]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
8 | * (C) 2007 SGI, Christoph Lameter <clameter@sgi.com> | |
9 | */ | |
10 | ||
11 | #include <linux/mm.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/bit_spinlock.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/bitops.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/seq_file.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/cpuset.h> | |
20 | #include <linux/mempolicy.h> | |
21 | #include <linux/ctype.h> | |
22 | #include <linux/kallsyms.h> | |
b9049e23 | 23 | #include <linux/memory.h> |
81819f0f CL |
24 | |
25 | /* | |
26 | * Lock order: | |
27 | * 1. slab_lock(page) | |
28 | * 2. slab->list_lock | |
29 | * | |
30 | * The slab_lock protects operations on the object of a particular | |
31 | * slab and its metadata in the page struct. If the slab lock | |
32 | * has been taken then no allocations nor frees can be performed | |
33 | * on the objects in the slab nor can the slab be added or removed | |
34 | * from the partial or full lists since this would mean modifying | |
35 | * the page_struct of the slab. | |
36 | * | |
37 | * The list_lock protects the partial and full list on each node and | |
38 | * the partial slab counter. If taken then no new slabs may be added or | |
39 | * removed from the lists nor make the number of partial slabs be modified. | |
40 | * (Note that the total number of slabs is an atomic value that may be | |
41 | * modified without taking the list lock). | |
42 | * | |
43 | * The list_lock is a centralized lock and thus we avoid taking it as | |
44 | * much as possible. As long as SLUB does not have to handle partial | |
45 | * slabs, operations can continue without any centralized lock. F.e. | |
46 | * allocating a long series of objects that fill up slabs does not require | |
47 | * the list lock. | |
48 | * | |
49 | * The lock order is sometimes inverted when we are trying to get a slab | |
50 | * off a list. We take the list_lock and then look for a page on the list | |
51 | * to use. While we do that objects in the slabs may be freed. We can | |
52 | * only operate on the slab if we have also taken the slab_lock. So we use | |
53 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
54 | * can occur anymore and we can use the slab for allocations etc. If the | |
55 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
56 | * we must stay away from it for a while since we may cause a bouncing | |
57 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
58 | * If all pages are busy then we may allocate a new slab instead of reusing | |
59 | * a partial slab. A new slab has noone operating on it and thus there is | |
60 | * no danger of cacheline contention. | |
61 | * | |
62 | * Interrupts are disabled during allocation and deallocation in order to | |
63 | * make the slab allocator safe to use in the context of an irq. In addition | |
64 | * interrupts are disabled to ensure that the processor does not change | |
65 | * while handling per_cpu slabs, due to kernel preemption. | |
66 | * | |
67 | * SLUB assigns one slab for allocation to each processor. | |
68 | * Allocations only occur from these slabs called cpu slabs. | |
69 | * | |
672bba3a CL |
70 | * Slabs with free elements are kept on a partial list and during regular |
71 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 72 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
73 | * We track full slabs for debugging purposes though because otherwise we |
74 | * cannot scan all objects. | |
81819f0f CL |
75 | * |
76 | * Slabs are freed when they become empty. Teardown and setup is | |
77 | * minimal so we rely on the page allocators per cpu caches for | |
78 | * fast frees and allocs. | |
79 | * | |
80 | * Overloading of page flags that are otherwise used for LRU management. | |
81 | * | |
4b6f0750 CL |
82 | * PageActive The slab is frozen and exempt from list processing. |
83 | * This means that the slab is dedicated to a purpose | |
84 | * such as satisfying allocations for a specific | |
85 | * processor. Objects may be freed in the slab while | |
86 | * it is frozen but slab_free will then skip the usual | |
87 | * list operations. It is up to the processor holding | |
88 | * the slab to integrate the slab into the slab lists | |
89 | * when the slab is no longer needed. | |
90 | * | |
91 | * One use of this flag is to mark slabs that are | |
92 | * used for allocations. Then such a slab becomes a cpu | |
93 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 94 | * freelist that allows lockless access to |
894b8788 CL |
95 | * free objects in addition to the regular freelist |
96 | * that requires the slab lock. | |
81819f0f CL |
97 | * |
98 | * PageError Slab requires special handling due to debug | |
99 | * options set. This moves slab handling out of | |
894b8788 | 100 | * the fast path and disables lockless freelists. |
81819f0f CL |
101 | */ |
102 | ||
5577bd8a CL |
103 | #define FROZEN (1 << PG_active) |
104 | ||
105 | #ifdef CONFIG_SLUB_DEBUG | |
106 | #define SLABDEBUG (1 << PG_error) | |
107 | #else | |
108 | #define SLABDEBUG 0 | |
109 | #endif | |
110 | ||
4b6f0750 CL |
111 | static inline int SlabFrozen(struct page *page) |
112 | { | |
5577bd8a | 113 | return page->flags & FROZEN; |
4b6f0750 CL |
114 | } |
115 | ||
116 | static inline void SetSlabFrozen(struct page *page) | |
117 | { | |
5577bd8a | 118 | page->flags |= FROZEN; |
4b6f0750 CL |
119 | } |
120 | ||
121 | static inline void ClearSlabFrozen(struct page *page) | |
122 | { | |
5577bd8a | 123 | page->flags &= ~FROZEN; |
4b6f0750 CL |
124 | } |
125 | ||
35e5d7ee CL |
126 | static inline int SlabDebug(struct page *page) |
127 | { | |
5577bd8a | 128 | return page->flags & SLABDEBUG; |
35e5d7ee CL |
129 | } |
130 | ||
131 | static inline void SetSlabDebug(struct page *page) | |
132 | { | |
5577bd8a | 133 | page->flags |= SLABDEBUG; |
35e5d7ee CL |
134 | } |
135 | ||
136 | static inline void ClearSlabDebug(struct page *page) | |
137 | { | |
5577bd8a | 138 | page->flags &= ~SLABDEBUG; |
35e5d7ee CL |
139 | } |
140 | ||
81819f0f CL |
141 | /* |
142 | * Issues still to be resolved: | |
143 | * | |
81819f0f CL |
144 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
145 | * | |
81819f0f CL |
146 | * - Variable sizing of the per node arrays |
147 | */ | |
148 | ||
149 | /* Enable to test recovery from slab corruption on boot */ | |
150 | #undef SLUB_RESILIENCY_TEST | |
151 | ||
152 | #if PAGE_SHIFT <= 12 | |
153 | ||
154 | /* | |
155 | * Small page size. Make sure that we do not fragment memory | |
156 | */ | |
157 | #define DEFAULT_MAX_ORDER 1 | |
158 | #define DEFAULT_MIN_OBJECTS 4 | |
159 | ||
160 | #else | |
161 | ||
162 | /* | |
163 | * Large page machines are customarily able to handle larger | |
164 | * page orders. | |
165 | */ | |
166 | #define DEFAULT_MAX_ORDER 2 | |
167 | #define DEFAULT_MIN_OBJECTS 8 | |
168 | ||
169 | #endif | |
170 | ||
2086d26a CL |
171 | /* |
172 | * Mininum number of partial slabs. These will be left on the partial | |
173 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
174 | */ | |
76be8950 | 175 | #define MIN_PARTIAL 5 |
e95eed57 | 176 | |
2086d26a CL |
177 | /* |
178 | * Maximum number of desirable partial slabs. | |
179 | * The existence of more partial slabs makes kmem_cache_shrink | |
180 | * sort the partial list by the number of objects in the. | |
181 | */ | |
182 | #define MAX_PARTIAL 10 | |
183 | ||
81819f0f CL |
184 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
185 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 186 | |
81819f0f CL |
187 | /* |
188 | * Set of flags that will prevent slab merging | |
189 | */ | |
190 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
191 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
192 | ||
193 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
194 | SLAB_CACHE_DMA) | |
195 | ||
196 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 197 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
198 | #endif |
199 | ||
200 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 201 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
202 | #endif |
203 | ||
204 | /* Internal SLUB flags */ | |
1ceef402 CL |
205 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
206 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
71c7a06f CL |
207 | #define __KMALLOC_CACHE 0x20000000 /* objects freed using kfree */ |
208 | #define __PAGE_ALLOC_FALLBACK 0x10000000 /* Allow fallback to page alloc */ | |
81819f0f | 209 | |
65c02d4c CL |
210 | /* Not all arches define cache_line_size */ |
211 | #ifndef cache_line_size | |
212 | #define cache_line_size() L1_CACHE_BYTES | |
213 | #endif | |
214 | ||
81819f0f CL |
215 | static int kmem_size = sizeof(struct kmem_cache); |
216 | ||
217 | #ifdef CONFIG_SMP | |
218 | static struct notifier_block slab_notifier; | |
219 | #endif | |
220 | ||
221 | static enum { | |
222 | DOWN, /* No slab functionality available */ | |
223 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 224 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
225 | SYSFS /* Sysfs up */ |
226 | } slab_state = DOWN; | |
227 | ||
228 | /* A list of all slab caches on the system */ | |
229 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 230 | static LIST_HEAD(slab_caches); |
81819f0f | 231 | |
02cbc874 CL |
232 | /* |
233 | * Tracking user of a slab. | |
234 | */ | |
235 | struct track { | |
236 | void *addr; /* Called from address */ | |
237 | int cpu; /* Was running on cpu */ | |
238 | int pid; /* Pid context */ | |
239 | unsigned long when; /* When did the operation occur */ | |
240 | }; | |
241 | ||
242 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
243 | ||
41ecc55b | 244 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
81819f0f CL |
245 | static int sysfs_slab_add(struct kmem_cache *); |
246 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
247 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 248 | |
81819f0f | 249 | #else |
0c710013 CL |
250 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
251 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
252 | { return 0; } | |
151c602f CL |
253 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
254 | { | |
255 | kfree(s); | |
256 | } | |
8ff12cfc | 257 | |
81819f0f CL |
258 | #endif |
259 | ||
8ff12cfc CL |
260 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
261 | { | |
262 | #ifdef CONFIG_SLUB_STATS | |
263 | c->stat[si]++; | |
264 | #endif | |
265 | } | |
266 | ||
81819f0f CL |
267 | /******************************************************************** |
268 | * Core slab cache functions | |
269 | *******************************************************************/ | |
270 | ||
271 | int slab_is_available(void) | |
272 | { | |
273 | return slab_state >= UP; | |
274 | } | |
275 | ||
276 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
277 | { | |
278 | #ifdef CONFIG_NUMA | |
279 | return s->node[node]; | |
280 | #else | |
281 | return &s->local_node; | |
282 | #endif | |
283 | } | |
284 | ||
dfb4f096 CL |
285 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
286 | { | |
4c93c355 CL |
287 | #ifdef CONFIG_SMP |
288 | return s->cpu_slab[cpu]; | |
289 | #else | |
290 | return &s->cpu_slab; | |
291 | #endif | |
dfb4f096 CL |
292 | } |
293 | ||
6446faa2 | 294 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
295 | static inline int check_valid_pointer(struct kmem_cache *s, |
296 | struct page *page, const void *object) | |
297 | { | |
298 | void *base; | |
299 | ||
a973e9dd | 300 | if (!object) |
02cbc874 CL |
301 | return 1; |
302 | ||
a973e9dd | 303 | base = page_address(page); |
39b26464 | 304 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
305 | (object - base) % s->size) { |
306 | return 0; | |
307 | } | |
308 | ||
309 | return 1; | |
310 | } | |
311 | ||
7656c72b CL |
312 | /* |
313 | * Slow version of get and set free pointer. | |
314 | * | |
315 | * This version requires touching the cache lines of kmem_cache which | |
316 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
317 | * from the page struct. | |
318 | */ | |
319 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
320 | { | |
321 | return *(void **)(object + s->offset); | |
322 | } | |
323 | ||
324 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
325 | { | |
326 | *(void **)(object + s->offset) = fp; | |
327 | } | |
328 | ||
329 | /* Loop over all objects in a slab */ | |
224a88be CL |
330 | #define for_each_object(__p, __s, __addr, __objects) \ |
331 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
332 | __p += (__s)->size) |
333 | ||
334 | /* Scan freelist */ | |
335 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 336 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
337 | |
338 | /* Determine object index from a given position */ | |
339 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
340 | { | |
341 | return (p - addr) / s->size; | |
342 | } | |
343 | ||
834f3d11 CL |
344 | static inline struct kmem_cache_order_objects oo_make(int order, |
345 | unsigned long size) | |
346 | { | |
347 | struct kmem_cache_order_objects x = { | |
348 | (order << 16) + (PAGE_SIZE << order) / size | |
349 | }; | |
350 | ||
351 | return x; | |
352 | } | |
353 | ||
354 | static inline int oo_order(struct kmem_cache_order_objects x) | |
355 | { | |
356 | return x.x >> 16; | |
357 | } | |
358 | ||
359 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
360 | { | |
361 | return x.x & ((1 << 16) - 1); | |
362 | } | |
363 | ||
41ecc55b CL |
364 | #ifdef CONFIG_SLUB_DEBUG |
365 | /* | |
366 | * Debug settings: | |
367 | */ | |
f0630fff CL |
368 | #ifdef CONFIG_SLUB_DEBUG_ON |
369 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
370 | #else | |
41ecc55b | 371 | static int slub_debug; |
f0630fff | 372 | #endif |
41ecc55b CL |
373 | |
374 | static char *slub_debug_slabs; | |
375 | ||
81819f0f CL |
376 | /* |
377 | * Object debugging | |
378 | */ | |
379 | static void print_section(char *text, u8 *addr, unsigned int length) | |
380 | { | |
381 | int i, offset; | |
382 | int newline = 1; | |
383 | char ascii[17]; | |
384 | ||
385 | ascii[16] = 0; | |
386 | ||
387 | for (i = 0; i < length; i++) { | |
388 | if (newline) { | |
24922684 | 389 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
390 | newline = 0; |
391 | } | |
06428780 | 392 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
393 | offset = i % 16; |
394 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
395 | if (offset == 15) { | |
06428780 | 396 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
397 | newline = 1; |
398 | } | |
399 | } | |
400 | if (!newline) { | |
401 | i %= 16; | |
402 | while (i < 16) { | |
06428780 | 403 | printk(KERN_CONT " "); |
81819f0f CL |
404 | ascii[i] = ' '; |
405 | i++; | |
406 | } | |
06428780 | 407 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
408 | } |
409 | } | |
410 | ||
81819f0f CL |
411 | static struct track *get_track(struct kmem_cache *s, void *object, |
412 | enum track_item alloc) | |
413 | { | |
414 | struct track *p; | |
415 | ||
416 | if (s->offset) | |
417 | p = object + s->offset + sizeof(void *); | |
418 | else | |
419 | p = object + s->inuse; | |
420 | ||
421 | return p + alloc; | |
422 | } | |
423 | ||
424 | static void set_track(struct kmem_cache *s, void *object, | |
425 | enum track_item alloc, void *addr) | |
426 | { | |
427 | struct track *p; | |
428 | ||
429 | if (s->offset) | |
430 | p = object + s->offset + sizeof(void *); | |
431 | else | |
432 | p = object + s->inuse; | |
433 | ||
434 | p += alloc; | |
435 | if (addr) { | |
436 | p->addr = addr; | |
437 | p->cpu = smp_processor_id(); | |
438 | p->pid = current ? current->pid : -1; | |
439 | p->when = jiffies; | |
440 | } else | |
441 | memset(p, 0, sizeof(struct track)); | |
442 | } | |
443 | ||
81819f0f CL |
444 | static void init_tracking(struct kmem_cache *s, void *object) |
445 | { | |
24922684 CL |
446 | if (!(s->flags & SLAB_STORE_USER)) |
447 | return; | |
448 | ||
449 | set_track(s, object, TRACK_FREE, NULL); | |
450 | set_track(s, object, TRACK_ALLOC, NULL); | |
81819f0f CL |
451 | } |
452 | ||
453 | static void print_track(const char *s, struct track *t) | |
454 | { | |
455 | if (!t->addr) | |
456 | return; | |
457 | ||
24922684 | 458 | printk(KERN_ERR "INFO: %s in ", s); |
81819f0f | 459 | __print_symbol("%s", (unsigned long)t->addr); |
24922684 CL |
460 | printk(" age=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); |
461 | } | |
462 | ||
463 | static void print_tracking(struct kmem_cache *s, void *object) | |
464 | { | |
465 | if (!(s->flags & SLAB_STORE_USER)) | |
466 | return; | |
467 | ||
468 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
469 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
470 | } | |
471 | ||
472 | static void print_page_info(struct page *page) | |
473 | { | |
39b26464 CL |
474 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
475 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
476 | |
477 | } | |
478 | ||
479 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
480 | { | |
481 | va_list args; | |
482 | char buf[100]; | |
483 | ||
484 | va_start(args, fmt); | |
485 | vsnprintf(buf, sizeof(buf), fmt, args); | |
486 | va_end(args); | |
487 | printk(KERN_ERR "========================================" | |
488 | "=====================================\n"); | |
489 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
490 | printk(KERN_ERR "----------------------------------------" | |
491 | "-------------------------------------\n\n"); | |
81819f0f CL |
492 | } |
493 | ||
24922684 CL |
494 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
495 | { | |
496 | va_list args; | |
497 | char buf[100]; | |
498 | ||
499 | va_start(args, fmt); | |
500 | vsnprintf(buf, sizeof(buf), fmt, args); | |
501 | va_end(args); | |
502 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
503 | } | |
504 | ||
505 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
506 | { |
507 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 508 | u8 *addr = page_address(page); |
24922684 CL |
509 | |
510 | print_tracking(s, p); | |
511 | ||
512 | print_page_info(page); | |
513 | ||
514 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
515 | p, p - addr, get_freepointer(s, p)); | |
516 | ||
517 | if (p > addr + 16) | |
518 | print_section("Bytes b4", p - 16, 16); | |
519 | ||
520 | print_section("Object", p, min(s->objsize, 128)); | |
81819f0f CL |
521 | |
522 | if (s->flags & SLAB_RED_ZONE) | |
523 | print_section("Redzone", p + s->objsize, | |
524 | s->inuse - s->objsize); | |
525 | ||
81819f0f CL |
526 | if (s->offset) |
527 | off = s->offset + sizeof(void *); | |
528 | else | |
529 | off = s->inuse; | |
530 | ||
24922684 | 531 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 532 | off += 2 * sizeof(struct track); |
81819f0f CL |
533 | |
534 | if (off != s->size) | |
535 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
536 | print_section("Padding", p + off, s->size - off); |
537 | ||
538 | dump_stack(); | |
81819f0f CL |
539 | } |
540 | ||
541 | static void object_err(struct kmem_cache *s, struct page *page, | |
542 | u8 *object, char *reason) | |
543 | { | |
3dc50637 | 544 | slab_bug(s, "%s", reason); |
24922684 | 545 | print_trailer(s, page, object); |
81819f0f CL |
546 | } |
547 | ||
24922684 | 548 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
549 | { |
550 | va_list args; | |
551 | char buf[100]; | |
552 | ||
24922684 CL |
553 | va_start(args, fmt); |
554 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 555 | va_end(args); |
