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 */ | |
71c7a06f CL |
214 | #define __KMALLOC_CACHE 0x20000000 /* objects freed using kfree */ |
215 | #define __PAGE_ALLOC_FALLBACK 0x10000000 /* Allow fallback to page alloc */ | |
81819f0f | 216 | |
65c02d4c CL |
217 | /* Not all arches define cache_line_size */ |
218 | #ifndef cache_line_size | |
219 | #define cache_line_size() L1_CACHE_BYTES | |
220 | #endif | |
221 | ||
81819f0f CL |
222 | static int kmem_size = sizeof(struct kmem_cache); |
223 | ||
224 | #ifdef CONFIG_SMP | |
225 | static struct notifier_block slab_notifier; | |
226 | #endif | |
227 | ||
228 | static enum { | |
229 | DOWN, /* No slab functionality available */ | |
230 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 231 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
232 | SYSFS /* Sysfs up */ |
233 | } slab_state = DOWN; | |
234 | ||
235 | /* A list of all slab caches on the system */ | |
236 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 237 | static LIST_HEAD(slab_caches); |
81819f0f | 238 | |
02cbc874 CL |
239 | /* |
240 | * Tracking user of a slab. | |
241 | */ | |
242 | struct track { | |
243 | void *addr; /* Called from address */ | |
244 | int cpu; /* Was running on cpu */ | |
245 | int pid; /* Pid context */ | |
246 | unsigned long when; /* When did the operation occur */ | |
247 | }; | |
248 | ||
249 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
250 | ||
41ecc55b | 251 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
252 | static int sysfs_slab_add(struct kmem_cache *); |
253 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
254 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 255 | |
81819f0f | 256 | #else |
0c710013 CL |
257 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
258 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
259 | { return 0; } | |
151c602f CL |
260 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
261 | { | |
262 | kfree(s); | |
263 | } | |
8ff12cfc | 264 | |
81819f0f CL |
265 | #endif |
266 | ||
8ff12cfc CL |
267 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
268 | { | |
269 | #ifdef CONFIG_SLUB_STATS | |
270 | c->stat[si]++; | |
271 | #endif | |
272 | } | |
273 | ||
81819f0f CL |
274 | /******************************************************************** |
275 | * Core slab cache functions | |
276 | *******************************************************************/ | |
277 | ||
278 | int slab_is_available(void) | |
279 | { | |
280 | return slab_state >= UP; | |
281 | } | |
282 | ||
283 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
284 | { | |
285 | #ifdef CONFIG_NUMA | |
286 | return s->node[node]; | |
287 | #else | |
288 | return &s->local_node; | |
289 | #endif | |
290 | } | |
291 | ||
dfb4f096 CL |
292 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
293 | { | |
4c93c355 CL |
294 | #ifdef CONFIG_SMP |
295 | return s->cpu_slab[cpu]; | |
296 | #else | |
297 | return &s->cpu_slab; | |
298 | #endif | |
dfb4f096 CL |
299 | } |
300 | ||
683d0baa CL |
301 | /* |
302 | * The end pointer in a slab is special. It points to the first object in the | |
303 | * slab but has bit 0 set to mark it. | |
304 | * | |
305 | * Note that SLUB relies on page_mapping returning NULL for pages with bit 0 | |
306 | * in the mapping set. | |
307 | */ | |
308 | static inline int is_end(void *addr) | |
309 | { | |
310 | return (unsigned long)addr & PAGE_MAPPING_ANON; | |
311 | } | |
312 | ||
dada123d | 313 | static void *slab_address(struct page *page) |
683d0baa CL |
314 | { |
315 | return page->end - PAGE_MAPPING_ANON; | |
316 | } | |
317 | ||
02cbc874 CL |
318 | static inline int check_valid_pointer(struct kmem_cache *s, |
319 | struct page *page, const void *object) | |
320 | { | |
321 | void *base; | |
322 | ||
683d0baa | 323 | if (object == page->end) |
02cbc874 CL |
324 | return 1; |
325 | ||
683d0baa | 326 | base = slab_address(page); |
02cbc874 CL |
327 | if (object < base || object >= base + s->objects * s->size || |
328 | (object - base) % s->size) { | |
329 | return 0; | |
330 | } | |
331 | ||
332 | return 1; | |
333 | } | |
334 | ||
7656c72b CL |
335 | /* |
336 | * Slow version of get and set free pointer. | |
337 | * | |
338 | * This version requires touching the cache lines of kmem_cache which | |
339 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
340 | * from the page struct. | |
341 | */ | |
342 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
343 | { | |
344 | return *(void **)(object + s->offset); | |
345 | } | |
346 | ||
347 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
348 | { | |
349 | *(void **)(object + s->offset) = fp; | |
350 | } | |
351 | ||
352 | /* Loop over all objects in a slab */ | |
353 | #define for_each_object(__p, __s, __addr) \ | |
354 | for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\ | |
355 | __p += (__s)->size) | |
356 | ||
357 | /* Scan freelist */ | |
358 | #define for_each_free_object(__p, __s, __free) \ | |
683d0baa CL |
359 | for (__p = (__free); (__p) != page->end; __p = get_freepointer((__s),\ |
360 | __p)) | |
7656c72b CL |
361 | |
362 | /* Determine object index from a given position */ | |
363 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
364 | { | |
365 | return (p - addr) / s->size; | |
366 | } | |
367 | ||
41ecc55b CL |
368 | #ifdef CONFIG_SLUB_DEBUG |
369 | /* | |
370 | * Debug settings: | |
371 | */ | |
f0630fff CL |
372 | #ifdef CONFIG_SLUB_DEBUG_ON |
373 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
374 | #else | |
41ecc55b | 375 | static int slub_debug; |
f0630fff | 376 | #endif |
41ecc55b CL |
377 | |
378 | static char *slub_debug_slabs; | |
379 | ||
81819f0f CL |
380 | /* |
381 | * Object debugging | |
382 | */ | |
383 | static void print_section(char *text, u8 *addr, unsigned int length) | |
384 | { | |
385 | int i, offset; | |
386 | int newline = 1; | |
387 | char ascii[17]; | |
388 | ||
389 | ascii[16] = 0; | |
390 | ||
391 | for (i = 0; i < length; i++) { | |
392 | if (newline) { | |
24922684 | 393 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
394 | newline = 0; |
395 | } | |
06428780 | 396 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
397 | offset = i % 16; |
398 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
399 | if (offset == 15) { | |
06428780 | 400 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
401 | newline = 1; |
402 | } | |
403 | } | |
404 | if (!newline) { | |
405 | i %= 16; | |
406 | while (i < 16) { | |
06428780 | 407 | printk(KERN_CONT " "); |
81819f0f CL |
408 | ascii[i] = ' '; |
409 | i++; | |
410 | } | |
06428780 | 411 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
412 | } |
413 | } | |
414 | ||
81819f0f CL |
415 | static struct track *get_track(struct kmem_cache *s, void *object, |
416 | enum track_item alloc) | |
417 | { | |
418 | struct track *p; | |
419 | ||
420 | if (s->offset) | |
421 | p = object + s->offset + sizeof(void *); | |
422 | else | |
423 | p = object + s->inuse; | |
424 | ||
425 | return p + alloc; | |
426 | } | |
427 | ||
428 | static void set_track(struct kmem_cache *s, void *object, | |
429 | enum track_item alloc, void *addr) | |
430 | { | |
431 | struct track *p; | |
432 | ||
433 | if (s->offset) | |
434 | p = object + s->offset + sizeof(void *); | |
435 | else | |
436 | p = object + s->inuse; | |
437 | ||
438 | p += alloc; | |
439 | if (addr) { | |
440 | p->addr = addr; | |
441 | p->cpu = smp_processor_id(); | |
442 | p->pid = current ? current->pid : -1; | |
443 | p->when = jiffies; | |
444 | } else | |
445 | memset(p, 0, sizeof(struct track)); | |
446 | } | |
447 | ||
81819f0f CL |
448 | static void init_tracking(struct kmem_cache *s, void *object) |
449 | { | |
24922684 CL |
450 | if (!(s->flags & SLAB_STORE_USER)) |
451 | return; | |
452 | ||
453 | set_track(s, object, TRACK_FREE, NULL); | |
454 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
455 | } |
456 | ||
457 | static void print_track(const char *s, struct track *t) | |
458 | { | |
459 | if (!t->addr) | |
460 | return; | |
461 | ||
24922684 | 462 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 463 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
464 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
465 | } | |
466 | ||
467 | static void print_tracking(struct kmem_cache *s, void *object) | |
468 | { | |
469 | if (!(s->flags & SLAB_STORE_USER)) | |
470 | return; | |
471 | ||
472 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
473 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
474 | } | |
475 | ||
476 | static void print_page_info(struct page *page) | |
477 | { | |
478 | printk(KERN_ERR "INFO: Slab 0x%p used=%u fp=0x%p flags=0x%04lx\n", | |
479 | page, page->inuse, page->freelist, page->flags); | |
480 | ||
481 | } | |
482 | ||
483 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
484 | { | |
485 | va_list args; | |
486 | char buf[100]; | |
487 | ||
488 | va_start(args, fmt); | |
489 | vsnprintf(buf, sizeof(buf), fmt, args); | |
490 | va_end(args); | |
491 | printk(KERN_ERR "========================================" | |
492 | "=====================================\n"); | |
493 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
494 | printk(KERN_ERR "----------------------------------------" | |
495 | "-------------------------------------\n\n"); | |
81819f0f CL |
496 | } |
497 | ||
24922684 CL |
498 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
499 | { | |
500 | va_list args; | |
501 | char buf[100]; | |
502 | ||
503 | va_start(args, fmt); | |
504 | vsnprintf(buf, sizeof(buf), fmt, args); | |
505 | va_end(args); | |
506 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
507 | } | |
508 | ||
509 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
510 | { |
511 | unsigned int off; /* Offset of last byte */ | |
683d0baa | 512 | u8 *addr = slab_address(page); |
24922684 CL |
513 | |
514 | print_tracking(s, p); | |
515 | ||
516 | print_page_info(page); | |
517 | ||
518 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
519 | p, p - addr, get_freepointer(s, p)); | |
520 | ||
521 | if (p > addr + 16) | |
522 | print_section("Bytes b4", p - 16, 16); | |
523 | ||
524 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
525 | |
526 | if (s->flags & SLAB_RED_ZONE) | |
527 | print_section("Redzone", p + s->objsize, | |
528 | s->inuse - s->objsize); | |
529 | ||
81819f0f CL |
530 | if (s->offset) |
531 | off = s->offset + sizeof(void *); | |
532 | else | |
533 | off = s->inuse; | |
534 | ||
24922684 | 535 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 536 | off += 2 * sizeof(struct track); |
81819f0f CL |
537 | |
538 | if (off != s->size) | |
539 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
540 | print_section("Padding", p + off, s->size - off); |
541 | ||
542 | dump_stack(); | |
81819f0f CL |
543 | } |
544 | ||
545 | static void object_err(struct kmem_cache *s, struct page *page, | |
546 | u8 *object, char *reason) | |
547 | { | |
24922684 CL |
548 | slab_bug(s, reason); |
549 | print_trailer(s, page, object); | |
81819f0f CL |
550 | } |
551 | ||
24922684 | 552 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
553 | { |
554 | va_list args; | |
555 | char buf[100]; | |
556 | ||
24922684 CL |
557 | va_start(args, fmt); |
558 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 559 | va_end(args); |
24922684 CL |
560 | slab_bug(s, fmt); |
561 | print_page_info(page); | |
81819f0f CL |
562 | dump_stack(); |
563 | } | |
564 | ||
565 | static void init_object(struct kmem_cache *s, void *object, int active) | |
566 | { | |
567 | u8 *p = object; | |
568 | ||
569 | if (s->flags & __OBJECT_POISON) { | |
570 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 571 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
572 | } |
573 | ||
574 | if (s->flags & SLAB_RED_ZONE) | |
575 | memset(p + s->objsize, | |
576 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
577 | s->inuse - s->objsize); | |
578 | } | |
579 | ||
24922684 | 580 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
581 | { |
582 | while (bytes) { | |
583 | if (*start != (u8)value) | |
24922684 | 584 | return start; |
81819f0f CL |
585 | start++; |
586 | bytes--; | |
587 | } | |
24922684 CL |
588 | return NULL; |
589 | } | |
590 | ||
591 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
592 | void *from, void *to) | |
593 | { | |
594 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
595 | memset(from, data, to - from); | |
596 | } | |
597 | ||
598 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
599 | u8 *object, char *what, | |
06428780 | 600 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
601 | { |
602 | u8 *fault; | |
603 | u8 *end; | |
604 | ||
605 | fault = check_bytes(start, value, bytes); | |
606 | if (!fault) | |
607 | return 1; | |
608 | ||
609 | end = start + bytes; | |
610 | while (end > fault && end[-1] == value) | |
611 | end--; | |
612 | ||
613 | slab_bug(s, "%s overwritten", what); | |
614 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
615 | fault, end - 1, fault[0], value); | |
616 | print_trailer(s, page, object); | |
617 | ||
618 | restore_bytes(s, what, value, fault, end); | |
619 | return 0; | |
81819f0f CL |
620 | } |
621 | ||
81819f0f CL |
622 | /* |
623 | * Object layout: | |
624 | * | |
625 | * object address | |
626 | * Bytes of the object to be managed. | |
627 | * If the freepointer may overlay the object then the free | |
628 | * pointer is the first word of the object. | |
672bba3a | 629 | * |
81819f0f CL |
630 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
631 | * 0xa5 (POISON_END) | |
632 | * | |
633 | * object + s->objsize | |
634 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
635 | * Padding is extended by another word if Redzoning is enabled and |
636 | * objsize == inuse. | |
637 | * | |
81819f0f CL |
638 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
639 | * 0xcc (RED_ACTIVE) for objects in use. | |
640 | * | |
641 | * object + s->inuse | |
672bba3a CL |
642 | * Meta data starts here. |
643 | * | |
81819f0f CL |
644 | * A. Free pointer (if we cannot overwrite object on free) |
645 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a CL |
646 | * C. Padding to reach required alignment boundary or at mininum |
647 | * one word if debuggin is on to be able to detect writes | |
648 | * before the word boundary. | |
649 | * | |
650 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
651 | * |
652 | * object + s->size | |
672bba3a | 653 | * Nothing is used beyond s->size. |
81819f0f | 654 | * |
672bba3a CL |
655 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
656 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
657 | * may be used with merged slabcaches. |
658 | */ | |
659 | ||
81819f0f CL |
660 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
661 | { | |
662 | unsigned long off = s->inuse; /* The end of info */ | |
663 | ||
664 | if (s->offset) | |
665 | /* Freepointer is placed after the object. */ | |
666 | off += sizeof(void *); | |
667 | ||
668 | if (s->flags & SLAB_STORE_USER) | |
669 | /* We also have user information there */ | |
670 | off += 2 * sizeof(struct track); | |
671 | ||
672 | if (s->size == off) | |
673 | return 1; | |
674 | ||
24922684 CL |
675 | return check_bytes_and_report(s, page, p, "Object padding", |
676 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
677 | } |
678 | ||
679 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
680 | { | |
24922684 CL |
681 | u8 *start; |
682 | u8 *fault; | |
683 | u8 *end; | |
684 | int length; | |
685 | int remainder; | |
81819f0f CL |
686 | |
687 | if (!(s->flags & SLAB_POISON)) | |
688 | return 1; | |
689 | ||
683d0baa | 690 | start = slab_address(page); |
24922684 | 691 | end = start + (PAGE_SIZE << s->order); |
81819f0f | 692 | length = s->objects * s->size; |
24922684 | 693 | remainder = end - (start + length); |
81819f0f CL |
694 | if (!remainder) |
695 | return 1; | |
696 | ||
24922684 CL |
697 | fault = check_bytes(start + length, POISON_INUSE, remainder); |
698 | if (!fault) | |
699 | return 1; | |
700 | while (end > fault && end[-1] == POISON_INUSE) | |
701 | end--; | |
702 | ||
703 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
704 | print_section("Padding", start, length); | |
705 | ||
706 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
707 | return 0; | |
81819f0f CL |
708 | } |
709 | ||
710 | static int check_object(struct kmem_cache *s, struct page *page, | |
711 | void *object, int active) | |
712 | { | |
713 | u8 *p = object; | |
714 | u8 *endobject = object + s->objsize; | |
715 | ||
716 | if (s->flags & SLAB_RED_ZONE) { | |
717 | unsigned int red = | |
718 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
719 | ||
24922684 CL |
720 | if (!check_bytes_and_report(s, page, object, "Redzone", |
721 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 722 | return 0; |
81819f0f | 723 | } else { |
3adbefee IM |
724 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
725 | check_bytes_and_report(s, page, p, "Alignment padding", | |
726 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
727 | } | |
81819f0f CL |
728 | } |
729 | ||
730 | if (s->flags & SLAB_POISON) { | |
731 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
732 | (!check_bytes_and_report(s, page, p, "Poison", p, |
733 | POISON_FREE, s->objsize - 1) || | |
734 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 735 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 736 | return 0; |
81819f0f CL |
737 | /* |
738 | * check_pad_bytes cleans up on its own. | |
739 | */ | |
740 | check_pad_bytes(s, page, p); | |
741 | } | |
742 | ||
743 | if (!s->offset && active) | |
744 | /* | |
745 | * Object and freepointer overlap. Cannot check | |
746 | * freepointer while object is allocated. | |
747 | */ | |
748 | return 1; | |
749 | ||
750 | /* Check free pointer validity */ | |
751 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
752 | object_err(s, page, p, "Freepointer corrupt"); | |
753 | /* | |
754 | * No choice but to zap it and thus loose the remainder | |
755 | * of the free objects in this slab. May cause | |
672bba3a | 756 | * another error because the object count is now wrong. |
81819f0f | 757 | */ |
