]>
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 | * | |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
9 | */ |
10 | ||
11 | #include <linux/mm.h> | |
1eb5ac64 | 12 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
13 | #include <linux/module.h> |
14 | #include <linux/bit_spinlock.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/bitops.h> | |
17 | #include <linux/slab.h> | |
7b3c3a50 | 18 | #include <linux/proc_fs.h> |
81819f0f | 19 | #include <linux/seq_file.h> |
5a896d9e | 20 | #include <linux/kmemcheck.h> |
81819f0f CL |
21 | #include <linux/cpu.h> |
22 | #include <linux/cpuset.h> | |
23 | #include <linux/mempolicy.h> | |
24 | #include <linux/ctype.h> | |
3ac7fe5a | 25 | #include <linux/debugobjects.h> |
81819f0f | 26 | #include <linux/kallsyms.h> |
b9049e23 | 27 | #include <linux/memory.h> |
f8bd2258 | 28 | #include <linux/math64.h> |
773ff60e | 29 | #include <linux/fault-inject.h> |
81819f0f CL |
30 | |
31 | /* | |
32 | * Lock order: | |
33 | * 1. slab_lock(page) | |
34 | * 2. slab->list_lock | |
35 | * | |
36 | * The slab_lock protects operations on the object of a particular | |
37 | * slab and its metadata in the page struct. If the slab lock | |
38 | * has been taken then no allocations nor frees can be performed | |
39 | * on the objects in the slab nor can the slab be added or removed | |
40 | * from the partial or full lists since this would mean modifying | |
41 | * the page_struct of the slab. | |
42 | * | |
43 | * The list_lock protects the partial and full list on each node and | |
44 | * the partial slab counter. If taken then no new slabs may be added or | |
45 | * removed from the lists nor make the number of partial slabs be modified. | |
46 | * (Note that the total number of slabs is an atomic value that may be | |
47 | * modified without taking the list lock). | |
48 | * | |
49 | * The list_lock is a centralized lock and thus we avoid taking it as | |
50 | * much as possible. As long as SLUB does not have to handle partial | |
51 | * slabs, operations can continue without any centralized lock. F.e. | |
52 | * allocating a long series of objects that fill up slabs does not require | |
53 | * the list lock. | |
54 | * | |
55 | * The lock order is sometimes inverted when we are trying to get a slab | |
56 | * off a list. We take the list_lock and then look for a page on the list | |
57 | * to use. While we do that objects in the slabs may be freed. We can | |
58 | * only operate on the slab if we have also taken the slab_lock. So we use | |
59 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
60 | * can occur anymore and we can use the slab for allocations etc. If the | |
61 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
62 | * we must stay away from it for a while since we may cause a bouncing | |
63 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
64 | * If all pages are busy then we may allocate a new slab instead of reusing | |
65 | * a partial slab. A new slab has noone operating on it and thus there is | |
66 | * no danger of cacheline contention. | |
67 | * | |
68 | * Interrupts are disabled during allocation and deallocation in order to | |
69 | * make the slab allocator safe to use in the context of an irq. In addition | |
70 | * interrupts are disabled to ensure that the processor does not change | |
71 | * while handling per_cpu slabs, due to kernel preemption. | |
72 | * | |
73 | * SLUB assigns one slab for allocation to each processor. | |
74 | * Allocations only occur from these slabs called cpu slabs. | |
75 | * | |
672bba3a CL |
76 | * Slabs with free elements are kept on a partial list and during regular |
77 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 78 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
79 | * We track full slabs for debugging purposes though because otherwise we |
80 | * cannot scan all objects. | |
81819f0f CL |
81 | * |
82 | * Slabs are freed when they become empty. Teardown and setup is | |
83 | * minimal so we rely on the page allocators per cpu caches for | |
84 | * fast frees and allocs. | |
85 | * | |
86 | * Overloading of page flags that are otherwise used for LRU management. | |
87 | * | |
4b6f0750 CL |
88 | * PageActive The slab is frozen and exempt from list processing. |
89 | * This means that the slab is dedicated to a purpose | |
90 | * such as satisfying allocations for a specific | |
91 | * processor. Objects may be freed in the slab while | |
92 | * it is frozen but slab_free will then skip the usual | |
93 | * list operations. It is up to the processor holding | |
94 | * the slab to integrate the slab into the slab lists | |
95 | * when the slab is no longer needed. | |
96 | * | |
97 | * One use of this flag is to mark slabs that are | |
98 | * used for allocations. Then such a slab becomes a cpu | |
99 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 100 | * freelist that allows lockless access to |
894b8788 CL |
101 | * free objects in addition to the regular freelist |
102 | * that requires the slab lock. | |
81819f0f CL |
103 | * |
104 | * PageError Slab requires special handling due to debug | |
105 | * options set. This moves slab handling out of | |
894b8788 | 106 | * the fast path and disables lockless freelists. |
81819f0f CL |
107 | */ |
108 | ||
5577bd8a | 109 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 110 | #define SLABDEBUG 1 |
5577bd8a CL |
111 | #else |
112 | #define SLABDEBUG 0 | |
113 | #endif | |
114 | ||
81819f0f CL |
115 | /* |
116 | * Issues still to be resolved: | |
117 | * | |
81819f0f CL |
118 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
119 | * | |
81819f0f CL |
120 | * - Variable sizing of the per node arrays |
121 | */ | |
122 | ||
123 | /* Enable to test recovery from slab corruption on boot */ | |
124 | #undef SLUB_RESILIENCY_TEST | |
125 | ||
2086d26a CL |
126 | /* |
127 | * Mininum number of partial slabs. These will be left on the partial | |
128 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
129 | */ | |
76be8950 | 130 | #define MIN_PARTIAL 5 |
e95eed57 | 131 | |
2086d26a CL |
132 | /* |
133 | * Maximum number of desirable partial slabs. | |
134 | * The existence of more partial slabs makes kmem_cache_shrink | |
135 | * sort the partial list by the number of objects in the. | |
136 | */ | |
137 | #define MAX_PARTIAL 10 | |
138 | ||
81819f0f CL |
139 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
140 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 141 | |
fa5ec8a1 | 142 | /* |
3de47213 DR |
143 | * Debugging flags that require metadata to be stored in the slab. These get |
144 | * disabled when slub_debug=O is used and a cache's min order increases with | |
145 | * metadata. | |
fa5ec8a1 | 146 | */ |
3de47213 | 147 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 148 | |
81819f0f CL |
149 | /* |
150 | * Set of flags that will prevent slab merging | |
151 | */ | |
152 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
153 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
154 | SLAB_FAILSLAB) | |
81819f0f CL |
155 | |
156 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 157 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 158 | |
210b5c06 CG |
159 | #define OO_SHIFT 16 |
160 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
161 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
162 | ||
81819f0f | 163 | /* Internal SLUB flags */ |
1ceef402 CL |
164 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
165 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
166 | |
167 | static int kmem_size = sizeof(struct kmem_cache); | |
168 | ||
169 | #ifdef CONFIG_SMP | |
170 | static struct notifier_block slab_notifier; | |
171 | #endif | |
172 | ||
173 | static enum { | |
174 | DOWN, /* No slab functionality available */ | |
175 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 176 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
177 | SYSFS /* Sysfs up */ |
178 | } slab_state = DOWN; | |
179 | ||
180 | /* A list of all slab caches on the system */ | |
181 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 182 | static LIST_HEAD(slab_caches); |
81819f0f | 183 | |
02cbc874 CL |
184 | /* |
185 | * Tracking user of a slab. | |
186 | */ | |
187 | struct track { | |
ce71e27c | 188 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
189 | int cpu; /* Was running on cpu */ |
190 | int pid; /* Pid context */ | |
191 | unsigned long when; /* When did the operation occur */ | |
192 | }; | |
193 | ||
194 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
195 | ||
f6acb635 | 196 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
197 | static int sysfs_slab_add(struct kmem_cache *); |
198 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
199 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 200 | |
81819f0f | 201 | #else |
0c710013 CL |
202 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
203 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
204 | { return 0; } | |
151c602f CL |
205 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
206 | { | |
207 | kfree(s); | |
208 | } | |
8ff12cfc | 209 | |
81819f0f CL |
210 | #endif |
211 | ||
84e554e6 | 212 | static inline void stat(struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
213 | { |
214 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 215 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
216 | #endif |
217 | } | |
218 | ||
81819f0f CL |
219 | /******************************************************************** |
220 | * Core slab cache functions | |
221 | *******************************************************************/ | |
222 | ||
223 | int slab_is_available(void) | |
224 | { | |
225 | return slab_state >= UP; | |
226 | } | |
227 | ||
228 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
229 | { | |
230 | #ifdef CONFIG_NUMA | |
231 | return s->node[node]; | |
232 | #else | |
233 | return &s->local_node; | |
234 | #endif | |
235 | } | |
236 | ||
6446faa2 | 237 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
238 | static inline int check_valid_pointer(struct kmem_cache *s, |
239 | struct page *page, const void *object) | |
240 | { | |
241 | void *base; | |
242 | ||
a973e9dd | 243 | if (!object) |
02cbc874 CL |
244 | return 1; |
245 | ||
a973e9dd | 246 | base = page_address(page); |
39b26464 | 247 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
248 | (object - base) % s->size) { |
249 | return 0; | |
250 | } | |
251 | ||
252 | return 1; | |
253 | } | |
254 | ||
7656c72b CL |
255 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
256 | { | |
257 | return *(void **)(object + s->offset); | |
258 | } | |
259 | ||
260 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
261 | { | |
262 | *(void **)(object + s->offset) = fp; | |
263 | } | |
264 | ||
265 | /* Loop over all objects in a slab */ | |
224a88be CL |
266 | #define for_each_object(__p, __s, __addr, __objects) \ |
267 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
268 | __p += (__s)->size) |
269 | ||
270 | /* Scan freelist */ | |
271 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 272 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
273 | |
274 | /* Determine object index from a given position */ | |
275 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
276 | { | |
277 | return (p - addr) / s->size; | |
278 | } | |
279 | ||
834f3d11 CL |
280 | static inline struct kmem_cache_order_objects oo_make(int order, |
281 | unsigned long size) | |
282 | { | |
283 | struct kmem_cache_order_objects x = { | |
210b5c06 | 284 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
285 | }; |
286 | ||
287 | return x; | |
288 | } | |
289 | ||
290 | static inline int oo_order(struct kmem_cache_order_objects x) | |
291 | { | |
210b5c06 | 292 | return x.x >> OO_SHIFT; |
834f3d11 CL |
293 | } |
294 | ||
295 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
296 | { | |
210b5c06 | 297 | return x.x & OO_MASK; |
834f3d11 CL |
298 | } |
299 | ||
41ecc55b CL |
300 | #ifdef CONFIG_SLUB_DEBUG |
301 | /* | |
302 | * Debug settings: | |
303 | */ | |
f0630fff CL |
304 | #ifdef CONFIG_SLUB_DEBUG_ON |
305 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
306 | #else | |
41ecc55b | 307 | static int slub_debug; |
f0630fff | 308 | #endif |
41ecc55b CL |
309 | |
310 | static char *slub_debug_slabs; | |
fa5ec8a1 | 311 | static int disable_higher_order_debug; |
41ecc55b | 312 | |
81819f0f CL |
313 | /* |
314 | * Object debugging | |
315 | */ | |
316 | static void print_section(char *text, u8 *addr, unsigned int length) | |
317 | { | |
318 | int i, offset; | |
319 | int newline = 1; | |
320 | char ascii[17]; | |
321 | ||
322 | ascii[16] = 0; | |
323 | ||
324 | for (i = 0; i < length; i++) { | |
325 | if (newline) { | |
24922684 | 326 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
327 | newline = 0; |
328 | } | |
06428780 | 329 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
330 | offset = i % 16; |
331 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
332 | if (offset == 15) { | |
06428780 | 333 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
334 | newline = 1; |
335 | } | |
336 | } | |
337 | if (!newline) { | |
338 | i %= 16; | |
339 | while (i < 16) { | |
06428780 | 340 | printk(KERN_CONT " "); |
81819f0f CL |
341 | ascii[i] = ' '; |
342 | i++; | |
343 | } | |
06428780 | 344 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
345 | } |
346 | } | |
347 | ||
81819f0f CL |
348 | static struct track *get_track(struct kmem_cache *s, void *object, |
349 | enum track_item alloc) | |
350 | { | |
351 | struct track *p; | |
352 | ||
353 | if (s->offset) | |
354 | p = object + s->offset + sizeof(void *); | |
355 | else | |
356 | p = object + s->inuse; | |
357 | ||
358 | return p + alloc; | |
359 | } | |
360 | ||
361 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 362 | enum track_item alloc, unsigned long addr) |
81819f0f | 363 | { |
1a00df4a | 364 | struct track *p = get_track(s, object, alloc); |
81819f0f | 365 | |
81819f0f CL |
366 | if (addr) { |
367 | p->addr = addr; | |
368 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 369 | p->pid = current->pid; |
81819f0f CL |
370 | p->when = jiffies; |
371 | } else | |
372 | memset(p, 0, sizeof(struct track)); | |
373 | } | |
374 | ||
81819f0f CL |
375 | static void init_tracking(struct kmem_cache *s, void *object) |
376 | { | |
24922684 CL |
377 | if (!(s->flags & SLAB_STORE_USER)) |
378 | return; | |
379 | ||
ce71e27c EGM |
380 | set_track(s, object, TRACK_FREE, 0UL); |
381 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
382 | } |
383 | ||
384 | static void print_track(const char *s, struct track *t) | |
385 | { | |
386 | if (!t->addr) | |
387 | return; | |
388 | ||
7daf705f | 389 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 390 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
391 | } |
392 | ||
393 | static void print_tracking(struct kmem_cache *s, void *object) | |
394 | { | |
395 | if (!(s->flags & SLAB_STORE_USER)) | |
396 | return; | |
397 | ||
398 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
399 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
400 | } | |
401 | ||
402 | static void print_page_info(struct page *page) | |
403 | { | |
39b26464 CL |
404 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
405 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
406 | |
407 | } | |
408 | ||
409 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
410 | { | |
411 | va_list args; | |
412 | char buf[100]; | |
413 | ||
414 | va_start(args, fmt); | |
415 | vsnprintf(buf, sizeof(buf), fmt, args); | |
416 | va_end(args); | |
417 | printk(KERN_ERR "========================================" | |
418 | "=====================================\n"); | |
419 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
420 | printk(KERN_ERR "----------------------------------------" | |
421 | "-------------------------------------\n\n"); | |
81819f0f CL |
422 | } |
423 | ||
24922684 CL |
424 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
425 | { | |
426 | va_list args; | |
427 | char buf[100]; | |
428 | ||
429 | va_start(args, fmt); | |
430 | vsnprintf(buf, sizeof(buf), fmt, args); | |
431 | va_end(args); | |
432 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
433 | } | |
434 | ||
435 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
436 | { |
437 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 438 | u8 *addr = page_address(page); |
24922684 CL |
439 | |
440 | print_tracking(s, p); | |
441 | ||
442 | print_page_info(page); | |
443 | ||
444 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
445 | p, p - addr, get_freepointer(s, p)); | |
446 | ||
447 | if (p > addr + 16) | |
448 | print_section("Bytes b4", p - 16, 16); | |
449 | ||
0ebd652b | 450 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
451 | |
452 | if (s->flags & SLAB_RED_ZONE) | |
453 | print_section("Redzone", p + s->objsize, | |
454 | s->inuse - s->objsize); | |
455 | ||
81819f0f CL |
456 | if (s->offset) |
457 | off = s->offset + sizeof(void *); | |
458 | else | |
459 | off = s->inuse; | |
460 | ||
24922684 | 461 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 462 | off += 2 * sizeof(struct track); |
81819f0f CL |
463 | |
464 | if (off != s->size) | |
465 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
466 | print_section("Padding", p + off, s->size - off); |
467 | ||
468 | dump_stack(); | |
81819f0f CL |
469 | } |
470 | ||
471 | static void object_err(struct kmem_cache *s, struct page *page, | |
472 | u8 *object, char *reason) | |
473 | { | |
3dc50637 | 474 | slab_bug(s, "%s", reason); |
24922684 | 475 | print_trailer(s, page, object); |
81819f0f CL |
476 | } |
477 | ||
24922684 | 478 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
479 | { |
480 | va_list args; | |
481 | char buf[100]; | |
482 | ||
24922684 CL |
483 | va_start(args, fmt); |
484 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 485 | va_end(args); |
3dc50637 | 486 | slab_bug(s, "%s", buf); |
24922684 | 487 | print_page_info(page); |
81819f0f CL |
488 | dump_stack(); |
489 | } | |
490 | ||
491 | static void init_object(struct kmem_cache *s, void *object, int active) | |
492 | { | |
493 | u8 *p = object; | |
494 | ||
495 | if (s->flags & __OBJECT_POISON) { | |
496 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 497 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
498 | } |
499 | ||
500 | if (s->flags & SLAB_RED_ZONE) | |
501 | memset(p + s->objsize, | |
502 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
503 | s->inuse - s->objsize); | |
504 | } | |
505 | ||
24922684 | 506 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
507 | { |
508 | while (bytes) { | |
509 | if (*start != (u8)value) | |
24922684 | 510 | return start; |
81819f0f CL |
511 | start++; |
512 | bytes--; | |
513 | } | |
24922684 CL |
514 | return NULL; |
515 | } | |
516 | ||
517 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
518 | void *from, void *to) | |
519 | { | |
520 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
521 | memset(from, data, to - from); | |
522 | } | |
523 | ||
524 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
525 | u8 *object, char *what, | |
06428780 | 526 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
527 | { |
528 | u8 *fault; | |
529 | u8 *end; | |
530 | ||
531 | fault = check_bytes(start, value, bytes); | |
532 | if (!fault) | |
533 | return 1; | |
534 | ||
535 | end = start + bytes; | |
536 | while (end > fault && end[-1] == value) | |
537 | end--; | |
538 | ||
539 | slab_bug(s, "%s overwritten", what); | |
540 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
541 | fault, end - 1, fault[0], value); | |
542 | print_trailer(s, page, object); | |
543 | ||
544 | restore_bytes(s, what, value, fault, end); | |
545 | return 0; | |
81819f0f CL |
546 | } |
547 | ||
81819f0f CL |
548 | /* |
549 | * Object layout: | |
550 | * | |
551 | * object address | |
552 | * Bytes of the object to be managed. | |
553 | * If the freepointer may overlay the object then the free | |
554 | * pointer is the first word of the object. | |
672bba3a | 555 | * |
81819f0f CL |
556 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
557 | * 0xa5 (POISON_END) | |
558 | * | |
559 | * object + s->objsize | |
560 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
561 | * Padding is extended by another word if Redzoning is enabled and |
562 | * objsize == inuse. | |
563 | * | |
81819f0f CL |
564 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
