]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
9 | */ |
10 | ||
11 | #include <linux/mm.h> | |
1eb5ac64 | 12 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
13 | #include <linux/module.h> |
14 | #include <linux/bit_spinlock.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/bitops.h> | |
17 | #include <linux/slab.h> | |
7b3c3a50 | 18 | #include <linux/proc_fs.h> |
81819f0f | 19 | #include <linux/seq_file.h> |
5a896d9e | 20 | #include <linux/kmemcheck.h> |
81819f0f CL |
21 | #include <linux/cpu.h> |
22 | #include <linux/cpuset.h> | |
23 | #include <linux/mempolicy.h> | |
24 | #include <linux/ctype.h> | |
3ac7fe5a | 25 | #include <linux/debugobjects.h> |
81819f0f | 26 | #include <linux/kallsyms.h> |
b9049e23 | 27 | #include <linux/memory.h> |
f8bd2258 | 28 | #include <linux/math64.h> |
773ff60e | 29 | #include <linux/fault-inject.h> |
81819f0f CL |
30 | |
31 | /* | |
32 | * Lock order: | |
33 | * 1. slab_lock(page) | |
34 | * 2. slab->list_lock | |
35 | * | |
36 | * The slab_lock protects operations on the object of a particular | |
37 | * slab and its metadata in the page struct. If the slab lock | |
38 | * has been taken then no allocations nor frees can be performed | |
39 | * on the objects in the slab nor can the slab be added or removed | |
40 | * from the partial or full lists since this would mean modifying | |
41 | * the page_struct of the slab. | |
42 | * | |
43 | * The list_lock protects the partial and full list on each node and | |
44 | * the partial slab counter. If taken then no new slabs may be added or | |
45 | * removed from the lists nor make the number of partial slabs be modified. | |
46 | * (Note that the total number of slabs is an atomic value that may be | |
47 | * modified without taking the list lock). | |
48 | * | |
49 | * The list_lock is a centralized lock and thus we avoid taking it as | |
50 | * much as possible. As long as SLUB does not have to handle partial | |
51 | * slabs, operations can continue without any centralized lock. F.e. | |
52 | * allocating a long series of objects that fill up slabs does not require | |
53 | * the list lock. | |
54 | * | |
55 | * The lock order is sometimes inverted when we are trying to get a slab | |
56 | * off a list. We take the list_lock and then look for a page on the list | |
57 | * to use. While we do that objects in the slabs may be freed. We can | |
58 | * only operate on the slab if we have also taken the slab_lock. So we use | |
59 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
60 | * can occur anymore and we can use the slab for allocations etc. If the | |
61 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
62 | * we must stay away from it for a while since we may cause a bouncing | |
63 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
64 | * If all pages are busy then we may allocate a new slab instead of reusing | |
65 | * a partial slab. A new slab has noone operating on it and thus there is | |
66 | * no danger of cacheline contention. | |
67 | * | |
68 | * Interrupts are disabled during allocation and deallocation in order to | |
69 | * make the slab allocator safe to use in the context of an irq. In addition | |
70 | * interrupts are disabled to ensure that the processor does not change | |
71 | * while handling per_cpu slabs, due to kernel preemption. | |
72 | * | |
73 | * SLUB assigns one slab for allocation to each processor. | |
74 | * Allocations only occur from these slabs called cpu slabs. | |
75 | * | |
672bba3a CL |
76 | * Slabs with free elements are kept on a partial list and during regular |
77 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 78 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
79 | * We track full slabs for debugging purposes though because otherwise we |
80 | * cannot scan all objects. | |
81819f0f CL |
81 | * |
82 | * Slabs are freed when they become empty. Teardown and setup is | |
83 | * minimal so we rely on the page allocators per cpu caches for | |
84 | * fast frees and allocs. | |
85 | * | |
86 | * Overloading of page flags that are otherwise used for LRU management. | |
87 | * | |
4b6f0750 CL |
88 | * PageActive The slab is frozen and exempt from list processing. |
89 | * This means that the slab is dedicated to a purpose | |
90 | * such as satisfying allocations for a specific | |
91 | * processor. Objects may be freed in the slab while | |
92 | * it is frozen but slab_free will then skip the usual | |
93 | * list operations. It is up to the processor holding | |
94 | * the slab to integrate the slab into the slab lists | |
95 | * when the slab is no longer needed. | |
96 | * | |
97 | * One use of this flag is to mark slabs that are | |
98 | * used for allocations. Then such a slab becomes a cpu | |
99 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 100 | * freelist that allows lockless access to |
894b8788 CL |
101 | * free objects in addition to the regular freelist |
102 | * that requires the slab lock. | |
81819f0f CL |
103 | * |
104 | * PageError Slab requires special handling due to debug | |
105 | * options set. This moves slab handling out of | |
894b8788 | 106 | * the fast path and disables lockless freelists. |
81819f0f CL |
107 | */ |
108 | ||
af537b0a CL |
109 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
110 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
111 | ||
112 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
113 | { | |
5577bd8a | 114 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 115 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 116 | #else |
af537b0a | 117 | return 0; |
5577bd8a | 118 | #endif |
af537b0a | 119 | } |
5577bd8a | 120 | |
81819f0f CL |
121 | /* |
122 | * Issues still to be resolved: | |
123 | * | |
81819f0f CL |
124 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
125 | * | |
81819f0f CL |
126 | * - Variable sizing of the per node arrays |
127 | */ | |
128 | ||
129 | /* Enable to test recovery from slab corruption on boot */ | |
130 | #undef SLUB_RESILIENCY_TEST | |
131 | ||
2086d26a CL |
132 | /* |
133 | * Mininum number of partial slabs. These will be left on the partial | |
134 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
135 | */ | |
76be8950 | 136 | #define MIN_PARTIAL 5 |
e95eed57 | 137 | |
2086d26a CL |
138 | /* |
139 | * Maximum number of desirable partial slabs. | |
140 | * The existence of more partial slabs makes kmem_cache_shrink | |
141 | * sort the partial list by the number of objects in the. | |
142 | */ | |
143 | #define MAX_PARTIAL 10 | |
144 | ||
81819f0f CL |
145 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
146 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 147 | |
fa5ec8a1 | 148 | /* |
3de47213 DR |
149 | * Debugging flags that require metadata to be stored in the slab. These get |
150 | * disabled when slub_debug=O is used and a cache's min order increases with | |
151 | * metadata. | |
fa5ec8a1 | 152 | */ |
3de47213 | 153 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 154 | |
81819f0f CL |
155 | /* |
156 | * Set of flags that will prevent slab merging | |
157 | */ | |
158 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
159 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
160 | SLAB_FAILSLAB) | |
81819f0f CL |
161 | |
162 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 163 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 164 | |
210b5c06 CG |
165 | #define OO_SHIFT 16 |
166 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
167 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
168 | ||
81819f0f | 169 | /* Internal SLUB flags */ |
f90ec390 | 170 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
81819f0f CL |
171 | |
172 | static int kmem_size = sizeof(struct kmem_cache); | |
173 | ||
174 | #ifdef CONFIG_SMP | |
175 | static struct notifier_block slab_notifier; | |
176 | #endif | |
177 | ||
178 | static enum { | |
179 | DOWN, /* No slab functionality available */ | |
51df1142 | 180 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 181 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
182 | SYSFS /* Sysfs up */ |
183 | } slab_state = DOWN; | |
184 | ||
185 | /* A list of all slab caches on the system */ | |
186 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 187 | static LIST_HEAD(slab_caches); |
81819f0f | 188 | |
02cbc874 CL |
189 | /* |
190 | * Tracking user of a slab. | |
191 | */ | |
192 | struct track { | |
ce71e27c | 193 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
194 | int cpu; /* Was running on cpu */ |
195 | int pid; /* Pid context */ | |
196 | unsigned long when; /* When did the operation occur */ | |
197 | }; | |
198 | ||
199 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
200 | ||
f6acb635 | 201 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
202 | static int sysfs_slab_add(struct kmem_cache *); |
203 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
204 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 205 | |
81819f0f | 206 | #else |
0c710013 CL |
207 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
208 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
209 | { return 0; } | |
151c602f CL |
