]>
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 | * | |
69 | * Slabs with free elements are kept on a partial list. | |
70 | * There is no list for full slabs. If an object in a full slab is | |
71 | * freed then the slab will show up again on the partial lists. | |
72 | * Otherwise there is no need to track full slabs unless we have to | |
73 | * track full slabs for debugging purposes. | |
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 | * | |
81 | * PageActive The slab is used as a cpu cache. Allocations | |
82 | * may be performed from the slab. The slab is not | |
83 | * on any slab list and cannot be moved onto one. | |
84 | * | |
85 | * PageError Slab requires special handling due to debug | |
86 | * options set. This moves slab handling out of | |
87 | * the fast path. | |
88 | */ | |
89 | ||
90 | /* | |
91 | * Issues still to be resolved: | |
92 | * | |
93 | * - The per cpu array is updated for each new slab and and is a remote | |
94 | * cacheline for most nodes. This could become a bouncing cacheline given | |
95 | * enough frequent updates. There are 16 pointers in a cacheline.so at | |
96 | * max 16 cpus could compete. Likely okay. | |
97 | * | |
98 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. | |
99 | * | |
81819f0f CL |
100 | * - Variable sizing of the per node arrays |
101 | */ | |
102 | ||
103 | /* Enable to test recovery from slab corruption on boot */ | |
104 | #undef SLUB_RESILIENCY_TEST | |
105 | ||
106 | #if PAGE_SHIFT <= 12 | |
107 | ||
108 | /* | |
109 | * Small page size. Make sure that we do not fragment memory | |
110 | */ | |
111 | #define DEFAULT_MAX_ORDER 1 | |
112 | #define DEFAULT_MIN_OBJECTS 4 | |
113 | ||
114 | #else | |
115 | ||
116 | /* | |
117 | * Large page machines are customarily able to handle larger | |
118 | * page orders. | |
119 | */ | |
120 | #define DEFAULT_MAX_ORDER 2 | |
121 | #define DEFAULT_MIN_OBJECTS 8 | |
122 | ||
123 | #endif | |
124 | ||
2086d26a CL |
125 | /* |
126 | * Mininum number of partial slabs. These will be left on the partial | |
127 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
128 | */ | |
e95eed57 CL |
129 | #define MIN_PARTIAL 2 |
130 | ||
2086d26a CL |
131 | /* |
132 | * Maximum number of desirable partial slabs. | |
133 | * The existence of more partial slabs makes kmem_cache_shrink | |
134 | * sort the partial list by the number of objects in the. | |
135 | */ | |
136 | #define MAX_PARTIAL 10 | |
137 | ||
81819f0f CL |
138 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
139 | SLAB_POISON | SLAB_STORE_USER) | |
140 | /* | |
141 | * Set of flags that will prevent slab merging | |
142 | */ | |
143 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
144 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
145 | ||
146 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
147 | SLAB_CACHE_DMA) | |
148 | ||
149 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 150 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
151 | #endif |
152 | ||
153 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 154 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
155 | #endif |
156 | ||
157 | /* Internal SLUB flags */ | |
158 | #define __OBJECT_POISON 0x80000000 /* Poison object */ | |
159 | ||
160 | static int kmem_size = sizeof(struct kmem_cache); | |
161 | ||
162 | #ifdef CONFIG_SMP | |
163 | static struct notifier_block slab_notifier; | |
164 | #endif | |
165 | ||
166 | static enum { | |
167 | DOWN, /* No slab functionality available */ | |
168 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
169 | UP, /* Everything works */ | |
170 | SYSFS /* Sysfs up */ | |
171 | } slab_state = DOWN; | |
172 | ||
173 | /* A list of all slab caches on the system */ | |
174 | static DECLARE_RWSEM(slub_lock); | |
175 | LIST_HEAD(slab_caches); | |
176 | ||
177 | #ifdef CONFIG_SYSFS | |
178 | static int sysfs_slab_add(struct kmem_cache *); | |
179 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
180 | static void sysfs_slab_remove(struct kmem_cache *); | |
181 | #else | |
182 | static int sysfs_slab_add(struct kmem_cache *s) { return 0; } | |
183 | static int sysfs_slab_alias(struct kmem_cache *s, const char *p) { return 0; } | |
184 | static void sysfs_slab_remove(struct kmem_cache *s) {} | |
185 | #endif | |
186 | ||
187 | /******************************************************************** | |
188 | * Core slab cache functions | |
189 | *******************************************************************/ | |
190 | ||
191 | int slab_is_available(void) | |
192 | { | |
193 | return slab_state >= UP; | |
194 | } | |
195 | ||
196 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
197 | { | |
198 | #ifdef CONFIG_NUMA | |
199 | return s->node[node]; | |
200 | #else | |
201 | return &s->local_node; | |
202 | #endif | |
203 | } | |
204 | ||
205 | /* | |
206 | * Object debugging | |
207 | */ | |
208 | static void print_section(char *text, u8 *addr, unsigned int length) | |
209 | { | |
210 | int i, offset; | |
211 | int newline = 1; | |
212 | char ascii[17]; | |
213 | ||
214 | ascii[16] = 0; | |
215 | ||
216 | for (i = 0; i < length; i++) { | |
217 | if (newline) { | |
218 | printk(KERN_ERR "%10s 0x%p: ", text, addr + i); | |
219 | newline = 0; | |
220 | } | |
221 | printk(" %02x", addr[i]); | |
222 | offset = i % 16; | |
223 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
224 | if (offset == 15) { | |
225 | printk(" %s\n",ascii); | |
226 | newline = 1; | |
227 | } | |
228 | } | |
229 | if (!newline) { | |
230 | i %= 16; | |
231 | while (i < 16) { | |
232 | printk(" "); | |
233 | ascii[i] = ' '; | |
234 | i++; | |
235 | } | |
236 | printk(" %s\n", ascii); | |
237 | } | |
238 | } | |
239 | ||
240 | /* | |
241 | * Slow version of get and set free pointer. | |
242 | * | |
243 | * This requires touching the cache lines of kmem_cache. | |
244 | * The offset can also be obtained from the page. In that | |
245 | * case it is in the cacheline that we already need to touch. | |
246 | */ | |
247 | static void *get_freepointer(struct kmem_cache *s, void *object) | |
248 | { | |
249 | return *(void **)(object + s->offset); | |
250 | } | |
251 | ||
252 | static void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
253 | { | |
254 | *(void **)(object + s->offset) = fp; | |
255 | } | |
256 | ||
257 | /* | |
258 | * Tracking user of a slab. | |
259 | */ | |
260 | struct track { | |
261 | void *addr; /* Called from address */ | |
262 | int cpu; /* Was running on cpu */ | |
263 | int pid; /* Pid context */ | |
264 | unsigned long when; /* When did the operation occur */ | |
265 | }; | |
266 | ||
267 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
268 | ||
269 | static struct track *get_track(struct kmem_cache *s, void *object, | |
270 | enum track_item alloc) | |
271 | { | |
272 | struct track *p; | |
273 | ||
274 | if (s->offset) | |
275 | p = object + s->offset + sizeof(void *); | |
276 | else | |
277 | p = object + s->inuse; | |
278 | ||
279 | return p + alloc; | |
280 | } | |
281 | ||
282 | static void set_track(struct kmem_cache *s, void *object, | |
283 | enum track_item alloc, void *addr) | |
284 | { | |
285 | struct track *p; | |
286 | ||
287 | if (s->offset) | |
288 | p = object + s->offset + sizeof(void *); | |
289 | else | |
290 | p = object + s->inuse; | |
291 | ||
292 | p += alloc; | |
293 | if (addr) { | |
294 | p->addr = addr; | |
295 | p->cpu = smp_processor_id(); | |
296 | p->pid = current ? current->pid : -1; | |
297 | p->when = jiffies; | |
298 | } else | |
299 | memset(p, 0, sizeof(struct track)); | |
300 | } | |
301 | ||
81819f0f CL |
302 | static void init_tracking(struct kmem_cache *s, void *object) |
303 | { | |
304 | if (s->flags & SLAB_STORE_USER) { | |
305 | set_track(s, object, TRACK_FREE, NULL); | |
306 | set_track(s, object, TRACK_ALLOC, NULL); | |
307 | } | |
308 | } | |
309 | ||
310 | static void print_track(const char *s, struct track *t) | |
311 | { | |
312 | if (!t->addr) | |
313 | return; | |
314 | ||
315 | printk(KERN_ERR "%s: ", s); | |
316 | __print_symbol("%s", (unsigned long)t->addr); | |
317 | printk(" jiffies_ago=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); | |
318 | } | |
319 | ||
320 | static void print_trailer(struct kmem_cache *s, u8 *p) | |
321 | { | |
322 | unsigned int off; /* Offset of last byte */ | |
323 | ||
324 | if (s->flags & SLAB_RED_ZONE) | |
325 | print_section("Redzone", p + s->objsize, | |
326 | s->inuse - s->objsize); | |
327 | ||
328 | printk(KERN_ERR "FreePointer 0x%p -> 0x%p\n", | |
329 | p + s->offset, | |
330 | get_freepointer(s, p)); | |
331 | ||
332 | if (s->offset) | |
333 | off = s->offset + sizeof(void *); | |
334 | else | |
335 | off = s->inuse; | |
336 | ||
337 | if (s->flags & SLAB_STORE_USER) { | |
338 | print_track("Last alloc", get_track(s, p, TRACK_ALLOC)); | |
339 | print_track("Last free ", get_track(s, p, TRACK_FREE)); | |
340 | off += 2 * sizeof(struct track); | |
341 | } | |
342 | ||
343 | if (off != s->size) | |
344 | /* Beginning of the filler is the free pointer */ | |
345 | print_section("Filler", p + off, s->size - off); | |
346 | } | |
347 | ||
348 | static void object_err(struct kmem_cache *s, struct page *page, | |
349 | u8 *object, char *reason) | |
350 | { | |
351 | u8 *addr = page_address(page); | |
352 | ||
353 | printk(KERN_ERR "*** SLUB %s: %s@0x%p slab 0x%p\n", | |
354 | s->name, reason, object, page); | |
355 | printk(KERN_ERR " offset=%tu flags=0x%04lx inuse=%u freelist=0x%p\n", | |
356 | object - addr, page->flags, page->inuse, page->freelist); | |
357 | if (object > addr + 16) | |
358 | print_section("Bytes b4", object - 16, 16); | |
359 | print_section("Object", object, min(s->objsize, 128)); | |
360 | print_trailer(s, object); | |
361 | dump_stack(); | |
362 | } | |
363 | ||
364 | static void slab_err(struct kmem_cache *s, struct page *page, char *reason, ...) | |
365 | { | |
366 | va_list args; | |
367 | char buf[100]; | |
368 | ||
369 | va_start(args, reason); | |
370 | vsnprintf(buf, sizeof(buf), reason, args); | |
371 | va_end(args); | |
372 | printk(KERN_ERR "*** SLUB %s: %s in slab @0x%p\n", s->name, buf, | |
373 | page); | |
374 | dump_stack(); | |
375 | } | |
376 | ||
377 | static void init_object(struct kmem_cache *s, void *object, int active) | |
378 | { | |
379 | u8 *p = object; | |
380 | ||
381 | if (s->flags & __OBJECT_POISON) { | |
382 | memset(p, POISON_FREE, s->objsize - 1); | |
383 | p[s->objsize -1] = POISON_END; | |
384 | } | |
385 | ||
386 | if (s->flags & SLAB_RED_ZONE) | |
387 | memset(p + s->objsize, | |
388 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
389 | s->inuse - s->objsize); | |
390 | } | |
391 | ||
392 | static int check_bytes(u8 *start, unsigned int value, unsigned int bytes) | |
393 | { | |
394 | while (bytes) { | |
395 | if (*start != (u8)value) | |
396 | return 0; | |
397 | start++; | |
398 | bytes--; | |
399 | } | |
400 | return 1; | |
401 | } | |
402 | ||
403 | ||
404 | static int check_valid_pointer(struct kmem_cache *s, struct page *page, | |
405 | void *object) | |
406 | { | |
407 | void *base; | |
408 | ||
409 | if (!object) | |
410 | return 1; | |
411 | ||
412 | base = page_address(page); | |
413 | if (object < base || object >= base + s->objects * s->size || | |
414 | (object - base) % s->size) { | |
415 | return 0; | |
416 | } | |
417 | ||
418 | return 1; | |
419 | } | |
420 | ||
421 | /* | |
422 | * Object layout: | |
423 | * | |
424 | * object address | |
425 | * Bytes of the object to be managed. | |
426 | * If the freepointer may overlay the object then the free | |
427 | * pointer is the first word of the object. | |
428 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is | |
429 | * 0xa5 (POISON_END) | |
430 | * | |
431 | * object + s->objsize | |
432 | * Padding to reach word boundary. This is also used for Redzoning. | |
433 | * Padding is extended to word size if Redzoning is enabled | |
434 | * and objsize == inuse. | |
435 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with | |
436 | * 0xcc (RED_ACTIVE) for objects in use. | |
437 | * | |
438 | * object + s->inuse | |
439 | * A. Free pointer (if we cannot overwrite object on free) | |
440 | * B. Tracking data for SLAB_STORE_USER | |
441 | * C. Padding to reach required alignment boundary | |
442 | * Padding is done using 0x5a (POISON_INUSE) | |
443 | * | |
444 | * object + s->size | |
445 | * | |
446 | * If slabcaches are merged then the objsize and inuse boundaries are to | |
447 | * be ignored. And therefore no slab options that rely on these boundaries | |
448 | * may be used with merged slabcaches. | |
449 | */ | |
450 | ||
451 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
452 | void *from, void *to) | |
453 | { | |
70d71228 | 454 | printk(KERN_ERR "@@@ SLUB %s: Restoring %s (0x%x) from 0x%p-0x%p\n", |
81819f0f CL |
455 | s->name, message, data, from, to - 1); |
456 | memset(from, data, to - from); | |
457 | } | |
458 | ||
459 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) | |
460 | { | |
461 | unsigned long off = s->inuse; /* The end of info */ | |
462 | ||
463 | if (s->offset) | |
464 | /* Freepointer is placed after the object. */ | |
465 | off += sizeof(void *); | |
466 | ||
467 | if (s->flags & SLAB_STORE_USER) | |
468 | /* We also have user information there */ | |
469 | off += 2 * sizeof(struct track); | |
470 | ||
471 | if (s->size == off) | |
472 | return 1; | |
473 | ||
474 | if (check_bytes(p + off, POISON_INUSE, s->size - off)) | |
475 | return 1; | |
476 | ||
477 | object_err(s, page, p, "Object padding check fails"); | |
478 | ||
479 | /* | |
480 | * Restore padding | |
481 | */ | |
482 | restore_bytes(s, "object padding", POISON_INUSE, p + off, p + s->size); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
487 | { | |
488 | u8 *p; | |
489 | int length, remainder; | |
490 | ||
491 | if (!(s->flags & SLAB_POISON)) | |
492 | return 1; | |
493 | ||
494 | p = page_address(page); | |
