]>
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> | |
12 | #include <linux/module.h> | |
13 | #include <linux/bit_spinlock.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/bitops.h> | |
16 | #include <linux/slab.h> | |
7b3c3a50 | 17 | #include <linux/proc_fs.h> |
81819f0f CL |
18 | #include <linux/seq_file.h> |
19 | #include <linux/cpu.h> | |
20 | #include <linux/cpuset.h> | |
21 | #include <linux/mempolicy.h> | |
22 | #include <linux/ctype.h> | |
3ac7fe5a | 23 | #include <linux/debugobjects.h> |
81819f0f | 24 | #include <linux/kallsyms.h> |
b9049e23 | 25 | #include <linux/memory.h> |
f8bd2258 | 26 | #include <linux/math64.h> |
773ff60e | 27 | #include <linux/fault-inject.h> |
81819f0f CL |
28 | |
29 | /* | |
30 | * Lock order: | |
31 | * 1. slab_lock(page) | |
32 | * 2. slab->list_lock | |
33 | * | |
34 | * The slab_lock protects operations on the object of a particular | |
35 | * slab and its metadata in the page struct. If the slab lock | |
36 | * has been taken then no allocations nor frees can be performed | |
37 | * on the objects in the slab nor can the slab be added or removed | |
38 | * from the partial or full lists since this would mean modifying | |
39 | * the page_struct of the slab. | |
40 | * | |
41 | * The list_lock protects the partial and full list on each node and | |
42 | * the partial slab counter. If taken then no new slabs may be added or | |
43 | * removed from the lists nor make the number of partial slabs be modified. | |
44 | * (Note that the total number of slabs is an atomic value that may be | |
45 | * modified without taking the list lock). | |
46 | * | |
47 | * The list_lock is a centralized lock and thus we avoid taking it as | |
48 | * much as possible. As long as SLUB does not have to handle partial | |
49 | * slabs, operations can continue without any centralized lock. F.e. | |
50 | * allocating a long series of objects that fill up slabs does not require | |
51 | * the list lock. | |
52 | * | |
53 | * The lock order is sometimes inverted when we are trying to get a slab | |
54 | * off a list. We take the list_lock and then look for a page on the list | |
55 | * to use. While we do that objects in the slabs may be freed. We can | |
56 | * only operate on the slab if we have also taken the slab_lock. So we use | |
57 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
58 | * can occur anymore and we can use the slab for allocations etc. If the | |
59 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
60 | * we must stay away from it for a while since we may cause a bouncing | |
61 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
62 | * If all pages are busy then we may allocate a new slab instead of reusing | |
63 | * a partial slab. A new slab has noone operating on it and thus there is | |
64 | * no danger of cacheline contention. | |
65 | * | |
66 | * Interrupts are disabled during allocation and deallocation in order to | |
67 | * make the slab allocator safe to use in the context of an irq. In addition | |
68 | * interrupts are disabled to ensure that the processor does not change | |
69 | * while handling per_cpu slabs, due to kernel preemption. | |
70 | * | |
71 | * SLUB assigns one slab for allocation to each processor. | |
72 | * Allocations only occur from these slabs called cpu slabs. | |
73 | * | |
672bba3a CL |
74 | * Slabs with free elements are kept on a partial list and during regular |
75 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 76 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
77 | * We track full slabs for debugging purposes though because otherwise we |
78 | * cannot scan all objects. | |
81819f0f CL |
79 | * |
80 | * Slabs are freed when they become empty. Teardown and setup is | |
81 | * minimal so we rely on the page allocators per cpu caches for | |
82 | * fast frees and allocs. | |
83 | * | |
84 | * Overloading of page flags that are otherwise used for LRU management. | |
85 | * | |
4b6f0750 CL |
86 | * PageActive The slab is frozen and exempt from list processing. |
87 | * This means that the slab is dedicated to a purpose | |
88 | * such as satisfying allocations for a specific | |
89 | * processor. Objects may be freed in the slab while | |
90 | * it is frozen but slab_free will then skip the usual | |
91 | * list operations. It is up to the processor holding | |
92 | * the slab to integrate the slab into the slab lists | |
93 | * when the slab is no longer needed. | |
94 | * | |
95 | * One use of this flag is to mark slabs that are | |
96 | * used for allocations. Then such a slab becomes a cpu | |
97 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 98 | * freelist that allows lockless access to |
894b8788 CL |
99 | * free objects in addition to the regular freelist |
100 | * that requires the slab lock. | |
81819f0f CL |
101 | * |
102 | * PageError Slab requires special handling due to debug | |
103 | * options set. This moves slab handling out of | |
894b8788 | 104 | * the fast path and disables lockless freelists. |
81819f0f CL |
105 | */ |
106 | ||
5577bd8a | 107 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 108 | #define SLABDEBUG 1 |
5577bd8a CL |
109 | #else |
110 | #define SLABDEBUG 0 | |
111 | #endif | |
112 | ||
81819f0f CL |
113 | /* |
114 | * Issues still to be resolved: | |
115 | * | |
81819f0f CL |
116 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
117 | * | |
81819f0f CL |
118 | * - Variable sizing of the per node arrays |
119 | */ | |
120 | ||
121 | /* Enable to test recovery from slab corruption on boot */ | |
122 | #undef SLUB_RESILIENCY_TEST | |
123 | ||
2086d26a CL |
124 | /* |
125 | * Mininum number of partial slabs. These will be left on the partial | |
126 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
127 | */ | |
76be8950 | 128 | #define MIN_PARTIAL 5 |
e95eed57 | 129 | |
2086d26a CL |
130 | /* |
131 | * Maximum number of desirable partial slabs. | |
132 | * The existence of more partial slabs makes kmem_cache_shrink | |
133 | * sort the partial list by the number of objects in the. | |
134 | */ | |
135 | #define MAX_PARTIAL 10 | |
136 | ||
81819f0f CL |
137 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
138 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 139 | |
81819f0f CL |
140 | /* |
141 | * Set of flags that will prevent slab merging | |
142 | */ | |
143 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
144 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
145 | ||
146 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
147 | SLAB_CACHE_DMA) | |
148 | ||
149 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 150 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
151 | #endif |
152 | ||
153 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 154 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
155 | #endif |
156 | ||
210b5c06 CG |
157 | #define OO_SHIFT 16 |
158 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
159 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
160 | ||
81819f0f | 161 | /* Internal SLUB flags */ |
1ceef402 CL |
162 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
163 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
164 | |
165 | static int kmem_size = sizeof(struct kmem_cache); | |
166 | ||
167 | #ifdef CONFIG_SMP | |
168 | static struct notifier_block slab_notifier; | |
169 | #endif | |
170 | ||
171 | static enum { | |
172 | DOWN, /* No slab functionality available */ | |
173 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 174 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
175 | SYSFS /* Sysfs up */ |
176 | } slab_state = DOWN; | |
177 | ||
178 | /* A list of all slab caches on the system */ | |
179 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 180 | static LIST_HEAD(slab_caches); |
81819f0f | 181 | |
02cbc874 CL |
182 | /* |
183 | * Tracking user of a slab. | |
184 | */ | |
185 | struct track { | |
ce71e27c | 186 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
187 | int cpu; /* Was running on cpu */ |
188 | int pid; /* Pid context */ | |
189 | unsigned long when; /* When did the operation occur */ | |
190 | }; | |
191 | ||
192 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
193 | ||
f6acb635 | 194 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
195 | static int sysfs_slab_add(struct kmem_cache *); |
196 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
197 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 198 | |
81819f0f | 199 | #else |
0c710013 CL |
200 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
201 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
202 | { return 0; } | |
151c602f CL |
203 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
204 | { | |
205 | kfree(s); | |
206 | } | |
8ff12cfc | 207 | |
81819f0f CL |
208 | #endif |
209 | ||
8ff12cfc CL |
210 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
211 | { | |
212 | #ifdef CONFIG_SLUB_STATS | |
213 | c->stat[si]++; | |
214 | #endif | |
215 | } | |
216 | ||
81819f0f CL |
217 | /******************************************************************** |
218 | * Core slab cache functions | |
219 | *******************************************************************/ | |
220 | ||
221 | int slab_is_available(void) | |
222 | { | |
223 | return slab_state >= UP; | |
224 | } | |
225 | ||
226 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
227 | { | |
228 | #ifdef CONFIG_NUMA | |
229 | return s->node[node]; | |
230 | #else | |
231 | return &s->local_node; | |
232 | #endif | |
233 | } | |
234 | ||
dfb4f096 CL |
235 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
236 | { | |
4c93c355 CL |
237 | #ifdef CONFIG_SMP |
238 | return s->cpu_slab[cpu]; | |
239 | #else | |
240 | return &s->cpu_slab; | |
241 | #endif | |
dfb4f096 CL |
242 | } |
243 | ||
6446faa2 | 244 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
245 | static inline int check_valid_pointer(struct kmem_cache *s, |
246 | struct page *page, const void *object) | |
247 | { | |
248 | void *base; | |
249 | ||
a973e9dd | 250 | if (!object) |
02cbc874 CL |
251 | return 1; |
252 | ||
a973e9dd | 253 | base = page_address(page); |
39b26464 | 254 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
255 | (object - base) % s->size) { |
256 | return 0; | |
257 | } | |
258 | ||
259 | return 1; | |
260 | } | |
261 | ||
7656c72b CL |
262 | /* |
263 | * Slow version of get and set free pointer. | |
264 | * | |
265 | * This version requires touching the cache lines of kmem_cache which | |
266 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
267 | * from the page struct. | |
268 | */ | |
269 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
270 | { | |
271 | return *(void **)(object + s->offset); | |
272 | } | |
273 | ||
274 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
275 | { | |
276 | *(void **)(object + s->offset) = fp; | |
277 | } | |
278 | ||
279 | /* Loop over all objects in a slab */ | |
224a88be CL |
280 | #define for_each_object(__p, __s, __addr, __objects) \ |
281 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
282 | __p += (__s)->size) |
283 | ||
284 | /* Scan freelist */ | |
285 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 286 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
287 | |
288 | /* Determine object index from a given position */ | |
289 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
290 | { | |
291 | return (p - addr) / s->size; | |
292 | } | |
293 | ||
834f3d11 CL |
294 | static inline struct kmem_cache_order_objects oo_make(int order, |
295 | unsigned long size) | |
296 | { | |
297 | struct kmem_cache_order_objects x = { | |
210b5c06 | 298 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
299 | }; |
300 | ||
301 | return x; | |
302 | } | |
303 | ||
304 | static inline int oo_order(struct kmem_cache_order_objects x) | |
305 | { | |
210b5c06 | 306 | return x.x >> OO_SHIFT; |
834f3d11 CL |
307 | } |
308 | ||
309 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
310 | { | |
210b5c06 | 311 | return x.x & OO_MASK; |
834f3d11 CL |
312 | } |
313 | ||
41ecc55b CL |
314 | #ifdef CONFIG_SLUB_DEBUG |
315 | /* | |
316 | * Debug settings: | |
317 | */ | |
f0630fff CL |
318 | #ifdef CONFIG_SLUB_DEBUG_ON |
319 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
320 | #else | |
41ecc55b | 321 | static int slub_debug; |
f0630fff | 322 | #endif |
41ecc55b CL |
323 | |
324 | static char *slub_debug_slabs; | |
325 | ||
81819f0f CL |
326 | /* |
327 | * Object debugging | |
328 | */ | |
329 | static void print_section(char *text, u8 *addr, unsigned int length) | |
330 | { | |
331 | int i, offset; | |
332 | int newline = 1; | |
333 | char ascii[17]; | |
334 | ||
335 | ascii[16] = 0; | |
336 | ||
337 | for (i = 0; i < length; i++) { | |
338 | if (newline) { | |
24922684 | 339 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
340 | newline = 0; |
341 | } | |
06428780 | 342 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
343 | offset = i % 16; |
344 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
345 | if (offset == 15) { | |
06428780 | 346 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
347 | newline = 1; |
348 | } | |
349 | } | |
350 | if (!newline) { | |
351 | i %= 16; | |
352 | while (i < 16) { | |
06428780 | 353 | printk(KERN_CONT " "); |
81819f0f CL |
354 | ascii[i] = ' '; |
355 | i++; | |
356 | } | |
06428780 | 357 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
358 | } |
359 | } | |
360 | ||
81819f0f CL |
361 | static struct track *get_track(struct kmem_cache *s, void *object, |
362 | enum track_item alloc) | |
363 | { | |
364 | struct track *p; | |
365 | ||
366 | if (s->offset) | |
367 | p = object + s->offset + sizeof(void *); | |
368 | else | |
369 | p = object + s->inuse; | |
370 | ||
371 | return p + alloc; | |
372 | } | |
373 | ||
374 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 375 | enum track_item alloc, unsigned long addr) |
81819f0f CL |
376 | { |
377 | struct track *p; | |
378 | ||
379 | if (s->offset) | |
380 | p = object + s->offset + sizeof(void *); | |
381 | else | |
382 | p = object + s->inuse; | |
383 | ||
384 | p += alloc; | |
385 | if (addr) { | |
386 | p->addr = addr; | |
387 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 388 | p->pid = current->pid; |
81819f0f CL |
389 | p->when = jiffies; |
390 | } else | |
391 | memset(p, 0, sizeof(struct track)); | |
392 | } | |
393 | ||
81819f0f CL |
394 | static void init_tracking(struct kmem_cache *s, void *object) |
395 | { | |
24922684 CL |
396 | if (!(s->flags & SLAB_STORE_USER)) |
397 | return; | |
398 | ||
ce71e27c EGM |
399 | set_track(s, object, TRACK_FREE, 0UL); |
400 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
401 | } |
402 | ||
403 | static void print_track(const char *s, struct track *t) | |
404 | { | |
405 | if (!t->addr) | |
406 | return; | |
407 | ||
7daf705f | 408 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 409 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
410 | } |
411 | ||
412 | static void print_tracking(struct kmem_cache *s, void *object) | |
413 | { | |
414 | if (!(s->flags & SLAB_STORE_USER)) | |
415 | return; | |
416 | ||
417 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
418 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
419 | } | |
420 | ||
421 | static void print_page_info(struct page *page) | |
422 | { | |
39b26464 CL |
423 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
424 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
425 | |
426 | } | |
427 | ||
428 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
429 | { | |
430 | va_list args; | |
431 | char buf[100]; | |
432 | ||
433 | va_start(args, fmt); | |
434 | vsnprintf(buf, sizeof(buf), fmt, args); | |
435 | va_end(args); | |
436 | printk(KERN_ERR "========================================" | |
437 | "=====================================\n"); | |
438 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
439 | printk(KERN_ERR "----------------------------------------" | |
440 | "-------------------------------------\n\n"); | |
81819f0f CL |
441 | } |
442 | ||
24922684 CL |
443 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
444 | { | |
445 | va_list args; | |
446 | char buf[100]; | |
447 | ||
448 | va_start(args, fmt); | |
449 | vsnprintf(buf, sizeof(buf), fmt, args); | |
450 | va_end(args); | |
451 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
452 | } | |
453 | ||
454 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
455 | { |
456 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 457 | u8 *addr = page_address(page); |
24922684 CL |
458 | |
459 | print_tracking(s, p); | |
460 | ||
461 | print_page_info(page); | |
462 | ||
463 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
464 | p, p - addr, get_freepointer(s, p)); | |
465 | ||
466 | if (p > addr + 16) | |
467 | print_section("Bytes b4", p - 16, 16); | |
468 | ||
0ebd652b | 469 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
470 | |
471 | if (s->flags & SLAB_RED_ZONE) | |
472 | print_section("Redzone", p + s->objsize, | |
473 | s->inuse - s->objsize); | |
474 | ||
81819f0f CL |
475 | if (s->offset) |
476 | off = s->offset + sizeof(void *); | |
477 | else | |
478 | off = s->inuse; | |
479 | ||
24922684 | 480 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 481 | off += 2 * sizeof(struct track); |
81819f0f CL |
482 | |
483 | if (off != s->size) | |
484 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
485 | print_section("Padding", p + off, s->size - off); |
486 | ||
487 | dump_stack(); | |
81819f0f CL |
488 | } |
489 | ||
490 | static void object_err(struct kmem_cache *s, struct page *page, | |
491 | u8 *object, char *reason) | |
492 | { | |
3dc50637 | 493 | slab_bug(s, "%s", reason); |
24922684 | 494 | print_trailer(s, page, object); |
81819f0f CL |
495 | } |
496 | ||
24922684 | 497 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
498 | { |
499 | va_list args; | |
500 | char buf[100]; | |
501 | ||
24922684 CL |
502 | va_start(args, fmt); |
503 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 504 | va_end(args); |
3dc50637 | 505 | slab_bug(s, "%s", buf); |
24922684 | 506 | print_page_info(page); |
81819f0f CL |
507 | dump_stack(); |
508 | } | |
509 | ||
510 | static void init_object(struct kmem_cache *s, void *object, int active) | |
511 | { | |
512 | u8 *p = object; | |
513 | ||
514 | if (s->flags & __OBJECT_POISON) { | |
515 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 516 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
517 | } |
518 | ||
519 | if (s->flags & SLAB_RED_ZONE) | |
520 | memset(p + s->objsize, | |
521 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
522 | s->inuse - s->objsize); | |
523 | } | |
524 | ||
24922684 | 525 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
526 | { |
527 | while (bytes) { | |
528 | if (*start != (u8)value) | |
24922684 | 529 | return start; |
81819f0f CL |
530 | start++; |
531 | bytes--; | |
532 | } | |
24922684 CL |
533 | return NULL; |
534 | } | |
535 | ||
536 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
537 | void *from, void *to) | |
538 | { | |
539 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
540 | memset(from, data, to - from); | |
541 | } | |
542 | ||
543 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
544 | u8 *object, char *what, | |
06428780 | 545 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
546 | { |
547 | u8 *fault; | |
548 | u8 *end; | |
549 | ||
550 | fault = check_bytes(start, value, bytes); | |
551 | if (!fault) | |
552 | return 1; | |
553 | ||
554 | end = start + bytes; | |
555 | while (end > fault && end[-1] == value) | |
556 | end--; | |