3dc50637 | 556 | slab_bug(s, "%s", buf); |
24922684 | 557 | print_page_info(page); |
81819f0f CL |
558 | dump_stack(); |
559 | } | |
560 | ||
561 | static void init_object(struct kmem_cache *s, void *object, int active) | |
562 | { | |
563 | u8 *p = object; | |
564 | ||
565 | if (s->flags & __OBJECT_POISON) { | |
566 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 567 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
568 | } |
569 | ||
570 | if (s->flags & SLAB_RED_ZONE) | |
571 | memset(p + s->objsize, | |
572 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
573 | s->inuse - s->objsize); | |
574 | } | |
575 | ||
24922684 | 576 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
577 | { |
578 | while (bytes) { | |
579 | if (*start != (u8)value) | |
24922684 | 580 | return start; |
81819f0f CL |
581 | start++; |
582 | bytes--; | |
583 | } | |
24922684 CL |
584 | return NULL; |
585 | } | |
586 | ||
587 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
588 | void *from, void *to) | |
589 | { | |
590 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
591 | memset(from, data, to - from); | |
592 | } | |
593 | ||
594 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
595 | u8 *object, char *what, | |
06428780 | 596 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
597 | { |
598 | u8 *fault; | |
599 | u8 *end; | |
600 | ||
601 | fault = check_bytes(start, value, bytes); | |
602 | if (!fault) | |
603 | return 1; | |
604 | ||
605 | end = start + bytes; | |
606 | while (end > fault && end[-1] == value) | |
607 | end--; | |
608 | ||
609 | slab_bug(s, "%s overwritten", what); | |
610 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
611 | fault, end - 1, fault[0], value); | |
612 | print_trailer(s, page, object); | |
613 | ||
614 | restore_bytes(s, what, value, fault, end); | |
615 | return 0; | |
81819f0f CL |
616 | } |
617 | ||
81819f0f CL |
618 | /* |
619 | * Object layout: | |
620 | * | |
621 | * object address | |
622 | * Bytes of the object to be managed. | |
623 | * If the freepointer may overlay the object then the free | |
624 | * pointer is the first word of the object. | |
672bba3a | 625 | * |
81819f0f CL |
626 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
627 | * 0xa5 (POISON_END) | |
628 | * | |
629 | * object + s->objsize | |
630 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
631 | * Padding is extended by another word if Redzoning is enabled and |
632 | * objsize == inuse. | |
633 | * | |
81819f0f CL |
634 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
635 | * 0xcc (RED_ACTIVE) for objects in use. | |
636 | * | |
637 | * object + s->inuse | |
672bba3a CL |
638 | * Meta data starts here. |
639 | * | |
81819f0f CL |
640 | * A. Free pointer (if we cannot overwrite object on free) |
641 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 642 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 643 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
644 | * before the word boundary. |
645 | * | |
646 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
647 | * |
648 | * object + s->size | |
672bba3a | 649 | * Nothing is used beyond s->size. |
81819f0f | 650 | * |
672bba3a CL |
651 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
652 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
653 | * may be used with merged slabcaches. |
654 | */ | |
655 | ||
81819f0f CL |
656 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
657 | { | |
658 | unsigned long off = s->inuse; /* The end of info */ | |
659 | ||
660 | if (s->offset) | |
661 | /* Freepointer is placed after the object. */ | |
662 | off += sizeof(void *); | |
663 | ||
664 | if (s->flags & SLAB_STORE_USER) | |
665 | /* We also have user information there */ | |
666 | off += 2 * sizeof(struct track); | |
667 | ||
668 | if (s->size == off) | |
669 | return 1; | |
670 | ||
24922684 CL |
671 | return check_bytes_and_report(s, page, p, "Object padding", |
672 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
673 | } |
674 | ||
39b26464 | 675 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
676 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
677 | { | |
24922684 CL |
678 | u8 *start; |
679 | u8 *fault; | |
680 | u8 *end; | |
681 | int length; | |
682 | int remainder; | |
81819f0f CL |
683 | |
684 | if (!(s->flags & SLAB_POISON)) | |
685 | return 1; | |
686 | ||
a973e9dd | 687 | start = page_address(page); |
834f3d11 | 688 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
689 | end = start + length; |
690 | remainder = length % s->size; | |
81819f0f CL |
691 | if (!remainder) |
692 | return 1; | |
693 | ||
39b26464 | 694 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
695 | if (!fault) |
696 | return 1; | |
697 | while (end > fault && end[-1] == POISON_INUSE) | |
698 | end--; | |
699 | ||
700 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 701 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
702 | |
703 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
704 | return 0; | |
81819f0f CL |
705 | } |
706 | ||
707 | static int check_object(struct kmem_cache *s, struct page *page, | |
708 | void *object, int active) | |
709 | { | |
710 | u8 *p = object; | |
711 | u8 *endobject = object + s->objsize; | |
712 | ||
713 | if (s->flags & SLAB_RED_ZONE) { | |
714 | unsigned int red = | |
715 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
716 | ||
24922684 CL |
717 | if (!check_bytes_and_report(s, page, object, "Redzone", |
718 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 719 | return 0; |
81819f0f | 720 | } else { |
3adbefee IM |
721 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
722 | check_bytes_and_report(s, page, p, "Alignment padding", | |
723 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
724 | } | |
81819f0f CL |
725 | } |
726 | ||
727 | if (s->flags & SLAB_POISON) { | |
728 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
729 | (!check_bytes_and_report(s, page, p, "Poison", p, |
730 | POISON_FREE, s->objsize - 1) || | |
731 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 732 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 733 | return 0; |
81819f0f CL |
734 | /* |
735 | * check_pad_bytes cleans up on its own. | |
736 | */ | |
737 | check_pad_bytes(s, page, p); | |
738 | } | |
739 | ||
740 | if (!s->offset && active) | |
741 | /* | |
742 | * Object and freepointer overlap. Cannot check | |
743 | * freepointer while object is allocated. | |
744 | */ | |
745 | return 1; | |
746 | ||
747 | /* Check free pointer validity */ | |
748 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
749 | object_err(s, page, p, "Freepointer corrupt"); | |
750 | /* | |
751 | * No choice but to zap it and thus loose the remainder | |
752 | * of the free objects in this slab. May cause | |
672bba3a | 753 | * another error because the object count is now wrong. |
81819f0f | 754 | */ |
a973e9dd | 755 | set_freepointer(s, p, NULL); |
81819f0f CL |
756 | return 0; |
757 | } | |
758 | return 1; | |
759 | } | |
760 | ||
761 | static int check_slab(struct kmem_cache *s, struct page *page) | |
762 | { | |
39b26464 CL |
763 | int maxobj; |
764 | ||
81819f0f CL |
765 | VM_BUG_ON(!irqs_disabled()); |
766 | ||
767 | if (!PageSlab(page)) { | |
24922684 | 768 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
769 | return 0; |
770 | } | |
39b26464 CL |
771 | |
772 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
773 | if (page->objects > maxobj) { | |
774 | slab_err(s, page, "objects %u > max %u", | |
775 | s->name, page->objects, maxobj); | |
776 | return 0; | |
777 | } | |
778 | if (page->inuse > page->objects) { | |
24922684 | 779 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 780 | s->name, page->inuse, page->objects); |
81819f0f CL |
781 | return 0; |
782 | } | |
783 | /* Slab_pad_check fixes things up after itself */ | |
784 | slab_pad_check(s, page); | |
785 | return 1; | |
786 | } | |
787 | ||
788 | /* | |
672bba3a CL |
789 | * Determine if a certain object on a page is on the freelist. Must hold the |
790 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
791 | */ |
792 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
793 | { | |
794 | int nr = 0; | |
795 | void *fp = page->freelist; | |
796 | void *object = NULL; | |
224a88be | 797 | unsigned long max_objects; |
81819f0f | 798 | |
39b26464 | 799 | while (fp && nr <= page->objects) { |
81819f0f CL |
800 | if (fp == search) |
801 | return 1; | |
802 | if (!check_valid_pointer(s, page, fp)) { | |
803 | if (object) { | |
804 | object_err(s, page, object, | |
805 | "Freechain corrupt"); | |
a973e9dd | 806 | set_freepointer(s, object, NULL); |
81819f0f CL |
807 | break; |
808 | } else { | |
24922684 | 809 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 810 | page->freelist = NULL; |
39b26464 | 811 | page->inuse = page->objects; |
24922684 | 812 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
813 | return 0; |
814 | } | |
815 | break; | |
816 | } | |
817 | object = fp; | |
818 | fp = get_freepointer(s, object); | |
819 | nr++; | |
820 | } | |
821 | ||
224a88be CL |
822 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
823 | if (max_objects > 65535) | |
824 | max_objects = 65535; | |
825 | ||
826 | if (page->objects != max_objects) { | |
827 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
828 | "should be %d", page->objects, max_objects); | |
829 | page->objects = max_objects; | |
830 | slab_fix(s, "Number of objects adjusted."); | |
831 | } | |
39b26464 | 832 | if (page->inuse != page->objects - nr) { |
70d71228 | 833 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
834 | "counted were %d", page->inuse, page->objects - nr); |
835 | page->inuse = page->objects - nr; | |
24922684 | 836 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
837 | } |
838 | return search == NULL; | |
839 | } | |
840 | ||
3ec09742 CL |
841 | static void trace(struct kmem_cache *s, struct page *page, void *object, int alloc) |
842 | { | |
843 | if (s->flags & SLAB_TRACE) { | |
844 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
845 | s->name, | |
846 | alloc ? "alloc" : "free", | |
847 | object, page->inuse, | |
848 | page->freelist); | |
849 | ||
850 | if (!alloc) | |
851 | print_section("Object", (void *)object, s->objsize); | |
852 | ||
853 | dump_stack(); | |
854 | } | |
855 | } | |
856 | ||
643b1138 | 857 | /* |
672bba3a | 858 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 859 | */ |
e95eed57 | 860 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 861 | { |
643b1138 CL |
862 | spin_lock(&n->list_lock); |
863 | list_add(&page->lru, &n->full); | |
864 | spin_unlock(&n->list_lock); | |
865 | } | |
866 | ||
867 | static void remove_full(struct kmem_cache *s, struct page *page) | |
868 | { | |
869 | struct kmem_cache_node *n; | |
870 | ||
871 | if (!(s->flags & SLAB_STORE_USER)) | |
872 | return; | |
873 | ||
874 | n = get_node(s, page_to_nid(page)); | |
875 | ||
876 | spin_lock(&n->list_lock); | |
877 | list_del(&page->lru); | |
878 | spin_unlock(&n->list_lock); | |
879 | } | |
880 | ||
0f389ec6 CL |
881 | /* Tracking of the number of slabs for debugging purposes */ |
882 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
883 | { | |
884 | struct kmem_cache_node *n = get_node(s, node); | |
885 | ||
886 | return atomic_long_read(&n->nr_slabs); | |
887 | } | |
888 | ||
205ab99d | 889 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
890 | { |
891 | struct kmem_cache_node *n = get_node(s, node); | |
892 | ||
893 | /* | |
894 | * May be called early in order to allocate a slab for the | |
895 | * kmem_cache_node structure. Solve the chicken-egg | |
896 | * dilemma by deferring the increment of the count during | |
897 | * bootstrap (see early_kmem_cache_node_alloc). | |
898 | */ | |
205ab99d | 899 | if (!NUMA_BUILD || n) { |
0f389ec6 | 900 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
901 | atomic_long_add(objects, &n->total_objects); |
902 | } | |
0f389ec6 | 903 | } |
205ab99d | 904 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
905 | { |
906 | struct kmem_cache_node *n = get_node(s, node); | |
907 | ||
908 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 909 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
910 | } |
911 | ||
912 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
913 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
914 | void *object) | |
915 | { | |
916 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
917 | return; | |
918 | ||
919 | init_object(s, object, 0); | |
920 | init_tracking(s, object); | |
921 | } | |
922 | ||
923 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
924 | void *object, void *addr) | |
81819f0f CL |
925 | { |
926 | if (!check_slab(s, page)) | |
927 | goto bad; | |
928 | ||
d692ef6d | 929 | if (!on_freelist(s, page, object)) { |
24922684 | 930 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 931 | goto bad; |
81819f0f CL |
932 | } |
933 | ||
934 | if (!check_valid_pointer(s, page, object)) { | |
935 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 936 | goto bad; |
81819f0f CL |
937 | } |
938 | ||
d692ef6d | 939 | if (!check_object(s, page, object, 0)) |
81819f0f | 940 | goto bad; |
81819f0f | 941 | |
3ec09742 CL |
942 | /* Success perform special debug activities for allocs */ |
943 | if (s->flags & SLAB_STORE_USER) | |
944 | set_track(s, object, TRACK_ALLOC, addr); | |
945 | trace(s, page, object, 1); | |
946 | init_object(s, object, 1); | |
81819f0f | 947 | return 1; |
3ec09742 | 948 | |
81819f0f CL |
949 | bad: |
950 | if (PageSlab(page)) { | |
951 | /* | |
952 | * If this is a slab page then lets do the best we can | |
953 | * to avoid issues in the future. Marking all objects | |
672bba3a | 954 | * as used avoids touching the remaining objects. |
81819f0f | 955 | */ |
24922684 | 956 | slab_fix(s, "Marking all objects used"); |
39b26464 | 957 | page->inuse = page->objects; |
a973e9dd | 958 | page->freelist = NULL; |
81819f0f CL |
959 | } |
960 | return 0; | |
961 | } | |
962 | ||
3ec09742 CL |
963 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
964 | void *object, void *addr) | |
81819f0f CL |
965 | { |
966 | if (!check_slab(s, page)) | |
967 | goto fail; | |
968 | ||
969 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 970 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
971 | goto fail; |
972 | } | |
973 | ||
974 | if (on_freelist(s, page, object)) { | |
24922684 | 975 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
976 | goto fail; |
977 | } | |
978 | ||
979 | if (!check_object(s, page, object, 1)) | |
980 | return 0; | |
981 | ||
982 | if (unlikely(s != page->slab)) { | |
3adbefee | 983 | if (!PageSlab(page)) { |
70d71228 CL |
984 | slab_err(s, page, "Attempt to free object(0x%p) " |
985 | "outside of slab", object); | |
3adbefee | 986 | } else if (!page->slab) { |
81819f0f | 987 | printk(KERN_ERR |
70d71228 | 988 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 989 | object); |
70d71228 | 990 | dump_stack(); |
06428780 | 991 | } else |
24922684 CL |
992 | object_err(s, page, object, |
993 | "page slab pointer corrupt."); | |
81819f0f CL |
994 | goto fail; |
995 | } | |
3ec09742 CL |
996 | |
997 | /* Special debug activities for freeing objects */ | |
a973e9dd | 998 | if (!SlabFrozen(page) && !page->freelist) |
3ec09742 CL |
999 | remove_full(s, page); |
1000 | if (s->flags & SLAB_STORE_USER) | |
1001 | set_track(s, object, TRACK_FREE, addr); | |
1002 | trace(s, page, object, 0); | |
1003 | init_object(s, object, 0); | |
81819f0f | 1004 | return 1; |
3ec09742 | 1005 | |
81819f0f | 1006 | fail: |
24922684 | 1007 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
1008 | return 0; |
1009 | } | |
1010 | ||
41ecc55b CL |
1011 | static int __init setup_slub_debug(char *str) |
1012 | { | |
f0630fff CL |
1013 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1014 | if (*str++ != '=' || !*str) | |
1015 | /* | |
1016 | * No options specified. Switch on full debugging. | |
1017 | */ | |
1018 | goto out; | |
1019 | ||
1020 | if (*str == ',') | |
1021 | /* | |
1022 | * No options but restriction on slabs. This means full | |
1023 | * debugging for slabs matching a pattern. | |
1024 | */ | |
1025 | goto check_slabs; | |
1026 | ||
1027 | slub_debug = 0; | |
1028 | if (*str == '-') | |
1029 | /* | |
1030 | * Switch off all debugging measures. | |
1031 | */ | |
1032 | goto out; | |
1033 | ||
1034 | /* | |
1035 | * Determine which debug features should be switched on | |
1036 | */ | |
06428780 | 1037 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1038 | switch (tolower(*str)) { |
1039 | case 'f': | |
1040 | slub_debug |= SLAB_DEBUG_FREE; | |
1041 | break; | |
1042 | case 'z': | |
1043 | slub_debug |= SLAB_RED_ZONE; | |
1044 | break; | |
1045 | case 'p': | |
1046 | slub_debug |= SLAB_POISON; | |
1047 | break; | |
1048 | case 'u': | |
1049 | slub_debug |= SLAB_STORE_USER; | |
1050 | break; | |
1051 | case 't': | |
1052 | slub_debug |= SLAB_TRACE; | |
1053 | break; | |
1054 | default: | |
1055 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1056 | "unknown. skipped\n", *str); |
f0630fff | 1057 | } |
41ecc55b CL |
1058 | } |
1059 | ||
f0630fff | 1060 | check_slabs: |
41ecc55b CL |
1061 | if (*str == ',') |
1062 | slub_debug_slabs = str + 1; | |
f0630fff | 1063 | out: |
41ecc55b CL |
1064 | return 1; |
1065 | } | |
1066 | ||
1067 | __setup("slub_debug", setup_slub_debug); | |
1068 | ||
ba0268a8 CL |
1069 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1070 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1071 | void (*ctor)(struct kmem_cache *, void *)) |
41ecc55b CL |
1072 | { |
1073 | /* | |
e153362a | 1074 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1075 | */ |
e153362a CL |
1076 | if (slub_debug && (!slub_debug_slabs || |
1077 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1078 | flags |= slub_debug; | |
ba0268a8 CL |
1079 | |
1080 | return flags; | |
41ecc55b CL |
1081 | } |
1082 | #else | |
3ec09742 CL |
1083 | static inline void setup_object_debug(struct kmem_cache *s, |
1084 | struct page *page, void *object) {} | |
41ecc55b | 1085 | |
3ec09742 CL |
1086 | static inline int alloc_debug_processing(struct kmem_cache *s, |
1087 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1088 | |
3ec09742 CL |
1089 | static inline int free_debug_processing(struct kmem_cache *s, |
1090 | struct page *page, void *object, void *addr) { return 0; } | |
41ecc55b | 1091 | |
41ecc55b CL |
1092 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1093 | { return 1; } | |
1094 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1095 | void *object, int active) { return 1; } | |
3ec09742 | 1096 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1097 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1098 | unsigned long flags, const char *name, | |
4ba9b9d0 | 1099 | void (*ctor)(struct kmem_cache *, void *)) |
ba0268a8 CL |
1100 | { |
1101 | return flags; | |
1102 | } | |
41ecc55b | 1103 | #define slub_debug 0 |
0f389ec6 CL |
1104 | |
1105 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1106 | { return 0; } | |
205ab99d CL |
1107 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1108 | int objects) {} | |
1109 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1110 | int objects) {} | |
41ecc55b | 1111 | #endif |
205ab99d | 1112 | |
81819f0f CL |
1113 | /* |
1114 | * Slab allocation and freeing | |
1115 | */ | |
1116 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1117 | { | |
06428780 | 1118 | struct page *page; |
834f3d11 CL |
1119 | struct kmem_cache_order_objects oo = s->oo; |
1120 | int order = oo_order(oo); | |
