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 | ||
152 | #if PAGE_SHIFT <= 12 | |
153 | ||
154 | /* | |
155 | * Small page size. Make sure that we do not fragment memory | |
156 | */ | |
157 | #define DEFAULT_MAX_ORDER 1 | |
158 | #define DEFAULT_MIN_OBJECTS 4 | |
159 | ||
160 | #else | |
161 | ||
162 | /* | |
163 | * Large page machines are customarily able to handle larger | |
164 | * page orders. | |
165 | */ | |
166 | #define DEFAULT_MAX_ORDER 2 | |
167 | #define DEFAULT_MIN_OBJECTS 8 | |
168 | ||
169 | #endif | |
170 | ||
2086d26a CL |
171 | /* |
172 | * Mininum number of partial slabs. These will be left on the partial | |
173 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
174 | */ | |
76be8950 | 175 | #define MIN_PARTIAL 5 |
e95eed57 | 176 | |
2086d26a CL |
177 | /* |
178 | * Maximum number of desirable partial slabs. | |
179 | * The existence of more partial slabs makes kmem_cache_shrink | |
180 | * sort the partial list by the number of objects in the. | |
181 | */ | |
182 | #define MAX_PARTIAL 10 | |
183 | ||
81819f0f CL |
184 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
185 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 186 | |
81819f0f CL |
187 | /* |
188 | * Set of flags that will prevent slab merging | |
189 | */ | |
190 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
191 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
192 | ||
193 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
194 | SLAB_CACHE_DMA) | |
195 | ||
196 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 197 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
198 | #endif |
199 | ||
200 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 201 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
202 | #endif |
203 | ||
204 | /* Internal SLUB flags */ | |
1ceef402 CL |
205 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
206 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
71c7a06f CL |
207 | #define __KMALLOC_CACHE 0x20000000 /* objects freed using kfree */ |
208 | #define __PAGE_ALLOC_FALLBACK 0x10000000 /* Allow fallback to page alloc */ | |
81819f0f | 209 | |
65c02d4c CL |
210 | /* Not all arches define cache_line_size */ |
211 | #ifndef cache_line_size | |
212 | #define cache_line_size() L1_CACHE_BYTES | |
213 | #endif | |
214 | ||
81819f0f CL |
215 | static int kmem_size = sizeof(struct kmem_cache); |
216 | ||
217 | #ifdef CONFIG_SMP | |
218 | static struct notifier_block slab_notifier; | |
219 | #endif | |
220 | ||
221 | static enum { | |
222 | DOWN, /* No slab functionality available */ | |
223 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 224 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
225 | SYSFS /* Sysfs up */ |
226 | } slab_state = DOWN; | |
227 | ||
228 | /* A list of all slab caches on the system */ | |
229 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 230 | static LIST_HEAD(slab_caches); |
81819f0f | 231 | |
02cbc874 CL |
232 | /* |
233 | * Tracking user of a slab. | |
234 | */ | |
235 | struct track { | |
236 | void *addr; /* Called from address */ | |
237 | int cpu; /* Was running on cpu */ | |
238 | int pid; /* Pid context */ | |
239 | unsigned long when; /* When did the operation occur */ | |
240 | }; | |
241 | ||
242 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
243 | ||
41ecc55b | 244 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
245 | static int sysfs_slab_add(struct kmem_cache *); |
246 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
247 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 248 | |
81819f0f | 249 | #else |
0c710013 CL |
250 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
251 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
252 | { return 0; } | |
151c602f CL |
253 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
254 | { | |
255 | kfree(s); | |
256 | } | |
8ff12cfc | 257 | |
81819f0f CL |
258 | #endif |
259 | ||
8ff12cfc CL |
260 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
261 | { | |
262 | #ifdef CONFIG_SLUB_STATS | |
263 | c->stat[si]++; | |
264 | #endif | |
265 | } | |
266 | ||
81819f0f CL |
267 | /******************************************************************** |
268 | * Core slab cache functions | |
269 | *******************************************************************/ | |
270 | ||
271 | int slab_is_available(void) | |
272 | { | |
273 | return slab_state >= UP; | |
274 | } | |
275 | ||
276 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
277 | { | |
278 | #ifdef CONFIG_NUMA | |
279 | return s->node[node]; | |
280 | #else | |
281 | return &s->local_node; | |
282 | #endif | |
283 | } | |
284 | ||
dfb4f096 CL |
285 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
286 | { | |
4c93c355 CL |
287 | #ifdef CONFIG_SMP |
288 | return s->cpu_slab[cpu]; | |
289 | #else | |
290 | return &s->cpu_slab; | |
291 | #endif | |
dfb4f096 CL |
292 | } |
293 | ||
02cbc874 CL |
294 | static inline int check_valid_pointer(struct kmem_cache *s, |
295 | struct page *page, const void *object) | |
296 | { | |
297 | void *base; | |
298 | ||
a973e9dd | 299 | if (!object) |
02cbc874 CL |
300 | return 1; |
301 | ||
a973e9dd | 302 | base = page_address(page); |
02cbc874 CL |
303 | if (object < base || object >= base + s->objects * s->size || |
304 | (object - base) % s->size) { | |
305 | return 0; | |
306 | } | |
307 | ||
308 | return 1; | |
309 | } | |
310 | ||
7656c72b CL |
311 | /* |
312 | * Slow version of get and set free pointer. | |
313 | * | |
314 | * This version requires touching the cache lines of kmem_cache which | |
315 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
316 | * from the page struct. | |
317 | */ | |
318 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
319 | { | |
320 | return *(void **)(object + s->offset); | |
321 | } | |
322 | ||
323 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
324 | { | |
325 | *(void **)(object + s->offset) = fp; | |
326 | } | |
327 | ||
328 | /* Loop over all objects in a slab */ | |
329 | #define for_each_object(__p, __s, __addr) \ | |
330 | for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\ | |
331 | __p += (__s)->size) | |
332 | ||
333 | /* Scan freelist */ | |
334 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 335 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
336 | |
337 | /* Determine object index from a given position */ | |
338 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
339 | { | |
340 | return (p - addr) / s->size; | |
341 | } | |
342 | ||
41ecc55b CL |
343 | #ifdef CONFIG_SLUB_DEBUG |
344 | /* | |
345 | * Debug settings: | |
346 | */ | |
f0630fff CL |
347 | #ifdef CONFIG_SLUB_DEBUG_ON |
348 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
349 | #else | |
41ecc55b | 350 | static int slub_debug; |
f0630fff | 351 | #endif |
41ecc55b CL |
352 | |
353 | static char *slub_debug_slabs; | |
354 | ||
81819f0f CL |
355 | /* |
356 | * Object debugging | |
357 | */ | |
358 | static void print_section(char *text, u8 *addr, unsigned int length) | |
359 | { | |
360 | int i, offset; | |
361 | int newline = 1; | |
362 | char ascii[17]; | |
363 | ||
364 | ascii[16] = 0; | |
365 | ||
366 | for (i = 0; i < length; i++) { | |
367 | if (newline) { | |
24922684 | 368 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
369 | newline = 0; |
370 | } | |
06428780 | 371 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
372 | offset = i % 16; |
373 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
374 | if (offset == 15) { | |
06428780 | 375 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
376 | newline = 1; |
377 | } | |
378 | } | |
379 | if (!newline) { | |
380 | i %= 16; | |
381 | while (i < 16) { | |
06428780 | 382 | printk(KERN_CONT " "); |
81819f0f CL |
383 | ascii[i] = ' '; |
384 | i++; | |
385 | } | |
06428780 | 386 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
387 | } |
388 | } | |
389 | ||
81819f0f CL |
390 | static struct track *get_track(struct kmem_cache *s, void *object, |
391 | enum track_item alloc) | |
392 | { | |
393 | struct track *p; | |
394 | ||
395 | if (s->offset) | |
396 | p = object + s->offset + sizeof(void *); | |
397 | else | |
398 | p = object + s->inuse; | |
399 | ||
400 | return p + alloc; | |
401 | } | |
402 | ||
403 | static void set_track(struct kmem_cache *s, void *object, | |
404 | enum track_item alloc, void *addr) | |
405 | { | |
406 | struct track *p; | |
407 | ||
408 | if (s->offset) | |
409 | p = object + s->offset + sizeof(void *); | |
410 | else | |
411 | p = object + s->inuse; | |
412 | ||
413 | p += alloc; | |
414 | if (addr) { | |
415 | p->addr = addr; | |
416 | p->cpu = smp_processor_id(); | |
417 | p->pid = current ? current->pid : -1; | |
418 | p->when = jiffies; | |
419 | } else | |
420 | memset(p, 0, sizeof(struct track)); | |
421 | } | |
422 | ||
81819f0f CL |
423 | static void init_tracking(struct kmem_cache *s, void *object) |
424 | { | |
24922684 CL |
425 | if (!(s->flags & SLAB_STORE_USER)) |
426 | return; | |
427 | ||
428 | set_track(s, object, TRACK_FREE, NULL); | |
429 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
430 | } |
431 | ||
432 | static void print_track(const char *s, struct track *t) | |
433 | { | |
434 | if (!t->addr) | |
435 | return; | |
436 | ||
24922684 | 437 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 438 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
439 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
440 | } | |
441 | ||
442 | static void print_tracking(struct kmem_cache *s, void *object) | |
443 | { | |
444 | if (!(s->flags & SLAB_STORE_USER)) | |
445 | return; | |
446 | ||
447 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
448 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
449 | } | |
450 | ||
451 | static void print_page_info(struct page *page) | |
452 | { | |
453 | printk(KERN_ERR "INFO: Slab 0x%p used=%u fp=0x%p flags=0x%04lx\n", | |
454 | page, page->inuse, page->freelist, page->flags); | |
455 | ||
456 | } | |
457 | ||
458 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
459 | { | |
460 | va_list args; | |
461 | char buf[100]; | |
462 | ||
463 | va_start(args, fmt); | |
464 | vsnprintf(buf, sizeof(buf), fmt, args); | |
465 | va_end(args); | |
466 | printk(KERN_ERR "========================================" | |
467 | "=====================================\n"); | |
468 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
469 | printk(KERN_ERR "----------------------------------------" | |
470 | "-------------------------------------\n\n"); | |
81819f0f CL |
471 | } |
472 | ||
24922684 CL |
473 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
474 | { | |
475 | va_list args; | |
476 | char buf[100]; | |
477 | ||
478 | va_start(args, fmt); | |
479 | vsnprintf(buf, sizeof(buf), fmt, args); | |
480 | va_end(args); | |
481 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
482 | } | |
483 | ||
484 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
485 | { |
486 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 487 | u8 *addr = page_address(page); |
24922684 CL |
488 | |
489 | print_tracking(s, p); | |
490 | ||
491 | print_page_info(page); | |
492 | ||
493 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
494 | p, p - addr, get_freepointer(s, p)); | |
495 | ||
496 | if (p > addr + 16) | |
497 | print_section("Bytes b4", p - 16, 16); | |
498 | ||
499 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
500 | |
501 | if (s->flags & SLAB_RED_ZONE) | |
502 | print_section("Redzone", p + s->objsize, | |
503 | s->inuse - s->objsize); | |
504 | ||
81819f0f CL |
505 | if (s->offset) |
506 | off = s->offset + sizeof(void *); | |
507 | else | |
508 | off = s->inuse; | |
509 | ||
24922684 | 510 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 511 | off += 2 * sizeof(struct track); |
81819f0f CL |
512 | |
513 | if (off != s->size) | |
514 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
515 | print_section("Padding", p + off, s->size - off); |
516 | ||
517 | dump_stack(); | |
81819f0f CL |
518 | } |
519 | ||
520 | static void object_err(struct kmem_cache *s, struct page *page, | |
521 | u8 *object, char *reason) | |
522 | { | |
24922684 CL |
523 | slab_bug(s, reason); |
524 | print_trailer(s, page, object); | |
81819f0f CL |
525 | } |
526 | ||
24922684 | 527 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
528 | { |
529 | va_list args; | |
530 | char buf[100]; | |
531 | ||
24922684 CL |
532 | va_start(args, fmt); |
533 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 534 | va_end(args); |
24922684 CL |
535 | slab_bug(s, fmt); |
536 | print_page_info(page); | |
81819f0f CL |
537 | dump_stack(); |
538 | } | |
539 | ||
540 | static void init_object(struct kmem_cache *s, void *object, int active) | |
541 | { | |
542 | u8 *p = object; | |
543 | ||
544 | if (s->flags & __OBJECT_POISON) { | |
545 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 546 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
547 | } |
548 | ||
549 | if (s->flags & SLAB_RED_ZONE) | |
550 | memset(p + s->objsize, | |
551 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
552 | s->inuse - s->objsize); | |
553 | } | |
554 | ||
24922684 | 555 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
556 | { |
557 | while (bytes) { | |
558 | if (*start != (u8)value) | |
24922684 | 559 | return start; |
81819f0f CL |
560 | start++; |
561 | bytes--; | |
562 | } | |
24922684 CL |
563 | return NULL; |
564 | } | |
565 | ||
566 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
567 | void *from, void *to) | |
568 | { | |
569 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
570 | memset(from, data, to - from); | |
571 | } | |
572 | ||
573 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
574 | u8 *object, char *what, | |
06428780 | 575 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
576 | { |
577 | u8 *fault; | |
578 | u8 *end; | |
579 | ||
580 | fault = check_bytes(start, value, bytes); | |
581 | if (!fault) | |
582 | return 1; | |
583 | ||
584 | end = start + bytes; | |
585 | while (end > fault && end[-1] == value) | |
586 | end--; | |
587 | ||
588 | slab_bug(s, "%s overwritten", what); | |
589 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
590 | fault, end - 1, fault[0], value); | |
591 | print_trailer(s, page, object); | |
592 | ||
593 | restore_bytes(s, what, value, fault, end); | |
594 | return 0; | |
81819f0f CL |
595 | } |
596 | ||
81819f0f CL |
597 | /* |
598 | * Object layout: | |
599 | * | |
600 | * object address | |
601 | * Bytes of the object to be managed. | |
602 | * If the freepointer may overlay the object then the free | |
603 | * pointer is the first word of the object. | |
672bba3a | 604 | * |
81819f0f CL |
605 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
606 | * 0xa5 (POISON_END) | |
607 | * | |
608 | * object + s->objsize | |
609 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
610 | * Padding is extended by another word if Redzoning is enabled and |
611 | * objsize == inuse. | |
612 | * | |
81819f0f CL |
613 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
614 | * 0xcc (RED_ACTIVE) for objects in use. | |
615 | * | |
616 | * object + s->inuse | |
672bba3a CL |
617 | * Meta data starts here. |
618 | * | |
81819f0f CL |
619 | * A. Free pointer (if we cannot overwrite object on free) |
620 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a CL |
621 | * C. Padding to reach required alignment boundary or at mininum |
622 | * one word if debuggin is on to be able to detect writes | |
623 | * before the word boundary. | |
624 | * | |
625 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
626 | * |
627 | * object + s->size | |
672bba3a | 628 | * Nothing is used beyond s->size. |
81819f0f | 629 | * |
672bba3a CL |
630 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
631 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
632 | * may be used with merged slabcaches. |
633 | */ | |
634 | ||
81819f0f CL |
635 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
636 | { | |
637 | unsigned long off = s->inuse; /* The end of info */ | |
638 | ||
639 | if (s->offset) | |
640 | /* Freepointer is placed after the object. */ | |
641 | off += sizeof(void *); | |
642 | ||
643 | if (s->flags & SLAB_STORE_USER) | |
644 | /* We also have user information there */ | |
645 | off += 2 * sizeof(struct track); | |
646 | ||
647 | if (s->size == off) | |
648 | return 1; | |
649 | ||
24922684 CL |
650 | return check_bytes_and_report(s, page, p, "Object padding", |
651 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
652 | } |
653 | ||
654 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
655 | { | |
24922684 CL |
656 | u8 *start; |
657 | u8 *fault; | |
658 | u8 *end; | |
659 | int length; | |
660 | int remainder; | |
81819f0f CL |
661 | |
662 | if (!(s->flags & SLAB_POISON)) | |
663 | return 1; | |
664 | ||
a973e9dd | 665 | start = page_address(page); |
24922684 | 666 | end = start + (PAGE_SIZE << s->order); |
81819f0f | 667 | length = s->objects * s->size; |
24922684 | 668 | remainder = end - (start + length); |
81819f0f CL |
669 | if (!remainder) |
670 | return 1; | |
671 | ||
24922684 CL |
672 | fault = check_bytes(start + length, POISON_INUSE, remainder); |
673 | if (!fault) | |
674 | return 1; | |
675 | while (end > fault && end[-1] == POISON_INUSE) | |
676 | end--; | |
677 | ||
678 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
679 | print_section("Padding", start, length); | |
680 | ||
681 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
682 | return 0; | |
81819f0f CL |
683 | } |
684 | ||
685 | static int check_object(struct kmem_cache *s, struct page *page, | |
686 | void *object, int active) | |
687 | { | |
688 | u8 *p = object; | |
689 | u8 *endobject = object + s->objsize; | |
690 | ||
691 | if (s->flags & SLAB_RED_ZONE) { | |
692 | unsigned int red = | |
693 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