495 | length = s->objects * s->size; | |
496 | remainder = (PAGE_SIZE << s->order) - length; | |
497 | if (!remainder) | |
498 | return 1; | |
499 | ||
500 | if (!check_bytes(p + length, POISON_INUSE, remainder)) { | |
70d71228 | 501 | slab_err(s, page, "Padding check failed"); |
81819f0f CL |
502 | restore_bytes(s, "slab padding", POISON_INUSE, p + length, |
503 | p + length + remainder); | |
504 | return 0; | |
505 | } | |
506 | return 1; | |
507 | } | |
508 | ||
509 | static int check_object(struct kmem_cache *s, struct page *page, | |
510 | void *object, int active) | |
511 | { | |
512 | u8 *p = object; | |
513 | u8 *endobject = object + s->objsize; | |
514 | ||
515 | if (s->flags & SLAB_RED_ZONE) { | |
516 | unsigned int red = | |
517 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
518 | ||
519 | if (!check_bytes(endobject, red, s->inuse - s->objsize)) { | |
520 | object_err(s, page, object, | |
521 | active ? "Redzone Active" : "Redzone Inactive"); | |
522 | restore_bytes(s, "redzone", red, | |
523 | endobject, object + s->inuse); | |
524 | return 0; | |
525 | } | |
526 | } else { | |
527 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse && | |
528 | !check_bytes(endobject, POISON_INUSE, | |
529 | s->inuse - s->objsize)) { | |
530 | object_err(s, page, p, "Alignment padding check fails"); | |
531 | /* | |
532 | * Fix it so that there will not be another report. | |
533 | * | |
534 | * Hmmm... We may be corrupting an object that now expects | |
535 | * to be longer than allowed. | |
536 | */ | |
537 | restore_bytes(s, "alignment padding", POISON_INUSE, | |
538 | endobject, object + s->inuse); | |
539 | } | |
540 | } | |
541 | ||
542 | if (s->flags & SLAB_POISON) { | |
543 | if (!active && (s->flags & __OBJECT_POISON) && | |
544 | (!check_bytes(p, POISON_FREE, s->objsize - 1) || | |
545 | p[s->objsize - 1] != POISON_END)) { | |
546 | ||
547 | object_err(s, page, p, "Poison check failed"); | |
548 | restore_bytes(s, "Poison", POISON_FREE, | |
549 | p, p + s->objsize -1); | |
550 | restore_bytes(s, "Poison", POISON_END, | |
551 | p + s->objsize - 1, p + s->objsize); | |
552 | return 0; | |
553 | } | |
554 | /* | |
555 | * check_pad_bytes cleans up on its own. | |
556 | */ | |
557 | check_pad_bytes(s, page, p); | |
558 | } | |
559 | ||
560 | if (!s->offset && active) | |
561 | /* | |
562 | * Object and freepointer overlap. Cannot check | |
563 | * freepointer while object is allocated. | |
564 | */ | |
565 | return 1; | |
566 | ||
567 | /* Check free pointer validity */ | |
568 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
569 | object_err(s, page, p, "Freepointer corrupt"); | |
570 | /* | |
571 | * No choice but to zap it and thus loose the remainder | |
572 | * of the free objects in this slab. May cause | |
573 | * another error because the object count maybe | |
574 | * wrong now. | |
575 | */ | |
576 | set_freepointer(s, p, NULL); | |
577 | return 0; | |
578 | } | |
579 | return 1; | |
580 | } | |
581 | ||
582 | static int check_slab(struct kmem_cache *s, struct page *page) | |
583 | { | |
584 | VM_BUG_ON(!irqs_disabled()); | |
585 | ||
586 | if (!PageSlab(page)) { | |
70d71228 CL |
587 | slab_err(s, page, "Not a valid slab page flags=%lx " |
588 | "mapping=0x%p count=%d", page->flags, page->mapping, | |
81819f0f CL |
589 | page_count(page)); |
590 | return 0; | |
591 | } | |
592 | if (page->offset * sizeof(void *) != s->offset) { | |
70d71228 CL |
593 | slab_err(s, page, "Corrupted offset %lu flags=0x%lx " |
594 | "mapping=0x%p count=%d", | |
81819f0f | 595 | (unsigned long)(page->offset * sizeof(void *)), |
81819f0f CL |
596 | page->flags, |
597 | page->mapping, | |
598 | page_count(page)); | |
81819f0f CL |
599 | return 0; |
600 | } | |
601 | if (page->inuse > s->objects) { | |
70d71228 CL |
602 | slab_err(s, page, "inuse %u > max %u @0x%p flags=%lx " |
603 | "mapping=0x%p count=%d", | |
604 | s->name, page->inuse, s->objects, page->flags, | |
81819f0f | 605 | page->mapping, page_count(page)); |
81819f0f CL |
606 | return 0; |
607 | } | |
608 | /* Slab_pad_check fixes things up after itself */ | |
609 | slab_pad_check(s, page); | |
610 | return 1; | |
611 | } | |
612 | ||
613 | /* | |
614 | * Determine if a certain object on a page is on the freelist and | |
615 | * therefore free. Must hold the slab lock for cpu slabs to | |
616 | * guarantee that the chains are consistent. | |
617 | */ | |
618 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
619 | { | |
620 | int nr = 0; | |
621 | void *fp = page->freelist; | |
622 | void *object = NULL; | |
623 | ||
624 | while (fp && nr <= s->objects) { | |
625 | if (fp == search) | |
626 | return 1; | |
627 | if (!check_valid_pointer(s, page, fp)) { | |
628 | if (object) { | |
629 | object_err(s, page, object, | |
630 | "Freechain corrupt"); | |
631 | set_freepointer(s, object, NULL); | |
632 | break; | |
633 | } else { | |
70d71228 CL |
634 | slab_err(s, page, "Freepointer 0x%p corrupt", |
635 | fp); | |
81819f0f CL |
636 | page->freelist = NULL; |
637 | page->inuse = s->objects; | |
70d71228 CL |
638 | printk(KERN_ERR "@@@ SLUB %s: Freelist " |
639 | "cleared. Slab 0x%p\n", | |
640 | s->name, page); | |
81819f0f CL |
641 | return 0; |
642 | } | |
643 | break; | |
644 | } | |
645 | object = fp; | |
646 | fp = get_freepointer(s, object); | |
647 | nr++; | |
648 | } | |
649 | ||
650 | if (page->inuse != s->objects - nr) { | |
70d71228 CL |
651 | slab_err(s, page, "Wrong object count. Counter is %d but " |
652 | "counted were %d", s, page, page->inuse, | |
653 | s->objects - nr); | |
81819f0f | 654 | page->inuse = s->objects - nr; |
70d71228 CL |
655 | printk(KERN_ERR "@@@ SLUB %s: Object count adjusted. " |
656 | "Slab @0x%p\n", s->name, page); | |
81819f0f CL |
657 | } |
658 | return search == NULL; | |
659 | } | |
660 | ||
643b1138 CL |
661 | /* |
662 | * Tracking of fully allocated slabs for debugging | |
663 | */ | |
e95eed57 | 664 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 665 | { |
643b1138 CL |
666 | spin_lock(&n->list_lock); |
667 | list_add(&page->lru, &n->full); | |
668 | spin_unlock(&n->list_lock); | |
669 | } | |
670 | ||
671 | static void remove_full(struct kmem_cache *s, struct page *page) | |
672 | { | |
673 | struct kmem_cache_node *n; | |
674 | ||
675 | if (!(s->flags & SLAB_STORE_USER)) | |
676 | return; | |
677 | ||
678 | n = get_node(s, page_to_nid(page)); | |
679 | ||
680 | spin_lock(&n->list_lock); | |
681 | list_del(&page->lru); | |
682 | spin_unlock(&n->list_lock); | |
683 | } | |
684 | ||
81819f0f CL |
685 | static int alloc_object_checks(struct kmem_cache *s, struct page *page, |
686 | void *object) | |
687 | { | |
688 | if (!check_slab(s, page)) | |
689 | goto bad; | |
690 | ||
691 | if (object && !on_freelist(s, page, object)) { | |
70d71228 CL |
692 | slab_err(s, page, "Object 0x%p already allocated", object); |
693 | goto bad; | |
81819f0f CL |
694 | } |
695 | ||
696 | if (!check_valid_pointer(s, page, object)) { | |
697 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 698 | goto bad; |
81819f0f CL |
699 | } |
700 | ||
701 | if (!object) | |
702 | return 1; | |
703 | ||
704 | if (!check_object(s, page, object, 0)) | |
705 | goto bad; | |
81819f0f | 706 | |
81819f0f | 707 | return 1; |
81819f0f CL |
708 | bad: |
709 | if (PageSlab(page)) { | |
710 | /* | |
711 | * If this is a slab page then lets do the best we can | |
712 | * to avoid issues in the future. Marking all objects | |
713 | * as used avoids touching the remainder. | |
714 | */ | |
715 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p. Marking all objects used.\n", | |
716 | s->name, page); | |
717 | page->inuse = s->objects; | |
718 | page->freelist = NULL; | |
719 | /* Fix up fields that may be corrupted */ | |
720 | page->offset = s->offset / sizeof(void *); | |
721 | } | |
722 | return 0; | |
723 | } | |
724 | ||
725 | static int free_object_checks(struct kmem_cache *s, struct page *page, | |
726 | void *object) | |
727 | { | |
728 | if (!check_slab(s, page)) | |
729 | goto fail; | |
730 | ||
731 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 732 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
733 | goto fail; |
734 | } | |
735 | ||
736 | if (on_freelist(s, page, object)) { | |
70d71228 | 737 | slab_err(s, page, "Object 0x%p already free", object); |
81819f0f CL |
738 | goto fail; |
739 | } | |
740 | ||
741 | if (!check_object(s, page, object, 1)) | |
742 | return 0; | |
743 | ||
744 | if (unlikely(s != page->slab)) { | |
745 | if (!PageSlab(page)) | |
70d71228 CL |
746 | slab_err(s, page, "Attempt to free object(0x%p) " |
747 | "outside of slab", object); | |
81819f0f | 748 | else |
70d71228 | 749 | if (!page->slab) { |
81819f0f | 750 | printk(KERN_ERR |
70d71228 | 751 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 752 | object); |
70d71228 CL |
753 | dump_stack(); |
754 | } | |
81819f0f | 755 | else |
70d71228 CL |
756 | slab_err(s, page, "object at 0x%p belongs " |
757 | "to slab %s", object, page->slab->name); | |
81819f0f CL |
758 | goto fail; |
759 | } | |
81819f0f CL |
760 | return 1; |
761 | fail: | |
81819f0f CL |
762 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p object at 0x%p not freed.\n", |
763 | s->name, page, object); | |
764 | return 0; | |
765 | } | |
766 | ||
767 | /* | |
768 | * Slab allocation and freeing | |
769 | */ | |
770 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
771 | { | |
772 | struct page * page; | |
773 | int pages = 1 << s->order; | |
774 | ||
775 | if (s->order) | |
776 | flags |= __GFP_COMP; | |
777 | ||
778 | if (s->flags & SLAB_CACHE_DMA) | |
779 | flags |= SLUB_DMA; | |
780 | ||
781 | if (node == -1) | |
782 | page = alloc_pages(flags, s->order); | |
783 | else | |
784 | page = alloc_pages_node(node, flags, s->order); | |
785 | ||
786 | if (!page) | |
787 | return NULL; | |
788 | ||
789 | mod_zone_page_state(page_zone(page), | |
790 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
791 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
792 | pages); | |
793 | ||
794 | return page; | |
795 | } | |
796 | ||
797 | static void setup_object(struct kmem_cache *s, struct page *page, | |
798 | void *object) | |
799 | { | |
800 | if (PageError(page)) { | |
801 | init_object(s, object, 0); | |
802 | init_tracking(s, object); | |
803 | } | |
804 | ||
805 | if (unlikely(s->ctor)) { | |
806 | int mode = SLAB_CTOR_CONSTRUCTOR; | |
807 | ||
808 | if (!(s->flags & __GFP_WAIT)) | |
809 | mode |= SLAB_CTOR_ATOMIC; | |
810 | ||
811 | s->ctor(object, s, mode); | |
812 | } | |
813 | } | |
814 | ||
815 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
816 | { | |
817 | struct page *page; | |
818 | struct kmem_cache_node *n; | |
819 | void *start; | |
820 | void *end; | |
821 | void *last; | |
822 | void *p; | |
823 | ||
824 | if (flags & __GFP_NO_GROW) | |
825 | return NULL; | |
826 | ||
827 | BUG_ON(flags & ~(GFP_DMA | GFP_LEVEL_MASK)); | |
828 | ||
829 | if (flags & __GFP_WAIT) | |
830 | local_irq_enable(); | |
831 | ||
832 | page = allocate_slab(s, flags & GFP_LEVEL_MASK, node); | |
833 | if (!page) | |
834 | goto out; | |
835 | ||
836 | n = get_node(s, page_to_nid(page)); | |
837 | if (n) | |
838 | atomic_long_inc(&n->nr_slabs); | |
839 | page->offset = s->offset / sizeof(void *); | |
840 | page->slab = s; | |
841 | page->flags |= 1 << PG_slab; | |
842 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
843 | SLAB_STORE_USER | SLAB_TRACE)) | |
844 | page->flags |= 1 << PG_error; | |
845 | ||
846 | start = page_address(page); | |
847 | end = start + s->objects * s->size; | |
848 | ||
849 | if (unlikely(s->flags & SLAB_POISON)) | |
850 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
851 | ||
852 | last = start; | |
853 | for (p = start + s->size; p < end; p += s->size) { | |
854 | setup_object(s, page, last); | |
855 | set_freepointer(s, last, p); | |
856 | last = p; | |
857 | } | |
858 | setup_object(s, page, last); | |
859 | set_freepointer(s, last, NULL); | |
860 | ||
861 | page->freelist = start; | |
862 | page->inuse = 0; | |
863 | out: | |
864 | if (flags & __GFP_WAIT) | |
865 | local_irq_disable(); | |
866 | return page; | |
867 | } | |
868 | ||
869 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
870 | { | |
871 | int pages = 1 << s->order; | |
872 | ||
873 | if (unlikely(PageError(page) || s->dtor)) { | |
874 | void *start = page_address(page); | |
875 | void *end = start + (pages << PAGE_SHIFT); | |
876 | void *p; | |
877 | ||
878 | slab_pad_check(s, page); | |
879 | for (p = start; p <= end - s->size; p += s->size) { | |
880 | if (s->dtor) | |
881 | s->dtor(p, s, 0); | |
882 | check_object(s, page, p, 0); | |
883 | } | |
884 | } | |
885 | ||
886 | mod_zone_page_state(page_zone(page), | |
887 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
888 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
889 | - pages); | |
890 | ||
891 | page->mapping = NULL; | |
892 | __free_pages(page, s->order); | |
893 | } | |
894 | ||
895 | static void rcu_free_slab(struct rcu_head *h) | |
896 | { | |
897 | struct page *page; | |
898 | ||
899 | page = container_of((struct list_head *)h, struct page, lru); | |
900 | __free_slab(page->slab, page); | |
901 | } | |
902 | ||
903 | static void free_slab(struct kmem_cache *s, struct page *page) | |
904 | { | |
905 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
906 | /* | |
907 | * RCU free overloads the RCU head over the LRU | |
908 | */ | |
909 | struct rcu_head *head = (void *)&page->lru; | |
910 | ||
911 | call_rcu(head, rcu_free_slab); | |
912 | } else | |
913 | __free_slab(s, page); | |
914 | } | |
915 | ||
916 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
917 | { | |
918 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
919 | ||
920 | atomic_long_dec(&n->nr_slabs); | |
921 | reset_page_mapcount(page); | |
922 | page->flags &= ~(1 << PG_slab | 1 << PG_error); | |
923 | free_slab(s, page); | |
924 | } | |