565 | * 0xcc (RED_ACTIVE) for objects in use. | |
566 | * | |
567 | * object + s->inuse | |
672bba3a CL |
568 | * Meta data starts here. |
569 | * | |
81819f0f CL |
570 | * A. Free pointer (if we cannot overwrite object on free) |
571 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 572 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 573 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
574 | * before the word boundary. |
575 | * | |
576 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
577 | * |
578 | * object + s->size | |
672bba3a | 579 | * Nothing is used beyond s->size. |
81819f0f | 580 | * |
672bba3a CL |
581 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
582 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
583 | * may be used with merged slabcaches. |
584 | */ | |
585 | ||
81819f0f CL |
586 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
587 | { | |
588 | unsigned long off = s->inuse; /* The end of info */ | |
589 | ||
590 | if (s->offset) | |
591 | /* Freepointer is placed after the object. */ | |
592 | off += sizeof(void *); | |
593 | ||
594 | if (s->flags & SLAB_STORE_USER) | |
595 | /* We also have user information there */ | |
596 | off += 2 * sizeof(struct track); | |
597 | ||
598 | if (s->size == off) | |
599 | return 1; | |
600 | ||
24922684 CL |
601 | return check_bytes_and_report(s, page, p, "Object padding", |
602 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
603 | } |
604 | ||
39b26464 | 605 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
606 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
607 | { | |
24922684 CL |
608 | u8 *start; |
609 | u8 *fault; | |
610 | u8 *end; | |
611 | int length; | |
612 | int remainder; | |
81819f0f CL |
613 | |
614 | if (!(s->flags & SLAB_POISON)) | |
615 | return 1; | |
616 | ||
a973e9dd | 617 | start = page_address(page); |
834f3d11 | 618 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
619 | end = start + length; |
620 | remainder = length % s->size; | |
81819f0f CL |
621 | if (!remainder) |
622 | return 1; | |
623 | ||
39b26464 | 624 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
625 | if (!fault) |
626 | return 1; | |
627 | while (end > fault && end[-1] == POISON_INUSE) | |
628 | end--; | |
629 | ||
630 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 631 | print_section("Padding", end - remainder, remainder); |
24922684 | 632 | |
8a3d271d | 633 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 634 | return 0; |
81819f0f CL |
635 | } |
636 | ||
637 | static int check_object(struct kmem_cache *s, struct page *page, | |
638 | void *object, int active) | |
639 | { | |
640 | u8 *p = object; | |
641 | u8 *endobject = object + s->objsize; | |
642 | ||
643 | if (s->flags & SLAB_RED_ZONE) { | |
644 | unsigned int red = | |
645 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
646 | ||
24922684 CL |
647 | if (!check_bytes_and_report(s, page, object, "Redzone", |
648 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 649 | return 0; |
81819f0f | 650 | } else { |
3adbefee IM |
651 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
652 | check_bytes_and_report(s, page, p, "Alignment padding", | |
653 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
654 | } | |
81819f0f CL |
655 | } |
656 | ||
657 | if (s->flags & SLAB_POISON) { | |
658 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
659 | (!check_bytes_and_report(s, page, p, "Poison", p, |
660 | POISON_FREE, s->objsize - 1) || | |
661 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 662 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 663 | return 0; |
81819f0f CL |
664 | /* |
665 | * check_pad_bytes cleans up on its own. | |
666 | */ | |
667 | check_pad_bytes(s, page, p); | |
668 | } | |
669 | ||
670 | if (!s->offset && active) | |
671 | /* | |
672 | * Object and freepointer overlap. Cannot check | |
673 | * freepointer while object is allocated. | |
674 | */ | |
675 | return 1; | |
676 | ||
677 | /* Check free pointer validity */ | |
678 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
679 | object_err(s, page, p, "Freepointer corrupt"); | |
680 | /* | |
9f6c708e | 681 | * No choice but to zap it and thus lose the remainder |
81819f0f | 682 | * of the free objects in this slab. May cause |
672bba3a | 683 | * another error because the object count is now wrong. |
81819f0f | 684 | */ |
a973e9dd | 685 | set_freepointer(s, p, NULL); |
81819f0f CL |
686 | return 0; |
687 | } | |
688 | return 1; | |
689 | } | |
690 | ||
691 | static int check_slab(struct kmem_cache *s, struct page *page) | |
692 | { | |
39b26464 CL |
693 | int maxobj; |
694 | ||
81819f0f CL |
695 | VM_BUG_ON(!irqs_disabled()); |
696 | ||
697 | if (!PageSlab(page)) { | |
24922684 | 698 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
699 | return 0; |
700 | } | |
39b26464 CL |
701 | |
702 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
703 | if (page->objects > maxobj) { | |
704 | slab_err(s, page, "objects %u > max %u", | |
705 | s->name, page->objects, maxobj); | |
706 | return 0; | |
707 | } | |
708 | if (page->inuse > page->objects) { | |
24922684 | 709 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 710 | s->name, page->inuse, page->objects); |
81819f0f CL |
711 | return 0; |
712 | } | |
713 | /* Slab_pad_check fixes things up after itself */ | |
714 | slab_pad_check(s, page); | |
715 | return 1; | |
716 | } | |
717 | ||
718 | /* | |
672bba3a CL |
719 | * Determine if a certain object on a page is on the freelist. Must hold the |
720 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
721 | */ |
722 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
723 | { | |
724 | int nr = 0; | |
725 | void *fp = page->freelist; | |
726 | void *object = NULL; | |
224a88be | 727 | unsigned long max_objects; |
81819f0f | 728 | |
39b26464 | 729 | while (fp && nr <= page->objects) { |
81819f0f CL |
730 | if (fp == search) |
731 | return 1; | |
732 | if (!check_valid_pointer(s, page, fp)) { | |
733 | if (object) { | |
734 | object_err(s, page, object, | |
735 | "Freechain corrupt"); | |
a973e9dd | 736 | set_freepointer(s, object, NULL); |
81819f0f CL |
737 | break; |
738 | } else { | |
24922684 | 739 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 740 | page->freelist = NULL; |
39b26464 | 741 | page->inuse = page->objects; |
24922684 | 742 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
743 | return 0; |
744 | } | |
745 | break; | |
746 | } | |
747 | object = fp; | |
748 | fp = get_freepointer(s, object); | |
749 | nr++; | |
750 | } | |
751 | ||
224a88be | 752 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
753 | if (max_objects > MAX_OBJS_PER_PAGE) |
754 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
755 | |
756 | if (page->objects != max_objects) { | |
757 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
758 | "should be %d", page->objects, max_objects); | |
759 | page->objects = max_objects; | |
760 | slab_fix(s, "Number of objects adjusted."); | |
761 | } | |
39b26464 | 762 | if (page->inuse != page->objects - nr) { |
70d71228 | 763 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
764 | "counted were %d", page->inuse, page->objects - nr); |
765 | page->inuse = page->objects - nr; | |
24922684 | 766 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
767 | } |
768 | return search == NULL; | |
769 | } | |
770 | ||
0121c619 CL |
771 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
772 | int alloc) | |
3ec09742 CL |
773 | { |
774 | if (s->flags & SLAB_TRACE) { | |
775 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
776 | s->name, | |
777 | alloc ? "alloc" : "free", | |
778 | object, page->inuse, | |
779 | page->freelist); | |
780 | ||
781 | if (!alloc) | |
782 | print_section("Object", (void *)object, s->objsize); | |
783 | ||
784 | dump_stack(); | |
785 | } | |
786 | } | |
787 | ||
643b1138 | 788 | /* |
672bba3a | 789 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 790 | */ |
e95eed57 | 791 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 792 | { |
643b1138 CL |
793 | spin_lock(&n->list_lock); |
794 | list_add(&page->lru, &n->full); | |
795 | spin_unlock(&n->list_lock); | |
796 | } | |
797 | ||
798 | static void remove_full(struct kmem_cache *s, struct page *page) | |
799 | { | |
800 | struct kmem_cache_node *n; | |
801 | ||
802 | if (!(s->flags & SLAB_STORE_USER)) | |
803 | return; | |
804 | ||
805 | n = get_node(s, page_to_nid(page)); | |
806 | ||
807 | spin_lock(&n->list_lock); | |
808 | list_del(&page->lru); | |
809 | spin_unlock(&n->list_lock); | |
810 | } | |
811 | ||
0f389ec6 CL |
812 | /* Tracking of the number of slabs for debugging purposes */ |
813 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
814 | { | |
815 | struct kmem_cache_node *n = get_node(s, node); | |
816 | ||
817 | return atomic_long_read(&n->nr_slabs); | |
818 | } | |
819 | ||
26c02cf0 AB |
820 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
821 | { | |
822 | return atomic_long_read(&n->nr_slabs); | |
823 | } | |
824 | ||
205ab99d | 825 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
826 | { |
827 | struct kmem_cache_node *n = get_node(s, node); | |
828 | ||
829 | /* | |
830 | * May be called early in order to allocate a slab for the | |
831 | * kmem_cache_node structure. Solve the chicken-egg | |
832 | * dilemma by deferring the increment of the count during | |
833 | * bootstrap (see early_kmem_cache_node_alloc). | |
834 | */ | |
205ab99d | 835 | if (!NUMA_BUILD || n) { |
0f389ec6 | 836 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
837 | atomic_long_add(objects, &n->total_objects); |
838 | } | |
0f389ec6 | 839 | } |
205ab99d | 840 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
841 | { |
842 | struct kmem_cache_node *n = get_node(s, node); | |
843 | ||
844 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 845 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
846 | } |
847 | ||
848 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
849 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
850 | void *object) | |
851 | { | |
852 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
853 | return; | |
854 | ||
855 | init_object(s, object, 0); | |
856 | init_tracking(s, object); | |
857 | } | |
858 | ||
859 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 860 | void *object, unsigned long addr) |
81819f0f CL |
861 | { |
862 | if (!check_slab(s, page)) | |
863 | goto bad; | |
864 | ||
d692ef6d | 865 | if (!on_freelist(s, page, object)) { |
24922684 | 866 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 867 | goto bad; |
81819f0f CL |
868 | } |
869 | ||
870 | if (!check_valid_pointer(s, page, object)) { | |
871 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 872 | goto bad; |
81819f0f CL |
873 | } |
874 | ||
d692ef6d | 875 | if (!check_object(s, page, object, 0)) |
81819f0f | 876 | goto bad; |
81819f0f | 877 | |
3ec09742 CL |
878 | /* Success perform special debug activities for allocs */ |
879 | if (s->flags & SLAB_STORE_USER) | |
880 | set_track(s, object, TRACK_ALLOC, addr); | |
881 | trace(s, page, object, 1); | |
882 | init_object(s, object, 1); | |
81819f0f | 883 | return 1; |
3ec09742 | 884 | |
81819f0f CL |
885 | bad: |
886 | if (PageSlab(page)) { | |
887 | /* | |
888 | * If this is a slab page then lets do the best we can | |
889 | * to avoid issues in the future. Marking all objects | |
672bba3a | 890 | * as used avoids touching the remaining objects. |
81819f0f | 891 | */ |
24922684 | 892 | slab_fix(s, "Marking all objects used"); |
39b26464 | 893 | page->inuse = page->objects; |
a973e9dd | 894 | page->freelist = NULL; |
81819f0f CL |
895 | } |
896 | return 0; | |
897 | } | |
898 | ||
3ec09742 | 899 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 900 | void *object, unsigned long addr) |
81819f0f CL |
901 | { |
902 | if (!check_slab(s, page)) | |
903 | goto fail; | |
904 | ||
905 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 906 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
907 | goto fail; |
908 | } | |
909 | ||
910 | if (on_freelist(s, page, object)) { | |
24922684 | 911 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
912 | goto fail; |
913 | } | |
914 | ||
915 | if (!check_object(s, page, object, 1)) | |
916 | return 0; | |
917 | ||
918 | if (unlikely(s != page->slab)) { | |
3adbefee | 919 | if (!PageSlab(page)) { |
70d71228 CL |
920 | slab_err(s, page, "Attempt to free object(0x%p) " |
921 | "outside of slab", object); | |
3adbefee | 922 | } else if (!page->slab) { |
81819f0f | 923 | printk(KERN_ERR |
70d71228 | 924 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 925 | object); |
70d71228 | 926 | dump_stack(); |
06428780 | 927 | } else |
24922684 CL |
928 | object_err(s, page, object, |
929 | "page slab pointer corrupt."); | |
81819f0f CL |
930 | goto fail; |
931 | } | |
3ec09742 CL |
932 | |
933 | /* Special debug activities for freeing objects */ | |
8a38082d | 934 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
935 | remove_full(s, page); |
936 | if (s->flags & SLAB_STORE_USER) | |
937 | set_track(s, object, TRACK_FREE, addr); | |
938 | trace(s, page, object, 0); | |
939 | init_object(s, object, 0); | |
81819f0f | 940 | return 1; |
3ec09742 | 941 | |
81819f0f | 942 | fail: |
24922684 | 943 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
944 | return 0; |
945 | } | |
946 | ||
41ecc55b CL |
947 | static int __init setup_slub_debug(char *str) |
948 | { | |
f0630fff CL |
949 | slub_debug = DEBUG_DEFAULT_FLAGS; |
950 | if (*str++ != '=' || !*str) | |
951 | /* | |
952 | * No options specified. Switch on full debugging. | |
953 | */ | |
954 | goto out; | |
955 | ||
956 | if (*str == ',') | |
957 | /* | |
958 | * No options but restriction on slabs. This means full | |
959 | * debugging for slabs matching a pattern. | |
960 | */ | |
961 | goto check_slabs; | |
962 | ||
fa5ec8a1 DR |
963 | if (tolower(*str) == 'o') { |
964 | /* | |
965 | * Avoid enabling debugging on caches if its minimum order | |
966 | * would increase as a result. | |
967 | */ | |
968 | disable_higher_order_debug = 1; | |
969 | goto out; | |
970 | } | |
971 | ||
f0630fff CL |
972 | slub_debug = 0; |
973 | if (*str == '-') | |
974 | /* | |
975 | * Switch off all debugging measures. | |
976 | */ | |
977 | goto out; | |
978 | ||
979 | /* | |
980 | * Determine which debug features should be switched on | |
981 | */ | |
06428780 | 982 | for (; *str && *str != ','; str++) { |
f0630fff CL |
983 | switch (tolower(*str)) { |
984 | case 'f': | |
985 | slub_debug |= SLAB_DEBUG_FREE; | |
986 | break; | |
987 | case 'z': | |
988 | slub_debug |= SLAB_RED_ZONE; | |
989 | break; | |
990 | case 'p': | |
991 | slub_debug |= SLAB_POISON; | |
992 | break; | |
993 | case 'u': | |
994 | slub_debug |= SLAB_STORE_USER; | |
995 | break; | |
996 | case 't': | |
997 | slub_debug |= SLAB_TRACE; | |
998 | break; | |
4c13dd3b DM |
999 | case 'a': |
1000 | slub_debug |= SLAB_FAILSLAB; | |
1001 | break; | |
f0630fff CL |
1002 | default: |
1003 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1004 | "unknown. skipped\n", *str); |
f0630fff | 1005 | } |
41ecc55b CL |
1006 | } |
1007 | ||
f0630fff | 1008 | check_slabs: |
41ecc55b CL |
1009 | if (*str == ',') |
1010 | slub_debug_slabs = str + 1; | |
f0630fff | 1011 | out: |
41ecc55b CL |
1012 | return 1; |
1013 | } | |
1014 | ||
1015 | __setup("slub_debug", setup_slub_debug); | |
1016 | ||
ba0268a8 CL |
1017 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1018 | unsigned long flags, const char *name, | |
51cc5068 | 1019 | void (*ctor)(void *)) |
41ecc55b CL |
1020 | { |
1021 | /* | |
e153362a | 1022 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1023 | */ |
e153362a | 1024 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1025 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1026 | flags |= slub_debug; | |
ba0268a8 CL |
1027 | |
1028 | return flags; | |
41ecc55b CL |
1029 | } |
1030 | #else | |
3ec09742 CL |
1031 | static inline void setup_object_debug(struct kmem_cache *s, |
1032 | struct page *page, void *object) {} | |
41ecc55b | 1033 | |
3ec09742 | 1034 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1035 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1036 | |
3ec09742 | 1037 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1038 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1039 | |
41ecc55b CL |
1040 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1041 | { return 1; } | |
1042 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1043 | void *object, int active) { return 1; } | |
3ec09742 | 1044 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1045 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1046 | unsigned long flags, const char *name, | |
51cc5068 | 1047 | void (*ctor)(void *)) |
ba0268a8 CL |
1048 | { |
1049 | return flags; | |
1050 | } | |
41ecc55b | 1051 | #define slub_debug 0 |
0f389ec6 | 1052 | |
fdaa45e9 IM |
1053 | #define disable_higher_order_debug 0 |
1054 | ||
0f389ec6 CL |
1055 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1056 | { return 0; } | |
26c02cf0 AB |
1057 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1058 | { return 0; } | |
205ab99d CL |
1059 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1060 | int objects) {} | |
1061 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1062 | int objects) {} | |
41ecc55b | 1063 | #endif |
205ab99d | 1064 | |
81819f0f CL |
1065 | /* |
1066 | * Slab allocation and freeing | |
1067 | */ | |
65c3376a CL |
1068 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1069 | struct kmem_cache_order_objects oo) | |
1070 | { | |
1071 | int order = oo_order(oo); | |
1072 | ||
b1eeab67 VN |
1073 | flags |= __GFP_NOTRACK; |
1074 | ||
65c3376a CL |
1075 | if (node == -1) |
1076 | return alloc_pages(flags, order); | |
1077 | else | |
6b65aaf3 | 1078 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1079 | } |
1080 | ||
81819f0f CL |
1081 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1082 | { | |
06428780 | 1083 | struct page *page; |
834f3d11 | 1084 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1085 | gfp_t alloc_gfp; |
81819f0f | 1086 | |
b7a49f0d | 1087 | flags |= s->allocflags; |
e12ba74d | 1088 | |
ba52270d PE |
1089 | /* |
1090 | * Let the initial higher-order allocation fail under memory pressure | |
1091 | * so we fall-back to the minimum order allocation. | |
1092 | */ | |
1093 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1094 | ||
1095 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1096 | if (unlikely(!page)) { |
1097 | oo = s->min; | |
1098 | /* | |
1099 | * Allocation may have failed due to fragmentation. | |
1100 | * Try a lower order alloc if possible | |
1101 | */ | |
1102 | page = alloc_slab_page(flags, node, oo); | |
1103 | if (!page) | |
1104 | return NULL; | |
81819f0f | 1105 | |
84e554e6 | 1106 | stat(s, ORDER_FALLBACK); |
65c3376a | 1107 | } |
5a896d9e VN |
1108 | |
1109 | if (kmemcheck_enabled | |
5086c389 | 1110 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1111 | int pages = 1 << oo_order(oo); |
1112 | ||
1113 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1114 | ||
1115 | /* | |
1116 | * Objects from caches that have a constructor don't get | |
1117 | * cleared when they're allocated, so we need to do it here. | |
1118 | */ | |
1119 | if (s->ctor) | |
1120 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1121 | else | |
1122 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1123 | } |
1124 | ||
834f3d11 | 1125 | page->objects = oo_objects(oo); |
81819f0f CL |
1126 | mod_zone_page_state(page_zone(page), |
1127 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1128 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1129 | 1 << oo_order(oo)); |
81819f0f CL |
1130 | |
1131 | return page; | |
1132 | } | |
1133 | ||
1134 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1135 | void *object) | |
1136 | { | |
3ec09742 | 1137 | setup_object_debug(s, page, object); |
4f104934 | 1138 | if (unlikely(s->ctor)) |
51cc5068 | 1139 | s->ctor(object); |
81819f0f CL |
1140 | } |
1141 | ||
1142 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1143 | { | |
1144 | struct page *page; | |
81819f0f | 1145 | void *start; |
81819f0f CL |
1146 | void *last; |
1147 | void *p; | |
1148 | ||
6cb06229 | 1149 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1150 | |
6cb06229 CL |
1151 | page = allocate_slab(s, |
1152 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1153 | if (!page) |
1154 | goto out; | |
1155 | ||