1121 | int pages = 1 << order; | |
81819f0f | 1122 | |
b7a49f0d | 1123 | flags |= s->allocflags; |
e12ba74d | 1124 | |
81819f0f | 1125 | if (node == -1) |
834f3d11 | 1126 | page = alloc_pages(flags, order); |
81819f0f | 1127 | else |
834f3d11 | 1128 | page = alloc_pages_node(node, flags, order); |
81819f0f CL |
1129 | |
1130 | if (!page) | |
1131 | return NULL; | |
1132 | ||
834f3d11 | 1133 | page->objects = oo_objects(oo); |
81819f0f CL |
1134 | mod_zone_page_state(page_zone(page), |
1135 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1136 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1137 | pages); | |
1138 | ||
1139 | return page; | |
1140 | } | |
1141 | ||
1142 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1143 | void *object) | |
1144 | { | |
3ec09742 | 1145 | setup_object_debug(s, page, object); |
4f104934 | 1146 | if (unlikely(s->ctor)) |
4ba9b9d0 | 1147 | s->ctor(s, object); |
81819f0f CL |
1148 | } |
1149 | ||
1150 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1151 | { | |
1152 | struct page *page; | |
81819f0f | 1153 | void *start; |
81819f0f CL |
1154 | void *last; |
1155 | void *p; | |
1156 | ||
6cb06229 | 1157 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1158 | |
6cb06229 CL |
1159 | page = allocate_slab(s, |
1160 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1161 | if (!page) |
1162 | goto out; | |
1163 | ||
205ab99d | 1164 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1165 | page->slab = s; |
1166 | page->flags |= 1 << PG_slab; | |
1167 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1168 | SLAB_STORE_USER | SLAB_TRACE)) | |
35e5d7ee | 1169 | SetSlabDebug(page); |
81819f0f CL |
1170 | |
1171 | start = page_address(page); | |
81819f0f CL |
1172 | |
1173 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1174 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1175 | |
1176 | last = start; | |
224a88be | 1177 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1178 | setup_object(s, page, last); |
1179 | set_freepointer(s, last, p); | |
1180 | last = p; | |
1181 | } | |
1182 | setup_object(s, page, last); | |
a973e9dd | 1183 | set_freepointer(s, last, NULL); |
81819f0f CL |
1184 | |
1185 | page->freelist = start; | |
1186 | page->inuse = 0; | |
1187 | out: | |
81819f0f CL |
1188 | return page; |
1189 | } | |
1190 | ||
1191 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1192 | { | |
834f3d11 CL |
1193 | int order = compound_order(page); |
1194 | int pages = 1 << order; | |
81819f0f | 1195 | |
c59def9f | 1196 | if (unlikely(SlabDebug(page))) { |
81819f0f CL |
1197 | void *p; |
1198 | ||
1199 | slab_pad_check(s, page); | |
224a88be CL |
1200 | for_each_object(p, s, page_address(page), |
1201 | page->objects) | |
81819f0f | 1202 | check_object(s, page, p, 0); |
2208b764 | 1203 | ClearSlabDebug(page); |
81819f0f CL |
1204 | } |
1205 | ||
1206 | mod_zone_page_state(page_zone(page), | |
1207 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1208 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1209 | -pages); |
81819f0f | 1210 | |
49bd5221 CL |
1211 | __ClearPageSlab(page); |
1212 | reset_page_mapcount(page); | |
834f3d11 | 1213 | __free_pages(page, order); |
81819f0f CL |
1214 | } |
1215 | ||
1216 | static void rcu_free_slab(struct rcu_head *h) | |
1217 | { | |
1218 | struct page *page; | |
1219 | ||
1220 | page = container_of((struct list_head *)h, struct page, lru); | |
1221 | __free_slab(page->slab, page); | |
1222 | } | |
1223 | ||
1224 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1225 | { | |
1226 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1227 | /* | |
1228 | * RCU free overloads the RCU head over the LRU | |
1229 | */ | |
1230 | struct rcu_head *head = (void *)&page->lru; | |
1231 | ||
1232 | call_rcu(head, rcu_free_slab); | |
1233 | } else | |
1234 | __free_slab(s, page); | |
1235 | } | |
1236 | ||
1237 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1238 | { | |
205ab99d | 1239 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1240 | free_slab(s, page); |
1241 | } | |
1242 | ||
1243 | /* | |
1244 | * Per slab locking using the pagelock | |
1245 | */ | |
1246 | static __always_inline void slab_lock(struct page *page) | |
1247 | { | |
1248 | bit_spin_lock(PG_locked, &page->flags); | |
1249 | } | |
1250 | ||
1251 | static __always_inline void slab_unlock(struct page *page) | |
1252 | { | |
a76d3546 | 1253 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1254 | } |
1255 | ||
1256 | static __always_inline int slab_trylock(struct page *page) | |
1257 | { | |
1258 | int rc = 1; | |
1259 | ||
1260 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1261 | return rc; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * Management of partially allocated slabs | |
1266 | */ | |
7c2e132c CL |
1267 | static void add_partial(struct kmem_cache_node *n, |
1268 | struct page *page, int tail) | |
81819f0f | 1269 | { |
e95eed57 CL |
1270 | spin_lock(&n->list_lock); |
1271 | n->nr_partial++; | |
7c2e132c CL |
1272 | if (tail) |
1273 | list_add_tail(&page->lru, &n->partial); | |
1274 | else | |
1275 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1276 | spin_unlock(&n->list_lock); |
1277 | } | |
1278 | ||
1279 | static void remove_partial(struct kmem_cache *s, | |
1280 | struct page *page) | |
1281 | { | |
1282 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1283 | ||
1284 | spin_lock(&n->list_lock); | |
1285 | list_del(&page->lru); | |
1286 | n->nr_partial--; | |
1287 | spin_unlock(&n->list_lock); | |
1288 | } | |
1289 | ||
1290 | /* | |
672bba3a | 1291 | * Lock slab and remove from the partial list. |
81819f0f | 1292 | * |
672bba3a | 1293 | * Must hold list_lock. |
81819f0f | 1294 | */ |
4b6f0750 | 1295 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, struct page *page) |
81819f0f CL |
1296 | { |
1297 | if (slab_trylock(page)) { | |
1298 | list_del(&page->lru); | |
1299 | n->nr_partial--; | |
4b6f0750 | 1300 | SetSlabFrozen(page); |
81819f0f CL |
1301 | return 1; |
1302 | } | |
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | /* | |
672bba3a | 1307 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1308 | */ |
1309 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1310 | { | |
1311 | struct page *page; | |
1312 | ||
1313 | /* | |
1314 | * Racy check. If we mistakenly see no partial slabs then we | |
1315 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1316 | * partial slab and there is none available then get_partials() |
1317 | * will return NULL. | |
81819f0f CL |
1318 | */ |
1319 | if (!n || !n->nr_partial) | |
1320 | return NULL; | |
1321 | ||
1322 | spin_lock(&n->list_lock); | |
1323 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1324 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1325 | goto out; |
1326 | page = NULL; | |
1327 | out: | |
1328 | spin_unlock(&n->list_lock); | |
1329 | return page; | |
1330 | } | |
1331 | ||
1332 | /* | |
672bba3a | 1333 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1334 | */ |
1335 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1336 | { | |
1337 | #ifdef CONFIG_NUMA | |
1338 | struct zonelist *zonelist; | |
1339 | struct zone **z; | |
1340 | struct page *page; | |
1341 | ||
1342 | /* | |
672bba3a CL |
1343 | * The defrag ratio allows a configuration of the tradeoffs between |
1344 | * inter node defragmentation and node local allocations. A lower | |
1345 | * defrag_ratio increases the tendency to do local allocations | |
1346 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1347 | * |
672bba3a CL |
1348 | * If the defrag_ratio is set to 0 then kmalloc() always |
1349 | * returns node local objects. If the ratio is higher then kmalloc() | |
1350 | * may return off node objects because partial slabs are obtained | |
1351 | * from other nodes and filled up. | |
81819f0f | 1352 | * |
6446faa2 | 1353 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1354 | * defrag_ratio = 1000) then every (well almost) allocation will |
1355 | * first attempt to defrag slab caches on other nodes. This means | |
1356 | * scanning over all nodes to look for partial slabs which may be | |
1357 | * expensive if we do it every time we are trying to find a slab | |
1358 | * with available objects. | |
81819f0f | 1359 | */ |
9824601e CL |
1360 | if (!s->remote_node_defrag_ratio || |
1361 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1362 | return NULL; |
1363 | ||
3adbefee IM |
1364 | zonelist = &NODE_DATA( |
1365 | slab_node(current->mempolicy))->node_zonelists[gfp_zone(flags)]; | |
81819f0f CL |
1366 | for (z = zonelist->zones; *z; z++) { |
1367 | struct kmem_cache_node *n; | |
1368 | ||
1369 | n = get_node(s, zone_to_nid(*z)); | |
1370 | ||
1371 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
e95eed57 | 1372 | n->nr_partial > MIN_PARTIAL) { |
81819f0f CL |
1373 | page = get_partial_node(n); |
1374 | if (page) | |
1375 | return page; | |
1376 | } | |
1377 | } | |
1378 | #endif | |
1379 | return NULL; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * Get a partial page, lock it and return it. | |
1384 | */ | |
1385 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1386 | { | |
1387 | struct page *page; | |
1388 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1389 | ||
1390 | page = get_partial_node(get_node(s, searchnode)); | |
1391 | if (page || (flags & __GFP_THISNODE)) | |
1392 | return page; | |
1393 | ||
1394 | return get_any_partial(s, flags); | |
1395 | } | |
1396 | ||
1397 | /* | |
1398 | * Move a page back to the lists. | |
1399 | * | |
1400 | * Must be called with the slab lock held. | |
1401 | * | |
1402 | * On exit the slab lock will have been dropped. | |
1403 | */ | |
7c2e132c | 1404 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1405 | { |
e95eed57 | 1406 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1407 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1408 | |
4b6f0750 | 1409 | ClearSlabFrozen(page); |
81819f0f | 1410 | if (page->inuse) { |
e95eed57 | 1411 | |
a973e9dd | 1412 | if (page->freelist) { |
7c2e132c | 1413 | add_partial(n, page, tail); |
8ff12cfc CL |
1414 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1415 | } else { | |
1416 | stat(c, DEACTIVATE_FULL); | |
1417 | if (SlabDebug(page) && (s->flags & SLAB_STORE_USER)) | |
1418 | add_full(n, page); | |
1419 | } | |
81819f0f CL |
1420 | slab_unlock(page); |
1421 | } else { | |
8ff12cfc | 1422 | stat(c, DEACTIVATE_EMPTY); |
e95eed57 CL |
1423 | if (n->nr_partial < MIN_PARTIAL) { |
1424 | /* | |
672bba3a CL |
1425 | * Adding an empty slab to the partial slabs in order |
1426 | * to avoid page allocator overhead. This slab needs | |
1427 | * to come after the other slabs with objects in | |
6446faa2 CL |
1428 | * so that the others get filled first. That way the |
1429 | * size of the partial list stays small. | |
1430 | * | |
1431 | * kmem_cache_shrink can reclaim any empty slabs from the | |
1432 | * partial list. | |
e95eed57 | 1433 | */ |
7c2e132c | 1434 | add_partial(n, page, 1); |
e95eed57 CL |
1435 | slab_unlock(page); |
1436 | } else { | |
1437 | slab_unlock(page); | |
8ff12cfc | 1438 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1439 | discard_slab(s, page); |
1440 | } | |
81819f0f CL |
1441 | } |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * Remove the cpu slab | |
1446 | */ | |
dfb4f096 | 1447 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1448 | { |
dfb4f096 | 1449 | struct page *page = c->page; |
7c2e132c | 1450 | int tail = 1; |
8ff12cfc | 1451 | |
b773ad73 | 1452 | if (page->freelist) |
8ff12cfc | 1453 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1454 | /* |
6446faa2 | 1455 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1456 | * because both freelists are empty. So this is unlikely |
1457 | * to occur. | |
1458 | */ | |
a973e9dd | 1459 | while (unlikely(c->freelist)) { |
894b8788 CL |
1460 | void **object; |
1461 | ||
7c2e132c CL |
1462 | tail = 0; /* Hot objects. Put the slab first */ |
1463 | ||
894b8788 | 1464 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1465 | object = c->freelist; |
b3fba8da | 1466 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1467 | |
1468 | /* And put onto the regular freelist */ | |
b3fba8da | 1469 | object[c->offset] = page->freelist; |
894b8788 CL |
1470 | page->freelist = object; |
1471 | page->inuse--; | |
1472 | } | |
dfb4f096 | 1473 | c->page = NULL; |
7c2e132c | 1474 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1475 | } |
1476 | ||
dfb4f096 | 1477 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1478 | { |
8ff12cfc | 1479 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1480 | slab_lock(c->page); |
1481 | deactivate_slab(s, c); | |
81819f0f CL |
1482 | } |
1483 | ||
1484 | /* | |
1485 | * Flush cpu slab. | |
6446faa2 | 1486 | * |
81819f0f CL |
1487 | * Called from IPI handler with interrupts disabled. |
1488 | */ | |
0c710013 | 1489 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1490 | { |
dfb4f096 | 1491 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1492 | |
dfb4f096 CL |
1493 | if (likely(c && c->page)) |
1494 | flush_slab(s, c); | |
81819f0f CL |
1495 | } |
1496 | ||
1497 | static void flush_cpu_slab(void *d) | |
1498 | { | |
1499 | struct kmem_cache *s = d; | |
81819f0f | 1500 | |
dfb4f096 | 1501 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1502 | } |
1503 | ||
1504 | static void flush_all(struct kmem_cache *s) | |
1505 | { | |
1506 | #ifdef CONFIG_SMP | |
1507 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1508 | #else | |
1509 | unsigned long flags; | |
1510 | ||
1511 | local_irq_save(flags); | |
1512 | flush_cpu_slab(s); | |
1513 | local_irq_restore(flags); | |
1514 | #endif | |
1515 | } | |
1516 | ||
dfb4f096 CL |
1517 | /* |
1518 | * Check if the objects in a per cpu structure fit numa | |
1519 | * locality expectations. | |
1520 | */ | |
1521 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1522 | { | |
1523 | #ifdef CONFIG_NUMA | |
1524 | if (node != -1 && c->node != node) | |
1525 | return 0; | |
1526 | #endif | |
1527 | return 1; | |
1528 | } | |
1529 | ||
81819f0f | 1530 | /* |
894b8788 CL |
1531 | * Slow path. The lockless freelist is empty or we need to perform |
1532 | * debugging duties. | |
1533 | * | |
1534 | * Interrupts are disabled. | |
81819f0f | 1535 | * |
894b8788 CL |
1536 | * Processing is still very fast if new objects have been freed to the |
1537 | * regular freelist. In that case we simply take over the regular freelist | |
1538 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1539 | * |
894b8788 CL |
1540 | * If that is not working then we fall back to the partial lists. We take the |
1541 | * first element of the freelist as the object to allocate now and move the | |
1542 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1543 | * |
894b8788 | 1544 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1545 | * we need to allocate a new slab. This is the slowest path since it involves |
1546 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1547 | */ |
894b8788 | 1548 | static void *__slab_alloc(struct kmem_cache *s, |
dfb4f096 | 1549 | gfp_t gfpflags, int node, void *addr, struct kmem_cache_cpu *c) |
81819f0f | 1550 | { |
81819f0f | 1551 | void **object; |
dfb4f096 | 1552 | struct page *new; |
81819f0f | 1553 | |
e72e9c23 LT |
1554 | /* We handle __GFP_ZERO in the caller */ |
1555 | gfpflags &= ~__GFP_ZERO; | |
1556 | ||
dfb4f096 | 1557 | if (!c->page) |
81819f0f CL |
1558 | goto new_slab; |
1559 | ||
dfb4f096 CL |
1560 | slab_lock(c->page); |
1561 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1562 | goto another_slab; |
6446faa2 | 1563 | |
8ff12cfc | 1564 | stat(c, ALLOC_REFILL); |
6446faa2 | 1565 | |
894b8788 | 1566 | load_freelist: |
dfb4f096 | 1567 | object = c->page->freelist; |
a973e9dd | 1568 | if (unlikely(!object)) |
81819f0f | 1569 | goto another_slab; |
dfb4f096 | 1570 | if (unlikely(SlabDebug(c->page))) |
81819f0f CL |
1571 | goto debug; |
1572 | ||
b3fba8da | 1573 | c->freelist = object[c->offset]; |
39b26464 | 1574 | c->page->inuse = c->page->objects; |
a973e9dd | 1575 | c->page->freelist = NULL; |
dfb4f096 | 1576 | c->node = page_to_nid(c->page); |
1f84260c | 1577 | unlock_out: |
dfb4f096 | 1578 | slab_unlock(c->page); |
8ff12cfc | 1579 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1580 | return object; |
1581 | ||
1582 | another_slab: | |
dfb4f096 | 1583 | deactivate_slab(s, c); |
81819f0f CL |
1584 | |
1585 | new_slab: | |
dfb4f096 CL |
1586 | new = get_partial(s, gfpflags, node); |
1587 | if (new) { | |
1588 | c->page = new; | |
8ff12cfc | 1589 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1590 | goto load_freelist; |
81819f0f CL |
1591 | } |
1592 | ||
b811c202 CL |
1593 | if (gfpflags & __GFP_WAIT) |
1594 | local_irq_enable(); | |
1595 | ||
dfb4f096 | 1596 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1597 | |
1598 | if (gfpflags & __GFP_WAIT) | |
1599 | local_irq_disable(); | |
1600 | ||
dfb4f096 CL |
1601 | if (new) { |
1602 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1603 | stat(c, ALLOC_SLAB); |
05aa3450 | 1604 | if (c->page) |
dfb4f096 | 1605 | flush_slab(s, c); |
dfb4f096 CL |
1606 | slab_lock(new); |
1607 | SetSlabFrozen(new); | |
1608 | c->page = new; | |
4b6f0750 | 1609 | goto load_freelist; |
81819f0f | 1610 | } |
00e962c5 | 1611 | |
71c7a06f CL |
1612 | /* |
1613 | * No memory available. | |
1614 | * | |
1615 | * If the slab uses higher order allocs but the object is | |
1616 | * smaller than a page size then we can fallback in emergencies | |
1617 | * to the page allocator via kmalloc_large. The page allocator may | |
1618 | * have failed to obtain a higher order page and we can try to | |
1619 | * allocate a single page if the object fits into a single page. | |
1620 | * That is only possible if certain conditions are met that are being | |
1621 | * checked when a slab is created. | |
1622 | */ | |
caeab084 CL |
1623 | if (!(gfpflags & __GFP_NORETRY) && |
1624 | (s->flags & __PAGE_ALLOC_FALLBACK)) { | |
1625 | if (gfpflags & __GFP_WAIT) | |
1626 | local_irq_enable(); | |
1627 | object = kmalloc_large(s->objsize, gfpflags); | |
1628 | if (gfpflags & __GFP_WAIT) | |
1629 | local_irq_disable(); | |
1630 | return object; | |
1631 | } | |
71c7a06f | 1632 | return NULL; |
81819f0f | 1633 | debug: |
dfb4f096 | 1634 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1635 | goto another_slab; |
894b8788 | 1636 | |
dfb4f096 | 1637 | c->page->inuse++; |
b3fba8da | 1638 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1639 | c->node = -1; |
1f84260c | 1640 | goto unlock_out; |
894b8788 CL |
1641 | } |
1642 | ||
1643 | /* | |
1644 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1645 | * have the fastpath folded into their functions. So no function call | |
1646 | * overhead for requests that can be satisfied on the fastpath. | |
1647 | * | |
1648 | * The fastpath works by first checking if the lockless freelist can be used. | |
1649 | * If not then __slab_alloc is called for slow processing. | |
1650 | * | |
1651 | * Otherwise we can simply pick the next object from the lockless free list. | |
1652 | */ | |
06428780 | 1653 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce15fea8 | 1654 | gfp_t gfpflags, int node, void *addr) |
894b8788 | 1655 | { |
894b8788 | 1656 | void **object; |
dfb4f096 | 1657 | struct kmem_cache_cpu *c; |
1f84260c CL |
1658 | unsigned long flags; |
1659 | ||
894b8788 | 1660 | local_irq_save(flags); |
dfb4f096 | 1661 | c = get_cpu_slab(s, smp_processor_id()); |
a973e9dd | 1662 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1663 | |