683d0baa | 758 | set_freepointer(s, p, page->end); |
81819f0f CL |
759 | return 0; |
760 | } | |
761 | return 1; | |
762 | } | |
763 | ||
764 | static int check_slab(struct kmem_cache *s, struct page *page) | |
765 | { | |
766 | VM_BUG_ON(!irqs_disabled()); | |
767 | ||
768 | if (!PageSlab(page)) { | |
24922684 | 769 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
770 | return 0; |
771 | } | |
81819f0f | 772 | if (page->inuse > s->objects) { |
24922684 CL |
773 | slab_err(s, page, "inuse %u > max %u", |
774 | s->name, page->inuse, s->objects); | |
81819f0f CL |
775 | return 0; |
776 | } | |
777 | /* Slab_pad_check fixes things up after itself */ | |
778 | slab_pad_check(s, page); | |
779 | return 1; | |
780 | } | |
781 | ||
782 | /* | |
672bba3a CL |
783 | * Determine if a certain object on a page is on the freelist. Must hold the |
784 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
785 | */ |
786 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
787 | { | |
788 | int nr = 0; | |
789 | void *fp = page->freelist; | |
790 | void *object = NULL; | |
791 | ||
683d0baa | 792 | while (fp != page->end && nr <= s->objects) { |
81819f0f CL |
793 | if (fp == search) |
794 | return 1; | |
795 | if (!check_valid_pointer(s, page, fp)) { | |
796 | if (object) { | |
797 | object_err(s, page, object, | |
798 | "Freechain corrupt"); | |
683d0baa | 799 | set_freepointer(s, object, page->end); |
81819f0f CL |
800 | break; |
801 | } else { | |
24922684 | 802 | slab_err(s, page, "Freepointer corrupt"); |
683d0baa | 803 | page->freelist = page->end; |
81819f0f | 804 | page->inuse = s->objects; |
24922684 | 805 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
806 | return 0; |
807 | } | |
808 | break; | |
809 | } | |
810 | object = fp; | |
811 | fp = get_freepointer(s, object); | |
812 | nr++; | |
813 | } | |
814 | ||
815 | if (page->inuse != s->objects - nr) { | |
70d71228 | 816 | slab_err(s, page, "Wrong object count. Counter is %d but " |
24922684 | 817 | "counted were %d", page->inuse, s->objects - nr); |
81819f0f | 818 | page->inuse = s->objects - nr; |
24922684 | 819 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
820 | } |
821 | return search == NULL; | |
822 | } | |
823 | ||
3ec09742 CL |
824 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
825 | { | |
826 | if (s->flags & SLAB_TRACE) { | |
827 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
828 | s->name, | |
829 | alloc ? "alloc" : "free", | |
830 | object, page->inuse, | |
831 | page->freelist); | |
832 | ||
833 | if (!alloc) | |
834 | print_section("Object", (void *)object, s->objsize); | |
835 | ||
836 | dump_stack(); | |
837 | } | |
838 | } | |
839 | ||
643b1138 | 840 | /* |
672bba3a | 841 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 842 | */ |
e95eed57 | 843 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 844 | { |
643b1138 CL |
845 | spin_lock(&n->list_lock); |
846 | list_add(&page->lru, &n->full); | |
847 | spin_unlock(&n->list_lock); | |
848 | } | |
849 | ||
850 | static void remove_full(struct kmem_cache *s, struct page *page) | |
851 | { | |
852 | struct kmem_cache_node *n; | |
853 | ||
854 | if (!(s->flags & SLAB_STORE_USER)) | |
855 | return; | |
856 | ||
857 | n = get_node(s, page_to_nid(page)); | |
858 | ||
859 | spin_lock(&n->list_lock); | |
860 | list_del(&page->lru); | |
861 | spin_unlock(&n->list_lock); | |
862 | } | |
863 | ||
3ec09742 CL |
864 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
865 | void *object) | |
866 | { | |
867 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
868 | return; | |
869 | ||
870 | init_object(s, object, 0); | |
871 | init_tracking(s, object); | |
872 | } | |
873 | ||
874 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
875 | void *object, void *addr) | |
81819f0f CL |
876 | { |
877 | if (!check_slab(s, page)) | |
878 | goto bad; | |
879 | ||
880 | if (object && !on_freelist(s, page, object)) { | |
24922684 | 881 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 882 | goto bad; |
81819f0f CL |
883 | } |
884 | ||
885 | if (!check_valid_pointer(s, page, object)) { | |
886 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 887 | goto bad; |
81819f0f CL |
888 | } |
889 | ||
3ec09742 | 890 | if (object && !check_object(s, page, object, 0)) |
81819f0f | 891 | goto bad; |
81819f0f | 892 | |
3ec09742 CL |
893 | /* Success perform special debug activities for allocs */ |
894 | if (s->flags & SLAB_STORE_USER) | |
895 | set_track(s, object, TRACK_ALLOC, addr); | |
896 | trace(s, page, object, 1); | |
897 | init_object(s, object, 1); | |
81819f0f | 898 | return 1; |
3ec09742 | 899 | |
81819f0f CL |
900 | bad: |
901 | if (PageSlab(page)) { | |
902 | /* | |
903 | * If this is a slab page then lets do the best we can | |
904 | * to avoid issues in the future. Marking all objects | |
672bba3a | 905 | * as used avoids touching the remaining objects. |
81819f0f | 906 | */ |
24922684 | 907 | slab_fix(s, "Marking all objects used"); |
81819f0f | 908 | page->inuse = s->objects; |
683d0baa | 909 | page->freelist = page->end; |
81819f0f CL |
910 | } |
911 | return 0; | |
912 | } | |
913 | ||
3ec09742 CL |
914 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
915 | void *object, void *addr) | |
81819f0f CL |
916 | { |
917 | if (!check_slab(s, page)) | |
918 | goto fail; | |
919 | ||
920 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 921 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
922 | goto fail; |
923 | } | |
924 | ||
925 | if (on_freelist(s, page, object)) { | |
24922684 | 926 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
927 | goto fail; |
928 | } | |
929 | ||
930 | if (!check_object(s, page, object, 1)) | |
931 | return 0; | |
932 | ||
933 | if (unlikely(s != page->slab)) { | |
3adbefee | 934 | if (!PageSlab(page)) { |
70d71228 CL |
935 | slab_err(s, page, "Attempt to free object(0x%p) " |
936 | "outside of slab", object); | |
3adbefee | 937 | } else if (!page->slab) { |
81819f0f | 938 | printk(KERN_ERR |
70d71228 | 939 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 940 | object); |
70d71228 | 941 | dump_stack(); |
06428780 | 942 | } else |
24922684 CL |
943 | object_err(s, page, object, |
944 | "page slab pointer corrupt."); | |
81819f0f CL |
945 | goto fail; |
946 | } | |
3ec09742 CL |
947 | |
948 | /* Special debug activities for freeing objects */ | |
683d0baa | 949 | if (!SlabFrozen(page) && page->freelist == page->end) |
3ec09742 CL |
950 | remove_full(s, page); |
951 | if (s->flags & SLAB_STORE_USER) | |
952 | set_track(s, object, TRACK_FREE, addr); | |
953 | trace(s, page, object, 0); | |
954 | init_object(s, object, 0); | |
81819f0f | 955 | return 1; |
3ec09742 | 956 | |
81819f0f | 957 | fail: |
24922684 | 958 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
959 | return 0; |
960 | } | |
961 | ||
41ecc55b CL |
962 | static int __init setup_slub_debug(char *str) |
963 | { | |
f0630fff CL |
964 | slub_debug = DEBUG_DEFAULT_FLAGS; |
965 | if (*str++ != '=' || !*str) | |
966 | /* | |
967 | * No options specified. Switch on full debugging. | |
968 | */ | |
969 | goto out; | |
970 | ||
971 | if (*str == ',') | |
972 | /* | |
973 | * No options but restriction on slabs. This means full | |
974 | * debugging for slabs matching a pattern. | |
975 | */ | |
976 | goto check_slabs; | |
977 | ||
978 | slub_debug = 0; | |
979 | if (*str == '-') | |
980 | /* | |
981 | * Switch off all debugging measures. | |
982 | */ | |
983 | goto out; | |
984 | ||
985 | /* | |
986 | * Determine which debug features should be switched on | |
987 | */ | |
06428780 | 988 | for (; *str && *str != ','; str++) { |
f0630fff CL |
989 | switch (tolower(*str)) { |
990 | case 'f': | |
991 | slub_debug |= SLAB_DEBUG_FREE; | |
992 | break; | |
993 | case 'z': | |
994 | slub_debug |= SLAB_RED_ZONE; | |
995 | break; | |
996 | case 'p': | |
997 | slub_debug |= SLAB_POISON; | |
998 | break; | |
999 | case 'u': | |
1000 | slub_debug |= SLAB_STORE_USER; | |
1001 | break; | |
1002 | case 't': | |
1003 | slub_debug |= SLAB_TRACE; | |
1004 | break; | |
1005 | default: | |
1006 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1007 | "unknown. skipped\n", *str); |
f0630fff | 1008 | } |
41ecc55b CL |
1009 | } |
1010 | ||
f0630fff | 1011 | check_slabs: |
41ecc55b CL |
1012 | if (*str == ',') |
1013 | slub_debug_slabs = str + 1; | |
f0630fff | 1014 | out: |
41ecc55b CL |
1015 | return 1; |
1016 | } | |
1017 | ||
1018 | __setup("slub_debug", setup_slub_debug); | |
1019 | ||
ba0268a8 CL |
1020 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1021 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1022 | void (*ctor)(struct kmem_cache *, void *)) |
41ecc55b CL |
1023 | { |
1024 | /* | |
1025 | * The page->offset field is only 16 bit wide. This is an offset | |
1026 | * in units of words from the beginning of an object. If the slab | |
1027 | * size is bigger then we cannot move the free pointer behind the | |
1028 | * object anymore. | |
1029 | * | |
1030 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
1031 | * the limit is 512k. | |
1032 | * | |
c59def9f | 1033 | * Debugging or ctor may create a need to move the free |
41ecc55b CL |
1034 | * pointer. Fail if this happens. |
1035 | */ | |
ba0268a8 CL |
1036 | if (objsize >= 65535 * sizeof(void *)) { |
1037 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
41ecc55b | 1038 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); |
ba0268a8 CL |
1039 | BUG_ON(ctor); |
1040 | } else { | |
41ecc55b CL |
1041 | /* |
1042 | * Enable debugging if selected on the kernel commandline. | |
1043 | */ | |
1044 | if (slub_debug && (!slub_debug_slabs || | |
ba0268a8 | 1045 | strncmp(slub_debug_slabs, name, |
3adbefee | 1046 | strlen(slub_debug_slabs)) == 0)) |
ba0268a8 CL |
1047 | flags |= slub_debug; |
1048 | } | |
1049 | ||
1050 | return flags; | |
41ecc55b CL |
1051 | } |
1052 | #else | |
3ec09742 CL |
1053 | static inline void setup_object_debug(struct kmem_cache *s, |
1054 | struct page *page, void *object) {} | |
41ecc55b | 1055 | |
3ec09742 CL |
1056 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1057 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1058 | |
3ec09742 CL |
1059 | static inline int free_debug_processing(struct kmem_cache *s, |
1060 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1061 | |
41ecc55b CL |
1062 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1063 | { return 1; } | |
1064 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1065 | void *object, int active) { return 1; } | |
3ec09742 | 1066 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1067 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1068 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1069 | void (*ctor)(struct kmem_cache *, void *)) |
ba0268a8 CL |
1070 | { |
1071 | return flags; | |
1072 | } | |
41ecc55b CL |
1073 | #define slub_debug 0 |
1074 | #endif | |
81819f0f CL |
1075 | /* |
1076 | * Slab allocation and freeing | |
1077 | */ | |
1078 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1079 | { | |
06428780 | 1080 | struct page *page; |
81819f0f CL |
1081 | int pages = 1 << s->order; |
1082 | ||
b7a49f0d | 1083 | flags |= s->allocflags; |
e12ba74d | 1084 | |
81819f0f CL |
1085 | if (node == -1) |
1086 | page = alloc_pages(flags, s->order); | |
1087 | else | |
1088 | page = alloc_pages_node(node, flags, s->order); | |
1089 | ||
1090 | if (!page) | |
1091 | return NULL; | |
1092 | ||
1093 | mod_zone_page_state(page_zone(page), | |
1094 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1095 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1096 | pages); | |
1097 | ||
1098 | return page; | |
1099 | } | |
1100 | ||
1101 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1102 | void *object) | |
1103 | { | |
3ec09742 | 1104 | setup_object_debug(s, page, object); |
4f104934 | 1105 | if (unlikely(s->ctor)) |
4ba9b9d0 | 1106 | s->ctor(s, object); |
81819f0f CL |
1107 | } |
1108 | ||
1109 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1110 | { | |
1111 | struct page *page; | |
1112 | struct kmem_cache_node *n; | |
1113 | void *start; | |
81819f0f CL |
1114 | void *last; |
1115 | void *p; | |
1116 | ||
6cb06229 | 1117 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1118 | |
6cb06229 CL |
1119 | page = allocate_slab(s, |
1120 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1121 | if (!page) |
1122 | goto out; | |
1123 | ||
1124 | n = get_node(s, page_to_nid(page)); | |
1125 | if (n) | |
1126 | atomic_long_inc(&n->nr_slabs); | |
81819f0f CL |
1127 | page->slab = s; |
1128 | page->flags |= 1 << PG_slab; | |
1129 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1130 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1131 | SetSlabDebug(page); |
81819f0f CL |
1132 | |
1133 | start = page_address(page); | |
683d0baa | 1134 | page->end = start + 1; |
81819f0f CL |
1135 | |
1136 | if (unlikely(s->flags & SLAB_POISON)) | |
1137 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
1138 | ||
1139 | last = start; | |
7656c72b | 1140 | for_each_object(p, s, start) { |
81819f0f CL |
1141 | setup_object(s, page, last); |
1142 | set_freepointer(s, last, p); | |
1143 | last = p; | |
1144 | } | |
1145 | setup_object(s, page, last); | |
683d0baa | 1146 | set_freepointer(s, last, page->end); |
81819f0f CL |
1147 | |
1148 | page->freelist = start; | |
1149 | page->inuse = 0; | |
1150 | out: | |
81819f0f CL |
1151 | return page; |
1152 | } | |
1153 | ||
1154 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1155 | { | |
1156 | int pages = 1 << s->order; | |
1157 | ||
c59def9f | 1158 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1159 | void *p; |
1160 | ||
1161 | slab_pad_check(s, page); | |
683d0baa | 1162 | for_each_object(p, s, slab_address(page)) |
81819f0f | 1163 | check_object(s, page, p, 0); |
2208b764 | 1164 | ClearSlabDebug(page); |
81819f0f CL |
1165 | } |
1166 | ||
1167 | mod_zone_page_state(page_zone(page), | |
1168 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1169 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1170 | -pages); |
81819f0f | 1171 | |
683d0baa | 1172 | page->mapping = NULL; |
81819f0f CL |
1173 | __free_pages(page, s->order); |
1174 | } | |
1175 | ||
1176 | static void rcu_free_slab(struct rcu_head *h) | |
1177 | { | |
1178 | struct page *page; | |
1179 | ||
1180 | page = container_of((struct list_head *)h, struct page, lru); | |
1181 | __free_slab(page->slab, page); | |
1182 | } | |
1183 | ||
1184 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1185 | { | |
1186 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1187 | /* | |
1188 | * RCU free overloads the RCU head over the LRU | |
1189 | */ | |
1190 | struct rcu_head *head = (void *)&page->lru; | |
1191 | ||
1192 | call_rcu(head, rcu_free_slab); | |
1193 | } else | |
1194 | __free_slab(s, page); | |
1195 | } | |
1196 | ||
1197 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1198 | { | |
1199 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1200 | ||
1201 | atomic_long_dec(&n->nr_slabs); | |
1202 | reset_page_mapcount(page); | |
35e5d7ee | 1203 | __ClearPageSlab(page); |
81819f0f CL |
1204 | free_slab(s, page); |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * Per slab locking using the pagelock | |
1209 | */ | |
1210 | static __always_inline void slab_lock(struct page *page) | |
1211 | { | |
1212 | bit_spin_lock(PG_locked, &page->flags); | |
1213 | } | |
1214 | ||
1215 | static __always_inline void slab_unlock(struct page *page) | |
1216 | { | |
a76d3546 | 1217 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1218 | } |
1219 | ||
1220 | static __always_inline int slab_trylock(struct page *page) | |
1221 | { | |
1222 | int rc = 1; | |
1223 | ||
1224 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1225 | return rc; | |
1226 | } | |
1227 | ||
1228 | /* | |
1229 | * Management of partially allocated slabs | |
1230 | */ | |
7c2e132c CL |
1231 | static void add_partial(struct kmem_cache_node *n, |
1232 | struct page *page, int tail) | |
81819f0f | 1233 | { |
e95eed57 CL |
1234 | spin_lock(&n->list_lock); |
1235 | n->nr_partial++; | |
7c2e132c CL |
1236 | if (tail) |
1237 | list_add_tail(&page->lru, &n->partial); | |
1238 | else | |
1239 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1240 | spin_unlock(&n->list_lock); |
1241 | } | |
1242 | ||
1243 | static void remove_partial(struct kmem_cache *s, | |
1244 | struct page *page) | |
1245 | { | |
1246 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1247 | ||
1248 | spin_lock(&n->list_lock); | |
1249 | list_del(&page->lru); | |
1250 | n->nr_partial--; | |
1251 | spin_unlock(&n->list_lock); | |
1252 | } | |
1253 | ||
1254 | /* | |
672bba3a | 1255 | * Lock slab and remove from the partial list. |
81819f0f | 1256 | * |
672bba3a | 1257 | * Must hold list_lock. |
81819f0f | 1258 | */ |
4b6f0750 | 1259 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1260 | { |
1261 | if (slab_trylock(page)) { | |
1262 | list_del(&page->lru); | |
1263 | n->nr_partial--; | |
4b6f0750 | 1264 | SetSlabFrozen(page); |
81819f0f CL |
1265 | return 1; |
1266 | } | |
1267 | return 0; | |
1268 | } | |
1269 | ||
1270 | /* | |
672bba3a | 1271 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1272 | */ |
1273 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1274 | { | |
1275 | struct page *page; | |
1276 | ||
1277 | /* | |
1278 | * Racy check. If we mistakenly see no partial slabs then we | |
1279 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1280 | * partial slab and there is none available then get_partials() |
1281 | * will return NULL. | |
81819f0f CL |
1282 | */ |
1283 | if (!n || !n->nr_partial) | |
1284 | return NULL; | |
1285 | ||
1286 | spin_lock(&n->list_lock); | |
1287 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1288 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1289 | goto out; |
1290 | page = NULL; | |
1291 | out: | |
1292 | spin_unlock(&n->list_lock); | |
1293 | return page; | |
1294 | } | |
1295 | ||
1296 | /* | |
672bba3a | 1297 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1298 | */ |
1299 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1300 | { | |
1301 | #ifdef CONFIG_NUMA | |
1302 | struct zonelist *zonelist; | |
