]>
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> |
02af61bb | 20 | #include <linux/kmemtrace.h> |
5a896d9e | 21 | #include <linux/kmemcheck.h> |
81819f0f CL |
22 | #include <linux/cpu.h> |
23 | #include <linux/cpuset.h> | |
24 | #include <linux/mempolicy.h> | |
25 | #include <linux/ctype.h> | |
3ac7fe5a | 26 | #include <linux/debugobjects.h> |
81819f0f | 27 | #include <linux/kallsyms.h> |
b9049e23 | 28 | #include <linux/memory.h> |
f8bd2258 | 29 | #include <linux/math64.h> |
773ff60e | 30 | #include <linux/fault-inject.h> |
81819f0f CL |
31 | |
32 | /* | |
33 | * Lock order: | |
34 | * 1. slab_lock(page) | |
35 | * 2. slab->list_lock | |
36 | * | |
37 | * The slab_lock protects operations on the object of a particular | |
38 | * slab and its metadata in the page struct. If the slab lock | |
39 | * has been taken then no allocations nor frees can be performed | |
40 | * on the objects in the slab nor can the slab be added or removed | |
41 | * from the partial or full lists since this would mean modifying | |
42 | * the page_struct of the slab. | |
43 | * | |
44 | * The list_lock protects the partial and full list on each node and | |
45 | * the partial slab counter. If taken then no new slabs may be added or | |
46 | * removed from the lists nor make the number of partial slabs be modified. | |
47 | * (Note that the total number of slabs is an atomic value that may be | |
48 | * modified without taking the list lock). | |
49 | * | |
50 | * The list_lock is a centralized lock and thus we avoid taking it as | |
51 | * much as possible. As long as SLUB does not have to handle partial | |
52 | * slabs, operations can continue without any centralized lock. F.e. | |
53 | * allocating a long series of objects that fill up slabs does not require | |
54 | * the list lock. | |
55 | * | |
56 | * The lock order is sometimes inverted when we are trying to get a slab | |
57 | * off a list. We take the list_lock and then look for a page on the list | |
58 | * to use. While we do that objects in the slabs may be freed. We can | |
59 | * only operate on the slab if we have also taken the slab_lock. So we use | |
60 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
61 | * can occur anymore and we can use the slab for allocations etc. If the | |
62 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
63 | * we must stay away from it for a while since we may cause a bouncing | |
64 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
65 | * If all pages are busy then we may allocate a new slab instead of reusing | |
66 | * a partial slab. A new slab has noone operating on it and thus there is | |
67 | * no danger of cacheline contention. | |
68 | * | |
69 | * Interrupts are disabled during allocation and deallocation in order to | |
70 | * make the slab allocator safe to use in the context of an irq. In addition | |
71 | * interrupts are disabled to ensure that the processor does not change | |
72 | * while handling per_cpu slabs, due to kernel preemption. | |
73 | * | |
74 | * SLUB assigns one slab for allocation to each processor. | |
75 | * Allocations only occur from these slabs called cpu slabs. | |
76 | * | |
672bba3a CL |
77 | * Slabs with free elements are kept on a partial list and during regular |
78 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 79 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
80 | * We track full slabs for debugging purposes though because otherwise we |
81 | * cannot scan all objects. | |
81819f0f CL |
82 | * |
83 | * Slabs are freed when they become empty. Teardown and setup is | |
84 | * minimal so we rely on the page allocators per cpu caches for | |
85 | * fast frees and allocs. | |
86 | * | |
87 | * Overloading of page flags that are otherwise used for LRU management. | |
88 | * | |
4b6f0750 CL |
89 | * PageActive The slab is frozen and exempt from list processing. |
90 | * This means that the slab is dedicated to a purpose | |
91 | * such as satisfying allocations for a specific | |
92 | * processor. Objects may be freed in the slab while | |
93 | * it is frozen but slab_free will then skip the usual | |
94 | * list operations. It is up to the processor holding | |
95 | * the slab to integrate the slab into the slab lists | |
96 | * when the slab is no longer needed. | |
97 | * | |
98 | * One use of this flag is to mark slabs that are | |
99 | * used for allocations. Then such a slab becomes a cpu | |
100 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 101 | * freelist that allows lockless access to |
894b8788 CL |
102 | * free objects in addition to the regular freelist |
103 | * that requires the slab lock. | |
81819f0f CL |
104 | * |
105 | * PageError Slab requires special handling due to debug | |
106 | * options set. This moves slab handling out of | |
894b8788 | 107 | * the fast path and disables lockless freelists. |
81819f0f CL |
108 | */ |
109 | ||
5577bd8a | 110 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 111 | #define SLABDEBUG 1 |
5577bd8a CL |
112 | #else |
113 | #define SLABDEBUG 0 | |
114 | #endif | |
115 | ||
81819f0f CL |
116 | /* |
117 | * Issues still to be resolved: | |
118 | * | |
81819f0f CL |
119 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
120 | * | |
81819f0f CL |
121 | * - Variable sizing of the per node arrays |
122 | */ | |
123 | ||
124 | /* Enable to test recovery from slab corruption on boot */ | |
125 | #undef SLUB_RESILIENCY_TEST | |
126 | ||
2086d26a CL |
127 | /* |
128 | * Mininum number of partial slabs. These will be left on the partial | |
129 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
130 | */ | |
76be8950 | 131 | #define MIN_PARTIAL 5 |
e95eed57 | 132 | |
2086d26a CL |
133 | /* |
134 | * Maximum number of desirable partial slabs. | |
135 | * The existence of more partial slabs makes kmem_cache_shrink | |
136 | * sort the partial list by the number of objects in the. | |
137 | */ | |
138 | #define MAX_PARTIAL 10 | |
139 | ||
81819f0f CL |
140 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
141 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 142 | |
fa5ec8a1 | 143 | /* |
3de47213 DR |
144 | * Debugging flags that require metadata to be stored in the slab. These get |
145 | * disabled when slub_debug=O is used and a cache's min order increases with | |
146 | * metadata. | |
fa5ec8a1 | 147 | */ |
3de47213 | 148 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 149 | |
81819f0f CL |
150 | /* |
151 | * Set of flags that will prevent slab merging | |
152 | */ | |
153 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
06f22f13 | 154 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE) |
81819f0f CL |
155 | |
156 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 157 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f CL |
158 | |
159 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 160 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
161 | #endif |
162 | ||
163 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 164 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
165 | #endif |
166 | ||
210b5c06 CG |
167 | #define OO_SHIFT 16 |
168 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
169 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
170 | ||
81819f0f | 171 | /* Internal SLUB flags */ |
1ceef402 CL |
172 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
173 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
174 | |
175 | static int kmem_size = sizeof(struct kmem_cache); | |
176 | ||
177 | #ifdef CONFIG_SMP | |
178 | static struct notifier_block slab_notifier; | |
179 | #endif | |
180 | ||
181 | static enum { | |
182 | DOWN, /* No slab functionality available */ | |
183 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 184 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
185 | SYSFS /* Sysfs up */ |
186 | } slab_state = DOWN; | |
187 | ||
188 | /* A list of all slab caches on the system */ | |
189 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 190 | static LIST_HEAD(slab_caches); |
81819f0f | 191 | |
02cbc874 CL |
192 | /* |
193 | * Tracking user of a slab. | |
194 | */ | |
195 | struct track { | |
ce71e27c | 196 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
197 | int cpu; /* Was running on cpu */ |
198 | int pid; /* Pid context */ | |
199 | unsigned long when; /* When did the operation occur */ | |
200 | }; | |
201 | ||
202 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
203 | ||
f6acb635 | 204 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
205 | static int sysfs_slab_add(struct kmem_cache *); |
206 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
207 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 208 | |
81819f0f | 209 | #else |
0c710013 CL |
210 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
211 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
212 | { return 0; } | |
151c602f CL |
213 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
214 | { | |
215 | kfree(s); | |
216 | } | |
8ff12cfc | 217 | |
81819f0f CL |
218 | #endif |
219 | ||
8ff12cfc CL |
220 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
221 | { | |
222 | #ifdef CONFIG_SLUB_STATS | |
223 | c->stat[si]++; | |
224 | #endif | |
225 | } | |
226 | ||
81819f0f CL |
227 | /******************************************************************** |
228 | * Core slab cache functions | |
229 | *******************************************************************/ | |
230 | ||
231 | int slab_is_available(void) | |
232 | { | |
233 | return slab_state >= UP; | |
234 | } | |
235 | ||
236 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
237 | { | |
238 | #ifdef CONFIG_NUMA | |
239 | return s->node[node]; | |
240 | #else | |
241 | return &s->local_node; | |
242 | #endif | |
243 | } | |
244 | ||
dfb4f096 CL |
245 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
246 | { | |
4c93c355 CL |
247 | #ifdef CONFIG_SMP |
248 | return s->cpu_slab[cpu]; | |
249 | #else | |
250 | return &s->cpu_slab; | |
251 | #endif | |
dfb4f096 CL |
252 | } |
253 | ||
6446faa2 | 254 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
255 | static inline int check_valid_pointer(struct kmem_cache *s, |
256 | struct page *page, const void *object) | |
257 | { | |
258 | void *base; | |
259 | ||
a973e9dd | 260 | if (!object) |
02cbc874 CL |
261 | return 1; |
262 | ||
a973e9dd | 263 | base = page_address(page); |
39b26464 | 264 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
265 | (object - base) % s->size) { |
266 | return 0; | |
267 | } | |
268 | ||
269 | return 1; | |
270 | } | |
271 | ||
7656c72b CL |
272 | /* |
273 | * Slow version of get and set free pointer. | |
274 | * | |
275 | * This version requires touching the cache lines of kmem_cache which | |
276 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
277 | * from the page struct. | |
278 | */ | |
279 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
280 | { | |
281 | return *(void **)(object + s->offset); | |
282 | } | |
283 | ||
284 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
285 | { | |
286 | *(void **)(object + s->offset) = fp; | |
287 | } | |
288 | ||
289 | /* Loop over all objects in a slab */ | |
224a88be CL |
290 | #define for_each_object(__p, __s, __addr, __objects) \ |
291 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
292 | __p += (__s)->size) |
293 | ||
294 | /* Scan freelist */ | |
295 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 296 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
297 | |
298 | /* Determine object index from a given position */ | |
299 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
300 | { | |
301 | return (p - addr) / s->size; | |
302 | } | |
303 | ||
834f3d11 CL |
304 | static inline struct kmem_cache_order_objects oo_make(int order, |
305 | unsigned long size) | |
306 | { | |
307 | struct kmem_cache_order_objects x = { | |
210b5c06 | 308 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
309 | }; |
310 | ||
311 | return x; | |
312 | } | |
313 | ||
314 | static inline int oo_order(struct kmem_cache_order_objects x) | |
315 | { | |
210b5c06 | 316 | return x.x >> OO_SHIFT; |
834f3d11 CL |
317 | } |
318 | ||
319 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
320 | { | |
210b5c06 | 321 | return x.x & OO_MASK; |
834f3d11 CL |
322 | } |
323 | ||
41ecc55b CL |
324 | #ifdef CONFIG_SLUB_DEBUG |
325 | /* | |
326 | * Debug settings: | |
327 | */ | |
f0630fff CL |
328 | #ifdef CONFIG_SLUB_DEBUG_ON |
329 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
330 | #else | |
41ecc55b | 331 | static int slub_debug; |
f0630fff | 332 | #endif |
41ecc55b CL |
333 | |
334 | static char *slub_debug_slabs; | |
fa5ec8a1 | 335 | static int disable_higher_order_debug; |
41ecc55b | 336 | |
81819f0f CL |
337 | /* |
338 | * Object debugging | |
339 | */ | |
340 | static void print_section(char *text, u8 *addr, unsigned int length) | |
341 | { | |
342 | int i, offset; | |
343 | int newline = 1; | |
344 | char ascii[17]; | |
345 | ||
346 | ascii[16] = 0; | |
347 | ||
348 | for (i = 0; i < length; i++) { | |
349 | if (newline) { | |
24922684 | 350 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
351 | newline = 0; |
352 | } | |
06428780 | 353 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
354 | offset = i % 16; |
355 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
356 | if (offset == 15) { | |
06428780 | 357 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
358 | newline = 1; |
359 | } | |
360 | } | |
361 | if (!newline) { | |
362 | i %= 16; | |
363 | while (i < 16) { | |
06428780 | 364 | printk(KERN_CONT " "); |
81819f0f CL |
365 | ascii[i] = ' '; |
366 | i++; | |
367 | } | |
06428780 | 368 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
369 | } |
370 | } | |
371 | ||
81819f0f CL |
372 | static struct track *get_track(struct kmem_cache *s, void *object, |
373 | enum track_item alloc) | |
374 | { | |
375 | struct track *p; | |
376 | ||
377 | if (s->offset) | |
378 | p = object + s->offset + sizeof(void *); | |
379 | else | |
380 | p = object + s->inuse; | |
381 | ||
382 | return p + alloc; | |
383 | } | |
384 | ||
385 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 386 | enum track_item alloc, unsigned long addr) |
81819f0f | 387 | { |
1a00df4a | 388 | struct track *p = get_track(s, object, alloc); |
81819f0f | 389 | |
81819f0f CL |
390 | if (addr) { |
391 | p->addr = addr; | |
392 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 393 | p->pid = current->pid; |
81819f0f CL |
394 | p->when = jiffies; |
395 | } else | |
396 | memset(p, 0, sizeof(struct track)); | |
397 | } | |
398 | ||
81819f0f CL |
399 | static void init_tracking(struct kmem_cache *s, void *object) |
400 | { | |
24922684 CL |
401 | if (!(s->flags & SLAB_STORE_USER)) |
402 | return; | |
403 | ||
ce71e27c EGM |
404 | set_track(s, object, TRACK_FREE, 0UL); |
405 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
406 | } |
407 | ||
408 | static void print_track(const char *s, struct track *t) | |
409 | { | |
410 | if (!t->addr) | |
411 | return; | |
412 | ||
7daf705f | 413 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 414 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
415 | } |
416 | ||
417 | static void print_tracking(struct kmem_cache *s, void *object) | |
418 | { | |
419 | if (!(s->flags & SLAB_STORE_USER)) | |
420 | return; | |
421 | ||
422 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
423 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
424 | } | |
425 | ||
426 | static void print_page_info(struct page *page) | |
427 | { | |
39b26464 CL |
428 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
429 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
430 | |
431 | } | |
432 | ||
433 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
434 | { | |
435 | va_list args; | |
436 | char buf[100]; | |
437 | ||
438 | va_start(args, fmt); | |
439 | vsnprintf(buf, sizeof(buf), fmt, args); | |
440 | va_end(args); | |
441 | printk(KERN_ERR "========================================" | |
442 | "=====================================\n"); | |
443 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
444 | printk(KERN_ERR "----------------------------------------" | |
445 | "-------------------------------------\n\n"); | |
81819f0f CL |
446 | } |
447 | ||
24922684 CL |
448 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
449 | { | |
450 | va_list args; | |
451 | char buf[100]; | |
452 | ||
453 | va_start(args, fmt); | |
454 | vsnprintf(buf, sizeof(buf), fmt, args); | |
455 | va_end(args); | |
456 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
457 | } | |
458 | ||
459 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
460 | { |
461 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 462 | u8 *addr = page_address(page); |
24922684 CL |
463 | |
464 | print_tracking(s, p); | |
465 | ||
466 | print_page_info(page); | |
467 | ||
468 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
469 | p, p - addr, get_freepointer(s, p)); | |
470 | ||
471 | if (p > addr + 16) | |
472 | print_section("Bytes b4", p - 16, 16); | |
473 | ||
0ebd652b | 474 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
475 | |
476 | if (s->flags & SLAB_RED_ZONE) | |
477 | print_section("Redzone", p + s->objsize, | |
478 | s->inuse - s->objsize); | |
479 | ||
81819f0f CL |
480 | if (s->offset) |
481 | off = s->offset + sizeof(void *); | |
482 | else | |
483 | off = s->inuse; | |
484 | ||
24922684 | 485 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 486 | off += 2 * sizeof(struct track); |
81819f0f CL |
487 | |
488 | if (off != s->size) | |
489 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
490 | print_section("Padding", p + off, s->size - off); |
491 | ||
492 | dump_stack(); | |
81819f0f CL |
493 | } |
494 | ||
495 | static void object_err(struct kmem_cache *s, struct page *page, | |
496 | u8 *object, char *reason) | |
497 | { | |
3dc50637 | 498 | slab_bug(s, "%s", reason); |
24922684 | 499 | print_trailer(s, page, object); |
81819f0f CL |
500 | } |
501 | ||
24922684 | 502 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
503 | { |
504 | va_list args; | |
505 | char buf[100]; | |
506 | ||
24922684 CL |
507 | va_start(args, fmt); |
508 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 509 | va_end(args); |
3dc50637 | 510 | slab_bug(s, "%s", buf); |
24922684 | 511 | print_page_info(page); |
81819f0f CL |
512 | dump_stack(); |
513 | } | |
514 | ||
515 | static void init_object(struct kmem_cache *s, void *object, int active) | |
516 | { | |
517 | u8 *p = object; | |
518 | ||
519 | if (s->flags & __OBJECT_POISON) { | |
520 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 521 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
522 | } |
523 | ||
524 | if (s->flags & SLAB_RED_ZONE) | |
525 | memset(p + s->objsize, | |
526 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
527 | s->inuse - s->objsize); | |
528 | } | |
529 | ||
24922684 | 530 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
531 | { |
532 | while (bytes) { | |
533 | if (*start != (u8)value) | |
24922684 | 534 | return start; |
81819f0f CL |
535 | start++; |
536 | bytes--; | |
537 | } | |
24922684 CL |
538 | return NULL; |
539 | } | |
540 | ||
541 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
542 | void *from, void *to) | |
543 | { | |
544 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
545 | memset(from, data, to - from); | |
546 | } | |
547 | ||
548 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
549 | u8 *object, char *what, | |
06428780 | 550 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
551 | { |
552 | u8 *fault; | |
553 | u8 *end; | |
554 | ||
555 | fault = check_bytes(start, value, bytes); | |
556 | if (!fault) | |
557 | return 1; | |
558 | ||
559 | end = start + bytes; | |
560 | while (end > fault && end[-1] == value) | |
561 | end--; | |
562 | ||
563 | slab_bug(s, "%s overwritten", what); | |
564 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
565 | fault, end - 1, fault[0], value); | |
566 | print_trailer(s, page, object); | |
567 | ||
568 | restore_bytes(s, what, value, fault, end); | |
569 | return 0; | |
81819f0f CL |
570 | } |
571 | ||
81819f0f CL |
572 | /* |
573 | * Object layout: | |
574 | * | |
575 | * object address | |
576 | * Bytes of the object to be managed. | |
577 | * If the freepointer may overlay the object then the free | |
578 | * pointer is the first word of the object. | |
672bba3a | 579 | * |
81819f0f CL |
580 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
581 | * 0xa5 (POISON_END) | |
582 | * | |
583 | * object + s->objsize | |
584 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
585 | * Padding is extended by another word if Redzoning is enabled and |
586 | * objsize == inuse. | |
587 | * | |
81819f0f CL |
588 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