205ab99d | 1156 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1157 | page->slab = s; |
1158 | page->flags |= 1 << PG_slab; | |
1159 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1160 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1161 | __SetPageSlubDebug(page); |
81819f0f CL |
1162 | |
1163 | start = page_address(page); | |
81819f0f CL |
1164 | |
1165 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1166 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1167 | |
1168 | last = start; | |
224a88be | 1169 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1170 | setup_object(s, page, last); |
1171 | set_freepointer(s, last, p); | |
1172 | last = p; | |
1173 | } | |
1174 | setup_object(s, page, last); | |
a973e9dd | 1175 | set_freepointer(s, last, NULL); |
81819f0f CL |
1176 | |
1177 | page->freelist = start; | |
1178 | page->inuse = 0; | |
1179 | out: | |
81819f0f CL |
1180 | return page; |
1181 | } | |
1182 | ||
1183 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1184 | { | |
834f3d11 CL |
1185 | int order = compound_order(page); |
1186 | int pages = 1 << order; | |
81819f0f | 1187 | |
8a38082d | 1188 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1189 | void *p; |
1190 | ||
1191 | slab_pad_check(s, page); | |
224a88be CL |
1192 | for_each_object(p, s, page_address(page), |
1193 | page->objects) | |
81819f0f | 1194 | check_object(s, page, p, 0); |
8a38082d | 1195 | __ClearPageSlubDebug(page); |
81819f0f CL |
1196 | } |
1197 | ||
b1eeab67 | 1198 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1199 | |
81819f0f CL |
1200 | mod_zone_page_state(page_zone(page), |
1201 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1202 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1203 | -pages); |
81819f0f | 1204 | |
49bd5221 CL |
1205 | __ClearPageSlab(page); |
1206 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1207 | if (current->reclaim_state) |
1208 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1209 | __free_pages(page, order); |
81819f0f CL |
1210 | } |
1211 | ||
1212 | static void rcu_free_slab(struct rcu_head *h) | |
1213 | { | |
1214 | struct page *page; | |
1215 | ||
1216 | page = container_of((struct list_head *)h, struct page, lru); | |
1217 | __free_slab(page->slab, page); | |
1218 | } | |
1219 | ||
1220 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1221 | { | |
1222 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1223 | /* | |
1224 | * RCU free overloads the RCU head over the LRU | |
1225 | */ | |
1226 | struct rcu_head *head = (void *)&page->lru; | |
1227 | ||
1228 | call_rcu(head, rcu_free_slab); | |
1229 | } else | |
1230 | __free_slab(s, page); | |
1231 | } | |
1232 | ||
1233 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1234 | { | |
205ab99d | 1235 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1236 | free_slab(s, page); |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Per slab locking using the pagelock | |
1241 | */ | |
1242 | static __always_inline void slab_lock(struct page *page) | |
1243 | { | |
1244 | bit_spin_lock(PG_locked, &page->flags); | |
1245 | } | |
1246 | ||
1247 | static __always_inline void slab_unlock(struct page *page) | |
1248 | { | |
a76d3546 | 1249 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1250 | } |
1251 | ||
1252 | static __always_inline int slab_trylock(struct page *page) | |
1253 | { | |
1254 | int rc = 1; | |
1255 | ||
1256 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1257 | return rc; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * Management of partially allocated slabs | |
1262 | */ | |
7c2e132c CL |
1263 | static void add_partial(struct kmem_cache_node *n, |
1264 | struct page *page, int tail) | |
81819f0f | 1265 | { |
e95eed57 CL |
1266 | spin_lock(&n->list_lock); |
1267 | n->nr_partial++; | |
7c2e132c CL |
1268 | if (tail) |
1269 | list_add_tail(&page->lru, &n->partial); | |
1270 | else | |
1271 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1272 | spin_unlock(&n->list_lock); |
1273 | } | |
1274 | ||
0121c619 | 1275 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1276 | { |
1277 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1278 | ||
1279 | spin_lock(&n->list_lock); | |
1280 | list_del(&page->lru); | |
1281 | n->nr_partial--; | |
1282 | spin_unlock(&n->list_lock); | |
1283 | } | |
1284 | ||
1285 | /* | |
672bba3a | 1286 | * Lock slab and remove from the partial list. |
81819f0f | 1287 | * |
672bba3a | 1288 | * Must hold list_lock. |
81819f0f | 1289 | */ |
0121c619 CL |
1290 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1291 | struct page *page) | |
81819f0f CL |
1292 | { |
1293 | if (slab_trylock(page)) { | |
1294 | list_del(&page->lru); | |
1295 | n->nr_partial--; | |
8a38082d | 1296 | __SetPageSlubFrozen(page); |
81819f0f CL |
1297 | return 1; |
1298 | } | |
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | /* | |
672bba3a | 1303 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1304 | */ |
1305 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1306 | { | |
1307 | struct page *page; | |
1308 | ||
1309 | /* | |
1310 | * Racy check. If we mistakenly see no partial slabs then we | |
1311 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1312 | * partial slab and there is none available then get_partials() |
1313 | * will return NULL. | |
81819f0f CL |
1314 | */ |
1315 | if (!n || !n->nr_partial) | |
1316 | return NULL; | |
1317 | ||
1318 | spin_lock(&n->list_lock); | |
1319 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1320 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1321 | goto out; |
1322 | page = NULL; | |
1323 | out: | |
1324 | spin_unlock(&n->list_lock); | |
1325 | return page; | |
1326 | } | |
1327 | ||
1328 | /* | |
672bba3a | 1329 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1330 | */ |
1331 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1332 | { | |
1333 | #ifdef CONFIG_NUMA | |
1334 | struct zonelist *zonelist; | |
dd1a239f | 1335 | struct zoneref *z; |
54a6eb5c MG |
1336 | struct zone *zone; |
1337 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1338 | struct page *page; |
1339 | ||
1340 | /* | |
672bba3a CL |
1341 | * The defrag ratio allows a configuration of the tradeoffs between |
1342 | * inter node defragmentation and node local allocations. A lower | |
1343 | * defrag_ratio increases the tendency to do local allocations | |
1344 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1345 | * |
672bba3a CL |
1346 | * If the defrag_ratio is set to 0 then kmalloc() always |
1347 | * returns node local objects. If the ratio is higher then kmalloc() | |
1348 | * may return off node objects because partial slabs are obtained | |
1349 | * from other nodes and filled up. | |
81819f0f | 1350 | * |
6446faa2 | 1351 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1352 | * defrag_ratio = 1000) then every (well almost) allocation will |
1353 | * first attempt to defrag slab caches on other nodes. This means | |
1354 | * scanning over all nodes to look for partial slabs which may be | |
1355 | * expensive if we do it every time we are trying to find a slab | |
1356 | * with available objects. | |
81819f0f | 1357 | */ |
9824601e CL |
1358 | if (!s->remote_node_defrag_ratio || |
1359 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1360 | return NULL; |
1361 | ||
c0ff7453 | 1362 | get_mems_allowed(); |
0e88460d | 1363 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1364 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1365 | struct kmem_cache_node *n; |
1366 | ||
54a6eb5c | 1367 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1368 | |
54a6eb5c | 1369 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1370 | n->nr_partial > s->min_partial) { |
81819f0f | 1371 | page = get_partial_node(n); |
c0ff7453 MX |
1372 | if (page) { |
1373 | put_mems_allowed(); | |
81819f0f | 1374 | return page; |
c0ff7453 | 1375 | } |
81819f0f CL |
1376 | } |
1377 | } | |
c0ff7453 | 1378 | put_mems_allowed(); |
81819f0f CL |
1379 | #endif |
1380 | return NULL; | |
1381 | } | |
1382 | ||
1383 | /* | |
1384 | * Get a partial page, lock it and return it. | |
1385 | */ | |
1386 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1387 | { | |
1388 | struct page *page; | |
1389 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1390 | ||
1391 | page = get_partial_node(get_node(s, searchnode)); | |
1392 | if (page || (flags & __GFP_THISNODE)) | |
1393 | return page; | |
1394 | ||
1395 | return get_any_partial(s, flags); | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Move a page back to the lists. | |
1400 | * | |
1401 | * Must be called with the slab lock held. | |
1402 | * | |
1403 | * On exit the slab lock will have been dropped. | |
1404 | */ | |
7c2e132c | 1405 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1406 | { |
e95eed57 CL |
1407 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1408 | ||
8a38082d | 1409 | __ClearPageSlubFrozen(page); |
81819f0f | 1410 | if (page->inuse) { |
e95eed57 | 1411 | |
a973e9dd | 1412 | if (page->freelist) { |
7c2e132c | 1413 | add_partial(n, page, tail); |
84e554e6 | 1414 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1415 | } else { |
84e554e6 | 1416 | stat(s, DEACTIVATE_FULL); |
8a38082d AW |
1417 | if (SLABDEBUG && PageSlubDebug(page) && |
1418 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1419 | add_full(n, page); |
1420 | } | |
81819f0f CL |
1421 | slab_unlock(page); |
1422 | } else { | |
84e554e6 | 1423 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1424 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1425 | /* |
672bba3a CL |
1426 | * Adding an empty slab to the partial slabs in order |
1427 | * to avoid page allocator overhead. This slab needs | |
1428 | * to come after the other slabs with objects in | |
6446faa2 CL |
1429 | * so that the others get filled first. That way the |
1430 | * size of the partial list stays small. | |
1431 | * | |
0121c619 CL |
1432 | * kmem_cache_shrink can reclaim any empty slabs from |
1433 | * the partial list. | |
e95eed57 | 1434 | */ |
7c2e132c | 1435 | add_partial(n, page, 1); |
e95eed57 CL |
1436 | slab_unlock(page); |
1437 | } else { | |
1438 | slab_unlock(page); | |
84e554e6 | 1439 | stat(s, FREE_SLAB); |
e95eed57 CL |
1440 | discard_slab(s, page); |
1441 | } | |
81819f0f CL |
1442 | } |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * Remove the cpu slab | |
1447 | */ | |
dfb4f096 | 1448 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1449 | { |
dfb4f096 | 1450 | struct page *page = c->page; |
7c2e132c | 1451 | int tail = 1; |
8ff12cfc | 1452 | |
b773ad73 | 1453 | if (page->freelist) |
84e554e6 | 1454 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1455 | /* |
6446faa2 | 1456 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1457 | * because both freelists are empty. So this is unlikely |
1458 | * to occur. | |
1459 | */ | |
a973e9dd | 1460 | while (unlikely(c->freelist)) { |
894b8788 CL |
1461 | void **object; |
1462 | ||
7c2e132c CL |
1463 | tail = 0; /* Hot objects. Put the slab first */ |
1464 | ||
894b8788 | 1465 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1466 | object = c->freelist; |
ff12059e | 1467 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1468 | |
1469 | /* And put onto the regular freelist */ | |
ff12059e | 1470 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1471 | page->freelist = object; |
1472 | page->inuse--; | |
1473 | } | |
dfb4f096 | 1474 | c->page = NULL; |
7c2e132c | 1475 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1476 | } |
1477 | ||
dfb4f096 | 1478 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1479 | { |
84e554e6 | 1480 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1481 | slab_lock(c->page); |
1482 | deactivate_slab(s, c); | |
81819f0f CL |
1483 | } |
1484 | ||
1485 | /* | |
1486 | * Flush cpu slab. | |
6446faa2 | 1487 | * |
81819f0f CL |
1488 | * Called from IPI handler with interrupts disabled. |
1489 | */ | |
0c710013 | 1490 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1491 | { |
9dfc6e68 | 1492 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1493 | |
dfb4f096 CL |
1494 | if (likely(c && c->page)) |
1495 | flush_slab(s, c); | |
81819f0f CL |
1496 | } |
1497 | ||
1498 | static void flush_cpu_slab(void *d) | |
1499 | { | |
1500 | struct kmem_cache *s = d; | |
81819f0f | 1501 | |
dfb4f096 | 1502 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1503 | } |
1504 | ||
1505 | static void flush_all(struct kmem_cache *s) | |
1506 | { | |
15c8b6c1 | 1507 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1508 | } |
1509 | ||
dfb4f096 CL |
1510 | /* |
1511 | * Check if the objects in a per cpu structure fit numa | |
1512 | * locality expectations. | |
1513 | */ | |
1514 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1515 | { | |
1516 | #ifdef CONFIG_NUMA | |
1517 | if (node != -1 && c->node != node) | |
1518 | return 0; | |
1519 | #endif | |
1520 | return 1; | |
1521 | } | |
1522 | ||
781b2ba6 PE |
1523 | static int count_free(struct page *page) |
1524 | { | |
1525 | return page->objects - page->inuse; | |
1526 | } | |
1527 | ||
1528 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1529 | int (*get_count)(struct page *)) | |
1530 | { | |
1531 | unsigned long flags; | |
1532 | unsigned long x = 0; | |
1533 | struct page *page; | |
1534 | ||
1535 | spin_lock_irqsave(&n->list_lock, flags); | |
1536 | list_for_each_entry(page, &n->partial, lru) | |
1537 | x += get_count(page); | |
1538 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1539 | return x; | |
1540 | } | |
1541 | ||
26c02cf0 AB |
1542 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1543 | { | |
1544 | #ifdef CONFIG_SLUB_DEBUG | |
1545 | return atomic_long_read(&n->total_objects); | |
1546 | #else | |
1547 | return 0; | |
1548 | #endif | |
1549 | } | |
1550 | ||
781b2ba6 PE |
1551 | static noinline void |
1552 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1553 | { | |
1554 | int node; | |
1555 | ||
1556 | printk(KERN_WARNING | |
1557 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1558 | nid, gfpflags); | |
1559 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1560 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1561 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1562 | ||
fa5ec8a1 DR |
1563 | if (oo_order(s->min) > get_order(s->objsize)) |
1564 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1565 | "slub_debug=O to disable.\n", s->name); | |
1566 | ||
781b2ba6 PE |
1567 | for_each_online_node(node) { |
1568 | struct kmem_cache_node *n = get_node(s, node); | |
1569 | unsigned long nr_slabs; | |
1570 | unsigned long nr_objs; | |
1571 | unsigned long nr_free; | |
1572 | ||
1573 | if (!n) | |
1574 | continue; | |
1575 | ||
26c02cf0 AB |
1576 | nr_free = count_partial(n, count_free); |
1577 | nr_slabs = node_nr_slabs(n); | |
1578 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1579 | |
1580 | printk(KERN_WARNING | |
1581 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1582 | node, nr_slabs, nr_objs, nr_free); | |
1583 | } | |
1584 | } | |
1585 | ||
81819f0f | 1586 | /* |
894b8788 CL |
1587 | * Slow path. The lockless freelist is empty or we need to perform |
1588 | * debugging duties. | |
1589 | * | |
1590 | * Interrupts are disabled. | |
81819f0f | 1591 | * |
894b8788 CL |
1592 | * Processing is still very fast if new objects have been freed to the |
1593 | * regular freelist. In that case we simply take over the regular freelist | |
1594 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1595 | * |
894b8788 CL |
1596 | * If that is not working then we fall back to the partial lists. We take the |
1597 | * first element of the freelist as the object to allocate now and move the | |
1598 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1599 | * |
894b8788 | 1600 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1601 | * we need to allocate a new slab. This is the slowest path since it involves |
1602 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1603 | */ |
ce71e27c EGM |
1604 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1605 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1606 | { |
81819f0f | 1607 | void **object; |
dfb4f096 | 1608 | struct page *new; |
81819f0f | 1609 | |
e72e9c23 LT |
1610 | /* We handle __GFP_ZERO in the caller */ |
1611 | gfpflags &= ~__GFP_ZERO; | |
1612 | ||
dfb4f096 | 1613 | if (!c->page) |
81819f0f CL |
1614 | goto new_slab; |
1615 | ||
dfb4f096 CL |
1616 | slab_lock(c->page); |
1617 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1618 | goto another_slab; |
6446faa2 | 1619 | |
84e554e6 | 1620 | stat(s, ALLOC_REFILL); |
6446faa2 | 1621 | |
894b8788 | 1622 | load_freelist: |
dfb4f096 | 1623 | object = c->page->freelist; |
a973e9dd | 1624 | if (unlikely(!object)) |
81819f0f | 1625 | goto another_slab; |
8a38082d | 1626 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1627 | goto debug; |
1628 | ||
ff12059e | 1629 | c->freelist = get_freepointer(s, object); |
39b26464 | 1630 | c->page->inuse = c->page->objects; |
a973e9dd | 1631 | c->page->freelist = NULL; |
dfb4f096 | 1632 | c->node = page_to_nid(c->page); |
1f84260c | 1633 | unlock_out: |
dfb4f096 | 1634 | slab_unlock(c->page); |
84e554e6 | 1635 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1636 | return object; |
1637 | ||
1638 | another_slab: | |
dfb4f096 | 1639 | deactivate_slab(s, c); |
81819f0f CL |
1640 | |
1641 | new_slab: | |
dfb4f096 CL |
1642 | new = get_partial(s, gfpflags, node); |
1643 | if (new) { | |
1644 | c->page = new; | |
84e554e6 | 1645 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1646 | goto load_freelist; |
81819f0f CL |
1647 | } |
1648 | ||
b811c202 CL |
1649 | if (gfpflags & __GFP_WAIT) |
1650 | local_irq_enable(); | |
1651 | ||
dfb4f096 | 1652 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1653 | |
1654 | if (gfpflags & __GFP_WAIT) | |
1655 | local_irq_disable(); | |
1656 | ||
dfb4f096 | 1657 | if (new) { |
9dfc6e68 | 1658 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1659 | stat(s, ALLOC_SLAB); |
05aa3450 | 1660 | if (c->page) |
dfb4f096 | 1661 | flush_slab(s, c); |
dfb4f096 | 1662 | slab_lock(new); |
8a38082d | 1663 | __SetPageSlubFrozen(new); |
dfb4f096 | 1664 | c->page = new; |
4b6f0750 | 1665 | goto load_freelist; |
81819f0f | 1666 | } |
95f85989 PE |
1667 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1668 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1669 | return NULL; |
81819f0f | 1670 | debug: |
dfb4f096 | 1671 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1672 | goto another_slab; |
894b8788 | 1673 | |
dfb4f096 | 1674 | c->page->inuse++; |
ff12059e | 1675 | c->page->freelist = get_freepointer(s, object); |
ee3c72a1 | 1676 | c->node = -1; |
1f84260c | 1677 | goto unlock_out; |
894b8788 CL |
1678 | } |
1679 | ||
1680 | /* | |
1681 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1682 | * have the fastpath folded into their functions. So no function call | |
1683 | * overhead for requests that can be satisfied on the fastpath. | |
1684 | * | |
1685 | * The fastpath works by first checking if the lockless freelist can be used. | |
1686 | * If not then __slab_alloc is called for slow processing. | |
1687 | * | |
1688 | * Otherwise we can simply pick the next object from the lockless free list. | |
1689 | */ | |
06428780 | 1690 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1691 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1692 | { |
894b8788 | 1693 | void **object; |
dfb4f096 | 1694 | struct kmem_cache_cpu *c; |
1f84260c CL |
1695 | unsigned long flags; |
1696 | ||
dcce284a | 1697 | gfpflags &= gfp_allowed_mask; |
7e85ee0c | 1698 | |
cf40bd16 | 1699 | lockdep_trace_alloc(gfpflags); |
89124d70 | 1700 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1701 | |
4c13dd3b | 1702 | if (should_failslab(s->objsize, gfpflags, s->flags)) |
773ff60e | 1703 | return NULL; |
1f84260c | 1704 | |
894b8788 | 1705 | local_irq_save(flags); |
9dfc6e68 CL |
1706 | c = __this_cpu_ptr(s->cpu_slab); |
1707 | object = c->freelist; | |
9dfc6e68 | 1708 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1709 | |
dfb4f096 | 1710 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1711 | |
1712 | else { | |
ff12059e | 1713 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1714 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1715 | } |
1716 | local_irq_restore(flags); | |
d07dbea4 | 1717 | |
74e2134f | 1718 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1719 | memset(object, 0, s->objsize); |