557 | ||
558 | slab_bug(s, "%s overwritten", what); | |
559 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
560 | fault, end - 1, fault[0], value); | |
561 | print_trailer(s, page, object); | |
562 | ||
563 | restore_bytes(s, what, value, fault, end); | |
564 | return 0; | |
81819f0f CL |
565 | } |
566 | ||
81819f0f CL |
567 | /* |
568 | * Object layout: | |
569 | * | |
570 | * object address | |
571 | * Bytes of the object to be managed. | |
572 | * If the freepointer may overlay the object then the free | |
573 | * pointer is the first word of the object. | |
672bba3a | 574 | * |
81819f0f CL |
575 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
576 | * 0xa5 (POISON_END) | |
577 | * | |
578 | * object + s->objsize | |
579 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
580 | * Padding is extended by another word if Redzoning is enabled and |
581 | * objsize == inuse. | |
582 | * | |
81819f0f CL |
583 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
584 | * 0xcc (RED_ACTIVE) for objects in use. | |
585 | * | |
586 | * object + s->inuse | |
672bba3a CL |
587 | * Meta data starts here. |
588 | * | |
81819f0f CL |
589 | * A. Free pointer (if we cannot overwrite object on free) |
590 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 591 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 592 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
593 | * before the word boundary. |
594 | * | |
595 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
596 | * |
597 | * object + s->size | |
672bba3a | 598 | * Nothing is used beyond s->size. |
81819f0f | 599 | * |
672bba3a CL |
600 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
601 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
602 | * may be used with merged slabcaches. |
603 | */ | |
604 | ||
81819f0f CL |
605 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
606 | { | |
607 | unsigned long off = s->inuse; /* The end of info */ | |
608 | ||
609 | if (s->offset) | |
610 | /* Freepointer is placed after the object. */ | |
611 | off += sizeof(void *); | |
612 | ||
613 | if (s->flags & SLAB_STORE_USER) | |
614 | /* We also have user information there */ | |
615 | off += 2 * sizeof(struct track); | |
616 | ||
617 | if (s->size == off) | |
618 | return 1; | |
619 | ||
24922684 CL |
620 | return check_bytes_and_report(s, page, p, "Object padding", |
621 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
622 | } |
623 | ||
39b26464 | 624 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
625 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
626 | { | |
24922684 CL |
627 | u8 *start; |
628 | u8 *fault; | |
629 | u8 *end; | |
630 | int length; | |
631 | int remainder; | |
81819f0f CL |
632 | |
633 | if (!(s->flags & SLAB_POISON)) | |
634 | return 1; | |
635 | ||
a973e9dd | 636 | start = page_address(page); |
834f3d11 | 637 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
638 | end = start + length; |
639 | remainder = length % s->size; | |
81819f0f CL |
640 | if (!remainder) |
641 | return 1; | |
642 | ||
39b26464 | 643 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
644 | if (!fault) |
645 | return 1; | |
646 | while (end > fault && end[-1] == POISON_INUSE) | |
647 | end--; | |
648 | ||
649 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 650 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
651 | |
652 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
653 | return 0; | |
81819f0f CL |
654 | } |
655 | ||
656 | static int check_object(struct kmem_cache *s, struct page *page, | |
657 | void *object, int active) | |
658 | { | |
659 | u8 *p = object; | |
660 | u8 *endobject = object + s->objsize; | |
661 | ||
662 | if (s->flags & SLAB_RED_ZONE) { | |
663 | unsigned int red = | |
664 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
665 | ||
24922684 CL |
666 | if (!check_bytes_and_report(s, page, object, "Redzone", |
667 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 668 | return 0; |
81819f0f | 669 | } else { |
3adbefee IM |
670 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
671 | check_bytes_and_report(s, page, p, "Alignment padding", | |
672 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
673 | } | |
81819f0f CL |
674 | } |
675 | ||
676 | if (s->flags & SLAB_POISON) { | |
677 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
678 | (!check_bytes_and_report(s, page, p, "Poison", p, |
679 | POISON_FREE, s->objsize - 1) || | |
680 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 681 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 682 | return 0; |
81819f0f CL |
683 | /* |
684 | * check_pad_bytes cleans up on its own. | |
685 | */ | |
686 | check_pad_bytes(s, page, p); | |
687 | } | |
688 | ||
689 | if (!s->offset && active) | |
690 | /* | |
691 | * Object and freepointer overlap. Cannot check | |
692 | * freepointer while object is allocated. | |
693 | */ | |
694 | return 1; | |
695 | ||
696 | /* Check free pointer validity */ | |
697 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
698 | object_err(s, page, p, "Freepointer corrupt"); | |
699 | /* | |
9f6c708e | 700 | * No choice but to zap it and thus lose the remainder |
81819f0f | 701 | * of the free objects in this slab. May cause |
672bba3a | 702 | * another error because the object count is now wrong. |
81819f0f | 703 | */ |
a973e9dd | 704 | set_freepointer(s, p, NULL); |
81819f0f CL |
705 | return 0; |
706 | } | |
707 | return 1; | |
708 | } | |
709 | ||
710 | static int check_slab(struct kmem_cache *s, struct page *page) | |
711 | { | |
39b26464 CL |
712 | int maxobj; |
713 | ||
81819f0f CL |
714 | VM_BUG_ON(!irqs_disabled()); |
715 | ||
716 | if (!PageSlab(page)) { | |
24922684 | 717 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
718 | return 0; |
719 | } | |
39b26464 CL |
720 | |
721 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
722 | if (page->objects > maxobj) { | |
723 | slab_err(s, page, "objects %u > max %u", | |
724 | s->name, page->objects, maxobj); | |
725 | return 0; | |
726 | } | |
727 | if (page->inuse > page->objects) { | |
24922684 | 728 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 729 | s->name, page->inuse, page->objects); |
81819f0f CL |
730 | return 0; |
731 | } | |
732 | /* Slab_pad_check fixes things up after itself */ | |
733 | slab_pad_check(s, page); | |
734 | return 1; | |
735 | } | |
736 | ||
737 | /* | |
672bba3a CL |
738 | * Determine if a certain object on a page is on the freelist. Must hold the |
739 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
740 | */ |
741 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
742 | { | |
743 | int nr = 0; | |
744 | void *fp = page->freelist; | |
745 | void *object = NULL; | |
224a88be | 746 | unsigned long max_objects; |
81819f0f | 747 | |
39b26464 | 748 | while (fp && nr <= page->objects) { |
81819f0f CL |
749 | if (fp == search) |
750 | return 1; | |
751 | if (!check_valid_pointer(s, page, fp)) { | |
752 | if (object) { | |
753 | object_err(s, page, object, | |
754 | "Freechain corrupt"); | |
a973e9dd | 755 | set_freepointer(s, object, NULL); |
81819f0f CL |
756 | break; |
757 | } else { | |
24922684 | 758 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 759 | page->freelist = NULL; |
39b26464 | 760 | page->inuse = page->objects; |
24922684 | 761 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
762 | return 0; |
763 | } | |
764 | break; | |
765 | } | |
766 | object = fp; | |
767 | fp = get_freepointer(s, object); | |
768 | nr++; | |
769 | } | |
770 | ||
224a88be | 771 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
772 | if (max_objects > MAX_OBJS_PER_PAGE) |
773 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
774 | |
775 | if (page->objects != max_objects) { | |
776 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
777 | "should be %d", page->objects, max_objects); | |
778 | page->objects = max_objects; | |
779 | slab_fix(s, "Number of objects adjusted."); | |
780 | } | |
39b26464 | 781 | if (page->inuse != page->objects - nr) { |
70d71228 | 782 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
783 | "counted were %d", page->inuse, page->objects - nr); |
784 | page->inuse = page->objects - nr; | |
24922684 | 785 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
786 | } |
787 | return search == NULL; | |
788 | } | |
789 | ||
0121c619 CL |
790 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
791 | int alloc) | |
3ec09742 CL |
792 | { |
793 | if (s->flags & SLAB_TRACE) { | |
794 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
795 | s->name, | |
796 | alloc ? "alloc" : "free", | |
797 | object, page->inuse, | |
798 | page->freelist); | |
799 | ||
800 | if (!alloc) | |
801 | print_section("Object", (void *)object, s->objsize); | |
802 | ||
803 | dump_stack(); | |
804 | } | |
805 | } | |
806 | ||
643b1138 | 807 | /* |
672bba3a | 808 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 809 | */ |
e95eed57 | 810 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 811 | { |
643b1138 CL |
812 | spin_lock(&n->list_lock); |
813 | list_add(&page->lru, &n->full); | |
814 | spin_unlock(&n->list_lock); | |
815 | } | |
816 | ||
817 | static void remove_full(struct kmem_cache *s, struct page *page) | |
818 | { | |
819 | struct kmem_cache_node *n; | |
820 | ||
821 | if (!(s->flags & SLAB_STORE_USER)) | |
822 | return; | |
823 | ||
824 | n = get_node(s, page_to_nid(page)); | |
825 | ||
826 | spin_lock(&n->list_lock); | |
827 | list_del(&page->lru); | |
828 | spin_unlock(&n->list_lock); | |
829 | } | |
830 | ||
0f389ec6 CL |
831 | /* Tracking of the number of slabs for debugging purposes */ |
832 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
833 | { | |
834 | struct kmem_cache_node *n = get_node(s, node); | |
835 | ||
836 | return atomic_long_read(&n->nr_slabs); | |
837 | } | |
838 | ||
205ab99d | 839 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
840 | { |
841 | struct kmem_cache_node *n = get_node(s, node); | |
842 | ||
843 | /* | |
844 | * May be called early in order to allocate a slab for the | |
845 | * kmem_cache_node structure. Solve the chicken-egg | |
846 | * dilemma by deferring the increment of the count during | |
847 | * bootstrap (see early_kmem_cache_node_alloc). | |
848 | */ | |
205ab99d | 849 | if (!NUMA_BUILD || n) { |
0f389ec6 | 850 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
851 | atomic_long_add(objects, &n->total_objects); |
852 | } | |
0f389ec6 | 853 | } |
205ab99d | 854 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
855 | { |
856 | struct kmem_cache_node *n = get_node(s, node); | |
857 | ||
858 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 859 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
860 | } |
861 | ||
862 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
863 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
864 | void *object) | |
865 | { | |
866 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
867 | return; | |
868 | ||
869 | init_object(s, object, 0); | |
870 | init_tracking(s, object); | |
871 | } | |
872 | ||
873 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 874 | void *object, unsigned long addr) |
81819f0f CL |
875 | { |
876 | if (!check_slab(s, page)) | |
877 | goto bad; | |
878 | ||
d692ef6d | 879 | if (!on_freelist(s, page, object)) { |
24922684 | 880 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 881 | goto bad; |
81819f0f CL |
882 | } |
883 | ||
884 | if (!check_valid_pointer(s, page, object)) { | |
885 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 886 | goto bad; |
81819f0f CL |
887 | } |
888 | ||
d692ef6d | 889 | if (!check_object(s, page, object, 0)) |
81819f0f | 890 | goto bad; |
81819f0f | 891 | |
3ec09742 CL |
892 | /* Success perform special debug activities for allocs */ |
893 | if (s->flags & SLAB_STORE_USER) | |
894 | set_track(s, object, TRACK_ALLOC, addr); | |
895 | trace(s, page, object, 1); | |
896 | init_object(s, object, 1); | |
81819f0f | 897 | return 1; |
3ec09742 | 898 | |
81819f0f CL |
899 | bad: |
900 | if (PageSlab(page)) { | |
901 | /* | |
902 | * If this is a slab page then lets do the best we can | |
903 | * to avoid issues in the future. Marking all objects | |
672bba3a | 904 | * as used avoids touching the remaining objects. |
81819f0f | 905 | */ |
24922684 | 906 | slab_fix(s, "Marking all objects used"); |
39b26464 | 907 | page->inuse = page->objects; |
a973e9dd | 908 | page->freelist = NULL; |
81819f0f CL |
909 | } |
910 | return 0; | |
911 | } | |
912 | ||
3ec09742 | 913 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 914 | void *object, unsigned long addr) |
81819f0f CL |
915 | { |
916 | if (!check_slab(s, page)) | |
917 | goto fail; | |
918 | ||
919 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 920 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
921 | goto fail; |
922 | } | |
923 | ||
924 | if (on_freelist(s, page, object)) { | |
24922684 | 925 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
926 | goto fail; |
927 | } | |
928 | ||
929 | if (!check_object(s, page, object, 1)) | |
930 | return 0; | |
931 | ||
932 | if (unlikely(s != page->slab)) { | |
3adbefee | 933 | if (!PageSlab(page)) { |
70d71228 CL |
934 | slab_err(s, page, "Attempt to free object(0x%p) " |
935 | "outside of slab", object); | |
3adbefee | 936 | } else if (!page->slab) { |
81819f0f | 937 | printk(KERN_ERR |
70d71228 | 938 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 939 | object); |
70d71228 | 940 | dump_stack(); |
06428780 | 941 | } else |
24922684 CL |
942 | object_err(s, page, object, |
943 | "page slab pointer corrupt."); | |
81819f0f CL |
944 | goto fail; |
945 | } | |
3ec09742 CL |
946 | |
947 | /* Special debug activities for freeing objects */ | |
8a38082d | 948 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
949 | remove_full(s, page); |
950 | if (s->flags & SLAB_STORE_USER) | |
951 | set_track(s, object, TRACK_FREE, addr); | |
952 | trace(s, page, object, 0); | |
953 | init_object(s, object, 0); | |
81819f0f | 954 | return 1; |
3ec09742 | 955 | |
81819f0f | 956 | fail: |
24922684 | 957 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
958 | return 0; |
959 | } | |
960 | ||
41ecc55b CL |
961 | static int __init setup_slub_debug(char *str) |
962 | { | |
f0630fff CL |
963 | slub_debug = DEBUG_DEFAULT_FLAGS; |
964 | if (*str++ != '=' || !*str) | |
965 | /* | |
966 | * No options specified. Switch on full debugging. | |
967 | */ | |
968 | goto out; | |
969 | ||
970 | if (*str == ',') | |
971 | /* | |
972 | * No options but restriction on slabs. This means full | |
973 | * debugging for slabs matching a pattern. | |
974 | */ | |
975 | goto check_slabs; | |
976 | ||
977 | slub_debug = 0; | |
978 | if (*str == '-') | |
979 | /* | |
980 | * Switch off all debugging measures. | |
981 | */ | |
982 | goto out; | |
983 | ||
984 | /* | |
985 | * Determine which debug features should be switched on | |
986 | */ | |
06428780 | 987 | for (; *str && *str != ','; str++) { |
f0630fff CL |
988 | switch (tolower(*str)) { |
989 | case 'f': | |
990 | slub_debug |= SLAB_DEBUG_FREE; | |
991 | break; | |
992 | case 'z': | |
993 | slub_debug |= SLAB_RED_ZONE; | |
994 | break; | |
995 | case 'p': | |
996 | slub_debug |= SLAB_POISON; | |
997 | break; | |
998 | case 'u': | |
999 | slub_debug |= SLAB_STORE_USER; | |
1000 | break; | |
1001 | case 't': | |
1002 | slub_debug |= SLAB_TRACE; | |
1003 | break; | |
1004 | default: | |
1005 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1006 | "unknown. skipped\n", *str); |
f0630fff | 1007 | } |
41ecc55b CL |
1008 | } |
1009 | ||
f0630fff | 1010 | check_slabs: |
41ecc55b CL |
1011 | if (*str == ',') |
1012 | slub_debug_slabs = str + 1; | |
f0630fff | 1013 | out: |
41ecc55b CL |
1014 | return 1; |
1015 | } | |
1016 | ||
1017 | __setup("slub_debug", setup_slub_debug); | |
1018 | ||
ba0268a8 CL |
1019 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1020 | unsigned long flags, const char *name, | |
51cc5068 | 1021 | void (*ctor)(void *)) |
41ecc55b CL |
1022 | { |
1023 | /* | |
e153362a | 1024 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1025 | */ |
e153362a CL |
1026 | if (slub_debug && (!slub_debug_slabs || |
1027 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1028 | flags |= slub_debug; | |
ba0268a8 CL |
1029 | |
1030 | return flags; | |
41ecc55b CL |
1031 | } |
1032 | #else | |
3ec09742 CL |
1033 | static inline void setup_object_debug(struct kmem_cache *s, |
1034 | struct page *page, void *object) {} | |
41ecc55b | 1035 | |
3ec09742 | 1036 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1037 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1038 | |
3ec09742 | 1039 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1040 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1041 | |
41ecc55b CL |
1042 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1043 | { return 1; } | |
1044 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1045 | void *object, int active) { return 1; } | |
3ec09742 | 1046 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1047 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1048 | unsigned long flags, const char *name, | |
51cc5068 | 1049 | void (*ctor)(void *)) |
ba0268a8 CL |
1050 | { |
1051 | return flags; | |
1052 | } | |
41ecc55b | 1053 | #define slub_debug 0 |
0f389ec6 CL |
1054 | |
1055 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1056 | { return 0; } | |
205ab99d CL |
1057 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1058 | int objects) {} | |
1059 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1060 | int objects) {} | |
41ecc55b | 1061 | #endif |
205ab99d | 1062 | |
81819f0f CL |
1063 | /* |
1064 | * Slab allocation and freeing | |
1065 | */ | |
65c3376a CL |
1066 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1067 | struct kmem_cache_order_objects oo) | |
1068 | { | |
1069 | int order = oo_order(oo); | |
1070 | ||
1071 | if (node == -1) | |
1072 | return alloc_pages(flags, order); | |
1073 | else | |
1074 | return alloc_pages_node(node, flags, order); | |
1075 | } | |
1076 | ||
81819f0f CL |
1077 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1078 | { | |
06428780 | 1079 | struct page *page; |
834f3d11 | 1080 | struct kmem_cache_order_objects oo = s->oo; |
81819f0f | 1081 | |
b7a49f0d | 1082 | flags |= s->allocflags; |
e12ba74d | 1083 | |
65c3376a CL |
1084 | page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node, |
1085 | oo); | |
1086 | if (unlikely(!page)) { | |
1087 | oo = s->min; | |
1088 | /* | |
1089 | * Allocation may have failed due to fragmentation. | |
1090 | * Try a lower order alloc if possible | |
1091 | */ | |
1092 | page = alloc_slab_page(flags, node, oo); | |
1093 | if (!page) | |
1094 | return NULL; | |
81819f0f | 1095 | |
65c3376a CL |
1096 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1097 | } | |
834f3d11 | 1098 | page->objects = oo_objects(oo); |
81819f0f CL |
1099 | mod_zone_page_state(page_zone(page), |
1100 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1101 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1102 | 1 << oo_order(oo)); |
81819f0f CL |
1103 | |
1104 | return page; | |
1105 | } | |
1106 | ||
1107 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1108 | void *object) | |
1109 | { | |
3ec09742 | 1110 | setup_object_debug(s, page, object); |
4f104934 | 1111 | if (unlikely(s->ctor)) |
51cc5068 | 1112 | s->ctor(object); |
81819f0f CL |
1113 | } |
1114 | ||
1115 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1116 | { | |
1117 | struct page *page; | |
81819f0f | 1118 | void *start; |
81819f0f CL |
1119 | void *last; |
1120 | void *p; | |