210 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
211 | { | |
212 | kfree(s); | |
213 | } | |
8ff12cfc | 214 | |
81819f0f CL |
215 | #endif |
216 | ||
84e554e6 | 217 | static inline void stat(struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
218 | { |
219 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 220 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
221 | #endif |
222 | } | |
223 | ||
81819f0f CL |
224 | /******************************************************************** |
225 | * Core slab cache functions | |
226 | *******************************************************************/ | |
227 | ||
228 | int slab_is_available(void) | |
229 | { | |
230 | return slab_state >= UP; | |
231 | } | |
232 | ||
233 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
234 | { | |
235 | #ifdef CONFIG_NUMA | |
236 | return s->node[node]; | |
237 | #else | |
238 | return &s->local_node; | |
239 | #endif | |
240 | } | |
241 | ||
6446faa2 | 242 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
243 | static inline int check_valid_pointer(struct kmem_cache *s, |
244 | struct page *page, const void *object) | |
245 | { | |
246 | void *base; | |
247 | ||
a973e9dd | 248 | if (!object) |
02cbc874 CL |
249 | return 1; |
250 | ||
a973e9dd | 251 | base = page_address(page); |
39b26464 | 252 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
253 | (object - base) % s->size) { |
254 | return 0; | |
255 | } | |
256 | ||
257 | return 1; | |
258 | } | |
259 | ||
7656c72b CL |
260 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
261 | { | |
262 | return *(void **)(object + s->offset); | |
263 | } | |
264 | ||
265 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
266 | { | |
267 | *(void **)(object + s->offset) = fp; | |
268 | } | |
269 | ||
270 | /* Loop over all objects in a slab */ | |
224a88be CL |
271 | #define for_each_object(__p, __s, __addr, __objects) \ |
272 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
273 | __p += (__s)->size) |
274 | ||
275 | /* Scan freelist */ | |
276 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 277 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
278 | |
279 | /* Determine object index from a given position */ | |
280 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
281 | { | |
282 | return (p - addr) / s->size; | |
283 | } | |
284 | ||
834f3d11 CL |
285 | static inline struct kmem_cache_order_objects oo_make(int order, |
286 | unsigned long size) | |
287 | { | |
288 | struct kmem_cache_order_objects x = { | |
210b5c06 | 289 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
290 | }; |
291 | ||
292 | return x; | |
293 | } | |
294 | ||
295 | static inline int oo_order(struct kmem_cache_order_objects x) | |
296 | { | |
210b5c06 | 297 | return x.x >> OO_SHIFT; |
834f3d11 CL |
298 | } |
299 | ||
300 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
301 | { | |
210b5c06 | 302 | return x.x & OO_MASK; |
834f3d11 CL |
303 | } |
304 | ||
41ecc55b CL |
305 | #ifdef CONFIG_SLUB_DEBUG |
306 | /* | |
307 | * Debug settings: | |
308 | */ | |
f0630fff CL |
309 | #ifdef CONFIG_SLUB_DEBUG_ON |
310 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
311 | #else | |
41ecc55b | 312 | static int slub_debug; |
f0630fff | 313 | #endif |
41ecc55b CL |
314 | |
315 | static char *slub_debug_slabs; | |
fa5ec8a1 | 316 | static int disable_higher_order_debug; |
41ecc55b | 317 | |
81819f0f CL |
318 | /* |
319 | * Object debugging | |
320 | */ | |
321 | static void print_section(char *text, u8 *addr, unsigned int length) | |
322 | { | |
323 | int i, offset; | |
324 | int newline = 1; | |
325 | char ascii[17]; | |
326 | ||
327 | ascii[16] = 0; | |
328 | ||
329 | for (i = 0; i < length; i++) { | |
330 | if (newline) { | |
24922684 | 331 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
332 | newline = 0; |
333 | } | |
06428780 | 334 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
335 | offset = i % 16; |
336 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
337 | if (offset == 15) { | |
06428780 | 338 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
339 | newline = 1; |
340 | } | |
341 | } | |
342 | if (!newline) { | |
343 | i %= 16; | |
344 | while (i < 16) { | |
06428780 | 345 | printk(KERN_CONT " "); |
81819f0f CL |
346 | ascii[i] = ' '; |
347 | i++; | |
348 | } | |
06428780 | 349 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
350 | } |
351 | } | |
352 | ||
81819f0f CL |
353 | static struct track *get_track(struct kmem_cache *s, void *object, |
354 | enum track_item alloc) | |
355 | { | |
356 | struct track *p; | |
357 | ||
358 | if (s->offset) | |
359 | p = object + s->offset + sizeof(void *); | |
360 | else | |
361 | p = object + s->inuse; | |
362 | ||
363 | return p + alloc; | |
364 | } | |
365 | ||
366 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 367 | enum track_item alloc, unsigned long addr) |
81819f0f | 368 | { |
1a00df4a | 369 | struct track *p = get_track(s, object, alloc); |
81819f0f | 370 | |
81819f0f CL |
371 | if (addr) { |
372 | p->addr = addr; | |
373 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 374 | p->pid = current->pid; |
81819f0f CL |
375 | p->when = jiffies; |
376 | } else | |
377 | memset(p, 0, sizeof(struct track)); | |
378 | } | |
379 | ||
81819f0f CL |
380 | static void init_tracking(struct kmem_cache *s, void *object) |
381 | { | |
24922684 CL |
382 | if (!(s->flags & SLAB_STORE_USER)) |
383 | return; | |
384 | ||
ce71e27c EGM |
385 | set_track(s, object, TRACK_FREE, 0UL); |
386 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
387 | } |
388 | ||
389 | static void print_track(const char *s, struct track *t) | |
390 | { | |
391 | if (!t->addr) | |
392 | return; | |
393 | ||
7daf705f | 394 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 395 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
396 | } |
397 | ||
398 | static void print_tracking(struct kmem_cache *s, void *object) | |
399 | { | |
400 | if (!(s->flags & SLAB_STORE_USER)) | |
401 | return; | |
402 | ||
403 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
404 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
405 | } | |
406 | ||
407 | static void print_page_info(struct page *page) | |
408 | { | |
39b26464 CL |
409 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
410 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
411 | |
412 | } | |
413 | ||
414 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
415 | { | |
416 | va_list args; | |
417 | char buf[100]; | |
418 | ||
419 | va_start(args, fmt); | |
420 | vsnprintf(buf, sizeof(buf), fmt, args); | |
421 | va_end(args); | |
422 | printk(KERN_ERR "========================================" | |
423 | "=====================================\n"); | |
424 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
425 | printk(KERN_ERR "----------------------------------------" | |
426 | "-------------------------------------\n\n"); | |
81819f0f CL |
427 | } |
428 | ||
24922684 CL |
429 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
430 | { | |
431 | va_list args; | |
432 | char buf[100]; | |
433 | ||
434 | va_start(args, fmt); | |
435 | vsnprintf(buf, sizeof(buf), fmt, args); | |
436 | va_end(args); | |
437 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
438 | } | |
439 | ||
440 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
441 | { |
442 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 443 | u8 *addr = page_address(page); |
24922684 CL |
444 | |
445 | print_tracking(s, p); | |
446 | ||
447 | print_page_info(page); | |
448 | ||
449 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
450 | p, p - addr, get_freepointer(s, p)); | |
451 | ||
452 | if (p > addr + 16) | |
453 | print_section("Bytes b4", p - 16, 16); | |
454 | ||
0ebd652b | 455 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
456 | |
457 | if (s->flags & SLAB_RED_ZONE) | |
458 | print_section("Redzone", p + s->objsize, | |
459 | s->inuse - s->objsize); | |
460 | ||
81819f0f CL |
461 | if (s->offset) |
462 | off = s->offset + sizeof(void *); | |
463 | else | |
464 | off = s->inuse; | |
465 | ||
24922684 | 466 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 467 | off += 2 * sizeof(struct track); |
81819f0f CL |
468 | |
469 | if (off != s->size) | |
470 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
471 | print_section("Padding", p + off, s->size - off); |
472 | ||
473 | dump_stack(); | |
81819f0f CL |
474 | } |
475 | ||
476 | static void object_err(struct kmem_cache *s, struct page *page, | |
477 | u8 *object, char *reason) | |
478 | { | |
3dc50637 | 479 | slab_bug(s, "%s", reason); |
24922684 | 480 | print_trailer(s, page, object); |
81819f0f CL |
481 | } |
482 | ||
24922684 | 483 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
484 | { |
485 | va_list args; | |
486 | char buf[100]; | |
487 | ||
24922684 CL |
488 | va_start(args, fmt); |
489 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 490 | va_end(args); |
3dc50637 | 491 | slab_bug(s, "%s", buf); |
24922684 | 492 | print_page_info(page); |
81819f0f CL |
493 | dump_stack(); |
494 | } | |
495 | ||
496 | static void init_object(struct kmem_cache *s, void *object, int active) | |
497 | { | |
498 | u8 *p = object; | |