1303 | struct zone **z; | |
1304 | struct page *page; | |
1305 | ||
1306 | /* | |
672bba3a CL |
1307 | * The defrag ratio allows a configuration of the tradeoffs between |
1308 | * inter node defragmentation and node local allocations. A lower | |
1309 | * defrag_ratio increases the tendency to do local allocations | |
1310 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1311 | * |
672bba3a CL |
1312 | * If the defrag_ratio is set to 0 then kmalloc() always |
1313 | * returns node local objects. If the ratio is higher then kmalloc() | |
1314 | * may return off node objects because partial slabs are obtained | |
1315 | * from other nodes and filled up. | |
81819f0f CL |
1316 | * |
1317 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
672bba3a CL |
1318 | * defrag_ratio = 1000) then every (well almost) allocation will |
1319 | * first attempt to defrag slab caches on other nodes. This means | |
1320 | * scanning over all nodes to look for partial slabs which may be | |
1321 | * expensive if we do it every time we are trying to find a slab | |
1322 | * with available objects. | |
81819f0f | 1323 | */ |
9824601e CL |
1324 | if (!s->remote_node_defrag_ratio || |
1325 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1326 | return NULL; |
1327 | ||
3adbefee IM |
1328 | zonelist = &NODE_DATA( |
1329 | slab_node(current->mempolicy))->node_zonelists[gfp_zone(flags)]; | |
81819f0f CL |
1330 | for (z = zonelist->zones; *z; z++) { |
1331 | struct kmem_cache_node *n; | |
1332 | ||
1333 | n = get_node(s, zone_to_nid(*z)); | |
1334 | ||
1335 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1336 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1337 | page = get_partial_node(n); |
1338 | if (page) | |
1339 | return page; | |
1340 | } | |
1341 | } | |
1342 | #endif | |
1343 | return NULL; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Get a partial page, lock it and return it. | |
1348 | */ | |
1349 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1350 | { | |
1351 | struct page *page; | |
1352 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1353 | ||
1354 | page = get_partial_node(get_node(s, searchnode)); | |
1355 | if (page || (flags & __GFP_THISNODE)) | |
1356 | return page; | |
1357 | ||
1358 | return get_any_partial(s, flags); | |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * Move a page back to the lists. | |
1363 | * | |
1364 | * Must be called with the slab lock held. | |
1365 | * | |
1366 | * On exit the slab lock will have been dropped. | |
1367 | */ | |
7c2e132c | 1368 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1369 | { |
e95eed57 | 1370 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1371 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1372 | |
4b6f0750 | 1373 | ClearSlabFrozen(page); |
81819f0f | 1374 | if (page->inuse) { |
e95eed57 | 1375 | |
8ff12cfc | 1376 | if (page->freelist != page->end) { |
7c2e132c | 1377 | add_partial(n, page, tail); |
8ff12cfc CL |
1378 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1379 | } else { | |
1380 | stat(c, DEACTIVATE_FULL); | |
1381 | if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) | |
1382 | add_full(n, page); | |
1383 | } | |
81819f0f CL |
1384 | slab_unlock(page); |
1385 | } else { | |
8ff12cfc | 1386 | stat(c, DEACTIVATE_EMPTY); |
e95eed57 CL |
1387 | if (n->nr_partial < MIN_PARTIAL) { |
1388 | /* | |
672bba3a CL |
1389 | * Adding an empty slab to the partial slabs in order |
1390 | * to avoid page allocator overhead. This slab needs | |
1391 | * to come after the other slabs with objects in | |
1392 | * order to fill them up. That way the size of the | |
1393 | * partial list stays small. kmem_cache_shrink can | |
1394 | * reclaim empty slabs from the partial list. | |
e95eed57 | 1395 | */ |
7c2e132c | 1396 | add_partial(n, page, 1); |
e95eed57 CL |
1397 | slab_unlock(page); |
1398 | } else { | |
1399 | slab_unlock(page); | |
8ff12cfc | 1400 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1401 | discard_slab(s, page); |
1402 | } | |
81819f0f CL |
1403 | } |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * Remove the cpu slab | |
1408 | */ | |
dfb4f096 | 1409 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1410 | { |
dfb4f096 | 1411 | struct page *page = c->page; |
7c2e132c | 1412 | int tail = 1; |
8ff12cfc CL |
1413 | |
1414 | if (c->freelist) | |
1415 | stat(c, DEACTIVATE_REMOTE_FREES); | |
894b8788 CL |
1416 | /* |
1417 | * Merge cpu freelist into freelist. Typically we get here | |
1418 | * because both freelists are empty. So this is unlikely | |
1419 | * to occur. | |
683d0baa CL |
1420 | * |
1421 | * We need to use _is_end here because deactivate slab may | |
1422 | * be called for a debug slab. Then c->freelist may contain | |
1423 | * a dummy pointer. | |
894b8788 | 1424 | */ |
683d0baa | 1425 | while (unlikely(!is_end(c->freelist))) { |
894b8788 CL |
1426 | void **object; |
1427 | ||
7c2e132c CL |
1428 | tail = 0; /* Hot objects. Put the slab first */ |
1429 | ||
894b8788 | 1430 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1431 | object = c->freelist; |
b3fba8da | 1432 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1433 | |
1434 | /* And put onto the regular freelist */ | |
b3fba8da | 1435 | object[c->offset] = page->freelist; |
894b8788 CL |
1436 | page->freelist = object; |
1437 | page->inuse--; | |
1438 | } | |
dfb4f096 | 1439 | c->page = NULL; |
7c2e132c | 1440 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1441 | } |
1442 | ||
dfb4f096 | 1443 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1444 | { |
8ff12cfc | 1445 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1446 | slab_lock(c->page); |
1447 | deactivate_slab(s, c); | |
81819f0f CL |
1448 | } |
1449 | ||
1450 | /* | |
1451 | * Flush cpu slab. | |
1452 | * Called from IPI handler with interrupts disabled. | |
1453 | */ | |
0c710013 | 1454 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1455 | { |
dfb4f096 | 1456 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1457 | |
dfb4f096 CL |
1458 | if (likely(c && c->page)) |
1459 | flush_slab(s, c); | |
81819f0f CL |
1460 | } |
1461 | ||
1462 | static void flush_cpu_slab(void *d) | |
1463 | { | |
1464 | struct kmem_cache *s = d; | |
81819f0f | 1465 | |
dfb4f096 | 1466 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1467 | } |
1468 | ||
1469 | static void flush_all(struct kmem_cache *s) | |
1470 | { | |
1471 | #ifdef CONFIG_SMP | |
1472 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1473 | #else | |
1474 | unsigned long flags; | |
1475 | ||
1476 | local_irq_save(flags); | |
1477 | flush_cpu_slab(s); | |
1478 | local_irq_restore(flags); | |
1479 | #endif | |
1480 | } | |
1481 | ||
dfb4f096 CL |
1482 | /* |
1483 | * Check if the objects in a per cpu structure fit numa | |
1484 | * locality expectations. | |
1485 | */ | |
1486 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1487 | { | |
1488 | #ifdef CONFIG_NUMA | |
1489 | if (node != -1 && c->node != node) | |
1490 | return 0; | |
1491 | #endif | |
1492 | return 1; | |
1493 | } | |
1494 | ||
81819f0f | 1495 | /* |
894b8788 CL |
1496 | * Slow path. The lockless freelist is empty or we need to perform |
1497 | * debugging duties. | |
1498 | * | |
1499 | * Interrupts are disabled. | |
81819f0f | 1500 | * |
894b8788 CL |
1501 | * Processing is still very fast if new objects have been freed to the |
1502 | * regular freelist. In that case we simply take over the regular freelist | |
1503 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1504 | * |
894b8788 CL |
1505 | * If that is not working then we fall back to the partial lists. We take the |
1506 | * first element of the freelist as the object to allocate now and move the | |
1507 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1508 | * |
894b8788 CL |
1509 | * And if we were unable to get a new slab from the partial slab lists then |
1510 | * we need to allocate a new slab. This is slowest path since we may sleep. | |
81819f0f | 1511 | */ |
894b8788 | 1512 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1513 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1514 | { |
81819f0f | 1515 | void **object; |
dfb4f096 | 1516 | struct page *new; |
1f84260c CL |
1517 | #ifdef SLUB_FASTPATH |
1518 | unsigned long flags; | |
81819f0f | 1519 | |
1f84260c CL |
1520 | local_irq_save(flags); |
1521 | #endif | |
dfb4f096 | 1522 | if (!c->page) |
81819f0f CL |
1523 | goto new_slab; |
1524 | ||
dfb4f096 CL |
1525 | slab_lock(c->page); |
1526 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1527 | goto another_slab; |
8ff12cfc | 1528 | stat(c, ALLOC_REFILL); |
894b8788 | 1529 | load_freelist: |
dfb4f096 | 1530 | object = c->page->freelist; |
683d0baa | 1531 | if (unlikely(object == c->page->end)) |
81819f0f | 1532 | goto another_slab; |
dfb4f096 | 1533 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1534 | goto debug; |
1535 | ||
dfb4f096 | 1536 | object = c->page->freelist; |
b3fba8da | 1537 | c->freelist = object[c->offset]; |
dfb4f096 | 1538 | c->page->inuse = s->objects; |
683d0baa | 1539 | c->page->freelist = c->page->end; |
dfb4f096 | 1540 | c->node = page_to_nid(c->page); |
1f84260c | 1541 | unlock_out: |
dfb4f096 | 1542 | slab_unlock(c->page); |
8ff12cfc | 1543 | stat(c, ALLOC_SLOWPATH); |
1f84260c CL |
1544 | #ifdef SLUB_FASTPATH |
1545 | local_irq_restore(flags); | |
1546 | #endif | |
81819f0f CL |
1547 | return object; |
1548 | ||
1549 | another_slab: | |
dfb4f096 | 1550 | deactivate_slab(s, c); |
81819f0f CL |
1551 | |
1552 | new_slab: | |
dfb4f096 CL |
1553 | new = get_partial(s, gfpflags, node); |
1554 | if (new) { | |
1555 | c->page = new; | |
8ff12cfc | 1556 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1557 | goto load_freelist; |
81819f0f CL |
1558 | } |
1559 | ||
b811c202 CL |
1560 | if (gfpflags & __GFP_WAIT) |
1561 | local_irq_enable(); | |
1562 | ||
dfb4f096 | 1563 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1564 | |
1565 | if (gfpflags & __GFP_WAIT) | |
1566 | local_irq_disable(); | |
1567 | ||
dfb4f096 CL |
1568 | if (new) { |
1569 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1570 | stat(c, ALLOC_SLAB); |
05aa3450 | 1571 | if (c->page) |
dfb4f096 | 1572 | flush_slab(s, c); |
dfb4f096 CL |
1573 | slab_lock(new); |
1574 | SetSlabFrozen(new); | |
1575 | c->page = new; | |
4b6f0750 | 1576 | goto load_freelist; |
81819f0f | 1577 | } |
71c7a06f CL |
1578 | #ifdef SLUB_FASTPATH |
1579 | local_irq_restore(flags); | |
1580 | #endif | |
1581 | /* | |
1582 | * No memory available. | |
1583 | * | |
1584 | * If the slab uses higher order allocs but the object is | |
1585 | * smaller than a page size then we can fallback in emergencies | |
1586 | * to the page allocator via kmalloc_large. The page allocator may | |
1587 | * have failed to obtain a higher order page and we can try to | |
1588 | * allocate a single page if the object fits into a single page. | |
1589 | * That is only possible if certain conditions are met that are being | |
1590 | * checked when a slab is created. | |
1591 | */ | |
1592 | if (!(gfpflags & __GFP_NORETRY) && (s->flags & __PAGE_ALLOC_FALLBACK)) | |
1593 | return kmalloc_large(s->objsize, gfpflags); | |
1594 | ||
1595 | return NULL; | |
81819f0f | 1596 | debug: |
dfb4f096 CL |
1597 | object = c->page->freelist; |
1598 | if (!alloc_debug_processing(s, c->page, object, addr)) | |
81819f0f | 1599 | goto another_slab; |
894b8788 | 1600 | |
dfb4f096 | 1601 | c->page->inuse++; |
b3fba8da | 1602 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1603 | c->node = -1; |
1f84260c | 1604 | goto unlock_out; |
894b8788 CL |
1605 | } |
1606 | ||
1607 | /* | |
1608 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1609 | * have the fastpath folded into their functions. So no function call | |
1610 | * overhead for requests that can be satisfied on the fastpath. | |
1611 | * | |
1612 | * The fastpath works by first checking if the lockless freelist can be used. | |
1613 | * If not then __slab_alloc is called for slow processing. | |
1614 | * | |
1615 | * Otherwise we can simply pick the next object from the lockless free list. | |
1616 | */ | |
06428780 | 1617 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce15fea8 | 1618 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1619 | { |
894b8788 | 1620 | void **object; |
dfb4f096 | 1621 | struct kmem_cache_cpu *c; |
894b8788 | 1622 | |
1f84260c CL |
1623 | /* |
1624 | * The SLUB_FASTPATH path is provisional and is currently disabled if the | |
1625 | * kernel is compiled with preemption or if the arch does not support | |
1626 | * fast cmpxchg operations. There are a couple of coming changes that will | |
1627 | * simplify matters and allow preemption. Ultimately we may end up making | |
1628 | * SLUB_FASTPATH the default. | |
1629 | * | |
1630 | * 1. The introduction of the per cpu allocator will avoid array lookups | |
1631 | * through get_cpu_slab(). A special register can be used instead. | |
1632 | * | |
1633 | * 2. The introduction of per cpu atomic operations (cpu_ops) means that | |
1634 | * we can realize the logic here entirely with per cpu atomics. The | |
1635 | * per cpu atomic ops will take care of the preemption issues. | |
1636 | */ | |
1637 | ||
1638 | #ifdef SLUB_FASTPATH | |
1639 | c = get_cpu_slab(s, raw_smp_processor_id()); | |
1640 | do { | |
1641 | object = c->freelist; | |
1642 | if (unlikely(is_end(object) || !node_match(c, node))) { | |
1643 | object = __slab_alloc(s, gfpflags, node, addr, c); | |
1644 | break; | |
1645 | } | |
8ff12cfc | 1646 | stat(c, ALLOC_FASTPATH); |
1f84260c CL |
1647 | } while (cmpxchg_local(&c->freelist, object, object[c->offset]) |
1648 | != object); | |
1649 | #else | |
1650 | unsigned long flags; | |
1651 | ||
894b8788 | 1652 | local_irq_save(flags); |
dfb4f096 | 1653 | c = get_cpu_slab(s, smp_processor_id()); |
683d0baa | 1654 | if (unlikely(is_end(c->freelist) || !node_match(c, node))) |
894b8788 | 1655 | |
dfb4f096 | 1656 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1657 | |
1658 | else { | |
dfb4f096 | 1659 | object = c->freelist; |
b3fba8da | 1660 | c->freelist = object[c->offset]; |
8ff12cfc | 1661 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1662 | } |
1663 | local_irq_restore(flags); | |
1f84260c | 1664 | #endif |
d07dbea4 CL |
1665 | |
1666 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
42a9fdbb | 1667 | memset(object, 0, c->objsize); |
d07dbea4 | 1668 | |
894b8788 | 1669 | return object; |
81819f0f CL |
1670 | } |
1671 | ||
1672 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1673 | { | |
ce15fea8 | 1674 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1675 | } |
1676 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1677 | ||
1678 | #ifdef CONFIG_NUMA | |
1679 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1680 | { | |
ce15fea8 | 1681 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1682 | } |
1683 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1684 | #endif | |
1685 | ||
1686 | /* | |
894b8788 CL |
1687 | * Slow patch handling. This may still be called frequently since objects |
1688 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1689 | * |
894b8788 CL |
1690 | * So we still attempt to reduce cache line usage. Just take the slab |
1691 | * lock and free the item. If there is no additional partial page | |
1692 | * handling required then we can return immediately. | |
81819f0f | 1693 | */ |
894b8788 | 1694 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1695 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1696 | { |
1697 | void *prior; | |
1698 | void **object = (void *)x; | |
8ff12cfc | 1699 | struct kmem_cache_cpu *c; |
81819f0f | 1700 | |
1f84260c CL |
1701 | #ifdef SLUB_FASTPATH |
1702 | unsigned long flags; | |
1703 | ||
1704 | local_irq_save(flags); | |
1705 | #endif | |
8ff12cfc CL |
1706 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1707 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1708 | slab_lock(page); |
1709 | ||
35e5d7ee | 1710 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1711 | goto debug; |
1712 | checks_ok: | |
b3fba8da | 1713 | prior = object[offset] = page->freelist; |
81819f0f CL |
1714 | page->freelist = object; |
1715 | page->inuse--; | |
1716 | ||
8ff12cfc CL |
1717 | if (unlikely(SlabFrozen(page))) { |
1718 | stat(c, FREE_FROZEN); | |
81819f0f | 1719 | goto out_unlock; |
8ff12cfc | 1720 | } |
81819f0f CL |
1721 | |
1722 | if (unlikely(!page->inuse)) | |
1723 | goto slab_empty; | |
1724 | ||
1725 | /* | |
1726 | * Objects left in the slab. If it | |
1727 | * was not on the partial list before | |
1728 | * then add it. | |
1729 | */ | |
8ff12cfc | 1730 | if (unlikely(prior == page->end)) { |
7c2e132c | 1731 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1732 | stat(c, FREE_ADD_PARTIAL); |
1733 | } | |
81819f0f CL |
1734 | |
1735 | out_unlock: | |
1736 | slab_unlock(page); | |
1f84260c CL |
1737 | #ifdef SLUB_FASTPATH |
1738 | local_irq_restore(flags); | |
1739 | #endif | |
81819f0f CL |
1740 | return; |
1741 | ||
1742 | slab_empty: | |
8ff12cfc | 1743 | if (prior != page->end) { |
81819f0f | 1744 | /* |
672bba3a | 1745 | * Slab still on the partial list. |
81819f0f CL |
1746 | */ |
1747 | remove_partial(s, page); | |
8ff12cfc CL |
1748 | stat(c, FREE_REMOVE_PARTIAL); |
1749 | } | |
81819f0f | 1750 | slab_unlock(page); |
8ff12cfc | 1751 | stat(c, FREE_SLAB); |
1f84260c CL |
1752 | #ifdef SLUB_FASTPATH |
1753 | local_irq_restore(flags); | |
1754 | #endif | |
81819f0f | 1755 | discard_slab(s, page); |
81819f0f CL |
1756 | return; |
1757 | ||
1758 | debug: | |
3ec09742 | 1759 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1760 | goto out_unlock; |
77c5e2d0 | 1761 | goto checks_ok; |
81819f0f CL |
1762 | } |
1763 | ||
894b8788 CL |
1764 | /* |
1765 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1766 | * can perform fastpath freeing without additional function calls. | |
1767 | * | |
1768 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1769 | * of this processor. This typically the case if we have just allocated | |
1770 | * the item before. | |
1771 | * | |
1772 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1773 | * with all sorts of special processing. | |