1121 | ||
6cb06229 | 1122 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1123 | |
6cb06229 CL |
1124 | page = allocate_slab(s, |
1125 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1126 | if (!page) |
1127 | goto out; | |
1128 | ||
205ab99d | 1129 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1130 | page->slab = s; |
1131 | page->flags |= 1 << PG_slab; | |
1132 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1133 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1134 | __SetPageSlubDebug(page); |
81819f0f CL |
1135 | |
1136 | start = page_address(page); | |
81819f0f CL |
1137 | |
1138 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1139 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1140 | |
1141 | last = start; | |
224a88be | 1142 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1143 | setup_object(s, page, last); |
1144 | set_freepointer(s, last, p); | |
1145 | last = p; | |
1146 | } | |
1147 | setup_object(s, page, last); | |
a973e9dd | 1148 | set_freepointer(s, last, NULL); |
81819f0f CL |
1149 | |
1150 | page->freelist = start; | |
1151 | page->inuse = 0; | |
1152 | out: | |
81819f0f CL |
1153 | return page; |
1154 | } | |
1155 | ||
1156 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1157 | { | |
834f3d11 CL |
1158 | int order = compound_order(page); |
1159 | int pages = 1 << order; | |
81819f0f | 1160 | |
8a38082d | 1161 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1162 | void *p; |
1163 | ||
1164 | slab_pad_check(s, page); | |
224a88be CL |
1165 | for_each_object(p, s, page_address(page), |
1166 | page->objects) | |
81819f0f | 1167 | check_object(s, page, p, 0); |
8a38082d | 1168 | __ClearPageSlubDebug(page); |
81819f0f CL |
1169 | } |
1170 | ||
1171 | mod_zone_page_state(page_zone(page), | |
1172 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1173 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1174 | -pages); |
81819f0f | 1175 | |
49bd5221 CL |
1176 | __ClearPageSlab(page); |
1177 | reset_page_mapcount(page); | |
834f3d11 | 1178 | __free_pages(page, order); |
81819f0f CL |
1179 | } |
1180 | ||
1181 | static void rcu_free_slab(struct rcu_head *h) | |
1182 | { | |
1183 | struct page *page; | |
1184 | ||
1185 | page = container_of((struct list_head *)h, struct page, lru); | |
1186 | __free_slab(page->slab, page); | |
1187 | } | |
1188 | ||
1189 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1190 | { | |
1191 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1192 | /* | |
1193 | * RCU free overloads the RCU head over the LRU | |
1194 | */ | |
1195 | struct rcu_head *head = (void *)&page->lru; | |
1196 | ||
1197 | call_rcu(head, rcu_free_slab); | |
1198 | } else | |
1199 | __free_slab(s, page); | |
1200 | } | |
1201 | ||
1202 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1203 | { | |
205ab99d | 1204 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1205 | free_slab(s, page); |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Per slab locking using the pagelock | |
1210 | */ | |
1211 | static __always_inline void slab_lock(struct page *page) | |
1212 | { | |
1213 | bit_spin_lock(PG_locked, &page->flags); | |
1214 | } | |
1215 | ||
1216 | static __always_inline void slab_unlock(struct page *page) | |
1217 | { | |
a76d3546 | 1218 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1219 | } |
1220 | ||
1221 | static __always_inline int slab_trylock(struct page *page) | |
1222 | { | |
1223 | int rc = 1; | |
1224 | ||
1225 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1226 | return rc; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Management of partially allocated slabs | |
1231 | */ | |
7c2e132c CL |
1232 | static void add_partial(struct kmem_cache_node *n, |
1233 | struct page *page, int tail) | |
81819f0f | 1234 | { |
e95eed57 CL |
1235 | spin_lock(&n->list_lock); |
1236 | n->nr_partial++; | |
7c2e132c CL |
1237 | if (tail) |
1238 | list_add_tail(&page->lru, &n->partial); | |
1239 | else | |
1240 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1241 | spin_unlock(&n->list_lock); |
1242 | } | |
1243 | ||
0121c619 | 1244 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1245 | { |
1246 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1247 | ||
1248 | spin_lock(&n->list_lock); | |
1249 | list_del(&page->lru); | |
1250 | n->nr_partial--; | |
1251 | spin_unlock(&n->list_lock); | |
1252 | } | |
1253 | ||
1254 | /* | |
672bba3a | 1255 | * Lock slab and remove from the partial list. |
81819f0f | 1256 | * |
672bba3a | 1257 | * Must hold list_lock. |
81819f0f | 1258 | */ |
0121c619 CL |
1259 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1260 | struct page *page) | |
81819f0f CL |
1261 | { |
1262 | if (slab_trylock(page)) { | |
1263 | list_del(&page->lru); | |
1264 | n->nr_partial--; | |
8a38082d | 1265 | __SetPageSlubFrozen(page); |
81819f0f CL |
1266 | return 1; |
1267 | } | |
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | /* | |
672bba3a | 1272 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1273 | */ |
1274 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1275 | { | |
1276 | struct page *page; | |
1277 | ||
1278 | /* | |
1279 | * Racy check. If we mistakenly see no partial slabs then we | |
1280 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1281 | * partial slab and there is none available then get_partials() |
1282 | * will return NULL. | |
81819f0f CL |
1283 | */ |
1284 | if (!n || !n->nr_partial) | |
1285 | return NULL; | |
1286 | ||
1287 | spin_lock(&n->list_lock); | |
1288 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1289 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1290 | goto out; |
1291 | page = NULL; | |
1292 | out: | |
1293 | spin_unlock(&n->list_lock); | |
1294 | return page; | |
1295 | } | |
1296 | ||
1297 | /* | |
672bba3a | 1298 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1299 | */ |
1300 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1301 | { | |
1302 | #ifdef CONFIG_NUMA | |
1303 | struct zonelist *zonelist; | |
dd1a239f | 1304 | struct zoneref *z; |
54a6eb5c MG |
1305 | struct zone *zone; |
1306 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1307 | struct page *page; |
1308 | ||
1309 | /* | |
672bba3a CL |
1310 | * The defrag ratio allows a configuration of the tradeoffs between |
1311 | * inter node defragmentation and node local allocations. A lower | |
1312 | * defrag_ratio increases the tendency to do local allocations | |
1313 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1314 | * |
672bba3a CL |
1315 | * If the defrag_ratio is set to 0 then kmalloc() always |
1316 | * returns node local objects. If the ratio is higher then kmalloc() | |
1317 | * may return off node objects because partial slabs are obtained | |
1318 | * from other nodes and filled up. | |
81819f0f | 1319 | * |
6446faa2 | 1320 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1321 | * defrag_ratio = 1000) then every (well almost) allocation will |
1322 | * first attempt to defrag slab caches on other nodes. This means | |
1323 | * scanning over all nodes to look for partial slabs which may be | |
1324 | * expensive if we do it every time we are trying to find a slab | |
1325 | * with available objects. | |
81819f0f | 1326 | */ |
9824601e CL |
1327 | if (!s->remote_node_defrag_ratio || |
1328 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1329 | return NULL; |
1330 | ||
0e88460d | 1331 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1332 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1333 | struct kmem_cache_node *n; |
1334 | ||
54a6eb5c | 1335 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1336 | |
54a6eb5c | 1337 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
5595cffc | 1338 | n->nr_partial > n->min_partial) { |
81819f0f CL |
1339 | page = get_partial_node(n); |
1340 | if (page) | |
1341 | return page; | |
1342 | } | |
1343 | } | |
1344 | #endif | |
1345 | return NULL; | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Get a partial page, lock it and return it. | |
1350 | */ | |
1351 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1352 | { | |
1353 | struct page *page; | |
1354 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1355 | ||
1356 | page = get_partial_node(get_node(s, searchnode)); | |
1357 | if (page || (flags & __GFP_THISNODE)) | |
1358 | return page; | |
1359 | ||
1360 | return get_any_partial(s, flags); | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * Move a page back to the lists. | |
1365 | * | |
1366 | * Must be called with the slab lock held. | |
1367 | * | |
1368 | * On exit the slab lock will have been dropped. | |
1369 | */ | |
7c2e132c | 1370 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1371 | { |
e95eed57 | 1372 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1373 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1374 | |
8a38082d | 1375 | __ClearPageSlubFrozen(page); |
81819f0f | 1376 | if (page->inuse) { |
e95eed57 | 1377 | |
a973e9dd | 1378 | if (page->freelist) { |
7c2e132c | 1379 | add_partial(n, page, tail); |
8ff12cfc CL |
1380 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1381 | } else { | |
1382 | stat(c, DEACTIVATE_FULL); | |
8a38082d AW |
1383 | if (SLABDEBUG && PageSlubDebug(page) && |
1384 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1385 | add_full(n, page); |
1386 | } | |
81819f0f CL |
1387 | slab_unlock(page); |
1388 | } else { | |
8ff12cfc | 1389 | stat(c, DEACTIVATE_EMPTY); |
5595cffc | 1390 | if (n->nr_partial < n->min_partial) { |
e95eed57 | 1391 | /* |
672bba3a CL |
1392 | * Adding an empty slab to the partial slabs in order |
1393 | * to avoid page allocator overhead. This slab needs | |
1394 | * to come after the other slabs with objects in | |
6446faa2 CL |
1395 | * so that the others get filled first. That way the |
1396 | * size of the partial list stays small. | |
1397 | * | |
0121c619 CL |
1398 | * kmem_cache_shrink can reclaim any empty slabs from |
1399 | * the partial list. | |
e95eed57 | 1400 | */ |
7c2e132c | 1401 | add_partial(n, page, 1); |
e95eed57 CL |
1402 | slab_unlock(page); |
1403 | } else { | |
1404 | slab_unlock(page); | |
8ff12cfc | 1405 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1406 | discard_slab(s, page); |
1407 | } | |
81819f0f CL |
1408 | } |
1409 | } | |
1410 | ||
1411 | /* | |
1412 | * Remove the cpu slab | |
1413 | */ | |
dfb4f096 | 1414 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1415 | { |
dfb4f096 | 1416 | struct page *page = c->page; |
7c2e132c | 1417 | int tail = 1; |
8ff12cfc | 1418 | |
b773ad73 | 1419 | if (page->freelist) |
8ff12cfc | 1420 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1421 | /* |
6446faa2 | 1422 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1423 | * because both freelists are empty. So this is unlikely |
1424 | * to occur. | |
1425 | */ | |
a973e9dd | 1426 | while (unlikely(c->freelist)) { |
894b8788 CL |
1427 | void **object; |
1428 | ||
7c2e132c CL |
1429 | tail = 0; /* Hot objects. Put the slab first */ |
1430 | ||
894b8788 | 1431 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1432 | object = c->freelist; |
b3fba8da | 1433 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1434 | |
1435 | /* And put onto the regular freelist */ | |
b3fba8da | 1436 | object[c->offset] = page->freelist; |
894b8788 CL |
1437 | page->freelist = object; |
1438 | page->inuse--; | |
1439 | } | |
dfb4f096 | 1440 | c->page = NULL; |
7c2e132c | 1441 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1442 | } |
1443 | ||
dfb4f096 | 1444 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1445 | { |
8ff12cfc | 1446 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1447 | slab_lock(c->page); |
1448 | deactivate_slab(s, c); | |
81819f0f CL |
1449 | } |
1450 | ||
1451 | /* | |
1452 | * Flush cpu slab. | |
6446faa2 | 1453 | * |
81819f0f CL |
1454 | * Called from IPI handler with interrupts disabled. |
1455 | */ | |
0c710013 | 1456 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1457 | { |
dfb4f096 | 1458 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1459 | |
dfb4f096 CL |
1460 | if (likely(c && c->page)) |
1461 | flush_slab(s, c); | |
81819f0f CL |
1462 | } |
1463 | ||
1464 | static void flush_cpu_slab(void *d) | |
1465 | { | |
1466 | struct kmem_cache *s = d; | |
81819f0f | 1467 | |
dfb4f096 | 1468 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1469 | } |
1470 | ||
1471 | static void flush_all(struct kmem_cache *s) | |
1472 | { | |
15c8b6c1 | 1473 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1474 | } |
1475 | ||
dfb4f096 CL |
1476 | /* |
1477 | * Check if the objects in a per cpu structure fit numa | |
1478 | * locality expectations. | |
1479 | */ | |
1480 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1481 | { | |
1482 | #ifdef CONFIG_NUMA | |
1483 | if (node != -1 && c->node != node) | |
1484 | return 0; | |
1485 | #endif | |
1486 | return 1; | |
1487 | } | |
1488 | ||
81819f0f | 1489 | /* |
894b8788 CL |
1490 | * Slow path. The lockless freelist is empty or we need to perform |
1491 | * debugging duties. | |
1492 | * | |
1493 | * Interrupts are disabled. | |
81819f0f | 1494 | * |
894b8788 CL |
1495 | * Processing is still very fast if new objects have been freed to the |
1496 | * regular freelist. In that case we simply take over the regular freelist | |
1497 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1498 | * |
894b8788 CL |
1499 | * If that is not working then we fall back to the partial lists. We take the |
1500 | * first element of the freelist as the object to allocate now and move the | |
1501 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1502 | * |
894b8788 | 1503 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1504 | * we need to allocate a new slab. This is the slowest path since it involves |
1505 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1506 | */ |
ce71e27c EGM |
1507 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1508 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1509 | { |
81819f0f | 1510 | void **object; |
dfb4f096 | 1511 | struct page *new; |
81819f0f | 1512 | |
e72e9c23 LT |
1513 | /* We handle __GFP_ZERO in the caller */ |
1514 | gfpflags &= ~__GFP_ZERO; | |
1515 | ||
dfb4f096 | 1516 | if (!c->page) |
81819f0f CL |
1517 | goto new_slab; |
1518 | ||
dfb4f096 CL |
1519 | slab_lock(c->page); |
1520 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1521 | goto another_slab; |
6446faa2 | 1522 | |
8ff12cfc | 1523 | stat(c, ALLOC_REFILL); |
6446faa2 | 1524 | |
894b8788 | 1525 | load_freelist: |
dfb4f096 | 1526 | object = c->page->freelist; |
a973e9dd | 1527 | if (unlikely(!object)) |
81819f0f | 1528 | goto another_slab; |
8a38082d | 1529 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1530 | goto debug; |
1531 | ||
b3fba8da | 1532 | c->freelist = object[c->offset]; |
39b26464 | 1533 | c->page->inuse = c->page->objects; |
a973e9dd | 1534 | c->page->freelist = NULL; |
dfb4f096 | 1535 | c->node = page_to_nid(c->page); |
1f84260c | 1536 | unlock_out: |
dfb4f096 | 1537 | slab_unlock(c->page); |
8ff12cfc | 1538 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1539 | return object; |
1540 | ||
1541 | another_slab: | |
dfb4f096 | 1542 | deactivate_slab(s, c); |
81819f0f CL |
1543 | |
1544 | new_slab: | |
dfb4f096 CL |
1545 | new = get_partial(s, gfpflags, node); |
1546 | if (new) { | |
1547 | c->page = new; | |
8ff12cfc | 1548 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1549 | goto load_freelist; |
81819f0f CL |
1550 | } |
1551 | ||
b811c202 CL |
1552 | if (gfpflags & __GFP_WAIT) |
1553 | local_irq_enable(); | |
1554 | ||
dfb4f096 | 1555 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1556 | |
1557 | if (gfpflags & __GFP_WAIT) | |
1558 | local_irq_disable(); | |
1559 | ||
dfb4f096 CL |
1560 | if (new) { |
1561 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1562 | stat(c, ALLOC_SLAB); |
05aa3450 | 1563 | if (c->page) |
dfb4f096 | 1564 | flush_slab(s, c); |
dfb4f096 | 1565 | slab_lock(new); |
8a38082d | 1566 | __SetPageSlubFrozen(new); |
dfb4f096 | 1567 | c->page = new; |
4b6f0750 | 1568 | goto load_freelist; |
81819f0f | 1569 | } |
71c7a06f | 1570 | return NULL; |
81819f0f | 1571 | debug: |
dfb4f096 | 1572 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1573 | goto another_slab; |
894b8788 | 1574 | |
dfb4f096 | 1575 | c->page->inuse++; |
b3fba8da | 1576 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1577 | c->node = -1; |
1f84260c | 1578 | goto unlock_out; |
894b8788 CL |
1579 | } |
1580 | ||
1581 | /* | |
1582 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1583 | * have the fastpath folded into their functions. So no function call | |
1584 | * overhead for requests that can be satisfied on the fastpath. | |
1585 | * | |
1586 | * The fastpath works by first checking if the lockless freelist can be used. | |
1587 | * If not then __slab_alloc is called for slow processing. | |
1588 | * | |
1589 | * Otherwise we can simply pick the next object from the lockless free list. | |
1590 | */ | |
06428780 | 1591 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1592 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1593 | { |
894b8788 | 1594 | void **object; |
dfb4f096 | 1595 | struct kmem_cache_cpu *c; |
1f84260c | 1596 | unsigned long flags; |
bdb21928 | 1597 | unsigned int objsize; |
1f84260c | 1598 | |
89124d70 | 1599 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1600 | |
773ff60e AM |
1601 | if (should_failslab(s->objsize, gfpflags)) |
1602 | return NULL; | |
1f84260c | 1603 | |
894b8788 | 1604 | local_irq_save(flags); |
dfb4f096 | 1605 | c = get_cpu_slab(s, smp_processor_id()); |
bdb21928 | 1606 | objsize = c->objsize; |
a973e9dd | 1607 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1608 | |
dfb4f096 | 1609 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1610 | |
1611 | else { | |
dfb4f096 | 1612 | object = c->freelist; |
b3fba8da | 1613 | c->freelist = object[c->offset]; |
8ff12cfc | 1614 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1615 | } |
1616 | local_irq_restore(flags); | |
d07dbea4 CL |
1617 | |
1618 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
bdb21928 | 1619 | memset(object, 0, objsize); |
d07dbea4 | 1620 | |
894b8788 | 1621 | return object; |
81819f0f CL |
1622 | } |
1623 | ||
1624 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1625 | { | |
ce71e27c | 1626 | return slab_alloc(s, gfpflags, -1, _RET_IP_); |
81819f0f CL |
1627 | } |
1628 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1629 | ||
1630 | #ifdef CONFIG_NUMA | |
1631 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1632 | { | |
ce71e27c | 1633 | return slab_alloc(s, gfpflags, node, _RET_IP_); |
81819f0f CL |
1634 | } |
1635 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1636 | #endif | |
1637 | ||
1638 | /* | |
894b8788 CL |
1639 | * Slow patch handling. This may still be called frequently since objects |
1640 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1641 | * |
894b8788 CL |
1642 | * So we still attempt to reduce cache line usage. Just take the slab |
1643 | * lock and free the item. If there is no additional partial page | |
1644 | * handling required then we can return immediately. | |
81819f0f | 1645 | */ |
894b8788 | 1646 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ce71e27c | 1647 | void *x, unsigned long addr, unsigned int offset) |
81819f0f CL |
1648 | { |
1649 | void *prior; | |
1650 | void **object = (void *)x; | |
8ff12cfc | 1651 | struct kmem_cache_cpu *c; |
81819f0f | 1652 | |
8ff12cfc CL |
1653 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1654 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1655 | slab_lock(page); |
