]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
9 | */ |
10 | ||
11 | #include <linux/mm.h> | |
1eb5ac64 | 12 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
13 | #include <linux/module.h> |
14 | #include <linux/bit_spinlock.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/bitops.h> | |
17 | #include <linux/slab.h> | |
7b3c3a50 | 18 | #include <linux/proc_fs.h> |
81819f0f | 19 | #include <linux/seq_file.h> |
02af61bb | 20 | #include <linux/kmemtrace.h> |
5a896d9e | 21 | #include <linux/kmemcheck.h> |
81819f0f CL |
22 | #include <linux/cpu.h> |
23 | #include <linux/cpuset.h> | |
24 | #include <linux/mempolicy.h> | |
25 | #include <linux/ctype.h> | |
3ac7fe5a | 26 | #include <linux/debugobjects.h> |
81819f0f | 27 | #include <linux/kallsyms.h> |
b9049e23 | 28 | #include <linux/memory.h> |
f8bd2258 | 29 | #include <linux/math64.h> |
773ff60e | 30 | #include <linux/fault-inject.h> |
81819f0f CL |
31 | |
32 | /* | |
33 | * Lock order: | |
34 | * 1. slab_lock(page) | |
35 | * 2. slab->list_lock | |
36 | * | |
37 | * The slab_lock protects operations on the object of a particular | |
38 | * slab and its metadata in the page struct. If the slab lock | |
39 | * has been taken then no allocations nor frees can be performed | |
40 | * on the objects in the slab nor can the slab be added or removed | |
41 | * from the partial or full lists since this would mean modifying | |
42 | * the page_struct of the slab. | |
43 | * | |
44 | * The list_lock protects the partial and full list on each node and | |
45 | * the partial slab counter. If taken then no new slabs may be added or | |
46 | * removed from the lists nor make the number of partial slabs be modified. | |
47 | * (Note that the total number of slabs is an atomic value that may be | |
48 | * modified without taking the list lock). | |
49 | * | |
50 | * The list_lock is a centralized lock and thus we avoid taking it as | |
51 | * much as possible. As long as SLUB does not have to handle partial | |
52 | * slabs, operations can continue without any centralized lock. F.e. | |
53 | * allocating a long series of objects that fill up slabs does not require | |
54 | * the list lock. | |
55 | * | |
56 | * The lock order is sometimes inverted when we are trying to get a slab | |
57 | * off a list. We take the list_lock and then look for a page on the list | |
58 | * to use. While we do that objects in the slabs may be freed. We can | |
59 | * only operate on the slab if we have also taken the slab_lock. So we use | |
60 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
61 | * can occur anymore and we can use the slab for allocations etc. If the | |
62 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
63 | * we must stay away from it for a while since we may cause a bouncing | |
64 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
65 | * If all pages are busy then we may allocate a new slab instead of reusing | |
66 | * a partial slab. A new slab has noone operating on it and thus there is | |
67 | * no danger of cacheline contention. | |
68 | * | |
69 | * Interrupts are disabled during allocation and deallocation in order to | |
70 | * make the slab allocator safe to use in the context of an irq. In addition | |
71 | * interrupts are disabled to ensure that the processor does not change | |
72 | * while handling per_cpu slabs, due to kernel preemption. | |
73 | * | |
74 | * SLUB assigns one slab for allocation to each processor. | |
75 | * Allocations only occur from these slabs called cpu slabs. | |
76 | * | |
672bba3a CL |
77 | * Slabs with free elements are kept on a partial list and during regular |
78 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 79 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
80 | * We track full slabs for debugging purposes though because otherwise we |
81 | * cannot scan all objects. | |
81819f0f CL |
82 | * |
83 | * Slabs are freed when they become empty. Teardown and setup is | |
84 | * minimal so we rely on the page allocators per cpu caches for | |
85 | * fast frees and allocs. | |
86 | * | |
87 | * Overloading of page flags that are otherwise used for LRU management. | |
88 | * | |
4b6f0750 CL |
89 | * PageActive The slab is frozen and exempt from list processing. |
90 | * This means that the slab is dedicated to a purpose | |
91 | * such as satisfying allocations for a specific | |
92 | * processor. Objects may be freed in the slab while | |
93 | * it is frozen but slab_free will then skip the usual | |
94 | * list operations. It is up to the processor holding | |
95 | * the slab to integrate the slab into the slab lists | |
96 | * when the slab is no longer needed. | |
97 | * | |
98 | * One use of this flag is to mark slabs that are | |
99 | * used for allocations. Then such a slab becomes a cpu | |
100 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 101 | * freelist that allows lockless access to |
894b8788 CL |
102 | * free objects in addition to the regular freelist |
103 | * that requires the slab lock. | |
81819f0f CL |
104 | * |
105 | * PageError Slab requires special handling due to debug | |
106 | * options set. This moves slab handling out of | |
894b8788 | 107 | * the fast path and disables lockless freelists. |
81819f0f CL |
108 | */ |
109 | ||
5577bd8a | 110 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 111 | #define SLABDEBUG 1 |
5577bd8a CL |
112 | #else |
113 | #define SLABDEBUG 0 | |
114 | #endif | |
115 | ||
81819f0f CL |
116 | /* |
117 | * Issues still to be resolved: | |
118 | * | |
81819f0f CL |
119 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
120 | * | |
81819f0f CL |
121 | * - Variable sizing of the per node arrays |
122 | */ | |
123 | ||
124 | /* Enable to test recovery from slab corruption on boot */ | |
125 | #undef SLUB_RESILIENCY_TEST | |
126 | ||
2086d26a CL |
127 | /* |
128 | * Mininum number of partial slabs. These will be left on the partial | |
129 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
130 | */ | |
76be8950 | 131 | #define MIN_PARTIAL 5 |
e95eed57 | 132 | |
2086d26a CL |
133 | /* |
134 | * Maximum number of desirable partial slabs. | |
135 | * The existence of more partial slabs makes kmem_cache_shrink | |
136 | * sort the partial list by the number of objects in the. | |
137 | */ | |
138 | #define MAX_PARTIAL 10 | |
139 | ||
81819f0f CL |
140 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
141 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 142 | |
81819f0f CL |
143 | /* |
144 | * Set of flags that will prevent slab merging | |
145 | */ | |
146 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
06f22f13 | 147 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE) |
81819f0f CL |
148 | |
149 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 150 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f CL |
151 | |
152 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 153 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
154 | #endif |
155 | ||
156 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 157 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
158 | #endif |
159 | ||
210b5c06 CG |
160 | #define OO_SHIFT 16 |
161 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
162 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
163 | ||
81819f0f | 164 | /* Internal SLUB flags */ |
1ceef402 CL |
165 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
166 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
167 | |
168 | static int kmem_size = sizeof(struct kmem_cache); | |
169 | ||
170 | #ifdef CONFIG_SMP | |
171 | static struct notifier_block slab_notifier; | |
172 | #endif | |
173 | ||
174 | static enum { | |
175 | DOWN, /* No slab functionality available */ | |
176 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 177 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
178 | SYSFS /* Sysfs up */ |
179 | } slab_state = DOWN; | |
180 | ||
181 | /* A list of all slab caches on the system */ | |
182 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 183 | static LIST_HEAD(slab_caches); |
81819f0f | 184 | |
02cbc874 CL |
185 | /* |
186 | * Tracking user of a slab. | |
187 | */ | |
188 | struct track { | |
ce71e27c | 189 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
190 | int cpu; /* Was running on cpu */ |
191 | int pid; /* Pid context */ | |
192 | unsigned long when; /* When did the operation occur */ | |
193 | }; | |
194 | ||
195 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
196 | ||
f6acb635 | 197 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
198 | static int sysfs_slab_add(struct kmem_cache *); |
199 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
200 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 201 | |
81819f0f | 202 | #else |
0c710013 CL |
203 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
204 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
205 | { return 0; } | |
151c602f CL |
206 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
207 | { | |
208 | kfree(s); | |
209 | } | |
8ff12cfc | 210 | |
81819f0f CL |
211 | #endif |
212 | ||
8ff12cfc CL |
213 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
214 | { | |
215 | #ifdef CONFIG_SLUB_STATS | |
216 | c->stat[si]++; | |
217 | #endif | |
218 | } | |
219 | ||
81819f0f CL |
220 | /******************************************************************** |
221 | * Core slab cache functions | |
222 | *******************************************************************/ | |
223 | ||
224 | int slab_is_available(void) | |
225 | { | |
226 | return slab_state >= UP; | |
227 | } | |
228 | ||
229 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
230 | { | |
231 | #ifdef CONFIG_NUMA | |
232 | return s->node[node]; | |
233 | #else | |
234 | return &s->local_node; | |
235 | #endif | |
236 | } | |
237 | ||
dfb4f096 CL |
238 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
239 | { | |
4c93c355 CL |
240 | #ifdef CONFIG_SMP |
241 | return s->cpu_slab[cpu]; | |
242 | #else | |
243 | return &s->cpu_slab; | |
244 | #endif | |
dfb4f096 CL |
245 | } |
246 | ||
6446faa2 | 247 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
248 | static inline int check_valid_pointer(struct kmem_cache *s, |
249 | struct page *page, const void *object) | |
250 | { | |
251 | void *base; | |
252 | ||
a973e9dd | 253 | if (!object) |
02cbc874 CL |
254 | return 1; |
255 | ||
a973e9dd | 256 | base = page_address(page); |
39b26464 | 257 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
258 | (object - base) % s->size) { |
259 | return 0; | |
260 | } | |
261 | ||
262 | return 1; | |
263 | } | |
264 | ||
7656c72b CL |
265 | /* |
266 | * Slow version of get and set free pointer. | |
267 | * | |
268 | * This version requires touching the cache lines of kmem_cache which | |
269 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
270 | * from the page struct. | |
271 | */ | |
272 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
273 | { | |
274 | return *(void **)(object + s->offset); | |
275 | } | |
276 | ||
277 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
278 | { | |
279 | *(void **)(object + s->offset) = fp; | |
280 | } | |
281 | ||
282 | /* Loop over all objects in a slab */ | |
224a88be CL |
283 | #define for_each_object(__p, __s, __addr, __objects) \ |
284 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
285 | __p += (__s)->size) |
286 | ||
287 | /* Scan freelist */ | |
288 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 289 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
290 | |
291 | /* Determine object index from a given position */ | |
292 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
293 | { | |
294 | return (p - addr) / s->size; | |
295 | } | |
296 | ||
834f3d11 CL |
297 | static inline struct kmem_cache_order_objects oo_make(int order, |
298 | unsigned long size) | |
299 | { | |
300 | struct kmem_cache_order_objects x = { | |
210b5c06 | 301 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
302 | }; |
303 | ||
304 | return x; | |
305 | } | |
306 | ||
307 | static inline int oo_order(struct kmem_cache_order_objects x) | |
308 | { | |
210b5c06 | 309 | return x.x >> OO_SHIFT; |
834f3d11 CL |
310 | } |
311 | ||
312 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
313 | { | |
210b5c06 | 314 | return x.x & OO_MASK; |
834f3d11 CL |
315 | } |
316 | ||
41ecc55b CL |
317 | #ifdef CONFIG_SLUB_DEBUG |
318 | /* | |
319 | * Debug settings: | |
320 | */ | |
f0630fff CL |
321 | #ifdef CONFIG_SLUB_DEBUG_ON |
322 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
323 | #else | |
41ecc55b | 324 | static int slub_debug; |
f0630fff | 325 | #endif |
41ecc55b CL |
326 | |
327 | static char *slub_debug_slabs; | |
328 | ||
81819f0f CL |
329 | /* |
330 | * Object debugging | |
331 | */ | |
332 | static void print_section(char *text, u8 *addr, unsigned int length) | |
333 | { | |
334 | int i, offset; | |
335 | int newline = 1; | |
336 | char ascii[17]; | |
337 | ||
338 | ascii[16] = 0; | |
339 | ||
340 | for (i = 0; i < length; i++) { | |
341 | if (newline) { | |
24922684 | 342 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
343 | newline = 0; |
344 | } | |
06428780 | 345 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
346 | offset = i % 16; |
347 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
348 | if (offset == 15) { | |
06428780 | 349 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
350 | newline = 1; |
351 | } | |
352 | } | |
353 | if (!newline) { | |
354 | i %= 16; | |
355 | while (i < 16) { | |
06428780 | 356 | printk(KERN_CONT " "); |
81819f0f CL |
357 | ascii[i] = ' '; |
358 | i++; | |
359 | } | |
06428780 | 360 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
361 | } |
362 | } | |
363 | ||
81819f0f CL |
364 | static struct track *get_track(struct kmem_cache *s, void *object, |
365 | enum track_item alloc) | |
366 | { | |
367 | struct track *p; | |
368 | ||
369 | if (s->offset) | |
370 | p = object + s->offset + sizeof(void *); | |
371 | else | |
372 | p = object + s->inuse; | |
373 | ||
374 | return p + alloc; | |
375 | } | |
376 | ||
377 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 378 | enum track_item alloc, unsigned long addr) |
81819f0f | 379 | { |
1a00df4a | 380 | struct track *p = get_track(s, object, alloc); |
81819f0f | 381 | |
81819f0f CL |
382 | if (addr) { |
383 | p->addr = addr; | |
384 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 385 | p->pid = current->pid; |
81819f0f CL |
386 | p->when = jiffies; |
387 | } else | |
388 | memset(p, 0, sizeof(struct track)); | |
389 | } | |
390 | ||
81819f0f CL |
391 | static void init_tracking(struct kmem_cache *s, void *object) |
392 | { | |
24922684 CL |
393 | if (!(s->flags & SLAB_STORE_USER)) |
394 | return; | |
395 | ||
ce71e27c EGM |
396 | set_track(s, object, TRACK_FREE, 0UL); |
397 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
398 | } |
399 | ||
400 | static void print_track(const char *s, struct track *t) | |
401 | { | |
402 | if (!t->addr) | |
403 | return; | |
404 | ||
7daf705f | 405 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 406 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
407 | } |
408 | ||
409 | static void print_tracking(struct kmem_cache *s, void *object) | |
410 | { | |
411 | if (!(s->flags & SLAB_STORE_USER)) | |
412 | return; | |
413 | ||
414 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
415 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
416 | } | |
417 | ||
418 | static void print_page_info(struct page *page) | |
419 | { | |
39b26464 CL |
420 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
421 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
422 | |
423 | } | |
424 | ||
425 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
426 | { | |
427 | va_list args; | |
428 | char buf[100]; | |
429 | ||
430 | va_start(args, fmt); | |
431 | vsnprintf(buf, sizeof(buf), fmt, args); | |
432 | va_end(args); | |
433 | printk(KERN_ERR "========================================" | |
434 | "=====================================\n"); | |
435 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
436 | printk(KERN_ERR "----------------------------------------" | |
437 | "-------------------------------------\n\n"); | |
81819f0f CL |
438 | } |
439 | ||
24922684 CL |
440 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
441 | { | |
442 | va_list args; | |
443 | char buf[100]; | |
444 | ||
445 | va_start(args, fmt); | |
446 | vsnprintf(buf, sizeof(buf), fmt, args); | |
447 | va_end(args); | |
448 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
449 | } | |
450 | ||
451 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
452 | { |
453 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 454 | u8 *addr = page_address(page); |
24922684 CL |
455 | |
456 | print_tracking(s, p); | |
457 | ||
458 | print_page_info(page); | |
459 | ||
460 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
461 | p, p - addr, get_freepointer(s, p)); | |
462 | ||
463 | if (p > addr + 16) | |
464 | print_section("Bytes b4", p - 16, 16); | |
465 | ||
0ebd652b | 466 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
467 | |
468 | if (s->flags & SLAB_RED_ZONE) | |
469 | print_section("Redzone", p + s->objsize, | |
470 | s->inuse - s->objsize); | |
471 | ||
81819f0f CL |
472 | if (s->offset) |
473 | off = s->offset + sizeof(void *); | |
474 | else | |
475 | off = s->inuse; | |
476 | ||
24922684 | 477 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 478 | off += 2 * sizeof(struct track); |
81819f0f CL |
479 | |
480 | if (off != s->size) | |
481 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
482 | print_section("Padding", p + off, s->size - off); |
483 | ||
484 | dump_stack(); | |
81819f0f CL |
485 | } |
486 | ||
487 | static void object_err(struct kmem_cache *s, struct page *page, | |
488 | u8 *object, char *reason) | |
489 | { | |
3dc50637 | 490 | slab_bug(s, "%s", reason); |
24922684 | 491 | print_trailer(s, page, object); |
81819f0f CL |
492 | } |
493 | ||
24922684 | 494 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
495 | { |
496 | va_list args; | |
497 | char buf[100]; | |
498 | ||
24922684 CL |
499 | va_start(args, fmt); |
500 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 501 | va_end(args); |
3dc50637 | 502 | slab_bug(s, "%s", buf); |
24922684 | 503 | print_page_info(page); |
81819f0f CL |
504 | dump_stack(); |
505 | } | |
506 | ||
507 | static void init_object(struct kmem_cache *s, void *object, int active) | |
508 | { | |
509 | u8 *p = object; | |
510 | ||
511 | if (s->flags & __OBJECT_POISON) { | |
512 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 513 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
514 | } |
515 | ||
516 | if (s->flags & SLAB_RED_ZONE) | |
517 | memset(p + s->objsize, | |
518 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
519 | s->inuse - s->objsize); | |
520 | } | |
521 | ||
24922684 | 522 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
523 | { |
524 | while (bytes) { | |
525 | if (*start != (u8)value) | |
24922684 | 526 | return start; |
81819f0f CL |
527 | start++; |
528 | bytes--; | |
529 | } | |
24922684 CL |
530 | return NULL; |
531 | } | |
532 | ||
533 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
534 | void *from, void *to) | |
535 | { | |
536 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
537 | memset(from, data, to - from); | |
538 | } | |
539 | ||
540 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
541 | u8 *object, char *what, | |
06428780 | 542 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
543 | { |
544 | u8 *fault; | |
545 | u8 *end; | |
546 | ||
547 | fault = check_bytes(start, value, bytes); | |
548 | if (!fault) | |
549 | return 1; | |
550 | ||
551 | end = start + bytes; | |
552 | while (end > fault && end[-1] == value) | |
553 | end--; | |
554 | ||
555 | slab_bug(s, "%s overwritten", what); | |
556 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
557 | fault, end - 1, fault[0], value); | |
558 | print_trailer(s, page, object); | |
559 | ||
560 | restore_bytes(s, what, value, fault, end); | |
561 | return 0; | |
81819f0f CL |
562 | } |
563 | ||
81819f0f CL |
564 | /* |
565 | * Object layout: | |
566 | * | |
567 | * object address | |
568 | * Bytes of the object to be managed. | |
569 | * If the freepointer may overlay the object then the free | |
570 | * pointer is the first word of the object. | |
672bba3a | 571 | * |
81819f0f CL |
572 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
573 | * 0xa5 (POISON_END) | |
574 | * | |
575 | * object + s->objsize | |