694 | ||
24922684 CL |
695 | if (!check_bytes_and_report(s, page, object, "Redzone", |
696 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 697 | return 0; |
81819f0f | 698 | } else { |
3adbefee IM |
699 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
700 | check_bytes_and_report(s, page, p, "Alignment padding", | |
701 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
702 | } | |
81819f0f CL |
703 | } |
704 | ||
705 | if (s->flags & SLAB_POISON) { | |
706 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
707 | (!check_bytes_and_report(s, page, p, "Poison", p, |
708 | POISON_FREE, s->objsize - 1) || | |
709 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 710 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 711 | return 0; |
81819f0f CL |
712 | /* |
713 | * check_pad_bytes cleans up on its own. | |
714 | */ | |
715 | check_pad_bytes(s, page, p); | |
716 | } | |
717 | ||
718 | if (!s->offset && active) | |
719 | /* | |
720 | * Object and freepointer overlap. Cannot check | |
721 | * freepointer while object is allocated. | |
722 | */ | |
723 | return 1; | |
724 | ||
725 | /* Check free pointer validity */ | |
726 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
727 | object_err(s, page, p, "Freepointer corrupt"); | |
728 | /* | |
729 | * No choice but to zap it and thus loose the remainder | |
730 | * of the free objects in this slab. May cause | |
672bba3a | 731 | * another error because the object count is now wrong. |
81819f0f | 732 | */ |
a973e9dd | 733 | set_freepointer(s, p, NULL); |
81819f0f CL |
734 | return 0; |
735 | } | |
736 | return 1; | |
737 | } | |
738 | ||
739 | static int check_slab(struct kmem_cache *s, struct page *page) | |
740 | { | |
741 | VM_BUG_ON(!irqs_disabled()); | |
742 | ||
743 | if (!PageSlab(page)) { | |
24922684 | 744 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
745 | return 0; |
746 | } | |
81819f0f | 747 | if (page->inuse > s->objects) { |
24922684 CL |
748 | slab_err(s, page, "inuse %u > max %u", |
749 | s->name, page->inuse, s->objects); | |
81819f0f CL |
750 | return 0; |
751 | } | |
752 | /* Slab_pad_check fixes things up after itself */ | |
753 | slab_pad_check(s, page); | |
754 | return 1; | |
755 | } | |
756 | ||
757 | /* | |
672bba3a CL |
758 | * Determine if a certain object on a page is on the freelist. Must hold the |
759 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
760 | */ |
761 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
762 | { | |
763 | int nr = 0; | |
764 | void *fp = page->freelist; | |
765 | void *object = NULL; | |
766 | ||
a973e9dd | 767 | while (fp && nr <= s->objects) { |
81819f0f CL |
768 | if (fp == search) |
769 | return 1; | |
770 | if (!check_valid_pointer(s, page, fp)) { | |
771 | if (object) { | |
772 | object_err(s, page, object, | |
773 | "Freechain corrupt"); | |
a973e9dd | 774 | set_freepointer(s, object, NULL); |
81819f0f CL |
775 | break; |
776 | } else { | |
24922684 | 777 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 778 | page->freelist = NULL; |
81819f0f | 779 | page->inuse = s->objects; |
24922684 | 780 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
781 | return 0; |
782 | } | |
783 | break; | |
784 | } | |
785 | object = fp; | |
786 | fp = get_freepointer(s, object); | |
787 | nr++; | |
788 | } | |
789 | ||
790 | if (page->inuse != s->objects - nr) { | |
70d71228 | 791 | slab_err(s, page, "Wrong object count. Counter is %d but " |
24922684 | 792 | "counted were %d", page->inuse, s->objects - nr); |
81819f0f | 793 | page->inuse = s->objects - nr; |
24922684 | 794 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
795 | } |
796 | return search == NULL; | |
797 | } | |
798 | ||
3ec09742 CL |
799 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
800 | { | |
801 | if (s->flags & SLAB_TRACE) { | |
802 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
803 | s->name, | |
804 | alloc ? "alloc" : "free", | |
805 | object, page->inuse, | |
806 | page->freelist); | |
807 | ||
808 | if (!alloc) | |
809 | print_section("Object", (void *)object, s->objsize); | |
810 | ||
811 | dump_stack(); | |
812 | } | |
813 | } | |
814 | ||
643b1138 | 815 | /* |
672bba3a | 816 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 817 | */ |
e95eed57 | 818 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 819 | { |
643b1138 CL |
820 | spin_lock(&n->list_lock); |
821 | list_add(&page->lru, &n->full); | |
822 | spin_unlock(&n->list_lock); | |
823 | } | |
824 | ||
825 | static void remove_full(struct kmem_cache *s, struct page *page) | |
826 | { | |
827 | struct kmem_cache_node *n; | |
828 | ||
829 | if (!(s->flags & SLAB_STORE_USER)) | |
830 | return; | |
831 | ||
832 | n = get_node(s, page_to_nid(page)); | |
833 | ||
834 | spin_lock(&n->list_lock); | |
835 | list_del(&page->lru); | |
836 | spin_unlock(&n->list_lock); | |
837 | } | |
838 | ||
3ec09742 CL |
839 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
840 | void *object) | |
841 | { | |
842 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
843 | return; | |
844 | ||
845 | init_object(s, object, 0); | |
846 | init_tracking(s, object); | |
847 | } | |
848 | ||
849 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
850 | void *object, void *addr) | |
81819f0f CL |
851 | { |
852 | if (!check_slab(s, page)) | |
853 | goto bad; | |
854 | ||
855 | if (object && !on_freelist(s, page, object)) { | |
24922684 | 856 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 857 | goto bad; |
81819f0f CL |
858 | } |
859 | ||
860 | if (!check_valid_pointer(s, page, object)) { | |
861 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 862 | goto bad; |
81819f0f CL |
863 | } |
864 | ||
3ec09742 | 865 | if (object && !check_object(s, page, object, 0)) |
81819f0f | 866 | goto bad; |
81819f0f | 867 | |
3ec09742 CL |
868 | /* Success perform special debug activities for allocs */ |
869 | if (s->flags & SLAB_STORE_USER) | |
870 | set_track(s, object, TRACK_ALLOC, addr); | |
871 | trace(s, page, object, 1); | |
872 | init_object(s, object, 1); | |
81819f0f | 873 | return 1; |
3ec09742 | 874 | |
81819f0f CL |
875 | bad: |
876 | if (PageSlab(page)) { | |
877 | /* | |
878 | * If this is a slab page then lets do the best we can | |
879 | * to avoid issues in the future. Marking all objects | |
672bba3a | 880 | * as used avoids touching the remaining objects. |
81819f0f | 881 | */ |
24922684 | 882 | slab_fix(s, "Marking all objects used"); |
81819f0f | 883 | page->inuse = s->objects; |
a973e9dd | 884 | page->freelist = NULL; |
81819f0f CL |
885 | } |
886 | return 0; | |
887 | } | |
888 | ||
3ec09742 CL |
889 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
890 | void *object, void *addr) | |
81819f0f CL |
891 | { |
892 | if (!check_slab(s, page)) | |
893 | goto fail; | |
894 | ||
895 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 896 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
897 | goto fail; |
898 | } | |
899 | ||
900 | if (on_freelist(s, page, object)) { | |
24922684 | 901 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
902 | goto fail; |
903 | } | |
904 | ||
905 | if (!check_object(s, page, object, 1)) | |
906 | return 0; | |
907 | ||
908 | if (unlikely(s != page->slab)) { | |
3adbefee | 909 | if (!PageSlab(page)) { |
70d71228 CL |
910 | slab_err(s, page, "Attempt to free object(0x%p) " |
911 | "outside of slab", object); | |
3adbefee | 912 | } else if (!page->slab) { |
81819f0f | 913 | printk(KERN_ERR |
70d71228 | 914 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 915 | object); |
70d71228 | 916 | dump_stack(); |
06428780 | 917 | } else |
24922684 CL |
918 | object_err(s, page, object, |
919 | "page slab pointer corrupt."); | |
81819f0f CL |
920 | goto fail; |
921 | } | |
3ec09742 CL |
922 | |
923 | /* Special debug activities for freeing objects */ | |
a973e9dd | 924 | if (!SlabFrozen(page) && !page->freelist) |
3ec09742 CL |
925 | remove_full(s, page); |
926 | if (s->flags & SLAB_STORE_USER) | |
927 | set_track(s, object, TRACK_FREE, addr); | |
928 | trace(s, page, object, 0); | |
929 | init_object(s, object, 0); | |
81819f0f | 930 | return 1; |
3ec09742 | 931 | |
81819f0f | 932 | fail: |
24922684 | 933 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
934 | return 0; |
935 | } | |
936 | ||
41ecc55b CL |
937 | static int __init setup_slub_debug(char *str) |
938 | { | |
f0630fff CL |
939 | slub_debug = DEBUG_DEFAULT_FLAGS; |
940 | if (*str++ != '=' || !*str) | |
941 | /* | |
942 | * No options specified. Switch on full debugging. | |
943 | */ | |
944 | goto out; | |
945 | ||
946 | if (*str == ',') | |
947 | /* | |
948 | * No options but restriction on slabs. This means full | |
949 | * debugging for slabs matching a pattern. | |
950 | */ | |
951 | goto check_slabs; | |
952 | ||
953 | slub_debug = 0; | |
954 | if (*str == '-') | |
955 | /* | |
956 | * Switch off all debugging measures. | |
957 | */ | |
958 | goto out; | |
959 | ||
960 | /* | |
961 | * Determine which debug features should be switched on | |
962 | */ | |
06428780 | 963 | for (; *str && *str != ','; str++) { |
f0630fff CL |
964 | switch (tolower(*str)) { |
965 | case 'f': | |
966 | slub_debug |= SLAB_DEBUG_FREE; | |
967 | break; | |
968 | case 'z': | |
969 | slub_debug |= SLAB_RED_ZONE; | |
970 | break; | |
971 | case 'p': | |
972 | slub_debug |= SLAB_POISON; | |
973 | break; | |
974 | case 'u': | |
975 | slub_debug |= SLAB_STORE_USER; | |
976 | break; | |
977 | case 't': | |
978 | slub_debug |= SLAB_TRACE; | |
979 | break; | |
980 | default: | |
981 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 982 | "unknown. skipped\n", *str); |
f0630fff | 983 | } |
41ecc55b CL |
984 | } |
985 | ||
f0630fff | 986 | check_slabs: |
41ecc55b CL |
987 | if (*str == ',') |
988 | slub_debug_slabs = str + 1; | |
f0630fff | 989 | out: |
41ecc55b CL |
990 | return 1; |
991 | } | |
992 | ||
993 | __setup("slub_debug", setup_slub_debug); | |
994 | ||
ba0268a8 CL |
995 | static unsigned long kmem_cache_flags(unsigned long objsize, |
996 | unsigned long flags, const char *name, | |
4ba9b9d0 | 997 | void (*ctor)(struct kmem_cache *, void *)) |
41ecc55b CL |
998 | { |
999 | /* | |
1000 | * The page->offset field is only 16 bit wide. This is an offset | |
1001 | * in units of words from the beginning of an object. If the slab | |
1002 | * size is bigger then we cannot move the free pointer behind the | |
1003 | * object anymore. | |
1004 | * | |
1005 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
1006 | * the limit is 512k. | |
1007 | * | |
c59def9f | 1008 | * Debugging or ctor may create a need to move the free |
41ecc55b CL |
1009 | * pointer. Fail if this happens. |
1010 | */ | |
ba0268a8 CL |
1011 | if (objsize >= 65535 * sizeof(void *)) { |
1012 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
41ecc55b | 1013 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); |
ba0268a8 CL |
1014 | BUG_ON(ctor); |
1015 | } else { | |
41ecc55b CL |
1016 | /* |
1017 | * Enable debugging if selected on the kernel commandline. | |
1018 | */ | |
1019 | if (slub_debug && (!slub_debug_slabs || | |
ba0268a8 | 1020 | strncmp(slub_debug_slabs, name, |
3adbefee | 1021 | strlen(slub_debug_slabs)) == 0)) |
ba0268a8 CL |
1022 | flags |= slub_debug; |
1023 | } | |
1024 | ||
1025 | return flags; | |
41ecc55b CL |
1026 | } |
1027 | #else | |
3ec09742 CL |
1028 | static inline void setup_object_debug(struct kmem_cache *s, |
1029 | struct page *page, void *object) {} | |
41ecc55b | 1030 | |
3ec09742 CL |
1031 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1032 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1033 | |
3ec09742 CL |
1034 | static inline int free_debug_processing(struct kmem_cache *s, |
1035 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1036 | |
41ecc55b CL |
1037 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1038 | { return 1; } | |
1039 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1040 | void *object, int active) { return 1; } | |
3ec09742 | 1041 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1042 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1043 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1044 | void (*ctor)(struct kmem_cache *, void *)) |
ba0268a8 CL |
1045 | { |
1046 | return flags; | |
1047 | } | |
41ecc55b CL |
1048 | #define slub_debug 0 |
1049 | #endif | |
81819f0f CL |
1050 | /* |
1051 | * Slab allocation and freeing | |
1052 | */ | |
1053 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1054 | { | |
06428780 | 1055 | struct page *page; |
81819f0f CL |
1056 | int pages = 1 << s->order; |
1057 | ||
b7a49f0d | 1058 | flags |= s->allocflags; |
e12ba74d | 1059 | |
81819f0f CL |
1060 | if (node == -1) |
1061 | page = alloc_pages(flags, s->order); | |
1062 | else | |
1063 | page = alloc_pages_node(node, flags, s->order); | |
1064 | ||
1065 | if (!page) | |
1066 | return NULL; | |
1067 | ||
1068 | mod_zone_page_state(page_zone(page), | |
1069 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1070 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1071 | pages); | |
1072 | ||
1073 | return page; | |
1074 | } | |
1075 | ||
1076 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1077 | void *object) | |
1078 | { | |
3ec09742 | 1079 | setup_object_debug(s, page, object); |
4f104934 | 1080 | if (unlikely(s->ctor)) |
4ba9b9d0 | 1081 | s->ctor(s, object); |
81819f0f CL |
1082 | } |
1083 | ||
1084 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1085 | { | |
1086 | struct page *page; | |
1087 | struct kmem_cache_node *n; | |
1088 | void *start; | |
81819f0f CL |
1089 | void *last; |
1090 | void *p; | |
1091 | ||
6cb06229 | 1092 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1093 | |
6cb06229 CL |
1094 | page = allocate_slab(s, |
1095 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1096 | if (!page) |
1097 | goto out; | |
1098 | ||
1099 | n = get_node(s, page_to_nid(page)); | |
1100 | if (n) | |
1101 | atomic_long_inc(&n->nr_slabs); | |
81819f0f CL |
1102 | page->slab = s; |
1103 | page->flags |= 1 << PG_slab; | |
1104 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1105 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1106 | SetSlabDebug(page); |
81819f0f CL |
1107 | |
1108 | start = page_address(page); | |
81819f0f CL |
1109 | |
1110 | if (unlikely(s->flags & SLAB_POISON)) | |
1111 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
1112 | ||
1113 | last = start; | |
7656c72b | 1114 | for_each_object(p, s, start) { |
81819f0f CL |
1115 | setup_object(s, page, last); |
1116 | set_freepointer(s, last, p); | |
1117 | last = p; | |
1118 | } | |
1119 | setup_object(s, page, last); | |
a973e9dd | 1120 | set_freepointer(s, last, NULL); |
81819f0f CL |
1121 | |
1122 | page->freelist = start; | |
1123 | page->inuse = 0; | |
1124 | out: | |
81819f0f CL |
1125 | return page; |
1126 | } | |
1127 | ||
1128 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1129 | { | |
1130 | int pages = 1 << s->order; | |
1131 | ||
c59def9f | 1132 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1133 | void *p; |
1134 | ||
1135 | slab_pad_check(s, page); | |
a973e9dd | 1136 | for_each_object(p, s, page_address(page)) |
81819f0f | 1137 | check_object(s, page, p, 0); |
2208b764 | 1138 | ClearSlabDebug(page); |
81819f0f CL |
1139 | } |
1140 | ||
1141 | mod_zone_page_state(page_zone(page), | |
1142 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1143 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1144 | -pages); |
81819f0f | 1145 | |
81819f0f CL |
1146 | __free_pages(page, s->order); |
1147 | } | |
1148 | ||
1149 | static void rcu_free_slab(struct rcu_head *h) | |
1150 | { | |
1151 | struct page *page; | |
1152 | ||
1153 | page = container_of((struct list_head *)h, struct page, lru); | |
1154 | __free_slab(page->slab, page); | |
1155 | } | |
1156 | ||
1157 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1158 | { | |
1159 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1160 | /* | |
1161 | * RCU free overloads the RCU head over the LRU | |
1162 | */ | |
1163 | struct rcu_head *head = (void *)&page->lru; | |
1164 | ||
1165 | call_rcu(head, rcu_free_slab); | |
1166 | } else | |
1167 | __free_slab(s, page); | |
1168 | } | |
1169 | ||
1170 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1171 | { | |
1172 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1173 | ||
1174 | atomic_long_dec(&n->nr_slabs); | |
1175 | reset_page_mapcount(page); | |
35e5d7ee | 1176 | __ClearPageSlab(page); |
81819f0f CL |
1177 | free_slab(s, page); |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * Per slab locking using the pagelock | |
1182 | */ | |
1183 | static __always_inline void slab_lock(struct page *page) | |
1184 | { | |
1185 | bit_spin_lock(PG_locked, &page->flags); | |
1186 | } | |
1187 | ||
1188 | static __always_inline void slab_unlock(struct page *page) | |
1189 | { | |
a76d3546 | 1190 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1191 | } |
1192 | ||
1193 | static __always_inline int slab_trylock(struct page *page) | |
1194 | { | |
1195 | int rc = 1; | |
1196 | ||
1197 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1198 | return rc; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Management of partially allocated slabs | |
1203 | */ | |
7c2e132c CL |
1204 | static void add_partial(struct kmem_cache_node *n, |
1205 | struct page *page, int tail) | |
81819f0f | 1206 | { |
e95eed57 CL |
1207 | spin_lock(&n->list_lock); |
1208 | n->nr_partial++; | |
7c2e132c CL |
1209 | if (tail) |
1210 | list_add_tail(&page->lru, &n->partial); | |
1211 | else | |
1212 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1213 | spin_unlock(&n->list_lock); |
1214 | } | |
1215 | ||
1216 | static void remove_partial(struct kmem_cache *s, | |
1217 | struct page *page) | |
1218 | { | |
1219 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1220 | ||
1221 | spin_lock(&n->list_lock); | |
