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> | |
3ac7fe5a | 22 | #include <linux/debugobjects.h> |
81819f0f | 23 | #include <linux/kallsyms.h> |
b9049e23 | 24 | #include <linux/memory.h> |
81819f0f CL |
25 | |
26 | /* | |
27 | * Lock order: | |
28 | * 1. slab_lock(page) | |
29 | * 2. slab->list_lock | |
30 | * | |
31 | * The slab_lock protects operations on the object of a particular | |
32 | * slab and its metadata in the page struct. If the slab lock | |
33 | * has been taken then no allocations nor frees can be performed | |
34 | * on the objects in the slab nor can the slab be added or removed | |
35 | * from the partial or full lists since this would mean modifying | |
36 | * the page_struct of the slab. | |
37 | * | |
38 | * The list_lock protects the partial and full list on each node and | |
39 | * the partial slab counter. If taken then no new slabs may be added or | |
40 | * removed from the lists nor make the number of partial slabs be modified. | |
41 | * (Note that the total number of slabs is an atomic value that may be | |
42 | * modified without taking the list lock). | |
43 | * | |
44 | * The list_lock is a centralized lock and thus we avoid taking it as | |
45 | * much as possible. As long as SLUB does not have to handle partial | |
46 | * slabs, operations can continue without any centralized lock. F.e. | |
47 | * allocating a long series of objects that fill up slabs does not require | |
48 | * the list lock. | |
49 | * | |
50 | * The lock order is sometimes inverted when we are trying to get a slab | |
51 | * off a list. We take the list_lock and then look for a page on the list | |
52 | * to use. While we do that objects in the slabs may be freed. We can | |
53 | * only operate on the slab if we have also taken the slab_lock. So we use | |
54 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
55 | * can occur anymore and we can use the slab for allocations etc. If the | |
56 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
57 | * we must stay away from it for a while since we may cause a bouncing | |
58 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
59 | * If all pages are busy then we may allocate a new slab instead of reusing | |
60 | * a partial slab. A new slab has noone operating on it and thus there is | |
61 | * no danger of cacheline contention. | |
62 | * | |
63 | * Interrupts are disabled during allocation and deallocation in order to | |
64 | * make the slab allocator safe to use in the context of an irq. In addition | |
65 | * interrupts are disabled to ensure that the processor does not change | |
66 | * while handling per_cpu slabs, due to kernel preemption. | |
67 | * | |
68 | * SLUB assigns one slab for allocation to each processor. | |
69 | * Allocations only occur from these slabs called cpu slabs. | |
70 | * | |
672bba3a CL |
71 | * Slabs with free elements are kept on a partial list and during regular |
72 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 73 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
74 | * We track full slabs for debugging purposes though because otherwise we |
75 | * cannot scan all objects. | |
81819f0f CL |
76 | * |
77 | * Slabs are freed when they become empty. Teardown and setup is | |
78 | * minimal so we rely on the page allocators per cpu caches for | |
79 | * fast frees and allocs. | |
80 | * | |
81 | * Overloading of page flags that are otherwise used for LRU management. | |
82 | * | |
4b6f0750 CL |
83 | * PageActive The slab is frozen and exempt from list processing. |
84 | * This means that the slab is dedicated to a purpose | |
85 | * such as satisfying allocations for a specific | |
86 | * processor. Objects may be freed in the slab while | |
87 | * it is frozen but slab_free will then skip the usual | |
88 | * list operations. It is up to the processor holding | |
89 | * the slab to integrate the slab into the slab lists | |
90 | * when the slab is no longer needed. | |
91 | * | |
92 | * One use of this flag is to mark slabs that are | |
93 | * used for allocations. Then such a slab becomes a cpu | |
94 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 95 | * freelist that allows lockless access to |
894b8788 CL |
96 | * free objects in addition to the regular freelist |
97 | * that requires the slab lock. | |
81819f0f CL |
98 | * |
99 | * PageError Slab requires special handling due to debug | |
100 | * options set. This moves slab handling out of | |
894b8788 | 101 | * the fast path and disables lockless freelists. |
81819f0f CL |
102 | */ |
103 | ||
5577bd8a CL |
104 | #define FROZEN (1 << PG_active) |
105 | ||
106 | #ifdef CONFIG_SLUB_DEBUG | |
107 | #define SLABDEBUG (1 << PG_error) | |
108 | #else | |
109 | #define SLABDEBUG 0 | |
110 | #endif | |
111 | ||
4b6f0750 CL |
112 | static inline int SlabFrozen(struct page *page) |
113 | { | |
5577bd8a | 114 | return page->flags & FROZEN; |
4b6f0750 CL |
115 | } |
116 | ||
117 | static inline void SetSlabFrozen(struct page *page) | |
118 | { | |
5577bd8a | 119 | page->flags |= FROZEN; |
4b6f0750 CL |
120 | } |
121 | ||
122 | static inline void ClearSlabFrozen(struct page *page) | |
123 | { | |
5577bd8a | 124 | page->flags &= ~FROZEN; |
4b6f0750 CL |
125 | } |
126 | ||
35e5d7ee CL |
127 | static inline int SlabDebug(struct page *page) |
128 | { | |
5577bd8a | 129 | return page->flags & SLABDEBUG; |
35e5d7ee CL |
130 | } |
131 | ||
132 | static inline void SetSlabDebug(struct page *page) | |
133 | { | |
5577bd8a | 134 | page->flags |= SLABDEBUG; |
35e5d7ee CL |
135 | } |
136 | ||
137 | static inline void ClearSlabDebug(struct page *page) | |
138 | { | |
5577bd8a | 139 | page->flags &= ~SLABDEBUG; |
35e5d7ee CL |
140 | } |
141 | ||
81819f0f CL |
142 | /* |
143 | * Issues still to be resolved: | |
144 | * | |
81819f0f CL |
145 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
146 | * | |
81819f0f CL |
147 | * - Variable sizing of the per node arrays |
148 | */ | |
149 | ||
150 | /* Enable to test recovery from slab corruption on boot */ | |
151 | #undef SLUB_RESILIENCY_TEST | |
152 | ||
2086d26a CL |
153 | /* |
154 | * Mininum number of partial slabs. These will be left on the partial | |
155 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
156 | */ | |
76be8950 | 157 | #define MIN_PARTIAL 5 |
e95eed57 | 158 | |
2086d26a CL |
159 | /* |
160 | * Maximum number of desirable partial slabs. | |
161 | * The existence of more partial slabs makes kmem_cache_shrink | |
162 | * sort the partial list by the number of objects in the. | |
163 | */ | |
164 | #define MAX_PARTIAL 10 | |
165 | ||
81819f0f CL |
166 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
167 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 168 | |
81819f0f CL |
169 | /* |
170 | * Set of flags that will prevent slab merging | |
171 | */ | |
172 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
173 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
174 | ||
175 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
176 | SLAB_CACHE_DMA) | |
177 | ||
178 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 179 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
180 | #endif |
181 | ||
182 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 183 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
184 | #endif |
185 | ||
186 | /* Internal SLUB flags */ | |
1ceef402 CL |
187 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
188 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
189 | |
190 | static int kmem_size = sizeof(struct kmem_cache); | |
191 | ||
192 | #ifdef CONFIG_SMP | |
193 | static struct notifier_block slab_notifier; | |
194 | #endif | |
195 | ||
196 | static enum { | |
197 | DOWN, /* No slab functionality available */ | |
198 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 199 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
200 | SYSFS /* Sysfs up */ |
201 | } slab_state = DOWN; | |
202 | ||
203 | /* A list of all slab caches on the system */ | |
204 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 205 | static LIST_HEAD(slab_caches); |
81819f0f | 206 | |
02cbc874 CL |
207 | /* |
208 | * Tracking user of a slab. | |
209 | */ | |
210 | struct track { | |
211 | void *addr; /* Called from address */ | |
212 | int cpu; /* Was running on cpu */ | |
213 | int pid; /* Pid context */ | |
214 | unsigned long when; /* When did the operation occur */ | |
215 | }; | |
216 | ||
217 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
218 | ||
41ecc55b | 219 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
220 | static int sysfs_slab_add(struct kmem_cache *); |
221 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
222 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 223 | |
81819f0f | 224 | #else |
0c710013 CL |
225 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
226 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
227 | { return 0; } | |
151c602f CL |
228 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
229 | { | |
230 | kfree(s); | |
231 | } | |
8ff12cfc | 232 | |
81819f0f CL |
233 | #endif |
234 | ||
8ff12cfc CL |
235 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
236 | { | |
237 | #ifdef CONFIG_SLUB_STATS | |
238 | c->stat[si]++; | |
239 | #endif | |
240 | } | |
241 | ||
81819f0f CL |
242 | /******************************************************************** |
243 | * Core slab cache functions | |
244 | *******************************************************************/ | |
245 | ||
246 | int slab_is_available(void) | |
247 | { | |
248 | return slab_state >= UP; | |
249 | } | |
250 | ||
251 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
252 | { | |
253 | #ifdef CONFIG_NUMA | |
254 | return s->node[node]; | |
255 | #else | |
256 | return &s->local_node; | |
257 | #endif | |
258 | } | |
259 | ||
dfb4f096 CL |
260 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
261 | { | |
4c93c355 CL |
262 | #ifdef CONFIG_SMP |
263 | return s->cpu_slab[cpu]; | |
264 | #else | |
265 | return &s->cpu_slab; | |
266 | #endif | |
dfb4f096 CL |
267 | } |
268 | ||
6446faa2 | 269 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
270 | static inline int check_valid_pointer(struct kmem_cache *s, |
271 | struct page *page, const void *object) | |
272 | { | |
273 | void *base; | |
274 | ||
a973e9dd | 275 | if (!object) |
02cbc874 CL |
276 | return 1; |
277 | ||
a973e9dd | 278 | base = page_address(page); |
39b26464 | 279 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
280 | (object - base) % s->size) { |
281 | return 0; | |
282 | } | |
283 | ||
284 | return 1; | |
285 | } | |
286 | ||
7656c72b CL |
287 | /* |
288 | * Slow version of get and set free pointer. | |
289 | * | |
290 | * This version requires touching the cache lines of kmem_cache which | |
291 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
292 | * from the page struct. | |
293 | */ | |
294 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
295 | { | |
296 | return *(void **)(object + s->offset); | |
297 | } | |
298 | ||
299 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
300 | { | |
301 | *(void **)(object + s->offset) = fp; | |
302 | } | |
303 | ||
304 | /* Loop over all objects in a slab */ | |
224a88be CL |
305 | #define for_each_object(__p, __s, __addr, __objects) \ |
306 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
307 | __p += (__s)->size) |
308 | ||
309 | /* Scan freelist */ | |
310 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 311 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
312 | |
313 | /* Determine object index from a given position */ | |
314 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
315 | { | |
316 | return (p - addr) / s->size; | |
317 | } | |
318 | ||
834f3d11 CL |
319 | static inline struct kmem_cache_order_objects oo_make(int order, |
320 | unsigned long size) | |
321 | { | |
322 | struct kmem_cache_order_objects x = { | |
323 | (order << 16) + (PAGE_SIZE << order) / size | |
324 | }; | |
325 | ||
326 | return x; | |
327 | } | |
328 | ||
329 | static inline int oo_order(struct kmem_cache_order_objects x) | |
330 | { | |
331 | return x.x >> 16; | |
332 | } | |
333 | ||
334 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
335 | { | |
336 | return x.x & ((1 << 16) - 1); | |
337 | } | |
338 | ||
41ecc55b CL |
339 | #ifdef CONFIG_SLUB_DEBUG |
340 | /* | |
341 | * Debug settings: | |
342 | */ | |
f0630fff CL |
343 | #ifdef CONFIG_SLUB_DEBUG_ON |
344 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
345 | #else | |
41ecc55b | 346 | static int slub_debug; |
f0630fff | 347 | #endif |
41ecc55b CL |
348 | |
349 | static char *slub_debug_slabs; | |
350 | ||
81819f0f CL |
351 | /* |
352 | * Object debugging | |
353 | */ | |
354 | static void print_section(char *text, u8 *addr, unsigned int length) | |
355 | { | |
356 | int i, offset; | |
357 | int newline = 1; | |
358 | char ascii[17]; | |
359 | ||
360 | ascii[16] = 0; | |
361 | ||
362 | for (i = 0; i < length; i++) { | |
363 | if (newline) { | |
24922684 | 364 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
365 | newline = 0; |
366 | } | |
06428780 | 367 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
368 | offset = i % 16; |
369 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
370 | if (offset == 15) { | |
06428780 | 371 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
372 | newline = 1; |
373 | } | |
374 | } | |
375 | if (!newline) { | |
376 | i %= 16; | |
377 | while (i < 16) { | |
06428780 | 378 | printk(KERN_CONT " "); |
81819f0f CL |
379 | ascii[i] = ' '; |
380 | i++; | |
381 | } | |
06428780 | 382 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
383 | } |
384 | } | |
385 | ||
81819f0f CL |
386 | static struct track *get_track(struct kmem_cache *s, void *object, |
387 | enum track_item alloc) | |
388 | { | |
389 | struct track *p; | |
390 | ||
391 | if (s->offset) | |
392 | p = object + s->offset + sizeof(void *); | |
393 | else | |
394 | p = object + s->inuse; | |
395 | ||
396 | return p + alloc; | |
397 | } | |
398 | ||
399 | static void set_track(struct kmem_cache *s, void *object, | |
400 | enum track_item alloc, void *addr) | |
401 | { | |
402 | struct track *p; | |
403 | ||
404 | if (s->offset) | |
405 | p = object + s->offset + sizeof(void *); | |
406 | else | |
407 | p = object + s->inuse; | |
408 | ||
409 | p += alloc; | |
410 | if (addr) { | |
411 | p->addr = addr; | |
412 | p->cpu = smp_processor_id(); | |
413 | p->pid = current ? current->pid : -1; | |
414 | p->when = jiffies; | |
415 | } else | |
416 | memset(p, 0, sizeof(struct track)); | |
417 | } | |
418 | ||
81819f0f CL |
419 | static void init_tracking(struct kmem_cache *s, void *object) |
420 | { | |
24922684 CL |
421 | if (!(s->flags & SLAB_STORE_USER)) |
422 | return; | |
423 | ||
424 | set_track(s, object, TRACK_FREE, NULL); | |
425 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
426 | } |
427 | ||
428 | static void print_track(const char *s, struct track *t) | |
429 | { | |
430 | if (!t->addr) | |
431 | return; | |
432 | ||
24922684 | 433 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 434 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
435 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
436 | } | |
437 | ||
438 | static void print_tracking(struct kmem_cache *s, void *object) | |
439 | { | |
440 | if (!(s->flags & SLAB_STORE_USER)) | |
441 | return; | |
442 | ||
443 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
444 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
445 | } | |
446 | ||
447 | static void print_page_info(struct page *page) | |
448 | { | |
39b26464 CL |
449 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
450 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
451 | |
452 | } | |
453 | ||
454 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
455 | { | |
456 | va_list args; | |
457 | char buf[100]; | |
458 | ||
459 | va_start(args, fmt); | |
460 | vsnprintf(buf, sizeof(buf), fmt, args); | |
461 | va_end(args); | |
462 | printk(KERN_ERR "========================================" | |
463 | "=====================================\n"); | |
464 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
465 | printk(KERN_ERR "----------------------------------------" | |
466 | "-------------------------------------\n\n"); | |
81819f0f CL |
467 | } |
468 | ||
24922684 CL |
469 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
470 | { | |
471 | va_list args; | |
472 | char buf[100]; | |
473 | ||
474 | va_start(args, fmt); | |
475 | vsnprintf(buf, sizeof(buf), fmt, args); | |
476 | va_end(args); | |
477 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
478 | } | |
479 | ||
480 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
481 | { |
482 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 483 | u8 *addr = page_address(page); |
24922684 CL |
484 | |
485 | print_tracking(s, p); | |
486 | ||
487 | print_page_info(page); | |
488 | ||
489 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
490 | p, p - addr, get_freepointer(s, p)); | |
491 | ||
492 | if (p > addr + 16) | |
493 | print_section("Bytes b4", p - 16, 16); | |
494 | ||
495 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
496 | |
497 | if (s->flags & SLAB_RED_ZONE) | |
498 | print_section("Redzone", p + s->objsize, | |
499 | s->inuse - s->objsize); | |
500 | ||
81819f0f CL |
501 | if (s->offset) |
502 | off = s->offset + sizeof(void *); | |
503 | else | |
504 | off = s->inuse; | |
505 | ||
24922684 | 506 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 507 | off += 2 * sizeof(struct track); |
81819f0f CL |
508 | |
509 | if (off != s->size) | |
510 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
511 | print_section("Padding", p + off, s->size - off); |
512 | ||
513 | dump_stack(); | |
81819f0f CL |
514 | } |
515 | ||
516 | static void object_err(struct kmem_cache *s, struct page *page, | |
517 | u8 *object, char *reason) | |
518 | { | |
3dc50637 | 519 | slab_bug(s, "%s", reason); |
24922684 | 520 | print_trailer(s, page, object); |
81819f0f CL |
521 | } |
522 | ||
24922684 | 523 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
524 | { |
525 | va_list args; | |
526 | char buf[100]; | |
527 | ||
24922684 CL |
528 | va_start(args, fmt); |
529 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 530 | va_end(args); |
3dc50637 | 531 | slab_bug(s, "%s", buf); |
24922684 | 532 | print_page_info(page); |
81819f0f CL |
533 | dump_stack(); |
534 | } | |
535 | ||
536 | static void init_object(struct kmem_cache *s, void *object, int active) | |
537 | { | |
538 | u8 *p = object; | |
539 | ||
540 | if (s->flags & __OBJECT_POISON) { | |
541 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 542 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
543 | } |
544 | ||
545 | if (s->flags & SLAB_RED_ZONE) | |
546 | memset(p + s->objsize, | |
547 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
548 | s->inuse - s->objsize); | |
549 | } | |
550 | ||
24922684 | 551 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
552 | { |
553 | while (bytes) { | |
554 | if (*start != (u8)value) | |
24922684 | 555 | return start; |
81819f0f CL |
556 | start++; |
557 | bytes--; | |
558 | } | |
24922684 CL |
559 | return NULL; |
560 | } | |
561 | ||
562 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
563 | void *from, void *to) | |
564 | { | |
565 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
