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