1222 | list_del(&page->lru); | |
1223 | n->nr_partial--; | |
1224 | spin_unlock(&n->list_lock); | |
1225 | } | |
1226 | ||
1227 | /* | |
672bba3a | 1228 | * Lock slab and remove from the partial list. |
81819f0f | 1229 | * |
672bba3a | 1230 | * Must hold list_lock. |
81819f0f | 1231 | */ |
4b6f0750 | 1232 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1233 | { |
1234 | if (slab_trylock(page)) { | |
1235 | list_del(&page->lru); | |
1236 | n->nr_partial--; | |
4b6f0750 | 1237 | SetSlabFrozen(page); |
81819f0f CL |
1238 | return 1; |
1239 | } | |
1240 | return 0; | |
1241 | } | |
1242 | ||
1243 | /* | |
672bba3a | 1244 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1245 | */ |
1246 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1247 | { | |
1248 | struct page *page; | |
1249 | ||
1250 | /* | |
1251 | * Racy check. If we mistakenly see no partial slabs then we | |
1252 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1253 | * partial slab and there is none available then get_partials() |
1254 | * will return NULL. | |
81819f0f CL |
1255 | */ |
1256 | if (!n || !n->nr_partial) | |
1257 | return NULL; | |
1258 | ||
1259 | spin_lock(&n->list_lock); | |
1260 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1261 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1262 | goto out; |
1263 | page = NULL; | |
1264 | out: | |
1265 | spin_unlock(&n->list_lock); | |
1266 | return page; | |
1267 | } | |
1268 | ||
1269 | /* | |
672bba3a | 1270 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1271 | */ |
1272 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1273 | { | |
1274 | #ifdef CONFIG_NUMA | |
1275 | struct zonelist *zonelist; | |
1276 | struct zone **z; | |
1277 | struct page *page; | |
1278 | ||
1279 | /* | |
672bba3a CL |
1280 | * The defrag ratio allows a configuration of the tradeoffs between |
1281 | * inter node defragmentation and node local allocations. A lower | |
1282 | * defrag_ratio increases the tendency to do local allocations | |
1283 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1284 | * |
672bba3a CL |
1285 | * If the defrag_ratio is set to 0 then kmalloc() always |
1286 | * returns node local objects. If the ratio is higher then kmalloc() | |
1287 | * may return off node objects because partial slabs are obtained | |
1288 | * from other nodes and filled up. | |
81819f0f CL |
1289 | * |
1290 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
672bba3a CL |
1291 | * defrag_ratio = 1000) then every (well almost) allocation will |
1292 | * first attempt to defrag slab caches on other nodes. This means | |
1293 | * scanning over all nodes to look for partial slabs which may be | |
1294 | * expensive if we do it every time we are trying to find a slab | |
1295 | * with available objects. | |
81819f0f | 1296 | */ |
9824601e CL |
1297 | if (!s->remote_node_defrag_ratio || |
1298 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1299 | return NULL; |
1300 | ||
3adbefee IM |
1301 | zonelist = &NODE_DATA( |
1302 | slab_node(current->mempolicy))->node_zonelists[gfp_zone(flags)]; | |
81819f0f CL |
1303 | for (z = zonelist->zones; *z; z++) { |
1304 | struct kmem_cache_node *n; | |
1305 | ||
1306 | n = get_node(s, zone_to_nid(*z)); | |
1307 | ||
1308 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1309 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1310 | page = get_partial_node(n); |
1311 | if (page) | |
1312 | return page; | |
1313 | } | |
1314 | } | |
1315 | #endif | |
1316 | return NULL; | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * Get a partial page, lock it and return it. | |
1321 | */ | |
1322 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1323 | { | |
1324 | struct page *page; | |
1325 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1326 | ||
1327 | page = get_partial_node(get_node(s, searchnode)); | |
1328 | if (page || (flags & __GFP_THISNODE)) | |
1329 | return page; | |
1330 | ||
1331 | return get_any_partial(s, flags); | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Move a page back to the lists. | |
1336 | * | |
1337 | * Must be called with the slab lock held. | |
1338 | * | |
1339 | * On exit the slab lock will have been dropped. | |
1340 | */ | |
7c2e132c | 1341 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1342 | { |
e95eed57 | 1343 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1344 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1345 | |
4b6f0750 | 1346 | ClearSlabFrozen(page); |
81819f0f | 1347 | if (page->inuse) { |
e95eed57 | 1348 | |
a973e9dd | 1349 | if (page->freelist) { |
7c2e132c | 1350 | add_partial(n, page, tail); |
8ff12cfc CL |
1351 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1352 | } else { | |
1353 | stat(c, DEACTIVATE_FULL); | |
1354 | if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) | |
1355 | add_full(n, page); | |
1356 | } | |
81819f0f CL |
1357 | slab_unlock(page); |
1358 | } else { | |
8ff12cfc | 1359 | stat(c, DEACTIVATE_EMPTY); |
e95eed57 CL |
1360 | if (n->nr_partial < MIN_PARTIAL) { |
1361 | /* | |
672bba3a CL |
1362 | * Adding an empty slab to the partial slabs in order |
1363 | * to avoid page allocator overhead. This slab needs | |
1364 | * to come after the other slabs with objects in | |
1365 | * order to fill them up. That way the size of the | |
1366 | * partial list stays small. kmem_cache_shrink can | |
1367 | * reclaim empty slabs from the partial list. | |
e95eed57 | 1368 | */ |
7c2e132c | 1369 | add_partial(n, page, 1); |
e95eed57 CL |
1370 | slab_unlock(page); |
1371 | } else { | |
1372 | slab_unlock(page); | |
8ff12cfc | 1373 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1374 | discard_slab(s, page); |
1375 | } | |
81819f0f CL |
1376 | } |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Remove the cpu slab | |
1381 | */ | |
dfb4f096 | 1382 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1383 | { |
dfb4f096 | 1384 | struct page *page = c->page; |
7c2e132c | 1385 | int tail = 1; |
8ff12cfc CL |
1386 | |
1387 | if (c->freelist) | |
1388 | stat(c, DEACTIVATE_REMOTE_FREES); | |
894b8788 CL |
1389 | /* |
1390 | * Merge cpu freelist into freelist. Typically we get here | |
1391 | * because both freelists are empty. So this is unlikely | |
1392 | * to occur. | |
1393 | */ | |
a973e9dd | 1394 | while (unlikely(c->freelist)) { |
894b8788 CL |
1395 | void **object; |
1396 | ||
7c2e132c CL |
1397 | tail = 0; /* Hot objects. Put the slab first */ |
1398 | ||
894b8788 | 1399 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1400 | object = c->freelist; |
b3fba8da | 1401 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1402 | |
1403 | /* And put onto the regular freelist */ | |
b3fba8da | 1404 | object[c->offset] = page->freelist; |
894b8788 CL |
1405 | page->freelist = object; |
1406 | page->inuse--; | |
1407 | } | |
dfb4f096 | 1408 | c->page = NULL; |
7c2e132c | 1409 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1410 | } |
1411 | ||
dfb4f096 | 1412 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1413 | { |
8ff12cfc | 1414 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1415 | slab_lock(c->page); |
1416 | deactivate_slab(s, c); | |
81819f0f CL |
1417 | } |
1418 | ||
1419 | /* | |
1420 | * Flush cpu slab. | |
1421 | * Called from IPI handler with interrupts disabled. | |
1422 | */ | |
0c710013 | 1423 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1424 | { |
dfb4f096 | 1425 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1426 | |
dfb4f096 CL |
1427 | if (likely(c && c->page)) |
1428 | flush_slab(s, c); | |
81819f0f CL |
1429 | } |
1430 | ||
1431 | static void flush_cpu_slab(void *d) | |
1432 | { | |
1433 | struct kmem_cache *s = d; | |
81819f0f | 1434 | |
dfb4f096 | 1435 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1436 | } |
1437 | ||
1438 | static void flush_all(struct kmem_cache *s) | |
1439 | { | |
1440 | #ifdef CONFIG_SMP | |
1441 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1442 | #else | |
1443 | unsigned long flags; | |
1444 | ||
1445 | local_irq_save(flags); | |
1446 | flush_cpu_slab(s); | |
1447 | local_irq_restore(flags); | |
1448 | #endif | |
1449 | } | |
1450 | ||
dfb4f096 CL |
1451 | /* |
1452 | * Check if the objects in a per cpu structure fit numa | |
1453 | * locality expectations. | |
1454 | */ | |
1455 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1456 | { | |
1457 | #ifdef CONFIG_NUMA | |
1458 | if (node != -1 && c->node != node) | |
1459 | return 0; | |
1460 | #endif | |
1461 | return 1; | |
1462 | } | |
1463 | ||
81819f0f | 1464 | /* |
894b8788 CL |
1465 | * Slow path. The lockless freelist is empty or we need to perform |
1466 | * debugging duties. | |
1467 | * | |
1468 | * Interrupts are disabled. | |
81819f0f | 1469 | * |
894b8788 CL |
1470 | * Processing is still very fast if new objects have been freed to the |
1471 | * regular freelist. In that case we simply take over the regular freelist | |
1472 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1473 | * |
894b8788 CL |
1474 | * If that is not working then we fall back to the partial lists. We take the |
1475 | * first element of the freelist as the object to allocate now and move the | |
1476 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1477 | * |
894b8788 CL |
1478 | * And if we were unable to get a new slab from the partial slab lists then |
1479 | * we need to allocate a new slab. This is slowest path since we may sleep. | |
81819f0f | 1480 | */ |
894b8788 | 1481 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1482 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1483 | { |
81819f0f | 1484 | void **object; |
dfb4f096 | 1485 | struct page *new; |
81819f0f | 1486 | |
dfb4f096 | 1487 | if (!c->page) |
81819f0f CL |
1488 | goto new_slab; |
1489 | ||
dfb4f096 CL |
1490 | slab_lock(c->page); |
1491 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1492 | goto another_slab; |
8ff12cfc | 1493 | stat(c, ALLOC_REFILL); |
894b8788 | 1494 | load_freelist: |
dfb4f096 | 1495 | object = c->page->freelist; |
a973e9dd | 1496 | if (unlikely(!object)) |
81819f0f | 1497 | goto another_slab; |
dfb4f096 | 1498 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1499 | goto debug; |
1500 | ||
dfb4f096 | 1501 | object = c->page->freelist; |
b3fba8da | 1502 | c->freelist = object[c->offset]; |
dfb4f096 | 1503 | c->page->inuse = s->objects; |
a973e9dd | 1504 | c->page->freelist = NULL; |
dfb4f096 | 1505 | c->node = page_to_nid(c->page); |
1f84260c | 1506 | unlock_out: |
dfb4f096 | 1507 | slab_unlock(c->page); |
8ff12cfc | 1508 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1509 | return object; |
1510 | ||
1511 | another_slab: | |
dfb4f096 | 1512 | deactivate_slab(s, c); |
81819f0f CL |
1513 | |
1514 | new_slab: | |
dfb4f096 CL |
1515 | new = get_partial(s, gfpflags, node); |
1516 | if (new) { | |
1517 | c->page = new; | |
8ff12cfc | 1518 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1519 | goto load_freelist; |
81819f0f CL |
1520 | } |
1521 | ||
b811c202 CL |
1522 | if (gfpflags & __GFP_WAIT) |
1523 | local_irq_enable(); | |
1524 | ||
dfb4f096 | 1525 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1526 | |
1527 | if (gfpflags & __GFP_WAIT) | |
1528 | local_irq_disable(); | |
1529 | ||
dfb4f096 CL |
1530 | if (new) { |
1531 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1532 | stat(c, ALLOC_SLAB); |
05aa3450 | 1533 | if (c->page) |
dfb4f096 | 1534 | flush_slab(s, c); |
dfb4f096 CL |
1535 | slab_lock(new); |
1536 | SetSlabFrozen(new); | |
1537 | c->page = new; | |
4b6f0750 | 1538 | goto load_freelist; |
81819f0f | 1539 | } |
00e962c5 | 1540 | |
71c7a06f CL |
1541 | /* |
1542 | * No memory available. | |
1543 | * | |
1544 | * If the slab uses higher order allocs but the object is | |
1545 | * smaller than a page size then we can fallback in emergencies | |
1546 | * to the page allocator via kmalloc_large. The page allocator may | |
1547 | * have failed to obtain a higher order page and we can try to | |
1548 | * allocate a single page if the object fits into a single page. | |
1549 | * That is only possible if certain conditions are met that are being | |
1550 | * checked when a slab is created. | |
1551 | */ | |
1552 | if (!(gfpflags & __GFP_NORETRY) && (s->flags & __PAGE_ALLOC_FALLBACK)) | |
1553 | return kmalloc_large(s->objsize, gfpflags); | |
1554 | ||
1555 | return NULL; | |
81819f0f | 1556 | debug: |
dfb4f096 CL |
1557 | object = c->page->freelist; |
1558 | if (!alloc_debug_processing(s, c->page, object, addr)) | |
81819f0f | 1559 | goto another_slab; |
894b8788 | 1560 | |
dfb4f096 | 1561 | c->page->inuse++; |
b3fba8da | 1562 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1563 | c->node = -1; |
1f84260c | 1564 | goto unlock_out; |
894b8788 CL |
1565 | } |
1566 | ||
1567 | /* | |
1568 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1569 | * have the fastpath folded into their functions. So no function call | |
1570 | * overhead for requests that can be satisfied on the fastpath. | |
1571 | * | |
1572 | * The fastpath works by first checking if the lockless freelist can be used. | |
1573 | * If not then __slab_alloc is called for slow processing. | |
1574 | * | |
1575 | * Otherwise we can simply pick the next object from the lockless free list. | |
1576 | */ | |
06428780 | 1577 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce15fea8 | 1578 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1579 | { |
894b8788 | 1580 | void **object; |
dfb4f096 | 1581 | struct kmem_cache_cpu *c; |
1f84260c CL |
1582 | unsigned long flags; |
1583 | ||
894b8788 | 1584 | local_irq_save(flags); |
dfb4f096 | 1585 | c = get_cpu_slab(s, smp_processor_id()); |
a973e9dd | 1586 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1587 | |
dfb4f096 | 1588 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1589 | |
1590 | else { | |
dfb4f096 | 1591 | object = c->freelist; |
b3fba8da | 1592 | c->freelist = object[c->offset]; |
8ff12cfc | 1593 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1594 | } |
1595 | local_irq_restore(flags); | |
d07dbea4 CL |
1596 | |
1597 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
42a9fdbb | 1598 | memset(object, 0, c->objsize); |
d07dbea4 | 1599 | |
894b8788 | 1600 | return object; |
81819f0f CL |
1601 | } |
1602 | ||
1603 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1604 | { | |
ce15fea8 | 1605 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1606 | } |
1607 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1608 | ||
1609 | #ifdef CONFIG_NUMA | |
1610 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1611 | { | |
ce15fea8 | 1612 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1613 | } |
1614 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1615 | #endif | |
1616 | ||
1617 | /* | |
894b8788 CL |
1618 | * Slow patch handling. This may still be called frequently since objects |
1619 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1620 | * |
894b8788 CL |
1621 | * So we still attempt to reduce cache line usage. Just take the slab |
1622 | * lock and free the item. If there is no additional partial page | |
1623 | * handling required then we can return immediately. | |
81819f0f | 1624 | */ |
894b8788 | 1625 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1626 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1627 | { |
1628 | void *prior; | |
1629 | void **object = (void *)x; | |
8ff12cfc | 1630 | struct kmem_cache_cpu *c; |
81819f0f | 1631 | |
8ff12cfc CL |
1632 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1633 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1634 | slab_lock(page); |
1635 | ||
35e5d7ee | 1636 | if (unlikely(SlabDebug(page))) |
81819f0f CL |
1637 | goto debug; |
1638 | checks_ok: | |
b3fba8da | 1639 | prior = object[offset] = page->freelist; |
81819f0f CL |
1640 | page->freelist = object; |
1641 | page->inuse--; | |
1642 | ||
8ff12cfc CL |
1643 | if (unlikely(SlabFrozen(page))) { |
1644 | stat(c, FREE_FROZEN); | |
81819f0f | 1645 | goto out_unlock; |
8ff12cfc | 1646 | } |
81819f0f CL |
1647 | |
1648 | if (unlikely(!page->inuse)) | |
1649 | goto slab_empty; | |
1650 | ||
1651 | /* | |
1652 | * Objects left in the slab. If it | |
1653 | * was not on the partial list before | |
1654 | * then add it. | |
1655 | */ | |
a973e9dd | 1656 | if (unlikely(!prior)) { |
7c2e132c | 1657 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1658 | stat(c, FREE_ADD_PARTIAL); |
1659 | } | |
81819f0f CL |
1660 | |
1661 | out_unlock: | |
1662 | slab_unlock(page); | |
81819f0f CL |
1663 | return; |
1664 | ||
1665 | slab_empty: | |
a973e9dd | 1666 | if (prior) { |
81819f0f | 1667 | /* |
672bba3a | 1668 | * Slab still on the partial list. |
81819f0f CL |
1669 | */ |
1670 | remove_partial(s, page); | |
8ff12cfc CL |
1671 | stat(c, FREE_REMOVE_PARTIAL); |
1672 | } | |
81819f0f | 1673 | slab_unlock(page); |
8ff12cfc | 1674 | stat(c, FREE_SLAB); |
81819f0f | 1675 | discard_slab(s, page); |
81819f0f CL |
1676 | return; |
1677 | ||
1678 | debug: | |
3ec09742 | 1679 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1680 | goto out_unlock; |
77c5e2d0 | 1681 | goto checks_ok; |
81819f0f CL |
1682 | } |
1683 | ||
894b8788 CL |
1684 | /* |
1685 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1686 | * can perform fastpath freeing without additional function calls. | |
1687 | * | |
1688 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1689 | * of this processor. This typically the case if we have just allocated | |
1690 | * the item before. | |
1691 | * | |
1692 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1693 | * with all sorts of special processing. | |
1694 | */ | |
06428780 | 1695 | static __always_inline void slab_free(struct kmem_cache *s, |
894b8788 CL |
1696 | struct page *page, void *x, void *addr) |
1697 | { | |
1698 | void **object = (void *)x; | |
dfb4f096 | 1699 | struct kmem_cache_cpu *c; |
1f84260c CL |
1700 | unsigned long flags; |