576 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
577 | * Padding is extended by another word if Redzoning is enabled and |
578 | * objsize == inuse. | |
579 | * | |
81819f0f CL |
580 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
581 | * 0xcc (RED_ACTIVE) for objects in use. | |
582 | * | |
583 | * object + s->inuse | |
672bba3a CL |
584 | * Meta data starts here. |
585 | * | |
81819f0f CL |
586 | * A. Free pointer (if we cannot overwrite object on free) |
587 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 588 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 589 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
590 | * before the word boundary. |
591 | * | |
592 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
593 | * |
594 | * object + s->size | |
672bba3a | 595 | * Nothing is used beyond s->size. |
81819f0f | 596 | * |
672bba3a CL |
597 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
598 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
599 | * may be used with merged slabcaches. |
600 | */ | |
601 | ||
81819f0f CL |
602 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
603 | { | |
604 | unsigned long off = s->inuse; /* The end of info */ | |
605 | ||
606 | if (s->offset) | |
607 | /* Freepointer is placed after the object. */ | |
608 | off += sizeof(void *); | |
609 | ||
610 | if (s->flags & SLAB_STORE_USER) | |
611 | /* We also have user information there */ | |
612 | off += 2 * sizeof(struct track); | |
613 | ||
614 | if (s->size == off) | |
615 | return 1; | |
616 | ||
24922684 CL |
617 | return check_bytes_and_report(s, page, p, "Object padding", |
618 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
619 | } |
620 | ||
39b26464 | 621 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
622 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
623 | { | |
24922684 CL |
624 | u8 *start; |
625 | u8 *fault; | |
626 | u8 *end; | |
627 | int length; | |
628 | int remainder; | |
81819f0f CL |
629 | |
630 | if (!(s->flags & SLAB_POISON)) | |
631 | return 1; | |
632 | ||
a973e9dd | 633 | start = page_address(page); |
834f3d11 | 634 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
635 | end = start + length; |
636 | remainder = length % s->size; | |
81819f0f CL |
637 | if (!remainder) |
638 | return 1; | |
639 | ||
39b26464 | 640 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
641 | if (!fault) |
642 | return 1; | |
643 | while (end > fault && end[-1] == POISON_INUSE) | |
644 | end--; | |
645 | ||
646 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 647 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
648 | |
649 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
650 | return 0; | |
81819f0f CL |
651 | } |
652 | ||
653 | static int check_object(struct kmem_cache *s, struct page *page, | |
654 | void *object, int active) | |
655 | { | |
656 | u8 *p = object; | |
657 | u8 *endobject = object + s->objsize; | |
658 | ||
659 | if (s->flags & SLAB_RED_ZONE) { | |
660 | unsigned int red = | |
661 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
662 | ||
24922684 CL |
663 | if (!check_bytes_and_report(s, page, object, "Redzone", |
664 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 665 | return 0; |
81819f0f | 666 | } else { |
3adbefee IM |
667 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
668 | check_bytes_and_report(s, page, p, "Alignment padding", | |
669 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
670 | } | |
81819f0f CL |
671 | } |
672 | ||
673 | if (s->flags & SLAB_POISON) { | |
674 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
675 | (!check_bytes_and_report(s, page, p, "Poison", p, |
676 | POISON_FREE, s->objsize - 1) || | |
677 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 678 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 679 | return 0; |
81819f0f CL |
680 | /* |
681 | * check_pad_bytes cleans up on its own. | |
682 | */ | |
683 | check_pad_bytes(s, page, p); | |
684 | } | |
685 | ||
686 | if (!s->offset && active) | |
687 | /* | |
688 | * Object and freepointer overlap. Cannot check | |
689 | * freepointer while object is allocated. | |
690 | */ | |
691 | return 1; | |
692 | ||
693 | /* Check free pointer validity */ | |
694 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
695 | object_err(s, page, p, "Freepointer corrupt"); | |
696 | /* | |
9f6c708e | 697 | * No choice but to zap it and thus lose the remainder |
81819f0f | 698 | * of the free objects in this slab. May cause |
672bba3a | 699 | * another error because the object count is now wrong. |
81819f0f | 700 | */ |
a973e9dd | 701 | set_freepointer(s, p, NULL); |
81819f0f CL |
702 | return 0; |
703 | } | |
704 | return 1; | |
705 | } | |
706 | ||
707 | static int check_slab(struct kmem_cache *s, struct page *page) | |
708 | { | |
39b26464 CL |
709 | int maxobj; |
710 | ||
81819f0f CL |
711 | VM_BUG_ON(!irqs_disabled()); |
712 | ||
713 | if (!PageSlab(page)) { | |
24922684 | 714 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
715 | return 0; |
716 | } | |
39b26464 CL |
717 | |
718 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
719 | if (page->objects > maxobj) { | |
720 | slab_err(s, page, "objects %u > max %u", | |
721 | s->name, page->objects, maxobj); | |
722 | return 0; | |
723 | } | |
724 | if (page->inuse > page->objects) { | |
24922684 | 725 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 726 | s->name, page->inuse, page->objects); |
81819f0f CL |
727 | return 0; |
728 | } | |
729 | /* Slab_pad_check fixes things up after itself */ | |
730 | slab_pad_check(s, page); | |
731 | return 1; | |
732 | } | |
733 | ||
734 | /* | |
672bba3a CL |
735 | * Determine if a certain object on a page is on the freelist. Must hold the |
736 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
737 | */ |
738 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
739 | { | |
740 | int nr = 0; | |
741 | void *fp = page->freelist; | |
742 | void *object = NULL; | |
224a88be | 743 | unsigned long max_objects; |
81819f0f | 744 | |
39b26464 | 745 | while (fp && nr <= page->objects) { |
81819f0f CL |
746 | if (fp == search) |
747 | return 1; | |
748 | if (!check_valid_pointer(s, page, fp)) { | |
749 | if (object) { | |
750 | object_err(s, page, object, | |
751 | "Freechain corrupt"); | |
a973e9dd | 752 | set_freepointer(s, object, NULL); |
81819f0f CL |
753 | break; |
754 | } else { | |
24922684 | 755 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 756 | page->freelist = NULL; |
39b26464 | 757 | page->inuse = page->objects; |
24922684 | 758 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
759 | return 0; |
760 | } | |
761 | break; | |
762 | } | |
763 | object = fp; | |
764 | fp = get_freepointer(s, object); | |
765 | nr++; | |
766 | } | |
767 | ||
224a88be | 768 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
769 | if (max_objects > MAX_OBJS_PER_PAGE) |
770 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
771 | |
772 | if (page->objects != max_objects) { | |
773 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
774 | "should be %d", page->objects, max_objects); | |
775 | page->objects = max_objects; | |
776 | slab_fix(s, "Number of objects adjusted."); | |
777 | } | |
39b26464 | 778 | if (page->inuse != page->objects - nr) { |
70d71228 | 779 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
780 | "counted were %d", page->inuse, page->objects - nr); |
781 | page->inuse = page->objects - nr; | |
24922684 | 782 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
783 | } |
784 | return search == NULL; | |
785 | } | |
786 | ||
0121c619 CL |
787 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
788 | int alloc) | |
3ec09742 CL |
789 | { |
790 | if (s->flags & SLAB_TRACE) { | |
791 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
792 | s->name, | |
793 | alloc ? "alloc" : "free", | |
794 | object, page->inuse, | |
795 | page->freelist); | |
796 | ||
797 | if (!alloc) | |
798 | print_section("Object", (void *)object, s->objsize); | |
799 | ||
800 | dump_stack(); | |
801 | } | |
802 | } | |
803 | ||
643b1138 | 804 | /* |
672bba3a | 805 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 806 | */ |
e95eed57 | 807 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 808 | { |
643b1138 CL |
809 | spin_lock(&n->list_lock); |
810 | list_add(&page->lru, &n->full); | |
811 | spin_unlock(&n->list_lock); | |
812 | } | |
813 | ||
814 | static void remove_full(struct kmem_cache *s, struct page *page) | |
815 | { | |
816 | struct kmem_cache_node *n; | |
817 | ||
818 | if (!(s->flags & SLAB_STORE_USER)) | |
819 | return; | |
820 | ||
821 | n = get_node(s, page_to_nid(page)); | |
822 | ||
823 | spin_lock(&n->list_lock); | |
824 | list_del(&page->lru); | |
825 | spin_unlock(&n->list_lock); | |
826 | } | |
827 | ||
0f389ec6 CL |
828 | /* Tracking of the number of slabs for debugging purposes */ |
829 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
830 | { | |
831 | struct kmem_cache_node *n = get_node(s, node); | |
832 | ||
833 | return atomic_long_read(&n->nr_slabs); | |
834 | } | |
835 | ||
26c02cf0 AB |
836 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
837 | { | |
838 | return atomic_long_read(&n->nr_slabs); | |
839 | } | |
840 | ||
205ab99d | 841 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
842 | { |
843 | struct kmem_cache_node *n = get_node(s, node); | |
844 | ||
845 | /* | |
846 | * May be called early in order to allocate a slab for the | |
847 | * kmem_cache_node structure. Solve the chicken-egg | |
848 | * dilemma by deferring the increment of the count during | |
849 | * bootstrap (see early_kmem_cache_node_alloc). | |
850 | */ | |
205ab99d | 851 | if (!NUMA_BUILD || n) { |
0f389ec6 | 852 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
853 | atomic_long_add(objects, &n->total_objects); |
854 | } | |
0f389ec6 | 855 | } |
205ab99d | 856 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
857 | { |
858 | struct kmem_cache_node *n = get_node(s, node); | |
859 | ||
860 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 861 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
862 | } |
863 | ||
864 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
865 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
866 | void *object) | |
867 | { | |
868 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
869 | return; | |
870 | ||
871 | init_object(s, object, 0); | |
872 | init_tracking(s, object); | |
873 | } | |
874 | ||
875 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 876 | void *object, unsigned long addr) |
81819f0f CL |
877 | { |
878 | if (!check_slab(s, page)) | |
879 | goto bad; | |
880 | ||
d692ef6d | 881 | if (!on_freelist(s, page, object)) { |
24922684 | 882 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 883 | goto bad; |
81819f0f CL |
884 | } |
885 | ||
886 | if (!check_valid_pointer(s, page, object)) { | |
887 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 888 | goto bad; |
81819f0f CL |
889 | } |
890 | ||
d692ef6d | 891 | if (!check_object(s, page, object, 0)) |
81819f0f | 892 | goto bad; |
81819f0f | 893 | |
3ec09742 CL |
894 | /* Success perform special debug activities for allocs */ |
895 | if (s->flags & SLAB_STORE_USER) | |
896 | set_track(s, object, TRACK_ALLOC, addr); | |
897 | trace(s, page, object, 1); | |
898 | init_object(s, object, 1); | |
81819f0f | 899 | return 1; |
3ec09742 | 900 | |
81819f0f CL |
901 | bad: |
902 | if (PageSlab(page)) { | |
903 | /* | |
904 | * If this is a slab page then lets do the best we can | |
905 | * to avoid issues in the future. Marking all objects | |
672bba3a | 906 | * as used avoids touching the remaining objects. |
81819f0f | 907 | */ |
24922684 | 908 | slab_fix(s, "Marking all objects used"); |
39b26464 | 909 | page->inuse = page->objects; |
a973e9dd | 910 | page->freelist = NULL; |
81819f0f CL |
911 | } |
912 | return 0; | |
913 | } | |
914 | ||
3ec09742 | 915 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 916 | void *object, unsigned long addr) |
81819f0f CL |
917 | { |
918 | if (!check_slab(s, page)) | |
919 | goto fail; | |
920 | ||
921 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 922 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
923 | goto fail; |
924 | } | |
925 | ||
926 | if (on_freelist(s, page, object)) { | |
24922684 | 927 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
928 | goto fail; |
929 | } | |
930 | ||
931 | if (!check_object(s, page, object, 1)) | |
932 | return 0; | |
933 | ||
934 | if (unlikely(s != page->slab)) { | |
3adbefee | 935 | if (!PageSlab(page)) { |
70d71228 CL |
936 | slab_err(s, page, "Attempt to free object(0x%p) " |
937 | "outside of slab", object); | |
3adbefee | 938 | } else if (!page->slab) { |
81819f0f | 939 | printk(KERN_ERR |
70d71228 | 940 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 941 | object); |
70d71228 | 942 | dump_stack(); |
06428780 | 943 | } else |
24922684 CL |
944 | object_err(s, page, object, |
945 | "page slab pointer corrupt."); | |
81819f0f CL |
946 | goto fail; |
947 | } | |
3ec09742 CL |
948 | |
949 | /* Special debug activities for freeing objects */ | |
8a38082d | 950 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
951 | remove_full(s, page); |
952 | if (s->flags & SLAB_STORE_USER) | |
953 | set_track(s, object, TRACK_FREE, addr); | |
954 | trace(s, page, object, 0); | |
955 | init_object(s, object, 0); | |
81819f0f | 956 | return 1; |
3ec09742 | 957 | |
81819f0f | 958 | fail: |
24922684 | 959 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
960 | return 0; |
961 | } | |
962 | ||
41ecc55b CL |
963 | static int __init setup_slub_debug(char *str) |
964 | { | |
f0630fff CL |
965 | slub_debug = DEBUG_DEFAULT_FLAGS; |
966 | if (*str++ != '=' || !*str) | |
967 | /* | |
968 | * No options specified. Switch on full debugging. | |
969 | */ | |
970 | goto out; | |
971 | ||
972 | if (*str == ',') | |
973 | /* | |
974 | * No options but restriction on slabs. This means full | |
975 | * debugging for slabs matching a pattern. | |
976 | */ | |
977 | goto check_slabs; | |
978 | ||
979 | slub_debug = 0; | |
980 | if (*str == '-') | |
981 | /* | |
982 | * Switch off all debugging measures. | |
983 | */ | |
984 | goto out; | |
985 | ||
986 | /* | |
987 | * Determine which debug features should be switched on | |
988 | */ | |
06428780 | 989 | for (; *str && *str != ','; str++) { |
f0630fff CL |
990 | switch (tolower(*str)) { |
991 | case 'f': | |
992 | slub_debug |= SLAB_DEBUG_FREE; | |
993 | break; | |
994 | case 'z': | |
995 | slub_debug |= SLAB_RED_ZONE; | |
996 | break; | |
997 | case 'p': | |
998 | slub_debug |= SLAB_POISON; | |
999 | break; | |
1000 | case 'u': | |
1001 | slub_debug |= SLAB_STORE_USER; | |
1002 | break; | |
1003 | case 't': | |
1004 | slub_debug |= SLAB_TRACE; | |
1005 | break; | |
1006 | default: | |
1007 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1008 | "unknown. skipped\n", *str); |
f0630fff | 1009 | } |
41ecc55b CL |
1010 | } |
1011 | ||
f0630fff | 1012 | check_slabs: |
41ecc55b CL |
1013 | if (*str == ',') |
1014 | slub_debug_slabs = str + 1; | |
f0630fff | 1015 | out: |
41ecc55b CL |
1016 | return 1; |
1017 | } | |
1018 | ||
1019 | __setup("slub_debug", setup_slub_debug); | |
1020 | ||
ba0268a8 CL |
1021 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1022 | unsigned long flags, const char *name, | |
51cc5068 | 1023 | void (*ctor)(void *)) |
41ecc55b CL |
1024 | { |
1025 | /* | |
e153362a | 1026 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1027 | */ |
e153362a CL |
1028 | if (slub_debug && (!slub_debug_slabs || |
1029 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1030 | flags |= slub_debug; | |
ba0268a8 CL |
1031 | |
1032 | return flags; | |
41ecc55b CL |
1033 | } |
1034 | #else | |
3ec09742 CL |
1035 | static inline void setup_object_debug(struct kmem_cache *s, |
1036 | struct page *page, void *object) {} | |
41ecc55b | 1037 | |
3ec09742 | 1038 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1039 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1040 | |
3ec09742 | 1041 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1042 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1043 | |
41ecc55b CL |
1044 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1045 | { return 1; } | |
1046 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1047 | void *object, int active) { return 1; } | |
3ec09742 | 1048 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1049 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1050 | unsigned long flags, const char *name, | |
51cc5068 | 1051 | void (*ctor)(void *)) |
ba0268a8 CL |
1052 | { |
1053 | return flags; | |
1054 | } | |
41ecc55b | 1055 | #define slub_debug 0 |
0f389ec6 CL |
1056 | |
1057 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1058 | { return 0; } | |
26c02cf0 AB |
1059 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1060 | { return 0; } | |
205ab99d CL |
1061 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1062 | int objects) {} | |
1063 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1064 | int objects) {} | |
41ecc55b | 1065 | #endif |
205ab99d | 1066 | |
81819f0f CL |
1067 | /* |
1068 | * Slab allocation and freeing | |
1069 | */ | |
65c3376a CL |
1070 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1071 | struct kmem_cache_order_objects oo) | |
1072 | { | |
1073 | int order = oo_order(oo); | |
1074 | ||
b1eeab67 VN |
1075 | flags |= __GFP_NOTRACK; |
1076 | ||
65c3376a CL |
1077 | if (node == -1) |
1078 | return alloc_pages(flags, order); | |
1079 | else | |
1080 | return alloc_pages_node(node, flags, order); | |
1081 | } | |
1082 | ||
81819f0f CL |
1083 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1084 | { | |
06428780 | 1085 | struct page *page; |
834f3d11 | 1086 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1087 | gfp_t alloc_gfp; |
81819f0f | 1088 | |
b7a49f0d | 1089 | flags |= s->allocflags; |
e12ba74d | 1090 | |
ba52270d PE |
1091 | /* |
1092 | * Let the initial higher-order allocation fail under memory pressure | |
1093 | * so we fall-back to the minimum order allocation. | |
1094 | */ | |
1095 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1096 | ||
1097 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1098 | if (unlikely(!page)) { |
1099 | oo = s->min; | |
1100 | /* | |
1101 | * Allocation may have failed due to fragmentation. | |
1102 | * Try a lower order alloc if possible | |
1103 | */ | |
1104 | page = alloc_slab_page(flags, node, oo); | |
1105 | if (!page) | |
1106 | return NULL; | |
81819f0f | 1107 | |
65c3376a CL |
1108 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1109 | } | |
5a896d9e VN |
1110 | |
1111 | if (kmemcheck_enabled | |
5086c389 | 1112 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1113 | int pages = 1 << oo_order(oo); |
1114 | ||
1115 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1116 | ||
1117 | /* | |
1118 | * Objects from caches that have a constructor don't get | |
1119 | * cleared when they're allocated, so we need to do it here. | |
1120 | */ | |
1121 | if (s->ctor) | |
1122 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1123 | else | |
1124 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1125 | } |
1126 | ||
834f3d11 | 1127 | page->objects = oo_objects(oo); |
81819f0f CL |
1128 | mod_zone_page_state(page_zone(page), |
1129 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1130 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1131 | 1 << oo_order(oo)); |
81819f0f CL |
1132 | |
1133 | return page; | |
1134 | } | |
1135 | ||
1136 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1137 | void *object) | |
1138 | { | |
3ec09742 | 1139 | setup_object_debug(s, page, object); |
4f104934 | 1140 | if (unlikely(s->ctor)) |
51cc5068 | 1141 | s->ctor(object); |
81819f0f CL |
1142 | } |
1143 | ||
1144 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1145 | { | |
1146 | struct page *page; | |
81819f0f | 1147 | void *start; |
81819f0f CL |
1148 | void *last; |
1149 | void *p; | |
1150 | ||
6cb06229 | 1151 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1152 | |
6cb06229 CL |
1153 | page = allocate_slab(s, |
1154 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1155 | if (!page) |
1156 | goto out; | |
1157 | ||
205ab99d | 1158 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1159 | page->slab = s; |
1160 | page->flags |= 1 << PG_slab; | |
1161 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1162 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1163 | __SetPageSlubDebug(page); |
81819f0f CL |
1164 | |
1165 | start = page_address(page); | |
