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> | |
b9049e23 | 23 | #include <linux/memory.h> |
81819f0f CL |
24 | |
25 | /* | |
26 | * Lock order: | |
27 | * 1. slab_lock(page) | |
28 | * 2. slab->list_lock | |
29 | * | |
30 | * The slab_lock protects operations on the object of a particular | |
31 | * slab and its metadata in the page struct. If the slab lock | |
32 | * has been taken then no allocations nor frees can be performed | |
33 | * on the objects in the slab nor can the slab be added or removed | |
34 | * from the partial or full lists since this would mean modifying | |
35 | * the page_struct of the slab. | |
36 | * | |
37 | * The list_lock protects the partial and full list on each node and | |
38 | * the partial slab counter. If taken then no new slabs may be added or | |
39 | * removed from the lists nor make the number of partial slabs be modified. | |
40 | * (Note that the total number of slabs is an atomic value that may be | |
41 | * modified without taking the list lock). | |
42 | * | |
43 | * The list_lock is a centralized lock and thus we avoid taking it as | |
44 | * much as possible. As long as SLUB does not have to handle partial | |
45 | * slabs, operations can continue without any centralized lock. F.e. | |
46 | * allocating a long series of objects that fill up slabs does not require | |
47 | * the list lock. | |
48 | * | |
49 | * The lock order is sometimes inverted when we are trying to get a slab | |
50 | * off a list. We take the list_lock and then look for a page on the list | |
51 | * to use. While we do that objects in the slabs may be freed. We can | |
52 | * only operate on the slab if we have also taken the slab_lock. So we use | |
53 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
54 | * can occur anymore and we can use the slab for allocations etc. If the | |
55 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
56 | * we must stay away from it for a while since we may cause a bouncing | |
57 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
58 | * If all pages are busy then we may allocate a new slab instead of reusing | |
59 | * a partial slab. A new slab has noone operating on it and thus there is | |
60 | * no danger of cacheline contention. | |
61 | * | |
62 | * Interrupts are disabled during allocation and deallocation in order to | |
63 | * make the slab allocator safe to use in the context of an irq. In addition | |
64 | * interrupts are disabled to ensure that the processor does not change | |
65 | * while handling per_cpu slabs, due to kernel preemption. | |
66 | * | |
67 | * SLUB assigns one slab for allocation to each processor. | |
68 | * Allocations only occur from these slabs called cpu slabs. | |
69 | * | |
672bba3a CL |
70 | * Slabs with free elements are kept on a partial list and during regular |
71 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 72 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
73 | * We track full slabs for debugging purposes though because otherwise we |
74 | * cannot scan all objects. | |
81819f0f CL |
75 | * |
76 | * Slabs are freed when they become empty. Teardown and setup is | |
77 | * minimal so we rely on the page allocators per cpu caches for | |
78 | * fast frees and allocs. | |
79 | * | |
80 | * Overloading of page flags that are otherwise used for LRU management. | |
81 | * | |
4b6f0750 CL |
82 | * PageActive The slab is frozen and exempt from list processing. |
83 | * This means that the slab is dedicated to a purpose | |
84 | * such as satisfying allocations for a specific | |
85 | * processor. Objects may be freed in the slab while | |
86 | * it is frozen but slab_free will then skip the usual | |
87 | * list operations. It is up to the processor holding | |
88 | * the slab to integrate the slab into the slab lists | |
89 | * when the slab is no longer needed. | |
90 | * | |
91 | * One use of this flag is to mark slabs that are | |
92 | * used for allocations. Then such a slab becomes a cpu | |
93 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 94 | * freelist that allows lockless access to |
894b8788 CL |
95 | * free objects in addition to the regular freelist |
96 | * that requires the slab lock. | |
81819f0f CL |
97 | * |
98 | * PageError Slab requires special handling due to debug | |
99 | * options set. This moves slab handling out of | |
894b8788 | 100 | * the fast path and disables lockless freelists. |
81819f0f CL |
101 | */ |
102 | ||
5577bd8a CL |
103 | #define FROZEN (1 << PG_active) |
104 | ||
105 | #ifdef CONFIG_SLUB_DEBUG | |
106 | #define SLABDEBUG (1 << PG_error) | |
107 | #else | |
108 | #define SLABDEBUG 0 | |
109 | #endif | |
110 | ||
4b6f0750 CL |
111 | static inline int SlabFrozen(struct page *page) |
112 | { | |
5577bd8a | 113 | return page->flags & FROZEN; |
4b6f0750 CL |
114 | } |
115 | ||
116 | static inline void SetSlabFrozen(struct page *page) | |
117 | { | |
5577bd8a | 118 | page->flags |= FROZEN; |
4b6f0750 CL |
119 | } |
120 | ||
121 | static inline void ClearSlabFrozen(struct page *page) | |
122 | { | |
5577bd8a | 123 | page->flags &= ~FROZEN; |
4b6f0750 CL |
124 | } |
125 | ||
35e5d7ee CL |
126 | static inline int SlabDebug(struct page *page) |
127 | { | |
5577bd8a | 128 | return page->flags & SLABDEBUG; |
35e5d7ee CL |
129 | } |
130 | ||
131 | static inline void SetSlabDebug(struct page *page) | |
132 | { | |
5577bd8a | 133 | page->flags |= SLABDEBUG; |
35e5d7ee CL |
134 | } |
135 | ||
136 | static inline void ClearSlabDebug(struct page *page) | |
137 | { | |
5577bd8a | 138 | page->flags &= ~SLABDEBUG; |
35e5d7ee CL |
139 | } |
140 | ||
81819f0f CL |
141 | /* |
142 | * Issues still to be resolved: | |
143 | * | |
81819f0f CL |
144 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
145 | * | |
81819f0f CL |
146 | * - Variable sizing of the per node arrays |
147 | */ | |
148 | ||
149 | /* Enable to test recovery from slab corruption on boot */ | |
150 | #undef SLUB_RESILIENCY_TEST | |
151 | ||
1f84260c CL |
152 | /* |
153 | * Currently fastpath is not supported if preemption is enabled. | |
154 | */ | |
155 | #if defined(CONFIG_FAST_CMPXCHG_LOCAL) && !defined(CONFIG_PREEMPT) | |
156 | #define SLUB_FASTPATH | |
157 | #endif | |
158 | ||
81819f0f CL |
159 | #if PAGE_SHIFT <= 12 |
160 | ||
161 | /* | |
162 | * Small page size. Make sure that we do not fragment memory | |
163 | */ | |
164 | #define DEFAULT_MAX_ORDER 1 | |
165 | #define DEFAULT_MIN_OBJECTS 4 | |
166 | ||
167 | #else | |
168 | ||
169 | /* | |
170 | * Large page machines are customarily able to handle larger | |
171 | * page orders. | |
172 | */ | |
173 | #define DEFAULT_MAX_ORDER 2 | |
174 | #define DEFAULT_MIN_OBJECTS 8 | |
175 | ||
176 | #endif | |
177 | ||
2086d26a CL |
178 | /* |
179 | * Mininum number of partial slabs. These will be left on the partial | |
180 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
181 | */ | |
76be8950 | 182 | #define MIN_PARTIAL 5 |
e95eed57 | 183 | |
2086d26a CL |
184 | /* |
185 | * Maximum number of desirable partial slabs. | |
186 | * The existence of more partial slabs makes kmem_cache_shrink | |
187 | * sort the partial list by the number of objects in the. | |
188 | */ | |
189 | #define MAX_PARTIAL 10 | |
190 | ||
81819f0f CL |
191 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
192 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 193 | |
81819f0f CL |
194 | /* |
195 | * Set of flags that will prevent slab merging | |
196 | */ | |
197 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
198 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
199 | ||
200 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
201 | SLAB_CACHE_DMA) | |
202 | ||
203 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 204 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
205 | #endif |
206 | ||
207 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 208 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
209 | #endif |
210 | ||
211 | /* Internal SLUB flags */ | |
1ceef402 CL |
212 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
213 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f | 214 | |
65c02d4c CL |
215 | /* Not all arches define cache_line_size */ |
216 | #ifndef cache_line_size | |
217 | #define cache_line_size() L1_CACHE_BYTES | |
218 | #endif | |
219 | ||
81819f0f CL |
220 | static int kmem_size = sizeof(struct kmem_cache); |
221 | ||
222 | #ifdef CONFIG_SMP | |
223 | static struct notifier_block slab_notifier; | |
224 | #endif | |
225 | ||
226 | static enum { | |
227 | DOWN, /* No slab functionality available */ | |
228 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 229 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
230 | SYSFS /* Sysfs up */ |
231 | } slab_state = DOWN; | |
232 | ||
233 | /* A list of all slab caches on the system */ | |
234 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 235 | static LIST_HEAD(slab_caches); |
81819f0f | 236 | |
02cbc874 CL |
237 | /* |
238 | * Tracking user of a slab. | |
239 | */ | |
240 | struct track { | |
241 | void *addr; /* Called from address */ | |
242 | int cpu; /* Was running on cpu */ | |
243 | int pid; /* Pid context */ | |
244 | unsigned long when; /* When did the operation occur */ | |
245 | }; | |
246 | ||
247 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
248 | ||
41ecc55b | 249 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
250 | static int sysfs_slab_add(struct kmem_cache *); |
251 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
252 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 253 | |
81819f0f | 254 | #else |
0c710013 CL |
255 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
256 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
257 | { return 0; } | |
151c602f CL |
258 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
259 | { | |
260 | kfree(s); | |
261 | } | |
8ff12cfc | 262 | |
81819f0f CL |
263 | #endif |
264 | ||
8ff12cfc CL |
265 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
266 | { | |
267 | #ifdef CONFIG_SLUB_STATS | |
268 | c->stat[si]++; | |
269 | #endif | |
270 | } | |
271 | ||
81819f0f CL |
272 | /******************************************************************** |
273 | * Core slab cache functions | |
274 | *******************************************************************/ | |
275 | ||
276 | int slab_is_available(void) | |
277 | { | |
278 | return slab_state >= UP; | |
279 | } | |
280 | ||
281 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
282 | { | |
283 | #ifdef CONFIG_NUMA | |
284 | return s->node[node]; | |
285 | #else | |
286 | return &s->local_node; | |
287 | #endif | |
288 | } | |
289 | ||
dfb4f096 CL |
290 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
291 | { | |
4c93c355 CL |
292 | #ifdef CONFIG_SMP |
293 | return s->cpu_slab[cpu]; | |
294 | #else | |
295 | return &s->cpu_slab; | |
296 | #endif | |
dfb4f096 CL |
297 | } |
298 | ||
683d0baa CL |
299 | /* |
300 | * The end pointer in a slab is special. It points to the first object in the | |
301 | * slab but has bit 0 set to mark it. | |
302 | * | |
303 | * Note that SLUB relies on page_mapping returning NULL for pages with bit 0 | |
304 | * in the mapping set. | |
305 | */ | |
306 | static inline int is_end(void *addr) | |
307 | { | |
308 | return (unsigned long)addr & PAGE_MAPPING_ANON; | |
309 | } | |
310 | ||
dada123d | 311 | static void *slab_address(struct page *page) |
683d0baa CL |
312 | { |
313 | return page->end - PAGE_MAPPING_ANON; | |
314 | } | |
315 | ||
02cbc874 CL |
316 | static inline int check_valid_pointer(struct kmem_cache *s, |
317 | struct page *page, const void *object) | |
318 | { | |
319 | void *base; | |
320 | ||
683d0baa | 321 | if (object == page->end) |
02cbc874 CL |
322 | return 1; |
323 | ||
683d0baa | 324 | base = slab_address(page); |
02cbc874 CL |
325 | if (object < base || object >= base + s->objects * s->size || |
326 | (object - base) % s->size) { | |
327 | return 0; | |
328 | } | |
329 | ||
330 | return 1; | |
331 | } | |
332 | ||
7656c72b CL |
333 | /* |
334 | * Slow version of get and set free pointer. | |
335 | * | |
336 | * This version requires touching the cache lines of kmem_cache which | |
337 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
338 | * from the page struct. | |
339 | */ | |
340 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
341 | { | |
342 | return *(void **)(object + s->offset); | |
343 | } | |
344 | ||
345 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
346 | { | |
347 | *(void **)(object + s->offset) = fp; | |
348 | } | |
349 | ||
350 | /* Loop over all objects in a slab */ | |
351 | #define for_each_object(__p, __s, __addr) \ | |
352 | for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\ | |
353 | __p += (__s)->size) | |
354 | ||
355 | /* Scan freelist */ | |
356 | #define for_each_free_object(__p, __s, __free) \ | |
683d0baa CL |
357 | for (__p = (__free); (__p) != page->end; __p = get_freepointer((__s),\ |
358 | __p)) | |
7656c72b CL |
359 | |
360 | /* Determine object index from a given position */ | |
361 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
362 | { | |
363 | return (p - addr) / s->size; | |
364 | } | |
365 | ||
41ecc55b CL |
366 | #ifdef CONFIG_SLUB_DEBUG |
367 | /* | |
368 | * Debug settings: | |
369 | */ | |
f0630fff CL |
370 | #ifdef CONFIG_SLUB_DEBUG_ON |
371 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
372 | #else | |
41ecc55b | 373 | static int slub_debug; |
f0630fff | 374 | #endif |
41ecc55b CL |
375 | |
376 | static char *slub_debug_slabs; | |
377 | ||
81819f0f CL |
378 | /* |
379 | * Object debugging | |
380 | */ | |
381 | static void print_section(char *text, u8 *addr, unsigned int length) | |
382 | { | |
383 | int i, offset; | |
384 | int newline = 1; | |
385 | char ascii[17]; | |
386 | ||
387 | ascii[16] = 0; | |
388 | ||
389 | for (i = 0; i < length; i++) { | |
390 | if (newline) { | |
24922684 | 391 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
392 | newline = 0; |
393 | } | |
06428780 | 394 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
395 | offset = i % 16; |
396 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
397 | if (offset == 15) { | |
06428780 | 398 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
399 | newline = 1; |
400 | } | |
401 | } | |
402 | if (!newline) { | |
403 | i %= 16; | |
404 | while (i < 16) { | |
06428780 | 405 | printk(KERN_CONT " "); |
81819f0f CL |
406 | ascii[i] = ' '; |
407 | i++; | |
408 | } | |
06428780 | 409 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
410 | } |
411 | } | |
412 | ||
81819f0f CL |
413 | static struct track *get_track(struct kmem_cache *s, void *object, |
414 | enum track_item alloc) | |
415 | { | |
416 | struct track *p; | |
417 | ||
418 | if (s->offset) | |
419 | p = object + s->offset + sizeof(void *); | |
420 | else | |
421 | p = object + s->inuse; | |
422 | ||
423 | return p + alloc; | |
424 | } | |
425 | ||
426 | static void set_track(struct kmem_cache *s, void *object, | |
427 | enum track_item alloc, void *addr) | |
428 | { | |
429 | struct track *p; | |
430 | ||
431 | if (s->offset) | |
432 | p = object + s->offset + sizeof(void *); | |
433 | else | |
434 | p = object + s->inuse; | |
435 | ||
436 | p += alloc; | |
437 | if (addr) { | |
438 | p->addr = addr; | |
439 | p->cpu = smp_processor_id(); | |
440 | p->pid = current ? current->pid : -1; | |
441 | p->when = jiffies; | |
442 | } else | |
443 | memset(p, 0, sizeof(struct track)); | |
444 | } | |
445 | ||
81819f0f CL |
446 | static void init_tracking(struct kmem_cache *s, void *object) |
447 | { | |
24922684 CL |
448 | if (!(s->flags & SLAB_STORE_USER)) |
449 | return; | |
450 | ||
451 | set_track(s, object, TRACK_FREE, NULL); | |
452 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
453 | } |
454 | ||
455 | static void print_track(const char *s, struct track *t) | |
456 | { | |
457 | if (!t->addr) | |
458 | return; | |
459 | ||
24922684 | 460 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 461 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
462 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
463 | } | |
464 | ||
465 | static void print_tracking(struct kmem_cache *s, void *object) | |
466 | { | |
467 | if (!(s->flags & SLAB_STORE_USER)) | |
468 | return; | |
469 | ||
470 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
471 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
472 | } | |
473 | ||
474 | static void print_page_info(struct page *page) | |
475 | { | |
476 | printk(KERN_ERR "INFO: Slab 0x%p used=%u fp=0x%p flags=0x%04lx\n", | |
477 | page, page->inuse, page->freelist, page->flags); | |
478 | ||
479 | } | |
480 | ||
481 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
482 | { | |
483 | va_list args; | |
484 | char buf[100]; | |
485 | ||
486 | va_start(args, fmt); | |
487 | vsnprintf(buf, sizeof(buf), fmt, args); | |
488 | va_end(args); | |
489 | printk(KERN_ERR "========================================" | |
490 | "=====================================\n"); | |
491 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
492 | printk(KERN_ERR "----------------------------------------" | |
493 | "-------------------------------------\n\n"); | |
81819f0f CL |
494 | } |
495 | ||
24922684 CL |
496 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
497 | { | |
498 | va_list args; | |
499 | char buf[100]; | |
500 | ||
501 | va_start(args, fmt); | |
502 | vsnprintf(buf, sizeof(buf), fmt, args); | |
503 | va_end(args); | |
504 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
505 | } | |
506 | ||
507 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
508 | { |
509 | unsigned int off; /* Offset of last byte */ | |
683d0baa | 510 | u8 *addr = slab_address(page); |
24922684 CL |
511 | |
512 | print_tracking(s, p); | |
513 | ||
514 | print_page_info(page); | |
515 | ||
516 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
517 | p, p - addr, get_freepointer(s, p)); | |
518 | ||
519 | if (p > addr + 16) | |
520 | print_section("Bytes b4", p - 16, 16); | |
521 | ||
522 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
523 | |
524 | if (s->flags & SLAB_RED_ZONE) | |
525 | print_section("Redzone", p + s->objsize, | |
526 | s->inuse - s->objsize); | |
527 | ||
81819f0f CL |
528 | if (s->offset) |
529 | off = s->offset + sizeof(void *); | |
530 | else | |
531 | off = s->inuse; | |
532 | ||
24922684 | 533 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 534 | off += 2 * sizeof(struct track); |
81819f0f CL |
535 | |
536 | if (off != s->size) | |
537 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
538 | print_section("Padding", p + off, s->size - off); |
539 | ||
540 | dump_stack(); | |
81819f0f CL |
541 | } |
542 | ||
543 | static void object_err(struct kmem_cache *s, struct page *page, | |
544 | u8 *object, char *reason) | |
545 | { | |
24922684 CL |
546 | slab_bug(s, reason); |
547 | print_trailer(s, page, object); | |
81819f0f CL |
548 | } |
549 | ||
24922684 | 550 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