1701 | ||
894b8788 | 1702 | local_irq_save(flags); |
02febdf7 | 1703 | debug_check_no_locks_freed(object, s->objsize); |
dfb4f096 | 1704 | c = get_cpu_slab(s, smp_processor_id()); |
ee3c72a1 | 1705 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1706 | object[c->offset] = c->freelist; |
dfb4f096 | 1707 | c->freelist = object; |
8ff12cfc | 1708 | stat(c, FREE_FASTPATH); |
894b8788 | 1709 | } else |
b3fba8da | 1710 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1711 | |
1712 | local_irq_restore(flags); | |
1713 | } | |
1714 | ||
81819f0f CL |
1715 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1716 | { | |
77c5e2d0 | 1717 | struct page *page; |
81819f0f | 1718 | |
b49af68f | 1719 | page = virt_to_head_page(x); |
81819f0f | 1720 | |
77c5e2d0 | 1721 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1722 | } |
1723 | EXPORT_SYMBOL(kmem_cache_free); | |
1724 | ||
1725 | /* Figure out on which slab object the object resides */ | |
1726 | static struct page *get_object_page(const void *x) | |
1727 | { | |
b49af68f | 1728 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1729 | |
1730 | if (!PageSlab(page)) | |
1731 | return NULL; | |
1732 | ||
1733 | return page; | |
1734 | } | |
1735 | ||
1736 | /* | |
672bba3a CL |
1737 | * Object placement in a slab is made very easy because we always start at |
1738 | * offset 0. If we tune the size of the object to the alignment then we can | |
1739 | * get the required alignment by putting one properly sized object after | |
1740 | * another. | |
81819f0f CL |
1741 | * |
1742 | * Notice that the allocation order determines the sizes of the per cpu | |
1743 | * caches. Each processor has always one slab available for allocations. | |
1744 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1745 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1746 | * locking overhead. |
81819f0f CL |
1747 | */ |
1748 | ||
1749 | /* | |
1750 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1751 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1752 | * and increases the number of allocations possible without having to | |
1753 | * take the list_lock. | |
1754 | */ | |
1755 | static int slub_min_order; | |
1756 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1757 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1758 | ||
1759 | /* | |
1760 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1761 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1762 | */ |
1763 | static int slub_nomerge; | |
1764 | ||
81819f0f CL |
1765 | /* |
1766 | * Calculate the order of allocation given an slab object size. | |
1767 | * | |
672bba3a CL |
1768 | * The order of allocation has significant impact on performance and other |
1769 | * system components. Generally order 0 allocations should be preferred since | |
1770 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1771 | * be problematic to put into order 0 slabs because there may be too much | |
1772 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1773 | * would be wasted. | |
1774 | * | |
1775 | * In order to reach satisfactory performance we must ensure that a minimum | |
1776 | * number of objects is in one slab. Otherwise we may generate too much | |
1777 | * activity on the partial lists which requires taking the list_lock. This is | |
1778 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1779 | * |
672bba3a CL |
1780 | * slub_max_order specifies the order where we begin to stop considering the |
1781 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1782 | * we try to keep the page order as low as possible. So we accept more waste | |
1783 | * of space in favor of a small page order. | |
81819f0f | 1784 | * |
672bba3a CL |
1785 | * Higher order allocations also allow the placement of more objects in a |
1786 | * slab and thereby reduce object handling overhead. If the user has | |
1787 | * requested a higher mininum order then we start with that one instead of | |
1788 | * the smallest order which will fit the object. | |
81819f0f | 1789 | */ |
5e6d444e CL |
1790 | static inline int slab_order(int size, int min_objects, |
1791 | int max_order, int fract_leftover) | |
81819f0f CL |
1792 | { |
1793 | int order; | |
1794 | int rem; | |
6300ea75 | 1795 | int min_order = slub_min_order; |
81819f0f | 1796 | |
6300ea75 | 1797 | for (order = max(min_order, |
5e6d444e CL |
1798 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1799 | order <= max_order; order++) { | |
81819f0f | 1800 | |
5e6d444e | 1801 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1802 | |
5e6d444e | 1803 | if (slab_size < min_objects * size) |
81819f0f CL |
1804 | continue; |
1805 | ||
1806 | rem = slab_size % size; | |
1807 | ||
5e6d444e | 1808 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1809 | break; |
1810 | ||
1811 | } | |
672bba3a | 1812 | |
81819f0f CL |
1813 | return order; |
1814 | } | |
1815 | ||
5e6d444e CL |
1816 | static inline int calculate_order(int size) |
1817 | { | |
1818 | int order; | |
1819 | int min_objects; | |
1820 | int fraction; | |
1821 | ||
1822 | /* | |
1823 | * Attempt to find best configuration for a slab. This | |
1824 | * works by first attempting to generate a layout with | |
1825 | * the best configuration and backing off gradually. | |
1826 | * | |
1827 | * First we reduce the acceptable waste in a slab. Then | |
1828 | * we reduce the minimum objects required in a slab. | |
1829 | */ | |
1830 | min_objects = slub_min_objects; | |
1831 | while (min_objects > 1) { | |
1832 | fraction = 8; | |
1833 | while (fraction >= 4) { | |
1834 | order = slab_order(size, min_objects, | |
1835 | slub_max_order, fraction); | |
1836 | if (order <= slub_max_order) | |
1837 | return order; | |
1838 | fraction /= 2; | |
1839 | } | |
1840 | min_objects /= 2; | |
1841 | } | |
1842 | ||
1843 | /* | |
1844 | * We were unable to place multiple objects in a slab. Now | |
1845 | * lets see if we can place a single object there. | |
1846 | */ | |
1847 | order = slab_order(size, 1, slub_max_order, 1); | |
1848 | if (order <= slub_max_order) | |
1849 | return order; | |
1850 | ||
1851 | /* | |
1852 | * Doh this slab cannot be placed using slub_max_order. | |
1853 | */ | |
1854 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1855 | if (order <= MAX_ORDER) | |
1856 | return order; | |
1857 | return -ENOSYS; | |
1858 | } | |
1859 | ||
81819f0f | 1860 | /* |
672bba3a | 1861 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1862 | */ |
1863 | static unsigned long calculate_alignment(unsigned long flags, | |
1864 | unsigned long align, unsigned long size) | |
1865 | { | |
1866 | /* | |
1867 | * If the user wants hardware cache aligned objects then | |
1868 | * follow that suggestion if the object is sufficiently | |
1869 | * large. | |
1870 | * | |
1871 | * The hardware cache alignment cannot override the | |
1872 | * specified alignment though. If that is greater | |
1873 | * then use it. | |
1874 | */ | |
5af60839 | 1875 | if ((flags & SLAB_HWCACHE_ALIGN) && |
65c02d4c CL |
1876 | size > cache_line_size() / 2) |
1877 | return max_t(unsigned long, align, cache_line_size()); | |
81819f0f CL |
1878 | |
1879 | if (align < ARCH_SLAB_MINALIGN) | |
1880 | return ARCH_SLAB_MINALIGN; | |
1881 | ||
1882 | return ALIGN(align, sizeof(void *)); | |
1883 | } | |
1884 | ||
dfb4f096 CL |
1885 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1886 | struct kmem_cache_cpu *c) | |
1887 | { | |
1888 | c->page = NULL; | |
a973e9dd | 1889 | c->freelist = NULL; |
dfb4f096 | 1890 | c->node = 0; |
42a9fdbb CL |
1891 | c->offset = s->offset / sizeof(void *); |
1892 | c->objsize = s->objsize; | |
dfb4f096 CL |
1893 | } |
1894 | ||
81819f0f CL |
1895 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
1896 | { | |
1897 | n->nr_partial = 0; | |
1898 | atomic_long_set(&n->nr_slabs, 0); | |
1899 | spin_lock_init(&n->list_lock); | |
1900 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1901 | #ifdef CONFIG_SLUB_DEBUG |
643b1138 | 1902 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1903 | #endif |
81819f0f CL |
1904 | } |
1905 | ||
4c93c355 CL |
1906 | #ifdef CONFIG_SMP |
1907 | /* | |
1908 | * Per cpu array for per cpu structures. | |
1909 | * | |
1910 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1911 | * close together meaning that it becomes possible that multiple per cpu | |
1912 | * structures are contained in one cacheline. This may be particularly | |
1913 | * beneficial for the kmalloc caches. | |
1914 | * | |
1915 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1916 | * likely able to get per cpu structures for all caches from the array defined | |
1917 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
1918 | * | |
1919 | * If the per cpu array is exhausted then fall back to kmalloc | |
1920 | * of individual cachelines. No sharing is possible then. | |
1921 | */ | |
1922 | #define NR_KMEM_CACHE_CPU 100 | |
1923 | ||
1924 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
1925 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
1926 | ||
1927 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
1928 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
1929 | ||
1930 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
1931 | int cpu, gfp_t flags) | |
1932 | { | |
1933 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
1934 | ||
1935 | if (c) | |
1936 | per_cpu(kmem_cache_cpu_free, cpu) = | |
1937 | (void *)c->freelist; | |
1938 | else { | |
1939 | /* Table overflow: So allocate ourselves */ | |
1940 | c = kmalloc_node( | |
1941 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
1942 | flags, cpu_to_node(cpu)); | |
1943 | if (!c) | |
1944 | return NULL; | |
1945 | } | |
1946 | ||
1947 | init_kmem_cache_cpu(s, c); | |
1948 | return c; | |
1949 | } | |
1950 | ||
1951 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
1952 | { | |
1953 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
1954 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
1955 | kfree(c); | |
1956 | return; | |
1957 | } | |
1958 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
1959 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
1960 | } | |
1961 | ||
1962 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
1963 | { | |
1964 | int cpu; | |
1965 | ||
1966 | for_each_online_cpu(cpu) { | |
1967 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
1968 | ||
1969 | if (c) { | |
1970 | s->cpu_slab[cpu] = NULL; | |
1971 | free_kmem_cache_cpu(c, cpu); | |
1972 | } | |
1973 | } | |
1974 | } | |
1975 | ||
1976 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
1977 | { | |
1978 | int cpu; | |
1979 | ||
1980 | for_each_online_cpu(cpu) { | |
1981 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
1982 | ||
1983 | if (c) | |
1984 | continue; | |
1985 | ||
1986 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
1987 | if (!c) { | |
1988 | free_kmem_cache_cpus(s); | |
1989 | return 0; | |
1990 | } | |
1991 | s->cpu_slab[cpu] = c; | |
1992 | } | |
1993 | return 1; | |
1994 | } | |
1995 | ||
1996 | /* | |
1997 | * Initialize the per cpu array. | |
1998 | */ | |
1999 | static void init_alloc_cpu_cpu(int cpu) | |
2000 | { | |
2001 | int i; | |
2002 | ||
2003 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
2004 | return; | |
2005 | ||
2006 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2007 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2008 | ||
2009 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
2010 | } | |
2011 | ||
2012 | static void __init init_alloc_cpu(void) | |
2013 | { | |
2014 | int cpu; | |
2015 | ||
2016 | for_each_online_cpu(cpu) | |
2017 | init_alloc_cpu_cpu(cpu); | |
2018 | } | |
2019 | ||
2020 | #else | |
2021 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2022 | static inline void init_alloc_cpu(void) {} | |
2023 | ||
2024 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2025 | { | |
2026 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2027 | return 1; | |
2028 | } | |
2029 | #endif | |
2030 | ||
81819f0f CL |
2031 | #ifdef CONFIG_NUMA |
2032 | /* | |
2033 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2034 | * slab on the node for this slabcache. There are no concurrent accesses | |
2035 | * possible. | |
2036 | * | |
2037 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2038 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2039 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2040 | */ |
1cd7daa5 AB |
2041 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2042 | int node) | |
81819f0f CL |
2043 | { |
2044 | struct page *page; | |
2045 | struct kmem_cache_node *n; | |
ba84c73c | 2046 | unsigned long flags; |
81819f0f CL |
2047 | |
2048 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2049 | ||
a2f92ee7 | 2050 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2051 | |
2052 | BUG_ON(!page); | |
a2f92ee7 CL |
2053 | if (page_to_nid(page) != node) { |
2054 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2055 | "node %d\n", node); | |
2056 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2057 | "in order to be able to continue\n"); | |
2058 | } | |
2059 | ||
81819f0f CL |
2060 | n = page->freelist; |
2061 | BUG_ON(!n); | |
2062 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2063 | page->inuse++; | |
2064 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2065 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2066 | init_object(kmalloc_caches, n, 1); |
2067 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2068 | #endif |
81819f0f CL |
2069 | init_kmem_cache_node(n); |
2070 | atomic_long_inc(&n->nr_slabs); | |
ba84c73c | 2071 | /* |
2072 | * lockdep requires consistent irq usage for each lock | |
2073 | * so even though there cannot be a race this early in | |
2074 | * the boot sequence, we still disable irqs. | |
2075 | */ | |
2076 | local_irq_save(flags); | |
7c2e132c | 2077 | add_partial(n, page, 0); |
ba84c73c | 2078 | local_irq_restore(flags); |
81819f0f CL |
2079 | return n; |
2080 | } | |
2081 | ||
2082 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2083 | { | |
2084 | int node; | |
2085 | ||
f64dc58c | 2086 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2087 | struct kmem_cache_node *n = s->node[node]; |
2088 | if (n && n != &s->local_node) | |
2089 | kmem_cache_free(kmalloc_caches, n); | |
2090 | s->node[node] = NULL; | |
2091 | } | |
2092 | } | |
2093 | ||
2094 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2095 | { | |
2096 | int node; | |
2097 | int local_node; | |
2098 | ||
2099 | if (slab_state >= UP) | |
2100 | local_node = page_to_nid(virt_to_page(s)); | |
2101 | else | |
2102 | local_node = 0; | |
2103 | ||
f64dc58c | 2104 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2105 | struct kmem_cache_node *n; |
2106 | ||
2107 | if (local_node == node) | |
2108 | n = &s->local_node; | |
2109 | else { | |
2110 | if (slab_state == DOWN) { | |
2111 | n = early_kmem_cache_node_alloc(gfpflags, | |
2112 | node); | |
2113 | continue; | |
2114 | } | |
2115 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2116 | gfpflags, node); | |
2117 | ||
2118 | if (!n) { | |
2119 | free_kmem_cache_nodes(s); | |
2120 | return 0; | |
2121 | } | |
2122 | ||
2123 | } | |
2124 | s->node[node] = n; | |
2125 | init_kmem_cache_node(n); | |
2126 | } | |
2127 | return 1; | |
2128 | } | |
2129 | #else | |
2130 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2131 | { | |
2132 | } | |
2133 | ||
2134 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2135 | { | |
2136 | init_kmem_cache_node(&s->local_node); | |
2137 | return 1; | |
2138 | } | |
2139 | #endif | |
2140 | ||
2141 | /* | |
2142 | * calculate_sizes() determines the order and the distribution of data within | |
2143 | * a slab object. | |
2144 | */ | |
2145 | static int calculate_sizes(struct kmem_cache *s) | |
2146 | { | |
2147 | unsigned long flags = s->flags; | |
2148 | unsigned long size = s->objsize; | |
2149 | unsigned long align = s->align; | |
2150 | ||
2151 | /* | |
2152 | * Determine if we can poison the object itself. If the user of | |
2153 | * the slab may touch the object after free or before allocation | |
2154 | * then we should never poison the object itself. | |
2155 | */ | |
2156 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2157 | !s->ctor) |
81819f0f CL |
2158 | s->flags |= __OBJECT_POISON; |
2159 | else | |
2160 | s->flags &= ~__OBJECT_POISON; | |
2161 | ||
2162 | /* | |
2163 | * Round up object size to the next word boundary. We can only | |
2164 | * place the free pointer at word boundaries and this determines | |
2165 | * the possible location of the free pointer. | |
2166 | */ | |
2167 | size = ALIGN(size, sizeof(void *)); | |
2168 | ||
41ecc55b | 2169 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f | 2170 | /* |
672bba3a | 2171 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2172 | * end of the object and the free pointer. If not then add an |
672bba3a | 2173 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2174 | */ |
2175 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2176 | size += sizeof(void *); | |
41ecc55b | 2177 | #endif |
81819f0f CL |
2178 | |
2179 | /* | |
672bba3a CL |
2180 | * With that we have determined the number of bytes in actual use |
2181 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2182 | */ |
2183 | s->inuse = size; | |
2184 | ||
2185 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2186 | s->ctor)) { |
81819f0f CL |
2187 | /* |
2188 | * Relocate free pointer after the object if it is not | |
2189 | * permitted to overwrite the first word of the object on | |
2190 | * kmem_cache_free. | |
2191 | * | |
2192 | * This is the case if we do RCU, have a constructor or | |
2193 | * destructor or are poisoning the objects. | |
2194 | */ | |
2195 | s->offset = size; | |
2196 | size += sizeof(void *); | |
2197 | } | |
2198 | ||
c12b3c62 | 2199 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2200 | if (flags & SLAB_STORE_USER) |