81819f0f CL |
1166 | |
1167 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1168 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1169 | |
1170 | last = start; | |
224a88be | 1171 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1172 | setup_object(s, page, last); |
1173 | set_freepointer(s, last, p); | |
1174 | last = p; | |
1175 | } | |
1176 | setup_object(s, page, last); | |
a973e9dd | 1177 | set_freepointer(s, last, NULL); |
81819f0f CL |
1178 | |
1179 | page->freelist = start; | |
1180 | page->inuse = 0; | |
1181 | out: | |
81819f0f CL |
1182 | return page; |
1183 | } | |
1184 | ||
1185 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1186 | { | |
834f3d11 CL |
1187 | int order = compound_order(page); |
1188 | int pages = 1 << order; | |
81819f0f | 1189 | |
8a38082d | 1190 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1191 | void *p; |
1192 | ||
1193 | slab_pad_check(s, page); | |
224a88be CL |
1194 | for_each_object(p, s, page_address(page), |
1195 | page->objects) | |
81819f0f | 1196 | check_object(s, page, p, 0); |
8a38082d | 1197 | __ClearPageSlubDebug(page); |
81819f0f CL |
1198 | } |
1199 | ||
b1eeab67 | 1200 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1201 | |
81819f0f CL |
1202 | mod_zone_page_state(page_zone(page), |
1203 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1204 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1205 | -pages); |
81819f0f | 1206 | |
49bd5221 CL |
1207 | __ClearPageSlab(page); |
1208 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1209 | if (current->reclaim_state) |
1210 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1211 | __free_pages(page, order); |
81819f0f CL |
1212 | } |
1213 | ||
1214 | static void rcu_free_slab(struct rcu_head *h) | |
1215 | { | |
1216 | struct page *page; | |
1217 | ||
1218 | page = container_of((struct list_head *)h, struct page, lru); | |
1219 | __free_slab(page->slab, page); | |
1220 | } | |
1221 | ||
1222 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1223 | { | |
1224 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1225 | /* | |
1226 | * RCU free overloads the RCU head over the LRU | |
1227 | */ | |
1228 | struct rcu_head *head = (void *)&page->lru; | |
1229 | ||
1230 | call_rcu(head, rcu_free_slab); | |
1231 | } else | |
1232 | __free_slab(s, page); | |
1233 | } | |
1234 | ||
1235 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1236 | { | |
205ab99d | 1237 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1238 | free_slab(s, page); |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Per slab locking using the pagelock | |
1243 | */ | |
1244 | static __always_inline void slab_lock(struct page *page) | |
1245 | { | |
1246 | bit_spin_lock(PG_locked, &page->flags); | |
1247 | } | |
1248 | ||
1249 | static __always_inline void slab_unlock(struct page *page) | |
1250 | { | |
a76d3546 | 1251 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1252 | } |
1253 | ||
1254 | static __always_inline int slab_trylock(struct page *page) | |
1255 | { | |
1256 | int rc = 1; | |
1257 | ||
1258 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1259 | return rc; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * Management of partially allocated slabs | |
1264 | */ | |
7c2e132c CL |
1265 | static void add_partial(struct kmem_cache_node *n, |
1266 | struct page *page, int tail) | |
81819f0f | 1267 | { |
e95eed57 CL |
1268 | spin_lock(&n->list_lock); |
1269 | n->nr_partial++; | |
7c2e132c CL |
1270 | if (tail) |
1271 | list_add_tail(&page->lru, &n->partial); | |
1272 | else | |
1273 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1274 | spin_unlock(&n->list_lock); |
1275 | } | |
1276 | ||
0121c619 | 1277 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1278 | { |
1279 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1280 | ||
1281 | spin_lock(&n->list_lock); | |
1282 | list_del(&page->lru); | |
1283 | n->nr_partial--; | |
1284 | spin_unlock(&n->list_lock); | |
1285 | } | |
1286 | ||
1287 | /* | |
672bba3a | 1288 | * Lock slab and remove from the partial list. |
81819f0f | 1289 | * |
672bba3a | 1290 | * Must hold list_lock. |
81819f0f | 1291 | */ |
0121c619 CL |
1292 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1293 | struct page *page) | |
81819f0f CL |
1294 | { |
1295 | if (slab_trylock(page)) { | |
1296 | list_del(&page->lru); | |
1297 | n->nr_partial--; | |
8a38082d | 1298 | __SetPageSlubFrozen(page); |
81819f0f CL |
1299 | return 1; |
1300 | } | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | /* | |
672bba3a | 1305 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1306 | */ |
1307 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1308 | { | |
1309 | struct page *page; | |
1310 | ||
1311 | /* | |
1312 | * Racy check. If we mistakenly see no partial slabs then we | |
1313 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1314 | * partial slab and there is none available then get_partials() |
1315 | * will return NULL. | |
81819f0f CL |
1316 | */ |
1317 | if (!n || !n->nr_partial) | |
1318 | return NULL; | |
1319 | ||
1320 | spin_lock(&n->list_lock); | |
1321 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1322 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1323 | goto out; |
1324 | page = NULL; | |
1325 | out: | |
1326 | spin_unlock(&n->list_lock); | |
1327 | return page; | |
1328 | } | |
1329 | ||
1330 | /* | |
672bba3a | 1331 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1332 | */ |
1333 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1334 | { | |
1335 | #ifdef CONFIG_NUMA | |
1336 | struct zonelist *zonelist; | |
dd1a239f | 1337 | struct zoneref *z; |
54a6eb5c MG |
1338 | struct zone *zone; |
1339 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1340 | struct page *page; |
1341 | ||
1342 | /* | |
672bba3a CL |
1343 | * The defrag ratio allows a configuration of the tradeoffs between |
1344 | * inter node defragmentation and node local allocations. A lower | |
1345 | * defrag_ratio increases the tendency to do local allocations | |
1346 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1347 | * |
672bba3a CL |
1348 | * If the defrag_ratio is set to 0 then kmalloc() always |
1349 | * returns node local objects. If the ratio is higher then kmalloc() | |
1350 | * may return off node objects because partial slabs are obtained | |
1351 | * from other nodes and filled up. | |
81819f0f | 1352 | * |
6446faa2 | 1353 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1354 | * defrag_ratio = 1000) then every (well almost) allocation will |
1355 | * first attempt to defrag slab caches on other nodes. This means | |
1356 | * scanning over all nodes to look for partial slabs which may be | |
1357 | * expensive if we do it every time we are trying to find a slab | |
1358 | * with available objects. | |
81819f0f | 1359 | */ |
9824601e CL |
1360 | if (!s->remote_node_defrag_ratio || |
1361 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1362 | return NULL; |
1363 | ||
0e88460d | 1364 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1365 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1366 | struct kmem_cache_node *n; |
1367 | ||
54a6eb5c | 1368 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1369 | |
54a6eb5c | 1370 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1371 | n->nr_partial > s->min_partial) { |
81819f0f CL |
1372 | page = get_partial_node(n); |
1373 | if (page) | |
1374 | return page; | |
1375 | } | |
1376 | } | |
1377 | #endif | |
1378 | return NULL; | |
1379 | } | |
1380 | ||
1381 | /* | |
1382 | * Get a partial page, lock it and return it. | |
1383 | */ | |
1384 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1385 | { | |
1386 | struct page *page; | |
1387 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1388 | ||
1389 | page = get_partial_node(get_node(s, searchnode)); | |
1390 | if (page || (flags & __GFP_THISNODE)) | |
1391 | return page; | |
1392 | ||
1393 | return get_any_partial(s, flags); | |
1394 | } | |
1395 | ||
1396 | /* | |
1397 | * Move a page back to the lists. | |
1398 | * | |
1399 | * Must be called with the slab lock held. | |
1400 | * | |
1401 | * On exit the slab lock will have been dropped. | |
1402 | */ | |
7c2e132c | 1403 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1404 | { |
e95eed57 | 1405 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1406 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1407 | |
8a38082d | 1408 | __ClearPageSlubFrozen(page); |
81819f0f | 1409 | if (page->inuse) { |
e95eed57 | 1410 | |
a973e9dd | 1411 | if (page->freelist) { |
7c2e132c | 1412 | add_partial(n, page, tail); |
8ff12cfc CL |
1413 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1414 | } else { | |
1415 | stat(c, DEACTIVATE_FULL); | |
8a38082d AW |
1416 | if (SLABDEBUG && PageSlubDebug(page) && |
1417 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1418 | add_full(n, page); |
1419 | } | |
81819f0f CL |
1420 | slab_unlock(page); |
1421 | } else { | |
8ff12cfc | 1422 | stat(c, DEACTIVATE_EMPTY); |
3b89d7d8 | 1423 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1424 | /* |
672bba3a CL |
1425 | * Adding an empty slab to the partial slabs in order |
1426 | * to avoid page allocator overhead. This slab needs | |
1427 | * to come after the other slabs with objects in | |
6446faa2 CL |
1428 | * so that the others get filled first. That way the |
1429 | * size of the partial list stays small. | |
1430 | * | |
0121c619 CL |
1431 | * kmem_cache_shrink can reclaim any empty slabs from |
1432 | * the partial list. | |
e95eed57 | 1433 | */ |
7c2e132c | 1434 | add_partial(n, page, 1); |
e95eed57 CL |
1435 | slab_unlock(page); |
1436 | } else { | |
1437 | slab_unlock(page); | |
8ff12cfc | 1438 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1439 | discard_slab(s, page); |
1440 | } | |
81819f0f CL |
1441 | } |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * Remove the cpu slab | |
1446 | */ | |
dfb4f096 | 1447 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1448 | { |
dfb4f096 | 1449 | struct page *page = c->page; |
7c2e132c | 1450 | int tail = 1; |
8ff12cfc | 1451 | |
b773ad73 | 1452 | if (page->freelist) |
8ff12cfc | 1453 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1454 | /* |
6446faa2 | 1455 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1456 | * because both freelists are empty. So this is unlikely |
1457 | * to occur. | |
1458 | */ | |
a973e9dd | 1459 | while (unlikely(c->freelist)) { |
894b8788 CL |
1460 | void **object; |
1461 | ||
7c2e132c CL |
1462 | tail = 0; /* Hot objects. Put the slab first */ |
1463 | ||
894b8788 | 1464 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1465 | object = c->freelist; |
b3fba8da | 1466 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1467 | |
1468 | /* And put onto the regular freelist */ | |
b3fba8da | 1469 | object[c->offset] = page->freelist; |
894b8788 CL |
1470 | page->freelist = object; |
1471 | page->inuse--; | |
1472 | } | |
dfb4f096 | 1473 | c->page = NULL; |
7c2e132c | 1474 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1475 | } |
1476 | ||
dfb4f096 | 1477 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1478 | { |
8ff12cfc | 1479 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1480 | slab_lock(c->page); |
1481 | deactivate_slab(s, c); | |
81819f0f CL |
1482 | } |
1483 | ||
1484 | /* | |
1485 | * Flush cpu slab. | |
6446faa2 | 1486 | * |
81819f0f CL |
1487 | * Called from IPI handler with interrupts disabled. |
1488 | */ | |
0c710013 | 1489 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1490 | { |
dfb4f096 | 1491 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1492 | |
dfb4f096 CL |
1493 | if (likely(c && c->page)) |
1494 | flush_slab(s, c); | |
81819f0f CL |
1495 | } |
1496 | ||
1497 | static void flush_cpu_slab(void *d) | |
1498 | { | |
1499 | struct kmem_cache *s = d; | |
81819f0f | 1500 | |
dfb4f096 | 1501 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1502 | } |
1503 | ||
1504 | static void flush_all(struct kmem_cache *s) | |
1505 | { | |
15c8b6c1 | 1506 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1507 | } |
1508 | ||
dfb4f096 CL |
1509 | /* |
1510 | * Check if the objects in a per cpu structure fit numa | |
1511 | * locality expectations. | |
1512 | */ | |
1513 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1514 | { | |
1515 | #ifdef CONFIG_NUMA | |
1516 | if (node != -1 && c->node != node) | |
1517 | return 0; | |
1518 | #endif | |
1519 | return 1; | |
1520 | } | |
1521 | ||
781b2ba6 PE |
1522 | static int count_free(struct page *page) |
1523 | { | |
1524 | return page->objects - page->inuse; | |
1525 | } | |
1526 | ||
1527 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1528 | int (*get_count)(struct page *)) | |
1529 | { | |
1530 | unsigned long flags; | |
1531 | unsigned long x = 0; | |
1532 | struct page *page; | |
1533 | ||
1534 | spin_lock_irqsave(&n->list_lock, flags); | |
1535 | list_for_each_entry(page, &n->partial, lru) | |
1536 | x += get_count(page); | |
1537 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1538 | return x; | |
1539 | } | |
1540 | ||
26c02cf0 AB |
1541 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1542 | { | |
1543 | #ifdef CONFIG_SLUB_DEBUG | |
1544 | return atomic_long_read(&n->total_objects); | |
1545 | #else | |
1546 | return 0; | |
1547 | #endif | |
1548 | } | |
1549 | ||
781b2ba6 PE |
1550 | static noinline void |
1551 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1552 | { | |
1553 | int node; | |
1554 | ||
1555 | printk(KERN_WARNING | |
1556 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1557 | nid, gfpflags); | |
1558 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1559 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1560 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1561 | ||
1562 | for_each_online_node(node) { | |
1563 | struct kmem_cache_node *n = get_node(s, node); | |
1564 | unsigned long nr_slabs; | |
1565 | unsigned long nr_objs; | |
1566 | unsigned long nr_free; | |
1567 | ||
1568 | if (!n) | |
1569 | continue; | |
1570 | ||
26c02cf0 AB |
1571 | nr_free = count_partial(n, count_free); |
1572 | nr_slabs = node_nr_slabs(n); | |
1573 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1574 | |
1575 | printk(KERN_WARNING | |
1576 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1577 | node, nr_slabs, nr_objs, nr_free); | |
1578 | } | |
1579 | } | |
1580 | ||
81819f0f | 1581 | /* |
894b8788 CL |
1582 | * Slow path. The lockless freelist is empty or we need to perform |
1583 | * debugging duties. | |
1584 | * | |
1585 | * Interrupts are disabled. | |
81819f0f | 1586 | * |
894b8788 CL |
1587 | * Processing is still very fast if new objects have been freed to the |
1588 | * regular freelist. In that case we simply take over the regular freelist | |
1589 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1590 | * |
894b8788 CL |
1591 | * If that is not working then we fall back to the partial lists. We take the |
1592 | * first element of the freelist as the object to allocate now and move the | |
1593 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1594 | * |
894b8788 | 1595 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1596 | * we need to allocate a new slab. This is the slowest path since it involves |
1597 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1598 | */ |
ce71e27c EGM |
1599 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1600 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1601 | { |
81819f0f | 1602 | void **object; |
dfb4f096 | 1603 | struct page *new; |
81819f0f | 1604 | |
e72e9c23 LT |
1605 | /* We handle __GFP_ZERO in the caller */ |
1606 | gfpflags &= ~__GFP_ZERO; | |
1607 | ||
dfb4f096 | 1608 | if (!c->page) |
81819f0f CL |
1609 | goto new_slab; |
1610 | ||
dfb4f096 CL |
1611 | slab_lock(c->page); |
1612 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1613 | goto another_slab; |
6446faa2 | 1614 | |
8ff12cfc | 1615 | stat(c, ALLOC_REFILL); |
6446faa2 | 1616 | |
894b8788 | 1617 | load_freelist: |
dfb4f096 | 1618 | object = c->page->freelist; |
a973e9dd | 1619 | if (unlikely(!object)) |
81819f0f | 1620 | goto another_slab; |
8a38082d | 1621 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1622 | goto debug; |
1623 | ||
b3fba8da | 1624 | c->freelist = object[c->offset]; |
39b26464 | 1625 | c->page->inuse = c->page->objects; |
a973e9dd | 1626 | c->page->freelist = NULL; |
dfb4f096 | 1627 | c->node = page_to_nid(c->page); |
1f84260c | 1628 | unlock_out: |
dfb4f096 | 1629 | slab_unlock(c->page); |
8ff12cfc | 1630 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1631 | return object; |
1632 | ||
1633 | another_slab: | |
dfb4f096 | 1634 | deactivate_slab(s, c); |
81819f0f CL |
1635 | |
1636 | new_slab: | |
dfb4f096 CL |
1637 | new = get_partial(s, gfpflags, node); |
1638 | if (new) { | |
1639 | c->page = new; | |
8ff12cfc | 1640 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1641 | goto load_freelist; |
81819f0f CL |
1642 | } |
1643 | ||
b811c202 CL |
1644 | if (gfpflags & __GFP_WAIT) |
1645 | local_irq_enable(); | |
1646 | ||
dfb4f096 | 1647 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1648 | |
1649 | if (gfpflags & __GFP_WAIT) | |
1650 | local_irq_disable(); | |
1651 | ||
dfb4f096 CL |
1652 | if (new) { |
1653 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1654 | stat(c, ALLOC_SLAB); |
05aa3450 | 1655 | if (c->page) |
dfb4f096 | 1656 | flush_slab(s, c); |
dfb4f096 | 1657 | slab_lock(new); |
8a38082d | 1658 | __SetPageSlubFrozen(new); |
dfb4f096 | 1659 | c->page = new; |
4b6f0750 | 1660 | goto load_freelist; |
81819f0f | 1661 | } |
95f85989 PE |
1662 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1663 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1664 | return NULL; |
81819f0f | 1665 | debug: |
dfb4f096 | 1666 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1667 | goto another_slab; |
894b8788 | 1668 | |
dfb4f096 | 1669 | c->page->inuse++; |
b3fba8da | 1670 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1671 | c->node = -1; |
1f84260c | 1672 | goto unlock_out; |
894b8788 CL |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1677 | * have the fastpath folded into their functions. So no function call | |
1678 | * overhead for requests that can be satisfied on the fastpath. | |
1679 | * | |
1680 | * The fastpath works by first checking if the lockless freelist can be used. | |
1681 | * If not then __slab_alloc is called for slow processing. | |
1682 | * | |
1683 | * Otherwise we can simply pick the next object from the lockless free list. | |
1684 | */ | |
06428780 | 1685 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1686 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1687 | { |
894b8788 | 1688 | void **object; |
dfb4f096 | 1689 | struct kmem_cache_cpu *c; |
1f84260c | 1690 | unsigned long flags; |
bdb21928 | 1691 | unsigned int objsize; |
1f84260c | 1692 | |
dcce284a | 1693 | gfpflags &= gfp_allowed_mask; |
7e85ee0c | 1694 | |
cf40bd16 | 1695 | lockdep_trace_alloc(gfpflags); |
89124d70 | 1696 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1697 | |
773ff60e AM |
1698 | if (should_failslab(s->objsize, gfpflags)) |
1699 | return NULL; | |
1f84260c | 1700 | |
894b8788 | 1701 | local_irq_save(flags); |
dfb4f096 | 1702 | c = get_cpu_slab(s, smp_processor_id()); |
bdb21928 | 1703 | objsize = c->objsize; |
a973e9dd | 1704 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1705 | |
dfb4f096 | 1706 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1707 | |
1708 | else { | |
dfb4f096 | 1709 | object = c->freelist; |
b3fba8da | 1710 | c->freelist = object[c->offset]; |
8ff12cfc | 1711 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1712 | } |
1713 | local_irq_restore(flags); | |
d07dbea4 CL |
1714 | |
1715 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
bdb21928 | 1716 | memset(object, 0, objsize); |
d07dbea4 | 1717 | |
5a896d9e | 1718 | kmemcheck_slab_alloc(s, gfpflags, object, c->objsize); |
06f22f13 | 1719 | kmemleak_alloc_recursive(object, objsize, 1, s->flags, gfpflags); |
5a896d9e | 1720 | |
894b8788 | 1721 | return object; |
81819f0f CL |
1722 | } |
1723 | ||
1724 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1725 | { | |
5b882be4 EGM |
1726 | void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_); |
1727 | ||
ca2b84cb | 1728 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1729 | |
1730 | return ret; | |
81819f0f CL |
1731 | } |
1732 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1733 | ||
5b882be4 EGM |
1734 | #ifdef CONFIG_KMEMTRACE |