1656 | ||
8a38082d | 1657 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1658 | goto debug; |
6446faa2 | 1659 | |
81819f0f | 1660 | checks_ok: |
b3fba8da | 1661 | prior = object[offset] = page->freelist; |
81819f0f CL |
1662 | page->freelist = object; |
1663 | page->inuse--; | |
1664 | ||
8a38082d | 1665 | if (unlikely(PageSlubFrozen(page))) { |
8ff12cfc | 1666 | stat(c, FREE_FROZEN); |
81819f0f | 1667 | goto out_unlock; |
8ff12cfc | 1668 | } |
81819f0f CL |
1669 | |
1670 | if (unlikely(!page->inuse)) | |
1671 | goto slab_empty; | |
1672 | ||
1673 | /* | |
6446faa2 | 1674 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1675 | * then add it. |
1676 | */ | |
a973e9dd | 1677 | if (unlikely(!prior)) { |
7c2e132c | 1678 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1679 | stat(c, FREE_ADD_PARTIAL); |
1680 | } | |
81819f0f CL |
1681 | |
1682 | out_unlock: | |
1683 | slab_unlock(page); | |
81819f0f CL |
1684 | return; |
1685 | ||
1686 | slab_empty: | |
a973e9dd | 1687 | if (prior) { |
81819f0f | 1688 | /* |
672bba3a | 1689 | * Slab still on the partial list. |
81819f0f CL |
1690 | */ |
1691 | remove_partial(s, page); | |
8ff12cfc CL |
1692 | stat(c, FREE_REMOVE_PARTIAL); |
1693 | } | |
81819f0f | 1694 | slab_unlock(page); |
8ff12cfc | 1695 | stat(c, FREE_SLAB); |
81819f0f | 1696 | discard_slab(s, page); |
81819f0f CL |
1697 | return; |
1698 | ||
1699 | debug: | |
3ec09742 | 1700 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1701 | goto out_unlock; |
77c5e2d0 | 1702 | goto checks_ok; |
81819f0f CL |
1703 | } |
1704 | ||
894b8788 CL |
1705 | /* |
1706 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1707 | * can perform fastpath freeing without additional function calls. | |
1708 | * | |
1709 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1710 | * of this processor. This typically the case if we have just allocated | |
1711 | * the item before. | |
1712 | * | |
1713 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1714 | * with all sorts of special processing. | |
1715 | */ | |
06428780 | 1716 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1717 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1718 | { |
1719 | void **object = (void *)x; | |
dfb4f096 | 1720 | struct kmem_cache_cpu *c; |
1f84260c CL |
1721 | unsigned long flags; |
1722 | ||
894b8788 | 1723 | local_irq_save(flags); |
dfb4f096 | 1724 | c = get_cpu_slab(s, smp_processor_id()); |
27d9e4e9 | 1725 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a | 1726 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
6047a007 | 1727 | debug_check_no_obj_freed(object, c->objsize); |
ee3c72a1 | 1728 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1729 | object[c->offset] = c->freelist; |
dfb4f096 | 1730 | c->freelist = object; |
8ff12cfc | 1731 | stat(c, FREE_FASTPATH); |
894b8788 | 1732 | } else |
b3fba8da | 1733 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1734 | |
1735 | local_irq_restore(flags); | |
1736 | } | |
1737 | ||
81819f0f CL |
1738 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1739 | { | |
77c5e2d0 | 1740 | struct page *page; |
81819f0f | 1741 | |
b49af68f | 1742 | page = virt_to_head_page(x); |
81819f0f | 1743 | |
ce71e27c | 1744 | slab_free(s, page, x, _RET_IP_); |
81819f0f CL |
1745 | } |
1746 | EXPORT_SYMBOL(kmem_cache_free); | |
1747 | ||
e9beef18 | 1748 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1749 | static struct page *get_object_page(const void *x) |
1750 | { | |
b49af68f | 1751 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1752 | |
1753 | if (!PageSlab(page)) | |
1754 | return NULL; | |
1755 | ||
1756 | return page; | |
1757 | } | |
1758 | ||
1759 | /* | |
672bba3a CL |
1760 | * Object placement in a slab is made very easy because we always start at |
1761 | * offset 0. If we tune the size of the object to the alignment then we can | |
1762 | * get the required alignment by putting one properly sized object after | |
1763 | * another. | |
81819f0f CL |
1764 | * |
1765 | * Notice that the allocation order determines the sizes of the per cpu | |
1766 | * caches. Each processor has always one slab available for allocations. | |
1767 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1768 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1769 | * locking overhead. |
81819f0f CL |
1770 | */ |
1771 | ||
1772 | /* | |
1773 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1774 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1775 | * and increases the number of allocations possible without having to | |
1776 | * take the list_lock. | |
1777 | */ | |
1778 | static int slub_min_order; | |
114e9e89 | 1779 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1780 | static int slub_min_objects; |
81819f0f CL |
1781 | |
1782 | /* | |
1783 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1784 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1785 | */ |
1786 | static int slub_nomerge; | |
1787 | ||
81819f0f CL |
1788 | /* |
1789 | * Calculate the order of allocation given an slab object size. | |
1790 | * | |
672bba3a CL |
1791 | * The order of allocation has significant impact on performance and other |
1792 | * system components. Generally order 0 allocations should be preferred since | |
1793 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1794 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1795 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1796 | * would be wasted. |
1797 | * | |
1798 | * In order to reach satisfactory performance we must ensure that a minimum | |
1799 | * number of objects is in one slab. Otherwise we may generate too much | |
1800 | * activity on the partial lists which requires taking the list_lock. This is | |
1801 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1802 | * |
672bba3a CL |
1803 | * slub_max_order specifies the order where we begin to stop considering the |
1804 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1805 | * we try to keep the page order as low as possible. So we accept more waste | |
1806 | * of space in favor of a small page order. | |
81819f0f | 1807 | * |
672bba3a CL |
1808 | * Higher order allocations also allow the placement of more objects in a |
1809 | * slab and thereby reduce object handling overhead. If the user has | |
1810 | * requested a higher mininum order then we start with that one instead of | |
1811 | * the smallest order which will fit the object. | |
81819f0f | 1812 | */ |
5e6d444e CL |
1813 | static inline int slab_order(int size, int min_objects, |
1814 | int max_order, int fract_leftover) | |
81819f0f CL |
1815 | { |
1816 | int order; | |
1817 | int rem; | |
6300ea75 | 1818 | int min_order = slub_min_order; |
81819f0f | 1819 | |
210b5c06 CG |
1820 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1821 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1822 | |
6300ea75 | 1823 | for (order = max(min_order, |
5e6d444e CL |
1824 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1825 | order <= max_order; order++) { | |
81819f0f | 1826 | |
5e6d444e | 1827 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1828 | |
5e6d444e | 1829 | if (slab_size < min_objects * size) |
81819f0f CL |
1830 | continue; |
1831 | ||
1832 | rem = slab_size % size; | |
1833 | ||
5e6d444e | 1834 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1835 | break; |
1836 | ||
1837 | } | |
672bba3a | 1838 | |
81819f0f CL |
1839 | return order; |
1840 | } | |
1841 | ||
5e6d444e CL |
1842 | static inline int calculate_order(int size) |
1843 | { | |
1844 | int order; | |
1845 | int min_objects; | |
1846 | int fraction; | |
1847 | ||
1848 | /* | |
1849 | * Attempt to find best configuration for a slab. This | |
1850 | * works by first attempting to generate a layout with | |
1851 | * the best configuration and backing off gradually. | |
1852 | * | |
1853 | * First we reduce the acceptable waste in a slab. Then | |
1854 | * we reduce the minimum objects required in a slab. | |
1855 | */ | |
1856 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1857 | if (!min_objects) |
1858 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
5e6d444e | 1859 | while (min_objects > 1) { |
c124f5b5 | 1860 | fraction = 16; |
5e6d444e CL |
1861 | while (fraction >= 4) { |
1862 | order = slab_order(size, min_objects, | |
1863 | slub_max_order, fraction); | |
1864 | if (order <= slub_max_order) | |
1865 | return order; | |
1866 | fraction /= 2; | |
1867 | } | |
1868 | min_objects /= 2; | |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * We were unable to place multiple objects in a slab. Now | |
1873 | * lets see if we can place a single object there. | |
1874 | */ | |
1875 | order = slab_order(size, 1, slub_max_order, 1); | |
1876 | if (order <= slub_max_order) | |
1877 | return order; | |
1878 | ||
1879 | /* | |
1880 | * Doh this slab cannot be placed using slub_max_order. | |
1881 | */ | |
1882 | order = slab_order(size, 1, MAX_ORDER, 1); | |
1883 | if (order <= MAX_ORDER) | |
1884 | return order; | |
1885 | return -ENOSYS; | |
1886 | } | |
1887 | ||
81819f0f | 1888 | /* |
672bba3a | 1889 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1890 | */ |
1891 | static unsigned long calculate_alignment(unsigned long flags, | |
1892 | unsigned long align, unsigned long size) | |
1893 | { | |
1894 | /* | |
6446faa2 CL |
1895 | * If the user wants hardware cache aligned objects then follow that |
1896 | * suggestion if the object is sufficiently large. | |
81819f0f | 1897 | * |
6446faa2 CL |
1898 | * The hardware cache alignment cannot override the specified |
1899 | * alignment though. If that is greater then use it. | |
81819f0f | 1900 | */ |
b6210386 NP |
1901 | if (flags & SLAB_HWCACHE_ALIGN) { |
1902 | unsigned long ralign = cache_line_size(); | |
1903 | while (size <= ralign / 2) | |
1904 | ralign /= 2; | |
1905 | align = max(align, ralign); | |
1906 | } | |
81819f0f CL |
1907 | |
1908 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 1909 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
1910 | |
1911 | return ALIGN(align, sizeof(void *)); | |
1912 | } | |
1913 | ||
dfb4f096 CL |
1914 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1915 | struct kmem_cache_cpu *c) | |
1916 | { | |
1917 | c->page = NULL; | |
a973e9dd | 1918 | c->freelist = NULL; |
dfb4f096 | 1919 | c->node = 0; |
42a9fdbb CL |
1920 | c->offset = s->offset / sizeof(void *); |
1921 | c->objsize = s->objsize; | |
62f75532 PE |
1922 | #ifdef CONFIG_SLUB_STATS |
1923 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
1924 | #endif | |
dfb4f096 CL |
1925 | } |
1926 | ||
5595cffc PE |
1927 | static void |
1928 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
1929 | { |
1930 | n->nr_partial = 0; | |
5595cffc PE |
1931 | |
1932 | /* | |
1933 | * The larger the object size is, the more pages we want on the partial | |
1934 | * list to avoid pounding the page allocator excessively. | |
1935 | */ | |
1936 | n->min_partial = ilog2(s->size); | |
1937 | if (n->min_partial < MIN_PARTIAL) | |
1938 | n->min_partial = MIN_PARTIAL; | |
1939 | else if (n->min_partial > MAX_PARTIAL) | |
1940 | n->min_partial = MAX_PARTIAL; | |
1941 | ||
81819f0f CL |
1942 | spin_lock_init(&n->list_lock); |
1943 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1944 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 1945 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 1946 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 1947 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1948 | #endif |
81819f0f CL |
1949 | } |
1950 | ||
4c93c355 CL |
1951 | #ifdef CONFIG_SMP |
1952 | /* | |
1953 | * Per cpu array for per cpu structures. | |
1954 | * | |
1955 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1956 | * close together meaning that it becomes possible that multiple per cpu | |
1957 | * structures are contained in one cacheline. This may be particularly | |
1958 | * beneficial for the kmalloc caches. | |
1959 | * | |
1960 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1961 | * likely able to get per cpu structures for all caches from the array defined | |
1962 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
1963 | * | |
1964 | * If the per cpu array is exhausted then fall back to kmalloc | |
1965 | * of individual cachelines. No sharing is possible then. | |
1966 | */ | |
1967 | #define NR_KMEM_CACHE_CPU 100 | |
1968 | ||
1969 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
1970 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
1971 | ||
1972 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
174596a0 | 1973 | static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS); |
4c93c355 CL |
1974 | |
1975 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
1976 | int cpu, gfp_t flags) | |
1977 | { | |
1978 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
1979 | ||
1980 | if (c) | |
1981 | per_cpu(kmem_cache_cpu_free, cpu) = | |
1982 | (void *)c->freelist; | |
1983 | else { | |
1984 | /* Table overflow: So allocate ourselves */ | |
1985 | c = kmalloc_node( | |
1986 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
1987 | flags, cpu_to_node(cpu)); | |
1988 | if (!c) | |
1989 | return NULL; | |
1990 | } | |
1991 | ||
1992 | init_kmem_cache_cpu(s, c); | |
1993 | return c; | |
1994 | } | |
1995 | ||
1996 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
1997 | { | |
1998 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
1999 | c > per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { | |
2000 | kfree(c); | |
2001 | return; | |
2002 | } | |
2003 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2004 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2005 | } | |
2006 | ||
2007 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2008 | { | |
2009 | int cpu; | |
2010 | ||
2011 | for_each_online_cpu(cpu) { | |
2012 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2013 | ||
2014 | if (c) { | |
2015 | s->cpu_slab[cpu] = NULL; | |
2016 | free_kmem_cache_cpu(c, cpu); | |
2017 | } | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2022 | { | |
2023 | int cpu; | |
2024 | ||
2025 | for_each_online_cpu(cpu) { | |
2026 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2027 | ||
2028 | if (c) | |
2029 | continue; | |
2030 | ||
2031 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2032 | if (!c) { | |
2033 | free_kmem_cache_cpus(s); | |
2034 | return 0; | |
2035 | } | |
2036 | s->cpu_slab[cpu] = c; | |
2037 | } | |
2038 | return 1; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Initialize the per cpu array. | |
2043 | */ | |
2044 | static void init_alloc_cpu_cpu(int cpu) | |
2045 | { | |
2046 | int i; | |
2047 | ||
174596a0 | 2048 | if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once))) |
4c93c355 CL |
2049 | return; |
2050 | ||
2051 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2052 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2053 | ||
174596a0 | 2054 | cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)); |
4c93c355 CL |
2055 | } |
2056 | ||
2057 | static void __init init_alloc_cpu(void) | |
2058 | { | |
2059 | int cpu; | |
2060 | ||
2061 | for_each_online_cpu(cpu) | |
2062 | init_alloc_cpu_cpu(cpu); | |
2063 | } | |
2064 | ||
2065 | #else | |
2066 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2067 | static inline void init_alloc_cpu(void) {} | |
2068 | ||
2069 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2070 | { | |
2071 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2072 | return 1; | |
2073 | } | |
2074 | #endif | |
2075 | ||
81819f0f CL |
2076 | #ifdef CONFIG_NUMA |
2077 | /* | |
2078 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2079 | * slab on the node for this slabcache. There are no concurrent accesses | |
2080 | * possible. | |
2081 | * | |
2082 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2083 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2084 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2085 | */ |
0094de92 | 2086 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2087 | { |
2088 | struct page *page; | |
2089 | struct kmem_cache_node *n; | |
ba84c73c | 2090 | unsigned long flags; |
81819f0f CL |
2091 | |
2092 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2093 | ||
a2f92ee7 | 2094 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2095 | |
2096 | BUG_ON(!page); | |
a2f92ee7 CL |
2097 | if (page_to_nid(page) != node) { |
2098 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2099 | "node %d\n", node); | |
2100 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2101 | "in order to be able to continue\n"); | |
2102 | } | |
2103 | ||
81819f0f CL |
2104 | n = page->freelist; |
2105 | BUG_ON(!n); | |
2106 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2107 | page->inuse++; | |
2108 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2109 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2110 | init_object(kmalloc_caches, n, 1); |
2111 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2112 | #endif |
5595cffc | 2113 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2114 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2115 | |
ba84c73c | 2116 | /* |
2117 | * lockdep requires consistent irq usage for each lock | |
2118 | * so even though there cannot be a race this early in | |
2119 | * the boot sequence, we still disable irqs. | |
2120 | */ | |
2121 | local_irq_save(flags); | |
7c2e132c | 2122 | add_partial(n, page, 0); |
ba84c73c | 2123 | local_irq_restore(flags); |
81819f0f CL |
2124 | } |
2125 | ||
2126 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2127 | { | |
2128 | int node; | |
2129 | ||
f64dc58c | 2130 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2131 | struct kmem_cache_node *n = s->node[node]; |
2132 | if (n && n != &s->local_node) | |
2133 | kmem_cache_free(kmalloc_caches, n); | |
2134 | s->node[node] = NULL; | |
2135 | } | |
2136 | } | |
2137 | ||
2138 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2139 | { | |
2140 | int node; | |
2141 | int local_node; | |
2142 | ||
2143 | if (slab_state >= UP) | |
2144 | local_node = page_to_nid(virt_to_page(s)); | |
2145 | else | |
2146 | local_node = 0; | |
2147 | ||
f64dc58c | 2148 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2149 | struct kmem_cache_node *n; |
2150 | ||
2151 | if (local_node == node) | |
2152 | n = &s->local_node; | |
2153 | else { | |
2154 | if (slab_state == DOWN) { | |
0094de92 | 2155 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2156 | continue; |
2157 | } | |
2158 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2159 | gfpflags, node); | |
2160 | ||
2161 | if (!n) { | |
2162 | free_kmem_cache_nodes(s); | |
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | } | |
2167 | s->node[node] = n; | |
5595cffc | 2168 | init_kmem_cache_node(n, s); |
81819f0f CL |
2169 | } |
2170 | return 1; | |
2171 | } | |
2172 | #else | |
2173 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2174 | { | |
2175 | } | |
2176 | ||
2177 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2178 | { | |
5595cffc | 2179 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2180 | return 1; |
2181 | } | |
2182 | #endif | |
2183 | ||
2184 | /* | |
2185 | * calculate_sizes() determines the order and the distribution of data within | |
2186 | * a slab object. | |
2187 | */ | |
06b285dc | 2188 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2189 | { |
2190 | unsigned long flags = s->flags; | |
2191 | unsigned long size = s->objsize; | |
2192 | unsigned long align = s->align; | |
834f3d11 | 2193 | int order; |
81819f0f | 2194 | |
d8b42bf5 CL |
2195 | /* |