dfb4f096 | 1664 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1665 | |
1666 | else { | |
dfb4f096 | 1667 | object = c->freelist; |
b3fba8da | 1668 | c->freelist = object[c->offset]; |
8ff12cfc | 1669 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1670 | } |
1671 | local_irq_restore(flags); | |
d07dbea4 CL |
1672 | |
1673 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
42a9fdbb | 1674 | memset(object, 0, c->objsize); |
d07dbea4 | 1675 | |
894b8788 | 1676 | return object; |
81819f0f CL |
1677 | } |
1678 | ||
1679 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1680 | { | |
ce15fea8 | 1681 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1682 | } |
1683 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1684 | ||
1685 | #ifdef CONFIG_NUMA | |
1686 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1687 | { | |
ce15fea8 | 1688 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1689 | } |
1690 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1691 | #endif | |
1692 | ||
1693 | /* | |
894b8788 CL |
1694 | * Slow patch handling. This may still be called frequently since objects |
1695 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1696 | * |
894b8788 CL |
1697 | * So we still attempt to reduce cache line usage. Just take the slab |
1698 | * lock and free the item. If there is no additional partial page | |
1699 | * handling required then we can return immediately. | |
81819f0f | 1700 | */ |
894b8788 | 1701 | static void __slab_free(struct kmem_cache *s, struct page *page, |
b3fba8da | 1702 | void *x, void *addr, unsigned int offset) |
81819f0f CL |
1703 | { |
1704 | void *prior; | |
1705 | void **object = (void *)x; | |
8ff12cfc | 1706 | struct kmem_cache_cpu *c; |
81819f0f | 1707 | |
8ff12cfc CL |
1708 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1709 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1710 | slab_lock(page); |
1711 | ||
35e5d7ee | 1712 | if (unlikely(SlabDebug(page))) |
81819f0f | 1713 | goto debug; |
6446faa2 | 1714 | |
81819f0f | 1715 | checks_ok: |
b3fba8da | 1716 | prior = object[offset] = page->freelist; |
81819f0f CL |
1717 | page->freelist = object; |
1718 | page->inuse--; | |
1719 | ||
8ff12cfc CL |
1720 | if (unlikely(SlabFrozen(page))) { |
1721 | stat(c, FREE_FROZEN); | |
81819f0f | 1722 | goto out_unlock; |
8ff12cfc | 1723 | } |
81819f0f CL |
1724 | |
1725 | if (unlikely(!page->inuse)) | |
1726 | goto slab_empty; | |
1727 | ||
1728 | /* | |
6446faa2 | 1729 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1730 | * then add it. |
1731 | */ | |
a973e9dd | 1732 | if (unlikely(!prior)) { |
7c2e132c | 1733 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1734 | stat(c, FREE_ADD_PARTIAL); |
1735 | } | |
81819f0f CL |
1736 | |
1737 | out_unlock: | |
1738 | slab_unlock(page); | |
81819f0f CL |
1739 | return; |
1740 | ||
1741 | slab_empty: | |
a973e9dd | 1742 | if (prior) { |
81819f0f | 1743 | /* |
672bba3a | 1744 | * Slab still on the partial list. |
81819f0f CL |
1745 | */ |
1746 | remove_partial(s, page); | |
8ff12cfc CL |
1747 | stat(c, FREE_REMOVE_PARTIAL); |
1748 | } | |
81819f0f | 1749 | slab_unlock(page); |
8ff12cfc | 1750 | stat(c, FREE_SLAB); |
81819f0f | 1751 | discard_slab(s, page); |
81819f0f CL |
1752 | return; |
1753 | ||
1754 | debug: | |
3ec09742 | 1755 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1756 | goto out_unlock; |
77c5e2d0 | 1757 | goto checks_ok; |
81819f0f CL |
1758 | } |
1759 | ||
894b8788 CL |
1760 | /* |
1761 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1762 | * can perform fastpath freeing without additional function calls. | |
1763 | * | |
1764 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1765 | * of this processor. This typically the case if we have just allocated | |
1766 | * the item before. | |
1767 | * | |
1768 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1769 | * with all sorts of special processing. | |
1770 | */ | |
06428780 | 1771 | static __always_inline void slab_free(struct kmem_cache *s, |
894b8788 CL |
1772 | struct page *page, void *x, void *addr) |
1773 | { | |
1774 | void **object = (void *)x; | |
dfb4f096 | 1775 | struct kmem_cache_cpu *c; |
1f84260c CL |
1776 | unsigned long flags; |
1777 | ||
894b8788 | 1778 | local_irq_save(flags); |
dfb4f096 | 1779 | c = get_cpu_slab(s, smp_processor_id()); |
27d9e4e9 | 1780 | debug_check_no_locks_freed(object, c->objsize); |
ee3c72a1 | 1781 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1782 | object[c->offset] = c->freelist; |
dfb4f096 | 1783 | c->freelist = object; |
8ff12cfc | 1784 | stat(c, FREE_FASTPATH); |
894b8788 | 1785 | } else |
b3fba8da | 1786 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1787 | |
1788 | local_irq_restore(flags); | |
1789 | } | |
1790 | ||
81819f0f CL |
1791 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1792 | { | |
77c5e2d0 | 1793 | struct page *page; |
81819f0f | 1794 | |
b49af68f | 1795 | page = virt_to_head_page(x); |
81819f0f | 1796 | |
77c5e2d0 | 1797 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1798 | } |
1799 | EXPORT_SYMBOL(kmem_cache_free); | |
1800 | ||
1801 | /* Figure out on which slab object the object resides */ | |
1802 | static struct page *get_object_page(const void *x) | |
1803 | { | |
b49af68f | 1804 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1805 | |
1806 | if (!PageSlab(page)) | |
1807 | return NULL; | |
1808 | ||
1809 | return page; | |
1810 | } | |
1811 | ||
1812 | /* | |
672bba3a CL |
1813 | * Object placement in a slab is made very easy because we always start at |
1814 | * offset 0. If we tune the size of the object to the alignment then we can | |
1815 | * get the required alignment by putting one properly sized object after | |
1816 | * another. | |
81819f0f CL |
1817 | * |
1818 | * Notice that the allocation order determines the sizes of the per cpu | |
1819 | * caches. Each processor has always one slab available for allocations. | |
1820 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1821 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1822 | * locking overhead. |
81819f0f CL |
1823 | */ |
1824 | ||
1825 | /* | |
1826 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1827 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1828 | * and increases the number of allocations possible without having to | |
1829 | * take the list_lock. | |
1830 | */ | |
1831 | static int slub_min_order; | |
1832 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
81819f0f CL |
1833 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; |
1834 | ||
1835 | /* | |
1836 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1837 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1838 | */ |
1839 | static int slub_nomerge; | |
1840 | ||
81819f0f CL |
1841 | /* |
1842 | * Calculate the order of allocation given an slab object size. | |
1843 | * | |
672bba3a CL |
1844 | * The order of allocation has significant impact on performance and other |
1845 | * system components. Generally order 0 allocations should be preferred since | |
1846 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1847 | * be problematic to put into order 0 slabs because there may be too much | |
1848 | * unused space left. We go to a higher order if more than 1/8th of the slab | |
1849 | * would be wasted. | |
1850 | * | |
1851 | * In order to reach satisfactory performance we must ensure that a minimum | |
1852 | * number of objects is in one slab. Otherwise we may generate too much | |
1853 | * activity on the partial lists which requires taking the list_lock. This is | |
1854 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1855 | * |
672bba3a CL |
1856 | * slub_max_order specifies the order where we begin to stop considering the |
1857 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1858 | * we try to keep the page order as low as possible. So we accept more waste | |
1859 | * of space in favor of a small page order. | |
81819f0f | 1860 | * |
672bba3a CL |
1861 | * Higher order allocations also allow the placement of more objects in a |
1862 | * slab and thereby reduce object handling overhead. If the user has | |
1863 | * requested a higher mininum order then we start with that one instead of | |
1864 | * the smallest order which will fit the object. | |
81819f0f | 1865 | */ |
5e6d444e CL |
1866 | static inline int slab_order(int size, int min_objects, |
1867 | int max_order, int fract_leftover) | |
81819f0f CL |
1868 | { |
1869 | int order; | |
1870 | int rem; | |
6300ea75 | 1871 | int min_order = slub_min_order; |
81819f0f | 1872 | |
39b26464 CL |
1873 | if ((PAGE_SIZE << min_order) / size > 65535) |
1874 | return get_order(size * 65535) - 1; | |
1875 | ||
6300ea75 | 1876 | for (order = max(min_order, |
5e6d444e CL |
1877 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1878 | order <= max_order; order++) { | |
81819f0f | 1879 | |
5e6d444e | 1880 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1881 | |
5e6d444e | 1882 | if (slab_size < min_objects * size) |
81819f0f CL |
1883 | continue; |
1884 | ||
1885 | rem = slab_size % size; | |
1886 | ||
5e6d444e | 1887 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1888 | break; |
1889 | ||
1890 | } | |
672bba3a | 1891 | |
81819f0f CL |
1892 | return order; |
1893 | } | |
1894 | ||
5e6d444e CL |
1895 | static inline int calculate_order(int size) |
1896 | { | |
1897 | int order; | |
1898 | int min_objects; | |
1899 | int fraction; | |
1900 | ||
1901 | /* | |
1902 | * Attempt to find best configuration for a slab. This | |
1903 | * works by first attempting to generate a layout with | |
1904 | * the best configuration and backing off gradually. | |
1905 | * | |
1906 | * First we reduce the acceptable waste in a slab. Then | |
1907 | * we reduce the minimum objects required in a slab. | |
1908 | */ | |
1909 | min_objects = slub_min_objects; | |
1910 | while (min_objects > 1) { | |
1911 | fraction = 8; | |
1912 | while (fraction >= 4) { | |
1913 | order = slab_order(size, min_objects, | |
1914 | slub_max_order, fraction); | |
1915 | if (order <= slub_max_order) | |
1916 | return order; | |
1917 | fraction /= 2; | |
1918 | } | |
1919 | min_objects /= 2; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * We were unable to place multiple objects in a slab. Now | |
1924 | * lets see if we can place a single object there. | |
1925 | */ | |
1926 | order = slab_order(size, 1, slub_max_order, 1); | |
1927 | if (order <= slub_max_order) | |
1928 | return order; | |
1929 | ||
1930 | /* | |
1931 | * Doh this slab cannot be placed using slub_max_order. | |
1932 | */ | |
1933 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1934 | if (order <= MAX_ORDER) | |
1935 | return order; | |
1936 | return -ENOSYS; | |
1937 | } | |
1938 | ||
81819f0f | 1939 | /* |
672bba3a | 1940 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1941 | */ |
1942 | static unsigned long calculate_alignment(unsigned long flags, | |
1943 | unsigned long align, unsigned long size) | |
1944 | { | |
1945 | /* | |
6446faa2 CL |
1946 | * If the user wants hardware cache aligned objects then follow that |
1947 | * suggestion if the object is sufficiently large. | |
81819f0f | 1948 | * |
6446faa2 CL |
1949 | * The hardware cache alignment cannot override the specified |
1950 | * alignment though. If that is greater then use it. | |
81819f0f | 1951 | */ |
b6210386 NP |
1952 | if (flags & SLAB_HWCACHE_ALIGN) { |
1953 | unsigned long ralign = cache_line_size(); | |
1954 | while (size <= ralign / 2) | |
1955 | ralign /= 2; | |
1956 | align = max(align, ralign); | |
1957 | } | |
81819f0f CL |
1958 | |
1959 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 1960 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
1961 | |
1962 | return ALIGN(align, sizeof(void *)); | |
1963 | } | |
1964 | ||
dfb4f096 CL |
1965 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1966 | struct kmem_cache_cpu *c) | |
1967 | { | |
1968 | c->page = NULL; | |
a973e9dd | 1969 | c->freelist = NULL; |
dfb4f096 | 1970 | c->node = 0; |
42a9fdbb CL |
1971 | c->offset = s->offset / sizeof(void *); |
1972 | c->objsize = s->objsize; | |
62f75532 PE |
1973 | #ifdef CONFIG_SLUB_STATS |
1974 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
1975 | #endif | |
dfb4f096 CL |
1976 | } |
1977 | ||
81819f0f CL |
1978 | static void init_kmem_cache_node(struct kmem_cache_node *n) |
1979 | { | |
1980 | n->nr_partial = 0; | |
81819f0f CL |
1981 | spin_lock_init(&n->list_lock); |
1982 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1983 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 1984 | atomic_long_set(&n->nr_slabs, 0); |
643b1138 | 1985 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1986 | #endif |
81819f0f CL |
1987 | } |
1988 | ||
4c93c355 CL |
1989 | #ifdef CONFIG_SMP |
1990 | /* | |
1991 | * Per cpu array for per cpu structures. | |
1992 | * | |
1993 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1994 | * close together meaning that it becomes possible that multiple per cpu | |
1995 | * structures are contained in one cacheline. This may be particularly | |
1996 | * beneficial for the kmalloc caches. | |
1997 | * | |
1998 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1999 | * likely able to get per cpu structures for all caches from the array defined | |
2000 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2001 | * | |
2002 | * If the per cpu array is exhausted then fall back to kmalloc | |
2003 | * of individual cachelines. No sharing is possible then. | |
2004 | */ | |
2005 | #define NR_KMEM_CACHE_CPU 100 | |
2006 | ||
2007 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2008 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2009 | ||
2010 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
2011 | static cpumask_t kmem_cach_cpu_free_init_once = CPU_MASK_NONE; | |
2012 | ||
2013 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2014 | int cpu, gfp_t flags) | |
2015 | { | |
2016 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2017 | ||
2018 | if (c) | |
2019 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2020 | (void *)c->freelist; | |
2021 | else { | |
2022 | /* Table overflow: So allocate ourselves */ | |
2023 | c = kmalloc_node( | |
2024 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2025 | flags, cpu_to_node(cpu)); | |
2026 | if (!c) | |
2027 | return NULL; | |
2028 | } | |
2029 | ||
2030 | init_kmem_cache_cpu(s, c); | |
2031 | return c; | |
2032 | } | |
2033 | ||
2034 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2035 | { | |
2036 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
2037 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
2038 | kfree(c); | |
2039 | return; | |
2040 | } | |
2041 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2042 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2043 | } | |
2044 | ||
2045 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2046 | { | |
2047 | int cpu; | |
2048 | ||
2049 | for_each_online_cpu(cpu) { | |
2050 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2051 | ||
2052 | if (c) { | |
2053 | s->cpu_slab[cpu] = NULL; | |
2054 | free_kmem_cache_cpu(c, cpu); | |
2055 | } | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2060 | { | |
2061 | int cpu; | |
2062 | ||
2063 | for_each_online_cpu(cpu) { | |
2064 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2065 | ||
2066 | if (c) | |
2067 | continue; | |
2068 | ||
2069 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2070 | if (!c) { | |
2071 | free_kmem_cache_cpus(s); | |
2072 | return 0; | |
2073 | } | |
2074 | s->cpu_slab[cpu] = c; | |
2075 | } | |
2076 | return 1; | |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * Initialize the per cpu array. | |
2081 | */ | |
2082 | static void init_alloc_cpu_cpu(int cpu) | |
2083 | { | |
2084 | int i; | |
2085 | ||
2086 | if (cpu_isset(cpu, kmem_cach_cpu_free_init_once)) | |
2087 | return; | |
2088 | ||
2089 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2090 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2091 | ||
2092 | cpu_set(cpu, kmem_cach_cpu_free_init_once); | |
2093 | } | |
2094 | ||
2095 | static void __init init_alloc_cpu(void) | |
2096 | { | |
2097 | int cpu; | |
2098 | ||
2099 | for_each_online_cpu(cpu) | |
2100 | init_alloc_cpu_cpu(cpu); | |
2101 | } | |
2102 | ||
2103 | #else | |
2104 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2105 | static inline void init_alloc_cpu(void) {} | |
2106 | ||
2107 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2108 | { | |
2109 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2110 | return 1; | |
2111 | } | |
2112 | #endif | |
2113 | ||
81819f0f CL |
2114 | #ifdef CONFIG_NUMA |
2115 | /* | |
2116 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2117 | * slab on the node for this slabcache. There are no concurrent accesses | |
2118 | * possible. | |
2119 | * | |
2120 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2121 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2122 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2123 | */ |
1cd7daa5 AB |
2124 | static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags, |
2125 | int node) | |
81819f0f CL |
2126 | { |
2127 | struct page *page; | |
2128 | struct kmem_cache_node *n; | |
ba84c73c | 2129 | unsigned long flags; |
81819f0f CL |
2130 | |
2131 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2132 | ||
a2f92ee7 | 2133 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2134 | |
2135 | BUG_ON(!page); | |
a2f92ee7 CL |
2136 | if (page_to_nid(page) != node) { |
2137 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2138 | "node %d\n", node); | |
2139 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2140 | "in order to be able to continue\n"); | |
2141 | } | |
2142 | ||
81819f0f CL |
2143 | n = page->freelist; |
2144 | BUG_ON(!n); | |
2145 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2146 | page->inuse++; | |
2147 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2148 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2149 | init_object(kmalloc_caches, n, 1); |
2150 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2151 | #endif |
81819f0f | 2152 | init_kmem_cache_node(n); |
205ab99d | 2153 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2154 | |
ba84c73c | 2155 | /* |
2156 | * lockdep requires consistent irq usage for each lock | |
2157 | * so even though there cannot be a race this early in | |
2158 | * the boot sequence, we still disable irqs. | |
2159 | */ | |
2160 | local_irq_save(flags); | |
7c2e132c | 2161 | add_partial(n, page, 0); |
ba84c73c | 2162 | local_irq_restore(flags); |
81819f0f CL |
2163 | return n; |
2164 | } | |
2165 | ||
2166 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2167 | { | |
2168 | int node; | |
2169 | ||
f64dc58c | 2170 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2171 | struct kmem_cache_node *n = s->node[node]; |
2172 | if (n && n != &s->local_node) | |
2173 | kmem_cache_free(kmalloc_caches, n); | |
2174 | s->node[node] = NULL; | |
2175 | } | |
2176 | } | |
2177 | ||
2178 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2179 | { | |
2180 | int node; | |
2181 | int local_node; | |
2182 | ||
2183 | if (slab_state >= UP) | |
2184 | local_node = page_to_nid(virt_to_page(s)); | |
2185 | else | |
2186 | local_node = 0; | |
2187 | ||
f64dc58c | 2188 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2189 | struct kmem_cache_node *n; |
2190 | ||
2191 | if (local_node == node) | |
2192 | n = &s->local_node; | |
2193 | else { | |
2194 | if (slab_state == DOWN) { | |
2195 | n = early_kmem_cache_node_alloc(gfpflags, | |