1735 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) | |
1736 | { | |
1737 | return slab_alloc(s, gfpflags, -1, _RET_IP_); | |
1738 | } | |
1739 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1740 | #endif | |
1741 | ||
81819f0f CL |
1742 | #ifdef CONFIG_NUMA |
1743 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1744 | { | |
5b882be4 EGM |
1745 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1746 | ||
ca2b84cb EGM |
1747 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1748 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1749 | |
1750 | return ret; | |
81819f0f CL |
1751 | } |
1752 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1753 | #endif | |
1754 | ||
5b882be4 EGM |
1755 | #ifdef CONFIG_KMEMTRACE |
1756 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, | |
1757 | gfp_t gfpflags, | |
1758 | int node) | |
1759 | { | |
1760 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1761 | } | |
1762 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1763 | #endif | |
1764 | ||
81819f0f | 1765 | /* |
894b8788 CL |
1766 | * Slow patch handling. This may still be called frequently since objects |
1767 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1768 | * |
894b8788 CL |
1769 | * So we still attempt to reduce cache line usage. Just take the slab |
1770 | * lock and free the item. If there is no additional partial page | |
1771 | * handling required then we can return immediately. | |
81819f0f | 1772 | */ |
894b8788 | 1773 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ce71e27c | 1774 | void *x, unsigned long addr, unsigned int offset) |
81819f0f CL |
1775 | { |
1776 | void *prior; | |
1777 | void **object = (void *)x; | |
8ff12cfc | 1778 | struct kmem_cache_cpu *c; |
81819f0f | 1779 | |
8ff12cfc CL |
1780 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1781 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1782 | slab_lock(page); |
1783 | ||
8a38082d | 1784 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1785 | goto debug; |
6446faa2 | 1786 | |
81819f0f | 1787 | checks_ok: |
b3fba8da | 1788 | prior = object[offset] = page->freelist; |
81819f0f CL |
1789 | page->freelist = object; |
1790 | page->inuse--; | |
1791 | ||
8a38082d | 1792 | if (unlikely(PageSlubFrozen(page))) { |
8ff12cfc | 1793 | stat(c, FREE_FROZEN); |
81819f0f | 1794 | goto out_unlock; |
8ff12cfc | 1795 | } |
81819f0f CL |
1796 | |
1797 | if (unlikely(!page->inuse)) | |
1798 | goto slab_empty; | |
1799 | ||
1800 | /* | |
6446faa2 | 1801 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1802 | * then add it. |
1803 | */ | |
a973e9dd | 1804 | if (unlikely(!prior)) { |
7c2e132c | 1805 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1806 | stat(c, FREE_ADD_PARTIAL); |
1807 | } | |
81819f0f CL |
1808 | |
1809 | out_unlock: | |
1810 | slab_unlock(page); | |
81819f0f CL |
1811 | return; |
1812 | ||
1813 | slab_empty: | |
a973e9dd | 1814 | if (prior) { |
81819f0f | 1815 | /* |
672bba3a | 1816 | * Slab still on the partial list. |
81819f0f CL |
1817 | */ |
1818 | remove_partial(s, page); | |
8ff12cfc CL |
1819 | stat(c, FREE_REMOVE_PARTIAL); |
1820 | } | |
81819f0f | 1821 | slab_unlock(page); |
8ff12cfc | 1822 | stat(c, FREE_SLAB); |
81819f0f | 1823 | discard_slab(s, page); |
81819f0f CL |
1824 | return; |
1825 | ||
1826 | debug: | |
3ec09742 | 1827 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1828 | goto out_unlock; |
77c5e2d0 | 1829 | goto checks_ok; |
81819f0f CL |
1830 | } |
1831 | ||
894b8788 CL |
1832 | /* |
1833 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1834 | * can perform fastpath freeing without additional function calls. | |
1835 | * | |
1836 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1837 | * of this processor. This typically the case if we have just allocated | |
1838 | * the item before. | |
1839 | * | |
1840 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1841 | * with all sorts of special processing. | |
1842 | */ | |
06428780 | 1843 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1844 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1845 | { |
1846 | void **object = (void *)x; | |
dfb4f096 | 1847 | struct kmem_cache_cpu *c; |
1f84260c CL |
1848 | unsigned long flags; |
1849 | ||
06f22f13 | 1850 | kmemleak_free_recursive(x, s->flags); |
894b8788 | 1851 | local_irq_save(flags); |
dfb4f096 | 1852 | c = get_cpu_slab(s, smp_processor_id()); |
5a896d9e | 1853 | kmemcheck_slab_free(s, object, c->objsize); |
27d9e4e9 | 1854 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a | 1855 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
6047a007 | 1856 | debug_check_no_obj_freed(object, c->objsize); |
ee3c72a1 | 1857 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1858 | object[c->offset] = c->freelist; |
dfb4f096 | 1859 | c->freelist = object; |
8ff12cfc | 1860 | stat(c, FREE_FASTPATH); |
894b8788 | 1861 | } else |
b3fba8da | 1862 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1863 | |
1864 | local_irq_restore(flags); | |
1865 | } | |
1866 | ||
81819f0f CL |
1867 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1868 | { | |
77c5e2d0 | 1869 | struct page *page; |
81819f0f | 1870 | |
b49af68f | 1871 | page = virt_to_head_page(x); |
81819f0f | 1872 | |
ce71e27c | 1873 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1874 | |
ca2b84cb | 1875 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1876 | } |
1877 | EXPORT_SYMBOL(kmem_cache_free); | |
1878 | ||
e9beef18 | 1879 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1880 | static struct page *get_object_page(const void *x) |
1881 | { | |
b49af68f | 1882 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1883 | |
1884 | if (!PageSlab(page)) | |
1885 | return NULL; | |
1886 | ||
1887 | return page; | |
1888 | } | |
1889 | ||
1890 | /* | |
672bba3a CL |
1891 | * Object placement in a slab is made very easy because we always start at |
1892 | * offset 0. If we tune the size of the object to the alignment then we can | |
1893 | * get the required alignment by putting one properly sized object after | |
1894 | * another. | |
81819f0f CL |
1895 | * |
1896 | * Notice that the allocation order determines the sizes of the per cpu | |
1897 | * caches. Each processor has always one slab available for allocations. | |
1898 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1899 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1900 | * locking overhead. |
81819f0f CL |
1901 | */ |
1902 | ||
1903 | /* | |
1904 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1905 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1906 | * and increases the number of allocations possible without having to | |
1907 | * take the list_lock. | |
1908 | */ | |
1909 | static int slub_min_order; | |
114e9e89 | 1910 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1911 | static int slub_min_objects; |
81819f0f CL |
1912 | |
1913 | /* | |
1914 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1915 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1916 | */ |
1917 | static int slub_nomerge; | |
1918 | ||
81819f0f CL |
1919 | /* |
1920 | * Calculate the order of allocation given an slab object size. | |
1921 | * | |
672bba3a CL |
1922 | * The order of allocation has significant impact on performance and other |
1923 | * system components. Generally order 0 allocations should be preferred since | |
1924 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1925 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1926 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1927 | * would be wasted. |
1928 | * | |
1929 | * In order to reach satisfactory performance we must ensure that a minimum | |
1930 | * number of objects is in one slab. Otherwise we may generate too much | |
1931 | * activity on the partial lists which requires taking the list_lock. This is | |
1932 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1933 | * |
672bba3a CL |
1934 | * slub_max_order specifies the order where we begin to stop considering the |
1935 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1936 | * we try to keep the page order as low as possible. So we accept more waste | |
1937 | * of space in favor of a small page order. | |
81819f0f | 1938 | * |
672bba3a CL |
1939 | * Higher order allocations also allow the placement of more objects in a |
1940 | * slab and thereby reduce object handling overhead. If the user has | |
1941 | * requested a higher mininum order then we start with that one instead of | |
1942 | * the smallest order which will fit the object. | |
81819f0f | 1943 | */ |
5e6d444e CL |
1944 | static inline int slab_order(int size, int min_objects, |
1945 | int max_order, int fract_leftover) | |
81819f0f CL |
1946 | { |
1947 | int order; | |
1948 | int rem; | |
6300ea75 | 1949 | int min_order = slub_min_order; |
81819f0f | 1950 | |
210b5c06 CG |
1951 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1952 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1953 | |
6300ea75 | 1954 | for (order = max(min_order, |
5e6d444e CL |
1955 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1956 | order <= max_order; order++) { | |
81819f0f | 1957 | |
5e6d444e | 1958 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1959 | |
5e6d444e | 1960 | if (slab_size < min_objects * size) |
81819f0f CL |
1961 | continue; |
1962 | ||
1963 | rem = slab_size % size; | |
1964 | ||
5e6d444e | 1965 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1966 | break; |
1967 | ||
1968 | } | |
672bba3a | 1969 | |
81819f0f CL |
1970 | return order; |
1971 | } | |
1972 | ||
5e6d444e CL |
1973 | static inline int calculate_order(int size) |
1974 | { | |
1975 | int order; | |
1976 | int min_objects; | |
1977 | int fraction; | |
e8120ff1 | 1978 | int max_objects; |
5e6d444e CL |
1979 | |
1980 | /* | |
1981 | * Attempt to find best configuration for a slab. This | |
1982 | * works by first attempting to generate a layout with | |
1983 | * the best configuration and backing off gradually. | |
1984 | * | |
1985 | * First we reduce the acceptable waste in a slab. Then | |
1986 | * we reduce the minimum objects required in a slab. | |
1987 | */ | |
1988 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1989 | if (!min_objects) |
1990 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
1991 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
1992 | min_objects = min(min_objects, max_objects); | |
1993 | ||
5e6d444e | 1994 | while (min_objects > 1) { |
c124f5b5 | 1995 | fraction = 16; |
5e6d444e CL |
1996 | while (fraction >= 4) { |
1997 | order = slab_order(size, min_objects, | |
1998 | slub_max_order, fraction); | |
1999 | if (order <= slub_max_order) | |
2000 | return order; | |
2001 | fraction /= 2; | |
2002 | } | |
5086c389 | 2003 | min_objects--; |
5e6d444e CL |
2004 | } |
2005 | ||
2006 | /* | |
2007 | * We were unable to place multiple objects in a slab. Now | |
2008 | * lets see if we can place a single object there. | |
2009 | */ | |
2010 | order = slab_order(size, 1, slub_max_order, 1); | |
2011 | if (order <= slub_max_order) | |
2012 | return order; | |
2013 | ||
2014 | /* | |
2015 | * Doh this slab cannot be placed using slub_max_order. | |
2016 | */ | |
2017 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2018 | if (order < MAX_ORDER) |
5e6d444e CL |
2019 | return order; |
2020 | return -ENOSYS; | |
2021 | } | |
2022 | ||
81819f0f | 2023 | /* |
672bba3a | 2024 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2025 | */ |
2026 | static unsigned long calculate_alignment(unsigned long flags, | |
2027 | unsigned long align, unsigned long size) | |
2028 | { | |
2029 | /* | |
6446faa2 CL |
2030 | * If the user wants hardware cache aligned objects then follow that |
2031 | * suggestion if the object is sufficiently large. | |
81819f0f | 2032 | * |
6446faa2 CL |
2033 | * The hardware cache alignment cannot override the specified |
2034 | * alignment though. If that is greater then use it. | |
81819f0f | 2035 | */ |
b6210386 NP |
2036 | if (flags & SLAB_HWCACHE_ALIGN) { |
2037 | unsigned long ralign = cache_line_size(); | |
2038 | while (size <= ralign / 2) | |
2039 | ralign /= 2; | |
2040 | align = max(align, ralign); | |
2041 | } | |
81819f0f CL |
2042 | |
2043 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2044 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2045 | |
2046 | return ALIGN(align, sizeof(void *)); | |
2047 | } | |
2048 | ||
dfb4f096 CL |
2049 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
2050 | struct kmem_cache_cpu *c) | |
2051 | { | |
2052 | c->page = NULL; | |
a973e9dd | 2053 | c->freelist = NULL; |
dfb4f096 | 2054 | c->node = 0; |
42a9fdbb CL |
2055 | c->offset = s->offset / sizeof(void *); |
2056 | c->objsize = s->objsize; | |
62f75532 PE |
2057 | #ifdef CONFIG_SLUB_STATS |
2058 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
2059 | #endif | |
dfb4f096 CL |
2060 | } |
2061 | ||
5595cffc PE |
2062 | static void |
2063 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2064 | { |
2065 | n->nr_partial = 0; | |
81819f0f CL |
2066 | spin_lock_init(&n->list_lock); |
2067 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2068 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2069 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2070 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2071 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2072 | #endif |
81819f0f CL |
2073 | } |
2074 | ||
4c93c355 CL |
2075 | #ifdef CONFIG_SMP |
2076 | /* | |
2077 | * Per cpu array for per cpu structures. | |
2078 | * | |
2079 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
2080 | * close together meaning that it becomes possible that multiple per cpu | |
2081 | * structures are contained in one cacheline. This may be particularly | |
2082 | * beneficial for the kmalloc caches. | |
2083 | * | |
2084 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
2085 | * likely able to get per cpu structures for all caches from the array defined | |
2086 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
2087 | * | |
2088 | * If the per cpu array is exhausted then fall back to kmalloc | |
2089 | * of individual cachelines. No sharing is possible then. | |
2090 | */ | |
2091 | #define NR_KMEM_CACHE_CPU 100 | |
2092 | ||
2093 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
2094 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
2095 | ||
2096 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
174596a0 | 2097 | static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS); |
4c93c355 CL |
2098 | |
2099 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
2100 | int cpu, gfp_t flags) | |
2101 | { | |
2102 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2103 | ||
2104 | if (c) | |
2105 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2106 | (void *)c->freelist; | |
2107 | else { | |
2108 | /* Table overflow: So allocate ourselves */ | |
2109 | c = kmalloc_node( | |
2110 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2111 | flags, cpu_to_node(cpu)); | |
2112 | if (!c) | |
2113 | return NULL; | |
2114 | } | |
2115 | ||
2116 | init_kmem_cache_cpu(s, c); | |
2117 | return c; | |
2118 | } | |
2119 | ||
2120 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2121 | { | |
2122 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
37189094 | 2123 | c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { |
4c93c355 CL |
2124 | kfree(c); |
2125 | return; | |
2126 | } | |
2127 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2128 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2129 | } | |
2130 | ||
2131 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2132 | { | |
2133 | int cpu; | |
2134 | ||
2135 | for_each_online_cpu(cpu) { | |
2136 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2137 | ||
2138 | if (c) { | |
2139 | s->cpu_slab[cpu] = NULL; | |
2140 | free_kmem_cache_cpu(c, cpu); | |
2141 | } | |
2142 | } | |
2143 | } | |
2144 | ||
2145 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2146 | { | |
2147 | int cpu; | |
2148 | ||
2149 | for_each_online_cpu(cpu) { | |
2150 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2151 | ||
2152 | if (c) | |
2153 | continue; | |
2154 | ||
2155 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2156 | if (!c) { | |
2157 | free_kmem_cache_cpus(s); | |
2158 | return 0; | |
2159 | } | |
2160 | s->cpu_slab[cpu] = c; | |
2161 | } | |
2162 | return 1; | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * Initialize the per cpu array. | |
2167 | */ | |
2168 | static void init_alloc_cpu_cpu(int cpu) | |
2169 | { | |
2170 | int i; | |
2171 | ||
174596a0 | 2172 | if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once))) |
4c93c355 CL |
2173 | return; |
2174 | ||
2175 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2176 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2177 | ||
174596a0 | 2178 | cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)); |
4c93c355 CL |
2179 | } |
2180 | ||
2181 | static void __init init_alloc_cpu(void) | |
2182 | { | |
2183 | int cpu; | |
2184 | ||
2185 | for_each_online_cpu(cpu) | |
2186 | init_alloc_cpu_cpu(cpu); | |
2187 | } | |
2188 | ||
2189 | #else | |
2190 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2191 | static inline void init_alloc_cpu(void) {} | |
2192 | ||
2193 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2194 | { | |
2195 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2196 | return 1; | |
2197 | } | |
2198 | #endif | |
2199 | ||
81819f0f CL |
2200 | #ifdef CONFIG_NUMA |
2201 | /* | |
2202 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2203 | * slab on the node for this slabcache. There are no concurrent accesses | |
2204 | * possible. | |
2205 | * | |
2206 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2207 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2208 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2209 | */ |
0094de92 | 2210 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2211 | { |
2212 | struct page *page; | |
2213 | struct kmem_cache_node *n; | |
ba84c73c | 2214 | unsigned long flags; |
81819f0f CL |
2215 | |
2216 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2217 | ||
a2f92ee7 | 2218 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2219 | |
2220 | BUG_ON(!page); | |
a2f92ee7 CL |
2221 | if (page_to_nid(page) != node) { |
2222 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2223 | "node %d\n", node); | |
2224 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2225 | "in order to be able to continue\n"); | |
2226 | } | |
2227 | ||
81819f0f CL |
2228 | n = page->freelist; |
2229 | BUG_ON(!n); | |
2230 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2231 | page->inuse++; | |
2232 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2233 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2234 | init_object(kmalloc_caches, n, 1); |
2235 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2236 | #endif |
5595cffc | 2237 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2238 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2239 | |
ba84c73c | 2240 | /* |
2241 | * lockdep requires consistent irq usage for each lock | |
2242 | * so even though there cannot be a race this early in | |
2243 | * the boot sequence, we still disable irqs. | |
2244 | */ | |
2245 | local_irq_save(flags); | |
7c2e132c | 2246 | add_partial(n, page, 0); |
ba84c73c | 2247 | local_irq_restore(flags); |
81819f0f CL |
2248 | } |
2249 | ||
2250 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2251 | { | |
2252 | int node; | |
2253 | ||
f64dc58c | 2254 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2255 | struct kmem_cache_node *n = s->node[node]; |
2256 | if (n && n != &s->local_node) | |
2257 | kmem_cache_free(kmalloc_caches, n); | |
2258 | s->node[node] = NULL; | |
2259 | } | |
2260 | } | |
2261 | ||
2262 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2263 | { | |
2264 | int node; | |
2265 | int local_node; | |
2266 | ||
2267 | if (slab_state >= UP) | |
2268 | local_node = page_to_nid(virt_to_page(s)); | |
2269 | else | |
2270 | local_node = 0; | |
2271 | ||
f64dc58c | 2272 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2273 | struct kmem_cache_node *n; |
2274 | ||
2275 | if (local_node == node) | |
2276 | n = &s->local_node; | |
2277 | else { | |
2278 | if (slab_state == DOWN) { | |
0094de92 | 2279 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2280 | continue; |
2281 | } | |
2282 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2283 | gfpflags, node); | |
2284 | ||
2285 | if (!n) { | |
2286 | free_kmem_cache_nodes(s); | |