925 | ||
926 | /* | |
927 | * Per slab locking using the pagelock | |
928 | */ | |
929 | static __always_inline void slab_lock(struct page *page) | |
930 | { | |
931 | bit_spin_lock(PG_locked, &page->flags); | |
932 | } | |
933 | ||
934 | static __always_inline void slab_unlock(struct page *page) | |
935 | { | |
936 | bit_spin_unlock(PG_locked, &page->flags); | |
937 | } | |
938 | ||
939 | static __always_inline int slab_trylock(struct page *page) | |
940 | { | |
941 | int rc = 1; | |
942 | ||
943 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
944 | return rc; | |
945 | } | |
946 | ||
947 | /* | |
948 | * Management of partially allocated slabs | |
949 | */ | |
e95eed57 | 950 | static void add_partial_tail(struct kmem_cache_node *n, struct page *page) |
81819f0f | 951 | { |
e95eed57 CL |
952 | spin_lock(&n->list_lock); |
953 | n->nr_partial++; | |
954 | list_add_tail(&page->lru, &n->partial); | |
955 | spin_unlock(&n->list_lock); | |
956 | } | |
81819f0f | 957 | |
e95eed57 CL |
958 | static void add_partial(struct kmem_cache_node *n, struct page *page) |
959 | { | |
81819f0f CL |
960 | spin_lock(&n->list_lock); |
961 | n->nr_partial++; | |
962 | list_add(&page->lru, &n->partial); | |
963 | spin_unlock(&n->list_lock); | |
964 | } | |
965 | ||
966 | static void remove_partial(struct kmem_cache *s, | |
967 | struct page *page) | |
968 | { | |
969 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
970 | ||
971 | spin_lock(&n->list_lock); | |
972 | list_del(&page->lru); | |
973 | n->nr_partial--; | |
974 | spin_unlock(&n->list_lock); | |
975 | } | |
976 | ||
977 | /* | |
978 | * Lock page and remove it from the partial list | |
979 | * | |
980 | * Must hold list_lock | |
981 | */ | |
982 | static int lock_and_del_slab(struct kmem_cache_node *n, struct page *page) | |
983 | { | |
984 | if (slab_trylock(page)) { | |
985 | list_del(&page->lru); | |
986 | n->nr_partial--; | |
987 | return 1; | |
988 | } | |
989 | return 0; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Try to get a partial slab from a specific node | |
994 | */ | |
995 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
996 | { | |
997 | struct page *page; | |
998 | ||
999 | /* | |
1000 | * Racy check. If we mistakenly see no partial slabs then we | |
1001 | * just allocate an empty slab. If we mistakenly try to get a | |
1002 | * partial slab then get_partials() will return NULL. | |
1003 | */ | |
1004 | if (!n || !n->nr_partial) | |
1005 | return NULL; | |
1006 | ||
1007 | spin_lock(&n->list_lock); | |
1008 | list_for_each_entry(page, &n->partial, lru) | |
1009 | if (lock_and_del_slab(n, page)) | |
1010 | goto out; | |
1011 | page = NULL; | |
1012 | out: | |
1013 | spin_unlock(&n->list_lock); | |
1014 | return page; | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Get a page from somewhere. Search in increasing NUMA | |
1019 | * distances. | |
1020 | */ | |
1021 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1022 | { | |
1023 | #ifdef CONFIG_NUMA | |
1024 | struct zonelist *zonelist; | |
1025 | struct zone **z; | |
1026 | struct page *page; | |
1027 | ||
1028 | /* | |
1029 | * The defrag ratio allows to configure the tradeoffs between | |
1030 | * inter node defragmentation and node local allocations. | |
1031 | * A lower defrag_ratio increases the tendency to do local | |
1032 | * allocations instead of scanning throught the partial | |
1033 | * lists on other nodes. | |
1034 | * | |
1035 | * If defrag_ratio is set to 0 then kmalloc() always | |
1036 | * returns node local objects. If its higher then kmalloc() | |
1037 | * may return off node objects in order to avoid fragmentation. | |
1038 | * | |
1039 | * A higher ratio means slabs may be taken from other nodes | |
1040 | * thus reducing the number of partial slabs on those nodes. | |
1041 | * | |
1042 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
1043 | * defrag_ratio = 1000) then every (well almost) allocation | |
1044 | * will first attempt to defrag slab caches on other nodes. This | |
1045 | * means scanning over all nodes to look for partial slabs which | |
1046 | * may be a bit expensive to do on every slab allocation. | |
1047 | */ | |
1048 | if (!s->defrag_ratio || get_cycles() % 1024 > s->defrag_ratio) | |
1049 | return NULL; | |
1050 | ||
1051 | zonelist = &NODE_DATA(slab_node(current->mempolicy)) | |
1052 | ->node_zonelists[gfp_zone(flags)]; | |
1053 | for (z = zonelist->zones; *z; z++) { | |
1054 | struct kmem_cache_node *n; | |
1055 | ||
1056 | n = get_node(s, zone_to_nid(*z)); | |
1057 | ||
1058 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1059 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1060 | page = get_partial_node(n); |
1061 | if (page) | |
1062 | return page; | |
1063 | } | |
1064 | } | |
1065 | #endif | |
1066 | return NULL; | |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * Get a partial page, lock it and return it. | |
1071 | */ | |
1072 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1073 | { | |
1074 | struct page *page; | |
1075 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1076 | ||
1077 | page = get_partial_node(get_node(s, searchnode)); | |
1078 | if (page || (flags & __GFP_THISNODE)) | |
1079 | return page; | |
1080 | ||
1081 | return get_any_partial(s, flags); | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Move a page back to the lists. | |
1086 | * | |
1087 | * Must be called with the slab lock held. | |
1088 | * | |
1089 | * On exit the slab lock will have been dropped. | |
1090 | */ | |
1091 | static void putback_slab(struct kmem_cache *s, struct page *page) | |
1092 | { | |
e95eed57 CL |
1093 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1094 | ||
81819f0f | 1095 | if (page->inuse) { |
e95eed57 | 1096 | |
81819f0f | 1097 | if (page->freelist) |
e95eed57 CL |
1098 | add_partial(n, page); |
1099 | else if (PageError(page) && (s->flags & SLAB_STORE_USER)) | |
1100 | add_full(n, page); | |
81819f0f | 1101 | slab_unlock(page); |
e95eed57 | 1102 | |
81819f0f | 1103 | } else { |
e95eed57 CL |
1104 | if (n->nr_partial < MIN_PARTIAL) { |
1105 | /* | |
1106 | * Adding an empty page to the partial slabs in order | |
1107 | * to avoid page allocator overhead. This page needs to | |
1108 | * come after all the others that are not fully empty | |
1109 | * in order to make sure that we do maximum | |
1110 | * defragmentation. | |
1111 | */ | |
1112 | add_partial_tail(n, page); | |
1113 | slab_unlock(page); | |
1114 | } else { | |
1115 | slab_unlock(page); | |
1116 | discard_slab(s, page); | |
1117 | } | |
81819f0f CL |
1118 | } |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * Remove the cpu slab | |
1123 | */ | |
1124 | static void deactivate_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1125 | { | |
1126 | s->cpu_slab[cpu] = NULL; | |
1127 | ClearPageActive(page); | |
1128 | ||
1129 | putback_slab(s, page); | |
1130 | } | |
1131 | ||
1132 | static void flush_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1133 | { | |
1134 | slab_lock(page); | |
1135 | deactivate_slab(s, page, cpu); | |
1136 | } | |
1137 | ||
1138 | /* | |
1139 | * Flush cpu slab. | |
1140 | * Called from IPI handler with interrupts disabled. | |
1141 | */ | |
1142 | static void __flush_cpu_slab(struct kmem_cache *s, int cpu) | |
1143 | { | |
1144 | struct page *page = s->cpu_slab[cpu]; | |
1145 | ||
1146 | if (likely(page)) | |
1147 | flush_slab(s, page, cpu); | |
1148 | } | |
1149 | ||
1150 | static void flush_cpu_slab(void *d) | |
1151 | { | |
1152 | struct kmem_cache *s = d; | |
1153 | int cpu = smp_processor_id(); | |
1154 | ||
1155 | __flush_cpu_slab(s, cpu); | |
1156 | } | |
1157 | ||
1158 | static void flush_all(struct kmem_cache *s) | |
1159 | { | |
1160 | #ifdef CONFIG_SMP | |
1161 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1162 | #else | |
1163 | unsigned long flags; | |
1164 | ||
1165 | local_irq_save(flags); | |
1166 | flush_cpu_slab(s); | |
1167 | local_irq_restore(flags); | |
1168 | #endif | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * slab_alloc is optimized to only modify two cachelines on the fast path | |
1173 | * (aside from the stack): | |
1174 | * | |
1175 | * 1. The page struct | |
1176 | * 2. The first cacheline of the object to be allocated. | |
1177 | * | |
1178 | * The only cache lines that are read (apart from code) is the | |
1179 | * per cpu array in the kmem_cache struct. | |
1180 | * | |
1181 | * Fastpath is not possible if we need to get a new slab or have | |
1182 | * debugging enabled (which means all slabs are marked with PageError) | |
1183 | */ | |
77c5e2d0 CL |
1184 | static void *slab_alloc(struct kmem_cache *s, |
1185 | gfp_t gfpflags, int node, void *addr) | |
81819f0f CL |
1186 | { |
1187 | struct page *page; | |
1188 | void **object; | |
1189 | unsigned long flags; | |
1190 | int cpu; | |
1191 | ||
1192 | local_irq_save(flags); | |
1193 | cpu = smp_processor_id(); | |
1194 | page = s->cpu_slab[cpu]; | |
1195 | if (!page) | |
1196 | goto new_slab; | |
1197 | ||
1198 | slab_lock(page); | |
1199 | if (unlikely(node != -1 && page_to_nid(page) != node)) | |
1200 | goto another_slab; | |
1201 | redo: | |
1202 | object = page->freelist; | |
1203 | if (unlikely(!object)) | |
1204 | goto another_slab; | |
1205 | if (unlikely(PageError(page))) | |
1206 | goto debug; | |
1207 | ||
1208 | have_object: | |
1209 | page->inuse++; | |
1210 | page->freelist = object[page->offset]; | |
1211 | slab_unlock(page); | |
1212 | local_irq_restore(flags); | |
1213 | return object; | |
1214 | ||
1215 | another_slab: | |
1216 | deactivate_slab(s, page, cpu); | |
1217 | ||
1218 | new_slab: | |
1219 | page = get_partial(s, gfpflags, node); | |
1220 | if (likely(page)) { | |
1221 | have_slab: | |
1222 | s->cpu_slab[cpu] = page; | |
1223 | SetPageActive(page); | |
1224 | goto redo; | |
1225 | } | |
1226 | ||
1227 | page = new_slab(s, gfpflags, node); | |
1228 | if (page) { | |
1229 | cpu = smp_processor_id(); | |
1230 | if (s->cpu_slab[cpu]) { | |
1231 | /* | |
1232 | * Someone else populated the cpu_slab while we enabled | |
1233 | * interrupts, or we have got scheduled on another cpu. | |
1234 | * The page may not be on the requested node. | |
1235 | */ | |
1236 | if (node == -1 || | |
1237 | page_to_nid(s->cpu_slab[cpu]) == node) { | |
1238 | /* | |
1239 | * Current cpuslab is acceptable and we | |
1240 | * want the current one since its cache hot | |
1241 | */ | |
1242 | discard_slab(s, page); | |
1243 | page = s->cpu_slab[cpu]; | |
1244 | slab_lock(page); | |
1245 | goto redo; | |
1246 | } | |
1247 | /* Dump the current slab */ | |
1248 | flush_slab(s, s->cpu_slab[cpu], cpu); | |
1249 | } | |
1250 | slab_lock(page); | |
1251 | goto have_slab; | |
1252 | } | |
1253 | local_irq_restore(flags); | |
1254 | return NULL; | |
1255 | debug: | |
1256 | if (!alloc_object_checks(s, page, object)) | |
1257 | goto another_slab; | |
1258 | if (s->flags & SLAB_STORE_USER) | |
77c5e2d0 | 1259 | set_track(s, object, TRACK_ALLOC, addr); |
70d71228 CL |
1260 | if (s->flags & SLAB_TRACE) { |
1261 | printk(KERN_INFO "TRACE %s alloc 0x%p inuse=%d fp=0x%p\n", | |
1262 | s->name, object, page->inuse, | |
1263 | page->freelist); | |
1264 | dump_stack(); | |
1265 | } | |
1266 | init_object(s, object, 1); | |
81819f0f CL |
1267 | goto have_object; |
1268 | } | |
1269 | ||
1270 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1271 | { | |
77c5e2d0 | 1272 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1273 | } |
1274 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1275 | ||
1276 | #ifdef CONFIG_NUMA | |
1277 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1278 | { | |
77c5e2d0 | 1279 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1280 | } |
1281 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1282 | #endif | |
1283 | ||
1284 | /* | |
1285 | * The fastpath only writes the cacheline of the page struct and the first | |
1286 | * cacheline of the object. | |
1287 | * | |
1288 | * No special cachelines need to be read | |
1289 | */ | |
77c5e2d0 CL |
1290 | static void slab_free(struct kmem_cache *s, struct page *page, |
1291 | void *x, void *addr) | |
81819f0f CL |
1292 | { |
1293 | void *prior; | |
1294 | void **object = (void *)x; | |
1295 | unsigned long flags; | |
1296 | ||
1297 | local_irq_save(flags); | |
1298 | slab_lock(page); | |
1299 | ||
1300 | if (unlikely(PageError(page))) | |
1301 | goto debug; | |
1302 | checks_ok: | |
1303 | prior = object[page->offset] = page->freelist; | |
1304 | page->freelist = object; | |
1305 | page->inuse--; | |
1306 | ||
1307 | if (unlikely(PageActive(page))) | |
1308 | /* | |
1309 | * Cpu slabs are never on partial lists and are | |
1310 | * never freed. | |
1311 | */ | |
1312 | goto out_unlock; | |
1313 | ||
1314 | if (unlikely(!page->inuse)) | |
1315 | goto slab_empty; | |
1316 | ||
1317 | /* | |
1318 | * Objects left in the slab. If it | |
1319 | * was not on the partial list before | |
1320 | * then add it. | |
1321 | */ | |
1322 | if (unlikely(!prior)) | |
e95eed57 | 1323 | add_partial(get_node(s, page_to_nid(page)), page); |
81819f0f CL |
1324 | |
1325 | out_unlock: | |
1326 | slab_unlock(page); | |
1327 | local_irq_restore(flags); | |
1328 | return; | |
1329 | ||
1330 | slab_empty: | |
1331 | if (prior) | |
1332 | /* | |
643b1138 | 1333 | * Slab on the partial list. |
81819f0f CL |
1334 | */ |
1335 | remove_partial(s, page); | |
1336 | ||
1337 | slab_unlock(page); | |
1338 | discard_slab(s, page); | |
1339 | local_irq_restore(flags); | |
1340 | return; | |
1341 | ||
1342 | debug: | |
77c5e2d0 CL |
1343 | if (!free_object_checks(s, page, x)) |
1344 | goto out_unlock; | |
643b1138 CL |
1345 | if (!PageActive(page) && !page->freelist) |
1346 | remove_full(s, page); | |
77c5e2d0 CL |
1347 | if (s->flags & SLAB_STORE_USER) |
1348 | set_track(s, x, TRACK_FREE, addr); | |
70d71228 CL |
1349 | if (s->flags & SLAB_TRACE) { |
1350 | printk(KERN_INFO "TRACE %s free 0x%p inuse=%d fp=0x%p\n", | |