d07dbea4 | 1720 | |
ff12059e CL |
1721 | kmemcheck_slab_alloc(s, gfpflags, object, s->objsize); |
1722 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, gfpflags); | |
5a896d9e | 1723 | |
894b8788 | 1724 | return object; |
81819f0f CL |
1725 | } |
1726 | ||
1727 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1728 | { | |
5b882be4 EGM |
1729 | void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_); |
1730 | ||
ca2b84cb | 1731 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1732 | |
1733 | return ret; | |
81819f0f CL |
1734 | } |
1735 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1736 | ||
0f24f128 | 1737 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1738 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) |
1739 | { | |
1740 | return slab_alloc(s, gfpflags, -1, _RET_IP_); | |
1741 | } | |
1742 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1743 | #endif | |
1744 | ||
81819f0f CL |
1745 | #ifdef CONFIG_NUMA |
1746 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1747 | { | |
5b882be4 EGM |
1748 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1749 | ||
ca2b84cb EGM |
1750 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1751 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1752 | |
1753 | return ret; | |
81819f0f CL |
1754 | } |
1755 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1756 | #endif | |
1757 | ||
0f24f128 | 1758 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1759 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, |
1760 | gfp_t gfpflags, | |
1761 | int node) | |
1762 | { | |
1763 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1764 | } | |
1765 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1766 | #endif | |
1767 | ||
81819f0f | 1768 | /* |
894b8788 CL |
1769 | * Slow patch handling. This may still be called frequently since objects |
1770 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1771 | * |
894b8788 CL |
1772 | * So we still attempt to reduce cache line usage. Just take the slab |
1773 | * lock and free the item. If there is no additional partial page | |
1774 | * handling required then we can return immediately. | |
81819f0f | 1775 | */ |
894b8788 | 1776 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1777 | void *x, unsigned long addr) |
81819f0f CL |
1778 | { |
1779 | void *prior; | |
1780 | void **object = (void *)x; | |
81819f0f | 1781 | |
84e554e6 | 1782 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1783 | slab_lock(page); |
1784 | ||
8a38082d | 1785 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1786 | goto debug; |
6446faa2 | 1787 | |
81819f0f | 1788 | checks_ok: |
ff12059e CL |
1789 | prior = page->freelist; |
1790 | set_freepointer(s, object, prior); | |
81819f0f CL |
1791 | page->freelist = object; |
1792 | page->inuse--; | |
1793 | ||
8a38082d | 1794 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1795 | stat(s, FREE_FROZEN); |
81819f0f | 1796 | goto out_unlock; |
8ff12cfc | 1797 | } |
81819f0f CL |
1798 | |
1799 | if (unlikely(!page->inuse)) | |
1800 | goto slab_empty; | |
1801 | ||
1802 | /* | |
6446faa2 | 1803 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1804 | * then add it. |
1805 | */ | |
a973e9dd | 1806 | if (unlikely(!prior)) { |
7c2e132c | 1807 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1808 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1809 | } |
81819f0f CL |
1810 | |
1811 | out_unlock: | |
1812 | slab_unlock(page); | |
81819f0f CL |
1813 | return; |
1814 | ||
1815 | slab_empty: | |
a973e9dd | 1816 | if (prior) { |
81819f0f | 1817 | /* |
672bba3a | 1818 | * Slab still on the partial list. |
81819f0f CL |
1819 | */ |
1820 | remove_partial(s, page); | |
84e554e6 | 1821 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1822 | } |
81819f0f | 1823 | slab_unlock(page); |
84e554e6 | 1824 | stat(s, FREE_SLAB); |
81819f0f | 1825 | discard_slab(s, page); |
81819f0f CL |
1826 | return; |
1827 | ||
1828 | debug: | |
3ec09742 | 1829 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1830 | goto out_unlock; |
77c5e2d0 | 1831 | goto checks_ok; |
81819f0f CL |
1832 | } |
1833 | ||
894b8788 CL |
1834 | /* |
1835 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1836 | * can perform fastpath freeing without additional function calls. | |
1837 | * | |
1838 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1839 | * of this processor. This typically the case if we have just allocated | |
1840 | * the item before. | |
1841 | * | |
1842 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1843 | * with all sorts of special processing. | |
1844 | */ | |
06428780 | 1845 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1846 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1847 | { |
1848 | void **object = (void *)x; | |
dfb4f096 | 1849 | struct kmem_cache_cpu *c; |
1f84260c CL |
1850 | unsigned long flags; |
1851 | ||
06f22f13 | 1852 | kmemleak_free_recursive(x, s->flags); |
894b8788 | 1853 | local_irq_save(flags); |
9dfc6e68 | 1854 | c = __this_cpu_ptr(s->cpu_slab); |
ff12059e CL |
1855 | kmemcheck_slab_free(s, object, s->objsize); |
1856 | debug_check_no_locks_freed(object, s->objsize); | |
3ac7fe5a | 1857 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
ff12059e | 1858 | debug_check_no_obj_freed(object, s->objsize); |
ee3c72a1 | 1859 | if (likely(page == c->page && c->node >= 0)) { |
ff12059e | 1860 | set_freepointer(s, object, c->freelist); |
dfb4f096 | 1861 | c->freelist = object; |
84e554e6 | 1862 | stat(s, FREE_FASTPATH); |
894b8788 | 1863 | } else |
ff12059e | 1864 | __slab_free(s, page, x, addr); |
894b8788 CL |
1865 | |
1866 | local_irq_restore(flags); | |
1867 | } | |
1868 | ||
81819f0f CL |
1869 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1870 | { | |
77c5e2d0 | 1871 | struct page *page; |
81819f0f | 1872 | |
b49af68f | 1873 | page = virt_to_head_page(x); |
81819f0f | 1874 | |
ce71e27c | 1875 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1876 | |
ca2b84cb | 1877 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1878 | } |
1879 | EXPORT_SYMBOL(kmem_cache_free); | |
1880 | ||
e9beef18 | 1881 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1882 | static struct page *get_object_page(const void *x) |
1883 | { | |
b49af68f | 1884 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1885 | |
1886 | if (!PageSlab(page)) | |
1887 | return NULL; | |
1888 | ||
1889 | return page; | |
1890 | } | |
1891 | ||
1892 | /* | |
672bba3a CL |
1893 | * Object placement in a slab is made very easy because we always start at |
1894 | * offset 0. If we tune the size of the object to the alignment then we can | |
1895 | * get the required alignment by putting one properly sized object after | |
1896 | * another. | |
81819f0f CL |
1897 | * |
1898 | * Notice that the allocation order determines the sizes of the per cpu | |
1899 | * caches. Each processor has always one slab available for allocations. | |
1900 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1901 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1902 | * locking overhead. |
81819f0f CL |
1903 | */ |
1904 | ||
1905 | /* | |
1906 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1907 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1908 | * and increases the number of allocations possible without having to | |
1909 | * take the list_lock. | |
1910 | */ | |
1911 | static int slub_min_order; | |
114e9e89 | 1912 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1913 | static int slub_min_objects; |
81819f0f CL |
1914 | |
1915 | /* | |
1916 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1917 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1918 | */ |
1919 | static int slub_nomerge; | |
1920 | ||
81819f0f CL |
1921 | /* |
1922 | * Calculate the order of allocation given an slab object size. | |
1923 | * | |
672bba3a CL |
1924 | * The order of allocation has significant impact on performance and other |
1925 | * system components. Generally order 0 allocations should be preferred since | |
1926 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1927 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1928 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1929 | * would be wasted. |
1930 | * | |
1931 | * In order to reach satisfactory performance we must ensure that a minimum | |
1932 | * number of objects is in one slab. Otherwise we may generate too much | |
1933 | * activity on the partial lists which requires taking the list_lock. This is | |
1934 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1935 | * |
672bba3a CL |
1936 | * slub_max_order specifies the order where we begin to stop considering the |
1937 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1938 | * we try to keep the page order as low as possible. So we accept more waste | |
1939 | * of space in favor of a small page order. | |
81819f0f | 1940 | * |
672bba3a CL |
1941 | * Higher order allocations also allow the placement of more objects in a |
1942 | * slab and thereby reduce object handling overhead. If the user has | |
1943 | * requested a higher mininum order then we start with that one instead of | |
1944 | * the smallest order which will fit the object. | |
81819f0f | 1945 | */ |
5e6d444e CL |
1946 | static inline int slab_order(int size, int min_objects, |
1947 | int max_order, int fract_leftover) | |
81819f0f CL |
1948 | { |
1949 | int order; | |
1950 | int rem; | |
6300ea75 | 1951 | int min_order = slub_min_order; |
81819f0f | 1952 | |
210b5c06 CG |
1953 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1954 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1955 | |
6300ea75 | 1956 | for (order = max(min_order, |
5e6d444e CL |
1957 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1958 | order <= max_order; order++) { | |
81819f0f | 1959 | |
5e6d444e | 1960 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1961 | |
5e6d444e | 1962 | if (slab_size < min_objects * size) |
81819f0f CL |
1963 | continue; |
1964 | ||
1965 | rem = slab_size % size; | |
1966 | ||
5e6d444e | 1967 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1968 | break; |
1969 | ||
1970 | } | |
672bba3a | 1971 | |
81819f0f CL |
1972 | return order; |
1973 | } | |
1974 | ||
5e6d444e CL |
1975 | static inline int calculate_order(int size) |
1976 | { | |
1977 | int order; | |
1978 | int min_objects; | |
1979 | int fraction; | |
e8120ff1 | 1980 | int max_objects; |
5e6d444e CL |
1981 | |
1982 | /* | |
1983 | * Attempt to find best configuration for a slab. This | |
1984 | * works by first attempting to generate a layout with | |
1985 | * the best configuration and backing off gradually. | |
1986 | * | |
1987 | * First we reduce the acceptable waste in a slab. Then | |
1988 | * we reduce the minimum objects required in a slab. | |
1989 | */ | |
1990 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1991 | if (!min_objects) |
1992 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
1993 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
1994 | min_objects = min(min_objects, max_objects); | |
1995 | ||
5e6d444e | 1996 | while (min_objects > 1) { |
c124f5b5 | 1997 | fraction = 16; |
5e6d444e CL |
1998 | while (fraction >= 4) { |
1999 | order = slab_order(size, min_objects, | |
2000 | slub_max_order, fraction); | |
2001 | if (order <= slub_max_order) | |
2002 | return order; | |
2003 | fraction /= 2; | |
2004 | } | |
5086c389 | 2005 | min_objects--; |
5e6d444e CL |
2006 | } |
2007 | ||
2008 | /* | |
2009 | * We were unable to place multiple objects in a slab. Now | |
2010 | * lets see if we can place a single object there. | |
2011 | */ | |
2012 | order = slab_order(size, 1, slub_max_order, 1); | |
2013 | if (order <= slub_max_order) | |
2014 | return order; | |
2015 | ||
2016 | /* | |
2017 | * Doh this slab cannot be placed using slub_max_order. | |
2018 | */ | |
2019 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2020 | if (order < MAX_ORDER) |
5e6d444e CL |
2021 | return order; |
2022 | return -ENOSYS; | |
2023 | } | |
2024 | ||
81819f0f | 2025 | /* |
672bba3a | 2026 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2027 | */ |
2028 | static unsigned long calculate_alignment(unsigned long flags, | |
2029 | unsigned long align, unsigned long size) | |
2030 | { | |
2031 | /* | |
6446faa2 CL |
2032 | * If the user wants hardware cache aligned objects then follow that |
2033 | * suggestion if the object is sufficiently large. | |
81819f0f | 2034 | * |
6446faa2 CL |
2035 | * The hardware cache alignment cannot override the specified |
2036 | * alignment though. If that is greater then use it. | |
81819f0f | 2037 | */ |
b6210386 NP |
2038 | if (flags & SLAB_HWCACHE_ALIGN) { |
2039 | unsigned long ralign = cache_line_size(); | |
2040 | while (size <= ralign / 2) | |
2041 | ralign /= 2; | |
2042 | align = max(align, ralign); | |
2043 | } | |
81819f0f CL |
2044 | |
2045 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2046 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2047 | |
2048 | return ALIGN(align, sizeof(void *)); | |
2049 | } | |
2050 | ||
5595cffc PE |
2051 | static void |
2052 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2053 | { |
2054 | n->nr_partial = 0; | |
81819f0f CL |
2055 | spin_lock_init(&n->list_lock); |
2056 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2057 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2058 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2059 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2060 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2061 | #endif |
81819f0f CL |
2062 | } |
2063 | ||
91efd773 | 2064 | static DEFINE_PER_CPU(struct kmem_cache_cpu, kmalloc_percpu[KMALLOC_CACHES]); |
4c93c355 | 2065 | |
9dfc6e68 | 2066 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) |
4c93c355 | 2067 | { |
756dee75 | 2068 | if (s < kmalloc_caches + KMALLOC_CACHES && s >= kmalloc_caches) |
9dfc6e68 CL |
2069 | /* |
2070 | * Boot time creation of the kmalloc array. Use static per cpu data | |
2071 | * since the per cpu allocator is not available yet. | |
2072 | */ | |
1154fab7 | 2073 | s->cpu_slab = kmalloc_percpu + (s - kmalloc_caches); |
9dfc6e68 CL |
2074 | else |
2075 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); | |
4c93c355 | 2076 | |
9dfc6e68 CL |
2077 | if (!s->cpu_slab) |
2078 | return 0; | |
4c93c355 | 2079 | |
4c93c355 CL |
2080 | return 1; |
2081 | } | |
4c93c355 | 2082 | |
81819f0f CL |
2083 | #ifdef CONFIG_NUMA |
2084 | /* | |
2085 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2086 | * slab on the node for this slabcache. There are no concurrent accesses | |
2087 | * possible. | |
2088 | * | |
2089 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2090 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2091 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2092 | */ |
0094de92 | 2093 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2094 | { |
2095 | struct page *page; | |
2096 | struct kmem_cache_node *n; | |
ba84c73c | 2097 | unsigned long flags; |
81819f0f CL |
2098 | |
2099 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2100 | ||
a2f92ee7 | 2101 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2102 | |
2103 | BUG_ON(!page); | |
a2f92ee7 CL |
2104 | if (page_to_nid(page) != node) { |
2105 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2106 | "node %d\n", node); | |
2107 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2108 | "in order to be able to continue\n"); | |
2109 | } | |
2110 | ||
81819f0f CL |
2111 | n = page->freelist; |
2112 | BUG_ON(!n); | |
2113 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2114 | page->inuse++; | |
2115 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2116 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2117 | init_object(kmalloc_caches, n, 1); |
2118 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2119 | #endif |
5595cffc | 2120 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2121 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2122 | |
ba84c73c | 2123 | /* |
2124 | * lockdep requires consistent irq usage for each lock | |
2125 | * so even though there cannot be a race this early in | |
2126 | * the boot sequence, we still disable irqs. | |
2127 | */ | |
2128 | local_irq_save(flags); | |
7c2e132c | 2129 | add_partial(n, page, 0); |
ba84c73c | 2130 | local_irq_restore(flags); |
81819f0f CL |
2131 | } |
2132 | ||
2133 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2134 | { | |
2135 | int node; | |
2136 | ||
f64dc58c | 2137 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2138 | struct kmem_cache_node *n = s->node[node]; |
2139 | if (n && n != &s->local_node) | |
2140 | kmem_cache_free(kmalloc_caches, n); | |
2141 | s->node[node] = NULL; | |
2142 | } | |
2143 | } | |
2144 | ||
2145 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2146 | { | |
2147 | int node; | |
2148 | int local_node; | |
2149 | ||
91efd773 | 2150 | if (slab_state >= UP && (s < kmalloc_caches || |
111c7d82 | 2151 | s >= kmalloc_caches + KMALLOC_CACHES)) |
81819f0f CL |
2152 | local_node = page_to_nid(virt_to_page(s)); |
2153 | else | |
2154 | local_node = 0; | |
2155 | ||
f64dc58c | 2156 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2157 | struct kmem_cache_node *n; |
2158 | ||
2159 | if (local_node == node) | |
2160 | n = &s->local_node; | |
2161 | else { | |
2162 | if (slab_state == DOWN) { | |
0094de92 | 2163 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2164 | continue; |
2165 | } | |
2166 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2167 | gfpflags, node); | |
2168 | ||
2169 | if (!n) { | |
2170 | free_kmem_cache_nodes(s); | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | } | |
2175 | s->node[node] = n; | |
5595cffc | 2176 | init_kmem_cache_node(n, s); |
81819f0f CL |
2177 | } |
2178 | return 1; | |
2179 | } | |
2180 | #else | |
2181 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2182 | { | |
2183 | } | |
2184 | ||
2185 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2186 | { | |
5595cffc | 2187 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2188 | return 1; |
2189 | } | |
2190 | #endif | |
2191 | ||
c0bdb232 | 2192 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2193 | { |
2194 | if (min < MIN_PARTIAL) | |
2195 | min = MIN_PARTIAL; | |
2196 | else if (min > MAX_PARTIAL) | |
2197 | min = MAX_PARTIAL; | |
2198 | s->min_partial = min; | |
2199 | } | |
2200 | ||
81819f0f CL |
2201 | /* |
2202 | * calculate_sizes() determines the order and the distribution of data within | |
2203 | * a slab object. | |
2204 | */ | |
06b285dc | 2205 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2206 | { |
2207 | unsigned long flags = s->flags; | |
2208 | unsigned long size = s->objsize; | |
2209 | unsigned long align = s->align; | |
834f3d11 | 2210 | int order; |
81819f0f | 2211 | |
d8b42bf5 CL |
2212 | /* |
2213 | * Round up object size to the next word boundary. We can only | |
2214 | * place the free pointer at word boundaries and this determines | |
2215 | * the possible location of the free pointer. | |
2216 | */ | |
2217 | size = ALIGN(size, sizeof(void *)); | |
2218 | ||
2219 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2220 | /* |
2221 | * Determine if we can poison the object itself. If the user of | |
2222 | * the slab may touch the object after free or before allocation | |
2223 | * then we should never poison the object itself. | |
2224 | */ | |
2225 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2226 | !s->ctor) |
81819f0f CL |
2227 | s->flags |= __OBJECT_POISON; |
2228 | else | |
2229 | s->flags &= ~__OBJECT_POISON; | |
2230 | ||
81819f0f CL |
2231 | |
2232 | /* | |
672bba3a | 2233 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2234 | * end of the object and the free pointer. If not then add an |
672bba3a | 2235 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2236 | */ |
2237 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2238 | size += sizeof(void *); | |
41ecc55b | 2239 | #endif |
81819f0f CL |
2240 | |
2241 | /* | |
672bba3a CL |
2242 | * With that we have determined the number of bytes in actual use |
2243 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2244 | */ |
2245 | s->inuse = size; | |
2246 | ||
2247 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2248 | s->ctor)) { |
81819f0f CL |
2249 | /* |
2250 | * Relocate free pointer after the object if it is not | |
2251 | * permitted to overwrite the first word of the object on | |
2252 | * kmem_cache_free. | |
2253 | * | |
2254 | * This is the case if we do RCU, have a constructor or | |