2196 | * Round up object size to the next word boundary. We can only | |
2197 | * place the free pointer at word boundaries and this determines | |
2198 | * the possible location of the free pointer. | |
2199 | */ | |
2200 | size = ALIGN(size, sizeof(void *)); | |
2201 | ||
2202 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2203 | /* |
2204 | * Determine if we can poison the object itself. If the user of | |
2205 | * the slab may touch the object after free or before allocation | |
2206 | * then we should never poison the object itself. | |
2207 | */ | |
2208 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2209 | !s->ctor) |
81819f0f CL |
2210 | s->flags |= __OBJECT_POISON; |
2211 | else | |
2212 | s->flags &= ~__OBJECT_POISON; | |
2213 | ||
81819f0f CL |
2214 | |
2215 | /* | |
672bba3a | 2216 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2217 | * end of the object and the free pointer. If not then add an |
672bba3a | 2218 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2219 | */ |
2220 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2221 | size += sizeof(void *); | |
41ecc55b | 2222 | #endif |
81819f0f CL |
2223 | |
2224 | /* | |
672bba3a CL |
2225 | * With that we have determined the number of bytes in actual use |
2226 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2227 | */ |
2228 | s->inuse = size; | |
2229 | ||
2230 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2231 | s->ctor)) { |
81819f0f CL |
2232 | /* |
2233 | * Relocate free pointer after the object if it is not | |
2234 | * permitted to overwrite the first word of the object on | |
2235 | * kmem_cache_free. | |
2236 | * | |
2237 | * This is the case if we do RCU, have a constructor or | |
2238 | * destructor or are poisoning the objects. | |
2239 | */ | |
2240 | s->offset = size; | |
2241 | size += sizeof(void *); | |
2242 | } | |
2243 | ||
c12b3c62 | 2244 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2245 | if (flags & SLAB_STORE_USER) |
2246 | /* | |
2247 | * Need to store information about allocs and frees after | |
2248 | * the object. | |
2249 | */ | |
2250 | size += 2 * sizeof(struct track); | |
2251 | ||
be7b3fbc | 2252 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2253 | /* |
2254 | * Add some empty padding so that we can catch | |
2255 | * overwrites from earlier objects rather than let | |
2256 | * tracking information or the free pointer be | |
0211a9c8 | 2257 | * corrupted if a user writes before the start |
81819f0f CL |
2258 | * of the object. |
2259 | */ | |
2260 | size += sizeof(void *); | |
41ecc55b | 2261 | #endif |
672bba3a | 2262 | |
81819f0f CL |
2263 | /* |
2264 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2265 | * user specified and the dynamic determination of cache line size |
2266 | * on bootup. | |
81819f0f CL |
2267 | */ |
2268 | align = calculate_alignment(flags, align, s->objsize); | |
2269 | ||
2270 | /* | |
2271 | * SLUB stores one object immediately after another beginning from | |
2272 | * offset 0. In order to align the objects we have to simply size | |
2273 | * each object to conform to the alignment. | |
2274 | */ | |
2275 | size = ALIGN(size, align); | |
2276 | s->size = size; | |
06b285dc CL |
2277 | if (forced_order >= 0) |
2278 | order = forced_order; | |
2279 | else | |
2280 | order = calculate_order(size); | |
81819f0f | 2281 | |
834f3d11 | 2282 | if (order < 0) |
81819f0f CL |
2283 | return 0; |
2284 | ||
b7a49f0d | 2285 | s->allocflags = 0; |
834f3d11 | 2286 | if (order) |
b7a49f0d CL |
2287 | s->allocflags |= __GFP_COMP; |
2288 | ||
2289 | if (s->flags & SLAB_CACHE_DMA) | |
2290 | s->allocflags |= SLUB_DMA; | |
2291 | ||
2292 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2293 | s->allocflags |= __GFP_RECLAIMABLE; | |
2294 | ||
81819f0f CL |
2295 | /* |
2296 | * Determine the number of objects per slab | |
2297 | */ | |
834f3d11 | 2298 | s->oo = oo_make(order, size); |
65c3376a | 2299 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2300 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2301 | s->max = s->oo; | |
81819f0f | 2302 | |
834f3d11 | 2303 | return !!oo_objects(s->oo); |
81819f0f CL |
2304 | |
2305 | } | |
2306 | ||
81819f0f CL |
2307 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2308 | const char *name, size_t size, | |
2309 | size_t align, unsigned long flags, | |
51cc5068 | 2310 | void (*ctor)(void *)) |
81819f0f CL |
2311 | { |
2312 | memset(s, 0, kmem_size); | |
2313 | s->name = name; | |
2314 | s->ctor = ctor; | |
81819f0f | 2315 | s->objsize = size; |
81819f0f | 2316 | s->align = align; |
ba0268a8 | 2317 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2318 | |
06b285dc | 2319 | if (!calculate_sizes(s, -1)) |
81819f0f CL |
2320 | goto error; |
2321 | ||
2322 | s->refcount = 1; | |
2323 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2324 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2325 | #endif |
dfb4f096 CL |
2326 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2327 | goto error; | |
81819f0f | 2328 | |
dfb4f096 | 2329 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2330 | return 1; |
4c93c355 | 2331 | free_kmem_cache_nodes(s); |
81819f0f CL |
2332 | error: |
2333 | if (flags & SLAB_PANIC) | |
2334 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2335 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2336 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2337 | s->offset, flags); |
2338 | return 0; | |
2339 | } | |
81819f0f CL |
2340 | |
2341 | /* | |
2342 | * Check if a given pointer is valid | |
2343 | */ | |
2344 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2345 | { | |
06428780 | 2346 | struct page *page; |
81819f0f CL |
2347 | |
2348 | page = get_object_page(object); | |
2349 | ||
2350 | if (!page || s != page->slab) | |
2351 | /* No slab or wrong slab */ | |
2352 | return 0; | |
2353 | ||
abcd08a6 | 2354 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2355 | return 0; |
2356 | ||
2357 | /* | |
2358 | * We could also check if the object is on the slabs freelist. | |
2359 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2360 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2361 | * to a certain slab. |
2362 | */ | |
2363 | return 1; | |
2364 | } | |
2365 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2366 | ||
2367 | /* | |
2368 | * Determine the size of a slab object | |
2369 | */ | |
2370 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2371 | { | |
2372 | return s->objsize; | |
2373 | } | |
2374 | EXPORT_SYMBOL(kmem_cache_size); | |
2375 | ||
2376 | const char *kmem_cache_name(struct kmem_cache *s) | |
2377 | { | |
2378 | return s->name; | |
2379 | } | |
2380 | EXPORT_SYMBOL(kmem_cache_name); | |
2381 | ||
33b12c38 CL |
2382 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2383 | const char *text) | |
2384 | { | |
2385 | #ifdef CONFIG_SLUB_DEBUG | |
2386 | void *addr = page_address(page); | |
2387 | void *p; | |
2388 | DECLARE_BITMAP(map, page->objects); | |
2389 | ||
2390 | bitmap_zero(map, page->objects); | |
2391 | slab_err(s, page, "%s", text); | |
2392 | slab_lock(page); | |
2393 | for_each_free_object(p, s, page->freelist) | |
2394 | set_bit(slab_index(p, s, addr), map); | |
2395 | ||
2396 | for_each_object(p, s, addr, page->objects) { | |
2397 | ||
2398 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2399 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2400 | p, p - addr); | |
2401 | print_tracking(s, p); | |
2402 | } | |
2403 | } | |
2404 | slab_unlock(page); | |
2405 | #endif | |
2406 | } | |
2407 | ||
81819f0f | 2408 | /* |
599870b1 | 2409 | * Attempt to free all partial slabs on a node. |
81819f0f | 2410 | */ |
599870b1 | 2411 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2412 | { |
81819f0f CL |
2413 | unsigned long flags; |
2414 | struct page *page, *h; | |
2415 | ||
2416 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2417 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2418 | if (!page->inuse) { |
2419 | list_del(&page->lru); | |
2420 | discard_slab(s, page); | |
599870b1 | 2421 | n->nr_partial--; |
33b12c38 CL |
2422 | } else { |
2423 | list_slab_objects(s, page, | |
2424 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2425 | } |
33b12c38 | 2426 | } |
81819f0f | 2427 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2428 | } |
2429 | ||
2430 | /* | |
672bba3a | 2431 | * Release all resources used by a slab cache. |
81819f0f | 2432 | */ |
0c710013 | 2433 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2434 | { |
2435 | int node; | |
2436 | ||
2437 | flush_all(s); | |
2438 | ||
2439 | /* Attempt to free all objects */ | |
4c93c355 | 2440 | free_kmem_cache_cpus(s); |
f64dc58c | 2441 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2442 | struct kmem_cache_node *n = get_node(s, node); |
2443 | ||
599870b1 CL |
2444 | free_partial(s, n); |
2445 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2446 | return 1; |
2447 | } | |
2448 | free_kmem_cache_nodes(s); | |
2449 | return 0; | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * Close a cache and release the kmem_cache structure | |
2454 | * (must be used for caches created using kmem_cache_create) | |
2455 | */ | |
2456 | void kmem_cache_destroy(struct kmem_cache *s) | |
2457 | { | |
2458 | down_write(&slub_lock); | |
2459 | s->refcount--; | |
2460 | if (!s->refcount) { | |
2461 | list_del(&s->list); | |
a0e1d1be | 2462 | up_write(&slub_lock); |
d629d819 PE |
2463 | if (kmem_cache_close(s)) { |
2464 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2465 | "still has objects.\n", s->name, __func__); | |
2466 | dump_stack(); | |
2467 | } | |
81819f0f | 2468 | sysfs_slab_remove(s); |
a0e1d1be CL |
2469 | } else |
2470 | up_write(&slub_lock); | |
81819f0f CL |
2471 | } |
2472 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2473 | ||
2474 | /******************************************************************** | |
2475 | * Kmalloc subsystem | |
2476 | *******************************************************************/ | |
2477 | ||
331dc558 | 2478 | struct kmem_cache kmalloc_caches[PAGE_SHIFT + 1] __cacheline_aligned; |
81819f0f CL |
2479 | EXPORT_SYMBOL(kmalloc_caches); |
2480 | ||
81819f0f CL |
2481 | static int __init setup_slub_min_order(char *str) |
2482 | { | |
06428780 | 2483 | get_option(&str, &slub_min_order); |
81819f0f CL |
2484 | |
2485 | return 1; | |
2486 | } | |
2487 | ||
2488 | __setup("slub_min_order=", setup_slub_min_order); | |
2489 | ||
2490 | static int __init setup_slub_max_order(char *str) | |
2491 | { | |
06428780 | 2492 | get_option(&str, &slub_max_order); |
81819f0f CL |
2493 | |
2494 | return 1; | |
2495 | } | |
2496 | ||
2497 | __setup("slub_max_order=", setup_slub_max_order); | |
2498 | ||
2499 | static int __init setup_slub_min_objects(char *str) | |
2500 | { | |
06428780 | 2501 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2502 | |
2503 | return 1; | |
2504 | } | |
2505 | ||
2506 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2507 | ||
2508 | static int __init setup_slub_nomerge(char *str) | |
2509 | { | |
2510 | slub_nomerge = 1; | |
2511 | return 1; | |
2512 | } | |
2513 | ||
2514 | __setup("slub_nomerge", setup_slub_nomerge); | |
2515 | ||
81819f0f CL |
2516 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2517 | const char *name, int size, gfp_t gfp_flags) | |
2518 | { | |
2519 | unsigned int flags = 0; | |
2520 | ||
2521 | if (gfp_flags & SLUB_DMA) | |
2522 | flags = SLAB_CACHE_DMA; | |
2523 | ||
2524 | down_write(&slub_lock); | |
2525 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
319d1e24 | 2526 | flags, NULL)) |
81819f0f CL |
2527 | goto panic; |
2528 | ||
2529 | list_add(&s->list, &slab_caches); | |
2530 | up_write(&slub_lock); | |
2531 | if (sysfs_slab_add(s)) | |
2532 | goto panic; | |
2533 | return s; | |
2534 | ||
2535 | panic: | |
2536 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2537 | } | |
2538 | ||
2e443fd0 | 2539 | #ifdef CONFIG_ZONE_DMA |
4097d601 | 2540 | static struct kmem_cache *kmalloc_caches_dma[PAGE_SHIFT + 1]; |
1ceef402 CL |
2541 | |
2542 | static void sysfs_add_func(struct work_struct *w) | |
2543 | { | |
2544 | struct kmem_cache *s; | |
2545 | ||
2546 | down_write(&slub_lock); | |
2547 | list_for_each_entry(s, &slab_caches, list) { | |
2548 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2549 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2550 | sysfs_slab_add(s); | |
2551 | } | |
2552 | } | |
2553 | up_write(&slub_lock); | |
2554 | } | |
2555 | ||
2556 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2557 | ||
2e443fd0 CL |
2558 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2559 | { | |
2560 | struct kmem_cache *s; | |
2e443fd0 CL |
2561 | char *text; |
2562 | size_t realsize; | |
2563 | ||
2564 | s = kmalloc_caches_dma[index]; | |
2565 | if (s) | |
2566 | return s; | |
2567 | ||
2568 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2569 | if (flags & __GFP_WAIT) |
2570 | down_write(&slub_lock); | |
2571 | else { | |
2572 | if (!down_write_trylock(&slub_lock)) | |
2573 | goto out; | |
2574 | } | |
2575 | ||
2576 | if (kmalloc_caches_dma[index]) | |
2577 | goto unlock_out; | |
2e443fd0 | 2578 | |
7b55f620 | 2579 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2580 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2581 | (unsigned int)realsize); | |
1ceef402 CL |
2582 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2583 | ||
2584 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2585 | realsize, ARCH_KMALLOC_MINALIGN, | |
2586 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2587 | kfree(s); | |
2588 | kfree(text); | |
2589 | goto unlock_out; | |
dfce8648 | 2590 | } |
1ceef402 CL |
2591 | |
2592 | list_add(&s->list, &slab_caches); | |
2593 | kmalloc_caches_dma[index] = s; | |
2594 | ||
2595 | schedule_work(&sysfs_add_work); | |
2596 | ||
2597 | unlock_out: | |
dfce8648 | 2598 | up_write(&slub_lock); |
1ceef402 | 2599 | out: |
dfce8648 | 2600 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2601 | } |
2602 | #endif | |
2603 | ||
f1b26339 CL |
2604 | /* |
2605 | * Conversion table for small slabs sizes / 8 to the index in the | |
2606 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2607 | * of two cache sizes there. The size of larger slabs can be determined using | |
2608 | * fls. | |
2609 | */ | |
2610 | static s8 size_index[24] = { | |
2611 | 3, /* 8 */ | |
2612 | 4, /* 16 */ | |
2613 | 5, /* 24 */ | |
2614 | 5, /* 32 */ | |
2615 | 6, /* 40 */ | |
2616 | 6, /* 48 */ | |
2617 | 6, /* 56 */ | |
2618 | 6, /* 64 */ | |
2619 | 1, /* 72 */ | |
2620 | 1, /* 80 */ | |
2621 | 1, /* 88 */ | |
2622 | 1, /* 96 */ | |
2623 | 7, /* 104 */ | |
2624 | 7, /* 112 */ | |
2625 | 7, /* 120 */ | |
2626 | 7, /* 128 */ | |
2627 | 2, /* 136 */ | |
2628 | 2, /* 144 */ | |
2629 | 2, /* 152 */ | |
2630 | 2, /* 160 */ | |
2631 | 2, /* 168 */ | |
2632 | 2, /* 176 */ | |
2633 | 2, /* 184 */ | |
2634 | 2 /* 192 */ | |
2635 | }; | |
2636 | ||
81819f0f CL |
2637 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2638 | { | |
f1b26339 | 2639 | int index; |
81819f0f | 2640 | |
f1b26339 CL |
2641 | if (size <= 192) { |
2642 | if (!size) | |
2643 | return ZERO_SIZE_PTR; | |
81819f0f | 2644 | |
f1b26339 | 2645 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2646 | } else |
f1b26339 | 2647 | index = fls(size - 1); |
81819f0f CL |
2648 | |
2649 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2650 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2651 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2652 | |
81819f0f CL |
2653 | #endif |
2654 | return &kmalloc_caches[index]; | |
2655 | } | |
2656 | ||
2657 | void *__kmalloc(size_t size, gfp_t flags) | |
2658 | { | |
aadb4bc4 | 2659 | struct kmem_cache *s; |
81819f0f | 2660 | |
331dc558 | 2661 | if (unlikely(size > PAGE_SIZE)) |
eada35ef | 2662 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2663 | |
2664 | s = get_slab(size, flags); | |
2665 | ||
2666 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2667 | return s; |
2668 | ||
ce71e27c | 2669 | return slab_alloc(s, flags, -1, _RET_IP_); |
81819f0f CL |
2670 | } |
2671 | EXPORT_SYMBOL(__kmalloc); | |
2672 | ||
f619cfe1 CL |
2673 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2674 | { | |
2675 | struct page *page = alloc_pages_node(node, flags | __GFP_COMP, | |
2676 | get_order(size)); | |
2677 | ||
2678 | if (page) | |
2679 | return page_address(page); | |
2680 | else | |
2681 | return NULL; | |
2682 | } | |
2683 | ||
81819f0f CL |
2684 | #ifdef CONFIG_NUMA |
2685 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2686 | { | |
aadb4bc4 | 2687 | struct kmem_cache *s; |
81819f0f | 2688 | |
331dc558 | 2689 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 2690 | return kmalloc_large_node(size, flags, node); |
aadb4bc4 CL |
2691 | |
2692 | s = get_slab(size, flags); | |
2693 | ||
2694 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2695 | return s; |
2696 | ||
ce71e27c | 2697 | return slab_alloc(s, flags, node, _RET_IP_); |
81819f0f CL |
2698 | } |
2699 | EXPORT_SYMBOL(__kmalloc_node); | |
2700 | #endif | |
2701 | ||
2702 | size_t ksize(const void *object) | |
2703 | { | |
272c1d21 | 2704 | struct page *page; |
81819f0f CL |
2705 | struct kmem_cache *s; |
2706 | ||
ef8b4520 | 2707 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2708 | return 0; |
2709 | ||
294a80a8 | 2710 | page = virt_to_head_page(object); |
294a80a8 | 2711 | |
76994412 PE |
2712 | if (unlikely(!PageSlab(page))) { |
2713 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2714 | return PAGE_SIZE << compound_order(page); |
76994412 | 2715 | } |
81819f0f | 2716 | s = page->slab; |
81819f0f | 2717 | |
ae20bfda | 2718 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2719 | /* |
2720 | * Debugging requires use of the padding between object | |
2721 | * and whatever may come after it. | |
2722 | */ | |
2723 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2724 | return s->objsize; | |
2725 | ||
ae20bfda | 2726 | #endif |
81819f0f CL |
2727 | /* |
2728 | * If we have the need to store the freelist pointer | |
2729 | * back there or track user information then we can | |
2730 | * only use the space before that information. | |
2731 | */ | |
2732 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2733 | return s->inuse; | |
81819f0f CL |
2734 | /* |
2735 | * Else we can use all the padding etc for the allocation | |
2736 | */ | |
2737 | return s->size; | |
2738 | } | |
81819f0f CL |
2739 | |
2740 | void kfree(const void *x) | |
2741 | { | |
81819f0f | 2742 | struct page *page; |
5bb983b0 | 2743 | void *object = (void *)x; |
81819f0f | 2744 | |
2408c550 | 2745 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2746 | return; |
2747 | ||
b49af68f | 2748 | page = virt_to_head_page(x); |
aadb4bc4 | 2749 | if (unlikely(!PageSlab(page))) { |
0937502a | 2750 | BUG_ON(!PageCompound(page)); |
aadb4bc4 CL |
2751 | put_page(page); |
2752 | return; | |
2753 | } | |
ce71e27c | 2754 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2755 | } |
2756 | EXPORT_SYMBOL(kfree); | |
2757 | ||
2086d26a | 2758 | /* |
672bba3a CL |
2759 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2760 | * the remaining slabs by the number of items in use. The slabs with the | |
2761 | * most items in use come first. New allocations will then fill those up | |
2762 | * and thus they can be removed from the partial lists. | |
2763 | * | |
2764 | * The slabs with the least items are placed last. This results in them | |
2765 | * being allocated from last increasing the chance that the last objects | |
2766 | * are freed in them. | |
2086d26a CL |