1351 | s->name, object, page->inuse, | |
1352 | page->freelist); | |
1353 | print_section("Object", (void *)object, s->objsize); | |
1354 | dump_stack(); | |
1355 | } | |
1356 | init_object(s, object, 0); | |
77c5e2d0 | 1357 | goto checks_ok; |
81819f0f CL |
1358 | } |
1359 | ||
1360 | void kmem_cache_free(struct kmem_cache *s, void *x) | |
1361 | { | |
77c5e2d0 | 1362 | struct page *page; |
81819f0f | 1363 | |
b49af68f | 1364 | page = virt_to_head_page(x); |
81819f0f | 1365 | |
77c5e2d0 | 1366 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1367 | } |
1368 | EXPORT_SYMBOL(kmem_cache_free); | |
1369 | ||
1370 | /* Figure out on which slab object the object resides */ | |
1371 | static struct page *get_object_page(const void *x) | |
1372 | { | |
b49af68f | 1373 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1374 | |
1375 | if (!PageSlab(page)) | |
1376 | return NULL; | |
1377 | ||
1378 | return page; | |
1379 | } | |
1380 | ||
1381 | /* | |
1382 | * kmem_cache_open produces objects aligned at "size" and the first object | |
1383 | * is placed at offset 0 in the slab (We have no metainformation on the | |
1384 | * slab, all slabs are in essence "off slab"). | |
1385 | * | |
1386 | * In order to get the desired alignment one just needs to align the | |
1387 | * size. | |
1388 | * | |
1389 | * Notice that the allocation order determines the sizes of the per cpu | |
1390 | * caches. Each processor has always one slab available for allocations. | |
1391 | * Increasing the allocation order reduces the number of times that slabs | |
1392 | * must be moved on and off the partial lists and therefore may influence | |
1393 | * locking overhead. | |
1394 | * | |
1395 | * The offset is used to relocate the free list link in each object. It is | |
1396 | * therefore possible to move the free list link behind the object. This | |
1397 | * is necessary for RCU to work properly and also useful for debugging. | |
1398 | */ | |
1399 | ||
1400 | /* | |
1401 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1402 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1403 | * and increases the number of allocations possible without having to | |
1404 | * take the list_lock. | |
1405 | */ | |
1406 | static int slub_min_order; | |
1407 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
1408 | ||
1409 | /* | |
1410 | * Minimum number of objects per slab. This is necessary in order to | |
1411 | * reduce locking overhead. Similar to the queue size in SLAB. | |
1412 | */ | |
1413 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; | |
1414 | ||
1415 | /* | |
1416 | * Merge control. If this is set then no merging of slab caches will occur. | |
1417 | */ | |
1418 | static int slub_nomerge; | |
1419 | ||
1420 | /* | |
1421 | * Debug settings: | |
1422 | */ | |
1423 | static int slub_debug; | |
1424 | ||
1425 | static char *slub_debug_slabs; | |
1426 | ||
1427 | /* | |
1428 | * Calculate the order of allocation given an slab object size. | |
1429 | * | |
1430 | * The order of allocation has significant impact on other elements | |
1431 | * of the system. Generally order 0 allocations should be preferred | |
1432 | * since they do not cause fragmentation in the page allocator. Larger | |
1433 | * objects may have problems with order 0 because there may be too much | |
1434 | * space left unused in a slab. We go to a higher order if more than 1/8th | |
1435 | * of the slab would be wasted. | |
1436 | * | |
1437 | * In order to reach satisfactory performance we must ensure that | |
1438 | * a minimum number of objects is in one slab. Otherwise we may | |
1439 | * generate too much activity on the partial lists. This is less a | |
1440 | * concern for large slabs though. slub_max_order specifies the order | |
1441 | * where we begin to stop considering the number of objects in a slab. | |
1442 | * | |
1443 | * Higher order allocations also allow the placement of more objects | |
1444 | * in a slab and thereby reduce object handling overhead. If the user | |
1445 | * has requested a higher mininum order then we start with that one | |
1446 | * instead of zero. | |
1447 | */ | |
1448 | static int calculate_order(int size) | |
1449 | { | |
1450 | int order; | |
1451 | int rem; | |
1452 | ||
1453 | for (order = max(slub_min_order, fls(size - 1) - PAGE_SHIFT); | |
1454 | order < MAX_ORDER; order++) { | |
1455 | unsigned long slab_size = PAGE_SIZE << order; | |
1456 | ||
1457 | if (slub_max_order > order && | |
1458 | slab_size < slub_min_objects * size) | |
1459 | continue; | |
1460 | ||
1461 | if (slab_size < size) | |
1462 | continue; | |
1463 | ||
1464 | rem = slab_size % size; | |
1465 | ||
1466 | if (rem <= (PAGE_SIZE << order) / 8) | |
1467 | break; | |
1468 | ||
1469 | } | |
1470 | if (order >= MAX_ORDER) | |
1471 | return -E2BIG; | |
1472 | return order; | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * Function to figure out which alignment to use from the | |
1477 | * various ways of specifying it. | |
1478 | */ | |
1479 | static unsigned long calculate_alignment(unsigned long flags, | |
1480 | unsigned long align, unsigned long size) | |
1481 | { | |
1482 | /* | |
1483 | * If the user wants hardware cache aligned objects then | |
1484 | * follow that suggestion if the object is sufficiently | |
1485 | * large. | |
1486 | * | |
1487 | * The hardware cache alignment cannot override the | |
1488 | * specified alignment though. If that is greater | |
1489 | * then use it. | |
1490 | */ | |
5af60839 | 1491 | if ((flags & SLAB_HWCACHE_ALIGN) && |
81819f0f CL |
1492 | size > L1_CACHE_BYTES / 2) |
1493 | return max_t(unsigned long, align, L1_CACHE_BYTES); | |
1494 | ||
1495 | if (align < ARCH_SLAB_MINALIGN) | |
1496 | return ARCH_SLAB_MINALIGN; | |
1497 | ||
1498 | return ALIGN(align, sizeof(void *)); | |
1499 | } | |
1500 | ||
1501 | static void init_kmem_cache_node(struct kmem_cache_node *n) | |
1502 | { | |
1503 | n->nr_partial = 0; | |
1504 | atomic_long_set(&n->nr_slabs, 0); | |
1505 | spin_lock_init(&n->list_lock); | |
1506 | INIT_LIST_HEAD(&n->partial); | |
643b1138 | 1507 | INIT_LIST_HEAD(&n->full); |
81819f0f CL |
1508 | } |
1509 | ||
1510 | #ifdef CONFIG_NUMA | |
1511 | /* | |
1512 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
1513 | * slab on the node for this slabcache. There are no concurrent accesses | |
1514 | * possible. | |
1515 | * | |
1516 | * Note that this function only works on the kmalloc_node_cache | |
1517 | * when allocating for the kmalloc_node_cache. | |
1518 | */ | |
1519 | static struct kmem_cache_node * __init early_kmem_cache_node_alloc(gfp_t gfpflags, | |
1520 | int node) | |
1521 | { | |
1522 | struct page *page; | |
1523 | struct kmem_cache_node *n; | |
1524 | ||
1525 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
1526 | ||
1527 | page = new_slab(kmalloc_caches, gfpflags | GFP_THISNODE, node); | |
1528 | /* new_slab() disables interupts */ | |
1529 | local_irq_enable(); | |
1530 | ||
1531 | BUG_ON(!page); | |
1532 | n = page->freelist; | |
1533 | BUG_ON(!n); | |
1534 | page->freelist = get_freepointer(kmalloc_caches, n); | |
1535 | page->inuse++; | |
1536 | kmalloc_caches->node[node] = n; | |
1537 | init_object(kmalloc_caches, n, 1); | |
1538 | init_kmem_cache_node(n); | |
1539 | atomic_long_inc(&n->nr_slabs); | |
e95eed57 | 1540 | add_partial(n, page); |
81819f0f CL |
1541 | return n; |
1542 | } | |
1543 | ||
1544 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1545 | { | |
1546 | int node; | |
1547 | ||
1548 | for_each_online_node(node) { | |
1549 | struct kmem_cache_node *n = s->node[node]; | |
1550 | if (n && n != &s->local_node) | |
1551 | kmem_cache_free(kmalloc_caches, n); | |
1552 | s->node[node] = NULL; | |
1553 | } | |
1554 | } | |
1555 | ||
1556 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1557 | { | |
1558 | int node; | |
1559 | int local_node; | |
1560 | ||
1561 | if (slab_state >= UP) | |
1562 | local_node = page_to_nid(virt_to_page(s)); | |
1563 | else | |
1564 | local_node = 0; | |
1565 | ||
1566 | for_each_online_node(node) { | |
1567 | struct kmem_cache_node *n; | |
1568 | ||
1569 | if (local_node == node) | |
1570 | n = &s->local_node; | |
1571 | else { | |
1572 | if (slab_state == DOWN) { | |
1573 | n = early_kmem_cache_node_alloc(gfpflags, | |
1574 | node); | |
1575 | continue; | |
1576 | } | |
1577 | n = kmem_cache_alloc_node(kmalloc_caches, | |
1578 | gfpflags, node); | |
1579 | ||
1580 | if (!n) { | |
1581 | free_kmem_cache_nodes(s); | |
1582 | return 0; | |
1583 | } | |
1584 | ||
1585 | } | |
1586 | s->node[node] = n; | |
1587 | init_kmem_cache_node(n); | |
1588 | } | |
1589 | return 1; | |
1590 | } | |
1591 | #else | |
1592 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1593 | { | |
1594 | } | |
1595 | ||
1596 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1597 | { | |
1598 | init_kmem_cache_node(&s->local_node); | |
1599 | return 1; | |
1600 | } | |
1601 | #endif | |
1602 | ||
1603 | /* | |
1604 | * calculate_sizes() determines the order and the distribution of data within | |
1605 | * a slab object. | |
1606 | */ | |
1607 | static int calculate_sizes(struct kmem_cache *s) | |
1608 | { | |
1609 | unsigned long flags = s->flags; | |
1610 | unsigned long size = s->objsize; | |
1611 | unsigned long align = s->align; | |
1612 | ||
1613 | /* | |
1614 | * Determine if we can poison the object itself. If the user of | |
1615 | * the slab may touch the object after free or before allocation | |
1616 | * then we should never poison the object itself. | |
1617 | */ | |
1618 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
1619 | !s->ctor && !s->dtor) | |
1620 | s->flags |= __OBJECT_POISON; | |
1621 | else | |
1622 | s->flags &= ~__OBJECT_POISON; | |
1623 | ||
1624 | /* | |
1625 | * Round up object size to the next word boundary. We can only | |
1626 | * place the free pointer at word boundaries and this determines | |
1627 | * the possible location of the free pointer. | |
1628 | */ | |
1629 | size = ALIGN(size, sizeof(void *)); | |
1630 | ||
1631 | /* | |
1632 | * If we are redzoning then check if there is some space between the | |
1633 | * end of the object and the free pointer. If not then add an | |
1634 | * additional word, so that we can establish a redzone between | |
1635 | * the object and the freepointer to be able to check for overwrites. | |
1636 | */ | |
1637 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
1638 | size += sizeof(void *); | |
1639 | ||
1640 | /* | |
1641 | * With that we have determined how much of the slab is in actual | |
1642 | * use by the object. This is the potential offset to the free | |
1643 | * pointer. | |
1644 | */ | |
1645 | s->inuse = size; | |
1646 | ||
1647 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
1648 | s->ctor || s->dtor)) { | |
1649 | /* | |
1650 | * Relocate free pointer after the object if it is not | |
1651 | * permitted to overwrite the first word of the object on | |
1652 | * kmem_cache_free. | |
1653 | * | |
1654 | * This is the case if we do RCU, have a constructor or | |
1655 | * destructor or are poisoning the objects. | |
1656 | */ | |
1657 | s->offset = size; | |
1658 | size += sizeof(void *); | |
1659 | } | |
1660 | ||
1661 | if (flags & SLAB_STORE_USER) | |
1662 | /* | |
1663 | * Need to store information about allocs and frees after | |
1664 | * the object. | |
1665 | */ | |
1666 | size += 2 * sizeof(struct track); | |
1667 | ||
1668 | if (flags & DEBUG_DEFAULT_FLAGS) | |
1669 | /* | |
1670 | * Add some empty padding so that we can catch | |
1671 | * overwrites from earlier objects rather than let | |
1672 | * tracking information or the free pointer be | |
1673 | * corrupted if an user writes before the start | |
1674 | * of the object. | |
1675 | */ | |
1676 | size += sizeof(void *); | |
1677 | /* | |
1678 | * Determine the alignment based on various parameters that the | |
1679 | * user specified (this is unecessarily complex due to the attempt | |
1680 | * to be compatible with SLAB. Should be cleaned up some day). | |
1681 | */ | |
1682 | align = calculate_alignment(flags, align, s->objsize); | |
1683 | ||
1684 | /* | |
1685 | * SLUB stores one object immediately after another beginning from | |
1686 | * offset 0. In order to align the objects we have to simply size | |
1687 | * each object to conform to the alignment. | |
1688 | */ | |
1689 | size = ALIGN(size, align); | |
1690 | s->size = size; | |
1691 | ||
1692 | s->order = calculate_order(size); | |
1693 | if (s->order < 0) | |
1694 | return 0; | |
1695 | ||
1696 | /* | |
1697 | * Determine the number of objects per slab | |
1698 | */ | |
1699 | s->objects = (PAGE_SIZE << s->order) / size; | |
1700 | ||
1701 | /* | |
1702 | * Verify that the number of objects is within permitted limits. | |
1703 | * The page->inuse field is only 16 bit wide! So we cannot have | |
1704 | * more than 64k objects per slab. | |
1705 | */ | |
1706 | if (!s->objects || s->objects > 65535) | |
1707 | return 0; | |
1708 | return 1; | |
1709 | ||
1710 | } | |
1711 | ||
1712 | static int __init finish_bootstrap(void) | |
1713 | { | |
1714 | struct list_head *h; | |
1715 | int err; | |
1716 | ||
1717 | slab_state = SYSFS; | |
1718 | ||
1719 | list_for_each(h, &slab_caches) { | |
1720 | struct kmem_cache *s = | |
1721 | container_of(h, struct kmem_cache, list); | |
1722 | ||
1723 | err = sysfs_slab_add(s); | |
1724 | BUG_ON(err); | |
1725 | } | |
1726 | return 0; | |
1727 | } | |
1728 | ||
1729 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, | |
1730 | const char *name, size_t size, | |
1731 | size_t align, unsigned long flags, | |
1732 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
1733 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
1734 | { | |
1735 | memset(s, 0, kmem_size); | |
1736 | s->name = name; | |
1737 | s->ctor = ctor; | |
1738 | s->dtor = dtor; | |
1739 | s->objsize = size; | |