551 | { |
552 | va_list args; | |
553 | char buf[100]; | |
554 | ||
24922684 CL |
555 | va_start(args, fmt); |
556 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 557 | va_end(args); |
24922684 CL |
558 | slab_bug(s, fmt); |
559 | print_page_info(page); | |
81819f0f CL |
560 | dump_stack(); |
561 | } | |
562 | ||
563 | static void init_object(struct kmem_cache *s, void *object, int active) | |
564 | { | |
565 | u8 *p = object; | |
566 | ||
567 | if (s->flags & __OBJECT_POISON) { | |
568 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 569 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
570 | } |
571 | ||
572 | if (s->flags & SLAB_RED_ZONE) | |
573 | memset(p + s->objsize, | |
574 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
575 | s->inuse - s->objsize); | |
576 | } | |
577 | ||
24922684 | 578 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
579 | { |
580 | while (bytes) { | |
581 | if (*start != (u8)value) | |
24922684 | 582 | return start; |
81819f0f CL |
583 | start++; |
584 | bytes--; | |
585 | } | |
24922684 CL |
586 | return NULL; |
587 | } | |
588 | ||
589 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
590 | void *from, void *to) | |
591 | { | |
592 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
593 | memset(from, data, to - from); | |
594 | } | |
595 | ||
596 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
597 | u8 *object, char *what, | |
06428780 | 598 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
599 | { |
600 | u8 *fault; | |
601 | u8 *end; | |
602 | ||
603 | fault = check_bytes(start, value, bytes); | |
604 | if (!fault) | |
605 | return 1; | |
606 | ||
607 | end = start + bytes; | |
608 | while (end > fault && end[-1] == value) | |
609 | end--; | |
610 | ||
611 | slab_bug(s, "%s overwritten", what); | |
612 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
613 | fault, end - 1, fault[0], value); | |
614 | print_trailer(s, page, object); | |
615 | ||
616 | restore_bytes(s, what, value, fault, end); | |
617 | return 0; | |
81819f0f CL |
618 | } |
619 | ||
81819f0f CL |
620 | /* |
621 | * Object layout: | |
622 | * | |
623 | * object address | |
624 | * Bytes of the object to be managed. | |
625 | * If the freepointer may overlay the object then the free | |
626 | * pointer is the first word of the object. | |
672bba3a | 627 | * |
81819f0f CL |
628 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
629 | * 0xa5 (POISON_END) | |
630 | * | |
631 | * object + s->objsize | |
632 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
633 | * Padding is extended by another word if Redzoning is enabled and |
634 | * objsize == inuse. | |
635 | * | |
81819f0f CL |
636 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
637 | * 0xcc (RED_ACTIVE) for objects in use. | |
638 | * | |
639 | * object + s->inuse | |
672bba3a CL |
640 | * Meta data starts here. |
641 | * | |
81819f0f CL |
642 | * A. Free pointer (if we cannot overwrite object on free) |
643 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a CL |
644 | * C. Padding to reach required alignment boundary or at mininum |
645 | * one word if debuggin is on to be able to detect writes | |
646 | * before the word boundary. | |
647 | * | |
648 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
649 | * |
650 | * object + s->size | |
672bba3a | 651 | * Nothing is used beyond s->size. |
81819f0f | 652 | * |
672bba3a CL |
653 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
654 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
655 | * may be used with merged slabcaches. |
656 | */ | |
657 | ||
81819f0f CL |
658 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
659 | { | |
660 | unsigned long off = s->inuse; /* The end of info */ | |
661 | ||
662 | if (s->offset) | |
663 | /* Freepointer is placed after the object. */ | |
664 | off += sizeof(void *); | |
665 | ||
666 | if (s->flags & SLAB_STORE_USER) | |
667 | /* We also have user information there */ | |
668 | off += 2 * sizeof(struct track); | |
669 | ||
670 | if (s->size == off) | |
671 | return 1; | |
672 | ||
24922684 CL |
673 | return check_bytes_and_report(s, page, p, "Object padding", |
674 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
675 | } |
676 | ||
677 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
678 | { | |
24922684 CL |
679 | u8 *start; |
680 | u8 *fault; | |
681 | u8 *end; | |
682 | int length; | |
683 | int remainder; | |
81819f0f CL |
684 | |
685 | if (!(s->flags & SLAB_POISON)) | |
686 | return 1; | |
687 | ||
683d0baa | 688 | start = slab_address(page); |
24922684 | 689 | end = start + (PAGE_SIZE << s->order); |
81819f0f | 690 | length = s->objects * s->size; |
24922684 | 691 | remainder = end - (start + length); |
81819f0f CL |
692 | if (!remainder) |
693 | return 1; | |
694 | ||
24922684 CL |
695 | fault = check_bytes(start + length, POISON_INUSE, remainder); |
696 | if (!fault) | |
697 | return 1; | |
698 | while (end > fault && end[-1] == POISON_INUSE) | |
699 | end--; | |
700 | ||
701 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
702 | print_section("Padding", start, length); | |
703 | ||
704 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
705 | return 0; | |
81819f0f CL |
706 | } |
707 | ||
708 | static int check_object(struct kmem_cache *s, struct page *page, | |
709 | void *object, int active) | |
710 | { | |
711 | u8 *p = object; | |
712 | u8 *endobject = object + s->objsize; | |
713 | ||
714 | if (s->flags & SLAB_RED_ZONE) { | |
715 | unsigned int red = | |
716 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
717 | ||
24922684 CL |
718 | if (!check_bytes_and_report(s, page, object, "Redzone", |
719 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 720 | return 0; |
81819f0f | 721 | } else { |
3adbefee IM |
722 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
723 | check_bytes_and_report(s, page, p, "Alignment padding", | |
724 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
725 | } | |
81819f0f CL |
726 | } |
727 | ||
728 | if (s->flags & SLAB_POISON) { | |
729 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
730 | (!check_bytes_and_report(s, page, p, "Poison", p, |
731 | POISON_FREE, s->objsize - 1) || | |
732 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 733 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 734 | return 0; |
81819f0f CL |
735 | /* |
736 | * check_pad_bytes cleans up on its own. | |
737 | */ | |
738 | check_pad_bytes(s, page, p); | |
739 | } | |
740 | ||
741 | if (!s->offset && active) | |
742 | /* | |
743 | * Object and freepointer overlap. Cannot check | |
744 | * freepointer while object is allocated. | |
745 | */ | |
746 | return 1; | |
747 | ||
748 | /* Check free pointer validity */ | |
749 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
750 | object_err(s, page, p, "Freepointer corrupt"); | |
751 | /* | |
752 | * No choice but to zap it and thus loose the remainder | |
753 | * of the free objects in this slab. May cause | |
672bba3a | 754 | * another error because the object count is now wrong. |
81819f0f | 755 | */ |
683d0baa | 756 | set_freepointer(s, p, page->end); |
81819f0f CL |
757 | return 0; |
758 | } | |
759 | return 1; | |
760 | } | |
761 | ||
762 | static int check_slab(struct kmem_cache *s, struct page *page) | |
763 | { | |
764 | VM_BUG_ON(!irqs_disabled()); | |
765 | ||
766 | if (!PageSlab(page)) { | |
24922684 | 767 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
768 | return 0; |
769 | } | |
81819f0f | 770 | if (page->inuse > s->objects) { |
24922684 CL |
771 | slab_err(s, page, "inuse %u > max %u", |
772 | s->name, page->inuse, s->objects); | |
81819f0f CL |
773 | return 0; |
774 | } | |
775 | /* Slab_pad_check fixes things up after itself */ | |
776 | slab_pad_check(s, page); | |
777 | return 1; | |
778 | } | |
779 | ||
780 | /* | |
672bba3a CL |
781 | * Determine if a certain object on a page is on the freelist. Must hold the |
782 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
783 | */ |
784 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
785 | { | |
786 | int nr = 0; | |
787 | void *fp = page->freelist; | |
788 | void *object = NULL; | |
789 | ||
683d0baa | 790 | while (fp != page->end && nr <= s->objects) { |
81819f0f CL |
791 | if (fp == search) |
792 | return 1; | |
793 | if (!check_valid_pointer(s, page, fp)) { | |
794 | if (object) { | |
795 | object_err(s, page, object, | |
796 | "Freechain corrupt"); | |
683d0baa | 797 | set_freepointer(s, object, page->end); |
81819f0f CL |
798 | break; |
799 | } else { | |
24922684 | 800 | slab_err(s, page, "Freepointer corrupt"); |
683d0baa | 801 | page->freelist = page->end; |
81819f0f | 802 | page->inuse = s->objects; |
24922684 | 803 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
804 | return 0; |
805 | } | |
806 | break; | |
807 | } | |
808 | object = fp; | |
809 | fp = get_freepointer(s, object); | |
810 | nr++; | |
811 | } | |
812 | ||
813 | if (page->inuse != s->objects - nr) { | |
70d71228 | 814 | slab_err(s, page, "Wrong object count. Counter is %d but " |
24922684 | 815 | "counted were %d", page->inuse, s->objects - nr); |
81819f0f | 816 | page->inuse = s->objects - nr; |
24922684 | 817 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
818 | } |
819 | return search == NULL; | |
820 | } | |
821 | ||
3ec09742 CL |
822 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
823 | { | |
824 | if (s->flags & SLAB_TRACE) { | |
825 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
826 | s->name, | |
827 | alloc ? "alloc" : "free", | |
828 | object, page->inuse, | |
829 | page->freelist); | |
830 | ||
831 | if (!alloc) | |
832 | print_section("Object", (void *)object, s->objsize); | |
833 | ||
834 | dump_stack(); | |
835 | } | |
836 | } | |
837 | ||
643b1138 | 838 | /* |
672bba3a | 839 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 840 | */ |
e95eed57 | 841 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 842 | { |
643b1138 CL |
843 | spin_lock(&n->list_lock); |
844 | list_add(&page->lru, &n->full); | |
845 | spin_unlock(&n->list_lock); | |
846 | } | |
847 | ||
848 | static void remove_full(struct kmem_cache *s, struct page *page) | |
849 | { | |
850 | struct kmem_cache_node *n; | |
851 | ||
852 | if (!(s->flags & SLAB_STORE_USER)) | |
853 | return; | |
854 | ||
855 | n = get_node(s, page_to_nid(page)); | |
856 | ||
857 | spin_lock(&n->list_lock); | |
858 | list_del(&page->lru); | |
859 | spin_unlock(&n->list_lock); | |
860 | } | |
861 | ||
3ec09742 CL |
862 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
863 | void *object) | |
864 | { | |
865 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
866 | return; | |
867 | ||
868 | init_object(s, object, 0); | |
869 | init_tracking(s, object); | |
870 | } | |
871 | ||
872 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
873 | void *object, void *addr) | |
81819f0f CL |
874 | { |
875 | if (!check_slab(s, page)) | |
876 | goto bad; | |
877 | ||
878 | if (object && !on_freelist(s, page, object)) { | |
24922684 | 879 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 880 | goto bad; |
81819f0f CL |
881 | } |
882 | ||
883 | if (!check_valid_pointer(s, page, object)) { | |
884 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 885 | goto bad; |
81819f0f CL |
886 | } |
887 | ||
3ec09742 | 888 | if (object && !check_object(s, page, object, 0)) |
81819f0f | 889 | goto bad; |
81819f0f | 890 | |
3ec09742 CL |
891 | /* Success perform special debug activities for allocs */ |
892 | if (s->flags & SLAB_STORE_USER) | |
893 | set_track(s, object, TRACK_ALLOC, addr); | |
894 | trace(s, page, object, 1); | |
895 | init_object(s, object, 1); | |
81819f0f | 896 | return 1; |
3ec09742 | 897 | |
81819f0f CL |
898 | bad: |
899 | if (PageSlab(page)) { | |
900 | /* | |
901 | * If this is a slab page then lets do the best we can | |
902 | * to avoid issues in the future. Marking all objects | |
672bba3a | 903 | * as used avoids touching the remaining objects. |
81819f0f | 904 | */ |
24922684 | 905 | slab_fix(s, "Marking all objects used"); |
81819f0f | 906 | page->inuse = s->objects; |
683d0baa | 907 | page->freelist = page->end; |
81819f0f CL |
908 | } |
909 | return 0; | |
910 | } | |
911 | ||
3ec09742 CL |
912 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
913 | void *object, void *addr) | |
81819f0f CL |
914 | { |
915 | if (!check_slab(s, page)) | |
916 | goto fail; | |
917 | ||
918 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 919 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
920 | goto fail; |
921 | } | |
922 | ||
923 | if (on_freelist(s, page, object)) { | |
24922684 | 924 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
925 | goto fail; |
926 | } | |
927 | ||
928 | if (!check_object(s, page, object, 1)) | |
929 | return 0; | |
930 | ||
931 | if (unlikely(s != page->slab)) { | |
3adbefee | 932 | if (!PageSlab(page)) { |
70d71228 CL |
933 | slab_err(s, page, "Attempt to free object(0x%p) " |
934 | "outside of slab", object); | |
3adbefee | 935 | } else if (!page->slab) { |
81819f0f | 936 | printk(KERN_ERR |
70d71228 | 937 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 938 | object); |
70d71228 | 939 | dump_stack(); |
06428780 | 940 | } else |
24922684 CL |
941 | object_err(s, page, object, |
942 | "page slab pointer corrupt."); | |
81819f0f CL |
943 | goto fail; |
944 | } | |
3ec09742 CL |
945 | |
946 | /* Special debug activities for freeing objects */ | |
683d0baa | 947 | if (!SlabFrozen(page) && page->freelist == page->end) |
3ec09742 CL |
948 | remove_full(s, page); |
949 | if (s->flags & SLAB_STORE_USER) | |
950 | set_track(s, object, TRACK_FREE, addr); | |
951 | trace(s, page, object, 0); | |
952 | init_object(s, object, 0); | |
81819f0f | 953 | return 1; |
3ec09742 | 954 | |
81819f0f | 955 | fail: |
24922684 | 956 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
957 | return 0; |
958 | } | |
959 | ||
41ecc55b CL |
960 | static int __init setup_slub_debug(char *str) |
961 | { | |
f0630fff CL |
962 | slub_debug = DEBUG_DEFAULT_FLAGS; |
963 | if (*str++ != '=' || !*str) | |
964 | /* | |
965 | * No options specified. Switch on full debugging. | |
966 | */ | |
967 | goto out; | |
968 | ||
969 | if (*str == ',') | |
970 | /* | |
971 | * No options but restriction on slabs. This means full | |
972 | * debugging for slabs matching a pattern. | |
973 | */ | |
974 | goto check_slabs; | |
975 | ||
976 | slub_debug = 0; | |
977 | if (*str == '-') | |
978 | /* | |
979 | * Switch off all debugging measures. | |
980 | */ | |
981 | goto out; | |
982 | ||
983 | /* | |
984 | * Determine which debug features should be switched on | |
985 | */ | |
06428780 | 986 | for (; *str && *str != ','; str++) { |
f0630fff CL |
987 | switch (tolower(*str)) { |
988 | case 'f': | |
989 | slub_debug |= SLAB_DEBUG_FREE; | |
990 | break; | |
991 | case 'z': | |
992 | slub_debug |= SLAB_RED_ZONE; | |
993 | break; | |
994 | case 'p': | |
995 | slub_debug |= SLAB_POISON; | |
996 | break; | |
997 | case 'u': | |
998 | slub_debug |= SLAB_STORE_USER; | |
999 | break; | |
1000 | case 't': | |
1001 | slub_debug |= SLAB_TRACE; | |
1002 | break; | |
1003 | default: | |
1004 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1005 | "unknown. skipped\n", *str); |
f0630fff | 1006 | } |
41ecc55b CL |
1007 | } |
1008 | ||
f0630fff | 1009 | check_slabs: |
41ecc55b CL |
1010 | if (*str == ',') |
1011 | slub_debug_slabs = str + 1; | |
f0630fff | 1012 | out: |
41ecc55b CL |
1013 | return 1; |
1014 | } | |
1015 | ||
1016 | __setup("slub_debug", setup_slub_debug); | |
1017 | ||
ba0268a8 CL |
1018 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1019 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1020 | void (*ctor)(struct kmem_cache *, void *)) |
41ecc55b CL |
1021 | { |
1022 | /* | |
1023 | * The page->offset field is only 16 bit wide. This is an offset | |
1024 | * in units of words from the beginning of an object. If the slab | |
1025 | * size is bigger then we cannot move the free pointer behind the | |
1026 | * object anymore. | |
1027 | * | |
1028 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
1029 | * the limit is 512k. | |
1030 | * | |
c59def9f | 1031 | * Debugging or ctor may create a need to move the free |
41ecc55b CL |
1032 | * pointer. Fail if this happens. |
1033 | */ | |
ba0268a8 CL |
1034 | if (objsize >= 65535 * sizeof(void *)) { |
1035 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
41ecc55b | 1036 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); |
ba0268a8 CL |
1037 | BUG_ON(ctor); |
1038 | } else { | |
41ecc55b CL |
1039 | /* |
1040 | * Enable debugging if selected on the kernel commandline. | |
1041 | */ | |
1042 | if (slub_debug && (!slub_debug_slabs || | |
ba0268a8 | 1043 | strncmp(slub_debug_slabs, name, |
3adbefee | 1044 | strlen(slub_debug_slabs)) == 0)) |
ba0268a8 CL |
1045 | flags |= slub_debug; |
1046 | } | |
1047 | ||
1048 | return flags; | |
41ecc55b CL |
1049 | } |
1050 | #else | |
3ec09742 CL |
1051 | static inline void setup_object_debug(struct kmem_cache *s, |
1052 | struct page *page, void *object) {} | |
41ecc55b | 1053 | |
3ec09742 CL |
1054 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1055 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1056 | |
3ec09742 CL |
1057 | static inline int free_debug_processing(struct kmem_cache *s, |
1058 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1059 | |
41ecc55b CL |
1060 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1061 | { return 1; } | |
1062 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1063 | void *object, int active) { return 1; } | |
3ec09742 | 1064 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1065 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1066 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1067 | void (*ctor)(struct kmem_cache *, void *)) |
ba0268a8 CL |
1068 | { |
1069 | return flags; | |
1070 | } | |
41ecc55b CL |
1071 | #define slub_debug 0 |
1072 | #endif | |
81819f0f CL |
1073 | /* |
1074 | * Slab allocation and freeing | |
1075 | */ | |
1076 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1077 | { | |
06428780 | 1078 | struct page *page; |
81819f0f CL |
1079 | int pages = 1 << s->order; |
1080 | ||
1081 | if (s->order) | |
1082 | flags |= __GFP_COMP; | |
1083 | ||
1084 | if (s->flags & SLAB_CACHE_DMA) | |
1085 | flags |= SLUB_DMA; | |
1086 | ||
e12ba74d MG |
1087 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
1088 | flags |= __GFP_RECLAIMABLE; | |
1089 | ||
81819f0f CL |
1090 | if (node == -1) |
1091 | page = alloc_pages(flags, s->order); | |
1092 | else | |
1093 | page = alloc_pages_node(node, flags, s->order); | |
1094 | ||
1095 | if (!page) | |
1096 | return NULL; | |
1097 | ||
1098 | mod_zone_page_state(page_zone(page), | |
1099 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1100 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1101 | pages); | |
1102 | ||
1103 | return page; | |
1104 | } | |
1105 | ||