2201 | /* | |
2202 | * Need to store information about allocs and frees after | |
2203 | * the object. | |
2204 | */ | |
2205 | size += 2 * sizeof(struct track); | |
2206 | ||
be7b3fbc | 2207 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2208 | /* |
2209 | * Add some empty padding so that we can catch | |
2210 | * overwrites from earlier objects rather than let | |
2211 | * tracking information or the free pointer be | |
2212 | * corrupted if an user writes before the start | |
2213 | * of the object. | |
2214 | */ | |
2215 | size += sizeof(void *); | |
41ecc55b | 2216 | #endif |
672bba3a | 2217 | |
81819f0f CL |
2218 | /* |
2219 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2220 | * user specified and the dynamic determination of cache line size |
2221 | * on bootup. | |
81819f0f CL |
2222 | */ |
2223 | align = calculate_alignment(flags, align, s->objsize); | |
2224 | ||
2225 | /* | |
2226 | * SLUB stores one object immediately after another beginning from | |
2227 | * offset 0. In order to align the objects we have to simply size | |
2228 | * each object to conform to the alignment. | |
2229 | */ | |
2230 | size = ALIGN(size, align); | |
2231 | s->size = size; | |
2232 | ||
71c7a06f CL |
2233 | if ((flags & __KMALLOC_CACHE) && |
2234 | PAGE_SIZE / size < slub_min_objects) { | |
2235 | /* | |
2236 | * Kmalloc cache that would not have enough objects in | |
2237 | * an order 0 page. Kmalloc slabs can fallback to | |
2238 | * page allocator order 0 allocs so take a reasonably large | |
2239 | * order that will allows us a good number of objects. | |
2240 | */ | |
2241 | s->order = max(slub_max_order, PAGE_ALLOC_COSTLY_ORDER); | |
2242 | s->flags |= __PAGE_ALLOC_FALLBACK; | |
2243 | s->allocflags |= __GFP_NOWARN; | |
2244 | } else | |
2245 | s->order = calculate_order(size); | |
2246 | ||
81819f0f CL |
2247 | if (s->order < 0) |
2248 | return 0; | |
2249 | ||
b7a49f0d CL |
2250 | s->allocflags = 0; |
2251 | if (s->order) | |
2252 | s->allocflags |= __GFP_COMP; | |
2253 | ||
2254 | if (s->flags & SLAB_CACHE_DMA) | |
2255 | s->allocflags |= SLUB_DMA; | |
2256 | ||
2257 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2258 | s->allocflags |= __GFP_RECLAIMABLE; | |
2259 | ||
81819f0f CL |
2260 | /* |
2261 | * Determine the number of objects per slab | |
2262 | */ | |
2263 | s->objects = (PAGE_SIZE << s->order) / size; | |
2264 | ||
b3fba8da | 2265 | return !!s->objects; |
81819f0f CL |
2266 | |
2267 | } | |
2268 | ||
81819f0f CL |
2269 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2270 | const char *name, size_t size, | |
2271 | size_t align, unsigned long flags, | |
4ba9b9d0 | 2272 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
2273 | { |
2274 | memset(s, 0, kmem_size); | |
2275 | s->name = name; | |
2276 | s->ctor = ctor; | |
81819f0f | 2277 | s->objsize = size; |
81819f0f | 2278 | s->align = align; |
ba0268a8 | 2279 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f CL |
2280 | |
2281 | if (!calculate_sizes(s)) | |
2282 | goto error; | |
2283 | ||
2284 | s->refcount = 1; | |
2285 | #ifdef CONFIG_NUMA | |
9824601e | 2286 | s->remote_node_defrag_ratio = 100; |
81819f0f | 2287 | #endif |
dfb4f096 CL |
2288 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2289 | goto error; | |
81819f0f | 2290 | |
dfb4f096 | 2291 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2292 | return 1; |
4c93c355 | 2293 | free_kmem_cache_nodes(s); |
81819f0f CL |
2294 | error: |
2295 | if (flags & SLAB_PANIC) | |
2296 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2297 | "order=%u offset=%u flags=%lx\n", | |
2298 | s->name, (unsigned long)size, s->size, s->order, | |
2299 | s->offset, flags); | |
2300 | return 0; | |
2301 | } | |
81819f0f CL |
2302 | |
2303 | /* | |
2304 | * Check if a given pointer is valid | |
2305 | */ | |
2306 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2307 | { | |
06428780 | 2308 | struct page *page; |
81819f0f CL |
2309 | |
2310 | page = get_object_page(object); | |
2311 | ||
2312 | if (!page || s != page->slab) | |
2313 | /* No slab or wrong slab */ | |
2314 | return 0; | |
2315 | ||
abcd08a6 | 2316 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2317 | return 0; |
2318 | ||
2319 | /* | |
2320 | * We could also check if the object is on the slabs freelist. | |
2321 | * But this would be too expensive and it seems that the main | |
2322 | * purpose of kmem_ptr_valid is to check if the object belongs | |
2323 | * to a certain slab. | |
2324 | */ | |
2325 | return 1; | |
2326 | } | |
2327 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2328 | ||
2329 | /* | |
2330 | * Determine the size of a slab object | |
2331 | */ | |
2332 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2333 | { | |
2334 | return s->objsize; | |
2335 | } | |
2336 | EXPORT_SYMBOL(kmem_cache_size); | |
2337 | ||
2338 | const char *kmem_cache_name(struct kmem_cache *s) | |
2339 | { | |
2340 | return s->name; | |
2341 | } | |
2342 | EXPORT_SYMBOL(kmem_cache_name); | |
2343 | ||
2344 | /* | |
672bba3a CL |
2345 | * Attempt to free all slabs on a node. Return the number of slabs we |
2346 | * were unable to free. | |
81819f0f CL |
2347 | */ |
2348 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
2349 | struct list_head *list) | |
2350 | { | |
2351 | int slabs_inuse = 0; | |
2352 | unsigned long flags; | |
2353 | struct page *page, *h; | |
2354 | ||
2355 | spin_lock_irqsave(&n->list_lock, flags); | |
2356 | list_for_each_entry_safe(page, h, list, lru) | |
2357 | if (!page->inuse) { | |
2358 | list_del(&page->lru); | |
2359 | discard_slab(s, page); | |
2360 | } else | |
2361 | slabs_inuse++; | |
2362 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2363 | return slabs_inuse; | |
2364 | } | |
2365 | ||
2366 | /* | |
672bba3a | 2367 | * Release all resources used by a slab cache. |
81819f0f | 2368 | */ |
0c710013 | 2369 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2370 | { |
2371 | int node; | |
2372 | ||
2373 | flush_all(s); | |
2374 | ||
2375 | /* Attempt to free all objects */ | |
4c93c355 | 2376 | free_kmem_cache_cpus(s); |
f64dc58c | 2377 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2378 | struct kmem_cache_node *n = get_node(s, node); |
2379 | ||
2086d26a | 2380 | n->nr_partial -= free_list(s, n, &n->partial); |
81819f0f CL |
2381 | if (atomic_long_read(&n->nr_slabs)) |
2382 | return 1; | |
2383 | } | |
2384 | free_kmem_cache_nodes(s); | |
2385 | return 0; | |
2386 | } | |
2387 | ||
2388 | /* | |
2389 | * Close a cache and release the kmem_cache structure | |
2390 | * (must be used for caches created using kmem_cache_create) | |
2391 | */ | |
2392 | void kmem_cache_destroy(struct kmem_cache *s) | |
2393 | { | |
2394 | down_write(&slub_lock); | |
2395 | s->refcount--; | |
2396 | if (!s->refcount) { | |
2397 | list_del(&s->list); | |
a0e1d1be | 2398 | up_write(&slub_lock); |
81819f0f CL |
2399 | if (kmem_cache_close(s)) |
2400 | WARN_ON(1); | |
2401 | sysfs_slab_remove(s); | |
a0e1d1be CL |
2402 | } else |
2403 | up_write(&slub_lock); | |
81819f0f CL |
2404 | } |
2405 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2406 | ||
2407 | /******************************************************************** | |
2408 | * Kmalloc subsystem | |
2409 | *******************************************************************/ | |
2410 | ||
331dc558 | 2411 | struct kmem_cache kmalloc_caches[PAGE_SHIFT + 1] __cacheline_aligned; |
81819f0f CL |
2412 | EXPORT_SYMBOL(kmalloc_caches); |
2413 | ||
2414 | #ifdef CONFIG_ZONE_DMA | |
331dc558 | 2415 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT + 1]; |
81819f0f CL |
2416 | #endif |
2417 | ||
2418 | static int __init setup_slub_min_order(char *str) | |
2419 | { | |
06428780 | 2420 | get_option(&str, &slub_min_order); |
81819f0f CL |
2421 | |
2422 | return 1; | |
2423 | } | |
2424 | ||
2425 | __setup("slub_min_order=", setup_slub_min_order); | |
2426 | ||
2427 | static int __init setup_slub_max_order(char *str) | |
2428 | { | |
06428780 | 2429 | get_option(&str, &slub_max_order); |
81819f0f CL |
2430 | |
2431 | return 1; | |
2432 | } | |
2433 | ||
2434 | __setup("slub_max_order=", setup_slub_max_order); | |
2435 | ||
2436 | static int __init setup_slub_min_objects(char *str) | |
2437 | { | |
06428780 | 2438 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2439 | |
2440 | return 1; | |
2441 | } | |
2442 | ||
2443 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2444 | ||
2445 | static int __init setup_slub_nomerge(char *str) | |
2446 | { | |
2447 | slub_nomerge = 1; | |
2448 | return 1; | |
2449 | } | |
2450 | ||
2451 | __setup("slub_nomerge", setup_slub_nomerge); | |
2452 | ||
81819f0f CL |
2453 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2454 | const char *name, int size, gfp_t gfp_flags) | |
2455 | { | |
2456 | unsigned int flags = 0; | |
2457 | ||
2458 | if (gfp_flags & SLUB_DMA) | |
2459 | flags = SLAB_CACHE_DMA; | |
2460 | ||
2461 | down_write(&slub_lock); | |
2462 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
71c7a06f | 2463 | flags | __KMALLOC_CACHE, NULL)) |
81819f0f CL |
2464 | goto panic; |
2465 | ||
2466 | list_add(&s->list, &slab_caches); | |
2467 | up_write(&slub_lock); | |
2468 | if (sysfs_slab_add(s)) | |
2469 | goto panic; | |
2470 | return s; | |
2471 | ||
2472 | panic: | |
2473 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2474 | } | |
2475 | ||
2e443fd0 | 2476 | #ifdef CONFIG_ZONE_DMA |
1ceef402 CL |
2477 | |
2478 | static void sysfs_add_func(struct work_struct *w) | |
2479 | { | |
2480 | struct kmem_cache *s; | |
2481 | ||
2482 | down_write(&slub_lock); | |
2483 | list_for_each_entry(s, &slab_caches, list) { | |
2484 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2485 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2486 | sysfs_slab_add(s); | |
2487 | } | |
2488 | } | |
2489 | up_write(&slub_lock); | |
2490 | } | |
2491 | ||
2492 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2493 | ||
2e443fd0 CL |
2494 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2495 | { | |
2496 | struct kmem_cache *s; | |
2e443fd0 CL |
2497 | char *text; |
2498 | size_t realsize; | |
2499 | ||
2500 | s = kmalloc_caches_dma[index]; | |
2501 | if (s) | |
2502 | return s; | |
2503 | ||
2504 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2505 | if (flags & __GFP_WAIT) |
2506 | down_write(&slub_lock); | |
2507 | else { | |
2508 | if (!down_write_trylock(&slub_lock)) | |
2509 | goto out; | |
2510 | } | |
2511 | ||
2512 | if (kmalloc_caches_dma[index]) | |
2513 | goto unlock_out; | |
2e443fd0 | 2514 | |
7b55f620 | 2515 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2516 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2517 | (unsigned int)realsize); | |
1ceef402 CL |
2518 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2519 | ||
2520 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2521 | realsize, ARCH_KMALLOC_MINALIGN, | |
2522 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2523 | kfree(s); | |
2524 | kfree(text); | |
2525 | goto unlock_out; | |
dfce8648 | 2526 | } |
1ceef402 CL |
2527 | |
2528 | list_add(&s->list, &slab_caches); | |
2529 | kmalloc_caches_dma[index] = s; | |
2530 | ||
2531 | schedule_work(&sysfs_add_work); | |
2532 | ||
2533 | unlock_out: | |
dfce8648 | 2534 | up_write(&slub_lock); |
1ceef402 | 2535 | out: |
dfce8648 | 2536 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2537 | } |
2538 | #endif | |
2539 | ||
f1b26339 CL |
2540 | /* |
2541 | * Conversion table for small slabs sizes / 8 to the index in the | |
2542 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2543 | * of two cache sizes there. The size of larger slabs can be determined using | |
2544 | * fls. | |
2545 | */ | |
2546 | static s8 size_index[24] = { | |
2547 | 3, /* 8 */ | |
2548 | 4, /* 16 */ | |
2549 | 5, /* 24 */ | |
2550 | 5, /* 32 */ | |
2551 | 6, /* 40 */ | |
2552 | 6, /* 48 */ | |
2553 | 6, /* 56 */ | |
2554 | 6, /* 64 */ | |
2555 | 1, /* 72 */ | |
2556 | 1, /* 80 */ | |
2557 | 1, /* 88 */ | |
2558 | 1, /* 96 */ | |
2559 | 7, /* 104 */ | |
2560 | 7, /* 112 */ | |
2561 | 7, /* 120 */ | |
2562 | 7, /* 128 */ | |
2563 | 2, /* 136 */ | |
2564 | 2, /* 144 */ | |
2565 | 2, /* 152 */ | |
2566 | 2, /* 160 */ | |
2567 | 2, /* 168 */ | |
2568 | 2, /* 176 */ | |
2569 | 2, /* 184 */ | |
2570 | 2 /* 192 */ | |
2571 | }; | |
2572 | ||
81819f0f CL |
2573 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2574 | { | |
f1b26339 | 2575 | int index; |
81819f0f | 2576 | |
f1b26339 CL |
2577 | if (size <= 192) { |
2578 | if (!size) | |
2579 | return ZERO_SIZE_PTR; | |
81819f0f | 2580 | |
f1b26339 | 2581 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2582 | } else |
f1b26339 | 2583 | index = fls(size - 1); |
81819f0f CL |
2584 | |
2585 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2586 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2587 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2588 | |
81819f0f CL |
2589 | #endif |
2590 | return &kmalloc_caches[index]; | |
2591 | } | |
2592 | ||
2593 | void *__kmalloc(size_t size, gfp_t flags) | |
2594 | { | |
aadb4bc4 | 2595 | struct kmem_cache *s; |
81819f0f | 2596 | |
331dc558 | 2597 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2598 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2599 | |
2600 | s = get_slab(size, flags); | |
2601 | ||
2602 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2603 | return s; |
2604 | ||
ce15fea8 | 2605 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2606 | } |
2607 | EXPORT_SYMBOL(__kmalloc); | |
2608 | ||
2609 | #ifdef CONFIG_NUMA | |
2610 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2611 | { | |
aadb4bc4 | 2612 | struct kmem_cache *s; |
81819f0f | 2613 | |
331dc558 | 2614 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2615 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2616 | |
2617 | s = get_slab(size, flags); | |
2618 | ||
2619 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2620 | return s; |
2621 | ||
ce15fea8 | 2622 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2623 | } |
2624 | EXPORT_SYMBOL(__kmalloc_node); | |
2625 | #endif | |
2626 | ||
2627 | size_t ksize(const void *object) | |
2628 | { | |
272c1d21 | 2629 | struct page *page; |
81819f0f CL |
2630 | struct kmem_cache *s; |
2631 | ||
ef8b4520 CL |
2632 | BUG_ON(!object); |
2633 | if (unlikely(object == ZERO_SIZE_PTR)) | |
272c1d21 CL |
2634 | return 0; |
2635 | ||
294a80a8 | 2636 | page = virt_to_head_page(object); |
81819f0f | 2637 | BUG_ON(!page); |
294a80a8 VN |
2638 | |
2639 | if (unlikely(!PageSlab(page))) | |
2640 | return PAGE_SIZE << compound_order(page); | |
2641 | ||
81819f0f CL |
2642 | s = page->slab; |
2643 | BUG_ON(!s); | |
2644 | ||
2645 | /* | |
2646 | * Debugging requires use of the padding between object | |
2647 | * and whatever may come after it. | |
2648 | */ | |
2649 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2650 | return s->objsize; | |
2651 | ||
2652 | /* | |
2653 | * If we have the need to store the freelist pointer | |
2654 | * back there or track user information then we can | |
2655 | * only use the space before that information. | |
2656 | */ | |
2657 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2658 | return s->inuse; | |
2659 | ||
2660 | /* | |
2661 | * Else we can use all the padding etc for the allocation | |
2662 | */ | |
2663 | return s->size; | |
2664 | } | |
2665 | EXPORT_SYMBOL(ksize); | |
2666 | ||
2667 | void kfree(const void *x) | |
2668 | { | |
81819f0f | 2669 | struct page *page; |
5bb983b0 | 2670 | void *object = (void *)x; |
81819f0f | 2671 | |
2408c550 | 2672 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2673 | return; |
2674 | ||
b49af68f | 2675 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2676 | if (unlikely(!PageSlab(page))) { |
2677 | put_page(page); | |
2678 | return; | |
2679 | } | |
5bb983b0 | 2680 | slab_free(page->slab, page, object, __builtin_return_address(0)); |
81819f0f CL |
2681 | } |
2682 | EXPORT_SYMBOL(kfree); | |
2683 | ||
f61396ae CL |
2684 | static unsigned long count_partial(struct kmem_cache_node *n) |
2685 | { | |
2686 | unsigned long flags; | |
2687 | unsigned long x = 0; | |
2688 | struct page *page; | |
2689 | ||
2690 | spin_lock_irqsave(&n->list_lock, flags); | |
2691 | list_for_each_entry(page, &n->partial, lru) | |
2692 | x += page->inuse; | |
2693 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2694 | return x; | |
2695 | } | |
2696 | ||
2086d26a | 2697 | /* |
672bba3a CL |
2698 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2699 | * the remaining slabs by the number of items in use. The slabs with the | |
2700 | * most items in use come first. New allocations will then fill those up | |
2701 | * and thus they can be removed from the partial lists. | |
2702 | * | |
2703 | * The slabs with the least items are placed last. This results in them | |
2704 | * being allocated from last increasing the chance that the last objects | |
2705 | * are freed in them. | |
2086d26a CL |
2706 | */ |
2707 | int kmem_cache_shrink(struct kmem_cache *s) | |
2708 | { | |
2709 | int node; | |
2710 | int i; | |
2711 | struct kmem_cache_node *n; | |
2712 | struct page *page; | |
2713 | struct page *t; | |
2714 | struct list_head *slabs_by_inuse = | |
2715 | kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL); | |
2716 | unsigned long flags; | |
2717 | ||
2718 | if (!slabs_by_inuse) | |
2719 | return -ENOMEM; | |
2720 | ||
2721 | flush_all(s); | |
f64dc58c | 2722 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2723 | n = get_node(s, node); |
2724 | ||
2725 | if (!n->nr_partial) | |
2726 | continue; | |
2727 | ||
2728 | for (i = 0; i < s->objects; i++) | |
2729 | INIT_LIST_HEAD(slabs_by_inuse + i); | |
2730 | ||
2731 | spin_lock_irqsave(&n->list_lock, flags); | |
2732 | ||
2733 | /* | |
672bba3a | 2734 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2735 | * |