499 | ||
500 | if (s->flags & __OBJECT_POISON) { | |
501 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 502 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
503 | } |
504 | ||
505 | if (s->flags & SLAB_RED_ZONE) | |
506 | memset(p + s->objsize, | |
507 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
508 | s->inuse - s->objsize); | |
509 | } | |
510 | ||
24922684 | 511 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
512 | { |
513 | while (bytes) { | |
514 | if (*start != (u8)value) | |
24922684 | 515 | return start; |
81819f0f CL |
516 | start++; |
517 | bytes--; | |
518 | } | |
24922684 CL |
519 | return NULL; |
520 | } | |
521 | ||
522 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
523 | void *from, void *to) | |
524 | { | |
525 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
526 | memset(from, data, to - from); | |
527 | } | |
528 | ||
529 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
530 | u8 *object, char *what, | |
06428780 | 531 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
532 | { |
533 | u8 *fault; | |
534 | u8 *end; | |
535 | ||
536 | fault = check_bytes(start, value, bytes); | |
537 | if (!fault) | |
538 | return 1; | |
539 | ||
540 | end = start + bytes; | |
541 | while (end > fault && end[-1] == value) | |
542 | end--; | |
543 | ||
544 | slab_bug(s, "%s overwritten", what); | |
545 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
546 | fault, end - 1, fault[0], value); | |
547 | print_trailer(s, page, object); | |
548 | ||
549 | restore_bytes(s, what, value, fault, end); | |
550 | return 0; | |
81819f0f CL |
551 | } |
552 | ||
81819f0f CL |
553 | /* |
554 | * Object layout: | |
555 | * | |
556 | * object address | |
557 | * Bytes of the object to be managed. | |
558 | * If the freepointer may overlay the object then the free | |
559 | * pointer is the first word of the object. | |
672bba3a | 560 | * |
81819f0f CL |
561 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
562 | * 0xa5 (POISON_END) | |
563 | * | |
564 | * object + s->objsize | |
565 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
566 | * Padding is extended by another word if Redzoning is enabled and |
567 | * objsize == inuse. | |
568 | * | |
81819f0f CL |
569 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
570 | * 0xcc (RED_ACTIVE) for objects in use. | |
571 | * | |
572 | * object + s->inuse | |
672bba3a CL |
573 | * Meta data starts here. |
574 | * | |
81819f0f CL |
575 | * A. Free pointer (if we cannot overwrite object on free) |
576 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 577 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 578 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
579 | * before the word boundary. |
580 | * | |
581 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
582 | * |
583 | * object + s->size | |
672bba3a | 584 | * Nothing is used beyond s->size. |
81819f0f | 585 | * |
672bba3a CL |
586 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
587 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
588 | * may be used with merged slabcaches. |
589 | */ | |
590 | ||
81819f0f CL |
591 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
592 | { | |
593 | unsigned long off = s->inuse; /* The end of info */ | |
594 | ||
595 | if (s->offset) | |
596 | /* Freepointer is placed after the object. */ | |
597 | off += sizeof(void *); | |
598 | ||
599 | if (s->flags & SLAB_STORE_USER) | |
600 | /* We also have user information there */ | |
601 | off += 2 * sizeof(struct track); | |
602 | ||
603 | if (s->size == off) | |
604 | return 1; | |
605 | ||
24922684 CL |
606 | return check_bytes_and_report(s, page, p, "Object padding", |
607 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
608 | } |
609 | ||
39b26464 | 610 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
611 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
612 | { | |
24922684 CL |
613 | u8 *start; |
614 | u8 *fault; | |
615 | u8 *end; | |
616 | int length; | |
617 | int remainder; | |
81819f0f CL |
618 | |
619 | if (!(s->flags & SLAB_POISON)) | |
620 | return 1; | |
621 | ||
a973e9dd | 622 | start = page_address(page); |
834f3d11 | 623 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
624 | end = start + length; |
625 | remainder = length % s->size; | |
81819f0f CL |
626 | if (!remainder) |
627 | return 1; | |
628 | ||
39b26464 | 629 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
630 | if (!fault) |
631 | return 1; | |
632 | while (end > fault && end[-1] == POISON_INUSE) | |
633 | end--; | |
634 | ||
635 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 636 | print_section("Padding", end - remainder, remainder); |
24922684 | 637 | |
8a3d271d | 638 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 639 | return 0; |
81819f0f CL |
640 | } |
641 | ||
642 | static int check_object(struct kmem_cache *s, struct page *page, | |
643 | void *object, int active) | |
644 | { | |
645 | u8 *p = object; | |
646 | u8 *endobject = object + s->objsize; | |
647 | ||
648 | if (s->flags & SLAB_RED_ZONE) { | |
649 | unsigned int red = | |
650 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
651 | ||
24922684 CL |
652 | if (!check_bytes_and_report(s, page, object, "Redzone", |
653 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 654 | return 0; |
81819f0f | 655 | } else { |
3adbefee IM |
656 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
657 | check_bytes_and_report(s, page, p, "Alignment padding", | |
658 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
659 | } | |
81819f0f CL |
660 | } |
661 | ||
662 | if (s->flags & SLAB_POISON) { | |
663 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
664 | (!check_bytes_and_report(s, page, p, "Poison", p, |
665 | POISON_FREE, s->objsize - 1) || | |
666 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 667 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 668 | return 0; |
81819f0f CL |
669 | /* |
670 | * check_pad_bytes cleans up on its own. | |
671 | */ | |
672 | check_pad_bytes(s, page, p); | |
673 | } | |
674 | ||
675 | if (!s->offset && active) | |
676 | /* | |
677 | * Object and freepointer overlap. Cannot check | |
678 | * freepointer while object is allocated. | |
679 | */ | |
680 | return 1; | |
681 | ||
682 | /* Check free pointer validity */ | |
683 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
684 | object_err(s, page, p, "Freepointer corrupt"); | |
685 | /* | |
9f6c708e | 686 | * No choice but to zap it and thus lose the remainder |
81819f0f | 687 | * of the free objects in this slab. May cause |
672bba3a | 688 | * another error because the object count is now wrong. |
81819f0f | 689 | */ |
a973e9dd | 690 | set_freepointer(s, p, NULL); |
81819f0f CL |
691 | return 0; |
692 | } | |
693 | return 1; | |
694 | } | |
695 | ||
696 | static int check_slab(struct kmem_cache *s, struct page *page) | |
697 | { | |
39b26464 CL |
698 | int maxobj; |
699 | ||
81819f0f CL |
700 | VM_BUG_ON(!irqs_disabled()); |
701 | ||
702 | if (!PageSlab(page)) { | |
24922684 | 703 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
704 | return 0; |
705 | } | |
39b26464 CL |
706 | |
707 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
708 | if (page->objects > maxobj) { | |
709 | slab_err(s, page, "objects %u > max %u", | |
710 | s->name, page->objects, maxobj); | |
711 | return 0; | |
712 | } | |
713 | if (page->inuse > page->objects) { | |
24922684 | 714 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 715 | s->name, page->inuse, page->objects); |
81819f0f CL |
716 | return 0; |
717 | } | |
718 | /* Slab_pad_check fixes things up after itself */ | |
719 | slab_pad_check(s, page); | |
720 | return 1; | |
721 | } | |
722 | ||
723 | /* | |
672bba3a CL |
724 | * Determine if a certain object on a page is on the freelist. Must hold the |
725 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
726 | */ |
727 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
728 | { | |
729 | int nr = 0; | |
730 | void *fp = page->freelist; | |
731 | void *object = NULL; | |
224a88be | 732 | unsigned long max_objects; |
81819f0f | 733 | |
39b26464 | 734 | while (fp && nr <= page->objects) { |
81819f0f CL |
735 | if (fp == search) |
736 | return 1; | |
737 | if (!check_valid_pointer(s, page, fp)) { | |
738 | if (object) { | |
739 | object_err(s, page, object, | |
740 | "Freechain corrupt"); | |
a973e9dd | 741 | set_freepointer(s, object, NULL); |
81819f0f CL |
742 | break; |
743 | } else { | |
24922684 | 744 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 745 | page->freelist = NULL; |
39b26464 | 746 | page->inuse = page->objects; |
24922684 | 747 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
748 | return 0; |
749 | } | |
750 | break; | |
751 | } | |
752 | object = fp; | |
753 | fp = get_freepointer(s, object); | |
754 | nr++; | |
755 | } | |
756 | ||
224a88be | 757 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
758 | if (max_objects > MAX_OBJS_PER_PAGE) |
759 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
760 | |
761 | if (page->objects != max_objects) { | |
762 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
763 | "should be %d", page->objects, max_objects); | |