1106 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1107 | void *object) | |
1108 | { | |
3ec09742 | 1109 | setup_object_debug(s, page, object); |
4f104934 | 1110 | if (unlikely(s->ctor)) |
4ba9b9d0 | 1111 | s->ctor(s, object); |
81819f0f CL |
1112 | } |
1113 | ||
1114 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1115 | { | |
1116 | struct page *page; | |
1117 | struct kmem_cache_node *n; | |
1118 | void *start; | |
81819f0f CL |
1119 | void *last; |
1120 | void *p; | |
1121 | ||
6cb06229 | 1122 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1123 | |
6cb06229 CL |
1124 | page = allocate_slab(s, |
1125 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1126 | if (!page) |
1127 | goto out; | |
1128 | ||
1129 | n = get_node(s, page_to_nid(page)); | |
1130 | if (n) | |
1131 | atomic_long_inc(&n->nr_slabs); | |
81819f0f CL |
1132 | page->slab = s; |
1133 | page->flags |= 1 << PG_slab; | |
1134 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1135 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1136 | SetSlabDebug(page); |
81819f0f CL |
1137 | |
1138 | start = page_address(page); | |
683d0baa | 1139 | page->end = start + 1; |
81819f0f CL |
1140 | |
1141 | if (unlikely(s->flags & SLAB_POISON)) | |
1142 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
1143 | ||
1144 | last = start; | |
7656c72b | 1145 | for_each_object(p, s, start) { |
81819f0f CL |
1146 | setup_object(s, page, last); |
1147 | set_freepointer(s, last, p); | |
1148 | last = p; | |
1149 | } | |
1150 | setup_object(s, page, last); | |
683d0baa | 1151 | set_freepointer(s, last, page->end); |
81819f0f CL |
1152 | |
1153 | page->freelist = start; | |
1154 | page->inuse = 0; | |
1155 | out: | |
81819f0f CL |
1156 | return page; |
1157 | } | |
1158 | ||
1159 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1160 | { | |
1161 | int pages = 1 << s->order; | |
1162 | ||
c59def9f | 1163 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1164 | void *p; |
1165 | ||
1166 | slab_pad_check(s, page); | |
683d0baa | 1167 | for_each_object(p, s, slab_address(page)) |
81819f0f | 1168 | check_object(s, page, p, 0); |
2208b764 | 1169 | ClearSlabDebug(page); |
81819f0f CL |
1170 | } |
1171 | ||
1172 | mod_zone_page_state(page_zone(page), | |
1173 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1174 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1175 | -pages); |
81819f0f | 1176 | |
683d0baa | 1177 | page->mapping = NULL; |
81819f0f CL |
1178 | __free_pages(page, s->order); |
1179 | } | |
1180 | ||
1181 | static void rcu_free_slab(struct rcu_head *h) | |
1182 | { | |
1183 | struct page *page; | |
1184 | ||
1185 | page = container_of((struct list_head *)h, struct page, lru); | |
1186 | __free_slab(page->slab, page); | |
1187 | } | |
1188 | ||
1189 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1190 | { | |
1191 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1192 | /* | |
1193 | * RCU free overloads the RCU head over the LRU | |
1194 | */ | |
1195 | struct rcu_head *head = (void *)&page->lru; | |
1196 | ||
1197 | call_rcu(head, rcu_free_slab); | |
1198 | } else | |
1199 | __free_slab(s, page); | |
1200 | } | |
1201 | ||
1202 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1203 | { | |
1204 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1205 | ||
1206 | atomic_long_dec(&n->nr_slabs); | |
1207 | reset_page_mapcount(page); | |
35e5d7ee | 1208 | __ClearPageSlab(page); |
81819f0f CL |
1209 | free_slab(s, page); |
1210 | } | |
1211 | ||
1212 | /* | |
1213 | * Per slab locking using the pagelock | |
1214 | */ | |
1215 | static __always_inline void slab_lock(struct page *page) | |
1216 | { | |
1217 | bit_spin_lock(PG_locked, &page->flags); | |
1218 | } | |
1219 | ||
1220 | static __always_inline void slab_unlock(struct page *page) | |
1221 | { | |
a76d3546 | 1222 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1223 | } |
1224 | ||
1225 | static __always_inline int slab_trylock(struct page *page) | |
1226 | { | |
1227 | int rc = 1; | |
1228 | ||
1229 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1230 | return rc; | |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Management of partially allocated slabs | |
1235 | */ | |
7c2e132c CL |
1236 | static void add_partial(struct kmem_cache_node *n, |
1237 | struct page *page, int tail) | |
81819f0f | 1238 | { |
e95eed57 CL |
1239 | spin_lock(&n->list_lock); |
1240 | n->nr_partial++; | |
7c2e132c CL |
1241 | if (tail) |
1242 | list_add_tail(&page->lru, &n->partial); | |
1243 | else | |
1244 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1245 | spin_unlock(&n->list_lock); |
1246 | } | |
1247 | ||
1248 | static void remove_partial(struct kmem_cache *s, | |
1249 | struct page *page) | |
1250 | { | |
1251 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1252 | ||
1253 | spin_lock(&n->list_lock); | |
1254 | list_del(&page->lru); | |
1255 | n->nr_partial--; | |
1256 | spin_unlock(&n->list_lock); | |
1257 | } | |
1258 | ||
1259 | /* | |
672bba3a | 1260 | * Lock slab and remove from the partial list. |
81819f0f | 1261 | * |
672bba3a | 1262 | * Must hold list_lock. |
81819f0f | 1263 | */ |
4b6f0750 | 1264 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1265 | { |
1266 | if (slab_trylock(page)) { | |
1267 | list_del(&page->lru); | |
1268 | n->nr_partial--; | |
4b6f0750 | 1269 | SetSlabFrozen(page); |
81819f0f CL |
1270 | return 1; |
1271 | } | |
1272 | return 0; | |
1273 | } | |
1274 | ||
1275 | /* | |
672bba3a | 1276 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1277 | */ |
1278 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1279 | { | |
1280 | struct page *page; | |
1281 | ||
1282 | /* | |
1283 | * Racy check. If we mistakenly see no partial slabs then we | |
1284 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1285 | * partial slab and there is none available then get_partials() |
1286 | * will return NULL. | |
81819f0f CL |
1287 | */ |
1288 | if (!n || !n->nr_partial) | |
1289 | return NULL; | |
1290 | ||
1291 | spin_lock(&n->list_lock); | |
1292 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1293 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1294 | goto out; |
1295 | page = NULL; | |
1296 | out: | |
1297 | spin_unlock(&n->list_lock); | |
1298 | return page; | |
1299 | } | |
1300 | ||
1301 | /* | |
672bba3a | 1302 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1303 | */ |
1304 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1305 | { | |
1306 | #ifdef CONFIG_NUMA | |
1307 | struct zonelist *zonelist; | |
1308 | struct zone **z; | |
1309 | struct page *page; | |
1310 | ||
1311 | /* | |
672bba3a CL |
1312 | * The defrag ratio allows a configuration of the tradeoffs between |
1313 | * inter node defragmentation and node local allocations. A lower | |
1314 | * defrag_ratio increases the tendency to do local allocations | |
1315 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1316 | * |
672bba3a CL |
1317 | * If the defrag_ratio is set to 0 then kmalloc() always |
1318 | * returns node local objects. If the ratio is higher then kmalloc() | |
1319 | * may return off node objects because partial slabs are obtained | |
1320 | * from other nodes and filled up. | |
81819f0f CL |
1321 | * |
1322 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
672bba3a CL |
1323 | * defrag_ratio = 1000) then every (well almost) allocation will |
1324 | * first attempt to defrag slab caches on other nodes. This means | |
1325 | * scanning over all nodes to look for partial slabs which may be | |
1326 | * expensive if we do it every time we are trying to find a slab | |
1327 | * with available objects. | |
81819f0f | 1328 | */ |
9824601e CL |
1329 | if (!s->remote_node_defrag_ratio || |
1330 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1331 | return NULL; |
1332 | ||
3adbefee IM |
1333 | zonelist = &NODE_DATA( |
1334 | slab_node(current->mempolicy))->node_zonelists[gfp_zone(flags)]; | |
81819f0f CL |
1335 | for (z = zonelist->zones; *z; z++) { |
1336 | struct kmem_cache_node *n; | |
1337 | ||
1338 | n = get_node(s, zone_to_nid(*z)); | |
1339 | ||
1340 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1341 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1342 | page = get_partial_node(n); |
1343 | if (page) | |
1344 | return page; | |
1345 | } | |
1346 | } | |
1347 | #endif | |
1348 | return NULL; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * Get a partial page, lock it and return it. | |
1353 | */ | |
1354 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1355 | { | |
1356 | struct page *page; | |
1357 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1358 | ||
1359 | page = get_partial_node(get_node(s, searchnode)); | |
1360 | if (page || (flags & __GFP_THISNODE)) | |
1361 | return page; | |
1362 | ||
1363 | return get_any_partial(s, flags); | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Move a page back to the lists. | |
1368 | * | |
1369 | * Must be called with the slab lock held. | |
1370 | * | |
1371 | * On exit the slab lock will have been dropped. | |
1372 | */ | |
7c2e132c | 1373 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1374 | { |
e95eed57 | 1375 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1376 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1377 | |
4b6f0750 | 1378 | ClearSlabFrozen(page); |
81819f0f | 1379 | if (page->inuse) { |
e95eed57 | 1380 | |
8ff12cfc | 1381 | if (page->freelist != page->end) { |
7c2e132c | 1382 | add_partial(n, page, tail); |
8ff12cfc CL |
1383 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1384 | } else { | |
1385 | stat(c, DEACTIVATE_FULL); | |
1386 | if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) | |
1387 | add_full(n, page); | |
1388 | } | |
81819f0f CL |
1389 | slab_unlock(page); |
1390 | } else { | |
8ff12cfc | 1391 | stat(c, DEACTIVATE_EMPTY); |
e95eed57 CL |
1392 | if (n->nr_partial < MIN_PARTIAL) { |
1393 | /* | |
672bba3a CL |
1394 | * Adding an empty slab to the partial slabs in order |
1395 | * to avoid page allocator overhead. This slab needs | |
1396 | * to come after the other slabs with objects in | |
1397 | * order to fill them up. That way the size of the | |
1398 | * partial list stays small. kmem_cache_shrink can | |
1399 | * reclaim empty slabs from the partial list. | |
e95eed57 | 1400 | */ |
7c2e132c | 1401 | add_partial(n, page, 1); |
e95eed57 CL |
1402 | slab_unlock(page); |
1403 | } else { | |
1404 | slab_unlock(page); | |
8ff12cfc | 1405 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1406 | discard_slab(s, page); |
1407 | } | |
81819f0f CL |
1408 | } |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * Remove the cpu slab | |
1413 | */ | |
dfb4f096 | 1414 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1415 | { |
dfb4f096 | 1416 | struct page *page = c->page; |
7c2e132c | 1417 | int tail = 1; |
8ff12cfc CL |
1418 | |
1419 | if (c->freelist) | |
1420 | stat(c, DEACTIVATE_REMOTE_FREES); | |
894b8788 CL |
1421 | /* |
1422 | * Merge cpu freelist into freelist. Typically we get here | |
1423 | * because both freelists are empty. So this is unlikely | |
1424 | * to occur. | |
683d0baa CL |
1425 | * |
1426 | * We need to use _is_end here because deactivate slab may | |
1427 | * be called for a debug slab. Then c->freelist may contain | |
1428 | * a dummy pointer. | |
894b8788 | 1429 | */ |
683d0baa | 1430 | while (unlikely(!is_end(c->freelist))) { |
894b8788 CL |
1431 | void **object; |
1432 | ||
7c2e132c CL |
1433 | tail = 0; /* Hot objects. Put the slab first */ |
1434 | ||
894b8788 | 1435 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1436 | object = c->freelist; |
b3fba8da | 1437 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1438 | |
1439 | /* And put onto the regular freelist */ | |
b3fba8da | 1440 | object[c->offset] = page->freelist; |
894b8788 CL |
1441 | page->freelist = object; |
1442 | page->inuse--; | |
1443 | } | |
dfb4f096 | 1444 | c->page = NULL; |
7c2e132c | 1445 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1446 | } |
1447 | ||
dfb4f096 | 1448 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1449 | { |
8ff12cfc | 1450 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1451 | slab_lock(c->page); |
1452 | deactivate_slab(s, c); | |
81819f0f CL |
1453 | } |
1454 | ||
1455 | /* | |
1456 | * Flush cpu slab. | |
1457 | * Called from IPI handler with interrupts disabled. | |
1458 | */ | |
0c710013 | 1459 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1460 | { |
dfb4f096 | 1461 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1462 | |
dfb4f096 CL |
1463 | if (likely(c && c->page)) |
1464 | flush_slab(s, c); | |
81819f0f CL |
1465 | } |
1466 | ||
1467 | static void flush_cpu_slab(void *d) | |
1468 | { | |
1469 | struct kmem_cache *s = d; | |
81819f0f | 1470 | |
dfb4f096 | 1471 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1472 | } |
1473 | ||
1474 | static void flush_all(struct kmem_cache *s) | |
1475 | { | |
1476 | #ifdef CONFIG_SMP | |
1477 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1478 | #else | |
1479 | unsigned long flags; | |
1480 | ||
1481 | local_irq_save(flags); | |
1482 | flush_cpu_slab(s); | |
1483 | local_irq_restore(flags); | |
1484 | #endif | |
1485 | } | |
1486 | ||
dfb4f096 CL |
1487 | /* |
1488 | * Check if the objects in a per cpu structure fit numa | |
1489 | * locality expectations. | |
1490 | */ | |
1491 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1492 | { | |
1493 | #ifdef CONFIG_NUMA | |
1494 | if (node != -1 && c->node != node) | |
1495 | return 0; | |
1496 | #endif | |
1497 | return 1; | |
1498 | } | |
1499 | ||
81819f0f | 1500 | /* |
894b8788 CL |
1501 | * Slow path. The lockless freelist is empty or we need to perform |
1502 | * debugging duties. | |
1503 | * | |
1504 | * Interrupts are disabled. | |
81819f0f | 1505 | * |
894b8788 CL |
1506 | * Processing is still very fast if new objects have been freed to the |
1507 | * regular freelist. In that case we simply take over the regular freelist | |
1508 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1509 | * |
894b8788 CL |
1510 | * If that is not working then we fall back to the partial lists. We take the |
1511 | * first element of the freelist as the object to allocate now and move the | |
1512 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1513 | * |
894b8788 CL |
1514 | * And if we were unable to get a new slab from the partial slab lists then |
1515 | * we need to allocate a new slab. This is slowest path since we may sleep. | |
81819f0f | 1516 | */ |
894b8788 | 1517 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1518 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1519 | { |
81819f0f | 1520 | void **object; |
dfb4f096 | 1521 | struct page *new; |
1f84260c CL |
1522 | #ifdef SLUB_FASTPATH |
1523 | unsigned long flags; | |
81819f0f | 1524 | |
1f84260c CL |
1525 | local_irq_save(flags); |
1526 | #endif | |
dfb4f096 | 1527 | if (!c->page) |
81819f0f CL |
1528 | goto new_slab; |
1529 | ||
dfb4f096 CL |
1530 | slab_lock(c->page); |
1531 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1532 | goto another_slab; |
8ff12cfc | 1533 | stat(c, ALLOC_REFILL); |
894b8788 | 1534 | load_freelist: |
dfb4f096 | 1535 | object = c->page->freelist; |
683d0baa | 1536 | if (unlikely(object == c->page->end)) |
81819f0f | 1537 | goto another_slab; |
dfb4f096 | 1538 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1539 | goto debug; |
1540 | ||
dfb4f096 | 1541 | object = c->page->freelist; |
b3fba8da | 1542 | c->freelist = object[c->offset]; |
dfb4f096 | 1543 | c->page->inuse = s->objects; |
683d0baa | 1544 | c->page->freelist = c->page->end; |
dfb4f096 | 1545 | c->node = page_to_nid(c->page); |
1f84260c | 1546 | unlock_out: |
dfb4f096 | 1547 | slab_unlock(c->page); |
8ff12cfc | 1548 | stat(c, ALLOC_SLOWPATH); |
1f84260c CL |
1549 | out: |
1550 | #ifdef SLUB_FASTPATH | |
1551 | local_irq_restore(flags); | |
1552 | #endif | |
81819f0f CL |
1553 | return object; |
1554 | ||
1555 | another_slab: | |
dfb4f096 | 1556 | deactivate_slab(s, c); |
81819f0f CL |
1557 | |
1558 | new_slab: | |
dfb4f096 CL |
1559 | new = get_partial(s, gfpflags, node); |
1560 | if (new) { | |
1561 | c->page = new; | |
8ff12cfc | 1562 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1563 | goto load_freelist; |
81819f0f CL |
1564 | } |
1565 | ||
b811c202 CL |
1566 | if (gfpflags & __GFP_WAIT) |
1567 | local_irq_enable(); | |
1568 | ||
dfb4f096 | 1569 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1570 | |
1571 | if (gfpflags & __GFP_WAIT) | |
1572 | local_irq_disable(); | |
1573 | ||
dfb4f096 CL |
1574 | if (new) { |
1575 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1576 | stat(c, ALLOC_SLAB); |
05aa3450 | 1577 | if (c->page) |
dfb4f096 | 1578 | flush_slab(s, c); |
dfb4f096 CL |
1579 | slab_lock(new); |
1580 | SetSlabFrozen(new); | |
1581 | c->page = new; | |
4b6f0750 | 1582 | goto load_freelist; |
81819f0f | 1583 | } |
1f84260c CL |
1584 | object = NULL; |
1585 | goto out; | |
81819f0f | 1586 | debug: |
dfb4f096 CL |
1587 | object = c->page->freelist; |
1588 | if (!alloc_debug_processing(s, c->page, object, addr)) | |
81819f0f | 1589 | goto another_slab; |
894b8788 | 1590 | |
dfb4f096 | 1591 | c->page->inuse++; |
b3fba8da | 1592 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1593 | c->node = -1; |
1f84260c | 1594 | goto unlock_out; |
894b8788 CL |
1595 | } |
1596 | ||
1597 | /* | |
1598 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1599 | * have the fastpath folded into their functions. So no function call | |
1600 | * overhead for requests that can be satisfied on the fastpath. | |
1601 | * | |
1602 | * The fastpath works by first checking if the lockless freelist can be used. | |
1603 | * If not then __slab_alloc is called for slow processing. | |
1604 | * | |
1605 | * Otherwise we can simply pick the next object from the lockless free list. | |
1606 | */ | |
06428780 | 1607 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce15fea8 | 1608 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1609 | { |
894b8788 | 1610 | void **object; |
dfb4f096 | 1611 | struct kmem_cache_cpu *c; |
894b8788 | 1612 | |
1f84260c CL |
1613 | /* |
1614 | * The SLUB_FASTPATH path is provisional and is currently disabled if the | |
1615 | * kernel is compiled with preemption or if the arch does not support | |
1616 | * fast cmpxchg operations. There are a couple of coming changes that will | |
1617 | * simplify matters and allow preemption. Ultimately we may end up making | |
1618 | * SLUB_FASTPATH the default. | |
1619 | * | |
1620 | * 1. The introduction of the per cpu allocator will avoid array lookups | |
1621 | * through get_cpu_slab(). A special register can be used instead. | |
1622 | * | |
1623 | * 2. The introduction of per cpu atomic operations (cpu_ops) means that | |
1624 | * we can realize the logic here entirely with per cpu atomics. The | |
1625 | * per cpu atomic ops will take care of the preemption issues. | |
1626 | */ | |
1627 | ||