2767 | */ |
2768 | int kmem_cache_shrink(struct kmem_cache *s) | |
2769 | { | |
2770 | int node; | |
2771 | int i; | |
2772 | struct kmem_cache_node *n; | |
2773 | struct page *page; | |
2774 | struct page *t; | |
205ab99d | 2775 | int objects = oo_objects(s->max); |
2086d26a | 2776 | struct list_head *slabs_by_inuse = |
834f3d11 | 2777 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2778 | unsigned long flags; |
2779 | ||
2780 | if (!slabs_by_inuse) | |
2781 | return -ENOMEM; | |
2782 | ||
2783 | flush_all(s); | |
f64dc58c | 2784 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2785 | n = get_node(s, node); |
2786 | ||
2787 | if (!n->nr_partial) | |
2788 | continue; | |
2789 | ||
834f3d11 | 2790 | for (i = 0; i < objects; i++) |
2086d26a CL |
2791 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2792 | ||
2793 | spin_lock_irqsave(&n->list_lock, flags); | |
2794 | ||
2795 | /* | |
672bba3a | 2796 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2797 | * |
672bba3a CL |
2798 | * Note that concurrent frees may occur while we hold the |
2799 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2800 | */ |
2801 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2802 | if (!page->inuse && slab_trylock(page)) { | |
2803 | /* | |
2804 | * Must hold slab lock here because slab_free | |
2805 | * may have freed the last object and be | |
2806 | * waiting to release the slab. | |
2807 | */ | |
2808 | list_del(&page->lru); | |
2809 | n->nr_partial--; | |
2810 | slab_unlock(page); | |
2811 | discard_slab(s, page); | |
2812 | } else { | |
fcda3d89 CL |
2813 | list_move(&page->lru, |
2814 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2815 | } |
2816 | } | |
2817 | ||
2086d26a | 2818 | /* |
672bba3a CL |
2819 | * Rebuild the partial list with the slabs filled up most |
2820 | * first and the least used slabs at the end. | |
2086d26a | 2821 | */ |
834f3d11 | 2822 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2823 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2824 | ||
2086d26a CL |
2825 | spin_unlock_irqrestore(&n->list_lock, flags); |
2826 | } | |
2827 | ||
2828 | kfree(slabs_by_inuse); | |
2829 | return 0; | |
2830 | } | |
2831 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2832 | ||
b9049e23 YG |
2833 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2834 | static int slab_mem_going_offline_callback(void *arg) | |
2835 | { | |
2836 | struct kmem_cache *s; | |
2837 | ||
2838 | down_read(&slub_lock); | |
2839 | list_for_each_entry(s, &slab_caches, list) | |
2840 | kmem_cache_shrink(s); | |
2841 | up_read(&slub_lock); | |
2842 | ||
2843 | return 0; | |
2844 | } | |
2845 | ||
2846 | static void slab_mem_offline_callback(void *arg) | |
2847 | { | |
2848 | struct kmem_cache_node *n; | |
2849 | struct kmem_cache *s; | |
2850 | struct memory_notify *marg = arg; | |
2851 | int offline_node; | |
2852 | ||
2853 | offline_node = marg->status_change_nid; | |
2854 | ||
2855 | /* | |
2856 | * If the node still has available memory. we need kmem_cache_node | |
2857 | * for it yet. | |
2858 | */ | |
2859 | if (offline_node < 0) | |
2860 | return; | |
2861 | ||
2862 | down_read(&slub_lock); | |
2863 | list_for_each_entry(s, &slab_caches, list) { | |
2864 | n = get_node(s, offline_node); | |
2865 | if (n) { | |
2866 | /* | |
2867 | * if n->nr_slabs > 0, slabs still exist on the node | |
2868 | * that is going down. We were unable to free them, | |
2869 | * and offline_pages() function shoudn't call this | |
2870 | * callback. So, we must fail. | |
2871 | */ | |
0f389ec6 | 2872 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2873 | |
2874 | s->node[offline_node] = NULL; | |
2875 | kmem_cache_free(kmalloc_caches, n); | |
2876 | } | |
2877 | } | |
2878 | up_read(&slub_lock); | |
2879 | } | |
2880 | ||
2881 | static int slab_mem_going_online_callback(void *arg) | |
2882 | { | |
2883 | struct kmem_cache_node *n; | |
2884 | struct kmem_cache *s; | |
2885 | struct memory_notify *marg = arg; | |
2886 | int nid = marg->status_change_nid; | |
2887 | int ret = 0; | |
2888 | ||
2889 | /* | |
2890 | * If the node's memory is already available, then kmem_cache_node is | |
2891 | * already created. Nothing to do. | |
2892 | */ | |
2893 | if (nid < 0) | |
2894 | return 0; | |
2895 | ||
2896 | /* | |
0121c619 | 2897 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2898 | * allocate a kmem_cache_node structure in order to bring the node |
2899 | * online. | |
2900 | */ | |
2901 | down_read(&slub_lock); | |
2902 | list_for_each_entry(s, &slab_caches, list) { | |
2903 | /* | |
2904 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2905 | * since memory is not yet available from the node that | |
2906 | * is brought up. | |
2907 | */ | |
2908 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2909 | if (!n) { | |
2910 | ret = -ENOMEM; | |
2911 | goto out; | |
2912 | } | |
5595cffc | 2913 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2914 | s->node[nid] = n; |
2915 | } | |
2916 | out: | |
2917 | up_read(&slub_lock); | |
2918 | return ret; | |
2919 | } | |
2920 | ||
2921 | static int slab_memory_callback(struct notifier_block *self, | |
2922 | unsigned long action, void *arg) | |
2923 | { | |
2924 | int ret = 0; | |
2925 | ||
2926 | switch (action) { | |
2927 | case MEM_GOING_ONLINE: | |
2928 | ret = slab_mem_going_online_callback(arg); | |
2929 | break; | |
2930 | case MEM_GOING_OFFLINE: | |
2931 | ret = slab_mem_going_offline_callback(arg); | |
2932 | break; | |
2933 | case MEM_OFFLINE: | |
2934 | case MEM_CANCEL_ONLINE: | |
2935 | slab_mem_offline_callback(arg); | |
2936 | break; | |
2937 | case MEM_ONLINE: | |
2938 | case MEM_CANCEL_OFFLINE: | |
2939 | break; | |
2940 | } | |
dc19f9db KH |
2941 | if (ret) |
2942 | ret = notifier_from_errno(ret); | |
2943 | else | |
2944 | ret = NOTIFY_OK; | |
b9049e23 YG |
2945 | return ret; |
2946 | } | |
2947 | ||
2948 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2949 | ||
81819f0f CL |
2950 | /******************************************************************** |
2951 | * Basic setup of slabs | |
2952 | *******************************************************************/ | |
2953 | ||
2954 | void __init kmem_cache_init(void) | |
2955 | { | |
2956 | int i; | |
4b356be0 | 2957 | int caches = 0; |
81819f0f | 2958 | |
4c93c355 CL |
2959 | init_alloc_cpu(); |
2960 | ||
81819f0f CL |
2961 | #ifdef CONFIG_NUMA |
2962 | /* | |
2963 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 2964 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
2965 | * kmem_cache_open for slab_state == DOWN. |
2966 | */ | |
2967 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2968 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 2969 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 2970 | caches++; |
b9049e23 | 2971 | |
0c40ba4f | 2972 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
2973 | #endif |
2974 | ||
2975 | /* Able to allocate the per node structures */ | |
2976 | slab_state = PARTIAL; | |
2977 | ||
2978 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
2979 | if (KMALLOC_MIN_SIZE <= 64) { |
2980 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 2981 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 | 2982 | caches++; |
4b356be0 | 2983 | create_kmalloc_cache(&kmalloc_caches[2], |
81819f0f | 2984 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
2985 | caches++; |
2986 | } | |
81819f0f | 2987 | |
331dc558 | 2988 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) { |
81819f0f CL |
2989 | create_kmalloc_cache(&kmalloc_caches[i], |
2990 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
2991 | caches++; |
2992 | } | |
81819f0f | 2993 | |
f1b26339 CL |
2994 | |
2995 | /* | |
2996 | * Patch up the size_index table if we have strange large alignment | |
2997 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 2998 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
2999 | * |
3000 | * Largest permitted alignment is 256 bytes due to the way we | |
3001 | * handle the index determination for the smaller caches. | |
3002 | * | |
3003 | * Make sure that nothing crazy happens if someone starts tinkering | |
3004 | * around with ARCH_KMALLOC_MINALIGN | |
3005 | */ | |
3006 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3007 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3008 | ||
12ad6843 | 3009 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3010 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3011 | ||
41d54d3b CL |
3012 | if (KMALLOC_MIN_SIZE == 128) { |
3013 | /* | |
3014 | * The 192 byte sized cache is not used if the alignment | |
3015 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3016 | * instead. | |
3017 | */ | |
3018 | for (i = 128 + 8; i <= 192; i += 8) | |
3019 | size_index[(i - 1) / 8] = 8; | |
3020 | } | |
3021 | ||
81819f0f CL |
3022 | slab_state = UP; |
3023 | ||
3024 | /* Provide the correct kmalloc names now that the caches are up */ | |
331dc558 | 3025 | for (i = KMALLOC_SHIFT_LOW; i <= PAGE_SHIFT; i++) |
81819f0f CL |
3026 | kmalloc_caches[i]. name = |
3027 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3028 | ||
3029 | #ifdef CONFIG_SMP | |
3030 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3031 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3032 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3033 | #else | |
3034 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3035 | #endif |
3036 | ||
3adbefee IM |
3037 | printk(KERN_INFO |
3038 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3039 | " CPUs=%d, Nodes=%d\n", |
3040 | caches, cache_line_size(), | |
81819f0f CL |
3041 | slub_min_order, slub_max_order, slub_min_objects, |
3042 | nr_cpu_ids, nr_node_ids); | |
3043 | } | |
3044 | ||
3045 | /* | |
3046 | * Find a mergeable slab cache | |
3047 | */ | |
3048 | static int slab_unmergeable(struct kmem_cache *s) | |
3049 | { | |
3050 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3051 | return 1; | |
3052 | ||
c59def9f | 3053 | if (s->ctor) |
81819f0f CL |
3054 | return 1; |
3055 | ||
8ffa6875 CL |
3056 | /* |
3057 | * We may have set a slab to be unmergeable during bootstrap. | |
3058 | */ | |
3059 | if (s->refcount < 0) | |
3060 | return 1; | |
3061 | ||
81819f0f CL |
3062 | return 0; |
3063 | } | |
3064 | ||
3065 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3066 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3067 | void (*ctor)(void *)) |
81819f0f | 3068 | { |
5b95a4ac | 3069 | struct kmem_cache *s; |
81819f0f CL |
3070 | |
3071 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3072 | return NULL; | |
3073 | ||
c59def9f | 3074 | if (ctor) |
81819f0f CL |
3075 | return NULL; |
3076 | ||
3077 | size = ALIGN(size, sizeof(void *)); | |
3078 | align = calculate_alignment(flags, align, size); | |
3079 | size = ALIGN(size, align); | |
ba0268a8 | 3080 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3081 | |
5b95a4ac | 3082 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3083 | if (slab_unmergeable(s)) |
3084 | continue; | |
3085 | ||
3086 | if (size > s->size) | |
3087 | continue; | |
3088 | ||
ba0268a8 | 3089 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3090 | continue; |
3091 | /* | |
3092 | * Check if alignment is compatible. | |
3093 | * Courtesy of Adrian Drzewiecki | |
3094 | */ | |
06428780 | 3095 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3096 | continue; |
3097 | ||
3098 | if (s->size - size >= sizeof(void *)) | |
3099 | continue; | |
3100 | ||
3101 | return s; | |
3102 | } | |
3103 | return NULL; | |
3104 | } | |
3105 | ||
3106 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3107 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3108 | { |
3109 | struct kmem_cache *s; | |
3110 | ||
3111 | down_write(&slub_lock); | |
ba0268a8 | 3112 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3113 | if (s) { |
42a9fdbb CL |
3114 | int cpu; |
3115 | ||
81819f0f CL |
3116 | s->refcount++; |
3117 | /* | |
3118 | * Adjust the object sizes so that we clear | |
3119 | * the complete object on kzalloc. | |
3120 | */ | |
3121 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3122 | |
3123 | /* | |
3124 | * And then we need to update the object size in the | |
3125 | * per cpu structures | |
3126 | */ | |
3127 | for_each_online_cpu(cpu) | |
3128 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3129 | |
81819f0f | 3130 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3131 | up_write(&slub_lock); |
6446faa2 | 3132 | |
7b8f3b66 DR |
3133 | if (sysfs_slab_alias(s, name)) { |
3134 | down_write(&slub_lock); | |
3135 | s->refcount--; | |
3136 | up_write(&slub_lock); | |
81819f0f | 3137 | goto err; |
7b8f3b66 | 3138 | } |
a0e1d1be CL |
3139 | return s; |
3140 | } | |
6446faa2 | 3141 | |
a0e1d1be CL |
3142 | s = kmalloc(kmem_size, GFP_KERNEL); |
3143 | if (s) { | |
3144 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3145 | size, align, flags, ctor)) { |
81819f0f | 3146 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3147 | up_write(&slub_lock); |
7b8f3b66 DR |
3148 | if (sysfs_slab_add(s)) { |
3149 | down_write(&slub_lock); | |
3150 | list_del(&s->list); | |
3151 | up_write(&slub_lock); | |
3152 | kfree(s); | |
a0e1d1be | 3153 | goto err; |
7b8f3b66 | 3154 | } |
a0e1d1be CL |
3155 | return s; |
3156 | } | |
3157 | kfree(s); | |
81819f0f CL |
3158 | } |
3159 | up_write(&slub_lock); | |
81819f0f CL |
3160 | |
3161 | err: | |
81819f0f CL |
3162 | if (flags & SLAB_PANIC) |
3163 | panic("Cannot create slabcache %s\n", name); | |
3164 | else | |
3165 | s = NULL; | |
3166 | return s; | |
3167 | } | |
3168 | EXPORT_SYMBOL(kmem_cache_create); | |
3169 | ||
81819f0f | 3170 | #ifdef CONFIG_SMP |
81819f0f | 3171 | /* |
672bba3a CL |
3172 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3173 | * necessary. | |
81819f0f CL |
3174 | */ |
3175 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3176 | unsigned long action, void *hcpu) | |
3177 | { | |
3178 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3179 | struct kmem_cache *s; |
3180 | unsigned long flags; | |
81819f0f CL |
3181 | |
3182 | switch (action) { | |
4c93c355 CL |
3183 | case CPU_UP_PREPARE: |
3184 | case CPU_UP_PREPARE_FROZEN: | |
3185 | init_alloc_cpu_cpu(cpu); | |
3186 | down_read(&slub_lock); | |
3187 | list_for_each_entry(s, &slab_caches, list) | |
3188 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3189 | GFP_KERNEL); | |
3190 | up_read(&slub_lock); | |
3191 | break; | |
3192 | ||
81819f0f | 3193 | case CPU_UP_CANCELED: |
8bb78442 | 3194 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3195 | case CPU_DEAD: |
8bb78442 | 3196 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3197 | down_read(&slub_lock); |
3198 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3199 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3200 | ||
5b95a4ac CL |
3201 | local_irq_save(flags); |
3202 | __flush_cpu_slab(s, cpu); | |
3203 | local_irq_restore(flags); | |
4c93c355 CL |
3204 | free_kmem_cache_cpu(c, cpu); |
3205 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3206 | } |
3207 | up_read(&slub_lock); | |
81819f0f CL |
3208 | break; |
3209 | default: | |
3210 | break; | |
3211 | } | |
3212 | return NOTIFY_OK; | |
3213 | } | |
3214 | ||
06428780 | 3215 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3216 | .notifier_call = slab_cpuup_callback |
06428780 | 3217 | }; |
81819f0f CL |
3218 | |
3219 | #endif | |
3220 | ||
ce71e27c | 3221 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3222 | { |
aadb4bc4 CL |
3223 | struct kmem_cache *s; |
3224 | ||
331dc558 | 3225 | if (unlikely(size > PAGE_SIZE)) |
eada35ef PE |
3226 | return kmalloc_large(size, gfpflags); |
3227 | ||
aadb4bc4 | 3228 | s = get_slab(size, gfpflags); |
81819f0f | 3229 | |
2408c550 | 3230 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3231 | return s; |
81819f0f | 3232 | |
ce15fea8 | 3233 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
3234 | } |
3235 | ||
3236 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3237 | int node, unsigned long caller) |
81819f0f | 3238 | { |
aadb4bc4 CL |
3239 | struct kmem_cache *s; |
3240 | ||
331dc558 | 3241 | if (unlikely(size > PAGE_SIZE)) |
f619cfe1 | 3242 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3243 | |
aadb4bc4 | 3244 | s = get_slab(size, gfpflags); |
81819f0f | 3245 | |
2408c550 | 3246 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3247 | return s; |
81819f0f | 3248 | |
ce15fea8 | 3249 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
3250 | } |
3251 | ||
f6acb635 | 3252 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3253 | static unsigned long count_partial(struct kmem_cache_node *n, |
3254 | int (*get_count)(struct page *)) | |
5b06c853 CL |
3255 | { |
3256 | unsigned long flags; | |
3257 | unsigned long x = 0; | |
3258 | struct page *page; | |
3259 | ||
3260 | spin_lock_irqsave(&n->list_lock, flags); | |
3261 | list_for_each_entry(page, &n->partial, lru) | |
205ab99d | 3262 | x += get_count(page); |
5b06c853 CL |
3263 | spin_unlock_irqrestore(&n->list_lock, flags); |
3264 | return x; | |
3265 | } | |
205ab99d CL |
3266 | |
3267 | static int count_inuse(struct page *page) | |
3268 | { | |
3269 | return page->inuse; | |
3270 | } | |
3271 | ||
3272 | static int count_total(struct page *page) | |
3273 | { | |
3274 | return page->objects; | |
3275 | } | |
3276 | ||
3277 | static int count_free(struct page *page) | |
3278 | { | |
3279 | return page->objects - page->inuse; | |
3280 | } | |
5b06c853 | 3281 | |
434e245d CL |
3282 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3283 | unsigned long *map) | |
53e15af0 CL |
3284 | { |
3285 | void *p; | |
a973e9dd | 3286 | void *addr = page_address(page); |
53e15af0 CL |
3287 | |
3288 | if (!check_slab(s, page) || | |
3289 | !on_freelist(s, page, NULL)) | |
3290 | return 0; | |
3291 | ||
3292 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3293 | bitmap_zero(map, page->objects); |
53e15af0 | 3294 | |
7656c72b CL |
3295 | for_each_free_object(p, s, page->freelist) { |
3296 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3297 | if (!check_object(s, page, p, 0)) |
3298 | return 0; | |
3299 | } | |
3300 | ||
224a88be | 3301 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3302 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3303 | if (!check_object(s, page, p, 1)) |
3304 | return 0; | |
3305 | return 1; | |
3306 | } | |
3307 | ||
434e245d CL |
3308 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3309 | unsigned long *map) | |
53e15af0 CL |
3310 | { |
3311 | if (slab_trylock(page)) { | |
434e245d | 3312 | validate_slab(s, page, map); |
53e15af0 CL |
3313 | slab_unlock(page); |
3314 | } else | |
3315 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3316 | s->name, page); | |
3317 | ||
3318 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3319 | if (!PageSlubDebug(page)) |
3320 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3321 | "on slab 0x%p\n", s->name, page); |
3322 | } else { | |
8a38082d AW |
3323 | if (PageSlubDebug(page)) |
3324 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3325 | "slab 0x%p\n", s->name, page); |
3326 | } | |
3327 | } | |
3328 | ||
434e245d CL |
3329 | static int validate_slab_node(struct kmem_cache *s, |
3330 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3331 | { |
3332 | unsigned long count = 0; | |
3333 | struct page *page; | |
3334 | unsigned long flags; | |
3335 | ||
3336 | spin_lock_irqsave(&n->list_lock, flags); | |
3337 | ||
3338 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3339 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3340 | count++; |
3341 | } | |
3342 | if (count != n->nr_partial) | |
