size_t ks; |
81819f0f | 2410 | |
272c1d21 | 2411 | if (unlikely(!p || p == ZERO_SIZE_PTR)) |
81819f0f CL |
2412 | return kmalloc(new_size, flags); |
2413 | ||
2414 | if (unlikely(!new_size)) { | |
2415 | kfree(p); | |
272c1d21 | 2416 | return ZERO_SIZE_PTR; |
81819f0f CL |
2417 | } |
2418 | ||
1f99a283 CL |
2419 | ks = ksize(p); |
2420 | if (ks >= new_size) | |
81819f0f CL |
2421 | return (void *)p; |
2422 | ||
2423 | ret = kmalloc(new_size, flags); | |
2424 | if (ret) { | |
1f99a283 | 2425 | memcpy(ret, p, min(new_size, ks)); |
81819f0f CL |
2426 | kfree(p); |
2427 | } | |
2428 | return ret; | |
2429 | } | |
2430 | EXPORT_SYMBOL(krealloc); | |
2431 | ||
2432 | /******************************************************************** | |
2433 | * Basic setup of slabs | |
2434 | *******************************************************************/ | |
2435 | ||
2436 | void __init kmem_cache_init(void) | |
2437 | { | |
2438 | int i; | |
2439 | ||
2440 | #ifdef CONFIG_NUMA | |
2441 | /* | |
2442 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2443 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2444 | * kmem_cache_open for slab_state == DOWN. |
2445 | */ | |
2446 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2447 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2448 | kmalloc_caches[0].refcount = -1; |
81819f0f CL |
2449 | #endif |
2450 | ||
2451 | /* Able to allocate the per node structures */ | |
2452 | slab_state = PARTIAL; | |
2453 | ||
2454 | /* Caches that are not of the two-to-the-power-of size */ | |
2455 | create_kmalloc_cache(&kmalloc_caches[1], | |
2456 | "kmalloc-96", 96, GFP_KERNEL); | |
2457 | create_kmalloc_cache(&kmalloc_caches[2], | |
2458 | "kmalloc-192", 192, GFP_KERNEL); | |
2459 | ||
2460 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2461 | create_kmalloc_cache(&kmalloc_caches[i], | |
2462 | "kmalloc", 1 << i, GFP_KERNEL); | |
2463 | ||
2464 | slab_state = UP; | |
2465 | ||
2466 | /* Provide the correct kmalloc names now that the caches are up */ | |
2467 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2468 | kmalloc_caches[i]. name = | |
2469 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
2470 | ||
2471 | #ifdef CONFIG_SMP | |
2472 | register_cpu_notifier(&slab_notifier); | |
2473 | #endif | |
2474 | ||
bcf889f9 CL |
2475 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
2476 | nr_cpu_ids * sizeof(struct page *); | |
81819f0f CL |
2477 | |
2478 | printk(KERN_INFO "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
2479 | " Processors=%d, Nodes=%d\n", | |
65c02d4c | 2480 | KMALLOC_SHIFT_HIGH, cache_line_size(), |
81819f0f CL |
2481 | slub_min_order, slub_max_order, slub_min_objects, |
2482 | nr_cpu_ids, nr_node_ids); | |
2483 | } | |
2484 | ||
2485 | /* | |
2486 | * Find a mergeable slab cache | |
2487 | */ | |
2488 | static int slab_unmergeable(struct kmem_cache *s) | |
2489 | { | |
2490 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
2491 | return 1; | |
2492 | ||
c59def9f | 2493 | if (s->ctor) |
81819f0f CL |
2494 | return 1; |
2495 | ||
8ffa6875 CL |
2496 | /* |
2497 | * We may have set a slab to be unmergeable during bootstrap. | |
2498 | */ | |
2499 | if (s->refcount < 0) | |
2500 | return 1; | |
2501 | ||
81819f0f CL |
2502 | return 0; |
2503 | } | |
2504 | ||
2505 | static struct kmem_cache *find_mergeable(size_t size, | |
2506 | size_t align, unsigned long flags, | |
c59def9f | 2507 | void (*ctor)(void *, struct kmem_cache *, unsigned long)) |
81819f0f CL |
2508 | { |
2509 | struct list_head *h; | |
2510 | ||
2511 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
2512 | return NULL; | |
2513 | ||
c59def9f | 2514 | if (ctor) |
81819f0f CL |
2515 | return NULL; |
2516 | ||
2517 | size = ALIGN(size, sizeof(void *)); | |
2518 | align = calculate_alignment(flags, align, size); | |
2519 | size = ALIGN(size, align); | |
2520 | ||
2521 | list_for_each(h, &slab_caches) { | |
2522 | struct kmem_cache *s = | |
2523 | container_of(h, struct kmem_cache, list); | |
2524 | ||
2525 | if (slab_unmergeable(s)) | |
2526 | continue; | |
2527 | ||
2528 | if (size > s->size) | |
2529 | continue; | |
2530 | ||
2531 | if (((flags | slub_debug) & SLUB_MERGE_SAME) != | |
2532 | (s->flags & SLUB_MERGE_SAME)) | |
2533 | continue; | |
2534 | /* | |
2535 | * Check if alignment is compatible. | |
2536 | * Courtesy of Adrian Drzewiecki | |
2537 | */ | |
2538 | if ((s->size & ~(align -1)) != s->size) | |
2539 | continue; | |
2540 | ||
2541 | if (s->size - size >= sizeof(void *)) | |
2542 | continue; | |
2543 | ||
2544 | return s; | |
2545 | } | |
2546 | return NULL; | |
2547 | } | |
2548 | ||
2549 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
2550 | size_t align, unsigned long flags, | |
2551 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
2552 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
2553 | { | |
2554 | struct kmem_cache *s; | |
2555 | ||
c59def9f | 2556 | BUG_ON(dtor); |
81819f0f | 2557 | down_write(&slub_lock); |
c59def9f | 2558 | s = find_mergeable(size, align, flags, ctor); |