1774 | */ | |
06428780 | 1775 | static __always_inline void slab_free(struct kmem_cache *s, |
894b8788 CL |
1776 | struct page *page, void *x, void *addr) |
1777 | { | |
1778 | void **object = (void *)x; | |
dfb4f096 | 1779 | struct kmem_cache_cpu *c; |
894b8788 | 1780 | |
1f84260c CL |
1781 | #ifdef SLUB_FASTPATH |
1782 | void **freelist; | |
1783 | ||
1784 | c = get_cpu_slab(s, raw_smp_processor_id()); | |
1785 | debug_check_no_locks_freed(object, s->objsize); | |
1786 | do { | |
1787 | freelist = c->freelist; | |
1788 | barrier(); | |
1789 | /* | |
1790 | * If the compiler would reorder the retrieval of c->page to | |
1791 | * come before c->freelist then an interrupt could | |
1792 | * change the cpu slab before we retrieve c->freelist. We | |
1793 | * could be matching on a page no longer active and put the | |
1794 | * object onto the freelist of the wrong slab. | |
1795 | * | |
1796 | * On the other hand: If we already have the freelist pointer | |
1797 | * then any change of cpu_slab will cause the cmpxchg to fail | |
1798 | * since the freelist pointers are unique per slab. | |
1799 | */ | |
1800 | if (unlikely(page != c->page || c->node < 0)) { | |
1801 | __slab_free(s, page, x, addr, c->offset); | |
1802 | break; | |
1803 | } | |
1804 | object[c->offset] = freelist; | |
8ff12cfc | 1805 | stat(c, FREE_FASTPATH); |
1f84260c CL |
1806 | } while (cmpxchg_local(&c->freelist, freelist, object) != freelist); |
1807 | #else | |
1808 | unsigned long flags; | |
1809 | ||
894b8788 | 1810 | local_irq_save(flags); |
02febdf7 | 1811 | debug_check_no_locks_freed(object, s->objsize); |
dfb4f096 | 1812 | c = get_cpu_slab(s, smp_processor_id()); |
ee3c72a1 | 1813 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1814 | object[c->offset] = c->freelist; |
dfb4f096 | 1815 | c->freelist = object; |
8ff12cfc | 1816 | stat(c, FREE_FASTPATH); |
894b8788 | 1817 | } else |
b3fba8da | 1818 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1819 | |
1820 | local_irq_restore(flags); | |
1f84260c | 1821 | #endif |
894b8788 CL |
1822 | } |
1823 | ||
81819f0f CL |
1824 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1825 | { | |
77c5e2d0 | 1826 | struct page *page; |
81819f0f | 1827 | |
b49af68f | 1828 | page = virt_to_head_page(x); |
81819f0f | 1829 | |
77c5e2d0 | 1830 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1831 | } |
1832 | EXPORT_SYMBOL(kmem_cache_free); | |
1833 | ||
1834 | /* Figure out on which slab object the object resides */ | |
1835 | static struct page *get_object_page(const void *x) | |
1836 | { | |
b49af68f | 1837 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1838 | |
1839 | if (!PageSlab(page)) | |
1840 | return NULL; | |
1841 | ||
1842 | return page; | |
1843 | } | |
1844 | ||
1845 | /* | |
672bba3a CL |
1846 | * Object placement in a slab is made very easy because we always start at |
1847 | * offset 0. If we tune the size of the object to the alignment then we can | |
1848 | * get the required alignment by putting one properly sized object after | |
1849 | * another. | |
81819f0f CL |
1850 | * |
1851 | * Notice that the allocation order determines the sizes of the per cpu | |
1852 | * caches. Each processor has always one slab available for allocations. | |
1853 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1854 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1855 | * locking overhead. |
81819f0f CL |
1856 | */ |
1857 | ||
1858 | /* | |
1859 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1860 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1861 | * and increases the number of allocations possible without having to | |
1862 | * take the list_lock. | |
1863 | */ | |
1864 | static int slub_min_order; | |
1865 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1866 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1867 | ||
1868 | /* | |
1869 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1870 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1871 | */ |
1872 | static int slub_nomerge; | |
1873 | ||
81819f0f CL |
1874 | /* |
1875 | * Calculate the order of allocation given an slab object size. | |
1876 | * | |
672bba3a CL |
1877 | * The order of allocation has significant impact on performance and other |
1878 | * system components. Generally order 0 allocations should be preferred since | |
1879 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1880 | * be problematic to put into order 0 slabs because there may be too much | |
1881 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1882 | * would be wasted. | |
1883 | * | |
1884 | * In order to reach satisfactory performance we must ensure that a minimum | |
1885 | * number of objects is in one slab. Otherwise we may generate too much | |
1886 | * activity on the partial lists which requires taking the list_lock. This is | |
1887 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1888 | * |
672bba3a CL |
1889 | * slub_max_order specifies the order where we begin to stop considering the |
1890 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1891 | * we try to keep the page order as low as possible. So we accept more waste | |
1892 | * of space in favor of a small page order. | |
81819f0f | 1893 | * |
672bba3a CL |
1894 | * Higher order allocations also allow the placement of more objects in a |
1895 | * slab and thereby reduce object handling overhead. If the user has | |
1896 | * requested a higher mininum order then we start with that one instead of | |
1897 | * the smallest order which will fit the object. | |
81819f0f | 1898 | */ |
5e6d444e CL |
1899 | static inline int slab_order(int size, int min_objects, |
1900 | int max_order, int fract_leftover) | |
81819f0f CL |
1901 | { |
1902 | int order; | |
1903 | int rem; | |
6300ea75 | 1904 | int min_order = slub_min_order; |
81819f0f | 1905 | |
6300ea75 | 1906 | for (order = max(min_order, |
5e6d444e CL |
1907 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1908 | order <= max_order; order++) { | |
81819f0f | 1909 | |
5e6d444e | 1910 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1911 | |
5e6d444e | 1912 | if (slab_size < min_objects * size) |
81819f0f CL |
1913 | continue; |
1914 | ||
1915 | rem = slab_size % size; | |
1916 | ||
5e6d444e | 1917 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1918 | break; |
1919 | ||
1920 | } | |
672bba3a | 1921 | |
81819f0f CL |
1922 | return order; |
1923 | } | |
1924 | ||
5e6d444e CL |
1925 | static inline int calculate_order(int size) |
1926 | { | |
1927 | int order; | |
1928 | int min_objects; | |
1929 | int fraction; | |
1930 | ||
1931 | /* | |
1932 | * Attempt to find best configuration for a slab. This | |
1933 | * works by first attempting to generate a layout with | |
1934 | * the best configuration and backing off gradually. | |
1935 | * | |
1936 | * First we reduce the acceptable waste in a slab. Then | |
1937 | * we reduce the minimum objects required in a slab. | |
1938 | */ | |
1939 | min_objects = slub_min_objects; | |
1940 | while (min_objects > 1) { | |
1941 | fraction = 8; | |
1942 | while (fraction >= 4) { | |
1943 | order = slab_order(size, min_objects, | |
1944 | slub_max_order, fraction); | |
1945 | if (order <= slub_max_order) | |
1946 | return order; | |
1947 | fraction /= 2; | |
1948 | } | |
1949 | min_objects /= 2; | |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * We were unable to place multiple objects in a slab. Now | |
1954 | * lets see if we can place a single object there. | |
1955 | */ | |
1956 | order = slab_order(size, 1, slub_max_order, 1); | |
1957 | if (order <= slub_max_order) | |
1958 | return order; | |
1959 | ||
1960 | /* | |
1961 | * Doh this slab cannot be placed using slub_max_order. | |
1962 | */ | |
1963 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1964 | if (order <= MAX_ORDER) | |
1965 | return order; | |
1966 | return -ENOSYS; | |
1967 | } | |
1968 | ||
81819f0f | 1969 | /* |
672bba3a | 1970 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1971 | */ |
1972 | static unsigned long calculate_alignment(unsigned long flags, | |
1973 | unsigned long align, unsigned long size) | |
1974 | { | |
1975 | /* | |
1976 | * If the user wants hardware cache aligned objects then | |
1977 | * follow that suggestion if the object is sufficiently | |
1978 | * large. | |
1979 | * | |
1980 | * The hardware cache alignment cannot override the | |
1981 | * specified alignment though. If that is greater | |
1982 | * then use it. | |
1983 | */ | |
5af60839 | 1984 | if ((flags & SLAB_HWCACHE_ALIGN) && |
65c02d4c CL |
1985 | size > cache_line_size() / 2) |
1986 | return max_t(unsigned long, align, cache_line_size()); | |
81819f0f CL |
1987 | |
1988 | if (align < ARCH_SLAB_MINALIGN) | |
1989 | return ARCH_SLAB_MINALIGN; | |
1990 | ||
1991 | return ALIGN(align, sizeof(void *)); | |
1992 | } | |
1993 | ||
dfb4f096 CL |
1994 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1995 | struct kmem_cache_cpu *c) | |
1996 | { | |
1997 | c->page = NULL; | |
683d0baa | 1998 | c->freelist = (void *)PAGE_MAPPING_ANON; |
dfb4f096 | 1999 | c->node = 0; |
42a9fdbb CL |
2000 | c->offset = s->offset / sizeof(void *); |
2001 | c->objsize = s->objsize; | |
dfb4f096 CL |
2002 | } |
2003 | ||
81819f0f CL |
2004 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
2005 | { | |
2006 | n->nr_partial = 0; | |
2007 | atomic_long_set(&n->nr_slabs, 0); | |
2008 | spin_lock_init(&n->list_lock); | |
2009 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2010 | #ifdef CONFIG_SLUB_DEBUG |
643b1138 | 2011 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2012 | #endif |
81819f0f CL |
2013 | } |
2014 | ||
4c93c355 CL |
2015 | #ifdef CONFIG_SMP |
2016 | /* | |
2017 | * Per cpu array for per cpu structures. | |
2018 | * | |
2019 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
2020 | * close together meaning that it becomes possible that multiple per cpu | |
2021 | * structures are contained in one cacheline. This may be particularly | |
2022 | * beneficial for the kmalloc caches. | |
2023 | * | |
2024 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
2025 | * likely able to get per cpu structures for all caches from the array defined | |
2026 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2027 | * | |
2028 | * If the per cpu array is exhausted then fall back to kmalloc | |
2029 | * of individual cachelines. No sharing is possible then. | |
2030 | */ | |
2031 | #define NR_KMEM_CACHE_CPU 100 | |
2032 | ||
2033 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2034 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2035 | ||
2036 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
2037 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
2038 | ||
2039 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2040 | int cpu, gfp_t flags) | |
2041 | { | |
2042 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2043 | ||
2044 | if (c) | |
2045 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2046 | (void *)c->freelist; | |
2047 | else { | |
2048 | /* Table overflow: So allocate ourselves */ | |
2049 | c = kmalloc_node( | |
2050 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2051 | flags, cpu_to_node(cpu)); | |
2052 | if (!c) | |
2053 | return NULL; | |
2054 | } | |
2055 | ||
2056 | init_kmem_cache_cpu(s, c); | |
2057 | return c; | |
2058 | } | |
2059 | ||
2060 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2061 | { | |
2062 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
2063 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
2064 | kfree(c); | |
2065 | return; | |
2066 | } | |
2067 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2068 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2069 | } | |
2070 | ||
2071 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2072 | { | |
2073 | int cpu; | |
2074 | ||
2075 | for_each_online_cpu(cpu) { | |
2076 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2077 | ||
2078 | if (c) { | |
2079 | s->cpu_slab[cpu] = NULL; | |
2080 | free_kmem_cache_cpu(c, cpu); | |
2081 | } | |
2082 | } | |
2083 | } | |
2084 | ||
2085 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2086 | { | |
2087 | int cpu; | |
2088 | ||
2089 | for_each_online_cpu(cpu) { | |
2090 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2091 | ||
2092 | if (c) | |
2093 | continue; | |
2094 | ||
2095 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2096 | if (!c) { | |
2097 | free_kmem_cache_cpus(s); | |
2098 | return 0; | |
2099 | } | |
2100 | s->cpu_slab[cpu] = c; | |
2101 | } | |
2102 | return 1; | |
2103 | } | |
2104 | ||
2105 | /* | |
2106 | * Initialize the per cpu array. | |
2107 | */ | |
2108 | static void init_alloc_cpu_cpu(int cpu) | |
2109 | { | |
2110 | int i; | |
2111 | ||
2112 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
2113 | return; | |
2114 | ||
2115 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2116 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2117 | ||
2118 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
2119 | } | |
2120 | ||
2121 | static void __init init_alloc_cpu(void) | |
2122 | { | |
2123 | int cpu; | |
2124 | ||
2125 | for_each_online_cpu(cpu) | |
2126 | init_alloc_cpu_cpu(cpu); | |
2127 | } | |
2128 | ||
2129 | #else | |
2130 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2131 | static inline void init_alloc_cpu(void) {} | |
2132 | ||
2133 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2134 | { | |
2135 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2136 | return 1; | |
2137 | } | |
2138 | #endif | |
2139 | ||
81819f0f CL |
2140 | #ifdef CONFIG_NUMA |
2141 | /* | |
2142 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2143 | * slab on the node for this slabcache. There are no concurrent accesses | |
2144 | * possible. | |
2145 | * | |
2146 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2147 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2148 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2149 | */ |
1cd7daa5 AB |
2150 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2151 | int node) | |
81819f0f CL |
2152 | { |
2153 | struct page *page; | |
2154 | struct kmem_cache_node *n; | |
ba84c73c | 2155 | unsigned long flags; |
81819f0f CL |
2156 | |
2157 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2158 | ||
a2f92ee7 | 2159 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2160 | |
2161 | BUG_ON(!page); | |
a2f92ee7 CL |
2162 | if (page_to_nid(page) != node) { |
2163 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2164 | "node %d\n", node); | |
2165 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2166 | "in order to be able to continue\n"); | |
2167 | } | |
2168 | ||
81819f0f CL |
2169 | n = page->freelist; |
2170 | BUG_ON(!n); | |
2171 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2172 | page->inuse++; | |
2173 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2174 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2175 | init_object(kmalloc_caches, n, 1); |
2176 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2177 | #endif |
81819f0f CL |
2178 | init_kmem_cache_node(n); |
2179 | atomic_long_inc(&n->nr_slabs); | |
ba84c73c | 2180 | /* |
2181 | * lockdep requires consistent irq usage for each lock | |
2182 | * so even though there cannot be a race this early in | |
2183 | * the boot sequence, we still disable irqs. | |
2184 | */ | |
2185 | local_irq_save(flags); | |
7c2e132c | 2186 | add_partial(n, page, 0); |
ba84c73c | 2187 | local_irq_restore(flags); |
81819f0f CL |
2188 | return n; |
2189 | } | |
2190 | ||
2191 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2192 | { | |
2193 | int node; | |
2194 | ||
f64dc58c | 2195 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2196 | struct kmem_cache_node *n = s->node[node]; |
2197 | if (n && n != &s->local_node) | |
2198 | kmem_cache_free(kmalloc_caches, n); | |
2199 | s->node[node] = NULL; | |
2200 | } | |
2201 | } | |
2202 | ||
2203 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2204 | { | |
2205 | int node; | |
2206 | int local_node; | |
2207 | ||
2208 | if (slab_state >= UP) | |
2209 | local_node = page_to_nid(virt_to_page(s)); | |
2210 | else | |
2211 | local_node = 0; | |
2212 | ||
f64dc58c | 2213 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2214 | struct kmem_cache_node *n; |
2215 | ||
2216 | if (local_node == node) | |
2217 | n = &s->local_node; | |
2218 | else { | |
2219 | if (slab_state == DOWN) { | |
2220 | n = early_kmem_cache_node_alloc(gfpflags, | |
2221 | node); | |
2222 | continue; | |
2223 | } | |
2224 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2225 | gfpflags, node); | |
2226 | ||
2227 | if (!n) { | |
2228 | free_kmem_cache_nodes(s); | |
2229 | return 0; | |
2230 | } | |
2231 | ||
2232 | } | |
2233 | s->node[node] = n; | |
2234 | init_kmem_cache_node(n); | |
2235 | } | |
2236 | return 1; | |
2237 | } | |
2238 | #else | |
2239 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2240 | { | |
2241 | } | |
2242 | ||
2243 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2244 | { | |
2245 | init_kmem_cache_node(&s->local_node); | |
2246 | return 1; | |
2247 | } | |
2248 | #endif | |
2249 | ||
2250 | /* | |
2251 | * calculate_sizes() determines the order and the distribution of data within | |
2252 | * a slab object. | |
2253 | */ | |
2254 | static int calculate_sizes(struct kmem_cache *s) | |
2255 | { | |
2256 | unsigned long flags = s->flags; | |
2257 | unsigned long size = s->objsize; | |
2258 | unsigned long align = s->align; | |
2259 | ||
2260 | /* | |
2261 | * Determine if we can poison the object itself. If the user of | |
2262 | * the slab may touch the object after free or before allocation | |
2263 | * then we should never poison the object itself. | |
2264 | */ | |
2265 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2266 | !s->ctor) |
81819f0f CL |
2267 | s->flags |= __OBJECT_POISON; |
2268 | else | |
2269 | s->flags &= ~__OBJECT_POISON; | |
2270 | ||
2271 | /* | |
2272 | * Round up object size to the next word boundary. We can only | |
2273 | * place the free pointer at word boundaries and this determines | |
2274 | * the possible location of the free pointer. | |
2275 | */ | |