2255 | * destructor or are poisoning the objects. | |
2256 | */ | |
2257 | s->offset = size; | |
2258 | size += sizeof(void *); | |
2259 | } | |
2260 | ||
c12b3c62 | 2261 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2262 | if (flags & SLAB_STORE_USER) |
2263 | /* | |
2264 | * Need to store information about allocs and frees after | |
2265 | * the object. | |
2266 | */ | |
2267 | size += 2 * sizeof(struct track); | |
2268 | ||
be7b3fbc | 2269 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2270 | /* |
2271 | * Add some empty padding so that we can catch | |
2272 | * overwrites from earlier objects rather than let | |
2273 | * tracking information or the free pointer be | |
0211a9c8 | 2274 | * corrupted if a user writes before the start |
81819f0f CL |
2275 | * of the object. |
2276 | */ | |
2277 | size += sizeof(void *); | |
41ecc55b | 2278 | #endif |
672bba3a | 2279 | |
81819f0f CL |
2280 | /* |
2281 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2282 | * user specified and the dynamic determination of cache line size |
2283 | * on bootup. | |
81819f0f CL |
2284 | */ |
2285 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2286 | s->align = align; |
81819f0f CL |
2287 | |
2288 | /* | |
2289 | * SLUB stores one object immediately after another beginning from | |
2290 | * offset 0. In order to align the objects we have to simply size | |
2291 | * each object to conform to the alignment. | |
2292 | */ | |
2293 | size = ALIGN(size, align); | |
2294 | s->size = size; | |
06b285dc CL |
2295 | if (forced_order >= 0) |
2296 | order = forced_order; | |
2297 | else | |
2298 | order = calculate_order(size); | |
81819f0f | 2299 | |
834f3d11 | 2300 | if (order < 0) |
81819f0f CL |
2301 | return 0; |
2302 | ||
b7a49f0d | 2303 | s->allocflags = 0; |
834f3d11 | 2304 | if (order) |
b7a49f0d CL |
2305 | s->allocflags |= __GFP_COMP; |
2306 | ||
2307 | if (s->flags & SLAB_CACHE_DMA) | |
2308 | s->allocflags |= SLUB_DMA; | |
2309 | ||
2310 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2311 | s->allocflags |= __GFP_RECLAIMABLE; | |
2312 | ||
81819f0f CL |
2313 | /* |
2314 | * Determine the number of objects per slab | |
2315 | */ | |
834f3d11 | 2316 | s->oo = oo_make(order, size); |
65c3376a | 2317 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2318 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2319 | s->max = s->oo; | |
81819f0f | 2320 | |
834f3d11 | 2321 | return !!oo_objects(s->oo); |
81819f0f CL |
2322 | |
2323 | } | |
2324 | ||
81819f0f CL |
2325 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2326 | const char *name, size_t size, | |
2327 | size_t align, unsigned long flags, | |
51cc5068 | 2328 | void (*ctor)(void *)) |
81819f0f CL |
2329 | { |
2330 | memset(s, 0, kmem_size); | |
2331 | s->name = name; | |
2332 | s->ctor = ctor; | |
81819f0f | 2333 | s->objsize = size; |
81819f0f | 2334 | s->align = align; |
ba0268a8 | 2335 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2336 | |
06b285dc | 2337 | if (!calculate_sizes(s, -1)) |
81819f0f | 2338 | goto error; |
3de47213 DR |
2339 | if (disable_higher_order_debug) { |
2340 | /* | |
2341 | * Disable debugging flags that store metadata if the min slab | |
2342 | * order increased. | |
2343 | */ | |
2344 | if (get_order(s->size) > get_order(s->objsize)) { | |
2345 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2346 | s->offset = 0; | |
2347 | if (!calculate_sizes(s, -1)) | |
2348 | goto error; | |
2349 | } | |
2350 | } | |
81819f0f | 2351 | |
3b89d7d8 DR |
2352 | /* |
2353 | * The larger the object size is, the more pages we want on the partial | |
2354 | * list to avoid pounding the page allocator excessively. | |
2355 | */ | |
c0bdb232 | 2356 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2357 | s->refcount = 1; |
2358 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2359 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2360 | #endif |
dfb4f096 CL |
2361 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2362 | goto error; | |
81819f0f | 2363 | |
dfb4f096 | 2364 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2365 | return 1; |
ff12059e | 2366 | |
4c93c355 | 2367 | free_kmem_cache_nodes(s); |
81819f0f CL |
2368 | error: |
2369 | if (flags & SLAB_PANIC) | |
2370 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2371 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2372 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2373 | s->offset, flags); |
2374 | return 0; | |
2375 | } | |
81819f0f CL |
2376 | |
2377 | /* | |
2378 | * Check if a given pointer is valid | |
2379 | */ | |
2380 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2381 | { | |
06428780 | 2382 | struct page *page; |
81819f0f | 2383 | |
d3e06e2b PE |
2384 | if (!kern_ptr_validate(object, s->size)) |
2385 | return 0; | |
2386 | ||
81819f0f CL |
2387 | page = get_object_page(object); |
2388 | ||
2389 | if (!page || s != page->slab) | |
2390 | /* No slab or wrong slab */ | |
2391 | return 0; | |
2392 | ||
abcd08a6 | 2393 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2394 | return 0; |
2395 | ||
2396 | /* | |
2397 | * We could also check if the object is on the slabs freelist. | |
2398 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2399 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2400 | * to a certain slab. |
2401 | */ | |
2402 | return 1; | |
2403 | } | |
2404 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2405 | ||
2406 | /* | |
2407 | * Determine the size of a slab object | |
2408 | */ | |
2409 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2410 | { | |
2411 | return s->objsize; | |
2412 | } | |
2413 | EXPORT_SYMBOL(kmem_cache_size); | |
2414 | ||
2415 | const char *kmem_cache_name(struct kmem_cache *s) | |
2416 | { | |
2417 | return s->name; | |
2418 | } | |
2419 | EXPORT_SYMBOL(kmem_cache_name); | |
2420 | ||
33b12c38 CL |
2421 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2422 | const char *text) | |
2423 | { | |
2424 | #ifdef CONFIG_SLUB_DEBUG | |
2425 | void *addr = page_address(page); | |
2426 | void *p; | |
bbd7d57b ED |
2427 | long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long), |
2428 | GFP_ATOMIC); | |
33b12c38 | 2429 | |
bbd7d57b ED |
2430 | if (!map) |
2431 | return; | |
33b12c38 CL |
2432 | slab_err(s, page, "%s", text); |
2433 | slab_lock(page); | |
2434 | for_each_free_object(p, s, page->freelist) | |
2435 | set_bit(slab_index(p, s, addr), map); | |
2436 | ||
2437 | for_each_object(p, s, addr, page->objects) { | |
2438 | ||
2439 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2440 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2441 | p, p - addr); | |
2442 | print_tracking(s, p); | |
2443 | } | |
2444 | } | |
2445 | slab_unlock(page); | |
bbd7d57b | 2446 | kfree(map); |
33b12c38 CL |
2447 | #endif |
2448 | } | |
2449 | ||
81819f0f | 2450 | /* |
599870b1 | 2451 | * Attempt to free all partial slabs on a node. |
81819f0f | 2452 | */ |
599870b1 | 2453 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2454 | { |
81819f0f CL |
2455 | unsigned long flags; |
2456 | struct page *page, *h; | |
2457 | ||
2458 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2459 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2460 | if (!page->inuse) { |
2461 | list_del(&page->lru); | |
2462 | discard_slab(s, page); | |
599870b1 | 2463 | n->nr_partial--; |
33b12c38 CL |
2464 | } else { |
2465 | list_slab_objects(s, page, | |
2466 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2467 | } |
33b12c38 | 2468 | } |
81819f0f | 2469 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2470 | } |
2471 | ||
2472 | /* | |
672bba3a | 2473 | * Release all resources used by a slab cache. |
81819f0f | 2474 | */ |
0c710013 | 2475 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2476 | { |
2477 | int node; | |
2478 | ||
2479 | flush_all(s); | |
9dfc6e68 | 2480 | free_percpu(s->cpu_slab); |
81819f0f | 2481 | /* Attempt to free all objects */ |
f64dc58c | 2482 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2483 | struct kmem_cache_node *n = get_node(s, node); |
2484 | ||
599870b1 CL |
2485 | free_partial(s, n); |
2486 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2487 | return 1; |
2488 | } | |
2489 | free_kmem_cache_nodes(s); | |
2490 | return 0; | |
2491 | } | |
2492 | ||
2493 | /* | |
2494 | * Close a cache and release the kmem_cache structure | |
2495 | * (must be used for caches created using kmem_cache_create) | |
2496 | */ | |
2497 | void kmem_cache_destroy(struct kmem_cache *s) | |
2498 | { | |
2499 | down_write(&slub_lock); | |
2500 | s->refcount--; | |
2501 | if (!s->refcount) { | |
2502 | list_del(&s->list); | |
a0e1d1be | 2503 | up_write(&slub_lock); |
d629d819 PE |
2504 | if (kmem_cache_close(s)) { |
2505 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2506 | "still has objects.\n", s->name, __func__); | |
2507 | dump_stack(); | |
2508 | } | |
d76b1590 ED |
2509 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2510 | rcu_barrier(); | |
81819f0f | 2511 | sysfs_slab_remove(s); |
a0e1d1be CL |
2512 | } else |
2513 | up_write(&slub_lock); | |
81819f0f CL |
2514 | } |
2515 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2516 | ||
2517 | /******************************************************************** | |
2518 | * Kmalloc subsystem | |
2519 | *******************************************************************/ | |
2520 | ||
756dee75 | 2521 | struct kmem_cache kmalloc_caches[KMALLOC_CACHES] __cacheline_aligned; |
81819f0f CL |
2522 | EXPORT_SYMBOL(kmalloc_caches); |
2523 | ||
81819f0f CL |
2524 | static int __init setup_slub_min_order(char *str) |
2525 | { | |
06428780 | 2526 | get_option(&str, &slub_min_order); |
81819f0f CL |
2527 | |
2528 | return 1; | |
2529 | } | |
2530 | ||
2531 | __setup("slub_min_order=", setup_slub_min_order); | |
2532 | ||
2533 | static int __init setup_slub_max_order(char *str) | |
2534 | { | |
06428780 | 2535 | get_option(&str, &slub_max_order); |
818cf590 | 2536 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2537 | |
2538 | return 1; | |
2539 | } | |
2540 | ||
2541 | __setup("slub_max_order=", setup_slub_max_order); | |
2542 | ||
2543 | static int __init setup_slub_min_objects(char *str) | |
2544 | { | |
06428780 | 2545 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2546 | |
2547 | return 1; | |
2548 | } | |
2549 | ||
2550 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2551 | ||
2552 | static int __init setup_slub_nomerge(char *str) | |
2553 | { | |
2554 | slub_nomerge = 1; | |
2555 | return 1; | |
2556 | } | |
2557 | ||
2558 | __setup("slub_nomerge", setup_slub_nomerge); | |
2559 | ||
81819f0f CL |
2560 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2561 | const char *name, int size, gfp_t gfp_flags) | |
2562 | { | |
2563 | unsigned int flags = 0; | |
2564 | ||
2565 | if (gfp_flags & SLUB_DMA) | |
2566 | flags = SLAB_CACHE_DMA; | |
2567 | ||
83b519e8 PE |
2568 | /* |
2569 | * This function is called with IRQs disabled during early-boot on | |
2570 | * single CPU so there's no need to take slub_lock here. | |
2571 | */ | |
81819f0f | 2572 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2573 | flags, NULL)) |
81819f0f CL |
2574 | goto panic; |
2575 | ||
2576 | list_add(&s->list, &slab_caches); | |
83b519e8 | 2577 | |
81819f0f CL |
2578 | if (sysfs_slab_add(s)) |
2579 | goto panic; | |
2580 | return s; | |
2581 | ||
2582 | panic: | |
2583 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2584 | } | |
2585 | ||
2e443fd0 | 2586 | #ifdef CONFIG_ZONE_DMA |
ffadd4d0 | 2587 | static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT]; |
1ceef402 CL |
2588 | |
2589 | static void sysfs_add_func(struct work_struct *w) | |
2590 | { | |
2591 | struct kmem_cache *s; | |
2592 | ||
2593 | down_write(&slub_lock); | |
2594 | list_for_each_entry(s, &slab_caches, list) { | |
2595 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2596 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2597 | sysfs_slab_add(s); | |
2598 | } | |
2599 | } | |
2600 | up_write(&slub_lock); | |
2601 | } | |
2602 | ||
2603 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2604 | ||
2e443fd0 CL |
2605 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2606 | { | |
2607 | struct kmem_cache *s; | |
2e443fd0 CL |
2608 | char *text; |
2609 | size_t realsize; | |
964cf35c | 2610 | unsigned long slabflags; |
756dee75 | 2611 | int i; |
2e443fd0 CL |
2612 | |
2613 | s = kmalloc_caches_dma[index]; | |
2614 | if (s) | |
2615 | return s; | |
2616 | ||
2617 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2618 | if (flags & __GFP_WAIT) |
2619 | down_write(&slub_lock); | |
2620 | else { | |
2621 | if (!down_write_trylock(&slub_lock)) | |
2622 | goto out; | |
2623 | } | |
2624 | ||
2625 | if (kmalloc_caches_dma[index]) | |
2626 | goto unlock_out; | |
2e443fd0 | 2627 | |
7b55f620 | 2628 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2629 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2630 | (unsigned int)realsize); | |
9dfc6e68 | 2631 | |
756dee75 CL |
2632 | s = NULL; |
2633 | for (i = 0; i < KMALLOC_CACHES; i++) | |
2634 | if (!kmalloc_caches[i].size) | |
2635 | break; | |
9dfc6e68 | 2636 | |
756dee75 CL |
2637 | BUG_ON(i >= KMALLOC_CACHES); |
2638 | s = kmalloc_caches + i; | |
1ceef402 | 2639 | |
964cf35c NP |
2640 | /* |
2641 | * Must defer sysfs creation to a workqueue because we don't know | |
2642 | * what context we are called from. Before sysfs comes up, we don't | |
2643 | * need to do anything because our sysfs initcall will start by | |
2644 | * adding all existing slabs to sysfs. | |
2645 | */ | |
5caf5c7d | 2646 | slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK; |
964cf35c NP |
2647 | if (slab_state >= SYSFS) |
2648 | slabflags |= __SYSFS_ADD_DEFERRED; | |
2649 | ||
7738dd9e | 2650 | if (!text || !kmem_cache_open(s, flags, text, |
964cf35c | 2651 | realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) { |
756dee75 | 2652 | s->size = 0; |
1ceef402 CL |
2653 | kfree(text); |
2654 | goto unlock_out; | |
dfce8648 | 2655 | } |
1ceef402 CL |
2656 | |
2657 | list_add(&s->list, &slab_caches); | |
2658 | kmalloc_caches_dma[index] = s; | |
2659 | ||
964cf35c NP |
2660 | if (slab_state >= SYSFS) |
2661 | schedule_work(&sysfs_add_work); | |
1ceef402 CL |
2662 | |
2663 | unlock_out: | |
dfce8648 | 2664 | up_write(&slub_lock); |
1ceef402 | 2665 | out: |
dfce8648 | 2666 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2667 | } |
2668 | #endif | |
2669 | ||
f1b26339 CL |
2670 | /* |
2671 | * Conversion table for small slabs sizes / 8 to the index in the | |
2672 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2673 | * of two cache sizes there. The size of larger slabs can be determined using | |
2674 | * fls. | |
2675 | */ | |
2676 | static s8 size_index[24] = { | |
2677 | 3, /* 8 */ | |
2678 | 4, /* 16 */ | |
2679 | 5, /* 24 */ | |
2680 | 5, /* 32 */ | |
2681 | 6, /* 40 */ | |
2682 | 6, /* 48 */ | |
2683 | 6, /* 56 */ | |
2684 | 6, /* 64 */ | |
2685 | 1, /* 72 */ | |
2686 | 1, /* 80 */ | |
2687 | 1, /* 88 */ | |
2688 | 1, /* 96 */ | |
2689 | 7, /* 104 */ | |
2690 | 7, /* 112 */ | |
2691 | 7, /* 120 */ | |
2692 | 7, /* 128 */ | |
2693 | 2, /* 136 */ | |
2694 | 2, /* 144 */ | |
2695 | 2, /* 152 */ | |
2696 | 2, /* 160 */ | |
2697 | 2, /* 168 */ | |
2698 | 2, /* 176 */ | |
2699 | 2, /* 184 */ | |
2700 | 2 /* 192 */ | |
2701 | }; | |
2702 | ||
acdfcd04 AK |
2703 | static inline int size_index_elem(size_t bytes) |
2704 | { | |
2705 | return (bytes - 1) / 8; | |
2706 | } | |
2707 | ||
81819f0f CL |
2708 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2709 | { | |
f1b26339 | 2710 | int index; |
81819f0f | 2711 | |
f1b26339 CL |
2712 | if (size <= 192) { |
2713 | if (!size) | |
2714 | return ZERO_SIZE_PTR; | |
81819f0f | 2715 | |
acdfcd04 | 2716 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2717 | } else |
f1b26339 | 2718 | index = fls(size - 1); |
81819f0f CL |
2719 | |
2720 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2721 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2722 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2723 | |
81819f0f CL |
2724 | #endif |
2725 | return &kmalloc_caches[index]; | |
2726 | } | |
2727 | ||
2728 | void *__kmalloc(size_t size, gfp_t flags) | |
2729 | { | |
aadb4bc4 | 2730 | struct kmem_cache *s; |
5b882be4 | 2731 | void *ret; |
81819f0f | 2732 | |
ffadd4d0 | 2733 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2734 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2735 | |
2736 | s = get_slab(size, flags); | |
2737 | ||
2738 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2739 | return s; |
2740 | ||
5b882be4 EGM |
2741 | ret = slab_alloc(s, flags, -1, _RET_IP_); |
2742 | ||
ca2b84cb | 2743 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2744 | |
2745 | return ret; | |
81819f0f CL |
2746 | } |
2747 | EXPORT_SYMBOL(__kmalloc); | |
2748 | ||
f619cfe1 CL |
2749 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2750 | { | |
b1eeab67 | 2751 | struct page *page; |
e4f7c0b4 | 2752 | void *ptr = NULL; |
f619cfe1 | 2753 | |
b1eeab67 VN |
2754 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2755 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2756 | if (page) |
e4f7c0b4 CM |
2757 | ptr = page_address(page); |
2758 | ||
2759 | kmemleak_alloc(ptr, size, 1, flags); | |
2760 | return ptr; | |
f619cfe1 CL |
2761 | } |
2762 | ||
81819f0f CL |
2763 | #ifdef CONFIG_NUMA |
2764 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2765 | { | |
aadb4bc4 | 2766 | struct kmem_cache *s; |
5b882be4 | 2767 | void *ret; |
81819f0f | 2768 | |
057685cf | 2769 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2770 | ret = kmalloc_large_node(size, flags, node); |
2771 | ||
ca2b84cb EGM |
2772 | trace_kmalloc_node(_RET_IP_, ret, |
2773 | size, PAGE_SIZE << get_order(size), | |
2774 | flags, node); | |
5b882be4 EGM |
2775 | |
2776 | return ret; | |
2777 | } | |
aadb4bc4 CL |
2778 | |
2779 | s = get_slab(size, flags); | |
2780 | ||
2781 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2782 | return s; |
2783 | ||
5b882be4 EGM |
2784 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2785 | ||
ca2b84cb | 2786 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2787 | |
2788 | return ret; | |
81819f0f CL |
2789 | } |
2790 | EXPORT_SYMBOL(__kmalloc_node); | |
2791 | #endif | |
2792 | ||
2793 | size_t ksize(const void *object) | |
2794 | { | |
272c1d21 | 2795 | struct page *page; |
81819f0f CL |
2796 | struct kmem_cache *s; |
2797 | ||
ef8b4520 | 2798 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2799 | return 0; |
2800 | ||
294a80a8 | 2801 | page = virt_to_head_page(object); |
294a80a8 | 2802 | |
76994412 PE |
2803 | if (unlikely(!PageSlab(page))) { |
2804 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2805 | return PAGE_SIZE << compound_order(page); |
76994412 | 2806 | } |
81819f0f | 2807 | s = page->slab; |
81819f0f | 2808 | |
ae20bfda | 2809 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2810 | /* |
2811 | * Debugging requires use of the padding between object | |
2812 | * and whatever may come after it. | |
2813 | */ | |
2814 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2815 | return s->objsize; | |
2816 | ||
ae20bfda | 2817 | #endif |
81819f0f CL |
2818 | /* |
2819 | * If we have the need to store the freelist pointer | |
2820 | * back there or track user information then we can | |
2821 | * only use the space before that information. | |
2822 | */ | |
2823 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2824 | return s->inuse; | |
81819f0f CL |
2825 | /* |
2826 | * Else we can use all the padding etc for the allocation | |
2827 | */ | |
2828 | return s->size; | |
2829 | } | |
b1aabecd | 2830 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2831 | |
2832 | void kfree(const void *x) | |
2833 | { | |
81819f0f | 2834 | struct page *page; |
5bb983b0 | 2835 | void *object = (void *)x; |
81819f0f | 2836 | |
2121db74 PE |
2837 | trace_kfree(_RET_IP_, x); |
2838 | ||
2408c550 | 2839 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2840 | return; |
2841 | ||
b49af68f | 2842 | page = virt_to_head_page(x); |
aadb4bc4 | 2843 | if (unlikely(!PageSlab(page))) { |
0937502a | 2844 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2845 | kmemleak_free(x); |
aadb4bc4 CL |
2846 | put_page(page); |
2847 | return; | |
2848 | } | |
ce71e27c | 2849 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2850 | } |
2851 | EXPORT_SYMBOL(kfree); | |
2852 | ||
2086d26a | 2853 | /* |