566 | memset(from, data, to - from); | |
567 | } | |
568 | ||
569 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
570 | u8 *object, char *what, | |
06428780 | 571 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
572 | { |
573 | u8 *fault; | |
574 | u8 *end; | |
575 | ||
576 | fault = check_bytes(start, value, bytes); | |
577 | if (!fault) | |
578 | return 1; | |
579 | ||
580 | end = start + bytes; | |
581 | while (end > fault && end[-1] == value) | |
582 | end--; | |
583 | ||
584 | slab_bug(s, "%s overwritten", what); | |
585 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
586 | fault, end - 1, fault[0], value); | |
587 | print_trailer(s, page, object); | |
588 | ||
589 | restore_bytes(s, what, value, fault, end); | |
590 | return 0; | |
81819f0f CL |
591 | } |
592 | ||
81819f0f CL |
593 | /* |
594 | * Object layout: | |
595 | * | |
596 | * object address | |
597 | * Bytes of the object to be managed. | |
598 | * If the freepointer may overlay the object then the free | |
599 | * pointer is the first word of the object. | |
672bba3a | 600 | * |
81819f0f CL |
601 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
602 | * 0xa5 (POISON_END) | |
603 | * | |
604 | * object + s->objsize | |
605 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
606 | * Padding is extended by another word if Redzoning is enabled and |
607 | * objsize == inuse. | |
608 | * | |
81819f0f CL |
609 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
610 | * 0xcc (RED_ACTIVE) for objects in use. | |
611 | * | |
612 | * object + s->inuse | |
672bba3a CL |
613 | * Meta data starts here. |
614 | * | |
81819f0f CL |
615 | * A. Free pointer (if we cannot overwrite object on free) |
616 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 617 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 618 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
619 | * before the word boundary. |
620 | * | |
621 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
622 | * |
623 | * object + s->size | |
672bba3a | 624 | * Nothing is used beyond s->size. |
81819f0f | 625 | * |
672bba3a CL |
626 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
627 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
628 | * may be used with merged slabcaches. |
629 | */ | |
630 | ||
81819f0f CL |
631 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
632 | { | |
633 | unsigned long off = s->inuse; /* The end of info */ | |
634 | ||
635 | if (s->offset) | |
636 | /* Freepointer is placed after the object. */ | |
637 | off += sizeof(void *); | |
638 | ||
639 | if (s->flags & SLAB_STORE_USER) | |
640 | /* We also have user information there */ | |
641 | off += 2 * sizeof(struct track); | |
642 | ||
643 | if (s->size == off) | |
644 | return 1; | |
645 | ||
24922684 CL |
646 | return check_bytes_and_report(s, page, p, "Object padding", |
647 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
648 | } |
649 | ||
39b26464 | 650 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
651 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
652 | { | |
24922684 CL |
653 | u8 *start; |
654 | u8 *fault; | |
655 | u8 *end; | |
656 | int length; | |
657 | int remainder; | |
81819f0f CL |
658 | |
659 | if (!(s->flags & SLAB_POISON)) | |
660 | return 1; | |
661 | ||
a973e9dd | 662 | start = page_address(page); |
834f3d11 | 663 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
664 | end = start + length; |
665 | remainder = length % s->size; | |
81819f0f CL |
666 | if (!remainder) |
667 | return 1; | |
668 | ||
39b26464 | 669 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
670 | if (!fault) |
671 | return 1; | |
672 | while (end > fault && end[-1] == POISON_INUSE) | |
673 | end--; | |
674 | ||
675 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 676 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
677 | |
678 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
679 | return 0; | |
81819f0f CL |
680 | } |
681 | ||
682 | static int check_object(struct kmem_cache *s, struct page *page, | |
683 | void *object, int active) | |
684 | { | |
685 | u8 *p = object; | |
686 | u8 *endobject = object + s->objsize; | |
687 | ||
688 | if (s->flags & SLAB_RED_ZONE) { | |
689 | unsigned int red = | |
690 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
691 | ||
24922684 CL |
692 | if (!check_bytes_and_report(s, page, object, "Redzone", |
693 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 694 | return 0; |
81819f0f | 695 | } else { |
3adbefee IM |
696 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
697 | check_bytes_and_report(s, page, p, "Alignment padding", | |
698 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
699 | } | |
81819f0f CL |
700 | } |
701 | ||
702 | if (s->flags & SLAB_POISON) { | |
703 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
704 | (!check_bytes_and_report(s, page, p, "Poison", p, |
705 | POISON_FREE, s->objsize - 1) || | |
706 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 707 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 708 | return 0; |
81819f0f CL |
709 | /* |
710 | * check_pad_bytes cleans up on its own. | |
711 | */ | |
712 | check_pad_bytes(s, page, p); | |
713 | } | |
714 | ||
715 | if (!s->offset && active) | |
716 | /* | |
717 | * Object and freepointer overlap. Cannot check | |
718 | * freepointer while object is allocated. | |
719 | */ | |
720 | return 1; | |
721 | ||
722 | /* Check free pointer validity */ | |
723 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
724 | object_err(s, page, p, "Freepointer corrupt"); | |
725 | /* | |
726 | * No choice but to zap it and thus loose the remainder | |
727 | * of the free objects in this slab. May cause | |
672bba3a | 728 | * another error because the object count is now wrong. |
81819f0f | 729 | */ |
a973e9dd | 730 | set_freepointer(s, p, NULL); |
81819f0f CL |
731 | return 0; |
732 | } | |
733 | return 1; | |
734 | } | |
735 | ||
736 | static int check_slab(struct kmem_cache *s, struct page *page) | |
737 | { | |
39b26464 CL |
738 | int maxobj; |
739 | ||
81819f0f CL |
740 | VM_BUG_ON(!irqs_disabled()); |
741 | ||
742 | if (!PageSlab(page)) { | |
24922684 | 743 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
744 | return 0; |
745 | } | |
39b26464 CL |
746 | |
747 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
748 | if (page->objects > maxobj) { | |
749 | slab_err(s, page, "objects %u > max %u", | |
750 | s->name, page->objects, maxobj); | |
751 | return 0; | |
752 | } | |
753 | if (page->inuse > page->objects) { | |
24922684 | 754 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 755 | s->name, page->inuse, page->objects); |
81819f0f CL |
756 | return 0; |
757 | } | |
758 | /* Slab_pad_check fixes things up after itself */ | |
759 | slab_pad_check(s, page); | |
760 | return 1; | |
761 | } | |
762 | ||
763 | /* | |
672bba3a CL |
764 | * Determine if a certain object on a page is on the freelist. Must hold the |
765 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
766 | */ |
767 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
768 | { | |
769 | int nr = 0; | |
770 | void *fp = page->freelist; | |
771 | void *object = NULL; | |
224a88be | 772 | unsigned long max_objects; |
81819f0f | 773 | |
39b26464 | 774 | while (fp && nr <= page->objects) { |
81819f0f CL |
775 | if (fp == search) |
776 | return 1; | |
777 | if (!check_valid_pointer(s, page, fp)) { | |
778 | if (object) { | |
779 | object_err(s, page, object, | |
780 | "Freechain corrupt"); | |
a973e9dd | 781 | set_freepointer(s, object, NULL); |
81819f0f CL |
782 | break; |
783 | } else { | |
24922684 | 784 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 785 | page->freelist = NULL; |
39b26464 | 786 | page->inuse = page->objects; |
24922684 | 787 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
788 | return 0; |
789 | } | |
790 | break; | |
791 | } | |
792 | object = fp; | |
793 | fp = get_freepointer(s, object); | |
794 | nr++; | |
795 | } | |
796 | ||
224a88be CL |
797 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
798 | if (max_objects > 65535) | |
799 | max_objects = 65535; | |
800 | ||
801 | if (page->objects != max_objects) { | |
802 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
803 | "should be %d", page->objects, max_objects); | |
804 | page->objects = max_objects; | |
805 | slab_fix(s, "Number of objects adjusted."); | |
806 | } | |
39b26464 | 807 | if (page->inuse != page->objects - nr) { |
70d71228 | 808 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
809 | "counted were %d", page->inuse, page->objects - nr); |
810 | page->inuse = page->objects - nr; | |
24922684 | 811 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
812 | } |
813 | return search == NULL; | |
814 | } | |
815 | ||
3ec09742 CL |
816 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
817 | { | |
818 | if (s->flags & SLAB_TRACE) { | |
819 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
820 | s->name, | |
821 | alloc ? "alloc" : "free", | |
822 | object, page->inuse, | |
823 | page->freelist); | |
824 | ||
825 | if (!alloc) | |
826 | print_section("Object", (void *)object, s->objsize); | |
827 | ||
828 | dump_stack(); | |
829 | } | |
830 | } | |
831 | ||
643b1138 | 832 | /* |
672bba3a | 833 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 834 | */ |
e95eed57 | 835 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 836 | { |
643b1138 CL |
837 | spin_lock(&n->list_lock); |
838 | list_add(&page->lru, &n->full); | |
839 | spin_unlock(&n->list_lock); | |
840 | } | |
841 | ||
842 | static void remove_full(struct kmem_cache *s, struct page *page) | |
843 | { | |
844 | struct kmem_cache_node *n; | |
845 | ||
846 | if (!(s->flags & SLAB_STORE_USER)) | |
847 | return; | |
848 | ||
849 | n = get_node(s, page_to_nid(page)); | |
850 | ||
851 | spin_lock(&n->list_lock); | |
852 | list_del(&page->lru); | |
853 | spin_unlock(&n->list_lock); | |
854 | } | |
855 | ||
0f389ec6 CL |
856 | /* Tracking of the number of slabs for debugging purposes */ |
857 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
858 | { | |
859 | struct kmem_cache_node *n = get_node(s, node); | |
860 | ||
861 | return atomic_long_read(&n->nr_slabs); | |
862 | } | |
863 | ||
205ab99d | 864 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
865 | { |
866 | struct kmem_cache_node *n = get_node(s, node); | |
867 | ||
868 | /* | |
869 | * May be called early in order to allocate a slab for the | |
870 | * kmem_cache_node structure. Solve the chicken-egg | |
871 | * dilemma by deferring the increment of the count during | |
872 | * bootstrap (see early_kmem_cache_node_alloc). | |
873 | */ | |
205ab99d | 874 | if (!NUMA_BUILD || n) { |
0f389ec6 | 875 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
876 | atomic_long_add(objects, &n->total_objects); |
877 | } | |
0f389ec6 | 878 | } |
205ab99d | 879 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
880 | { |
881 | struct kmem_cache_node *n = get_node(s, node); | |
882 | ||
883 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 884 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
885 | } |
886 | ||
887 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
888 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
889 | void *object) | |
890 | { | |
891 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
892 | return; | |
893 | ||
894 | init_object(s, object, 0); | |
895 | init_tracking(s, object); | |
896 | } | |
897 | ||
898 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
899 | void *object, void *addr) | |
81819f0f CL |
900 | { |
901 | if (!check_slab(s, page)) | |
902 | goto bad; | |
903 | ||
d692ef6d | 904 | if (!on_freelist(s, page, object)) { |
24922684 | 905 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 906 | goto bad; |
81819f0f CL |
907 | } |
908 | ||
909 | if (!check_valid_pointer(s, page, object)) { | |
910 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 911 | goto bad; |
81819f0f CL |
912 | } |
913 | ||
d692ef6d | 914 | if (!check_object(s, page, object, 0)) |
81819f0f | 915 | goto bad; |
81819f0f | 916 | |
3ec09742 CL |
917 | /* Success perform special debug activities for allocs */ |
918 | if (s->flags & SLAB_STORE_USER) | |
919 | set_track(s, object, TRACK_ALLOC, addr); | |
920 | trace(s, page, object, 1); | |
921 | init_object(s, object, 1); | |
81819f0f | 922 | return 1; |
3ec09742 | 923 | |
81819f0f CL |
924 | bad: |
925 | if (PageSlab(page)) { | |
926 | /* | |
927 | * If this is a slab page then lets do the best we can | |
928 | * to avoid issues in the future. Marking all objects | |
672bba3a | 929 | * as used avoids touching the remaining objects. |
81819f0f | 930 | */ |
24922684 | 931 | slab_fix(s, "Marking all objects used"); |
39b26464 | 932 | page->inuse = page->objects; |
a973e9dd | 933 | page->freelist = NULL; |
81819f0f CL |
934 | } |
935 | return 0; | |
936 | } | |
937 | ||
3ec09742 CL |
938 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
939 | void *object, void *addr) | |
81819f0f CL |
940 | { |
941 | if (!check_slab(s, page)) | |
942 | goto fail; | |
943 | ||
944 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 945 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
946 | goto fail; |
947 | } | |
948 | ||
949 | if (on_freelist(s, page, object)) { | |
24922684 | 950 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
951 | goto fail; |
952 | } | |
953 | ||
954 | if (!check_object(s, page, object, 1)) | |
955 | return 0; | |
956 | ||
957 | if (unlikely(s != page->slab)) { | |
3adbefee | 958 | if (!PageSlab(page)) { |
70d71228 CL |
959 | slab_err(s, page, "Attempt to free object(0x%p) " |
960 | "outside of slab", object); | |
3adbefee | 961 | } else if (!page->slab) { |
81819f0f | 962 | printk(KERN_ERR |
70d71228 | 963 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 964 | object); |
70d71228 | 965 | dump_stack(); |
06428780 | 966 | } else |
24922684 CL |
967 | object_err(s, page, object, |
968 | "page slab pointer corrupt."); | |
81819f0f CL |
969 | goto fail; |
970 | } | |
3ec09742 CL |
971 | |
972 | /* Special debug activities for freeing objects */ | |
a973e9dd | 973 | if (!SlabFrozen(page) && !page->freelist) |
3ec09742 CL |
974 | remove_full(s, page); |
975 | if (s->flags & SLAB_STORE_USER) | |
976 | set_track(s, object, TRACK_FREE, addr); | |
977 | trace(s, page, object, 0); | |
978 | init_object(s, object, 0); | |
81819f0f | 979 | return 1; |
3ec09742 | 980 | |
81819f0f | 981 | fail: |
24922684 | 982 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
983 | return 0; |
984 | } | |
985 | ||
41ecc55b CL |
986 | static int __init setup_slub_debug(char *str) |
987 | { | |
f0630fff CL |
988 | slub_debug = DEBUG_DEFAULT_FLAGS; |
989 | if (*str++ != '=' || !*str) | |
990 | /* | |
991 | * No options specified. Switch on full debugging. | |
992 | */ | |
993 | goto out; | |
994 | ||
995 | if (*str == ',') | |
996 | /* | |
997 | * No options but restriction on slabs. This means full | |
998 | * debugging for slabs matching a pattern. | |
999 | */ | |
1000 | goto check_slabs; | |
1001 | ||
1002 | slub_debug = 0; | |
1003 | if (*str == '-') | |
1004 | /* | |
1005 | * Switch off all debugging measures. | |
1006 | */ | |
1007 | goto out; | |
1008 | ||
1009 | /* | |
1010 | * Determine which debug features should be switched on | |
1011 | */ | |
06428780 | 1012 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1013 | switch (tolower(*str)) { |
1014 | case 'f': | |
1015 | slub_debug |= SLAB_DEBUG_FREE; | |
1016 | break; | |
1017 | case 'z': | |
1018 | slub_debug |= SLAB_RED_ZONE; | |
1019 | break; | |
1020 | case 'p': | |
1021 | slub_debug |= SLAB_POISON; | |
1022 | break; | |
1023 | case 'u': | |
1024 | slub_debug |= SLAB_STORE_USER; | |
1025 | break; | |
1026 | case 't': | |
1027 | slub_debug |= SLAB_TRACE; | |
1028 | break; | |
1029 | default: | |
1030 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1031 | "unknown. skipped\n", *str); |
f0630fff | 1032 | } |
41ecc55b CL |
1033 | } |
1034 | ||
f0630fff | 1035 | check_slabs: |
41ecc55b CL |
1036 | if (*str == ',') |
1037 | slub_debug_slabs = str + 1; | |
f0630fff | 1038 | out: |
41ecc55b CL |
1039 | return 1; |
1040 | } | |
1041 | ||
1042 | __setup("slub_debug", setup_slub_debug); | |
1043 | ||
ba0268a8 CL |
1044 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1045 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1046 | void (*ctor)(struct kmem_cache *, void *)) |
41ecc55b CL |
1047 | { |
1048 | /* | |
e153362a | 1049 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1050 | */ |
e153362a CL |
1051 | if (slub_debug && (!slub_debug_slabs || |
1052 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1053 | flags |= slub_debug; | |
ba0268a8 CL |
1054 | |
1055 | return flags; | |
41ecc55b CL |
1056 | } |
1057 | #else | |
3ec09742 CL |
1058 | static inline void setup_object_debug(struct kmem_cache *s, |
1059 | struct page *page, void *object) {} | |
41ecc55b | 1060 | |
3ec09742 CL |
1061 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1062 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1063 | |
3ec09742 CL |
1064 | static inline int free_debug_processing(struct kmem_cache *s, |
1065 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1066 | |
41ecc55b CL |
1067 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1068 | { return 1; } | |
1069 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1070 | void *object, int active) { return 1; } | |
3ec09742 | 1071 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1072 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1073 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1074 | void (*ctor)(struct kmem_cache *, void *)) |
ba0268a8 CL |
1075 | { |
1076 | return flags; | |
1077 | } | |
41ecc55b | 1078 | #define slub_debug 0 |
0f389ec6 CL |
1079 | |
1080 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1081 | { return 0; } | |
205ab99d CL |
1082 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1083 | int objects) {} | |
1084 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1085 | int objects) {} | |
41ecc55b | 1086 | #endif |
205ab99d | 1087 | |
81819f0f CL |
1088 | /* |
1089 | * Slab allocation and freeing | |
1090 | */ | |
65c3376a CL |
1091 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1092 | struct kmem_cache_order_objects oo) | |
1093 | { | |
1094 | int order = oo_order(oo); | |
1095 | ||
1096 | if (node == -1) | |
1097 | return alloc_pages(flags, order); | |
1098 | else | |
1099 | return alloc_pages_node(node, flags, order); | |
1100 | } | |
1101 | ||
81819f0f CL |
1102 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1103 | { | |
06428780 | 1104 | struct page *page; |
834f3d11 | 1105 | struct kmem_cache_order_objects oo = s->oo; |
81819f0f | 1106 | |
b7a49f0d | 1107 | flags |= s->allocflags; |
e12ba74d | 1108 | |
65c3376a CL |
1109 | page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node, |
1110 | oo); | |
1111 | if (unlikely(!page)) { | |
1112 | oo = s->min; | |
1113 | /* | |
1114 | * Allocation may have failed due to fragmentation. | |
1115 | * Try a lower order alloc if possible | |
1116 | */ | |
1117 | page = alloc_slab_page(flags, node, oo); | |
1118 | if (!page) | |
1119 | return NULL; | |
81819f0f | 1120 | |
65c3376a CL |
1121 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1122 | } | |