2287 | return 0; | |
2288 | } | |
2289 | ||
2290 | } | |
2291 | s->node[node] = n; | |
5595cffc | 2292 | init_kmem_cache_node(n, s); |
81819f0f CL |
2293 | } |
2294 | return 1; | |
2295 | } | |
2296 | #else | |
2297 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2298 | { | |
2299 | } | |
2300 | ||
2301 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2302 | { | |
5595cffc | 2303 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2304 | return 1; |
2305 | } | |
2306 | #endif | |
2307 | ||
c0bdb232 | 2308 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2309 | { |
2310 | if (min < MIN_PARTIAL) | |
2311 | min = MIN_PARTIAL; | |
2312 | else if (min > MAX_PARTIAL) | |
2313 | min = MAX_PARTIAL; | |
2314 | s->min_partial = min; | |
2315 | } | |
2316 | ||
81819f0f CL |
2317 | /* |
2318 | * calculate_sizes() determines the order and the distribution of data within | |
2319 | * a slab object. | |
2320 | */ | |
06b285dc | 2321 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2322 | { |
2323 | unsigned long flags = s->flags; | |
2324 | unsigned long size = s->objsize; | |
2325 | unsigned long align = s->align; | |
834f3d11 | 2326 | int order; |
81819f0f | 2327 | |
d8b42bf5 CL |
2328 | /* |
2329 | * Round up object size to the next word boundary. We can only | |
2330 | * place the free pointer at word boundaries and this determines | |
2331 | * the possible location of the free pointer. | |
2332 | */ | |
2333 | size = ALIGN(size, sizeof(void *)); | |
2334 | ||
2335 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2336 | /* |
2337 | * Determine if we can poison the object itself. If the user of | |
2338 | * the slab may touch the object after free or before allocation | |
2339 | * then we should never poison the object itself. | |
2340 | */ | |
2341 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2342 | !s->ctor) |
81819f0f CL |
2343 | s->flags |= __OBJECT_POISON; |
2344 | else | |
2345 | s->flags &= ~__OBJECT_POISON; | |
2346 | ||
81819f0f CL |
2347 | |
2348 | /* | |
672bba3a | 2349 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2350 | * end of the object and the free pointer. If not then add an |
672bba3a | 2351 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2352 | */ |
2353 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2354 | size += sizeof(void *); | |
41ecc55b | 2355 | #endif |
81819f0f CL |
2356 | |
2357 | /* | |
672bba3a CL |
2358 | * With that we have determined the number of bytes in actual use |
2359 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2360 | */ |
2361 | s->inuse = size; | |
2362 | ||
2363 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2364 | s->ctor)) { |
81819f0f CL |
2365 | /* |
2366 | * Relocate free pointer after the object if it is not | |
2367 | * permitted to overwrite the first word of the object on | |
2368 | * kmem_cache_free. | |
2369 | * | |
2370 | * This is the case if we do RCU, have a constructor or | |
2371 | * destructor or are poisoning the objects. | |
2372 | */ | |
2373 | s->offset = size; | |
2374 | size += sizeof(void *); | |
2375 | } | |
2376 | ||
c12b3c62 | 2377 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2378 | if (flags & SLAB_STORE_USER) |
2379 | /* | |
2380 | * Need to store information about allocs and frees after | |
2381 | * the object. | |
2382 | */ | |
2383 | size += 2 * sizeof(struct track); | |
2384 | ||
be7b3fbc | 2385 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2386 | /* |
2387 | * Add some empty padding so that we can catch | |
2388 | * overwrites from earlier objects rather than let | |
2389 | * tracking information or the free pointer be | |
0211a9c8 | 2390 | * corrupted if a user writes before the start |
81819f0f CL |
2391 | * of the object. |
2392 | */ | |
2393 | size += sizeof(void *); | |
41ecc55b | 2394 | #endif |
672bba3a | 2395 | |
81819f0f CL |
2396 | /* |
2397 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2398 | * user specified and the dynamic determination of cache line size |
2399 | * on bootup. | |
81819f0f CL |
2400 | */ |
2401 | align = calculate_alignment(flags, align, s->objsize); | |
2402 | ||
2403 | /* | |
2404 | * SLUB stores one object immediately after another beginning from | |
2405 | * offset 0. In order to align the objects we have to simply size | |
2406 | * each object to conform to the alignment. | |
2407 | */ | |
2408 | size = ALIGN(size, align); | |
2409 | s->size = size; | |
06b285dc CL |
2410 | if (forced_order >= 0) |
2411 | order = forced_order; | |
2412 | else | |
2413 | order = calculate_order(size); | |
81819f0f | 2414 | |
834f3d11 | 2415 | if (order < 0) |
81819f0f CL |
2416 | return 0; |
2417 | ||
b7a49f0d | 2418 | s->allocflags = 0; |
834f3d11 | 2419 | if (order) |
b7a49f0d CL |
2420 | s->allocflags |= __GFP_COMP; |
2421 | ||
2422 | if (s->flags & SLAB_CACHE_DMA) | |
2423 | s->allocflags |= SLUB_DMA; | |
2424 | ||
2425 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2426 | s->allocflags |= __GFP_RECLAIMABLE; | |
2427 | ||
81819f0f CL |
2428 | /* |
2429 | * Determine the number of objects per slab | |
2430 | */ | |
834f3d11 | 2431 | s->oo = oo_make(order, size); |
65c3376a | 2432 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2433 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2434 | s->max = s->oo; | |
81819f0f | 2435 | |
834f3d11 | 2436 | return !!oo_objects(s->oo); |
81819f0f CL |
2437 | |
2438 | } | |
2439 | ||
81819f0f CL |
2440 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2441 | const char *name, size_t size, | |
2442 | size_t align, unsigned long flags, | |
51cc5068 | 2443 | void (*ctor)(void *)) |
81819f0f CL |
2444 | { |
2445 | memset(s, 0, kmem_size); | |
2446 | s->name = name; | |
2447 | s->ctor = ctor; | |
81819f0f | 2448 | s->objsize = size; |
81819f0f | 2449 | s->align = align; |
ba0268a8 | 2450 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2451 | |
06b285dc | 2452 | if (!calculate_sizes(s, -1)) |
81819f0f CL |
2453 | goto error; |
2454 | ||
3b89d7d8 DR |
2455 | /* |
2456 | * The larger the object size is, the more pages we want on the partial | |
2457 | * list to avoid pounding the page allocator excessively. | |
2458 | */ | |
c0bdb232 | 2459 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2460 | s->refcount = 1; |
2461 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2462 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2463 | #endif |
dfb4f096 CL |
2464 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2465 | goto error; | |
81819f0f | 2466 | |
dfb4f096 | 2467 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2468 | return 1; |
4c93c355 | 2469 | free_kmem_cache_nodes(s); |
81819f0f CL |
2470 | error: |
2471 | if (flags & SLAB_PANIC) | |
2472 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2473 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2474 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2475 | s->offset, flags); |
2476 | return 0; | |
2477 | } | |
81819f0f CL |
2478 | |
2479 | /* | |
2480 | * Check if a given pointer is valid | |
2481 | */ | |
2482 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2483 | { | |
06428780 | 2484 | struct page *page; |
81819f0f CL |
2485 | |
2486 | page = get_object_page(object); | |
2487 | ||
2488 | if (!page || s != page->slab) | |
2489 | /* No slab or wrong slab */ | |
2490 | return 0; | |
2491 | ||
abcd08a6 | 2492 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2493 | return 0; |
2494 | ||
2495 | /* | |
2496 | * We could also check if the object is on the slabs freelist. | |
2497 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2498 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2499 | * to a certain slab. |
2500 | */ | |
2501 | return 1; | |
2502 | } | |
2503 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2504 | ||
2505 | /* | |
2506 | * Determine the size of a slab object | |
2507 | */ | |
2508 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2509 | { | |
2510 | return s->objsize; | |
2511 | } | |
2512 | EXPORT_SYMBOL(kmem_cache_size); | |
2513 | ||
2514 | const char *kmem_cache_name(struct kmem_cache *s) | |
2515 | { | |
2516 | return s->name; | |
2517 | } | |
2518 | EXPORT_SYMBOL(kmem_cache_name); | |
2519 | ||
33b12c38 CL |
2520 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2521 | const char *text) | |
2522 | { | |
2523 | #ifdef CONFIG_SLUB_DEBUG | |
2524 | void *addr = page_address(page); | |
2525 | void *p; | |
2526 | DECLARE_BITMAP(map, page->objects); | |
2527 | ||
2528 | bitmap_zero(map, page->objects); | |
2529 | slab_err(s, page, "%s", text); | |
2530 | slab_lock(page); | |
2531 | for_each_free_object(p, s, page->freelist) | |
2532 | set_bit(slab_index(p, s, addr), map); | |
2533 | ||
2534 | for_each_object(p, s, addr, page->objects) { | |
2535 | ||
2536 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2537 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2538 | p, p - addr); | |
2539 | print_tracking(s, p); | |
2540 | } | |
2541 | } | |
2542 | slab_unlock(page); | |
2543 | #endif | |
2544 | } | |
2545 | ||
81819f0f | 2546 | /* |
599870b1 | 2547 | * Attempt to free all partial slabs on a node. |
81819f0f | 2548 | */ |
599870b1 | 2549 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2550 | { |
81819f0f CL |
2551 | unsigned long flags; |
2552 | struct page *page, *h; | |
2553 | ||
2554 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2555 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2556 | if (!page->inuse) { |
2557 | list_del(&page->lru); | |
2558 | discard_slab(s, page); | |
599870b1 | 2559 | n->nr_partial--; |
33b12c38 CL |
2560 | } else { |
2561 | list_slab_objects(s, page, | |
2562 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2563 | } |
33b12c38 | 2564 | } |
81819f0f | 2565 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2566 | } |
2567 | ||
2568 | /* | |
672bba3a | 2569 | * Release all resources used by a slab cache. |
81819f0f | 2570 | */ |
0c710013 | 2571 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2572 | { |
2573 | int node; | |
2574 | ||
2575 | flush_all(s); | |
2576 | ||
2577 | /* Attempt to free all objects */ | |
4c93c355 | 2578 | free_kmem_cache_cpus(s); |
f64dc58c | 2579 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2580 | struct kmem_cache_node *n = get_node(s, node); |
2581 | ||
599870b1 CL |
2582 | free_partial(s, n); |
2583 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2584 | return 1; |
2585 | } | |
2586 | free_kmem_cache_nodes(s); | |
2587 | return 0; | |
2588 | } | |
2589 | ||
2590 | /* | |
2591 | * Close a cache and release the kmem_cache structure | |
2592 | * (must be used for caches created using kmem_cache_create) | |
2593 | */ | |
2594 | void kmem_cache_destroy(struct kmem_cache *s) | |
2595 | { | |
7ed9f7e5 PM |
2596 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2597 | rcu_barrier(); | |
81819f0f CL |
2598 | down_write(&slub_lock); |
2599 | s->refcount--; | |
2600 | if (!s->refcount) { | |
2601 | list_del(&s->list); | |
a0e1d1be | 2602 | up_write(&slub_lock); |
d629d819 PE |
2603 | if (kmem_cache_close(s)) { |
2604 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2605 | "still has objects.\n", s->name, __func__); | |
2606 | dump_stack(); | |
2607 | } | |
81819f0f | 2608 | sysfs_slab_remove(s); |
a0e1d1be CL |
2609 | } else |
2610 | up_write(&slub_lock); | |
81819f0f CL |
2611 | } |
2612 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2613 | ||
2614 | /******************************************************************** | |
2615 | * Kmalloc subsystem | |
2616 | *******************************************************************/ | |
2617 | ||
ffadd4d0 | 2618 | struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2619 | EXPORT_SYMBOL(kmalloc_caches); |
2620 | ||
81819f0f CL |
2621 | static int __init setup_slub_min_order(char *str) |
2622 | { | |
06428780 | 2623 | get_option(&str, &slub_min_order); |
81819f0f CL |
2624 | |
2625 | return 1; | |
2626 | } | |
2627 | ||
2628 | __setup("slub_min_order=", setup_slub_min_order); | |
2629 | ||
2630 | static int __init setup_slub_max_order(char *str) | |
2631 | { | |
06428780 | 2632 | get_option(&str, &slub_max_order); |
818cf590 | 2633 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2634 | |
2635 | return 1; | |
2636 | } | |
2637 | ||
2638 | __setup("slub_max_order=", setup_slub_max_order); | |
2639 | ||
2640 | static int __init setup_slub_min_objects(char *str) | |
2641 | { | |
06428780 | 2642 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2643 | |
2644 | return 1; | |
2645 | } | |
2646 | ||
2647 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2648 | ||
2649 | static int __init setup_slub_nomerge(char *str) | |
2650 | { | |
2651 | slub_nomerge = 1; | |
2652 | return 1; | |
2653 | } | |
2654 | ||
2655 | __setup("slub_nomerge", setup_slub_nomerge); | |
2656 | ||
81819f0f CL |
2657 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2658 | const char *name, int size, gfp_t gfp_flags) | |
2659 | { | |
2660 | unsigned int flags = 0; | |
2661 | ||
2662 | if (gfp_flags & SLUB_DMA) | |
2663 | flags = SLAB_CACHE_DMA; | |
2664 | ||
83b519e8 PE |
2665 | /* |
2666 | * This function is called with IRQs disabled during early-boot on | |
2667 | * single CPU so there's no need to take slub_lock here. | |
2668 | */ | |
81819f0f | 2669 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2670 | flags, NULL)) |
81819f0f CL |
2671 | goto panic; |
2672 | ||
2673 | list_add(&s->list, &slab_caches); | |
83b519e8 | 2674 | |
81819f0f CL |
2675 | if (sysfs_slab_add(s)) |
2676 | goto panic; | |
2677 | return s; | |
2678 | ||
2679 | panic: | |
2680 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2681 | } | |
2682 | ||
2e443fd0 | 2683 | #ifdef CONFIG_ZONE_DMA |
ffadd4d0 | 2684 | static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT]; |
1ceef402 CL |
2685 | |
2686 | static void sysfs_add_func(struct work_struct *w) | |
2687 | { | |
2688 | struct kmem_cache *s; | |
2689 | ||
2690 | down_write(&slub_lock); | |
2691 | list_for_each_entry(s, &slab_caches, list) { | |
2692 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2693 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2694 | sysfs_slab_add(s); | |
2695 | } | |
2696 | } | |
2697 | up_write(&slub_lock); | |
2698 | } | |
2699 | ||
2700 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2701 | ||
2e443fd0 CL |
2702 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2703 | { | |
2704 | struct kmem_cache *s; | |
2e443fd0 CL |
2705 | char *text; |
2706 | size_t realsize; | |
964cf35c | 2707 | unsigned long slabflags; |
2e443fd0 CL |
2708 | |
2709 | s = kmalloc_caches_dma[index]; | |
2710 | if (s) | |
2711 | return s; | |
2712 | ||
2713 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2714 | if (flags & __GFP_WAIT) |
2715 | down_write(&slub_lock); | |
2716 | else { | |
2717 | if (!down_write_trylock(&slub_lock)) | |
2718 | goto out; | |
2719 | } | |
2720 | ||
2721 | if (kmalloc_caches_dma[index]) | |
2722 | goto unlock_out; | |
2e443fd0 | 2723 | |
7b55f620 | 2724 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2725 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2726 | (unsigned int)realsize); | |
1ceef402 CL |
2727 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2728 | ||
964cf35c NP |
2729 | /* |
2730 | * Must defer sysfs creation to a workqueue because we don't know | |
2731 | * what context we are called from. Before sysfs comes up, we don't | |
2732 | * need to do anything because our sysfs initcall will start by | |
2733 | * adding all existing slabs to sysfs. | |
2734 | */ | |
5caf5c7d | 2735 | slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK; |
964cf35c NP |
2736 | if (slab_state >= SYSFS) |
2737 | slabflags |= __SYSFS_ADD_DEFERRED; | |
2738 | ||
1ceef402 | 2739 | if (!s || !text || !kmem_cache_open(s, flags, text, |
964cf35c | 2740 | realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) { |
1ceef402 CL |
2741 | kfree(s); |
2742 | kfree(text); | |
2743 | goto unlock_out; | |
dfce8648 | 2744 | } |
1ceef402 CL |
2745 | |
2746 | list_add(&s->list, &slab_caches); | |
2747 | kmalloc_caches_dma[index] = s; | |
2748 | ||
964cf35c NP |
2749 | if (slab_state >= SYSFS) |
2750 | schedule_work(&sysfs_add_work); | |
1ceef402 CL |
2751 | |
2752 | unlock_out: | |
dfce8648 | 2753 | up_write(&slub_lock); |
1ceef402 | 2754 | out: |
dfce8648 | 2755 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2756 | } |
2757 | #endif | |
2758 | ||
f1b26339 CL |
2759 | /* |
2760 | * Conversion table for small slabs sizes / 8 to the index in the | |
2761 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2762 | * of two cache sizes there. The size of larger slabs can be determined using | |
2763 | * fls. | |
2764 | */ | |
2765 | static s8 size_index[24] = { | |
2766 | 3, /* 8 */ | |
2767 | 4, /* 16 */ | |
2768 | 5, /* 24 */ | |
2769 | 5, /* 32 */ | |
2770 | 6, /* 40 */ | |
2771 | 6, /* 48 */ | |
2772 | 6, /* 56 */ | |
2773 | 6, /* 64 */ | |
2774 | 1, /* 72 */ | |
2775 | 1, /* 80 */ | |
2776 | 1, /* 88 */ | |
2777 | 1, /* 96 */ | |
2778 | 7, /* 104 */ | |
2779 | 7, /* 112 */ | |
2780 | 7, /* 120 */ | |
2781 | 7, /* 128 */ | |
2782 | 2, /* 136 */ | |
2783 | 2, /* 144 */ | |
2784 | 2, /* 152 */ | |
2785 | 2, /* 160 */ | |
2786 | 2, /* 168 */ | |
2787 | 2, /* 176 */ | |
2788 | 2, /* 184 */ | |
2789 | 2 /* 192 */ | |
2790 | }; | |
2791 | ||
81819f0f CL |
2792 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2793 | { | |
f1b26339 | 2794 | int index; |
81819f0f | 2795 | |
f1b26339 CL |
2796 | if (size <= 192) { |
2797 | if (!size) | |
2798 | return ZERO_SIZE_PTR; | |
81819f0f | 2799 | |
f1b26339 | 2800 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2801 | } else |
f1b26339 | 2802 | index = fls(size - 1); |
81819f0f CL |
2803 | |
2804 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2805 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2806 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2807 | |
81819f0f CL |
2808 | #endif |
2809 | return &kmalloc_caches[index]; | |
2810 | } | |
2811 | ||
2812 | void *__kmalloc(size_t size, gfp_t flags) | |
2813 | { | |
aadb4bc4 | 2814 | struct kmem_cache *s; |
5b882be4 | 2815 | void *ret; |
81819f0f | 2816 | |
ffadd4d0 | 2817 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2818 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2819 | |
2820 | s = get_slab(size, flags); | |
2821 | ||
2822 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2823 | return s; |
2824 | ||
5b882be4 EGM |
2825 | ret = slab_alloc(s, flags, -1, _RET_IP_); |
2826 | ||
ca2b84cb | 2827 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2828 | |
2829 | return ret; | |
81819f0f CL |
2830 | } |
2831 | EXPORT_SYMBOL(__kmalloc); | |
2832 | ||
f619cfe1 CL |
2833 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2834 | { | |
b1eeab67 | 2835 | struct page *page; |
e4f7c0b4 | 2836 | void *ptr = NULL; |
f619cfe1 | 2837 | |
b1eeab67 VN |
2838 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2839 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2840 | if (page) |
e4f7c0b4 CM |
2841 | ptr = page_address(page); |
2842 | ||
2843 | kmemleak_alloc(ptr, size, 1, flags); | |
2844 | return ptr; | |
f619cfe1 CL |
2845 | } |
2846 | ||
81819f0f CL |
2847 | #ifdef CONFIG_NUMA |
2848 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2849 | { | |
aadb4bc4 | 2850 | struct kmem_cache *s; |
5b882be4 | 2851 | void *ret; |
81819f0f | 2852 | |
057685cf | 2853 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2854 | ret = kmalloc_large_node(size, flags, node); |
2855 | ||
ca2b84cb EGM |
2856 | trace_kmalloc_node(_RET_IP_, ret, |
2857 | size, PAGE_SIZE << get_order(size), | |
2858 | flags, node); | |
5b882be4 EGM |
2859 | |
2860 | return ret; | |
2861 | } | |
aadb4bc4 CL |
2862 | |
2863 | s = get_slab(size, flags); | |
2864 | ||
2865 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2866 | return s; |
2867 | ||
5b882be4 EGM |
2868 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2869 | ||
ca2b84cb | 2870 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2871 | |
2872 | return ret; | |
81819f0f CL |
2873 | } |
2874 | EXPORT_SYMBOL(__kmalloc_node); | |
2875 | #endif | |
2876 | ||
2877 | size_t ksize(const void *object) | |
2878 | { | |
272c1d21 | 2879 | struct page *page; |
81819f0f CL |
2880 | struct kmem_cache *s; |
2881 | ||
ef8b4520 | 2882 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2883 | return 0; |
2884 | ||
294a80a8 | 2885 | page = virt_to_head_page(object); |
294a80a8 | 2886 | |
76994412 PE |
2887 | if (unlikely(!PageSlab(page))) { |