764 | page->objects = max_objects; | |
765 | slab_fix(s, "Number of objects adjusted."); | |
766 | } | |
39b26464 | 767 | if (page->inuse != page->objects - nr) { |
70d71228 | 768 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
769 | "counted were %d", page->inuse, page->objects - nr); |
770 | page->inuse = page->objects - nr; | |
24922684 | 771 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
772 | } |
773 | return search == NULL; | |
774 | } | |
775 | ||
0121c619 CL |
776 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
777 | int alloc) | |
3ec09742 CL |
778 | { |
779 | if (s->flags & SLAB_TRACE) { | |
780 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
781 | s->name, | |
782 | alloc ? "alloc" : "free", | |
783 | object, page->inuse, | |
784 | page->freelist); | |
785 | ||
786 | if (!alloc) | |
787 | print_section("Object", (void *)object, s->objsize); | |
788 | ||
789 | dump_stack(); | |
790 | } | |
791 | } | |
792 | ||
c016b0bd CL |
793 | /* |
794 | * Hooks for other subsystems that check memory allocations. In a typical | |
795 | * production configuration these hooks all should produce no code at all. | |
796 | */ | |
797 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
798 | { | |
c1d50836 | 799 | flags &= gfp_allowed_mask; |
c016b0bd CL |
800 | lockdep_trace_alloc(flags); |
801 | might_sleep_if(flags & __GFP_WAIT); | |
802 | ||
803 | return should_failslab(s->objsize, flags, s->flags); | |
804 | } | |
805 | ||
806 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
807 | { | |
c1d50836 | 808 | flags &= gfp_allowed_mask; |
c016b0bd CL |
809 | kmemcheck_slab_alloc(s, flags, object, s->objsize); |
810 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | |
811 | } | |
812 | ||
813 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
814 | { | |
815 | kmemleak_free_recursive(x, s->flags); | |
816 | } | |
817 | ||
818 | static inline void slab_free_hook_irq(struct kmem_cache *s, void *object) | |
819 | { | |
820 | kmemcheck_slab_free(s, object, s->objsize); | |
821 | debug_check_no_locks_freed(object, s->objsize); | |
822 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
823 | debug_check_no_obj_freed(object, s->objsize); | |
824 | } | |
825 | ||
643b1138 | 826 | /* |
672bba3a | 827 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 828 | */ |
e95eed57 | 829 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 830 | { |
643b1138 CL |
831 | spin_lock(&n->list_lock); |
832 | list_add(&page->lru, &n->full); | |
833 | spin_unlock(&n->list_lock); | |
834 | } | |
835 | ||
836 | static void remove_full(struct kmem_cache *s, struct page *page) | |
837 | { | |
838 | struct kmem_cache_node *n; | |
839 | ||
840 | if (!(s->flags & SLAB_STORE_USER)) | |
841 | return; | |
842 | ||
843 | n = get_node(s, page_to_nid(page)); | |
844 | ||
845 | spin_lock(&n->list_lock); | |
846 | list_del(&page->lru); | |
847 | spin_unlock(&n->list_lock); | |
848 | } | |
849 | ||
0f389ec6 CL |
850 | /* Tracking of the number of slabs for debugging purposes */ |
851 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
852 | { | |
853 | struct kmem_cache_node *n = get_node(s, node); | |
854 | ||
855 | return atomic_long_read(&n->nr_slabs); | |
856 | } | |
857 | ||
26c02cf0 AB |
858 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
859 | { | |
860 | return atomic_long_read(&n->nr_slabs); | |
861 | } | |
862 | ||
205ab99d | 863 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
864 | { |
865 | struct kmem_cache_node *n = get_node(s, node); | |
866 | ||
867 | /* | |
868 | * May be called early in order to allocate a slab for the | |
869 | * kmem_cache_node structure. Solve the chicken-egg | |
870 | * dilemma by deferring the increment of the count during | |
871 | * bootstrap (see early_kmem_cache_node_alloc). | |
872 | */ | |
205ab99d | 873 | if (!NUMA_BUILD || n) { |
0f389ec6 | 874 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
875 | atomic_long_add(objects, &n->total_objects); |
876 | } | |
0f389ec6 | 877 | } |
205ab99d | 878 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
879 | { |
880 | struct kmem_cache_node *n = get_node(s, node); | |
881 | ||
882 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 883 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
884 | } |
885 | ||
886 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
887 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
888 | void *object) | |
889 | { | |
890 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
891 | return; | |
892 | ||
893 | init_object(s, object, 0); | |
894 | init_tracking(s, object); | |
895 | } | |
896 | ||
1537066c | 897 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 898 | void *object, unsigned long addr) |
81819f0f CL |
899 | { |
900 | if (!check_slab(s, page)) | |
901 | goto bad; | |
902 | ||
d692ef6d | 903 | if (!on_freelist(s, page, object)) { |
24922684 | 904 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 905 | goto bad; |
81819f0f CL |
906 | } |
907 | ||
908 | if (!check_valid_pointer(s, page, object)) { | |
909 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 910 | goto bad; |
81819f0f CL |
911 | } |
912 | ||
d692ef6d | 913 | if (!check_object(s, page, object, 0)) |
81819f0f | 914 | goto bad; |
81819f0f | 915 | |
3ec09742 CL |
916 | /* Success perform special debug activities for allocs */ |
917 | if (s->flags & SLAB_STORE_USER) | |
918 | set_track(s, object, TRACK_ALLOC, addr); | |
919 | trace(s, page, object, 1); | |
920 | init_object(s, object, 1); | |
81819f0f | 921 | return 1; |
3ec09742 | 922 | |
81819f0f CL |
923 | bad: |
924 | if (PageSlab(page)) { | |
925 | /* | |
926 | * If this is a slab page then lets do the best we can | |
927 | * to avoid issues in the future. Marking all objects | |
672bba3a | 928 | * as used avoids touching the remaining objects. |
81819f0f | 929 | */ |
24922684 | 930 | slab_fix(s, "Marking all objects used"); |
39b26464 | 931 | page->inuse = page->objects; |
a973e9dd | 932 | page->freelist = NULL; |
81819f0f CL |
933 | } |
934 | return 0; | |
935 | } | |
936 | ||
1537066c CL |
937 | static noinline int free_debug_processing(struct kmem_cache *s, |
938 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
939 | { |
940 | if (!check_slab(s, page)) | |
941 | goto fail; | |
942 | ||
943 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 944 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
945 | goto fail; |
946 | } | |
947 | ||
948 | if (on_freelist(s, page, object)) { | |
24922684 | 949 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
950 | goto fail; |
951 | } | |
952 | ||
953 | if (!check_object(s, page, object, 1)) | |
954 | return 0; | |
955 | ||
956 | if (unlikely(s != page->slab)) { | |
3adbefee | 957 | if (!PageSlab(page)) { |
70d71228 CL |
958 | slab_err(s, page, "Attempt to free object(0x%p) " |
959 | "outside of slab", object); | |
3adbefee | 960 | } else if (!page->slab) { |
81819f0f | 961 | printk(KERN_ERR |
70d71228 | 962 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 963 | object); |
70d71228 | 964 | dump_stack(); |
06428780 | 965 | } else |
24922684 CL |
966 | object_err(s, page, object, |
967 | "page slab pointer corrupt."); | |
81819f0f CL |
968 | goto fail; |
969 | } | |
3ec09742 CL |
970 | |
971 | /* Special debug activities for freeing objects */ | |
8a38082d | 972 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
973 | remove_full(s, page); |
974 | if (s->flags & SLAB_STORE_USER) | |
975 | set_track(s, object, TRACK_FREE, addr); | |
976 | trace(s, page, object, 0); | |
977 | init_object(s, object, 0); | |
81819f0f | 978 | return 1; |
3ec09742 | 979 | |
81819f0f | 980 | fail: |
24922684 | 981 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
982 | return 0; |
983 | } | |
984 | ||
41ecc55b CL |
985 | static int __init setup_slub_debug(char *str) |
986 | { | |
f0630fff CL |
987 | slub_debug = DEBUG_DEFAULT_FLAGS; |
988 | if (*str++ != '=' || !*str) | |
989 | /* | |
990 | * No options specified. Switch on full debugging. | |
991 | */ | |
992 | goto out; | |
993 | ||
994 | if (*str == ',') | |
995 | /* | |
996 | * No options but restriction on slabs. This means full | |
997 | * debugging for slabs matching a pattern. | |
998 | */ | |
999 | goto check_slabs; | |
1000 | ||
fa5ec8a1 DR |
1001 | if (tolower(*str) == 'o') { |
1002 | /* | |
1003 | * Avoid enabling debugging on caches if its minimum order | |
1004 | * would increase as a result. | |
1005 | */ | |
1006 | disable_higher_order_debug = 1; | |
1007 | goto out; | |
1008 | } | |
1009 | ||
f0630fff CL |
1010 | slub_debug = 0; |
1011 | if (*str == '-') | |
1012 | /* | |
1013 | * Switch off all debugging measures. | |
1014 | */ | |
1015 | goto out; | |
1016 | ||
1017 | /* | |
1018 | * Determine which debug features should be switched on | |
1019 | */ | |
06428780 | 1020 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1021 | switch (tolower(*str)) { |
1022 | case 'f': | |
1023 | slub_debug |= SLAB_DEBUG_FREE; | |
1024 | break; | |
1025 | case 'z': | |
1026 | slub_debug |= SLAB_RED_ZONE; | |
1027 | break; | |
1028 | case 'p': | |
1029 | slub_debug |= SLAB_POISON; | |
1030 | break; | |
1031 | case 'u': | |