1740 | s->flags = flags; | |
1741 | s->align = align; | |
1742 | ||
81819f0f CL |
1743 | /* |
1744 | * The page->offset field is only 16 bit wide. This is an offset | |
1745 | * in units of words from the beginning of an object. If the slab | |
1746 | * size is bigger then we cannot move the free pointer behind the | |
1747 | * object anymore. | |
1748 | * | |
1749 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
1750 | * the limit is 512k. | |
1751 | * | |
1752 | * Debugging or ctor/dtors may create a need to move the free | |
1753 | * pointer. Fail if this happens. | |
1754 | */ | |
1755 | if (s->size >= 65535 * sizeof(void *)) { | |
1756 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
1757 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); | |
1758 | BUG_ON(ctor || dtor); | |
1759 | } | |
1760 | else | |
1761 | /* | |
1762 | * Enable debugging if selected on the kernel commandline. | |
1763 | */ | |
1764 | if (slub_debug && (!slub_debug_slabs || | |
1765 | strncmp(slub_debug_slabs, name, | |
1766 | strlen(slub_debug_slabs)) == 0)) | |
1767 | s->flags |= slub_debug; | |
1768 | ||
1769 | if (!calculate_sizes(s)) | |
1770 | goto error; | |
1771 | ||
1772 | s->refcount = 1; | |
1773 | #ifdef CONFIG_NUMA | |
1774 | s->defrag_ratio = 100; | |
1775 | #endif | |
1776 | ||
1777 | if (init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) | |
1778 | return 1; | |
1779 | error: | |
1780 | if (flags & SLAB_PANIC) | |
1781 | panic("Cannot create slab %s size=%lu realsize=%u " | |
1782 | "order=%u offset=%u flags=%lx\n", | |
1783 | s->name, (unsigned long)size, s->size, s->order, | |
1784 | s->offset, flags); | |
1785 | return 0; | |
1786 | } | |
1787 | EXPORT_SYMBOL(kmem_cache_open); | |
1788 | ||
1789 | /* | |
1790 | * Check if a given pointer is valid | |
1791 | */ | |
1792 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
1793 | { | |
1794 | struct page * page; | |
1795 | void *addr; | |
1796 | ||
1797 | page = get_object_page(object); | |
1798 | ||
1799 | if (!page || s != page->slab) | |
1800 | /* No slab or wrong slab */ | |
1801 | return 0; | |
1802 | ||
1803 | addr = page_address(page); | |
1804 | if (object < addr || object >= addr + s->objects * s->size) | |
1805 | /* Out of bounds */ | |
1806 | return 0; | |
1807 | ||
1808 | if ((object - addr) % s->size) | |
1809 | /* Improperly aligned */ | |
1810 | return 0; | |
1811 | ||
1812 | /* | |
1813 | * We could also check if the object is on the slabs freelist. | |
1814 | * But this would be too expensive and it seems that the main | |
1815 | * purpose of kmem_ptr_valid is to check if the object belongs | |
1816 | * to a certain slab. | |
1817 | */ | |
1818 | return 1; | |
1819 | } | |
1820 | EXPORT_SYMBOL(kmem_ptr_validate); | |
1821 | ||
1822 | /* | |
1823 | * Determine the size of a slab object | |
1824 | */ | |
1825 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
1826 | { | |
1827 | return s->objsize; | |
1828 | } | |
1829 | EXPORT_SYMBOL(kmem_cache_size); | |
1830 | ||
1831 | const char *kmem_cache_name(struct kmem_cache *s) | |
1832 | { | |
1833 | return s->name; | |
1834 | } | |
1835 | EXPORT_SYMBOL(kmem_cache_name); | |
1836 | ||
1837 | /* | |
1838 | * Attempt to free all slabs on a node | |
1839 | */ | |
1840 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
1841 | struct list_head *list) | |
1842 | { | |
1843 | int slabs_inuse = 0; | |
1844 | unsigned long flags; | |
1845 | struct page *page, *h; | |
1846 | ||
1847 | spin_lock_irqsave(&n->list_lock, flags); | |
1848 | list_for_each_entry_safe(page, h, list, lru) | |
1849 | if (!page->inuse) { | |
1850 | list_del(&page->lru); | |
1851 | discard_slab(s, page); | |
1852 | } else | |
1853 | slabs_inuse++; | |
1854 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1855 | return slabs_inuse; | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * Release all resources used by slab cache | |
1860 | */ | |
1861 | static int kmem_cache_close(struct kmem_cache *s) | |
1862 | { | |
1863 | int node; | |
1864 | ||
1865 | flush_all(s); | |
1866 | ||
1867 | /* Attempt to free all objects */ | |
1868 | for_each_online_node(node) { | |
1869 | struct kmem_cache_node *n = get_node(s, node); | |
1870 | ||
2086d26a | 1871 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
1872 | if (atomic_long_read(&n->nr_slabs)) |
1873 | return 1; | |
1874 | } | |
1875 | free_kmem_cache_nodes(s); | |
1876 | return 0; | |
1877 | } | |
1878 | ||
1879 | /* | |
1880 | * Close a cache and release the kmem_cache structure | |
1881 | * (must be used for caches created using kmem_cache_create) | |
1882 | */ | |
1883 | void kmem_cache_destroy(struct kmem_cache *s) | |
1884 | { | |
1885 | down_write(&slub_lock); | |
1886 | s->refcount--; | |
1887 | if (!s->refcount) { | |
1888 | list_del(&s->list); | |
1889 | if (kmem_cache_close(s)) | |
1890 | WARN_ON(1); | |
1891 | sysfs_slab_remove(s); | |
1892 | kfree(s); | |
1893 | } | |
1894 | up_write(&slub_lock); | |
1895 | } | |
1896 | EXPORT_SYMBOL(kmem_cache_destroy); | |
1897 | ||
1898 | /******************************************************************** | |
1899 | * Kmalloc subsystem | |
1900 | *******************************************************************/ | |
1901 | ||
1902 | struct kmem_cache kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __cacheline_aligned; | |
1903 | EXPORT_SYMBOL(kmalloc_caches); | |
1904 | ||
1905 | #ifdef CONFIG_ZONE_DMA | |
1906 | static struct kmem_cache *kmalloc_caches_dma[KMALLOC_SHIFT_HIGH + 1]; | |
1907 | #endif | |
1908 | ||
1909 | static int __init setup_slub_min_order(char *str) | |
1910 | { | |
1911 | get_option (&str, &slub_min_order); | |
1912 | ||
1913 | return 1; | |
1914 | } | |
1915 | ||
1916 | __setup("slub_min_order=", setup_slub_min_order); | |
1917 | ||
1918 | static int __init setup_slub_max_order(char *str) | |
1919 | { | |
1920 | get_option (&str, &slub_max_order); | |
1921 | ||
1922 | return 1; | |
1923 | } | |
1924 | ||
1925 | __setup("slub_max_order=", setup_slub_max_order); | |
1926 | ||
1927 | static int __init setup_slub_min_objects(char *str) | |
1928 | { | |
1929 | get_option (&str, &slub_min_objects); | |
1930 | ||
1931 | return 1; | |
1932 | } | |
1933 | ||
1934 | __setup("slub_min_objects=", setup_slub_min_objects); | |
1935 | ||
1936 | static int __init setup_slub_nomerge(char *str) | |
1937 | { | |
1938 | slub_nomerge = 1; | |
1939 | return 1; | |
1940 | } | |
1941 | ||
1942 | __setup("slub_nomerge", setup_slub_nomerge); | |
1943 | ||
1944 | static int __init setup_slub_debug(char *str) | |
1945 | { | |
1946 | if (!str || *str != '=') | |
1947 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1948 | else { | |
1949 | str++; | |
1950 | if (*str == 0 || *str == ',') | |
1951 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1952 | else | |
1953 | for( ;*str && *str != ','; str++) | |
1954 | switch (*str) { | |
1955 | case 'f' : case 'F' : | |
1956 | slub_debug |= SLAB_DEBUG_FREE; | |
1957 | break; | |
1958 | case 'z' : case 'Z' : | |
1959 | slub_debug |= SLAB_RED_ZONE; | |
1960 | break; | |
1961 | case 'p' : case 'P' : | |
1962 | slub_debug |= SLAB_POISON; | |
1963 | break; | |
1964 | case 'u' : case 'U' : | |
1965 | slub_debug |= SLAB_STORE_USER; | |
1966 | break; | |
1967 | case 't' : case 'T' : | |
1968 | slub_debug |= SLAB_TRACE; | |
1969 | break; | |
1970 | default: | |
1971 | printk(KERN_ERR "slub_debug option '%c' " | |
1972 | "unknown. skipped\n",*str); | |
1973 | } | |
1974 | } | |
1975 | ||
1976 | if (*str == ',') | |
1977 | slub_debug_slabs = str + 1; | |
1978 | return 1; | |
1979 | } | |
1980 | ||
1981 | __setup("slub_debug", setup_slub_debug); | |
1982 | ||
1983 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, | |
1984 | const char *name, int size, gfp_t gfp_flags) | |
1985 | { | |
1986 | unsigned int flags = 0; | |
1987 | ||
1988 | if (gfp_flags & SLUB_DMA) | |
1989 | flags = SLAB_CACHE_DMA; | |
1990 | ||
1991 | down_write(&slub_lock); | |
1992 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
1993 | flags, NULL, NULL)) | |
1994 | goto panic; | |
1995 | ||
1996 | list_add(&s->list, &slab_caches); | |
1997 | up_write(&slub_lock); | |
1998 | if (sysfs_slab_add(s)) | |
1999 | goto panic; | |
2000 | return s; | |
2001 | ||
2002 | panic: | |
2003 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2004 | } | |
2005 | ||
2006 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) | |
2007 | { | |
2008 | int index = kmalloc_index(size); | |
2009 | ||
614410d5 | 2010 | if (!index) |
81819f0f CL |
2011 | return NULL; |
2012 | ||
2013 | /* Allocation too large? */ | |
2014 | BUG_ON(index < 0); | |
2015 | ||
2016 | #ifdef CONFIG_ZONE_DMA | |
2017 | if ((flags & SLUB_DMA)) { | |
2018 | struct kmem_cache *s; | |
2019 | struct kmem_cache *x; | |
2020 | char *text; | |
2021 | size_t realsize; | |
2022 | ||
2023 | s = kmalloc_caches_dma[index]; | |
2024 | if (s) | |
2025 | return s; | |
2026 | ||
2027 | /* Dynamically create dma cache */ | |
2028 | x = kmalloc(kmem_size, flags & ~SLUB_DMA); | |
2029 | if (!x) | |
2030 | panic("Unable to allocate memory for dma cache\n"); | |
2031 | ||
2032 | if (index <= KMALLOC_SHIFT_HIGH) | |
2033 | realsize = 1 << index; | |
2034 | else { | |
2035 | if (index == 1) | |
2036 | realsize = 96; | |
2037 | else | |
2038 | realsize = 192; | |
2039 | } | |
2040 | ||
2041 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", | |
2042 | (unsigned int)realsize); | |
2043 | s = create_kmalloc_cache(x, text, realsize, flags); | |
2044 | kmalloc_caches_dma[index] = s; | |
2045 | return s; | |
2046 | } | |
2047 | #endif | |
2048 | return &kmalloc_caches[index]; | |
2049 | } | |
2050 | ||
2051 | void *__kmalloc(size_t size, gfp_t flags) | |
2052 | { | |
2053 | struct kmem_cache *s = get_slab(size, flags); | |
2054 | ||
2055 | if (s) | |
77c5e2d0 | 2056 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2057 | return NULL; |
2058 | } | |
2059 | EXPORT_SYMBOL(__kmalloc); | |
2060 | ||
2061 | #ifdef CONFIG_NUMA | |
2062 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2063 | { | |
2064 | struct kmem_cache *s = get_slab(size, flags); | |
2065 | ||
2066 | if (s) | |
77c5e2d0 | 2067 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2068 | return NULL; |
2069 | } | |
2070 | EXPORT_SYMBOL(__kmalloc_node); | |
2071 | #endif | |
2072 | ||
2073 | size_t ksize(const void *object) | |
2074 | { | |
2075 | struct page *page = get_object_page(object); | |
2076 | struct kmem_cache *s; | |
2077 | ||
2078 | BUG_ON(!page); | |
2079 | s = page->slab; | |
2080 | BUG_ON(!s); | |
2081 | ||
2082 | /* | |
2083 | * Debugging requires use of the padding between object | |
2084 | * and whatever may come after it. | |
2085 | */ | |
2086 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2087 | return s->objsize; | |
2088 | ||
2089 | /* | |
2090 | * If we have the need to store the freelist pointer | |
2091 | * back there or track user information then we can | |
2092 | * only use the space before that information. | |
2093 | */ | |
2094 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2095 | return s->inuse; | |
2096 | ||
2097 | /* | |
2098 | * Else we can use all the padding etc for the allocation | |
2099 | */ | |
2100 | return s->size; | |
2101 | } | |
2102 | EXPORT_SYMBOL(ksize); | |
2103 | ||
2104 | void kfree(const void *x) | |
2105 | { | |
2106 | struct kmem_cache *s; | |
2107 | struct page *page; | |
2108 | ||
2109 | if (!x) | |
2110 | return; | |
2111 | ||
b49af68f | 2112 | page = virt_to_head_page(x); |
81819f0f CL |
2113 | s = page->slab; |
2114 | ||
77c5e2d0 | 2115 | slab_free(s, page, (void *)x, __builtin_return_address(0)); |
81819f0f CL |
2116 | } |
2117 | EXPORT_SYMBOL(kfree); | |
2118 | ||
2086d26a CL |
2119 | /* |
2120 | * kmem_cache_shrink removes empty slabs from the partial lists | |
2121 | * and then sorts the partially allocated slabs by the number | |
2122 | * of items in use. The slabs with the most items in use | |
2123 | * come first. New allocations will remove these from the | |
2124 | * partial list because they are full. The slabs with the | |
2125 | * least items are placed last. If it happens that the objects | |
2126 | * are freed then the page can be returned to the page allocator. | |
2127 | */ | |
2128 | int kmem_cache_shrink(struct kmem_cache *s) | |
2129 | { | |
2130 | int node; | |
2131 | int i; | |
2132 | struct kmem_cache_node *n; | |
2133 | struct page *page; | |
2134 | struct page *t; | |
2135 | struct list_head *slabs_by_inuse = | |
2136 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2137 | unsigned long flags; | |
2138 | ||
2139 | if (!slabs_by_inuse) | |
2140 | return -ENOMEM; | |
2141 | ||
2142 | flush_all(s); | |
2143 | for_each_online_node(node) { | |
2144 | n = get_node(s, node); | |
2145 | ||
2146 | if (!n->nr_partial) | |
2147 | continue; | |
2148 | ||
2149 | for (i = 0; i < s->objects; i++) | |
2150 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2151 | ||
2152 | spin_lock_irqsave(&n->list_lock, flags); | |
2153 | ||
2154 | /* | |
2155 | * Build lists indexed by the items in use in | |
2156 | * each slab or free slabs if empty. | |
2157 | * | |
2158 | * Note that concurrent frees may occur while | |
2159 | * we hold the list_lock. page->inuse here is | |
2160 | * the upper limit. | |
2161 | */ | |
2162 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2163 | if (!page->inuse && slab_trylock(page)) { | |
2164 | /* | |
2165 | * Must hold slab lock here because slab_free | |
2166 | * may have freed the last object and be | |
2167 | * waiting to release the slab. | |
2168 | */ | |
2169 | list_del(&page->lru); | |
2170 | n->nr_partial--; | |
2171 | slab_unlock(page); | |
2172 | discard_slab(s, page); | |
2173 | } else { | |
2174 | if (n->nr_partial > MAX_PARTIAL) | |
2175 | list_move(&page->lru, | |
2176 | slabs_by_inuse + page->inuse); | |
2177 | } | |
2178 | } | |
2179 | ||
2180 | if (n->nr_partial <= MAX_PARTIAL) | |
2181 | goto out; | |