2888 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2889 | return PAGE_SIZE << compound_order(page); |
76994412 | 2890 | } |
81819f0f | 2891 | s = page->slab; |
81819f0f | 2892 | |
ae20bfda | 2893 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2894 | /* |
2895 | * Debugging requires use of the padding between object | |
2896 | * and whatever may come after it. | |
2897 | */ | |
2898 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2899 | return s->objsize; | |
2900 | ||
ae20bfda | 2901 | #endif |
81819f0f CL |
2902 | /* |
2903 | * If we have the need to store the freelist pointer | |
2904 | * back there or track user information then we can | |
2905 | * only use the space before that information. | |
2906 | */ | |
2907 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2908 | return s->inuse; | |
81819f0f CL |
2909 | /* |
2910 | * Else we can use all the padding etc for the allocation | |
2911 | */ | |
2912 | return s->size; | |
2913 | } | |
b1aabecd | 2914 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2915 | |
2916 | void kfree(const void *x) | |
2917 | { | |
81819f0f | 2918 | struct page *page; |
5bb983b0 | 2919 | void *object = (void *)x; |
81819f0f | 2920 | |
2121db74 PE |
2921 | trace_kfree(_RET_IP_, x); |
2922 | ||
2408c550 | 2923 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2924 | return; |
2925 | ||
b49af68f | 2926 | page = virt_to_head_page(x); |
aadb4bc4 | 2927 | if (unlikely(!PageSlab(page))) { |
0937502a | 2928 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2929 | kmemleak_free(x); |
aadb4bc4 CL |
2930 | put_page(page); |
2931 | return; | |
2932 | } | |
ce71e27c | 2933 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2934 | } |
2935 | EXPORT_SYMBOL(kfree); | |
2936 | ||
2086d26a | 2937 | /* |
672bba3a CL |
2938 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2939 | * the remaining slabs by the number of items in use. The slabs with the | |
2940 | * most items in use come first. New allocations will then fill those up | |
2941 | * and thus they can be removed from the partial lists. | |
2942 | * | |
2943 | * The slabs with the least items are placed last. This results in them | |
2944 | * being allocated from last increasing the chance that the last objects | |
2945 | * are freed in them. | |
2086d26a CL |
2946 | */ |
2947 | int kmem_cache_shrink(struct kmem_cache *s) | |
2948 | { | |
2949 | int node; | |
2950 | int i; | |
2951 | struct kmem_cache_node *n; | |
2952 | struct page *page; | |
2953 | struct page *t; | |
205ab99d | 2954 | int objects = oo_objects(s->max); |
2086d26a | 2955 | struct list_head *slabs_by_inuse = |
834f3d11 | 2956 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2957 | unsigned long flags; |
2958 | ||
2959 | if (!slabs_by_inuse) | |
2960 | return -ENOMEM; | |
2961 | ||
2962 | flush_all(s); | |
f64dc58c | 2963 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2964 | n = get_node(s, node); |
2965 | ||
2966 | if (!n->nr_partial) | |
2967 | continue; | |
2968 | ||
834f3d11 | 2969 | for (i = 0; i < objects; i++) |
2086d26a CL |
2970 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2971 | ||
2972 | spin_lock_irqsave(&n->list_lock, flags); | |
2973 | ||
2974 | /* | |
672bba3a | 2975 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2976 | * |
672bba3a CL |
2977 | * Note that concurrent frees may occur while we hold the |
2978 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2979 | */ |
2980 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2981 | if (!page->inuse && slab_trylock(page)) { | |
2982 | /* | |
2983 | * Must hold slab lock here because slab_free | |
2984 | * may have freed the last object and be | |
2985 | * waiting to release the slab. | |
2986 | */ | |
2987 | list_del(&page->lru); | |
2988 | n->nr_partial--; | |
2989 | slab_unlock(page); | |
2990 | discard_slab(s, page); | |
2991 | } else { | |
fcda3d89 CL |
2992 | list_move(&page->lru, |
2993 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2994 | } |
2995 | } | |
2996 | ||
2086d26a | 2997 | /* |
672bba3a CL |
2998 | * Rebuild the partial list with the slabs filled up most |
2999 | * first and the least used slabs at the end. | |
2086d26a | 3000 | */ |
834f3d11 | 3001 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3002 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3003 | ||
2086d26a CL |
3004 | spin_unlock_irqrestore(&n->list_lock, flags); |
3005 | } | |
3006 | ||
3007 | kfree(slabs_by_inuse); | |
3008 | return 0; | |
3009 | } | |
3010 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3011 | ||
b9049e23 YG |
3012 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
3013 | static int slab_mem_going_offline_callback(void *arg) | |
3014 | { | |
3015 | struct kmem_cache *s; | |
3016 | ||
3017 | down_read(&slub_lock); | |
3018 | list_for_each_entry(s, &slab_caches, list) | |
3019 | kmem_cache_shrink(s); | |
3020 | up_read(&slub_lock); | |
3021 | ||
3022 | return 0; | |
3023 | } | |
3024 | ||
3025 | static void slab_mem_offline_callback(void *arg) | |
3026 | { | |
3027 | struct kmem_cache_node *n; | |
3028 | struct kmem_cache *s; | |
3029 | struct memory_notify *marg = arg; | |
3030 | int offline_node; | |
3031 | ||
3032 | offline_node = marg->status_change_nid; | |
3033 | ||
3034 | /* | |
3035 | * If the node still has available memory. we need kmem_cache_node | |
3036 | * for it yet. | |
3037 | */ | |
3038 | if (offline_node < 0) | |
3039 | return; | |
3040 | ||
3041 | down_read(&slub_lock); | |
3042 | list_for_each_entry(s, &slab_caches, list) { | |
3043 | n = get_node(s, offline_node); | |
3044 | if (n) { | |
3045 | /* | |
3046 | * if n->nr_slabs > 0, slabs still exist on the node | |
3047 | * that is going down. We were unable to free them, | |
3048 | * and offline_pages() function shoudn't call this | |
3049 | * callback. So, we must fail. | |
3050 | */ | |
0f389ec6 | 3051 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3052 | |
3053 | s->node[offline_node] = NULL; | |
3054 | kmem_cache_free(kmalloc_caches, n); | |
3055 | } | |
3056 | } | |
3057 | up_read(&slub_lock); | |
3058 | } | |
3059 | ||
3060 | static int slab_mem_going_online_callback(void *arg) | |
3061 | { | |
3062 | struct kmem_cache_node *n; | |
3063 | struct kmem_cache *s; | |
3064 | struct memory_notify *marg = arg; | |
3065 | int nid = marg->status_change_nid; | |
3066 | int ret = 0; | |
3067 | ||
3068 | /* | |
3069 | * If the node's memory is already available, then kmem_cache_node is | |
3070 | * already created. Nothing to do. | |
3071 | */ | |
3072 | if (nid < 0) | |
3073 | return 0; | |
3074 | ||
3075 | /* | |
0121c619 | 3076 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3077 | * allocate a kmem_cache_node structure in order to bring the node |
3078 | * online. | |
3079 | */ | |
3080 | down_read(&slub_lock); | |
3081 | list_for_each_entry(s, &slab_caches, list) { | |
3082 | /* | |
3083 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3084 | * since memory is not yet available from the node that | |
3085 | * is brought up. | |
3086 | */ | |
3087 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
3088 | if (!n) { | |
3089 | ret = -ENOMEM; | |
3090 | goto out; | |
3091 | } | |
5595cffc | 3092 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3093 | s->node[nid] = n; |
3094 | } | |
3095 | out: | |
3096 | up_read(&slub_lock); | |
3097 | return ret; | |
3098 | } | |
3099 | ||
3100 | static int slab_memory_callback(struct notifier_block *self, | |
3101 | unsigned long action, void *arg) | |
3102 | { | |
3103 | int ret = 0; | |
3104 | ||
3105 | switch (action) { | |
3106 | case MEM_GOING_ONLINE: | |
3107 | ret = slab_mem_going_online_callback(arg); | |
3108 | break; | |
3109 | case MEM_GOING_OFFLINE: | |
3110 | ret = slab_mem_going_offline_callback(arg); | |
3111 | break; | |
3112 | case MEM_OFFLINE: | |
3113 | case MEM_CANCEL_ONLINE: | |
3114 | slab_mem_offline_callback(arg); | |
3115 | break; | |
3116 | case MEM_ONLINE: | |
3117 | case MEM_CANCEL_OFFLINE: | |
3118 | break; | |
3119 | } | |
dc19f9db KH |
3120 | if (ret) |
3121 | ret = notifier_from_errno(ret); | |
3122 | else | |
3123 | ret = NOTIFY_OK; | |
b9049e23 YG |
3124 | return ret; |
3125 | } | |
3126 | ||
3127 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3128 | ||
81819f0f CL |
3129 | /******************************************************************** |
3130 | * Basic setup of slabs | |
3131 | *******************************************************************/ | |
3132 | ||
3133 | void __init kmem_cache_init(void) | |
3134 | { | |
3135 | int i; | |
4b356be0 | 3136 | int caches = 0; |
81819f0f | 3137 | |
4c93c355 CL |
3138 | init_alloc_cpu(); |
3139 | ||
81819f0f CL |
3140 | #ifdef CONFIG_NUMA |
3141 | /* | |
3142 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3143 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3144 | * kmem_cache_open for slab_state == DOWN. |
3145 | */ | |
3146 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
83b519e8 | 3147 | sizeof(struct kmem_cache_node), GFP_NOWAIT); |
8ffa6875 | 3148 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3149 | caches++; |
b9049e23 | 3150 | |
0c40ba4f | 3151 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3152 | #endif |
3153 | ||
3154 | /* Able to allocate the per node structures */ | |
3155 | slab_state = PARTIAL; | |
3156 | ||
3157 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
3158 | if (KMALLOC_MIN_SIZE <= 64) { |
3159 | create_kmalloc_cache(&kmalloc_caches[1], | |
83b519e8 | 3160 | "kmalloc-96", 96, GFP_NOWAIT); |
4b356be0 | 3161 | caches++; |
4b356be0 | 3162 | create_kmalloc_cache(&kmalloc_caches[2], |
83b519e8 | 3163 | "kmalloc-192", 192, GFP_NOWAIT); |
4b356be0 CL |
3164 | caches++; |
3165 | } | |
81819f0f | 3166 | |
ffadd4d0 | 3167 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
81819f0f | 3168 | create_kmalloc_cache(&kmalloc_caches[i], |
83b519e8 | 3169 | "kmalloc", 1 << i, GFP_NOWAIT); |
4b356be0 CL |
3170 | caches++; |
3171 | } | |
81819f0f | 3172 | |
f1b26339 CL |
3173 | |
3174 | /* | |
3175 | * Patch up the size_index table if we have strange large alignment | |
3176 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3177 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3178 | * |
3179 | * Largest permitted alignment is 256 bytes due to the way we | |
3180 | * handle the index determination for the smaller caches. | |
3181 | * | |
3182 | * Make sure that nothing crazy happens if someone starts tinkering | |
3183 | * around with ARCH_KMALLOC_MINALIGN | |
3184 | */ | |
3185 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3186 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3187 | ||
12ad6843 | 3188 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3189 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3190 | ||
41d54d3b CL |
3191 | if (KMALLOC_MIN_SIZE == 128) { |
3192 | /* | |
3193 | * The 192 byte sized cache is not used if the alignment | |
3194 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3195 | * instead. | |
3196 | */ | |
3197 | for (i = 128 + 8; i <= 192; i += 8) | |
3198 | size_index[(i - 1) / 8] = 8; | |
3199 | } | |
3200 | ||
81819f0f CL |
3201 | slab_state = UP; |
3202 | ||
3203 | /* Provide the correct kmalloc names now that the caches are up */ | |
ffadd4d0 | 3204 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) |
81819f0f | 3205 | kmalloc_caches[i]. name = |
83b519e8 | 3206 | kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); |
81819f0f CL |
3207 | |
3208 | #ifdef CONFIG_SMP | |
3209 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3210 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3211 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3212 | #else | |
3213 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3214 | #endif |
3215 | ||
3adbefee IM |
3216 | printk(KERN_INFO |
3217 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3218 | " CPUs=%d, Nodes=%d\n", |
3219 | caches, cache_line_size(), | |
81819f0f CL |
3220 | slub_min_order, slub_max_order, slub_min_objects, |
3221 | nr_cpu_ids, nr_node_ids); | |
3222 | } | |
3223 | ||
7e85ee0c PE |
3224 | void __init kmem_cache_init_late(void) |
3225 | { | |
7e85ee0c PE |
3226 | } |
3227 | ||
81819f0f CL |
3228 | /* |
3229 | * Find a mergeable slab cache | |
3230 | */ | |
3231 | static int slab_unmergeable(struct kmem_cache *s) | |
3232 | { | |
3233 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3234 | return 1; | |
3235 | ||
c59def9f | 3236 | if (s->ctor) |
81819f0f CL |
3237 | return 1; |
3238 | ||
8ffa6875 CL |
3239 | /* |
3240 | * We may have set a slab to be unmergeable during bootstrap. | |
3241 | */ | |
3242 | if (s->refcount < 0) | |
3243 | return 1; | |
3244 | ||
81819f0f CL |
3245 | return 0; |
3246 | } | |
3247 | ||
3248 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3249 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3250 | void (*ctor)(void *)) |
81819f0f | 3251 | { |
5b95a4ac | 3252 | struct kmem_cache *s; |
81819f0f CL |
3253 | |
3254 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3255 | return NULL; | |
3256 | ||
c59def9f | 3257 | if (ctor) |
81819f0f CL |
3258 | return NULL; |
3259 | ||
3260 | size = ALIGN(size, sizeof(void *)); | |
3261 | align = calculate_alignment(flags, align, size); | |
3262 | size = ALIGN(size, align); | |
ba0268a8 | 3263 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3264 | |
5b95a4ac | 3265 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3266 | if (slab_unmergeable(s)) |
3267 | continue; | |
3268 | ||
3269 | if (size > s->size) | |
3270 | continue; | |
3271 | ||
ba0268a8 | 3272 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3273 | continue; |
3274 | /* | |
3275 | * Check if alignment is compatible. | |
3276 | * Courtesy of Adrian Drzewiecki | |
3277 | */ | |
06428780 | 3278 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3279 | continue; |
3280 | ||
3281 | if (s->size - size >= sizeof(void *)) | |
3282 | continue; | |
3283 | ||
3284 | return s; | |
3285 | } | |
3286 | return NULL; | |
3287 | } | |
3288 | ||
3289 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3290 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3291 | { |
3292 | struct kmem_cache *s; | |
3293 | ||
3294 | down_write(&slub_lock); | |
ba0268a8 | 3295 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3296 | if (s) { |
42a9fdbb CL |
3297 | int cpu; |
3298 | ||
81819f0f CL |
3299 | s->refcount++; |
3300 | /* | |
3301 | * Adjust the object sizes so that we clear | |
3302 | * the complete object on kzalloc. | |
3303 | */ | |
3304 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3305 | |
3306 | /* | |
3307 | * And then we need to update the object size in the | |
3308 | * per cpu structures | |
3309 | */ | |
3310 | for_each_online_cpu(cpu) | |
3311 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3312 | |
81819f0f | 3313 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3314 | up_write(&slub_lock); |
6446faa2 | 3315 | |
7b8f3b66 DR |
3316 | if (sysfs_slab_alias(s, name)) { |
3317 | down_write(&slub_lock); | |
3318 | s->refcount--; | |
3319 | up_write(&slub_lock); | |
81819f0f | 3320 | goto err; |
7b8f3b66 | 3321 | } |
a0e1d1be CL |
3322 | return s; |
3323 | } | |
6446faa2 | 3324 | |
a0e1d1be CL |
3325 | s = kmalloc(kmem_size, GFP_KERNEL); |
3326 | if (s) { | |
3327 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3328 | size, align, flags, ctor)) { |
81819f0f | 3329 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3330 | up_write(&slub_lock); |
7b8f3b66 DR |
3331 | if (sysfs_slab_add(s)) { |
3332 | down_write(&slub_lock); | |
3333 | list_del(&s->list); | |
3334 | up_write(&slub_lock); | |
3335 | kfree(s); | |
a0e1d1be | 3336 | goto err; |
7b8f3b66 | 3337 | } |
a0e1d1be CL |
3338 | return s; |
3339 | } | |
3340 | kfree(s); | |
81819f0f CL |
3341 | } |
3342 | up_write(&slub_lock); | |
81819f0f CL |
3343 | |
3344 | err: | |
81819f0f CL |
3345 | if (flags & SLAB_PANIC) |
3346 | panic("Cannot create slabcache %s\n", name); | |
3347 | else | |
3348 | s = NULL; | |
3349 | return s; | |
3350 | } | |
3351 | EXPORT_SYMBOL(kmem_cache_create); | |
3352 | ||
81819f0f | 3353 | #ifdef CONFIG_SMP |
81819f0f | 3354 | /* |
672bba3a CL |
3355 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3356 | * necessary. | |
81819f0f CL |
3357 | */ |
3358 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3359 | unsigned long action, void *hcpu) | |
3360 | { | |
3361 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3362 | struct kmem_cache *s; |
3363 | unsigned long flags; | |
81819f0f CL |
3364 | |
3365 | switch (action) { | |
4c93c355 CL |
3366 | case CPU_UP_PREPARE: |
3367 | case CPU_UP_PREPARE_FROZEN: | |
3368 | init_alloc_cpu_cpu(cpu); | |
3369 | down_read(&slub_lock); | |
3370 | list_for_each_entry(s, &slab_caches, list) | |
3371 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3372 | GFP_KERNEL); | |
3373 | up_read(&slub_lock); | |
3374 | break; | |
3375 | ||
81819f0f | 3376 | case CPU_UP_CANCELED: |
8bb78442 | 3377 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3378 | case CPU_DEAD: |
8bb78442 | 3379 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3380 | down_read(&slub_lock); |
3381 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3382 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3383 | ||
5b95a4ac CL |
3384 | local_irq_save(flags); |
3385 | __flush_cpu_slab(s, cpu); | |
3386 | local_irq_restore(flags); | |
4c93c355 CL |
3387 | free_kmem_cache_cpu(c, cpu); |
3388 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3389 | } |
3390 | up_read(&slub_lock); | |
81819f0f CL |
3391 | break; |
3392 | default: | |
3393 | break; | |
3394 | } | |
3395 | return NOTIFY_OK; | |
3396 | } | |
3397 | ||
06428780 | 3398 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3399 | .notifier_call = slab_cpuup_callback |
06428780 | 3400 | }; |
81819f0f CL |
3401 | |
3402 | #endif | |
3403 | ||
ce71e27c | 3404 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3405 | { |
aadb4bc4 | 3406 | struct kmem_cache *s; |
94b528d0 | 3407 | void *ret; |
aadb4bc4 | 3408 | |
ffadd4d0 | 3409 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3410 | return kmalloc_large(size, gfpflags); |
3411 | ||
aadb4bc4 | 3412 | s = get_slab(size, gfpflags); |
81819f0f | 3413 | |
2408c550 | 3414 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3415 | return s; |
81819f0f | 3416 | |
94b528d0 EGM |
3417 | ret = slab_alloc(s, gfpflags, -1, caller); |
3418 | ||
3419 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3420 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3421 | |
3422 | return ret; | |
81819f0f CL |
3423 | } |
3424 | ||
3425 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3426 | int node, unsigned long caller) |
81819f0f | 3427 | { |
aadb4bc4 | 3428 | struct kmem_cache *s; |
94b528d0 | 3429 | void *ret; |
aadb4bc4 | 3430 | |
ffadd4d0 | 3431 | if (unlikely(size > SLUB_MAX_SIZE)) |
f619cfe1 | 3432 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3433 | |
aadb4bc4 | 3434 | s = get_slab(size, gfpflags); |
81819f0f | 3435 | |
2408c550 | 3436 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3437 | return s; |
81819f0f | 3438 | |
94b528d0 EGM |
3439 | ret = slab_alloc(s, gfpflags, node, caller); |
3440 | ||
3441 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3442 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3443 | |
3444 | return ret; | |
81819f0f CL |
3445 | } |
3446 | ||
f6acb635 | 3447 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3448 | static int count_inuse(struct page *page) |
3449 | { | |
3450 | return page->inuse; | |
3451 | } | |
3452 | ||
3453 | static int count_total(struct page *page) | |
3454 | { | |
3455 | return page->objects; | |
3456 | } | |
3457 | ||
434e245d CL |
3458 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3459 | unsigned long *map) | |
53e15af0 CL |
3460 | { |
3461 | void *p; | |
a973e9dd | 3462 | void *addr = page_address(page); |
53e15af0 CL |
3463 | |
3464 | if (!check_slab(s, page) || | |
3465 | !on_freelist(s, page, NULL)) | |
3466 | return 0; | |
3467 | ||
3468 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3469 | bitmap_zero(map, page->objects); |
53e15af0 | 3470 | |
7656c72b CL |
3471 | for_each_free_object(p, s, page->freelist) { |
3472 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3473 | if (!check_object(s, page, p, 0)) |
3474 | return 0; | |
3475 | } | |
3476 | ||
224a88be | 3477 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3478 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3479 | if (!check_object(s, page, p, 1)) |
3480 | return 0; | |
3481 | return 1; | |
3482 | } | |
3483 | ||
434e245d CL |