1628 | #ifdef SLUB_FASTPATH | |
1629 | c = get_cpu_slab(s, raw_smp_processor_id()); | |
1630 | do { | |
1631 | object = c->freelist; | |
1632 | if (unlikely(is_end(object) || !node_match(c, node))) { | |
1633 | object = __slab_alloc(s, gfpflags, node, addr, c); | |
1634 | break; | |
1635 | } | |
8ff12cfc | 1636 | stat(c, ALLOC_FASTPATH); |
1f84260c CL |
1637 | } while (cmpxchg_local(&c->freelist, object, object[c->offset]) |
1638 | != object); | |
1639 | #else | |
1640 | unsigned long flags; | |
1641 | ||
894b8788 | 1642 | local_irq_save(flags); |
dfb4f096 | 1643 | c = get_cpu_slab(s, smp_processor_id()); |
683d0baa | 1644 | if (unlikely(is_end(c->freelist) || !node_match(c, node))) |
894b8788 | 1645 | |
dfb4f096 | 1646 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1647 | |
1648 | else { | |
dfb4f096 | 1649 | object = c->freelist; |
b3fba8da | 1650 | c->freelist = object[c->offset]; |
8ff12cfc | 1651 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1652 | } |
1653 | local_irq_restore(flags); | |
1f84260c | 1654 | #endif |
d07dbea4 CL |
1655 | |
1656 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
42a9fdbb | 1657 | memset(object, 0, c->objsize); |
d07dbea4 | 1658 | |
894b8788 | 1659 | return object; |
81819f0f CL |
1660 | } |
1661 | ||
1662 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1663 | { | |
ce15fea8 | 1664 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1665 | } |
1666 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1667 | ||
1668 | #ifdef CONFIG_NUMA | |
1669 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1670 | { | |
ce15fea8 | 1671 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1672 | } |
1673 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1674 | #endif | |
1675 | ||
1676 | /* | |
894b8788 CL |
1677 | * Slow patch handling. This may still be called frequently since objects |
1678 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1679 | * |
894b8788 CL |
1680 | * So we still attempt to reduce cache line usage. Just take the slab |
1681 | * lock and free the item. If there is no additional partial page | |
1682 | * handling required then we can return immediately. | |
81819f0f | 1683 | */ |
894b8788 | 1684 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1685 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1686 | { |
1687 | void *prior; | |
1688 | void **object = (void *)x; | |
8ff12cfc | 1689 | struct kmem_cache_cpu *c; |
81819f0f | 1690 | |
1f84260c CL |
1691 | #ifdef SLUB_FASTPATH |
1692 | unsigned long flags; | |
1693 | ||
1694 | local_irq_save(flags); | |
1695 | #endif | |
8ff12cfc CL |
1696 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1697 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1698 | slab_lock(page); |
1699 | ||
35e5d7ee | 1700 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1701 | goto debug; |
1702 | checks_ok: | |
b3fba8da | 1703 | prior = object[offset] = page->freelist; |
81819f0f CL |
1704 | page->freelist = object; |
1705 | page->inuse--; | |
1706 | ||
8ff12cfc CL |
1707 | if (unlikely(SlabFrozen(page))) { |
1708 | stat(c, FREE_FROZEN); | |
81819f0f | 1709 | goto out_unlock; |
8ff12cfc | 1710 | } |
81819f0f CL |
1711 | |
1712 | if (unlikely(!page->inuse)) | |
1713 | goto slab_empty; | |
1714 | ||
1715 | /* | |
1716 | * Objects left in the slab. If it | |
1717 | * was not on the partial list before | |
1718 | * then add it. | |
1719 | */ | |
8ff12cfc | 1720 | if (unlikely(prior == page->end)) { |
7c2e132c | 1721 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1722 | stat(c, FREE_ADD_PARTIAL); |
1723 | } | |
81819f0f CL |
1724 | |
1725 | out_unlock: | |
1726 | slab_unlock(page); | |
1f84260c CL |
1727 | #ifdef SLUB_FASTPATH |
1728 | local_irq_restore(flags); | |
1729 | #endif | |
81819f0f CL |
1730 | return; |
1731 | ||
1732 | slab_empty: | |
8ff12cfc | 1733 | if (prior != page->end) { |
81819f0f | 1734 | /* |
672bba3a | 1735 | * Slab still on the partial list. |
81819f0f CL |
1736 | */ |
1737 | remove_partial(s, page); | |
8ff12cfc CL |
1738 | stat(c, FREE_REMOVE_PARTIAL); |
1739 | } | |
81819f0f | 1740 | slab_unlock(page); |
8ff12cfc | 1741 | stat(c, FREE_SLAB); |
1f84260c CL |
1742 | #ifdef SLUB_FASTPATH |
1743 | local_irq_restore(flags); | |
1744 | #endif | |
81819f0f | 1745 | discard_slab(s, page); |
81819f0f CL |
1746 | return; |
1747 | ||
1748 | debug: | |
3ec09742 | 1749 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1750 | goto out_unlock; |
77c5e2d0 | 1751 | goto checks_ok; |
81819f0f CL |
1752 | } |
1753 | ||
894b8788 CL |
1754 | /* |
1755 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1756 | * can perform fastpath freeing without additional function calls. | |
1757 | * | |
1758 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1759 | * of this processor. This typically the case if we have just allocated | |
1760 | * the item before. | |
1761 | * | |
1762 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1763 | * with all sorts of special processing. | |
1764 | */ | |
06428780 | 1765 | static __always_inline void slab_free(struct kmem_cache *s, |
894b8788 CL |
1766 | struct page *page, void *x, void *addr) |
1767 | { | |
1768 | void **object = (void *)x; | |
dfb4f096 | 1769 | struct kmem_cache_cpu *c; |
894b8788 | 1770 | |
1f84260c CL |
1771 | #ifdef SLUB_FASTPATH |
1772 | void **freelist; | |
1773 | ||
1774 | c = get_cpu_slab(s, raw_smp_processor_id()); | |
1775 | debug_check_no_locks_freed(object, s->objsize); | |
1776 | do { | |
1777 | freelist = c->freelist; | |
1778 | barrier(); | |
1779 | /* | |
1780 | * If the compiler would reorder the retrieval of c->page to | |
1781 | * come before c->freelist then an interrupt could | |
1782 | * change the cpu slab before we retrieve c->freelist. We | |
1783 | * could be matching on a page no longer active and put the | |
1784 | * object onto the freelist of the wrong slab. | |
1785 | * | |
1786 | * On the other hand: If we already have the freelist pointer | |
1787 | * then any change of cpu_slab will cause the cmpxchg to fail | |
1788 | * since the freelist pointers are unique per slab. | |
1789 | */ | |
1790 | if (unlikely(page != c->page || c->node < 0)) { | |
1791 | __slab_free(s, page, x, addr, c->offset); | |
1792 | break; | |
1793 | } | |
1794 | object[c->offset] = freelist; | |
8ff12cfc | 1795 | stat(c, FREE_FASTPATH); |
1f84260c CL |
1796 | } while (cmpxchg_local(&c->freelist, freelist, object) != freelist); |
1797 | #else | |
1798 | unsigned long flags; | |
1799 | ||
894b8788 | 1800 | local_irq_save(flags); |
02febdf7 | 1801 | debug_check_no_locks_freed(object, s->objsize); |
dfb4f096 | 1802 | c = get_cpu_slab(s, smp_processor_id()); |
ee3c72a1 | 1803 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1804 | object[c->offset] = c->freelist; |
dfb4f096 | 1805 | c->freelist = object; |
8ff12cfc | 1806 | stat(c, FREE_FASTPATH); |
894b8788 | 1807 | } else |
b3fba8da | 1808 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1809 | |
1810 | local_irq_restore(flags); | |
1f84260c | 1811 | #endif |
894b8788 CL |
1812 | } |
1813 | ||
81819f0f CL |
1814 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1815 | { | |
77c5e2d0 | 1816 | struct page *page; |
81819f0f | 1817 | |
b49af68f | 1818 | page = virt_to_head_page(x); |
81819f0f | 1819 | |
77c5e2d0 | 1820 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1821 | } |
1822 | EXPORT_SYMBOL(kmem_cache_free); | |
1823 | ||
1824 | /* Figure out on which slab object the object resides */ | |
1825 | static struct page *get_object_page(const void *x) | |
1826 | { | |
b49af68f | 1827 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1828 | |
1829 | if (!PageSlab(page)) | |
1830 | return NULL; | |
1831 | ||
1832 | return page; | |
1833 | } | |
1834 | ||
1835 | /* | |
672bba3a CL |
1836 | * Object placement in a slab is made very easy because we always start at |
1837 | * offset 0. If we tune the size of the object to the alignment then we can | |
1838 | * get the required alignment by putting one properly sized object after | |
1839 | * another. | |
81819f0f CL |
1840 | * |
1841 | * Notice that the allocation order determines the sizes of the per cpu | |
1842 | * caches. Each processor has always one slab available for allocations. | |
1843 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1844 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1845 | * locking overhead. |
81819f0f CL |
1846 | */ |
1847 | ||
1848 | /* | |
1849 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1850 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1851 | * and increases the number of allocations possible without having to | |
1852 | * take the list_lock. | |
1853 | */ | |
1854 | static int slub_min_order; | |
1855 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1856 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1857 | ||
1858 | /* | |
1859 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1860 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1861 | */ |
1862 | static int slub_nomerge; | |
1863 | ||
81819f0f CL |
1864 | /* |
1865 | * Calculate the order of allocation given an slab object size. | |
1866 | * | |
672bba3a CL |
1867 | * The order of allocation has significant impact on performance and other |
1868 | * system components. Generally order 0 allocations should be preferred since | |
1869 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1870 | * be problematic to put into order 0 slabs because there may be too much | |
1871 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1872 | * would be wasted. | |
1873 | * | |
1874 | * In order to reach satisfactory performance we must ensure that a minimum | |
1875 | * number of objects is in one slab. Otherwise we may generate too much | |
1876 | * activity on the partial lists which requires taking the list_lock. This is | |
1877 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1878 | * |
672bba3a CL |
1879 | * slub_max_order specifies the order where we begin to stop considering the |
1880 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1881 | * we try to keep the page order as low as possible. So we accept more waste | |
1882 | * of space in favor of a small page order. | |
81819f0f | 1883 | * |
672bba3a CL |
1884 | * Higher order allocations also allow the placement of more objects in a |
1885 | * slab and thereby reduce object handling overhead. If the user has | |
1886 | * requested a higher mininum order then we start with that one instead of | |
1887 | * the smallest order which will fit the object. | |
81819f0f | 1888 | */ |
5e6d444e CL |
1889 | static inline int slab_order(int size, int min_objects, |
1890 | int max_order, int fract_leftover) | |
81819f0f CL |
1891 | { |
1892 | int order; | |
1893 | int rem; | |
6300ea75 | 1894 | int min_order = slub_min_order; |
81819f0f | 1895 | |
6300ea75 | 1896 | for (order = max(min_order, |
5e6d444e CL |
1897 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1898 | order <= max_order; order++) { | |
81819f0f | 1899 | |
5e6d444e | 1900 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1901 | |
5e6d444e | 1902 | if (slab_size < min_objects * size) |
81819f0f CL |
1903 | continue; |
1904 | ||
1905 | rem = slab_size % size; | |
1906 | ||
5e6d444e | 1907 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1908 | break; |
1909 | ||
1910 | } | |
672bba3a | 1911 | |
81819f0f CL |
1912 | return order; |
1913 | } | |
1914 | ||
5e6d444e CL |
1915 | static inline int calculate_order(int size) |
1916 | { | |
1917 | int order; | |
1918 | int min_objects; | |
1919 | int fraction; | |
1920 | ||
1921 | /* | |
1922 | * Attempt to find best configuration for a slab. This | |
1923 | * works by first attempting to generate a layout with | |
1924 | * the best configuration and backing off gradually. | |
1925 | * | |
1926 | * First we reduce the acceptable waste in a slab. Then | |
1927 | * we reduce the minimum objects required in a slab. | |
1928 | */ | |
1929 | min_objects = slub_min_objects; | |
1930 | while (min_objects > 1) { | |
1931 | fraction = 8; | |
1932 | while (fraction >= 4) { | |
1933 | order = slab_order(size, min_objects, | |
1934 | slub_max_order, fraction); | |
1935 | if (order <= slub_max_order) | |
1936 | return order; | |
1937 | fraction /= 2; | |
1938 | } | |
1939 | min_objects /= 2; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * We were unable to place multiple objects in a slab. Now | |
1944 | * lets see if we can place a single object there. | |
1945 | */ | |
1946 | order = slab_order(size, 1, slub_max_order, 1); | |
1947 | if (order <= slub_max_order) | |
1948 | return order; | |
1949 | ||
1950 | /* | |
1951 | * Doh this slab cannot be placed using slub_max_order. | |
1952 | */ | |
1953 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1954 | if (order <= MAX_ORDER) | |
1955 | return order; | |
1956 | return -ENOSYS; | |
1957 | } | |
1958 | ||
81819f0f | 1959 | /* |
672bba3a | 1960 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1961 | */ |
1962 | static unsigned long calculate_alignment(unsigned long flags, | |
1963 | unsigned long align, unsigned long size) | |
1964 | { | |
1965 | /* | |
1966 | * If the user wants hardware cache aligned objects then | |
1967 | * follow that suggestion if the object is sufficiently | |
1968 | * large. | |
1969 | * | |
1970 | * The hardware cache alignment cannot override the | |
1971 | * specified alignment though. If that is greater | |
1972 | * then use it. | |
1973 | */ | |
5af60839 | 1974 | if ((flags & SLAB_HWCACHE_ALIGN) && |
65c02d4c CL |
1975 | size > cache_line_size() / 2) |
1976 | return max_t(unsigned long, align, cache_line_size()); | |
81819f0f CL |
1977 | |
1978 | if (align < ARCH_SLAB_MINALIGN) | |
1979 | return ARCH_SLAB_MINALIGN; | |
1980 | ||
1981 | return ALIGN(align, sizeof(void *)); | |
1982 | } | |
1983 | ||
dfb4f096 CL |
1984 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1985 | struct kmem_cache_cpu *c) | |
1986 | { | |
1987 | c->page = NULL; | |
683d0baa | 1988 | c->freelist = (void *)PAGE_MAPPING_ANON; |
dfb4f096 | 1989 | c->node = 0; |
42a9fdbb CL |
1990 | c->offset = s->offset / sizeof(void *); |
1991 | c->objsize = s->objsize; | |
dfb4f096 CL |
1992 | } |
1993 | ||
81819f0f CL |
1994 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
1995 | { | |
1996 | n->nr_partial = 0; | |
1997 | atomic_long_set(&n->nr_slabs, 0); | |
1998 | spin_lock_init(&n->list_lock); | |
1999 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2000 | #ifdef CONFIG_SLUB_DEBUG |
643b1138 | 2001 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2002 | #endif |
81819f0f CL |
2003 | } |
2004 | ||
4c93c355 CL |
2005 | #ifdef CONFIG_SMP |
2006 | /* | |
2007 | * Per cpu array for per cpu structures. | |
2008 | * | |
2009 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
2010 | * close together meaning that it becomes possible that multiple per cpu | |
2011 | * structures are contained in one cacheline. This may be particularly | |
2012 | * beneficial for the kmalloc caches. | |
2013 | * | |
2014 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
2015 | * likely able to get per cpu structures for all caches from the array defined | |
2016 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2017 | * | |
2018 | * If the per cpu array is exhausted then fall back to kmalloc | |
2019 | * of individual cachelines. No sharing is possible then. | |
2020 | */ | |
2021 | #define NR_KMEM_CACHE_CPU 100 | |
2022 | ||
2023 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2024 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2025 | ||
2026 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
2027 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
2028 | ||
2029 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2030 | int cpu, gfp_t flags) | |
2031 | { | |
2032 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2033 | ||
2034 | if (c) | |
2035 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2036 | (void *)c->freelist; | |
2037 | else { | |
2038 | /* Table overflow: So allocate ourselves */ | |
2039 | c = kmalloc_node( | |
2040 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2041 | flags, cpu_to_node(cpu)); | |
2042 | if (!c) | |
2043 | return NULL; | |
2044 | } | |
2045 | ||
2046 | init_kmem_cache_cpu(s, c); | |
2047 | return c; | |
2048 | } | |
2049 | ||
2050 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2051 | { | |
2052 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
2053 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
2054 | kfree(c); | |
2055 | return; | |
2056 | } | |
2057 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2058 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2059 | } | |
2060 | ||
2061 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2062 | { | |
2063 | int cpu; | |
2064 | ||
2065 | for_each_online_cpu(cpu) { | |
2066 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2067 | ||
2068 | if (c) { | |
2069 | s->cpu_slab[cpu] = NULL; | |
2070 | free_kmem_cache_cpu(c, cpu); | |
2071 | } | |
2072 | } | |
2073 | } | |
2074 | ||
2075 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2076 | { | |
2077 | int cpu; | |
2078 | ||
2079 | for_each_online_cpu(cpu) { | |
2080 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2081 | ||
2082 | if (c) | |
2083 | continue; | |
2084 | ||
2085 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2086 | if (!c) { | |
2087 | free_kmem_cache_cpus(s); | |
2088 | return 0; | |
2089 | } | |
2090 | s->cpu_slab[cpu] = c; | |
2091 | } | |
2092 | return 1; | |
2093 | } | |
2094 | ||
2095 | /* | |
2096 | * Initialize the per cpu array. | |
2097 | */ | |
2098 | static void init_alloc_cpu_cpu(int cpu) | |
2099 | { | |
2100 | int i; | |
2101 | ||
2102 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
2103 | return; | |
2104 | ||
2105 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2106 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2107 | ||
2108 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
2109 | } | |
2110 | ||
2111 | static void __init init_alloc_cpu(void) | |
2112 | { | |
2113 | int cpu; | |
2114 | ||
2115 | for_each_online_cpu(cpu) | |
2116 | init_alloc_cpu_cpu(cpu); | |
2117 | } | |
2118 | ||
2119 | #else | |
2120 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2121 | static inline void init_alloc_cpu(void) {} | |
2122 | ||
2123 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2124 | { | |
2125 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2126 | return 1; | |
2127 | } | |
2128 | #endif | |
2129 | ||
81819f0f CL |
2130 | #ifdef CONFIG_NUMA |
2131 | /* | |
2132 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2133 | * slab on the node for this slabcache. There are no concurrent accesses | |
2134 | * possible. | |
2135 | * | |
2136 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2137 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2138 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2139 | */ |
1cd7daa5 AB |
