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