589 | * 0xcc (RED_ACTIVE) for objects in use. | |
590 | * | |
591 | * object + s->inuse | |
672bba3a CL |
592 | * Meta data starts here. |
593 | * | |
81819f0f CL |
594 | * A. Free pointer (if we cannot overwrite object on free) |
595 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 596 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 597 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
598 | * before the word boundary. |
599 | * | |
600 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
601 | * |
602 | * object + s->size | |
672bba3a | 603 | * Nothing is used beyond s->size. |
81819f0f | 604 | * |
672bba3a CL |
605 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
606 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
607 | * may be used with merged slabcaches. |
608 | */ | |
609 | ||
81819f0f CL |
610 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
611 | { | |
612 | unsigned long off = s->inuse; /* The end of info */ | |
613 | ||
614 | if (s->offset) | |
615 | /* Freepointer is placed after the object. */ | |
616 | off += sizeof(void *); | |
617 | ||
618 | if (s->flags & SLAB_STORE_USER) | |
619 | /* We also have user information there */ | |
620 | off += 2 * sizeof(struct track); | |
621 | ||
622 | if (s->size == off) | |
623 | return 1; | |
624 | ||
24922684 CL |
625 | return check_bytes_and_report(s, page, p, "Object padding", |
626 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
627 | } |
628 | ||
39b26464 | 629 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
630 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
631 | { | |
24922684 CL |
632 | u8 *start; |
633 | u8 *fault; | |
634 | u8 *end; | |
635 | int length; | |
636 | int remainder; | |
81819f0f CL |
637 | |
638 | if (!(s->flags & SLAB_POISON)) | |
639 | return 1; | |
640 | ||
a973e9dd | 641 | start = page_address(page); |
834f3d11 | 642 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
643 | end = start + length; |
644 | remainder = length % s->size; | |
81819f0f CL |
645 | if (!remainder) |
646 | return 1; | |
647 | ||
39b26464 | 648 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
649 | if (!fault) |
650 | return 1; | |
651 | while (end > fault && end[-1] == POISON_INUSE) | |
652 | end--; | |
653 | ||
654 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 655 | print_section("Padding", end - remainder, remainder); |
24922684 | 656 | |
8a3d271d | 657 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 658 | return 0; |
81819f0f CL |
659 | } |
660 | ||
661 | static int check_object(struct kmem_cache *s, struct page *page, | |
662 | void *object, int active) | |
663 | { | |
664 | u8 *p = object; | |
665 | u8 *endobject = object + s->objsize; | |
666 | ||
667 | if (s->flags & SLAB_RED_ZONE) { | |
668 | unsigned int red = | |
669 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
670 | ||
24922684 CL |
671 | if (!check_bytes_and_report(s, page, object, "Redzone", |
672 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 673 | return 0; |
81819f0f | 674 | } else { |
3adbefee IM |
675 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
676 | check_bytes_and_report(s, page, p, "Alignment padding", | |
677 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
678 | } | |
81819f0f CL |
679 | } |
680 | ||
681 | if (s->flags & SLAB_POISON) { | |
682 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
683 | (!check_bytes_and_report(s, page, p, "Poison", p, |
684 | POISON_FREE, s->objsize - 1) || | |
685 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 686 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 687 | return 0; |
81819f0f CL |
688 | /* |
689 | * check_pad_bytes cleans up on its own. | |
690 | */ | |
691 | check_pad_bytes(s, page, p); | |
692 | } | |
693 | ||
694 | if (!s->offset && active) | |
695 | /* | |
696 | * Object and freepointer overlap. Cannot check | |
697 | * freepointer while object is allocated. | |
698 | */ | |
699 | return 1; | |
700 | ||
701 | /* Check free pointer validity */ | |
702 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
703 | object_err(s, page, p, "Freepointer corrupt"); | |
704 | /* | |
9f6c708e | 705 | * No choice but to zap it and thus lose the remainder |
81819f0f | 706 | * of the free objects in this slab. May cause |
672bba3a | 707 | * another error because the object count is now wrong. |
81819f0f | 708 | */ |
a973e9dd | 709 | set_freepointer(s, p, NULL); |
81819f0f CL |
710 | return 0; |
711 | } | |
712 | return 1; | |
713 | } | |
714 | ||
715 | static int check_slab(struct kmem_cache *s, struct page *page) | |
716 | { | |
39b26464 CL |
717 | int maxobj; |
718 | ||
81819f0f CL |
719 | VM_BUG_ON(!irqs_disabled()); |
720 | ||
721 | if (!PageSlab(page)) { | |
24922684 | 722 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
723 | return 0; |
724 | } | |
39b26464 CL |
725 | |
726 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
727 | if (page->objects > maxobj) { | |
728 | slab_err(s, page, "objects %u > max %u", | |
729 | s->name, page->objects, maxobj); | |
730 | return 0; | |
731 | } | |
732 | if (page->inuse > page->objects) { | |
24922684 | 733 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 734 | s->name, page->inuse, page->objects); |
81819f0f CL |
735 | return 0; |
736 | } | |
737 | /* Slab_pad_check fixes things up after itself */ | |
738 | slab_pad_check(s, page); | |
739 | return 1; | |
740 | } | |
741 | ||
742 | /* | |
672bba3a CL |
743 | * Determine if a certain object on a page is on the freelist. Must hold the |
744 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
745 | */ |
746 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
747 | { | |
748 | int nr = 0; | |
749 | void *fp = page->freelist; | |
750 | void *object = NULL; | |
224a88be | 751 | unsigned long max_objects; |
81819f0f | 752 | |
39b26464 | 753 | while (fp && nr <= page->objects) { |
81819f0f CL |
754 | if (fp == search) |
755 | return 1; | |
756 | if (!check_valid_pointer(s, page, fp)) { | |
757 | if (object) { | |
758 | object_err(s, page, object, | |
759 | "Freechain corrupt"); | |
a973e9dd | 760 | set_freepointer(s, object, NULL); |
81819f0f CL |
761 | break; |
762 | } else { | |
24922684 | 763 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 764 | page->freelist = NULL; |
39b26464 | 765 | page->inuse = page->objects; |
24922684 | 766 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
767 | return 0; |
768 | } | |
769 | break; | |
770 | } | |
771 | object = fp; | |
772 | fp = get_freepointer(s, object); | |
773 | nr++; | |
774 | } | |
775 | ||
224a88be | 776 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
777 | if (max_objects > MAX_OBJS_PER_PAGE) |
778 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
779 | |
780 | if (page->objects != max_objects) { | |
781 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
782 | "should be %d", page->objects, max_objects); | |
783 | page->objects = max_objects; | |
784 | slab_fix(s, "Number of objects adjusted."); | |
785 | } | |
39b26464 | 786 | if (page->inuse != page->objects - nr) { |
70d71228 | 787 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
788 | "counted were %d", page->inuse, page->objects - nr); |
789 | page->inuse = page->objects - nr; | |
24922684 | 790 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
791 | } |
792 | return search == NULL; | |
793 | } | |
794 | ||
0121c619 CL |
795 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
796 | int alloc) | |
3ec09742 CL |
797 | { |
798 | if (s->flags & SLAB_TRACE) { | |
799 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
800 | s->name, | |
801 | alloc ? "alloc" : "free", | |
802 | object, page->inuse, | |
803 | page->freelist); | |
804 | ||
805 | if (!alloc) | |
806 | print_section("Object", (void *)object, s->objsize); | |
807 | ||
808 | dump_stack(); | |
809 | } | |
810 | } | |
811 | ||
643b1138 | 812 | /* |
672bba3a | 813 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 814 | */ |
e95eed57 | 815 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 816 | { |
643b1138 CL |
817 | spin_lock(&n->list_lock); |
818 | list_add(&page->lru, &n->full); | |
819 | spin_unlock(&n->list_lock); | |
820 | } | |
821 | ||
822 | static void remove_full(struct kmem_cache *s, struct page *page) | |
823 | { | |
824 | struct kmem_cache_node *n; | |
825 | ||
826 | if (!(s->flags & SLAB_STORE_USER)) | |
827 | return; | |
828 | ||
829 | n = get_node(s, page_to_nid(page)); | |
830 | ||
831 | spin_lock(&n->list_lock); | |
832 | list_del(&page->lru); | |
833 | spin_unlock(&n->list_lock); | |
834 | } | |
835 | ||
0f389ec6 CL |
836 | /* Tracking of the number of slabs for debugging purposes */ |
837 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
838 | { | |
839 | struct kmem_cache_node *n = get_node(s, node); | |
840 | ||
841 | return atomic_long_read(&n->nr_slabs); | |
842 | } | |
843 | ||
26c02cf0 AB |
844 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
845 | { | |
846 | return atomic_long_read(&n->nr_slabs); | |
847 | } | |
848 | ||
205ab99d | 849 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
850 | { |
851 | struct kmem_cache_node *n = get_node(s, node); | |
852 | ||
853 | /* | |
854 | * May be called early in order to allocate a slab for the | |
855 | * kmem_cache_node structure. Solve the chicken-egg | |
856 | * dilemma by deferring the increment of the count during | |
857 | * bootstrap (see early_kmem_cache_node_alloc). | |
858 | */ | |
205ab99d | 859 | if (!NUMA_BUILD || n) { |
0f389ec6 | 860 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
861 | atomic_long_add(objects, &n->total_objects); |
862 | } | |
0f389ec6 | 863 | } |
205ab99d | 864 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
865 | { |
866 | struct kmem_cache_node *n = get_node(s, node); | |
867 | ||
868 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 869 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
870 | } |
871 | ||
872 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
873 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
874 | void *object) | |
875 | { | |
876 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
877 | return; | |
878 | ||
879 | init_object(s, object, 0); | |
880 | init_tracking(s, object); | |
881 | } | |
882 | ||
883 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 884 | void *object, unsigned long addr) |
81819f0f CL |
885 | { |
886 | if (!check_slab(s, page)) | |
887 | goto bad; | |
888 | ||
d692ef6d | 889 | if (!on_freelist(s, page, object)) { |
24922684 | 890 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 891 | goto bad; |
81819f0f CL |
892 | } |
893 | ||
894 | if (!check_valid_pointer(s, page, object)) { | |
895 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 896 | goto bad; |
81819f0f CL |
897 | } |
898 | ||
d692ef6d | 899 | if (!check_object(s, page, object, 0)) |
81819f0f | 900 | goto bad; |
81819f0f | 901 | |
3ec09742 CL |
902 | /* Success perform special debug activities for allocs */ |
903 | if (s->flags & SLAB_STORE_USER) | |
904 | set_track(s, object, TRACK_ALLOC, addr); | |
905 | trace(s, page, object, 1); | |
906 | init_object(s, object, 1); | |
81819f0f | 907 | return 1; |
3ec09742 | 908 | |
81819f0f CL |
909 | bad: |
910 | if (PageSlab(page)) { | |
911 | /* | |
912 | * If this is a slab page then lets do the best we can | |
913 | * to avoid issues in the future. Marking all objects | |
672bba3a | 914 | * as used avoids touching the remaining objects. |
81819f0f | 915 | */ |
24922684 | 916 | slab_fix(s, "Marking all objects used"); |
39b26464 | 917 | page->inuse = page->objects; |
a973e9dd | 918 | page->freelist = NULL; |
81819f0f CL |
919 | } |
920 | return 0; | |
921 | } | |
922 | ||
3ec09742 | 923 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 924 | void *object, unsigned long addr) |
81819f0f CL |
925 | { |
926 | if (!check_slab(s, page)) | |
927 | goto fail; | |
928 | ||
929 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 930 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
931 | goto fail; |
932 | } | |
933 | ||
934 | if (on_freelist(s, page, object)) { | |
24922684 | 935 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
936 | goto fail; |
937 | } | |
938 | ||
939 | if (!check_object(s, page, object, 1)) | |
940 | return 0; | |
941 | ||
942 | if (unlikely(s != page->slab)) { | |
3adbefee | 943 | if (!PageSlab(page)) { |
70d71228 CL |
944 | slab_err(s, page, "Attempt to free object(0x%p) " |
945 | "outside of slab", object); | |
3adbefee | 946 | } else if (!page->slab) { |
81819f0f | 947 | printk(KERN_ERR |
70d71228 | 948 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 949 | object); |
70d71228 | 950 | dump_stack(); |
06428780 | 951 | } else |
24922684 CL |
952 | object_err(s, page, object, |
953 | "page slab pointer corrupt."); | |
81819f0f CL |
954 | goto fail; |
955 | } | |
3ec09742 CL |
956 | |
957 | /* Special debug activities for freeing objects */ | |
8a38082d | 958 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
959 | remove_full(s, page); |
960 | if (s->flags & SLAB_STORE_USER) | |
961 | set_track(s, object, TRACK_FREE, addr); | |
962 | trace(s, page, object, 0); | |
963 | init_object(s, object, 0); | |
81819f0f | 964 | return 1; |
3ec09742 | 965 | |
81819f0f | 966 | fail: |
24922684 | 967 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
968 | return 0; |
969 | } | |
970 | ||
41ecc55b CL |
971 | static int __init setup_slub_debug(char *str) |
972 | { | |
f0630fff CL |
973 | slub_debug = DEBUG_DEFAULT_FLAGS; |
974 | if (*str++ != '=' || !*str) | |
975 | /* | |
976 | * No options specified. Switch on full debugging. | |
977 | */ | |
978 | goto out; | |
979 | ||
980 | if (*str == ',') | |
981 | /* | |
982 | * No options but restriction on slabs. This means full | |
983 | * debugging for slabs matching a pattern. | |
984 | */ | |
985 | goto check_slabs; | |
986 | ||
fa5ec8a1 DR |
987 | if (tolower(*str) == 'o') { |
988 | /* | |
989 | * Avoid enabling debugging on caches if its minimum order | |
990 | * would increase as a result. | |
991 | */ | |
992 | disable_higher_order_debug = 1; | |
993 | goto out; | |
994 | } | |
995 | ||
f0630fff CL |
996 | slub_debug = 0; |
997 | if (*str == '-') | |
998 | /* | |
999 | * Switch off all debugging measures. | |
1000 | */ | |
1001 | goto out; | |
1002 | ||
1003 | /* | |
1004 | * Determine which debug features should be switched on | |
1005 | */ | |
06428780 | 1006 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1007 | switch (tolower(*str)) { |
1008 | case 'f': | |
1009 | slub_debug |= SLAB_DEBUG_FREE; | |
1010 | break; | |
1011 | case 'z': | |
1012 | slub_debug |= SLAB_RED_ZONE; | |
1013 | break; | |
1014 | case 'p': | |
1015 | slub_debug |= SLAB_POISON; | |
1016 | break; | |
1017 | case 'u': | |
1018 | slub_debug |= SLAB_STORE_USER; | |
1019 | break; | |
1020 | case 't': | |
1021 | slub_debug |= SLAB_TRACE; | |
1022 | break; | |
1023 | default: | |
1024 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1025 | "unknown. skipped\n", *str); |
f0630fff | 1026 | } |
41ecc55b CL |
1027 | } |
1028 | ||
f0630fff | 1029 | check_slabs: |
41ecc55b CL |
1030 | if (*str == ',') |
1031 | slub_debug_slabs = str + 1; | |
f0630fff | 1032 | out: |
41ecc55b CL |
1033 | return 1; |
1034 | } | |
1035 | ||
1036 | __setup("slub_debug", setup_slub_debug); | |
1037 | ||
ba0268a8 CL |
1038 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1039 | unsigned long flags, const char *name, | |
51cc5068 | 1040 | void (*ctor)(void *)) |
41ecc55b CL |
1041 | { |
1042 | /* | |
e153362a | 1043 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1044 | */ |
e153362a | 1045 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1046 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1047 | flags |= slub_debug; | |
ba0268a8 CL |
1048 | |
1049 | return flags; | |
41ecc55b CL |
1050 | } |
1051 | #else | |
3ec09742 CL |
1052 | static inline void setup_object_debug(struct kmem_cache *s, |
1053 | struct page *page, void *object) {} | |
41ecc55b | 1054 | |
3ec09742 | 1055 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1056 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1057 | |
3ec09742 | 1058 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1059 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1060 | |
41ecc55b CL |
1061 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1062 | { return 1; } | |
1063 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1064 | void *object, int active) { return 1; } | |
3ec09742 | 1065 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1066 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1067 | unsigned long flags, const char *name, | |
51cc5068 | 1068 | void (*ctor)(void *)) |
ba0268a8 CL |
1069 | { |
1070 | return flags; | |
1071 | } | |
41ecc55b | 1072 | #define slub_debug 0 |
0f389ec6 CL |
1073 | |
1074 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1075 | { return 0; } | |
26c02cf0 AB |
1076 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1077 | { return 0; } | |
205ab99d CL |
1078 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1079 | int objects) {} | |
1080 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1081 | int objects) {} | |
41ecc55b | 1082 | #endif |
205ab99d | 1083 | |
81819f0f CL |
1084 | /* |
1085 | * Slab allocation and freeing | |
1086 | */ | |
65c3376a CL |
1087 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1088 | struct kmem_cache_order_objects oo) | |
1089 | { | |
1090 | int order = oo_order(oo); | |
1091 | ||
b1eeab67 VN |
1092 | flags |= __GFP_NOTRACK; |
1093 | ||
65c3376a CL |
1094 | if (node == -1) |
1095 | return alloc_pages(flags, order); | |
1096 | else | |
1097 | return alloc_pages_node(node, flags, order); | |
1098 | } | |
1099 | ||
81819f0f CL |
1100 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1101 | { | |
06428780 | 1102 | struct page *page; |
834f3d11 | 1103 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1104 | gfp_t alloc_gfp; |
81819f0f | 1105 | |
b7a49f0d | 1106 | flags |= s->allocflags; |
e12ba74d | 1107 | |
ba52270d PE |
1108 | /* |
1109 | * Let the initial higher-order allocation fail under memory pressure | |
1110 | * so we fall-back to the minimum order allocation. | |
1111 | */ | |
1112 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1113 | ||
1114 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1115 | if (unlikely(!page)) { |
1116 | oo = s->min; | |
1117 | /* | |
1118 | * Allocation may have failed due to fragmentation. | |
1119 | * Try a lower order alloc if possible | |
1120 | */ | |
1121 | page = alloc_slab_page(flags, node, oo); | |
1122 | if (!page) | |
1123 | return NULL; | |
81819f0f | 1124 | |
65c3376a CL |
1125 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1126 | } | |
5a896d9e VN |
1127 | |
1128 | if (kmemcheck_enabled | |
5086c389 | 1129 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1130 | int pages = 1 << oo_order(oo); |
1131 | ||
1132 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1133 | ||
1134 | /* | |
1135 | * Objects from caches that have a constructor don't get | |
1136 | * cleared when they're allocated, so we need to do it here. | |
1137 | */ | |
1138 | if (s->ctor) | |
1139 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1140 | else | |
1141 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1142 | } |
1143 | ||
834f3d11 | 1144 | page->objects = oo_objects(oo); |
81819f0f CL |
1145 | mod_zone_page_state(page_zone(page), |
1146 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1147 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1148 | 1 << oo_order(oo)); |
81819f0f CL |
1149 | |
1150 | return page; | |
1151 | } | |
1152 | ||
1153 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1154 | void *object) | |
1155 | { | |
3ec09742 | 1156 | setup_object_debug(s, page, object); |
4f104934 | 1157 | if (unlikely(s->ctor)) |
51cc5068 | 1158 | s->ctor(object); |
81819f0f CL |
1159 | } |
1160 | ||
1161 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1162 | { | |
1163 | struct page *page; | |
81819f0f | 1164 | void *start; |
81819f0f CL |
1165 | void *last; |
1166 | void *p; | |
1167 | ||
6cb06229 | 1168 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1169 | |
6cb06229 CL |
1170 | page = allocate_slab(s, |
1171 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1172 | if (!page) |
1173 | goto out; | |
1174 | ||
205ab99d | 1175 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1176 | page->slab = s; |
1177 | page->flags |= 1 << PG_slab; | |
1178 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1179 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1180 | __SetPageSlubDebug(page); |
81819f0f CL |
1181 | |
1182 | start = page_address(page); | |
81819f0f CL |
1183 | |
1184 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1185 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1186 | |