81819f0f CL |
2559 | if (s) { |
2560 | s->refcount++; | |
2561 | /* | |
2562 | * Adjust the object sizes so that we clear | |
2563 | * the complete object on kzalloc. | |
2564 | */ | |
2565 | s->objsize = max(s->objsize, (int)size); | |
2566 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
2567 | if (sysfs_slab_alias(s, name)) | |
2568 | goto err; | |
2569 | } else { | |
2570 | s = kmalloc(kmem_size, GFP_KERNEL); | |
2571 | if (s && kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 2572 | size, align, flags, ctor)) { |
81819f0f CL |
2573 | if (sysfs_slab_add(s)) { |
2574 | kfree(s); | |
2575 | goto err; | |
2576 | } | |
2577 | list_add(&s->list, &slab_caches); | |
2578 | } else | |
2579 | kfree(s); | |
2580 | } | |
2581 | up_write(&slub_lock); | |
2582 | return s; | |
2583 | ||
2584 | err: | |
2585 | up_write(&slub_lock); | |
2586 | if (flags & SLAB_PANIC) | |
2587 | panic("Cannot create slabcache %s\n", name); | |
2588 | else | |
2589 | s = NULL; | |
2590 | return s; | |
2591 | } | |
2592 | EXPORT_SYMBOL(kmem_cache_create); | |
2593 | ||
2594 | void *kmem_cache_zalloc(struct kmem_cache *s, gfp_t flags) | |
2595 | { | |
2596 | void *x; | |
2597 | ||
77c5e2d0 | 2598 | x = slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2599 | if (x) |
2600 | memset(x, 0, s->objsize); | |
2601 | return x; | |
2602 | } | |
2603 | EXPORT_SYMBOL(kmem_cache_zalloc); | |
2604 | ||
2605 | #ifdef CONFIG_SMP | |
2606 | static void for_all_slabs(void (*func)(struct kmem_cache *, int), int cpu) | |
2607 | { | |
2608 | struct list_head *h; | |
2609 | ||
2610 | down_read(&slub_lock); | |
2611 | list_for_each(h, &slab_caches) { | |
2612 | struct kmem_cache *s = | |
2613 | container_of(h, struct kmem_cache, list); | |
2614 | ||
2615 | func(s, cpu); | |
2616 | } | |
2617 | up_read(&slub_lock); | |
2618 | } | |
2619 | ||
27390bc3 CL |
2620 | /* |
2621 | * Version of __flush_cpu_slab for the case that interrupts | |
2622 | * are enabled. | |
2623 | */ | |
2624 | static void cpu_slab_flush(struct kmem_cache *s, int cpu) | |
2625 | { | |
2626 | unsigned long flags; | |
2627 | ||
2628 | local_irq_save(flags); | |
2629 | __flush_cpu_slab(s, cpu); | |
2630 | local_irq_restore(flags); | |
2631 | } | |
2632 | ||
81819f0f | 2633 | /* |
672bba3a CL |
2634 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
2635 | * necessary. | |
81819f0f CL |
2636 | */ |
2637 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
2638 | unsigned long action, void *hcpu) | |
2639 | { | |
2640 | long cpu = (long)hcpu; | |
2641 | ||
2642 | switch (action) { | |
2643 | case CPU_UP_CANCELED: | |
8bb78442 | 2644 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 2645 | case CPU_DEAD: |
8bb78442 | 2646 | case CPU_DEAD_FROZEN: |
27390bc3 | 2647 | for_all_slabs(cpu_slab_flush, cpu); |
81819f0f CL |
2648 | break; |
2649 | default: | |
2650 | break; | |
2651 | } | |
2652 | return NOTIFY_OK; | |
2653 | } | |
2654 | ||
2655 | static struct notifier_block __cpuinitdata slab_notifier = | |
2656 | { &slab_cpuup_callback, NULL, 0 }; | |
2657 | ||
2658 | #endif | |
2659 | ||
81819f0f CL |
2660 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
2661 | { | |
2662 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2663 | |
2664 | if (!s) | |
272c1d21 | 2665 | return ZERO_SIZE_PTR; |
81819f0f | 2666 | |
77c5e2d0 | 2667 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
2668 | } |
2669 | ||
2670 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
2671 | int node, void *caller) | |
2672 | { | |
2673 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2674 | |
2675 | if (!s) | |
272c1d21 | 2676 | return ZERO_SIZE_PTR; |
81819f0f | 2677 | |
77c5e2d0 | 2678 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
2679 | } |
2680 | ||
41ecc55b | 2681 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
53e15af0 CL |
2682 | static int validate_slab(struct kmem_cache *s, struct page *page) |
2683 | { | |
2684 | void *p; | |
2685 | void *addr = page_address(page); | |
7656c72b | 2686 | DECLARE_BITMAP(map, s->objects); |
53e15af0 CL |
2687 | |
2688 | if (!check_slab(s, page) || | |
2689 | !on_freelist(s, page, NULL)) | |
2690 | return 0; | |
2691 | ||
2692 | /* Now we know that a valid freelist exists */ | |
2693 | bitmap_zero(map, s->objects); | |
2694 | ||
7656c72b CL |
2695 | for_each_free_object(p, s, page->freelist) { |
2696 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
2697 | if (!check_object(s, page, p, 0)) |
2698 | return 0; | |
2699 | } | |
2700 | ||
7656c72b CL |
2701 | for_each_object(p, s, addr) |
2702 | if (!test_bit(slab_index(p, s, addr), map)) | |
53e15af0 CL |
2703 | if (!check_object(s, page, p, 1)) |
2704 | return 0; | |
2705 | return 1; | |
2706 | } | |
2707 | ||
2708 | static void validate_slab_slab(struct kmem_cache *s, struct page *page) | |
2709 | { | |
2710 | if (slab_trylock(page)) { | |
2711 | validate_slab(s, page); | |
2712 | slab_unlock(page); | |
2713 | } else | |
2714 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
2715 | s->name, page); | |
2716 | ||