834f3d11 | 1123 | page->objects = oo_objects(oo); |
81819f0f CL |
1124 | mod_zone_page_state(page_zone(page), |
1125 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1126 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1127 | 1 << oo_order(oo)); |
81819f0f CL |
1128 | |
1129 | return page; | |
1130 | } | |
1131 | ||
1132 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1133 | void *object) | |
1134 | { | |
3ec09742 | 1135 | setup_object_debug(s, page, object); |
4f104934 | 1136 | if (unlikely(s->ctor)) |
4ba9b9d0 | 1137 | s->ctor(s, object); |
81819f0f CL |
1138 | } |
1139 | ||
1140 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1141 | { | |
1142 | struct page *page; | |
81819f0f | 1143 | void *start; |
81819f0f CL |
1144 | void *last; |
1145 | void *p; | |
1146 | ||
6cb06229 | 1147 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1148 | |
6cb06229 CL |
1149 | page = allocate_slab(s, |
1150 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1151 | if (!page) |
1152 | goto out; | |
1153 | ||
205ab99d | 1154 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1155 | page->slab = s; |
1156 | page->flags |= 1 << PG_slab; | |
1157 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1158 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1159 | SetSlabDebug(page); |
81819f0f CL |
1160 | |
1161 | start = page_address(page); | |
81819f0f CL |
1162 | |
1163 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1164 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1165 | |
1166 | last = start; | |
224a88be | 1167 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1168 | setup_object(s, page, last); |
1169 | set_freepointer(s, last, p); | |
1170 | last = p; | |
1171 | } | |
1172 | setup_object(s, page, last); | |
a973e9dd | 1173 | set_freepointer(s, last, NULL); |
81819f0f CL |
1174 | |
1175 | page->freelist = start; | |
1176 | page->inuse = 0; | |
1177 | out: | |
81819f0f CL |
1178 | return page; |
1179 | } | |
1180 | ||
1181 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1182 | { | |
834f3d11 CL |
1183 | int order = compound_order(page); |
1184 | int pages = 1 << order; | |
81819f0f | 1185 | |
c59def9f | 1186 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1187 | void *p; |
1188 | ||
1189 | slab_pad_check(s, page); | |
224a88be CL |
1190 | for_each_object(p, s, page_address(page), |
1191 | page->objects) | |
81819f0f | 1192 | check_object(s, page, p, 0); |
2208b764 | 1193 | ClearSlabDebug(page); |
81819f0f CL |
1194 | } |
1195 | ||
1196 | mod_zone_page_state(page_zone(page), | |
1197 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1198 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1199 | -pages); |
81819f0f | 1200 | |
49bd5221 CL |
1201 | __ClearPageSlab(page); |
1202 | reset_page_mapcount(page); | |
834f3d11 | 1203 | __free_pages(page, order); |
81819f0f CL |
1204 | } |
1205 | ||
1206 | static void rcu_free_slab(struct rcu_head *h) | |
1207 | { | |
1208 | struct page *page; | |
1209 | ||
1210 | page = container_of((struct list_head *)h, struct page, lru); | |
1211 | __free_slab(page->slab, page); | |
1212 | } | |
1213 | ||
1214 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1215 | { | |
1216 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1217 | /* | |
1218 | * RCU free overloads the RCU head over the LRU | |
1219 | */ | |
1220 | struct rcu_head *head = (void *)&page->lru; | |
1221 | ||
1222 | call_rcu(head, rcu_free_slab); | |
1223 | } else | |
1224 | __free_slab(s, page); | |
1225 | } | |
1226 | ||
1227 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1228 | { | |
205ab99d | 1229 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1230 | free_slab(s, page); |
1231 | } | |
1232 | ||
1233 | /* | |
1234 | * Per slab locking using the pagelock | |
1235 | */ | |
1236 | static __always_inline void slab_lock(struct page *page) | |
1237 | { | |
1238 | bit_spin_lock(PG_locked, &page->flags); | |
1239 | } | |
1240 | ||
1241 | static __always_inline void slab_unlock(struct page *page) | |
1242 | { | |
a76d3546 | 1243 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1244 | } |
1245 | ||
1246 | static __always_inline int slab_trylock(struct page *page) | |
1247 | { | |
1248 | int rc = 1; | |
1249 | ||
1250 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1251 | return rc; | |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * Management of partially allocated slabs | |
1256 | */ | |
7c2e132c CL |
1257 | static void add_partial(struct kmem_cache_node *n, |
1258 | struct page *page, int tail) | |
81819f0f | 1259 | { |
e95eed57 CL |
1260 | spin_lock(&n->list_lock); |
1261 | n->nr_partial++; | |
7c2e132c CL |
1262 | if (tail) |
1263 | list_add_tail(&page->lru, &n->partial); | |
1264 | else | |
1265 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1266 | spin_unlock(&n->list_lock); |
1267 | } | |
1268 | ||
1269 | static void remove_partial(struct kmem_cache *s, | |
1270 | struct page *page) | |
1271 | { | |
1272 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1273 | ||
1274 | spin_lock(&n->list_lock); | |
1275 | list_del(&page->lru); | |
1276 | n->nr_partial--; | |
1277 | spin_unlock(&n->list_lock); | |
1278 | } | |
1279 | ||
1280 | /* | |
672bba3a | 1281 | * Lock slab and remove from the partial list. |
81819f0f | 1282 | * |
672bba3a | 1283 | * Must hold list_lock. |
81819f0f | 1284 | */ |
4b6f0750 | 1285 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1286 | { |
1287 | if (slab_trylock(page)) { | |
1288 | list_del(&page->lru); | |
1289 | n->nr_partial--; | |
4b6f0750 | 1290 | SetSlabFrozen(page); |
81819f0f CL |
1291 | return 1; |
1292 | } | |
1293 | return 0; | |
1294 | } | |
1295 | ||
1296 | /* | |
672bba3a | 1297 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1298 | */ |
1299 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1300 | { | |
1301 | struct page *page; | |
1302 | ||
1303 | /* | |
1304 | * Racy check. If we mistakenly see no partial slabs then we | |
1305 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1306 | * partial slab and there is none available then get_partials() |
1307 | * will return NULL. | |
81819f0f CL |
1308 | */ |
1309 | if (!n || !n->nr_partial) | |
1310 | return NULL; | |
1311 | ||
1312 | spin_lock(&n->list_lock); | |
1313 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1314 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1315 | goto out; |
1316 | page = NULL; | |
1317 | out: | |
1318 | spin_unlock(&n->list_lock); | |
1319 | return page; | |
1320 | } | |
1321 | ||
1322 | /* | |
672bba3a | 1323 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1324 | */ |
1325 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1326 | { | |
1327 | #ifdef CONFIG_NUMA | |
1328 | struct zonelist *zonelist; | |
dd1a239f | 1329 | struct zoneref *z; |
54a6eb5c MG |
1330 | struct zone *zone; |
1331 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1332 | struct page *page; |
1333 | ||
1334 | /* | |
672bba3a CL |
1335 | * The defrag ratio allows a configuration of the tradeoffs between |
1336 | * inter node defragmentation and node local allocations. A lower | |
1337 | * defrag_ratio increases the tendency to do local allocations | |
1338 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1339 | * |
672bba3a CL |
1340 | * If the defrag_ratio is set to 0 then kmalloc() always |
1341 | * returns node local objects. If the ratio is higher then kmalloc() | |
1342 | * may return off node objects because partial slabs are obtained | |
1343 | * from other nodes and filled up. | |
81819f0f | 1344 | * |
6446faa2 | 1345 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1346 | * defrag_ratio = 1000) then every (well almost) allocation will |
1347 | * first attempt to defrag slab caches on other nodes. This means | |
1348 | * scanning over all nodes to look for partial slabs which may be | |
1349 | * expensive if we do it every time we are trying to find a slab | |
1350 | * with available objects. | |
81819f0f | 1351 | */ |
9824601e CL |
1352 | if (!s->remote_node_defrag_ratio || |
1353 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1354 | return NULL; |
1355 | ||
0e88460d | 1356 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1357 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1358 | struct kmem_cache_node *n; |
1359 | ||
54a6eb5c | 1360 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1361 | |
54a6eb5c | 1362 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
e95eed57 | 1363 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1364 | page = get_partial_node(n); |
1365 | if (page) | |
1366 | return page; | |
1367 | } | |
1368 | } | |
1369 | #endif | |
1370 | return NULL; | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * Get a partial page, lock it and return it. | |
1375 | */ | |
1376 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1377 | { | |
1378 | struct page *page; | |
1379 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1380 | ||
1381 | page = get_partial_node(get_node(s, searchnode)); | |
1382 | if (page || (flags & __GFP_THISNODE)) | |
1383 | return page; | |
1384 | ||
1385 | return get_any_partial(s, flags); | |
1386 | } | |
1387 | ||
1388 | /* | |
1389 | * Move a page back to the lists. | |
1390 | * | |
1391 | * Must be called with the slab lock held. | |
1392 | * | |
1393 | * On exit the slab lock will have been dropped. | |
1394 | */ | |
7c2e132c | 1395 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1396 | { |
e95eed57 | 1397 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1398 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1399 | |
4b6f0750 | 1400 | ClearSlabFrozen(page); |
81819f0f | 1401 | if (page->inuse) { |
e95eed57 | 1402 | |
a973e9dd | 1403 | if (page->freelist) { |
7c2e132c | 1404 | add_partial(n, page, tail); |
8ff12cfc CL |
1405 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1406 | } else { | |
1407 | stat(c, DEACTIVATE_FULL); | |
1408 | if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) | |
1409 | add_full(n, page); | |
1410 | } | |
81819f0f CL |
1411 | slab_unlock(page); |
1412 | } else { | |
8ff12cfc | 1413 | stat(c, DEACTIVATE_EMPTY); |
e95eed57 CL |
1414 | if (n->nr_partial < MIN_PARTIAL) { |
1415 | /* | |
672bba3a CL |
1416 | * Adding an empty slab to the partial slabs in order |
1417 | * to avoid page allocator overhead. This slab needs | |
1418 | * to come after the other slabs with objects in | |
6446faa2 CL |
1419 | * so that the others get filled first. That way the |
1420 | * size of the partial list stays small. | |
1421 | * | |
1422 | * kmem_cache_shrink can reclaim any empty slabs from the | |
1423 | * partial list. | |
e95eed57 | 1424 | */ |
7c2e132c | 1425 | add_partial(n, page, 1); |
e95eed57 CL |
1426 | slab_unlock(page); |
1427 | } else { | |
1428 | slab_unlock(page); | |
8ff12cfc | 1429 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1430 | discard_slab(s, page); |
1431 | } | |
81819f0f CL |
1432 | } |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Remove the cpu slab | |
1437 | */ | |
dfb4f096 | 1438 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1439 | { |
dfb4f096 | 1440 | struct page *page = c->page; |
7c2e132c | 1441 | int tail = 1; |
8ff12cfc | 1442 | |
b773ad73 | 1443 | if (page->freelist) |
8ff12cfc | 1444 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1445 | /* |
6446faa2 | 1446 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1447 | * because both freelists are empty. So this is unlikely |
1448 | * to occur. | |
1449 | */ | |
a973e9dd | 1450 | while (unlikely(c->freelist)) { |
894b8788 CL |
1451 | void **object; |
1452 | ||
7c2e132c CL |
1453 | tail = 0; /* Hot objects. Put the slab first */ |
1454 | ||
894b8788 | 1455 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1456 | object = c->freelist; |
b3fba8da | 1457 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1458 | |
1459 | /* And put onto the regular freelist */ | |
b3fba8da | 1460 | object[c->offset] = page->freelist; |
894b8788 CL |
1461 | page->freelist = object; |
1462 | page->inuse--; | |
1463 | } | |
dfb4f096 | 1464 | c->page = NULL; |
7c2e132c | 1465 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1466 | } |
1467 | ||
dfb4f096 | 1468 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1469 | { |
8ff12cfc | 1470 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1471 | slab_lock(c->page); |
1472 | deactivate_slab(s, c); | |
81819f0f CL |
1473 | } |
1474 | ||
1475 | /* | |
1476 | * Flush cpu slab. | |
6446faa2 | 1477 | * |
81819f0f CL |
1478 | * Called from IPI handler with interrupts disabled. |
1479 | */ | |
0c710013 | 1480 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1481 | { |
dfb4f096 | 1482 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1483 | |
dfb4f096 CL |
1484 | if (likely(c && c->page)) |
1485 | flush_slab(s, c); | |
81819f0f CL |
1486 | } |
1487 | ||
1488 | static void flush_cpu_slab(void *d) | |
1489 | { | |
1490 | struct kmem_cache *s = d; | |
81819f0f | 1491 | |
dfb4f096 | 1492 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1493 | } |
1494 | ||
1495 | static void flush_all(struct kmem_cache *s) | |
1496 | { | |
1497 | #ifdef CONFIG_SMP | |
1498 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1499 | #else | |
1500 | unsigned long flags; | |
1501 | ||
1502 | local_irq_save(flags); | |
1503 | flush_cpu_slab(s); | |
1504 | local_irq_restore(flags); | |
1505 | #endif | |
1506 | } | |
1507 | ||
dfb4f096 CL |
1508 | /* |
1509 | * Check if the objects in a per cpu structure fit numa | |
1510 | * locality expectations. | |
1511 | */ | |
1512 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1513 | { | |
1514 | #ifdef CONFIG_NUMA | |
1515 | if (node != -1 && c->node != node) | |
1516 | return 0; | |
1517 | #endif | |
1518 | return 1; | |
1519 | } | |
1520 | ||
81819f0f | 1521 | /* |
894b8788 CL |
1522 | * Slow path. The lockless freelist is empty or we need to perform |
1523 | * debugging duties. | |
1524 | * | |
1525 | * Interrupts are disabled. | |
81819f0f | 1526 | * |
894b8788 CL |
1527 | * Processing is still very fast if new objects have been freed to the |
1528 | * regular freelist. In that case we simply take over the regular freelist | |
1529 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1530 | * |
894b8788 CL |
1531 | * If that is not working then we fall back to the partial lists. We take the |
1532 | * first element of the freelist as the object to allocate now and move the | |
1533 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1534 | * |
894b8788 | 1535 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1536 | * we need to allocate a new slab. This is the slowest path since it involves |
1537 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1538 | */ |
894b8788 | 1539 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1540 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1541 | { |
81819f0f | 1542 | void **object; |
dfb4f096 | 1543 | struct page *new; |
81819f0f | 1544 | |
e72e9c23 LT |
1545 | /* We handle __GFP_ZERO in the caller */ |
1546 | gfpflags &= ~__GFP_ZERO; | |
1547 | ||
dfb4f096 | 1548 | if (!c->page) |
81819f0f CL |
1549 | goto new_slab; |
1550 | ||
dfb4f096 CL |
1551 | slab_lock(c->page); |
1552 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1553 | goto another_slab; |
6446faa2 | 1554 | |
8ff12cfc | 1555 | stat(c, ALLOC_REFILL); |
6446faa2 | 1556 | |
894b8788 | 1557 | load_freelist: |
dfb4f096 | 1558 | object = c->page->freelist; |
a973e9dd | 1559 | if (unlikely(!object)) |
81819f0f | 1560 | goto another_slab; |
dfb4f096 | 1561 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1562 | goto debug; |
1563 | ||
b3fba8da | 1564 | c->freelist = object[c->offset]; |
39b26464 | 1565 | c->page->inuse = c->page->objects; |
a973e9dd | 1566 | c->page->freelist = NULL; |
dfb4f096 | 1567 | c->node = page_to_nid(c->page); |
1f84260c | 1568 | unlock_out: |
dfb4f096 | 1569 | slab_unlock(c->page); |
8ff12cfc | 1570 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1571 | return object; |
1572 | ||
1573 | another_slab: | |
dfb4f096 | 1574 | deactivate_slab(s, c); |
81819f0f CL |
1575 | |
1576 | new_slab: | |
dfb4f096 CL |
1577 | new = get_partial(s, gfpflags, node); |
1578 | if (new) { | |
1579 | c->page = new; | |
8ff12cfc | 1580 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1581 | goto load_freelist; |
81819f0f CL |
1582 | } |
1583 | ||
b811c202 CL |
1584 | if (gfpflags & __GFP_WAIT) |
1585 | local_irq_enable(); | |
1586 | ||
dfb4f096 | 1587 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1588 | |
1589 | if (gfpflags & __GFP_WAIT) | |
1590 | local_irq_disable(); | |
1591 | ||
dfb4f096 CL |
1592 | if (new) { |
1593 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1594 | stat(c, ALLOC_SLAB); |
05aa3450 | 1595 | if (c->page) |
dfb4f096 | 1596 | flush_slab(s, c); |
dfb4f096 CL |
1597 | slab_lock(new); |
1598 | SetSlabFrozen(new); | |
1599 | c->page = new; | |
4b6f0750 | 1600 | goto load_freelist; |
81819f0f | 1601 | } |
71c7a06f | 1602 | return NULL; |
81819f0f | 1603 | debug: |
dfb4f096 | 1604 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1605 | goto another_slab; |
894b8788 | 1606 | |
dfb4f096 | 1607 | c->page->inuse++; |
b3fba8da | 1608 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1609 | c->node = -1; |
1f84260c | 1610 | goto unlock_out; |
894b8788 CL |
1611 | } |
1612 | ||
1613 | /* | |
1614 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1615 | * have the fastpath folded into their functions. So no function call | |
1616 | * overhead for requests that can be satisfied on the fastpath. | |
1617 | * | |
1618 | * The fastpath works by first checking if the lockless freelist can be used. | |
1619 | * If not then __slab_alloc is called for slow processing. | |
1620 | * | |
1621 | * Otherwise we can simply pick the next object from the lockless free list. | |
1622 | */ | |
06428780 | 1623 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce15fea8 | 1624 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1625 | { |
894b8788 | 1626 | void **object; |
dfb4f096 | 1627 | struct kmem_cache_cpu *c; |
1f84260c CL |
1628 | unsigned long flags; |
1629 | ||
894b8788 | 1630 | local_irq_save(flags); |
dfb4f096 | 1631 | c = get_cpu_slab(s, smp_processor_id()); |
a973e9dd | 1632 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1633 | |
dfb4f096 | 1634 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1635 | |
1636 | else { | |
dfb4f096 | 1637 | object = c->freelist; |
b3fba8da | 1638 | c->freelist = object[c->offset]; |
8ff12cfc | 1639 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1640 | } |
1641 | local_irq_restore(flags); | |
d07dbea4 CL |
1642 | |
1643 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
42a9fdbb | 1644 | memset(object, 0, c->objsize); |
d07dbea4 | 1645 | |
894b8788 | 1646 | return object; |
81819f0f CL |
1647 | } |
1648 | ||
1649 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1650 | { | |
ce15fea8 | 1651 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1652 | } |
1653 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1654 | ||
1655 | #ifdef CONFIG_NUMA | |
1656 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1657 | { | |
ce15fea8 | 1658 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1659 | } |
1660 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1661 | #endif | |
1662 | ||
1663 | /* | |
894b8788 CL |