3484 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3485 | unsigned long *map) | |
53e15af0 CL |
3486 | { |
3487 | if (slab_trylock(page)) { | |
434e245d | 3488 | validate_slab(s, page, map); |
53e15af0 CL |
3489 | slab_unlock(page); |
3490 | } else | |
3491 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3492 | s->name, page); | |
3493 | ||
3494 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3495 | if (!PageSlubDebug(page)) |
3496 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3497 | "on slab 0x%p\n", s->name, page); |
3498 | } else { | |
8a38082d AW |
3499 | if (PageSlubDebug(page)) |
3500 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3501 | "slab 0x%p\n", s->name, page); |
3502 | } | |
3503 | } | |
3504 | ||
434e245d CL |
3505 | static int validate_slab_node(struct kmem_cache *s, |
3506 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3507 | { |
3508 | unsigned long count = 0; | |
3509 | struct page *page; | |
3510 | unsigned long flags; | |
3511 | ||
3512 | spin_lock_irqsave(&n->list_lock, flags); | |
3513 | ||
3514 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3515 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3516 | count++; |
3517 | } | |
3518 | if (count != n->nr_partial) | |
3519 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3520 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3521 | ||
3522 | if (!(s->flags & SLAB_STORE_USER)) | |
3523 | goto out; | |
3524 | ||
3525 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3526 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3527 | count++; |
3528 | } | |
3529 | if (count != atomic_long_read(&n->nr_slabs)) | |
3530 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3531 | "counter=%ld\n", s->name, count, | |
3532 | atomic_long_read(&n->nr_slabs)); | |
3533 | ||
3534 | out: | |
3535 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3536 | return count; | |
3537 | } | |
3538 | ||
434e245d | 3539 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3540 | { |
3541 | int node; | |
3542 | unsigned long count = 0; | |
205ab99d | 3543 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3544 | sizeof(unsigned long), GFP_KERNEL); |
3545 | ||
3546 | if (!map) | |
3547 | return -ENOMEM; | |
53e15af0 CL |
3548 | |
3549 | flush_all(s); | |
f64dc58c | 3550 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3551 | struct kmem_cache_node *n = get_node(s, node); |
3552 | ||
434e245d | 3553 | count += validate_slab_node(s, n, map); |
53e15af0 | 3554 | } |
434e245d | 3555 | kfree(map); |
53e15af0 CL |
3556 | return count; |
3557 | } | |
3558 | ||
b3459709 CL |
3559 | #ifdef SLUB_RESILIENCY_TEST |
3560 | static void resiliency_test(void) | |
3561 | { | |
3562 | u8 *p; | |
3563 | ||
3564 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3565 | printk(KERN_ERR "-----------------------\n"); | |
3566 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3567 | ||
3568 | p = kzalloc(16, GFP_KERNEL); | |
3569 | p[16] = 0x12; | |
3570 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3571 | " 0x12->0x%p\n\n", p + 16); | |
3572 | ||
3573 | validate_slab_cache(kmalloc_caches + 4); | |
3574 | ||
3575 | /* Hmmm... The next two are dangerous */ | |
3576 | p = kzalloc(32, GFP_KERNEL); | |
3577 | p[32 + sizeof(void *)] = 0x34; | |
3578 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3579 | " 0x34 -> -0x%p\n", p); |
3580 | printk(KERN_ERR | |
3581 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3582 | |
3583 | validate_slab_cache(kmalloc_caches + 5); | |
3584 | p = kzalloc(64, GFP_KERNEL); | |
3585 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3586 | *p = 0x56; | |
3587 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3588 | p); | |
3adbefee IM |
3589 | printk(KERN_ERR |
3590 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3591 | validate_slab_cache(kmalloc_caches + 6); |
3592 | ||
3593 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3594 | p = kzalloc(128, GFP_KERNEL); | |
3595 | kfree(p); | |
3596 | *p = 0x78; | |
3597 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3598 | validate_slab_cache(kmalloc_caches + 7); | |
3599 | ||
3600 | p = kzalloc(256, GFP_KERNEL); | |
3601 | kfree(p); | |
3602 | p[50] = 0x9a; | |
3adbefee IM |
3603 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3604 | p); | |
b3459709 CL |
3605 | validate_slab_cache(kmalloc_caches + 8); |
3606 | ||
3607 | p = kzalloc(512, GFP_KERNEL); | |
3608 | kfree(p); | |
3609 | p[512] = 0xab; | |
3610 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3611 | validate_slab_cache(kmalloc_caches + 9); | |
3612 | } | |
3613 | #else | |
3614 | static void resiliency_test(void) {}; | |
3615 | #endif | |
3616 | ||
88a420e4 | 3617 | /* |
672bba3a | 3618 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3619 | * and freed. |
3620 | */ | |
3621 | ||
3622 | struct location { | |
3623 | unsigned long count; | |
ce71e27c | 3624 | unsigned long addr; |
45edfa58 CL |
3625 | long long sum_time; |
3626 | long min_time; | |
3627 | long max_time; | |
3628 | long min_pid; | |
3629 | long max_pid; | |
174596a0 | 3630 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3631 | nodemask_t nodes; |
88a420e4 CL |
3632 | }; |
3633 | ||
3634 | struct loc_track { | |
3635 | unsigned long max; | |
3636 | unsigned long count; | |
3637 | struct location *loc; | |
3638 | }; | |
3639 | ||
3640 | static void free_loc_track(struct loc_track *t) | |
3641 | { | |
3642 | if (t->max) | |
3643 | free_pages((unsigned long)t->loc, | |
3644 | get_order(sizeof(struct location) * t->max)); | |
3645 | } | |
3646 | ||
68dff6a9 | 3647 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3648 | { |
3649 | struct location *l; | |
3650 | int order; | |
3651 | ||
88a420e4 CL |
3652 | order = get_order(sizeof(struct location) * max); |
3653 | ||
68dff6a9 | 3654 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3655 | if (!l) |
3656 | return 0; | |
3657 | ||
3658 | if (t->count) { | |
3659 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3660 | free_loc_track(t); | |
3661 | } | |
3662 | t->max = max; | |
3663 | t->loc = l; | |
3664 | return 1; | |
3665 | } | |
3666 | ||
3667 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3668 | const struct track *track) |
88a420e4 CL |
3669 | { |
3670 | long start, end, pos; | |
3671 | struct location *l; | |
ce71e27c | 3672 | unsigned long caddr; |
45edfa58 | 3673 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3674 | |
3675 | start = -1; | |
3676 | end = t->count; | |
3677 | ||
3678 | for ( ; ; ) { | |
3679 | pos = start + (end - start + 1) / 2; | |
3680 | ||
3681 | /* | |
3682 | * There is nothing at "end". If we end up there | |
3683 | * we need to add something to before end. | |
3684 | */ | |
3685 | if (pos == end) | |
3686 | break; | |
3687 | ||
3688 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3689 | if (track->addr == caddr) { |
3690 | ||
3691 | l = &t->loc[pos]; | |
3692 | l->count++; | |
3693 | if (track->when) { | |
3694 | l->sum_time += age; | |
3695 | if (age < l->min_time) | |
3696 | l->min_time = age; | |
3697 | if (age > l->max_time) | |
3698 | l->max_time = age; | |
3699 | ||
3700 | if (track->pid < l->min_pid) | |
3701 | l->min_pid = track->pid; | |
3702 | if (track->pid > l->max_pid) | |
3703 | l->max_pid = track->pid; | |
3704 | ||
174596a0 RR |
3705 | cpumask_set_cpu(track->cpu, |
3706 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3707 | } |
3708 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3709 | return 1; |
3710 | } | |
3711 | ||
45edfa58 | 3712 | if (track->addr < caddr) |
88a420e4 CL |
3713 | end = pos; |
3714 | else | |
3715 | start = pos; | |
3716 | } | |
3717 | ||
3718 | /* | |
672bba3a | 3719 | * Not found. Insert new tracking element. |
88a420e4 | 3720 | */ |
68dff6a9 | 3721 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3722 | return 0; |
3723 | ||
3724 | l = t->loc + pos; | |
3725 | if (pos < t->count) | |
3726 | memmove(l + 1, l, | |
3727 | (t->count - pos) * sizeof(struct location)); | |
3728 | t->count++; | |
3729 | l->count = 1; | |
45edfa58 CL |
3730 | l->addr = track->addr; |
3731 | l->sum_time = age; | |
3732 | l->min_time = age; | |
3733 | l->max_time = age; | |
3734 | l->min_pid = track->pid; | |
3735 | l->max_pid = track->pid; | |
174596a0 RR |
3736 | cpumask_clear(to_cpumask(l->cpus)); |
3737 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3738 | nodes_clear(l->nodes); |
3739 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3740 | return 1; |
3741 | } | |
3742 | ||
3743 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3744 | struct page *page, enum track_item alloc) | |
3745 | { | |
a973e9dd | 3746 | void *addr = page_address(page); |
39b26464 | 3747 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3748 | void *p; |
3749 | ||
39b26464 | 3750 | bitmap_zero(map, page->objects); |
7656c72b CL |
3751 | for_each_free_object(p, s, page->freelist) |
3752 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3753 | |
224a88be | 3754 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3755 | if (!test_bit(slab_index(p, s, addr), map)) |
3756 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3757 | } |
3758 | ||
3759 | static int list_locations(struct kmem_cache *s, char *buf, | |
3760 | enum track_item alloc) | |
3761 | { | |
e374d483 | 3762 | int len = 0; |
88a420e4 | 3763 | unsigned long i; |
68dff6a9 | 3764 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3765 | int node; |
3766 | ||
68dff6a9 | 3767 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3768 | GFP_TEMPORARY)) |
68dff6a9 | 3769 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3770 | |
3771 | /* Push back cpu slabs */ | |
3772 | flush_all(s); | |
3773 | ||
f64dc58c | 3774 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3775 | struct kmem_cache_node *n = get_node(s, node); |
3776 | unsigned long flags; | |
3777 | struct page *page; | |
3778 | ||
9e86943b | 3779 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3780 | continue; |
3781 | ||
3782 | spin_lock_irqsave(&n->list_lock, flags); | |
3783 | list_for_each_entry(page, &n->partial, lru) | |
3784 | process_slab(&t, s, page, alloc); | |
3785 | list_for_each_entry(page, &n->full, lru) | |
3786 | process_slab(&t, s, page, alloc); | |
3787 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3788 | } | |
3789 | ||
3790 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3791 | struct location *l = &t.loc[i]; |
88a420e4 | 3792 | |
9c246247 | 3793 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3794 | break; |
e374d483 | 3795 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3796 | |
3797 | if (l->addr) | |
e374d483 | 3798 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3799 | else |
e374d483 | 3800 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3801 | |
3802 | if (l->sum_time != l->min_time) { | |
e374d483 | 3803 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3804 | l->min_time, |
3805 | (long)div_u64(l->sum_time, l->count), | |
3806 | l->max_time); | |
45edfa58 | 3807 | } else |
e374d483 | 3808 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3809 | l->min_time); |
3810 | ||
3811 | if (l->min_pid != l->max_pid) | |
e374d483 | 3812 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3813 | l->min_pid, l->max_pid); |
3814 | else | |
e374d483 | 3815 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3816 | l->min_pid); |
3817 | ||
174596a0 RR |
3818 | if (num_online_cpus() > 1 && |
3819 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3820 | len < PAGE_SIZE - 60) { |
3821 | len += sprintf(buf + len, " cpus="); | |
3822 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3823 | to_cpumask(l->cpus)); |
45edfa58 CL |
3824 | } |
3825 | ||
62bc62a8 | 3826 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3827 | len < PAGE_SIZE - 60) { |
3828 | len += sprintf(buf + len, " nodes="); | |
3829 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3830 | l->nodes); |
3831 | } | |
3832 | ||
e374d483 | 3833 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3834 | } |
3835 | ||
3836 | free_loc_track(&t); | |
3837 | if (!t.count) | |
e374d483 HH |
3838 | len += sprintf(buf, "No data\n"); |
3839 | return len; | |
88a420e4 CL |
3840 | } |
3841 | ||
81819f0f | 3842 | enum slab_stat_type { |
205ab99d CL |
3843 | SL_ALL, /* All slabs */ |
3844 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3845 | SL_CPU, /* Only slabs used for cpu caches */ | |
3846 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3847 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3848 | }; |
3849 | ||
205ab99d | 3850 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3851 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3852 | #define SO_CPU (1 << SL_CPU) | |
3853 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3854 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3855 | |
62e5c4b4 CG |
3856 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3857 | char *buf, unsigned long flags) | |
81819f0f CL |
3858 | { |
3859 | unsigned long total = 0; | |
81819f0f CL |
3860 | int node; |
3861 | int x; | |
3862 | unsigned long *nodes; | |
3863 | unsigned long *per_cpu; | |
3864 | ||
3865 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3866 | if (!nodes) |
3867 | return -ENOMEM; | |
81819f0f CL |
3868 | per_cpu = nodes + nr_node_ids; |
3869 | ||
205ab99d CL |
3870 | if (flags & SO_CPU) { |
3871 | int cpu; | |
81819f0f | 3872 | |
205ab99d CL |
3873 | for_each_possible_cpu(cpu) { |
3874 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3875 | |
205ab99d CL |
3876 | if (!c || c->node < 0) |
3877 | continue; | |
3878 | ||
3879 | if (c->page) { | |
3880 | if (flags & SO_TOTAL) | |
3881 | x = c->page->objects; | |
3882 | else if (flags & SO_OBJECTS) | |
3883 | x = c->page->inuse; | |
81819f0f CL |
3884 | else |
3885 | x = 1; | |
205ab99d | 3886 | |
81819f0f | 3887 | total += x; |
205ab99d | 3888 | nodes[c->node] += x; |
81819f0f | 3889 | } |
205ab99d | 3890 | per_cpu[c->node]++; |
81819f0f CL |
3891 | } |
3892 | } | |
3893 | ||
205ab99d CL |
3894 | if (flags & SO_ALL) { |
3895 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3896 | struct kmem_cache_node *n = get_node(s, node); | |
3897 | ||
3898 | if (flags & SO_TOTAL) | |
3899 | x = atomic_long_read(&n->total_objects); | |
3900 | else if (flags & SO_OBJECTS) | |
3901 | x = atomic_long_read(&n->total_objects) - | |
3902 | count_partial(n, count_free); | |
81819f0f | 3903 | |
81819f0f | 3904 | else |
205ab99d | 3905 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3906 | total += x; |
3907 | nodes[node] += x; | |
3908 | } | |
3909 | ||
205ab99d CL |
3910 | } else if (flags & SO_PARTIAL) { |
3911 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3912 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3913 | |
205ab99d CL |
3914 | if (flags & SO_TOTAL) |
3915 | x = count_partial(n, count_total); | |
3916 | else if (flags & SO_OBJECTS) | |
3917 | x = count_partial(n, count_inuse); | |
81819f0f | 3918 | else |
205ab99d | 3919 | x = n->nr_partial; |
81819f0f CL |
3920 | total += x; |
3921 | nodes[node] += x; | |
3922 | } | |
3923 | } | |
81819f0f CL |
3924 | x = sprintf(buf, "%lu", total); |
3925 | #ifdef CONFIG_NUMA | |
f64dc58c | 3926 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3927 | if (nodes[node]) |
3928 | x += sprintf(buf + x, " N%d=%lu", | |
3929 | node, nodes[node]); | |
3930 | #endif | |
3931 | kfree(nodes); | |
3932 | return x + sprintf(buf + x, "\n"); | |
3933 | } | |
3934 | ||
3935 | static int any_slab_objects(struct kmem_cache *s) | |
3936 | { | |
3937 | int node; | |
81819f0f | 3938 | |
dfb4f096 | 3939 | for_each_online_node(node) { |
81819f0f CL |
3940 | struct kmem_cache_node *n = get_node(s, node); |
3941 | ||
dfb4f096 CL |
3942 | if (!n) |
3943 | continue; | |
3944 | ||
4ea33e2d | 3945 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3946 | return 1; |
3947 | } | |
3948 | return 0; | |
3949 | } | |
3950 | ||
3951 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3952 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3953 | ||
3954 | struct slab_attribute { | |
3955 | struct attribute attr; | |
3956 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3957 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3958 | }; | |
3959 | ||
3960 | #define SLAB_ATTR_RO(_name) \ | |
3961 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3962 | ||
3963 | #define SLAB_ATTR(_name) \ | |
3964 | static struct slab_attribute _name##_attr = \ | |
3965 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3966 | ||
81819f0f CL |
3967 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3968 | { | |
3969 | return sprintf(buf, "%d\n", s->size); | |
3970 | } | |
3971 | SLAB_ATTR_RO(slab_size); | |
3972 | ||
3973 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3974 | { | |
3975 | return sprintf(buf, "%d\n", s->align); | |
3976 | } | |
3977 | SLAB_ATTR_RO(align); | |
3978 | ||
3979 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3980 | { | |
3981 | return sprintf(buf, "%d\n", s->objsize); | |
3982 | } | |
3983 | SLAB_ATTR_RO(object_size); | |
3984 | ||
3985 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3986 | { | |
834f3d11 | 3987 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3988 | } |
3989 | SLAB_ATTR_RO(objs_per_slab); | |
3990 | ||
06b285dc CL |
3991 | static ssize_t order_store(struct kmem_cache *s, |
3992 | const char *buf, size_t length) | |
3993 | { | |
0121c619 CL |
3994 | unsigned long order; |
3995 | int err; | |
3996 | ||
3997 | err = strict_strtoul(buf, 10, &order); | |
3998 | if (err) | |
3999 | return err; | |
06b285dc CL |
4000 | |
4001 | if (order > slub_max_order || order < slub_min_order) | |
4002 | return -EINVAL; | |
4003 | ||
4004 | calculate_sizes(s, order); | |
4005 | return length; | |
4006 | } | |
4007 | ||
81819f0f CL |
4008 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4009 | { | |
834f3d11 | 4010 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4011 | } |
06b285dc | 4012 | SLAB_ATTR(order); |
81819f0f | 4013 | |
73d342b1 DR |
4014 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4015 | { | |
4016 | return sprintf(buf, "%lu\n", s->min_partial); | |
4017 | } | |
4018 | ||
4019 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4020 | size_t length) | |
4021 | { | |
4022 | unsigned long min; | |
4023 | int err; | |
4024 | ||
4025 | err = strict_strtoul(buf, 10, &min); | |
4026 | if (err) | |
4027 | return err; | |
4028 | ||
c0bdb232 | 4029 | set_min_partial(s, min); |
73d342b1 DR |
4030 | return length; |
4031 | } | |
4032 | SLAB_ATTR(min_partial); | |
4033 | ||
81819f0f CL |
4034 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4035 | { | |
4036 | if (s->ctor) { | |
4037 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
4038 | ||
4039 | return n + sprintf(buf + n, "\n"); | |
4040 | } | |
4041 | return 0; | |
4042 | } | |
4043 | SLAB_ATTR_RO(ctor); | |
4044 | ||
81819f0f CL |
4045 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4046 | { | |
4047 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4048 | } | |
4049 | SLAB_ATTR_RO(aliases); | |
4050 | ||
4051 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
4052 | { | |
205ab99d | 4053 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
4054 | } |
4055 | SLAB_ATTR_RO(slabs); | |
4056 | ||
4057 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
4058 | { | |
d9acf4b7 | 4059 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4060 | } |
4061 | SLAB_ATTR_RO(partial); | |
4062 | ||
4063 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4064 | { | |
d9acf4b7 | 4065 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4066 | } |
4067 | SLAB_ATTR_RO(cpu_slabs); | |
4068 | ||
4069 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4070 | { | |
205ab99d | 4071 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4072 | } |
4073 | SLAB_ATTR_RO(objects); | |
4074 | ||
205ab99d CL |
4075 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4076 | { | |
4077 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4078 | } | |
4079 | SLAB_ATTR_RO(objects_partial); | |
4080 | ||
4081 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4082 | { | |
4083 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4084 | } | |
4085 | SLAB_ATTR_RO(total_objects); | |
4086 | ||
81819f0f CL |
4087 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4088 | { | |
4089 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4090 | } | |
4091 | ||
4092 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4093 | const char *buf, size_t length) | |
4094 | { | |
4095 | s->flags &= ~SLAB_DEBUG_FREE; | |