672bba3a CL |
2854 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2855 | * the remaining slabs by the number of items in use. The slabs with the | |
2856 | * most items in use come first. New allocations will then fill those up | |
2857 | * and thus they can be removed from the partial lists. | |
2858 | * | |
2859 | * The slabs with the least items are placed last. This results in them | |
2860 | * being allocated from last increasing the chance that the last objects | |
2861 | * are freed in them. | |
2086d26a CL |
2862 | */ |
2863 | int kmem_cache_shrink(struct kmem_cache *s) | |
2864 | { | |
2865 | int node; | |
2866 | int i; | |
2867 | struct kmem_cache_node *n; | |
2868 | struct page *page; | |
2869 | struct page *t; | |
205ab99d | 2870 | int objects = oo_objects(s->max); |
2086d26a | 2871 | struct list_head *slabs_by_inuse = |
834f3d11 | 2872 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2873 | unsigned long flags; |
2874 | ||
2875 | if (!slabs_by_inuse) | |
2876 | return -ENOMEM; | |
2877 | ||
2878 | flush_all(s); | |
f64dc58c | 2879 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2880 | n = get_node(s, node); |
2881 | ||
2882 | if (!n->nr_partial) | |
2883 | continue; | |
2884 | ||
834f3d11 | 2885 | for (i = 0; i < objects; i++) |
2086d26a CL |
2886 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2887 | ||
2888 | spin_lock_irqsave(&n->list_lock, flags); | |
2889 | ||
2890 | /* | |
672bba3a | 2891 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2892 | * |
672bba3a CL |
2893 | * Note that concurrent frees may occur while we hold the |
2894 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2895 | */ |
2896 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2897 | if (!page->inuse && slab_trylock(page)) { | |
2898 | /* | |
2899 | * Must hold slab lock here because slab_free | |
2900 | * may have freed the last object and be | |
2901 | * waiting to release the slab. | |
2902 | */ | |
2903 | list_del(&page->lru); | |
2904 | n->nr_partial--; | |
2905 | slab_unlock(page); | |
2906 | discard_slab(s, page); | |
2907 | } else { | |
fcda3d89 CL |
2908 | list_move(&page->lru, |
2909 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2910 | } |
2911 | } | |
2912 | ||
2086d26a | 2913 | /* |
672bba3a CL |
2914 | * Rebuild the partial list with the slabs filled up most |
2915 | * first and the least used slabs at the end. | |
2086d26a | 2916 | */ |
834f3d11 | 2917 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2918 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2919 | ||
2086d26a CL |
2920 | spin_unlock_irqrestore(&n->list_lock, flags); |
2921 | } | |
2922 | ||
2923 | kfree(slabs_by_inuse); | |
2924 | return 0; | |
2925 | } | |
2926 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2927 | ||
b9049e23 YG |
2928 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2929 | static int slab_mem_going_offline_callback(void *arg) | |
2930 | { | |
2931 | struct kmem_cache *s; | |
2932 | ||
2933 | down_read(&slub_lock); | |
2934 | list_for_each_entry(s, &slab_caches, list) | |
2935 | kmem_cache_shrink(s); | |
2936 | up_read(&slub_lock); | |
2937 | ||
2938 | return 0; | |
2939 | } | |
2940 | ||
2941 | static void slab_mem_offline_callback(void *arg) | |
2942 | { | |
2943 | struct kmem_cache_node *n; | |
2944 | struct kmem_cache *s; | |
2945 | struct memory_notify *marg = arg; | |
2946 | int offline_node; | |
2947 | ||
2948 | offline_node = marg->status_change_nid; | |
2949 | ||
2950 | /* | |
2951 | * If the node still has available memory. we need kmem_cache_node | |
2952 | * for it yet. | |
2953 | */ | |
2954 | if (offline_node < 0) | |
2955 | return; | |
2956 | ||
2957 | down_read(&slub_lock); | |
2958 | list_for_each_entry(s, &slab_caches, list) { | |
2959 | n = get_node(s, offline_node); | |
2960 | if (n) { | |
2961 | /* | |
2962 | * if n->nr_slabs > 0, slabs still exist on the node | |
2963 | * that is going down. We were unable to free them, | |
c9404c9c | 2964 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2965 | * callback. So, we must fail. |
2966 | */ | |
0f389ec6 | 2967 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2968 | |
2969 | s->node[offline_node] = NULL; | |
2970 | kmem_cache_free(kmalloc_caches, n); | |
2971 | } | |
2972 | } | |
2973 | up_read(&slub_lock); | |
2974 | } | |
2975 | ||
2976 | static int slab_mem_going_online_callback(void *arg) | |
2977 | { | |
2978 | struct kmem_cache_node *n; | |
2979 | struct kmem_cache *s; | |
2980 | struct memory_notify *marg = arg; | |
2981 | int nid = marg->status_change_nid; | |
2982 | int ret = 0; | |
2983 | ||
2984 | /* | |
2985 | * If the node's memory is already available, then kmem_cache_node is | |
2986 | * already created. Nothing to do. | |
2987 | */ | |
2988 | if (nid < 0) | |
2989 | return 0; | |
2990 | ||
2991 | /* | |
0121c619 | 2992 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2993 | * allocate a kmem_cache_node structure in order to bring the node |
2994 | * online. | |
2995 | */ | |
2996 | down_read(&slub_lock); | |
2997 | list_for_each_entry(s, &slab_caches, list) { | |
2998 | /* | |
2999 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3000 | * since memory is not yet available from the node that | |
3001 | * is brought up. | |
3002 | */ | |
3003 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
3004 | if (!n) { | |
3005 | ret = -ENOMEM; | |
3006 | goto out; | |
3007 | } | |
5595cffc | 3008 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3009 | s->node[nid] = n; |
3010 | } | |
3011 | out: | |
3012 | up_read(&slub_lock); | |
3013 | return ret; | |
3014 | } | |
3015 | ||
3016 | static int slab_memory_callback(struct notifier_block *self, | |
3017 | unsigned long action, void *arg) | |
3018 | { | |
3019 | int ret = 0; | |
3020 | ||
3021 | switch (action) { | |
3022 | case MEM_GOING_ONLINE: | |
3023 | ret = slab_mem_going_online_callback(arg); | |
3024 | break; | |
3025 | case MEM_GOING_OFFLINE: | |
3026 | ret = slab_mem_going_offline_callback(arg); | |
3027 | break; | |
3028 | case MEM_OFFLINE: | |
3029 | case MEM_CANCEL_ONLINE: | |
3030 | slab_mem_offline_callback(arg); | |
3031 | break; | |
3032 | case MEM_ONLINE: | |
3033 | case MEM_CANCEL_OFFLINE: | |
3034 | break; | |
3035 | } | |
dc19f9db KH |
3036 | if (ret) |
3037 | ret = notifier_from_errno(ret); | |
3038 | else | |
3039 | ret = NOTIFY_OK; | |
b9049e23 YG |
3040 | return ret; |
3041 | } | |
3042 | ||
3043 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3044 | ||
81819f0f CL |
3045 | /******************************************************************** |
3046 | * Basic setup of slabs | |
3047 | *******************************************************************/ | |
3048 | ||
3049 | void __init kmem_cache_init(void) | |
3050 | { | |
3051 | int i; | |
4b356be0 | 3052 | int caches = 0; |
81819f0f CL |
3053 | |
3054 | #ifdef CONFIG_NUMA | |
3055 | /* | |
3056 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3057 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3058 | * kmem_cache_open for slab_state == DOWN. |
3059 | */ | |
3060 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
83b519e8 | 3061 | sizeof(struct kmem_cache_node), GFP_NOWAIT); |
8ffa6875 | 3062 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3063 | caches++; |
b9049e23 | 3064 | |
0c40ba4f | 3065 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3066 | #endif |
3067 | ||
3068 | /* Able to allocate the per node structures */ | |
3069 | slab_state = PARTIAL; | |
3070 | ||
3071 | /* Caches that are not of the two-to-the-power-of size */ | |
acdfcd04 | 3072 | if (KMALLOC_MIN_SIZE <= 32) { |
4b356be0 | 3073 | create_kmalloc_cache(&kmalloc_caches[1], |
83b519e8 | 3074 | "kmalloc-96", 96, GFP_NOWAIT); |
4b356be0 | 3075 | caches++; |
acdfcd04 AK |
3076 | } |
3077 | if (KMALLOC_MIN_SIZE <= 64) { | |
4b356be0 | 3078 | create_kmalloc_cache(&kmalloc_caches[2], |
83b519e8 | 3079 | "kmalloc-192", 192, GFP_NOWAIT); |
4b356be0 CL |
3080 | caches++; |
3081 | } | |
81819f0f | 3082 | |
ffadd4d0 | 3083 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
81819f0f | 3084 | create_kmalloc_cache(&kmalloc_caches[i], |
83b519e8 | 3085 | "kmalloc", 1 << i, GFP_NOWAIT); |
4b356be0 CL |
3086 | caches++; |
3087 | } | |
81819f0f | 3088 | |
f1b26339 CL |
3089 | |
3090 | /* | |
3091 | * Patch up the size_index table if we have strange large alignment | |
3092 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3093 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3094 | * |
3095 | * Largest permitted alignment is 256 bytes due to the way we | |
3096 | * handle the index determination for the smaller caches. | |
3097 | * | |
3098 | * Make sure that nothing crazy happens if someone starts tinkering | |
3099 | * around with ARCH_KMALLOC_MINALIGN | |
3100 | */ | |
3101 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3102 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3103 | ||
acdfcd04 AK |
3104 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3105 | int elem = size_index_elem(i); | |
3106 | if (elem >= ARRAY_SIZE(size_index)) | |
3107 | break; | |
3108 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3109 | } | |
f1b26339 | 3110 | |
acdfcd04 AK |
3111 | if (KMALLOC_MIN_SIZE == 64) { |
3112 | /* | |
3113 | * The 96 byte size cache is not used if the alignment | |
3114 | * is 64 byte. | |
3115 | */ | |
3116 | for (i = 64 + 8; i <= 96; i += 8) | |
3117 | size_index[size_index_elem(i)] = 7; | |
3118 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3119 | /* |
3120 | * The 192 byte sized cache is not used if the alignment | |
3121 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3122 | * instead. | |
3123 | */ | |
3124 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3125 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3126 | } |
3127 | ||
81819f0f CL |
3128 | slab_state = UP; |
3129 | ||
3130 | /* Provide the correct kmalloc names now that the caches are up */ | |
ffadd4d0 | 3131 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) |
81819f0f | 3132 | kmalloc_caches[i]. name = |
83b519e8 | 3133 | kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); |
81819f0f CL |
3134 | |
3135 | #ifdef CONFIG_SMP | |
3136 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 CL |
3137 | #endif |
3138 | #ifdef CONFIG_NUMA | |
3139 | kmem_size = offsetof(struct kmem_cache, node) + | |
3140 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
4c93c355 CL |
3141 | #else |
3142 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3143 | #endif |
3144 | ||
3adbefee IM |
3145 | printk(KERN_INFO |
3146 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3147 | " CPUs=%d, Nodes=%d\n", |
3148 | caches, cache_line_size(), | |
81819f0f CL |
3149 | slub_min_order, slub_max_order, slub_min_objects, |
3150 | nr_cpu_ids, nr_node_ids); | |
3151 | } | |
3152 | ||
7e85ee0c PE |
3153 | void __init kmem_cache_init_late(void) |
3154 | { | |
7e85ee0c PE |
3155 | } |
3156 | ||
81819f0f CL |
3157 | /* |
3158 | * Find a mergeable slab cache | |
3159 | */ | |
3160 | static int slab_unmergeable(struct kmem_cache *s) | |
3161 | { | |
3162 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3163 | return 1; | |
3164 | ||
c59def9f | 3165 | if (s->ctor) |
81819f0f CL |
3166 | return 1; |
3167 | ||
8ffa6875 CL |
3168 | /* |
3169 | * We may have set a slab to be unmergeable during bootstrap. | |
3170 | */ | |
3171 | if (s->refcount < 0) | |
3172 | return 1; | |
3173 | ||
81819f0f CL |
3174 | return 0; |
3175 | } | |
3176 | ||
3177 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3178 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3179 | void (*ctor)(void *)) |
81819f0f | 3180 | { |
5b95a4ac | 3181 | struct kmem_cache *s; |
81819f0f CL |
3182 | |
3183 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3184 | return NULL; | |
3185 | ||
c59def9f | 3186 | if (ctor) |
81819f0f CL |
3187 | return NULL; |
3188 | ||
3189 | size = ALIGN(size, sizeof(void *)); | |
3190 | align = calculate_alignment(flags, align, size); | |
3191 | size = ALIGN(size, align); | |
ba0268a8 | 3192 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3193 | |
5b95a4ac | 3194 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3195 | if (slab_unmergeable(s)) |
3196 | continue; | |
3197 | ||
3198 | if (size > s->size) | |
3199 | continue; | |
3200 | ||
ba0268a8 | 3201 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3202 | continue; |
3203 | /* | |
3204 | * Check if alignment is compatible. | |
3205 | * Courtesy of Adrian Drzewiecki | |
3206 | */ | |
06428780 | 3207 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3208 | continue; |
3209 | ||
3210 | if (s->size - size >= sizeof(void *)) | |
3211 | continue; | |
3212 | ||
3213 | return s; | |
3214 | } | |
3215 | return NULL; | |
3216 | } | |
3217 | ||
3218 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3219 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3220 | { |
3221 | struct kmem_cache *s; | |
3222 | ||
fe1ff49d BH |
3223 | if (WARN_ON(!name)) |
3224 | return NULL; | |
3225 | ||
81819f0f | 3226 | down_write(&slub_lock); |
ba0268a8 | 3227 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3228 | if (s) { |
3229 | s->refcount++; | |
3230 | /* | |
3231 | * Adjust the object sizes so that we clear | |
3232 | * the complete object on kzalloc. | |
3233 | */ | |
3234 | s->objsize = max(s->objsize, (int)size); | |
3235 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
a0e1d1be | 3236 | up_write(&slub_lock); |
6446faa2 | 3237 | |
7b8f3b66 DR |
3238 | if (sysfs_slab_alias(s, name)) { |
3239 | down_write(&slub_lock); | |
3240 | s->refcount--; | |
3241 | up_write(&slub_lock); | |
81819f0f | 3242 | goto err; |
7b8f3b66 | 3243 | } |
a0e1d1be CL |
3244 | return s; |
3245 | } | |
6446faa2 | 3246 | |
a0e1d1be CL |
3247 | s = kmalloc(kmem_size, GFP_KERNEL); |
3248 | if (s) { | |
3249 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3250 | size, align, flags, ctor)) { |
81819f0f | 3251 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3252 | up_write(&slub_lock); |
7b8f3b66 DR |
3253 | if (sysfs_slab_add(s)) { |
3254 | down_write(&slub_lock); | |
3255 | list_del(&s->list); | |
3256 | up_write(&slub_lock); | |
3257 | kfree(s); | |
a0e1d1be | 3258 | goto err; |
7b8f3b66 | 3259 | } |
a0e1d1be CL |
3260 | return s; |
3261 | } | |
3262 | kfree(s); | |
81819f0f CL |
3263 | } |
3264 | up_write(&slub_lock); | |
81819f0f CL |
3265 | |
3266 | err: | |
81819f0f CL |
3267 | if (flags & SLAB_PANIC) |
3268 | panic("Cannot create slabcache %s\n", name); | |
3269 | else | |
3270 | s = NULL; | |
3271 | return s; | |
3272 | } | |
3273 | EXPORT_SYMBOL(kmem_cache_create); | |
3274 | ||
81819f0f | 3275 | #ifdef CONFIG_SMP |
81819f0f | 3276 | /* |
672bba3a CL |
3277 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3278 | * necessary. | |
81819f0f CL |
3279 | */ |
3280 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3281 | unsigned long action, void *hcpu) | |
3282 | { | |
3283 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3284 | struct kmem_cache *s; |
3285 | unsigned long flags; | |
81819f0f CL |
3286 | |
3287 | switch (action) { | |
3288 | case CPU_UP_CANCELED: | |
8bb78442 | 3289 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3290 | case CPU_DEAD: |
8bb78442 | 3291 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3292 | down_read(&slub_lock); |
3293 | list_for_each_entry(s, &slab_caches, list) { | |
3294 | local_irq_save(flags); | |
3295 | __flush_cpu_slab(s, cpu); | |
3296 | local_irq_restore(flags); | |
3297 | } | |
3298 | up_read(&slub_lock); | |
81819f0f CL |
3299 | break; |
3300 | default: | |
3301 | break; | |
3302 | } | |
3303 | return NOTIFY_OK; | |
3304 | } | |
3305 | ||
06428780 | 3306 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3307 | .notifier_call = slab_cpuup_callback |
06428780 | 3308 | }; |
81819f0f CL |
3309 | |
3310 | #endif | |
3311 | ||
ce71e27c | 3312 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3313 | { |
aadb4bc4 | 3314 | struct kmem_cache *s; |
94b528d0 | 3315 | void *ret; |
aadb4bc4 | 3316 | |
ffadd4d0 | 3317 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3318 | return kmalloc_large(size, gfpflags); |
3319 | ||
aadb4bc4 | 3320 | s = get_slab(size, gfpflags); |
81819f0f | 3321 | |
2408c550 | 3322 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3323 | return s; |
81819f0f | 3324 | |
94b528d0 EGM |
3325 | ret = slab_alloc(s, gfpflags, -1, caller); |
3326 | ||
3327 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3328 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3329 | |
3330 | return ret; | |
81819f0f CL |
3331 | } |
3332 | ||
3333 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3334 | int node, unsigned long caller) |
81819f0f | 3335 | { |
aadb4bc4 | 3336 | struct kmem_cache *s; |
94b528d0 | 3337 | void *ret; |
aadb4bc4 | 3338 | |
d3e14aa3 XF |
3339 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3340 | ret = kmalloc_large_node(size, gfpflags, node); | |
3341 | ||
3342 | trace_kmalloc_node(caller, ret, | |
3343 | size, PAGE_SIZE << get_order(size), | |
3344 | gfpflags, node); | |
3345 | ||
3346 | return ret; | |
3347 | } | |
eada35ef | 3348 | |
aadb4bc4 | 3349 | s = get_slab(size, gfpflags); |
81819f0f | 3350 | |
2408c550 | 3351 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3352 | return s; |
81819f0f | 3353 | |
94b528d0 EGM |
3354 | ret = slab_alloc(s, gfpflags, node, caller); |
3355 | ||
3356 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3357 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3358 | |
3359 | return ret; | |
81819f0f CL |
3360 | } |
3361 | ||
f6acb635 | 3362 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3363 | static int count_inuse(struct page *page) |
3364 | { | |
3365 | return page->inuse; | |
3366 | } | |
3367 | ||
3368 | static int count_total(struct page *page) | |
3369 | { | |
3370 | return page->objects; | |
3371 | } | |
3372 | ||
434e245d CL |
3373 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3374 | unsigned long *map) | |
53e15af0 CL |
3375 | { |
3376 | void *p; | |
a973e9dd | 3377 | void *addr = page_address(page); |
53e15af0 CL |
3378 | |
3379 | if (!check_slab(s, page) || | |
3380 | !on_freelist(s, page, NULL)) | |
3381 | return 0; | |
3382 | ||
3383 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3384 | bitmap_zero(map, page->objects); |
53e15af0 | 3385 | |
7656c72b CL |
3386 | for_each_free_object(p, s, page->freelist) { |
3387 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3388 | if (!check_object(s, page, p, 0)) |
3389 | return 0; | |
3390 | } | |
3391 | ||
224a88be | 3392 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3393 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3394 | if (!check_object(s, page, p, 1)) |
3395 | return 0; | |
3396 | return 1; | |
3397 | } | |
3398 | ||
434e245d CL |
3399 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3400 | unsigned long *map) | |
53e15af0 CL |
3401 | { |
3402 | if (slab_trylock(page)) { | |
434e245d | 3403 | validate_slab(s, page, map); |
53e15af0 CL |
3404 | slab_unlock(page); |
3405 | } else | |
3406 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3407 | s->name, page); | |
3408 | ||
3409 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3410 | if (!PageSlubDebug(page)) |
3411 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3412 | "on slab 0x%p\n", s->name, page); |
3413 | } else { | |
8a38082d AW |
3414 | if (PageSlubDebug(page)) |
3415 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3416 | "slab 0x%p\n", s->name, page); |
3417 | } | |
3418 | } | |
3419 | ||
434e245d CL |
3420 | static int validate_slab_node(struct kmem_cache *s, |
3421 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3422 | { |
3423 | unsigned long count = 0; | |
3424 | struct page *page; | |
3425 | unsigned long flags; | |
3426 | ||
3427 | spin_lock_irqsave(&n->list_lock, flags); | |
3428 | ||
3429 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3430 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3431 | count++; |
3432 | } | |
3433 | if (count != n->nr_partial) | |
3434 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3435 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3436 | ||
3437 | if (!(s->flags & SLAB_STORE_USER)) | |
3438 | goto out; | |
3439 | ||
3440 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3441 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3442 | count++; |
3443 | } | |
3444 | if (count != atomic_long_read(&n->nr_slabs)) | |
3445 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3446 | "counter=%ld\n", s->name, count, | |