672bba3a CL |
2736 | * Note that concurrent frees may occur while we hold the |
2737 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2738 | */ |
2739 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2740 | if (!page->inuse && slab_trylock(page)) { | |
2741 | /* | |
2742 | * Must hold slab lock here because slab_free | |
2743 | * may have freed the last object and be | |
2744 | * waiting to release the slab. | |
2745 | */ | |
2746 | list_del(&page->lru); | |
2747 | n->nr_partial--; | |
2748 | slab_unlock(page); | |
2749 | discard_slab(s, page); | |
2750 | } else { | |
fcda3d89 CL |
2751 | list_move(&page->lru, |
2752 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2753 | } |
2754 | } | |
2755 | ||
2086d26a | 2756 | /* |
672bba3a CL |
2757 | * Rebuild the partial list with the slabs filled up most |
2758 | * first and the least used slabs at the end. | |
2086d26a CL |
2759 | */ |
2760 | for (i = s->objects - 1; i >= 0; i--) | |
2761 | list_splice(slabs_by_inuse + i, n->partial.prev); | |
2762 | ||
2086d26a CL |
2763 | spin_unlock_irqrestore(&n->list_lock, flags); |
2764 | } | |
2765 | ||
2766 | kfree(slabs_by_inuse); | |
2767 | return 0; | |
2768 | } | |
2769 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2770 | ||
b9049e23 YG |
2771 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2772 | static int slab_mem_going_offline_callback(void *arg) | |
2773 | { | |
2774 | struct kmem_cache *s; | |
2775 | ||
2776 | down_read(&slub_lock); | |
2777 | list_for_each_entry(s, &slab_caches, list) | |
2778 | kmem_cache_shrink(s); | |
2779 | up_read(&slub_lock); | |
2780 | ||
2781 | return 0; | |
2782 | } | |
2783 | ||
2784 | static void slab_mem_offline_callback(void *arg) | |
2785 | { | |
2786 | struct kmem_cache_node *n; | |
2787 | struct kmem_cache *s; | |
2788 | struct memory_notify *marg = arg; | |
2789 | int offline_node; | |
2790 | ||
2791 | offline_node = marg->status_change_nid; | |
2792 | ||
2793 | /* | |
2794 | * If the node still has available memory. we need kmem_cache_node | |
2795 | * for it yet. | |
2796 | */ | |
2797 | if (offline_node < 0) | |
2798 | return; | |
2799 | ||
2800 | down_read(&slub_lock); | |
2801 | list_for_each_entry(s, &slab_caches, list) { | |
2802 | n = get_node(s, offline_node); | |
2803 | if (n) { | |
2804 | /* | |
2805 | * if n->nr_slabs > 0, slabs still exist on the node | |
2806 | * that is going down. We were unable to free them, | |
2807 | * and offline_pages() function shoudn't call this | |
2808 | * callback. So, we must fail. | |
2809 | */ | |
27bb628a | 2810 | BUG_ON(atomic_long_read(&n->nr_slabs)); |
b9049e23 YG |
2811 | |
2812 | s->node[offline_node] = NULL; | |
2813 | kmem_cache_free(kmalloc_caches, n); | |
2814 | } | |
2815 | } | |
2816 | up_read(&slub_lock); | |
2817 | } | |
2818 | ||
2819 | static int slab_mem_going_online_callback(void *arg) | |
2820 | { | |
2821 | struct kmem_cache_node *n; | |
2822 | struct kmem_cache *s; | |
2823 | struct memory_notify *marg = arg; | |
2824 | int nid = marg->status_change_nid; | |
2825 | int ret = 0; | |
2826 | ||
2827 | /* | |
2828 | * If the node's memory is already available, then kmem_cache_node is | |
2829 | * already created. Nothing to do. | |
2830 | */ | |
2831 | if (nid < 0) | |
2832 | return 0; | |
2833 | ||
2834 | /* | |
2835 | * We are bringing a node online. No memory is availabe yet. We must | |
2836 | * allocate a kmem_cache_node structure in order to bring the node | |
2837 | * online. | |
2838 | */ | |
2839 | down_read(&slub_lock); | |
2840 | list_for_each_entry(s, &slab_caches, list) { | |
2841 | /* | |
2842 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2843 | * since memory is not yet available from the node that | |
2844 | * is brought up. | |
2845 | */ | |
2846 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2847 | if (!n) { | |
2848 | ret = -ENOMEM; | |
2849 | goto out; | |
2850 | } | |
2851 | init_kmem_cache_node(n); | |
2852 | s->node[nid] = n; | |
2853 | } | |
2854 | out: | |
2855 | up_read(&slub_lock); | |
2856 | return ret; | |
2857 | } | |
2858 | ||
2859 | static int slab_memory_callback(struct notifier_block *self, | |
2860 | unsigned long action, void *arg) | |
2861 | { | |
2862 | int ret = 0; | |
2863 | ||
2864 | switch (action) { | |
2865 | case MEM_GOING_ONLINE: | |
2866 | ret = slab_mem_going_online_callback(arg); | |
2867 | break; | |
2868 | case MEM_GOING_OFFLINE: | |
2869 | ret = slab_mem_going_offline_callback(arg); | |
2870 | break; | |
2871 | case MEM_OFFLINE: | |
2872 | case MEM_CANCEL_ONLINE: | |
2873 | slab_mem_offline_callback(arg); | |
2874 | break; | |
2875 | case MEM_ONLINE: | |
2876 | case MEM_CANCEL_OFFLINE: | |
2877 | break; | |
2878 | } | |
2879 | ||
2880 | ret = notifier_from_errno(ret); | |
2881 | return ret; | |
2882 | } | |
2883 | ||
2884 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2885 | ||
81819f0f CL |
2886 | /******************************************************************** |
2887 | * Basic setup of slabs | |
2888 | *******************************************************************/ | |
2889 | ||
2890 | void __init kmem_cache_init(void) | |
2891 | { | |
2892 | int i; | |
4b356be0 | 2893 | int caches = 0; |
81819f0f | 2894 | |
4c93c355 CL |
2895 | init_alloc_cpu(); |
2896 | ||
81819f0f CL |
2897 | #ifdef CONFIG_NUMA |
2898 | /* | |
2899 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2900 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2901 | * kmem_cache_open for slab_state == DOWN. |
2902 | */ | |
2903 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2904 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2905 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 2906 | caches++; |
b9049e23 YG |
2907 | |
2908 | hotplug_memory_notifier(slab_memory_callback, 1); | |
81819f0f CL |
2909 | #endif |
2910 | ||
2911 | /* Able to allocate the per node structures */ | |
2912 | slab_state = PARTIAL; | |
2913 | ||
2914 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
2915 | if (KMALLOC_MIN_SIZE <= 64) { |
2916 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 2917 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
2918 | caches++; |
2919 | } | |
2920 | if (KMALLOC_MIN_SIZE <= 128) { | |
2921 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 2922 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
2923 | caches++; |
2924 | } | |
81819f0f | 2925 | |
331dc558 | 2926 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) { |
81819f0f CL |
2927 | create_kmalloc_cache(&kmalloc_caches[i], |
2928 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
2929 | caches++; |
2930 | } | |
81819f0f | 2931 | |
f1b26339 CL |
2932 | |
2933 | /* | |
2934 | * Patch up the size_index table if we have strange large alignment | |
2935 | * requirements for the kmalloc array. This is only the case for | |
2936 | * mips it seems. The standard arches will not generate any code here. | |
2937 | * | |
2938 | * Largest permitted alignment is 256 bytes due to the way we | |
2939 | * handle the index determination for the smaller caches. | |
2940 | * | |
2941 | * Make sure that nothing crazy happens if someone starts tinkering | |
2942 | * around with ARCH_KMALLOC_MINALIGN | |
2943 | */ | |
2944 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
2945 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
2946 | ||
12ad6843 | 2947 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
2948 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
2949 | ||
81819f0f CL |
2950 | slab_state = UP; |
2951 | ||
2952 | /* Provide the correct kmalloc names now that the caches are up */ | |
331dc558 | 2953 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) |
81819f0f CL |
2954 | kmalloc_caches[i]. name = |
2955 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
2956 | ||
2957 | #ifdef CONFIG_SMP | |
2958 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
2959 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
2960 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
2961 | #else | |
2962 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
2963 | #endif |
2964 | ||
81819f0f | 2965 | |
3adbefee IM |
2966 | printk(KERN_INFO |
2967 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
2968 | " CPUs=%d, Nodes=%d\n", |
2969 | caches, cache_line_size(), | |
81819f0f CL |
2970 | slub_min_order, slub_max_order, slub_min_objects, |
2971 | nr_cpu_ids, nr_node_ids); | |
2972 | } | |
2973 | ||
2974 | /* | |
2975 | * Find a mergeable slab cache | |
2976 | */ | |
2977 | static int slab_unmergeable(struct kmem_cache *s) | |
2978 | { | |
2979 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
2980 | return 1; | |
2981 | ||
331dc558 | 2982 | if ((s->flags & __PAGE_ALLOC_FALLBACK)) |
71c7a06f CL |
2983 | return 1; |
2984 | ||
c59def9f | 2985 | if (s->ctor) |
81819f0f CL |
2986 | return 1; |
2987 | ||
8ffa6875 CL |
2988 | /* |
2989 | * We may have set a slab to be unmergeable during bootstrap. | |
2990 | */ | |
2991 | if (s->refcount < 0) | |
2992 | return 1; | |
2993 | ||
81819f0f CL |
2994 | return 0; |
2995 | } | |
2996 | ||
2997 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 2998 | size_t align, unsigned long flags, const char *name, |
4ba9b9d0 | 2999 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f | 3000 | { |
5b95a4ac | 3001 | struct kmem_cache *s; |
81819f0f CL |
3002 | |
3003 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3004 | return NULL; | |
3005 | ||
c59def9f | 3006 | if (ctor) |
81819f0f CL |
3007 | return NULL; |
3008 | ||
3009 | size = ALIGN(size, sizeof(void *)); | |
3010 | align = calculate_alignment(flags, align, size); | |
3011 | size = ALIGN(size, align); | |
ba0268a8 | 3012 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3013 | |
5b95a4ac | 3014 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3015 | if (slab_unmergeable(s)) |
3016 | continue; | |
3017 | ||
3018 | if (size > s->size) | |
3019 | continue; | |
3020 | ||
ba0268a8 | 3021 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3022 | continue; |
3023 | /* | |
3024 | * Check if alignment is compatible. | |
3025 | * Courtesy of Adrian Drzewiecki | |
3026 | */ | |
06428780 | 3027 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3028 | continue; |
3029 | ||
3030 | if (s->size - size >= sizeof(void *)) | |
3031 | continue; | |
3032 | ||
3033 | return s; | |
3034 | } | |
3035 | return NULL; | |
3036 | } | |
3037 | ||
3038 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
3039 | size_t align, unsigned long flags, | |
4ba9b9d0 | 3040 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
3041 | { |
3042 | struct kmem_cache *s; | |
3043 | ||
3044 | down_write(&slub_lock); | |
ba0268a8 | 3045 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3046 | if (s) { |
42a9fdbb CL |
3047 | int cpu; |
3048 | ||
81819f0f CL |
3049 | s->refcount++; |
3050 | /* | |
3051 | * Adjust the object sizes so that we clear | |
3052 | * the complete object on kzalloc. | |
3053 | */ | |
3054 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3055 | |
3056 | /* | |
3057 | * And then we need to update the object size in the | |
3058 | * per cpu structures | |
3059 | */ | |
3060 | for_each_online_cpu(cpu) | |
3061 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
81819f0f | 3062 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3063 | up_write(&slub_lock); |
81819f0f CL |
3064 | if (sysfs_slab_alias(s, name)) |
3065 | goto err; | |
a0e1d1be CL |
3066 | return s; |
3067 | } | |
3068 | s = kmalloc(kmem_size, GFP_KERNEL); | |
3069 | if (s) { | |
3070 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3071 | size, align, flags, ctor)) { |
81819f0f | 3072 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
3073 | up_write(&slub_lock); |
3074 | if (sysfs_slab_add(s)) | |
3075 | goto err; | |
3076 | return s; | |
3077 | } | |
3078 | kfree(s); | |
81819f0f CL |
3079 | } |
3080 | up_write(&slub_lock); | |
81819f0f CL |
3081 | |
3082 | err: | |
81819f0f CL |
3083 | if (flags & SLAB_PANIC) |
3084 | panic("Cannot create slabcache %s\n", name); | |
3085 | else | |
3086 | s = NULL; | |
3087 | return s; | |
3088 | } | |
3089 | EXPORT_SYMBOL(kmem_cache_create); | |
3090 | ||
81819f0f | 3091 | #ifdef CONFIG_SMP |
81819f0f | 3092 | /* |
672bba3a CL |
3093 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3094 | * necessary. | |
81819f0f CL |
3095 | */ |
3096 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3097 | unsigned long action, void *hcpu) | |
3098 | { | |
3099 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3100 | struct kmem_cache *s; |
3101 | unsigned long flags; | |
81819f0f CL |
3102 | |
3103 | switch (action) { | |
4c93c355 CL |
3104 | case CPU_UP_PREPARE: |
3105 | case CPU_UP_PREPARE_FROZEN: | |
3106 | init_alloc_cpu_cpu(cpu); | |
3107 | down_read(&slub_lock); | |
3108 | list_for_each_entry(s, &slab_caches, list) | |
3109 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3110 | GFP_KERNEL); | |
3111 | up_read(&slub_lock); | |
3112 | break; | |
3113 | ||
81819f0f | 3114 | case CPU_UP_CANCELED: |
8bb78442 | 3115 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3116 | case CPU_DEAD: |
8bb78442 | 3117 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3118 | down_read(&slub_lock); |
3119 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3120 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3121 | ||
5b95a4ac CL |
3122 | local_irq_save(flags); |
3123 | __flush_cpu_slab(s, cpu); | |
3124 | local_irq_restore(flags); | |
4c93c355 CL |
3125 | free_kmem_cache_cpu(c, cpu); |
3126 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3127 | } |
3128 | up_read(&slub_lock); | |
81819f0f CL |
3129 | break; |
3130 | default: | |
3131 | break; | |
3132 | } | |
3133 | return NOTIFY_OK; | |
3134 | } | |
3135 | ||
06428780 | 3136 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3137 | .notifier_call = slab_cpuup_callback |
06428780 | 3138 | }; |
81819f0f CL |
3139 | |
3140 | #endif | |
3141 | ||
81819f0f CL |
3142 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
3143 | { | |
aadb4bc4 CL |
3144 | struct kmem_cache *s; |
3145 | ||
331dc558 | 3146 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3147 | return kmalloc_large(size, gfpflags); |
3148 | ||
aadb4bc4 | 3149 | s = get_slab(size, gfpflags); |
81819f0f | 3150 | |
2408c550 | 3151 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3152 | return s; |
81819f0f | 3153 | |
ce15fea8 | 3154 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3155 | } |
3156 | ||
3157 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
3158 | int node, void *caller) | |
3159 | { | |
aadb4bc4 CL |
3160 | struct kmem_cache *s; |
3161 | ||
331dc558 | 3162 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3163 | return kmalloc_large(size, gfpflags); |
3164 | ||
aadb4bc4 | 3165 | s = get_slab(size, gfpflags); |
81819f0f | 3166 | |
2408c550 | 3167 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3168 | return s; |
81819f0f | 3169 | |
ce15fea8 | 3170 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3171 | } |
3172 | ||
41ecc55b | 3173 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
3174 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3175 | unsigned long *map) | |
53e15af0 CL |
3176 | { |
3177 | void *p; | |
a973e9dd | 3178 | void *addr = page_address(page); |
53e15af0 CL |
3179 | |
3180 | if (!check_slab(s, page) || | |
3181 | !on_freelist(s, page, NULL)) | |
3182 | return 0; | |
3183 | ||
3184 | /* Now we know that a valid freelist exists */ | |
3185 | bitmap_zero(map, s->objects); | |
3186 | ||
7656c72b CL |
3187 | for_each_free_object(p, s, page->freelist) { |
3188 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3189 | if (!check_object(s, page, p, 0)) |
3190 | return 0; | |
3191 | } | |
3192 | ||
7656c72b CL |
3193 | for_each_object(p, s, addr) |
3194 | if (!test_bit(slab_index(p, s, addr), map)) | |
53e15af0 CL |
3195 | if (!check_object(s, page, p, 1)) |
3196 | return 0; | |
3197 | return 1; | |
3198 | } | |
3199 | ||
434e245d CL |
3200 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3201 | unsigned long *map) | |
53e15af0 CL |
3202 | { |
3203 | if (slab_trylock(page)) { | |
434e245d | 3204 | validate_slab(s, page, map); |
53e15af0 CL |
3205 | slab_unlock(page); |
3206 | } else | |
3207 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3208 | s->name, page); | |
3209 | ||
3210 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3211 | if (!SlabDebug(page)) |
3212 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3213 | "on slab 0x%p\n", s->name, page); |
3214 | } else { | |
35e5d7ee CL |
3215 | if (SlabDebug(page)) |
3216 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3217 | "slab 0x%p\n", s->name, page); |
3218 | } | |
3219 | } | |
3220 | ||
434e245d CL |
3221 | static int validate_slab_node(struct kmem_cache *s, |
3222 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3223 | { |
3224 | unsigned long count = 0; | |
3225 | struct page *page; | |
3226 | unsigned long flags; | |
3227 | ||
3228 | spin_lock_irqsave(&n->list_lock, flags); | |
3229 | ||
3230 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3231 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3232 | count++; |
3233 | } | |
3234 | if (count != n->nr_partial) | |
3235 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3236 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3237 | ||
3238 | if (!(s->flags & SLAB_STORE_USER)) | |
3239 | goto out; | |
3240 | ||
3241 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3242 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3243 | count++; |
3244 | } | |
3245 | if (count != atomic_long_read(&n->nr_slabs)) | |
3246 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3247 | "counter=%ld\n", s->name, count, | |
3248 | atomic_long_read(&n->nr_slabs)); | |
3249 | ||
3250 | out: | |
3251 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3252 | return count; | |
3253 | } | |