2276 | size = ALIGN(size, sizeof(void *)); | |
2277 | ||
41ecc55b | 2278 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f | 2279 | /* |
672bba3a | 2280 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2281 | * end of the object and the free pointer. If not then add an |
672bba3a | 2282 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2283 | */ |
2284 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2285 | size += sizeof(void *); | |
41ecc55b | 2286 | #endif |
81819f0f CL |
2287 | |
2288 | /* | |
672bba3a CL |
2289 | * With that we have determined the number of bytes in actual use |
2290 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2291 | */ |
2292 | s->inuse = size; | |
2293 | ||
2294 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2295 | s->ctor)) { |
81819f0f CL |
2296 | /* |
2297 | * Relocate free pointer after the object if it is not | |
2298 | * permitted to overwrite the first word of the object on | |
2299 | * kmem_cache_free. | |
2300 | * | |
2301 | * This is the case if we do RCU, have a constructor or | |
2302 | * destructor or are poisoning the objects. | |
2303 | */ | |
2304 | s->offset = size; | |
2305 | size += sizeof(void *); | |
2306 | } | |
2307 | ||
c12b3c62 | 2308 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2309 | if (flags & SLAB_STORE_USER) |
2310 | /* | |
2311 | * Need to store information about allocs and frees after | |
2312 | * the object. | |
2313 | */ | |
2314 | size += 2 * sizeof(struct track); | |
2315 | ||
be7b3fbc | 2316 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2317 | /* |
2318 | * Add some empty padding so that we can catch | |
2319 | * overwrites from earlier objects rather than let | |
2320 | * tracking information or the free pointer be | |
2321 | * corrupted if an user writes before the start | |
2322 | * of the object. | |
2323 | */ | |
2324 | size += sizeof(void *); | |
41ecc55b | 2325 | #endif |
672bba3a | 2326 | |
81819f0f CL |
2327 | /* |
2328 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2329 | * user specified and the dynamic determination of cache line size |
2330 | * on bootup. | |
81819f0f CL |
2331 | */ |
2332 | align = calculate_alignment(flags, align, s->objsize); | |
2333 | ||
2334 | /* | |
2335 | * SLUB stores one object immediately after another beginning from | |
2336 | * offset 0. In order to align the objects we have to simply size | |
2337 | * each object to conform to the alignment. | |
2338 | */ | |
2339 | size = ALIGN(size, align); | |
2340 | s->size = size; | |
2341 | ||
71c7a06f CL |
2342 | if ((flags & __KMALLOC_CACHE) && |
2343 | PAGE_SIZE / size < slub_min_objects) { | |
2344 | /* | |
2345 | * Kmalloc cache that would not have enough objects in | |
2346 | * an order 0 page. Kmalloc slabs can fallback to | |
2347 | * page allocator order 0 allocs so take a reasonably large | |
2348 | * order that will allows us a good number of objects. | |
2349 | */ | |
2350 | s->order = max(slub_max_order, PAGE_ALLOC_COSTLY_ORDER); | |
2351 | s->flags |= __PAGE_ALLOC_FALLBACK; | |
2352 | s->allocflags |= __GFP_NOWARN; | |
2353 | } else | |
2354 | s->order = calculate_order(size); | |
2355 | ||
81819f0f CL |
2356 | if (s->order < 0) |
2357 | return 0; | |
2358 | ||
b7a49f0d CL |
2359 | s->allocflags = 0; |
2360 | if (s->order) | |
2361 | s->allocflags |= __GFP_COMP; | |
2362 | ||
2363 | if (s->flags & SLAB_CACHE_DMA) | |
2364 | s->allocflags |= SLUB_DMA; | |
2365 | ||
2366 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2367 | s->allocflags |= __GFP_RECLAIMABLE; | |
2368 | ||
81819f0f CL |
2369 | /* |
2370 | * Determine the number of objects per slab | |
2371 | */ | |
2372 | s->objects = (PAGE_SIZE << s->order) / size; | |
2373 | ||
b3fba8da | 2374 | return !!s->objects; |
81819f0f CL |
2375 | |
2376 | } | |
2377 | ||
81819f0f CL |
2378 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2379 | const char *name, size_t size, | |
2380 | size_t align, unsigned long flags, | |
4ba9b9d0 | 2381 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
2382 | { |
2383 | memset(s, 0, kmem_size); | |
2384 | s->name = name; | |
2385 | s->ctor = ctor; | |
81819f0f | 2386 | s->objsize = size; |
81819f0f | 2387 | s->align = align; |
ba0268a8 | 2388 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f CL |
2389 | |
2390 | if (!calculate_sizes(s)) | |
2391 | goto error; | |
2392 | ||
2393 | s->refcount = 1; | |
2394 | #ifdef CONFIG_NUMA | |
9824601e | 2395 | s->remote_node_defrag_ratio = 100; |
81819f0f | 2396 | #endif |
dfb4f096 CL |
2397 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2398 | goto error; | |
81819f0f | 2399 | |
dfb4f096 | 2400 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2401 | return 1; |
4c93c355 | 2402 | free_kmem_cache_nodes(s); |
81819f0f CL |
2403 | error: |
2404 | if (flags & SLAB_PANIC) | |
2405 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2406 | "order=%u offset=%u flags=%lx\n", | |
2407 | s->name, (unsigned long)size, s->size, s->order, | |
2408 | s->offset, flags); | |
2409 | return 0; | |
2410 | } | |
81819f0f CL |
2411 | |
2412 | /* | |
2413 | * Check if a given pointer is valid | |
2414 | */ | |
2415 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2416 | { | |
06428780 | 2417 | struct page *page; |
81819f0f CL |
2418 | |
2419 | page = get_object_page(object); | |
2420 | ||
2421 | if (!page || s != page->slab) | |
2422 | /* No slab or wrong slab */ | |
2423 | return 0; | |
2424 | ||
abcd08a6 | 2425 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2426 | return 0; |
2427 | ||
2428 | /* | |
2429 | * We could also check if the object is on the slabs freelist. | |
2430 | * But this would be too expensive and it seems that the main | |
2431 | * purpose of kmem_ptr_valid is to check if the object belongs | |
2432 | * to a certain slab. | |
2433 | */ | |
2434 | return 1; | |
2435 | } | |
2436 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2437 | ||
2438 | /* | |
2439 | * Determine the size of a slab object | |
2440 | */ | |
2441 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2442 | { | |
2443 | return s->objsize; | |
2444 | } | |
2445 | EXPORT_SYMBOL(kmem_cache_size); | |
2446 | ||
2447 | const char *kmem_cache_name(struct kmem_cache *s) | |
2448 | { | |
2449 | return s->name; | |
2450 | } | |
2451 | EXPORT_SYMBOL(kmem_cache_name); | |
2452 | ||
2453 | /* | |
672bba3a CL |
2454 | * Attempt to free all slabs on a node. Return the number of slabs we |
2455 | * were unable to free. | |
81819f0f CL |
2456 | */ |
2457 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
2458 | struct list_head *list) | |
2459 | { | |
2460 | int slabs_inuse = 0; | |
2461 | unsigned long flags; | |
2462 | struct page *page, *h; | |
2463 | ||
2464 | spin_lock_irqsave(&n->list_lock, flags); | |
2465 | list_for_each_entry_safe(page, h, list, lru) | |
2466 | if (!page->inuse) { | |
2467 | list_del(&page->lru); | |
2468 | discard_slab(s, page); | |
2469 | } else | |
2470 | slabs_inuse++; | |
2471 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2472 | return slabs_inuse; | |
2473 | } | |
2474 | ||
2475 | /* | |
672bba3a | 2476 | * Release all resources used by a slab cache. |
81819f0f | 2477 | */ |
0c710013 | 2478 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2479 | { |
2480 | int node; | |
2481 | ||
2482 | flush_all(s); | |
2483 | ||
2484 | /* Attempt to free all objects */ | |
4c93c355 | 2485 | free_kmem_cache_cpus(s); |
f64dc58c | 2486 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2487 | struct kmem_cache_node *n = get_node(s, node); |
2488 | ||
2086d26a | 2489 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
2490 | if (atomic_long_read(&n->nr_slabs)) |
2491 | return 1; | |
2492 | } | |
2493 | free_kmem_cache_nodes(s); | |
2494 | return 0; | |
2495 | } | |
2496 | ||
2497 | /* | |
2498 | * Close a cache and release the kmem_cache structure | |
2499 | * (must be used for caches created using kmem_cache_create) | |
2500 | */ | |
2501 | void kmem_cache_destroy(struct kmem_cache *s) | |
2502 | { | |
2503 | down_write(&slub_lock); | |
2504 | s->refcount--; | |
2505 | if (!s->refcount) { | |
2506 | list_del(&s->list); | |
a0e1d1be | 2507 | up_write(&slub_lock); |
81819f0f CL |
2508 | if (kmem_cache_close(s)) |
2509 | WARN_ON(1); | |
2510 | sysfs_slab_remove(s); | |
a0e1d1be CL |
2511 | } else |
2512 | up_write(&slub_lock); | |
81819f0f CL |
2513 | } |
2514 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2515 | ||
2516 | /******************************************************************** | |
2517 | * Kmalloc subsystem | |
2518 | *******************************************************************/ | |
2519 | ||
331dc558 | 2520 | struct kmem_cache kmalloc_caches[PAGE_SHIFT + 1] __cacheline_aligned; |
81819f0f CL |
2521 | EXPORT_SYMBOL(kmalloc_caches); |
2522 | ||
2523 | #ifdef CONFIG_ZONE_DMA | |
331dc558 | 2524 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT + 1]; |
81819f0f CL |
2525 | #endif |
2526 | ||
2527 | static int __init setup_slub_min_order(char *str) | |
2528 | { | |
06428780 | 2529 | get_option(&str, &slub_min_order); |
81819f0f CL |
2530 | |
2531 | return 1; | |
2532 | } | |
2533 | ||
2534 | __setup("slub_min_order=", setup_slub_min_order); | |
2535 | ||
2536 | static int __init setup_slub_max_order(char *str) | |
2537 | { | |
06428780 | 2538 | get_option(&str, &slub_max_order); |
81819f0f CL |
2539 | |
2540 | return 1; | |
2541 | } | |
2542 | ||
2543 | __setup("slub_max_order=", setup_slub_max_order); | |
2544 | ||
2545 | static int __init setup_slub_min_objects(char *str) | |
2546 | { | |
06428780 | 2547 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2548 | |
2549 | return 1; | |
2550 | } | |
2551 | ||
2552 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2553 | ||
2554 | static int __init setup_slub_nomerge(char *str) | |
2555 | { | |
2556 | slub_nomerge = 1; | |
2557 | return 1; | |
2558 | } | |
2559 | ||
2560 | __setup("slub_nomerge", setup_slub_nomerge); | |
2561 | ||
81819f0f CL |
2562 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2563 | const char *name, int size, gfp_t gfp_flags) | |
2564 | { | |
2565 | unsigned int flags = 0; | |
2566 | ||
2567 | if (gfp_flags & SLUB_DMA) | |
2568 | flags = SLAB_CACHE_DMA; | |
2569 | ||
2570 | down_write(&slub_lock); | |
2571 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
71c7a06f | 2572 | flags | __KMALLOC_CACHE, NULL)) |
81819f0f CL |
2573 | goto panic; |
2574 | ||
2575 | list_add(&s->list, &slab_caches); | |
2576 | up_write(&slub_lock); | |
2577 | if (sysfs_slab_add(s)) | |
2578 | goto panic; | |
2579 | return s; | |
2580 | ||
2581 | panic: | |
2582 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2583 | } | |
2584 | ||
2e443fd0 | 2585 | #ifdef CONFIG_ZONE_DMA |
1ceef402 CL |
2586 | |
2587 | static void sysfs_add_func(struct work_struct *w) | |
2588 | { | |
2589 | struct kmem_cache *s; | |
2590 | ||
2591 | down_write(&slub_lock); | |
2592 | list_for_each_entry(s, &slab_caches, list) { | |
2593 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2594 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2595 | sysfs_slab_add(s); | |
2596 | } | |
2597 | } | |
2598 | up_write(&slub_lock); | |
2599 | } | |
2600 | ||
2601 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2602 | ||
2e443fd0 CL |
2603 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2604 | { | |
2605 | struct kmem_cache *s; | |
2e443fd0 CL |
2606 | char *text; |
2607 | size_t realsize; | |
2608 | ||
2609 | s = kmalloc_caches_dma[index]; | |
2610 | if (s) | |
2611 | return s; | |
2612 | ||
2613 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2614 | if (flags & __GFP_WAIT) |
2615 | down_write(&slub_lock); | |
2616 | else { | |
2617 | if (!down_write_trylock(&slub_lock)) | |
2618 | goto out; | |
2619 | } | |
2620 | ||
2621 | if (kmalloc_caches_dma[index]) | |
2622 | goto unlock_out; | |
2e443fd0 | 2623 | |
7b55f620 | 2624 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2625 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2626 | (unsigned int)realsize); | |
1ceef402 CL |
2627 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2628 | ||
2629 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2630 | realsize, ARCH_KMALLOC_MINALIGN, | |
2631 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2632 | kfree(s); | |
2633 | kfree(text); | |
2634 | goto unlock_out; | |
dfce8648 | 2635 | } |
1ceef402 CL |
2636 | |
2637 | list_add(&s->list, &slab_caches); | |
2638 | kmalloc_caches_dma[index] = s; | |
2639 | ||
2640 | schedule_work(&sysfs_add_work); | |
2641 | ||
2642 | unlock_out: | |
dfce8648 | 2643 | up_write(&slub_lock); |
1ceef402 | 2644 | out: |
dfce8648 | 2645 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2646 | } |
2647 | #endif | |
2648 | ||
f1b26339 CL |
2649 | /* |
2650 | * Conversion table for small slabs sizes / 8 to the index in the | |
2651 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2652 | * of two cache sizes there. The size of larger slabs can be determined using | |
2653 | * fls. | |
2654 | */ | |
2655 | static s8 size_index[24] = { | |
2656 | 3, /* 8 */ | |
2657 | 4, /* 16 */ | |
2658 | 5, /* 24 */ | |
2659 | 5, /* 32 */ | |
2660 | 6, /* 40 */ | |
2661 | 6, /* 48 */ | |
2662 | 6, /* 56 */ | |
2663 | 6, /* 64 */ | |
2664 | 1, /* 72 */ | |
2665 | 1, /* 80 */ | |
2666 | 1, /* 88 */ | |
2667 | 1, /* 96 */ | |
2668 | 7, /* 104 */ | |
2669 | 7, /* 112 */ | |
2670 | 7, /* 120 */ | |
2671 | 7, /* 128 */ | |
2672 | 2, /* 136 */ | |
2673 | 2, /* 144 */ | |
2674 | 2, /* 152 */ | |
2675 | 2, /* 160 */ | |
2676 | 2, /* 168 */ | |
2677 | 2, /* 176 */ | |
2678 | 2, /* 184 */ | |
2679 | 2 /* 192 */ | |
2680 | }; | |
2681 | ||
81819f0f CL |
2682 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2683 | { | |
f1b26339 | 2684 | int index; |
81819f0f | 2685 | |
f1b26339 CL |
2686 | if (size <= 192) { |
2687 | if (!size) | |
2688 | return ZERO_SIZE_PTR; | |
81819f0f | 2689 | |
f1b26339 | 2690 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2691 | } else |
f1b26339 | 2692 | index = fls(size - 1); |
81819f0f CL |
2693 | |
2694 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2695 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2696 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2697 | |
81819f0f CL |
2698 | #endif |
2699 | return &kmalloc_caches[index]; | |
2700 | } | |
2701 | ||
2702 | void *__kmalloc(size_t size, gfp_t flags) | |
2703 | { | |
aadb4bc4 | 2704 | struct kmem_cache *s; |
81819f0f | 2705 | |
331dc558 | 2706 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2707 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2708 | |
2709 | s = get_slab(size, flags); | |
2710 | ||
2711 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2712 | return s; |
2713 | ||
ce15fea8 | 2714 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2715 | } |
2716 | EXPORT_SYMBOL(__kmalloc); | |
2717 | ||
2718 | #ifdef CONFIG_NUMA | |
2719 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2720 | { | |
aadb4bc4 | 2721 | struct kmem_cache *s; |
81819f0f | 2722 | |
331dc558 | 2723 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2724 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2725 | |
2726 | s = get_slab(size, flags); | |
2727 | ||
2728 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2729 | return s; |
2730 | ||
ce15fea8 | 2731 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2732 | } |
2733 | EXPORT_SYMBOL(__kmalloc_node); | |
2734 | #endif | |
2735 | ||
2736 | size_t ksize(const void *object) | |
2737 | { | |
272c1d21 | 2738 | struct page *page; |
81819f0f CL |
2739 | struct kmem_cache *s; |
2740 | ||
ef8b4520 CL |
2741 | BUG_ON(!object); |
2742 | if (unlikely(object == ZERO_SIZE_PTR)) | |
272c1d21 CL |
2743 | return 0; |
2744 | ||
294a80a8 | 2745 | page = virt_to_head_page(object); |
81819f0f | 2746 | BUG_ON(!page); |
294a80a8 VN |
2747 | |
2748 | if (unlikely(!PageSlab(page))) | |
2749 | return PAGE_SIZE << compound_order(page); | |
2750 | ||
81819f0f CL |
2751 | s = page->slab; |
2752 | BUG_ON(!s); | |
2753 | ||
2754 | /* | |
2755 | * Debugging requires use of the padding between object | |
2756 | * and whatever may come after it. | |
2757 | */ | |
2758 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2759 | return s->objsize; | |
2760 | ||
2761 | /* | |
2762 | * If we have the need to store the freelist pointer | |
2763 | * back there or track user information then we can | |
2764 | * only use the space before that information. | |
2765 | */ | |
2766 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2767 | return s->inuse; | |
2768 | ||
2769 | /* | |
2770 | * Else we can use all the padding etc for the allocation | |
2771 | */ | |
2772 | return s->size; | |
2773 | } | |
2774 | EXPORT_SYMBOL(ksize); | |
2775 | ||
2776 | void kfree(const void *x) | |
2777 | { | |
81819f0f | 2778 | struct page *page; |
5bb983b0 | 2779 | void *object = (void *)x; |
81819f0f | 2780 | |
2408c550 | 2781 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2782 | return; |
2783 | ||
b49af68f | 2784 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2785 | if (unlikely(!PageSlab(page))) { |
2786 | put_page(page); | |
2787 | return; | |
2788 | } | |
5bb983b0 | 2789 | slab_free(page->slab, page, object, __builtin_return_address(0)); |
81819f0f CL |
2790 | } |
2791 | EXPORT_SYMBOL(kfree); | |
2792 | ||
f61396ae CL |
2793 | static unsigned long count_partial(struct kmem_cache_node *n) |
2794 | { | |
2795 | unsigned long flags; | |
2796 | unsigned long x = 0; | |
2797 | struct page *page; | |
2798 | ||
2799 | spin_lock_irqsave(&n->list_lock, flags); | |
2800 | list_for_each_entry(page, &n->partial, lru) | |
2801 | x += page->inuse; | |
2802 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2803 | return x; | |
2804 | } | |
2805 | ||
2086d26a | 2806 | /* |
672bba3a CL |
2807 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2808 | * the remaining slabs by the number of items in use. The slabs with the | |
2809 | * most items in use come first. New allocations will then fill those up | |
2810 | * and thus they can be removed from the partial lists. | |
2811 | * | |
2812 | * The slabs with the least items are placed last. This results in them | |
2813 | * being allocated from last increasing the chance that the last objects | |