1187 | last = start; | |
224a88be | 1188 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1189 | setup_object(s, page, last); |
1190 | set_freepointer(s, last, p); | |
1191 | last = p; | |
1192 | } | |
1193 | setup_object(s, page, last); | |
a973e9dd | 1194 | set_freepointer(s, last, NULL); |
81819f0f CL |
1195 | |
1196 | page->freelist = start; | |
1197 | page->inuse = 0; | |
1198 | out: | |
81819f0f CL |
1199 | return page; |
1200 | } | |
1201 | ||
1202 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1203 | { | |
834f3d11 CL |
1204 | int order = compound_order(page); |
1205 | int pages = 1 << order; | |
81819f0f | 1206 | |
8a38082d | 1207 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1208 | void *p; |
1209 | ||
1210 | slab_pad_check(s, page); | |
224a88be CL |
1211 | for_each_object(p, s, page_address(page), |
1212 | page->objects) | |
81819f0f | 1213 | check_object(s, page, p, 0); |
8a38082d | 1214 | __ClearPageSlubDebug(page); |
81819f0f CL |
1215 | } |
1216 | ||
b1eeab67 | 1217 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1218 | |
81819f0f CL |
1219 | mod_zone_page_state(page_zone(page), |
1220 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1221 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1222 | -pages); |
81819f0f | 1223 | |
49bd5221 CL |
1224 | __ClearPageSlab(page); |
1225 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1226 | if (current->reclaim_state) |
1227 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1228 | __free_pages(page, order); |
81819f0f CL |
1229 | } |
1230 | ||
1231 | static void rcu_free_slab(struct rcu_head *h) | |
1232 | { | |
1233 | struct page *page; | |
1234 | ||
1235 | page = container_of((struct list_head *)h, struct page, lru); | |
1236 | __free_slab(page->slab, page); | |
1237 | } | |
1238 | ||
1239 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1240 | { | |
1241 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1242 | /* | |
1243 | * RCU free overloads the RCU head over the LRU | |
1244 | */ | |
1245 | struct rcu_head *head = (void *)&page->lru; | |
1246 | ||
1247 | call_rcu(head, rcu_free_slab); | |
1248 | } else | |
1249 | __free_slab(s, page); | |
1250 | } | |
1251 | ||
1252 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1253 | { | |
205ab99d | 1254 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1255 | free_slab(s, page); |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Per slab locking using the pagelock | |
1260 | */ | |
1261 | static __always_inline void slab_lock(struct page *page) | |
1262 | { | |
1263 | bit_spin_lock(PG_locked, &page->flags); | |
1264 | } | |
1265 | ||
1266 | static __always_inline void slab_unlock(struct page *page) | |
1267 | { | |
a76d3546 | 1268 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1269 | } |
1270 | ||
1271 | static __always_inline int slab_trylock(struct page *page) | |
1272 | { | |
1273 | int rc = 1; | |
1274 | ||
1275 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1276 | return rc; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Management of partially allocated slabs | |
1281 | */ | |
7c2e132c CL |
1282 | static void add_partial(struct kmem_cache_node *n, |
1283 | struct page *page, int tail) | |
81819f0f | 1284 | { |
e95eed57 CL |
1285 | spin_lock(&n->list_lock); |
1286 | n->nr_partial++; | |
7c2e132c CL |
1287 | if (tail) |
1288 | list_add_tail(&page->lru, &n->partial); | |
1289 | else | |
1290 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1291 | spin_unlock(&n->list_lock); |
1292 | } | |
1293 | ||
0121c619 | 1294 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1295 | { |
1296 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1297 | ||
1298 | spin_lock(&n->list_lock); | |
1299 | list_del(&page->lru); | |
1300 | n->nr_partial--; | |
1301 | spin_unlock(&n->list_lock); | |
1302 | } | |
1303 | ||
1304 | /* | |
672bba3a | 1305 | * Lock slab and remove from the partial list. |
81819f0f | 1306 | * |
672bba3a | 1307 | * Must hold list_lock. |
81819f0f | 1308 | */ |
0121c619 CL |
1309 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1310 | struct page *page) | |
81819f0f CL |
1311 | { |
1312 | if (slab_trylock(page)) { | |
1313 | list_del(&page->lru); | |
1314 | n->nr_partial--; | |
8a38082d | 1315 | __SetPageSlubFrozen(page); |
81819f0f CL |
1316 | return 1; |
1317 | } | |
1318 | return 0; | |
1319 | } | |
1320 | ||
1321 | /* | |
672bba3a | 1322 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1323 | */ |
1324 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1325 | { | |
1326 | struct page *page; | |
1327 | ||
1328 | /* | |
1329 | * Racy check. If we mistakenly see no partial slabs then we | |
1330 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1331 | * partial slab and there is none available then get_partials() |
1332 | * will return NULL. | |
81819f0f CL |
1333 | */ |
1334 | if (!n || !n->nr_partial) | |
1335 | return NULL; | |
1336 | ||
1337 | spin_lock(&n->list_lock); | |
1338 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1339 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1340 | goto out; |
1341 | page = NULL; | |
1342 | out: | |
1343 | spin_unlock(&n->list_lock); | |
1344 | return page; | |
1345 | } | |
1346 | ||
1347 | /* | |
672bba3a | 1348 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1349 | */ |
1350 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1351 | { | |
1352 | #ifdef CONFIG_NUMA | |
1353 | struct zonelist *zonelist; | |
dd1a239f | 1354 | struct zoneref *z; |
54a6eb5c MG |
1355 | struct zone *zone; |
1356 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1357 | struct page *page; |
1358 | ||
1359 | /* | |
672bba3a CL |
1360 | * The defrag ratio allows a configuration of the tradeoffs between |
1361 | * inter node defragmentation and node local allocations. A lower | |
1362 | * defrag_ratio increases the tendency to do local allocations | |
1363 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1364 | * |
672bba3a CL |
1365 | * If the defrag_ratio is set to 0 then kmalloc() always |
1366 | * returns node local objects. If the ratio is higher then kmalloc() | |
1367 | * may return off node objects because partial slabs are obtained | |
1368 | * from other nodes and filled up. | |
81819f0f | 1369 | * |
6446faa2 | 1370 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1371 | * defrag_ratio = 1000) then every (well almost) allocation will |
1372 | * first attempt to defrag slab caches on other nodes. This means | |
1373 | * scanning over all nodes to look for partial slabs which may be | |
1374 | * expensive if we do it every time we are trying to find a slab | |
1375 | * with available objects. | |
81819f0f | 1376 | */ |
9824601e CL |
1377 | if (!s->remote_node_defrag_ratio || |
1378 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1379 | return NULL; |
1380 | ||
0e88460d | 1381 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1382 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1383 | struct kmem_cache_node *n; |
1384 | ||
54a6eb5c | 1385 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1386 | |
54a6eb5c | 1387 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1388 | n->nr_partial > s->min_partial) { |
81819f0f CL |
1389 | page = get_partial_node(n); |
1390 | if (page) | |
1391 | return page; | |
1392 | } | |
1393 | } | |
1394 | #endif | |
1395 | return NULL; | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Get a partial page, lock it and return it. | |
1400 | */ | |
1401 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1402 | { | |
1403 | struct page *page; | |
1404 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1405 | ||
1406 | page = get_partial_node(get_node(s, searchnode)); | |
1407 | if (page || (flags & __GFP_THISNODE)) | |
1408 | return page; | |
1409 | ||
1410 | return get_any_partial(s, flags); | |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * Move a page back to the lists. | |
1415 | * | |
1416 | * Must be called with the slab lock held. | |
1417 | * | |
1418 | * On exit the slab lock will have been dropped. | |
1419 | */ | |
7c2e132c | 1420 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1421 | { |
e95eed57 | 1422 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1423 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1424 | |
8a38082d | 1425 | __ClearPageSlubFrozen(page); |
81819f0f | 1426 | if (page->inuse) { |
e95eed57 | 1427 | |
a973e9dd | 1428 | if (page->freelist) { |
7c2e132c | 1429 | add_partial(n, page, tail); |
8ff12cfc CL |
1430 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1431 | } else { | |
1432 | stat(c, DEACTIVATE_FULL); | |
8a38082d AW |
1433 | if (SLABDEBUG && PageSlubDebug(page) && |
1434 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1435 | add_full(n, page); |
1436 | } | |
81819f0f CL |
1437 | slab_unlock(page); |
1438 | } else { | |
8ff12cfc | 1439 | stat(c, DEACTIVATE_EMPTY); |
3b89d7d8 | 1440 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1441 | /* |
672bba3a CL |
1442 | * Adding an empty slab to the partial slabs in order |
1443 | * to avoid page allocator overhead. This slab needs | |
1444 | * to come after the other slabs with objects in | |
6446faa2 CL |
1445 | * so that the others get filled first. That way the |
1446 | * size of the partial list stays small. | |
1447 | * | |
0121c619 CL |
1448 | * kmem_cache_shrink can reclaim any empty slabs from |
1449 | * the partial list. | |
e95eed57 | 1450 | */ |
7c2e132c | 1451 | add_partial(n, page, 1); |
e95eed57 CL |
1452 | slab_unlock(page); |
1453 | } else { | |
1454 | slab_unlock(page); | |
8ff12cfc | 1455 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1456 | discard_slab(s, page); |
1457 | } | |
81819f0f CL |
1458 | } |
1459 | } | |
1460 | ||
1461 | /* | |
1462 | * Remove the cpu slab | |
1463 | */ | |
dfb4f096 | 1464 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1465 | { |
dfb4f096 | 1466 | struct page *page = c->page; |
7c2e132c | 1467 | int tail = 1; |
8ff12cfc | 1468 | |
b773ad73 | 1469 | if (page->freelist) |
8ff12cfc | 1470 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1471 | /* |
6446faa2 | 1472 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1473 | * because both freelists are empty. So this is unlikely |
1474 | * to occur. | |
1475 | */ | |
a973e9dd | 1476 | while (unlikely(c->freelist)) { |
894b8788 CL |
1477 | void **object; |
1478 | ||
7c2e132c CL |
1479 | tail = 0; /* Hot objects. Put the slab first */ |
1480 | ||
894b8788 | 1481 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1482 | object = c->freelist; |
b3fba8da | 1483 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1484 | |
1485 | /* And put onto the regular freelist */ | |
b3fba8da | 1486 | object[c->offset] = page->freelist; |
894b8788 CL |
1487 | page->freelist = object; |
1488 | page->inuse--; | |
1489 | } | |
dfb4f096 | 1490 | c->page = NULL; |
7c2e132c | 1491 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1492 | } |
1493 | ||
dfb4f096 | 1494 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1495 | { |
8ff12cfc | 1496 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1497 | slab_lock(c->page); |
1498 | deactivate_slab(s, c); | |
81819f0f CL |
1499 | } |
1500 | ||
1501 | /* | |
1502 | * Flush cpu slab. | |
6446faa2 | 1503 | * |
81819f0f CL |
1504 | * Called from IPI handler with interrupts disabled. |
1505 | */ | |
0c710013 | 1506 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1507 | { |
dfb4f096 | 1508 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1509 | |
dfb4f096 CL |
1510 | if (likely(c && c->page)) |
1511 | flush_slab(s, c); | |
81819f0f CL |
1512 | } |
1513 | ||
1514 | static void flush_cpu_slab(void *d) | |
1515 | { | |
1516 | struct kmem_cache *s = d; | |
81819f0f | 1517 | |
dfb4f096 | 1518 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1519 | } |
1520 | ||
1521 | static void flush_all(struct kmem_cache *s) | |
1522 | { | |
15c8b6c1 | 1523 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1524 | } |
1525 | ||
dfb4f096 CL |
1526 | /* |
1527 | * Check if the objects in a per cpu structure fit numa | |
1528 | * locality expectations. | |
1529 | */ | |
1530 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1531 | { | |
1532 | #ifdef CONFIG_NUMA | |
1533 | if (node != -1 && c->node != node) | |
1534 | return 0; | |
1535 | #endif | |
1536 | return 1; | |
1537 | } | |
1538 | ||
781b2ba6 PE |
1539 | static int count_free(struct page *page) |
1540 | { | |
1541 | return page->objects - page->inuse; | |
1542 | } | |
1543 | ||
1544 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1545 | int (*get_count)(struct page *)) | |
1546 | { | |
1547 | unsigned long flags; | |
1548 | unsigned long x = 0; | |
1549 | struct page *page; | |
1550 | ||
1551 | spin_lock_irqsave(&n->list_lock, flags); | |
1552 | list_for_each_entry(page, &n->partial, lru) | |
1553 | x += get_count(page); | |
1554 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1555 | return x; | |
1556 | } | |
1557 | ||
26c02cf0 AB |
1558 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1559 | { | |
1560 | #ifdef CONFIG_SLUB_DEBUG | |
1561 | return atomic_long_read(&n->total_objects); | |
1562 | #else | |
1563 | return 0; | |
1564 | #endif | |
1565 | } | |
1566 | ||
781b2ba6 PE |
1567 | static noinline void |
1568 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1569 | { | |
1570 | int node; | |
1571 | ||
1572 | printk(KERN_WARNING | |
1573 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1574 | nid, gfpflags); | |
1575 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1576 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1577 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1578 | ||
fa5ec8a1 DR |
1579 | if (oo_order(s->min) > get_order(s->objsize)) |
1580 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1581 | "slub_debug=O to disable.\n", s->name); | |
1582 | ||
781b2ba6 PE |
1583 | for_each_online_node(node) { |
1584 | struct kmem_cache_node *n = get_node(s, node); | |
1585 | unsigned long nr_slabs; | |
1586 | unsigned long nr_objs; | |
1587 | unsigned long nr_free; | |
1588 | ||
1589 | if (!n) | |
1590 | continue; | |
1591 | ||
26c02cf0 AB |
1592 | nr_free = count_partial(n, count_free); |
1593 | nr_slabs = node_nr_slabs(n); | |
1594 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1595 | |
1596 | printk(KERN_WARNING | |
1597 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1598 | node, nr_slabs, nr_objs, nr_free); | |
1599 | } | |
1600 | } | |
1601 | ||
81819f0f | 1602 | /* |
894b8788 CL |
1603 | * Slow path. The lockless freelist is empty or we need to perform |
1604 | * debugging duties. | |
1605 | * | |
1606 | * Interrupts are disabled. | |
81819f0f | 1607 | * |
894b8788 CL |
1608 | * Processing is still very fast if new objects have been freed to the |
1609 | * regular freelist. In that case we simply take over the regular freelist | |
1610 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1611 | * |
894b8788 CL |
1612 | * If that is not working then we fall back to the partial lists. We take the |
1613 | * first element of the freelist as the object to allocate now and move the | |
1614 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1615 | * |
894b8788 | 1616 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1617 | * we need to allocate a new slab. This is the slowest path since it involves |
1618 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1619 | */ |
ce71e27c EGM |
1620 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1621 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1622 | { |
81819f0f | 1623 | void **object; |
dfb4f096 | 1624 | struct page *new; |
81819f0f | 1625 | |
e72e9c23 LT |
1626 | /* We handle __GFP_ZERO in the caller */ |
1627 | gfpflags &= ~__GFP_ZERO; | |
1628 | ||
dfb4f096 | 1629 | if (!c->page) |
81819f0f CL |
1630 | goto new_slab; |
1631 | ||
dfb4f096 CL |
1632 | slab_lock(c->page); |
1633 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1634 | goto another_slab; |
6446faa2 | 1635 | |
8ff12cfc | 1636 | stat(c, ALLOC_REFILL); |
6446faa2 | 1637 | |
894b8788 | 1638 | load_freelist: |
dfb4f096 | 1639 | object = c->page->freelist; |
a973e9dd | 1640 | if (unlikely(!object)) |
81819f0f | 1641 | goto another_slab; |
8a38082d | 1642 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1643 | goto debug; |
1644 | ||
b3fba8da | 1645 | c->freelist = object[c->offset]; |
39b26464 | 1646 | c->page->inuse = c->page->objects; |
a973e9dd | 1647 | c->page->freelist = NULL; |
dfb4f096 | 1648 | c->node = page_to_nid(c->page); |
1f84260c | 1649 | unlock_out: |
dfb4f096 | 1650 | slab_unlock(c->page); |
8ff12cfc | 1651 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1652 | return object; |
1653 | ||
1654 | another_slab: | |
dfb4f096 | 1655 | deactivate_slab(s, c); |
81819f0f CL |
1656 | |
1657 | new_slab: | |
dfb4f096 CL |
1658 | new = get_partial(s, gfpflags, node); |
1659 | if (new) { | |
1660 | c->page = new; | |
8ff12cfc | 1661 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1662 | goto load_freelist; |
81819f0f CL |
1663 | } |
1664 | ||
b811c202 CL |
1665 | if (gfpflags & __GFP_WAIT) |
1666 | local_irq_enable(); | |
1667 | ||
dfb4f096 | 1668 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1669 | |
1670 | if (gfpflags & __GFP_WAIT) | |
1671 | local_irq_disable(); | |
1672 | ||
dfb4f096 CL |
1673 | if (new) { |
1674 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1675 | stat(c, ALLOC_SLAB); |
05aa3450 | 1676 | if (c->page) |
dfb4f096 | 1677 | flush_slab(s, c); |
dfb4f096 | 1678 | slab_lock(new); |
8a38082d | 1679 | __SetPageSlubFrozen(new); |
dfb4f096 | 1680 | c->page = new; |
4b6f0750 | 1681 | goto load_freelist; |
81819f0f | 1682 | } |
95f85989 PE |
1683 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1684 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1685 | return NULL; |
81819f0f | 1686 | debug: |
dfb4f096 | 1687 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1688 | goto another_slab; |
894b8788 | 1689 | |
dfb4f096 | 1690 | c->page->inuse++; |
b3fba8da | 1691 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1692 | c->node = -1; |
1f84260c | 1693 | goto unlock_out; |
894b8788 CL |
1694 | } |
1695 | ||
1696 | /* | |
1697 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1698 | * have the fastpath folded into their functions. So no function call | |
1699 | * overhead for requests that can be satisfied on the fastpath. | |
1700 | * | |
1701 | * The fastpath works by first checking if the lockless freelist can be used. | |
1702 | * If not then __slab_alloc is called for slow processing. | |
1703 | * | |
1704 | * Otherwise we can simply pick the next object from the lockless free list. | |
1705 | */ | |
06428780 | 1706 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1707 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1708 | { |
894b8788 | 1709 | void **object; |
dfb4f096 | 1710 | struct kmem_cache_cpu *c; |
1f84260c | 1711 | unsigned long flags; |
bdb21928 | 1712 | unsigned int objsize; |
1f84260c | 1713 | |
dcce284a | 1714 | gfpflags &= gfp_allowed_mask; |
7e85ee0c | 1715 | |
cf40bd16 | 1716 | lockdep_trace_alloc(gfpflags); |
89124d70 | 1717 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1718 | |
773ff60e AM |
1719 | if (should_failslab(s->objsize, gfpflags)) |
1720 | return NULL; | |
1f84260c | 1721 | |
894b8788 | 1722 | local_irq_save(flags); |
dfb4f096 | 1723 | c = get_cpu_slab(s, smp_processor_id()); |
bdb21928 | 1724 | objsize = c->objsize; |
a973e9dd | 1725 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1726 | |
dfb4f096 | 1727 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1728 | |
1729 | else { | |
dfb4f096 | 1730 | object = c->freelist; |
b3fba8da | 1731 | c->freelist = object[c->offset]; |
8ff12cfc | 1732 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1733 | } |
1734 | local_irq_restore(flags); | |
d07dbea4 CL |
1735 | |
1736 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
bdb21928 | 1737 | memset(object, 0, objsize); |
d07dbea4 | 1738 | |
5a896d9e | 1739 | kmemcheck_slab_alloc(s, gfpflags, object, c->objsize); |
06f22f13 | 1740 | kmemleak_alloc_recursive(object, objsize, 1, s->flags, gfpflags); |
5a896d9e | 1741 | |
894b8788 | 1742 | return object; |
81819f0f CL |
1743 | } |
1744 | ||
1745 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1746 | { | |
5b882be4 EGM |
1747 | void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_); |
1748 | ||
ca2b84cb | 1749 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1750 | |
1751 | return ret; | |
81819f0f CL |
1752 | } |
1753 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1754 | ||
5b882be4 EGM |
1755 | #ifdef CONFIG_KMEMTRACE |
1756 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) | |
1757 | { | |