2717 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
2718 | if (!SlabDebug(page)) |
2719 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
2720 | "on slab 0x%p\n", s->name, page); |
2721 | } else { | |
35e5d7ee CL |
2722 | if (SlabDebug(page)) |
2723 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
2724 | "slab 0x%p\n", s->name, page); |
2725 | } | |
2726 | } | |
2727 | ||
2728 | static int validate_slab_node(struct kmem_cache *s, struct kmem_cache_node *n) | |
2729 | { | |
2730 | unsigned long count = 0; | |
2731 | struct page *page; | |
2732 | unsigned long flags; | |
2733 | ||
2734 | spin_lock_irqsave(&n->list_lock, flags); | |
2735 | ||
2736 | list_for_each_entry(page, &n->partial, lru) { | |
2737 | validate_slab_slab(s, page); | |
2738 | count++; | |
2739 | } | |
2740 | if (count != n->nr_partial) | |
2741 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
2742 | "counter=%ld\n", s->name, count, n->nr_partial); | |
2743 | ||
2744 | if (!(s->flags & SLAB_STORE_USER)) | |
2745 | goto out; | |
2746 | ||
2747 | list_for_each_entry(page, &n->full, lru) { | |
2748 | validate_slab_slab(s, page); | |
2749 | count++; | |
2750 | } | |
2751 | if (count != atomic_long_read(&n->nr_slabs)) | |
2752 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
2753 | "counter=%ld\n", s->name, count, | |
2754 | atomic_long_read(&n->nr_slabs)); | |
2755 | ||
2756 | out: | |
2757 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2758 | return count; | |
2759 | } | |
2760 | ||
2761 | static unsigned long validate_slab_cache(struct kmem_cache *s) | |
2762 | { | |
2763 | int node; | |
2764 | unsigned long count = 0; | |
2765 | ||
2766 | flush_all(s); | |
2767 | for_each_online_node(node) { | |
2768 | struct kmem_cache_node *n = get_node(s, node); | |
2769 | ||
2770 | count += validate_slab_node(s, n); | |
2771 | } | |
2772 | return count; | |
2773 | } | |
2774 | ||
b3459709 CL |
2775 | #ifdef SLUB_RESILIENCY_TEST |
2776 | static void resiliency_test(void) | |
2777 | { | |
2778 | u8 *p; | |
2779 | ||
2780 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
2781 | printk(KERN_ERR "-----------------------\n"); | |
2782 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
2783 | ||
2784 | p = kzalloc(16, GFP_KERNEL); | |
2785 | p[16] = 0x12; | |
2786 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
2787 | " 0x12->0x%p\n\n", p + 16); | |
2788 | ||
2789 | validate_slab_cache(kmalloc_caches + 4); | |
2790 | ||
2791 | /* Hmmm... The next two are dangerous */ | |
2792 | p = kzalloc(32, GFP_KERNEL); | |
2793 | p[32 + sizeof(void *)] = 0x34; | |
2794 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
2795 | " 0x34 -> -0x%p\n", p); | |
2796 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2797 | ||
2798 | validate_slab_cache(kmalloc_caches + 5); | |
2799 | p = kzalloc(64, GFP_KERNEL); | |
2800 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
2801 | *p = 0x56; | |
2802 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
2803 | p); | |
2804 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2805 | validate_slab_cache(kmalloc_caches + 6); | |
2806 | ||
2807 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
2808 | p = kzalloc(128, GFP_KERNEL); | |
2809 | kfree(p); | |
2810 | *p = 0x78; | |
2811 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
2812 | validate_slab_cache(kmalloc_caches + 7); | |
2813 | ||
2814 | p = kzalloc(256, GFP_KERNEL); | |
2815 | kfree(p); | |
2816 | p[50] = 0x9a; | |
2817 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); | |
2818 | validate_slab_cache(kmalloc_caches + 8); | |
2819 | ||
2820 | p = kzalloc(512, GFP_KERNEL); | |
2821 | kfree(p); | |
2822 | p[512] = 0xab; | |
2823 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
2824 | validate_slab_cache(kmalloc_caches + 9); | |
2825 | } | |
2826 | #else | |
2827 | static void resiliency_test(void) {}; | |
2828 | #endif | |
2829 | ||
88a420e4 | 2830 | /* |
672bba3a | 2831 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
2832 | * and freed. |
2833 | */ | |
2834 | ||
2835 | struct location { | |
2836 | unsigned long count; | |
2837 | void *addr; | |
45edfa58 CL |
2838 | long long sum_time; |
2839 | long min_time; | |
2840 | long max_time; | |
2841 | long min_pid; | |
2842 | long max_pid; | |
2843 | cpumask_t cpus; | |
2844 | nodemask_t nodes; | |
88a420e4 CL |
2845 | }; |
2846 | ||
2847 | struct loc_track { | |
2848 | unsigned long max; | |
2849 | unsigned long count; | |
2850 | struct location *loc; | |
2851 | }; | |
2852 | ||
2853 | static void free_loc_track(struct loc_track *t) | |
2854 | { | |
2855 | if (t->max) | |
2856 | free_pages((unsigned long)t->loc, | |
2857 | get_order(sizeof(struct location) * t->max)); | |
2858 | } | |
2859 | ||
2860 | static int alloc_loc_track(struct loc_track *t, unsigned long max) | |
2861 | { | |
2862 | struct location *l; | |
2863 | int order; | |
2864 | ||
2865 | if (!max) | |
2866 | max = PAGE_SIZE / sizeof(struct location); | |
2867 | ||
2868 | order = get_order(sizeof(struct location) * max); | |
2869 | ||