2196 | node); | |
2197 | continue; | |
2198 | } | |
2199 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2200 | gfpflags, node); | |
2201 | ||
2202 | if (!n) { | |
2203 | free_kmem_cache_nodes(s); | |
2204 | return 0; | |
2205 | } | |
2206 | ||
2207 | } | |
2208 | s->node[node] = n; | |
2209 | init_kmem_cache_node(n); | |
2210 | } | |
2211 | return 1; | |
2212 | } | |
2213 | #else | |
2214 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2215 | { | |
2216 | } | |
2217 | ||
2218 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2219 | { | |
2220 | init_kmem_cache_node(&s->local_node); | |
2221 | return 1; | |
2222 | } | |
2223 | #endif | |
2224 | ||
2225 | /* | |
2226 | * calculate_sizes() determines the order and the distribution of data within | |
2227 | * a slab object. | |
2228 | */ | |
2229 | static int calculate_sizes(struct kmem_cache *s) | |
2230 | { | |
2231 | unsigned long flags = s->flags; | |
2232 | unsigned long size = s->objsize; | |
2233 | unsigned long align = s->align; | |
834f3d11 | 2234 | int order; |
81819f0f | 2235 | |
d8b42bf5 CL |
2236 | /* |
2237 | * Round up object size to the next word boundary. We can only | |
2238 | * place the free pointer at word boundaries and this determines | |
2239 | * the possible location of the free pointer. | |
2240 | */ | |
2241 | size = ALIGN(size, sizeof(void *)); | |
2242 | ||
2243 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2244 | /* |
2245 | * Determine if we can poison the object itself. If the user of | |
2246 | * the slab may touch the object after free or before allocation | |
2247 | * then we should never poison the object itself. | |
2248 | */ | |
2249 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2250 | !s->ctor) |
81819f0f CL |
2251 | s->flags |= __OBJECT_POISON; |
2252 | else | |
2253 | s->flags &= ~__OBJECT_POISON; | |
2254 | ||
81819f0f CL |
2255 | |
2256 | /* | |
672bba3a | 2257 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2258 | * end of the object and the free pointer. If not then add an |
672bba3a | 2259 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2260 | */ |
2261 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2262 | size += sizeof(void *); | |
41ecc55b | 2263 | #endif |
81819f0f CL |
2264 | |
2265 | /* | |
672bba3a CL |
2266 | * With that we have determined the number of bytes in actual use |
2267 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2268 | */ |
2269 | s->inuse = size; | |
2270 | ||
2271 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2272 | s->ctor)) { |
81819f0f CL |
2273 | /* |
2274 | * Relocate free pointer after the object if it is not | |
2275 | * permitted to overwrite the first word of the object on | |
2276 | * kmem_cache_free. | |
2277 | * | |
2278 | * This is the case if we do RCU, have a constructor or | |
2279 | * destructor or are poisoning the objects. | |
2280 | */ | |
2281 | s->offset = size; | |
2282 | size += sizeof(void *); | |
2283 | } | |
2284 | ||
c12b3c62 | 2285 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2286 | if (flags & SLAB_STORE_USER) |
2287 | /* | |
2288 | * Need to store information about allocs and frees after | |
2289 | * the object. | |
2290 | */ | |
2291 | size += 2 * sizeof(struct track); | |
2292 | ||
be7b3fbc | 2293 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2294 | /* |
2295 | * Add some empty padding so that we can catch | |
2296 | * overwrites from earlier objects rather than let | |
2297 | * tracking information or the free pointer be | |
2298 | * corrupted if an user writes before the start | |
2299 | * of the object. | |
2300 | */ | |
2301 | size += sizeof(void *); | |
41ecc55b | 2302 | #endif |
672bba3a | 2303 | |
81819f0f CL |
2304 | /* |
2305 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2306 | * user specified and the dynamic determination of cache line size |
2307 | * on bootup. | |
81819f0f CL |
2308 | */ |
2309 | align = calculate_alignment(flags, align, s->objsize); | |
2310 | ||
2311 | /* | |
2312 | * SLUB stores one object immediately after another beginning from | |
2313 | * offset 0. In order to align the objects we have to simply size | |
2314 | * each object to conform to the alignment. | |
2315 | */ | |
2316 | size = ALIGN(size, align); | |
2317 | s->size = size; | |
2318 | ||
71c7a06f CL |
2319 | if ((flags & __KMALLOC_CACHE) && |
2320 | PAGE_SIZE / size < slub_min_objects) { | |
2321 | /* | |
2322 | * Kmalloc cache that would not have enough objects in | |
2323 | * an order 0 page. Kmalloc slabs can fallback to | |
2324 | * page allocator order 0 allocs so take a reasonably large | |
2325 | * order that will allows us a good number of objects. | |
2326 | */ | |
834f3d11 | 2327 | order = max(slub_max_order, PAGE_ALLOC_COSTLY_ORDER); |
71c7a06f CL |
2328 | s->flags |= __PAGE_ALLOC_FALLBACK; |
2329 | s->allocflags |= __GFP_NOWARN; | |
2330 | } else | |
834f3d11 | 2331 | order = calculate_order(size); |
71c7a06f | 2332 | |
834f3d11 | 2333 | if (order < 0) |
81819f0f CL |
2334 | return 0; |
2335 | ||
b7a49f0d | 2336 | s->allocflags = 0; |
834f3d11 | 2337 | if (order) |
b7a49f0d CL |
2338 | s->allocflags |= __GFP_COMP; |
2339 | ||
2340 | if (s->flags & SLAB_CACHE_DMA) | |
2341 | s->allocflags |= SLUB_DMA; | |
2342 | ||
2343 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2344 | s->allocflags |= __GFP_RECLAIMABLE; | |
2345 | ||
81819f0f CL |
2346 | /* |
2347 | * Determine the number of objects per slab | |
2348 | */ | |
834f3d11 | 2349 | s->oo = oo_make(order, size); |
205ab99d CL |
2350 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2351 | s->max = s->oo; | |
81819f0f | 2352 | |
834f3d11 | 2353 | return !!oo_objects(s->oo); |
81819f0f CL |
2354 | |
2355 | } | |
2356 | ||
81819f0f CL |
2357 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2358 | const char *name, size_t size, | |
2359 | size_t align, unsigned long flags, | |
4ba9b9d0 | 2360 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
2361 | { |
2362 | memset(s, 0, kmem_size); | |
2363 | s->name = name; | |
2364 | s->ctor = ctor; | |
81819f0f | 2365 | s->objsize = size; |
81819f0f | 2366 | s->align = align; |
ba0268a8 | 2367 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f CL |
2368 | |
2369 | if (!calculate_sizes(s)) | |
2370 | goto error; | |
2371 | ||
2372 | s->refcount = 1; | |
2373 | #ifdef CONFIG_NUMA | |
9824601e | 2374 | s->remote_node_defrag_ratio = 100; |
81819f0f | 2375 | #endif |
dfb4f096 CL |
2376 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2377 | goto error; | |
81819f0f | 2378 | |
dfb4f096 | 2379 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2380 | return 1; |
4c93c355 | 2381 | free_kmem_cache_nodes(s); |
81819f0f CL |
2382 | error: |
2383 | if (flags & SLAB_PANIC) | |
2384 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2385 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2386 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2387 | s->offset, flags); |
2388 | return 0; | |
2389 | } | |
81819f0f CL |
2390 | |
2391 | /* | |
2392 | * Check if a given pointer is valid | |
2393 | */ | |
2394 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2395 | { | |
06428780 | 2396 | struct page *page; |
81819f0f CL |
2397 | |
2398 | page = get_object_page(object); | |
2399 | ||
2400 | if (!page || s != page->slab) | |
2401 | /* No slab or wrong slab */ | |
2402 | return 0; | |
2403 | ||
abcd08a6 | 2404 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2405 | return 0; |
2406 | ||
2407 | /* | |
2408 | * We could also check if the object is on the slabs freelist. | |
2409 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2410 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2411 | * to a certain slab. |
2412 | */ | |
2413 | return 1; | |
2414 | } | |
2415 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2416 | ||
2417 | /* | |
2418 | * Determine the size of a slab object | |
2419 | */ | |
2420 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2421 | { | |
2422 | return s->objsize; | |
2423 | } | |
2424 | EXPORT_SYMBOL(kmem_cache_size); | |
2425 | ||
2426 | const char *kmem_cache_name(struct kmem_cache *s) | |
2427 | { | |
2428 | return s->name; | |
2429 | } | |
2430 | EXPORT_SYMBOL(kmem_cache_name); | |
2431 | ||
33b12c38 CL |
2432 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2433 | const char *text) | |
2434 | { | |
2435 | #ifdef CONFIG_SLUB_DEBUG | |
2436 | void *addr = page_address(page); | |
2437 | void *p; | |
2438 | DECLARE_BITMAP(map, page->objects); | |
2439 | ||
2440 | bitmap_zero(map, page->objects); | |
2441 | slab_err(s, page, "%s", text); | |
2442 | slab_lock(page); | |
2443 | for_each_free_object(p, s, page->freelist) | |
2444 | set_bit(slab_index(p, s, addr), map); | |
2445 | ||
2446 | for_each_object(p, s, addr, page->objects) { | |
2447 | ||
2448 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2449 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2450 | p, p - addr); | |
2451 | print_tracking(s, p); | |
2452 | } | |
2453 | } | |
2454 | slab_unlock(page); | |
2455 | #endif | |
2456 | } | |
2457 | ||
81819f0f | 2458 | /* |
599870b1 | 2459 | * Attempt to free all partial slabs on a node. |
81819f0f | 2460 | */ |
599870b1 | 2461 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2462 | { |
81819f0f CL |
2463 | unsigned long flags; |
2464 | struct page *page, *h; | |
2465 | ||
2466 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2467 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2468 | if (!page->inuse) { |
2469 | list_del(&page->lru); | |
2470 | discard_slab(s, page); | |
599870b1 | 2471 | n->nr_partial--; |
33b12c38 CL |
2472 | } else { |
2473 | list_slab_objects(s, page, | |
2474 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2475 | } |
33b12c38 | 2476 | } |
81819f0f | 2477 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2478 | } |
2479 | ||
2480 | /* | |
672bba3a | 2481 | * Release all resources used by a slab cache. |
81819f0f | 2482 | */ |
0c710013 | 2483 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2484 | { |
2485 | int node; | |
2486 | ||
2487 | flush_all(s); | |
2488 | ||
2489 | /* Attempt to free all objects */ | |
4c93c355 | 2490 | free_kmem_cache_cpus(s); |
f64dc58c | 2491 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2492 | struct kmem_cache_node *n = get_node(s, node); |
2493 | ||
599870b1 CL |
2494 | free_partial(s, n); |
2495 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2496 | return 1; |
2497 | } | |
2498 | free_kmem_cache_nodes(s); | |
2499 | return 0; | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * Close a cache and release the kmem_cache structure | |
2504 | * (must be used for caches created using kmem_cache_create) | |
2505 | */ | |
2506 | void kmem_cache_destroy(struct kmem_cache *s) | |
2507 | { | |
2508 | down_write(&slub_lock); | |
2509 | s->refcount--; | |
2510 | if (!s->refcount) { | |
2511 | list_del(&s->list); | |
a0e1d1be | 2512 | up_write(&slub_lock); |
d629d819 PE |
2513 | if (kmem_cache_close(s)) { |
2514 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2515 | "still has objects.\n", s->name, __func__); | |
2516 | dump_stack(); | |
2517 | } | |
81819f0f | 2518 | sysfs_slab_remove(s); |
a0e1d1be CL |
2519 | } else |
2520 | up_write(&slub_lock); | |
81819f0f CL |
2521 | } |
2522 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2523 | ||
2524 | /******************************************************************** | |
2525 | * Kmalloc subsystem | |
2526 | *******************************************************************/ | |
2527 | ||
331dc558 | 2528 | struct kmem_cache kmalloc_caches[PAGE_SHIFT + 1] __cacheline_aligned; |
81819f0f CL |
2529 | EXPORT_SYMBOL(kmalloc_caches); |
2530 | ||
81819f0f CL |
2531 | static int __init setup_slub_min_order(char *str) |
2532 | { | |
06428780 | 2533 | get_option(&str, &slub_min_order); |
81819f0f CL |
2534 | |
2535 | return 1; | |
2536 | } | |
2537 | ||
2538 | __setup("slub_min_order=", setup_slub_min_order); | |
2539 | ||
2540 | static int __init setup_slub_max_order(char *str) | |
2541 | { | |
06428780 | 2542 | get_option(&str, &slub_max_order); |
81819f0f CL |
2543 | |
2544 | return 1; | |
2545 | } | |
2546 | ||
2547 | __setup("slub_max_order=", setup_slub_max_order); | |
2548 | ||
2549 | static int __init setup_slub_min_objects(char *str) | |
2550 | { | |
06428780 | 2551 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2552 | |
2553 | return 1; | |
2554 | } | |
2555 | ||
2556 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2557 | ||
2558 | static int __init setup_slub_nomerge(char *str) | |
2559 | { | |
2560 | slub_nomerge = 1; | |
2561 | return 1; | |
2562 | } | |
2563 | ||
2564 | __setup("slub_nomerge", setup_slub_nomerge); | |
2565 | ||
81819f0f CL |
2566 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2567 | const char *name, int size, gfp_t gfp_flags) | |
2568 | { | |
2569 | unsigned int flags = 0; | |
2570 | ||
2571 | if (gfp_flags & SLUB_DMA) | |
2572 | flags = SLAB_CACHE_DMA; | |
2573 | ||
2574 | down_write(&slub_lock); | |
2575 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
71c7a06f | 2576 | flags | __KMALLOC_CACHE, NULL)) |
81819f0f CL |
2577 | goto panic; |
2578 | ||
2579 | list_add(&s->list, &slab_caches); | |
2580 | up_write(&slub_lock); | |
2581 | if (sysfs_slab_add(s)) | |
2582 | goto panic; | |
2583 | return s; | |
2584 | ||
2585 | panic: | |
2586 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2587 | } | |
2588 | ||
2e443fd0 | 2589 | #ifdef CONFIG_ZONE_DMA |
4097d601 | 2590 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT + 1]; |
1ceef402 CL |
2591 | |
2592 | static void sysfs_add_func(struct work_struct *w) | |
2593 | { | |
2594 | struct kmem_cache *s; | |
2595 | ||
2596 | down_write(&slub_lock); | |
2597 | list_for_each_entry(s, &slab_caches, list) { | |
2598 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2599 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2600 | sysfs_slab_add(s); | |
2601 | } | |
2602 | } | |
2603 | up_write(&slub_lock); | |
2604 | } | |
2605 | ||
2606 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2607 | ||
2e443fd0 CL |
2608 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2609 | { | |
2610 | struct kmem_cache *s; | |
2e443fd0 CL |
2611 | char *text; |
2612 | size_t realsize; | |
2613 | ||
2614 | s = kmalloc_caches_dma[index]; | |
2615 | if (s) | |
2616 | return s; | |
2617 | ||
2618 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2619 | if (flags & __GFP_WAIT) |
2620 | down_write(&slub_lock); | |
2621 | else { | |
2622 | if (!down_write_trylock(&slub_lock)) | |
2623 | goto out; | |
2624 | } | |
2625 | ||
2626 | if (kmalloc_caches_dma[index]) | |
2627 | goto unlock_out; | |
2e443fd0 | 2628 | |
7b55f620 | 2629 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2630 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2631 | (unsigned int)realsize); | |
1ceef402 CL |
2632 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2633 | ||
2634 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2635 | realsize, ARCH_KMALLOC_MINALIGN, | |
2636 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2637 | kfree(s); | |
2638 | kfree(text); | |
2639 | goto unlock_out; | |
dfce8648 | 2640 | } |
1ceef402 CL |
2641 | |
2642 | list_add(&s->list, &slab_caches); | |
2643 | kmalloc_caches_dma[index] = s; | |
2644 | ||
2645 | schedule_work(&sysfs_add_work); | |
2646 | ||
2647 | unlock_out: | |
dfce8648 | 2648 | up_write(&slub_lock); |
1ceef402 | 2649 | out: |
dfce8648 | 2650 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2651 | } |
2652 | #endif | |
2653 | ||
f1b26339 CL |
2654 | /* |
2655 | * Conversion table for small slabs sizes / 8 to the index in the | |
2656 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2657 | * of two cache sizes there. The size of larger slabs can be determined using | |
2658 | * fls. | |
2659 | */ | |
2660 | static s8 size_index[24] = { | |
2661 | 3, /* 8 */ | |
2662 | 4, /* 16 */ | |
2663 | 5, /* 24 */ | |
2664 | 5, /* 32 */ | |
2665 | 6, /* 40 */ | |
2666 | 6, /* 48 */ | |
2667 | 6, /* 56 */ | |
2668 | 6, /* 64 */ | |
2669 | 1, /* 72 */ | |
2670 | 1, /* 80 */ | |
2671 | 1, /* 88 */ | |
2672 | 1, /* 96 */ | |
2673 | 7, /* 104 */ | |
2674 | 7, /* 112 */ | |
2675 | 7, /* 120 */ | |
2676 | 7, /* 128 */ | |
2677 | 2, /* 136 */ | |
2678 | 2, /* 144 */ | |
2679 | 2, /* 152 */ | |
2680 | 2, /* 160 */ | |
2681 | 2, /* 168 */ | |
2682 | 2, /* 176 */ | |
2683 | 2, /* 184 */ | |
2684 | 2 /* 192 */ | |
2685 | }; | |
2686 | ||
81819f0f CL |
2687 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2688 | { | |
f1b26339 | 2689 | int index; |
81819f0f | 2690 | |
f1b26339 CL |
2691 | if (size <= 192) { |
2692 | if (!size) | |
2693 | return ZERO_SIZE_PTR; | |
81819f0f | 2694 | |
f1b26339 | 2695 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2696 | } else |
f1b26339 | 2697 | index = fls(size - 1); |
81819f0f CL |
2698 | |
2699 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2700 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2701 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2702 | |
81819f0f CL |
2703 | #endif |
2704 | return &kmalloc_caches[index]; | |
2705 | } | |
2706 | ||
2707 | void *__kmalloc(size_t size, gfp_t flags) | |
2708 | { | |
aadb4bc4 | 2709 | struct kmem_cache *s; |
81819f0f | 2710 | |
331dc558 | 2711 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2712 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2713 | |
2714 | s = get_slab(size, flags); | |
2715 | ||
2716 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2717 | return s; |
2718 | ||
ce15fea8 | 2719 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2720 | } |
2721 | EXPORT_SYMBOL(__kmalloc); | |
2722 | ||
f619cfe1 CL |
2723 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2724 | { | |
2725 | struct page *page = alloc_pages_node(node, flags | __GFP_COMP, | |
2726 | get_order(size)); | |
2727 | ||
2728 | if (page) | |
2729 | return page_address(page); | |
2730 | else | |
2731 | return NULL; | |
2732 | } | |
2733 | ||
81819f0f CL |
2734 | #ifdef CONFIG_NUMA |
2735 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2736 | { | |
aadb4bc4 | 2737 | struct kmem_cache *s; |
81819f0f | 2738 | |
331dc558 | 2739 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 2740 | return kmalloc_large_node(size, flags, node); |
aadb4bc4 CL |
2741 | |
2742 | s = get_slab(size, flags); | |
2743 | ||
2744 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2745 | return s; |
2746 | ||
ce15fea8 | 2747 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2748 | } |
2749 | EXPORT_SYMBOL(__kmalloc_node); | |
2750 | #endif | |
2751 | ||
2752 | size_t ksize(const void *object) | |
2753 | { | |
272c1d21 | 2754 | struct page *page; |
81819f0f CL |
2755 | struct kmem_cache *s; |
2756 | ||
ef8b4520 | 2757 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2758 | return 0; |
2759 | ||
294a80a8 | 2760 | page = virt_to_head_page(object); |
294a80a8 VN |
2761 | |
2762 | if (unlikely(!PageSlab(page))) | |
2763 | return PAGE_SIZE << compound_order(page); | |
2764 | ||
81819f0f | 2765 | s = page->slab; |
81819f0f | 2766 | |
ae20bfda | 2767 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2768 | /* |
2769 | * Debugging requires use of the padding between object | |
2770 | * and whatever may come after it. | |
2771 | */ | |
2772 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2773 | return s->objsize; | |
2774 | ||
ae20bfda | 2775 | #endif |
81819f0f CL |
2776 | /* |
2777 | * If we have the need to store the freelist pointer | |
2778 | * back there or track user information then we can | |