2140 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2141 | int node) | |
81819f0f CL |
2142 | { |
2143 | struct page *page; | |
2144 | struct kmem_cache_node *n; | |
ba84c73c | 2145 | unsigned long flags; |
81819f0f CL |
2146 | |
2147 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2148 | ||
a2f92ee7 | 2149 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2150 | |
2151 | BUG_ON(!page); | |
a2f92ee7 CL |
2152 | if (page_to_nid(page) != node) { |
2153 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2154 | "node %d\n", node); | |
2155 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2156 | "in order to be able to continue\n"); | |
2157 | } | |
2158 | ||
81819f0f CL |
2159 | n = page->freelist; |
2160 | BUG_ON(!n); | |
2161 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2162 | page->inuse++; | |
2163 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2164 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2165 | init_object(kmalloc_caches, n, 1); |
2166 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2167 | #endif |
81819f0f CL |
2168 | init_kmem_cache_node(n); |
2169 | atomic_long_inc(&n->nr_slabs); | |
ba84c73c | 2170 | /* |
2171 | * lockdep requires consistent irq usage for each lock | |
2172 | * so even though there cannot be a race this early in | |
2173 | * the boot sequence, we still disable irqs. | |
2174 | */ | |
2175 | local_irq_save(flags); | |
7c2e132c | 2176 | add_partial(n, page, 0); |
ba84c73c | 2177 | local_irq_restore(flags); |
81819f0f CL |
2178 | return n; |
2179 | } | |
2180 | ||
2181 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2182 | { | |
2183 | int node; | |
2184 | ||
f64dc58c | 2185 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2186 | struct kmem_cache_node *n = s->node[node]; |
2187 | if (n && n != &s->local_node) | |
2188 | kmem_cache_free(kmalloc_caches, n); | |
2189 | s->node[node] = NULL; | |
2190 | } | |
2191 | } | |
2192 | ||
2193 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2194 | { | |
2195 | int node; | |
2196 | int local_node; | |
2197 | ||
2198 | if (slab_state >= UP) | |
2199 | local_node = page_to_nid(virt_to_page(s)); | |
2200 | else | |
2201 | local_node = 0; | |
2202 | ||
f64dc58c | 2203 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2204 | struct kmem_cache_node *n; |
2205 | ||
2206 | if (local_node == node) | |
2207 | n = &s->local_node; | |
2208 | else { | |
2209 | if (slab_state == DOWN) { | |
2210 | n = early_kmem_cache_node_alloc(gfpflags, | |
2211 | node); | |
2212 | continue; | |
2213 | } | |
2214 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2215 | gfpflags, node); | |
2216 | ||
2217 | if (!n) { | |
2218 | free_kmem_cache_nodes(s); | |
2219 | return 0; | |
2220 | } | |
2221 | ||
2222 | } | |
2223 | s->node[node] = n; | |
2224 | init_kmem_cache_node(n); | |
2225 | } | |
2226 | return 1; | |
2227 | } | |
2228 | #else | |
2229 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2230 | { | |
2231 | } | |
2232 | ||
2233 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2234 | { | |
2235 | init_kmem_cache_node(&s->local_node); | |
2236 | return 1; | |
2237 | } | |
2238 | #endif | |
2239 | ||
2240 | /* | |
2241 | * calculate_sizes() determines the order and the distribution of data within | |
2242 | * a slab object. | |
2243 | */ | |
2244 | static int calculate_sizes(struct kmem_cache *s) | |
2245 | { | |
2246 | unsigned long flags = s->flags; | |
2247 | unsigned long size = s->objsize; | |
2248 | unsigned long align = s->align; | |
2249 | ||
2250 | /* | |
2251 | * Determine if we can poison the object itself. If the user of | |
2252 | * the slab may touch the object after free or before allocation | |
2253 | * then we should never poison the object itself. | |
2254 | */ | |
2255 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2256 | !s->ctor) |
81819f0f CL |
2257 | s->flags |= __OBJECT_POISON; |
2258 | else | |
2259 | s->flags &= ~__OBJECT_POISON; | |
2260 | ||
2261 | /* | |
2262 | * Round up object size to the next word boundary. We can only | |
2263 | * place the free pointer at word boundaries and this determines | |
2264 | * the possible location of the free pointer. | |
2265 | */ | |
2266 | size = ALIGN(size, sizeof(void *)); | |
2267 | ||
41ecc55b | 2268 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f | 2269 | /* |
672bba3a | 2270 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2271 | * end of the object and the free pointer. If not then add an |
672bba3a | 2272 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2273 | */ |
2274 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2275 | size += sizeof(void *); | |
41ecc55b | 2276 | #endif |
81819f0f CL |
2277 | |
2278 | /* | |
672bba3a CL |
2279 | * With that we have determined the number of bytes in actual use |
2280 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2281 | */ |
2282 | s->inuse = size; | |
2283 | ||
2284 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2285 | s->ctor)) { |
81819f0f CL |
2286 | /* |
2287 | * Relocate free pointer after the object if it is not | |
2288 | * permitted to overwrite the first word of the object on | |
2289 | * kmem_cache_free. | |
2290 | * | |
2291 | * This is the case if we do RCU, have a constructor or | |
2292 | * destructor or are poisoning the objects. | |
2293 | */ | |
2294 | s->offset = size; | |
2295 | size += sizeof(void *); | |
2296 | } | |
2297 | ||
c12b3c62 | 2298 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2299 | if (flags & SLAB_STORE_USER) |
2300 | /* | |
2301 | * Need to store information about allocs and frees after | |
2302 | * the object. | |
2303 | */ | |
2304 | size += 2 * sizeof(struct track); | |
2305 | ||
be7b3fbc | 2306 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2307 | /* |
2308 | * Add some empty padding so that we can catch | |
2309 | * overwrites from earlier objects rather than let | |
2310 | * tracking information or the free pointer be | |
2311 | * corrupted if an user writes before the start | |
2312 | * of the object. | |
2313 | */ | |
2314 | size += sizeof(void *); | |
41ecc55b | 2315 | #endif |
672bba3a | 2316 | |
81819f0f CL |
2317 | /* |
2318 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2319 | * user specified and the dynamic determination of cache line size |
2320 | * on bootup. | |
81819f0f CL |
2321 | */ |
2322 | align = calculate_alignment(flags, align, s->objsize); | |
2323 | ||
2324 | /* | |
2325 | * SLUB stores one object immediately after another beginning from | |
2326 | * offset 0. In order to align the objects we have to simply size | |
2327 | * each object to conform to the alignment. | |
2328 | */ | |
2329 | size = ALIGN(size, align); | |
2330 | s->size = size; | |
2331 | ||
2332 | s->order = calculate_order(size); | |
2333 | if (s->order < 0) | |
2334 | return 0; | |
2335 | ||
2336 | /* | |
2337 | * Determine the number of objects per slab | |
2338 | */ | |
2339 | s->objects = (PAGE_SIZE << s->order) / size; | |
2340 | ||
b3fba8da | 2341 | return !!s->objects; |
81819f0f CL |
2342 | |
2343 | } | |
2344 | ||
81819f0f CL |
2345 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2346 | const char *name, size_t size, | |
2347 | size_t align, unsigned long flags, | |
4ba9b9d0 | 2348 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
2349 | { |
2350 | memset(s, 0, kmem_size); | |
2351 | s->name = name; | |
2352 | s->ctor = ctor; | |
81819f0f | 2353 | s->objsize = size; |
81819f0f | 2354 | s->align = align; |
ba0268a8 | 2355 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f CL |
2356 | |
2357 | if (!calculate_sizes(s)) | |
2358 | goto error; | |
2359 | ||
2360 | s->refcount = 1; | |
2361 | #ifdef CONFIG_NUMA | |
9824601e | 2362 | s->remote_node_defrag_ratio = 100; |
81819f0f | 2363 | #endif |
dfb4f096 CL |
2364 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2365 | goto error; | |
81819f0f | 2366 | |
dfb4f096 | 2367 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2368 | return 1; |
4c93c355 | 2369 | free_kmem_cache_nodes(s); |
81819f0f CL |
2370 | error: |
2371 | if (flags & SLAB_PANIC) | |
2372 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2373 | "order=%u offset=%u flags=%lx\n", | |
2374 | s->name, (unsigned long)size, s->size, s->order, | |
2375 | s->offset, flags); | |
2376 | return 0; | |
2377 | } | |
81819f0f CL |
2378 | |
2379 | /* | |
2380 | * Check if a given pointer is valid | |
2381 | */ | |
2382 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2383 | { | |
06428780 | 2384 | struct page *page; |
81819f0f CL |
2385 | |
2386 | page = get_object_page(object); | |
2387 | ||
2388 | if (!page || s != page->slab) | |
2389 | /* No slab or wrong slab */ | |
2390 | return 0; | |
2391 | ||
abcd08a6 | 2392 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2393 | return 0; |
2394 | ||
2395 | /* | |
2396 | * We could also check if the object is on the slabs freelist. | |
2397 | * But this would be too expensive and it seems that the main | |
2398 | * purpose of kmem_ptr_valid is to check if the object belongs | |
2399 | * to a certain slab. | |
2400 | */ | |
2401 | return 1; | |
2402 | } | |
2403 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2404 | ||
2405 | /* | |
2406 | * Determine the size of a slab object | |
2407 | */ | |
2408 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2409 | { | |
2410 | return s->objsize; | |
2411 | } | |
2412 | EXPORT_SYMBOL(kmem_cache_size); | |
2413 | ||
2414 | const char *kmem_cache_name(struct kmem_cache *s) | |
2415 | { | |
2416 | return s->name; | |
2417 | } | |
2418 | EXPORT_SYMBOL(kmem_cache_name); | |
2419 | ||
2420 | /* | |
672bba3a CL |
2421 | * Attempt to free all slabs on a node. Return the number of slabs we |
2422 | * were unable to free. | |
81819f0f CL |
2423 | */ |
2424 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
2425 | struct list_head *list) | |
2426 | { | |
2427 | int slabs_inuse = 0; | |
2428 | unsigned long flags; | |
2429 | struct page *page, *h; | |
2430 | ||
2431 | spin_lock_irqsave(&n->list_lock, flags); | |
2432 | list_for_each_entry_safe(page, h, list, lru) | |
2433 | if (!page->inuse) { | |
2434 | list_del(&page->lru); | |
2435 | discard_slab(s, page); | |
2436 | } else | |
2437 | slabs_inuse++; | |
2438 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2439 | return slabs_inuse; | |
2440 | } | |
2441 | ||
2442 | /* | |
672bba3a | 2443 | * Release all resources used by a slab cache. |
81819f0f | 2444 | */ |
0c710013 | 2445 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2446 | { |
2447 | int node; | |
2448 | ||
2449 | flush_all(s); | |
2450 | ||
2451 | /* Attempt to free all objects */ | |
4c93c355 | 2452 | free_kmem_cache_cpus(s); |
f64dc58c | 2453 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2454 | struct kmem_cache_node *n = get_node(s, node); |
2455 | ||
2086d26a | 2456 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
2457 | if (atomic_long_read(&n->nr_slabs)) |
2458 | return 1; | |
2459 | } | |
2460 | free_kmem_cache_nodes(s); | |
2461 | return 0; | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * Close a cache and release the kmem_cache structure | |
2466 | * (must be used for caches created using kmem_cache_create) | |
2467 | */ | |
2468 | void kmem_cache_destroy(struct kmem_cache *s) | |
2469 | { | |
2470 | down_write(&slub_lock); | |
2471 | s->refcount--; | |
2472 | if (!s->refcount) { | |
2473 | list_del(&s->list); | |
a0e1d1be | 2474 | up_write(&slub_lock); |
81819f0f CL |
2475 | if (kmem_cache_close(s)) |
2476 | WARN_ON(1); | |
2477 | sysfs_slab_remove(s); | |
a0e1d1be CL |
2478 | } else |
2479 | up_write(&slub_lock); | |
81819f0f CL |
2480 | } |
2481 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2482 | ||
2483 | /******************************************************************** | |
2484 | * Kmalloc subsystem | |
2485 | *******************************************************************/ | |
2486 | ||
aadb4bc4 | 2487 | struct kmem_cache kmalloc_caches[PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2488 | EXPORT_SYMBOL(kmalloc_caches); |
2489 | ||
2490 | #ifdef CONFIG_ZONE_DMA | |
aadb4bc4 | 2491 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT]; |
81819f0f CL |
2492 | #endif |
2493 | ||
2494 | static int __init setup_slub_min_order(char *str) | |
2495 | { | |
06428780 | 2496 | get_option(&str, &slub_min_order); |
81819f0f CL |
2497 | |
2498 | return 1; | |
2499 | } | |
2500 | ||
2501 | __setup("slub_min_order=", setup_slub_min_order); | |
2502 | ||
2503 | static int __init setup_slub_max_order(char *str) | |
2504 | { | |
06428780 | 2505 | get_option(&str, &slub_max_order); |
81819f0f CL |
2506 | |
2507 | return 1; | |
2508 | } | |
2509 | ||
2510 | __setup("slub_max_order=", setup_slub_max_order); | |
2511 | ||
2512 | static int __init setup_slub_min_objects(char *str) | |
2513 | { | |
06428780 | 2514 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2515 | |
2516 | return 1; | |
2517 | } | |
2518 | ||
2519 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2520 | ||
2521 | static int __init setup_slub_nomerge(char *str) | |
2522 | { | |
2523 | slub_nomerge = 1; | |
2524 | return 1; | |
2525 | } | |
2526 | ||
2527 | __setup("slub_nomerge", setup_slub_nomerge); | |
2528 | ||
81819f0f CL |
2529 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2530 | const char *name, int size, gfp_t gfp_flags) | |
2531 | { | |
2532 | unsigned int flags = 0; | |
2533 | ||
2534 | if (gfp_flags & SLUB_DMA) | |
2535 | flags = SLAB_CACHE_DMA; | |
2536 | ||
2537 | down_write(&slub_lock); | |
2538 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
c59def9f | 2539 | flags, NULL)) |
81819f0f CL |
2540 | goto panic; |
2541 | ||
2542 | list_add(&s->list, &slab_caches); | |
2543 | up_write(&slub_lock); | |
2544 | if (sysfs_slab_add(s)) | |
2545 | goto panic; | |
2546 | return s; | |
2547 | ||
2548 | panic: | |
2549 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2550 | } | |
2551 | ||
2e443fd0 | 2552 | #ifdef CONFIG_ZONE_DMA |
1ceef402 CL |
2553 | |
2554 | static void sysfs_add_func(struct work_struct *w) | |
2555 | { | |
2556 | struct kmem_cache *s; | |
2557 | ||
2558 | down_write(&slub_lock); | |
2559 | list_for_each_entry(s, &slab_caches, list) { | |
2560 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2561 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2562 | sysfs_slab_add(s); | |
2563 | } | |
2564 | } | |
2565 | up_write(&slub_lock); | |
2566 | } | |
2567 | ||
2568 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2569 | ||
2e443fd0 CL |
2570 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2571 | { | |
2572 | struct kmem_cache *s; | |
2e443fd0 CL |
2573 | char *text; |
2574 | size_t realsize; | |
2575 | ||
2576 | s = kmalloc_caches_dma[index]; | |
2577 | if (s) | |
2578 | return s; | |
2579 | ||
2580 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2581 | if (flags & __GFP_WAIT) |
2582 | down_write(&slub_lock); | |
2583 | else { | |
2584 | if (!down_write_trylock(&slub_lock)) | |
2585 | goto out; | |
2586 | } | |
2587 | ||
2588 | if (kmalloc_caches_dma[index]) | |
2589 | goto unlock_out; | |
2e443fd0 | 2590 | |
7b55f620 | 2591 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2592 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2593 | (unsigned int)realsize); | |
1ceef402 CL |
2594 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2595 | ||
2596 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2597 | realsize, ARCH_KMALLOC_MINALIGN, | |
2598 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2599 | kfree(s); | |
2600 | kfree(text); | |
2601 | goto unlock_out; | |
dfce8648 | 2602 | } |
1ceef402 CL |
2603 | |
2604 | list_add(&s->list, &slab_caches); | |
2605 | kmalloc_caches_dma[index] = s; | |
2606 | ||
2607 | schedule_work(&sysfs_add_work); | |
2608 | ||
2609 | unlock_out: | |
dfce8648 | 2610 | up_write(&slub_lock); |
1ceef402 | 2611 | out: |
dfce8648 | 2612 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2613 | } |
2614 | #endif | |
2615 | ||
f1b26339 CL |
2616 | /* |
2617 | * Conversion table for small slabs sizes / 8 to the index in the | |
2618 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2619 | * of two cache sizes there. The size of larger slabs can be determined using | |
2620 | * fls. | |
2621 | */ | |
2622 | static s8 size_index[24] = { | |
2623 | 3, /* 8 */ | |
2624 | 4, /* 16 */ | |
2625 | 5, /* 24 */ | |
2626 | 5, /* 32 */ | |
2627 | 6, /* 40 */ | |
2628 | 6, /* 48 */ | |
2629 | 6, /* 56 */ | |
2630 | 6, /* 64 */ | |
2631 | 1, /* 72 */ | |
2632 | 1, /* 80 */ | |
2633 | 1, /* 88 */ | |
2634 | 1, /* 96 */ | |
2635 | 7, /* 104 */ | |
2636 | 7, /* 112 */ | |
2637 | 7, /* 120 */ | |
2638 | 7, /* 128 */ | |
2639 | 2, /* 136 */ | |
2640 | 2, /* 144 */ | |
2641 | 2, /* 152 */ | |
2642 | 2, /* 160 */ | |
2643 | 2, /* 168 */ | |
2644 | 2, /* 176 */ | |
2645 | 2, /* 184 */ | |
2646 | 2 /* 192 */ | |
2647 | }; | |
2648 | ||
81819f0f CL |
2649 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2650 | { | |
f1b26339 | 2651 | int index; |
81819f0f | 2652 | |
f1b26339 CL |
2653 | if (size <= 192) { |
2654 | if (!size) | |
2655 | return ZERO_SIZE_PTR; | |
81819f0f | 2656 | |
f1b26339 | 2657 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2658 | } else |
f1b26339 | 2659 | index = fls(size - 1); |
81819f0f CL |
2660 | |
2661 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2662 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2663 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2664 | |
81819f0f CL |
2665 | #endif |
2666 | return &kmalloc_caches[index]; | |
2667 | } | |
2668 | ||
2669 | void *__kmalloc(size_t size, gfp_t flags) | |
2670 | { | |
aadb4bc4 | 2671 | struct kmem_cache *s; |
81819f0f | 2672 | |
aadb4bc4 | 2673 | if (unlikely(size > PAGE_SIZE / 2)) |
eada35ef | 2674 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2675 | |
2676 | s = get_slab(size, flags); | |
2677 | ||
2678 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2679 | return s; |
2680 | ||
ce15fea8 | 2681 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2682 | } |
2683 | EXPORT_SYMBOL(__kmalloc); | |
2684 | ||
2685 | #ifdef CONFIG_NUMA | |
2686 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2687 | { | |
aadb4bc4 | 2688 | struct kmem_cache *s; |
81819f0f | 2689 | |
aadb4bc4 | 2690 | if (unlikely(size > PAGE_SIZE / 2)) |
eada35ef | 2691 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2692 | |
2693 | s = get_slab(size, flags); | |
2694 | ||
2695 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2696 | return s; |
2697 | ||
ce15fea8 | 2698 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2699 | } |
2700 | EXPORT_SYMBOL(__kmalloc_node); | |
2701 | #endif | |
2702 | ||
2703 | size_t ksize(const void *object) | |
2704 | { | |
272c1d21 | 2705 | struct page *page; |
81819f0f CL |
2706 | struct kmem_cache *s; |
2707 | ||
ef8b4520 CL |
2708 | BUG_ON(!object); |
2709 | if (unlikely(object == ZERO_SIZE_PTR)) | |
272c1d21 CL |
2710 | return 0; |
2711 | ||
294a80a8 | 2712 | page = virt_to_head_page(object); |