2870 | l = (void *)__get_free_pages(GFP_KERNEL, order); | |
2871 | ||
2872 | if (!l) | |
2873 | return 0; | |
2874 | ||
2875 | if (t->count) { | |
2876 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
2877 | free_loc_track(t); | |
2878 | } | |
2879 | t->max = max; | |
2880 | t->loc = l; | |
2881 | return 1; | |
2882 | } | |
2883 | ||
2884 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 2885 | const struct track *track) |
88a420e4 CL |
2886 | { |
2887 | long start, end, pos; | |
2888 | struct location *l; | |
2889 | void *caddr; | |
45edfa58 | 2890 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
2891 | |
2892 | start = -1; | |
2893 | end = t->count; | |
2894 | ||
2895 | for ( ; ; ) { | |
2896 | pos = start + (end - start + 1) / 2; | |
2897 | ||
2898 | /* | |
2899 | * There is nothing at "end". If we end up there | |
2900 | * we need to add something to before end. | |
2901 | */ | |
2902 | if (pos == end) | |
2903 | break; | |
2904 | ||
2905 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
2906 | if (track->addr == caddr) { |
2907 | ||
2908 | l = &t->loc[pos]; | |
2909 | l->count++; | |
2910 | if (track->when) { | |
2911 | l->sum_time += age; | |
2912 | if (age < l->min_time) | |
2913 | l->min_time = age; | |
2914 | if (age > l->max_time) | |
2915 | l->max_time = age; | |
2916 | ||
2917 | if (track->pid < l->min_pid) | |
2918 | l->min_pid = track->pid; | |
2919 | if (track->pid > l->max_pid) | |
2920 | l->max_pid = track->pid; | |
2921 | ||
2922 | cpu_set(track->cpu, l->cpus); | |
2923 | } | |
2924 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
2925 | return 1; |
2926 | } | |
2927 | ||
45edfa58 | 2928 | if (track->addr < caddr) |
88a420e4 CL |
2929 | end = pos; |
2930 | else | |
2931 | start = pos; | |
2932 | } | |
2933 | ||
2934 | /* | |
672bba3a | 2935 | * Not found. Insert new tracking element. |
88a420e4 CL |
2936 | */ |
2937 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max)) | |
2938 | return 0; | |
2939 | ||
2940 | l = t->loc + pos; | |
2941 | if (pos < t->count) | |
2942 | memmove(l + 1, l, | |
2943 | (t->count - pos) * sizeof(struct location)); | |
2944 | t->count++; | |
2945 | l->count = 1; | |
45edfa58 CL |
2946 | l->addr = track->addr; |
2947 | l->sum_time = age; | |
2948 | l->min_time = age; | |
2949 | l->max_time = age; | |
2950 | l->min_pid = track->pid; | |
2951 | l->max_pid = track->pid; | |
2952 | cpus_clear(l->cpus); | |
2953 | cpu_set(track->cpu, l->cpus); | |
2954 | nodes_clear(l->nodes); | |
2955 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
2956 | return 1; |
2957 | } | |
2958 | ||
2959 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
2960 | struct page *page, enum track_item alloc) | |
2961 | { | |
2962 | void *addr = page_address(page); | |
7656c72b | 2963 | DECLARE_BITMAP(map, s->objects); |
88a420e4 CL |
2964 | void *p; |
2965 | ||
2966 | bitmap_zero(map, s->objects); | |
7656c72b CL |
2967 | for_each_free_object(p, s, page->freelist) |
2968 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 2969 | |
7656c72b | 2970 | for_each_object(p, s, addr) |
45edfa58 CL |
2971 | if (!test_bit(slab_index(p, s, addr), map)) |
2972 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
2973 | } |
2974 | ||
2975 | static int list_locations(struct kmem_cache *s, char *buf, | |
2976 | enum track_item alloc) | |
2977 | { | |
2978 | int n = 0; | |
2979 | unsigned long i; | |
2980 | struct loc_track t; | |
2981 | int node; | |
2982 | ||
2983 | t.count = 0; | |
2984 | t.max = 0; | |
2985 | ||
2986 | /* Push back cpu slabs */ | |
2987 | flush_all(s); | |
2988 | ||
2989 | for_each_online_node(node) { | |
2990 | struct kmem_cache_node *n = get_node(s, node); | |
2991 | unsigned long flags; | |
2992 | struct page *page; | |
2993 | ||
2994 | if (!atomic_read(&n->nr_slabs)) | |
2995 | continue; | |
2996 | ||
2997 | spin_lock_irqsave(&n->list_lock, flags); | |
2998 | list_for_each_entry(page, &n->partial, lru) | |
2999 | process_slab(&t, s, page, alloc); | |
3000 | list_for_each_entry(page, &n->full, lru) | |
3001 | process_slab(&t, s, page, alloc); | |
3002 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3003 | } | |
3004 | ||
3005 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3006 | struct location *l = &t.loc[i]; |
88a420e4 CL |
3007 | |
3008 | if (n > PAGE_SIZE - 100) | |
3009 | break; | |
45edfa58 CL |
3010 | n += sprintf(buf + n, "%7ld ", l->count); |
3011 | ||
3012 | if (l->addr) | |
3013 | n += sprint_symbol(buf + n, (unsigned long)l->addr); | |
88a420e4 CL |
3014 | else |
3015 | n += sprintf(buf + n, "<not-available>"); | |
45edfa58 CL |
3016 | |
3017 | if (l->sum_time != l->min_time) { | |
3018 | unsigned long remainder; | |
3019 | ||
3020 | n += sprintf(buf + n, " age=%ld/%ld/%ld", | |
3021 | l->min_time, | |
3022 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3023 | l->max_time); | |
3024 | } else | |
3025 | n += sprintf(buf + n, " age=%ld", | |
3026 | l->min_time); | |
3027 | ||
3028 | if (l->min_pid != l->max_pid) | |
3029 | n += sprintf(buf + n, " pid=%ld-%ld", | |