3447 | atomic_long_read(&n->nr_slabs)); | |
3448 | ||
3449 | out: | |
3450 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3451 | return count; | |
3452 | } | |
3453 | ||
434e245d | 3454 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3455 | { |
3456 | int node; | |
3457 | unsigned long count = 0; | |
205ab99d | 3458 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3459 | sizeof(unsigned long), GFP_KERNEL); |
3460 | ||
3461 | if (!map) | |
3462 | return -ENOMEM; | |
53e15af0 CL |
3463 | |
3464 | flush_all(s); | |
f64dc58c | 3465 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3466 | struct kmem_cache_node *n = get_node(s, node); |
3467 | ||
434e245d | 3468 | count += validate_slab_node(s, n, map); |
53e15af0 | 3469 | } |
434e245d | 3470 | kfree(map); |
53e15af0 CL |
3471 | return count; |
3472 | } | |
3473 | ||
b3459709 CL |
3474 | #ifdef SLUB_RESILIENCY_TEST |
3475 | static void resiliency_test(void) | |
3476 | { | |
3477 | u8 *p; | |
3478 | ||
3479 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3480 | printk(KERN_ERR "-----------------------\n"); | |
3481 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3482 | ||
3483 | p = kzalloc(16, GFP_KERNEL); | |
3484 | p[16] = 0x12; | |
3485 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3486 | " 0x12->0x%p\n\n", p + 16); | |
3487 | ||
3488 | validate_slab_cache(kmalloc_caches + 4); | |
3489 | ||
3490 | /* Hmmm... The next two are dangerous */ | |
3491 | p = kzalloc(32, GFP_KERNEL); | |
3492 | p[32 + sizeof(void *)] = 0x34; | |
3493 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3494 | " 0x34 -> -0x%p\n", p); |
3495 | printk(KERN_ERR | |
3496 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3497 | |
3498 | validate_slab_cache(kmalloc_caches + 5); | |
3499 | p = kzalloc(64, GFP_KERNEL); | |
3500 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3501 | *p = 0x56; | |
3502 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3503 | p); | |
3adbefee IM |
3504 | printk(KERN_ERR |
3505 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3506 | validate_slab_cache(kmalloc_caches + 6); |
3507 | ||
3508 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3509 | p = kzalloc(128, GFP_KERNEL); | |
3510 | kfree(p); | |
3511 | *p = 0x78; | |
3512 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3513 | validate_slab_cache(kmalloc_caches + 7); | |
3514 | ||
3515 | p = kzalloc(256, GFP_KERNEL); | |
3516 | kfree(p); | |
3517 | p[50] = 0x9a; | |
3adbefee IM |
3518 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3519 | p); | |
b3459709 CL |
3520 | validate_slab_cache(kmalloc_caches + 8); |
3521 | ||
3522 | p = kzalloc(512, GFP_KERNEL); | |
3523 | kfree(p); | |
3524 | p[512] = 0xab; | |
3525 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3526 | validate_slab_cache(kmalloc_caches + 9); | |
3527 | } | |
3528 | #else | |
3529 | static void resiliency_test(void) {}; | |
3530 | #endif | |
3531 | ||
88a420e4 | 3532 | /* |
672bba3a | 3533 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3534 | * and freed. |
3535 | */ | |
3536 | ||
3537 | struct location { | |
3538 | unsigned long count; | |
ce71e27c | 3539 | unsigned long addr; |
45edfa58 CL |
3540 | long long sum_time; |
3541 | long min_time; | |
3542 | long max_time; | |
3543 | long min_pid; | |
3544 | long max_pid; | |
174596a0 | 3545 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3546 | nodemask_t nodes; |
88a420e4 CL |
3547 | }; |
3548 | ||
3549 | struct loc_track { | |
3550 | unsigned long max; | |
3551 | unsigned long count; | |
3552 | struct location *loc; | |
3553 | }; | |
3554 | ||
3555 | static void free_loc_track(struct loc_track *t) | |
3556 | { | |
3557 | if (t->max) | |
3558 | free_pages((unsigned long)t->loc, | |
3559 | get_order(sizeof(struct location) * t->max)); | |
3560 | } | |
3561 | ||
68dff6a9 | 3562 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3563 | { |
3564 | struct location *l; | |
3565 | int order; | |
3566 | ||
88a420e4 CL |
3567 | order = get_order(sizeof(struct location) * max); |
3568 | ||
68dff6a9 | 3569 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3570 | if (!l) |
3571 | return 0; | |
3572 | ||
3573 | if (t->count) { | |
3574 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3575 | free_loc_track(t); | |
3576 | } | |
3577 | t->max = max; | |
3578 | t->loc = l; | |
3579 | return 1; | |
3580 | } | |
3581 | ||
3582 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3583 | const struct track *track) |
88a420e4 CL |
3584 | { |
3585 | long start, end, pos; | |
3586 | struct location *l; | |
ce71e27c | 3587 | unsigned long caddr; |
45edfa58 | 3588 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3589 | |
3590 | start = -1; | |
3591 | end = t->count; | |
3592 | ||
3593 | for ( ; ; ) { | |
3594 | pos = start + (end - start + 1) / 2; | |
3595 | ||
3596 | /* | |
3597 | * There is nothing at "end". If we end up there | |
3598 | * we need to add something to before end. | |
3599 | */ | |
3600 | if (pos == end) | |
3601 | break; | |
3602 | ||
3603 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3604 | if (track->addr == caddr) { |
3605 | ||
3606 | l = &t->loc[pos]; | |
3607 | l->count++; | |
3608 | if (track->when) { | |
3609 | l->sum_time += age; | |
3610 | if (age < l->min_time) | |
3611 | l->min_time = age; | |
3612 | if (age > l->max_time) | |
3613 | l->max_time = age; | |
3614 | ||
3615 | if (track->pid < l->min_pid) | |
3616 | l->min_pid = track->pid; | |
3617 | if (track->pid > l->max_pid) | |
3618 | l->max_pid = track->pid; | |
3619 | ||
174596a0 RR |
3620 | cpumask_set_cpu(track->cpu, |
3621 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3622 | } |
3623 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3624 | return 1; |
3625 | } | |
3626 | ||
45edfa58 | 3627 | if (track->addr < caddr) |
88a420e4 CL |
3628 | end = pos; |
3629 | else | |
3630 | start = pos; | |
3631 | } | |
3632 | ||
3633 | /* | |
672bba3a | 3634 | * Not found. Insert new tracking element. |
88a420e4 | 3635 | */ |
68dff6a9 | 3636 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3637 | return 0; |
3638 | ||
3639 | l = t->loc + pos; | |
3640 | if (pos < t->count) | |
3641 | memmove(l + 1, l, | |
3642 | (t->count - pos) * sizeof(struct location)); | |
3643 | t->count++; | |
3644 | l->count = 1; | |
45edfa58 CL |
3645 | l->addr = track->addr; |
3646 | l->sum_time = age; | |
3647 | l->min_time = age; | |
3648 | l->max_time = age; | |
3649 | l->min_pid = track->pid; | |
3650 | l->max_pid = track->pid; | |
174596a0 RR |
3651 | cpumask_clear(to_cpumask(l->cpus)); |
3652 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3653 | nodes_clear(l->nodes); |
3654 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3655 | return 1; |
3656 | } | |
3657 | ||
3658 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b ED |
3659 | struct page *page, enum track_item alloc, |
3660 | long *map) | |
88a420e4 | 3661 | { |
a973e9dd | 3662 | void *addr = page_address(page); |
88a420e4 CL |
3663 | void *p; |
3664 | ||
39b26464 | 3665 | bitmap_zero(map, page->objects); |
7656c72b CL |
3666 | for_each_free_object(p, s, page->freelist) |
3667 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3668 | |
224a88be | 3669 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3670 | if (!test_bit(slab_index(p, s, addr), map)) |
3671 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3672 | } |
3673 | ||
3674 | static int list_locations(struct kmem_cache *s, char *buf, | |
3675 | enum track_item alloc) | |
3676 | { | |
e374d483 | 3677 | int len = 0; |
88a420e4 | 3678 | unsigned long i; |
68dff6a9 | 3679 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3680 | int node; |
bbd7d57b ED |
3681 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3682 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3683 | |
bbd7d57b ED |
3684 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3685 | GFP_TEMPORARY)) { | |
3686 | kfree(map); | |
68dff6a9 | 3687 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3688 | } |
88a420e4 CL |
3689 | /* Push back cpu slabs */ |
3690 | flush_all(s); | |
3691 | ||
f64dc58c | 3692 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3693 | struct kmem_cache_node *n = get_node(s, node); |
3694 | unsigned long flags; | |
3695 | struct page *page; | |
3696 | ||
9e86943b | 3697 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3698 | continue; |
3699 | ||
3700 | spin_lock_irqsave(&n->list_lock, flags); | |
3701 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3702 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3703 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3704 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3705 | spin_unlock_irqrestore(&n->list_lock, flags); |
3706 | } | |
3707 | ||
3708 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3709 | struct location *l = &t.loc[i]; |
88a420e4 | 3710 | |
9c246247 | 3711 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3712 | break; |
e374d483 | 3713 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3714 | |
3715 | if (l->addr) | |
e374d483 | 3716 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3717 | else |
e374d483 | 3718 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3719 | |
3720 | if (l->sum_time != l->min_time) { | |
e374d483 | 3721 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3722 | l->min_time, |
3723 | (long)div_u64(l->sum_time, l->count), | |
3724 | l->max_time); | |
45edfa58 | 3725 | } else |
e374d483 | 3726 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3727 | l->min_time); |
3728 | ||
3729 | if (l->min_pid != l->max_pid) | |
e374d483 | 3730 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3731 | l->min_pid, l->max_pid); |
3732 | else | |
e374d483 | 3733 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3734 | l->min_pid); |
3735 | ||
174596a0 RR |
3736 | if (num_online_cpus() > 1 && |
3737 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3738 | len < PAGE_SIZE - 60) { |
3739 | len += sprintf(buf + len, " cpus="); | |
3740 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3741 | to_cpumask(l->cpus)); |
45edfa58 CL |
3742 | } |
3743 | ||
62bc62a8 | 3744 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3745 | len < PAGE_SIZE - 60) { |
3746 | len += sprintf(buf + len, " nodes="); | |
3747 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3748 | l->nodes); |
3749 | } | |
3750 | ||
e374d483 | 3751 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3752 | } |
3753 | ||
3754 | free_loc_track(&t); | |
bbd7d57b | 3755 | kfree(map); |
88a420e4 | 3756 | if (!t.count) |
e374d483 HH |
3757 | len += sprintf(buf, "No data\n"); |
3758 | return len; | |
88a420e4 CL |
3759 | } |
3760 | ||
81819f0f | 3761 | enum slab_stat_type { |
205ab99d CL |
3762 | SL_ALL, /* All slabs */ |
3763 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3764 | SL_CPU, /* Only slabs used for cpu caches */ | |
3765 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3766 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3767 | }; |
3768 | ||
205ab99d | 3769 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3770 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3771 | #define SO_CPU (1 << SL_CPU) | |
3772 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3773 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3774 | |
62e5c4b4 CG |
3775 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3776 | char *buf, unsigned long flags) | |
81819f0f CL |
3777 | { |
3778 | unsigned long total = 0; | |
81819f0f CL |
3779 | int node; |
3780 | int x; | |
3781 | unsigned long *nodes; | |
3782 | unsigned long *per_cpu; | |
3783 | ||
3784 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3785 | if (!nodes) |
3786 | return -ENOMEM; | |
81819f0f CL |
3787 | per_cpu = nodes + nr_node_ids; |
3788 | ||
205ab99d CL |
3789 | if (flags & SO_CPU) { |
3790 | int cpu; | |
81819f0f | 3791 | |
205ab99d | 3792 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3793 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3794 | |
205ab99d CL |
3795 | if (!c || c->node < 0) |
3796 | continue; | |
3797 | ||
3798 | if (c->page) { | |
3799 | if (flags & SO_TOTAL) | |
3800 | x = c->page->objects; | |
3801 | else if (flags & SO_OBJECTS) | |
3802 | x = c->page->inuse; | |
81819f0f CL |
3803 | else |
3804 | x = 1; | |
205ab99d | 3805 | |
81819f0f | 3806 | total += x; |
205ab99d | 3807 | nodes[c->node] += x; |
81819f0f | 3808 | } |
205ab99d | 3809 | per_cpu[c->node]++; |
81819f0f CL |
3810 | } |
3811 | } | |
3812 | ||
205ab99d CL |
3813 | if (flags & SO_ALL) { |
3814 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3815 | struct kmem_cache_node *n = get_node(s, node); | |
3816 | ||
3817 | if (flags & SO_TOTAL) | |
3818 | x = atomic_long_read(&n->total_objects); | |
3819 | else if (flags & SO_OBJECTS) | |
3820 | x = atomic_long_read(&n->total_objects) - | |
3821 | count_partial(n, count_free); | |
81819f0f | 3822 | |
81819f0f | 3823 | else |
205ab99d | 3824 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3825 | total += x; |
3826 | nodes[node] += x; | |
3827 | } | |
3828 | ||
205ab99d CL |
3829 | } else if (flags & SO_PARTIAL) { |
3830 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3831 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3832 | |
205ab99d CL |
3833 | if (flags & SO_TOTAL) |
3834 | x = count_partial(n, count_total); | |
3835 | else if (flags & SO_OBJECTS) | |
3836 | x = count_partial(n, count_inuse); | |
81819f0f | 3837 | else |
205ab99d | 3838 | x = n->nr_partial; |
81819f0f CL |
3839 | total += x; |
3840 | nodes[node] += x; | |
3841 | } | |
3842 | } | |
81819f0f CL |
3843 | x = sprintf(buf, "%lu", total); |
3844 | #ifdef CONFIG_NUMA | |
f64dc58c | 3845 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3846 | if (nodes[node]) |
3847 | x += sprintf(buf + x, " N%d=%lu", | |
3848 | node, nodes[node]); | |
3849 | #endif | |
3850 | kfree(nodes); | |
3851 | return x + sprintf(buf + x, "\n"); | |
3852 | } | |
3853 | ||
3854 | static int any_slab_objects(struct kmem_cache *s) | |
3855 | { | |
3856 | int node; | |
81819f0f | 3857 | |
dfb4f096 | 3858 | for_each_online_node(node) { |
81819f0f CL |
3859 | struct kmem_cache_node *n = get_node(s, node); |
3860 | ||
dfb4f096 CL |
3861 | if (!n) |
3862 | continue; | |
3863 | ||
4ea33e2d | 3864 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3865 | return 1; |
3866 | } | |
3867 | return 0; | |
3868 | } | |
3869 | ||
3870 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3871 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3872 | ||
3873 | struct slab_attribute { | |
3874 | struct attribute attr; | |
3875 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3876 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3877 | }; | |
3878 | ||
3879 | #define SLAB_ATTR_RO(_name) \ | |
3880 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3881 | ||
3882 | #define SLAB_ATTR(_name) \ | |
3883 | static struct slab_attribute _name##_attr = \ | |
3884 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3885 | ||
81819f0f CL |
3886 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3887 | { | |
3888 | return sprintf(buf, "%d\n", s->size); | |
3889 | } | |
3890 | SLAB_ATTR_RO(slab_size); | |
3891 | ||
3892 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3893 | { | |
3894 | return sprintf(buf, "%d\n", s->align); | |
3895 | } | |
3896 | SLAB_ATTR_RO(align); | |
3897 | ||
3898 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3899 | { | |
3900 | return sprintf(buf, "%d\n", s->objsize); | |
3901 | } | |
3902 | SLAB_ATTR_RO(object_size); | |
3903 | ||
3904 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3905 | { | |
834f3d11 | 3906 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3907 | } |
3908 | SLAB_ATTR_RO(objs_per_slab); | |
3909 | ||
06b285dc CL |
3910 | static ssize_t order_store(struct kmem_cache *s, |
3911 | const char *buf, size_t length) | |
3912 | { | |
0121c619 CL |
3913 | unsigned long order; |
3914 | int err; | |
3915 | ||
3916 | err = strict_strtoul(buf, 10, &order); | |
3917 | if (err) | |
3918 | return err; | |
06b285dc CL |
3919 | |
3920 | if (order > slub_max_order || order < slub_min_order) | |
3921 | return -EINVAL; | |
3922 | ||
3923 | calculate_sizes(s, order); | |
3924 | return length; | |
3925 | } | |
3926 | ||
81819f0f CL |
3927 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3928 | { | |
834f3d11 | 3929 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3930 | } |
06b285dc | 3931 | SLAB_ATTR(order); |
81819f0f | 3932 | |
73d342b1 DR |
3933 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3934 | { | |
3935 | return sprintf(buf, "%lu\n", s->min_partial); | |
3936 | } | |
3937 | ||
3938 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3939 | size_t length) | |
3940 | { | |
3941 | unsigned long min; | |
3942 | int err; | |
3943 | ||
3944 | err = strict_strtoul(buf, 10, &min); | |
3945 | if (err) | |
3946 | return err; | |
3947 | ||
c0bdb232 | 3948 | set_min_partial(s, min); |
73d342b1 DR |
3949 | return length; |
3950 | } | |
3951 | SLAB_ATTR(min_partial); | |
3952 | ||
81819f0f CL |
3953 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3954 | { | |
3955 | if (s->ctor) { | |
3956 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3957 | ||
3958 | return n + sprintf(buf + n, "\n"); | |
3959 | } | |
3960 | return 0; | |
3961 | } | |
3962 | SLAB_ATTR_RO(ctor); | |
3963 | ||
81819f0f CL |
3964 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3965 | { | |
3966 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3967 | } | |
3968 | SLAB_ATTR_RO(aliases); | |
3969 | ||
3970 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3971 | { | |
205ab99d | 3972 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3973 | } |
3974 | SLAB_ATTR_RO(slabs); | |
3975 | ||
3976 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3977 | { | |
d9acf4b7 | 3978 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3979 | } |
3980 | SLAB_ATTR_RO(partial); | |
3981 | ||
3982 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3983 | { | |
d9acf4b7 | 3984 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3985 | } |
3986 | SLAB_ATTR_RO(cpu_slabs); | |
3987 | ||
3988 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3989 | { | |
205ab99d | 3990 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3991 | } |
3992 | SLAB_ATTR_RO(objects); | |
3993 | ||
205ab99d CL |
3994 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3995 | { | |
3996 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3997 | } | |
3998 | SLAB_ATTR_RO(objects_partial); | |
3999 | ||
4000 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4001 | { | |
4002 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4003 | } | |
4004 | SLAB_ATTR_RO(total_objects); | |
4005 | ||
81819f0f CL |
4006 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4007 | { | |
4008 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4009 | } | |
4010 | ||
4011 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4012 | const char *buf, size_t length) | |
4013 | { | |
4014 | s->flags &= ~SLAB_DEBUG_FREE; | |
4015 | if (buf[0] == '1') | |
4016 | s->flags |= SLAB_DEBUG_FREE; | |
4017 | return length; | |
4018 | } | |
4019 | SLAB_ATTR(sanity_checks); | |
4020 | ||
4021 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4022 | { | |
4023 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4024 | } | |
4025 | ||
4026 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4027 | size_t length) | |
4028 | { | |
4029 | s->flags &= ~SLAB_TRACE; | |
4030 | if (buf[0] == '1') | |
4031 | s->flags |= SLAB_TRACE; | |
4032 | return length; | |
4033 | } | |
4034 | SLAB_ATTR(trace); | |
4035 | ||
4c13dd3b DM |
4036 | #ifdef CONFIG_FAILSLAB |
4037 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4038 | { | |
4039 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4040 | } | |
4041 | ||
4042 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4043 | size_t length) | |
4044 | { | |
4045 | s->flags &= ~SLAB_FAILSLAB; | |
4046 | if (buf[0] == '1') | |
4047 | s->flags |= SLAB_FAILSLAB; | |
4048 | return length; | |
4049 | } | |
4050 | SLAB_ATTR(failslab); | |
4051 | #endif | |
4052 | ||
81819f0f CL |
4053 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4054 | { | |
4055 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4056 | } | |
4057 | ||
4058 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4059 | const char *buf, size_t length) | |
4060 | { | |