4096 | if (buf[0] == '1') | |
4097 | s->flags |= SLAB_DEBUG_FREE; | |
4098 | return length; | |
4099 | } | |
4100 | SLAB_ATTR(sanity_checks); | |
4101 | ||
4102 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4103 | { | |
4104 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4105 | } | |
4106 | ||
4107 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4108 | size_t length) | |
4109 | { | |
4110 | s->flags &= ~SLAB_TRACE; | |
4111 | if (buf[0] == '1') | |
4112 | s->flags |= SLAB_TRACE; | |
4113 | return length; | |
4114 | } | |
4115 | SLAB_ATTR(trace); | |
4116 | ||
4117 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
4118 | { | |
4119 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4120 | } | |
4121 | ||
4122 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4123 | const char *buf, size_t length) | |
4124 | { | |
4125 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4126 | if (buf[0] == '1') | |
4127 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4128 | return length; | |
4129 | } | |
4130 | SLAB_ATTR(reclaim_account); | |
4131 | ||
4132 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4133 | { | |
5af60839 | 4134 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4135 | } |
4136 | SLAB_ATTR_RO(hwcache_align); | |
4137 | ||
4138 | #ifdef CONFIG_ZONE_DMA | |
4139 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4140 | { | |
4141 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4142 | } | |
4143 | SLAB_ATTR_RO(cache_dma); | |
4144 | #endif | |
4145 | ||
4146 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4147 | { | |
4148 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4149 | } | |
4150 | SLAB_ATTR_RO(destroy_by_rcu); | |
4151 | ||
4152 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4153 | { | |
4154 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4155 | } | |
4156 | ||
4157 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4158 | const char *buf, size_t length) | |
4159 | { | |
4160 | if (any_slab_objects(s)) | |
4161 | return -EBUSY; | |
4162 | ||
4163 | s->flags &= ~SLAB_RED_ZONE; | |
4164 | if (buf[0] == '1') | |
4165 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4166 | calculate_sizes(s, -1); |
81819f0f CL |
4167 | return length; |
4168 | } | |
4169 | SLAB_ATTR(red_zone); | |
4170 | ||
4171 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4172 | { | |
4173 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4174 | } | |
4175 | ||
4176 | static ssize_t poison_store(struct kmem_cache *s, | |
4177 | const char *buf, size_t length) | |
4178 | { | |
4179 | if (any_slab_objects(s)) | |
4180 | return -EBUSY; | |
4181 | ||
4182 | s->flags &= ~SLAB_POISON; | |
4183 | if (buf[0] == '1') | |
4184 | s->flags |= SLAB_POISON; | |
06b285dc | 4185 | calculate_sizes(s, -1); |
81819f0f CL |
4186 | return length; |
4187 | } | |
4188 | SLAB_ATTR(poison); | |
4189 | ||
4190 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4191 | { | |
4192 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4193 | } | |
4194 | ||
4195 | static ssize_t store_user_store(struct kmem_cache *s, | |
4196 | const char *buf, size_t length) | |
4197 | { | |
4198 | if (any_slab_objects(s)) | |
4199 | return -EBUSY; | |
4200 | ||
4201 | s->flags &= ~SLAB_STORE_USER; | |
4202 | if (buf[0] == '1') | |
4203 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4204 | calculate_sizes(s, -1); |
81819f0f CL |
4205 | return length; |
4206 | } | |
4207 | SLAB_ATTR(store_user); | |
4208 | ||
53e15af0 CL |
4209 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4210 | { | |
4211 | return 0; | |
4212 | } | |
4213 | ||
4214 | static ssize_t validate_store(struct kmem_cache *s, | |
4215 | const char *buf, size_t length) | |
4216 | { | |
434e245d CL |
4217 | int ret = -EINVAL; |
4218 | ||
4219 | if (buf[0] == '1') { | |
4220 | ret = validate_slab_cache(s); | |
4221 | if (ret >= 0) | |
4222 | ret = length; | |
4223 | } | |
4224 | return ret; | |
53e15af0 CL |
4225 | } |
4226 | SLAB_ATTR(validate); | |
4227 | ||
2086d26a CL |
4228 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4229 | { | |
4230 | return 0; | |
4231 | } | |
4232 | ||
4233 | static ssize_t shrink_store(struct kmem_cache *s, | |
4234 | const char *buf, size_t length) | |
4235 | { | |
4236 | if (buf[0] == '1') { | |
4237 | int rc = kmem_cache_shrink(s); | |
4238 | ||
4239 | if (rc) | |
4240 | return rc; | |
4241 | } else | |
4242 | return -EINVAL; | |
4243 | return length; | |
4244 | } | |
4245 | SLAB_ATTR(shrink); | |
4246 | ||
88a420e4 CL |
4247 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4248 | { | |
4249 | if (!(s->flags & SLAB_STORE_USER)) | |
4250 | return -ENOSYS; | |
4251 | return list_locations(s, buf, TRACK_ALLOC); | |
4252 | } | |
4253 | SLAB_ATTR_RO(alloc_calls); | |
4254 | ||
4255 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4256 | { | |
4257 | if (!(s->flags & SLAB_STORE_USER)) | |
4258 | return -ENOSYS; | |
4259 | return list_locations(s, buf, TRACK_FREE); | |
4260 | } | |
4261 | SLAB_ATTR_RO(free_calls); | |
4262 | ||
81819f0f | 4263 | #ifdef CONFIG_NUMA |
9824601e | 4264 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4265 | { |
9824601e | 4266 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4267 | } |
4268 | ||
9824601e | 4269 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4270 | const char *buf, size_t length) |
4271 | { | |
0121c619 CL |
4272 | unsigned long ratio; |
4273 | int err; | |
4274 | ||
4275 | err = strict_strtoul(buf, 10, &ratio); | |
4276 | if (err) | |
4277 | return err; | |
4278 | ||
e2cb96b7 | 4279 | if (ratio <= 100) |
0121c619 | 4280 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4281 | |
81819f0f CL |
4282 | return length; |
4283 | } | |
9824601e | 4284 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4285 | #endif |
4286 | ||
8ff12cfc | 4287 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4288 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4289 | { | |
4290 | unsigned long sum = 0; | |
4291 | int cpu; | |
4292 | int len; | |
4293 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4294 | ||
4295 | if (!data) | |
4296 | return -ENOMEM; | |
4297 | ||
4298 | for_each_online_cpu(cpu) { | |
4299 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4300 | ||
4301 | data[cpu] = x; | |
4302 | sum += x; | |
4303 | } | |
4304 | ||
4305 | len = sprintf(buf, "%lu", sum); | |
4306 | ||
50ef37b9 | 4307 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4308 | for_each_online_cpu(cpu) { |
4309 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4310 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4311 | } |
50ef37b9 | 4312 | #endif |
8ff12cfc CL |
4313 | kfree(data); |
4314 | return len + sprintf(buf + len, "\n"); | |
4315 | } | |
4316 | ||
4317 | #define STAT_ATTR(si, text) \ | |
4318 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4319 | { \ | |
4320 | return show_stat(s, buf, si); \ | |
4321 | } \ | |
4322 | SLAB_ATTR_RO(text); \ | |
4323 | ||
4324 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4325 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4326 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4327 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4328 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4329 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4330 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4331 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4332 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4333 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4334 | STAT_ATTR(FREE_SLAB, free_slab); | |
4335 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4336 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4337 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4338 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4339 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4340 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4341 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4342 | #endif |
4343 | ||
06428780 | 4344 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4345 | &slab_size_attr.attr, |
4346 | &object_size_attr.attr, | |
4347 | &objs_per_slab_attr.attr, | |
4348 | &order_attr.attr, | |
73d342b1 | 4349 | &min_partial_attr.attr, |
81819f0f | 4350 | &objects_attr.attr, |
205ab99d CL |
4351 | &objects_partial_attr.attr, |
4352 | &total_objects_attr.attr, | |
81819f0f CL |
4353 | &slabs_attr.attr, |
4354 | &partial_attr.attr, | |
4355 | &cpu_slabs_attr.attr, | |
4356 | &ctor_attr.attr, | |
81819f0f CL |
4357 | &aliases_attr.attr, |
4358 | &align_attr.attr, | |
4359 | &sanity_checks_attr.attr, | |
4360 | &trace_attr.attr, | |
4361 | &hwcache_align_attr.attr, | |
4362 | &reclaim_account_attr.attr, | |
4363 | &destroy_by_rcu_attr.attr, | |
4364 | &red_zone_attr.attr, | |
4365 | &poison_attr.attr, | |
4366 | &store_user_attr.attr, | |
53e15af0 | 4367 | &validate_attr.attr, |
2086d26a | 4368 | &shrink_attr.attr, |
88a420e4 CL |
4369 | &alloc_calls_attr.attr, |
4370 | &free_calls_attr.attr, | |
81819f0f CL |
4371 | #ifdef CONFIG_ZONE_DMA |
4372 | &cache_dma_attr.attr, | |
4373 | #endif | |
4374 | #ifdef CONFIG_NUMA | |
9824601e | 4375 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4376 | #endif |
4377 | #ifdef CONFIG_SLUB_STATS | |
4378 | &alloc_fastpath_attr.attr, | |
4379 | &alloc_slowpath_attr.attr, | |
4380 | &free_fastpath_attr.attr, | |
4381 | &free_slowpath_attr.attr, | |
4382 | &free_frozen_attr.attr, | |
4383 | &free_add_partial_attr.attr, | |
4384 | &free_remove_partial_attr.attr, | |
4385 | &alloc_from_partial_attr.attr, | |
4386 | &alloc_slab_attr.attr, | |
4387 | &alloc_refill_attr.attr, | |
4388 | &free_slab_attr.attr, | |
4389 | &cpuslab_flush_attr.attr, | |
4390 | &deactivate_full_attr.attr, | |
4391 | &deactivate_empty_attr.attr, | |
4392 | &deactivate_to_head_attr.attr, | |
4393 | &deactivate_to_tail_attr.attr, | |
4394 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4395 | &order_fallback_attr.attr, |
81819f0f CL |
4396 | #endif |
4397 | NULL | |
4398 | }; | |
4399 | ||
4400 | static struct attribute_group slab_attr_group = { | |
4401 | .attrs = slab_attrs, | |
4402 | }; | |
4403 | ||
4404 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4405 | struct attribute *attr, | |
4406 | char *buf) | |
4407 | { | |
4408 | struct slab_attribute *attribute; | |
4409 | struct kmem_cache *s; | |
4410 | int err; | |
4411 | ||
4412 | attribute = to_slab_attr(attr); | |
4413 | s = to_slab(kobj); | |
4414 | ||
4415 | if (!attribute->show) | |
4416 | return -EIO; | |
4417 | ||
4418 | err = attribute->show(s, buf); | |
4419 | ||
4420 | return err; | |
4421 | } | |
4422 | ||
4423 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4424 | struct attribute *attr, | |
4425 | const char *buf, size_t len) | |
4426 | { | |
4427 | struct slab_attribute *attribute; | |
4428 | struct kmem_cache *s; | |
4429 | int err; | |
4430 | ||
4431 | attribute = to_slab_attr(attr); | |
4432 | s = to_slab(kobj); | |
4433 | ||
4434 | if (!attribute->store) | |
4435 | return -EIO; | |
4436 | ||
4437 | err = attribute->store(s, buf, len); | |
4438 | ||
4439 | return err; | |
4440 | } | |
4441 | ||
151c602f CL |
4442 | static void kmem_cache_release(struct kobject *kobj) |
4443 | { | |
4444 | struct kmem_cache *s = to_slab(kobj); | |
4445 | ||
4446 | kfree(s); | |
4447 | } | |
4448 | ||
81819f0f CL |
4449 | static struct sysfs_ops slab_sysfs_ops = { |
4450 | .show = slab_attr_show, | |
4451 | .store = slab_attr_store, | |
4452 | }; | |
4453 | ||
4454 | static struct kobj_type slab_ktype = { | |
4455 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4456 | .release = kmem_cache_release |
81819f0f CL |
4457 | }; |
4458 | ||
4459 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4460 | { | |
4461 | struct kobj_type *ktype = get_ktype(kobj); | |
4462 | ||
4463 | if (ktype == &slab_ktype) | |
4464 | return 1; | |
4465 | return 0; | |
4466 | } | |
4467 | ||
4468 | static struct kset_uevent_ops slab_uevent_ops = { | |
4469 | .filter = uevent_filter, | |
4470 | }; | |
4471 | ||
27c3a314 | 4472 | static struct kset *slab_kset; |
81819f0f CL |
4473 | |
4474 | #define ID_STR_LENGTH 64 | |
4475 | ||
4476 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4477 | * |
4478 | * Format :[flags-]size | |
81819f0f CL |
4479 | */ |
4480 | static char *create_unique_id(struct kmem_cache *s) | |
4481 | { | |
4482 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4483 | char *p = name; | |
4484 | ||
4485 | BUG_ON(!name); | |
4486 | ||
4487 | *p++ = ':'; | |
4488 | /* | |
4489 | * First flags affecting slabcache operations. We will only | |
4490 | * get here for aliasable slabs so we do not need to support | |
4491 | * too many flags. The flags here must cover all flags that | |
4492 | * are matched during merging to guarantee that the id is | |
4493 | * unique. | |
4494 | */ | |
4495 | if (s->flags & SLAB_CACHE_DMA) | |
4496 | *p++ = 'd'; | |
4497 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4498 | *p++ = 'a'; | |
4499 | if (s->flags & SLAB_DEBUG_FREE) | |
4500 | *p++ = 'F'; | |
5a896d9e VN |
4501 | if (!(s->flags & SLAB_NOTRACK)) |
4502 | *p++ = 't'; | |
81819f0f CL |
4503 | if (p != name + 1) |
4504 | *p++ = '-'; | |
4505 | p += sprintf(p, "%07d", s->size); | |
4506 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4507 | return name; | |
4508 | } | |
4509 | ||
4510 | static int sysfs_slab_add(struct kmem_cache *s) | |
4511 | { | |
4512 | int err; | |
4513 | const char *name; | |
4514 | int unmergeable; | |
4515 | ||
4516 | if (slab_state < SYSFS) | |
4517 | /* Defer until later */ | |
4518 | return 0; | |
4519 | ||
4520 | unmergeable = slab_unmergeable(s); | |
4521 | if (unmergeable) { | |
4522 | /* | |
4523 | * Slabcache can never be merged so we can use the name proper. | |
4524 | * This is typically the case for debug situations. In that | |
4525 | * case we can catch duplicate names easily. | |
4526 | */ | |
27c3a314 | 4527 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4528 | name = s->name; |
4529 | } else { | |
4530 | /* | |
4531 | * Create a unique name for the slab as a target | |
4532 | * for the symlinks. | |
4533 | */ | |
4534 | name = create_unique_id(s); | |
4535 | } | |
4536 | ||
27c3a314 | 4537 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4538 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4539 | if (err) { | |
4540 | kobject_put(&s->kobj); | |
81819f0f | 4541 | return err; |
1eada11c | 4542 | } |
81819f0f CL |
4543 | |
4544 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4545 | if (err) | |
4546 | return err; | |
4547 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4548 | if (!unmergeable) { | |
4549 | /* Setup first alias */ | |
4550 | sysfs_slab_alias(s, s->name); | |
4551 | kfree(name); | |
4552 | } | |
4553 | return 0; | |
4554 | } | |
4555 | ||
4556 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4557 | { | |
4558 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4559 | kobject_del(&s->kobj); | |
151c602f | 4560 | kobject_put(&s->kobj); |
81819f0f CL |
4561 | } |
4562 | ||
4563 | /* | |
4564 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4565 | * available lest we lose that information. |
81819f0f CL |
4566 | */ |
4567 | struct saved_alias { | |
4568 | struct kmem_cache *s; | |
4569 | const char *name; | |
4570 | struct saved_alias *next; | |
4571 | }; | |
4572 | ||
5af328a5 | 4573 | static struct saved_alias *alias_list; |
81819f0f CL |
4574 | |
4575 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4576 | { | |
4577 | struct saved_alias *al; | |
4578 | ||
4579 | if (slab_state == SYSFS) { | |
4580 | /* | |
4581 | * If we have a leftover link then remove it. | |
4582 | */ | |
27c3a314 GKH |
4583 | sysfs_remove_link(&slab_kset->kobj, name); |
4584 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4585 | } |
4586 | ||
4587 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4588 | if (!al) | |
4589 | return -ENOMEM; | |
4590 | ||
4591 | al->s = s; | |
4592 | al->name = name; | |
4593 | al->next = alias_list; | |
4594 | alias_list = al; | |
4595 | return 0; | |
4596 | } | |
4597 | ||
4598 | static int __init slab_sysfs_init(void) | |
4599 | { | |
5b95a4ac | 4600 | struct kmem_cache *s; |
81819f0f CL |
4601 | int err; |
4602 | ||
0ff21e46 | 4603 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4604 | if (!slab_kset) { |
81819f0f CL |
4605 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4606 | return -ENOSYS; | |
4607 | } | |
4608 | ||
26a7bd03 CL |
4609 | slab_state = SYSFS; |
4610 | ||
5b95a4ac | 4611 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4612 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4613 | if (err) |
4614 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4615 | " to sysfs\n", s->name); | |
26a7bd03 | 4616 | } |
81819f0f CL |
4617 | |
4618 | while (alias_list) { | |
4619 | struct saved_alias *al = alias_list; | |
4620 | ||
4621 | alias_list = alias_list->next; | |
4622 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4623 | if (err) |
4624 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4625 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4626 | kfree(al); |
4627 | } | |
4628 | ||
4629 | resiliency_test(); | |
4630 | return 0; | |
4631 | } | |
4632 | ||
4633 | __initcall(slab_sysfs_init); | |
81819f0f | 4634 | #endif |
57ed3eda PE |
4635 | |
4636 | /* | |
4637 | * The /proc/slabinfo ABI | |
4638 | */ | |
158a9624 | 4639 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4640 | static void print_slabinfo_header(struct seq_file *m) |
4641 | { | |
4642 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4643 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4644 | "<objperslab> <pagesperslab>"); | |
4645 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4646 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4647 | seq_putc(m, '\n'); | |
4648 | } | |
4649 | ||
4650 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4651 | { | |
4652 | loff_t n = *pos; | |
4653 | ||
4654 | down_read(&slub_lock); | |
4655 | if (!n) | |
4656 | print_slabinfo_header(m); | |
4657 | ||
4658 | return seq_list_start(&slab_caches, *pos); | |
4659 | } | |
4660 | ||
4661 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4662 | { | |
4663 | return seq_list_next(p, &slab_caches, pos); | |
4664 | } | |
4665 | ||
4666 | static void s_stop(struct seq_file *m, void *p) | |
4667 | { | |
4668 | up_read(&slub_lock); | |
4669 | } | |
4670 | ||
4671 | static int s_show(struct seq_file *m, void *p) | |
4672 | { | |
4673 | unsigned long nr_partials = 0; | |
4674 | unsigned long nr_slabs = 0; | |
4675 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4676 | unsigned long nr_objs = 0; |
4677 | unsigned long nr_free = 0; | |
57ed3eda PE |
4678 | struct kmem_cache *s; |
4679 | int node; | |
4680 | ||
4681 | s = list_entry(p, struct kmem_cache, list); | |
4682 | ||
4683 | for_each_online_node(node) { | |
4684 | struct kmem_cache_node *n = get_node(s, node); | |
4685 | ||
4686 | if (!n) | |
4687 | continue; | |
4688 | ||
4689 | nr_partials += n->nr_partial; | |
4690 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4691 | nr_objs += atomic_long_read(&n->total_objects); |
4692 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4693 | } |
4694 | ||
205ab99d | 4695 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4696 | |
4697 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4698 | nr_objs, s->size, oo_objects(s->oo), |
4699 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4700 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4701 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4702 | 0UL); | |
4703 | seq_putc(m, '\n'); | |
4704 | return 0; | |
4705 | } | |
4706 | ||
7b3c3a50 | 4707 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4708 | .start = s_start, |
4709 | .next = s_next, | |
4710 | .stop = s_stop, | |
4711 | .show = s_show, | |
4712 | }; | |
4713 | ||
7b3c3a50 AD |
4714 | static int slabinfo_open(struct inode *inode, struct file *file) |
4715 | { | |
4716 | return seq_open(file, &slabinfo_op); | |
4717 | } | |
4718 | ||
4719 | static const struct file_operations proc_slabinfo_operations = { | |
4720 | .open = slabinfo_open, | |
4721 | .read = seq_read, | |
4722 | .llseek = seq_lseek, | |
4723 | .release = seq_release, | |
4724 | }; | |
4725 | ||
4726 | static int __init slab_proc_init(void) | |
4727 | { | |
4728 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); | |
4729 | return 0; | |
4730 | } | |
4731 | module_init(slab_proc_init); | |
158a9624 | 4732 | #endif /* CONFIG_SLABINFO */ |