3030 | l->min_pid, l->max_pid); | |
3031 | else | |
3032 | n += sprintf(buf + n, " pid=%ld", | |
3033 | l->min_pid); | |
3034 | ||
3035 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus)) { | |
3036 | n += sprintf(buf + n, " cpus="); | |
3037 | n += cpulist_scnprintf(buf + n, PAGE_SIZE - n - 50, | |
3038 | l->cpus); | |
3039 | } | |
3040 | ||
3041 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes)) { | |
3042 | n += sprintf(buf + n, " nodes="); | |
3043 | n += nodelist_scnprintf(buf + n, PAGE_SIZE - n - 50, | |
3044 | l->nodes); | |
3045 | } | |
3046 | ||
88a420e4 CL |
3047 | n += sprintf(buf + n, "\n"); |
3048 | } | |
3049 | ||
3050 | free_loc_track(&t); | |
3051 | if (!t.count) | |
3052 | n += sprintf(buf, "No data\n"); | |
3053 | return n; | |
3054 | } | |
3055 | ||
81819f0f CL |
3056 | static unsigned long count_partial(struct kmem_cache_node *n) |
3057 | { | |
3058 | unsigned long flags; | |
3059 | unsigned long x = 0; | |
3060 | struct page *page; | |
3061 | ||
3062 | spin_lock_irqsave(&n->list_lock, flags); | |
3063 | list_for_each_entry(page, &n->partial, lru) | |
3064 | x += page->inuse; | |
3065 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3066 | return x; | |
3067 | } | |
3068 | ||
3069 | enum slab_stat_type { | |
3070 | SL_FULL, | |
3071 | SL_PARTIAL, | |
3072 | SL_CPU, | |
3073 | SL_OBJECTS | |
3074 | }; | |
3075 | ||
3076 | #define SO_FULL (1 << SL_FULL) | |
3077 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
3078 | #define SO_CPU (1 << SL_CPU) | |
3079 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
3080 | ||
3081 | static unsigned long slab_objects(struct kmem_cache *s, | |
3082 | char *buf, unsigned long flags) | |
3083 | { | |
3084 | unsigned long total = 0; | |
3085 | int cpu; | |
3086 | int node; | |
3087 | int x; | |
3088 | unsigned long *nodes; | |
3089 | unsigned long *per_cpu; | |
3090 | ||
3091 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
3092 | per_cpu = nodes + nr_node_ids; | |
3093 | ||
3094 | for_each_possible_cpu(cpu) { | |
3095 | struct page *page = s->cpu_slab[cpu]; | |
3096 | int node; | |
3097 | ||
3098 | if (page) { | |
3099 | node = page_to_nid(page); | |
3100 | if (flags & SO_CPU) { | |
3101 | int x = 0; | |
3102 | ||
3103 | if (flags & SO_OBJECTS) | |
3104 | x = page->inuse; | |
3105 | else | |
3106 | x = 1; | |
3107 | total += x; | |
3108 | nodes[node] += x; | |
3109 | } | |
3110 | per_cpu[node]++; | |
3111 | } | |
3112 | } | |
3113 | ||
3114 | for_each_online_node(node) { | |
3115 | struct kmem_cache_node *n = get_node(s, node); | |
3116 | ||
3117 | if (flags & SO_PARTIAL) { | |
3118 | if (flags & SO_OBJECTS) | |
3119 | x = count_partial(n); | |
3120 | else | |
3121 | x = n->nr_partial; | |
3122 | total += x; | |
3123 | nodes[node] += x; | |
3124 | } | |
3125 | ||
3126 | if (flags & SO_FULL) { | |
3127 | int full_slabs = atomic_read(&n->nr_slabs) | |
3128 | - per_cpu[node] | |
3129 | - n->nr_partial; | |
3130 | ||
3131 | if (flags & SO_OBJECTS) | |
3132 | x = full_slabs * s->objects; | |
3133 | else | |
3134 | x = full_slabs; | |
3135 | total += x; | |
3136 | nodes[node] += x; | |
3137 | } | |
3138 | } | |
3139 | ||
3140 | x = sprintf(buf, "%lu", total); | |
3141 | #ifdef CONFIG_NUMA | |
3142 | for_each_online_node(node) | |
3143 | if (nodes[node]) | |
3144 | x += sprintf(buf + x, " N%d=%lu", | |
3145 | node, nodes[node]); | |
3146 | #endif | |
3147 | kfree(nodes); | |
3148 | return x + sprintf(buf + x, "\n"); | |
3149 | } | |
3150 | ||
3151 | static int any_slab_objects(struct kmem_cache *s) | |
3152 | { | |
3153 | int node; | |
3154 | int cpu; | |
3155 | ||
3156 | for_each_possible_cpu(cpu) | |
3157 | if (s->cpu_slab[cpu]) | |
3158 | return 1; | |
3159 | ||
3160 | for_each_node(node) { | |
3161 | struct kmem_cache_node *n = get_node(s, node); | |
3162 | ||
3163 | if (n->nr_partial || atomic_read(&n->nr_slabs)) | |
3164 | return 1; | |
3165 | } | |
3166 | return 0; | |
3167 | } | |
3168 | ||
3169 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3170 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3171 | ||
3172 | struct slab_attribute { | |
3173 | struct attribute attr; | |
3174 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3175 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3176 | }; | |
3177 | ||
3178 | #define SLAB_ATTR_RO(_name) \ | |
3179 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3180 | ||
3181 | #define SLAB_ATTR(_name) \ | |
3182 | static struct slab_attribute _name##_attr = \ | |
3183 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3184 | ||
81819f0f CL |
3185 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3186 | { | |
3187 | return sprintf(buf, "%d\n", s->size); | |
3188 | } | |
3189 | SLAB_ATTR_RO(slab_size); | |
3190 | ||
3191 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3192 | { | |
3193 | return sprintf(buf, "%d\n", s->align); | |
3194 | } | |
3195 | SLAB_ATTR_RO(align); | |
3196 | ||
3197 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3198 | { | |
3199 | return sprintf(buf, "%d\n", s->objsize); | |
3200 | } | |