3254 | ||
434e245d | 3255 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3256 | { |
3257 | int node; | |
3258 | unsigned long count = 0; | |
434e245d CL |
3259 | unsigned long *map = kmalloc(BITS_TO_LONGS(s->objects) * |
3260 | sizeof(unsigned long), GFP_KERNEL); | |
3261 | ||
3262 | if (!map) | |
3263 | return -ENOMEM; | |
53e15af0 CL |
3264 | |
3265 | flush_all(s); | |
f64dc58c | 3266 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3267 | struct kmem_cache_node *n = get_node(s, node); |
3268 | ||
434e245d | 3269 | count += validate_slab_node(s, n, map); |
53e15af0 | 3270 | } |
434e245d | 3271 | kfree(map); |
53e15af0 CL |
3272 | return count; |
3273 | } | |
3274 | ||
b3459709 CL |
3275 | #ifdef SLUB_RESILIENCY_TEST |
3276 | static void resiliency_test(void) | |
3277 | { | |
3278 | u8 *p; | |
3279 | ||
3280 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3281 | printk(KERN_ERR "-----------------------\n"); | |
3282 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3283 | ||
3284 | p = kzalloc(16, GFP_KERNEL); | |
3285 | p[16] = 0x12; | |
3286 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3287 | " 0x12->0x%p\n\n", p + 16); | |
3288 | ||
3289 | validate_slab_cache(kmalloc_caches + 4); | |
3290 | ||
3291 | /* Hmmm... The next two are dangerous */ | |
3292 | p = kzalloc(32, GFP_KERNEL); | |
3293 | p[32 + sizeof(void *)] = 0x34; | |
3294 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3295 | " 0x34 -> -0x%p\n", p); |
3296 | printk(KERN_ERR | |
3297 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3298 | |
3299 | validate_slab_cache(kmalloc_caches + 5); | |
3300 | p = kzalloc(64, GFP_KERNEL); | |
3301 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3302 | *p = 0x56; | |
3303 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3304 | p); | |
3adbefee IM |
3305 | printk(KERN_ERR |
3306 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3307 | validate_slab_cache(kmalloc_caches + 6); |
3308 | ||
3309 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3310 | p = kzalloc(128, GFP_KERNEL); | |
3311 | kfree(p); | |
3312 | *p = 0x78; | |
3313 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3314 | validate_slab_cache(kmalloc_caches + 7); | |
3315 | ||
3316 | p = kzalloc(256, GFP_KERNEL); | |
3317 | kfree(p); | |
3318 | p[50] = 0x9a; | |
3adbefee IM |
3319 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3320 | p); | |
b3459709 CL |
3321 | validate_slab_cache(kmalloc_caches + 8); |
3322 | ||
3323 | p = kzalloc(512, GFP_KERNEL); | |
3324 | kfree(p); | |
3325 | p[512] = 0xab; | |
3326 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3327 | validate_slab_cache(kmalloc_caches + 9); | |
3328 | } | |
3329 | #else | |
3330 | static void resiliency_test(void) {}; | |
3331 | #endif | |
3332 | ||
88a420e4 | 3333 | /* |
672bba3a | 3334 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3335 | * and freed. |
3336 | */ | |
3337 | ||
3338 | struct location { | |
3339 | unsigned long count; | |
3340 | void *addr; | |
45edfa58 CL |
3341 | long long sum_time; |
3342 | long min_time; | |
3343 | long max_time; | |
3344 | long min_pid; | |
3345 | long max_pid; | |
3346 | cpumask_t cpus; | |
3347 | nodemask_t nodes; | |
88a420e4 CL |
3348 | }; |
3349 | ||
3350 | struct loc_track { | |
3351 | unsigned long max; | |
3352 | unsigned long count; | |
3353 | struct location *loc; | |
3354 | }; | |
3355 | ||
3356 | static void free_loc_track(struct loc_track *t) | |
3357 | { | |
3358 | if (t->max) | |
3359 | free_pages((unsigned long)t->loc, | |
3360 | get_order(sizeof(struct location) * t->max)); | |
3361 | } | |
3362 | ||
68dff6a9 | 3363 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3364 | { |
3365 | struct location *l; | |
3366 | int order; | |
3367 | ||
88a420e4 CL |
3368 | order = get_order(sizeof(struct location) * max); |
3369 | ||
68dff6a9 | 3370 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3371 | if (!l) |
3372 | return 0; | |
3373 | ||
3374 | if (t->count) { | |
3375 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3376 | free_loc_track(t); | |
3377 | } | |
3378 | t->max = max; | |
3379 | t->loc = l; | |
3380 | return 1; | |
3381 | } | |
3382 | ||
3383 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3384 | const struct track *track) |
88a420e4 CL |
3385 | { |
3386 | long start, end, pos; | |
3387 | struct location *l; | |
3388 | void *caddr; | |
45edfa58 | 3389 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3390 | |
3391 | start = -1; | |
3392 | end = t->count; | |
3393 | ||
3394 | for ( ; ; ) { | |
3395 | pos = start + (end - start + 1) / 2; | |
3396 | ||
3397 | /* | |
3398 | * There is nothing at "end". If we end up there | |
3399 | * we need to add something to before end. | |
3400 | */ | |
3401 | if (pos == end) | |
3402 | break; | |
3403 | ||
3404 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3405 | if (track->addr == caddr) { |
3406 | ||
3407 | l = &t->loc[pos]; | |
3408 | l->count++; | |
3409 | if (track->when) { | |
3410 | l->sum_time += age; | |
3411 | if (age < l->min_time) | |
3412 | l->min_time = age; | |
3413 | if (age > l->max_time) | |
3414 | l->max_time = age; | |
3415 | ||
3416 | if (track->pid < l->min_pid) | |
3417 | l->min_pid = track->pid; | |
3418 | if (track->pid > l->max_pid) | |
3419 | l->max_pid = track->pid; | |
3420 | ||
3421 | cpu_set(track->cpu, l->cpus); | |
3422 | } | |
3423 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3424 | return 1; |
3425 | } | |
3426 | ||
45edfa58 | 3427 | if (track->addr < caddr) |
88a420e4 CL |
3428 | end = pos; |
3429 | else | |
3430 | start = pos; | |
3431 | } | |
3432 | ||
3433 | /* | |
672bba3a | 3434 | * Not found. Insert new tracking element. |
88a420e4 | 3435 | */ |
68dff6a9 | 3436 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3437 | return 0; |
3438 | ||
3439 | l = t->loc + pos; | |
3440 | if (pos < t->count) | |
3441 | memmove(l + 1, l, | |
3442 | (t->count - pos) * sizeof(struct location)); | |
3443 | t->count++; | |
3444 | l->count = 1; | |
45edfa58 CL |
3445 | l->addr = track->addr; |
3446 | l->sum_time = age; | |
3447 | l->min_time = age; | |
3448 | l->max_time = age; | |
3449 | l->min_pid = track->pid; | |
3450 | l->max_pid = track->pid; | |
3451 | cpus_clear(l->cpus); | |
3452 | cpu_set(track->cpu, l->cpus); | |
3453 | nodes_clear(l->nodes); | |
3454 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3455 | return 1; |
3456 | } | |
3457 | ||
3458 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3459 | struct page *page, enum track_item alloc) | |
3460 | { | |
a973e9dd | 3461 | void *addr = page_address(page); |
7656c72b | 3462 | DECLARE_BITMAP(map, s->objects); |
88a420e4 CL |
3463 | void *p; |
3464 | ||
3465 | bitmap_zero(map, s->objects); | |
7656c72b CL |
3466 | for_each_free_object(p, s, page->freelist) |
3467 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3468 | |
7656c72b | 3469 | for_each_object(p, s, addr) |
45edfa58 CL |
3470 | if (!test_bit(slab_index(p, s, addr), map)) |
3471 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3472 | } |
3473 | ||
3474 | static int list_locations(struct kmem_cache *s, char *buf, | |
3475 | enum track_item alloc) | |
3476 | { | |
e374d483 | 3477 | int len = 0; |
88a420e4 | 3478 | unsigned long i; |
68dff6a9 | 3479 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3480 | int node; |
3481 | ||
68dff6a9 | 3482 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3483 | GFP_TEMPORARY)) |
68dff6a9 | 3484 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3485 | |
3486 | /* Push back cpu slabs */ | |
3487 | flush_all(s); | |
3488 | ||
f64dc58c | 3489 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3490 | struct kmem_cache_node *n = get_node(s, node); |
3491 | unsigned long flags; | |
3492 | struct page *page; | |
3493 | ||
9e86943b | 3494 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3495 | continue; |
3496 | ||
3497 | spin_lock_irqsave(&n->list_lock, flags); | |
3498 | list_for_each_entry(page, &n->partial, lru) | |
3499 | process_slab(&t, s, page, alloc); | |
3500 | list_for_each_entry(page, &n->full, lru) | |
3501 | process_slab(&t, s, page, alloc); | |
3502 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3503 | } | |
3504 | ||
3505 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3506 | struct location *l = &t.loc[i]; |
88a420e4 | 3507 | |
e374d483 | 3508 | if (len > PAGE_SIZE - 100) |
88a420e4 | 3509 | break; |
e374d483 | 3510 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3511 | |
3512 | if (l->addr) | |
e374d483 | 3513 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3514 | else |
e374d483 | 3515 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3516 | |
3517 | if (l->sum_time != l->min_time) { | |
3518 | unsigned long remainder; | |
3519 | ||
e374d483 | 3520 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
45edfa58 CL |
3521 | l->min_time, |
3522 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3523 | l->max_time); | |
3524 | } else | |
e374d483 | 3525 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3526 | l->min_time); |
3527 | ||
3528 | if (l->min_pid != l->max_pid) | |
e374d483 | 3529 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3530 | l->min_pid, l->max_pid); |
3531 | else | |
e374d483 | 3532 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3533 | l->min_pid); |
3534 | ||
84966343 | 3535 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
e374d483 HH |
3536 | len < PAGE_SIZE - 60) { |
3537 | len += sprintf(buf + len, " cpus="); | |
3538 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3539 | l->cpus); |
3540 | } | |
3541 | ||
84966343 | 3542 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3543 | len < PAGE_SIZE - 60) { |
3544 | len += sprintf(buf + len, " nodes="); | |
3545 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3546 | l->nodes); |
3547 | } | |
3548 | ||
e374d483 | 3549 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3550 | } |
3551 | ||
3552 | free_loc_track(&t); | |
3553 | if (!t.count) | |
e374d483 HH |
3554 | len += sprintf(buf, "No data\n"); |
3555 | return len; | |
88a420e4 CL |
3556 | } |
3557 | ||
81819f0f CL |
3558 | enum slab_stat_type { |
3559 | SL_FULL, | |
3560 | SL_PARTIAL, | |
3561 | SL_CPU, | |
3562 | SL_OBJECTS | |
3563 | }; | |
3564 | ||
3565 | #define SO_FULL (1 << SL_FULL) | |
3566 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
3567 | #define SO_CPU (1 << SL_CPU) | |
3568 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
3569 | ||
3570 | static unsigned long slab_objects(struct kmem_cache *s, | |
3571 | char *buf, unsigned long flags) | |
3572 | { | |
3573 | unsigned long total = 0; | |
3574 | int cpu; | |
3575 | int node; | |
3576 | int x; | |
3577 | unsigned long *nodes; | |
3578 | unsigned long *per_cpu; | |
3579 | ||
3580 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
3581 | per_cpu = nodes + nr_node_ids; | |
3582 | ||
3583 | for_each_possible_cpu(cpu) { | |
dfb4f096 CL |
3584 | struct page *page; |
3585 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
81819f0f | 3586 | |
dfb4f096 CL |
3587 | if (!c) |
3588 | continue; | |
3589 | ||
3590 | page = c->page; | |
ee3c72a1 CL |
3591 | node = c->node; |
3592 | if (node < 0) | |
3593 | continue; | |
81819f0f | 3594 | if (page) { |
81819f0f | 3595 | if (flags & SO_CPU) { |
81819f0f CL |
3596 | if (flags & SO_OBJECTS) |
3597 | x = page->inuse; | |
3598 | else | |
3599 | x = 1; | |
3600 | total += x; | |
ee3c72a1 | 3601 | nodes[node] += x; |
81819f0f | 3602 | } |
ee3c72a1 | 3603 | per_cpu[node]++; |
81819f0f CL |
3604 | } |
3605 | } | |
3606 | ||
f64dc58c | 3607 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3608 | struct kmem_cache_node *n = get_node(s, node); |
3609 | ||
3610 | if (flags & SO_PARTIAL) { | |
3611 | if (flags & SO_OBJECTS) | |
3612 | x = count_partial(n); | |
3613 | else | |
3614 | x = n->nr_partial; | |
3615 | total += x; | |
3616 | nodes[node] += x; | |
3617 | } | |
3618 | ||
3619 | if (flags & SO_FULL) { | |
9e86943b | 3620 | int full_slabs = atomic_long_read(&n->nr_slabs) |
81819f0f CL |
3621 | - per_cpu[node] |
3622 | - n->nr_partial; | |
3623 | ||
3624 | if (flags & SO_OBJECTS) | |
3625 | x = full_slabs * s->objects; | |
3626 | else | |
3627 | x = full_slabs; | |
3628 | total += x; | |
3629 | nodes[node] += x; | |
3630 | } | |
3631 | } | |
3632 | ||
3633 | x = sprintf(buf, "%lu", total); | |
3634 | #ifdef CONFIG_NUMA | |
f64dc58c | 3635 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3636 | if (nodes[node]) |
3637 | x += sprintf(buf + x, " N%d=%lu", | |
3638 | node, nodes[node]); | |
3639 | #endif | |
3640 | kfree(nodes); | |
3641 | return x + sprintf(buf + x, "\n"); | |
3642 | } | |
3643 | ||
3644 | static int any_slab_objects(struct kmem_cache *s) | |
3645 | { | |
3646 | int node; | |
3647 | int cpu; | |
3648 | ||
dfb4f096 CL |
3649 | for_each_possible_cpu(cpu) { |
3650 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
3651 | ||
3652 | if (c && c->page) | |
81819f0f | 3653 | return 1; |
dfb4f096 | 3654 | } |
81819f0f | 3655 | |
dfb4f096 | 3656 | for_each_online_node(node) { |
81819f0f CL |
3657 | struct kmem_cache_node *n = get_node(s, node); |
3658 | ||
dfb4f096 CL |
3659 | if (!n) |
3660 | continue; | |
3661 | ||
9e86943b | 3662 | if (n->nr_partial || atomic_long_read(&n->nr_slabs)) |
81819f0f CL |
3663 | return 1; |
3664 | } | |
3665 | return 0; | |
3666 | } | |
3667 | ||
3668 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3669 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3670 | ||
3671 | struct slab_attribute { | |
3672 | struct attribute attr; | |
3673 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3674 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3675 | }; | |
3676 | ||
3677 | #define SLAB_ATTR_RO(_name) \ | |
3678 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3679 | ||
3680 | #define SLAB_ATTR(_name) \ | |
3681 | static struct slab_attribute _name##_attr = \ | |
3682 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3683 | ||
81819f0f CL |
3684 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3685 | { | |
3686 | return sprintf(buf, "%d\n", s->size); | |
3687 | } | |
3688 | SLAB_ATTR_RO(slab_size); | |
3689 | ||
3690 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3691 | { | |
3692 | return sprintf(buf, "%d\n", s->align); | |
3693 | } | |
3694 | SLAB_ATTR_RO(align); | |
3695 | ||
3696 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3697 | { | |
3698 | return sprintf(buf, "%d\n", s->objsize); | |
3699 | } | |
3700 | SLAB_ATTR_RO(object_size); | |
3701 | ||
3702 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3703 | { | |
3704 | return sprintf(buf, "%d\n", s->objects); | |
3705 | } | |
3706 | SLAB_ATTR_RO(objs_per_slab); | |
3707 | ||
3708 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3709 | { | |
3710 | return sprintf(buf, "%d\n", s->order); | |
3711 | } | |
3712 | SLAB_ATTR_RO(order); | |
3713 | ||
3714 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3715 | { | |
3716 | if (s->ctor) { | |
3717 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3718 | ||
3719 | return n + sprintf(buf + n, "\n"); | |
3720 | } | |
3721 | return 0; | |
3722 | } | |
3723 | SLAB_ATTR_RO(ctor); | |
3724 | ||
81819f0f CL |
3725 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3726 | { | |
3727 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3728 | } | |
3729 | SLAB_ATTR_RO(aliases); | |
3730 | ||
3731 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3732 | { | |
3733 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
3734 | } | |
3735 | SLAB_ATTR_RO(slabs); | |
3736 | ||
3737 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3738 | { | |
3739 | return slab_objects(s, buf, SO_PARTIAL); | |
3740 | } | |
3741 | SLAB_ATTR_RO(partial); | |
3742 | ||
3743 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3744 | { | |
3745 | return slab_objects(s, buf, SO_CPU); | |
3746 | } | |
3747 | SLAB_ATTR_RO(cpu_slabs); | |
3748 | ||
3749 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3750 | { | |
3751 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
3752 | } | |
3753 | SLAB_ATTR_RO(objects); | |
3754 | ||
3755 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
3756 | { | |
3757 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3758 | } | |
3759 | ||
3760 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3761 | const char *buf, size_t length) | |
3762 | { | |
3763 | s->flags &= ~SLAB_DEBUG_FREE; | |
3764 | if (buf[0] == '1') | |
3765 | s->flags |= SLAB_DEBUG_FREE; | |
3766 | return length; | |
3767 | } | |
3768 | SLAB_ATTR(sanity_checks); | |
3769 | ||
3770 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3771 | { | |
3772 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3773 | } | |
3774 | ||
3775 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3776 | size_t length) | |
3777 | { | |
3778 | s->flags &= ~SLAB_TRACE; | |
3779 | if (buf[0] == '1') | |
3780 | s->flags |= SLAB_TRACE; | |
3781 | return length; | |
3782 | } | |
3783 | SLAB_ATTR(trace); | |
3784 | ||
3785 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3786 | { | |
3787 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3788 | } | |
3789 | ||
3790 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3791 | const char *buf, size_t length) | |
3792 | { | |
3793 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3794 | if (buf[0] == '1') | |
3795 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3796 | return length; | |
3797 | } | |
3798 | SLAB_ATTR(reclaim_account); | |
3799 | ||
3800 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3801 | { | |
5af60839 | 3802 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3803 | } |
3804 | SLAB_ATTR_RO(hwcache_align); | |
3805 | ||
3806 | #ifdef CONFIG_ZONE_DMA | |
3807 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3808 | { | |
3809 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3810 | } | |