1758 | return slab_alloc(s, gfpflags, -1, _RET_IP_); | |
1759 | } | |
1760 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1761 | #endif | |
1762 | ||
81819f0f CL |
1763 | #ifdef CONFIG_NUMA |
1764 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1765 | { | |
5b882be4 EGM |
1766 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1767 | ||
ca2b84cb EGM |
1768 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1769 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1770 | |
1771 | return ret; | |
81819f0f CL |
1772 | } |
1773 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1774 | #endif | |
1775 | ||
5b882be4 EGM |
1776 | #ifdef CONFIG_KMEMTRACE |
1777 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, | |
1778 | gfp_t gfpflags, | |
1779 | int node) | |
1780 | { | |
1781 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1782 | } | |
1783 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1784 | #endif | |
1785 | ||
81819f0f | 1786 | /* |
894b8788 CL |
1787 | * Slow patch handling. This may still be called frequently since objects |
1788 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1789 | * |
894b8788 CL |
1790 | * So we still attempt to reduce cache line usage. Just take the slab |
1791 | * lock and free the item. If there is no additional partial page | |
1792 | * handling required then we can return immediately. | |
81819f0f | 1793 | */ |
894b8788 | 1794 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ce71e27c | 1795 | void *x, unsigned long addr, unsigned int offset) |
81819f0f CL |
1796 | { |
1797 | void *prior; | |
1798 | void **object = (void *)x; | |
8ff12cfc | 1799 | struct kmem_cache_cpu *c; |
81819f0f | 1800 | |
8ff12cfc CL |
1801 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1802 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1803 | slab_lock(page); |
1804 | ||
8a38082d | 1805 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1806 | goto debug; |
6446faa2 | 1807 | |
81819f0f | 1808 | checks_ok: |
b3fba8da | 1809 | prior = object[offset] = page->freelist; |
81819f0f CL |
1810 | page->freelist = object; |
1811 | page->inuse--; | |
1812 | ||
8a38082d | 1813 | if (unlikely(PageSlubFrozen(page))) { |
8ff12cfc | 1814 | stat(c, FREE_FROZEN); |
81819f0f | 1815 | goto out_unlock; |
8ff12cfc | 1816 | } |
81819f0f CL |
1817 | |
1818 | if (unlikely(!page->inuse)) | |
1819 | goto slab_empty; | |
1820 | ||
1821 | /* | |
6446faa2 | 1822 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1823 | * then add it. |
1824 | */ | |
a973e9dd | 1825 | if (unlikely(!prior)) { |
7c2e132c | 1826 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1827 | stat(c, FREE_ADD_PARTIAL); |
1828 | } | |
81819f0f CL |
1829 | |
1830 | out_unlock: | |
1831 | slab_unlock(page); | |
81819f0f CL |
1832 | return; |
1833 | ||
1834 | slab_empty: | |
a973e9dd | 1835 | if (prior) { |
81819f0f | 1836 | /* |
672bba3a | 1837 | * Slab still on the partial list. |
81819f0f CL |
1838 | */ |
1839 | remove_partial(s, page); | |
8ff12cfc CL |
1840 | stat(c, FREE_REMOVE_PARTIAL); |
1841 | } | |
81819f0f | 1842 | slab_unlock(page); |
8ff12cfc | 1843 | stat(c, FREE_SLAB); |
81819f0f | 1844 | discard_slab(s, page); |
81819f0f CL |
1845 | return; |
1846 | ||
1847 | debug: | |
3ec09742 | 1848 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1849 | goto out_unlock; |
77c5e2d0 | 1850 | goto checks_ok; |
81819f0f CL |
1851 | } |
1852 | ||
894b8788 CL |
1853 | /* |
1854 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1855 | * can perform fastpath freeing without additional function calls. | |
1856 | * | |
1857 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1858 | * of this processor. This typically the case if we have just allocated | |
1859 | * the item before. | |
1860 | * | |
1861 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1862 | * with all sorts of special processing. | |
1863 | */ | |
06428780 | 1864 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1865 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1866 | { |
1867 | void **object = (void *)x; | |
dfb4f096 | 1868 | struct kmem_cache_cpu *c; |
1f84260c CL |
1869 | unsigned long flags; |
1870 | ||
06f22f13 | 1871 | kmemleak_free_recursive(x, s->flags); |
894b8788 | 1872 | local_irq_save(flags); |
dfb4f096 | 1873 | c = get_cpu_slab(s, smp_processor_id()); |
5a896d9e | 1874 | kmemcheck_slab_free(s, object, c->objsize); |
27d9e4e9 | 1875 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a | 1876 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
6047a007 | 1877 | debug_check_no_obj_freed(object, c->objsize); |
ee3c72a1 | 1878 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1879 | object[c->offset] = c->freelist; |
dfb4f096 | 1880 | c->freelist = object; |
8ff12cfc | 1881 | stat(c, FREE_FASTPATH); |
894b8788 | 1882 | } else |
b3fba8da | 1883 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1884 | |
1885 | local_irq_restore(flags); | |
1886 | } | |
1887 | ||
81819f0f CL |
1888 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1889 | { | |
77c5e2d0 | 1890 | struct page *page; |
81819f0f | 1891 | |
b49af68f | 1892 | page = virt_to_head_page(x); |
81819f0f | 1893 | |
ce71e27c | 1894 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1895 | |
ca2b84cb | 1896 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1897 | } |
1898 | EXPORT_SYMBOL(kmem_cache_free); | |
1899 | ||
e9beef18 | 1900 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1901 | static struct page *get_object_page(const void *x) |
1902 | { | |
b49af68f | 1903 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1904 | |
1905 | if (!PageSlab(page)) | |
1906 | return NULL; | |
1907 | ||
1908 | return page; | |
1909 | } | |
1910 | ||
1911 | /* | |
672bba3a CL |
1912 | * Object placement in a slab is made very easy because we always start at |
1913 | * offset 0. If we tune the size of the object to the alignment then we can | |
1914 | * get the required alignment by putting one properly sized object after | |
1915 | * another. | |
81819f0f CL |
1916 | * |
1917 | * Notice that the allocation order determines the sizes of the per cpu | |
1918 | * caches. Each processor has always one slab available for allocations. | |
1919 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1920 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1921 | * locking overhead. |
81819f0f CL |
1922 | */ |
1923 | ||
1924 | /* | |
1925 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1926 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1927 | * and increases the number of allocations possible without having to | |
1928 | * take the list_lock. | |
1929 | */ | |
1930 | static int slub_min_order; | |
114e9e89 | 1931 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1932 | static int slub_min_objects; |
81819f0f CL |
1933 | |
1934 | /* | |
1935 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1936 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1937 | */ |
1938 | static int slub_nomerge; | |
1939 | ||
81819f0f CL |
1940 | /* |
1941 | * Calculate the order of allocation given an slab object size. | |
1942 | * | |
672bba3a CL |
1943 | * The order of allocation has significant impact on performance and other |
1944 | * system components. Generally order 0 allocations should be preferred since | |
1945 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1946 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1947 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1948 | * would be wasted. |
1949 | * | |
1950 | * In order to reach satisfactory performance we must ensure that a minimum | |
1951 | * number of objects is in one slab. Otherwise we may generate too much | |
1952 | * activity on the partial lists which requires taking the list_lock. This is | |
1953 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1954 | * |
672bba3a CL |
1955 | * slub_max_order specifies the order where we begin to stop considering the |
1956 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1957 | * we try to keep the page order as low as possible. So we accept more waste | |
1958 | * of space in favor of a small page order. | |
81819f0f | 1959 | * |
672bba3a CL |
1960 | * Higher order allocations also allow the placement of more objects in a |
1961 | * slab and thereby reduce object handling overhead. If the user has | |
1962 | * requested a higher mininum order then we start with that one instead of | |
1963 | * the smallest order which will fit the object. | |
81819f0f | 1964 | */ |
5e6d444e CL |
1965 | static inline int slab_order(int size, int min_objects, |
1966 | int max_order, int fract_leftover) | |
81819f0f CL |
1967 | { |
1968 | int order; | |
1969 | int rem; | |
6300ea75 | 1970 | int min_order = slub_min_order; |
81819f0f | 1971 | |
210b5c06 CG |
1972 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1973 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1974 | |
6300ea75 | 1975 | for (order = max(min_order, |
5e6d444e CL |
1976 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1977 | order <= max_order; order++) { | |
81819f0f | 1978 | |
5e6d444e | 1979 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1980 | |
5e6d444e | 1981 | if (slab_size < min_objects * size) |
81819f0f CL |
1982 | continue; |
1983 | ||
1984 | rem = slab_size % size; | |
1985 | ||
5e6d444e | 1986 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1987 | break; |
1988 | ||
1989 | } | |
672bba3a | 1990 | |
81819f0f CL |
1991 | return order; |
1992 | } | |
1993 | ||
5e6d444e CL |
1994 | static inline int calculate_order(int size) |
1995 | { | |
1996 | int order; | |
1997 | int min_objects; | |
1998 | int fraction; | |
e8120ff1 | 1999 | int max_objects; |
5e6d444e CL |
2000 | |
2001 | /* | |
2002 | * Attempt to find best configuration for a slab. This | |
2003 | * works by first attempting to generate a layout with | |
2004 | * the best configuration and backing off gradually. | |
2005 | * | |
2006 | * First we reduce the acceptable waste in a slab. Then | |
2007 | * we reduce the minimum objects required in a slab. | |
2008 | */ | |
2009 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2010 | if (!min_objects) |
2011 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
2012 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
2013 | min_objects = min(min_objects, max_objects); | |
2014 | ||
5e6d444e | 2015 | while (min_objects > 1) { |
c124f5b5 | 2016 | fraction = 16; |
5e6d444e CL |
2017 | while (fraction >= 4) { |
2018 | order = slab_order(size, min_objects, | |
2019 | slub_max_order, fraction); | |
2020 | if (order <= slub_max_order) | |
2021 | return order; | |
2022 | fraction /= 2; | |
2023 | } | |
5086c389 | 2024 | min_objects--; |
5e6d444e CL |
2025 | } |
2026 | ||
2027 | /* | |
2028 | * We were unable to place multiple objects in a slab. Now | |
2029 | * lets see if we can place a single object there. | |
2030 | */ | |
2031 | order = slab_order(size, 1, slub_max_order, 1); | |
2032 | if (order <= slub_max_order) | |
2033 | return order; | |
2034 | ||
2035 | /* | |
2036 | * Doh this slab cannot be placed using slub_max_order. | |
2037 | */ | |
2038 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2039 | if (order < MAX_ORDER) |
5e6d444e CL |
2040 | return order; |
2041 | return -ENOSYS; | |
2042 | } | |
2043 | ||
81819f0f | 2044 | /* |
672bba3a | 2045 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2046 | */ |
2047 | static unsigned long calculate_alignment(unsigned long flags, | |
2048 | unsigned long align, unsigned long size) | |
2049 | { | |
2050 | /* | |
6446faa2 CL |
2051 | * If the user wants hardware cache aligned objects then follow that |
2052 | * suggestion if the object is sufficiently large. | |
81819f0f | 2053 | * |
6446faa2 CL |
2054 | * The hardware cache alignment cannot override the specified |
2055 | * alignment though. If that is greater then use it. | |
81819f0f | 2056 | */ |
b6210386 NP |
2057 | if (flags & SLAB_HWCACHE_ALIGN) { |
2058 | unsigned long ralign = cache_line_size(); | |
2059 | while (size <= ralign / 2) | |
2060 | ralign /= 2; | |
2061 | align = max(align, ralign); | |
2062 | } | |
81819f0f CL |
2063 | |
2064 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2065 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2066 | |
2067 | return ALIGN(align, sizeof(void *)); | |
2068 | } | |
2069 | ||
dfb4f096 CL |
2070 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
2071 | struct kmem_cache_cpu *c) | |
2072 | { | |
2073 | c->page = NULL; | |
a973e9dd | 2074 | c->freelist = NULL; |
dfb4f096 | 2075 | c->node = 0; |
42a9fdbb CL |
2076 | c->offset = s->offset / sizeof(void *); |
2077 | c->objsize = s->objsize; | |
62f75532 PE |
2078 | #ifdef CONFIG_SLUB_STATS |
2079 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
2080 | #endif | |
dfb4f096 CL |
2081 | } |
2082 | ||
5595cffc PE |
2083 | static void |
2084 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2085 | { |
2086 | n->nr_partial = 0; | |
81819f0f CL |
2087 | spin_lock_init(&n->list_lock); |
2088 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2089 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2090 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2091 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2092 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2093 | #endif |
81819f0f CL |
2094 | } |
2095 | ||
4c93c355 CL |
2096 | #ifdef CONFIG_SMP |
2097 | /* | |
2098 | * Per cpu array for per cpu structures. | |
2099 | * | |
2100 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
2101 | * close together meaning that it becomes possible that multiple per cpu | |
2102 | * structures are contained in one cacheline. This may be particularly | |
2103 | * beneficial for the kmalloc caches. | |
2104 | * | |
2105 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
2106 | * likely able to get per cpu structures for all caches from the array defined | |
2107 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2108 | * | |
2109 | * If the per cpu array is exhausted then fall back to kmalloc | |
2110 | * of individual cachelines. No sharing is possible then. | |
2111 | */ | |
2112 | #define NR_KMEM_CACHE_CPU 100 | |
2113 | ||
2114 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2115 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2116 | ||
2117 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
174596a0 | 2118 | static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS); |
4c93c355 CL |
2119 | |
2120 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2121 | int cpu, gfp_t flags) | |
2122 | { | |
2123 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2124 | ||
2125 | if (c) | |
2126 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2127 | (void *)c->freelist; | |
2128 | else { | |
2129 | /* Table overflow: So allocate ourselves */ | |
2130 | c = kmalloc_node( | |
2131 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2132 | flags, cpu_to_node(cpu)); | |
2133 | if (!c) | |
2134 | return NULL; | |
2135 | } | |
2136 | ||
2137 | init_kmem_cache_cpu(s, c); | |
2138 | return c; | |
2139 | } | |
2140 | ||
2141 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2142 | { | |
2143 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
37189094 | 2144 | c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { |
4c93c355 CL |
2145 | kfree(c); |
2146 | return; | |
2147 | } | |
2148 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2149 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2150 | } | |
2151 | ||
2152 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2153 | { | |
2154 | int cpu; | |
2155 | ||
2156 | for_each_online_cpu(cpu) { | |
2157 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2158 | ||
2159 | if (c) { | |
2160 | s->cpu_slab[cpu] = NULL; | |
2161 | free_kmem_cache_cpu(c, cpu); | |
2162 | } | |
2163 | } | |
2164 | } | |
2165 | ||
2166 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2167 | { | |
2168 | int cpu; | |
2169 | ||
2170 | for_each_online_cpu(cpu) { | |
2171 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2172 | ||
2173 | if (c) | |
2174 | continue; | |
2175 | ||
2176 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2177 | if (!c) { | |
2178 | free_kmem_cache_cpus(s); | |
2179 | return 0; | |
2180 | } | |
2181 | s->cpu_slab[cpu] = c; | |
2182 | } | |
2183 | return 1; | |
2184 | } | |
2185 | ||
2186 | /* | |
2187 | * Initialize the per cpu array. | |
2188 | */ | |
2189 | static void init_alloc_cpu_cpu(int cpu) | |
2190 | { | |
2191 | int i; | |
2192 | ||
174596a0 | 2193 | if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once))) |
4c93c355 CL |
2194 | return; |
2195 | ||
2196 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2197 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2198 | ||
174596a0 | 2199 | cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)); |
4c93c355 CL |
2200 | } |
2201 | ||
2202 | static void __init init_alloc_cpu(void) | |
2203 | { | |
2204 | int cpu; | |
2205 | ||
2206 | for_each_online_cpu(cpu) | |
2207 | init_alloc_cpu_cpu(cpu); | |
2208 | } | |
2209 | ||
2210 | #else | |
2211 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2212 | static inline void init_alloc_cpu(void) {} | |
2213 | ||
2214 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2215 | { | |
2216 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2217 | return 1; | |
2218 | } | |
2219 | #endif | |
2220 | ||
81819f0f CL |
2221 | #ifdef CONFIG_NUMA |
2222 | /* | |
2223 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2224 | * slab on the node for this slabcache. There are no concurrent accesses | |
2225 | * possible. | |
2226 | * | |
2227 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2228 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2229 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2230 | */ |
0094de92 | 2231 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2232 | { |
2233 | struct page *page; | |
2234 | struct kmem_cache_node *n; | |
ba84c73c | 2235 | unsigned long flags; |
81819f0f CL |
2236 | |
2237 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2238 | ||
a2f92ee7 | 2239 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2240 | |
2241 | BUG_ON(!page); | |
a2f92ee7 CL |
2242 | if (page_to_nid(page) != node) { |
2243 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2244 | "node %d\n", node); | |
2245 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2246 | "in order to be able to continue\n"); | |
2247 | } | |
2248 | ||
81819f0f CL |
2249 | n = page->freelist; |
2250 | BUG_ON(!n); | |
2251 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2252 | page->inuse++; | |
2253 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2254 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2255 | init_object(kmalloc_caches, n, 1); |
2256 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2257 | #endif |
5595cffc | 2258 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2259 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2260 | |
ba84c73c | 2261 | /* |
2262 | * lockdep requires consistent irq usage for each lock | |
2263 | * so even though there cannot be a race this early in | |
2264 | * the boot sequence, we still disable irqs. | |
2265 | */ | |
2266 | local_irq_save(flags); | |
7c2e132c | 2267 | add_partial(n, page, 0); |
ba84c73c | 2268 | local_irq_restore(flags); |
81819f0f CL |
2269 | } |
2270 | ||
2271 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2272 | { | |
2273 | int node; | |
2274 | ||
f64dc58c | 2275 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2276 | struct kmem_cache_node *n = s->node[node]; |
2277 | if (n && n != &s->local_node) | |
2278 | kmem_cache_free(kmalloc_caches, n); | |
2279 | s->node[node] = NULL; | |
2280 | } | |
2281 | } | |
2282 | ||
2283 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2284 | { | |
2285 | int node; | |
2286 | int local_node; | |
2287 | ||
2288 | if (slab_state >= UP) | |
2289 | local_node = page_to_nid(virt_to_page(s)); | |
2290 | else | |
2291 | local_node = 0; | |
2292 | ||
f64dc58c | 2293 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2294 | struct kmem_cache_node *n; |
2295 | ||
2296 | if (local_node == node) | |
2297 | n = &s->local_node; | |
2298 | else { | |
2299 | if (slab_state == DOWN) { | |
0094de92 | 2300 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2301 | continue; |
2302 | } | |
2303 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2304 | gfpflags, node); | |
2305 | ||
2306 | if (!n) { | |
2307 | free_kmem_cache_nodes(s); | |
2308 | return 0; | |
2309 | } | |
2310 | ||
2311 | } | |
2312 | s->node[node] = n; | |
5595cffc | 2313 | init_kmem_cache_node(n, s); |
81819f0f CL |
2314 | } |
2315 | return 1; | |
2316 | } | |
2317 | #else | |
2318 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2319 | { | |
2320 | } | |
2321 | ||