3201 | SLAB_ATTR_RO(object_size); | |
3202 | ||
3203 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3204 | { | |
3205 | return sprintf(buf, "%d\n", s->objects); | |
3206 | } | |
3207 | SLAB_ATTR_RO(objs_per_slab); | |
3208 | ||
3209 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3210 | { | |
3211 | return sprintf(buf, "%d\n", s->order); | |
3212 | } | |
3213 | SLAB_ATTR_RO(order); | |
3214 | ||
3215 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3216 | { | |
3217 | if (s->ctor) { | |
3218 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3219 | ||
3220 | return n + sprintf(buf + n, "\n"); | |
3221 | } | |
3222 | return 0; | |
3223 | } | |
3224 | SLAB_ATTR_RO(ctor); | |
3225 | ||
81819f0f CL |
3226 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3227 | { | |
3228 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3229 | } | |
3230 | SLAB_ATTR_RO(aliases); | |
3231 | ||
3232 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3233 | { | |
3234 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3235 | } | |
3236 | SLAB_ATTR_RO(slabs); | |
3237 | ||
3238 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3239 | { | |
3240 | return slab_objects(s, buf, SO_PARTIAL); | |
3241 | } | |
3242 | SLAB_ATTR_RO(partial); | |
3243 | ||
3244 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3245 | { | |
3246 | return slab_objects(s, buf, SO_CPU); | |
3247 | } | |
3248 | SLAB_ATTR_RO(cpu_slabs); | |
3249 | ||
3250 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3251 | { | |
3252 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3253 | } | |
3254 | SLAB_ATTR_RO(objects); | |
3255 | ||
3256 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3257 | { | |
3258 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3259 | } | |
3260 | ||
3261 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3262 | const char *buf, size_t length) | |
3263 | { | |
3264 | s->flags &= ~SLAB_DEBUG_FREE; | |
3265 | if (buf[0] == '1') | |
3266 | s->flags |= SLAB_DEBUG_FREE; | |
3267 | return length; | |
3268 | } | |
3269 | SLAB_ATTR(sanity_checks); | |
3270 | ||
3271 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3272 | { | |
3273 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3274 | } | |
3275 | ||
3276 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3277 | size_t length) | |
3278 | { | |
3279 | s->flags &= ~SLAB_TRACE; | |
3280 | if (buf[0] == '1') | |
3281 | s->flags |= SLAB_TRACE; | |
3282 | return length; | |
3283 | } | |
3284 | SLAB_ATTR(trace); | |
3285 | ||
3286 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3287 | { | |
3288 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3289 | } | |
3290 | ||
3291 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3292 | const char *buf, size_t length) | |
3293 | { | |
3294 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3295 | if (buf[0] == '1') | |
3296 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3297 | return length; | |
3298 | } | |
3299 | SLAB_ATTR(reclaim_account); | |
3300 | ||
3301 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3302 | { | |
5af60839 | 3303 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3304 | } |
3305 | SLAB_ATTR_RO(hwcache_align); | |
3306 | ||
3307 | #ifdef CONFIG_ZONE_DMA | |
3308 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3309 | { | |
3310 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3311 | } | |
3312 | SLAB_ATTR_RO(cache_dma); | |
3313 | #endif | |
3314 | ||
3315 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3316 | { | |
3317 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3318 | } | |
3319 | SLAB_ATTR_RO(destroy_by_rcu); | |
3320 | ||
3321 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3322 | { | |
3323 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3324 | } | |
3325 | ||
3326 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3327 | const char *buf, size_t length) | |
3328 | { | |
3329 | if (any_slab_objects(s)) | |
3330 | return -EBUSY; | |
3331 | ||
3332 | s->flags &= ~SLAB_RED_ZONE; | |
3333 | if (buf[0] == '1') | |
3334 | s->flags |= SLAB_RED_ZONE; | |
3335 | calculate_sizes(s); | |
3336 | return length; | |
3337 | } | |
3338 | SLAB_ATTR(red_zone); | |
3339 | ||
3340 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3341 | { | |
3342 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3343 | } | |
3344 | ||
3345 | static ssize_t poison_store(struct kmem_cache *s, | |
3346 | const char *buf, size_t length) | |
3347 | { | |
3348 | if (any_slab_objects(s)) | |
3349 | return -EBUSY; | |
3350 | ||
3351 | s->flags &= ~SLAB_POISON; | |
3352 | if (buf[0] == '1') | |
3353 | s->flags |= SLAB_POISON; | |
3354 | calculate_sizes(s); | |
3355 | return length; | |
3356 | } | |
3357 | SLAB_ATTR(poison); | |
3358 | ||
3359 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3360 | { | |
3361 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3362 | } | |
3363 | ||
3364 | static ssize_t store_user_store(struct kmem_cache *s, | |
3365 | const char *buf, size_t length) | |
3366 | { | |