2182 | ||
2183 | /* | |
2184 | * Rebuild the partial list with the slabs filled up | |
2185 | * most first and the least used slabs at the end. | |
2186 | */ | |
2187 | for (i = s->objects - 1; i >= 0; i--) | |
2188 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2189 | ||
2190 | out: | |
2191 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2192 | } | |
2193 | ||
2194 | kfree(slabs_by_inuse); | |
2195 | return 0; | |
2196 | } | |
2197 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2198 | ||
81819f0f CL |
2199 | /** |
2200 | * krealloc - reallocate memory. The contents will remain unchanged. | |
2201 | * | |
2202 | * @p: object to reallocate memory for. | |
2203 | * @new_size: how many bytes of memory are required. | |
2204 | * @flags: the type of memory to allocate. | |
2205 | * | |
2206 | * The contents of the object pointed to are preserved up to the | |
2207 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
2208 | * behaves exactly like kmalloc(). If @size is 0 and @p is not a | |
2209 | * %NULL pointer, the object pointed to is freed. | |
2210 | */ | |
2211 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
2212 | { | |
2213 | struct kmem_cache *new_cache; | |
2214 | void *ret; | |
2215 | struct page *page; | |
2216 | ||
2217 | if (unlikely(!p)) | |
2218 | return kmalloc(new_size, flags); | |
2219 | ||
2220 | if (unlikely(!new_size)) { | |
2221 | kfree(p); | |
2222 | return NULL; | |
2223 | } | |
2224 | ||
b49af68f | 2225 | page = virt_to_head_page(p); |
81819f0f CL |
2226 | |
2227 | new_cache = get_slab(new_size, flags); | |
2228 | ||
2229 | /* | |
2230 | * If new size fits in the current cache, bail out. | |
2231 | */ | |
2232 | if (likely(page->slab == new_cache)) | |
2233 | return (void *)p; | |
2234 | ||
2235 | ret = kmalloc(new_size, flags); | |
2236 | if (ret) { | |
2237 | memcpy(ret, p, min(new_size, ksize(p))); | |
2238 | kfree(p); | |
2239 | } | |
2240 | return ret; | |
2241 | } | |
2242 | EXPORT_SYMBOL(krealloc); | |
2243 | ||
2244 | /******************************************************************** | |
2245 | * Basic setup of slabs | |
2246 | *******************************************************************/ | |
2247 | ||
2248 | void __init kmem_cache_init(void) | |
2249 | { | |
2250 | int i; | |
2251 | ||
2252 | #ifdef CONFIG_NUMA | |
2253 | /* | |
2254 | * Must first have the slab cache available for the allocations of the | |
2255 | * struct kmalloc_cache_node's. There is special bootstrap code in | |
2256 | * kmem_cache_open for slab_state == DOWN. | |
2257 | */ | |
2258 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2259 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
2260 | #endif | |
2261 | ||
2262 | /* Able to allocate the per node structures */ | |
2263 | slab_state = PARTIAL; | |
2264 | ||
2265 | /* Caches that are not of the two-to-the-power-of size */ | |
2266 | create_kmalloc_cache(&kmalloc_caches[1], | |
2267 | "kmalloc-96", 96, GFP_KERNEL); | |
2268 | create_kmalloc_cache(&kmalloc_caches[2], | |
2269 | "kmalloc-192", 192, GFP_KERNEL); | |
2270 | ||
2271 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2272 | create_kmalloc_cache(&kmalloc_caches[i], | |
2273 | "kmalloc", 1 << i, GFP_KERNEL); | |
2274 | ||
2275 | slab_state = UP; | |
2276 | ||
2277 | /* Provide the correct kmalloc names now that the caches are up */ | |
2278 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2279 | kmalloc_caches[i]. name = | |
2280 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
2281 | ||
2282 | #ifdef CONFIG_SMP | |
2283 | register_cpu_notifier(&slab_notifier); | |
2284 | #endif | |
2285 | ||
2286 | if (nr_cpu_ids) /* Remove when nr_cpu_ids is fixed upstream ! */ | |
2287 | kmem_size = offsetof(struct kmem_cache, cpu_slab) | |
2288 | + nr_cpu_ids * sizeof(struct page *); | |
2289 | ||
2290 | printk(KERN_INFO "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
2291 | " Processors=%d, Nodes=%d\n", | |
2292 | KMALLOC_SHIFT_HIGH, L1_CACHE_BYTES, | |
2293 | slub_min_order, slub_max_order, slub_min_objects, | |
2294 | nr_cpu_ids, nr_node_ids); | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * Find a mergeable slab cache | |
2299 | */ | |
2300 | static int slab_unmergeable(struct kmem_cache *s) | |
2301 | { | |
2302 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
2303 | return 1; | |
2304 | ||
2305 | if (s->ctor || s->dtor) | |
2306 | return 1; | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | ||
2311 | static struct kmem_cache *find_mergeable(size_t size, | |
2312 | size_t align, unsigned long flags, | |
2313 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
2314 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
2315 | { | |
2316 | struct list_head *h; | |
2317 | ||
2318 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
2319 | return NULL; | |
2320 | ||
2321 | if (ctor || dtor) | |
2322 | return NULL; | |
2323 | ||
2324 | size = ALIGN(size, sizeof(void *)); | |
2325 | align = calculate_alignment(flags, align, size); | |
2326 | size = ALIGN(size, align); | |
2327 | ||
2328 | list_for_each(h, &slab_caches) { | |
2329 | struct kmem_cache *s = | |
2330 | container_of(h, struct kmem_cache, list); | |
2331 | ||
2332 | if (slab_unmergeable(s)) | |
2333 | continue; | |
2334 | ||
2335 | if (size > s->size) | |
2336 | continue; | |
2337 | ||
2338 | if (((flags | slub_debug) & SLUB_MERGE_SAME) != | |
2339 | (s->flags & SLUB_MERGE_SAME)) | |
2340 | continue; | |
2341 | /* | |
2342 | * Check if alignment is compatible. | |
2343 | * Courtesy of Adrian Drzewiecki | |
2344 | */ | |
2345 | if ((s->size & ~(align -1)) != s->size) | |
2346 | continue; | |
2347 | ||
2348 | if (s->size - size >= sizeof(void *)) | |
2349 | continue; | |
2350 | ||
2351 | return s; | |
2352 | } | |
2353 | return NULL; | |
2354 | } | |
2355 | ||
2356 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
2357 | size_t align, unsigned long flags, | |
2358 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
2359 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
2360 | { | |
2361 | struct kmem_cache *s; | |
2362 | ||
2363 | down_write(&slub_lock); | |
2364 | s = find_mergeable(size, align, flags, dtor, ctor); | |
2365 | if (s) { | |
2366 | s->refcount++; | |
2367 | /* | |
2368 | * Adjust the object sizes so that we clear | |
2369 | * the complete object on kzalloc. | |
2370 | */ | |
2371 | s->objsize = max(s->objsize, (int)size); | |
2372 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
2373 | if (sysfs_slab_alias(s, name)) | |
2374 | goto err; | |
2375 | } else { | |
2376 | s = kmalloc(kmem_size, GFP_KERNEL); | |
2377 | if (s && kmem_cache_open(s, GFP_KERNEL, name, | |
2378 | size, align, flags, ctor, dtor)) { | |
2379 | if (sysfs_slab_add(s)) { | |
2380 | kfree(s); | |
2381 | goto err; | |
2382 | } | |
2383 | list_add(&s->list, &slab_caches); | |
2384 | } else | |
2385 | kfree(s); | |
2386 | } | |
2387 | up_write(&slub_lock); | |
2388 | return s; | |
2389 | ||
2390 | err: | |
2391 | up_write(&slub_lock); | |
2392 | if (flags & SLAB_PANIC) | |
2393 | panic("Cannot create slabcache %s\n", name); | |
2394 | else | |
2395 | s = NULL; | |
2396 | return s; | |
2397 | } | |
2398 | EXPORT_SYMBOL(kmem_cache_create); | |
2399 | ||
2400 | void *kmem_cache_zalloc(struct kmem_cache *s, gfp_t flags) | |
2401 | { | |
2402 | void *x; | |
2403 | ||
77c5e2d0 | 2404 | x = slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2405 | if (x) |
2406 | memset(x, 0, s->objsize); | |
2407 | return x; | |
2408 | } | |
2409 | EXPORT_SYMBOL(kmem_cache_zalloc); | |
2410 | ||
2411 | #ifdef CONFIG_SMP | |
2412 | static void for_all_slabs(void (*func)(struct kmem_cache *, int), int cpu) | |
2413 | { | |
2414 | struct list_head *h; | |
2415 | ||
2416 | down_read(&slub_lock); | |
2417 | list_for_each(h, &slab_caches) { | |
2418 | struct kmem_cache *s = | |
2419 | container_of(h, struct kmem_cache, list); | |
2420 | ||
2421 | func(s, cpu); | |
2422 | } | |
2423 | up_read(&slub_lock); | |
2424 | } | |
2425 | ||
2426 | /* | |
2427 | * Use the cpu notifier to insure that the slab are flushed | |
2428 | * when necessary. | |
2429 | */ | |
2430 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
2431 | unsigned long action, void *hcpu) | |
2432 | { | |
2433 | long cpu = (long)hcpu; | |
2434 | ||
2435 | switch (action) { | |
2436 | case CPU_UP_CANCELED: | |
2437 | case CPU_DEAD: | |
2438 | for_all_slabs(__flush_cpu_slab, cpu); | |
2439 | break; | |
2440 | default: | |
2441 | break; | |
2442 | } | |
2443 | return NOTIFY_OK; | |
2444 | } | |
2445 | ||
2446 | static struct notifier_block __cpuinitdata slab_notifier = | |
2447 | { &slab_cpuup_callback, NULL, 0 }; | |
2448 | ||
2449 | #endif | |
2450 | ||
81819f0f CL |
2451 | #ifdef CONFIG_NUMA |
2452 | ||
2453 | /***************************************************************** | |
2454 | * Generic reaper used to support the page allocator | |
2455 | * (the cpu slabs are reaped by a per slab workqueue). | |
2456 | * | |
2457 | * Maybe move this to the page allocator? | |
2458 | ****************************************************************/ | |
2459 | ||
2460 | static DEFINE_PER_CPU(unsigned long, reap_node); | |
2461 | ||
2462 | static void init_reap_node(int cpu) | |
2463 | { | |
2464 | int node; | |
2465 | ||
2466 | node = next_node(cpu_to_node(cpu), node_online_map); | |
2467 | if (node == MAX_NUMNODES) | |
2468 | node = first_node(node_online_map); | |
2469 | ||
2470 | __get_cpu_var(reap_node) = node; | |
2471 | } | |
2472 | ||
2473 | static void next_reap_node(void) | |
2474 | { | |
2475 | int node = __get_cpu_var(reap_node); | |
2476 | ||
2477 | /* | |
2478 | * Also drain per cpu pages on remote zones | |
2479 | */ | |
2480 | if (node != numa_node_id()) | |
2481 | drain_node_pages(node); | |
2482 | ||
2483 | node = next_node(node, node_online_map); | |
2484 | if (unlikely(node >= MAX_NUMNODES)) | |
2485 | node = first_node(node_online_map); | |
2486 | __get_cpu_var(reap_node) = node; | |
2487 | } | |
2488 | #else | |
2489 | #define init_reap_node(cpu) do { } while (0) | |
2490 | #define next_reap_node(void) do { } while (0) | |
2491 | #endif | |
2492 | ||
2493 | #define REAPTIMEOUT_CPUC (2*HZ) | |
2494 | ||
2495 | #ifdef CONFIG_SMP | |
2496 | static DEFINE_PER_CPU(struct delayed_work, reap_work); | |
2497 | ||
2498 | static void cache_reap(struct work_struct *unused) | |
2499 | { | |
2500 | next_reap_node(); | |
2501 | refresh_cpu_vm_stats(smp_processor_id()); | |
2502 | schedule_delayed_work(&__get_cpu_var(reap_work), | |
2503 | REAPTIMEOUT_CPUC); | |
2504 | } | |
2505 | ||
2506 | static void __devinit start_cpu_timer(int cpu) | |
2507 | { | |
2508 | struct delayed_work *reap_work = &per_cpu(reap_work, cpu); | |
2509 | ||
2510 | /* | |
2511 | * When this gets called from do_initcalls via cpucache_init(), | |
2512 | * init_workqueues() has already run, so keventd will be setup | |
2513 | * at that time. | |
2514 | */ | |
2515 | if (keventd_up() && reap_work->work.func == NULL) { | |
2516 | init_reap_node(cpu); | |
2517 | INIT_DELAYED_WORK(reap_work, cache_reap); | |
2518 | schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu); | |
2519 | } | |
2520 | } | |
2521 | ||
2522 | static int __init cpucache_init(void) | |
2523 | { | |
2524 | int cpu; | |
2525 | ||
2526 | /* | |
2527 | * Register the timers that drain pcp pages and update vm statistics | |
2528 | */ | |
2529 | for_each_online_cpu(cpu) | |
2530 | start_cpu_timer(cpu); | |
2531 | return 0; | |
2532 | } | |
2533 | __initcall(cpucache_init); | |
2534 | #endif | |
2535 | ||
2536 | #ifdef SLUB_RESILIENCY_TEST | |
2537 | static unsigned long validate_slab_cache(struct kmem_cache *s); | |
2538 | ||
2539 | static void resiliency_test(void) | |
2540 | { | |
2541 | u8 *p; | |
2542 | ||
2543 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
2544 | printk(KERN_ERR "-----------------------\n"); | |
2545 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
2546 | ||
2547 | p = kzalloc(16, GFP_KERNEL); | |
2548 | p[16] = 0x12; | |
2549 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
2550 | " 0x12->0x%p\n\n", p + 16); | |
2551 | ||
2552 | validate_slab_cache(kmalloc_caches + 4); | |
2553 | ||
2554 | /* Hmmm... The next two are dangerous */ | |
2555 | p = kzalloc(32, GFP_KERNEL); | |
2556 | p[32 + sizeof(void *)] = 0x34; | |
2557 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
2558 | " 0x34 -> -0x%p\n", p); | |
2559 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2560 | ||
2561 | validate_slab_cache(kmalloc_caches + 5); | |
2562 | p = kzalloc(64, GFP_KERNEL); | |
2563 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
2564 | *p = 0x56; | |
2565 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
2566 | p); | |
2567 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2568 | validate_slab_cache(kmalloc_caches + 6); | |
2569 | ||
2570 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
2571 | p = kzalloc(128, GFP_KERNEL); | |
2572 | kfree(p); | |
2573 | *p = 0x78; | |
2574 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
2575 | validate_slab_cache(kmalloc_caches + 7); | |
2576 | ||
2577 | p = kzalloc(256, GFP_KERNEL); | |
2578 | kfree(p); | |
2579 | p[50] = 0x9a; | |
2580 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); | |
2581 | validate_slab_cache(kmalloc_caches + 8); | |
2582 | ||
2583 | p = kzalloc(512, GFP_KERNEL); | |
2584 | kfree(p); | |
2585 | p[512] = 0xab; | |
2586 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
2587 | validate_slab_cache(kmalloc_caches + 9); | |
2588 | } | |
2589 | #else | |
2590 | static void resiliency_test(void) {}; | |
2591 | #endif | |
2592 | ||
2593 | /* | |
2594 | * These are not as efficient as kmalloc for the non debug case. | |
2595 | * We do not have the page struct available so we have to touch one | |
2596 | * cacheline in struct kmem_cache to check slab flags. | |
2597 | */ | |
2598 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) | |