1664 | * Slow patch handling. This may still be called frequently since objects |
1665 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1666 | * |
894b8788 CL |
1667 | * So we still attempt to reduce cache line usage. Just take the slab |
1668 | * lock and free the item. If there is no additional partial page | |
1669 | * handling required then we can return immediately. | |
81819f0f | 1670 | */ |
894b8788 | 1671 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1672 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1673 | { |
1674 | void *prior; | |
1675 | void **object = (void *)x; | |
8ff12cfc | 1676 | struct kmem_cache_cpu *c; |
81819f0f | 1677 | |
8ff12cfc CL |
1678 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1679 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1680 | slab_lock(page); |
1681 | ||
35e5d7ee | 1682 | if (unlikely(SlabDebug(page))) |
81819f0f | 1683 | goto debug; |
6446faa2 | 1684 | |
81819f0f | 1685 | checks_ok: |
b3fba8da | 1686 | prior = object[offset] = page->freelist; |
81819f0f CL |
1687 | page->freelist = object; |
1688 | page->inuse--; | |
1689 | ||
8ff12cfc CL |
1690 | if (unlikely(SlabFrozen(page))) { |
1691 | stat(c, FREE_FROZEN); | |
81819f0f | 1692 | goto out_unlock; |
8ff12cfc | 1693 | } |
81819f0f CL |
1694 | |
1695 | if (unlikely(!page->inuse)) | |
1696 | goto slab_empty; | |
1697 | ||
1698 | /* | |
6446faa2 | 1699 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1700 | * then add it. |
1701 | */ | |
a973e9dd | 1702 | if (unlikely(!prior)) { |
7c2e132c | 1703 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1704 | stat(c, FREE_ADD_PARTIAL); |
1705 | } | |
81819f0f CL |
1706 | |
1707 | out_unlock: | |
1708 | slab_unlock(page); | |
81819f0f CL |
1709 | return; |
1710 | ||
1711 | slab_empty: | |
a973e9dd | 1712 | if (prior) { |
81819f0f | 1713 | /* |
672bba3a | 1714 | * Slab still on the partial list. |
81819f0f CL |
1715 | */ |
1716 | remove_partial(s, page); | |
8ff12cfc CL |
1717 | stat(c, FREE_REMOVE_PARTIAL); |
1718 | } | |
81819f0f | 1719 | slab_unlock(page); |
8ff12cfc | 1720 | stat(c, FREE_SLAB); |
81819f0f | 1721 | discard_slab(s, page); |
81819f0f CL |
1722 | return; |
1723 | ||
1724 | debug: | |
3ec09742 | 1725 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1726 | goto out_unlock; |
77c5e2d0 | 1727 | goto checks_ok; |
81819f0f CL |
1728 | } |
1729 | ||
894b8788 CL |
1730 | /* |
1731 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1732 | * can perform fastpath freeing without additional function calls. | |
1733 | * | |
1734 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1735 | * of this processor. This typically the case if we have just allocated | |
1736 | * the item before. | |
1737 | * | |
1738 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1739 | * with all sorts of special processing. | |
1740 | */ | |
06428780 | 1741 | static __always_inline void slab_free(struct kmem_cache *s, |
894b8788 CL |
1742 | struct page *page, void *x, void *addr) |
1743 | { | |
1744 | void **object = (void *)x; | |
dfb4f096 | 1745 | struct kmem_cache_cpu *c; |
1f84260c CL |
1746 | unsigned long flags; |
1747 | ||
894b8788 | 1748 | local_irq_save(flags); |
dfb4f096 | 1749 | c = get_cpu_slab(s, smp_processor_id()); |
27d9e4e9 | 1750 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a TG |
1751 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
1752 | debug_check_no_obj_freed(object, s->objsize); | |
ee3c72a1 | 1753 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1754 | object[c->offset] = c->freelist; |
dfb4f096 | 1755 | c->freelist = object; |
8ff12cfc | 1756 | stat(c, FREE_FASTPATH); |
894b8788 | 1757 | } else |
b3fba8da | 1758 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1759 | |
1760 | local_irq_restore(flags); | |
1761 | } | |
1762 | ||
81819f0f CL |
1763 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1764 | { | |
77c5e2d0 | 1765 | struct page *page; |
81819f0f | 1766 | |
b49af68f | 1767 | page = virt_to_head_page(x); |
81819f0f | 1768 | |
77c5e2d0 | 1769 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1770 | } |
1771 | EXPORT_SYMBOL(kmem_cache_free); | |
1772 | ||
1773 | /* Figure out on which slab object the object resides */ | |
1774 | static struct page *get_object_page(const void *x) | |
1775 | { | |
b49af68f | 1776 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1777 | |
1778 | if (!PageSlab(page)) | |
1779 | return NULL; | |
1780 | ||
1781 | return page; | |
1782 | } | |
1783 | ||
1784 | /* | |
672bba3a CL |
1785 | * Object placement in a slab is made very easy because we always start at |
1786 | * offset 0. If we tune the size of the object to the alignment then we can | |
1787 | * get the required alignment by putting one properly sized object after | |
1788 | * another. | |
81819f0f CL |
1789 | * |
1790 | * Notice that the allocation order determines the sizes of the per cpu | |
1791 | * caches. Each processor has always one slab available for allocations. | |
1792 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1793 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1794 | * locking overhead. |
81819f0f CL |
1795 | */ |
1796 | ||
1797 | /* | |
1798 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1799 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1800 | * and increases the number of allocations possible without having to | |
1801 | * take the list_lock. | |
1802 | */ | |
1803 | static int slub_min_order; | |
114e9e89 | 1804 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1805 | static int slub_min_objects; |
81819f0f CL |
1806 | |
1807 | /* | |
1808 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1809 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1810 | */ |
1811 | static int slub_nomerge; | |
1812 | ||
81819f0f CL |
1813 | /* |
1814 | * Calculate the order of allocation given an slab object size. | |
1815 | * | |
672bba3a CL |
1816 | * The order of allocation has significant impact on performance and other |
1817 | * system components. Generally order 0 allocations should be preferred since | |
1818 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1819 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1820 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1821 | * would be wasted. |
1822 | * | |
1823 | * In order to reach satisfactory performance we must ensure that a minimum | |
1824 | * number of objects is in one slab. Otherwise we may generate too much | |
1825 | * activity on the partial lists which requires taking the list_lock. This is | |
1826 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1827 | * |
672bba3a CL |
1828 | * slub_max_order specifies the order where we begin to stop considering the |
1829 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1830 | * we try to keep the page order as low as possible. So we accept more waste | |
1831 | * of space in favor of a small page order. | |
81819f0f | 1832 | * |
672bba3a CL |
1833 | * Higher order allocations also allow the placement of more objects in a |
1834 | * slab and thereby reduce object handling overhead. If the user has | |
1835 | * requested a higher mininum order then we start with that one instead of | |
1836 | * the smallest order which will fit the object. | |
81819f0f | 1837 | */ |
5e6d444e CL |
1838 | static inline int slab_order(int size, int min_objects, |
1839 | int max_order, int fract_leftover) | |
81819f0f CL |
1840 | { |
1841 | int order; | |
1842 | int rem; | |
6300ea75 | 1843 | int min_order = slub_min_order; |
81819f0f | 1844 | |
39b26464 CL |
1845 | if ((PAGE_SIZE << min_order) / size > 65535) |
1846 | return get_order(size * 65535) - 1; | |
1847 | ||
6300ea75 | 1848 | for (order = max(min_order, |
5e6d444e CL |
1849 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1850 | order <= max_order; order++) { | |
81819f0f | 1851 | |
5e6d444e | 1852 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1853 | |
5e6d444e | 1854 | if (slab_size < min_objects * size) |
81819f0f CL |
1855 | continue; |
1856 | ||
1857 | rem = slab_size % size; | |
1858 | ||
5e6d444e | 1859 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1860 | break; |
1861 | ||
1862 | } | |
672bba3a | 1863 | |
81819f0f CL |
1864 | return order; |
1865 | } | |
1866 | ||
5e6d444e CL |
1867 | static inline int calculate_order(int size) |
1868 | { | |
1869 | int order; | |
1870 | int min_objects; | |
1871 | int fraction; | |
1872 | ||
1873 | /* | |
1874 | * Attempt to find best configuration for a slab. This | |
1875 | * works by first attempting to generate a layout with | |
1876 | * the best configuration and backing off gradually. | |
1877 | * | |
1878 | * First we reduce the acceptable waste in a slab. Then | |
1879 | * we reduce the minimum objects required in a slab. | |
1880 | */ | |
1881 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1882 | if (!min_objects) |
1883 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
5e6d444e | 1884 | while (min_objects > 1) { |
c124f5b5 | 1885 | fraction = 16; |
5e6d444e CL |
1886 | while (fraction >= 4) { |
1887 | order = slab_order(size, min_objects, | |
1888 | slub_max_order, fraction); | |
1889 | if (order <= slub_max_order) | |
1890 | return order; | |
1891 | fraction /= 2; | |
1892 | } | |
1893 | min_objects /= 2; | |
1894 | } | |
1895 | ||
1896 | /* | |
1897 | * We were unable to place multiple objects in a slab. Now | |
1898 | * lets see if we can place a single object there. | |
1899 | */ | |
1900 | order = slab_order(size, 1, slub_max_order, 1); | |
1901 | if (order <= slub_max_order) | |
1902 | return order; | |
1903 | ||
1904 | /* | |
1905 | * Doh this slab cannot be placed using slub_max_order. | |
1906 | */ | |
1907 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1908 | if (order <= MAX_ORDER) | |
1909 | return order; | |
1910 | return -ENOSYS; | |
1911 | } | |
1912 | ||
81819f0f | 1913 | /* |
672bba3a | 1914 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1915 | */ |
1916 | static unsigned long calculate_alignment(unsigned long flags, | |
1917 | unsigned long align, unsigned long size) | |
1918 | { | |
1919 | /* | |
6446faa2 CL |
1920 | * If the user wants hardware cache aligned objects then follow that |
1921 | * suggestion if the object is sufficiently large. | |
81819f0f | 1922 | * |
6446faa2 CL |
1923 | * The hardware cache alignment cannot override the specified |
1924 | * alignment though. If that is greater then use it. | |
81819f0f | 1925 | */ |
b6210386 NP |
1926 | if (flags & SLAB_HWCACHE_ALIGN) { |
1927 | unsigned long ralign = cache_line_size(); | |
1928 | while (size <= ralign / 2) | |
1929 | ralign /= 2; | |
1930 | align = max(align, ralign); | |
1931 | } | |
81819f0f CL |
1932 | |
1933 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 1934 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
1935 | |
1936 | return ALIGN(align, sizeof(void *)); | |
1937 | } | |
1938 | ||
dfb4f096 CL |
1939 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1940 | struct kmem_cache_cpu *c) | |
1941 | { | |
1942 | c->page = NULL; | |
a973e9dd | 1943 | c->freelist = NULL; |
dfb4f096 | 1944 | c->node = 0; |
42a9fdbb CL |
1945 | c->offset = s->offset / sizeof(void *); |
1946 | c->objsize = s->objsize; | |
62f75532 PE |
1947 | #ifdef CONFIG_SLUB_STATS |
1948 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
1949 | #endif | |
dfb4f096 CL |
1950 | } |
1951 | ||
81819f0f CL |
1952 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
1953 | { | |
1954 | n->nr_partial = 0; | |
81819f0f CL |
1955 | spin_lock_init(&n->list_lock); |
1956 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1957 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 1958 | atomic_long_set(&n->nr_slabs, 0); |
643b1138 | 1959 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1960 | #endif |
81819f0f CL |
1961 | } |
1962 | ||
4c93c355 CL |
1963 | #ifdef CONFIG_SMP |
1964 | /* | |
1965 | * Per cpu array for per cpu structures. | |
1966 | * | |
1967 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1968 | * close together meaning that it becomes possible that multiple per cpu | |
1969 | * structures are contained in one cacheline. This may be particularly | |
1970 | * beneficial for the kmalloc caches. | |
1971 | * | |
1972 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1973 | * likely able to get per cpu structures for all caches from the array defined | |
1974 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
1975 | * | |
1976 | * If the per cpu array is exhausted then fall back to kmalloc | |
1977 | * of individual cachelines. No sharing is possible then. | |
1978 | */ | |
1979 | #define NR_KMEM_CACHE_CPU 100 | |
1980 | ||
1981 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
1982 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
1983 | ||
1984 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
1985 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
1986 | ||
1987 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
1988 | int cpu, gfp_t flags) | |
1989 | { | |
1990 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
1991 | ||
1992 | if (c) | |
1993 | per_cpu(kmem_cache_cpu_free, cpu) = | |
1994 | (void *)c->freelist; | |
1995 | else { | |
1996 | /* Table overflow: So allocate ourselves */ | |
1997 | c = kmalloc_node( | |
1998 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
1999 | flags, cpu_to_node(cpu)); | |
2000 | if (!c) | |
2001 | return NULL; | |
2002 | } | |
2003 | ||
2004 | init_kmem_cache_cpu(s, c); | |
2005 | return c; | |
2006 | } | |
2007 | ||
2008 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2009 | { | |
2010 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
2011 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
2012 | kfree(c); | |
2013 | return; | |
2014 | } | |
2015 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2016 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2017 | } | |
2018 | ||
2019 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2020 | { | |
2021 | int cpu; | |
2022 | ||
2023 | for_each_online_cpu(cpu) { | |
2024 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2025 | ||
2026 | if (c) { | |
2027 | s->cpu_slab[cpu] = NULL; | |
2028 | free_kmem_cache_cpu(c, cpu); | |
2029 | } | |
2030 | } | |
2031 | } | |
2032 | ||
2033 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2034 | { | |
2035 | int cpu; | |
2036 | ||
2037 | for_each_online_cpu(cpu) { | |
2038 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2039 | ||
2040 | if (c) | |
2041 | continue; | |
2042 | ||
2043 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2044 | if (!c) { | |
2045 | free_kmem_cache_cpus(s); | |
2046 | return 0; | |
2047 | } | |
2048 | s->cpu_slab[cpu] = c; | |
2049 | } | |
2050 | return 1; | |
2051 | } | |
2052 | ||
2053 | /* | |
2054 | * Initialize the per cpu array. | |
2055 | */ | |
2056 | static void init_alloc_cpu_cpu(int cpu) | |
2057 | { | |
2058 | int i; | |
2059 | ||
2060 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
2061 | return; | |
2062 | ||
2063 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2064 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2065 | ||
2066 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
2067 | } | |
2068 | ||
2069 | static void __init init_alloc_cpu(void) | |
2070 | { | |
2071 | int cpu; | |
2072 | ||
2073 | for_each_online_cpu(cpu) | |
2074 | init_alloc_cpu_cpu(cpu); | |
2075 | } | |
2076 | ||
2077 | #else | |
2078 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2079 | static inline void init_alloc_cpu(void) {} | |
2080 | ||
2081 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2082 | { | |
2083 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2084 | return 1; | |
2085 | } | |
2086 | #endif | |
2087 | ||
81819f0f CL |
2088 | #ifdef CONFIG_NUMA |
2089 | /* | |
2090 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2091 | * slab on the node for this slabcache. There are no concurrent accesses | |
2092 | * possible. | |
2093 | * | |
2094 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2095 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2096 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2097 | */ |
1cd7daa5 AB |
2098 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2099 | int node) | |
81819f0f CL |
2100 | { |
2101 | struct page *page; | |
2102 | struct kmem_cache_node *n; | |
ba84c73c | 2103 | unsigned long flags; |
81819f0f CL |
2104 | |
2105 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2106 | ||
a2f92ee7 | 2107 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2108 | |
2109 | BUG_ON(!page); | |
a2f92ee7 CL |
2110 | if (page_to_nid(page) != node) { |
2111 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2112 | "node %d\n", node); | |
2113 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2114 | "in order to be able to continue\n"); | |
2115 | } | |
2116 | ||
81819f0f CL |
2117 | n = page->freelist; |
2118 | BUG_ON(!n); | |
2119 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2120 | page->inuse++; | |
2121 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2122 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2123 | init_object(kmalloc_caches, n, 1); |
2124 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2125 | #endif |
81819f0f | 2126 | init_kmem_cache_node(n); |
205ab99d | 2127 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2128 | |
ba84c73c | 2129 | /* |
2130 | * lockdep requires consistent irq usage for each lock | |
2131 | * so even though there cannot be a race this early in | |
2132 | * the boot sequence, we still disable irqs. | |
2133 | */ | |
2134 | local_irq_save(flags); | |
7c2e132c | 2135 | add_partial(n, page, 0); |
ba84c73c | 2136 | local_irq_restore(flags); |
81819f0f CL |
2137 | return n; |
2138 | } | |
2139 | ||
2140 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2141 | { | |
2142 | int node; | |
2143 | ||
f64dc58c | 2144 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2145 | struct kmem_cache_node *n = s->node[node]; |
2146 | if (n && n != &s->local_node) | |
2147 | kmem_cache_free(kmalloc_caches, n); | |
2148 | s->node[node] = NULL; | |
2149 | } | |
2150 | } | |
2151 | ||
2152 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2153 | { | |
2154 | int node; | |
2155 | int local_node; | |
2156 | ||
2157 | if (slab_state >= UP) | |
2158 | local_node = page_to_nid(virt_to_page(s)); | |
2159 | else | |
2160 | local_node = 0; | |
2161 | ||
f64dc58c | 2162 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2163 | struct kmem_cache_node *n; |
2164 | ||
2165 | if (local_node == node) | |
2166 | n = &s->local_node; | |
2167 | else { | |
2168 | if (slab_state == DOWN) { | |
2169 | n = early_kmem_cache_node_alloc(gfpflags, | |
2170 | node); | |
2171 | continue; | |
2172 | } | |
2173 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2174 | gfpflags, node); | |
2175 | ||
2176 | if (!n) { | |
2177 | free_kmem_cache_nodes(s); | |
2178 | return 0; | |
2179 | } | |
2180 | ||
2181 | } | |
2182 | s->node[node] = n; | |
2183 | init_kmem_cache_node(n); | |
2184 | } | |
2185 | return 1; | |
2186 | } | |
2187 | #else | |
2188 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2189 | { | |
2190 | } | |
2191 | ||
2192 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2193 | { | |
2194 | init_kmem_cache_node(&s->local_node); | |
2195 | return 1; | |
2196 | } | |
2197 | #endif | |