1032 | slub_debug |= SLAB_STORE_USER; | |
1033 | break; | |
1034 | case 't': | |
1035 | slub_debug |= SLAB_TRACE; | |
1036 | break; | |
4c13dd3b DM |
1037 | case 'a': |
1038 | slub_debug |= SLAB_FAILSLAB; | |
1039 | break; | |
f0630fff CL |
1040 | default: |
1041 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1042 | "unknown. skipped\n", *str); |
f0630fff | 1043 | } |
41ecc55b CL |
1044 | } |
1045 | ||
f0630fff | 1046 | check_slabs: |
41ecc55b CL |
1047 | if (*str == ',') |
1048 | slub_debug_slabs = str + 1; | |
f0630fff | 1049 | out: |
41ecc55b CL |
1050 | return 1; |
1051 | } | |
1052 | ||
1053 | __setup("slub_debug", setup_slub_debug); | |
1054 | ||
ba0268a8 CL |
1055 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1056 | unsigned long flags, const char *name, | |
51cc5068 | 1057 | void (*ctor)(void *)) |
41ecc55b CL |
1058 | { |
1059 | /* | |
e153362a | 1060 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1061 | */ |
e153362a | 1062 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1063 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1064 | flags |= slub_debug; | |
ba0268a8 CL |
1065 | |
1066 | return flags; | |
41ecc55b CL |
1067 | } |
1068 | #else | |
3ec09742 CL |
1069 | static inline void setup_object_debug(struct kmem_cache *s, |
1070 | struct page *page, void *object) {} | |
41ecc55b | 1071 | |
3ec09742 | 1072 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1073 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1074 | |
3ec09742 | 1075 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1076 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1077 | |
41ecc55b CL |
1078 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1079 | { return 1; } | |
1080 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1081 | void *object, int active) { return 1; } | |
3ec09742 | 1082 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1083 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1084 | unsigned long flags, const char *name, | |
51cc5068 | 1085 | void (*ctor)(void *)) |
ba0268a8 CL |
1086 | { |
1087 | return flags; | |
1088 | } | |
41ecc55b | 1089 | #define slub_debug 0 |
0f389ec6 | 1090 | |
fdaa45e9 IM |
1091 | #define disable_higher_order_debug 0 |
1092 | ||
0f389ec6 CL |
1093 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1094 | { return 0; } | |
26c02cf0 AB |
1095 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1096 | { return 0; } | |
205ab99d CL |
1097 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1098 | int objects) {} | |
1099 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1100 | int objects) {} | |
41ecc55b | 1101 | #endif |
205ab99d | 1102 | |
81819f0f CL |
1103 | /* |
1104 | * Slab allocation and freeing | |
1105 | */ | |
65c3376a CL |
1106 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1107 | struct kmem_cache_order_objects oo) | |
1108 | { | |
1109 | int order = oo_order(oo); | |
1110 | ||
b1eeab67 VN |
1111 | flags |= __GFP_NOTRACK; |
1112 | ||
2154a336 | 1113 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1114 | return alloc_pages(flags, order); |
1115 | else | |
6b65aaf3 | 1116 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1117 | } |
1118 | ||
81819f0f CL |
1119 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1120 | { | |
06428780 | 1121 | struct page *page; |
834f3d11 | 1122 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1123 | gfp_t alloc_gfp; |
81819f0f | 1124 | |
b7a49f0d | 1125 | flags |= s->allocflags; |
e12ba74d | 1126 | |
ba52270d PE |
1127 | /* |
1128 | * Let the initial higher-order allocation fail under memory pressure | |
1129 | * so we fall-back to the minimum order allocation. | |
1130 | */ | |
1131 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1132 | ||
1133 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1134 | if (unlikely(!page)) { |
1135 | oo = s->min; | |
1136 | /* | |
1137 | * Allocation may have failed due to fragmentation. | |
1138 | * Try a lower order alloc if possible | |
1139 | */ | |
1140 | page = alloc_slab_page(flags, node, oo); | |
1141 | if (!page) | |
1142 | return NULL; | |
81819f0f | 1143 | |
84e554e6 | 1144 | stat(s, ORDER_FALLBACK); |
65c3376a | 1145 | } |
5a896d9e VN |
1146 | |
1147 | if (kmemcheck_enabled | |
5086c389 | 1148 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1149 | int pages = 1 << oo_order(oo); |
1150 | ||
1151 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1152 | ||
1153 | /* | |
1154 | * Objects from caches that have a constructor don't get | |
1155 | * cleared when they're allocated, so we need to do it here. | |
1156 | */ | |
1157 | if (s->ctor) | |
1158 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1159 | else | |
1160 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1161 | } |
1162 | ||
834f3d11 | 1163 | page->objects = oo_objects(oo); |
81819f0f CL |
1164 | mod_zone_page_state(page_zone(page), |
1165 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1166 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1167 | 1 << oo_order(oo)); |
81819f0f CL |
1168 | |
1169 | return page; | |
1170 | } | |
1171 | ||
1172 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1173 | void *object) | |
1174 | { | |
3ec09742 | 1175 | setup_object_debug(s, page, object); |
4f104934 | 1176 | if (unlikely(s->ctor)) |
51cc5068 | 1177 | s->ctor(object); |
81819f0f CL |
1178 | } |
1179 | ||
1180 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1181 | { | |
1182 | struct page *page; | |
81819f0f | 1183 | void *start; |
81819f0f CL |
1184 | void *last; |
1185 | void *p; | |
1186 | ||
6cb06229 | 1187 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1188 | |
6cb06229 CL |
1189 | page = allocate_slab(s, |
1190 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1191 | if (!page) |
1192 | goto out; | |
1193 | ||
205ab99d | 1194 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1195 | page->slab = s; |
1196 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1197 | |
1198 | start = page_address(page); | |
81819f0f CL |
1199 | |
1200 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1201 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1202 | |
1203 | last = start; | |
224a88be | 1204 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1205 | setup_object(s, page, last); |
1206 | set_freepointer(s, last, p); | |
1207 | last = p; | |
1208 | } | |
1209 | setup_object(s, page, last); | |
a973e9dd | 1210 | set_freepointer(s, last, NULL); |
81819f0f CL |
1211 | |
1212 | page->freelist = start; | |
1213 | page->inuse = 0; | |
1214 | out: | |
81819f0f CL |
1215 | return page; |
1216 | } | |
1217 | ||
1218 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1219 | { | |
834f3d11 CL |
1220 | int order = compound_order(page); |
1221 | int pages = 1 << order; | |
81819f0f | 1222 | |
af537b0a | 1223 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1224 | void *p; |
1225 | ||
1226 | slab_pad_check(s, page); | |
224a88be CL |
1227 | for_each_object(p, s, page_address(page), |
1228 | page->objects) | |
81819f0f | 1229 | check_object(s, page, p, 0); |
81819f0f CL |
1230 | } |
1231 | ||
b1eeab67 | 1232 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1233 | |
81819f0f CL |
1234 | mod_zone_page_state(page_zone(page), |
1235 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1236 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1237 | -pages); |
81819f0f | 1238 | |
49bd5221 CL |
1239 | __ClearPageSlab(page); |
1240 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1241 | if (current->reclaim_state) |
1242 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1243 | __free_pages(page, order); |
81819f0f CL |
1244 | } |
1245 | ||
1246 | static void rcu_free_slab(struct rcu_head *h) | |
1247 | { | |
1248 | struct page *page; | |
1249 | ||
1250 | page = container_of((struct list_head *)h, struct page, lru); | |
1251 | __free_slab(page->slab, page); | |
1252 | } | |
1253 | ||
1254 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1255 | { | |
1256 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1257 | /* | |
1258 | * RCU free overloads the RCU head over the LRU | |
1259 | */ | |
1260 | struct rcu_head *head = (void *)&page->lru; | |
1261 | ||
1262 | call_rcu(head, rcu_free_slab); | |
1263 | } else | |
1264 | __free_slab(s, page); | |
1265 | } | |
1266 | ||
1267 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1268 | { | |
205ab99d | 1269 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1270 | free_slab(s, page); |
1271 | } | |
1272 | ||
1273 | /* | |
1274 | * Per slab locking using the pagelock | |
1275 | */ | |
1276 | static __always_inline void slab_lock(struct page *page) | |
1277 | { | |
1278 | bit_spin_lock(PG_locked, &page->flags); | |
1279 | } | |
1280 | ||
1281 | static __always_inline void slab_unlock(struct page *page) | |
1282 | { | |
a76d3546 | 1283 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1284 | } |
1285 | ||
1286 | static __always_inline int slab_trylock(struct page *page) | |
1287 | { | |
1288 | int rc = 1; | |
1289 | ||
1290 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1291 | return rc; | |
1292 | } | |
1293 | ||
1294 | /* | |
1295 | * Management of partially allocated slabs | |