3367 | if (any_slab_objects(s)) | |
3368 | return -EBUSY; | |
3369 | ||
3370 | s->flags &= ~SLAB_STORE_USER; | |
3371 | if (buf[0] == '1') | |
3372 | s->flags |= SLAB_STORE_USER; | |
3373 | calculate_sizes(s); | |
3374 | return length; | |
3375 | } | |
3376 | SLAB_ATTR(store_user); | |
3377 | ||
53e15af0 CL |
3378 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3379 | { | |
3380 | return 0; | |
3381 | } | |
3382 | ||
3383 | static ssize_t validate_store(struct kmem_cache *s, | |
3384 | const char *buf, size_t length) | |
3385 | { | |
3386 | if (buf[0] == '1') | |
3387 | validate_slab_cache(s); | |
3388 | else | |
3389 | return -EINVAL; | |
3390 | return length; | |
3391 | } | |
3392 | SLAB_ATTR(validate); | |
3393 | ||
2086d26a CL |
3394 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
3395 | { | |
3396 | return 0; | |
3397 | } | |
3398 | ||
3399 | static ssize_t shrink_store(struct kmem_cache *s, | |
3400 | const char *buf, size_t length) | |
3401 | { | |
3402 | if (buf[0] == '1') { | |
3403 | int rc = kmem_cache_shrink(s); | |
3404 | ||
3405 | if (rc) | |
3406 | return rc; | |
3407 | } else | |
3408 | return -EINVAL; | |
3409 | return length; | |
3410 | } | |
3411 | SLAB_ATTR(shrink); | |
3412 | ||
88a420e4 CL |
3413 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
3414 | { | |
3415 | if (!(s->flags & SLAB_STORE_USER)) | |
3416 | return -ENOSYS; | |
3417 | return list_locations(s, buf, TRACK_ALLOC); | |
3418 | } | |
3419 | SLAB_ATTR_RO(alloc_calls); | |
3420 | ||
3421 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
3422 | { | |
3423 | if (!(s->flags & SLAB_STORE_USER)) | |
3424 | return -ENOSYS; | |
3425 | return list_locations(s, buf, TRACK_FREE); | |
3426 | } | |
3427 | SLAB_ATTR_RO(free_calls); | |
3428 | ||
81819f0f CL |
3429 | #ifdef CONFIG_NUMA |
3430 | static ssize_t defrag_ratio_show(struct kmem_cache *s, char *buf) | |
3431 | { | |
3432 | return sprintf(buf, "%d\n", s->defrag_ratio / 10); | |
3433 | } | |
3434 | ||
3435 | static ssize_t defrag_ratio_store(struct kmem_cache *s, | |
3436 | const char *buf, size_t length) | |
3437 | { | |
3438 | int n = simple_strtoul(buf, NULL, 10); | |
3439 | ||
3440 | if (n < 100) | |
3441 | s->defrag_ratio = n * 10; | |
3442 | return length; | |
3443 | } | |
3444 | SLAB_ATTR(defrag_ratio); | |
3445 | #endif | |
3446 | ||
3447 | static struct attribute * slab_attrs[] = { | |
3448 | &slab_size_attr.attr, | |
3449 | &object_size_attr.attr, | |
3450 | &objs_per_slab_attr.attr, | |
3451 | &order_attr.attr, | |
3452 | &objects_attr.attr, | |
3453 | &slabs_attr.attr, | |
3454 | &partial_attr.attr, | |
3455 | &cpu_slabs_attr.attr, | |
3456 | &ctor_attr.attr, | |
81819f0f CL |
3457 | &aliases_attr.attr, |
3458 | &align_attr.attr, | |
3459 | &sanity_checks_attr.attr, | |
3460 | &trace_attr.attr, | |
3461 | &hwcache_align_attr.attr, | |
3462 | &reclaim_account_attr.attr, | |
3463 | &destroy_by_rcu_attr.attr, | |
3464 | &red_zone_attr.attr, | |
3465 | &poison_attr.attr, | |
3466 | &store_user_attr.attr, | |
53e15af0 | 3467 | &validate_attr.attr, |
2086d26a | 3468 | &shrink_attr.attr, |
88a420e4 CL |
3469 | &alloc_calls_attr.attr, |
3470 | &free_calls_attr.attr, | |
81819f0f CL |
3471 | #ifdef CONFIG_ZONE_DMA |
3472 | &cache_dma_attr.attr, | |
3473 | #endif | |
3474 | #ifdef CONFIG_NUMA | |
3475 | &defrag_ratio_attr.attr, | |
3476 | #endif | |
3477 | NULL | |
3478 | }; | |
3479 | ||
3480 | static struct attribute_group slab_attr_group = { | |
3481 | .attrs = slab_attrs, | |
3482 | }; | |
3483 | ||
3484 | static ssize_t slab_attr_show(struct kobject *kobj, | |
3485 | struct attribute *attr, | |
3486 | char *buf) | |
3487 | { | |
3488 | struct slab_attribute *attribute; | |
3489 | struct kmem_cache *s; | |
3490 | int err; | |
3491 | ||
3492 | attribute = to_slab_attr(attr); | |
3493 | s = to_slab(kobj); | |
3494 | ||
3495 | if (!attribute->show) | |
3496 | return -EIO; | |
3497 | ||
3498 | err = attribute->show(s, buf); | |
3499 | ||
3500 | return err; | |
3501 | } | |
3502 | ||
3503 | static ssize_t slab_attr_store(struct kobject *kobj, | |
3504 | struct attribute *attr, | |
3505 | const char *buf, size_t len) | |
3506 | { | |
3507 | struct slab_attribute *attribute; | |
3508 | struct kmem_cache *s; | |
3509 | int err; | |
3510 | ||
3511 | attribute = to_slab_attr(attr); | |
3512 | s = to_slab(kobj); | |
3513 | ||
3514 | if (!attribute->store) | |
3515 | return -EIO; | |
3516 | ||
3517 | err = attribute->store(s, buf, len); | |
3518 | ||
3519 | return err; | |
3520 | } | |
3521 | ||
3522 | static struct sysfs_ops slab_sysfs_ops = { | |
3523 | .show = slab_attr_show, | |
3524 | .store = slab_attr_store, | |
3525 | }; | |
3526 | ||
3527 | static struct kobj_type slab_ktype = { | |
3528 | .sysfs_ops = &slab_sysfs_ops, | |
3529 | }; | |
3530 | ||
3531 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
3532 | { | |
3533 | struct kobj_type *ktype = get_ktype(kobj); | |
3534 | ||
3535 | if (ktype == &slab_ktype) | |
3536 | return 1; | |
3537 | return 0; | |
3538 | } | |
3539 | ||