81819f0f | 2713 | BUG_ON(!page); |
294a80a8 VN |
2714 | |
2715 | if (unlikely(!PageSlab(page))) | |
2716 | return PAGE_SIZE << compound_order(page); | |
2717 | ||
81819f0f CL |
2718 | s = page->slab; |
2719 | BUG_ON(!s); | |
2720 | ||
2721 | /* | |
2722 | * Debugging requires use of the padding between object | |
2723 | * and whatever may come after it. | |
2724 | */ | |
2725 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2726 | return s->objsize; | |
2727 | ||
2728 | /* | |
2729 | * If we have the need to store the freelist pointer | |
2730 | * back there or track user information then we can | |
2731 | * only use the space before that information. | |
2732 | */ | |
2733 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2734 | return s->inuse; | |
2735 | ||
2736 | /* | |
2737 | * Else we can use all the padding etc for the allocation | |
2738 | */ | |
2739 | return s->size; | |
2740 | } | |
2741 | EXPORT_SYMBOL(ksize); | |
2742 | ||
2743 | void kfree(const void *x) | |
2744 | { | |
81819f0f | 2745 | struct page *page; |
5bb983b0 | 2746 | void *object = (void *)x; |
81819f0f | 2747 | |
2408c550 | 2748 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2749 | return; |
2750 | ||
b49af68f | 2751 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2752 | if (unlikely(!PageSlab(page))) { |
2753 | put_page(page); | |
2754 | return; | |
2755 | } | |
5bb983b0 | 2756 | slab_free(page->slab, page, object, __builtin_return_address(0)); |
81819f0f CL |
2757 | } |
2758 | EXPORT_SYMBOL(kfree); | |
2759 | ||
f61396ae CL |
2760 | static unsigned long count_partial(struct kmem_cache_node *n) |
2761 | { | |
2762 | unsigned long flags; | |
2763 | unsigned long x = 0; | |
2764 | struct page *page; | |
2765 | ||
2766 | spin_lock_irqsave(&n->list_lock, flags); | |
2767 | list_for_each_entry(page, &n->partial, lru) | |
2768 | x += page->inuse; | |
2769 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2770 | return x; | |
2771 | } | |
2772 | ||
2086d26a | 2773 | /* |
672bba3a CL |
2774 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2775 | * the remaining slabs by the number of items in use. The slabs with the | |
2776 | * most items in use come first. New allocations will then fill those up | |
2777 | * and thus they can be removed from the partial lists. | |
2778 | * | |
2779 | * The slabs with the least items are placed last. This results in them | |
2780 | * being allocated from last increasing the chance that the last objects | |
2781 | * are freed in them. | |
2086d26a CL |
2782 | */ |
2783 | int kmem_cache_shrink(struct kmem_cache *s) | |
2784 | { | |
2785 | int node; | |
2786 | int i; | |
2787 | struct kmem_cache_node *n; | |
2788 | struct page *page; | |
2789 | struct page *t; | |
2790 | struct list_head *slabs_by_inuse = | |
2791 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2792 | unsigned long flags; | |
2793 | ||
2794 | if (!slabs_by_inuse) | |
2795 | return -ENOMEM; | |
2796 | ||
2797 | flush_all(s); | |
f64dc58c | 2798 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2799 | n = get_node(s, node); |
2800 | ||
2801 | if (!n->nr_partial) | |
2802 | continue; | |
2803 | ||
2804 | for (i = 0; i < s->objects; i++) | |
2805 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2806 | ||
2807 | spin_lock_irqsave(&n->list_lock, flags); | |
2808 | ||
2809 | /* | |
672bba3a | 2810 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2811 | * |
672bba3a CL |
2812 | * Note that concurrent frees may occur while we hold the |
2813 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2814 | */ |
2815 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2816 | if (!page->inuse && slab_trylock(page)) { | |
2817 | /* | |
2818 | * Must hold slab lock here because slab_free | |
2819 | * may have freed the last object and be | |
2820 | * waiting to release the slab. | |
2821 | */ | |
2822 | list_del(&page->lru); | |
2823 | n->nr_partial--; | |
2824 | slab_unlock(page); | |
2825 | discard_slab(s, page); | |
2826 | } else { | |
fcda3d89 CL |
2827 | list_move(&page->lru, |
2828 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2829 | } |
2830 | } | |
2831 | ||
2086d26a | 2832 | /* |
672bba3a CL |
2833 | * Rebuild the partial list with the slabs filled up most |
2834 | * first and the least used slabs at the end. | |
2086d26a CL |
2835 | */ |
2836 | for (i = s->objects - 1; i >= 0; i--) | |
2837 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2838 | ||
2086d26a CL |
2839 | spin_unlock_irqrestore(&n->list_lock, flags); |
2840 | } | |
2841 | ||
2842 | kfree(slabs_by_inuse); | |
2843 | return 0; | |
2844 | } | |
2845 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2846 | ||
b9049e23 YG |
2847 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2848 | static int slab_mem_going_offline_callback(void *arg) | |
2849 | { | |
2850 | struct kmem_cache *s; | |
2851 | ||
2852 | down_read(&slub_lock); | |
2853 | list_for_each_entry(s, &slab_caches, list) | |
2854 | kmem_cache_shrink(s); | |
2855 | up_read(&slub_lock); | |
2856 | ||
2857 | return 0; | |
2858 | } | |
2859 | ||
2860 | static void slab_mem_offline_callback(void *arg) | |
2861 | { | |
2862 | struct kmem_cache_node *n; | |
2863 | struct kmem_cache *s; | |
2864 | struct memory_notify *marg = arg; | |
2865 | int offline_node; | |
2866 | ||
2867 | offline_node = marg->status_change_nid; | |
2868 | ||
2869 | /* | |
2870 | * If the node still has available memory. we need kmem_cache_node | |
2871 | * for it yet. | |
2872 | */ | |
2873 | if (offline_node < 0) | |
2874 | return; | |
2875 | ||
2876 | down_read(&slub_lock); | |
2877 | list_for_each_entry(s, &slab_caches, list) { | |
2878 | n = get_node(s, offline_node); | |
2879 | if (n) { | |
2880 | /* | |
2881 | * if n->nr_slabs > 0, slabs still exist on the node | |
2882 | * that is going down. We were unable to free them, | |
2883 | * and offline_pages() function shoudn't call this | |
2884 | * callback. So, we must fail. | |
2885 | */ | |
27bb628a | 2886 | BUG_ON(atomic_long_read(&n->nr_slabs)); |
b9049e23 YG |
2887 | |
2888 | s->node[offline_node] = NULL; | |
2889 | kmem_cache_free(kmalloc_caches, n); | |
2890 | } | |
2891 | } | |
2892 | up_read(&slub_lock); | |
2893 | } | |
2894 | ||
2895 | static int slab_mem_going_online_callback(void *arg) | |
2896 | { | |
2897 | struct kmem_cache_node *n; | |
2898 | struct kmem_cache *s; | |
2899 | struct memory_notify *marg = arg; | |
2900 | int nid = marg->status_change_nid; | |
2901 | int ret = 0; | |
2902 | ||
2903 | /* | |
2904 | * If the node's memory is already available, then kmem_cache_node is | |
2905 | * already created. Nothing to do. | |
2906 | */ | |
2907 | if (nid < 0) | |
2908 | return 0; | |
2909 | ||
2910 | /* | |
2911 | * We are bringing a node online. No memory is availabe yet. We must | |
2912 | * allocate a kmem_cache_node structure in order to bring the node | |
2913 | * online. | |
2914 | */ | |
2915 | down_read(&slub_lock); | |
2916 | list_for_each_entry(s, &slab_caches, list) { | |
2917 | /* | |
2918 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2919 | * since memory is not yet available from the node that | |
2920 | * is brought up. | |
2921 | */ | |
2922 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2923 | if (!n) { | |
2924 | ret = -ENOMEM; | |
2925 | goto out; | |
2926 | } | |
2927 | init_kmem_cache_node(n); | |
2928 | s->node[nid] = n; | |
2929 | } | |
2930 | out: | |
2931 | up_read(&slub_lock); | |
2932 | return ret; | |
2933 | } | |
2934 | ||
2935 | static int slab_memory_callback(struct notifier_block *self, | |
2936 | unsigned long action, void *arg) | |
2937 | { | |
2938 | int ret = 0; | |
2939 | ||
2940 | switch (action) { | |
2941 | case MEM_GOING_ONLINE: | |
2942 | ret = slab_mem_going_online_callback(arg); | |
2943 | break; | |
2944 | case MEM_GOING_OFFLINE: | |
2945 | ret = slab_mem_going_offline_callback(arg); | |
2946 | break; | |
2947 | case MEM_OFFLINE: | |
2948 | case MEM_CANCEL_ONLINE: | |
2949 | slab_mem_offline_callback(arg); | |
2950 | break; | |
2951 | case MEM_ONLINE: | |
2952 | case MEM_CANCEL_OFFLINE: | |
2953 | break; | |
2954 | } | |
2955 | ||
2956 | ret = notifier_from_errno(ret); | |
2957 | return ret; | |
2958 | } | |
2959 | ||
2960 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2961 | ||
81819f0f CL |
2962 | /******************************************************************** |
2963 | * Basic setup of slabs | |
2964 | *******************************************************************/ | |
2965 | ||
2966 | void __init kmem_cache_init(void) | |
2967 | { | |
2968 | int i; | |
4b356be0 | 2969 | int caches = 0; |
81819f0f | 2970 | |
4c93c355 CL |
2971 | init_alloc_cpu(); |
2972 | ||
81819f0f CL |
2973 | #ifdef CONFIG_NUMA |
2974 | /* | |
2975 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2976 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2977 | * kmem_cache_open for slab_state == DOWN. |
2978 | */ | |
2979 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2980 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2981 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 2982 | caches++; |
b9049e23 YG |
2983 | |
2984 | hotplug_memory_notifier(slab_memory_callback, 1); | |
81819f0f CL |
2985 | #endif |
2986 | ||
2987 | /* Able to allocate the per node structures */ | |
2988 | slab_state = PARTIAL; | |
2989 | ||
2990 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
2991 | if (KMALLOC_MIN_SIZE <= 64) { |
2992 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 2993 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
2994 | caches++; |
2995 | } | |
2996 | if (KMALLOC_MIN_SIZE <= 128) { | |
2997 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 2998 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
2999 | caches++; |
3000 | } | |
81819f0f | 3001 | |
aadb4bc4 | 3002 | for (i = KMALLOC_SHIFT_LOW; i < PAGE_SHIFT; i++) { |
81819f0f CL |
3003 | create_kmalloc_cache(&kmalloc_caches[i], |
3004 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
3005 | caches++; |
3006 | } | |
81819f0f | 3007 | |
f1b26339 CL |
3008 | |
3009 | /* | |
3010 | * Patch up the size_index table if we have strange large alignment | |
3011 | * requirements for the kmalloc array. This is only the case for | |
3012 | * mips it seems. The standard arches will not generate any code here. | |
3013 | * | |
3014 | * Largest permitted alignment is 256 bytes due to the way we | |
3015 | * handle the index determination for the smaller caches. | |
3016 | * | |
3017 | * Make sure that nothing crazy happens if someone starts tinkering | |
3018 | * around with ARCH_KMALLOC_MINALIGN | |
3019 | */ | |
3020 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3021 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3022 | ||
12ad6843 | 3023 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3024 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3025 | ||
81819f0f CL |
3026 | slab_state = UP; |
3027 | ||
3028 | /* Provide the correct kmalloc names now that the caches are up */ | |
aadb4bc4 | 3029 | for (i = KMALLOC_SHIFT_LOW; i < PAGE_SHIFT; i++) |
81819f0f CL |
3030 | kmalloc_caches[i]. name = |
3031 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3032 | ||
3033 | #ifdef CONFIG_SMP | |
3034 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3035 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3036 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3037 | #else | |
3038 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3039 | #endif |
3040 | ||
81819f0f | 3041 | |
3adbefee IM |
3042 | printk(KERN_INFO |
3043 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3044 | " CPUs=%d, Nodes=%d\n", |
3045 | caches, cache_line_size(), | |
81819f0f CL |
3046 | slub_min_order, slub_max_order, slub_min_objects, |
3047 | nr_cpu_ids, nr_node_ids); | |
3048 | } | |
3049 | ||
3050 | /* | |
3051 | * Find a mergeable slab cache | |
3052 | */ | |
3053 | static int slab_unmergeable(struct kmem_cache *s) | |
3054 | { | |
3055 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3056 | return 1; | |
3057 | ||
c59def9f | 3058 | if (s->ctor) |
81819f0f CL |
3059 | return 1; |
3060 | ||
8ffa6875 CL |
3061 | /* |
3062 | * We may have set a slab to be unmergeable during bootstrap. | |
3063 | */ | |
3064 | if (s->refcount < 0) | |
3065 | return 1; | |
3066 | ||
81819f0f CL |
3067 | return 0; |
3068 | } | |
3069 | ||
3070 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3071 | size_t align, unsigned long flags, const char *name, |
4ba9b9d0 | 3072 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f | 3073 | { |
5b95a4ac | 3074 | struct kmem_cache *s; |
81819f0f CL |
3075 | |
3076 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3077 | return NULL; | |
3078 | ||
c59def9f | 3079 | if (ctor) |
81819f0f CL |
3080 | return NULL; |
3081 | ||
3082 | size = ALIGN(size, sizeof(void *)); | |
3083 | align = calculate_alignment(flags, align, size); | |
3084 | size = ALIGN(size, align); | |
ba0268a8 | 3085 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3086 | |
5b95a4ac | 3087 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3088 | if (slab_unmergeable(s)) |
3089 | continue; | |
3090 | ||
3091 | if (size > s->size) | |
3092 | continue; | |
3093 | ||
ba0268a8 | 3094 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3095 | continue; |
3096 | /* | |
3097 | * Check if alignment is compatible. | |
3098 | * Courtesy of Adrian Drzewiecki | |
3099 | */ | |
06428780 | 3100 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3101 | continue; |
3102 | ||
3103 | if (s->size - size >= sizeof(void *)) | |
3104 | continue; | |
3105 | ||
3106 | return s; | |
3107 | } | |
3108 | return NULL; | |
3109 | } | |
3110 | ||
3111 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
3112 | size_t align, unsigned long flags, | |
4ba9b9d0 | 3113 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
3114 | { |
3115 | struct kmem_cache *s; | |
3116 | ||
3117 | down_write(&slub_lock); | |
ba0268a8 | 3118 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3119 | if (s) { |
42a9fdbb CL |
3120 | int cpu; |
3121 | ||
81819f0f CL |
3122 | s->refcount++; |
3123 | /* | |
3124 | * Adjust the object sizes so that we clear | |
3125 | * the complete object on kzalloc. | |
3126 | */ | |
3127 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3128 | |
3129 | /* | |
3130 | * And then we need to update the object size in the | |
3131 | * per cpu structures | |
3132 | */ | |
3133 | for_each_online_cpu(cpu) | |
3134 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
81819f0f | 3135 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3136 | up_write(&slub_lock); |
81819f0f CL |
3137 | if (sysfs_slab_alias(s, name)) |
3138 | goto err; | |
a0e1d1be CL |
3139 | return s; |
3140 | } | |
3141 | s = kmalloc(kmem_size, GFP_KERNEL); | |
3142 | if (s) { | |
3143 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3144 | size, align, flags, ctor)) { |
81819f0f | 3145 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
3146 | up_write(&slub_lock); |
3147 | if (sysfs_slab_add(s)) | |
3148 | goto err; | |
3149 | return s; | |
3150 | } | |
3151 | kfree(s); | |
81819f0f CL |
3152 | } |
3153 | up_write(&slub_lock); | |
81819f0f CL |
3154 | |
3155 | err: | |
81819f0f CL |
3156 | if (flags & SLAB_PANIC) |
3157 | panic("Cannot create slabcache %s\n", name); | |
3158 | else | |
3159 | s = NULL; | |
3160 | return s; | |
3161 | } | |
3162 | EXPORT_SYMBOL(kmem_cache_create); | |
3163 | ||
81819f0f | 3164 | #ifdef CONFIG_SMP |
81819f0f | 3165 | /* |
672bba3a CL |
3166 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3167 | * necessary. | |
81819f0f CL |
3168 | */ |
3169 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3170 | unsigned long action, void *hcpu) | |
3171 | { | |
3172 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3173 | struct kmem_cache *s; |
3174 | unsigned long flags; | |
81819f0f CL |
3175 | |
3176 | switch (action) { | |
4c93c355 CL |
3177 | case CPU_UP_PREPARE: |
3178 | case CPU_UP_PREPARE_FROZEN: | |
3179 | init_alloc_cpu_cpu(cpu); | |
3180 | down_read(&slub_lock); | |
3181 | list_for_each_entry(s, &slab_caches, list) | |
3182 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3183 | GFP_KERNEL); | |
3184 | up_read(&slub_lock); | |
3185 | break; | |
3186 | ||
81819f0f | 3187 | case CPU_UP_CANCELED: |
8bb78442 | 3188 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3189 | case CPU_DEAD: |
8bb78442 | 3190 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3191 | down_read(&slub_lock); |
3192 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3193 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3194 | ||
5b95a4ac CL |
3195 | local_irq_save(flags); |
3196 | __flush_cpu_slab(s, cpu); | |
3197 | local_irq_restore(flags); | |
4c93c355 CL |
3198 | free_kmem_cache_cpu(c, cpu); |
3199 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3200 | } |
3201 | up_read(&slub_lock); | |
81819f0f CL |
3202 | break; |
3203 | default: | |
3204 | break; | |
3205 | } | |
3206 | return NOTIFY_OK; | |
3207 | } | |
3208 | ||
06428780 | 3209 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3210 | .notifier_call = slab_cpuup_callback |
06428780 | 3211 | }; |
81819f0f CL |
3212 | |
3213 | #endif | |
3214 | ||
81819f0f CL |
3215 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
3216 | { | |
aadb4bc4 CL |
3217 | struct kmem_cache *s; |
3218 | ||
3219 | if (unlikely(size > PAGE_SIZE / 2)) | |
eada35ef PE |
3220 | return kmalloc_large(size, gfpflags); |
3221 | ||
aadb4bc4 | 3222 | s = get_slab(size, gfpflags); |
81819f0f | 3223 | |
2408c550 | 3224 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3225 | return s; |
81819f0f | 3226 | |
ce15fea8 | 3227 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3228 | } |
3229 | ||
3230 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
3231 | int node, void *caller) | |
3232 | { | |
aadb4bc4 CL |
3233 | struct kmem_cache *s; |
3234 | ||
3235 | if (unlikely(size > PAGE_SIZE / 2)) | |
eada35ef PE |
3236 | return kmalloc_large(size, gfpflags); |
3237 | ||
aadb4bc4 | 3238 | s = get_slab(size, gfpflags); |
81819f0f | 3239 | |
2408c550 | 3240 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3241 | return s; |
81819f0f | 3242 | |
ce15fea8 | 3243 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3244 | } |
3245 | ||
41ecc55b | 3246 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
3247 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3248 | unsigned long *map) | |
53e15af0 CL |
3249 | { |
3250 | void *p; | |
683d0baa | 3251 | void *addr = slab_address(page); |
53e15af0 CL |
3252 | |
3253 | if (!check_slab(s, page) || | |
3254 | !on_freelist(s, page, NULL)) | |
3255 | return 0; | |
3256 | ||
3257 | /* Now we know that a valid freelist exists */ | |
3258 | bitmap_zero(map, s->objects); | |
3259 | ||
7656c72b CL |
3260 | for_each_free_object(p, s, page->freelist) { |
3261 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3262 | if (!check_object(s, page, p, 0)) |