2198 | ||
2199 | /* | |
2200 | * calculate_sizes() determines the order and the distribution of data within | |
2201 | * a slab object. | |
2202 | */ | |
06b285dc | 2203 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2204 | { |
2205 | unsigned long flags = s->flags; | |
2206 | unsigned long size = s->objsize; | |
2207 | unsigned long align = s->align; | |
834f3d11 | 2208 | int order; |
81819f0f | 2209 | |
d8b42bf5 CL |
2210 | /* |
2211 | * Round up object size to the next word boundary. We can only | |
2212 | * place the free pointer at word boundaries and this determines | |
2213 | * the possible location of the free pointer. | |
2214 | */ | |
2215 | size = ALIGN(size, sizeof(void *)); | |
2216 | ||
2217 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2218 | /* |
2219 | * Determine if we can poison the object itself. If the user of | |
2220 | * the slab may touch the object after free or before allocation | |
2221 | * then we should never poison the object itself. | |
2222 | */ | |
2223 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2224 | !s->ctor) |
81819f0f CL |
2225 | s->flags |= __OBJECT_POISON; |
2226 | else | |
2227 | s->flags &= ~__OBJECT_POISON; | |
2228 | ||
81819f0f CL |
2229 | |
2230 | /* | |
672bba3a | 2231 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2232 | * end of the object and the free pointer. If not then add an |
672bba3a | 2233 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2234 | */ |
2235 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2236 | size += sizeof(void *); | |
41ecc55b | 2237 | #endif |
81819f0f CL |
2238 | |
2239 | /* | |
672bba3a CL |
2240 | * With that we have determined the number of bytes in actual use |
2241 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2242 | */ |
2243 | s->inuse = size; | |
2244 | ||
2245 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2246 | s->ctor)) { |
81819f0f CL |
2247 | /* |
2248 | * Relocate free pointer after the object if it is not | |
2249 | * permitted to overwrite the first word of the object on | |
2250 | * kmem_cache_free. | |
2251 | * | |
2252 | * This is the case if we do RCU, have a constructor or | |
2253 | * destructor or are poisoning the objects. | |
2254 | */ | |
2255 | s->offset = size; | |
2256 | size += sizeof(void *); | |
2257 | } | |
2258 | ||
c12b3c62 | 2259 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2260 | if (flags & SLAB_STORE_USER) |
2261 | /* | |
2262 | * Need to store information about allocs and frees after | |
2263 | * the object. | |
2264 | */ | |
2265 | size += 2 * sizeof(struct track); | |
2266 | ||
be7b3fbc | 2267 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2268 | /* |
2269 | * Add some empty padding so that we can catch | |
2270 | * overwrites from earlier objects rather than let | |
2271 | * tracking information or the free pointer be | |
2272 | * corrupted if an user writes before the start | |
2273 | * of the object. | |
2274 | */ | |
2275 | size += sizeof(void *); | |
41ecc55b | 2276 | #endif |
672bba3a | 2277 | |
81819f0f CL |
2278 | /* |
2279 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2280 | * user specified and the dynamic determination of cache line size |
2281 | * on bootup. | |
81819f0f CL |
2282 | */ |
2283 | align = calculate_alignment(flags, align, s->objsize); | |
2284 | ||
2285 | /* | |
2286 | * SLUB stores one object immediately after another beginning from | |
2287 | * offset 0. In order to align the objects we have to simply size | |
2288 | * each object to conform to the alignment. | |
2289 | */ | |
2290 | size = ALIGN(size, align); | |
2291 | s->size = size; | |
06b285dc CL |
2292 | if (forced_order >= 0) |
2293 | order = forced_order; | |
2294 | else | |
2295 | order = calculate_order(size); | |
81819f0f | 2296 | |
834f3d11 | 2297 | if (order < 0) |
81819f0f CL |
2298 | return 0; |
2299 | ||
b7a49f0d | 2300 | s->allocflags = 0; |
834f3d11 | 2301 | if (order) |
b7a49f0d CL |
2302 | s->allocflags |= __GFP_COMP; |
2303 | ||
2304 | if (s->flags & SLAB_CACHE_DMA) | |
2305 | s->allocflags |= SLUB_DMA; | |
2306 | ||
2307 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2308 | s->allocflags |= __GFP_RECLAIMABLE; | |
2309 | ||
81819f0f CL |
2310 | /* |
2311 | * Determine the number of objects per slab | |
2312 | */ | |
834f3d11 | 2313 | s->oo = oo_make(order, size); |
65c3376a | 2314 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2315 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2316 | s->max = s->oo; | |
81819f0f | 2317 | |
834f3d11 | 2318 | return !!oo_objects(s->oo); |
81819f0f CL |
2319 | |
2320 | } | |
2321 | ||
81819f0f CL |
2322 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2323 | const char *name, size_t size, | |
2324 | size_t align, unsigned long flags, | |
4ba9b9d0 | 2325 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
2326 | { |
2327 | memset(s, 0, kmem_size); | |
2328 | s->name = name; | |
2329 | s->ctor = ctor; | |
81819f0f | 2330 | s->objsize = size; |
81819f0f | 2331 | s->align = align; |
ba0268a8 | 2332 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2333 | |
06b285dc | 2334 | if (!calculate_sizes(s, -1)) |
81819f0f CL |
2335 | goto error; |
2336 | ||
2337 | s->refcount = 1; | |
2338 | #ifdef CONFIG_NUMA | |
9824601e | 2339 | s->remote_node_defrag_ratio = 100; |
81819f0f | 2340 | #endif |
dfb4f096 CL |
2341 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2342 | goto error; | |
81819f0f | 2343 | |
dfb4f096 | 2344 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2345 | return 1; |
4c93c355 | 2346 | free_kmem_cache_nodes(s); |
81819f0f CL |
2347 | error: |
2348 | if (flags & SLAB_PANIC) | |
2349 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2350 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2351 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2352 | s->offset, flags); |
2353 | return 0; | |
2354 | } | |
81819f0f CL |
2355 | |
2356 | /* | |
2357 | * Check if a given pointer is valid | |
2358 | */ | |
2359 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2360 | { | |
06428780 | 2361 | struct page *page; |
81819f0f CL |
2362 | |
2363 | page = get_object_page(object); | |
2364 | ||
2365 | if (!page || s != page->slab) | |
2366 | /* No slab or wrong slab */ | |
2367 | return 0; | |
2368 | ||
abcd08a6 | 2369 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2370 | return 0; |
2371 | ||
2372 | /* | |
2373 | * We could also check if the object is on the slabs freelist. | |
2374 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2375 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2376 | * to a certain slab. |
2377 | */ | |
2378 | return 1; | |
2379 | } | |
2380 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2381 | ||
2382 | /* | |
2383 | * Determine the size of a slab object | |
2384 | */ | |
2385 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2386 | { | |
2387 | return s->objsize; | |
2388 | } | |
2389 | EXPORT_SYMBOL(kmem_cache_size); | |
2390 | ||
2391 | const char *kmem_cache_name(struct kmem_cache *s) | |
2392 | { | |
2393 | return s->name; | |
2394 | } | |
2395 | EXPORT_SYMBOL(kmem_cache_name); | |
2396 | ||
33b12c38 CL |
2397 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2398 | const char *text) | |
2399 | { | |
2400 | #ifdef CONFIG_SLUB_DEBUG | |
2401 | void *addr = page_address(page); | |
2402 | void *p; | |
2403 | DECLARE_BITMAP(map, page->objects); | |
2404 | ||
2405 | bitmap_zero(map, page->objects); | |
2406 | slab_err(s, page, "%s", text); | |
2407 | slab_lock(page); | |
2408 | for_each_free_object(p, s, page->freelist) | |
2409 | set_bit(slab_index(p, s, addr), map); | |
2410 | ||
2411 | for_each_object(p, s, addr, page->objects) { | |
2412 | ||
2413 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2414 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2415 | p, p - addr); | |
2416 | print_tracking(s, p); | |
2417 | } | |
2418 | } | |
2419 | slab_unlock(page); | |
2420 | #endif | |
2421 | } | |
2422 | ||
81819f0f | 2423 | /* |
599870b1 | 2424 | * Attempt to free all partial slabs on a node. |
81819f0f | 2425 | */ |
599870b1 | 2426 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2427 | { |
81819f0f CL |
2428 | unsigned long flags; |
2429 | struct page *page, *h; | |
2430 | ||
2431 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2432 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2433 | if (!page->inuse) { |
2434 | list_del(&page->lru); | |
2435 | discard_slab(s, page); | |
599870b1 | 2436 | n->nr_partial--; |
33b12c38 CL |
2437 | } else { |
2438 | list_slab_objects(s, page, | |
2439 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2440 | } |
33b12c38 | 2441 | } |
81819f0f | 2442 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2443 | } |
2444 | ||
2445 | /* | |
672bba3a | 2446 | * Release all resources used by a slab cache. |
81819f0f | 2447 | */ |
0c710013 | 2448 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2449 | { |
2450 | int node; | |
2451 | ||
2452 | flush_all(s); | |
2453 | ||
2454 | /* Attempt to free all objects */ | |
4c93c355 | 2455 | free_kmem_cache_cpus(s); |
f64dc58c | 2456 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2457 | struct kmem_cache_node *n = get_node(s, node); |
2458 | ||
599870b1 CL |
2459 | free_partial(s, n); |
2460 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2461 | return 1; |
2462 | } | |
2463 | free_kmem_cache_nodes(s); | |
2464 | return 0; | |
2465 | } | |
2466 | ||
2467 | /* | |
2468 | * Close a cache and release the kmem_cache structure | |
2469 | * (must be used for caches created using kmem_cache_create) | |
2470 | */ | |
2471 | void kmem_cache_destroy(struct kmem_cache *s) | |
2472 | { | |
2473 | down_write(&slub_lock); | |
2474 | s->refcount--; | |
2475 | if (!s->refcount) { | |
2476 | list_del(&s->list); | |
a0e1d1be | 2477 | up_write(&slub_lock); |
d629d819 PE |
2478 | if (kmem_cache_close(s)) { |
2479 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2480 | "still has objects.\n", s->name, __func__); | |
2481 | dump_stack(); | |
2482 | } | |
81819f0f | 2483 | sysfs_slab_remove(s); |
a0e1d1be CL |
2484 | } else |
2485 | up_write(&slub_lock); | |
81819f0f CL |
2486 | } |
2487 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2488 | ||
2489 | /******************************************************************** | |
2490 | * Kmalloc subsystem | |
2491 | *******************************************************************/ | |
2492 | ||
331dc558 | 2493 | struct kmem_cache kmalloc_caches[PAGE_SHIFT + 1] __cacheline_aligned; |
81819f0f CL |
2494 | EXPORT_SYMBOL(kmalloc_caches); |
2495 | ||
81819f0f CL |
2496 | static int __init setup_slub_min_order(char *str) |
2497 | { | |
06428780 | 2498 | get_option(&str, &slub_min_order); |
81819f0f CL |
2499 | |
2500 | return 1; | |
2501 | } | |
2502 | ||
2503 | __setup("slub_min_order=", setup_slub_min_order); | |
2504 | ||
2505 | static int __init setup_slub_max_order(char *str) | |
2506 | { | |
06428780 | 2507 | get_option(&str, &slub_max_order); |
81819f0f CL |
2508 | |
2509 | return 1; | |
2510 | } | |
2511 | ||
2512 | __setup("slub_max_order=", setup_slub_max_order); | |
2513 | ||
2514 | static int __init setup_slub_min_objects(char *str) | |
2515 | { | |
06428780 | 2516 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2517 | |
2518 | return 1; | |
2519 | } | |
2520 | ||
2521 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2522 | ||
2523 | static int __init setup_slub_nomerge(char *str) | |
2524 | { | |
2525 | slub_nomerge = 1; | |
2526 | return 1; | |
2527 | } | |
2528 | ||
2529 | __setup("slub_nomerge", setup_slub_nomerge); | |
2530 | ||
81819f0f CL |
2531 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2532 | const char *name, int size, gfp_t gfp_flags) | |
2533 | { | |
2534 | unsigned int flags = 0; | |
2535 | ||
2536 | if (gfp_flags & SLUB_DMA) | |
2537 | flags = SLAB_CACHE_DMA; | |
2538 | ||
2539 | down_write(&slub_lock); | |
2540 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
319d1e24 | 2541 | flags, NULL)) |
81819f0f CL |
2542 | goto panic; |
2543 | ||
2544 | list_add(&s->list, &slab_caches); | |
2545 | up_write(&slub_lock); | |
2546 | if (sysfs_slab_add(s)) | |
2547 | goto panic; | |
2548 | return s; | |
2549 | ||
2550 | panic: | |
2551 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2552 | } | |
2553 | ||
2e443fd0 | 2554 | #ifdef CONFIG_ZONE_DMA |
4097d601 | 2555 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT + 1]; |
1ceef402 CL |
2556 | |
2557 | static void sysfs_add_func(struct work_struct *w) | |
2558 | { | |
2559 | struct kmem_cache *s; | |
2560 | ||
2561 | down_write(&slub_lock); | |
2562 | list_for_each_entry(s, &slab_caches, list) { | |
2563 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2564 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2565 | sysfs_slab_add(s); | |
2566 | } | |
2567 | } | |
2568 | up_write(&slub_lock); | |
2569 | } | |
2570 | ||
2571 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2572 | ||
2e443fd0 CL |
2573 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2574 | { | |
2575 | struct kmem_cache *s; | |
2e443fd0 CL |
2576 | char *text; |
2577 | size_t realsize; | |
2578 | ||
2579 | s = kmalloc_caches_dma[index]; | |
2580 | if (s) | |
2581 | return s; | |
2582 | ||
2583 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2584 | if (flags & __GFP_WAIT) |
2585 | down_write(&slub_lock); | |
2586 | else { | |
2587 | if (!down_write_trylock(&slub_lock)) | |
2588 | goto out; | |
2589 | } | |
2590 | ||
2591 | if (kmalloc_caches_dma[index]) | |
2592 | goto unlock_out; | |
2e443fd0 | 2593 | |
7b55f620 | 2594 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2595 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2596 | (unsigned int)realsize); | |
1ceef402 CL |
2597 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2598 | ||
2599 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2600 | realsize, ARCH_KMALLOC_MINALIGN, | |
2601 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2602 | kfree(s); | |
2603 | kfree(text); | |
2604 | goto unlock_out; | |
dfce8648 | 2605 | } |
1ceef402 CL |
2606 | |
2607 | list_add(&s->list, &slab_caches); | |
2608 | kmalloc_caches_dma[index] = s; | |
2609 | ||
2610 | schedule_work(&sysfs_add_work); | |
2611 | ||
2612 | unlock_out: | |
dfce8648 | 2613 | up_write(&slub_lock); |
1ceef402 | 2614 | out: |
dfce8648 | 2615 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2616 | } |
2617 | #endif | |
2618 | ||
f1b26339 CL |
2619 | /* |
2620 | * Conversion table for small slabs sizes / 8 to the index in the | |
2621 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2622 | * of two cache sizes there. The size of larger slabs can be determined using | |
2623 | * fls. | |
2624 | */ | |
2625 | static s8 size_index[24] = { | |
2626 | 3, /* 8 */ | |
2627 | 4, /* 16 */ | |
2628 | 5, /* 24 */ | |
2629 | 5, /* 32 */ | |
2630 | 6, /* 40 */ | |
2631 | 6, /* 48 */ | |
2632 | 6, /* 56 */ | |
2633 | 6, /* 64 */ | |
2634 | 1, /* 72 */ | |
2635 | 1, /* 80 */ | |
2636 | 1, /* 88 */ | |
2637 | 1, /* 96 */ | |
2638 | 7, /* 104 */ | |
2639 | 7, /* 112 */ | |
2640 | 7, /* 120 */ | |
2641 | 7, /* 128 */ | |
2642 | 2, /* 136 */ | |
2643 | 2, /* 144 */ | |
2644 | 2, /* 152 */ | |
2645 | 2, /* 160 */ | |
2646 | 2, /* 168 */ | |
2647 | 2, /* 176 */ | |
2648 | 2, /* 184 */ | |
2649 | 2 /* 192 */ | |
2650 | }; | |
2651 | ||
81819f0f CL |
2652 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2653 | { | |
f1b26339 | 2654 | int index; |
81819f0f | 2655 | |
f1b26339 CL |
2656 | if (size <= 192) { |
2657 | if (!size) | |
2658 | return ZERO_SIZE_PTR; | |
81819f0f | 2659 | |
f1b26339 | 2660 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2661 | } else |
f1b26339 | 2662 | index = fls(size - 1); |
81819f0f CL |
2663 | |
2664 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2665 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2666 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2667 | |
81819f0f CL |
2668 | #endif |
2669 | return &kmalloc_caches[index]; | |
2670 | } | |
2671 | ||
2672 | void *__kmalloc(size_t size, gfp_t flags) | |
2673 | { | |
aadb4bc4 | 2674 | struct kmem_cache *s; |
81819f0f | 2675 | |
331dc558 | 2676 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2677 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2678 | |
2679 | s = get_slab(size, flags); | |
2680 | ||
2681 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2682 | return s; |
2683 | ||
ce15fea8 | 2684 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2685 | } |
2686 | EXPORT_SYMBOL(__kmalloc); | |
2687 | ||
f619cfe1 CL |
2688 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2689 | { | |
2690 | struct page *page = alloc_pages_node(node, flags | __GFP_COMP, | |
2691 | get_order(size)); | |
2692 | ||
2693 | if (page) | |
2694 | return page_address(page); | |
2695 | else | |
2696 | return NULL; | |
2697 | } | |
2698 | ||
81819f0f CL |
2699 | #ifdef CONFIG_NUMA |
2700 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2701 | { | |
aadb4bc4 | 2702 | struct kmem_cache *s; |
81819f0f | 2703 | |
331dc558 | 2704 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 2705 | return kmalloc_large_node(size, flags, node); |
aadb4bc4 CL |
2706 | |
2707 | s = get_slab(size, flags); | |
2708 | ||
2709 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2710 | return s; |
2711 | ||
ce15fea8 | 2712 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2713 | } |
2714 | EXPORT_SYMBOL(__kmalloc_node); | |
2715 | #endif | |
2716 | ||
2717 | size_t ksize(const void *object) | |
2718 | { | |
272c1d21 | 2719 | struct page *page; |
81819f0f CL |
2720 | struct kmem_cache *s; |
2721 | ||
ef8b4520 | 2722 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2723 | return 0; |
2724 | ||
294a80a8 | 2725 | page = virt_to_head_page(object); |
294a80a8 VN |
2726 | |
2727 | if (unlikely(!PageSlab(page))) | |
2728 | return PAGE_SIZE << compound_order(page); | |
2729 | ||
81819f0f | 2730 | s = page->slab; |
81819f0f | 2731 | |
ae20bfda | 2732 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2733 | /* |
2734 | * Debugging requires use of the padding between object | |
2735 | * and whatever may come after it. | |
2736 | */ | |
2737 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2738 | return s->objsize; | |
2739 | ||
ae20bfda | 2740 | #endif |
81819f0f CL |
2741 | /* |
2742 | * If we have the need to store the freelist pointer | |
2743 | * back there or track user information then we can | |
2744 | * only use the space before that information. | |
2745 | */ | |
2746 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2747 | return s->inuse; | |
81819f0f CL |
2748 | /* |
2749 | * Else we can use all the padding etc for the allocation | |
2750 | */ | |
2751 | return s->size; | |
2752 | } | |
2753 | EXPORT_SYMBOL(ksize); | |
2754 | ||
2755 | void kfree(const void *x) | |
2756 | { | |
81819f0f | 2757 | struct page *page; |
5bb983b0 | 2758 | void *object = (void *)x; |
81819f0f | 2759 | |
2408c550 | 2760 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2761 | return; |
2762 | ||
b49af68f | 2763 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2764 | if (unlikely(!PageSlab(page))) { |
2765 | put_page(page); | |
2766 | return; | |
2767 | } | |
5bb983b0 | 2768 | slab_free(page->slab, page, object, __builtin_return_address(0)); |
81819f0f CL |
2769 | } |