2322 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2323 | { | |
5595cffc | 2324 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2325 | return 1; |
2326 | } | |
2327 | #endif | |
2328 | ||
c0bdb232 | 2329 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2330 | { |
2331 | if (min < MIN_PARTIAL) | |
2332 | min = MIN_PARTIAL; | |
2333 | else if (min > MAX_PARTIAL) | |
2334 | min = MAX_PARTIAL; | |
2335 | s->min_partial = min; | |
2336 | } | |
2337 | ||
81819f0f CL |
2338 | /* |
2339 | * calculate_sizes() determines the order and the distribution of data within | |
2340 | * a slab object. | |
2341 | */ | |
06b285dc | 2342 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2343 | { |
2344 | unsigned long flags = s->flags; | |
2345 | unsigned long size = s->objsize; | |
2346 | unsigned long align = s->align; | |
834f3d11 | 2347 | int order; |
81819f0f | 2348 | |
d8b42bf5 CL |
2349 | /* |
2350 | * Round up object size to the next word boundary. We can only | |
2351 | * place the free pointer at word boundaries and this determines | |
2352 | * the possible location of the free pointer. | |
2353 | */ | |
2354 | size = ALIGN(size, sizeof(void *)); | |
2355 | ||
2356 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2357 | /* |
2358 | * Determine if we can poison the object itself. If the user of | |
2359 | * the slab may touch the object after free or before allocation | |
2360 | * then we should never poison the object itself. | |
2361 | */ | |
2362 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2363 | !s->ctor) |
81819f0f CL |
2364 | s->flags |= __OBJECT_POISON; |
2365 | else | |
2366 | s->flags &= ~__OBJECT_POISON; | |
2367 | ||
81819f0f CL |
2368 | |
2369 | /* | |
672bba3a | 2370 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2371 | * end of the object and the free pointer. If not then add an |
672bba3a | 2372 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2373 | */ |
2374 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2375 | size += sizeof(void *); | |
41ecc55b | 2376 | #endif |
81819f0f CL |
2377 | |
2378 | /* | |
672bba3a CL |
2379 | * With that we have determined the number of bytes in actual use |
2380 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2381 | */ |
2382 | s->inuse = size; | |
2383 | ||
2384 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2385 | s->ctor)) { |
81819f0f CL |
2386 | /* |
2387 | * Relocate free pointer after the object if it is not | |
2388 | * permitted to overwrite the first word of the object on | |
2389 | * kmem_cache_free. | |
2390 | * | |
2391 | * This is the case if we do RCU, have a constructor or | |
2392 | * destructor or are poisoning the objects. | |
2393 | */ | |
2394 | s->offset = size; | |
2395 | size += sizeof(void *); | |
2396 | } | |
2397 | ||
c12b3c62 | 2398 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2399 | if (flags & SLAB_STORE_USER) |
2400 | /* | |
2401 | * Need to store information about allocs and frees after | |
2402 | * the object. | |
2403 | */ | |
2404 | size += 2 * sizeof(struct track); | |
2405 | ||
be7b3fbc | 2406 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2407 | /* |
2408 | * Add some empty padding so that we can catch | |
2409 | * overwrites from earlier objects rather than let | |
2410 | * tracking information or the free pointer be | |
0211a9c8 | 2411 | * corrupted if a user writes before the start |
81819f0f CL |
2412 | * of the object. |
2413 | */ | |
2414 | size += sizeof(void *); | |
41ecc55b | 2415 | #endif |
672bba3a | 2416 | |
81819f0f CL |
2417 | /* |
2418 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2419 | * user specified and the dynamic determination of cache line size |
2420 | * on bootup. | |
81819f0f CL |
2421 | */ |
2422 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2423 | s->align = align; |
81819f0f CL |
2424 | |
2425 | /* | |
2426 | * SLUB stores one object immediately after another beginning from | |
2427 | * offset 0. In order to align the objects we have to simply size | |
2428 | * each object to conform to the alignment. | |
2429 | */ | |
2430 | size = ALIGN(size, align); | |
2431 | s->size = size; | |
06b285dc CL |
2432 | if (forced_order >= 0) |
2433 | order = forced_order; | |
2434 | else | |
2435 | order = calculate_order(size); | |
81819f0f | 2436 | |
834f3d11 | 2437 | if (order < 0) |
81819f0f CL |
2438 | return 0; |
2439 | ||
b7a49f0d | 2440 | s->allocflags = 0; |
834f3d11 | 2441 | if (order) |
b7a49f0d CL |
2442 | s->allocflags |= __GFP_COMP; |
2443 | ||
2444 | if (s->flags & SLAB_CACHE_DMA) | |
2445 | s->allocflags |= SLUB_DMA; | |
2446 | ||
2447 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2448 | s->allocflags |= __GFP_RECLAIMABLE; | |
2449 | ||
81819f0f CL |
2450 | /* |
2451 | * Determine the number of objects per slab | |
2452 | */ | |
834f3d11 | 2453 | s->oo = oo_make(order, size); |
65c3376a | 2454 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2455 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2456 | s->max = s->oo; | |
81819f0f | 2457 | |
834f3d11 | 2458 | return !!oo_objects(s->oo); |
81819f0f CL |
2459 | |
2460 | } | |
2461 | ||
81819f0f CL |
2462 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2463 | const char *name, size_t size, | |
2464 | size_t align, unsigned long flags, | |
51cc5068 | 2465 | void (*ctor)(void *)) |
81819f0f CL |
2466 | { |
2467 | memset(s, 0, kmem_size); | |
2468 | s->name = name; | |
2469 | s->ctor = ctor; | |
81819f0f | 2470 | s->objsize = size; |
81819f0f | 2471 | s->align = align; |
ba0268a8 | 2472 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2473 | |
06b285dc | 2474 | if (!calculate_sizes(s, -1)) |
81819f0f | 2475 | goto error; |
3de47213 DR |
2476 | if (disable_higher_order_debug) { |
2477 | /* | |
2478 | * Disable debugging flags that store metadata if the min slab | |
2479 | * order increased. | |
2480 | */ | |
2481 | if (get_order(s->size) > get_order(s->objsize)) { | |
2482 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2483 | s->offset = 0; | |
2484 | if (!calculate_sizes(s, -1)) | |
2485 | goto error; | |
2486 | } | |
2487 | } | |
81819f0f | 2488 | |
3b89d7d8 DR |
2489 | /* |
2490 | * The larger the object size is, the more pages we want on the partial | |
2491 | * list to avoid pounding the page allocator excessively. | |
2492 | */ | |
c0bdb232 | 2493 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2494 | s->refcount = 1; |
2495 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2496 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2497 | #endif |
dfb4f096 CL |
2498 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2499 | goto error; | |
81819f0f | 2500 | |
dfb4f096 | 2501 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2502 | return 1; |
4c93c355 | 2503 | free_kmem_cache_nodes(s); |
81819f0f CL |
2504 | error: |
2505 | if (flags & SLAB_PANIC) | |
2506 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2507 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2508 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2509 | s->offset, flags); |
2510 | return 0; | |
2511 | } | |
81819f0f CL |
2512 | |
2513 | /* | |
2514 | * Check if a given pointer is valid | |
2515 | */ | |
2516 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2517 | { | |
06428780 | 2518 | struct page *page; |
81819f0f CL |
2519 | |
2520 | page = get_object_page(object); | |
2521 | ||
2522 | if (!page || s != page->slab) | |
2523 | /* No slab or wrong slab */ | |
2524 | return 0; | |
2525 | ||
abcd08a6 | 2526 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2527 | return 0; |
2528 | ||
2529 | /* | |
2530 | * We could also check if the object is on the slabs freelist. | |
2531 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2532 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2533 | * to a certain slab. |
2534 | */ | |
2535 | return 1; | |
2536 | } | |
2537 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2538 | ||
2539 | /* | |
2540 | * Determine the size of a slab object | |
2541 | */ | |
2542 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2543 | { | |
2544 | return s->objsize; | |
2545 | } | |
2546 | EXPORT_SYMBOL(kmem_cache_size); | |
2547 | ||
2548 | const char *kmem_cache_name(struct kmem_cache *s) | |
2549 | { | |
2550 | return s->name; | |
2551 | } | |
2552 | EXPORT_SYMBOL(kmem_cache_name); | |
2553 | ||
33b12c38 CL |
2554 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2555 | const char *text) | |
2556 | { | |
2557 | #ifdef CONFIG_SLUB_DEBUG | |
2558 | void *addr = page_address(page); | |
2559 | void *p; | |
2560 | DECLARE_BITMAP(map, page->objects); | |
2561 | ||
2562 | bitmap_zero(map, page->objects); | |
2563 | slab_err(s, page, "%s", text); | |
2564 | slab_lock(page); | |
2565 | for_each_free_object(p, s, page->freelist) | |
2566 | set_bit(slab_index(p, s, addr), map); | |
2567 | ||
2568 | for_each_object(p, s, addr, page->objects) { | |
2569 | ||
2570 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2571 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2572 | p, p - addr); | |
2573 | print_tracking(s, p); | |
2574 | } | |
2575 | } | |
2576 | slab_unlock(page); | |
2577 | #endif | |
2578 | } | |
2579 | ||
81819f0f | 2580 | /* |
599870b1 | 2581 | * Attempt to free all partial slabs on a node. |
81819f0f | 2582 | */ |
599870b1 | 2583 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2584 | { |
81819f0f CL |
2585 | unsigned long flags; |
2586 | struct page *page, *h; | |
2587 | ||
2588 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2589 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2590 | if (!page->inuse) { |
2591 | list_del(&page->lru); | |
2592 | discard_slab(s, page); | |
599870b1 | 2593 | n->nr_partial--; |
33b12c38 CL |
2594 | } else { |
2595 | list_slab_objects(s, page, | |
2596 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2597 | } |
33b12c38 | 2598 | } |
81819f0f | 2599 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2600 | } |
2601 | ||
2602 | /* | |
672bba3a | 2603 | * Release all resources used by a slab cache. |
81819f0f | 2604 | */ |
0c710013 | 2605 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2606 | { |
2607 | int node; | |
2608 | ||
2609 | flush_all(s); | |
2610 | ||
2611 | /* Attempt to free all objects */ | |
4c93c355 | 2612 | free_kmem_cache_cpus(s); |
f64dc58c | 2613 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2614 | struct kmem_cache_node *n = get_node(s, node); |
2615 | ||
599870b1 CL |
2616 | free_partial(s, n); |
2617 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2618 | return 1; |
2619 | } | |
2620 | free_kmem_cache_nodes(s); | |
2621 | return 0; | |
2622 | } | |
2623 | ||
2624 | /* | |
2625 | * Close a cache and release the kmem_cache structure | |
2626 | * (must be used for caches created using kmem_cache_create) | |
2627 | */ | |
2628 | void kmem_cache_destroy(struct kmem_cache *s) | |
2629 | { | |
2630 | down_write(&slub_lock); | |
2631 | s->refcount--; | |
2632 | if (!s->refcount) { | |
2633 | list_del(&s->list); | |
a0e1d1be | 2634 | up_write(&slub_lock); |
d629d819 PE |
2635 | if (kmem_cache_close(s)) { |
2636 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2637 | "still has objects.\n", s->name, __func__); | |
2638 | dump_stack(); | |
2639 | } | |
d76b1590 ED |
2640 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2641 | rcu_barrier(); | |
81819f0f | 2642 | sysfs_slab_remove(s); |
a0e1d1be CL |
2643 | } else |
2644 | up_write(&slub_lock); | |
81819f0f CL |
2645 | } |
2646 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2647 | ||
2648 | /******************************************************************** | |
2649 | * Kmalloc subsystem | |
2650 | *******************************************************************/ | |
2651 | ||
ffadd4d0 | 2652 | struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2653 | EXPORT_SYMBOL(kmalloc_caches); |
2654 | ||
81819f0f CL |
2655 | static int __init setup_slub_min_order(char *str) |
2656 | { | |
06428780 | 2657 | get_option(&str, &slub_min_order); |
81819f0f CL |
2658 | |
2659 | return 1; | |
2660 | } | |
2661 | ||
2662 | __setup("slub_min_order=", setup_slub_min_order); | |
2663 | ||
2664 | static int __init setup_slub_max_order(char *str) | |
2665 | { | |
06428780 | 2666 | get_option(&str, &slub_max_order); |
818cf590 | 2667 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2668 | |
2669 | return 1; | |
2670 | } | |
2671 | ||
2672 | __setup("slub_max_order=", setup_slub_max_order); | |
2673 | ||
2674 | static int __init setup_slub_min_objects(char *str) | |
2675 | { | |
06428780 | 2676 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2677 | |
2678 | return 1; | |
2679 | } | |
2680 | ||
2681 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2682 | ||
2683 | static int __init setup_slub_nomerge(char *str) | |
2684 | { | |
2685 | slub_nomerge = 1; | |
2686 | return 1; | |
2687 | } | |
2688 | ||
2689 | __setup("slub_nomerge", setup_slub_nomerge); | |
2690 | ||
81819f0f CL |
2691 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2692 | const char *name, int size, gfp_t gfp_flags) | |
2693 | { | |
2694 | unsigned int flags = 0; | |
2695 | ||
2696 | if (gfp_flags & SLUB_DMA) | |
2697 | flags = SLAB_CACHE_DMA; | |
2698 | ||
83b519e8 PE |
2699 | /* |
2700 | * This function is called with IRQs disabled during early-boot on | |
2701 | * single CPU so there's no need to take slub_lock here. | |
2702 | */ | |
81819f0f | 2703 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2704 | flags, NULL)) |
81819f0f CL |
2705 | goto panic; |
2706 | ||
2707 | list_add(&s->list, &slab_caches); | |
83b519e8 | 2708 | |
81819f0f CL |
2709 | if (sysfs_slab_add(s)) |
2710 | goto panic; | |
2711 | return s; | |
2712 | ||
2713 | panic: | |
2714 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2715 | } | |
2716 | ||
2e443fd0 | 2717 | #ifdef CONFIG_ZONE_DMA |
ffadd4d0 | 2718 | static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT]; |
1ceef402 CL |
2719 | |
2720 | static void sysfs_add_func(struct work_struct *w) | |
2721 | { | |
2722 | struct kmem_cache *s; | |
2723 | ||
2724 | down_write(&slub_lock); | |
2725 | list_for_each_entry(s, &slab_caches, list) { | |
2726 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2727 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2728 | sysfs_slab_add(s); | |
2729 | } | |
2730 | } | |
2731 | up_write(&slub_lock); | |
2732 | } | |
2733 | ||
2734 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2735 | ||
2e443fd0 CL |
2736 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2737 | { | |
2738 | struct kmem_cache *s; | |
2e443fd0 CL |
2739 | char *text; |
2740 | size_t realsize; | |
964cf35c | 2741 | unsigned long slabflags; |
2e443fd0 CL |
2742 | |
2743 | s = kmalloc_caches_dma[index]; | |
2744 | if (s) | |
2745 | return s; | |
2746 | ||
2747 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2748 | if (flags & __GFP_WAIT) |
2749 | down_write(&slub_lock); | |
2750 | else { | |
2751 | if (!down_write_trylock(&slub_lock)) | |
2752 | goto out; | |
2753 | } | |
2754 | ||
2755 | if (kmalloc_caches_dma[index]) | |
2756 | goto unlock_out; | |
2e443fd0 | 2757 | |
7b55f620 | 2758 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2759 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2760 | (unsigned int)realsize); | |
1ceef402 CL |
2761 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2762 | ||
964cf35c NP |
2763 | /* |
2764 | * Must defer sysfs creation to a workqueue because we don't know | |
2765 | * what context we are called from. Before sysfs comes up, we don't | |
2766 | * need to do anything because our sysfs initcall will start by | |
2767 | * adding all existing slabs to sysfs. | |
2768 | */ | |
5caf5c7d | 2769 | slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK; |
964cf35c NP |
2770 | if (slab_state >= SYSFS) |
2771 | slabflags |= __SYSFS_ADD_DEFERRED; | |
2772 | ||
1ceef402 | 2773 | if (!s || !text || !kmem_cache_open(s, flags, text, |
964cf35c | 2774 | realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) { |
1ceef402 CL |
2775 | kfree(s); |
2776 | kfree(text); | |
2777 | goto unlock_out; | |
dfce8648 | 2778 | } |
1ceef402 CL |
2779 | |
2780 | list_add(&s->list, &slab_caches); | |
2781 | kmalloc_caches_dma[index] = s; | |
2782 | ||
964cf35c NP |
2783 | if (slab_state >= SYSFS) |
2784 | schedule_work(&sysfs_add_work); | |
1ceef402 CL |
2785 | |
2786 | unlock_out: | |
dfce8648 | 2787 | up_write(&slub_lock); |
1ceef402 | 2788 | out: |
dfce8648 | 2789 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2790 | } |
2791 | #endif | |
2792 | ||
f1b26339 CL |
2793 | /* |
2794 | * Conversion table for small slabs sizes / 8 to the index in the | |
2795 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2796 | * of two cache sizes there. The size of larger slabs can be determined using | |
2797 | * fls. | |
2798 | */ | |
2799 | static s8 size_index[24] = { | |
2800 | 3, /* 8 */ | |
2801 | 4, /* 16 */ | |
2802 | 5, /* 24 */ | |
2803 | 5, /* 32 */ | |
2804 | 6, /* 40 */ | |
2805 | 6, /* 48 */ | |
2806 | 6, /* 56 */ | |
2807 | 6, /* 64 */ | |
2808 | 1, /* 72 */ | |
2809 | 1, /* 80 */ | |
2810 | 1, /* 88 */ | |
2811 | 1, /* 96 */ | |
2812 | 7, /* 104 */ | |
2813 | 7, /* 112 */ | |
2814 | 7, /* 120 */ | |
2815 | 7, /* 128 */ | |
2816 | 2, /* 136 */ | |
2817 | 2, /* 144 */ | |
2818 | 2, /* 152 */ | |
2819 | 2, /* 160 */ | |
2820 | 2, /* 168 */ | |
2821 | 2, /* 176 */ | |
2822 | 2, /* 184 */ | |
2823 | 2 /* 192 */ | |
2824 | }; | |
2825 | ||
acdfcd04 AK |
2826 | static inline int size_index_elem(size_t bytes) |
2827 | { | |
2828 | return (bytes - 1) / 8; | |
2829 | } | |
2830 | ||
81819f0f CL |
2831 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2832 | { | |
f1b26339 | 2833 | int index; |
81819f0f | 2834 | |
f1b26339 CL |
2835 | if (size <= 192) { |
2836 | if (!size) | |
2837 | return ZERO_SIZE_PTR; | |
81819f0f | 2838 | |
acdfcd04 | 2839 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2840 | } else |
f1b26339 | 2841 | index = fls(size - 1); |
81819f0f CL |
2842 | |
2843 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2844 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2845 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2846 | |
81819f0f CL |
2847 | #endif |
2848 | return &kmalloc_caches[index]; | |
2849 | } | |
2850 | ||
2851 | void *__kmalloc(size_t size, gfp_t flags) | |
2852 | { | |
aadb4bc4 | 2853 | struct kmem_cache *s; |
5b882be4 | 2854 | void *ret; |
81819f0f | 2855 | |
ffadd4d0 | 2856 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2857 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2858 | |
2859 | s = get_slab(size, flags); | |
2860 | ||
2861 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2862 | return s; |
2863 | ||
5b882be4 EGM |
2864 | ret = slab_alloc(s, flags, -1, _RET_IP_); |
2865 | ||
ca2b84cb | 2866 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2867 | |
2868 | return ret; | |
81819f0f CL |
2869 | } |
2870 | EXPORT_SYMBOL(__kmalloc); | |
2871 | ||
f619cfe1 CL |
2872 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2873 | { | |
b1eeab67 | 2874 | struct page *page; |
e4f7c0b4 | 2875 | void *ptr = NULL; |
f619cfe1 | 2876 | |
b1eeab67 VN |
2877 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2878 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2879 | if (page) |
e4f7c0b4 CM |
2880 | ptr = page_address(page); |
2881 | ||
2882 | kmemleak_alloc(ptr, size, 1, flags); | |
2883 | return ptr; | |
f619cfe1 CL |
2884 | } |
2885 | ||
81819f0f CL |
2886 | #ifdef CONFIG_NUMA |
2887 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2888 | { | |
aadb4bc4 | 2889 | struct kmem_cache *s; |
5b882be4 | 2890 | void *ret; |
81819f0f | 2891 | |
057685cf | 2892 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2893 | ret = kmalloc_large_node(size, flags, node); |
2894 | ||
ca2b84cb EGM |
2895 | trace_kmalloc_node(_RET_IP_, ret, |
2896 | size, PAGE_SIZE << get_order(size), | |
2897 | flags, node); | |
5b882be4 EGM |
2898 | |
2899 | return ret; | |
2900 | } | |
aadb4bc4 CL |
2901 | |
2902 | s = get_slab(size, flags); | |
2903 | ||
2904 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2905 | return s; |
2906 | ||
5b882be4 EGM |
2907 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2908 | ||
ca2b84cb | 2909 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2910 | |
2911 | return ret; | |
81819f0f CL |
2912 | } |
2913 | EXPORT_SYMBOL(__kmalloc_node); | |
2914 | #endif | |
2915 | ||
2916 | size_t ksize(const void *object) | |
2917 | { | |
272c1d21 | 2918 | struct page *page; |
81819f0f CL |
2919 | struct kmem_cache *s; |
2920 | ||
ef8b4520 | 2921 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2922 | return 0; |
2923 | ||
294a80a8 | 2924 | page = virt_to_head_page(object); |
294a80a8 | 2925 | |
76994412 PE |
2926 | if (unlikely(!PageSlab(page))) { |