3263 | return 0; | |
3264 | } | |
3265 | ||
7656c72b CL |
3266 | for_each_object(p, s, addr) |
3267 | if (!test_bit(slab_index(p, s, addr), map)) | |
53e15af0 CL |
3268 | if (!check_object(s, page, p, 1)) |
3269 | return 0; | |
3270 | return 1; | |
3271 | } | |
3272 | ||
434e245d CL |
3273 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3274 | unsigned long *map) | |
53e15af0 CL |
3275 | { |
3276 | if (slab_trylock(page)) { | |
434e245d | 3277 | validate_slab(s, page, map); |
53e15af0 CL |
3278 | slab_unlock(page); |
3279 | } else | |
3280 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3281 | s->name, page); | |
3282 | ||
3283 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3284 | if (!SlabDebug(page)) |
3285 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3286 | "on slab 0x%p\n", s->name, page); |
3287 | } else { | |
35e5d7ee CL |
3288 | if (SlabDebug(page)) |
3289 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3290 | "slab 0x%p\n", s->name, page); |
3291 | } | |
3292 | } | |
3293 | ||
434e245d CL |
3294 | static int validate_slab_node(struct kmem_cache *s, |
3295 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3296 | { |
3297 | unsigned long count = 0; | |
3298 | struct page *page; | |
3299 | unsigned long flags; | |
3300 | ||
3301 | spin_lock_irqsave(&n->list_lock, flags); | |
3302 | ||
3303 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3304 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3305 | count++; |
3306 | } | |
3307 | if (count != n->nr_partial) | |
3308 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3309 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3310 | ||
3311 | if (!(s->flags & SLAB_STORE_USER)) | |
3312 | goto out; | |
3313 | ||
3314 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3315 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3316 | count++; |
3317 | } | |
3318 | if (count != atomic_long_read(&n->nr_slabs)) | |
3319 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3320 | "counter=%ld\n", s->name, count, | |
3321 | atomic_long_read(&n->nr_slabs)); | |
3322 | ||
3323 | out: | |
3324 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3325 | return count; | |
3326 | } | |
3327 | ||
434e245d | 3328 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3329 | { |
3330 | int node; | |
3331 | unsigned long count = 0; | |
434e245d CL |
3332 | unsigned long *map = kmalloc(BITS_TO_LONGS(s->objects) * |
3333 | sizeof(unsigned long), GFP_KERNEL); | |
3334 | ||
3335 | if (!map) | |
3336 | return -ENOMEM; | |
53e15af0 CL |
3337 | |
3338 | flush_all(s); | |
f64dc58c | 3339 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3340 | struct kmem_cache_node *n = get_node(s, node); |
3341 | ||
434e245d | 3342 | count += validate_slab_node(s, n, map); |
53e15af0 | 3343 | } |
434e245d | 3344 | kfree(map); |
53e15af0 CL |
3345 | return count; |
3346 | } | |
3347 | ||
b3459709 CL |
3348 | #ifdef SLUB_RESILIENCY_TEST |
3349 | static void resiliency_test(void) | |
3350 | { | |
3351 | u8 *p; | |
3352 | ||
3353 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3354 | printk(KERN_ERR "-----------------------\n"); | |
3355 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3356 | ||
3357 | p = kzalloc(16, GFP_KERNEL); | |
3358 | p[16] = 0x12; | |
3359 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3360 | " 0x12->0x%p\n\n", p + 16); | |
3361 | ||
3362 | validate_slab_cache(kmalloc_caches + 4); | |
3363 | ||
3364 | /* Hmmm... The next two are dangerous */ | |
3365 | p = kzalloc(32, GFP_KERNEL); | |
3366 | p[32 + sizeof(void *)] = 0x34; | |
3367 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3368 | " 0x34 -> -0x%p\n", p); |
3369 | printk(KERN_ERR | |
3370 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3371 | |
3372 | validate_slab_cache(kmalloc_caches + 5); | |
3373 | p = kzalloc(64, GFP_KERNEL); | |
3374 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3375 | *p = 0x56; | |
3376 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3377 | p); | |
3adbefee IM |
3378 | printk(KERN_ERR |
3379 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3380 | validate_slab_cache(kmalloc_caches + 6); |
3381 | ||
3382 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3383 | p = kzalloc(128, GFP_KERNEL); | |
3384 | kfree(p); | |
3385 | *p = 0x78; | |
3386 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3387 | validate_slab_cache(kmalloc_caches + 7); | |
3388 | ||
3389 | p = kzalloc(256, GFP_KERNEL); | |
3390 | kfree(p); | |
3391 | p[50] = 0x9a; | |
3adbefee IM |
3392 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3393 | p); | |
b3459709 CL |
3394 | validate_slab_cache(kmalloc_caches + 8); |
3395 | ||
3396 | p = kzalloc(512, GFP_KERNEL); | |
3397 | kfree(p); | |
3398 | p[512] = 0xab; | |
3399 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3400 | validate_slab_cache(kmalloc_caches + 9); | |
3401 | } | |
3402 | #else | |
3403 | static void resiliency_test(void) {}; | |
3404 | #endif | |
3405 | ||
88a420e4 | 3406 | /* |
672bba3a | 3407 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3408 | * and freed. |
3409 | */ | |
3410 | ||
3411 | struct location { | |
3412 | unsigned long count; | |
3413 | void *addr; | |
45edfa58 CL |
3414 | long long sum_time; |
3415 | long min_time; | |
3416 | long max_time; | |
3417 | long min_pid; | |
3418 | long max_pid; | |
3419 | cpumask_t cpus; | |
3420 | nodemask_t nodes; | |
88a420e4 CL |
3421 | }; |
3422 | ||
3423 | struct loc_track { | |
3424 | unsigned long max; | |
3425 | unsigned long count; | |
3426 | struct location *loc; | |
3427 | }; | |
3428 | ||
3429 | static void free_loc_track(struct loc_track *t) | |
3430 | { | |
3431 | if (t->max) | |
3432 | free_pages((unsigned long)t->loc, | |
3433 | get_order(sizeof(struct location) * t->max)); | |
3434 | } | |
3435 | ||
68dff6a9 | 3436 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3437 | { |
3438 | struct location *l; | |
3439 | int order; | |
3440 | ||
88a420e4 CL |
3441 | order = get_order(sizeof(struct location) * max); |
3442 | ||
68dff6a9 | 3443 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3444 | if (!l) |
3445 | return 0; | |
3446 | ||
3447 | if (t->count) { | |
3448 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3449 | free_loc_track(t); | |
3450 | } | |
3451 | t->max = max; | |
3452 | t->loc = l; | |
3453 | return 1; | |
3454 | } | |
3455 | ||
3456 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3457 | const struct track *track) |
88a420e4 CL |
3458 | { |
3459 | long start, end, pos; | |
3460 | struct location *l; | |
3461 | void *caddr; | |
45edfa58 | 3462 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3463 | |
3464 | start = -1; | |
3465 | end = t->count; | |
3466 | ||
3467 | for ( ; ; ) { | |
3468 | pos = start + (end - start + 1) / 2; | |
3469 | ||
3470 | /* | |
3471 | * There is nothing at "end". If we end up there | |
3472 | * we need to add something to before end. | |
3473 | */ | |
3474 | if (pos == end) | |
3475 | break; | |
3476 | ||
3477 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3478 | if (track->addr == caddr) { |
3479 | ||
3480 | l = &t->loc[pos]; | |
3481 | l->count++; | |
3482 | if (track->when) { | |
3483 | l->sum_time += age; | |
3484 | if (age < l->min_time) | |
3485 | l->min_time = age; | |
3486 | if (age > l->max_time) | |
3487 | l->max_time = age; | |
3488 | ||
3489 | if (track->pid < l->min_pid) | |
3490 | l->min_pid = track->pid; | |
3491 | if (track->pid > l->max_pid) | |
3492 | l->max_pid = track->pid; | |
3493 | ||
3494 | cpu_set(track->cpu, l->cpus); | |
3495 | } | |
3496 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3497 | return 1; |
3498 | } | |
3499 | ||
45edfa58 | 3500 | if (track->addr < caddr) |
88a420e4 CL |
3501 | end = pos; |
3502 | else | |
3503 | start = pos; | |
3504 | } | |
3505 | ||
3506 | /* | |
672bba3a | 3507 | * Not found. Insert new tracking element. |
88a420e4 | 3508 | */ |
68dff6a9 | 3509 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3510 | return 0; |
3511 | ||
3512 | l = t->loc + pos; | |
3513 | if (pos < t->count) | |
3514 | memmove(l + 1, l, | |
3515 | (t->count - pos) * sizeof(struct location)); | |
3516 | t->count++; | |
3517 | l->count = 1; | |
45edfa58 CL |
3518 | l->addr = track->addr; |
3519 | l->sum_time = age; | |
3520 | l->min_time = age; | |
3521 | l->max_time = age; | |
3522 | l->min_pid = track->pid; | |
3523 | l->max_pid = track->pid; | |
3524 | cpus_clear(l->cpus); | |
3525 | cpu_set(track->cpu, l->cpus); | |
3526 | nodes_clear(l->nodes); | |
3527 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3528 | return 1; |
3529 | } | |
3530 | ||
3531 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3532 | struct page *page, enum track_item alloc) | |
3533 | { | |
683d0baa | 3534 | void *addr = slab_address(page); |
7656c72b | 3535 | DECLARE_BITMAP(map, s->objects); |
88a420e4 CL |
3536 | void *p; |
3537 | ||
3538 | bitmap_zero(map, s->objects); | |
7656c72b CL |
3539 | for_each_free_object(p, s, page->freelist) |
3540 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3541 | |
7656c72b | 3542 | for_each_object(p, s, addr) |
45edfa58 CL |
3543 | if (!test_bit(slab_index(p, s, addr), map)) |
3544 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3545 | } |
3546 | ||
3547 | static int list_locations(struct kmem_cache *s, char *buf, | |
3548 | enum track_item alloc) | |
3549 | { | |
e374d483 | 3550 | int len = 0; |
88a420e4 | 3551 | unsigned long i; |
68dff6a9 | 3552 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3553 | int node; |
3554 | ||
68dff6a9 | 3555 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3556 | GFP_TEMPORARY)) |
68dff6a9 | 3557 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3558 | |
3559 | /* Push back cpu slabs */ | |
3560 | flush_all(s); | |
3561 | ||
f64dc58c | 3562 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3563 | struct kmem_cache_node *n = get_node(s, node); |
3564 | unsigned long flags; | |
3565 | struct page *page; | |
3566 | ||
9e86943b | 3567 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3568 | continue; |
3569 | ||
3570 | spin_lock_irqsave(&n->list_lock, flags); | |
3571 | list_for_each_entry(page, &n->partial, lru) | |
3572 | process_slab(&t, s, page, alloc); | |
3573 | list_for_each_entry(page, &n->full, lru) | |
3574 | process_slab(&t, s, page, alloc); | |
3575 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3576 | } | |
3577 | ||
3578 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3579 | struct location *l = &t.loc[i]; |
88a420e4 | 3580 | |
e374d483 | 3581 | if (len > PAGE_SIZE - 100) |
88a420e4 | 3582 | break; |
e374d483 | 3583 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3584 | |
3585 | if (l->addr) | |
e374d483 | 3586 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3587 | else |
e374d483 | 3588 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3589 | |
3590 | if (l->sum_time != l->min_time) { | |
3591 | unsigned long remainder; | |
3592 | ||
e374d483 | 3593 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
45edfa58 CL |
3594 | l->min_time, |
3595 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3596 | l->max_time); | |
3597 | } else | |
e374d483 | 3598 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3599 | l->min_time); |
3600 | ||
3601 | if (l->min_pid != l->max_pid) | |
e374d483 | 3602 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3603 | l->min_pid, l->max_pid); |
3604 | else | |
e374d483 | 3605 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3606 | l->min_pid); |
3607 | ||
84966343 | 3608 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
e374d483 HH |
3609 | len < PAGE_SIZE - 60) { |
3610 | len += sprintf(buf + len, " cpus="); | |
3611 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3612 | l->cpus); |
3613 | } | |
3614 | ||
84966343 | 3615 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3616 | len < PAGE_SIZE - 60) { |
3617 | len += sprintf(buf + len, " nodes="); | |
3618 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3619 | l->nodes); |
3620 | } | |
3621 | ||
e374d483 | 3622 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3623 | } |
3624 | ||
3625 | free_loc_track(&t); | |
3626 | if (!t.count) | |
e374d483 HH |
3627 | len += sprintf(buf, "No data\n"); |
3628 | return len; | |
88a420e4 CL |
3629 | } |
3630 | ||
81819f0f CL |
3631 | enum slab_stat_type { |
3632 | SL_FULL, | |
3633 | SL_PARTIAL, | |
3634 | SL_CPU, | |
3635 | SL_OBJECTS | |
3636 | }; | |
3637 | ||
3638 | #define SO_FULL (1 << SL_FULL) | |
3639 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
3640 | #define SO_CPU (1 << SL_CPU) | |
3641 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
3642 | ||
3643 | static unsigned long slab_objects(struct kmem_cache *s, | |
3644 | char *buf, unsigned long flags) | |
3645 | { | |
3646 | unsigned long total = 0; | |
3647 | int cpu; | |
3648 | int node; | |
3649 | int x; | |
3650 | unsigned long *nodes; | |
3651 | unsigned long *per_cpu; | |
3652 | ||
3653 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
3654 | per_cpu = nodes + nr_node_ids; | |
3655 | ||
3656 | for_each_possible_cpu(cpu) { | |
dfb4f096 CL |
3657 | struct page *page; |
3658 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
81819f0f | 3659 | |
dfb4f096 CL |
3660 | if (!c) |
3661 | continue; | |
3662 | ||
3663 | page = c->page; | |
ee3c72a1 CL |
3664 | node = c->node; |
3665 | if (node < 0) | |
3666 | continue; | |
81819f0f | 3667 | if (page) { |
81819f0f | 3668 | if (flags & SO_CPU) { |
81819f0f CL |
3669 | if (flags & SO_OBJECTS) |
3670 | x = page->inuse; | |
3671 | else | |
3672 | x = 1; | |
3673 | total += x; | |
ee3c72a1 | 3674 | nodes[node] += x; |
81819f0f | 3675 | } |
ee3c72a1 | 3676 | per_cpu[node]++; |
81819f0f CL |
3677 | } |
3678 | } | |
3679 | ||
f64dc58c | 3680 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3681 | struct kmem_cache_node *n = get_node(s, node); |
3682 | ||
3683 | if (flags & SO_PARTIAL) { | |
3684 | if (flags & SO_OBJECTS) | |
3685 | x = count_partial(n); | |
3686 | else | |
3687 | x = n->nr_partial; | |
3688 | total += x; | |
3689 | nodes[node] += x; | |
3690 | } | |
3691 | ||
3692 | if (flags & SO_FULL) { | |
9e86943b | 3693 | int full_slabs = atomic_long_read(&n->nr_slabs) |
81819f0f CL |
3694 | - per_cpu[node] |
3695 | - n->nr_partial; | |
3696 | ||
3697 | if (flags & SO_OBJECTS) | |
3698 | x = full_slabs * s->objects; | |
3699 | else | |
3700 | x = full_slabs; | |
3701 | total += x; | |
3702 | nodes[node] += x; | |
3703 | } | |
3704 | } | |
3705 | ||
3706 | x = sprintf(buf, "%lu", total); | |
3707 | #ifdef CONFIG_NUMA | |
f64dc58c | 3708 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3709 | if (nodes[node]) |
3710 | x += sprintf(buf + x, " N%d=%lu", | |
3711 | node, nodes[node]); | |
3712 | #endif | |
3713 | kfree(nodes); | |
3714 | return x + sprintf(buf + x, "\n"); | |
3715 | } | |
3716 | ||
3717 | static int any_slab_objects(struct kmem_cache *s) | |
3718 | { | |
3719 | int node; | |
3720 | int cpu; | |
3721 | ||
dfb4f096 CL |
3722 | for_each_possible_cpu(cpu) { |
3723 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
3724 | ||
3725 | if (c && c->page) | |
81819f0f | 3726 | return 1; |
dfb4f096 | 3727 | } |
81819f0f | 3728 | |
dfb4f096 | 3729 | for_each_online_node(node) { |
81819f0f CL |
3730 | struct kmem_cache_node *n = get_node(s, node); |
3731 | ||
dfb4f096 CL |
3732 | if (!n) |
3733 | continue; | |
3734 | ||
9e86943b | 3735 | if (n->nr_partial || atomic_long_read(&n->nr_slabs)) |
81819f0f CL |
3736 | return 1; |
3737 | } | |
3738 | return 0; | |
3739 | } | |
3740 | ||
3741 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3742 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3743 | ||
3744 | struct slab_attribute { | |
3745 | struct attribute attr; | |
3746 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3747 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3748 | }; | |
3749 | ||
3750 | #define SLAB_ATTR_RO(_name) \ | |
3751 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3752 | ||
3753 | #define SLAB_ATTR(_name) \ | |
3754 | static struct slab_attribute _name##_attr = \ | |
3755 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3756 | ||
81819f0f CL |
3757 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3758 | { | |
3759 | return sprintf(buf, "%d\n", s->size); | |
3760 | } | |
3761 | SLAB_ATTR_RO(slab_size); | |
3762 | ||
3763 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3764 | { | |
3765 | return sprintf(buf, "%d\n", s->align); | |
3766 | } | |
3767 | SLAB_ATTR_RO(align); | |
3768 | ||
3769 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3770 | { | |
3771 | return sprintf(buf, "%d\n", s->objsize); | |
3772 | } | |
3773 | SLAB_ATTR_RO(object_size); | |
3774 | ||
3775 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3776 | { | |
3777 | return sprintf(buf, "%d\n", s->objects); | |
3778 | } | |
3779 | SLAB_ATTR_RO(objs_per_slab); | |
3780 | ||
3781 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3782 | { | |
3783 | return sprintf(buf, "%d\n", s->order); | |
3784 | } | |
3785 | SLAB_ATTR_RO(order); | |
3786 | ||
3787 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3788 | { | |
3789 | if (s->ctor) { | |
3790 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3791 | ||
3792 | return n + sprintf(buf + n, "\n"); | |
3793 | } | |
3794 | return 0; | |
3795 | } | |
3796 | SLAB_ATTR_RO(ctor); | |
3797 | ||
81819f0f CL |
3798 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3799 | { | |
3800 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3801 | } | |
3802 | SLAB_ATTR_RO(aliases); | |
3803 | ||
3804 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3805 | { | |
3806 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3807 | } | |
3808 | SLAB_ATTR_RO(slabs); | |
3809 | ||
3810 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3811 | { | |
3812 | return slab_objects(s, buf, SO_PARTIAL); | |
3813 | } | |
3814 | SLAB_ATTR_RO(partial); | |
3815 | ||
3816 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3817 | { | |
3818 | return slab_objects(s, buf, SO_CPU); | |
3819 | } | |
3820 | SLAB_ATTR_RO(cpu_slabs); | |
3821 | ||
3822 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3823 | { | |
3824 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3825 | } | |
3826 | SLAB_ATTR_RO(objects); | |
3827 | ||
3828 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3829 | { | |
3830 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3831 | } | |
3832 | ||
3833 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3834 | const char *buf, size_t length) | |
3835 | { | |
3836 | s->flags &= ~SLAB_DEBUG_FREE; | |
3837 | if (buf[0] == '1') | |
3838 | s->flags |= SLAB_DEBUG_FREE; | |
3839 | return length; | |
3840 | } | |
3841 | SLAB_ATTR(sanity_checks); | |
3842 | ||
3843 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3844 | { | |
3845 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3846 | } | |
3847 | ||
3848 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3849 | size_t length) | |