2779 | * only use the space before that information. | |
2780 | */ | |
2781 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2782 | return s->inuse; | |
81819f0f CL |
2783 | /* |
2784 | * Else we can use all the padding etc for the allocation | |
2785 | */ | |
2786 | return s->size; | |
2787 | } | |
2788 | EXPORT_SYMBOL(ksize); | |
2789 | ||
2790 | void kfree(const void *x) | |
2791 | { | |
81819f0f | 2792 | struct page *page; |
5bb983b0 | 2793 | void *object = (void *)x; |
81819f0f | 2794 | |
2408c550 | 2795 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2796 | return; |
2797 | ||
b49af68f | 2798 | page = virt_to_head_page(x); |
aadb4bc4 CL |
2799 | if (unlikely(!PageSlab(page))) { |
2800 | put_page(page); | |
2801 | return; | |
2802 | } | |
5bb983b0 | 2803 | slab_free(page->slab, page, object, __builtin_return_address(0)); |
81819f0f CL |
2804 | } |
2805 | EXPORT_SYMBOL(kfree); | |
2806 | ||
2086d26a | 2807 | /* |
672bba3a CL |
2808 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2809 | * the remaining slabs by the number of items in use. The slabs with the | |
2810 | * most items in use come first. New allocations will then fill those up | |
2811 | * and thus they can be removed from the partial lists. | |
2812 | * | |
2813 | * The slabs with the least items are placed last. This results in them | |
2814 | * being allocated from last increasing the chance that the last objects | |
2815 | * are freed in them. | |
2086d26a CL |
2816 | */ |
2817 | int kmem_cache_shrink(struct kmem_cache *s) | |
2818 | { | |
2819 | int node; | |
2820 | int i; | |
2821 | struct kmem_cache_node *n; | |
2822 | struct page *page; | |
2823 | struct page *t; | |
205ab99d | 2824 | int objects = oo_objects(s->max); |
2086d26a | 2825 | struct list_head *slabs_by_inuse = |
834f3d11 | 2826 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2827 | unsigned long flags; |
2828 | ||
2829 | if (!slabs_by_inuse) | |
2830 | return -ENOMEM; | |
2831 | ||
2832 | flush_all(s); | |
f64dc58c | 2833 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2834 | n = get_node(s, node); |
2835 | ||
2836 | if (!n->nr_partial) | |
2837 | continue; | |
2838 | ||
834f3d11 | 2839 | for (i = 0; i < objects; i++) |
2086d26a CL |
2840 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2841 | ||
2842 | spin_lock_irqsave(&n->list_lock, flags); | |
2843 | ||
2844 | /* | |
672bba3a | 2845 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2846 | * |
672bba3a CL |
2847 | * Note that concurrent frees may occur while we hold the |
2848 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2849 | */ |
2850 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2851 | if (!page->inuse && slab_trylock(page)) { | |
2852 | /* | |
2853 | * Must hold slab lock here because slab_free | |
2854 | * may have freed the last object and be | |
2855 | * waiting to release the slab. | |
2856 | */ | |
2857 | list_del(&page->lru); | |
2858 | n->nr_partial--; | |
2859 | slab_unlock(page); | |
2860 | discard_slab(s, page); | |
2861 | } else { | |
fcda3d89 CL |
2862 | list_move(&page->lru, |
2863 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2864 | } |
2865 | } | |
2866 | ||
2086d26a | 2867 | /* |
672bba3a CL |
2868 | * Rebuild the partial list with the slabs filled up most |
2869 | * first and the least used slabs at the end. | |
2086d26a | 2870 | */ |
834f3d11 | 2871 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2872 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2873 | ||
2086d26a CL |
2874 | spin_unlock_irqrestore(&n->list_lock, flags); |
2875 | } | |
2876 | ||
2877 | kfree(slabs_by_inuse); | |
2878 | return 0; | |
2879 | } | |
2880 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2881 | ||
b9049e23 YG |
2882 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2883 | static int slab_mem_going_offline_callback(void *arg) | |
2884 | { | |
2885 | struct kmem_cache *s; | |
2886 | ||
2887 | down_read(&slub_lock); | |
2888 | list_for_each_entry(s, &slab_caches, list) | |
2889 | kmem_cache_shrink(s); | |
2890 | up_read(&slub_lock); | |
2891 | ||
2892 | return 0; | |
2893 | } | |
2894 | ||
2895 | static void slab_mem_offline_callback(void *arg) | |
2896 | { | |
2897 | struct kmem_cache_node *n; | |
2898 | struct kmem_cache *s; | |
2899 | struct memory_notify *marg = arg; | |
2900 | int offline_node; | |
2901 | ||
2902 | offline_node = marg->status_change_nid; | |
2903 | ||
2904 | /* | |
2905 | * If the node still has available memory. we need kmem_cache_node | |
2906 | * for it yet. | |
2907 | */ | |
2908 | if (offline_node < 0) | |
2909 | return; | |
2910 | ||
2911 | down_read(&slub_lock); | |
2912 | list_for_each_entry(s, &slab_caches, list) { | |
2913 | n = get_node(s, offline_node); | |
2914 | if (n) { | |
2915 | /* | |
2916 | * if n->nr_slabs > 0, slabs still exist on the node | |
2917 | * that is going down. We were unable to free them, | |
2918 | * and offline_pages() function shoudn't call this | |
2919 | * callback. So, we must fail. | |
2920 | */ | |
0f389ec6 | 2921 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2922 | |
2923 | s->node[offline_node] = NULL; | |
2924 | kmem_cache_free(kmalloc_caches, n); | |
2925 | } | |
2926 | } | |
2927 | up_read(&slub_lock); | |
2928 | } | |
2929 | ||
2930 | static int slab_mem_going_online_callback(void *arg) | |
2931 | { | |
2932 | struct kmem_cache_node *n; | |
2933 | struct kmem_cache *s; | |
2934 | struct memory_notify *marg = arg; | |
2935 | int nid = marg->status_change_nid; | |
2936 | int ret = 0; | |
2937 | ||
2938 | /* | |
2939 | * If the node's memory is already available, then kmem_cache_node is | |
2940 | * already created. Nothing to do. | |
2941 | */ | |
2942 | if (nid < 0) | |
2943 | return 0; | |
2944 | ||
2945 | /* | |
2946 | * We are bringing a node online. No memory is availabe yet. We must | |
2947 | * allocate a kmem_cache_node structure in order to bring the node | |
2948 | * online. | |
2949 | */ | |
2950 | down_read(&slub_lock); | |
2951 | list_for_each_entry(s, &slab_caches, list) { | |
2952 | /* | |
2953 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2954 | * since memory is not yet available from the node that | |
2955 | * is brought up. | |
2956 | */ | |
2957 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2958 | if (!n) { | |
2959 | ret = -ENOMEM; | |
2960 | goto out; | |
2961 | } | |
2962 | init_kmem_cache_node(n); | |
2963 | s->node[nid] = n; | |
2964 | } | |
2965 | out: | |
2966 | up_read(&slub_lock); | |
2967 | return ret; | |
2968 | } | |
2969 | ||
2970 | static int slab_memory_callback(struct notifier_block *self, | |
2971 | unsigned long action, void *arg) | |
2972 | { | |
2973 | int ret = 0; | |
2974 | ||
2975 | switch (action) { | |
2976 | case MEM_GOING_ONLINE: | |
2977 | ret = slab_mem_going_online_callback(arg); | |
2978 | break; | |
2979 | case MEM_GOING_OFFLINE: | |
2980 | ret = slab_mem_going_offline_callback(arg); | |
2981 | break; | |
2982 | case MEM_OFFLINE: | |
2983 | case MEM_CANCEL_ONLINE: | |
2984 | slab_mem_offline_callback(arg); | |
2985 | break; | |
2986 | case MEM_ONLINE: | |
2987 | case MEM_CANCEL_OFFLINE: | |
2988 | break; | |
2989 | } | |
2990 | ||
2991 | ret = notifier_from_errno(ret); | |
2992 | return ret; | |
2993 | } | |
2994 | ||
2995 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2996 | ||
81819f0f CL |
2997 | /******************************************************************** |
2998 | * Basic setup of slabs | |
2999 | *******************************************************************/ | |
3000 | ||
3001 | void __init kmem_cache_init(void) | |
3002 | { | |
3003 | int i; | |
4b356be0 | 3004 | int caches = 0; |
81819f0f | 3005 | |
4c93c355 CL |
3006 | init_alloc_cpu(); |
3007 | ||
81819f0f CL |
3008 | #ifdef CONFIG_NUMA |
3009 | /* | |
3010 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3011 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3012 | * kmem_cache_open for slab_state == DOWN. |
3013 | */ | |
3014 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
3015 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 3016 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3017 | caches++; |
b9049e23 YG |
3018 | |
3019 | hotplug_memory_notifier(slab_memory_callback, 1); | |
81819f0f CL |
3020 | #endif |
3021 | ||
3022 | /* Able to allocate the per node structures */ | |
3023 | slab_state = PARTIAL; | |
3024 | ||
3025 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
3026 | if (KMALLOC_MIN_SIZE <= 64) { |
3027 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 3028 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 CL |
3029 | caches++; |
3030 | } | |
3031 | if (KMALLOC_MIN_SIZE <= 128) { | |
3032 | create_kmalloc_cache(&kmalloc_caches[2], | |
81819f0f | 3033 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
3034 | caches++; |
3035 | } | |
81819f0f | 3036 | |
331dc558 | 3037 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) { |
81819f0f CL |
3038 | create_kmalloc_cache(&kmalloc_caches[i], |
3039 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
3040 | caches++; |
3041 | } | |
81819f0f | 3042 | |
f1b26339 CL |
3043 | |
3044 | /* | |
3045 | * Patch up the size_index table if we have strange large alignment | |
3046 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3047 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3048 | * |
3049 | * Largest permitted alignment is 256 bytes due to the way we | |
3050 | * handle the index determination for the smaller caches. | |
3051 | * | |
3052 | * Make sure that nothing crazy happens if someone starts tinkering | |
3053 | * around with ARCH_KMALLOC_MINALIGN | |
3054 | */ | |
3055 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3056 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3057 | ||
12ad6843 | 3058 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3059 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3060 | ||
81819f0f CL |
3061 | slab_state = UP; |
3062 | ||
3063 | /* Provide the correct kmalloc names now that the caches are up */ | |
331dc558 | 3064 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) |
81819f0f CL |
3065 | kmalloc_caches[i]. name = |
3066 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3067 | ||
3068 | #ifdef CONFIG_SMP | |
3069 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3070 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3071 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3072 | #else | |
3073 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3074 | #endif |
3075 | ||
3adbefee IM |
3076 | printk(KERN_INFO |
3077 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3078 | " CPUs=%d, Nodes=%d\n", |
3079 | caches, cache_line_size(), | |
81819f0f CL |
3080 | slub_min_order, slub_max_order, slub_min_objects, |
3081 | nr_cpu_ids, nr_node_ids); | |
3082 | } | |
3083 | ||
3084 | /* | |
3085 | * Find a mergeable slab cache | |
3086 | */ | |
3087 | static int slab_unmergeable(struct kmem_cache *s) | |
3088 | { | |
3089 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3090 | return 1; | |
3091 | ||
331dc558 | 3092 | if ((s->flags & __PAGE_ALLOC_FALLBACK)) |
71c7a06f CL |
3093 | return 1; |
3094 | ||
c59def9f | 3095 | if (s->ctor) |
81819f0f CL |
3096 | return 1; |
3097 | ||
8ffa6875 CL |
3098 | /* |
3099 | * We may have set a slab to be unmergeable during bootstrap. | |
3100 | */ | |
3101 | if (s->refcount < 0) | |
3102 | return 1; | |
3103 | ||
81819f0f CL |
3104 | return 0; |
3105 | } | |
3106 | ||
3107 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3108 | size_t align, unsigned long flags, const char *name, |
4ba9b9d0 | 3109 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f | 3110 | { |
5b95a4ac | 3111 | struct kmem_cache *s; |
81819f0f CL |
3112 | |
3113 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3114 | return NULL; | |
3115 | ||
c59def9f | 3116 | if (ctor) |
81819f0f CL |
3117 | return NULL; |
3118 | ||
3119 | size = ALIGN(size, sizeof(void *)); | |
3120 | align = calculate_alignment(flags, align, size); | |
3121 | size = ALIGN(size, align); | |
ba0268a8 | 3122 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3123 | |
5b95a4ac | 3124 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3125 | if (slab_unmergeable(s)) |
3126 | continue; | |
3127 | ||
3128 | if (size > s->size) | |
3129 | continue; | |
3130 | ||
ba0268a8 | 3131 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3132 | continue; |
3133 | /* | |
3134 | * Check if alignment is compatible. | |
3135 | * Courtesy of Adrian Drzewiecki | |
3136 | */ | |
06428780 | 3137 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3138 | continue; |
3139 | ||
3140 | if (s->size - size >= sizeof(void *)) | |
3141 | continue; | |
3142 | ||
3143 | return s; | |
3144 | } | |
3145 | return NULL; | |
3146 | } | |
3147 | ||
3148 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
3149 | size_t align, unsigned long flags, | |
4ba9b9d0 | 3150 | void (*ctor)(struct kmem_cache *, void *)) |
81819f0f CL |
3151 | { |
3152 | struct kmem_cache *s; | |
3153 | ||
3154 | down_write(&slub_lock); | |
ba0268a8 | 3155 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3156 | if (s) { |
42a9fdbb CL |
3157 | int cpu; |
3158 | ||
81819f0f CL |
3159 | s->refcount++; |
3160 | /* | |
3161 | * Adjust the object sizes so that we clear | |
3162 | * the complete object on kzalloc. | |
3163 | */ | |
3164 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3165 | |
3166 | /* | |
3167 | * And then we need to update the object size in the | |
3168 | * per cpu structures | |
3169 | */ | |
3170 | for_each_online_cpu(cpu) | |
3171 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3172 | |
81819f0f | 3173 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3174 | up_write(&slub_lock); |
6446faa2 | 3175 | |
81819f0f CL |
3176 | if (sysfs_slab_alias(s, name)) |
3177 | goto err; | |
a0e1d1be CL |
3178 | return s; |
3179 | } | |
6446faa2 | 3180 | |
a0e1d1be CL |
3181 | s = kmalloc(kmem_size, GFP_KERNEL); |
3182 | if (s) { | |
3183 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3184 | size, align, flags, ctor)) { |
81819f0f | 3185 | list_add(&s->list, &slab_caches); |
a0e1d1be CL |
3186 | up_write(&slub_lock); |
3187 | if (sysfs_slab_add(s)) | |
3188 | goto err; | |
3189 | return s; | |
3190 | } | |
3191 | kfree(s); | |
81819f0f CL |
3192 | } |
3193 | up_write(&slub_lock); | |
81819f0f CL |
3194 | |
3195 | err: | |
81819f0f CL |
3196 | if (flags & SLAB_PANIC) |
3197 | panic("Cannot create slabcache %s\n", name); | |
3198 | else | |
3199 | s = NULL; | |
3200 | return s; | |
3201 | } | |
3202 | EXPORT_SYMBOL(kmem_cache_create); | |
3203 | ||
81819f0f | 3204 | #ifdef CONFIG_SMP |
81819f0f | 3205 | /* |
672bba3a CL |
3206 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3207 | * necessary. | |
81819f0f CL |
3208 | */ |
3209 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3210 | unsigned long action, void *hcpu) | |
3211 | { | |
3212 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3213 | struct kmem_cache *s; |
3214 | unsigned long flags; | |
81819f0f CL |
3215 | |
3216 | switch (action) { | |
4c93c355 CL |
3217 | case CPU_UP_PREPARE: |
3218 | case CPU_UP_PREPARE_FROZEN: | |
3219 | init_alloc_cpu_cpu(cpu); | |
3220 | down_read(&slub_lock); | |
3221 | list_for_each_entry(s, &slab_caches, list) | |
3222 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3223 | GFP_KERNEL); | |
3224 | up_read(&slub_lock); | |
3225 | break; | |
3226 | ||
81819f0f | 3227 | case CPU_UP_CANCELED: |
8bb78442 | 3228 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3229 | case CPU_DEAD: |
8bb78442 | 3230 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3231 | down_read(&slub_lock); |
3232 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3233 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3234 | ||
5b95a4ac CL |
3235 | local_irq_save(flags); |
3236 | __flush_cpu_slab(s, cpu); | |
3237 | local_irq_restore(flags); | |
4c93c355 CL |
3238 | free_kmem_cache_cpu(c, cpu); |
3239 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3240 | } |
3241 | up_read(&slub_lock); | |
81819f0f CL |
3242 | break; |
3243 | default: | |
3244 | break; | |
3245 | } | |
3246 | return NOTIFY_OK; | |
3247 | } | |
3248 | ||
06428780 | 3249 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3250 | .notifier_call = slab_cpuup_callback |
06428780 | 3251 | }; |
81819f0f CL |
3252 | |
3253 | #endif | |
3254 | ||
81819f0f CL |
3255 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) |
3256 | { | |
aadb4bc4 CL |
3257 | struct kmem_cache *s; |
3258 | ||
331dc558 | 3259 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3260 | return kmalloc_large(size, gfpflags); |
3261 | ||
aadb4bc4 | 3262 | s = get_slab(size, gfpflags); |
81819f0f | 3263 | |
2408c550 | 3264 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3265 | return s; |
81819f0f | 3266 | |
ce15fea8 | 3267 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3268 | } |
3269 | ||
3270 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
3271 | int node, void *caller) | |
3272 | { | |
aadb4bc4 CL |
3273 | struct kmem_cache *s; |
3274 | ||
331dc558 | 3275 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 3276 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3277 | |
aadb4bc4 | 3278 | s = get_slab(size, gfpflags); |
81819f0f | 3279 | |
2408c550 | 3280 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3281 | return s; |
81819f0f | 3282 | |
ce15fea8 | 3283 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3284 | } |
3285 | ||
5b06c853 | 3286 | #if (defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)) || defined(CONFIG_SLABINFO) |
205ab99d CL |
3287 | static unsigned long count_partial(struct kmem_cache_node *n, |
3288 | int (*get_count)(struct page *)) | |
5b06c853 CL |
3289 | { |
3290 | unsigned long flags; | |
3291 | unsigned long x = 0; | |
3292 | struct page *page; | |
3293 | ||
3294 | spin_lock_irqsave(&n->list_lock, flags); | |
3295 | list_for_each_entry(page, &n->partial, lru) | |
205ab99d | 3296 | x += get_count(page); |
5b06c853 CL |
3297 | spin_unlock_irqrestore(&n->list_lock, flags); |
3298 | return x; | |
3299 | } | |
205ab99d CL |
3300 | |
3301 | static int count_inuse(struct page *page) | |
3302 | { | |
3303 | return page->inuse; | |
3304 | } | |
3305 | ||
3306 | static int count_total(struct page *page) | |
3307 | { | |
3308 | return page->objects; | |
3309 | } | |
3310 | ||
3311 | static int count_free(struct page *page) | |
3312 | { | |
3313 | return page->objects - page->inuse; | |
3314 | } | |
5b06c853 CL |
3315 | #endif |
3316 | ||
41ecc55b | 3317 | #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG) |
434e245d CL |
3318 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3319 | unsigned long *map) | |
53e15af0 CL |
3320 | { |
3321 | void *p; | |
a973e9dd | 3322 | void *addr = page_address(page); |
53e15af0 CL |
3323 | |
3324 | if (!check_slab(s, page) || | |
3325 | !on_freelist(s, page, NULL)) | |
3326 | return 0; | |
3327 | ||
3328 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3329 | bitmap_zero(map, page->objects); |
53e15af0 | 3330 | |
7656c72b CL |
3331 | for_each_free_object(p, s, page->freelist) { |
3332 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3333 | if (!check_object(s, page, p, 0)) |
3334 | return 0; | |
3335 | } | |
3336 | ||
224a88be | 3337 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3338 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3339 | if (!check_object(s, page, p, 1)) |
3340 | return 0; | |
3341 | return 1; | |
3342 | } | |
3343 | ||
434e245d CL |
3344 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3345 | unsigned long *map) | |