4061 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4062 | if (buf[0] == '1') | |
4063 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4064 | return length; | |
4065 | } | |
4066 | SLAB_ATTR(reclaim_account); | |
4067 | ||
4068 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4069 | { | |
5af60839 | 4070 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4071 | } |
4072 | SLAB_ATTR_RO(hwcache_align); | |
4073 | ||
4074 | #ifdef CONFIG_ZONE_DMA | |
4075 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4076 | { | |
4077 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4078 | } | |
4079 | SLAB_ATTR_RO(cache_dma); | |
4080 | #endif | |
4081 | ||
4082 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4083 | { | |
4084 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4085 | } | |
4086 | SLAB_ATTR_RO(destroy_by_rcu); | |
4087 | ||
4088 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4089 | { | |
4090 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4091 | } | |
4092 | ||
4093 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4094 | const char *buf, size_t length) | |
4095 | { | |
4096 | if (any_slab_objects(s)) | |
4097 | return -EBUSY; | |
4098 | ||
4099 | s->flags &= ~SLAB_RED_ZONE; | |
4100 | if (buf[0] == '1') | |
4101 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4102 | calculate_sizes(s, -1); |
81819f0f CL |
4103 | return length; |
4104 | } | |
4105 | SLAB_ATTR(red_zone); | |
4106 | ||
4107 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4108 | { | |
4109 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4110 | } | |
4111 | ||
4112 | static ssize_t poison_store(struct kmem_cache *s, | |
4113 | const char *buf, size_t length) | |
4114 | { | |
4115 | if (any_slab_objects(s)) | |
4116 | return -EBUSY; | |
4117 | ||
4118 | s->flags &= ~SLAB_POISON; | |
4119 | if (buf[0] == '1') | |
4120 | s->flags |= SLAB_POISON; | |
06b285dc | 4121 | calculate_sizes(s, -1); |
81819f0f CL |
4122 | return length; |
4123 | } | |
4124 | SLAB_ATTR(poison); | |
4125 | ||
4126 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4127 | { | |
4128 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4129 | } | |
4130 | ||
4131 | static ssize_t store_user_store(struct kmem_cache *s, | |
4132 | const char *buf, size_t length) | |
4133 | { | |
4134 | if (any_slab_objects(s)) | |
4135 | return -EBUSY; | |
4136 | ||
4137 | s->flags &= ~SLAB_STORE_USER; | |
4138 | if (buf[0] == '1') | |
4139 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4140 | calculate_sizes(s, -1); |
81819f0f CL |
4141 | return length; |
4142 | } | |
4143 | SLAB_ATTR(store_user); | |
4144 | ||
53e15af0 CL |
4145 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4146 | { | |
4147 | return 0; | |
4148 | } | |
4149 | ||
4150 | static ssize_t validate_store(struct kmem_cache *s, | |
4151 | const char *buf, size_t length) | |
4152 | { | |
434e245d CL |
4153 | int ret = -EINVAL; |
4154 | ||
4155 | if (buf[0] == '1') { | |
4156 | ret = validate_slab_cache(s); | |
4157 | if (ret >= 0) | |
4158 | ret = length; | |
4159 | } | |
4160 | return ret; | |
53e15af0 CL |
4161 | } |
4162 | SLAB_ATTR(validate); | |
4163 | ||
2086d26a CL |
4164 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4165 | { | |
4166 | return 0; | |
4167 | } | |
4168 | ||
4169 | static ssize_t shrink_store(struct kmem_cache *s, | |
4170 | const char *buf, size_t length) | |
4171 | { | |
4172 | if (buf[0] == '1') { | |
4173 | int rc = kmem_cache_shrink(s); | |
4174 | ||
4175 | if (rc) | |
4176 | return rc; | |
4177 | } else | |
4178 | return -EINVAL; | |
4179 | return length; | |
4180 | } | |
4181 | SLAB_ATTR(shrink); | |
4182 | ||
88a420e4 CL |
4183 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4184 | { | |
4185 | if (!(s->flags & SLAB_STORE_USER)) | |
4186 | return -ENOSYS; | |
4187 | return list_locations(s, buf, TRACK_ALLOC); | |
4188 | } | |
4189 | SLAB_ATTR_RO(alloc_calls); | |
4190 | ||
4191 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4192 | { | |
4193 | if (!(s->flags & SLAB_STORE_USER)) | |
4194 | return -ENOSYS; | |
4195 | return list_locations(s, buf, TRACK_FREE); | |
4196 | } | |
4197 | SLAB_ATTR_RO(free_calls); | |
4198 | ||
81819f0f | 4199 | #ifdef CONFIG_NUMA |
9824601e | 4200 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4201 | { |
9824601e | 4202 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4203 | } |
4204 | ||
9824601e | 4205 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4206 | const char *buf, size_t length) |
4207 | { | |
0121c619 CL |
4208 | unsigned long ratio; |
4209 | int err; | |
4210 | ||
4211 | err = strict_strtoul(buf, 10, &ratio); | |
4212 | if (err) | |
4213 | return err; | |
4214 | ||
e2cb96b7 | 4215 | if (ratio <= 100) |
0121c619 | 4216 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4217 | |
81819f0f CL |
4218 | return length; |
4219 | } | |
9824601e | 4220 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4221 | #endif |
4222 | ||
8ff12cfc | 4223 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4224 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4225 | { | |
4226 | unsigned long sum = 0; | |
4227 | int cpu; | |
4228 | int len; | |
4229 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4230 | ||
4231 | if (!data) | |
4232 | return -ENOMEM; | |
4233 | ||
4234 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4235 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4236 | |
4237 | data[cpu] = x; | |
4238 | sum += x; | |
4239 | } | |
4240 | ||
4241 | len = sprintf(buf, "%lu", sum); | |
4242 | ||
50ef37b9 | 4243 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4244 | for_each_online_cpu(cpu) { |
4245 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4246 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4247 | } |
50ef37b9 | 4248 | #endif |
8ff12cfc CL |
4249 | kfree(data); |
4250 | return len + sprintf(buf + len, "\n"); | |
4251 | } | |
4252 | ||
78eb00cc DR |
4253 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4254 | { | |
4255 | int cpu; | |
4256 | ||
4257 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4258 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4259 | } |
4260 | ||
8ff12cfc CL |
4261 | #define STAT_ATTR(si, text) \ |
4262 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4263 | { \ | |
4264 | return show_stat(s, buf, si); \ | |
4265 | } \ | |
78eb00cc DR |
4266 | static ssize_t text##_store(struct kmem_cache *s, \ |
4267 | const char *buf, size_t length) \ | |
4268 | { \ | |
4269 | if (buf[0] != '0') \ | |
4270 | return -EINVAL; \ | |
4271 | clear_stat(s, si); \ | |
4272 | return length; \ | |
4273 | } \ | |
4274 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4275 | |
4276 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4277 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4278 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4279 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4280 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4281 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4282 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4283 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4284 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4285 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4286 | STAT_ATTR(FREE_SLAB, free_slab); | |
4287 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4288 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4289 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4290 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4291 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4292 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4293 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4294 | #endif |
4295 | ||
06428780 | 4296 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4297 | &slab_size_attr.attr, |
4298 | &object_size_attr.attr, | |
4299 | &objs_per_slab_attr.attr, | |
4300 | &order_attr.attr, | |
73d342b1 | 4301 | &min_partial_attr.attr, |
81819f0f | 4302 | &objects_attr.attr, |
205ab99d CL |
4303 | &objects_partial_attr.attr, |
4304 | &total_objects_attr.attr, | |
81819f0f CL |
4305 | &slabs_attr.attr, |
4306 | &partial_attr.attr, | |
4307 | &cpu_slabs_attr.attr, | |
4308 | &ctor_attr.attr, | |
81819f0f CL |
4309 | &aliases_attr.attr, |
4310 | &align_attr.attr, | |
4311 | &sanity_checks_attr.attr, | |
4312 | &trace_attr.attr, | |
4313 | &hwcache_align_attr.attr, | |
4314 | &reclaim_account_attr.attr, | |
4315 | &destroy_by_rcu_attr.attr, | |
4316 | &red_zone_attr.attr, | |
4317 | &poison_attr.attr, | |
4318 | &store_user_attr.attr, | |
53e15af0 | 4319 | &validate_attr.attr, |
2086d26a | 4320 | &shrink_attr.attr, |
88a420e4 CL |
4321 | &alloc_calls_attr.attr, |
4322 | &free_calls_attr.attr, | |
81819f0f CL |
4323 | #ifdef CONFIG_ZONE_DMA |
4324 | &cache_dma_attr.attr, | |
4325 | #endif | |
4326 | #ifdef CONFIG_NUMA | |
9824601e | 4327 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4328 | #endif |
4329 | #ifdef CONFIG_SLUB_STATS | |
4330 | &alloc_fastpath_attr.attr, | |
4331 | &alloc_slowpath_attr.attr, | |
4332 | &free_fastpath_attr.attr, | |
4333 | &free_slowpath_attr.attr, | |
4334 | &free_frozen_attr.attr, | |
4335 | &free_add_partial_attr.attr, | |
4336 | &free_remove_partial_attr.attr, | |
4337 | &alloc_from_partial_attr.attr, | |
4338 | &alloc_slab_attr.attr, | |
4339 | &alloc_refill_attr.attr, | |
4340 | &free_slab_attr.attr, | |
4341 | &cpuslab_flush_attr.attr, | |
4342 | &deactivate_full_attr.attr, | |
4343 | &deactivate_empty_attr.attr, | |
4344 | &deactivate_to_head_attr.attr, | |
4345 | &deactivate_to_tail_attr.attr, | |
4346 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4347 | &order_fallback_attr.attr, |
81819f0f | 4348 | #endif |
4c13dd3b DM |
4349 | #ifdef CONFIG_FAILSLAB |
4350 | &failslab_attr.attr, | |
4351 | #endif | |
4352 | ||
81819f0f CL |
4353 | NULL |
4354 | }; | |
4355 | ||
4356 | static struct attribute_group slab_attr_group = { | |
4357 | .attrs = slab_attrs, | |
4358 | }; | |
4359 | ||
4360 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4361 | struct attribute *attr, | |
4362 | char *buf) | |
4363 | { | |
4364 | struct slab_attribute *attribute; | |
4365 | struct kmem_cache *s; | |
4366 | int err; | |
4367 | ||
4368 | attribute = to_slab_attr(attr); | |
4369 | s = to_slab(kobj); | |
4370 | ||
4371 | if (!attribute->show) | |
4372 | return -EIO; | |
4373 | ||
4374 | err = attribute->show(s, buf); | |
4375 | ||
4376 | return err; | |
4377 | } | |
4378 | ||
4379 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4380 | struct attribute *attr, | |
4381 | const char *buf, size_t len) | |
4382 | { | |
4383 | struct slab_attribute *attribute; | |
4384 | struct kmem_cache *s; | |
4385 | int err; | |
4386 | ||
4387 | attribute = to_slab_attr(attr); | |
4388 | s = to_slab(kobj); | |
4389 | ||
4390 | if (!attribute->store) | |
4391 | return -EIO; | |
4392 | ||
4393 | err = attribute->store(s, buf, len); | |
4394 | ||
4395 | return err; | |
4396 | } | |
4397 | ||
151c602f CL |
4398 | static void kmem_cache_release(struct kobject *kobj) |
4399 | { | |
4400 | struct kmem_cache *s = to_slab(kobj); | |
4401 | ||
4402 | kfree(s); | |
4403 | } | |
4404 | ||
52cf25d0 | 4405 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4406 | .show = slab_attr_show, |
4407 | .store = slab_attr_store, | |
4408 | }; | |
4409 | ||
4410 | static struct kobj_type slab_ktype = { | |
4411 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4412 | .release = kmem_cache_release |
81819f0f CL |
4413 | }; |
4414 | ||
4415 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4416 | { | |
4417 | struct kobj_type *ktype = get_ktype(kobj); | |
4418 | ||
4419 | if (ktype == &slab_ktype) | |
4420 | return 1; | |
4421 | return 0; | |
4422 | } | |
4423 | ||
9cd43611 | 4424 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4425 | .filter = uevent_filter, |
4426 | }; | |
4427 | ||
27c3a314 | 4428 | static struct kset *slab_kset; |
81819f0f CL |
4429 | |
4430 | #define ID_STR_LENGTH 64 | |
4431 | ||
4432 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4433 | * |
4434 | * Format :[flags-]size | |
81819f0f CL |
4435 | */ |
4436 | static char *create_unique_id(struct kmem_cache *s) | |
4437 | { | |
4438 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4439 | char *p = name; | |
4440 | ||
4441 | BUG_ON(!name); | |
4442 | ||
4443 | *p++ = ':'; | |
4444 | /* | |
4445 | * First flags affecting slabcache operations. We will only | |
4446 | * get here for aliasable slabs so we do not need to support | |
4447 | * too many flags. The flags here must cover all flags that | |
4448 | * are matched during merging to guarantee that the id is | |
4449 | * unique. | |
4450 | */ | |
4451 | if (s->flags & SLAB_CACHE_DMA) | |
4452 | *p++ = 'd'; | |
4453 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4454 | *p++ = 'a'; | |
4455 | if (s->flags & SLAB_DEBUG_FREE) | |
4456 | *p++ = 'F'; | |
5a896d9e VN |
4457 | if (!(s->flags & SLAB_NOTRACK)) |
4458 | *p++ = 't'; | |
81819f0f CL |
4459 | if (p != name + 1) |
4460 | *p++ = '-'; | |
4461 | p += sprintf(p, "%07d", s->size); | |
4462 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4463 | return name; | |
4464 | } | |
4465 | ||
4466 | static int sysfs_slab_add(struct kmem_cache *s) | |
4467 | { | |
4468 | int err; | |
4469 | const char *name; | |
4470 | int unmergeable; | |
4471 | ||
4472 | if (slab_state < SYSFS) | |
4473 | /* Defer until later */ | |
4474 | return 0; | |
4475 | ||
4476 | unmergeable = slab_unmergeable(s); | |
4477 | if (unmergeable) { | |
4478 | /* | |
4479 | * Slabcache can never be merged so we can use the name proper. | |
4480 | * This is typically the case for debug situations. In that | |
4481 | * case we can catch duplicate names easily. | |
4482 | */ | |
27c3a314 | 4483 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4484 | name = s->name; |
4485 | } else { | |
4486 | /* | |
4487 | * Create a unique name for the slab as a target | |
4488 | * for the symlinks. | |
4489 | */ | |
4490 | name = create_unique_id(s); | |
4491 | } | |
4492 | ||
27c3a314 | 4493 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4494 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4495 | if (err) { | |
4496 | kobject_put(&s->kobj); | |
81819f0f | 4497 | return err; |
1eada11c | 4498 | } |
81819f0f CL |
4499 | |
4500 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4501 | if (err) { |
4502 | kobject_del(&s->kobj); | |
4503 | kobject_put(&s->kobj); | |
81819f0f | 4504 | return err; |
5788d8ad | 4505 | } |
81819f0f CL |
4506 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4507 | if (!unmergeable) { | |
4508 | /* Setup first alias */ | |
4509 | sysfs_slab_alias(s, s->name); | |
4510 | kfree(name); | |
4511 | } | |
4512 | return 0; | |
4513 | } | |
4514 | ||
4515 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4516 | { | |
4517 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4518 | kobject_del(&s->kobj); | |
151c602f | 4519 | kobject_put(&s->kobj); |
81819f0f CL |
4520 | } |
4521 | ||
4522 | /* | |
4523 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4524 | * available lest we lose that information. |
81819f0f CL |
4525 | */ |
4526 | struct saved_alias { | |
4527 | struct kmem_cache *s; | |
4528 | const char *name; | |
4529 | struct saved_alias *next; | |
4530 | }; | |
4531 | ||
5af328a5 | 4532 | static struct saved_alias *alias_list; |
81819f0f CL |
4533 | |
4534 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4535 | { | |
4536 | struct saved_alias *al; | |
4537 | ||
4538 | if (slab_state == SYSFS) { | |
4539 | /* | |
4540 | * If we have a leftover link then remove it. | |
4541 | */ | |
27c3a314 GKH |
4542 | sysfs_remove_link(&slab_kset->kobj, name); |
4543 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4544 | } |
4545 | ||
4546 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4547 | if (!al) | |
4548 | return -ENOMEM; | |
4549 | ||
4550 | al->s = s; | |
4551 | al->name = name; | |
4552 | al->next = alias_list; | |
4553 | alias_list = al; | |
4554 | return 0; | |
4555 | } | |
4556 | ||
4557 | static int __init slab_sysfs_init(void) | |
4558 | { | |
5b95a4ac | 4559 | struct kmem_cache *s; |
81819f0f CL |
4560 | int err; |
4561 | ||
0ff21e46 | 4562 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4563 | if (!slab_kset) { |
81819f0f CL |
4564 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4565 | return -ENOSYS; | |
4566 | } | |
4567 | ||
26a7bd03 CL |
4568 | slab_state = SYSFS; |
4569 | ||
5b95a4ac | 4570 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4571 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4572 | if (err) |
4573 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4574 | " to sysfs\n", s->name); | |
26a7bd03 | 4575 | } |
81819f0f CL |
4576 | |
4577 | while (alias_list) { | |
4578 | struct saved_alias *al = alias_list; | |
4579 | ||
4580 | alias_list = alias_list->next; | |
4581 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4582 | if (err) |
4583 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4584 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4585 | kfree(al); |
4586 | } | |
4587 | ||
4588 | resiliency_test(); | |
4589 | return 0; | |
4590 | } | |
4591 | ||
4592 | __initcall(slab_sysfs_init); | |
81819f0f | 4593 | #endif |
57ed3eda PE |
4594 | |
4595 | /* | |
4596 | * The /proc/slabinfo ABI | |
4597 | */ | |
158a9624 | 4598 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4599 | static void print_slabinfo_header(struct seq_file *m) |
4600 | { | |
4601 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4602 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4603 | "<objperslab> <pagesperslab>"); | |
4604 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4605 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4606 | seq_putc(m, '\n'); | |
4607 | } | |
4608 | ||
4609 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4610 | { | |
4611 | loff_t n = *pos; | |
4612 | ||
4613 | down_read(&slub_lock); | |
4614 | if (!n) | |
4615 | print_slabinfo_header(m); | |
4616 | ||
4617 | return seq_list_start(&slab_caches, *pos); | |
4618 | } | |
4619 | ||
4620 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4621 | { | |
4622 | return seq_list_next(p, &slab_caches, pos); | |
4623 | } | |
4624 | ||
4625 | static void s_stop(struct seq_file *m, void *p) | |
4626 | { | |
4627 | up_read(&slub_lock); | |
4628 | } | |
4629 | ||
4630 | static int s_show(struct seq_file *m, void *p) | |
4631 | { | |
4632 | unsigned long nr_partials = 0; | |
4633 | unsigned long nr_slabs = 0; | |
4634 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4635 | unsigned long nr_objs = 0; |
4636 | unsigned long nr_free = 0; | |
57ed3eda PE |
4637 | struct kmem_cache *s; |
4638 | int node; | |
4639 | ||
4640 | s = list_entry(p, struct kmem_cache, list); | |
4641 | ||
4642 | for_each_online_node(node) { | |
4643 | struct kmem_cache_node *n = get_node(s, node); | |
4644 | ||
4645 | if (!n) | |
4646 | continue; | |
4647 | ||
4648 | nr_partials += n->nr_partial; | |
4649 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4650 | nr_objs += atomic_long_read(&n->total_objects); |
4651 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4652 | } |
4653 | ||
205ab99d | 4654 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4655 | |
4656 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4657 | nr_objs, s->size, oo_objects(s->oo), |
4658 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4659 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4660 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4661 | 0UL); | |
4662 | seq_putc(m, '\n'); | |
4663 | return 0; | |
4664 | } | |
4665 | ||
7b3c3a50 | 4666 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4667 | .start = s_start, |
4668 | .next = s_next, | |
4669 | .stop = s_stop, | |
4670 | .show = s_show, | |
4671 | }; | |
4672 | ||
7b3c3a50 AD |
4673 | static int slabinfo_open(struct inode *inode, struct file *file) |
4674 | { | |
4675 | return seq_open(file, &slabinfo_op); | |
4676 | } | |
4677 | ||
4678 | static const struct file_operations proc_slabinfo_operations = { | |
4679 | .open = slabinfo_open, | |
4680 | .read = seq_read, | |
4681 | .llseek = seq_lseek, | |
4682 | .release = seq_release, | |
4683 | }; | |
4684 | ||
4685 | static int __init slab_proc_init(void) | |
4686 | { | |
cf5d1131 | 4687 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4688 | return 0; |
4689 | } | |
4690 | module_init(slab_proc_init); | |
158a9624 | 4691 | #endif /* CONFIG_SLABINFO */ |