2814 | * are freed in them. | |
2086d26a CL |
2815 | */ |
2816 | int kmem_cache_shrink(struct kmem_cache *s) | |
2817 | { | |
2818 | int node; | |
2819 | int i; | |
2820 | struct kmem_cache_node *n; | |
2821 | struct page *page; | |
2822 | struct page *t; | |
2823 | struct list_head *slabs_by_inuse = | |
2824 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2825 | unsigned long flags; | |
2826 | ||
2827 | if (!slabs_by_inuse) | |
2828 | return -ENOMEM; | |
2829 | ||
2830 | flush_all(s); | |
f64dc58c | 2831 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2832 | n = get_node(s, node); |
2833 | ||
2834 | if (!n->nr_partial) | |
2835 | continue; | |
2836 | ||
2837 | for (i = 0; i < s->objects; i++) | |
2838 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2839 | ||
2840 | spin_lock_irqsave(&n->list_lock, flags); | |
2841 | ||
2842 | /* | |
672bba3a | 2843 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2844 | * |
672bba3a CL |
2845 | * Note that concurrent frees may occur while we hold the |
2846 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2847 | */ |
2848 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2849 | if (!page->inuse && slab_trylock(page)) { | |
2850 | /* | |
2851 | * Must hold slab lock here because slab_free | |
2852 | * may have freed the last object and be | |
2853 | * waiting to release the slab. | |
2854 | */ | |
2855 | list_del(&page->lru); | |
2856 | n->nr_partial--; | |
2857 | slab_unlock(page); | |
2858 | discard_slab(s, page); | |
2859 | } else { | |
fcda3d89 CL |
2860 | list_move(&page->lru, |
2861 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2862 | } |
2863 | } | |
2864 | ||
2086d26a | 2865 | /* |
672bba3a CL |
2866 | * Rebuild the partial list with the slabs filled up most |
2867 | * first and the least used slabs at the end. | |
2086d26a CL |
2868 | */ |
2869 | for (i = s->objects - 1; i >= 0; i--) | |
2870 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2871 | ||
2086d26a CL |
2872 | spin_unlock_irqrestore(&n->list_lock, flags); |
2873 | } | |
2874 | ||
2875 | kfree(slabs_by_inuse); | |
2876 | return 0; | |
2877 | } | |
2878 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2879 | ||
b9049e23 YG |
2880 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2881 | static int slab_mem_going_offline_callback(void *arg) | |
2882 | { | |
2883 | struct kmem_cache *s; | |
2884 | ||
2885 | down_read(&slub_lock); | |
2886 | list_for_each_entry(s, &slab_caches, list) | |
2887 | kmem_cache_shrink(s); | |
2888 | up_read(&slub_lock); | |
2889 | ||
2890 | return 0; | |
2891 | } | |
2892 | ||
2893 | static void slab_mem_offline_callback(void *arg) | |
2894 | { | |
2895 | struct kmem_cache_node *n; | |
2896 | struct kmem_cache *s; | |
2897 | struct memory_notify *marg = arg; | |
2898 | int offline_node; | |
2899 | ||
2900 | offline_node = marg->status_change_nid; | |
2901 | ||
2902 | /* | |
2903 | * If the node still has available memory. we need kmem_cache_node | |
2904 | * for it yet. | |
2905 | */ | |
2906 | if (offline_node < 0) | |
2907 | return; | |
2908 | ||
2909 | down_read(&slub_lock); | |
2910 | list_for_each_entry(s, &slab_caches, list) { | |
2911 | n = get_node(s, offline_node); | |
2912 | if (n) { | |
2913 | /* | |
2914 | * if n->nr_slabs > 0, slabs still exist on the node | |
2915 | * that is going down. We were unable to free them, | |
2916 | * and offline_pages() function shoudn't call this | |
2917 | * callback. So, we must fail. | |
2918 | */ | |
27bb628a | 2919 | BUG_ON(atomic_long_read(&n->nr_slabs)); |
b9049e23 YG |
2920 | |
2921 | s->node[offline_node] = NULL; | |
2922 | kmem_cache_free(kmalloc_caches, n); | |
2923 | } | |
2924 | } | |
2925 | up_read(&slub_lock); | |
2926 | } | |
2927 | ||
2928 | static int slab_mem_going_online_callback(void *arg) | |
2929 | { | |
2930 | struct kmem_cache_node *n; | |
2931 | struct kmem_cache *s; | |
2932 | struct memory_notify *marg = arg; | |
2933 | int nid = marg->status_change_nid; | |
2934 | int ret = 0; | |
2935 | ||
2936 | /* | |
2937 | * If the node's memory is already available, then kmem_cache_node is | |
2938 | * already created. Nothing to do. | |
2939 | */ | |
2940 | if (nid < 0) | |
2941 | return 0; | |
2942 | ||
2943 | /* | |
2944 | * We are bringing a node online. No memory is availabe yet. We must | |
2945 | * allocate a kmem_cache_node structure in order to bring the node | |
2946 | * online. | |
2947 | */ | |
2948 | down_read(&slub_lock); | |
2949 | list_for_each_entry(s, &slab_caches, list) { | |
2950 | /* | |
2951 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2952 | * since memory is not yet available from the node that | |
2953 | * is brought up. | |
2954 | */ | |
2955 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2956 | if (!n) { | |
2957 | ret = -ENOMEM; | |
2958 | goto out; | |
2959 | } | |
2960 | init_kmem_cache_node(n); | |
2961 | s->node[nid] = n; | |
2962 | } | |
2963 | out: | |
2964 | up_read(&slub_lock); | |
2965 | return ret; | |
2966 | } | |
2967 | ||
2968 | static int slab_memory_callback(struct notifier_block *self, | |
2969 | unsigned long action, void *arg) | |
2970 | { | |
2971 | int ret = 0; | |
2972 | ||
2973 | switch (action) { | |
2974 | case MEM_GOING_ONLINE: | |
2975 | ret = slab_mem_going_online_callback(arg); | |
2976 | break; | |
2977 | case MEM_GOING_OFFLINE: | |
2978 | ret = slab_mem_going_offline_callback(arg); | |
2979 | break; | |
2980 | case MEM_OFFLINE: | |
2981 | case MEM_CANCEL_ONLINE: | |
2982 | slab_mem_offline_callback(arg); | |
2983 | break; | |
2984 | case MEM_ONLINE: | |
2985 | case MEM_CANCEL_OFFLINE: | |
2986 | break; | |
2987 | } | |
2988 | ||
2989 | ret = notifier_from_errno(ret); | |
2990 | return ret; | |
2991 | } | |
2992 | ||
2993 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2994 | ||
81819f0f CL |
2995 | /******************************************************************** |
2996 | * Basic setup of slabs | |
2997 | *******************************************************************/ | |
2998 | ||
2999 | void __init kmem_cache_init(void) | |
3000 | { | |
3001 | int i; | |
4b356be0 | 3002 | int caches = 0; |
81819f0f | 3003 | |
4c93c355 CL |
3004 | init_alloc_cpu(); |
3005 | ||
81819f0f CL |
3006 | #ifdef CONFIG_NUMA |
3007 | /* | |
3008 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3009 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3010 | * kmem_cache_open for slab_state == DOWN. |
3011 | */ | |
3012 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
3013 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 3014 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3015 | caches++; |
b9049e23 YG |
3016 | |
3017 | hotplug_memory_notifier(slab_memory_callback, 1); | |
81819f0f CL |
3018 | #endif |
3019 | ||
3020 | /* Able to allocate the per node structures */ | |
3021 | slab_state = PARTIAL; | |
3022 | ||
3023 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
3024 | if (KMALLOC_MIN_SIZE <= 64) { |
3025 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 3026 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
3027 | caches++; |
3028 | } | |
3029 | if (KMALLOC_MIN_SIZE <= 128) { | |
3030 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 3031 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
3032 | caches++; |
3033 | } | |
81819f0f | 3034 | |
331dc558 | 3035 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) { |
81819f0f CL |
3036 | create_kmalloc_cache(&kmalloc_caches[i], |
3037 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
3038 | caches++; |
3039 | } | |
81819f0f | 3040 | |
f1b26339 CL |
3041 | |
3042 | /* | |
3043 | * Patch up the size_index table if we have strange large alignment | |
3044 | * requirements for the kmalloc array. This is only the case for | |
3045 | * mips it seems. The standard arches will not generate any code here. | |
3046 | * | |
3047 | * Largest permitted alignment is 256 bytes due to the way we | |
3048 | * handle the index determination for the smaller caches. | |
3049 | * | |
3050 | * Make sure that nothing crazy happens if someone starts tinkering | |
3051 | * around with ARCH_KMALLOC_MINALIGN | |
3052 | */ | |
3053 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3054 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3055 | ||
12ad6843 | 3056 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3057 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3058 | ||
81819f0f CL |
3059 | slab_state = UP; |
3060 | ||
3061 | /* Provide the correct kmalloc names now that the caches are up */ | |
331dc558 | 3062 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) |
81819f0f CL |
3063 | kmalloc_caches[i]. name = |
3064 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3065 | ||
3066 | #ifdef CONFIG_SMP | |
3067 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3068 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3069 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3070 | #else | |
3071 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3072 | #endif |
3073 | ||
81819f0f | 3074 | |
3adbefee IM |
3075 | printk(KERN_INFO |
3076 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3077 | " CPUs=%d, Nodes=%d\n", |
3078 | caches, cache_line_size(), | |
81819f0f CL |
3079 | slub_min_order, slub_max_order, slub_min_objects, |
3080 | nr_cpu_ids, nr_node_ids); | |
3081 | } | |
3082 | ||
3083 | /* | |
3084 | * Find a mergeable slab cache | |
3085 | */ | |
3086 | static int slab_unmergeable(struct kmem_cache *s) | |
3087 | { | |
3088 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3089 | return 1; | |
3090 | ||
331dc558 | 3091 | if ((s->flags & __PAGE_ALLOC_FALLBACK)) |
71c7a06f CL |
3092 | return 1; |
3093 | ||
c59def9f | 3094 | if (s->ctor) |
81819f0f CL |
3095 | return 1; |
3096 | ||
8ffa6875 CL |
3097 | /* |
3098 | * We may have set a slab to be unmergeable during bootstrap. | |
3099 | */ | |
3100 | if (s->refcount < 0) | |
3101 | return 1; | |
3102 | ||
81819f0f CL |
3103 | return 0; |
3104 | } | |
3105 | ||
3106 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3107 | size_t align, unsigned long flags, const char *name, |
4ba9b9d0 | 3108 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f | 3109 | { |
5b95a4ac | 3110 | struct kmem_cache *s; |
81819f0f CL |
3111 | |
3112 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3113 | return NULL; | |
3114 | ||
c59def9f | 3115 | if (ctor) |
81819f0f CL |
3116 | return NULL; |
3117 | ||
3118 | size = ALIGN(size, sizeof(void *)); | |
3119 | align = calculate_alignment(flags, align, size); | |
3120 | size = ALIGN(size, align); | |
ba0268a8 | 3121 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3122 | |
5b95a4ac | 3123 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3124 | if (slab_unmergeable(s)) |
3125 | continue; | |
3126 | ||
3127 | if (size > s->size) | |
3128 | continue; | |
3129 | ||
ba0268a8 | 3130 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3131 | continue; |
3132 | /* | |
3133 | * Check if alignment is compatible. | |
3134 | * Courtesy of Adrian Drzewiecki | |
3135 | */ | |
06428780 | 3136 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3137 | continue; |
3138 | ||
3139 | if (s->size - size >= sizeof(void *)) | |
3140 | continue; | |
3141 | ||
3142 | return s; | |
3143 | } | |
3144 | return NULL; | |
3145 | } | |
3146 | ||
3147 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
3148 | size_t align, unsigned long flags, | |
4ba9b9d0 | 3149 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
3150 | { |
3151 | struct kmem_cache *s; | |
3152 | ||
3153 | down_write(&slub_lock); | |
ba0268a8 | 3154 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3155 | if (s) { |
42a9fdbb CL |
3156 | int cpu; |
3157 | ||
81819f0f CL |
3158 | s->refcount++; |
3159 | /* | |
3160 | * Adjust the object sizes so that we clear | |
3161 | * the complete object on kzalloc. | |
3162 | */ | |
3163 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3164 | |
3165 | /* | |
3166 | * And then we need to update the object size in the | |
3167 | * per cpu structures | |
3168 | */ | |
3169 | for_each_online_cpu(cpu) | |
3170 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
81819f0f | 3171 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3172 | up_write(&slub_lock); |
81819f0f CL |
3173 | if (sysfs_slab_alias(s, name)) |
3174 | goto err; | |
a0e1d1be CL |
3175 | return s; |
3176 | } | |
3177 | s = kmalloc(kmem_size, GFP_KERNEL); | |
3178 | if (s) { | |
3179 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3180 | size, align, flags, ctor)) { |
81819f0f | 3181 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
3182 | up_write(&slub_lock); |
3183 | if (sysfs_slab_add(s)) | |
3184 | goto err; | |
3185 | return s; | |
3186 | } | |
3187 | kfree(s); | |
81819f0f CL |
3188 | } |
3189 | up_write(&slub_lock); | |
81819f0f CL |
3190 | |
3191 | err: | |
81819f0f CL |
3192 | if (flags & SLAB_PANIC) |
3193 | panic("Cannot create slabcache %s\n", name); | |
3194 | else | |
3195 | s = NULL; | |
3196 | return s; | |
3197 | } | |
3198 | EXPORT_SYMBOL(kmem_cache_create); | |
3199 | ||
81819f0f | 3200 | #ifdef CONFIG_SMP |
81819f0f | 3201 | /* |
672bba3a CL |
3202 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3203 | * necessary. | |
81819f0f CL |
3204 | */ |
3205 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3206 | unsigned long action, void *hcpu) | |
3207 | { | |
3208 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3209 | struct kmem_cache *s; |
3210 | unsigned long flags; | |
81819f0f CL |
3211 | |
3212 | switch (action) { | |
4c93c355 CL |
3213 | case CPU_UP_PREPARE: |
3214 | case CPU_UP_PREPARE_FROZEN: | |
3215 | init_alloc_cpu_cpu(cpu); | |
3216 | down_read(&slub_lock); | |
3217 | list_for_each_entry(s, &slab_caches, list) | |
3218 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3219 | GFP_KERNEL); | |
3220 | up_read(&slub_lock); | |
3221 | break; | |
3222 | ||
81819f0f | 3223 | case CPU_UP_CANCELED: |
8bb78442 | 3224 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3225 | case CPU_DEAD: |
8bb78442 | 3226 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3227 | down_read(&slub_lock); |
3228 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3229 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3230 | ||
5b95a4ac CL |
3231 | local_irq_save(flags); |
3232 | __flush_cpu_slab(s, cpu); | |
3233 | local_irq_restore(flags); | |
4c93c355 CL |
3234 | free_kmem_cache_cpu(c, cpu); |
3235 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3236 | } |
3237 | up_read(&slub_lock); | |
81819f0f CL |
3238 | break; |
3239 | default: | |
3240 | break; | |
3241 | } | |
3242 | return NOTIFY_OK; | |
3243 | } | |
3244 | ||
06428780 | 3245 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3246 | .notifier_call = slab_cpuup_callback |
06428780 | 3247 | }; |
81819f0f CL |
3248 | |
3249 | #endif | |
3250 | ||
81819f0f CL |
3251 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
3252 | { | |
aadb4bc4 CL |
3253 | struct kmem_cache *s; |
3254 | ||
331dc558 | 3255 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3256 | return kmalloc_large(size, gfpflags); |
3257 | ||
aadb4bc4 | 3258 | s = get_slab(size, gfpflags); |
81819f0f | 3259 | |
2408c550 | 3260 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3261 | return s; |
81819f0f | 3262 | |
ce15fea8 | 3263 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3264 | } |
3265 | ||
3266 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
3267 | int node, void *caller) | |
3268 | { | |
aadb4bc4 CL |
3269 | struct kmem_cache *s; |
3270 | ||
331dc558 | 3271 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3272 | return kmalloc_large(size, gfpflags); |
3273 | ||
aadb4bc4 | 3274 | s = get_slab(size, gfpflags); |
81819f0f | 3275 | |
2408c550 | 3276 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3277 | return s; |
81819f0f | 3278 | |
ce15fea8 | 3279 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3280 | } |
3281 | ||
41ecc55b | 3282 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
3283 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3284 | unsigned long *map) | |
53e15af0 CL |
3285 | { |
3286 | void *p; | |
683d0baa | 3287 | void *addr = slab_address(page); |
53e15af0 CL |
3288 | |
3289 | if (!check_slab(s, page) || | |
3290 | !on_freelist(s, page, NULL)) | |
3291 | return 0; | |
3292 | ||
3293 | /* Now we know that a valid freelist exists */ | |
3294 | bitmap_zero(map, s->objects); | |
3295 | ||
7656c72b CL |
3296 | for_each_free_object(p, s, page->freelist) { |
3297 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3298 | if (!check_object(s, page, p, 0)) |
3299 | return 0; | |
3300 | } | |
3301 | ||
7656c72b CL |
3302 | for_each_object(p, s, addr) |
3303 | if (!test_bit(slab_index(p, s, addr), map)) | |
53e15af0 CL |
3304 | if (!check_object(s, page, p, 1)) |
3305 | return 0; | |
3306 | return 1; | |
3307 | } | |
3308 | ||
434e245d CL |
3309 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3310 | unsigned long *map) | |
53e15af0 CL |
3311 | { |
3312 | if (slab_trylock(page)) { | |
434e245d | 3313 | validate_slab(s, page, map); |
53e15af0 CL |
3314 | slab_unlock(page); |
3315 | } else | |
3316 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3317 | s->name, page); | |
3318 | ||
3319 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3320 | if (!SlabDebug(page)) |
3321 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3322 | "on slab 0x%p\n", s->name, page); |
3323 | } else { | |
35e5d7ee CL |
3324 | if (SlabDebug(page)) |
3325 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3326 | "slab 0x%p\n", s->name, page); |
3327 | } | |
3328 | } | |
3329 | ||
434e245d CL |
3330 | static int validate_slab_node(struct kmem_cache *s, |
3331 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3332 | { |
3333 | unsigned long count = 0; | |
3334 | struct page *page; | |
3335 | unsigned long flags; | |
3336 | ||
3337 | spin_lock_irqsave(&n->list_lock, flags); | |
3338 | ||
3339 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3340 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3341 | count++; |
3342 | } | |
3343 | if (count != n->nr_partial) | |