3343 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3344 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3345 | ||
3346 | if (!(s->flags & SLAB_STORE_USER)) | |
3347 | goto out; | |
3348 | ||
3349 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3350 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3351 | count++; |
3352 | } | |
3353 | if (count != atomic_long_read(&n->nr_slabs)) | |
3354 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3355 | "counter=%ld\n", s->name, count, | |
3356 | atomic_long_read(&n->nr_slabs)); | |
3357 | ||
3358 | out: | |
3359 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3360 | return count; | |
3361 | } | |
3362 | ||
434e245d | 3363 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3364 | { |
3365 | int node; | |
3366 | unsigned long count = 0; | |
205ab99d | 3367 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3368 | sizeof(unsigned long), GFP_KERNEL); |
3369 | ||
3370 | if (!map) | |
3371 | return -ENOMEM; | |
53e15af0 CL |
3372 | |
3373 | flush_all(s); | |
f64dc58c | 3374 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3375 | struct kmem_cache_node *n = get_node(s, node); |
3376 | ||
434e245d | 3377 | count += validate_slab_node(s, n, map); |
53e15af0 | 3378 | } |
434e245d | 3379 | kfree(map); |
53e15af0 CL |
3380 | return count; |
3381 | } | |
3382 | ||
b3459709 CL |
3383 | #ifdef SLUB_RESILIENCY_TEST |
3384 | static void resiliency_test(void) | |
3385 | { | |
3386 | u8 *p; | |
3387 | ||
3388 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3389 | printk(KERN_ERR "-----------------------\n"); | |
3390 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3391 | ||
3392 | p = kzalloc(16, GFP_KERNEL); | |
3393 | p[16] = 0x12; | |
3394 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3395 | " 0x12->0x%p\n\n", p + 16); | |
3396 | ||
3397 | validate_slab_cache(kmalloc_caches + 4); | |
3398 | ||
3399 | /* Hmmm... The next two are dangerous */ | |
3400 | p = kzalloc(32, GFP_KERNEL); | |
3401 | p[32 + sizeof(void *)] = 0x34; | |
3402 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3403 | " 0x34 -> -0x%p\n", p); |
3404 | printk(KERN_ERR | |
3405 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3406 | |
3407 | validate_slab_cache(kmalloc_caches + 5); | |
3408 | p = kzalloc(64, GFP_KERNEL); | |
3409 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3410 | *p = 0x56; | |
3411 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3412 | p); | |
3adbefee IM |
3413 | printk(KERN_ERR |
3414 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3415 | validate_slab_cache(kmalloc_caches + 6); |
3416 | ||
3417 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3418 | p = kzalloc(128, GFP_KERNEL); | |
3419 | kfree(p); | |
3420 | *p = 0x78; | |
3421 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3422 | validate_slab_cache(kmalloc_caches + 7); | |
3423 | ||
3424 | p = kzalloc(256, GFP_KERNEL); | |
3425 | kfree(p); | |
3426 | p[50] = 0x9a; | |
3adbefee IM |
3427 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3428 | p); | |
b3459709 CL |
3429 | validate_slab_cache(kmalloc_caches + 8); |
3430 | ||
3431 | p = kzalloc(512, GFP_KERNEL); | |
3432 | kfree(p); | |
3433 | p[512] = 0xab; | |
3434 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3435 | validate_slab_cache(kmalloc_caches + 9); | |
3436 | } | |
3437 | #else | |
3438 | static void resiliency_test(void) {}; | |
3439 | #endif | |
3440 | ||
88a420e4 | 3441 | /* |
672bba3a | 3442 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3443 | * and freed. |
3444 | */ | |
3445 | ||
3446 | struct location { | |
3447 | unsigned long count; | |
ce71e27c | 3448 | unsigned long addr; |
45edfa58 CL |
3449 | long long sum_time; |
3450 | long min_time; | |
3451 | long max_time; | |
3452 | long min_pid; | |
3453 | long max_pid; | |
174596a0 | 3454 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3455 | nodemask_t nodes; |
88a420e4 CL |
3456 | }; |
3457 | ||
3458 | struct loc_track { | |
3459 | unsigned long max; | |
3460 | unsigned long count; | |
3461 | struct location *loc; | |
3462 | }; | |
3463 | ||
3464 | static void free_loc_track(struct loc_track *t) | |
3465 | { | |
3466 | if (t->max) | |
3467 | free_pages((unsigned long)t->loc, | |
3468 | get_order(sizeof(struct location) * t->max)); | |
3469 | } | |
3470 | ||
68dff6a9 | 3471 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3472 | { |
3473 | struct location *l; | |
3474 | int order; | |
3475 | ||
88a420e4 CL |
3476 | order = get_order(sizeof(struct location) * max); |
3477 | ||
68dff6a9 | 3478 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3479 | if (!l) |
3480 | return 0; | |
3481 | ||
3482 | if (t->count) { | |
3483 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3484 | free_loc_track(t); | |
3485 | } | |
3486 | t->max = max; | |
3487 | t->loc = l; | |
3488 | return 1; | |
3489 | } | |
3490 | ||
3491 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3492 | const struct track *track) |
88a420e4 CL |
3493 | { |
3494 | long start, end, pos; | |
3495 | struct location *l; | |
ce71e27c | 3496 | unsigned long caddr; |
45edfa58 | 3497 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3498 | |
3499 | start = -1; | |
3500 | end = t->count; | |
3501 | ||
3502 | for ( ; ; ) { | |
3503 | pos = start + (end - start + 1) / 2; | |
3504 | ||
3505 | /* | |
3506 | * There is nothing at "end". If we end up there | |
3507 | * we need to add something to before end. | |
3508 | */ | |
3509 | if (pos == end) | |
3510 | break; | |
3511 | ||
3512 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3513 | if (track->addr == caddr) { |
3514 | ||
3515 | l = &t->loc[pos]; | |
3516 | l->count++; | |
3517 | if (track->when) { | |
3518 | l->sum_time += age; | |
3519 | if (age < l->min_time) | |
3520 | l->min_time = age; | |
3521 | if (age > l->max_time) | |
3522 | l->max_time = age; | |
3523 | ||
3524 | if (track->pid < l->min_pid) | |
3525 | l->min_pid = track->pid; | |
3526 | if (track->pid > l->max_pid) | |
3527 | l->max_pid = track->pid; | |
3528 | ||
174596a0 RR |
3529 | cpumask_set_cpu(track->cpu, |
3530 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3531 | } |
3532 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3533 | return 1; |
3534 | } | |
3535 | ||
45edfa58 | 3536 | if (track->addr < caddr) |
88a420e4 CL |
3537 | end = pos; |
3538 | else | |
3539 | start = pos; | |
3540 | } | |
3541 | ||
3542 | /* | |
672bba3a | 3543 | * Not found. Insert new tracking element. |
88a420e4 | 3544 | */ |
68dff6a9 | 3545 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3546 | return 0; |
3547 | ||
3548 | l = t->loc + pos; | |
3549 | if (pos < t->count) | |
3550 | memmove(l + 1, l, | |
3551 | (t->count - pos) * sizeof(struct location)); | |
3552 | t->count++; | |
3553 | l->count = 1; | |
45edfa58 CL |
3554 | l->addr = track->addr; |
3555 | l->sum_time = age; | |
3556 | l->min_time = age; | |
3557 | l->max_time = age; | |
3558 | l->min_pid = track->pid; | |
3559 | l->max_pid = track->pid; | |
174596a0 RR |
3560 | cpumask_clear(to_cpumask(l->cpus)); |
3561 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3562 | nodes_clear(l->nodes); |
3563 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3564 | return 1; |
3565 | } | |
3566 | ||
3567 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3568 | struct page *page, enum track_item alloc) | |
3569 | { | |
a973e9dd | 3570 | void *addr = page_address(page); |
39b26464 | 3571 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3572 | void *p; |
3573 | ||
39b26464 | 3574 | bitmap_zero(map, page->objects); |
7656c72b CL |
3575 | for_each_free_object(p, s, page->freelist) |
3576 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3577 | |
224a88be | 3578 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3579 | if (!test_bit(slab_index(p, s, addr), map)) |
3580 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3581 | } |
3582 | ||
3583 | static int list_locations(struct kmem_cache *s, char *buf, | |
3584 | enum track_item alloc) | |
3585 | { | |
e374d483 | 3586 | int len = 0; |
88a420e4 | 3587 | unsigned long i; |
68dff6a9 | 3588 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3589 | int node; |
3590 | ||
68dff6a9 | 3591 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3592 | GFP_TEMPORARY)) |
68dff6a9 | 3593 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3594 | |
3595 | /* Push back cpu slabs */ | |
3596 | flush_all(s); | |
3597 | ||
f64dc58c | 3598 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3599 | struct kmem_cache_node *n = get_node(s, node); |
3600 | unsigned long flags; | |
3601 | struct page *page; | |
3602 | ||
9e86943b | 3603 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3604 | continue; |
3605 | ||
3606 | spin_lock_irqsave(&n->list_lock, flags); | |
3607 | list_for_each_entry(page, &n->partial, lru) | |
3608 | process_slab(&t, s, page, alloc); | |
3609 | list_for_each_entry(page, &n->full, lru) | |
3610 | process_slab(&t, s, page, alloc); | |
3611 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3612 | } | |
3613 | ||
3614 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3615 | struct location *l = &t.loc[i]; |
88a420e4 | 3616 | |
9c246247 | 3617 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3618 | break; |
e374d483 | 3619 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3620 | |
3621 | if (l->addr) | |
e374d483 | 3622 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3623 | else |
e374d483 | 3624 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3625 | |
3626 | if (l->sum_time != l->min_time) { | |
e374d483 | 3627 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3628 | l->min_time, |
3629 | (long)div_u64(l->sum_time, l->count), | |
3630 | l->max_time); | |
45edfa58 | 3631 | } else |
e374d483 | 3632 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3633 | l->min_time); |
3634 | ||
3635 | if (l->min_pid != l->max_pid) | |
e374d483 | 3636 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3637 | l->min_pid, l->max_pid); |
3638 | else | |
e374d483 | 3639 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3640 | l->min_pid); |
3641 | ||
174596a0 RR |
3642 | if (num_online_cpus() > 1 && |
3643 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3644 | len < PAGE_SIZE - 60) { |
3645 | len += sprintf(buf + len, " cpus="); | |
3646 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3647 | to_cpumask(l->cpus)); |
45edfa58 CL |
3648 | } |
3649 | ||
84966343 | 3650 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3651 | len < PAGE_SIZE - 60) { |
3652 | len += sprintf(buf + len, " nodes="); | |
3653 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3654 | l->nodes); |
3655 | } | |
3656 | ||
e374d483 | 3657 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3658 | } |
3659 | ||
3660 | free_loc_track(&t); | |
3661 | if (!t.count) | |
e374d483 HH |
3662 | len += sprintf(buf, "No data\n"); |
3663 | return len; | |
88a420e4 CL |
3664 | } |
3665 | ||
81819f0f | 3666 | enum slab_stat_type { |
205ab99d CL |
3667 | SL_ALL, /* All slabs */ |
3668 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3669 | SL_CPU, /* Only slabs used for cpu caches */ | |
3670 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3671 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3672 | }; |
3673 | ||
205ab99d | 3674 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3675 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3676 | #define SO_CPU (1 << SL_CPU) | |
3677 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3678 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3679 | |
62e5c4b4 CG |
3680 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3681 | char *buf, unsigned long flags) | |
81819f0f CL |
3682 | { |
3683 | unsigned long total = 0; | |
81819f0f CL |
3684 | int node; |
3685 | int x; | |
3686 | unsigned long *nodes; | |
3687 | unsigned long *per_cpu; | |
3688 | ||
3689 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3690 | if (!nodes) |
3691 | return -ENOMEM; | |
81819f0f CL |
3692 | per_cpu = nodes + nr_node_ids; |
3693 | ||
205ab99d CL |
3694 | if (flags & SO_CPU) { |
3695 | int cpu; | |
81819f0f | 3696 | |
205ab99d CL |
3697 | for_each_possible_cpu(cpu) { |
3698 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3699 | |
205ab99d CL |
3700 | if (!c || c->node < 0) |
3701 | continue; | |
3702 | ||
3703 | if (c->page) { | |
3704 | if (flags & SO_TOTAL) | |
3705 | x = c->page->objects; | |
3706 | else if (flags & SO_OBJECTS) | |
3707 | x = c->page->inuse; | |
81819f0f CL |
3708 | else |
3709 | x = 1; | |
205ab99d | 3710 | |
81819f0f | 3711 | total += x; |
205ab99d | 3712 | nodes[c->node] += x; |
81819f0f | 3713 | } |
205ab99d | 3714 | per_cpu[c->node]++; |
81819f0f CL |
3715 | } |
3716 | } | |
3717 | ||
205ab99d CL |
3718 | if (flags & SO_ALL) { |
3719 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3720 | struct kmem_cache_node *n = get_node(s, node); | |
3721 | ||
3722 | if (flags & SO_TOTAL) | |
3723 | x = atomic_long_read(&n->total_objects); | |
3724 | else if (flags & SO_OBJECTS) | |
3725 | x = atomic_long_read(&n->total_objects) - | |
3726 | count_partial(n, count_free); | |
81819f0f | 3727 | |
81819f0f | 3728 | else |
205ab99d | 3729 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3730 | total += x; |
3731 | nodes[node] += x; | |
3732 | } | |
3733 | ||
205ab99d CL |
3734 | } else if (flags & SO_PARTIAL) { |
3735 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3736 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3737 | |
205ab99d CL |
3738 | if (flags & SO_TOTAL) |
3739 | x = count_partial(n, count_total); | |
3740 | else if (flags & SO_OBJECTS) | |
3741 | x = count_partial(n, count_inuse); | |
81819f0f | 3742 | else |
205ab99d | 3743 | x = n->nr_partial; |
81819f0f CL |
3744 | total += x; |
3745 | nodes[node] += x; | |
3746 | } | |
3747 | } | |
81819f0f CL |
3748 | x = sprintf(buf, "%lu", total); |
3749 | #ifdef CONFIG_NUMA | |
f64dc58c | 3750 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3751 | if (nodes[node]) |
3752 | x += sprintf(buf + x, " N%d=%lu", | |
3753 | node, nodes[node]); | |
3754 | #endif | |
3755 | kfree(nodes); | |
3756 | return x + sprintf(buf + x, "\n"); | |
3757 | } | |
3758 | ||
3759 | static int any_slab_objects(struct kmem_cache *s) | |
3760 | { | |
3761 | int node; | |
81819f0f | 3762 | |
dfb4f096 | 3763 | for_each_online_node(node) { |
81819f0f CL |
3764 | struct kmem_cache_node *n = get_node(s, node); |
3765 | ||
dfb4f096 CL |
3766 | if (!n) |
3767 | continue; | |
3768 | ||
4ea33e2d | 3769 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3770 | return 1; |
3771 | } | |
3772 | return 0; | |
3773 | } | |
3774 | ||
3775 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3776 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3777 | ||
3778 | struct slab_attribute { | |
3779 | struct attribute attr; | |
3780 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3781 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3782 | }; | |
3783 | ||
3784 | #define SLAB_ATTR_RO(_name) \ | |
3785 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3786 | ||
3787 | #define SLAB_ATTR(_name) \ | |
3788 | static struct slab_attribute _name##_attr = \ | |
3789 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3790 | ||
81819f0f CL |
3791 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3792 | { | |
3793 | return sprintf(buf, "%d\n", s->size); | |
3794 | } | |
3795 | SLAB_ATTR_RO(slab_size); | |
3796 | ||
3797 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3798 | { | |
3799 | return sprintf(buf, "%d\n", s->align); | |
3800 | } | |
3801 | SLAB_ATTR_RO(align); | |
3802 | ||
3803 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3804 | { | |
3805 | return sprintf(buf, "%d\n", s->objsize); | |
3806 | } | |
3807 | SLAB_ATTR_RO(object_size); | |
3808 | ||
3809 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3810 | { | |
834f3d11 | 3811 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3812 | } |
3813 | SLAB_ATTR_RO(objs_per_slab); | |
3814 | ||
06b285dc CL |
3815 | static ssize_t order_store(struct kmem_cache *s, |
3816 | const char *buf, size_t length) | |
3817 | { | |
0121c619 CL |
3818 | unsigned long order; |
3819 | int err; | |
3820 | ||
3821 | err = strict_strtoul(buf, 10, &order); | |
3822 | if (err) | |
3823 | return err; | |
06b285dc CL |
3824 | |
3825 | if (order > slub_max_order || order < slub_min_order) | |
3826 | return -EINVAL; | |
3827 | ||
3828 | calculate_sizes(s, order); | |
3829 | return length; | |
3830 | } | |
3831 | ||
81819f0f CL |
3832 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3833 | { | |
834f3d11 | 3834 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3835 | } |
06b285dc | 3836 | SLAB_ATTR(order); |
81819f0f CL |
3837 | |
3838 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
3839 | { | |
3840 | if (s->ctor) { | |
3841 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3842 | ||
3843 | return n + sprintf(buf + n, "\n"); | |
3844 | } | |
3845 | return 0; | |
3846 | } | |
3847 | SLAB_ATTR_RO(ctor); | |
3848 | ||
81819f0f CL |
3849 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3850 | { | |
3851 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3852 | } | |
3853 | SLAB_ATTR_RO(aliases); | |
3854 | ||
3855 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3856 | { | |
205ab99d | 3857 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3858 | } |
3859 | SLAB_ATTR_RO(slabs); | |
3860 | ||
3861 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3862 | { | |
d9acf4b7 | 3863 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3864 | } |
3865 | SLAB_ATTR_RO(partial); | |
3866 | ||
3867 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3868 | { | |
d9acf4b7 | 3869 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3870 | } |
3871 | SLAB_ATTR_RO(cpu_slabs); | |
3872 | ||
3873 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3874 | { | |
205ab99d | 3875 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3876 | } |
3877 | SLAB_ATTR_RO(objects); | |
3878 | ||
205ab99d CL |
3879 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3880 | { | |
3881 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3882 | } | |
3883 | SLAB_ATTR_RO(objects_partial); | |
3884 | ||
3885 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
3886 | { | |
3887 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
3888 | } | |
3889 | SLAB_ATTR_RO(total_objects); | |
3890 | ||
81819f0f CL |
3891 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
3892 | { | |
3893 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3894 | } | |
3895 | ||
3896 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3897 | const char *buf, size_t length) | |
3898 | { | |
3899 | s->flags &= ~SLAB_DEBUG_FREE; | |
3900 | if (buf[0] == '1') | |
3901 | s->flags |= SLAB_DEBUG_FREE; | |
3902 | return length; | |
3903 | } | |
3904 | SLAB_ATTR(sanity_checks); | |
3905 | ||
3906 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3907 | { | |
3908 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3909 | } | |
3910 | ||
3911 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
3912 | size_t length) | |
3913 | { | |
3914 | s->flags &= ~SLAB_TRACE; | |
3915 | if (buf[0] == '1') | |
3916 | s->flags |= SLAB_TRACE; | |
3917 | return length; | |
3918 | } | |
3919 | SLAB_ATTR(trace); | |
3920 | ||
3921 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
3922 | { | |
3923 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
3924 | } | |
3925 | ||
3926 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
3927 | const char *buf, size_t length) | |