2599 | { | |
2600 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2601 | |
2602 | if (!s) | |
2603 | return NULL; | |
2604 | ||
77c5e2d0 | 2605 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
2606 | } |
2607 | ||
2608 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
2609 | int node, void *caller) | |
2610 | { | |
2611 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2612 | |
2613 | if (!s) | |
2614 | return NULL; | |
2615 | ||
77c5e2d0 | 2616 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
2617 | } |
2618 | ||
2619 | #ifdef CONFIG_SYSFS | |
2620 | ||
53e15af0 CL |
2621 | static int validate_slab(struct kmem_cache *s, struct page *page) |
2622 | { | |
2623 | void *p; | |
2624 | void *addr = page_address(page); | |
2625 | unsigned long map[BITS_TO_LONGS(s->objects)]; | |
2626 | ||
2627 | if (!check_slab(s, page) || | |
2628 | !on_freelist(s, page, NULL)) | |
2629 | return 0; | |
2630 | ||
2631 | /* Now we know that a valid freelist exists */ | |
2632 | bitmap_zero(map, s->objects); | |
2633 | ||
2634 | for(p = page->freelist; p; p = get_freepointer(s, p)) { | |
2635 | set_bit((p - addr) / s->size, map); | |
2636 | if (!check_object(s, page, p, 0)) | |
2637 | return 0; | |
2638 | } | |
2639 | ||
2640 | for(p = addr; p < addr + s->objects * s->size; p += s->size) | |
2641 | if (!test_bit((p - addr) / s->size, map)) | |
2642 | if (!check_object(s, page, p, 1)) | |
2643 | return 0; | |
2644 | return 1; | |
2645 | } | |
2646 | ||
2647 | static void validate_slab_slab(struct kmem_cache *s, struct page *page) | |
2648 | { | |
2649 | if (slab_trylock(page)) { | |
2650 | validate_slab(s, page); | |
2651 | slab_unlock(page); | |
2652 | } else | |
2653 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
2654 | s->name, page); | |
2655 | ||
2656 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
2657 | if (!PageError(page)) | |
2658 | printk(KERN_ERR "SLUB %s: PageError not set " | |
2659 | "on slab 0x%p\n", s->name, page); | |
2660 | } else { | |
2661 | if (PageError(page)) | |
2662 | printk(KERN_ERR "SLUB %s: PageError set on " | |
2663 | "slab 0x%p\n", s->name, page); | |
2664 | } | |
2665 | } | |
2666 | ||
2667 | static int validate_slab_node(struct kmem_cache *s, struct kmem_cache_node *n) | |
2668 | { | |
2669 | unsigned long count = 0; | |
2670 | struct page *page; | |
2671 | unsigned long flags; | |
2672 | ||
2673 | spin_lock_irqsave(&n->list_lock, flags); | |
2674 | ||
2675 | list_for_each_entry(page, &n->partial, lru) { | |
2676 | validate_slab_slab(s, page); | |
2677 | count++; | |
2678 | } | |
2679 | if (count != n->nr_partial) | |
2680 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
2681 | "counter=%ld\n", s->name, count, n->nr_partial); | |
2682 | ||
2683 | if (!(s->flags & SLAB_STORE_USER)) | |
2684 | goto out; | |
2685 | ||
2686 | list_for_each_entry(page, &n->full, lru) { | |
2687 | validate_slab_slab(s, page); | |
2688 | count++; | |
2689 | } | |
2690 | if (count != atomic_long_read(&n->nr_slabs)) | |
2691 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
2692 | "counter=%ld\n", s->name, count, | |
2693 | atomic_long_read(&n->nr_slabs)); | |
2694 | ||
2695 | out: | |
2696 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2697 | return count; | |
2698 | } | |
2699 | ||
2700 | static unsigned long validate_slab_cache(struct kmem_cache *s) | |
2701 | { | |
2702 | int node; | |
2703 | unsigned long count = 0; | |
2704 | ||
2705 | flush_all(s); | |
2706 | for_each_online_node(node) { | |
2707 | struct kmem_cache_node *n = get_node(s, node); | |
2708 | ||
2709 | count += validate_slab_node(s, n); | |
2710 | } | |
2711 | return count; | |
2712 | } | |
2713 | ||
88a420e4 CL |
2714 | /* |
2715 | * Generate lists of locations where slabcache objects are allocated | |
2716 | * and freed. | |
2717 | */ | |
2718 | ||
2719 | struct location { | |
2720 | unsigned long count; | |
2721 | void *addr; | |
2722 | }; | |
2723 | ||
2724 | struct loc_track { | |
2725 | unsigned long max; | |
2726 | unsigned long count; | |
2727 | struct location *loc; | |
2728 | }; | |
2729 | ||
2730 | static void free_loc_track(struct loc_track *t) | |
2731 | { | |
2732 | if (t->max) | |
2733 | free_pages((unsigned long)t->loc, | |
2734 | get_order(sizeof(struct location) * t->max)); | |
2735 | } | |
2736 | ||
2737 | static int alloc_loc_track(struct loc_track *t, unsigned long max) | |
2738 | { | |
2739 | struct location *l; | |
2740 | int order; | |
2741 | ||
2742 | if (!max) | |
2743 | max = PAGE_SIZE / sizeof(struct location); | |
2744 | ||
2745 | order = get_order(sizeof(struct location) * max); | |
2746 | ||
2747 | l = (void *)__get_free_pages(GFP_KERNEL, order); | |
2748 | ||
2749 | if (!l) | |
2750 | return 0; | |
2751 | ||
2752 | if (t->count) { | |
2753 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
2754 | free_loc_track(t); | |
2755 | } | |
2756 | t->max = max; | |
2757 | t->loc = l; | |
2758 | return 1; | |
2759 | } | |
2760 | ||
2761 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
2762 | void *addr) | |
2763 | { | |
2764 | long start, end, pos; | |
2765 | struct location *l; | |
2766 | void *caddr; | |
2767 | ||
2768 | start = -1; | |
2769 | end = t->count; | |
2770 | ||
2771 | for ( ; ; ) { | |
2772 | pos = start + (end - start + 1) / 2; | |
2773 | ||
2774 | /* | |
2775 | * There is nothing at "end". If we end up there | |
2776 | * we need to add something to before end. | |
2777 | */ | |
2778 | if (pos == end) | |
2779 | break; | |
2780 | ||
2781 | caddr = t->loc[pos].addr; | |
2782 | if (addr == caddr) { | |
2783 | t->loc[pos].count++; | |
2784 | return 1; | |
2785 | } | |
2786 | ||
2787 | if (addr < caddr) | |
2788 | end = pos; | |
2789 | else | |
2790 | start = pos; | |
2791 | } | |
2792 | ||
2793 | /* | |
2794 | * Not found. Insert new tracking element | |
2795 | */ | |
2796 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max)) | |
2797 | return 0; | |
2798 | ||
2799 | l = t->loc + pos; | |
2800 | if (pos < t->count) | |
2801 | memmove(l + 1, l, | |
2802 | (t->count - pos) * sizeof(struct location)); | |
2803 | t->count++; | |
2804 | l->count = 1; | |
2805 | l->addr = addr; | |
2806 | return 1; | |
2807 | } | |
2808 | ||
2809 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
2810 | struct page *page, enum track_item alloc) | |
2811 | { | |
2812 | void *addr = page_address(page); | |
2813 | unsigned long map[BITS_TO_LONGS(s->objects)]; | |
2814 | void *p; | |
2815 | ||
2816 | bitmap_zero(map, s->objects); | |
2817 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
2818 | set_bit((p - addr) / s->size, map); | |
2819 | ||
2820 | for (p = addr; p < addr + s->objects * s->size; p += s->size) | |
2821 | if (!test_bit((p - addr) / s->size, map)) { | |
2822 | void *addr = get_track(s, p, alloc)->addr; | |
2823 | ||
2824 | add_location(t, s, addr); | |
2825 | } | |
2826 | } | |
2827 | ||
2828 | static int list_locations(struct kmem_cache *s, char *buf, | |
2829 | enum track_item alloc) | |
2830 | { | |
2831 | int n = 0; | |
2832 | unsigned long i; | |
2833 | struct loc_track t; | |
2834 | int node; | |
2835 | ||
2836 | t.count = 0; | |
2837 | t.max = 0; | |
2838 | ||
2839 | /* Push back cpu slabs */ | |
2840 | flush_all(s); | |
2841 | ||
2842 | for_each_online_node(node) { | |
2843 | struct kmem_cache_node *n = get_node(s, node); | |
2844 | unsigned long flags; | |
2845 | struct page *page; | |
2846 | ||
2847 | if (!atomic_read(&n->nr_slabs)) | |
2848 | continue; | |
2849 | ||
2850 | spin_lock_irqsave(&n->list_lock, flags); | |
2851 | list_for_each_entry(page, &n->partial, lru) | |
2852 | process_slab(&t, s, page, alloc); | |
2853 | list_for_each_entry(page, &n->full, lru) | |
2854 | process_slab(&t, s, page, alloc); | |
2855 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2856 | } | |
2857 | ||
2858 | for (i = 0; i < t.count; i++) { | |
2859 | void *addr = t.loc[i].addr; | |
2860 | ||
2861 | if (n > PAGE_SIZE - 100) | |
2862 | break; | |
2863 | n += sprintf(buf + n, "%7ld ", t.loc[i].count); | |
2864 | if (addr) | |
2865 | n += sprint_symbol(buf + n, (unsigned long)t.loc[i].addr); | |
2866 | else | |
2867 | n += sprintf(buf + n, "<not-available>"); | |
2868 | n += sprintf(buf + n, "\n"); | |
2869 | } | |
2870 | ||
2871 | free_loc_track(&t); | |
2872 | if (!t.count) | |
2873 | n += sprintf(buf, "No data\n"); | |
2874 | return n; | |
2875 | } | |
2876 | ||
81819f0f CL |
2877 | static unsigned long count_partial(struct kmem_cache_node *n) |
2878 | { | |
2879 | unsigned long flags; | |
2880 | unsigned long x = 0; | |
2881 | struct page *page; | |
2882 | ||
2883 | spin_lock_irqsave(&n->list_lock, flags); | |
2884 | list_for_each_entry(page, &n->partial, lru) | |
2885 | x += page->inuse; | |
2886 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2887 | return x; | |
2888 | } | |
2889 | ||
2890 | enum slab_stat_type { | |
2891 | SL_FULL, | |
2892 | SL_PARTIAL, | |
2893 | SL_CPU, | |
2894 | SL_OBJECTS | |
2895 | }; | |
2896 | ||
2897 | #define SO_FULL (1 << SL_FULL) | |
2898 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
2899 | #define SO_CPU (1 << SL_CPU) | |
2900 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
2901 | ||
2902 | static unsigned long slab_objects(struct kmem_cache *s, | |
2903 | char *buf, unsigned long flags) | |
2904 | { | |
2905 | unsigned long total = 0; | |
2906 | int cpu; | |
2907 | int node; | |
2908 | int x; | |
2909 | unsigned long *nodes; | |
2910 | unsigned long *per_cpu; | |
2911 | ||
2912 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
2913 | per_cpu = nodes + nr_node_ids; | |
2914 | ||
2915 | for_each_possible_cpu(cpu) { | |
2916 | struct page *page = s->cpu_slab[cpu]; | |
2917 | int node; | |
2918 | ||
2919 | if (page) { | |
2920 | node = page_to_nid(page); | |
2921 | if (flags & SO_CPU) { | |
2922 | int x = 0; | |
2923 | ||
2924 | if (flags & SO_OBJECTS) | |
2925 | x = page->inuse; | |
2926 | else | |
2927 | x = 1; | |
2928 | total += x; | |
2929 | nodes[node] += x; | |
2930 | } | |
2931 | per_cpu[node]++; | |
2932 | } | |
2933 | } | |
2934 | ||
2935 | for_each_online_node(node) { | |
2936 | struct kmem_cache_node *n = get_node(s, node); | |
2937 | ||
2938 | if (flags & SO_PARTIAL) { | |
2939 | if (flags & SO_OBJECTS) | |
2940 | x = count_partial(n); | |
2941 | else | |
2942 | x = n->nr_partial; | |
2943 | total += x; | |
2944 | nodes[node] += x; | |
2945 | } | |
2946 | ||
2947 | if (flags & SO_FULL) { | |
2948 | int full_slabs = atomic_read(&n->nr_slabs) | |
2949 | - per_cpu[node] | |
2950 | - n->nr_partial; | |
2951 | ||
2952 | if (flags & SO_OBJECTS) | |
2953 | x = full_slabs * s->objects; | |
2954 | else | |
2955 | x = full_slabs; | |
2956 | total += x; | |
2957 | nodes[node] += x; | |
2958 | } | |
2959 | } | |
2960 | ||
2961 | x = sprintf(buf, "%lu", total); | |
2962 | #ifdef CONFIG_NUMA | |
2963 | for_each_online_node(node) | |
2964 | if (nodes[node]) | |
2965 | x += sprintf(buf + x, " N%d=%lu", | |
2966 | node, nodes[node]); | |
2967 | #endif | |
2968 | kfree(nodes); | |
2969 | return x + sprintf(buf + x, "\n"); | |
2970 | } | |
2971 | ||
2972 | static int any_slab_objects(struct kmem_cache *s) | |
2973 | { | |
2974 | int node; | |
2975 | int cpu; | |
2976 | ||
2977 | for_each_possible_cpu(cpu) | |
2978 | if (s->cpu_slab[cpu]) | |
2979 | return 1; | |
2980 | ||
2981 | for_each_node(node) { | |
2982 | struct kmem_cache_node *n = get_node(s, node); | |
2983 | ||
2984 | if (n->nr_partial || atomic_read(&n->nr_slabs)) | |
2985 | return 1; | |
2986 | } | |
2987 | return 0; | |
2988 | } | |
2989 | ||
2990 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
2991 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
2992 | ||
2993 | struct slab_attribute { | |
2994 | struct attribute attr; | |
2995 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
2996 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
2997 | }; | |
2998 | ||
2999 | #define SLAB_ATTR_RO(_name) \ | |
3000 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3001 | ||
3002 | #define SLAB_ATTR(_name) \ | |
3003 | static struct slab_attribute _name##_attr = \ | |
3004 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3005 | ||
81819f0f CL |
3006 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3007 | { | |
3008 | return sprintf(buf, "%d\n", s->size); | |
3009 | } | |
3010 | SLAB_ATTR_RO(slab_size); | |
3011 | ||
3012 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3013 | { | |
3014 | return sprintf(buf, "%d\n", s->align); | |
3015 | } | |
3016 | SLAB_ATTR_RO(align); | |
3017 | ||
3018 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3019 | { | |
3020 | return sprintf(buf, "%d\n", s->objsize); | |
3021 | } | |
3022 | SLAB_ATTR_RO(object_size); | |
3023 | ||
3024 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3025 | { | |
3026 | return sprintf(buf, "%d\n", s->objects); | |
3027 | } | |
3028 | SLAB_ATTR_RO(objs_per_slab); | |
3029 | ||
3030 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3031 | { | |
3032 | return sprintf(buf, "%d\n", s->order); | |
3033 | } | |
3034 | SLAB_ATTR_RO(order); | |
3035 | ||
3036 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3037 | { | |
3038 | if (s->ctor) { | |
3039 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3040 | ||
3041 | return n + sprintf(buf + n, "\n"); | |
3042 | } | |
3043 | return 0; | |
3044 | } | |
3045 | SLAB_ATTR_RO(ctor); | |
3046 | ||
3047 | static ssize_t dtor_show(struct kmem_cache *s, char *buf) | |
3048 | { | |
3049 | if (s->dtor) { | |
3050 | int n = sprint_symbol(buf, (unsigned long)s->dtor); | |
3051 | ||
3052 | return n + sprintf(buf + n, "\n"); | |
3053 | } | |
3054 | return 0; | |
3055 | } | |
3056 | SLAB_ATTR_RO(dtor); | |
3057 | ||
3058 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) | |
3059 | { | |
3060 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3061 | } | |
3062 | SLAB_ATTR_RO(aliases); | |
3063 | ||
3064 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3065 | { | |