53e15af0 CL |
3346 | { |
3347 | if (slab_trylock(page)) { | |
434e245d | 3348 | validate_slab(s, page, map); |
53e15af0 CL |
3349 | slab_unlock(page); |
3350 | } else | |
3351 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3352 | s->name, page); | |
3353 | ||
3354 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
35e5d7ee CL |
3355 | if (!SlabDebug(page)) |
3356 | printk(KERN_ERR "SLUB %s: SlabDebug not set " | |
53e15af0 CL |
3357 | "on slab 0x%p\n", s->name, page); |
3358 | } else { | |
35e5d7ee CL |
3359 | if (SlabDebug(page)) |
3360 | printk(KERN_ERR "SLUB %s: SlabDebug set on " | |
53e15af0 CL |
3361 | "slab 0x%p\n", s->name, page); |
3362 | } | |
3363 | } | |
3364 | ||
434e245d CL |
3365 | static int validate_slab_node(struct kmem_cache *s, |
3366 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3367 | { |
3368 | unsigned long count = 0; | |
3369 | struct page *page; | |
3370 | unsigned long flags; | |
3371 | ||
3372 | spin_lock_irqsave(&n->list_lock, flags); | |
3373 | ||
3374 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3375 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3376 | count++; |
3377 | } | |
3378 | if (count != n->nr_partial) | |
3379 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3380 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3381 | ||
3382 | if (!(s->flags & SLAB_STORE_USER)) | |
3383 | goto out; | |
3384 | ||
3385 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3386 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3387 | count++; |
3388 | } | |
3389 | if (count != atomic_long_read(&n->nr_slabs)) | |
3390 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3391 | "counter=%ld\n", s->name, count, | |
3392 | atomic_long_read(&n->nr_slabs)); | |
3393 | ||
3394 | out: | |
3395 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3396 | return count; | |
3397 | } | |
3398 | ||
434e245d | 3399 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3400 | { |
3401 | int node; | |
3402 | unsigned long count = 0; | |
205ab99d | 3403 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3404 | sizeof(unsigned long), GFP_KERNEL); |
3405 | ||
3406 | if (!map) | |
3407 | return -ENOMEM; | |
53e15af0 CL |
3408 | |
3409 | flush_all(s); | |
f64dc58c | 3410 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3411 | struct kmem_cache_node *n = get_node(s, node); |
3412 | ||
434e245d | 3413 | count += validate_slab_node(s, n, map); |
53e15af0 | 3414 | } |
434e245d | 3415 | kfree(map); |
53e15af0 CL |
3416 | return count; |
3417 | } | |
3418 | ||
b3459709 CL |
3419 | #ifdef SLUB_RESILIENCY_TEST |
3420 | static void resiliency_test(void) | |
3421 | { | |
3422 | u8 *p; | |
3423 | ||
3424 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3425 | printk(KERN_ERR "-----------------------\n"); | |
3426 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3427 | ||
3428 | p = kzalloc(16, GFP_KERNEL); | |
3429 | p[16] = 0x12; | |
3430 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3431 | " 0x12->0x%p\n\n", p + 16); | |
3432 | ||
3433 | validate_slab_cache(kmalloc_caches + 4); | |
3434 | ||
3435 | /* Hmmm... The next two are dangerous */ | |
3436 | p = kzalloc(32, GFP_KERNEL); | |
3437 | p[32 + sizeof(void *)] = 0x34; | |
3438 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3439 | " 0x34 -> -0x%p\n", p); |
3440 | printk(KERN_ERR | |
3441 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3442 | |
3443 | validate_slab_cache(kmalloc_caches + 5); | |
3444 | p = kzalloc(64, GFP_KERNEL); | |
3445 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3446 | *p = 0x56; | |
3447 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3448 | p); | |
3adbefee IM |
3449 | printk(KERN_ERR |
3450 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3451 | validate_slab_cache(kmalloc_caches + 6); |
3452 | ||
3453 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3454 | p = kzalloc(128, GFP_KERNEL); | |
3455 | kfree(p); | |
3456 | *p = 0x78; | |
3457 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3458 | validate_slab_cache(kmalloc_caches + 7); | |
3459 | ||
3460 | p = kzalloc(256, GFP_KERNEL); | |
3461 | kfree(p); | |
3462 | p[50] = 0x9a; | |
3adbefee IM |
3463 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3464 | p); | |
b3459709 CL |
3465 | validate_slab_cache(kmalloc_caches + 8); |
3466 | ||
3467 | p = kzalloc(512, GFP_KERNEL); | |
3468 | kfree(p); | |
3469 | p[512] = 0xab; | |
3470 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3471 | validate_slab_cache(kmalloc_caches + 9); | |
3472 | } | |
3473 | #else | |
3474 | static void resiliency_test(void) {}; | |
3475 | #endif | |
3476 | ||
88a420e4 | 3477 | /* |
672bba3a | 3478 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3479 | * and freed. |
3480 | */ | |
3481 | ||
3482 | struct location { | |
3483 | unsigned long count; | |
3484 | void *addr; | |
45edfa58 CL |
3485 | long long sum_time; |
3486 | long min_time; | |
3487 | long max_time; | |
3488 | long min_pid; | |
3489 | long max_pid; | |
3490 | cpumask_t cpus; | |
3491 | nodemask_t nodes; | |
88a420e4 CL |
3492 | }; |
3493 | ||
3494 | struct loc_track { | |
3495 | unsigned long max; | |
3496 | unsigned long count; | |
3497 | struct location *loc; | |
3498 | }; | |
3499 | ||
3500 | static void free_loc_track(struct loc_track *t) | |
3501 | { | |
3502 | if (t->max) | |
3503 | free_pages((unsigned long)t->loc, | |
3504 | get_order(sizeof(struct location) * t->max)); | |
3505 | } | |
3506 | ||
68dff6a9 | 3507 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3508 | { |
3509 | struct location *l; | |
3510 | int order; | |
3511 | ||
88a420e4 CL |
3512 | order = get_order(sizeof(struct location) * max); |
3513 | ||
68dff6a9 | 3514 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3515 | if (!l) |
3516 | return 0; | |
3517 | ||
3518 | if (t->count) { | |
3519 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3520 | free_loc_track(t); | |
3521 | } | |
3522 | t->max = max; | |
3523 | t->loc = l; | |
3524 | return 1; | |
3525 | } | |
3526 | ||
3527 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3528 | const struct track *track) |
88a420e4 CL |
3529 | { |
3530 | long start, end, pos; | |
3531 | struct location *l; | |
3532 | void *caddr; | |
45edfa58 | 3533 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3534 | |
3535 | start = -1; | |
3536 | end = t->count; | |
3537 | ||
3538 | for ( ; ; ) { | |
3539 | pos = start + (end - start + 1) / 2; | |
3540 | ||
3541 | /* | |
3542 | * There is nothing at "end". If we end up there | |
3543 | * we need to add something to before end. | |
3544 | */ | |
3545 | if (pos == end) | |
3546 | break; | |
3547 | ||
3548 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3549 | if (track->addr == caddr) { |
3550 | ||
3551 | l = &t->loc[pos]; | |
3552 | l->count++; | |
3553 | if (track->when) { | |
3554 | l->sum_time += age; | |
3555 | if (age < l->min_time) | |
3556 | l->min_time = age; | |
3557 | if (age > l->max_time) | |
3558 | l->max_time = age; | |
3559 | ||
3560 | if (track->pid < l->min_pid) | |
3561 | l->min_pid = track->pid; | |
3562 | if (track->pid > l->max_pid) | |
3563 | l->max_pid = track->pid; | |
3564 | ||
3565 | cpu_set(track->cpu, l->cpus); | |
3566 | } | |
3567 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3568 | return 1; |
3569 | } | |
3570 | ||
45edfa58 | 3571 | if (track->addr < caddr) |
88a420e4 CL |
3572 | end = pos; |
3573 | else | |
3574 | start = pos; | |
3575 | } | |
3576 | ||
3577 | /* | |
672bba3a | 3578 | * Not found. Insert new tracking element. |
88a420e4 | 3579 | */ |
68dff6a9 | 3580 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3581 | return 0; |
3582 | ||
3583 | l = t->loc + pos; | |
3584 | if (pos < t->count) | |
3585 | memmove(l + 1, l, | |
3586 | (t->count - pos) * sizeof(struct location)); | |
3587 | t->count++; | |
3588 | l->count = 1; | |
45edfa58 CL |
3589 | l->addr = track->addr; |
3590 | l->sum_time = age; | |
3591 | l->min_time = age; | |
3592 | l->max_time = age; | |
3593 | l->min_pid = track->pid; | |
3594 | l->max_pid = track->pid; | |
3595 | cpus_clear(l->cpus); | |
3596 | cpu_set(track->cpu, l->cpus); | |
3597 | nodes_clear(l->nodes); | |
3598 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3599 | return 1; |
3600 | } | |
3601 | ||
3602 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3603 | struct page *page, enum track_item alloc) | |
3604 | { | |
a973e9dd | 3605 | void *addr = page_address(page); |
39b26464 | 3606 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3607 | void *p; |
3608 | ||
39b26464 | 3609 | bitmap_zero(map, page->objects); |
7656c72b CL |
3610 | for_each_free_object(p, s, page->freelist) |
3611 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3612 | |
224a88be | 3613 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3614 | if (!test_bit(slab_index(p, s, addr), map)) |
3615 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3616 | } |
3617 | ||
3618 | static int list_locations(struct kmem_cache *s, char *buf, | |
3619 | enum track_item alloc) | |
3620 | { | |
e374d483 | 3621 | int len = 0; |
88a420e4 | 3622 | unsigned long i; |
68dff6a9 | 3623 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3624 | int node; |
3625 | ||
68dff6a9 | 3626 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3627 | GFP_TEMPORARY)) |
68dff6a9 | 3628 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3629 | |
3630 | /* Push back cpu slabs */ | |
3631 | flush_all(s); | |
3632 | ||
f64dc58c | 3633 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3634 | struct kmem_cache_node *n = get_node(s, node); |
3635 | unsigned long flags; | |
3636 | struct page *page; | |
3637 | ||
9e86943b | 3638 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3639 | continue; |
3640 | ||
3641 | spin_lock_irqsave(&n->list_lock, flags); | |
3642 | list_for_each_entry(page, &n->partial, lru) | |
3643 | process_slab(&t, s, page, alloc); | |
3644 | list_for_each_entry(page, &n->full, lru) | |
3645 | process_slab(&t, s, page, alloc); | |
3646 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3647 | } | |
3648 | ||
3649 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3650 | struct location *l = &t.loc[i]; |
88a420e4 | 3651 | |
e374d483 | 3652 | if (len > PAGE_SIZE - 100) |
88a420e4 | 3653 | break; |
e374d483 | 3654 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3655 | |
3656 | if (l->addr) | |
e374d483 | 3657 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3658 | else |
e374d483 | 3659 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3660 | |
3661 | if (l->sum_time != l->min_time) { | |
3662 | unsigned long remainder; | |
3663 | ||
e374d483 | 3664 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
45edfa58 CL |
3665 | l->min_time, |
3666 | div_long_long_rem(l->sum_time, l->count, &remainder), | |
3667 | l->max_time); | |
3668 | } else | |
e374d483 | 3669 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3670 | l->min_time); |
3671 | ||
3672 | if (l->min_pid != l->max_pid) | |
e374d483 | 3673 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3674 | l->min_pid, l->max_pid); |
3675 | else | |
e374d483 | 3676 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3677 | l->min_pid); |
3678 | ||
84966343 | 3679 | if (num_online_cpus() > 1 && !cpus_empty(l->cpus) && |
e374d483 HH |
3680 | len < PAGE_SIZE - 60) { |
3681 | len += sprintf(buf + len, " cpus="); | |
3682 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3683 | l->cpus); |
3684 | } | |
3685 | ||
84966343 | 3686 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3687 | len < PAGE_SIZE - 60) { |
3688 | len += sprintf(buf + len, " nodes="); | |
3689 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3690 | l->nodes); |
3691 | } | |
3692 | ||
e374d483 | 3693 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3694 | } |
3695 | ||
3696 | free_loc_track(&t); | |
3697 | if (!t.count) | |
e374d483 HH |
3698 | len += sprintf(buf, "No data\n"); |
3699 | return len; | |
88a420e4 CL |
3700 | } |
3701 | ||
81819f0f | 3702 | enum slab_stat_type { |
205ab99d CL |
3703 | SL_ALL, /* All slabs */ |
3704 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3705 | SL_CPU, /* Only slabs used for cpu caches */ | |
3706 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3707 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3708 | }; |
3709 | ||
205ab99d | 3710 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3711 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3712 | #define SO_CPU (1 << SL_CPU) | |
3713 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3714 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3715 | |
62e5c4b4 CG |
3716 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3717 | char *buf, unsigned long flags) | |
81819f0f CL |
3718 | { |
3719 | unsigned long total = 0; | |
81819f0f CL |
3720 | int node; |
3721 | int x; | |
3722 | unsigned long *nodes; | |
3723 | unsigned long *per_cpu; | |
3724 | ||
3725 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3726 | if (!nodes) |
3727 | return -ENOMEM; | |
81819f0f CL |
3728 | per_cpu = nodes + nr_node_ids; |
3729 | ||
205ab99d CL |
3730 | if (flags & SO_CPU) { |
3731 | int cpu; | |
81819f0f | 3732 | |
205ab99d CL |
3733 | for_each_possible_cpu(cpu) { |
3734 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3735 | |
205ab99d CL |
3736 | if (!c || c->node < 0) |
3737 | continue; | |
3738 | ||
3739 | if (c->page) { | |
3740 | if (flags & SO_TOTAL) | |
3741 | x = c->page->objects; | |
3742 | else if (flags & SO_OBJECTS) | |
3743 | x = c->page->inuse; | |
81819f0f CL |
3744 | else |
3745 | x = 1; | |
205ab99d | 3746 | |
81819f0f | 3747 | total += x; |
205ab99d | 3748 | nodes[c->node] += x; |
81819f0f | 3749 | } |
205ab99d | 3750 | per_cpu[c->node]++; |
81819f0f CL |
3751 | } |
3752 | } | |
3753 | ||
205ab99d CL |
3754 | if (flags & SO_ALL) { |
3755 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3756 | struct kmem_cache_node *n = get_node(s, node); | |
3757 | ||
3758 | if (flags & SO_TOTAL) | |
3759 | x = atomic_long_read(&n->total_objects); | |
3760 | else if (flags & SO_OBJECTS) | |
3761 | x = atomic_long_read(&n->total_objects) - | |
3762 | count_partial(n, count_free); | |
81819f0f | 3763 | |
81819f0f | 3764 | else |
205ab99d | 3765 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3766 | total += x; |
3767 | nodes[node] += x; | |
3768 | } | |
3769 | ||
205ab99d CL |
3770 | } else if (flags & SO_PARTIAL) { |
3771 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3772 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3773 | |
205ab99d CL |
3774 | if (flags & SO_TOTAL) |
3775 | x = count_partial(n, count_total); | |
3776 | else if (flags & SO_OBJECTS) | |
3777 | x = count_partial(n, count_inuse); | |
81819f0f | 3778 | else |
205ab99d | 3779 | x = n->nr_partial; |
81819f0f CL |
3780 | total += x; |
3781 | nodes[node] += x; | |
3782 | } | |
3783 | } | |
81819f0f CL |
3784 | x = sprintf(buf, "%lu", total); |
3785 | #ifdef CONFIG_NUMA | |
f64dc58c | 3786 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3787 | if (nodes[node]) |
3788 | x += sprintf(buf + x, " N%d=%lu", | |
3789 | node, nodes[node]); | |
3790 | #endif | |
3791 | kfree(nodes); | |
3792 | return x + sprintf(buf + x, "\n"); | |
3793 | } | |
3794 | ||
3795 | static int any_slab_objects(struct kmem_cache *s) | |
3796 | { | |
3797 | int node; | |
3798 | int cpu; | |
3799 | ||
dfb4f096 CL |
3800 | for_each_possible_cpu(cpu) { |
3801 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
3802 | ||
3803 | if (c && c->page) | |
81819f0f | 3804 | return 1; |
dfb4f096 | 3805 | } |
81819f0f | 3806 | |
dfb4f096 | 3807 | for_each_online_node(node) { |
81819f0f CL |
3808 | struct kmem_cache_node *n = get_node(s, node); |
3809 | ||
dfb4f096 CL |
3810 | if (!n) |
3811 | continue; | |
3812 | ||
9e86943b | 3813 | if (n->nr_partial || atomic_long_read(&n->nr_slabs)) |
81819f0f CL |
3814 | return 1; |
3815 | } | |
3816 | return 0; | |
3817 | } | |
3818 | ||
3819 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3820 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3821 | ||
3822 | struct slab_attribute { | |
3823 | struct attribute attr; | |
3824 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3825 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3826 | }; | |
3827 | ||
3828 | #define SLAB_ATTR_RO(_name) \ | |
3829 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3830 | ||
3831 | #define SLAB_ATTR(_name) \ | |
3832 | static struct slab_attribute _name##_attr = \ | |
3833 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3834 | ||
81819f0f CL |
3835 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3836 | { | |
3837 | return sprintf(buf, "%d\n", s->size); | |
3838 | } | |
3839 | SLAB_ATTR_RO(slab_size); | |
3840 | ||
3841 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3842 | { | |
3843 | return sprintf(buf, "%d\n", s->align); | |
3844 | } | |
3845 | SLAB_ATTR_RO(align); | |
3846 | ||
3847 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3848 | { | |
3849 | return sprintf(buf, "%d\n", s->objsize); | |
3850 | } | |
3851 | SLAB_ATTR_RO(object_size); | |
3852 | ||
3853 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3854 | { | |
834f3d11 | 3855 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3856 | } |
3857 | SLAB_ATTR_RO(objs_per_slab); | |
3858 | ||
3859 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
3860 | { | |
834f3d11 | 3861 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f CL |
3862 | } |
3863 | SLAB_ATTR_RO(order); | |
3864 | ||
3865 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3866 | { | |
3867 | if (s->ctor) { | |
3868 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3869 | ||
3870 | return n + sprintf(buf + n, "\n"); | |
3871 | } | |
3872 | return 0; | |
3873 | } | |
3874 | SLAB_ATTR_RO(ctor); | |
3875 | ||
81819f0f CL |
3876 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3877 | { | |
3878 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3879 | } | |
3880 | SLAB_ATTR_RO(aliases); | |
3881 | ||
3882 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3883 | { | |
205ab99d | 3884 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3885 | } |
3886 | SLAB_ATTR_RO(slabs); | |
3887 | ||
3888 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3889 | { | |
d9acf4b7 | 3890 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3891 | } |
3892 | SLAB_ATTR_RO(partial); | |
3893 | ||
3894 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3895 | { | |
d9acf4b7 | 3896 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3897 | } |
3898 | SLAB_ATTR_RO(cpu_slabs); | |
3899 | ||
3900 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3901 | { | |
205ab99d | 3902 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3903 | } |
3904 | SLAB_ATTR_RO(objects); | |
3905 | ||
205ab99d CL |
3906 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3907 | { | |
3908 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3909 | } | |
3910 | SLAB_ATTR_RO(objects_partial); | |
3911 | ||
3912 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
3913 | { | |
3914 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
3915 | } | |
3916 | SLAB_ATTR_RO(total_objects); | |
3917 | ||
81819f0f CL |
3918 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
3919 | { | |
3920 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3921 | } | |
3922 | ||
3923 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3924 | const char *buf, size_t length) | |
3925 | { | |
3926 | s->flags &= ~SLAB_DEBUG_FREE; | |
3927 | if (buf[0] == '1') | |
3928 | s->flags |= SLAB_DEBUG_FREE; | |