3344 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3345 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3346 | ||
3347 | if (!(s->flags & SLAB_STORE_USER)) | |
3348 | goto out; | |
3349 | ||
3350 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3351 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3352 | count++; |
3353 | } | |
3354 | if (count != atomic_long_read(&n->nr_slabs)) | |
3355 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3356 | "counter=%ld\n", s->name, count, | |
3357 | atomic_long_read(&n->nr_slabs)); | |
3358 | ||
3359 | out: | |
3360 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3361 | return count; | |
3362 | } | |
3363 | ||
434e245d | 3364 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3365 | { |
3366 | int node; | |
3367 | unsigned long count = 0; | |
434e245d CL |
3368 | unsigned long *map = kmalloc(BITS_TO_LONGS(s->objects) * |
3369 | sizeof(unsigned long), GFP_KERNEL); | |
3370 | ||
3371 | if (!map) | |
3372 | return -ENOMEM; | |
53e15af0 CL |
3373 | |
3374 | flush_all(s); | |
f64dc58c | 3375 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3376 | struct kmem_cache_node *n = get_node(s, node); |
3377 | ||
434e245d | 3378 | count += validate_slab_node(s, n, map); |
53e15af0 | 3379 | } |
434e245d | 3380 | kfree(map); |
53e15af0 CL |
3381 | return count; |
3382 | } | |
3383 | ||
b3459709 CL |
3384 | #ifdef SLUB_RESILIENCY_TEST |
3385 | static void resiliency_test(void) | |
3386 | { | |
3387 | u8 *p; | |
3388 | ||
3389 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3390 | printk(KERN_ERR "-----------------------\n"); | |
3391 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3392 | ||
3393 | p = kzalloc(16, GFP_KERNEL); | |
3394 | p[16] = 0x12; | |
3395 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3396 | " 0x12->0x%p\n\n", p + 16); | |
3397 | ||
3398 | validate_slab_cache(kmalloc_caches + 4); | |
3399 | ||
3400 | /* Hmmm... The next two are dangerous */ | |
3401 | p = kzalloc(32, GFP_KERNEL); | |
3402 | p[32 + sizeof(void *)] = 0x34; | |
3403 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3404 | " 0x34 -> -0x%p\n", p); |
3405 | printk(KERN_ERR | |
3406 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3407 | |
3408 | validate_slab_cache(kmalloc_caches + 5); | |
3409 | p = kzalloc(64, GFP_KERNEL); | |
3410 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3411 | *p = 0x56; | |
3412 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3413 | p); | |
3adbefee IM |
3414 | printk(KERN_ERR |
3415 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3416 | validate_slab_cache(kmalloc_caches + 6); |
3417 | ||
3418 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3419 | p = kzalloc(128, GFP_KERNEL); | |
3420 | kfree(p); | |
3421 | *p = 0x78; | |
3422 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3423 | validate_slab_cache(kmalloc_caches + 7); | |
3424 | ||
3425 | p = kzalloc(256, GFP_KERNEL); | |
3426 | kfree(p); | |
3427 | p[50] = 0x9a; | |
3adbefee IM |
3428 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3429 | p); | |
b3459709 CL |
3430 | validate_slab_cache(kmalloc_caches + 8); |
3431 | ||
3432 | p = kzalloc(512, GFP_KERNEL); | |
3433 | kfree(p); | |
3434 | p[512] = 0xab; | |
3435 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3436 | validate_slab_cache(kmalloc_caches + 9); | |
3437 | } | |
3438 | #else | |
3439 | static void resiliency_test(void) {}; | |
3440 | #endif | |
3441 | ||
88a420e4 | 3442 | /* |
672bba3a | 3443 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3444 | * and freed. |
3445 | */ | |
3446 | ||
3447 | struct location { | |
3448 | unsigned long count; | |
3449 | void *addr; | |
45edfa58 CL |
3450 | long long sum_time; |
3451 | long min_time; | |
3452 | long max_time; | |
3453 | long min_pid; | |
3454 | long max_pid; | |
3455 | cpumask_t cpus; | |
3456 | nodemask_t nodes; | |
88a420e4 CL |
3457 | }; |
3458 | ||
3459 | struct loc_track { | |
3460 | unsigned long max; | |
3461 | unsigned long count; | |
3462 | struct location *loc; | |
3463 | }; | |
3464 | ||
3465 | static void free_loc_track(struct loc_track *t) | |
3466 | { | |
3467 | if (t->max) | |
3468 | free_pages((unsigned long)t->loc, | |
3469 | get_order(sizeof(struct location) * t->max)); | |
3470 | } | |
3471 | ||
68dff6a9 | 3472 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3473 | { |
3474 | struct location *l; | |
3475 | int order; | |
3476 | ||
88a420e4 CL |
3477 | order = get_order(sizeof(struct location) * max); |
3478 | ||
68dff6a9 | 3479 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3480 | if (!l) |
3481 | return 0; | |
3482 | ||
3483 | if (t->count) { | |
3484 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3485 | free_loc_track(t); | |
3486 | } | |
3487 | t->max = max; | |
3488 | t->loc = l; | |
3489 | return 1; | |
3490 | } | |
3491 | ||
3492 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3493 | const struct track *track) |
88a420e4 CL |
3494 | { |
3495 | long start, end, pos; | |
3496 | struct location *l; | |
3497 | void *caddr; | |
45edfa58 | 3498 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3499 | |
3500 | start = -1; | |
3501 | end = t->count; | |
3502 | ||
3503 | for ( ; ; ) { | |
3504 | pos = start + (end - start + 1) / 2; | |
3505 | ||
3506 | /* | |
3507 | * There is nothing at "end". If we end up there | |
3508 | * we need to add something to before end. | |
3509 | */ | |
3510 | if (pos == end) | |
3511 | break; | |
3512 | ||
3513 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3514 | if (track->addr == caddr) { |
3515 | ||
3516 | l = &t->loc[pos]; | |
3517 | l->count++; | |
3518 | if (track->when) { | |
3519 | l->sum_time += age; | |
3520 | if (age < l->min_time) | |
3521 | l->min_time = age; | |
3522 | if (age > l->max_time) | |
3523 | l->max_time = age; | |
3524 | ||
3525 | if (track->pid < l->min_pid) | |
3526 | l->min_pid = track->pid; | |
3527 | if (track->pid > l->max_pid) | |
3528 | l->max_pid = track->pid; | |
3529 | ||
3530 | cpu_set(track->cpu, l->cpus); | |
3531 | } | |
3532 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3533 | return 1; |
3534 | } | |
3535 | ||
45edfa58 | 3536 | if (track->addr < caddr) |
88a420e4 CL |
3537 | end = pos; |
3538 | else | |
3539 | start = pos; | |
3540 | } | |
3541 | ||
3542 | /* | |
672bba3a | 3543 | * Not found. Insert new tracking element. |
88a420e4 | 3544 | */ |
68dff6a9 | 3545 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3546 | return 0; |
3547 | ||
3548 | l = t->loc + pos; | |
3549 | if (pos < t->count) | |
3550 | memmove(l + 1, l, | |
3551 | (t->count - pos) * sizeof(struct location)); | |
3552 | t->count++; | |
3553 | l->count = 1; | |
45edfa58 CL |
3554 | l->addr = track->addr; |
3555 | l->sum_time = age; | |
3556 | l->min_time = age; | |
3557 | l->max_time = age; | |
3558 | l->min_pid = track->pid; | |
3559 | l->max_pid = track->pid; | |
3560 | cpus_clear(l->cpus); | |
3561 | cpu_set(track->cpu, l->cpus); | |
3562 | nodes_clear(l->nodes); | |
3563 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3564 | return 1; |
3565 | } | |
3566 | ||
3567 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3568 | struct page *page, enum track_item alloc) | |
3569 | { | |
683d0baa | 3570 | void *addr = slab_address(page); |
7656c72b | 3571 | DECLARE_BITMAP(map, s->objects); |
88a420e4 CL |
3572 | void *p; |
3573 | ||
3574 | bitmap_zero(map, s->objects); | |
7656c72b CL |
3575 | for_each_free_object(p, s, page->freelist) |
3576 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3577 | |
7656c72b | 3578 | for_each_object(p, s, addr) |
45edfa58 CL |
3579 | if (!test_bit(slab_index(p, s, addr), map)) |
3580 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3581 | } |
3582 | ||
3583 | static int list_locations(struct kmem_cache *s, char *buf, | |
3584 | enum track_item alloc) | |
3585 | { | |
e374d483 | 3586 | int len = 0; |
88a420e4 | 3587 | unsigned long i; |
68dff6a9 | 3588 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3589 | int node; |
3590 | ||
68dff6a9 | 3591 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3592 | GFP_TEMPORARY)) |
68dff6a9 | 3593 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3594 | |
3595 | /* Push back cpu slabs */ | |
3596 | flush_all(s); | |
3597 | ||
f64dc58c | 3598 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3599 | struct kmem_cache_node *n = get_node(s, node); |
3600 | unsigned long flags; | |
3601 | struct page *page; | |
3602 | ||
9e86943b | 3603 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3604 | continue; |
3605 | ||
3606 | spin_lock_irqsave(&n->list_lock, flags); | |
3607 | list_for_each_entry(page, &n->partial, lru) | |
3608 | process_slab(&t, s, page, alloc); | |
3609 | list_for_each_entry(page, &n->full, lru) | |
3610 | process_slab(&t, s, page, alloc); | |
3611 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3612 | } | |
3613 | ||
3614 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3615 | struct location *l = &t.loc[i]; |
88a420e4 | 3616 | |
e374d483 | 3617 | if (len > PAGE_SIZE - 100) |
88a420e4 | 3618 | break; |
e374d483 | 3619 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3620 | |
3621 | if (l->addr) | |
e374d483 | 3622 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3623 | else |
e374d483 | 3624 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3625 | |
3626 | if (l->sum_time != l->min_time) { | |
3627 | unsigned long remainder; | |
3628 | ||
e374d483 | 3629 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
45edfa58 CL |
3630 | l->min_time, |
3631 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3632 | l->max_time); | |
3633 | } else | |
e374d483 | 3634 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3635 | l->min_time); |
3636 | ||
3637 | if (l->min_pid != l->max_pid) | |
e374d483 | 3638 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3639 | l->min_pid, l->max_pid); |
3640 | else | |
e374d483 | 3641 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3642 | l->min_pid); |
3643 | ||
84966343 | 3644 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
e374d483 HH |
3645 | len < PAGE_SIZE - 60) { |
3646 | len += sprintf(buf + len, " cpus="); | |
3647 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3648 | l->cpus); |
3649 | } | |
3650 | ||
84966343 | 3651 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3652 | len < PAGE_SIZE - 60) { |
3653 | len += sprintf(buf + len, " nodes="); | |
3654 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3655 | l->nodes); |
3656 | } | |
3657 | ||
e374d483 | 3658 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3659 | } |
3660 | ||
3661 | free_loc_track(&t); | |
3662 | if (!t.count) | |
e374d483 HH |
3663 | len += sprintf(buf, "No data\n"); |
3664 | return len; | |
88a420e4 CL |
3665 | } |
3666 | ||
81819f0f CL |
3667 | enum slab_stat_type { |
3668 | SL_FULL, | |
3669 | SL_PARTIAL, | |
3670 | SL_CPU, | |
3671 | SL_OBJECTS | |
3672 | }; | |
3673 | ||
3674 | #define SO_FULL (1 << SL_FULL) | |
3675 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
3676 | #define SO_CPU (1 << SL_CPU) | |
3677 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
3678 | ||
3679 | static unsigned long slab_objects(struct kmem_cache *s, | |
3680 | char *buf, unsigned long flags) | |
3681 | { | |
3682 | unsigned long total = 0; | |
3683 | int cpu; | |
3684 | int node; | |
3685 | int x; | |
3686 | unsigned long *nodes; | |
3687 | unsigned long *per_cpu; | |
3688 | ||
3689 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
3690 | per_cpu = nodes + nr_node_ids; | |
3691 | ||
3692 | for_each_possible_cpu(cpu) { | |
dfb4f096 CL |
3693 | struct page *page; |
3694 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
81819f0f | 3695 | |
dfb4f096 CL |
3696 | if (!c) |
3697 | continue; | |
3698 | ||
3699 | page = c->page; | |
ee3c72a1 CL |
3700 | node = c->node; |
3701 | if (node < 0) | |
3702 | continue; | |
81819f0f | 3703 | if (page) { |
81819f0f | 3704 | if (flags & SO_CPU) { |
81819f0f CL |
3705 | if (flags & SO_OBJECTS) |
3706 | x = page->inuse; | |
3707 | else | |
3708 | x = 1; | |
3709 | total += x; | |
ee3c72a1 | 3710 | nodes[node] += x; |
81819f0f | 3711 | } |
ee3c72a1 | 3712 | per_cpu[node]++; |
81819f0f CL |
3713 | } |
3714 | } | |
3715 | ||
f64dc58c | 3716 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3717 | struct kmem_cache_node *n = get_node(s, node); |
3718 | ||
3719 | if (flags & SO_PARTIAL) { | |
3720 | if (flags & SO_OBJECTS) | |
3721 | x = count_partial(n); | |
3722 | else | |
3723 | x = n->nr_partial; | |
3724 | total += x; | |
3725 | nodes[node] += x; | |
3726 | } | |
3727 | ||
3728 | if (flags & SO_FULL) { | |
9e86943b | 3729 | int full_slabs = atomic_long_read(&n->nr_slabs) |
81819f0f CL |
3730 | - per_cpu[node] |
3731 | - n->nr_partial; | |
3732 | ||
3733 | if (flags & SO_OBJECTS) | |
3734 | x = full_slabs * s->objects; | |
3735 | else | |
3736 | x = full_slabs; | |
3737 | total += x; | |
3738 | nodes[node] += x; | |
3739 | } | |
3740 | } | |
3741 | ||
3742 | x = sprintf(buf, "%lu", total); | |
3743 | #ifdef CONFIG_NUMA | |
f64dc58c | 3744 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3745 | if (nodes[node]) |
3746 | x += sprintf(buf + x, " N%d=%lu", | |
3747 | node, nodes[node]); | |
3748 | #endif | |
3749 | kfree(nodes); | |
3750 | return x + sprintf(buf + x, "\n"); | |
3751 | } | |
3752 | ||
3753 | static int any_slab_objects(struct kmem_cache *s) | |
3754 | { | |
3755 | int node; | |
3756 | int cpu; | |
3757 | ||
dfb4f096 CL |
3758 | for_each_possible_cpu(cpu) { |
3759 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
3760 | ||
3761 | if (c && c->page) | |
81819f0f | 3762 | return 1; |
dfb4f096 | 3763 | } |
81819f0f | 3764 | |
dfb4f096 | 3765 | for_each_online_node(node) { |
81819f0f CL |
3766 | struct kmem_cache_node *n = get_node(s, node); |
3767 | ||
dfb4f096 CL |
3768 | if (!n) |
3769 | continue; | |
3770 | ||
9e86943b | 3771 | if (n->nr_partial || atomic_long_read(&n->nr_slabs)) |
81819f0f CL |
3772 | return 1; |
3773 | } | |
3774 | return 0; | |
3775 | } | |
3776 | ||
3777 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3778 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3779 | ||
3780 | struct slab_attribute { | |
3781 | struct attribute attr; | |
3782 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3783 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3784 | }; | |
3785 | ||
3786 | #define SLAB_ATTR_RO(_name) \ | |
3787 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3788 | ||
3789 | #define SLAB_ATTR(_name) \ | |
3790 | static struct slab_attribute _name##_attr = \ | |
3791 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3792 | ||
81819f0f CL |
3793 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3794 | { | |
3795 | return sprintf(buf, "%d\n", s->size); | |
3796 | } | |
3797 | SLAB_ATTR_RO(slab_size); | |
3798 | ||
3799 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3800 | { | |
3801 | return sprintf(buf, "%d\n", s->align); | |
3802 | } | |
3803 | SLAB_ATTR_RO(align); | |
3804 | ||
3805 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3806 | { | |
3807 | return sprintf(buf, "%d\n", s->objsize); | |
3808 | } | |
3809 | SLAB_ATTR_RO(object_size); | |
3810 | ||
3811 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3812 | { | |
3813 | return sprintf(buf, "%d\n", s->objects); | |
3814 | } | |
3815 | SLAB_ATTR_RO(objs_per_slab); | |
3816 | ||
3817 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3818 | { | |
3819 | return sprintf(buf, "%d\n", s->order); | |
3820 | } | |
3821 | SLAB_ATTR_RO(order); | |
3822 | ||
3823 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3824 | { | |
3825 | if (s->ctor) { | |
3826 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3827 | ||
3828 | return n + sprintf(buf + n, "\n"); | |
3829 | } | |
3830 | return 0; | |
3831 | } | |
3832 | SLAB_ATTR_RO(ctor); | |
3833 | ||
81819f0f CL |
3834 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3835 | { | |
3836 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3837 | } | |
3838 | SLAB_ATTR_RO(aliases); | |
3839 | ||
3840 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3841 | { | |
3842 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3843 | } | |
3844 | SLAB_ATTR_RO(slabs); | |
3845 | ||
3846 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3847 | { | |
3848 | return slab_objects(s, buf, SO_PARTIAL); | |
3849 | } | |
3850 | SLAB_ATTR_RO(partial); | |
3851 | ||
3852 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3853 | { | |
3854 | return slab_objects(s, buf, SO_CPU); | |
3855 | } | |
3856 | SLAB_ATTR_RO(cpu_slabs); | |
3857 | ||
3858 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3859 | { | |
3860 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3861 | } | |
3862 | SLAB_ATTR_RO(objects); | |
3863 | ||
3864 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3865 | { | |
3866 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3867 | } | |
3868 | ||
3869 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3870 | const char *buf, size_t length) | |
3871 | { | |
3872 | s->flags &= ~SLAB_DEBUG_FREE; | |
3873 | if (buf[0] == '1') | |
3874 | s->flags |= SLAB_DEBUG_FREE; | |
3875 | return length; | |
3876 | } | |
3877 | SLAB_ATTR(sanity_checks); | |
3878 | ||
3879 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3880 | { | |
3881 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3882 | } | |
3883 | ||
3884 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3885 | size_t length) | |
3886 | { | |
3887 | s->flags &= ~SLAB_TRACE; | |
3888 | if (buf[0] == '1') | |
3889 | s->flags |= SLAB_TRACE; | |
3890 | return length; | |
3891 | } | |
3892 | SLAB_ATTR(trace); | |
3893 | ||
3894 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3895 | { | |
3896 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3897 | } | |
3898 | ||
3899 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3900 | const char *buf, size_t length) | |
3901 | { | |
3902 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3903 | if (buf[0] == '1') | |
3904 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3905 | return length; | |
3906 | } | |
3907 | SLAB_ATTR(reclaim_account); | |
3908 | ||
3909 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3910 | { | |