2927 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2928 | return PAGE_SIZE << compound_order(page); |
76994412 | 2929 | } |
81819f0f | 2930 | s = page->slab; |
81819f0f | 2931 | |
ae20bfda | 2932 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2933 | /* |
2934 | * Debugging requires use of the padding between object | |
2935 | * and whatever may come after it. | |
2936 | */ | |
2937 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2938 | return s->objsize; | |
2939 | ||
ae20bfda | 2940 | #endif |
81819f0f CL |
2941 | /* |
2942 | * If we have the need to store the freelist pointer | |
2943 | * back there or track user information then we can | |
2944 | * only use the space before that information. | |
2945 | */ | |
2946 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2947 | return s->inuse; | |
81819f0f CL |
2948 | /* |
2949 | * Else we can use all the padding etc for the allocation | |
2950 | */ | |
2951 | return s->size; | |
2952 | } | |
b1aabecd | 2953 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2954 | |
2955 | void kfree(const void *x) | |
2956 | { | |
81819f0f | 2957 | struct page *page; |
5bb983b0 | 2958 | void *object = (void *)x; |
81819f0f | 2959 | |
2121db74 PE |
2960 | trace_kfree(_RET_IP_, x); |
2961 | ||
2408c550 | 2962 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2963 | return; |
2964 | ||
b49af68f | 2965 | page = virt_to_head_page(x); |
aadb4bc4 | 2966 | if (unlikely(!PageSlab(page))) { |
0937502a | 2967 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2968 | kmemleak_free(x); |
aadb4bc4 CL |
2969 | put_page(page); |
2970 | return; | |
2971 | } | |
ce71e27c | 2972 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2973 | } |
2974 | EXPORT_SYMBOL(kfree); | |
2975 | ||
2086d26a | 2976 | /* |
672bba3a CL |
2977 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2978 | * the remaining slabs by the number of items in use. The slabs with the | |
2979 | * most items in use come first. New allocations will then fill those up | |
2980 | * and thus they can be removed from the partial lists. | |
2981 | * | |
2982 | * The slabs with the least items are placed last. This results in them | |
2983 | * being allocated from last increasing the chance that the last objects | |
2984 | * are freed in them. | |
2086d26a CL |
2985 | */ |
2986 | int kmem_cache_shrink(struct kmem_cache *s) | |
2987 | { | |
2988 | int node; | |
2989 | int i; | |
2990 | struct kmem_cache_node *n; | |
2991 | struct page *page; | |
2992 | struct page *t; | |
205ab99d | 2993 | int objects = oo_objects(s->max); |
2086d26a | 2994 | struct list_head *slabs_by_inuse = |
834f3d11 | 2995 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2996 | unsigned long flags; |
2997 | ||
2998 | if (!slabs_by_inuse) | |
2999 | return -ENOMEM; | |
3000 | ||
3001 | flush_all(s); | |
f64dc58c | 3002 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3003 | n = get_node(s, node); |
3004 | ||
3005 | if (!n->nr_partial) | |
3006 | continue; | |
3007 | ||
834f3d11 | 3008 | for (i = 0; i < objects; i++) |
2086d26a CL |
3009 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3010 | ||
3011 | spin_lock_irqsave(&n->list_lock, flags); | |
3012 | ||
3013 | /* | |
672bba3a | 3014 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3015 | * |
672bba3a CL |
3016 | * Note that concurrent frees may occur while we hold the |
3017 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3018 | */ |
3019 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
3020 | if (!page->inuse && slab_trylock(page)) { | |
3021 | /* | |
3022 | * Must hold slab lock here because slab_free | |
3023 | * may have freed the last object and be | |
3024 | * waiting to release the slab. | |
3025 | */ | |
3026 | list_del(&page->lru); | |
3027 | n->nr_partial--; | |
3028 | slab_unlock(page); | |
3029 | discard_slab(s, page); | |
3030 | } else { | |
fcda3d89 CL |
3031 | list_move(&page->lru, |
3032 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
3033 | } |
3034 | } | |
3035 | ||
2086d26a | 3036 | /* |
672bba3a CL |
3037 | * Rebuild the partial list with the slabs filled up most |
3038 | * first and the least used slabs at the end. | |
2086d26a | 3039 | */ |
834f3d11 | 3040 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3041 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3042 | ||
2086d26a CL |
3043 | spin_unlock_irqrestore(&n->list_lock, flags); |
3044 | } | |
3045 | ||
3046 | kfree(slabs_by_inuse); | |
3047 | return 0; | |
3048 | } | |
3049 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3050 | ||
b9049e23 YG |
3051 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
3052 | static int slab_mem_going_offline_callback(void *arg) | |
3053 | { | |
3054 | struct kmem_cache *s; | |
3055 | ||
3056 | down_read(&slub_lock); | |
3057 | list_for_each_entry(s, &slab_caches, list) | |
3058 | kmem_cache_shrink(s); | |
3059 | up_read(&slub_lock); | |
3060 | ||
3061 | return 0; | |
3062 | } | |
3063 | ||
3064 | static void slab_mem_offline_callback(void *arg) | |
3065 | { | |
3066 | struct kmem_cache_node *n; | |
3067 | struct kmem_cache *s; | |
3068 | struct memory_notify *marg = arg; | |
3069 | int offline_node; | |
3070 | ||
3071 | offline_node = marg->status_change_nid; | |
3072 | ||
3073 | /* | |
3074 | * If the node still has available memory. we need kmem_cache_node | |
3075 | * for it yet. | |
3076 | */ | |
3077 | if (offline_node < 0) | |
3078 | return; | |
3079 | ||
3080 | down_read(&slub_lock); | |
3081 | list_for_each_entry(s, &slab_caches, list) { | |
3082 | n = get_node(s, offline_node); | |
3083 | if (n) { | |
3084 | /* | |
3085 | * if n->nr_slabs > 0, slabs still exist on the node | |
3086 | * that is going down. We were unable to free them, | |
3087 | * and offline_pages() function shoudn't call this | |
3088 | * callback. So, we must fail. | |
3089 | */ | |
0f389ec6 | 3090 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3091 | |
3092 | s->node[offline_node] = NULL; | |
3093 | kmem_cache_free(kmalloc_caches, n); | |
3094 | } | |
3095 | } | |
3096 | up_read(&slub_lock); | |
3097 | } | |
3098 | ||
3099 | static int slab_mem_going_online_callback(void *arg) | |
3100 | { | |
3101 | struct kmem_cache_node *n; | |
3102 | struct kmem_cache *s; | |
3103 | struct memory_notify *marg = arg; | |
3104 | int nid = marg->status_change_nid; | |
3105 | int ret = 0; | |
3106 | ||
3107 | /* | |
3108 | * If the node's memory is already available, then kmem_cache_node is | |
3109 | * already created. Nothing to do. | |
3110 | */ | |
3111 | if (nid < 0) | |
3112 | return 0; | |
3113 | ||
3114 | /* | |
0121c619 | 3115 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3116 | * allocate a kmem_cache_node structure in order to bring the node |
3117 | * online. | |
3118 | */ | |
3119 | down_read(&slub_lock); | |
3120 | list_for_each_entry(s, &slab_caches, list) { | |
3121 | /* | |
3122 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3123 | * since memory is not yet available from the node that | |
3124 | * is brought up. | |
3125 | */ | |
3126 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
3127 | if (!n) { | |
3128 | ret = -ENOMEM; | |
3129 | goto out; | |
3130 | } | |
5595cffc | 3131 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3132 | s->node[nid] = n; |
3133 | } | |
3134 | out: | |
3135 | up_read(&slub_lock); | |
3136 | return ret; | |
3137 | } | |
3138 | ||
3139 | static int slab_memory_callback(struct notifier_block *self, | |
3140 | unsigned long action, void *arg) | |
3141 | { | |
3142 | int ret = 0; | |
3143 | ||
3144 | switch (action) { | |
3145 | case MEM_GOING_ONLINE: | |
3146 | ret = slab_mem_going_online_callback(arg); | |
3147 | break; | |
3148 | case MEM_GOING_OFFLINE: | |
3149 | ret = slab_mem_going_offline_callback(arg); | |
3150 | break; | |
3151 | case MEM_OFFLINE: | |
3152 | case MEM_CANCEL_ONLINE: | |
3153 | slab_mem_offline_callback(arg); | |
3154 | break; | |
3155 | case MEM_ONLINE: | |
3156 | case MEM_CANCEL_OFFLINE: | |
3157 | break; | |
3158 | } | |
dc19f9db KH |
3159 | if (ret) |
3160 | ret = notifier_from_errno(ret); | |
3161 | else | |
3162 | ret = NOTIFY_OK; | |
b9049e23 YG |
3163 | return ret; |
3164 | } | |
3165 | ||
3166 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3167 | ||
81819f0f CL |
3168 | /******************************************************************** |
3169 | * Basic setup of slabs | |
3170 | *******************************************************************/ | |
3171 | ||
3172 | void __init kmem_cache_init(void) | |
3173 | { | |
3174 | int i; | |
4b356be0 | 3175 | int caches = 0; |
81819f0f | 3176 | |
4c93c355 CL |
3177 | init_alloc_cpu(); |
3178 | ||
81819f0f CL |
3179 | #ifdef CONFIG_NUMA |
3180 | /* | |
3181 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3182 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3183 | * kmem_cache_open for slab_state == DOWN. |
3184 | */ | |
3185 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
83b519e8 | 3186 | sizeof(struct kmem_cache_node), GFP_NOWAIT); |
8ffa6875 | 3187 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3188 | caches++; |
b9049e23 | 3189 | |
0c40ba4f | 3190 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3191 | #endif |
3192 | ||
3193 | /* Able to allocate the per node structures */ | |
3194 | slab_state = PARTIAL; | |
3195 | ||
3196 | /* Caches that are not of the two-to-the-power-of size */ | |
acdfcd04 | 3197 | if (KMALLOC_MIN_SIZE <= 32) { |
4b356be0 | 3198 | create_kmalloc_cache(&kmalloc_caches[1], |
83b519e8 | 3199 | "kmalloc-96", 96, GFP_NOWAIT); |
4b356be0 | 3200 | caches++; |
acdfcd04 AK |
3201 | } |
3202 | if (KMALLOC_MIN_SIZE <= 64) { | |
4b356be0 | 3203 | create_kmalloc_cache(&kmalloc_caches[2], |
83b519e8 | 3204 | "kmalloc-192", 192, GFP_NOWAIT); |
4b356be0 CL |
3205 | caches++; |
3206 | } | |
81819f0f | 3207 | |
ffadd4d0 | 3208 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
81819f0f | 3209 | create_kmalloc_cache(&kmalloc_caches[i], |
83b519e8 | 3210 | "kmalloc", 1 << i, GFP_NOWAIT); |
4b356be0 CL |
3211 | caches++; |
3212 | } | |
81819f0f | 3213 | |
f1b26339 CL |
3214 | |
3215 | /* | |
3216 | * Patch up the size_index table if we have strange large alignment | |
3217 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3218 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3219 | * |
3220 | * Largest permitted alignment is 256 bytes due to the way we | |
3221 | * handle the index determination for the smaller caches. | |
3222 | * | |
3223 | * Make sure that nothing crazy happens if someone starts tinkering | |
3224 | * around with ARCH_KMALLOC_MINALIGN | |
3225 | */ | |
3226 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3227 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3228 | ||
acdfcd04 AK |
3229 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3230 | int elem = size_index_elem(i); | |
3231 | if (elem >= ARRAY_SIZE(size_index)) | |
3232 | break; | |
3233 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3234 | } | |
f1b26339 | 3235 | |
acdfcd04 AK |
3236 | if (KMALLOC_MIN_SIZE == 64) { |
3237 | /* | |
3238 | * The 96 byte size cache is not used if the alignment | |
3239 | * is 64 byte. | |
3240 | */ | |
3241 | for (i = 64 + 8; i <= 96; i += 8) | |
3242 | size_index[size_index_elem(i)] = 7; | |
3243 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3244 | /* |
3245 | * The 192 byte sized cache is not used if the alignment | |
3246 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3247 | * instead. | |
3248 | */ | |
3249 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3250 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3251 | } |
3252 | ||
81819f0f CL |
3253 | slab_state = UP; |
3254 | ||
3255 | /* Provide the correct kmalloc names now that the caches are up */ | |
ffadd4d0 | 3256 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) |
81819f0f | 3257 | kmalloc_caches[i]. name = |
83b519e8 | 3258 | kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); |
81819f0f CL |
3259 | |
3260 | #ifdef CONFIG_SMP | |
3261 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3262 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3263 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3264 | #else | |
3265 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3266 | #endif |
3267 | ||
3adbefee IM |
3268 | printk(KERN_INFO |
3269 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3270 | " CPUs=%d, Nodes=%d\n", |
3271 | caches, cache_line_size(), | |
81819f0f CL |
3272 | slub_min_order, slub_max_order, slub_min_objects, |
3273 | nr_cpu_ids, nr_node_ids); | |
3274 | } | |
3275 | ||
7e85ee0c PE |
3276 | void __init kmem_cache_init_late(void) |
3277 | { | |
7e85ee0c PE |
3278 | } |
3279 | ||
81819f0f CL |
3280 | /* |
3281 | * Find a mergeable slab cache | |
3282 | */ | |
3283 | static int slab_unmergeable(struct kmem_cache *s) | |
3284 | { | |
3285 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3286 | return 1; | |
3287 | ||
c59def9f | 3288 | if (s->ctor) |
81819f0f CL |
3289 | return 1; |
3290 | ||
8ffa6875 CL |
3291 | /* |
3292 | * We may have set a slab to be unmergeable during bootstrap. | |
3293 | */ | |
3294 | if (s->refcount < 0) | |
3295 | return 1; | |
3296 | ||
81819f0f CL |
3297 | return 0; |
3298 | } | |
3299 | ||
3300 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3301 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3302 | void (*ctor)(void *)) |
81819f0f | 3303 | { |
5b95a4ac | 3304 | struct kmem_cache *s; |
81819f0f CL |
3305 | |
3306 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3307 | return NULL; | |
3308 | ||
c59def9f | 3309 | if (ctor) |
81819f0f CL |
3310 | return NULL; |
3311 | ||
3312 | size = ALIGN(size, sizeof(void *)); | |
3313 | align = calculate_alignment(flags, align, size); | |
3314 | size = ALIGN(size, align); | |
ba0268a8 | 3315 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3316 | |
5b95a4ac | 3317 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3318 | if (slab_unmergeable(s)) |
3319 | continue; | |
3320 | ||
3321 | if (size > s->size) | |
3322 | continue; | |
3323 | ||
ba0268a8 | 3324 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3325 | continue; |
3326 | /* | |
3327 | * Check if alignment is compatible. | |
3328 | * Courtesy of Adrian Drzewiecki | |
3329 | */ | |
06428780 | 3330 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3331 | continue; |
3332 | ||
3333 | if (s->size - size >= sizeof(void *)) | |
3334 | continue; | |
3335 | ||
3336 | return s; | |
3337 | } | |
3338 | return NULL; | |
3339 | } | |
3340 | ||
3341 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3342 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3343 | { |
3344 | struct kmem_cache *s; | |
3345 | ||
3346 | down_write(&slub_lock); | |
ba0268a8 | 3347 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3348 | if (s) { |
42a9fdbb CL |
3349 | int cpu; |
3350 | ||
81819f0f CL |
3351 | s->refcount++; |
3352 | /* | |
3353 | * Adjust the object sizes so that we clear | |
3354 | * the complete object on kzalloc. | |
3355 | */ | |
3356 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3357 | |
3358 | /* | |
3359 | * And then we need to update the object size in the | |
3360 | * per cpu structures | |
3361 | */ | |
3362 | for_each_online_cpu(cpu) | |
3363 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3364 | |
81819f0f | 3365 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3366 | up_write(&slub_lock); |
6446faa2 | 3367 | |
7b8f3b66 DR |
3368 | if (sysfs_slab_alias(s, name)) { |
3369 | down_write(&slub_lock); | |
3370 | s->refcount--; | |
3371 | up_write(&slub_lock); | |
81819f0f | 3372 | goto err; |
7b8f3b66 | 3373 | } |
a0e1d1be CL |
3374 | return s; |
3375 | } | |
6446faa2 | 3376 | |
a0e1d1be CL |
3377 | s = kmalloc(kmem_size, GFP_KERNEL); |
3378 | if (s) { | |
3379 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3380 | size, align, flags, ctor)) { |
81819f0f | 3381 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3382 | up_write(&slub_lock); |
7b8f3b66 DR |
3383 | if (sysfs_slab_add(s)) { |
3384 | down_write(&slub_lock); | |
3385 | list_del(&s->list); | |
3386 | up_write(&slub_lock); | |
3387 | kfree(s); | |
a0e1d1be | 3388 | goto err; |
7b8f3b66 | 3389 | } |
a0e1d1be CL |
3390 | return s; |
3391 | } | |
3392 | kfree(s); | |
81819f0f CL |
3393 | } |
3394 | up_write(&slub_lock); | |
81819f0f CL |
3395 | |
3396 | err: | |
81819f0f CL |
3397 | if (flags & SLAB_PANIC) |
3398 | panic("Cannot create slabcache %s\n", name); | |
3399 | else | |
3400 | s = NULL; | |
3401 | return s; | |
3402 | } | |
3403 | EXPORT_SYMBOL(kmem_cache_create); | |
3404 | ||
81819f0f | 3405 | #ifdef CONFIG_SMP |
81819f0f | 3406 | /* |
672bba3a CL |
3407 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3408 | * necessary. | |
81819f0f CL |
3409 | */ |
3410 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3411 | unsigned long action, void *hcpu) | |
3412 | { | |
3413 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3414 | struct kmem_cache *s; |
3415 | unsigned long flags; | |
81819f0f CL |
3416 | |
3417 | switch (action) { | |
4c93c355 CL |
3418 | case CPU_UP_PREPARE: |
3419 | case CPU_UP_PREPARE_FROZEN: | |
3420 | init_alloc_cpu_cpu(cpu); | |
3421 | down_read(&slub_lock); | |
3422 | list_for_each_entry(s, &slab_caches, list) | |
3423 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3424 | GFP_KERNEL); | |
3425 | up_read(&slub_lock); | |
3426 | break; | |
3427 | ||
81819f0f | 3428 | case CPU_UP_CANCELED: |
8bb78442 | 3429 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3430 | case CPU_DEAD: |
8bb78442 | 3431 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3432 | down_read(&slub_lock); |
3433 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3434 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3435 | ||
5b95a4ac CL |
3436 | local_irq_save(flags); |
3437 | __flush_cpu_slab(s, cpu); | |
3438 | local_irq_restore(flags); | |
4c93c355 CL |
3439 | free_kmem_cache_cpu(c, cpu); |
3440 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3441 | } |
3442 | up_read(&slub_lock); | |
81819f0f CL |
3443 | break; |
3444 | default: | |
3445 | break; | |
3446 | } | |
3447 | return NOTIFY_OK; | |
3448 | } | |
3449 | ||
06428780 | 3450 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3451 | .notifier_call = slab_cpuup_callback |
06428780 | 3452 | }; |
81819f0f CL |
3453 | |
3454 | #endif | |
3455 | ||
ce71e27c | 3456 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3457 | { |
aadb4bc4 | 3458 | struct kmem_cache *s; |
94b528d0 | 3459 | void *ret; |
aadb4bc4 | 3460 | |
ffadd4d0 | 3461 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3462 | return kmalloc_large(size, gfpflags); |
3463 | ||
aadb4bc4 | 3464 | s = get_slab(size, gfpflags); |
81819f0f | 3465 | |
2408c550 | 3466 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3467 | return s; |
81819f0f | 3468 | |
94b528d0 EGM |
3469 | ret = slab_alloc(s, gfpflags, -1, caller); |
3470 | ||
3471 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3472 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3473 | |
3474 | return ret; | |
81819f0f CL |
3475 | } |
3476 | ||
3477 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3478 | int node, unsigned long caller) |
81819f0f | 3479 | { |
aadb4bc4 | 3480 | struct kmem_cache *s; |
94b528d0 | 3481 | void *ret; |
aadb4bc4 | 3482 | |
ffadd4d0 | 3483 | if (unlikely(size > SLUB_MAX_SIZE)) |
f619cfe1 | 3484 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3485 | |
aadb4bc4 | 3486 | s = get_slab(size, gfpflags); |
81819f0f | 3487 | |
2408c550 | 3488 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3489 | return s; |
81819f0f | 3490 | |
94b528d0 EGM |
3491 | ret = slab_alloc(s, gfpflags, node, caller); |
3492 | ||
3493 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3494 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3495 | |
3496 | return ret; | |
81819f0f CL |
3497 | } |
3498 | ||
f6acb635 | 3499 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3500 | static int count_inuse(struct page *page) |
3501 | { | |
3502 | return page->inuse; | |
3503 | } | |
3504 | ||
3505 | static int count_total(struct page *page) | |
3506 | { | |
3507 | return page->objects; | |
3508 | } | |
3509 | ||
434e245d CL |
3510 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3511 | unsigned long *map) | |
53e15af0 CL |
3512 | { |
3513 | void *p; | |
a973e9dd | 3514 | void *addr = page_address(page); |
53e15af0 CL |
3515 | |
3516 | if (!check_slab(s, page) || | |
3517 | !on_freelist(s, page, NULL)) | |
3518 | return 0; | |
3519 | ||
3520 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3521 | bitmap_zero(map, page->objects); |
53e15af0 | 3522 | |
7656c72b CL |
3523 | for_each_free_object(p, s, page->freelist) { |
3524 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3525 | if (!check_object(s, page, p, 0)) |