3540 | static struct kset_uevent_ops slab_uevent_ops = { | |
3541 | .filter = uevent_filter, | |
3542 | }; | |
3543 | ||
3544 | decl_subsys(slab, &slab_ktype, &slab_uevent_ops); | |
3545 | ||
3546 | #define ID_STR_LENGTH 64 | |
3547 | ||
3548 | /* Create a unique string id for a slab cache: | |
3549 | * format | |
3550 | * :[flags-]size:[memory address of kmemcache] | |
3551 | */ | |
3552 | static char *create_unique_id(struct kmem_cache *s) | |
3553 | { | |
3554 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
3555 | char *p = name; | |
3556 | ||
3557 | BUG_ON(!name); | |
3558 | ||
3559 | *p++ = ':'; | |
3560 | /* | |
3561 | * First flags affecting slabcache operations. We will only | |
3562 | * get here for aliasable slabs so we do not need to support | |
3563 | * too many flags. The flags here must cover all flags that | |
3564 | * are matched during merging to guarantee that the id is | |
3565 | * unique. | |
3566 | */ | |
3567 | if (s->flags & SLAB_CACHE_DMA) | |
3568 | *p++ = 'd'; | |
3569 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3570 | *p++ = 'a'; | |
3571 | if (s->flags & SLAB_DEBUG_FREE) | |
3572 | *p++ = 'F'; | |
3573 | if (p != name + 1) | |
3574 | *p++ = '-'; | |
3575 | p += sprintf(p, "%07d", s->size); | |
3576 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
3577 | return name; | |
3578 | } | |
3579 | ||
3580 | static int sysfs_slab_add(struct kmem_cache *s) | |
3581 | { | |
3582 | int err; | |
3583 | const char *name; | |
3584 | int unmergeable; | |
3585 | ||
3586 | if (slab_state < SYSFS) | |
3587 | /* Defer until later */ | |
3588 | return 0; | |
3589 | ||
3590 | unmergeable = slab_unmergeable(s); | |
3591 | if (unmergeable) { | |
3592 | /* | |
3593 | * Slabcache can never be merged so we can use the name proper. | |
3594 | * This is typically the case for debug situations. In that | |
3595 | * case we can catch duplicate names easily. | |
3596 | */ | |
0f9008ef | 3597 | sysfs_remove_link(&slab_subsys.kobj, s->name); |
81819f0f CL |
3598 | name = s->name; |
3599 | } else { | |
3600 | /* | |
3601 | * Create a unique name for the slab as a target | |
3602 | * for the symlinks. | |
3603 | */ | |
3604 | name = create_unique_id(s); | |
3605 | } | |
3606 | ||
3607 | kobj_set_kset_s(s, slab_subsys); | |
3608 | kobject_set_name(&s->kobj, name); | |
3609 | kobject_init(&s->kobj); | |
3610 | err = kobject_add(&s->kobj); | |
3611 | if (err) | |
3612 | return err; | |
3613 | ||
3614 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
3615 | if (err) | |
3616 | return err; | |
3617 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
3618 | if (!unmergeable) { | |
3619 | /* Setup first alias */ | |
3620 | sysfs_slab_alias(s, s->name); | |
3621 | kfree(name); | |
3622 | } | |
3623 | return 0; | |
3624 | } | |
3625 | ||
3626 | static void sysfs_slab_remove(struct kmem_cache *s) | |
3627 | { | |
3628 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
3629 | kobject_del(&s->kobj); | |
3630 | } | |
3631 | ||
3632 | /* | |
3633 | * Need to buffer aliases during bootup until sysfs becomes | |
3634 | * available lest we loose that information. | |
3635 | */ | |
3636 | struct saved_alias { | |
3637 | struct kmem_cache *s; | |
3638 | const char *name; | |
3639 | struct saved_alias *next; | |
3640 | }; | |
3641 | ||
3642 | struct saved_alias *alias_list; | |
3643 | ||
3644 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
3645 | { | |
3646 | struct saved_alias *al; | |
3647 | ||
3648 | if (slab_state == SYSFS) { | |
3649 | /* | |
3650 | * If we have a leftover link then remove it. | |
3651 | */ | |
0f9008ef LT |
3652 | sysfs_remove_link(&slab_subsys.kobj, name); |
3653 | return sysfs_create_link(&slab_subsys.kobj, | |
81819f0f CL |
3654 | &s->kobj, name); |
3655 | } | |
3656 | ||
3657 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
3658 | if (!al) | |
3659 | return -ENOMEM; | |
3660 | ||
3661 | al->s = s; | |
3662 | al->name = name; | |
3663 | al->next = alias_list; | |
3664 | alias_list = al; | |
3665 | return 0; | |
3666 | } | |
3667 | ||
3668 | static int __init slab_sysfs_init(void) | |
3669 | { | |
26a7bd03 | 3670 | struct list_head *h; |
81819f0f CL |
3671 | int err; |
3672 | ||
3673 | err = subsystem_register(&slab_subsys); | |
3674 | if (err) { | |
3675 | printk(KERN_ERR "Cannot register slab subsystem.\n"); | |
3676 | return -ENOSYS; | |
3677 | } | |
3678 | ||
26a7bd03 CL |
3679 | slab_state = SYSFS; |
3680 | ||
3681 | list_for_each(h, &slab_caches) { | |
3682 | struct kmem_cache *s = | |
3683 | container_of(h, struct kmem_cache, list); | |
3684 | ||
3685 | err = sysfs_slab_add(s); | |
3686 | BUG_ON(err); | |
3687 | } | |
81819f0f CL |
3688 | |
3689 | while (alias_list) { | |
3690 | struct saved_alias *al = alias_list; | |
3691 | ||
3692 | alias_list = alias_list->next; | |
3693 | err = sysfs_slab_alias(al->s, al->name); | |
3694 | BUG_ON(err); | |
3695 | kfree(al); | |
3696 | } | |
3697 | ||
3698 | resiliency_test(); | |
3699 | return 0; | |
3700 | } | |
3701 | ||
3702 | __initcall(slab_sysfs_init); | |
81819f0f | 3703 | #endif |