2770 | EXPORT_SYMBOL(kfree); | |
2771 | ||
2086d26a | 2772 | /* |
672bba3a CL |
2773 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2774 | * the remaining slabs by the number of items in use. The slabs with the | |
2775 | * most items in use come first. New allocations will then fill those up | |
2776 | * and thus they can be removed from the partial lists. | |
2777 | * | |
2778 | * The slabs with the least items are placed last. This results in them | |
2779 | * being allocated from last increasing the chance that the last objects | |
2780 | * are freed in them. | |
2086d26a CL |
2781 | */ |
2782 | int kmem_cache_shrink(struct kmem_cache *s) | |
2783 | { | |
2784 | int node; | |
2785 | int i; | |
2786 | struct kmem_cache_node *n; | |
2787 | struct page *page; | |
2788 | struct page *t; | |
205ab99d | 2789 | int objects = oo_objects(s->max); |
2086d26a | 2790 | struct list_head *slabs_by_inuse = |
834f3d11 | 2791 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2792 | unsigned long flags; |
2793 | ||
2794 | if (!slabs_by_inuse) | |
2795 | return -ENOMEM; | |
2796 | ||
2797 | flush_all(s); | |
f64dc58c | 2798 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2799 | n = get_node(s, node); |
2800 | ||
2801 | if (!n->nr_partial) | |
2802 | continue; | |
2803 | ||
834f3d11 | 2804 | for (i = 0; i < objects; i++) |
2086d26a CL |
2805 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2806 | ||
2807 | spin_lock_irqsave(&n->list_lock, flags); | |
2808 | ||
2809 | /* | |
672bba3a | 2810 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2811 | * |
672bba3a CL |
2812 | * Note that concurrent frees may occur while we hold the |
2813 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2814 | */ |
2815 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2816 | if (!page->inuse && slab_trylock(page)) { | |
2817 | /* | |
2818 | * Must hold slab lock here because slab_free | |
2819 | * may have freed the last object and be | |
2820 | * waiting to release the slab. | |
2821 | */ | |
2822 | list_del(&page->lru); | |
2823 | n->nr_partial--; | |
2824 | slab_unlock(page); | |
2825 | discard_slab(s, page); | |
2826 | } else { | |
fcda3d89 CL |
2827 | list_move(&page->lru, |
2828 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2829 | } |
2830 | } | |
2831 | ||
2086d26a | 2832 | /* |
672bba3a CL |
2833 | * Rebuild the partial list with the slabs filled up most |
2834 | * first and the least used slabs at the end. | |
2086d26a | 2835 | */ |
834f3d11 | 2836 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2837 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2838 | ||
2086d26a CL |
2839 | spin_unlock_irqrestore(&n->list_lock, flags); |
2840 | } | |
2841 | ||
2842 | kfree(slabs_by_inuse); | |
2843 | return 0; | |
2844 | } | |
2845 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2846 | ||
b9049e23 YG |
2847 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2848 | static int slab_mem_going_offline_callback(void *arg) | |
2849 | { | |
2850 | struct kmem_cache *s; | |
2851 | ||
2852 | down_read(&slub_lock); | |
2853 | list_for_each_entry(s, &slab_caches, list) | |
2854 | kmem_cache_shrink(s); | |
2855 | up_read(&slub_lock); | |
2856 | ||
2857 | return 0; | |
2858 | } | |
2859 | ||
2860 | static void slab_mem_offline_callback(void *arg) | |
2861 | { | |
2862 | struct kmem_cache_node *n; | |
2863 | struct kmem_cache *s; | |
2864 | struct memory_notify *marg = arg; | |
2865 | int offline_node; | |
2866 | ||
2867 | offline_node = marg->status_change_nid; | |
2868 | ||
2869 | /* | |
2870 | * If the node still has available memory. we need kmem_cache_node | |
2871 | * for it yet. | |
2872 | */ | |
2873 | if (offline_node < 0) | |
2874 | return; | |
2875 | ||
2876 | down_read(&slub_lock); | |
2877 | list_for_each_entry(s, &slab_caches, list) { | |
2878 | n = get_node(s, offline_node); | |
2879 | if (n) { | |
2880 | /* | |
2881 | * if n->nr_slabs > 0, slabs still exist on the node | |
2882 | * that is going down. We were unable to free them, | |
2883 | * and offline_pages() function shoudn't call this | |
2884 | * callback. So, we must fail. | |
2885 | */ | |
0f389ec6 | 2886 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2887 | |
2888 | s->node[offline_node] = NULL; | |
2889 | kmem_cache_free(kmalloc_caches, n); | |
2890 | } | |
2891 | } | |
2892 | up_read(&slub_lock); | |
2893 | } | |
2894 | ||
2895 | static int slab_mem_going_online_callback(void *arg) | |
2896 | { | |
2897 | struct kmem_cache_node *n; | |
2898 | struct kmem_cache *s; | |
2899 | struct memory_notify *marg = arg; | |
2900 | int nid = marg->status_change_nid; | |
2901 | int ret = 0; | |
2902 | ||
2903 | /* | |
2904 | * If the node's memory is already available, then kmem_cache_node is | |
2905 | * already created. Nothing to do. | |
2906 | */ | |
2907 | if (nid < 0) | |
2908 | return 0; | |
2909 | ||
2910 | /* | |
2911 | * We are bringing a node online. No memory is availabe yet. We must | |
2912 | * allocate a kmem_cache_node structure in order to bring the node | |
2913 | * online. | |
2914 | */ | |
2915 | down_read(&slub_lock); | |
2916 | list_for_each_entry(s, &slab_caches, list) { | |
2917 | /* | |
2918 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2919 | * since memory is not yet available from the node that | |
2920 | * is brought up. | |
2921 | */ | |
2922 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2923 | if (!n) { | |
2924 | ret = -ENOMEM; | |
2925 | goto out; | |
2926 | } | |
2927 | init_kmem_cache_node(n); | |
2928 | s->node[nid] = n; | |
2929 | } | |
2930 | out: | |
2931 | up_read(&slub_lock); | |
2932 | return ret; | |
2933 | } | |
2934 | ||
2935 | static int slab_memory_callback(struct notifier_block *self, | |
2936 | unsigned long action, void *arg) | |
2937 | { | |
2938 | int ret = 0; | |
2939 | ||
2940 | switch (action) { | |
2941 | case MEM_GOING_ONLINE: | |
2942 | ret = slab_mem_going_online_callback(arg); | |
2943 | break; | |
2944 | case MEM_GOING_OFFLINE: | |
2945 | ret = slab_mem_going_offline_callback(arg); | |
2946 | break; | |
2947 | case MEM_OFFLINE: | |
2948 | case MEM_CANCEL_ONLINE: | |
2949 | slab_mem_offline_callback(arg); | |
2950 | break; | |
2951 | case MEM_ONLINE: | |
2952 | case MEM_CANCEL_OFFLINE: | |
2953 | break; | |
2954 | } | |
2955 | ||
2956 | ret = notifier_from_errno(ret); | |
2957 | return ret; | |
2958 | } | |
2959 | ||
2960 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2961 | ||
81819f0f CL |
2962 | /******************************************************************** |
2963 | * Basic setup of slabs | |
2964 | *******************************************************************/ | |
2965 | ||
2966 | void __init kmem_cache_init(void) | |
2967 | { | |
2968 | int i; | |
4b356be0 | 2969 | int caches = 0; |
81819f0f | 2970 | |
4c93c355 CL |
2971 | init_alloc_cpu(); |
2972 | ||
81819f0f CL |
2973 | #ifdef CONFIG_NUMA |
2974 | /* | |
2975 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2976 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2977 | * kmem_cache_open for slab_state == DOWN. |
2978 | */ | |
2979 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2980 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2981 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 2982 | caches++; |
b9049e23 | 2983 | |
0c40ba4f | 2984 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
2985 | #endif |
2986 | ||
2987 | /* Able to allocate the per node structures */ | |
2988 | slab_state = PARTIAL; | |
2989 | ||
2990 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
2991 | if (KMALLOC_MIN_SIZE <= 64) { |
2992 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 2993 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
2994 | caches++; |
2995 | } | |
2996 | if (KMALLOC_MIN_SIZE <= 128) { | |
2997 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 2998 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
2999 | caches++; |
3000 | } | |
81819f0f | 3001 | |
331dc558 | 3002 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) { |
81819f0f CL |
3003 | create_kmalloc_cache(&kmalloc_caches[i], |
3004 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
3005 | caches++; |
3006 | } | |
81819f0f | 3007 | |
f1b26339 CL |
3008 | |
3009 | /* | |
3010 | * Patch up the size_index table if we have strange large alignment | |
3011 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3012 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3013 | * |
3014 | * Largest permitted alignment is 256 bytes due to the way we | |
3015 | * handle the index determination for the smaller caches. | |
3016 | * | |
3017 | * Make sure that nothing crazy happens if someone starts tinkering | |
3018 | * around with ARCH_KMALLOC_MINALIGN | |
3019 | */ | |
3020 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3021 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3022 | ||
12ad6843 | 3023 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3024 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3025 | ||
81819f0f CL |
3026 | slab_state = UP; |
3027 | ||
3028 | /* Provide the correct kmalloc names now that the caches are up */ | |
331dc558 | 3029 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) |
81819f0f CL |
3030 | kmalloc_caches[i]. name = |
3031 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3032 | ||
3033 | #ifdef CONFIG_SMP | |
3034 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3035 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3036 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3037 | #else | |
3038 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3039 | #endif |
3040 | ||
3adbefee IM |
3041 | printk(KERN_INFO |
3042 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3043 | " CPUs=%d, Nodes=%d\n", |
3044 | caches, cache_line_size(), | |
81819f0f CL |
3045 | slub_min_order, slub_max_order, slub_min_objects, |
3046 | nr_cpu_ids, nr_node_ids); | |
3047 | } | |
3048 | ||
3049 | /* | |
3050 | * Find a mergeable slab cache | |
3051 | */ | |
3052 | static int slab_unmergeable(struct kmem_cache *s) | |
3053 | { | |
3054 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3055 | return 1; | |
3056 | ||
c59def9f | 3057 | if (s->ctor) |
81819f0f CL |
3058 | return 1; |
3059 | ||
8ffa6875 CL |
3060 | /* |
3061 | * We may have set a slab to be unmergeable during bootstrap. | |
3062 | */ | |
3063 | if (s->refcount < 0) | |
3064 | return 1; | |
3065 | ||
81819f0f CL |
3066 | return 0; |
3067 | } | |
3068 | ||
3069 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3070 | size_t align, unsigned long flags, const char *name, |
4ba9b9d0 | 3071 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f | 3072 | { |
5b95a4ac | 3073 | struct kmem_cache *s; |
81819f0f CL |
3074 | |
3075 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3076 | return NULL; | |
3077 | ||
c59def9f | 3078 | if (ctor) |
81819f0f CL |
3079 | return NULL; |
3080 | ||
3081 | size = ALIGN(size, sizeof(void *)); | |
3082 | align = calculate_alignment(flags, align, size); | |
3083 | size = ALIGN(size, align); | |
ba0268a8 | 3084 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3085 | |
5b95a4ac | 3086 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3087 | if (slab_unmergeable(s)) |
3088 | continue; | |
3089 | ||
3090 | if (size > s->size) | |
3091 | continue; | |
3092 | ||
ba0268a8 | 3093 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3094 | continue; |
3095 | /* | |
3096 | * Check if alignment is compatible. | |
3097 | * Courtesy of Adrian Drzewiecki | |
3098 | */ | |
06428780 | 3099 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3100 | continue; |
3101 | ||
3102 | if (s->size - size >= sizeof(void *)) | |
3103 | continue; | |
3104 | ||
3105 | return s; | |
3106 | } | |
3107 | return NULL; | |
3108 | } | |
3109 | ||
3110 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
3111 | size_t align, unsigned long flags, | |
4ba9b9d0 | 3112 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
3113 | { |
3114 | struct kmem_cache *s; | |
3115 | ||
3116 | down_write(&slub_lock); | |
ba0268a8 | 3117 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3118 | if (s) { |
42a9fdbb CL |
3119 | int cpu; |
3120 | ||
81819f0f CL |
3121 | s->refcount++; |
3122 | /* | |
3123 | * Adjust the object sizes so that we clear | |
3124 | * the complete object on kzalloc. | |
3125 | */ | |
3126 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3127 | |
3128 | /* | |
3129 | * And then we need to update the object size in the | |
3130 | * per cpu structures | |
3131 | */ | |
3132 | for_each_online_cpu(cpu) | |
3133 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3134 | |
81819f0f | 3135 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3136 | up_write(&slub_lock); |
6446faa2 | 3137 | |
81819f0f CL |
3138 | if (sysfs_slab_alias(s, name)) |
3139 | goto err; | |
a0e1d1be CL |
3140 | return s; |
3141 | } | |
6446faa2 | 3142 | |
a0e1d1be CL |
3143 | s = kmalloc(kmem_size, GFP_KERNEL); |
3144 | if (s) { | |
3145 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3146 | size, align, flags, ctor)) { |
81819f0f | 3147 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
3148 | up_write(&slub_lock); |
3149 | if (sysfs_slab_add(s)) | |
3150 | goto err; | |
3151 | return s; | |
3152 | } | |
3153 | kfree(s); | |
81819f0f CL |
3154 | } |
3155 | up_write(&slub_lock); | |
81819f0f CL |
3156 | |
3157 | err: | |
81819f0f CL |
3158 | if (flags & SLAB_PANIC) |
3159 | panic("Cannot create slabcache %s\n", name); | |
3160 | else | |
3161 | s = NULL; | |
3162 | return s; | |
3163 | } | |
3164 | EXPORT_SYMBOL(kmem_cache_create); | |
3165 | ||
81819f0f | 3166 | #ifdef CONFIG_SMP |
81819f0f | 3167 | /* |
672bba3a CL |
3168 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3169 | * necessary. | |
81819f0f CL |
3170 | */ |
3171 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3172 | unsigned long action, void *hcpu) | |
3173 | { | |
3174 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3175 | struct kmem_cache *s; |
3176 | unsigned long flags; | |
81819f0f CL |
3177 | |
3178 | switch (action) { | |
4c93c355 CL |
3179 | case CPU_UP_PREPARE: |
3180 | case CPU_UP_PREPARE_FROZEN: | |
3181 | init_alloc_cpu_cpu(cpu); | |
3182 | down_read(&slub_lock); | |
3183 | list_for_each_entry(s, &slab_caches, list) | |
3184 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3185 | GFP_KERNEL); | |
3186 | up_read(&slub_lock); | |
3187 | break; | |
3188 | ||
81819f0f | 3189 | case CPU_UP_CANCELED: |
8bb78442 | 3190 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3191 | case CPU_DEAD: |
8bb78442 | 3192 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3193 | down_read(&slub_lock); |
3194 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3195 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3196 | ||
5b95a4ac CL |
3197 | local_irq_save(flags); |
3198 | __flush_cpu_slab(s, cpu); | |
3199 | local_irq_restore(flags); | |
4c93c355 CL |
3200 | free_kmem_cache_cpu(c, cpu); |
3201 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3202 | } |
3203 | up_read(&slub_lock); | |
81819f0f CL |
3204 | break; |
3205 | default: | |
3206 | break; | |
3207 | } | |
3208 | return NOTIFY_OK; | |
3209 | } | |
3210 | ||
06428780 | 3211 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3212 | .notifier_call = slab_cpuup_callback |
06428780 | 3213 | }; |
81819f0f CL |
3214 | |
3215 | #endif | |
3216 | ||
81819f0f CL |
3217 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
3218 | { | |
aadb4bc4 CL |
3219 | struct kmem_cache *s; |
3220 | ||
331dc558 | 3221 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3222 | return kmalloc_large(size, gfpflags); |
3223 | ||
aadb4bc4 | 3224 | s = get_slab(size, gfpflags); |
81819f0f | 3225 | |
2408c550 | 3226 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3227 | return s; |
81819f0f | 3228 | |
ce15fea8 | 3229 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3230 | } |
3231 | ||
3232 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
3233 | int node, void *caller) | |
3234 | { | |
aadb4bc4 CL |
3235 | struct kmem_cache *s; |
3236 | ||
331dc558 | 3237 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 3238 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3239 | |
aadb4bc4 | 3240 | s = get_slab(size, gfpflags); |
81819f0f | 3241 | |
2408c550 | 3242 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3243 | return s; |
81819f0f | 3244 | |
ce15fea8 | 3245 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3246 | } |
3247 | ||
5b06c853 | 3248 | #if (defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)) || defined(CONFIG_SLABINFO) |
205ab99d CL |
3249 | static unsigned long count_partial(struct kmem_cache_node *n, |
3250 | int (*get_count)(struct page *)) | |
5b06c853 CL |
3251 | { |
3252 | unsigned long flags; | |
3253 | unsigned long x = 0; | |
3254 | struct page *page; | |
3255 | ||
3256 | spin_lock_irqsave(&n->list_lock, flags); | |
3257 | list_for_each_entry(page, &n->partial, lru) | |
205ab99d | 3258 | x += get_count(page); |
5b06c853 CL |
3259 | spin_unlock_irqrestore(&n->list_lock, flags); |
3260 | return x; | |
3261 | } | |
205ab99d CL |
3262 | |
3263 | static int count_inuse(struct page *page) | |
3264 | { | |
3265 | return page->inuse; | |
3266 | } | |
3267 | ||
3268 | static int count_total(struct page *page) | |
3269 | { | |
3270 | return page->objects; | |
3271 | } | |
3272 | ||
3273 | static int count_free(struct page *page) | |
3274 | { | |
3275 | return page->objects - page->inuse; | |
3276 | } | |
5b06c853 CL |
3277 | #endif |
3278 | ||
41ecc55b | 3279 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
3280 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3281 | unsigned long *map) | |
53e15af0 CL |
3282 | { |
3283 | void *p; | |
a973e9dd | 3284 | void *addr = page_address(page); |
53e15af0 CL |
3285 | |
3286 | if (!check_slab(s, page) || | |
3287 | !on_freelist(s, page, NULL)) | |
3288 | return 0; | |
3289 | ||
3290 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3291 | bitmap_zero(map, page->objects); |
53e15af0 | 3292 | |
7656c72b CL |
3293 | for_each_free_object(p, s, page->freelist) { |
3294 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3295 | if (!check_object(s, page, p, 0)) |
3296 | return 0; | |
3297 | } | |
3298 | ||
224a88be | 3299 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3300 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3301 | if (!check_object(s, page, p, 1)) |
3302 | return 0; | |
3303 | return 1; | |
3304 | } | |
3305 | ||
434e245d CL |
3306 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3307 | unsigned long *map) | |
53e15af0 CL |
3308 | { |
3309 | if (slab_trylock(page)) { | |
434e245d | 3310 | validate_slab(s, page, map); |
53e15af0 CL |
3311 | slab_unlock(page); |
3312 | } else | |
3313 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3314 | s->name, page); | |
3315 | ||
3316 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3317 | if (!SlabDebug(page)) |
3318 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3319 | "on slab 0x%p\n", s->name, page); |
3320 | } else { | |
35e5d7ee CL |
3321 | if (SlabDebug(page)) |
3322 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3323 | "slab 0x%p\n", s->name, page); |
3324 | } | |
3325 | } | |
3326 | ||
434e245d CL |
3327 | static int validate_slab_node(struct kmem_cache *s, |