3929 | return length; | |
3930 | } | |
3931 | SLAB_ATTR(sanity_checks); | |
3932 | ||
3933 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3934 | { | |
3935 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3936 | } | |
3937 | ||
3938 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3939 | size_t length) | |
3940 | { | |
3941 | s->flags &= ~SLAB_TRACE; | |
3942 | if (buf[0] == '1') | |
3943 | s->flags |= SLAB_TRACE; | |
3944 | return length; | |
3945 | } | |
3946 | SLAB_ATTR(trace); | |
3947 | ||
3948 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3949 | { | |
3950 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3951 | } | |
3952 | ||
3953 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3954 | const char *buf, size_t length) | |
3955 | { | |
3956 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3957 | if (buf[0] == '1') | |
3958 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3959 | return length; | |
3960 | } | |
3961 | SLAB_ATTR(reclaim_account); | |
3962 | ||
3963 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3964 | { | |
5af60839 | 3965 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3966 | } |
3967 | SLAB_ATTR_RO(hwcache_align); | |
3968 | ||
3969 | #ifdef CONFIG_ZONE_DMA | |
3970 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3971 | { | |
3972 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3973 | } | |
3974 | SLAB_ATTR_RO(cache_dma); | |
3975 | #endif | |
3976 | ||
3977 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3978 | { | |
3979 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3980 | } | |
3981 | SLAB_ATTR_RO(destroy_by_rcu); | |
3982 | ||
3983 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3984 | { | |
3985 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3986 | } | |
3987 | ||
3988 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3989 | const char *buf, size_t length) | |
3990 | { | |
3991 | if (any_slab_objects(s)) | |
3992 | return -EBUSY; | |
3993 | ||
3994 | s->flags &= ~SLAB_RED_ZONE; | |
3995 | if (buf[0] == '1') | |
3996 | s->flags |= SLAB_RED_ZONE; | |
3997 | calculate_sizes(s); | |
3998 | return length; | |
3999 | } | |
4000 | SLAB_ATTR(red_zone); | |
4001 | ||
4002 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4003 | { | |
4004 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4005 | } | |
4006 | ||
4007 | static ssize_t poison_store(struct kmem_cache *s, | |
4008 | const char *buf, size_t length) | |
4009 | { | |
4010 | if (any_slab_objects(s)) | |
4011 | return -EBUSY; | |
4012 | ||
4013 | s->flags &= ~SLAB_POISON; | |
4014 | if (buf[0] == '1') | |
4015 | s->flags |= SLAB_POISON; | |
4016 | calculate_sizes(s); | |
4017 | return length; | |
4018 | } | |
4019 | SLAB_ATTR(poison); | |
4020 | ||
4021 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4022 | { | |
4023 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4024 | } | |
4025 | ||
4026 | static ssize_t store_user_store(struct kmem_cache *s, | |
4027 | const char *buf, size_t length) | |
4028 | { | |
4029 | if (any_slab_objects(s)) | |
4030 | return -EBUSY; | |
4031 | ||
4032 | s->flags &= ~SLAB_STORE_USER; | |
4033 | if (buf[0] == '1') | |
4034 | s->flags |= SLAB_STORE_USER; | |
4035 | calculate_sizes(s); | |
4036 | return length; | |
4037 | } | |
4038 | SLAB_ATTR(store_user); | |
4039 | ||
53e15af0 CL |
4040 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4041 | { | |
4042 | return 0; | |
4043 | } | |
4044 | ||
4045 | static ssize_t validate_store(struct kmem_cache *s, | |
4046 | const char *buf, size_t length) | |
4047 | { | |
434e245d CL |
4048 | int ret = -EINVAL; |
4049 | ||
4050 | if (buf[0] == '1') { | |
4051 | ret = validate_slab_cache(s); | |
4052 | if (ret >= 0) | |
4053 | ret = length; | |
4054 | } | |
4055 | return ret; | |
53e15af0 CL |
4056 | } |
4057 | SLAB_ATTR(validate); | |
4058 | ||
2086d26a CL |
4059 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4060 | { | |
4061 | return 0; | |
4062 | } | |
4063 | ||
4064 | static ssize_t shrink_store(struct kmem_cache *s, | |
4065 | const char *buf, size_t length) | |
4066 | { | |
4067 | if (buf[0] == '1') { | |
4068 | int rc = kmem_cache_shrink(s); | |
4069 | ||
4070 | if (rc) | |
4071 | return rc; | |
4072 | } else | |
4073 | return -EINVAL; | |
4074 | return length; | |
4075 | } | |
4076 | SLAB_ATTR(shrink); | |
4077 | ||
88a420e4 CL |
4078 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4079 | { | |
4080 | if (!(s->flags & SLAB_STORE_USER)) | |
4081 | return -ENOSYS; | |
4082 | return list_locations(s, buf, TRACK_ALLOC); | |
4083 | } | |
4084 | SLAB_ATTR_RO(alloc_calls); | |
4085 | ||
4086 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4087 | { | |
4088 | if (!(s->flags & SLAB_STORE_USER)) | |
4089 | return -ENOSYS; | |
4090 | return list_locations(s, buf, TRACK_FREE); | |
4091 | } | |
4092 | SLAB_ATTR_RO(free_calls); | |
4093 | ||
81819f0f | 4094 | #ifdef CONFIG_NUMA |
9824601e | 4095 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4096 | { |
9824601e | 4097 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4098 | } |
4099 | ||
9824601e | 4100 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4101 | const char *buf, size_t length) |
4102 | { | |
4103 | int n = simple_strtoul(buf, NULL, 10); | |
4104 | ||
4105 | if (n < 100) | |
9824601e | 4106 | s->remote_node_defrag_ratio = n * 10; |
81819f0f CL |
4107 | return length; |
4108 | } | |
9824601e | 4109 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4110 | #endif |
4111 | ||
8ff12cfc | 4112 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4113 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4114 | { | |
4115 | unsigned long sum = 0; | |
4116 | int cpu; | |
4117 | int len; | |
4118 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4119 | ||
4120 | if (!data) | |
4121 | return -ENOMEM; | |
4122 | ||
4123 | for_each_online_cpu(cpu) { | |
4124 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4125 | ||
4126 | data[cpu] = x; | |
4127 | sum += x; | |
4128 | } | |
4129 | ||
4130 | len = sprintf(buf, "%lu", sum); | |
4131 | ||
50ef37b9 | 4132 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4133 | for_each_online_cpu(cpu) { |
4134 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4135 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4136 | } |
50ef37b9 | 4137 | #endif |
8ff12cfc CL |
4138 | kfree(data); |
4139 | return len + sprintf(buf + len, "\n"); | |
4140 | } | |
4141 | ||
4142 | #define STAT_ATTR(si, text) \ | |
4143 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4144 | { \ | |
4145 | return show_stat(s, buf, si); \ | |
4146 | } \ | |
4147 | SLAB_ATTR_RO(text); \ | |
4148 | ||
4149 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4150 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4151 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4152 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4153 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4154 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4155 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4156 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4157 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4158 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4159 | STAT_ATTR(FREE_SLAB, free_slab); | |
4160 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4161 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4162 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4163 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4164 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4165 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
4166 | ||
4167 | #endif | |
4168 | ||
06428780 | 4169 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4170 | &slab_size_attr.attr, |
4171 | &object_size_attr.attr, | |
4172 | &objs_per_slab_attr.attr, | |
4173 | &order_attr.attr, | |
4174 | &objects_attr.attr, | |
205ab99d CL |
4175 | &objects_partial_attr.attr, |
4176 | &total_objects_attr.attr, | |
81819f0f CL |
4177 | &slabs_attr.attr, |
4178 | &partial_attr.attr, | |
4179 | &cpu_slabs_attr.attr, | |
4180 | &ctor_attr.attr, | |
81819f0f CL |
4181 | &aliases_attr.attr, |
4182 | &align_attr.attr, | |
4183 | &sanity_checks_attr.attr, | |
4184 | &trace_attr.attr, | |
4185 | &hwcache_align_attr.attr, | |
4186 | &reclaim_account_attr.attr, | |
4187 | &destroy_by_rcu_attr.attr, | |
4188 | &red_zone_attr.attr, | |
4189 | &poison_attr.attr, | |
4190 | &store_user_attr.attr, | |
53e15af0 | 4191 | &validate_attr.attr, |
2086d26a | 4192 | &shrink_attr.attr, |
88a420e4 CL |
4193 | &alloc_calls_attr.attr, |
4194 | &free_calls_attr.attr, | |
81819f0f CL |
4195 | #ifdef CONFIG_ZONE_DMA |
4196 | &cache_dma_attr.attr, | |
4197 | #endif | |
4198 | #ifdef CONFIG_NUMA | |
9824601e | 4199 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4200 | #endif |
4201 | #ifdef CONFIG_SLUB_STATS | |
4202 | &alloc_fastpath_attr.attr, | |
4203 | &alloc_slowpath_attr.attr, | |
4204 | &free_fastpath_attr.attr, | |
4205 | &free_slowpath_attr.attr, | |
4206 | &free_frozen_attr.attr, | |
4207 | &free_add_partial_attr.attr, | |
4208 | &free_remove_partial_attr.attr, | |
4209 | &alloc_from_partial_attr.attr, | |
4210 | &alloc_slab_attr.attr, | |
4211 | &alloc_refill_attr.attr, | |
4212 | &free_slab_attr.attr, | |
4213 | &cpuslab_flush_attr.attr, | |
4214 | &deactivate_full_attr.attr, | |
4215 | &deactivate_empty_attr.attr, | |
4216 | &deactivate_to_head_attr.attr, | |
4217 | &deactivate_to_tail_attr.attr, | |
4218 | &deactivate_remote_frees_attr.attr, | |
81819f0f CL |
4219 | #endif |
4220 | NULL | |
4221 | }; | |
4222 | ||
4223 | static struct attribute_group slab_attr_group = { | |
4224 | .attrs = slab_attrs, | |
4225 | }; | |
4226 | ||
4227 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4228 | struct attribute *attr, | |
4229 | char *buf) | |
4230 | { | |
4231 | struct slab_attribute *attribute; | |
4232 | struct kmem_cache *s; | |
4233 | int err; | |
4234 | ||
4235 | attribute = to_slab_attr(attr); | |
4236 | s = to_slab(kobj); | |
4237 | ||
4238 | if (!attribute->show) | |
4239 | return -EIO; | |
4240 | ||
4241 | err = attribute->show(s, buf); | |
4242 | ||
4243 | return err; | |
4244 | } | |
4245 | ||
4246 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4247 | struct attribute *attr, | |
4248 | const char *buf, size_t len) | |
4249 | { | |
4250 | struct slab_attribute *attribute; | |
4251 | struct kmem_cache *s; | |
4252 | int err; | |
4253 | ||
4254 | attribute = to_slab_attr(attr); | |
4255 | s = to_slab(kobj); | |
4256 | ||
4257 | if (!attribute->store) | |
4258 | return -EIO; | |
4259 | ||
4260 | err = attribute->store(s, buf, len); | |
4261 | ||
4262 | return err; | |
4263 | } | |
4264 | ||
151c602f CL |
4265 | static void kmem_cache_release(struct kobject *kobj) |
4266 | { | |
4267 | struct kmem_cache *s = to_slab(kobj); | |
4268 | ||
4269 | kfree(s); | |
4270 | } | |
4271 | ||
81819f0f CL |
4272 | static struct sysfs_ops slab_sysfs_ops = { |
4273 | .show = slab_attr_show, | |
4274 | .store = slab_attr_store, | |
4275 | }; | |
4276 | ||
4277 | static struct kobj_type slab_ktype = { | |
4278 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4279 | .release = kmem_cache_release |
81819f0f CL |
4280 | }; |
4281 | ||
4282 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4283 | { | |
4284 | struct kobj_type *ktype = get_ktype(kobj); | |
4285 | ||
4286 | if (ktype == &slab_ktype) | |
4287 | return 1; | |
4288 | return 0; | |
4289 | } | |
4290 | ||
4291 | static struct kset_uevent_ops slab_uevent_ops = { | |
4292 | .filter = uevent_filter, | |
4293 | }; | |
4294 | ||
27c3a314 | 4295 | static struct kset *slab_kset; |
81819f0f CL |
4296 | |
4297 | #define ID_STR_LENGTH 64 | |
4298 | ||
4299 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4300 | * |
4301 | * Format :[flags-]size | |
81819f0f CL |
4302 | */ |
4303 | static char *create_unique_id(struct kmem_cache *s) | |
4304 | { | |
4305 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4306 | char *p = name; | |
4307 | ||
4308 | BUG_ON(!name); | |
4309 | ||
4310 | *p++ = ':'; | |
4311 | /* | |
4312 | * First flags affecting slabcache operations. We will only | |
4313 | * get here for aliasable slabs so we do not need to support | |
4314 | * too many flags. The flags here must cover all flags that | |
4315 | * are matched during merging to guarantee that the id is | |
4316 | * unique. | |
4317 | */ | |
4318 | if (s->flags & SLAB_CACHE_DMA) | |
4319 | *p++ = 'd'; | |
4320 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4321 | *p++ = 'a'; | |
4322 | if (s->flags & SLAB_DEBUG_FREE) | |
4323 | *p++ = 'F'; | |
4324 | if (p != name + 1) | |
4325 | *p++ = '-'; | |
4326 | p += sprintf(p, "%07d", s->size); | |
4327 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4328 | return name; | |
4329 | } | |
4330 | ||
4331 | static int sysfs_slab_add(struct kmem_cache *s) | |
4332 | { | |
4333 | int err; | |
4334 | const char *name; | |
4335 | int unmergeable; | |
4336 | ||
4337 | if (slab_state < SYSFS) | |
4338 | /* Defer until later */ | |
4339 | return 0; | |
4340 | ||
4341 | unmergeable = slab_unmergeable(s); | |
4342 | if (unmergeable) { | |
4343 | /* | |
4344 | * Slabcache can never be merged so we can use the name proper. | |
4345 | * This is typically the case for debug situations. In that | |
4346 | * case we can catch duplicate names easily. | |
4347 | */ | |
27c3a314 | 4348 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4349 | name = s->name; |
4350 | } else { | |
4351 | /* | |
4352 | * Create a unique name for the slab as a target | |
4353 | * for the symlinks. | |
4354 | */ | |
4355 | name = create_unique_id(s); | |
4356 | } | |
4357 | ||
27c3a314 | 4358 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4359 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4360 | if (err) { | |
4361 | kobject_put(&s->kobj); | |
81819f0f | 4362 | return err; |
1eada11c | 4363 | } |
81819f0f CL |
4364 | |
4365 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4366 | if (err) | |
4367 | return err; | |
4368 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4369 | if (!unmergeable) { | |
4370 | /* Setup first alias */ | |
4371 | sysfs_slab_alias(s, s->name); | |
4372 | kfree(name); | |
4373 | } | |
4374 | return 0; | |
4375 | } | |
4376 | ||
4377 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4378 | { | |
4379 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4380 | kobject_del(&s->kobj); | |
151c602f | 4381 | kobject_put(&s->kobj); |
81819f0f CL |
4382 | } |
4383 | ||
4384 | /* | |
4385 | * Need to buffer aliases during bootup until sysfs becomes | |
4386 | * available lest we loose that information. | |
4387 | */ | |
4388 | struct saved_alias { | |
4389 | struct kmem_cache *s; | |
4390 | const char *name; | |
4391 | struct saved_alias *next; | |
4392 | }; | |
4393 | ||
5af328a5 | 4394 | static struct saved_alias *alias_list; |
81819f0f CL |
4395 | |
4396 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4397 | { | |
4398 | struct saved_alias *al; | |
4399 | ||
4400 | if (slab_state == SYSFS) { | |
4401 | /* | |
4402 | * If we have a leftover link then remove it. | |
4403 | */ | |
27c3a314 GKH |
4404 | sysfs_remove_link(&slab_kset->kobj, name); |
4405 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4406 | } |
4407 | ||
4408 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4409 | if (!al) | |
4410 | return -ENOMEM; | |
4411 | ||
4412 | al->s = s; | |
4413 | al->name = name; | |
4414 | al->next = alias_list; | |
4415 | alias_list = al; | |
4416 | return 0; | |
4417 | } | |
4418 | ||
4419 | static int __init slab_sysfs_init(void) | |
4420 | { | |
5b95a4ac | 4421 | struct kmem_cache *s; |
81819f0f CL |
4422 | int err; |
4423 | ||
0ff21e46 | 4424 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4425 | if (!slab_kset) { |
81819f0f CL |
4426 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4427 | return -ENOSYS; | |
4428 | } | |
4429 | ||
26a7bd03 CL |
4430 | slab_state = SYSFS; |
4431 | ||
5b95a4ac | 4432 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4433 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4434 | if (err) |
4435 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4436 | " to sysfs\n", s->name); | |
26a7bd03 | 4437 | } |
81819f0f CL |
4438 | |
4439 | while (alias_list) { | |
4440 | struct saved_alias *al = alias_list; | |
4441 | ||
4442 | alias_list = alias_list->next; | |
4443 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4444 | if (err) |
4445 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4446 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4447 | kfree(al); |
4448 | } | |
4449 | ||
4450 | resiliency_test(); | |
4451 | return 0; | |
4452 | } | |
4453 | ||
4454 | __initcall(slab_sysfs_init); | |
81819f0f | 4455 | #endif |
57ed3eda PE |
4456 | |
4457 | /* | |
4458 | * The /proc/slabinfo ABI | |
4459 | */ | |
158a9624 LT |
4460 | #ifdef CONFIG_SLABINFO |
4461 | ||
4462 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, | |
4463 | size_t count, loff_t *ppos) | |
4464 | { | |
4465 | return -EINVAL; | |
4466 | } | |
4467 | ||
57ed3eda PE |
4468 | |
4469 | static void print_slabinfo_header(struct seq_file *m) | |
4470 | { | |
4471 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4472 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4473 | "<objperslab> <pagesperslab>"); | |
4474 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4475 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4476 | seq_putc(m, '\n'); | |
4477 | } | |
4478 | ||
4479 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4480 | { | |
4481 | loff_t n = *pos; | |
4482 | ||
4483 | down_read(&slub_lock); | |
4484 | if (!n) | |
4485 | print_slabinfo_header(m); | |
4486 | ||
4487 | return seq_list_start(&slab_caches, *pos); | |
4488 | } | |
4489 | ||
4490 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4491 | { | |
4492 | return seq_list_next(p, &slab_caches, pos); | |
4493 | } | |
4494 | ||
4495 | static void s_stop(struct seq_file *m, void *p) | |
4496 | { | |
4497 | up_read(&slub_lock); | |
4498 | } | |
4499 | ||
4500 | static int s_show(struct seq_file *m, void *p) | |
4501 | { | |
4502 | unsigned long nr_partials = 0; | |
4503 | unsigned long nr_slabs = 0; | |
4504 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4505 | unsigned long nr_objs = 0; |
4506 | unsigned long nr_free = 0; | |
57ed3eda PE |
4507 | struct kmem_cache *s; |
4508 | int node; | |
4509 | ||
4510 | s = list_entry(p, struct kmem_cache, list); | |
4511 | ||
4512 | for_each_online_node(node) { | |
4513 | struct kmem_cache_node *n = get_node(s, node); | |
4514 | ||
4515 | if (!n) | |
4516 | continue; | |
4517 | ||
4518 | nr_partials += n->nr_partial; | |
4519 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4520 | nr_objs += atomic_long_read(&n->total_objects); |
4521 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4522 | } |
4523 | ||
205ab99d | 4524 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4525 | |
4526 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4527 | nr_objs, s->size, oo_objects(s->oo), |
4528 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4529 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4530 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4531 | 0UL); | |
4532 | seq_putc(m, '\n'); | |
4533 | return 0; | |
4534 | } | |
4535 | ||
4536 | const struct seq_operations slabinfo_op = { | |
4537 | .start = s_start, | |
4538 | .next = s_next, | |
4539 | .stop = s_stop, | |
4540 | .show = s_show, | |
4541 | }; | |
4542 | ||
158a9624 | 4543 | #endif /* CONFIG_SLABINFO */ |