3811 | SLAB_ATTR_RO(cache_dma); | |
3812 | #endif | |
3813 | ||
3814 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3815 | { | |
3816 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3817 | } | |
3818 | SLAB_ATTR_RO(destroy_by_rcu); | |
3819 | ||
3820 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3821 | { | |
3822 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3823 | } | |
3824 | ||
3825 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3826 | const char *buf, size_t length) | |
3827 | { | |
3828 | if (any_slab_objects(s)) | |
3829 | return -EBUSY; | |
3830 | ||
3831 | s->flags &= ~SLAB_RED_ZONE; | |
3832 | if (buf[0] == '1') | |
3833 | s->flags |= SLAB_RED_ZONE; | |
3834 | calculate_sizes(s); | |
3835 | return length; | |
3836 | } | |
3837 | SLAB_ATTR(red_zone); | |
3838 | ||
3839 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3840 | { | |
3841 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3842 | } | |
3843 | ||
3844 | static ssize_t poison_store(struct kmem_cache *s, | |
3845 | const char *buf, size_t length) | |
3846 | { | |
3847 | if (any_slab_objects(s)) | |
3848 | return -EBUSY; | |
3849 | ||
3850 | s->flags &= ~SLAB_POISON; | |
3851 | if (buf[0] == '1') | |
3852 | s->flags |= SLAB_POISON; | |
3853 | calculate_sizes(s); | |
3854 | return length; | |
3855 | } | |
3856 | SLAB_ATTR(poison); | |
3857 | ||
3858 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3859 | { | |
3860 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3861 | } | |
3862 | ||
3863 | static ssize_t store_user_store(struct kmem_cache *s, | |
3864 | const char *buf, size_t length) | |
3865 | { | |
3866 | if (any_slab_objects(s)) | |
3867 | return -EBUSY; | |
3868 | ||
3869 | s->flags &= ~SLAB_STORE_USER; | |
3870 | if (buf[0] == '1') | |
3871 | s->flags |= SLAB_STORE_USER; | |
3872 | calculate_sizes(s); | |
3873 | return length; | |
3874 | } | |
3875 | SLAB_ATTR(store_user); | |
3876 | ||
53e15af0 CL |
3877 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
3878 | { | |
3879 | return 0; | |
3880 | } | |
3881 | ||
3882 | static ssize_t validate_store(struct kmem_cache *s, | |
3883 | const char *buf, size_t length) | |
3884 | { | |
434e245d CL |
3885 | int ret = -EINVAL; |
3886 | ||
3887 | if (buf[0] == '1') { | |
3888 | ret = validate_slab_cache(s); | |
3889 | if (ret >= 0) | |
3890 | ret = length; | |
3891 | } | |
3892 | return ret; | |
53e15af0 CL |
3893 | } |
3894 | SLAB_ATTR(validate); | |
3895 | ||
2086d26a CL |
3896 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
3897 | { | |
3898 | return 0; | |
3899 | } | |
3900 | ||
3901 | static ssize_t shrink_store(struct kmem_cache *s, | |
3902 | const char *buf, size_t length) | |
3903 | { | |
3904 | if (buf[0] == '1') { | |
3905 | int rc = kmem_cache_shrink(s); | |
3906 | ||
3907 | if (rc) | |
3908 | return rc; | |
3909 | } else | |
3910 | return -EINVAL; | |
3911 | return length; | |
3912 | } | |
3913 | SLAB_ATTR(shrink); | |
3914 | ||
88a420e4 CL |
3915 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
3916 | { | |
3917 | if (!(s->flags & SLAB_STORE_USER)) | |
3918 | return -ENOSYS; | |
3919 | return list_locations(s, buf, TRACK_ALLOC); | |
3920 | } | |
3921 | SLAB_ATTR_RO(alloc_calls); | |
3922 | ||
3923 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
3924 | { | |
3925 | if (!(s->flags & SLAB_STORE_USER)) | |
3926 | return -ENOSYS; | |
3927 | return list_locations(s, buf, TRACK_FREE); | |
3928 | } | |
3929 | SLAB_ATTR_RO(free_calls); | |
3930 | ||
81819f0f | 3931 | #ifdef CONFIG_NUMA |
9824601e | 3932 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 3933 | { |
9824601e | 3934 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
3935 | } |
3936 | ||
9824601e | 3937 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
3938 | const char *buf, size_t length) |
3939 | { | |
3940 | int n = simple_strtoul(buf, NULL, 10); | |
3941 | ||
3942 | if (n < 100) | |
9824601e | 3943 | s->remote_node_defrag_ratio = n * 10; |
81819f0f CL |
3944 | return length; |
3945 | } | |
9824601e | 3946 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
3947 | #endif |
3948 | ||
8ff12cfc CL |
3949 | #ifdef CONFIG_SLUB_STATS |
3950 | ||
3951 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) | |
3952 | { | |
3953 | unsigned long sum = 0; | |
3954 | int cpu; | |
3955 | int len; | |
3956 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
3957 | ||
3958 | if (!data) | |
3959 | return -ENOMEM; | |
3960 | ||
3961 | for_each_online_cpu(cpu) { | |
3962 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
3963 | ||
3964 | data[cpu] = x; | |
3965 | sum += x; | |
3966 | } | |
3967 | ||
3968 | len = sprintf(buf, "%lu", sum); | |
3969 | ||
3970 | for_each_online_cpu(cpu) { | |
3971 | if (data[cpu] && len < PAGE_SIZE - 20) | |
3972 | len += sprintf(buf + len, " c%d=%u", cpu, data[cpu]); | |
3973 | } | |
3974 | kfree(data); | |
3975 | return len + sprintf(buf + len, "\n"); | |
3976 | } | |
3977 | ||
3978 | #define STAT_ATTR(si, text) \ | |
3979 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
3980 | { \ | |
3981 | return show_stat(s, buf, si); \ | |
3982 | } \ | |
3983 | SLAB_ATTR_RO(text); \ | |
3984 | ||
3985 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
3986 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
3987 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
3988 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
3989 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
3990 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
3991 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
3992 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
3993 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
3994 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
3995 | STAT_ATTR(FREE_SLAB, free_slab); | |
3996 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
3997 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
3998 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
3999 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4000 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4001 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
4002 | ||
4003 | #endif | |
4004 | ||
06428780 | 4005 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4006 | &slab_size_attr.attr, |
4007 | &object_size_attr.attr, | |
4008 | &objs_per_slab_attr.attr, | |
4009 | &order_attr.attr, | |
4010 | &objects_attr.attr, | |
4011 | &slabs_attr.attr, | |
4012 | &partial_attr.attr, | |
4013 | &cpu_slabs_attr.attr, | |
4014 | &ctor_attr.attr, | |
81819f0f CL |
4015 | &aliases_attr.attr, |
4016 | &align_attr.attr, | |
4017 | &sanity_checks_attr.attr, | |
4018 | &trace_attr.attr, | |
4019 | &hwcache_align_attr.attr, | |
4020 | &reclaim_account_attr.attr, | |
4021 | &destroy_by_rcu_attr.attr, | |
4022 | &red_zone_attr.attr, | |
4023 | &poison_attr.attr, | |
4024 | &store_user_attr.attr, | |
53e15af0 | 4025 | &validate_attr.attr, |
2086d26a | 4026 | &shrink_attr.attr, |
88a420e4 CL |
4027 | &alloc_calls_attr.attr, |
4028 | &free_calls_attr.attr, | |
81819f0f CL |
4029 | #ifdef CONFIG_ZONE_DMA |
4030 | &cache_dma_attr.attr, | |
4031 | #endif | |
4032 | #ifdef CONFIG_NUMA | |
9824601e | 4033 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4034 | #endif |
4035 | #ifdef CONFIG_SLUB_STATS | |
4036 | &alloc_fastpath_attr.attr, | |
4037 | &alloc_slowpath_attr.attr, | |
4038 | &free_fastpath_attr.attr, | |
4039 | &free_slowpath_attr.attr, | |
4040 | &free_frozen_attr.attr, | |
4041 | &free_add_partial_attr.attr, | |
4042 | &free_remove_partial_attr.attr, | |
4043 | &alloc_from_partial_attr.attr, | |
4044 | &alloc_slab_attr.attr, | |
4045 | &alloc_refill_attr.attr, | |
4046 | &free_slab_attr.attr, | |
4047 | &cpuslab_flush_attr.attr, | |
4048 | &deactivate_full_attr.attr, | |
4049 | &deactivate_empty_attr.attr, | |
4050 | &deactivate_to_head_attr.attr, | |
4051 | &deactivate_to_tail_attr.attr, | |
4052 | &deactivate_remote_frees_attr.attr, | |
81819f0f CL |
4053 | #endif |
4054 | NULL | |
4055 | }; | |
4056 | ||
4057 | static struct attribute_group slab_attr_group = { | |
4058 | .attrs = slab_attrs, | |
4059 | }; | |
4060 | ||
4061 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4062 | struct attribute *attr, | |
4063 | char *buf) | |
4064 | { | |
4065 | struct slab_attribute *attribute; | |
4066 | struct kmem_cache *s; | |
4067 | int err; | |
4068 | ||
4069 | attribute = to_slab_attr(attr); | |
4070 | s = to_slab(kobj); | |
4071 | ||
4072 | if (!attribute->show) | |
4073 | return -EIO; | |
4074 | ||
4075 | err = attribute->show(s, buf); | |
4076 | ||
4077 | return err; | |
4078 | } | |
4079 | ||
4080 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4081 | struct attribute *attr, | |
4082 | const char *buf, size_t len) | |
4083 | { | |
4084 | struct slab_attribute *attribute; | |
4085 | struct kmem_cache *s; | |
4086 | int err; | |
4087 | ||
4088 | attribute = to_slab_attr(attr); | |
4089 | s = to_slab(kobj); | |
4090 | ||
4091 | if (!attribute->store) | |
4092 | return -EIO; | |
4093 | ||
4094 | err = attribute->store(s, buf, len); | |
4095 | ||
4096 | return err; | |
4097 | } | |
4098 | ||
151c602f CL |
4099 | static void kmem_cache_release(struct kobject *kobj) |
4100 | { | |
4101 | struct kmem_cache *s = to_slab(kobj); | |
4102 | ||
4103 | kfree(s); | |
4104 | } | |
4105 | ||
81819f0f CL |
4106 | static struct sysfs_ops slab_sysfs_ops = { |
4107 | .show = slab_attr_show, | |
4108 | .store = slab_attr_store, | |
4109 | }; | |
4110 | ||
4111 | static struct kobj_type slab_ktype = { | |
4112 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4113 | .release = kmem_cache_release |
81819f0f CL |
4114 | }; |
4115 | ||
4116 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4117 | { | |
4118 | struct kobj_type *ktype = get_ktype(kobj); | |
4119 | ||
4120 | if (ktype == &slab_ktype) | |
4121 | return 1; | |
4122 | return 0; | |
4123 | } | |
4124 | ||
4125 | static struct kset_uevent_ops slab_uevent_ops = { | |
4126 | .filter = uevent_filter, | |
4127 | }; | |
4128 | ||
27c3a314 | 4129 | static struct kset *slab_kset; |
81819f0f CL |
4130 | |
4131 | #define ID_STR_LENGTH 64 | |
4132 | ||
4133 | /* Create a unique string id for a slab cache: | |
4134 | * format | |
4135 | * :[flags-]size:[memory address of kmemcache] | |
4136 | */ | |
4137 | static char *create_unique_id(struct kmem_cache *s) | |
4138 | { | |
4139 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4140 | char *p = name; | |
4141 | ||
4142 | BUG_ON(!name); | |
4143 | ||
4144 | *p++ = ':'; | |
4145 | /* | |
4146 | * First flags affecting slabcache operations. We will only | |
4147 | * get here for aliasable slabs so we do not need to support | |
4148 | * too many flags. The flags here must cover all flags that | |
4149 | * are matched during merging to guarantee that the id is | |
4150 | * unique. | |
4151 | */ | |
4152 | if (s->flags & SLAB_CACHE_DMA) | |
4153 | *p++ = 'd'; | |
4154 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4155 | *p++ = 'a'; | |
4156 | if (s->flags & SLAB_DEBUG_FREE) | |
4157 | *p++ = 'F'; | |
4158 | if (p != name + 1) | |
4159 | *p++ = '-'; | |
4160 | p += sprintf(p, "%07d", s->size); | |
4161 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4162 | return name; | |
4163 | } | |
4164 | ||
4165 | static int sysfs_slab_add(struct kmem_cache *s) | |
4166 | { | |
4167 | int err; | |
4168 | const char *name; | |
4169 | int unmergeable; | |
4170 | ||
4171 | if (slab_state < SYSFS) | |
4172 | /* Defer until later */ | |
4173 | return 0; | |
4174 | ||
4175 | unmergeable = slab_unmergeable(s); | |
4176 | if (unmergeable) { | |
4177 | /* | |
4178 | * Slabcache can never be merged so we can use the name proper. | |
4179 | * This is typically the case for debug situations. In that | |
4180 | * case we can catch duplicate names easily. | |
4181 | */ | |
27c3a314 | 4182 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4183 | name = s->name; |
4184 | } else { | |
4185 | /* | |
4186 | * Create a unique name for the slab as a target | |
4187 | * for the symlinks. | |
4188 | */ | |
4189 | name = create_unique_id(s); | |
4190 | } | |
4191 | ||
27c3a314 | 4192 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4193 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4194 | if (err) { | |
4195 | kobject_put(&s->kobj); | |
81819f0f | 4196 | return err; |
1eada11c | 4197 | } |
81819f0f CL |
4198 | |
4199 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4200 | if (err) | |
4201 | return err; | |
4202 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4203 | if (!unmergeable) { | |
4204 | /* Setup first alias */ | |
4205 | sysfs_slab_alias(s, s->name); | |
4206 | kfree(name); | |
4207 | } | |
4208 | return 0; | |
4209 | } | |
4210 | ||
4211 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4212 | { | |
4213 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4214 | kobject_del(&s->kobj); | |
151c602f | 4215 | kobject_put(&s->kobj); |
81819f0f CL |
4216 | } |
4217 | ||
4218 | /* | |
4219 | * Need to buffer aliases during bootup until sysfs becomes | |
4220 | * available lest we loose that information. | |
4221 | */ | |
4222 | struct saved_alias { | |
4223 | struct kmem_cache *s; | |
4224 | const char *name; | |
4225 | struct saved_alias *next; | |
4226 | }; | |
4227 | ||
5af328a5 | 4228 | static struct saved_alias *alias_list; |
81819f0f CL |
4229 | |
4230 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4231 | { | |
4232 | struct saved_alias *al; | |
4233 | ||
4234 | if (slab_state == SYSFS) { | |
4235 | /* | |
4236 | * If we have a leftover link then remove it. | |
4237 | */ | |
27c3a314 GKH |
4238 | sysfs_remove_link(&slab_kset->kobj, name); |
4239 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4240 | } |
4241 | ||
4242 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4243 | if (!al) | |
4244 | return -ENOMEM; | |
4245 | ||
4246 | al->s = s; | |
4247 | al->name = name; | |
4248 | al->next = alias_list; | |
4249 | alias_list = al; | |
4250 | return 0; | |
4251 | } | |
4252 | ||
4253 | static int __init slab_sysfs_init(void) | |
4254 | { | |
5b95a4ac | 4255 | struct kmem_cache *s; |
81819f0f CL |
4256 | int err; |
4257 | ||
0ff21e46 | 4258 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4259 | if (!slab_kset) { |
81819f0f CL |
4260 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4261 | return -ENOSYS; | |
4262 | } | |
4263 | ||
26a7bd03 CL |
4264 | slab_state = SYSFS; |
4265 | ||
5b95a4ac | 4266 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4267 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4268 | if (err) |
4269 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4270 | " to sysfs\n", s->name); | |
26a7bd03 | 4271 | } |
81819f0f CL |
4272 | |
4273 | while (alias_list) { | |
4274 | struct saved_alias *al = alias_list; | |
4275 | ||
4276 | alias_list = alias_list->next; | |
4277 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4278 | if (err) |
4279 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4280 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4281 | kfree(al); |
4282 | } | |
4283 | ||
4284 | resiliency_test(); | |
4285 | return 0; | |
4286 | } | |
4287 | ||
4288 | __initcall(slab_sysfs_init); | |
81819f0f | 4289 | #endif |
57ed3eda PE |
4290 | |
4291 | /* | |
4292 | * The /proc/slabinfo ABI | |
4293 | */ | |
158a9624 LT |
4294 | #ifdef CONFIG_SLABINFO |
4295 | ||
4296 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |
4297 | size_t count, loff_t *ppos) | |
4298 | { | |
4299 | return -EINVAL; | |
4300 | } | |
4301 | ||
57ed3eda PE |
4302 | |
4303 | static void print_slabinfo_header(struct seq_file *m) | |
4304 | { | |
4305 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4306 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4307 | "<objperslab> <pagesperslab>"); | |
4308 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4309 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4310 | seq_putc(m, '\n'); | |
4311 | } | |
4312 | ||
4313 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4314 | { | |
4315 | loff_t n = *pos; | |
4316 | ||
4317 | down_read(&slub_lock); | |
4318 | if (!n) | |
4319 | print_slabinfo_header(m); | |
4320 | ||
4321 | return seq_list_start(&slab_caches, *pos); | |
4322 | } | |
4323 | ||
4324 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4325 | { | |
4326 | return seq_list_next(p, &slab_caches, pos); | |
4327 | } | |
4328 | ||
4329 | static void s_stop(struct seq_file *m, void *p) | |
4330 | { | |
4331 | up_read(&slub_lock); | |
4332 | } | |
4333 | ||
4334 | static int s_show(struct seq_file *m, void *p) | |
4335 | { | |
4336 | unsigned long nr_partials = 0; | |
4337 | unsigned long nr_slabs = 0; | |
4338 | unsigned long nr_inuse = 0; | |
4339 | unsigned long nr_objs; | |
4340 | struct kmem_cache *s; | |
4341 | int node; | |
4342 | ||
4343 | s = list_entry(p, struct kmem_cache, list); | |
4344 | ||
4345 | for_each_online_node(node) { | |
4346 | struct kmem_cache_node *n = get_node(s, node); | |
4347 | ||
4348 | if (!n) | |
4349 | continue; | |
4350 | ||
4351 | nr_partials += n->nr_partial; | |
4352 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
4353 | nr_inuse += count_partial(n); | |
4354 | } | |
4355 | ||
4356 | nr_objs = nr_slabs * s->objects; | |
4357 | nr_inuse += (nr_slabs - nr_partials) * s->objects; | |
4358 | ||
4359 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
4360 | nr_objs, s->size, s->objects, (1 << s->order)); | |
4361 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); | |
4362 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4363 | 0UL); | |
4364 | seq_putc(m, '\n'); | |
4365 | return 0; | |
4366 | } | |
4367 | ||
4368 | const struct seq_operations slabinfo_op = { | |
4369 | .start = s_start, | |
4370 | .next = s_next, | |
4371 | .stop = s_stop, | |
4372 | .show = s_show, | |
4373 | }; | |
4374 | ||
158a9624 | 4375 | #endif /* CONFIG_SLABINFO */ |