3328 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3329 | { |
3330 | unsigned long count = 0; | |
3331 | struct page *page; | |
3332 | unsigned long flags; | |
3333 | ||
3334 | spin_lock_irqsave(&n->list_lock, flags); | |
3335 | ||
3336 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3337 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3338 | count++; |
3339 | } | |
3340 | if (count != n->nr_partial) | |
3341 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3342 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3343 | ||
3344 | if (!(s->flags & SLAB_STORE_USER)) | |
3345 | goto out; | |
3346 | ||
3347 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3348 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3349 | count++; |
3350 | } | |
3351 | if (count != atomic_long_read(&n->nr_slabs)) | |
3352 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3353 | "counter=%ld\n", s->name, count, | |
3354 | atomic_long_read(&n->nr_slabs)); | |
3355 | ||
3356 | out: | |
3357 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3358 | return count; | |
3359 | } | |
3360 | ||
434e245d | 3361 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3362 | { |
3363 | int node; | |
3364 | unsigned long count = 0; | |
205ab99d | 3365 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3366 | sizeof(unsigned long), GFP_KERNEL); |
3367 | ||
3368 | if (!map) | |
3369 | return -ENOMEM; | |
53e15af0 CL |
3370 | |
3371 | flush_all(s); | |
f64dc58c | 3372 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3373 | struct kmem_cache_node *n = get_node(s, node); |
3374 | ||
434e245d | 3375 | count += validate_slab_node(s, n, map); |
53e15af0 | 3376 | } |
434e245d | 3377 | kfree(map); |
53e15af0 CL |
3378 | return count; |
3379 | } | |
3380 | ||
b3459709 CL |
3381 | #ifdef SLUB_RESILIENCY_TEST |
3382 | static void resiliency_test(void) | |
3383 | { | |
3384 | u8 *p; | |
3385 | ||
3386 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3387 | printk(KERN_ERR "-----------------------\n"); | |
3388 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3389 | ||
3390 | p = kzalloc(16, GFP_KERNEL); | |
3391 | p[16] = 0x12; | |
3392 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3393 | " 0x12->0x%p\n\n", p + 16); | |
3394 | ||
3395 | validate_slab_cache(kmalloc_caches + 4); | |
3396 | ||
3397 | /* Hmmm... The next two are dangerous */ | |
3398 | p = kzalloc(32, GFP_KERNEL); | |
3399 | p[32 + sizeof(void *)] = 0x34; | |
3400 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3401 | " 0x34 -> -0x%p\n", p); |
3402 | printk(KERN_ERR | |
3403 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3404 | |
3405 | validate_slab_cache(kmalloc_caches + 5); | |
3406 | p = kzalloc(64, GFP_KERNEL); | |
3407 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3408 | *p = 0x56; | |
3409 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3410 | p); | |
3adbefee IM |
3411 | printk(KERN_ERR |
3412 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3413 | validate_slab_cache(kmalloc_caches + 6); |
3414 | ||
3415 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3416 | p = kzalloc(128, GFP_KERNEL); | |
3417 | kfree(p); | |
3418 | *p = 0x78; | |
3419 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3420 | validate_slab_cache(kmalloc_caches + 7); | |
3421 | ||
3422 | p = kzalloc(256, GFP_KERNEL); | |
3423 | kfree(p); | |
3424 | p[50] = 0x9a; | |
3adbefee IM |
3425 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3426 | p); | |
b3459709 CL |
3427 | validate_slab_cache(kmalloc_caches + 8); |
3428 | ||
3429 | p = kzalloc(512, GFP_KERNEL); | |
3430 | kfree(p); | |
3431 | p[512] = 0xab; | |
3432 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3433 | validate_slab_cache(kmalloc_caches + 9); | |
3434 | } | |
3435 | #else | |
3436 | static void resiliency_test(void) {}; | |
3437 | #endif | |
3438 | ||
88a420e4 | 3439 | /* |
672bba3a | 3440 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3441 | * and freed. |
3442 | */ | |
3443 | ||
3444 | struct location { | |
3445 | unsigned long count; | |
3446 | void *addr; | |
45edfa58 CL |
3447 | long long sum_time; |
3448 | long min_time; | |
3449 | long max_time; | |
3450 | long min_pid; | |
3451 | long max_pid; | |
3452 | cpumask_t cpus; | |
3453 | nodemask_t nodes; | |
88a420e4 CL |
3454 | }; |
3455 | ||
3456 | struct loc_track { | |
3457 | unsigned long max; | |
3458 | unsigned long count; | |
3459 | struct location *loc; | |
3460 | }; | |
3461 | ||
3462 | static void free_loc_track(struct loc_track *t) | |
3463 | { | |
3464 | if (t->max) | |
3465 | free_pages((unsigned long)t->loc, | |
3466 | get_order(sizeof(struct location) * t->max)); | |
3467 | } | |
3468 | ||
68dff6a9 | 3469 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3470 | { |
3471 | struct location *l; | |
3472 | int order; | |
3473 | ||
88a420e4 CL |
3474 | order = get_order(sizeof(struct location) * max); |
3475 | ||
68dff6a9 | 3476 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3477 | if (!l) |
3478 | return 0; | |
3479 | ||
3480 | if (t->count) { | |
3481 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3482 | free_loc_track(t); | |
3483 | } | |
3484 | t->max = max; | |
3485 | t->loc = l; | |
3486 | return 1; | |
3487 | } | |
3488 | ||
3489 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3490 | const struct track *track) |
88a420e4 CL |
3491 | { |
3492 | long start, end, pos; | |
3493 | struct location *l; | |
3494 | void *caddr; | |
45edfa58 | 3495 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3496 | |
3497 | start = -1; | |
3498 | end = t->count; | |
3499 | ||
3500 | for ( ; ; ) { | |
3501 | pos = start + (end - start + 1) / 2; | |
3502 | ||
3503 | /* | |
3504 | * There is nothing at "end". If we end up there | |
3505 | * we need to add something to before end. | |
3506 | */ | |
3507 | if (pos == end) | |
3508 | break; | |
3509 | ||
3510 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3511 | if (track->addr == caddr) { |
3512 | ||
3513 | l = &t->loc[pos]; | |
3514 | l->count++; | |
3515 | if (track->when) { | |
3516 | l->sum_time += age; | |
3517 | if (age < l->min_time) | |
3518 | l->min_time = age; | |
3519 | if (age > l->max_time) | |
3520 | l->max_time = age; | |
3521 | ||
3522 | if (track->pid < l->min_pid) | |
3523 | l->min_pid = track->pid; | |
3524 | if (track->pid > l->max_pid) | |
3525 | l->max_pid = track->pid; | |
3526 | ||
3527 | cpu_set(track->cpu, l->cpus); | |
3528 | } | |
3529 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3530 | return 1; |
3531 | } | |
3532 | ||
45edfa58 | 3533 | if (track->addr < caddr) |
88a420e4 CL |
3534 | end = pos; |
3535 | else | |
3536 | start = pos; | |
3537 | } | |
3538 | ||
3539 | /* | |
672bba3a | 3540 | * Not found. Insert new tracking element. |
88a420e4 | 3541 | */ |
68dff6a9 | 3542 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3543 | return 0; |
3544 | ||
3545 | l = t->loc + pos; | |
3546 | if (pos < t->count) | |
3547 | memmove(l + 1, l, | |
3548 | (t->count - pos) * sizeof(struct location)); | |
3549 | t->count++; | |
3550 | l->count = 1; | |
45edfa58 CL |
3551 | l->addr = track->addr; |
3552 | l->sum_time = age; | |
3553 | l->min_time = age; | |
3554 | l->max_time = age; | |
3555 | l->min_pid = track->pid; | |
3556 | l->max_pid = track->pid; | |
3557 | cpus_clear(l->cpus); | |
3558 | cpu_set(track->cpu, l->cpus); | |
3559 | nodes_clear(l->nodes); | |
3560 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3561 | return 1; |
3562 | } | |
3563 | ||
3564 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3565 | struct page *page, enum track_item alloc) | |
3566 | { | |
a973e9dd | 3567 | void *addr = page_address(page); |
39b26464 | 3568 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3569 | void *p; |
3570 | ||
39b26464 | 3571 | bitmap_zero(map, page->objects); |
7656c72b CL |
3572 | for_each_free_object(p, s, page->freelist) |
3573 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3574 | |
224a88be | 3575 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3576 | if (!test_bit(slab_index(p, s, addr), map)) |
3577 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3578 | } |
3579 | ||
3580 | static int list_locations(struct kmem_cache *s, char *buf, | |
3581 | enum track_item alloc) | |
3582 | { | |
e374d483 | 3583 | int len = 0; |
88a420e4 | 3584 | unsigned long i; |
68dff6a9 | 3585 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3586 | int node; |
3587 | ||
68dff6a9 | 3588 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3589 | GFP_TEMPORARY)) |
68dff6a9 | 3590 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3591 | |
3592 | /* Push back cpu slabs */ | |
3593 | flush_all(s); | |
3594 | ||
f64dc58c | 3595 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3596 | struct kmem_cache_node *n = get_node(s, node); |
3597 | unsigned long flags; | |
3598 | struct page *page; | |
3599 | ||
9e86943b | 3600 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3601 | continue; |
3602 | ||
3603 | spin_lock_irqsave(&n->list_lock, flags); | |
3604 | list_for_each_entry(page, &n->partial, lru) | |
3605 | process_slab(&t, s, page, alloc); | |
3606 | list_for_each_entry(page, &n->full, lru) | |
3607 | process_slab(&t, s, page, alloc); | |
3608 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3609 | } | |
3610 | ||
3611 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3612 | struct location *l = &t.loc[i]; |
88a420e4 | 3613 | |
e374d483 | 3614 | if (len > PAGE_SIZE - 100) |
88a420e4 | 3615 | break; |
e374d483 | 3616 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3617 | |
3618 | if (l->addr) | |
e374d483 | 3619 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3620 | else |
e374d483 | 3621 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3622 | |
3623 | if (l->sum_time != l->min_time) { | |
3624 | unsigned long remainder; | |
3625 | ||
e374d483 | 3626 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
45edfa58 CL |
3627 | l->min_time, |
3628 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3629 | l->max_time); | |
3630 | } else | |
e374d483 | 3631 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3632 | l->min_time); |
3633 | ||
3634 | if (l->min_pid != l->max_pid) | |
e374d483 | 3635 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3636 | l->min_pid, l->max_pid); |
3637 | else | |
e374d483 | 3638 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3639 | l->min_pid); |
3640 | ||
84966343 | 3641 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
e374d483 HH |
3642 | len < PAGE_SIZE - 60) { |
3643 | len += sprintf(buf + len, " cpus="); | |
3644 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3645 | l->cpus); |
3646 | } | |
3647 | ||
84966343 | 3648 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3649 | len < PAGE_SIZE - 60) { |
3650 | len += sprintf(buf + len, " nodes="); | |
3651 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3652 | l->nodes); |
3653 | } | |
3654 | ||
e374d483 | 3655 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3656 | } |
3657 | ||
3658 | free_loc_track(&t); | |
3659 | if (!t.count) | |
e374d483 HH |
3660 | len += sprintf(buf, "No data\n"); |
3661 | return len; | |
88a420e4 CL |
3662 | } |
3663 | ||
81819f0f | 3664 | enum slab_stat_type { |
205ab99d CL |
3665 | SL_ALL, /* All slabs */ |
3666 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3667 | SL_CPU, /* Only slabs used for cpu caches */ | |
3668 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3669 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3670 | }; |
3671 | ||
205ab99d | 3672 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3673 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3674 | #define SO_CPU (1 << SL_CPU) | |
3675 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3676 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3677 | |
62e5c4b4 CG |
3678 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3679 | char *buf, unsigned long flags) | |
81819f0f CL |
3680 | { |
3681 | unsigned long total = 0; | |
81819f0f CL |
3682 | int node; |
3683 | int x; | |
3684 | unsigned long *nodes; | |
3685 | unsigned long *per_cpu; | |
3686 | ||
3687 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3688 | if (!nodes) |
3689 | return -ENOMEM; | |
81819f0f CL |
3690 | per_cpu = nodes + nr_node_ids; |
3691 | ||
205ab99d CL |
3692 | if (flags & SO_CPU) { |
3693 | int cpu; | |
81819f0f | 3694 | |
205ab99d CL |
3695 | for_each_possible_cpu(cpu) { |
3696 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3697 | |
205ab99d CL |
3698 | if (!c || c->node < 0) |
3699 | continue; | |
3700 | ||
3701 | if (c->page) { | |
3702 | if (flags & SO_TOTAL) | |
3703 | x = c->page->objects; | |
3704 | else if (flags & SO_OBJECTS) | |
3705 | x = c->page->inuse; | |
81819f0f CL |
3706 | else |
3707 | x = 1; | |
205ab99d | 3708 | |
81819f0f | 3709 | total += x; |
205ab99d | 3710 | nodes[c->node] += x; |
81819f0f | 3711 | } |
205ab99d | 3712 | per_cpu[c->node]++; |
81819f0f CL |
3713 | } |
3714 | } | |
3715 | ||
205ab99d CL |
3716 | if (flags & SO_ALL) { |
3717 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3718 | struct kmem_cache_node *n = get_node(s, node); | |
3719 | ||
3720 | if (flags & SO_TOTAL) | |
3721 | x = atomic_long_read(&n->total_objects); | |
3722 | else if (flags & SO_OBJECTS) | |
3723 | x = atomic_long_read(&n->total_objects) - | |
3724 | count_partial(n, count_free); | |
81819f0f | 3725 | |
81819f0f | 3726 | else |
205ab99d | 3727 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3728 | total += x; |
3729 | nodes[node] += x; | |
3730 | } | |
3731 | ||
205ab99d CL |
3732 | } else if (flags & SO_PARTIAL) { |
3733 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3734 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3735 | |
205ab99d CL |
3736 | if (flags & SO_TOTAL) |
3737 | x = count_partial(n, count_total); | |
3738 | else if (flags & SO_OBJECTS) | |
3739 | x = count_partial(n, count_inuse); | |
81819f0f | 3740 | else |
205ab99d | 3741 | x = n->nr_partial; |
81819f0f CL |
3742 | total += x; |
3743 | nodes[node] += x; | |
3744 | } | |
3745 | } | |
81819f0f CL |
3746 | x = sprintf(buf, "%lu", total); |
3747 | #ifdef CONFIG_NUMA | |
f64dc58c | 3748 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3749 | if (nodes[node]) |
3750 | x += sprintf(buf + x, " N%d=%lu", | |
3751 | node, nodes[node]); | |
3752 | #endif | |
3753 | kfree(nodes); | |
3754 | return x + sprintf(buf + x, "\n"); | |
3755 | } | |
3756 | ||
3757 | static int any_slab_objects(struct kmem_cache *s) | |
3758 | { | |
3759 | int node; | |
81819f0f | 3760 | |
dfb4f096 | 3761 | for_each_online_node(node) { |
81819f0f CL |
3762 | struct kmem_cache_node *n = get_node(s, node); |
3763 | ||
dfb4f096 CL |
3764 | if (!n) |
3765 | continue; | |
3766 | ||
31d33baf | 3767 | if (atomic_read(&n->total_objects)) |
81819f0f CL |
3768 | return 1; |
3769 | } | |
3770 | return 0; | |
3771 | } | |
3772 | ||
3773 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3774 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3775 | ||
3776 | struct slab_attribute { | |
3777 | struct attribute attr; | |
3778 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3779 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3780 | }; | |
3781 | ||
3782 | #define SLAB_ATTR_RO(_name) \ | |
3783 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3784 | ||
3785 | #define SLAB_ATTR(_name) \ | |
3786 | static struct slab_attribute _name##_attr = \ | |
3787 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3788 | ||
81819f0f CL |
3789 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3790 | { | |
3791 | return sprintf(buf, "%d\n", s->size); | |
3792 | } | |
3793 | SLAB_ATTR_RO(slab_size); | |
3794 | ||
3795 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3796 | { | |
3797 | return sprintf(buf, "%d\n", s->align); | |
3798 | } | |
3799 | SLAB_ATTR_RO(align); | |
3800 | ||
3801 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3802 | { | |
3803 | return sprintf(buf, "%d\n", s->objsize); | |
3804 | } | |
3805 | SLAB_ATTR_RO(object_size); | |
3806 | ||
3807 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3808 | { | |
834f3d11 | 3809 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3810 | } |
3811 | SLAB_ATTR_RO(objs_per_slab); | |
3812 | ||
06b285dc CL |
3813 | static ssize_t order_store(struct kmem_cache *s, |
3814 | const char *buf, size_t length) | |
3815 | { | |
3816 | int order = simple_strtoul(buf, NULL, 10); | |
3817 | ||
3818 | if (order > slub_max_order || order < slub_min_order) | |
3819 | return -EINVAL; | |
3820 | ||
3821 | calculate_sizes(s, order); | |
3822 | return length; | |
3823 | } | |
3824 | ||
81819f0f CL |
3825 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3826 | { | |
834f3d11 | 3827 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3828 | } |
06b285dc | 3829 | SLAB_ATTR(order); |
81819f0f CL |
3830 | |
3831 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3832 | { | |
3833 | if (s->ctor) { | |
3834 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3835 | ||
3836 | return n + sprintf(buf + n, "\n"); | |
3837 | } | |
3838 | return 0; | |
3839 | } | |
3840 | SLAB_ATTR_RO(ctor); | |
3841 | ||
81819f0f CL |
3842 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3843 | { | |
3844 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3845 | } | |
3846 | SLAB_ATTR_RO(aliases); | |
3847 | ||
3848 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3849 | { | |
205ab99d | 3850 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3851 | } |
3852 | SLAB_ATTR_RO(slabs); | |
3853 | ||
3854 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3855 | { | |
d9acf4b7 | 3856 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3857 | } |
3858 | SLAB_ATTR_RO(partial); | |
3859 | ||
3860 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3861 | { | |
d9acf4b7 | 3862 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3863 | } |
3864 | SLAB_ATTR_RO(cpu_slabs); | |
3865 | ||
3866 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3867 | { | |
205ab99d | 3868 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3869 | } |
3870 | SLAB_ATTR_RO(objects); | |
3871 | ||
205ab99d CL |
3872 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3873 | { | |
3874 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3875 | } | |
3876 | SLAB_ATTR_RO(objects_partial); | |
3877 | ||
3878 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
3879 | { | |
3880 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
3881 | } | |
3882 | SLAB_ATTR_RO(total_objects); | |
3883 | ||
81819f0f CL |
3884 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
3885 | { | |
3886 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3887 | } | |
3888 | ||
3889 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3890 | const char *buf, size_t length) | |
3891 | { | |
3892 | s->flags &= ~SLAB_DEBUG_FREE; | |
3893 | if (buf[0] == '1') | |
3894 | s->flags |= SLAB_DEBUG_FREE; | |
3895 | return length; | |
3896 | } | |
3897 | SLAB_ATTR(sanity_checks); | |
3898 | ||
3899 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3900 | { | |
3901 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3902 | } | |
3903 | ||
3904 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3905 | size_t length) | |
3906 | { | |
3907 | s->flags &= ~SLAB_TRACE; | |
3908 | if (buf[0] == '1') | |
3909 | s->flags |= SLAB_TRACE; | |