3066 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3067 | } | |
3068 | SLAB_ATTR_RO(slabs); | |
3069 | ||
3070 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3071 | { | |
3072 | return slab_objects(s, buf, SO_PARTIAL); | |
3073 | } | |
3074 | SLAB_ATTR_RO(partial); | |
3075 | ||
3076 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3077 | { | |
3078 | return slab_objects(s, buf, SO_CPU); | |
3079 | } | |
3080 | SLAB_ATTR_RO(cpu_slabs); | |
3081 | ||
3082 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3083 | { | |
3084 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3085 | } | |
3086 | SLAB_ATTR_RO(objects); | |
3087 | ||
3088 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3089 | { | |
3090 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3091 | } | |
3092 | ||
3093 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3094 | const char *buf, size_t length) | |
3095 | { | |
3096 | s->flags &= ~SLAB_DEBUG_FREE; | |
3097 | if (buf[0] == '1') | |
3098 | s->flags |= SLAB_DEBUG_FREE; | |
3099 | return length; | |
3100 | } | |
3101 | SLAB_ATTR(sanity_checks); | |
3102 | ||
3103 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3104 | { | |
3105 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3106 | } | |
3107 | ||
3108 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3109 | size_t length) | |
3110 | { | |
3111 | s->flags &= ~SLAB_TRACE; | |
3112 | if (buf[0] == '1') | |
3113 | s->flags |= SLAB_TRACE; | |
3114 | return length; | |
3115 | } | |
3116 | SLAB_ATTR(trace); | |
3117 | ||
3118 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3119 | { | |
3120 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3121 | } | |
3122 | ||
3123 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3124 | const char *buf, size_t length) | |
3125 | { | |
3126 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3127 | if (buf[0] == '1') | |
3128 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3129 | return length; | |
3130 | } | |
3131 | SLAB_ATTR(reclaim_account); | |
3132 | ||
3133 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3134 | { | |
5af60839 | 3135 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3136 | } |
3137 | SLAB_ATTR_RO(hwcache_align); | |
3138 | ||
3139 | #ifdef CONFIG_ZONE_DMA | |
3140 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3141 | { | |
3142 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3143 | } | |
3144 | SLAB_ATTR_RO(cache_dma); | |
3145 | #endif | |
3146 | ||
3147 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3148 | { | |
3149 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3150 | } | |
3151 | SLAB_ATTR_RO(destroy_by_rcu); | |
3152 | ||
3153 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3154 | { | |
3155 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3156 | } | |
3157 | ||
3158 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3159 | const char *buf, size_t length) | |
3160 | { | |
3161 | if (any_slab_objects(s)) | |
3162 | return -EBUSY; | |
3163 | ||
3164 | s->flags &= ~SLAB_RED_ZONE; | |
3165 | if (buf[0] == '1') | |
3166 | s->flags |= SLAB_RED_ZONE; | |
3167 | calculate_sizes(s); | |
3168 | return length; | |
3169 | } | |
3170 | SLAB_ATTR(red_zone); | |
3171 | ||
3172 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3173 | { | |
3174 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3175 | } | |
3176 | ||
3177 | static ssize_t poison_store(struct kmem_cache *s, | |
3178 | const char *buf, size_t length) | |
3179 | { | |
3180 | if (any_slab_objects(s)) | |
3181 | return -EBUSY; | |
3182 | ||
3183 | s->flags &= ~SLAB_POISON; | |
3184 | if (buf[0] == '1') | |
3185 | s->flags |= SLAB_POISON; | |
3186 | calculate_sizes(s); | |
3187 | return length; | |
3188 | } | |
3189 | SLAB_ATTR(poison); | |
3190 | ||
3191 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3192 | { | |
3193 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3194 | } | |
3195 | ||
3196 | static ssize_t store_user_store(struct kmem_cache *s, | |
3197 | const char *buf, size_t length) | |
3198 | { | |
3199 | if (any_slab_objects(s)) | |
3200 | return -EBUSY; | |
3201 | ||
3202 | s->flags &= ~SLAB_STORE_USER; | |
3203 | if (buf[0] == '1') | |
3204 | s->flags |= SLAB_STORE_USER; | |
3205 | calculate_sizes(s); | |
3206 | return length; | |
3207 | } | |
3208 | SLAB_ATTR(store_user); | |
3209 | ||
53e15af0 CL |
3210 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3211 | { | |
3212 | return 0; | |
3213 | } | |
3214 | ||
3215 | static ssize_t validate_store(struct kmem_cache *s, | |
3216 | const char *buf, size_t length) | |
3217 | { | |
3218 | if (buf[0] == '1') | |
3219 | validate_slab_cache(s); | |
3220 | else | |
3221 | return -EINVAL; | |
3222 | return length; | |
3223 | } | |
3224 | SLAB_ATTR(validate); | |
3225 | ||
2086d26a CL |
3226 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
3227 | { | |
3228 | return 0; | |
3229 | } | |
3230 | ||
3231 | static ssize_t shrink_store(struct kmem_cache *s, | |
3232 | const char *buf, size_t length) | |
3233 | { | |
3234 | if (buf[0] == '1') { | |
3235 | int rc = kmem_cache_shrink(s); | |
3236 | ||
3237 | if (rc) | |
3238 | return rc; | |
3239 | } else | |
3240 | return -EINVAL; | |
3241 | return length; | |
3242 | } | |
3243 | SLAB_ATTR(shrink); | |
3244 | ||
88a420e4 CL |
3245 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
3246 | { | |
3247 | if (!(s->flags & SLAB_STORE_USER)) | |
3248 | return -ENOSYS; | |
3249 | return list_locations(s, buf, TRACK_ALLOC); | |
3250 | } | |
3251 | SLAB_ATTR_RO(alloc_calls); | |
3252 | ||
3253 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
3254 | { | |
3255 | if (!(s->flags & SLAB_STORE_USER)) | |
3256 | return -ENOSYS; | |
3257 | return list_locations(s, buf, TRACK_FREE); | |
3258 | } | |
3259 | SLAB_ATTR_RO(free_calls); | |
3260 | ||
81819f0f CL |
3261 | #ifdef CONFIG_NUMA |
3262 | static ssize_t defrag_ratio_show(struct kmem_cache *s, char *buf) | |
3263 | { | |
3264 | return sprintf(buf, "%d\n", s->defrag_ratio / 10); | |
3265 | } | |
3266 | ||
3267 | static ssize_t defrag_ratio_store(struct kmem_cache *s, | |
3268 | const char *buf, size_t length) | |
3269 | { | |
3270 | int n = simple_strtoul(buf, NULL, 10); | |
3271 | ||
3272 | if (n < 100) | |
3273 | s->defrag_ratio = n * 10; | |
3274 | return length; | |
3275 | } | |
3276 | SLAB_ATTR(defrag_ratio); | |
3277 | #endif | |
3278 | ||
3279 | static struct attribute * slab_attrs[] = { | |
3280 | &slab_size_attr.attr, | |
3281 | &object_size_attr.attr, | |
3282 | &objs_per_slab_attr.attr, | |
3283 | &order_attr.attr, | |
3284 | &objects_attr.attr, | |
3285 | &slabs_attr.attr, | |
3286 | &partial_attr.attr, | |
3287 | &cpu_slabs_attr.attr, | |
3288 | &ctor_attr.attr, | |
3289 | &dtor_attr.attr, | |
3290 | &aliases_attr.attr, | |
3291 | &align_attr.attr, | |
3292 | &sanity_checks_attr.attr, | |
3293 | &trace_attr.attr, | |
3294 | &hwcache_align_attr.attr, | |
3295 | &reclaim_account_attr.attr, | |
3296 | &destroy_by_rcu_attr.attr, | |
3297 | &red_zone_attr.attr, | |
3298 | &poison_attr.attr, | |
3299 | &store_user_attr.attr, | |
53e15af0 | 3300 | &validate_attr.attr, |
2086d26a | 3301 | &shrink_attr.attr, |
88a420e4 CL |
3302 | &alloc_calls_attr.attr, |
3303 | &free_calls_attr.attr, | |
81819f0f CL |
3304 | #ifdef CONFIG_ZONE_DMA |
3305 | &cache_dma_attr.attr, | |
3306 | #endif | |
3307 | #ifdef CONFIG_NUMA | |
3308 | &defrag_ratio_attr.attr, | |
3309 | #endif | |
3310 | NULL | |
3311 | }; | |
3312 | ||
3313 | static struct attribute_group slab_attr_group = { | |
3314 | .attrs = slab_attrs, | |
3315 | }; | |
3316 | ||
3317 | static ssize_t slab_attr_show(struct kobject *kobj, | |
3318 | struct attribute *attr, | |
3319 | char *buf) | |
3320 | { | |
3321 | struct slab_attribute *attribute; | |
3322 | struct kmem_cache *s; | |
3323 | int err; | |
3324 | ||
3325 | attribute = to_slab_attr(attr); | |
3326 | s = to_slab(kobj); | |
3327 | ||
3328 | if (!attribute->show) | |
3329 | return -EIO; | |
3330 | ||
3331 | err = attribute->show(s, buf); | |
3332 | ||
3333 | return err; | |
3334 | } | |
3335 | ||
3336 | static ssize_t slab_attr_store(struct kobject *kobj, | |
3337 | struct attribute *attr, | |
3338 | const char *buf, size_t len) | |
3339 | { | |
3340 | struct slab_attribute *attribute; | |
3341 | struct kmem_cache *s; | |
3342 | int err; | |
3343 | ||
3344 | attribute = to_slab_attr(attr); | |
3345 | s = to_slab(kobj); | |
3346 | ||
3347 | if (!attribute->store) | |
3348 | return -EIO; | |
3349 | ||
3350 | err = attribute->store(s, buf, len); | |
3351 | ||
3352 | return err; | |
3353 | } | |
3354 | ||
3355 | static struct sysfs_ops slab_sysfs_ops = { | |
3356 | .show = slab_attr_show, | |
3357 | .store = slab_attr_store, | |
3358 | }; | |
3359 | ||
3360 | static struct kobj_type slab_ktype = { | |
3361 | .sysfs_ops = &slab_sysfs_ops, | |
3362 | }; | |
3363 | ||
3364 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
3365 | { | |
3366 | struct kobj_type *ktype = get_ktype(kobj); | |
3367 | ||
3368 | if (ktype == &slab_ktype) | |
3369 | return 1; | |
3370 | return 0; | |
3371 | } | |
3372 | ||
3373 | static struct kset_uevent_ops slab_uevent_ops = { | |
3374 | .filter = uevent_filter, | |
3375 | }; | |
3376 | ||
3377 | decl_subsys(slab, &slab_ktype, &slab_uevent_ops); | |
3378 | ||
3379 | #define ID_STR_LENGTH 64 | |
3380 | ||
3381 | /* Create a unique string id for a slab cache: | |
3382 | * format | |
3383 | * :[flags-]size:[memory address of kmemcache] | |
3384 | */ | |
3385 | static char *create_unique_id(struct kmem_cache *s) | |
3386 | { | |
3387 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
3388 | char *p = name; | |
3389 | ||
3390 | BUG_ON(!name); | |
3391 | ||
3392 | *p++ = ':'; | |
3393 | /* | |
3394 | * First flags affecting slabcache operations. We will only | |
3395 | * get here for aliasable slabs so we do not need to support | |
3396 | * too many flags. The flags here must cover all flags that | |
3397 | * are matched during merging to guarantee that the id is | |
3398 | * unique. | |
3399 | */ | |
3400 | if (s->flags & SLAB_CACHE_DMA) | |
3401 | *p++ = 'd'; | |
3402 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3403 | *p++ = 'a'; | |
3404 | if (s->flags & SLAB_DEBUG_FREE) | |
3405 | *p++ = 'F'; | |
3406 | if (p != name + 1) | |
3407 | *p++ = '-'; | |
3408 | p += sprintf(p, "%07d", s->size); | |
3409 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
3410 | return name; | |
3411 | } | |
3412 | ||
3413 | static int sysfs_slab_add(struct kmem_cache *s) | |
3414 | { | |
3415 | int err; | |
3416 | const char *name; | |
3417 | int unmergeable; | |
3418 | ||
3419 | if (slab_state < SYSFS) | |
3420 | /* Defer until later */ | |
3421 | return 0; | |
3422 | ||
3423 | unmergeable = slab_unmergeable(s); | |
3424 | if (unmergeable) { | |
3425 | /* | |
3426 | * Slabcache can never be merged so we can use the name proper. | |
3427 | * This is typically the case for debug situations. In that | |
3428 | * case we can catch duplicate names easily. | |
3429 | */ | |
3430 | sysfs_remove_link(&slab_subsys.kset.kobj, s->name); | |
3431 | name = s->name; | |
3432 | } else { | |
3433 | /* | |
3434 | * Create a unique name for the slab as a target | |
3435 | * for the symlinks. | |
3436 | */ | |
3437 | name = create_unique_id(s); | |
3438 | } | |
3439 | ||
3440 | kobj_set_kset_s(s, slab_subsys); | |
3441 | kobject_set_name(&s->kobj, name); | |
3442 | kobject_init(&s->kobj); | |
3443 | err = kobject_add(&s->kobj); | |
3444 | if (err) | |
3445 | return err; | |
3446 | ||
3447 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
3448 | if (err) | |
3449 | return err; | |
3450 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
3451 | if (!unmergeable) { | |
3452 | /* Setup first alias */ | |
3453 | sysfs_slab_alias(s, s->name); | |
3454 | kfree(name); | |
3455 | } | |
3456 | return 0; | |
3457 | } | |
3458 | ||
3459 | static void sysfs_slab_remove(struct kmem_cache *s) | |
3460 | { | |
3461 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
3462 | kobject_del(&s->kobj); | |
3463 | } | |
3464 | ||
3465 | /* | |
3466 | * Need to buffer aliases during bootup until sysfs becomes | |
3467 | * available lest we loose that information. | |
3468 | */ | |
3469 | struct saved_alias { | |
3470 | struct kmem_cache *s; | |
3471 | const char *name; | |
3472 | struct saved_alias *next; | |
3473 | }; | |
3474 | ||
3475 | struct saved_alias *alias_list; | |
3476 | ||
3477 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
3478 | { | |
3479 | struct saved_alias *al; | |
3480 | ||
3481 | if (slab_state == SYSFS) { | |
3482 | /* | |
3483 | * If we have a leftover link then remove it. | |
3484 | */ | |
3485 | sysfs_remove_link(&slab_subsys.kset.kobj, name); | |
3486 | return sysfs_create_link(&slab_subsys.kset.kobj, | |
3487 | &s->kobj, name); | |
3488 | } | |
3489 | ||
3490 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
3491 | if (!al) | |
3492 | return -ENOMEM; | |
3493 | ||
3494 | al->s = s; | |
3495 | al->name = name; | |
3496 | al->next = alias_list; | |
3497 | alias_list = al; | |
3498 | return 0; | |
3499 | } | |
3500 | ||
3501 | static int __init slab_sysfs_init(void) | |
3502 | { | |
3503 | int err; | |
3504 | ||
3505 | err = subsystem_register(&slab_subsys); | |
3506 | if (err) { | |
3507 | printk(KERN_ERR "Cannot register slab subsystem.\n"); | |
3508 | return -ENOSYS; | |
3509 | } | |
3510 | ||
3511 | finish_bootstrap(); | |
3512 | ||
3513 | while (alias_list) { | |
3514 | struct saved_alias *al = alias_list; | |
3515 | ||
3516 | alias_list = alias_list->next; | |
3517 | err = sysfs_slab_alias(al->s, al->name); | |
3518 | BUG_ON(err); | |
3519 | kfree(al); | |
3520 | } | |
3521 | ||
3522 | resiliency_test(); | |
3523 | return 0; | |
3524 | } | |
3525 | ||
3526 | __initcall(slab_sysfs_init); | |
3527 | #else | |
3528 | __initcall(finish_bootstrap); | |
3529 | #endif |