1296 | */ | |
7c2e132c CL |
1297 | static void add_partial(struct kmem_cache_node *n, |
1298 | struct page *page, int tail) | |
81819f0f | 1299 | { |
e95eed57 CL |
1300 | spin_lock(&n->list_lock); |
1301 | n->nr_partial++; | |
7c2e132c CL |
1302 | if (tail) |
1303 | list_add_tail(&page->lru, &n->partial); | |
1304 | else | |
1305 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1306 | spin_unlock(&n->list_lock); |
1307 | } | |
1308 | ||
0121c619 | 1309 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1310 | { |
1311 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1312 | ||
1313 | spin_lock(&n->list_lock); | |
1314 | list_del(&page->lru); | |
1315 | n->nr_partial--; | |
1316 | spin_unlock(&n->list_lock); | |
1317 | } | |
1318 | ||
1319 | /* | |
672bba3a | 1320 | * Lock slab and remove from the partial list. |
81819f0f | 1321 | * |
672bba3a | 1322 | * Must hold list_lock. |
81819f0f | 1323 | */ |
0121c619 CL |
1324 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1325 | struct page *page) | |
81819f0f CL |
1326 | { |
1327 | if (slab_trylock(page)) { | |
1328 | list_del(&page->lru); | |
1329 | n->nr_partial--; | |
8a38082d | 1330 | __SetPageSlubFrozen(page); |
81819f0f CL |
1331 | return 1; |
1332 | } | |
1333 | return 0; | |
1334 | } | |
1335 | ||
1336 | /* | |
672bba3a | 1337 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1338 | */ |
1339 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1340 | { | |
1341 | struct page *page; | |
1342 | ||
1343 | /* | |
1344 | * Racy check. If we mistakenly see no partial slabs then we | |
1345 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1346 | * partial slab and there is none available then get_partials() |
1347 | * will return NULL. | |
81819f0f CL |
1348 | */ |
1349 | if (!n || !n->nr_partial) | |
1350 | return NULL; | |
1351 | ||
1352 | spin_lock(&n->list_lock); | |
1353 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1354 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1355 | goto out; |
1356 | page = NULL; | |
1357 | out: | |
1358 | spin_unlock(&n->list_lock); | |
1359 | return page; | |
1360 | } | |
1361 | ||
1362 | /* | |
672bba3a | 1363 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1364 | */ |
1365 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1366 | { | |
1367 | #ifdef CONFIG_NUMA | |
1368 | struct zonelist *zonelist; | |
dd1a239f | 1369 | struct zoneref *z; |
54a6eb5c MG |
1370 | struct zone *zone; |
1371 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1372 | struct page *page; |
1373 | ||
1374 | /* | |
672bba3a CL |
1375 | * The defrag ratio allows a configuration of the tradeoffs between |
1376 | * inter node defragmentation and node local allocations. A lower | |
1377 | * defrag_ratio increases the tendency to do local allocations | |
1378 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1379 | * |
672bba3a CL |
1380 | * If the defrag_ratio is set to 0 then kmalloc() always |
1381 | * returns node local objects. If the ratio is higher then kmalloc() | |
1382 | * may return off node objects because partial slabs are obtained | |
1383 | * from other nodes and filled up. | |
81819f0f | 1384 | * |
6446faa2 | 1385 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1386 | * defrag_ratio = 1000) then every (well almost) allocation will |
1387 | * first attempt to defrag slab caches on other nodes. This means | |
1388 | * scanning over all nodes to look for partial slabs which may be | |
1389 | * expensive if we do it every time we are trying to find a slab | |
1390 | * with available objects. | |
81819f0f | 1391 | */ |
9824601e CL |
1392 | if (!s->remote_node_defrag_ratio || |
1393 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1394 | return NULL; |
1395 | ||
c0ff7453 | 1396 | get_mems_allowed(); |
0e88460d | 1397 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1398 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1399 | struct kmem_cache_node *n; |
1400 | ||
54a6eb5c | 1401 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1402 | |
54a6eb5c | 1403 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1404 | n->nr_partial > s->min_partial) { |
81819f0f | 1405 | page = get_partial_node(n); |
c0ff7453 MX |
1406 | if (page) { |
1407 | put_mems_allowed(); | |
81819f0f | 1408 | return page; |
c0ff7453 | 1409 | } |
81819f0f CL |
1410 | } |
1411 | } | |
c0ff7453 | 1412 | put_mems_allowed(); |
81819f0f CL |
1413 | #endif |
1414 | return NULL; | |
1415 | } | |
1416 | ||
1417 | /* | |
1418 | * Get a partial page, lock it and return it. | |
1419 | */ | |
1420 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1421 | { | |
1422 | struct page *page; | |
2154a336 | 1423 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1424 | |
1425 | page = get_partial_node(get_node(s, searchnode)); | |
bc6488e9 | 1426 | if (page || node != -1) |
81819f0f CL |
1427 | return page; |
1428 | ||
1429 | return get_any_partial(s, flags); | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * Move a page back to the lists. | |
1434 | * | |
1435 | * Must be called with the slab lock held. | |
1436 | * | |
1437 | * On exit the slab lock will have been dropped. | |
1438 | */ | |
7c2e132c | 1439 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1440 | { |
e95eed57 CL |
1441 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1442 | ||
8a38082d | 1443 | __ClearPageSlubFrozen(page); |
81819f0f | 1444 | if (page->inuse) { |
e95eed57 | 1445 | |
a973e9dd | 1446 | if (page->freelist) { |
7c2e132c | 1447 | add_partial(n, page, tail); |
84e554e6 | 1448 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1449 | } else { |
84e554e6 | 1450 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1451 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1452 | add_full(n, page); |
1453 | } | |
81819f0f CL |
1454 | slab_unlock(page); |
1455 | } else { | |
84e554e6 | 1456 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1457 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1458 | /* |
672bba3a CL |
1459 | * Adding an empty slab to the partial slabs in order |
1460 | * to avoid page allocator overhead. This slab needs | |
1461 | * to come after the other slabs with objects in | |
6446faa2 CL |
1462 | * so that the others get filled first. That way the |
1463 | * size of the partial list stays small. | |
1464 | * | |
0121c619 CL |
1465 | * kmem_cache_shrink can reclaim any empty slabs from |
1466 | * the partial list. | |
e95eed57 | 1467 | */ |
7c2e132c | 1468 | add_partial(n, page, 1); |
e95eed57 CL |
1469 | slab_unlock(page); |
1470 | } else { | |
1471 | slab_unlock(page); | |
84e554e6 | 1472 | stat(s, FREE_SLAB); |
e95eed57 CL |
1473 | discard_slab(s, page); |
1474 | } | |
81819f0f CL |
1475 | } |
1476 | } | |
1477 | ||
1478 | /* | |
1479 | * Remove the cpu slab | |
1480 | */ | |
dfb4f096 | 1481 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1482 | { |
dfb4f096 | 1483 | struct page *page = c->page; |
7c2e132c | 1484 | int tail = 1; |
8ff12cfc | 1485 | |
b773ad73 | 1486 | if (page->freelist) |
84e554e6 | 1487 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1488 | /* |
6446faa2 | 1489 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1490 | * because both freelists are empty. So this is unlikely |
1491 | * to occur. | |
1492 | */ | |
a973e9dd | 1493 | while (unlikely(c->freelist)) { |
894b8788 CL |
1494 | void **object; |
1495 | ||
7c2e132c CL |
1496 | tail = 0; /* Hot objects. Put the slab first */ |
1497 | ||
894b8788 | 1498 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1499 | object = c->freelist; |
ff12059e | 1500 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1501 | |
1502 | /* And put onto the regular freelist */ | |
ff12059e | 1503 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1504 | page->freelist = object; |
1505 | page->inuse--; | |
1506 | } | |
dfb4f096 | 1507 | c->page = NULL; |
7c2e132c | 1508 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1509 | } |
1510 | ||
dfb4f096 | 1511 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1512 | { |
84e554e6 | 1513 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1514 | slab_lock(c->page); |
1515 | deactivate_slab(s, c); | |
81819f0f CL |
1516 | } |
1517 | ||
1518 | /* | |
1519 | * Flush cpu slab. | |
6446faa2 | 1520 | * |
81819f0f CL |
1521 | * Called from IPI handler with interrupts disabled. |
1522 | */ | |
0c710013 | 1523 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1524 | { |
9dfc6e68 | 1525 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1526 | |
dfb4f096 CL |
1527 | if (likely(c && c->page)) |
1528 | flush_slab(s, c); | |
81819f0f CL |
1529 | } |
1530 | ||
1531 | static void flush_cpu_slab(void *d) | |
1532 | { | |
1533 | struct kmem_cache *s = d; | |
81819f0f | 1534 | |
dfb4f096 | 1535 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1536 | } |
1537 | ||
1538 | static void flush_all(struct kmem_cache *s) | |
1539 | { | |
15c8b6c1 | 1540 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1541 | } |
1542 | ||
dfb4f096 CL |
1543 | /* |
1544 | * Check if the objects in a per cpu structure fit numa | |
1545 | * locality expectations. | |
1546 | */ | |
1547 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1548 | { | |