3928 | { | |
3929 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
3930 | if (buf[0] == '1') | |
3931 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
3932 | return length; | |
3933 | } | |
3934 | SLAB_ATTR(reclaim_account); | |
3935 | ||
3936 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
3937 | { | |
5af60839 | 3938 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
3939 | } |
3940 | SLAB_ATTR_RO(hwcache_align); | |
3941 | ||
3942 | #ifdef CONFIG_ZONE_DMA | |
3943 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
3944 | { | |
3945 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
3946 | } | |
3947 | SLAB_ATTR_RO(cache_dma); | |
3948 | #endif | |
3949 | ||
3950 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
3951 | { | |
3952 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
3953 | } | |
3954 | SLAB_ATTR_RO(destroy_by_rcu); | |
3955 | ||
3956 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
3957 | { | |
3958 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
3959 | } | |
3960 | ||
3961 | static ssize_t red_zone_store(struct kmem_cache *s, | |
3962 | const char *buf, size_t length) | |
3963 | { | |
3964 | if (any_slab_objects(s)) | |
3965 | return -EBUSY; | |
3966 | ||
3967 | s->flags &= ~SLAB_RED_ZONE; | |
3968 | if (buf[0] == '1') | |
3969 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 3970 | calculate_sizes(s, -1); |
81819f0f CL |
3971 | return length; |
3972 | } | |
3973 | SLAB_ATTR(red_zone); | |
3974 | ||
3975 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
3976 | { | |
3977 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
3978 | } | |
3979 | ||
3980 | static ssize_t poison_store(struct kmem_cache *s, | |
3981 | const char *buf, size_t length) | |
3982 | { | |
3983 | if (any_slab_objects(s)) | |
3984 | return -EBUSY; | |
3985 | ||
3986 | s->flags &= ~SLAB_POISON; | |
3987 | if (buf[0] == '1') | |
3988 | s->flags |= SLAB_POISON; | |
06b285dc | 3989 | calculate_sizes(s, -1); |
81819f0f CL |
3990 | return length; |
3991 | } | |
3992 | SLAB_ATTR(poison); | |
3993 | ||
3994 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
3995 | { | |
3996 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
3997 | } | |
3998 | ||
3999 | static ssize_t store_user_store(struct kmem_cache *s, | |
4000 | const char *buf, size_t length) | |
4001 | { | |
4002 | if (any_slab_objects(s)) | |
4003 | return -EBUSY; | |
4004 | ||
4005 | s->flags &= ~SLAB_STORE_USER; | |
4006 | if (buf[0] == '1') | |
4007 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4008 | calculate_sizes(s, -1); |
81819f0f CL |
4009 | return length; |
4010 | } | |
4011 | SLAB_ATTR(store_user); | |
4012 | ||
53e15af0 CL |
4013 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4014 | { | |
4015 | return 0; | |
4016 | } | |
4017 | ||
4018 | static ssize_t validate_store(struct kmem_cache *s, | |
4019 | const char *buf, size_t length) | |
4020 | { | |
434e245d CL |
4021 | int ret = -EINVAL; |
4022 | ||
4023 | if (buf[0] == '1') { | |
4024 | ret = validate_slab_cache(s); | |
4025 | if (ret >= 0) | |
4026 | ret = length; | |
4027 | } | |
4028 | return ret; | |
53e15af0 CL |
4029 | } |
4030 | SLAB_ATTR(validate); | |
4031 | ||
2086d26a CL |
4032 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4033 | { | |
4034 | return 0; | |
4035 | } | |
4036 | ||
4037 | static ssize_t shrink_store(struct kmem_cache *s, | |
4038 | const char *buf, size_t length) | |
4039 | { | |
4040 | if (buf[0] == '1') { | |
4041 | int rc = kmem_cache_shrink(s); | |
4042 | ||
4043 | if (rc) | |
4044 | return rc; | |
4045 | } else | |
4046 | return -EINVAL; | |
4047 | return length; | |
4048 | } | |
4049 | SLAB_ATTR(shrink); | |
4050 | ||
88a420e4 CL |
4051 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4052 | { | |
4053 | if (!(s->flags & SLAB_STORE_USER)) | |
4054 | return -ENOSYS; | |
4055 | return list_locations(s, buf, TRACK_ALLOC); | |
4056 | } | |
4057 | SLAB_ATTR_RO(alloc_calls); | |
4058 | ||
4059 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4060 | { | |
4061 | if (!(s->flags & SLAB_STORE_USER)) | |
4062 | return -ENOSYS; | |
4063 | return list_locations(s, buf, TRACK_FREE); | |
4064 | } | |
4065 | SLAB_ATTR_RO(free_calls); | |
4066 | ||
81819f0f | 4067 | #ifdef CONFIG_NUMA |
9824601e | 4068 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4069 | { |
9824601e | 4070 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4071 | } |
4072 | ||
9824601e | 4073 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4074 | const char *buf, size_t length) |
4075 | { | |
0121c619 CL |
4076 | unsigned long ratio; |
4077 | int err; | |
4078 | ||
4079 | err = strict_strtoul(buf, 10, &ratio); | |
4080 | if (err) | |
4081 | return err; | |
4082 | ||
e2cb96b7 | 4083 | if (ratio <= 100) |
0121c619 | 4084 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4085 | |
81819f0f CL |
4086 | return length; |
4087 | } | |
9824601e | 4088 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4089 | #endif |
4090 | ||
8ff12cfc | 4091 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4092 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4093 | { | |
4094 | unsigned long sum = 0; | |
4095 | int cpu; | |
4096 | int len; | |
4097 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4098 | ||
4099 | if (!data) | |
4100 | return -ENOMEM; | |
4101 | ||
4102 | for_each_online_cpu(cpu) { | |
4103 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4104 | ||
4105 | data[cpu] = x; | |
4106 | sum += x; | |
4107 | } | |
4108 | ||
4109 | len = sprintf(buf, "%lu", sum); | |
4110 | ||
50ef37b9 | 4111 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4112 | for_each_online_cpu(cpu) { |
4113 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4114 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4115 | } |
50ef37b9 | 4116 | #endif |
8ff12cfc CL |
4117 | kfree(data); |
4118 | return len + sprintf(buf + len, "\n"); | |
4119 | } | |
4120 | ||
4121 | #define STAT_ATTR(si, text) \ | |
4122 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4123 | { \ | |
4124 | return show_stat(s, buf, si); \ | |
4125 | } \ | |
4126 | SLAB_ATTR_RO(text); \ | |
4127 | ||
4128 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4129 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4130 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4131 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4132 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4133 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4134 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4135 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4136 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4137 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4138 | STAT_ATTR(FREE_SLAB, free_slab); | |
4139 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4140 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4141 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4142 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4143 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4144 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4145 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4146 | #endif |
4147 | ||
06428780 | 4148 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4149 | &slab_size_attr.attr, |
4150 | &object_size_attr.attr, | |
4151 | &objs_per_slab_attr.attr, | |
4152 | &order_attr.attr, | |
4153 | &objects_attr.attr, | |
205ab99d CL |
4154 | &objects_partial_attr.attr, |
4155 | &total_objects_attr.attr, | |
81819f0f CL |
4156 | &slabs_attr.attr, |
4157 | &partial_attr.attr, | |
4158 | &cpu_slabs_attr.attr, | |
4159 | &ctor_attr.attr, | |
81819f0f CL |
4160 | &aliases_attr.attr, |
4161 | &align_attr.attr, | |
4162 | &sanity_checks_attr.attr, | |
4163 | &trace_attr.attr, | |
4164 | &hwcache_align_attr.attr, | |
4165 | &reclaim_account_attr.attr, | |
4166 | &destroy_by_rcu_attr.attr, | |
4167 | &red_zone_attr.attr, | |
4168 | &poison_attr.attr, | |
4169 | &store_user_attr.attr, | |
53e15af0 | 4170 | &validate_attr.attr, |
2086d26a | 4171 | &shrink_attr.attr, |
88a420e4 CL |
4172 | &alloc_calls_attr.attr, |
4173 | &free_calls_attr.attr, | |
81819f0f CL |
4174 | #ifdef CONFIG_ZONE_DMA |
4175 | &cache_dma_attr.attr, | |
4176 | #endif | |
4177 | #ifdef CONFIG_NUMA | |
9824601e | 4178 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4179 | #endif |
4180 | #ifdef CONFIG_SLUB_STATS | |
4181 | &alloc_fastpath_attr.attr, | |
4182 | &alloc_slowpath_attr.attr, | |
4183 | &free_fastpath_attr.attr, | |
4184 | &free_slowpath_attr.attr, | |
4185 | &free_frozen_attr.attr, | |
4186 | &free_add_partial_attr.attr, | |
4187 | &free_remove_partial_attr.attr, | |
4188 | &alloc_from_partial_attr.attr, | |
4189 | &alloc_slab_attr.attr, | |
4190 | &alloc_refill_attr.attr, | |
4191 | &free_slab_attr.attr, | |
4192 | &cpuslab_flush_attr.attr, | |
4193 | &deactivate_full_attr.attr, | |
4194 | &deactivate_empty_attr.attr, | |
4195 | &deactivate_to_head_attr.attr, | |
4196 | &deactivate_to_tail_attr.attr, | |
4197 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4198 | &order_fallback_attr.attr, |
81819f0f CL |
4199 | #endif |
4200 | NULL | |
4201 | }; | |
4202 | ||
4203 | static struct attribute_group slab_attr_group = { | |
4204 | .attrs = slab_attrs, | |
4205 | }; | |
4206 | ||
4207 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4208 | struct attribute *attr, | |
4209 | char *buf) | |
4210 | { | |
4211 | struct slab_attribute *attribute; | |
4212 | struct kmem_cache *s; | |
4213 | int err; | |
4214 | ||
4215 | attribute = to_slab_attr(attr); | |
4216 | s = to_slab(kobj); | |
4217 | ||
4218 | if (!attribute->show) | |
4219 | return -EIO; | |
4220 | ||
4221 | err = attribute->show(s, buf); | |
4222 | ||
4223 | return err; | |
4224 | } | |
4225 | ||
4226 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4227 | struct attribute *attr, | |
4228 | const char *buf, size_t len) | |
4229 | { | |
4230 | struct slab_attribute *attribute; | |
4231 | struct kmem_cache *s; | |
4232 | int err; | |
4233 | ||
4234 | attribute = to_slab_attr(attr); | |
4235 | s = to_slab(kobj); | |
4236 | ||
4237 | if (!attribute->store) | |
4238 | return -EIO; | |
4239 | ||
4240 | err = attribute->store(s, buf, len); | |
4241 | ||
4242 | return err; | |
4243 | } | |
4244 | ||
151c602f CL |
4245 | static void kmem_cache_release(struct kobject *kobj) |
4246 | { | |
4247 | struct kmem_cache *s = to_slab(kobj); | |
4248 | ||
4249 | kfree(s); | |
4250 | } | |
4251 | ||
81819f0f CL |
4252 | static struct sysfs_ops slab_sysfs_ops = { |
4253 | .show = slab_attr_show, | |
4254 | .store = slab_attr_store, | |
4255 | }; | |
4256 | ||
4257 | static struct kobj_type slab_ktype = { | |
4258 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4259 | .release = kmem_cache_release |
81819f0f CL |
4260 | }; |
4261 | ||
4262 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4263 | { | |
4264 | struct kobj_type *ktype = get_ktype(kobj); | |
4265 | ||
4266 | if (ktype == &slab_ktype) | |
4267 | return 1; | |
4268 | return 0; | |
4269 | } | |
4270 | ||
4271 | static struct kset_uevent_ops slab_uevent_ops = { | |
4272 | .filter = uevent_filter, | |
4273 | }; | |
4274 | ||
27c3a314 | 4275 | static struct kset *slab_kset; |
81819f0f CL |
4276 | |
4277 | #define ID_STR_LENGTH 64 | |
4278 | ||
4279 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4280 | * |
4281 | * Format :[flags-]size | |
81819f0f CL |
4282 | */ |
4283 | static char *create_unique_id(struct kmem_cache *s) | |
4284 | { | |
4285 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4286 | char *p = name; | |
4287 | ||
4288 | BUG_ON(!name); | |
4289 | ||
4290 | *p++ = ':'; | |
4291 | /* | |
4292 | * First flags affecting slabcache operations. We will only | |
4293 | * get here for aliasable slabs so we do not need to support | |
4294 | * too many flags. The flags here must cover all flags that | |
4295 | * are matched during merging to guarantee that the id is | |
4296 | * unique. | |
4297 | */ | |
4298 | if (s->flags & SLAB_CACHE_DMA) | |
4299 | *p++ = 'd'; | |
4300 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4301 | *p++ = 'a'; | |
4302 | if (s->flags & SLAB_DEBUG_FREE) | |
4303 | *p++ = 'F'; | |
4304 | if (p != name + 1) | |
4305 | *p++ = '-'; | |
4306 | p += sprintf(p, "%07d", s->size); | |
4307 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4308 | return name; | |
4309 | } | |
4310 | ||
4311 | static int sysfs_slab_add(struct kmem_cache *s) | |
4312 | { | |
4313 | int err; | |
4314 | const char *name; | |
4315 | int unmergeable; | |
4316 | ||
4317 | if (slab_state < SYSFS) | |
4318 | /* Defer until later */ | |
4319 | return 0; | |
4320 | ||
4321 | unmergeable = slab_unmergeable(s); | |
4322 | if (unmergeable) { | |
4323 | /* | |
4324 | * Slabcache can never be merged so we can use the name proper. | |
4325 | * This is typically the case for debug situations. In that | |
4326 | * case we can catch duplicate names easily. | |
4327 | */ | |
27c3a314 | 4328 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4329 | name = s->name; |
4330 | } else { | |
4331 | /* | |
4332 | * Create a unique name for the slab as a target | |
4333 | * for the symlinks. | |
4334 | */ | |
4335 | name = create_unique_id(s); | |
4336 | } | |
4337 | ||
27c3a314 | 4338 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4339 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4340 | if (err) { | |
4341 | kobject_put(&s->kobj); | |
81819f0f | 4342 | return err; |
1eada11c | 4343 | } |
81819f0f CL |
4344 | |
4345 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4346 | if (err) | |
4347 | return err; | |
4348 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4349 | if (!unmergeable) { | |
4350 | /* Setup first alias */ | |
4351 | sysfs_slab_alias(s, s->name); | |
4352 | kfree(name); | |
4353 | } | |
4354 | return 0; | |
4355 | } | |
4356 | ||
4357 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4358 | { | |
4359 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4360 | kobject_del(&s->kobj); | |
151c602f | 4361 | kobject_put(&s->kobj); |
81819f0f CL |
4362 | } |
4363 | ||
4364 | /* | |
4365 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4366 | * available lest we lose that information. |
81819f0f CL |
4367 | */ |
4368 | struct saved_alias { | |
4369 | struct kmem_cache *s; | |
4370 | const char *name; | |
4371 | struct saved_alias *next; | |
4372 | }; | |
4373 | ||
5af328a5 | 4374 | static struct saved_alias *alias_list; |
81819f0f CL |
4375 | |
4376 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4377 | { | |
4378 | struct saved_alias *al; | |
4379 | ||
4380 | if (slab_state == SYSFS) { | |
4381 | /* | |
4382 | * If we have a leftover link then remove it. | |
4383 | */ | |
27c3a314 GKH |
4384 | sysfs_remove_link(&slab_kset->kobj, name); |
4385 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4386 | } |
4387 | ||
4388 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4389 | if (!al) | |
4390 | return -ENOMEM; | |
4391 | ||
4392 | al->s = s; | |
4393 | al->name = name; | |
4394 | al->next = alias_list; | |
4395 | alias_list = al; | |
4396 | return 0; | |
4397 | } | |
4398 | ||
4399 | static int __init slab_sysfs_init(void) | |
4400 | { | |
5b95a4ac | 4401 | struct kmem_cache *s; |
81819f0f CL |
4402 | int err; |
4403 | ||
0ff21e46 | 4404 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4405 | if (!slab_kset) { |
81819f0f CL |
4406 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4407 | return -ENOSYS; | |
4408 | } | |
4409 | ||
26a7bd03 CL |
4410 | slab_state = SYSFS; |
4411 | ||
5b95a4ac | 4412 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4413 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4414 | if (err) |
4415 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4416 | " to sysfs\n", s->name); | |
26a7bd03 | 4417 | } |
81819f0f CL |
4418 | |
4419 | while (alias_list) { | |
4420 | struct saved_alias *al = alias_list; | |
4421 | ||
4422 | alias_list = alias_list->next; | |
4423 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4424 | if (err) |
4425 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4426 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4427 | kfree(al); |
4428 | } | |
4429 | ||
4430 | resiliency_test(); | |
4431 | return 0; | |
4432 | } | |
4433 | ||
4434 | __initcall(slab_sysfs_init); | |
81819f0f | 4435 | #endif |
57ed3eda PE |
4436 | |
4437 | /* | |
4438 | * The /proc/slabinfo ABI | |
4439 | */ | |
158a9624 | 4440 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4441 | static void print_slabinfo_header(struct seq_file *m) |
4442 | { | |
4443 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4444 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4445 | "<objperslab> <pagesperslab>"); | |
4446 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4447 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4448 | seq_putc(m, '\n'); | |
4449 | } | |
4450 | ||
4451 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4452 | { | |
4453 | loff_t n = *pos; | |
4454 | ||
4455 | down_read(&slub_lock); | |
4456 | if (!n) | |
4457 | print_slabinfo_header(m); | |
4458 | ||
4459 | return seq_list_start(&slab_caches, *pos); | |
4460 | } | |
4461 | ||
4462 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4463 | { | |
4464 | return seq_list_next(p, &slab_caches, pos); | |
4465 | } | |
4466 | ||
4467 | static void s_stop(struct seq_file *m, void *p) | |
4468 | { | |
4469 | up_read(&slub_lock); | |
4470 | } | |
4471 | ||
4472 | static int s_show(struct seq_file *m, void *p) | |
4473 | { | |
4474 | unsigned long nr_partials = 0; | |
4475 | unsigned long nr_slabs = 0; | |
4476 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4477 | unsigned long nr_objs = 0; |
4478 | unsigned long nr_free = 0; | |
57ed3eda PE |
4479 | struct kmem_cache *s; |
4480 | int node; | |
4481 | ||
4482 | s = list_entry(p, struct kmem_cache, list); | |
4483 | ||
4484 | for_each_online_node(node) { | |
4485 | struct kmem_cache_node *n = get_node(s, node); | |
4486 | ||
4487 | if (!n) | |
4488 | continue; | |
4489 | ||
4490 | nr_partials += n->nr_partial; | |
4491 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4492 | nr_objs += atomic_long_read(&n->total_objects); |
4493 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4494 | } |
4495 | ||
205ab99d | 4496 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4497 | |
4498 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4499 | nr_objs, s->size, oo_objects(s->oo), |
4500 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4501 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4502 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4503 | 0UL); | |
4504 | seq_putc(m, '\n'); | |
4505 | return 0; | |
4506 | } | |
4507 | ||
7b3c3a50 | 4508 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4509 | .start = s_start, |
4510 | .next = s_next, | |
4511 | .stop = s_stop, | |
4512 | .show = s_show, | |
4513 | }; | |
4514 | ||
7b3c3a50 AD |
4515 | static int slabinfo_open(struct inode *inode, struct file *file) |
4516 | { | |
4517 | return seq_open(file, &slabinfo_op); | |
4518 | } | |
4519 | ||
4520 | static const struct file_operations proc_slabinfo_operations = { | |
4521 | .open = slabinfo_open, | |
4522 | .read = seq_read, | |
4523 | .llseek = seq_lseek, | |
4524 | .release = seq_release, | |
4525 | }; | |
4526 | ||
4527 | static int __init slab_proc_init(void) | |
4528 | { | |
4529 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); | |
4530 | return 0; | |
4531 | } | |
4532 | module_init(slab_proc_init); | |
158a9624 | 4533 | #endif /* CONFIG_SLABINFO */ |