5af60839 | 3911 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3912 | } |
3913 | SLAB_ATTR_RO(hwcache_align); | |
3914 | ||
3915 | #ifdef CONFIG_ZONE_DMA | |
3916 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3917 | { | |
3918 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3919 | } | |
3920 | SLAB_ATTR_RO(cache_dma); | |
3921 | #endif | |
3922 | ||
3923 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3924 | { | |
3925 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3926 | } | |
3927 | SLAB_ATTR_RO(destroy_by_rcu); | |
3928 | ||
3929 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3930 | { | |
3931 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3932 | } | |
3933 | ||
3934 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3935 | const char *buf, size_t length) | |
3936 | { | |
3937 | if (any_slab_objects(s)) | |
3938 | return -EBUSY; | |
3939 | ||
3940 | s->flags &= ~SLAB_RED_ZONE; | |
3941 | if (buf[0] == '1') | |
3942 | s->flags |= SLAB_RED_ZONE; | |
3943 | calculate_sizes(s); | |
3944 | return length; | |
3945 | } | |
3946 | SLAB_ATTR(red_zone); | |
3947 | ||
3948 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3949 | { | |
3950 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3951 | } | |
3952 | ||
3953 | static ssize_t poison_store(struct kmem_cache *s, | |
3954 | const char *buf, size_t length) | |
3955 | { | |
3956 | if (any_slab_objects(s)) | |
3957 | return -EBUSY; | |
3958 | ||
3959 | s->flags &= ~SLAB_POISON; | |
3960 | if (buf[0] == '1') | |
3961 | s->flags |= SLAB_POISON; | |
3962 | calculate_sizes(s); | |
3963 | return length; | |
3964 | } | |
3965 | SLAB_ATTR(poison); | |
3966 | ||
3967 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3968 | { | |
3969 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3970 | } | |
3971 | ||
3972 | static ssize_t store_user_store(struct kmem_cache *s, | |
3973 | const char *buf, size_t length) | |
3974 | { | |
3975 | if (any_slab_objects(s)) | |
3976 | return -EBUSY; | |
3977 | ||
3978 | s->flags &= ~SLAB_STORE_USER; | |
3979 | if (buf[0] == '1') | |
3980 | s->flags |= SLAB_STORE_USER; | |
3981 | calculate_sizes(s); | |
3982 | return length; | |
3983 | } | |
3984 | SLAB_ATTR(store_user); | |
3985 | ||
53e15af0 CL |
3986 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3987 | { | |
3988 | return 0; | |
3989 | } | |
3990 | ||
3991 | static ssize_t validate_store(struct kmem_cache *s, | |
3992 | const char *buf, size_t length) | |
3993 | { | |
434e245d CL |
3994 | int ret = -EINVAL; |
3995 | ||
3996 | if (buf[0] == '1') { | |
3997 | ret = validate_slab_cache(s); | |
3998 | if (ret >= 0) | |
3999 | ret = length; | |
4000 | } | |
4001 | return ret; | |
53e15af0 CL |
4002 | } |
4003 | SLAB_ATTR(validate); | |
4004 | ||
2086d26a CL |
4005 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4006 | { | |
4007 | return 0; | |
4008 | } | |
4009 | ||
4010 | static ssize_t shrink_store(struct kmem_cache *s, | |
4011 | const char *buf, size_t length) | |
4012 | { | |
4013 | if (buf[0] == '1') { | |
4014 | int rc = kmem_cache_shrink(s); | |
4015 | ||
4016 | if (rc) | |
4017 | return rc; | |
4018 | } else | |
4019 | return -EINVAL; | |
4020 | return length; | |
4021 | } | |
4022 | SLAB_ATTR(shrink); | |
4023 | ||
88a420e4 CL |
4024 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4025 | { | |
4026 | if (!(s->flags & SLAB_STORE_USER)) | |
4027 | return -ENOSYS; | |
4028 | return list_locations(s, buf, TRACK_ALLOC); | |
4029 | } | |
4030 | SLAB_ATTR_RO(alloc_calls); | |
4031 | ||
4032 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4033 | { | |
4034 | if (!(s->flags & SLAB_STORE_USER)) | |
4035 | return -ENOSYS; | |
4036 | return list_locations(s, buf, TRACK_FREE); | |
4037 | } | |
4038 | SLAB_ATTR_RO(free_calls); | |
4039 | ||
81819f0f | 4040 | #ifdef CONFIG_NUMA |
9824601e | 4041 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4042 | { |
9824601e | 4043 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4044 | } |
4045 | ||
9824601e | 4046 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4047 | const char *buf, size_t length) |
4048 | { | |
4049 | int n = simple_strtoul(buf, NULL, 10); | |
4050 | ||
4051 | if (n < 100) | |
9824601e | 4052 | s->remote_node_defrag_ratio = n * 10; |
81819f0f CL |
4053 | return length; |
4054 | } | |
9824601e | 4055 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4056 | #endif |
4057 | ||
8ff12cfc CL |
4058 | #ifdef CONFIG_SLUB_STATS |
4059 | ||
4060 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) | |
4061 | { | |
4062 | unsigned long sum = 0; | |
4063 | int cpu; | |
4064 | int len; | |
4065 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4066 | ||
4067 | if (!data) | |
4068 | return -ENOMEM; | |
4069 | ||
4070 | for_each_online_cpu(cpu) { | |
4071 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4072 | ||
4073 | data[cpu] = x; | |
4074 | sum += x; | |
4075 | } | |
4076 | ||
4077 | len = sprintf(buf, "%lu", sum); | |
4078 | ||
4079 | for_each_online_cpu(cpu) { | |
4080 | if (data[cpu] && len < PAGE_SIZE - 20) | |
4081 | len += sprintf(buf + len, " c%d=%u", cpu, data[cpu]); | |
4082 | } | |
4083 | kfree(data); | |
4084 | return len + sprintf(buf + len, "\n"); | |
4085 | } | |
4086 | ||
4087 | #define STAT_ATTR(si, text) \ | |
4088 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4089 | { \ | |
4090 | return show_stat(s, buf, si); \ | |
4091 | } \ | |
4092 | SLAB_ATTR_RO(text); \ | |
4093 | ||
4094 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4095 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4096 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4097 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4098 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4099 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4100 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4101 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4102 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4103 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4104 | STAT_ATTR(FREE_SLAB, free_slab); | |
4105 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4106 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4107 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4108 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4109 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4110 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
4111 | ||
4112 | #endif | |
4113 | ||
06428780 | 4114 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4115 | &slab_size_attr.attr, |
4116 | &object_size_attr.attr, | |
4117 | &objs_per_slab_attr.attr, | |
4118 | &order_attr.attr, | |
4119 | &objects_attr.attr, | |
4120 | &slabs_attr.attr, | |
4121 | &partial_attr.attr, | |
4122 | &cpu_slabs_attr.attr, | |
4123 | &ctor_attr.attr, | |
81819f0f CL |
4124 | &aliases_attr.attr, |
4125 | &align_attr.attr, | |
4126 | &sanity_checks_attr.attr, | |
4127 | &trace_attr.attr, | |
4128 | &hwcache_align_attr.attr, | |
4129 | &reclaim_account_attr.attr, | |
4130 | &destroy_by_rcu_attr.attr, | |
4131 | &red_zone_attr.attr, | |
4132 | &poison_attr.attr, | |
4133 | &store_user_attr.attr, | |
53e15af0 | 4134 | &validate_attr.attr, |
2086d26a | 4135 | &shrink_attr.attr, |
88a420e4 CL |
4136 | &alloc_calls_attr.attr, |
4137 | &free_calls_attr.attr, | |
81819f0f CL |
4138 | #ifdef CONFIG_ZONE_DMA |
4139 | &cache_dma_attr.attr, | |
4140 | #endif | |
4141 | #ifdef CONFIG_NUMA | |
9824601e | 4142 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4143 | #endif |
4144 | #ifdef CONFIG_SLUB_STATS | |
4145 | &alloc_fastpath_attr.attr, | |
4146 | &alloc_slowpath_attr.attr, | |
4147 | &free_fastpath_attr.attr, | |
4148 | &free_slowpath_attr.attr, | |
4149 | &free_frozen_attr.attr, | |
4150 | &free_add_partial_attr.attr, | |
4151 | &free_remove_partial_attr.attr, | |
4152 | &alloc_from_partial_attr.attr, | |
4153 | &alloc_slab_attr.attr, | |
4154 | &alloc_refill_attr.attr, | |
4155 | &free_slab_attr.attr, | |
4156 | &cpuslab_flush_attr.attr, | |
4157 | &deactivate_full_attr.attr, | |
4158 | &deactivate_empty_attr.attr, | |
4159 | &deactivate_to_head_attr.attr, | |
4160 | &deactivate_to_tail_attr.attr, | |
4161 | &deactivate_remote_frees_attr.attr, | |
81819f0f CL |
4162 | #endif |
4163 | NULL | |
4164 | }; | |
4165 | ||
4166 | static struct attribute_group slab_attr_group = { | |
4167 | .attrs = slab_attrs, | |
4168 | }; | |
4169 | ||
4170 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4171 | struct attribute *attr, | |
4172 | char *buf) | |
4173 | { | |
4174 | struct slab_attribute *attribute; | |
4175 | struct kmem_cache *s; | |
4176 | int err; | |
4177 | ||
4178 | attribute = to_slab_attr(attr); | |
4179 | s = to_slab(kobj); | |
4180 | ||
4181 | if (!attribute->show) | |
4182 | return -EIO; | |
4183 | ||
4184 | err = attribute->show(s, buf); | |
4185 | ||
4186 | return err; | |
4187 | } | |
4188 | ||
4189 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4190 | struct attribute *attr, | |
4191 | const char *buf, size_t len) | |
4192 | { | |
4193 | struct slab_attribute *attribute; | |
4194 | struct kmem_cache *s; | |
4195 | int err; | |
4196 | ||
4197 | attribute = to_slab_attr(attr); | |
4198 | s = to_slab(kobj); | |
4199 | ||
4200 | if (!attribute->store) | |
4201 | return -EIO; | |
4202 | ||
4203 | err = attribute->store(s, buf, len); | |
4204 | ||
4205 | return err; | |
4206 | } | |
4207 | ||
151c602f CL |
4208 | static void kmem_cache_release(struct kobject *kobj) |
4209 | { | |
4210 | struct kmem_cache *s = to_slab(kobj); | |
4211 | ||
4212 | kfree(s); | |
4213 | } | |
4214 | ||
81819f0f CL |
4215 | static struct sysfs_ops slab_sysfs_ops = { |
4216 | .show = slab_attr_show, | |
4217 | .store = slab_attr_store, | |
4218 | }; | |
4219 | ||
4220 | static struct kobj_type slab_ktype = { | |
4221 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4222 | .release = kmem_cache_release |
81819f0f CL |
4223 | }; |
4224 | ||
4225 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4226 | { | |
4227 | struct kobj_type *ktype = get_ktype(kobj); | |
4228 | ||
4229 | if (ktype == &slab_ktype) | |
4230 | return 1; | |
4231 | return 0; | |
4232 | } | |
4233 | ||
4234 | static struct kset_uevent_ops slab_uevent_ops = { | |
4235 | .filter = uevent_filter, | |
4236 | }; | |
4237 | ||
27c3a314 | 4238 | static struct kset *slab_kset; |
81819f0f CL |
4239 | |
4240 | #define ID_STR_LENGTH 64 | |
4241 | ||
4242 | /* Create a unique string id for a slab cache: | |
4243 | * format | |
4244 | * :[flags-]size:[memory address of kmemcache] | |
4245 | */ | |
4246 | static char *create_unique_id(struct kmem_cache *s) | |
4247 | { | |
4248 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4249 | char *p = name; | |
4250 | ||
4251 | BUG_ON(!name); | |
4252 | ||
4253 | *p++ = ':'; | |
4254 | /* | |
4255 | * First flags affecting slabcache operations. We will only | |
4256 | * get here for aliasable slabs so we do not need to support | |
4257 | * too many flags. The flags here must cover all flags that | |
4258 | * are matched during merging to guarantee that the id is | |
4259 | * unique. | |
4260 | */ | |
4261 | if (s->flags & SLAB_CACHE_DMA) | |
4262 | *p++ = 'd'; | |
4263 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4264 | *p++ = 'a'; | |
4265 | if (s->flags & SLAB_DEBUG_FREE) | |
4266 | *p++ = 'F'; | |
4267 | if (p != name + 1) | |
4268 | *p++ = '-'; | |
4269 | p += sprintf(p, "%07d", s->size); | |
4270 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4271 | return name; | |
4272 | } | |
4273 | ||
4274 | static int sysfs_slab_add(struct kmem_cache *s) | |
4275 | { | |
4276 | int err; | |
4277 | const char *name; | |
4278 | int unmergeable; | |
4279 | ||
4280 | if (slab_state < SYSFS) | |
4281 | /* Defer until later */ | |
4282 | return 0; | |
4283 | ||
4284 | unmergeable = slab_unmergeable(s); | |
4285 | if (unmergeable) { | |
4286 | /* | |
4287 | * Slabcache can never be merged so we can use the name proper. | |
4288 | * This is typically the case for debug situations. In that | |
4289 | * case we can catch duplicate names easily. | |
4290 | */ | |
27c3a314 | 4291 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4292 | name = s->name; |
4293 | } else { | |
4294 | /* | |
4295 | * Create a unique name for the slab as a target | |
4296 | * for the symlinks. | |
4297 | */ | |
4298 | name = create_unique_id(s); | |
4299 | } | |
4300 | ||
27c3a314 | 4301 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4302 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4303 | if (err) { | |
4304 | kobject_put(&s->kobj); | |
81819f0f | 4305 | return err; |
1eada11c | 4306 | } |
81819f0f CL |
4307 | |
4308 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4309 | if (err) | |
4310 | return err; | |
4311 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4312 | if (!unmergeable) { | |
4313 | /* Setup first alias */ | |
4314 | sysfs_slab_alias(s, s->name); | |
4315 | kfree(name); | |
4316 | } | |
4317 | return 0; | |
4318 | } | |
4319 | ||
4320 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4321 | { | |
4322 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4323 | kobject_del(&s->kobj); | |
151c602f | 4324 | kobject_put(&s->kobj); |
81819f0f CL |
4325 | } |
4326 | ||
4327 | /* | |
4328 | * Need to buffer aliases during bootup until sysfs becomes | |
4329 | * available lest we loose that information. | |
4330 | */ | |
4331 | struct saved_alias { | |
4332 | struct kmem_cache *s; | |
4333 | const char *name; | |
4334 | struct saved_alias *next; | |
4335 | }; | |
4336 | ||
5af328a5 | 4337 | static struct saved_alias *alias_list; |
81819f0f CL |
4338 | |
4339 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4340 | { | |
4341 | struct saved_alias *al; | |
4342 | ||
4343 | if (slab_state == SYSFS) { | |
4344 | /* | |
4345 | * If we have a leftover link then remove it. | |
4346 | */ | |
27c3a314 GKH |
4347 | sysfs_remove_link(&slab_kset->kobj, name); |
4348 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4349 | } |
4350 | ||
4351 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4352 | if (!al) | |
4353 | return -ENOMEM; | |
4354 | ||
4355 | al->s = s; | |
4356 | al->name = name; | |
4357 | al->next = alias_list; | |
4358 | alias_list = al; | |
4359 | return 0; | |
4360 | } | |
4361 | ||
4362 | static int __init slab_sysfs_init(void) | |
4363 | { | |
5b95a4ac | 4364 | struct kmem_cache *s; |
81819f0f CL |
4365 | int err; |
4366 | ||
0ff21e46 | 4367 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4368 | if (!slab_kset) { |
81819f0f CL |
4369 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4370 | return -ENOSYS; | |
4371 | } | |
4372 | ||
26a7bd03 CL |
4373 | slab_state = SYSFS; |
4374 | ||
5b95a4ac | 4375 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4376 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4377 | if (err) |
4378 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4379 | " to sysfs\n", s->name); | |
26a7bd03 | 4380 | } |
81819f0f CL |
4381 | |
4382 | while (alias_list) { | |
4383 | struct saved_alias *al = alias_list; | |
4384 | ||
4385 | alias_list = alias_list->next; | |
4386 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4387 | if (err) |
4388 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4389 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4390 | kfree(al); |
4391 | } | |
4392 | ||
4393 | resiliency_test(); | |
4394 | return 0; | |
4395 | } | |
4396 | ||
4397 | __initcall(slab_sysfs_init); | |
81819f0f | 4398 | #endif |
57ed3eda PE |
4399 | |
4400 | /* | |
4401 | * The /proc/slabinfo ABI | |
4402 | */ | |
158a9624 LT |
4403 | #ifdef CONFIG_SLABINFO |
4404 | ||
4405 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |
4406 | size_t count, loff_t *ppos) | |
4407 | { | |
4408 | return -EINVAL; | |
4409 | } | |
4410 | ||
57ed3eda PE |
4411 | |
4412 | static void print_slabinfo_header(struct seq_file *m) | |
4413 | { | |
4414 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4415 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4416 | "<objperslab> <pagesperslab>"); | |
4417 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4418 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4419 | seq_putc(m, '\n'); | |
4420 | } | |
4421 | ||
4422 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4423 | { | |
4424 | loff_t n = *pos; | |
4425 | ||
4426 | down_read(&slub_lock); | |
4427 | if (!n) | |
4428 | print_slabinfo_header(m); | |
4429 | ||
4430 | return seq_list_start(&slab_caches, *pos); | |
4431 | } | |
4432 | ||
4433 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4434 | { | |
4435 | return seq_list_next(p, &slab_caches, pos); | |
4436 | } | |
4437 | ||
4438 | static void s_stop(struct seq_file *m, void *p) | |
4439 | { | |
4440 | up_read(&slub_lock); | |
4441 | } | |
4442 | ||
4443 | static int s_show(struct seq_file *m, void *p) | |
4444 | { | |
4445 | unsigned long nr_partials = 0; | |
4446 | unsigned long nr_slabs = 0; | |
4447 | unsigned long nr_inuse = 0; | |
4448 | unsigned long nr_objs; | |
4449 | struct kmem_cache *s; | |
4450 | int node; | |
4451 | ||
4452 | s = list_entry(p, struct kmem_cache, list); | |
4453 | ||
4454 | for_each_online_node(node) { | |
4455 | struct kmem_cache_node *n = get_node(s, node); | |
4456 | ||
4457 | if (!n) | |
4458 | continue; | |
4459 | ||
4460 | nr_partials += n->nr_partial; | |
4461 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
4462 | nr_inuse += count_partial(n); | |
4463 | } | |
4464 | ||
4465 | nr_objs = nr_slabs * s->objects; | |
4466 | nr_inuse += (nr_slabs - nr_partials) * s->objects; | |
4467 | ||
4468 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
4469 | nr_objs, s->size, s->objects, (1 << s->order)); | |
4470 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); | |
4471 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4472 | 0UL); | |
4473 | seq_putc(m, '\n'); | |
4474 | return 0; | |
4475 | } | |
4476 | ||
4477 | const struct seq_operations slabinfo_op = { | |
4478 | .start = s_start, | |
4479 | .next = s_next, | |
4480 | .stop = s_stop, | |
4481 | .show = s_show, | |
4482 | }; | |
4483 | ||
158a9624 | 4484 | #endif /* CONFIG_SLABINFO */ |