3526 | return 0; | |
3527 | } | |
3528 | ||
224a88be | 3529 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3530 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3531 | if (!check_object(s, page, p, 1)) |
3532 | return 0; | |
3533 | return 1; | |
3534 | } | |
3535 | ||
434e245d CL |
3536 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3537 | unsigned long *map) | |
53e15af0 CL |
3538 | { |
3539 | if (slab_trylock(page)) { | |
434e245d | 3540 | validate_slab(s, page, map); |
53e15af0 CL |
3541 | slab_unlock(page); |
3542 | } else | |
3543 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3544 | s->name, page); | |
3545 | ||
3546 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3547 | if (!PageSlubDebug(page)) |
3548 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3549 | "on slab 0x%p\n", s->name, page); |
3550 | } else { | |
8a38082d AW |
3551 | if (PageSlubDebug(page)) |
3552 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3553 | "slab 0x%p\n", s->name, page); |
3554 | } | |
3555 | } | |
3556 | ||
434e245d CL |
3557 | static int validate_slab_node(struct kmem_cache *s, |
3558 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3559 | { |
3560 | unsigned long count = 0; | |
3561 | struct page *page; | |
3562 | unsigned long flags; | |
3563 | ||
3564 | spin_lock_irqsave(&n->list_lock, flags); | |
3565 | ||
3566 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3567 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3568 | count++; |
3569 | } | |
3570 | if (count != n->nr_partial) | |
3571 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3572 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3573 | ||
3574 | if (!(s->flags & SLAB_STORE_USER)) | |
3575 | goto out; | |
3576 | ||
3577 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3578 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3579 | count++; |
3580 | } | |
3581 | if (count != atomic_long_read(&n->nr_slabs)) | |
3582 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3583 | "counter=%ld\n", s->name, count, | |
3584 | atomic_long_read(&n->nr_slabs)); | |
3585 | ||
3586 | out: | |
3587 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3588 | return count; | |
3589 | } | |
3590 | ||
434e245d | 3591 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3592 | { |
3593 | int node; | |
3594 | unsigned long count = 0; | |
205ab99d | 3595 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3596 | sizeof(unsigned long), GFP_KERNEL); |
3597 | ||
3598 | if (!map) | |
3599 | return -ENOMEM; | |
53e15af0 CL |
3600 | |
3601 | flush_all(s); | |
f64dc58c | 3602 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3603 | struct kmem_cache_node *n = get_node(s, node); |
3604 | ||
434e245d | 3605 | count += validate_slab_node(s, n, map); |
53e15af0 | 3606 | } |
434e245d | 3607 | kfree(map); |
53e15af0 CL |
3608 | return count; |
3609 | } | |
3610 | ||
b3459709 CL |
3611 | #ifdef SLUB_RESILIENCY_TEST |
3612 | static void resiliency_test(void) | |
3613 | { | |
3614 | u8 *p; | |
3615 | ||
3616 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3617 | printk(KERN_ERR "-----------------------\n"); | |
3618 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3619 | ||
3620 | p = kzalloc(16, GFP_KERNEL); | |
3621 | p[16] = 0x12; | |
3622 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3623 | " 0x12->0x%p\n\n", p + 16); | |
3624 | ||
3625 | validate_slab_cache(kmalloc_caches + 4); | |
3626 | ||
3627 | /* Hmmm... The next two are dangerous */ | |
3628 | p = kzalloc(32, GFP_KERNEL); | |
3629 | p[32 + sizeof(void *)] = 0x34; | |
3630 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3631 | " 0x34 -> -0x%p\n", p); |
3632 | printk(KERN_ERR | |
3633 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3634 | |
3635 | validate_slab_cache(kmalloc_caches + 5); | |
3636 | p = kzalloc(64, GFP_KERNEL); | |
3637 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3638 | *p = 0x56; | |
3639 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3640 | p); | |
3adbefee IM |
3641 | printk(KERN_ERR |
3642 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3643 | validate_slab_cache(kmalloc_caches + 6); |
3644 | ||
3645 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3646 | p = kzalloc(128, GFP_KERNEL); | |
3647 | kfree(p); | |
3648 | *p = 0x78; | |
3649 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3650 | validate_slab_cache(kmalloc_caches + 7); | |
3651 | ||
3652 | p = kzalloc(256, GFP_KERNEL); | |
3653 | kfree(p); | |
3654 | p[50] = 0x9a; | |
3adbefee IM |
3655 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3656 | p); | |
b3459709 CL |
3657 | validate_slab_cache(kmalloc_caches + 8); |
3658 | ||
3659 | p = kzalloc(512, GFP_KERNEL); | |
3660 | kfree(p); | |
3661 | p[512] = 0xab; | |
3662 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3663 | validate_slab_cache(kmalloc_caches + 9); | |
3664 | } | |
3665 | #else | |
3666 | static void resiliency_test(void) {}; | |
3667 | #endif | |
3668 | ||
88a420e4 | 3669 | /* |
672bba3a | 3670 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3671 | * and freed. |
3672 | */ | |
3673 | ||
3674 | struct location { | |
3675 | unsigned long count; | |
ce71e27c | 3676 | unsigned long addr; |
45edfa58 CL |
3677 | long long sum_time; |
3678 | long min_time; | |
3679 | long max_time; | |
3680 | long min_pid; | |
3681 | long max_pid; | |
174596a0 | 3682 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3683 | nodemask_t nodes; |
88a420e4 CL |
3684 | }; |
3685 | ||
3686 | struct loc_track { | |
3687 | unsigned long max; | |
3688 | unsigned long count; | |
3689 | struct location *loc; | |
3690 | }; | |
3691 | ||
3692 | static void free_loc_track(struct loc_track *t) | |
3693 | { | |
3694 | if (t->max) | |
3695 | free_pages((unsigned long)t->loc, | |
3696 | get_order(sizeof(struct location) * t->max)); | |
3697 | } | |
3698 | ||
68dff6a9 | 3699 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3700 | { |
3701 | struct location *l; | |
3702 | int order; | |
3703 | ||
88a420e4 CL |
3704 | order = get_order(sizeof(struct location) * max); |
3705 | ||
68dff6a9 | 3706 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3707 | if (!l) |
3708 | return 0; | |
3709 | ||
3710 | if (t->count) { | |
3711 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3712 | free_loc_track(t); | |
3713 | } | |
3714 | t->max = max; | |
3715 | t->loc = l; | |
3716 | return 1; | |
3717 | } | |
3718 | ||
3719 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3720 | const struct track *track) |
88a420e4 CL |
3721 | { |
3722 | long start, end, pos; | |
3723 | struct location *l; | |
ce71e27c | 3724 | unsigned long caddr; |
45edfa58 | 3725 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3726 | |
3727 | start = -1; | |
3728 | end = t->count; | |
3729 | ||
3730 | for ( ; ; ) { | |
3731 | pos = start + (end - start + 1) / 2; | |
3732 | ||
3733 | /* | |
3734 | * There is nothing at "end". If we end up there | |
3735 | * we need to add something to before end. | |
3736 | */ | |
3737 | if (pos == end) | |
3738 | break; | |
3739 | ||
3740 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3741 | if (track->addr == caddr) { |
3742 | ||
3743 | l = &t->loc[pos]; | |
3744 | l->count++; | |
3745 | if (track->when) { | |
3746 | l->sum_time += age; | |
3747 | if (age < l->min_time) | |
3748 | l->min_time = age; | |
3749 | if (age > l->max_time) | |
3750 | l->max_time = age; | |
3751 | ||
3752 | if (track->pid < l->min_pid) | |
3753 | l->min_pid = track->pid; | |
3754 | if (track->pid > l->max_pid) | |
3755 | l->max_pid = track->pid; | |
3756 | ||
174596a0 RR |
3757 | cpumask_set_cpu(track->cpu, |
3758 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3759 | } |
3760 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3761 | return 1; |
3762 | } | |
3763 | ||
45edfa58 | 3764 | if (track->addr < caddr) |
88a420e4 CL |
3765 | end = pos; |
3766 | else | |
3767 | start = pos; | |
3768 | } | |
3769 | ||
3770 | /* | |
672bba3a | 3771 | * Not found. Insert new tracking element. |
88a420e4 | 3772 | */ |
68dff6a9 | 3773 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3774 | return 0; |
3775 | ||
3776 | l = t->loc + pos; | |
3777 | if (pos < t->count) | |
3778 | memmove(l + 1, l, | |
3779 | (t->count - pos) * sizeof(struct location)); | |
3780 | t->count++; | |
3781 | l->count = 1; | |
45edfa58 CL |
3782 | l->addr = track->addr; |
3783 | l->sum_time = age; | |
3784 | l->min_time = age; | |
3785 | l->max_time = age; | |
3786 | l->min_pid = track->pid; | |
3787 | l->max_pid = track->pid; | |
174596a0 RR |
3788 | cpumask_clear(to_cpumask(l->cpus)); |
3789 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3790 | nodes_clear(l->nodes); |
3791 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3792 | return 1; |
3793 | } | |
3794 | ||
3795 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3796 | struct page *page, enum track_item alloc) | |
3797 | { | |
a973e9dd | 3798 | void *addr = page_address(page); |
39b26464 | 3799 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3800 | void *p; |
3801 | ||
39b26464 | 3802 | bitmap_zero(map, page->objects); |
7656c72b CL |
3803 | for_each_free_object(p, s, page->freelist) |
3804 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3805 | |
224a88be | 3806 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3807 | if (!test_bit(slab_index(p, s, addr), map)) |
3808 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3809 | } |
3810 | ||
3811 | static int list_locations(struct kmem_cache *s, char *buf, | |
3812 | enum track_item alloc) | |
3813 | { | |
e374d483 | 3814 | int len = 0; |
88a420e4 | 3815 | unsigned long i; |
68dff6a9 | 3816 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3817 | int node; |
3818 | ||
68dff6a9 | 3819 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3820 | GFP_TEMPORARY)) |
68dff6a9 | 3821 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3822 | |
3823 | /* Push back cpu slabs */ | |
3824 | flush_all(s); | |
3825 | ||
f64dc58c | 3826 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3827 | struct kmem_cache_node *n = get_node(s, node); |
3828 | unsigned long flags; | |
3829 | struct page *page; | |
3830 | ||
9e86943b | 3831 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3832 | continue; |
3833 | ||
3834 | spin_lock_irqsave(&n->list_lock, flags); | |
3835 | list_for_each_entry(page, &n->partial, lru) | |
3836 | process_slab(&t, s, page, alloc); | |
3837 | list_for_each_entry(page, &n->full, lru) | |
3838 | process_slab(&t, s, page, alloc); | |
3839 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3840 | } | |
3841 | ||
3842 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3843 | struct location *l = &t.loc[i]; |
88a420e4 | 3844 | |
9c246247 | 3845 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3846 | break; |
e374d483 | 3847 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3848 | |
3849 | if (l->addr) | |
e374d483 | 3850 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3851 | else |
e374d483 | 3852 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3853 | |
3854 | if (l->sum_time != l->min_time) { | |
e374d483 | 3855 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3856 | l->min_time, |
3857 | (long)div_u64(l->sum_time, l->count), | |
3858 | l->max_time); | |
45edfa58 | 3859 | } else |
e374d483 | 3860 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3861 | l->min_time); |
3862 | ||
3863 | if (l->min_pid != l->max_pid) | |
e374d483 | 3864 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3865 | l->min_pid, l->max_pid); |
3866 | else | |
e374d483 | 3867 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3868 | l->min_pid); |
3869 | ||
174596a0 RR |
3870 | if (num_online_cpus() > 1 && |
3871 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3872 | len < PAGE_SIZE - 60) { |
3873 | len += sprintf(buf + len, " cpus="); | |
3874 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3875 | to_cpumask(l->cpus)); |
45edfa58 CL |
3876 | } |
3877 | ||
62bc62a8 | 3878 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3879 | len < PAGE_SIZE - 60) { |
3880 | len += sprintf(buf + len, " nodes="); | |
3881 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3882 | l->nodes); |
3883 | } | |
3884 | ||
e374d483 | 3885 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3886 | } |
3887 | ||
3888 | free_loc_track(&t); | |
3889 | if (!t.count) | |
e374d483 HH |
3890 | len += sprintf(buf, "No data\n"); |
3891 | return len; | |
88a420e4 CL |
3892 | } |
3893 | ||
81819f0f | 3894 | enum slab_stat_type { |
205ab99d CL |
3895 | SL_ALL, /* All slabs */ |
3896 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3897 | SL_CPU, /* Only slabs used for cpu caches */ | |
3898 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3899 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3900 | }; |
3901 | ||
205ab99d | 3902 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3903 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3904 | #define SO_CPU (1 << SL_CPU) | |
3905 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3906 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3907 | |
62e5c4b4 CG |
3908 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3909 | char *buf, unsigned long flags) | |
81819f0f CL |
3910 | { |
3911 | unsigned long total = 0; | |
81819f0f CL |
3912 | int node; |
3913 | int x; | |
3914 | unsigned long *nodes; | |
3915 | unsigned long *per_cpu; | |
3916 | ||
3917 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3918 | if (!nodes) |
3919 | return -ENOMEM; | |
81819f0f CL |
3920 | per_cpu = nodes + nr_node_ids; |
3921 | ||
205ab99d CL |
3922 | if (flags & SO_CPU) { |
3923 | int cpu; | |
81819f0f | 3924 | |
205ab99d CL |
3925 | for_each_possible_cpu(cpu) { |
3926 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3927 | |
205ab99d CL |
3928 | if (!c || c->node < 0) |
3929 | continue; | |
3930 | ||
3931 | if (c->page) { | |
3932 | if (flags & SO_TOTAL) | |
3933 | x = c->page->objects; | |
3934 | else if (flags & SO_OBJECTS) | |
3935 | x = c->page->inuse; | |
81819f0f CL |
3936 | else |
3937 | x = 1; | |
205ab99d | 3938 | |
81819f0f | 3939 | total += x; |
205ab99d | 3940 | nodes[c->node] += x; |
81819f0f | 3941 | } |
205ab99d | 3942 | per_cpu[c->node]++; |
81819f0f CL |
3943 | } |
3944 | } | |
3945 | ||
205ab99d CL |
3946 | if (flags & SO_ALL) { |
3947 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3948 | struct kmem_cache_node *n = get_node(s, node); | |
3949 | ||
3950 | if (flags & SO_TOTAL) | |
3951 | x = atomic_long_read(&n->total_objects); | |
3952 | else if (flags & SO_OBJECTS) | |
3953 | x = atomic_long_read(&n->total_objects) - | |
3954 | count_partial(n, count_free); | |
81819f0f | 3955 | |
81819f0f | 3956 | else |
205ab99d | 3957 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3958 | total += x; |
3959 | nodes[node] += x; | |
3960 | } | |
3961 | ||
205ab99d CL |
3962 | } else if (flags & SO_PARTIAL) { |
3963 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3964 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3965 | |
205ab99d CL |
3966 | if (flags & SO_TOTAL) |
3967 | x = count_partial(n, count_total); | |
3968 | else if (flags & SO_OBJECTS) | |
3969 | x = count_partial(n, count_inuse); | |
81819f0f | 3970 | else |
205ab99d | 3971 | x = n->nr_partial; |
81819f0f CL |
3972 | total += x; |
3973 | nodes[node] += x; | |
3974 | } | |
3975 | } | |
81819f0f CL |
3976 | x = sprintf(buf, "%lu", total); |
3977 | #ifdef CONFIG_NUMA | |
f64dc58c | 3978 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3979 | if (nodes[node]) |
3980 | x += sprintf(buf + x, " N%d=%lu", | |
3981 | node, nodes[node]); | |
3982 | #endif | |
3983 | kfree(nodes); | |
3984 | return x + sprintf(buf + x, "\n"); | |
3985 | } | |
3986 | ||
3987 | static int any_slab_objects(struct kmem_cache *s) | |
3988 | { | |
3989 | int node; | |
81819f0f | 3990 | |
dfb4f096 | 3991 | for_each_online_node(node) { |
81819f0f CL |
3992 | struct kmem_cache_node *n = get_node(s, node); |
3993 | ||
dfb4f096 CL |
3994 | if (!n) |
3995 | continue; | |
3996 | ||
4ea33e2d | 3997 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3998 | return 1; |
3999 | } | |
4000 | return 0; | |
4001 | } | |
4002 | ||
4003 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
4004 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
4005 | ||
4006 | struct slab_attribute { | |
4007 | struct attribute attr; | |
4008 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4009 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4010 | }; | |
4011 | ||
4012 | #define SLAB_ATTR_RO(_name) \ | |
4013 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4014 | ||
4015 | #define SLAB_ATTR(_name) \ | |
4016 | static struct slab_attribute _name##_attr = \ | |
4017 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4018 | ||
81819f0f CL |
4019 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4020 | { | |
4021 | return sprintf(buf, "%d\n", s->size); | |
4022 | } | |
4023 | SLAB_ATTR_RO(slab_size); | |
4024 | ||
4025 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4026 | { | |
4027 | return sprintf(buf, "%d\n", s->align); | |
4028 | } | |
4029 | SLAB_ATTR_RO(align); | |
4030 | ||
4031 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4032 | { | |
4033 | return sprintf(buf, "%d\n", s->objsize); | |
4034 | } | |
4035 | SLAB_ATTR_RO(object_size); | |
4036 | ||
4037 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4038 | { | |
834f3d11 | 4039 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4040 | } |
4041 | SLAB_ATTR_RO(objs_per_slab); | |
4042 | ||
06b285dc CL |
4043 | static ssize_t order_store(struct kmem_cache *s, |
4044 | const char *buf, size_t length) | |
4045 | { | |
0121c619 CL |
4046 | unsigned long order; |
4047 | int err; | |
4048 | ||
4049 | err = strict_strtoul(buf, 10, &order); | |
4050 | if (err) | |
4051 | return err; | |
06b285dc CL |
4052 | |
4053 | if (order > slub_max_order || order < slub_min_order) | |
4054 | return -EINVAL; | |
4055 | ||
4056 | calculate_sizes(s, order); | |
4057 | return length; | |
4058 | } | |
4059 | ||
81819f0f CL |
4060 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4061 | { | |
834f3d11 | 4062 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4063 | } |
06b285dc | 4064 | SLAB_ATTR(order); |
81819f0f | 4065 | |
73d342b1 DR |
4066 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4067 | { | |
4068 | return sprintf(buf, "%lu\n", s->min_partial); | |
4069 | } | |
4070 | ||
4071 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4072 | size_t length) | |
4073 | { | |
4074 | unsigned long min; | |
4075 | int err; | |
4076 | ||
4077 | err = strict_strtoul(buf, 10, &min); | |
4078 | if (err) | |
4079 | return err; | |
4080 | ||
c0bdb232 | 4081 | set_min_partial(s, min); |
73d342b1 DR |
4082 | return length; |
4083 | } | |
4084 | SLAB_ATTR(min_partial); | |
4085 | ||
81819f0f CL |
4086 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4087 | { | |
4088 | if (s->ctor) { | |
4089 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
4090 | ||
4091 | return n + sprintf(buf + n, "\n"); | |
4092 | } | |
4093 | return 0; | |
4094 | } | |
4095 | SLAB_ATTR_RO(ctor); | |
4096 | ||
81819f0f CL |
4097 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4098 | { | |
4099 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4100 | } | |
4101 | SLAB_ATTR_RO(aliases); | |
4102 | ||
4103 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
4104 | { | |
205ab99d | 4105 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
4106 | } |
4107 | SLAB_ATTR_RO(slabs); | |
4108 | ||
4109 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
4110 | { | |
d9acf4b7 | 4111 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4112 | } |
4113 | SLAB_ATTR_RO(partial); | |
4114 | ||
4115 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4116 | { | |
d9acf4b7 | 4117 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4118 | } |
4119 | SLAB_ATTR_RO(cpu_slabs); | |
4120 | ||
4121 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4122 | { | |
205ab99d | 4123 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4124 | } |
4125 | SLAB_ATTR_RO(objects); | |
4126 | ||
205ab99d CL |
4127 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4128 | { | |
4129 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4130 | } | |
4131 | SLAB_ATTR_RO(objects_partial); | |
4132 | ||
4133 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4134 | { | |
4135 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4136 | } | |
4137 | SLAB_ATTR_RO(total_objects); | |
4138 | ||
81819f0f CL |
4139 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4140 | { | |
4141 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4142 | } | |
4143 | ||