3910 | return length; | |
3911 | } | |
3912 | SLAB_ATTR(trace); | |
3913 | ||
3914 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3915 | { | |
3916 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3917 | } | |
3918 | ||
3919 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3920 | const char *buf, size_t length) | |
3921 | { | |
3922 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3923 | if (buf[0] == '1') | |
3924 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3925 | return length; | |
3926 | } | |
3927 | SLAB_ATTR(reclaim_account); | |
3928 | ||
3929 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3930 | { | |
5af60839 | 3931 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3932 | } |
3933 | SLAB_ATTR_RO(hwcache_align); | |
3934 | ||
3935 | #ifdef CONFIG_ZONE_DMA | |
3936 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3937 | { | |
3938 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3939 | } | |
3940 | SLAB_ATTR_RO(cache_dma); | |
3941 | #endif | |
3942 | ||
3943 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3944 | { | |
3945 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3946 | } | |
3947 | SLAB_ATTR_RO(destroy_by_rcu); | |
3948 | ||
3949 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3950 | { | |
3951 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3952 | } | |
3953 | ||
3954 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3955 | const char *buf, size_t length) | |
3956 | { | |
3957 | if (any_slab_objects(s)) | |
3958 | return -EBUSY; | |
3959 | ||
3960 | s->flags &= ~SLAB_RED_ZONE; | |
3961 | if (buf[0] == '1') | |
3962 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 3963 | calculate_sizes(s, -1); |
81819f0f CL |
3964 | return length; |
3965 | } | |
3966 | SLAB_ATTR(red_zone); | |
3967 | ||
3968 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3969 | { | |
3970 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3971 | } | |
3972 | ||
3973 | static ssize_t poison_store(struct kmem_cache *s, | |
3974 | const char *buf, size_t length) | |
3975 | { | |
3976 | if (any_slab_objects(s)) | |
3977 | return -EBUSY; | |
3978 | ||
3979 | s->flags &= ~SLAB_POISON; | |
3980 | if (buf[0] == '1') | |
3981 | s->flags |= SLAB_POISON; | |
06b285dc | 3982 | calculate_sizes(s, -1); |
81819f0f CL |
3983 | return length; |
3984 | } | |
3985 | SLAB_ATTR(poison); | |
3986 | ||
3987 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3988 | { | |
3989 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3990 | } | |
3991 | ||
3992 | static ssize_t store_user_store(struct kmem_cache *s, | |
3993 | const char *buf, size_t length) | |
3994 | { | |
3995 | if (any_slab_objects(s)) | |
3996 | return -EBUSY; | |
3997 | ||
3998 | s->flags &= ~SLAB_STORE_USER; | |
3999 | if (buf[0] == '1') | |
4000 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4001 | calculate_sizes(s, -1); |
81819f0f CL |
4002 | return length; |
4003 | } | |
4004 | SLAB_ATTR(store_user); | |
4005 | ||
53e15af0 CL |
4006 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4007 | { | |
4008 | return 0; | |
4009 | } | |
4010 | ||
4011 | static ssize_t validate_store(struct kmem_cache *s, | |
4012 | const char *buf, size_t length) | |
4013 | { | |
434e245d CL |
4014 | int ret = -EINVAL; |
4015 | ||
4016 | if (buf[0] == '1') { | |
4017 | ret = validate_slab_cache(s); | |
4018 | if (ret >= 0) | |
4019 | ret = length; | |
4020 | } | |
4021 | return ret; | |
53e15af0 CL |
4022 | } |
4023 | SLAB_ATTR(validate); | |
4024 | ||
2086d26a CL |
4025 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4026 | { | |
4027 | return 0; | |
4028 | } | |
4029 | ||
4030 | static ssize_t shrink_store(struct kmem_cache *s, | |
4031 | const char *buf, size_t length) | |
4032 | { | |
4033 | if (buf[0] == '1') { | |
4034 | int rc = kmem_cache_shrink(s); | |
4035 | ||
4036 | if (rc) | |
4037 | return rc; | |
4038 | } else | |
4039 | return -EINVAL; | |
4040 | return length; | |
4041 | } | |
4042 | SLAB_ATTR(shrink); | |
4043 | ||
88a420e4 CL |
4044 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4045 | { | |
4046 | if (!(s->flags & SLAB_STORE_USER)) | |
4047 | return -ENOSYS; | |
4048 | return list_locations(s, buf, TRACK_ALLOC); | |
4049 | } | |
4050 | SLAB_ATTR_RO(alloc_calls); | |
4051 | ||
4052 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4053 | { | |
4054 | if (!(s->flags & SLAB_STORE_USER)) | |
4055 | return -ENOSYS; | |
4056 | return list_locations(s, buf, TRACK_FREE); | |
4057 | } | |
4058 | SLAB_ATTR_RO(free_calls); | |
4059 | ||
81819f0f | 4060 | #ifdef CONFIG_NUMA |
9824601e | 4061 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4062 | { |
9824601e | 4063 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4064 | } |
4065 | ||
9824601e | 4066 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4067 | const char *buf, size_t length) |
4068 | { | |
4069 | int n = simple_strtoul(buf, NULL, 10); | |
4070 | ||
4071 | if (n < 100) | |
9824601e | 4072 | s->remote_node_defrag_ratio = n * 10; |
81819f0f CL |
4073 | return length; |
4074 | } | |
9824601e | 4075 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4076 | #endif |
4077 | ||
8ff12cfc | 4078 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4079 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4080 | { | |
4081 | unsigned long sum = 0; | |
4082 | int cpu; | |
4083 | int len; | |
4084 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4085 | ||
4086 | if (!data) | |
4087 | return -ENOMEM; | |
4088 | ||
4089 | for_each_online_cpu(cpu) { | |
4090 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4091 | ||
4092 | data[cpu] = x; | |
4093 | sum += x; | |
4094 | } | |
4095 | ||
4096 | len = sprintf(buf, "%lu", sum); | |
4097 | ||
50ef37b9 | 4098 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4099 | for_each_online_cpu(cpu) { |
4100 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4101 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4102 | } |
50ef37b9 | 4103 | #endif |
8ff12cfc CL |
4104 | kfree(data); |
4105 | return len + sprintf(buf + len, "\n"); | |
4106 | } | |
4107 | ||
4108 | #define STAT_ATTR(si, text) \ | |
4109 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4110 | { \ | |
4111 | return show_stat(s, buf, si); \ | |
4112 | } \ | |
4113 | SLAB_ATTR_RO(text); \ | |
4114 | ||
4115 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4116 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4117 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4118 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4119 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4120 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4121 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4122 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4123 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4124 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4125 | STAT_ATTR(FREE_SLAB, free_slab); | |
4126 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4127 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4128 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4129 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4130 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4131 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4132 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4133 | #endif |
4134 | ||
06428780 | 4135 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4136 | &slab_size_attr.attr, |
4137 | &object_size_attr.attr, | |
4138 | &objs_per_slab_attr.attr, | |
4139 | &order_attr.attr, | |
4140 | &objects_attr.attr, | |
205ab99d CL |
4141 | &objects_partial_attr.attr, |
4142 | &total_objects_attr.attr, | |
81819f0f CL |
4143 | &slabs_attr.attr, |
4144 | &partial_attr.attr, | |
4145 | &cpu_slabs_attr.attr, | |
4146 | &ctor_attr.attr, | |
81819f0f CL |
4147 | &aliases_attr.attr, |
4148 | &align_attr.attr, | |
4149 | &sanity_checks_attr.attr, | |
4150 | &trace_attr.attr, | |
4151 | &hwcache_align_attr.attr, | |
4152 | &reclaim_account_attr.attr, | |
4153 | &destroy_by_rcu_attr.attr, | |
4154 | &red_zone_attr.attr, | |
4155 | &poison_attr.attr, | |
4156 | &store_user_attr.attr, | |
53e15af0 | 4157 | &validate_attr.attr, |
2086d26a | 4158 | &shrink_attr.attr, |
88a420e4 CL |
4159 | &alloc_calls_attr.attr, |
4160 | &free_calls_attr.attr, | |
81819f0f CL |
4161 | #ifdef CONFIG_ZONE_DMA |
4162 | &cache_dma_attr.attr, | |
4163 | #endif | |
4164 | #ifdef CONFIG_NUMA | |
9824601e | 4165 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4166 | #endif |
4167 | #ifdef CONFIG_SLUB_STATS | |
4168 | &alloc_fastpath_attr.attr, | |
4169 | &alloc_slowpath_attr.attr, | |
4170 | &free_fastpath_attr.attr, | |
4171 | &free_slowpath_attr.attr, | |
4172 | &free_frozen_attr.attr, | |
4173 | &free_add_partial_attr.attr, | |
4174 | &free_remove_partial_attr.attr, | |
4175 | &alloc_from_partial_attr.attr, | |
4176 | &alloc_slab_attr.attr, | |
4177 | &alloc_refill_attr.attr, | |
4178 | &free_slab_attr.attr, | |
4179 | &cpuslab_flush_attr.attr, | |
4180 | &deactivate_full_attr.attr, | |
4181 | &deactivate_empty_attr.attr, | |
4182 | &deactivate_to_head_attr.attr, | |
4183 | &deactivate_to_tail_attr.attr, | |
4184 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4185 | &order_fallback_attr.attr, |
81819f0f CL |
4186 | #endif |
4187 | NULL | |
4188 | }; | |
4189 | ||
4190 | static struct attribute_group slab_attr_group = { | |
4191 | .attrs = slab_attrs, | |
4192 | }; | |
4193 | ||
4194 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4195 | struct attribute *attr, | |
4196 | char *buf) | |
4197 | { | |
4198 | struct slab_attribute *attribute; | |
4199 | struct kmem_cache *s; | |
4200 | int err; | |
4201 | ||
4202 | attribute = to_slab_attr(attr); | |
4203 | s = to_slab(kobj); | |
4204 | ||
4205 | if (!attribute->show) | |
4206 | return -EIO; | |
4207 | ||
4208 | err = attribute->show(s, buf); | |
4209 | ||
4210 | return err; | |
4211 | } | |
4212 | ||
4213 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4214 | struct attribute *attr, | |
4215 | const char *buf, size_t len) | |
4216 | { | |
4217 | struct slab_attribute *attribute; | |
4218 | struct kmem_cache *s; | |
4219 | int err; | |
4220 | ||
4221 | attribute = to_slab_attr(attr); | |
4222 | s = to_slab(kobj); | |
4223 | ||
4224 | if (!attribute->store) | |
4225 | return -EIO; | |
4226 | ||
4227 | err = attribute->store(s, buf, len); | |
4228 | ||
4229 | return err; | |
4230 | } | |
4231 | ||
151c602f CL |
4232 | static void kmem_cache_release(struct kobject *kobj) |
4233 | { | |
4234 | struct kmem_cache *s = to_slab(kobj); | |
4235 | ||
4236 | kfree(s); | |
4237 | } | |
4238 | ||
81819f0f CL |
4239 | static struct sysfs_ops slab_sysfs_ops = { |
4240 | .show = slab_attr_show, | |
4241 | .store = slab_attr_store, | |
4242 | }; | |
4243 | ||
4244 | static struct kobj_type slab_ktype = { | |
4245 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4246 | .release = kmem_cache_release |
81819f0f CL |
4247 | }; |
4248 | ||
4249 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4250 | { | |
4251 | struct kobj_type *ktype = get_ktype(kobj); | |
4252 | ||
4253 | if (ktype == &slab_ktype) | |
4254 | return 1; | |
4255 | return 0; | |
4256 | } | |
4257 | ||
4258 | static struct kset_uevent_ops slab_uevent_ops = { | |
4259 | .filter = uevent_filter, | |
4260 | }; | |
4261 | ||
27c3a314 | 4262 | static struct kset *slab_kset; |
81819f0f CL |
4263 | |
4264 | #define ID_STR_LENGTH 64 | |
4265 | ||
4266 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4267 | * |
4268 | * Format :[flags-]size | |
81819f0f CL |
4269 | */ |
4270 | static char *create_unique_id(struct kmem_cache *s) | |
4271 | { | |
4272 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4273 | char *p = name; | |
4274 | ||
4275 | BUG_ON(!name); | |
4276 | ||
4277 | *p++ = ':'; | |
4278 | /* | |
4279 | * First flags affecting slabcache operations. We will only | |
4280 | * get here for aliasable slabs so we do not need to support | |
4281 | * too many flags. The flags here must cover all flags that | |
4282 | * are matched during merging to guarantee that the id is | |
4283 | * unique. | |
4284 | */ | |
4285 | if (s->flags & SLAB_CACHE_DMA) | |
4286 | *p++ = 'd'; | |
4287 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4288 | *p++ = 'a'; | |
4289 | if (s->flags & SLAB_DEBUG_FREE) | |
4290 | *p++ = 'F'; | |
4291 | if (p != name + 1) | |
4292 | *p++ = '-'; | |
4293 | p += sprintf(p, "%07d", s->size); | |
4294 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4295 | return name; | |
4296 | } | |
4297 | ||
4298 | static int sysfs_slab_add(struct kmem_cache *s) | |
4299 | { | |
4300 | int err; | |
4301 | const char *name; | |
4302 | int unmergeable; | |
4303 | ||
4304 | if (slab_state < SYSFS) | |
4305 | /* Defer until later */ | |
4306 | return 0; | |
4307 | ||
4308 | unmergeable = slab_unmergeable(s); | |
4309 | if (unmergeable) { | |
4310 | /* | |
4311 | * Slabcache can never be merged so we can use the name proper. | |
4312 | * This is typically the case for debug situations. In that | |
4313 | * case we can catch duplicate names easily. | |
4314 | */ | |
27c3a314 | 4315 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4316 | name = s->name; |
4317 | } else { | |
4318 | /* | |
4319 | * Create a unique name for the slab as a target | |
4320 | * for the symlinks. | |
4321 | */ | |
4322 | name = create_unique_id(s); | |
4323 | } | |
4324 | ||
27c3a314 | 4325 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4326 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4327 | if (err) { | |
4328 | kobject_put(&s->kobj); | |
81819f0f | 4329 | return err; |
1eada11c | 4330 | } |
81819f0f CL |
4331 | |
4332 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4333 | if (err) | |
4334 | return err; | |
4335 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4336 | if (!unmergeable) { | |
4337 | /* Setup first alias */ | |
4338 | sysfs_slab_alias(s, s->name); | |
4339 | kfree(name); | |
4340 | } | |
4341 | return 0; | |
4342 | } | |
4343 | ||
4344 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4345 | { | |
4346 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4347 | kobject_del(&s->kobj); | |
151c602f | 4348 | kobject_put(&s->kobj); |
81819f0f CL |
4349 | } |
4350 | ||
4351 | /* | |
4352 | * Need to buffer aliases during bootup until sysfs becomes | |
4353 | * available lest we loose that information. | |
4354 | */ | |
4355 | struct saved_alias { | |
4356 | struct kmem_cache *s; | |
4357 | const char *name; | |
4358 | struct saved_alias *next; | |
4359 | }; | |
4360 | ||
5af328a5 | 4361 | static struct saved_alias *alias_list; |
81819f0f CL |
4362 | |
4363 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4364 | { | |
4365 | struct saved_alias *al; | |
4366 | ||
4367 | if (slab_state == SYSFS) { | |
4368 | /* | |
4369 | * If we have a leftover link then remove it. | |
4370 | */ | |
27c3a314 GKH |
4371 | sysfs_remove_link(&slab_kset->kobj, name); |
4372 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4373 | } |
4374 | ||
4375 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4376 | if (!al) | |
4377 | return -ENOMEM; | |
4378 | ||
4379 | al->s = s; | |
4380 | al->name = name; | |
4381 | al->next = alias_list; | |
4382 | alias_list = al; | |
4383 | return 0; | |
4384 | } | |
4385 | ||
4386 | static int __init slab_sysfs_init(void) | |
4387 | { | |
5b95a4ac | 4388 | struct kmem_cache *s; |
81819f0f CL |
4389 | int err; |
4390 | ||
0ff21e46 | 4391 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4392 | if (!slab_kset) { |
81819f0f CL |
4393 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4394 | return -ENOSYS; | |
4395 | } | |
4396 | ||
26a7bd03 CL |
4397 | slab_state = SYSFS; |
4398 | ||
5b95a4ac | 4399 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4400 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4401 | if (err) |
4402 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4403 | " to sysfs\n", s->name); | |
26a7bd03 | 4404 | } |
81819f0f CL |
4405 | |
4406 | while (alias_list) { | |
4407 | struct saved_alias *al = alias_list; | |
4408 | ||
4409 | alias_list = alias_list->next; | |
4410 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4411 | if (err) |
4412 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4413 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4414 | kfree(al); |
4415 | } | |
4416 | ||
4417 | resiliency_test(); | |
4418 | return 0; | |
4419 | } | |
4420 | ||
4421 | __initcall(slab_sysfs_init); | |
81819f0f | 4422 | #endif |
57ed3eda PE |
4423 | |
4424 | /* | |
4425 | * The /proc/slabinfo ABI | |
4426 | */ | |
158a9624 LT |
4427 | #ifdef CONFIG_SLABINFO |
4428 | ||
4429 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |
4430 | size_t count, loff_t *ppos) | |
4431 | { | |
4432 | return -EINVAL; | |
4433 | } | |
4434 | ||
57ed3eda PE |
4435 | |
4436 | static void print_slabinfo_header(struct seq_file *m) | |
4437 | { | |
4438 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4439 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4440 | "<objperslab> <pagesperslab>"); | |
4441 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4442 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4443 | seq_putc(m, '\n'); | |
4444 | } | |
4445 | ||
4446 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4447 | { | |
4448 | loff_t n = *pos; | |
4449 | ||
4450 | down_read(&slub_lock); | |
4451 | if (!n) | |
4452 | print_slabinfo_header(m); | |
4453 | ||
4454 | return seq_list_start(&slab_caches, *pos); | |
4455 | } | |
4456 | ||
4457 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4458 | { | |
4459 | return seq_list_next(p, &slab_caches, pos); | |
4460 | } | |
4461 | ||
4462 | static void s_stop(struct seq_file *m, void *p) | |
4463 | { | |
4464 | up_read(&slub_lock); | |
4465 | } | |
4466 | ||
4467 | static int s_show(struct seq_file *m, void *p) | |
4468 | { | |
4469 | unsigned long nr_partials = 0; | |
4470 | unsigned long nr_slabs = 0; | |
4471 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4472 | unsigned long nr_objs = 0; |
4473 | unsigned long nr_free = 0; | |
57ed3eda PE |
4474 | struct kmem_cache *s; |
4475 | int node; | |
4476 | ||
4477 | s = list_entry(p, struct kmem_cache, list); | |
4478 | ||
4479 | for_each_online_node(node) { | |
4480 | struct kmem_cache_node *n = get_node(s, node); | |
4481 | ||
4482 | if (!n) | |
4483 | continue; | |
4484 | ||
4485 | nr_partials += n->nr_partial; | |
4486 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4487 | nr_objs += atomic_long_read(&n->total_objects); |
4488 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4489 | } |
4490 | ||
205ab99d | 4491 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4492 | |
4493 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4494 | nr_objs, s->size, oo_objects(s->oo), |
4495 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4496 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4497 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4498 | 0UL); | |
4499 | seq_putc(m, '\n'); | |
4500 | return 0; | |
4501 | } | |
4502 | ||
4503 | const struct seq_operations slabinfo_op = { | |
4504 | .start = s_start, | |
4505 | .next = s_next, | |
4506 | .stop = s_stop, | |
4507 | .show = s_show, | |
4508 | }; | |
4509 | ||
158a9624 | 4510 | #endif /* CONFIG_SLABINFO */ |