1549 | #ifdef CONFIG_NUMA | |
2154a336 | 1550 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1551 | return 0; |
1552 | #endif | |
1553 | return 1; | |
1554 | } | |
1555 | ||
781b2ba6 PE |
1556 | static int count_free(struct page *page) |
1557 | { | |
1558 | return page->objects - page->inuse; | |
1559 | } | |
1560 | ||
1561 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1562 | int (*get_count)(struct page *)) | |
1563 | { | |
1564 | unsigned long flags; | |
1565 | unsigned long x = 0; | |
1566 | struct page *page; | |
1567 | ||
1568 | spin_lock_irqsave(&n->list_lock, flags); | |
1569 | list_for_each_entry(page, &n->partial, lru) | |
1570 | x += get_count(page); | |
1571 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1572 | return x; | |
1573 | } | |
1574 | ||
26c02cf0 AB |
1575 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1576 | { | |
1577 | #ifdef CONFIG_SLUB_DEBUG | |
1578 | return atomic_long_read(&n->total_objects); | |
1579 | #else | |
1580 | return 0; | |
1581 | #endif | |
1582 | } | |
1583 | ||
781b2ba6 PE |
1584 | static noinline void |
1585 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1586 | { | |
1587 | int node; | |
1588 | ||
1589 | printk(KERN_WARNING | |
1590 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1591 | nid, gfpflags); | |
1592 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1593 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1594 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1595 | ||
fa5ec8a1 DR |
1596 | if (oo_order(s->min) > get_order(s->objsize)) |
1597 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1598 | "slub_debug=O to disable.\n", s->name); | |
1599 | ||
781b2ba6 PE |
1600 | for_each_online_node(node) { |
1601 | struct kmem_cache_node *n = get_node(s, node); | |
1602 | unsigned long nr_slabs; | |
1603 | unsigned long nr_objs; | |
1604 | unsigned long nr_free; | |
1605 | ||
1606 | if (!n) | |
1607 | continue; | |
1608 | ||
26c02cf0 AB |
1609 | nr_free = count_partial(n, count_free); |
1610 | nr_slabs = node_nr_slabs(n); | |
1611 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1612 | |
1613 | printk(KERN_WARNING | |
1614 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1615 | node, nr_slabs, nr_objs, nr_free); | |
1616 | } | |
1617 | } | |
1618 | ||
81819f0f | 1619 | /* |
894b8788 CL |
1620 | * Slow path. The lockless freelist is empty or we need to perform |
1621 | * debugging duties. | |
1622 | * | |
1623 | * Interrupts are disabled. | |
81819f0f | 1624 | * |
894b8788 CL |
1625 | * Processing is still very fast if new objects have been freed to the |
1626 | * regular freelist. In that case we simply take over the regular freelist | |
1627 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1628 | * |
894b8788 CL |
1629 | * If that is not working then we fall back to the partial lists. We take the |
1630 | * first element of the freelist as the object to allocate now and move the | |
1631 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1632 | * |
894b8788 | 1633 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1634 | * we need to allocate a new slab. This is the slowest path since it involves |
1635 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1636 | */ |
ce71e27c EGM |
1637 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1638 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1639 | { |
81819f0f | 1640 | void **object; |
dfb4f096 | 1641 | struct page *new; |
81819f0f | 1642 | |
e72e9c23 LT |
1643 | /* We handle __GFP_ZERO in the caller */ |
1644 | gfpflags &= ~__GFP_ZERO; | |
1645 | ||
dfb4f096 | 1646 | if (!c->page) |
81819f0f CL |
1647 | goto new_slab; |
1648 | ||
dfb4f096 CL |
1649 | slab_lock(c->page); |
1650 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1651 | goto another_slab; |
6446faa2 | 1652 | |
84e554e6 | 1653 | stat(s, ALLOC_REFILL); |
6446faa2 | 1654 | |
894b8788 | 1655 | load_freelist: |
dfb4f096 | 1656 | object = c->page->freelist; |
a973e9dd | 1657 | if (unlikely(!object)) |
81819f0f | 1658 | goto another_slab; |
af537b0a | 1659 | if (kmem_cache_debug(s)) |
81819f0f CL |
1660 | goto debug; |
1661 | ||
ff12059e | 1662 | c->freelist = get_freepointer(s, object); |
39b26464 | 1663 | c->page->inuse = c->page->objects; |
a973e9dd | 1664 | c->page->freelist = NULL; |
dfb4f096 | 1665 | c->node = page_to_nid(c->page); |
1f84260c | 1666 | unlock_out: |
dfb4f096 | 1667 | slab_unlock(c->page); |
84e554e6 | 1668 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1669 | return object; |
1670 | ||
1671 | another_slab: | |
dfb4f096 | 1672 | deactivate_slab(s, c); |
81819f0f CL |
1673 | |
1674 | new_slab: | |
dfb4f096 CL |
1675 | new = get_partial(s, gfpflags, node); |
1676 | if (new) { | |
1677 | c->page = new; | |
84e554e6 | 1678 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1679 | goto load_freelist; |
81819f0f CL |
1680 | } |
1681 | ||
c1d50836 | 1682 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1683 | if (gfpflags & __GFP_WAIT) |
1684 | local_irq_enable(); | |
1685 | ||
dfb4f096 | 1686 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1687 | |
1688 | if (gfpflags & __GFP_WAIT) | |
1689 | local_irq_disable(); | |
1690 | ||
dfb4f096 | 1691 | if (new) { |
9dfc6e68 | 1692 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1693 | stat(s, ALLOC_SLAB); |
05aa3450 | 1694 | if (c->page) |
dfb4f096 | 1695 | flush_slab(s, c); |
dfb4f096 | 1696 | slab_lock(new); |
8a38082d | 1697 | __SetPageSlubFrozen(new); |
dfb4f096 | 1698 | c->page = new; |
4b6f0750 | 1699 | goto load_freelist; |
81819f0f | 1700 | } |
95f85989 PE |
1701 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1702 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1703 | return NULL; |
81819f0f | 1704 | debug: |
dfb4f096 | 1705 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1706 | goto another_slab; |
894b8788 | 1707 | |
dfb4f096 | 1708 | c->page->inuse++; |
ff12059e | 1709 | c->page->freelist = get_freepointer(s, object); |
ee3c72a1 | 1710 | c->node = -1; |
1f84260c | 1711 | goto unlock_out; |
894b8788 CL |
1712 | } |
1713 | ||
1714 | /* | |
1715 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1716 | * have the fastpath folded into their functions. So no function call | |
1717 | * overhead for requests that can be satisfied on the fastpath. | |
1718 | * | |
1719 | * The fastpath works by first checking if the lockless freelist can be used. | |
1720 | * If not then __slab_alloc is called for slow processing. | |
1721 | * | |
1722 | * Otherwise we can simply pick the next object from the lockless free list. | |
1723 | */ | |
06428780 | 1724 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1725 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1726 | { |
894b8788 | 1727 | void **object; |
dfb4f096 | 1728 | struct kmem_cache_cpu *c; |
1f84260c CL |
1729 | unsigned long flags; |
1730 | ||
c016b0bd | 1731 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1732 | return NULL; |
1f84260c | 1733 | |
894b8788 | 1734 | local_irq_save(flags); |
9dfc6e68 CL |
1735 | c = __this_cpu_ptr(s->cpu_slab); |
1736 | object = c->freelist; | |
9dfc6e68 | 1737 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1738 | |
dfb4f096 | 1739 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1740 | |
1741 | else { | |
ff12059e | 1742 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1743 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1744 | } |
1745 | local_irq_restore(flags); | |
d07dbea4 | 1746 | |
74e2134f | 1747 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1748 | memset(object, 0, s->objsize); |
d07dbea4 | 1749 | |
c016b0bd | 1750 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1751 | |
894b8788 | 1752 | return object; |
81819f0f CL |
1753 | } |
1754 | ||
1755 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1756 | { | |
2154a336 | 1757 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1758 | |
ca2b84cb | 1759 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1760 | |
1761 | return ret; | |
81819f0f CL |
1762 | } |
1763 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1764 | ||
0f24f128 | 1765 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1766 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) |
1767 | { | |
2154a336 | 1768 | return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 EGM |
1769 | } |
1770 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1771 | #endif | |
1772 | ||
81819f0f CL |
1773 | #ifdef CONFIG_NUMA |
1774 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1775 | { | |
5b882be4 EGM |
1776 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1777 | ||
ca2b84cb EGM |
1778 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1779 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1780 | |
1781 | return ret; | |
81819f0f CL |
1782 | } |
1783 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1784 | #endif | |
1785 | ||
0f24f128 | 1786 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1787 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, |
1788 | gfp_t gfpflags, | |
1789 | int node) | |
1790 | { | |
1791 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1792 | } | |