3850 | { | |
3851 | s->flags &= ~SLAB_TRACE; | |
3852 | if (buf[0] == '1') | |
3853 | s->flags |= SLAB_TRACE; | |
3854 | return length; | |
3855 | } | |
3856 | SLAB_ATTR(trace); | |
3857 | ||
3858 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3859 | { | |
3860 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3861 | } | |
3862 | ||
3863 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3864 | const char *buf, size_t length) | |
3865 | { | |
3866 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3867 | if (buf[0] == '1') | |
3868 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3869 | return length; | |
3870 | } | |
3871 | SLAB_ATTR(reclaim_account); | |
3872 | ||
3873 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3874 | { | |
5af60839 | 3875 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3876 | } |
3877 | SLAB_ATTR_RO(hwcache_align); | |
3878 | ||
3879 | #ifdef CONFIG_ZONE_DMA | |
3880 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3881 | { | |
3882 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3883 | } | |
3884 | SLAB_ATTR_RO(cache_dma); | |
3885 | #endif | |
3886 | ||
3887 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3888 | { | |
3889 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3890 | } | |
3891 | SLAB_ATTR_RO(destroy_by_rcu); | |
3892 | ||
3893 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3894 | { | |
3895 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3896 | } | |
3897 | ||
3898 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3899 | const char *buf, size_t length) | |
3900 | { | |
3901 | if (any_slab_objects(s)) | |
3902 | return -EBUSY; | |
3903 | ||
3904 | s->flags &= ~SLAB_RED_ZONE; | |
3905 | if (buf[0] == '1') | |
3906 | s->flags |= SLAB_RED_ZONE; | |
3907 | calculate_sizes(s); | |
3908 | return length; | |
3909 | } | |
3910 | SLAB_ATTR(red_zone); | |
3911 | ||
3912 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3913 | { | |
3914 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3915 | } | |
3916 | ||
3917 | static ssize_t poison_store(struct kmem_cache *s, | |
3918 | const char *buf, size_t length) | |
3919 | { | |
3920 | if (any_slab_objects(s)) | |
3921 | return -EBUSY; | |
3922 | ||
3923 | s->flags &= ~SLAB_POISON; | |
3924 | if (buf[0] == '1') | |
3925 | s->flags |= SLAB_POISON; | |
3926 | calculate_sizes(s); | |
3927 | return length; | |
3928 | } | |
3929 | SLAB_ATTR(poison); | |
3930 | ||
3931 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3932 | { | |
3933 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3934 | } | |
3935 | ||
3936 | static ssize_t store_user_store(struct kmem_cache *s, | |
3937 | const char *buf, size_t length) | |
3938 | { | |
3939 | if (any_slab_objects(s)) | |
3940 | return -EBUSY; | |
3941 | ||
3942 | s->flags &= ~SLAB_STORE_USER; | |
3943 | if (buf[0] == '1') | |
3944 | s->flags |= SLAB_STORE_USER; | |
3945 | calculate_sizes(s); | |
3946 | return length; | |
3947 | } | |
3948 | SLAB_ATTR(store_user); | |
3949 | ||
53e15af0 CL |
3950 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3951 | { | |
3952 | return 0; | |
3953 | } | |
3954 | ||
3955 | static ssize_t validate_store(struct kmem_cache *s, | |
3956 | const char *buf, size_t length) | |
3957 | { | |
434e245d CL |
3958 | int ret = -EINVAL; |
3959 | ||
3960 | if (buf[0] == '1') { | |
3961 | ret = validate_slab_cache(s); | |
3962 | if (ret >= 0) | |
3963 | ret = length; | |
3964 | } | |
3965 | return ret; | |
53e15af0 CL |
3966 | } |
3967 | SLAB_ATTR(validate); | |
3968 | ||
2086d26a CL |
3969 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
3970 | { | |
3971 | return 0; | |
3972 | } | |
3973 | ||
3974 | static ssize_t shrink_store(struct kmem_cache *s, | |
3975 | const char *buf, size_t length) | |
3976 | { | |
3977 | if (buf[0] == '1') { | |
3978 | int rc = kmem_cache_shrink(s); | |
3979 | ||
3980 | if (rc) | |
3981 | return rc; | |
3982 | } else | |
3983 | return -EINVAL; | |
3984 | return length; | |
3985 | } | |
3986 | SLAB_ATTR(shrink); | |
3987 | ||
88a420e4 CL |
3988 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
3989 | { | |
3990 | if (!(s->flags & SLAB_STORE_USER)) | |
3991 | return -ENOSYS; | |
3992 | return list_locations(s, buf, TRACK_ALLOC); | |
3993 | } | |
3994 | SLAB_ATTR_RO(alloc_calls); | |
3995 | ||
3996 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
3997 | { | |
3998 | if (!(s->flags & SLAB_STORE_USER)) | |
3999 | return -ENOSYS; | |
4000 | return list_locations(s, buf, TRACK_FREE); | |
4001 | } | |
4002 | SLAB_ATTR_RO(free_calls); | |
4003 | ||
81819f0f | 4004 | #ifdef CONFIG_NUMA |
9824601e | 4005 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4006 | { |
9824601e | 4007 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4008 | } |
4009 | ||
9824601e | 4010 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4011 | const char *buf, size_t length) |
4012 | { | |
4013 | int n = simple_strtoul(buf, NULL, 10); | |
4014 | ||
4015 | if (n < 100) | |
9824601e | 4016 | s->remote_node_defrag_ratio = n * 10; |
81819f0f CL |
4017 | return length; |
4018 | } | |
9824601e | 4019 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4020 | #endif |
4021 | ||
8ff12cfc CL |
4022 | #ifdef CONFIG_SLUB_STATS |
4023 | ||
4024 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) | |
4025 | { | |
4026 | unsigned long sum = 0; | |
4027 | int cpu; | |
4028 | int len; | |
4029 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4030 | ||
4031 | if (!data) | |
4032 | return -ENOMEM; | |
4033 | ||
4034 | for_each_online_cpu(cpu) { | |
4035 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4036 | ||
4037 | data[cpu] = x; | |
4038 | sum += x; | |
4039 | } | |
4040 | ||
4041 | len = sprintf(buf, "%lu", sum); | |
4042 | ||
4043 | for_each_online_cpu(cpu) { | |
4044 | if (data[cpu] && len < PAGE_SIZE - 20) | |
4045 | len += sprintf(buf + len, " c%d=%u", cpu, data[cpu]); | |
4046 | } | |
4047 | kfree(data); | |
4048 | return len + sprintf(buf + len, "\n"); | |
4049 | } | |
4050 | ||
4051 | #define STAT_ATTR(si, text) \ | |
4052 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4053 | { \ | |
4054 | return show_stat(s, buf, si); \ | |
4055 | } \ | |
4056 | SLAB_ATTR_RO(text); \ | |
4057 | ||
4058 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4059 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4060 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4061 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4062 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4063 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4064 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4065 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4066 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4067 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4068 | STAT_ATTR(FREE_SLAB, free_slab); | |
4069 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4070 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4071 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4072 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4073 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4074 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
4075 | ||
4076 | #endif | |
4077 | ||
06428780 | 4078 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4079 | &slab_size_attr.attr, |
4080 | &object_size_attr.attr, | |
4081 | &objs_per_slab_attr.attr, | |
4082 | &order_attr.attr, | |
4083 | &objects_attr.attr, | |
4084 | &slabs_attr.attr, | |
4085 | &partial_attr.attr, | |
4086 | &cpu_slabs_attr.attr, | |
4087 | &ctor_attr.attr, | |
81819f0f CL |
4088 | &aliases_attr.attr, |
4089 | &align_attr.attr, | |
4090 | &sanity_checks_attr.attr, | |
4091 | &trace_attr.attr, | |
4092 | &hwcache_align_attr.attr, | |
4093 | &reclaim_account_attr.attr, | |
4094 | &destroy_by_rcu_attr.attr, | |
4095 | &red_zone_attr.attr, | |
4096 | &poison_attr.attr, | |
4097 | &store_user_attr.attr, | |
53e15af0 | 4098 | &validate_attr.attr, |
2086d26a | 4099 | &shrink_attr.attr, |
88a420e4 CL |
4100 | &alloc_calls_attr.attr, |
4101 | &free_calls_attr.attr, | |
81819f0f CL |
4102 | #ifdef CONFIG_ZONE_DMA |
4103 | &cache_dma_attr.attr, | |
4104 | #endif | |
4105 | #ifdef CONFIG_NUMA | |
9824601e | 4106 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4107 | #endif |
4108 | #ifdef CONFIG_SLUB_STATS | |
4109 | &alloc_fastpath_attr.attr, | |
4110 | &alloc_slowpath_attr.attr, | |
4111 | &free_fastpath_attr.attr, | |
4112 | &free_slowpath_attr.attr, | |
4113 | &free_frozen_attr.attr, | |
4114 | &free_add_partial_attr.attr, | |
4115 | &free_remove_partial_attr.attr, | |
4116 | &alloc_from_partial_attr.attr, | |
4117 | &alloc_slab_attr.attr, | |
4118 | &alloc_refill_attr.attr, | |
4119 | &free_slab_attr.attr, | |
4120 | &cpuslab_flush_attr.attr, | |
4121 | &deactivate_full_attr.attr, | |
4122 | &deactivate_empty_attr.attr, | |
4123 | &deactivate_to_head_attr.attr, | |
4124 | &deactivate_to_tail_attr.attr, | |
4125 | &deactivate_remote_frees_attr.attr, | |
81819f0f CL |
4126 | #endif |
4127 | NULL | |
4128 | }; | |
4129 | ||
4130 | static struct attribute_group slab_attr_group = { | |
4131 | .attrs = slab_attrs, | |
4132 | }; | |
4133 | ||
4134 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4135 | struct attribute *attr, | |
4136 | char *buf) | |
4137 | { | |
4138 | struct slab_attribute *attribute; | |
4139 | struct kmem_cache *s; | |
4140 | int err; | |
4141 | ||
4142 | attribute = to_slab_attr(attr); | |
4143 | s = to_slab(kobj); | |
4144 | ||
4145 | if (!attribute->show) | |
4146 | return -EIO; | |
4147 | ||
4148 | err = attribute->show(s, buf); | |
4149 | ||
4150 | return err; | |
4151 | } | |
4152 | ||
4153 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4154 | struct attribute *attr, | |
4155 | const char *buf, size_t len) | |
4156 | { | |
4157 | struct slab_attribute *attribute; | |
4158 | struct kmem_cache *s; | |
4159 | int err; | |
4160 | ||
4161 | attribute = to_slab_attr(attr); | |
4162 | s = to_slab(kobj); | |
4163 | ||
4164 | if (!attribute->store) | |
4165 | return -EIO; | |
4166 | ||
4167 | err = attribute->store(s, buf, len); | |
4168 | ||
4169 | return err; | |
4170 | } | |
4171 | ||
151c602f CL |
4172 | static void kmem_cache_release(struct kobject *kobj) |
4173 | { | |
4174 | struct kmem_cache *s = to_slab(kobj); | |
4175 | ||
4176 | kfree(s); | |
4177 | } | |
4178 | ||
81819f0f CL |
4179 | static struct sysfs_ops slab_sysfs_ops = { |
4180 | .show = slab_attr_show, | |
4181 | .store = slab_attr_store, | |
4182 | }; | |
4183 | ||
4184 | static struct kobj_type slab_ktype = { | |
4185 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4186 | .release = kmem_cache_release |
81819f0f CL |
4187 | }; |
4188 | ||
4189 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4190 | { | |
4191 | struct kobj_type *ktype = get_ktype(kobj); | |
4192 | ||
4193 | if (ktype == &slab_ktype) | |
4194 | return 1; | |
4195 | return 0; | |
4196 | } | |
4197 | ||
4198 | static struct kset_uevent_ops slab_uevent_ops = { | |
4199 | .filter = uevent_filter, | |
4200 | }; | |
4201 | ||
27c3a314 | 4202 | static struct kset *slab_kset; |
81819f0f CL |
4203 | |
4204 | #define ID_STR_LENGTH 64 | |
4205 | ||
4206 | /* Create a unique string id for a slab cache: | |
4207 | * format | |
4208 | * :[flags-]size:[memory address of kmemcache] | |
4209 | */ | |
4210 | static char *create_unique_id(struct kmem_cache *s) | |
4211 | { | |
4212 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4213 | char *p = name; | |
4214 | ||
4215 | BUG_ON(!name); | |
4216 | ||
4217 | *p++ = ':'; | |
4218 | /* | |
4219 | * First flags affecting slabcache operations. We will only | |
4220 | * get here for aliasable slabs so we do not need to support | |
4221 | * too many flags. The flags here must cover all flags that | |
4222 | * are matched during merging to guarantee that the id is | |
4223 | * unique. | |
4224 | */ | |
4225 | if (s->flags & SLAB_CACHE_DMA) | |
4226 | *p++ = 'd'; | |
4227 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4228 | *p++ = 'a'; | |
4229 | if (s->flags & SLAB_DEBUG_FREE) | |
4230 | *p++ = 'F'; | |
4231 | if (p != name + 1) | |
4232 | *p++ = '-'; | |
4233 | p += sprintf(p, "%07d", s->size); | |
4234 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4235 | return name; | |
4236 | } | |
4237 | ||
4238 | static int sysfs_slab_add(struct kmem_cache *s) | |
4239 | { | |
4240 | int err; | |
4241 | const char *name; | |
4242 | int unmergeable; | |
4243 | ||
4244 | if (slab_state < SYSFS) | |
4245 | /* Defer until later */ | |
4246 | return 0; | |
4247 | ||
4248 | unmergeable = slab_unmergeable(s); | |
4249 | if (unmergeable) { | |
4250 | /* | |
4251 | * Slabcache can never be merged so we can use the name proper. | |
4252 | * This is typically the case for debug situations. In that | |
4253 | * case we can catch duplicate names easily. | |
4254 | */ | |
27c3a314 | 4255 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4256 | name = s->name; |
4257 | } else { | |
4258 | /* | |
4259 | * Create a unique name for the slab as a target | |
4260 | * for the symlinks. | |
4261 | */ | |
4262 | name = create_unique_id(s); | |
4263 | } | |
4264 | ||
27c3a314 | 4265 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4266 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4267 | if (err) { | |
4268 | kobject_put(&s->kobj); | |
81819f0f | 4269 | return err; |
1eada11c | 4270 | } |
81819f0f CL |
4271 | |
4272 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4273 | if (err) | |
4274 | return err; | |
4275 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4276 | if (!unmergeable) { | |
4277 | /* Setup first alias */ | |
4278 | sysfs_slab_alias(s, s->name); | |
4279 | kfree(name); | |
4280 | } | |
4281 | return 0; | |
4282 | } | |
4283 | ||
4284 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4285 | { | |
4286 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4287 | kobject_del(&s->kobj); | |
151c602f | 4288 | kobject_put(&s->kobj); |
81819f0f CL |
4289 | } |
4290 | ||
4291 | /* | |
4292 | * Need to buffer aliases during bootup until sysfs becomes | |
4293 | * available lest we loose that information. | |
4294 | */ | |
4295 | struct saved_alias { | |
4296 | struct kmem_cache *s; | |
4297 | const char *name; | |
4298 | struct saved_alias *next; | |
4299 | }; | |
4300 | ||
5af328a5 | 4301 | static struct saved_alias *alias_list; |
81819f0f CL |
4302 | |
4303 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4304 | { | |
4305 | struct saved_alias *al; | |
4306 | ||
4307 | if (slab_state == SYSFS) { | |
4308 | /* | |
4309 | * If we have a leftover link then remove it. | |
4310 | */ | |
27c3a314 GKH |
4311 | sysfs_remove_link(&slab_kset->kobj, name); |
4312 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4313 | } |
4314 | ||
4315 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4316 | if (!al) | |
4317 | return -ENOMEM; | |
4318 | ||
4319 | al->s = s; | |
4320 | al->name = name; | |
4321 | al->next = alias_list; | |
4322 | alias_list = al; | |
4323 | return 0; | |
4324 | } | |
4325 | ||
4326 | static int __init slab_sysfs_init(void) | |
4327 | { | |
5b95a4ac | 4328 | struct kmem_cache *s; |
81819f0f CL |
4329 | int err; |
4330 | ||
0ff21e46 | 4331 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4332 | if (!slab_kset) { |
81819f0f CL |
4333 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4334 | return -ENOSYS; | |
4335 | } | |
4336 | ||
26a7bd03 CL |
4337 | slab_state = SYSFS; |
4338 | ||
5b95a4ac | 4339 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4340 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4341 | if (err) |
4342 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4343 | " to sysfs\n", s->name); | |
26a7bd03 | 4344 | } |
81819f0f CL |
4345 | |
4346 | while (alias_list) { | |
4347 | struct saved_alias *al = alias_list; | |
4348 | ||
4349 | alias_list = alias_list->next; | |
4350 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4351 | if (err) |
4352 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4353 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4354 | kfree(al); |
4355 | } | |
4356 | ||
4357 | resiliency_test(); | |
4358 | return 0; | |
4359 | } | |
4360 | ||
4361 | __initcall(slab_sysfs_init); | |
81819f0f | 4362 | #endif |
57ed3eda PE |
4363 | |
4364 | /* | |
4365 | * The /proc/slabinfo ABI | |
4366 | */ | |
158a9624 LT |
4367 | #ifdef CONFIG_SLABINFO |
4368 | ||
4369 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |
4370 | size_t count, loff_t *ppos) | |
4371 | { | |
4372 | return -EINVAL; | |
4373 | } | |
4374 | ||
57ed3eda PE |
4375 | |
4376 | static void print_slabinfo_header(struct seq_file *m) | |
4377 | { | |
4378 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4379 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4380 | "<objperslab> <pagesperslab>"); | |
4381 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4382 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4383 | seq_putc(m, '\n'); | |
4384 | } | |
4385 | ||
4386 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4387 | { | |
4388 | loff_t n = *pos; | |
4389 | ||
4390 | down_read(&slub_lock); | |
4391 | if (!n) | |
4392 | print_slabinfo_header(m); | |
4393 | ||
4394 | return seq_list_start(&slab_caches, *pos); | |
4395 | } | |
4396 | ||
4397 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4398 | { | |
4399 | return seq_list_next(p, &slab_caches, pos); | |
4400 | } | |
4401 | ||
4402 | static void s_stop(struct seq_file *m, void *p) | |
4403 | { | |
4404 | up_read(&slub_lock); | |
4405 | } | |
4406 | ||
4407 | static int s_show(struct seq_file *m, void *p) | |
4408 | { | |
4409 | unsigned long nr_partials = 0; | |
4410 | unsigned long nr_slabs = 0; | |
4411 | unsigned long nr_inuse = 0; | |
4412 | unsigned long nr_objs; | |
4413 | struct kmem_cache *s; | |
4414 | int node; | |
4415 | ||
4416 | s = list_entry(p, struct kmem_cache, list); | |
4417 | ||
4418 | for_each_online_node(node) { | |
4419 | struct kmem_cache_node *n = get_node(s, node); | |
4420 | ||
4421 | if (!n) | |
4422 | continue; | |
4423 | ||
4424 | nr_partials += n->nr_partial; | |
4425 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
4426 | nr_inuse += count_partial(n); | |
4427 | } | |
4428 | ||
4429 | nr_objs = nr_slabs * s->objects; | |
4430 | nr_inuse += (nr_slabs - nr_partials) * s->objects; | |
4431 | ||
4432 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
4433 | nr_objs, s->size, s->objects, (1 << s->order)); | |
4434 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); | |
4435 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4436 | 0UL); | |
4437 | seq_putc(m, '\n'); | |
4438 | return 0; | |
4439 | } | |
4440 | ||
4441 | const struct seq_operations slabinfo_op = { | |
4442 | .start = s_start, | |
4443 | .next = s_next, | |
4444 | .stop = s_stop, | |
4445 | .show = s_show, | |
4446 | }; | |
4447 | ||
158a9624 | 4448 | #endif /* CONFIG_SLABINFO */ |