4144 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4145 | const char *buf, size_t length) | |
4146 | { | |
4147 | s->flags &= ~SLAB_DEBUG_FREE; | |
4148 | if (buf[0] == '1') | |
4149 | s->flags |= SLAB_DEBUG_FREE; | |
4150 | return length; | |
4151 | } | |
4152 | SLAB_ATTR(sanity_checks); | |
4153 | ||
4154 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4155 | { | |
4156 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4157 | } | |
4158 | ||
4159 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4160 | size_t length) | |
4161 | { | |
4162 | s->flags &= ~SLAB_TRACE; | |
4163 | if (buf[0] == '1') | |
4164 | s->flags |= SLAB_TRACE; | |
4165 | return length; | |
4166 | } | |
4167 | SLAB_ATTR(trace); | |
4168 | ||
4169 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
4170 | { | |
4171 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4172 | } | |
4173 | ||
4174 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4175 | const char *buf, size_t length) | |
4176 | { | |
4177 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4178 | if (buf[0] == '1') | |
4179 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4180 | return length; | |
4181 | } | |
4182 | SLAB_ATTR(reclaim_account); | |
4183 | ||
4184 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4185 | { | |
5af60839 | 4186 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4187 | } |
4188 | SLAB_ATTR_RO(hwcache_align); | |
4189 | ||
4190 | #ifdef CONFIG_ZONE_DMA | |
4191 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4192 | { | |
4193 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4194 | } | |
4195 | SLAB_ATTR_RO(cache_dma); | |
4196 | #endif | |
4197 | ||
4198 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4199 | { | |
4200 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4201 | } | |
4202 | SLAB_ATTR_RO(destroy_by_rcu); | |
4203 | ||
4204 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4205 | { | |
4206 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4207 | } | |
4208 | ||
4209 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4210 | const char *buf, size_t length) | |
4211 | { | |
4212 | if (any_slab_objects(s)) | |
4213 | return -EBUSY; | |
4214 | ||
4215 | s->flags &= ~SLAB_RED_ZONE; | |
4216 | if (buf[0] == '1') | |
4217 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4218 | calculate_sizes(s, -1); |
81819f0f CL |
4219 | return length; |
4220 | } | |
4221 | SLAB_ATTR(red_zone); | |
4222 | ||
4223 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4224 | { | |
4225 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4226 | } | |
4227 | ||
4228 | static ssize_t poison_store(struct kmem_cache *s, | |
4229 | const char *buf, size_t length) | |
4230 | { | |
4231 | if (any_slab_objects(s)) | |
4232 | return -EBUSY; | |
4233 | ||
4234 | s->flags &= ~SLAB_POISON; | |
4235 | if (buf[0] == '1') | |
4236 | s->flags |= SLAB_POISON; | |
06b285dc | 4237 | calculate_sizes(s, -1); |
81819f0f CL |
4238 | return length; |
4239 | } | |
4240 | SLAB_ATTR(poison); | |
4241 | ||
4242 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4243 | { | |
4244 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4245 | } | |
4246 | ||
4247 | static ssize_t store_user_store(struct kmem_cache *s, | |
4248 | const char *buf, size_t length) | |
4249 | { | |
4250 | if (any_slab_objects(s)) | |
4251 | return -EBUSY; | |
4252 | ||
4253 | s->flags &= ~SLAB_STORE_USER; | |
4254 | if (buf[0] == '1') | |
4255 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4256 | calculate_sizes(s, -1); |
81819f0f CL |
4257 | return length; |
4258 | } | |
4259 | SLAB_ATTR(store_user); | |
4260 | ||
53e15af0 CL |
4261 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4262 | { | |
4263 | return 0; | |
4264 | } | |
4265 | ||
4266 | static ssize_t validate_store(struct kmem_cache *s, | |
4267 | const char *buf, size_t length) | |
4268 | { | |
434e245d CL |
4269 | int ret = -EINVAL; |
4270 | ||
4271 | if (buf[0] == '1') { | |
4272 | ret = validate_slab_cache(s); | |
4273 | if (ret >= 0) | |
4274 | ret = length; | |
4275 | } | |
4276 | return ret; | |
53e15af0 CL |
4277 | } |
4278 | SLAB_ATTR(validate); | |
4279 | ||
2086d26a CL |
4280 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4281 | { | |
4282 | return 0; | |
4283 | } | |
4284 | ||
4285 | static ssize_t shrink_store(struct kmem_cache *s, | |
4286 | const char *buf, size_t length) | |
4287 | { | |
4288 | if (buf[0] == '1') { | |
4289 | int rc = kmem_cache_shrink(s); | |
4290 | ||
4291 | if (rc) | |
4292 | return rc; | |
4293 | } else | |
4294 | return -EINVAL; | |
4295 | return length; | |
4296 | } | |
4297 | SLAB_ATTR(shrink); | |
4298 | ||
88a420e4 CL |
4299 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4300 | { | |
4301 | if (!(s->flags & SLAB_STORE_USER)) | |
4302 | return -ENOSYS; | |
4303 | return list_locations(s, buf, TRACK_ALLOC); | |
4304 | } | |
4305 | SLAB_ATTR_RO(alloc_calls); | |
4306 | ||
4307 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4308 | { | |
4309 | if (!(s->flags & SLAB_STORE_USER)) | |
4310 | return -ENOSYS; | |
4311 | return list_locations(s, buf, TRACK_FREE); | |
4312 | } | |
4313 | SLAB_ATTR_RO(free_calls); | |
4314 | ||
81819f0f | 4315 | #ifdef CONFIG_NUMA |
9824601e | 4316 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4317 | { |
9824601e | 4318 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4319 | } |
4320 | ||
9824601e | 4321 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4322 | const char *buf, size_t length) |
4323 | { | |
0121c619 CL |
4324 | unsigned long ratio; |
4325 | int err; | |
4326 | ||
4327 | err = strict_strtoul(buf, 10, &ratio); | |
4328 | if (err) | |
4329 | return err; | |
4330 | ||
e2cb96b7 | 4331 | if (ratio <= 100) |
0121c619 | 4332 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4333 | |
81819f0f CL |
4334 | return length; |
4335 | } | |
9824601e | 4336 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4337 | #endif |
4338 | ||
8ff12cfc | 4339 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4340 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4341 | { | |
4342 | unsigned long sum = 0; | |
4343 | int cpu; | |
4344 | int len; | |
4345 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4346 | ||
4347 | if (!data) | |
4348 | return -ENOMEM; | |
4349 | ||
4350 | for_each_online_cpu(cpu) { | |
4351 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4352 | ||
4353 | data[cpu] = x; | |
4354 | sum += x; | |
4355 | } | |
4356 | ||
4357 | len = sprintf(buf, "%lu", sum); | |
4358 | ||
50ef37b9 | 4359 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4360 | for_each_online_cpu(cpu) { |
4361 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4362 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4363 | } |
50ef37b9 | 4364 | #endif |
8ff12cfc CL |
4365 | kfree(data); |
4366 | return len + sprintf(buf + len, "\n"); | |
4367 | } | |
4368 | ||
4369 | #define STAT_ATTR(si, text) \ | |
4370 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4371 | { \ | |
4372 | return show_stat(s, buf, si); \ | |
4373 | } \ | |
4374 | SLAB_ATTR_RO(text); \ | |
4375 | ||
4376 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4377 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4378 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4379 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4380 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4381 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4382 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4383 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4384 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4385 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4386 | STAT_ATTR(FREE_SLAB, free_slab); | |
4387 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4388 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4389 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4390 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4391 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4392 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4393 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4394 | #endif |
4395 | ||
06428780 | 4396 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4397 | &slab_size_attr.attr, |
4398 | &object_size_attr.attr, | |
4399 | &objs_per_slab_attr.attr, | |
4400 | &order_attr.attr, | |
73d342b1 | 4401 | &min_partial_attr.attr, |
81819f0f | 4402 | &objects_attr.attr, |
205ab99d CL |
4403 | &objects_partial_attr.attr, |
4404 | &total_objects_attr.attr, | |
81819f0f CL |
4405 | &slabs_attr.attr, |
4406 | &partial_attr.attr, | |
4407 | &cpu_slabs_attr.attr, | |
4408 | &ctor_attr.attr, | |
81819f0f CL |
4409 | &aliases_attr.attr, |
4410 | &align_attr.attr, | |
4411 | &sanity_checks_attr.attr, | |
4412 | &trace_attr.attr, | |
4413 | &hwcache_align_attr.attr, | |
4414 | &reclaim_account_attr.attr, | |
4415 | &destroy_by_rcu_attr.attr, | |
4416 | &red_zone_attr.attr, | |
4417 | &poison_attr.attr, | |
4418 | &store_user_attr.attr, | |
53e15af0 | 4419 | &validate_attr.attr, |
2086d26a | 4420 | &shrink_attr.attr, |
88a420e4 CL |
4421 | &alloc_calls_attr.attr, |
4422 | &free_calls_attr.attr, | |
81819f0f CL |
4423 | #ifdef CONFIG_ZONE_DMA |
4424 | &cache_dma_attr.attr, | |
4425 | #endif | |
4426 | #ifdef CONFIG_NUMA | |
9824601e | 4427 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4428 | #endif |
4429 | #ifdef CONFIG_SLUB_STATS | |
4430 | &alloc_fastpath_attr.attr, | |
4431 | &alloc_slowpath_attr.attr, | |
4432 | &free_fastpath_attr.attr, | |
4433 | &free_slowpath_attr.attr, | |
4434 | &free_frozen_attr.attr, | |
4435 | &free_add_partial_attr.attr, | |
4436 | &free_remove_partial_attr.attr, | |
4437 | &alloc_from_partial_attr.attr, | |
4438 | &alloc_slab_attr.attr, | |
4439 | &alloc_refill_attr.attr, | |
4440 | &free_slab_attr.attr, | |
4441 | &cpuslab_flush_attr.attr, | |
4442 | &deactivate_full_attr.attr, | |
4443 | &deactivate_empty_attr.attr, | |
4444 | &deactivate_to_head_attr.attr, | |
4445 | &deactivate_to_tail_attr.attr, | |
4446 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4447 | &order_fallback_attr.attr, |
81819f0f CL |
4448 | #endif |
4449 | NULL | |
4450 | }; | |
4451 | ||
4452 | static struct attribute_group slab_attr_group = { | |
4453 | .attrs = slab_attrs, | |
4454 | }; | |
4455 | ||
4456 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4457 | struct attribute *attr, | |
4458 | char *buf) | |
4459 | { | |
4460 | struct slab_attribute *attribute; | |
4461 | struct kmem_cache *s; | |
4462 | int err; | |
4463 | ||
4464 | attribute = to_slab_attr(attr); | |
4465 | s = to_slab(kobj); | |
4466 | ||
4467 | if (!attribute->show) | |
4468 | return -EIO; | |
4469 | ||
4470 | err = attribute->show(s, buf); | |
4471 | ||
4472 | return err; | |
4473 | } | |
4474 | ||
4475 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4476 | struct attribute *attr, | |
4477 | const char *buf, size_t len) | |
4478 | { | |
4479 | struct slab_attribute *attribute; | |
4480 | struct kmem_cache *s; | |
4481 | int err; | |
4482 | ||
4483 | attribute = to_slab_attr(attr); | |
4484 | s = to_slab(kobj); | |
4485 | ||
4486 | if (!attribute->store) | |
4487 | return -EIO; | |
4488 | ||
4489 | err = attribute->store(s, buf, len); | |
4490 | ||
4491 | return err; | |
4492 | } | |
4493 | ||
151c602f CL |
4494 | static void kmem_cache_release(struct kobject *kobj) |
4495 | { | |
4496 | struct kmem_cache *s = to_slab(kobj); | |
4497 | ||
4498 | kfree(s); | |
4499 | } | |
4500 | ||
81819f0f CL |
4501 | static struct sysfs_ops slab_sysfs_ops = { |
4502 | .show = slab_attr_show, | |
4503 | .store = slab_attr_store, | |
4504 | }; | |
4505 | ||
4506 | static struct kobj_type slab_ktype = { | |
4507 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4508 | .release = kmem_cache_release |
81819f0f CL |
4509 | }; |
4510 | ||
4511 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4512 | { | |
4513 | struct kobj_type *ktype = get_ktype(kobj); | |
4514 | ||
4515 | if (ktype == &slab_ktype) | |
4516 | return 1; | |
4517 | return 0; | |
4518 | } | |
4519 | ||
4520 | static struct kset_uevent_ops slab_uevent_ops = { | |
4521 | .filter = uevent_filter, | |
4522 | }; | |
4523 | ||
27c3a314 | 4524 | static struct kset *slab_kset; |
81819f0f CL |
4525 | |
4526 | #define ID_STR_LENGTH 64 | |
4527 | ||
4528 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4529 | * |
4530 | * Format :[flags-]size | |
81819f0f CL |
4531 | */ |
4532 | static char *create_unique_id(struct kmem_cache *s) | |
4533 | { | |
4534 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4535 | char *p = name; | |
4536 | ||
4537 | BUG_ON(!name); | |
4538 | ||
4539 | *p++ = ':'; | |
4540 | /* | |
4541 | * First flags affecting slabcache operations. We will only | |
4542 | * get here for aliasable slabs so we do not need to support | |
4543 | * too many flags. The flags here must cover all flags that | |
4544 | * are matched during merging to guarantee that the id is | |
4545 | * unique. | |
4546 | */ | |
4547 | if (s->flags & SLAB_CACHE_DMA) | |
4548 | *p++ = 'd'; | |
4549 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4550 | *p++ = 'a'; | |
4551 | if (s->flags & SLAB_DEBUG_FREE) | |
4552 | *p++ = 'F'; | |
5a896d9e VN |
4553 | if (!(s->flags & SLAB_NOTRACK)) |
4554 | *p++ = 't'; | |
81819f0f CL |
4555 | if (p != name + 1) |
4556 | *p++ = '-'; | |
4557 | p += sprintf(p, "%07d", s->size); | |
4558 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4559 | return name; | |
4560 | } | |
4561 | ||
4562 | static int sysfs_slab_add(struct kmem_cache *s) | |
4563 | { | |
4564 | int err; | |
4565 | const char *name; | |
4566 | int unmergeable; | |
4567 | ||
4568 | if (slab_state < SYSFS) | |
4569 | /* Defer until later */ | |
4570 | return 0; | |
4571 | ||
4572 | unmergeable = slab_unmergeable(s); | |
4573 | if (unmergeable) { | |
4574 | /* | |
4575 | * Slabcache can never be merged so we can use the name proper. | |
4576 | * This is typically the case for debug situations. In that | |
4577 | * case we can catch duplicate names easily. | |
4578 | */ | |
27c3a314 | 4579 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4580 | name = s->name; |
4581 | } else { | |
4582 | /* | |
4583 | * Create a unique name for the slab as a target | |
4584 | * for the symlinks. | |
4585 | */ | |
4586 | name = create_unique_id(s); | |
4587 | } | |
4588 | ||
27c3a314 | 4589 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4590 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4591 | if (err) { | |
4592 | kobject_put(&s->kobj); | |
81819f0f | 4593 | return err; |
1eada11c | 4594 | } |
81819f0f CL |
4595 | |
4596 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4597 | if (err) { |
4598 | kobject_del(&s->kobj); | |
4599 | kobject_put(&s->kobj); | |
81819f0f | 4600 | return err; |
5788d8ad | 4601 | } |
81819f0f CL |
4602 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4603 | if (!unmergeable) { | |
4604 | /* Setup first alias */ | |
4605 | sysfs_slab_alias(s, s->name); | |
4606 | kfree(name); | |
4607 | } | |
4608 | return 0; | |
4609 | } | |
4610 | ||
4611 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4612 | { | |
4613 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4614 | kobject_del(&s->kobj); | |
151c602f | 4615 | kobject_put(&s->kobj); |
81819f0f CL |
4616 | } |
4617 | ||
4618 | /* | |
4619 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4620 | * available lest we lose that information. |
81819f0f CL |
4621 | */ |
4622 | struct saved_alias { | |
4623 | struct kmem_cache *s; | |
4624 | const char *name; | |
4625 | struct saved_alias *next; | |
4626 | }; | |
4627 | ||
5af328a5 | 4628 | static struct saved_alias *alias_list; |
81819f0f CL |
4629 | |
4630 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4631 | { | |
4632 | struct saved_alias *al; | |
4633 | ||
4634 | if (slab_state == SYSFS) { | |
4635 | /* | |
4636 | * If we have a leftover link then remove it. | |
4637 | */ | |
27c3a314 GKH |
4638 | sysfs_remove_link(&slab_kset->kobj, name); |
4639 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4640 | } |
4641 | ||
4642 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4643 | if (!al) | |
4644 | return -ENOMEM; | |
4645 | ||
4646 | al->s = s; | |
4647 | al->name = name; | |
4648 | al->next = alias_list; | |
4649 | alias_list = al; | |
4650 | return 0; | |
4651 | } | |
4652 | ||
4653 | static int __init slab_sysfs_init(void) | |
4654 | { | |
5b95a4ac | 4655 | struct kmem_cache *s; |
81819f0f CL |
4656 | int err; |
4657 | ||
0ff21e46 | 4658 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4659 | if (!slab_kset) { |
81819f0f CL |
4660 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4661 | return -ENOSYS; | |
4662 | } | |
4663 | ||
26a7bd03 CL |
4664 | slab_state = SYSFS; |
4665 | ||
5b95a4ac | 4666 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4667 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4668 | if (err) |
4669 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4670 | " to sysfs\n", s->name); | |
26a7bd03 | 4671 | } |
81819f0f CL |
4672 | |
4673 | while (alias_list) { | |
4674 | struct saved_alias *al = alias_list; | |
4675 | ||
4676 | alias_list = alias_list->next; | |
4677 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4678 | if (err) |
4679 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4680 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4681 | kfree(al); |
4682 | } | |
4683 | ||
4684 | resiliency_test(); | |
4685 | return 0; | |
4686 | } | |
4687 | ||
4688 | __initcall(slab_sysfs_init); | |
81819f0f | 4689 | #endif |
57ed3eda PE |
4690 | |
4691 | /* | |
4692 | * The /proc/slabinfo ABI | |
4693 | */ | |
158a9624 | 4694 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4695 | static void print_slabinfo_header(struct seq_file *m) |
4696 | { | |
4697 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4698 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4699 | "<objperslab> <pagesperslab>"); | |
4700 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4701 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4702 | seq_putc(m, '\n'); | |
4703 | } | |
4704 | ||
4705 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4706 | { | |
4707 | loff_t n = *pos; | |
4708 | ||
4709 | down_read(&slub_lock); | |
4710 | if (!n) | |
4711 | print_slabinfo_header(m); | |
4712 | ||
4713 | return seq_list_start(&slab_caches, *pos); | |
4714 | } | |
4715 | ||
4716 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4717 | { | |
4718 | return seq_list_next(p, &slab_caches, pos); | |
4719 | } | |
4720 | ||
4721 | static void s_stop(struct seq_file *m, void *p) | |
4722 | { | |
4723 | up_read(&slub_lock); | |
4724 | } | |
4725 | ||
4726 | static int s_show(struct seq_file *m, void *p) | |
4727 | { | |
4728 | unsigned long nr_partials = 0; | |
4729 | unsigned long nr_slabs = 0; | |
4730 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4731 | unsigned long nr_objs = 0; |
4732 | unsigned long nr_free = 0; | |
57ed3eda PE |
4733 | struct kmem_cache *s; |
4734 | int node; | |
4735 | ||
4736 | s = list_entry(p, struct kmem_cache, list); | |
4737 | ||
4738 | for_each_online_node(node) { | |
4739 | struct kmem_cache_node *n = get_node(s, node); | |
4740 | ||
4741 | if (!n) | |
4742 | continue; | |
4743 | ||
4744 | nr_partials += n->nr_partial; | |
4745 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4746 | nr_objs += atomic_long_read(&n->total_objects); |
4747 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4748 | } |
4749 | ||
205ab99d | 4750 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4751 | |
4752 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4753 | nr_objs, s->size, oo_objects(s->oo), |
4754 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4755 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4756 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4757 | 0UL); | |
4758 | seq_putc(m, '\n'); | |
4759 | return 0; | |
4760 | } | |
4761 | ||
7b3c3a50 | 4762 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4763 | .start = s_start, |
4764 | .next = s_next, | |
4765 | .stop = s_stop, | |
4766 | .show = s_show, | |
4767 | }; | |
4768 | ||
7b3c3a50 AD |
4769 | static int slabinfo_open(struct inode *inode, struct file *file) |
4770 | { | |
4771 | return seq_open(file, &slabinfo_op); | |
4772 | } | |
4773 | ||
4774 | static const struct file_operations proc_slabinfo_operations = { | |
4775 | .open = slabinfo_open, | |
4776 | .read = seq_read, | |
4777 | .llseek = seq_lseek, | |
4778 | .release = seq_release, | |
4779 | }; | |
4780 | ||
4781 | static int __init slab_proc_init(void) | |
4782 | { | |
cf5d1131 | 4783 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4784 | return 0; |
4785 | } | |
4786 | module_init(slab_proc_init); | |
158a9624 | 4787 | #endif /* CONFIG_SLABINFO */ |