1793 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1794 | #endif | |
1795 | ||
81819f0f | 1796 | /* |
894b8788 CL |
1797 | * Slow patch handling. This may still be called frequently since objects |
1798 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1799 | * |
894b8788 CL |
1800 | * So we still attempt to reduce cache line usage. Just take the slab |
1801 | * lock and free the item. If there is no additional partial page | |
1802 | * handling required then we can return immediately. | |
81819f0f | 1803 | */ |
894b8788 | 1804 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1805 | void *x, unsigned long addr) |
81819f0f CL |
1806 | { |
1807 | void *prior; | |
1808 | void **object = (void *)x; | |
81819f0f | 1809 | |
84e554e6 | 1810 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1811 | slab_lock(page); |
1812 | ||
af537b0a | 1813 | if (kmem_cache_debug(s)) |
81819f0f | 1814 | goto debug; |
6446faa2 | 1815 | |
81819f0f | 1816 | checks_ok: |
ff12059e CL |
1817 | prior = page->freelist; |
1818 | set_freepointer(s, object, prior); | |
81819f0f CL |
1819 | page->freelist = object; |
1820 | page->inuse--; | |
1821 | ||
8a38082d | 1822 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1823 | stat(s, FREE_FROZEN); |
81819f0f | 1824 | goto out_unlock; |
8ff12cfc | 1825 | } |
81819f0f CL |
1826 | |
1827 | if (unlikely(!page->inuse)) | |
1828 | goto slab_empty; | |
1829 | ||
1830 | /* | |
6446faa2 | 1831 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1832 | * then add it. |
1833 | */ | |
a973e9dd | 1834 | if (unlikely(!prior)) { |
7c2e132c | 1835 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1836 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1837 | } |
81819f0f CL |
1838 | |
1839 | out_unlock: | |
1840 | slab_unlock(page); | |
81819f0f CL |
1841 | return; |
1842 | ||
1843 | slab_empty: | |
a973e9dd | 1844 | if (prior) { |
81819f0f | 1845 | /* |
672bba3a | 1846 | * Slab still on the partial list. |
81819f0f CL |
1847 | */ |
1848 | remove_partial(s, page); | |
84e554e6 | 1849 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1850 | } |
81819f0f | 1851 | slab_unlock(page); |
84e554e6 | 1852 | stat(s, FREE_SLAB); |
81819f0f | 1853 | discard_slab(s, page); |
81819f0f CL |
1854 | return; |
1855 | ||
1856 | debug: | |
3ec09742 | 1857 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1858 | goto out_unlock; |
77c5e2d0 | 1859 | goto checks_ok; |
81819f0f CL |
1860 | } |
1861 | ||
894b8788 CL |
1862 | /* |
1863 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1864 | * can perform fastpath freeing without additional function calls. | |
1865 | * | |
1866 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1867 | * of this processor. This typically the case if we have just allocated | |
1868 | * the item before. | |
1869 | * | |
1870 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1871 | * with all sorts of special processing. | |
1872 | */ | |
06428780 | 1873 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1874 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1875 | { |
1876 | void **object = (void *)x; | |
dfb4f096 | 1877 | struct kmem_cache_cpu *c; |
1f84260c CL |
1878 | unsigned long flags; |
1879 | ||
c016b0bd CL |
1880 | slab_free_hook(s, x); |
1881 | ||
894b8788 | 1882 | local_irq_save(flags); |
9dfc6e68 | 1883 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd CL |
1884 | |
1885 | slab_free_hook_irq(s, x); | |
1886 | ||
ee3c72a1 | 1887 | if (likely(page == c->page && c->node >= 0)) { |
ff12059e | 1888 | set_freepointer(s, object, c->freelist); |
dfb4f096 | 1889 | c->freelist = object; |
84e554e6 | 1890 | stat(s, FREE_FASTPATH); |
894b8788 | 1891 | } else |
ff12059e | 1892 | __slab_free(s, page, x, addr); |
894b8788 CL |
1893 | |
1894 | local_irq_restore(flags); | |
1895 | } | |
1896 | ||
81819f0f CL |
1897 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1898 | { | |
77c5e2d0 | 1899 | struct page *page; |
81819f0f | 1900 | |
b49af68f | 1901 | page = virt_to_head_page(x); |
81819f0f | 1902 | |
ce71e27c | 1903 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1904 | |
ca2b84cb | 1905 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1906 | } |
1907 | EXPORT_SYMBOL(kmem_cache_free); | |
1908 | ||
e9beef18 | 1909 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1910 | static struct page *get_object_page(const void *x) |
1911 | { | |
b49af68f | 1912 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1913 | |
1914 | if (!PageSlab(page)) | |
1915 | return NULL; | |
1916 | ||
1917 | return page; | |
1918 | } | |
1919 | ||
1920 | /* | |
672bba3a CL |
1921 | * Object placement in a slab is made very easy because we always start at |
1922 | * offset 0. If we tune the size of the object to the alignment then we can | |
1923 | * get the required alignment by putting one properly sized object after | |
1924 | * another. | |
81819f0f CL |
1925 | * |
1926 | * Notice that the allocation order determines the sizes of the per cpu | |
1927 | * caches. Each processor has always one slab available for allocations. | |
1928 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1929 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1930 | * locking overhead. |
81819f0f CL |
1931 | */ |
1932 | ||
1933 | /* | |
1934 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1935 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1936 | * and increases the number of allocations possible without having to | |
1937 | * take the list_lock. | |
1938 | */ | |
1939 | static int slub_min_order; | |
114e9e89 | 1940 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1941 | static int slub_min_objects; |
81819f0f CL |
1942 | |
1943 | /* | |
1944 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1945 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1946 | */ |
1947 | static int slub_nomerge; | |
1948 | ||
81819f0f CL |
1949 | /* |
1950 | * Calculate the order of allocation given an slab object size. | |
1951 | * | |
672bba3a CL |
1952 | * The order of allocation has significant impact on performance and other |
1953 | * system components. Generally order 0 allocations should be preferred since | |
1954 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1955 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1956 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1957 | * would be wasted. |
1958 | * | |
1959 | * In order to reach satisfactory performance we must ensure that a minimum | |
1960 | * number of objects is in one slab. Otherwise we may generate too much | |
1961 | * activity on the partial lists which requires taking the list_lock. This is | |
1962 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1963 | * |
672bba3a CL |
1964 | * slub_max_order specifies the order where we begin to stop considering the |
1965 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1966 | * we try to keep the page order as low as possible. So we accept more waste | |
1967 | * of space in favor of a small page order. | |
81819f0f | 1968 | * |
672bba3a CL |
1969 | * Higher order allocations also allow the placement of more objects in a |
1970 | * slab and thereby reduce object handling overhead. If the user has | |
1971 | * requested a higher mininum order then we start with that one instead of | |
1972 | * the smallest order which will fit the object. | |
81819f0f | 1973 | */ |
5e6d444e CL |
1974 | static inline int slab_order(int size, int min_objects, |
1975 | int max_order, int fract_leftover) | |
81819f0f CL |
1976 | { |
1977 | int order; | |
1978 | int rem; | |
6300ea75 | 1979 | int min_order = slub_min_order; |
81819f0f | 1980 | |
210b5c06 CG |
1981 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1982 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1983 | |
6300ea75 | 1984 | for (order = max(min_order, |
5e6d444e CL |
1985 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1986 | order <= max_order; order++) { | |
81819f0f | 1987 | |
5e6d444e | 1988 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1989 | |
5e6d444e | 1990 | if (slab_size < min_objects * size) |
81819f0f CL |
1991 | continue; |
1992 | ||
1993 | rem = slab_size % size; | |
1994 | ||
5e6d444e | 1995 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1996 | break; |
1997 | ||
1998 | } | |
672bba3a | 1999 | |
81819f0f CL |
2000 | return order; |
2001 | } | |
2002 | ||
5e6d444e CL |
2003 | static inline int calculate_order(int size) |
2004 | { | |
2005 | int order; | |
2006 | int min_objects; | |
2007 | int fraction; | |
e8120ff1 | 2008 | int max_objects; |
5e6d444e CL |
2009 | |
2010 | /* | |
2011 | * Attempt to find best configuration for a slab. This | |
2012 | * works by first attempting to generate a layout with | |
2013 | * the best configuration and backing off gradually. | |
2014 | * | |
2015 | * First we reduce the acceptable waste in a slab. Then | |
2016 | * we reduce the minimum objects required in a slab. | |
2017 | */ | |
2018 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2019 | if (!min_objects) |
2020 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
2021 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
2022 | min_objects = min(min_objects, max_objects); | |
2023 | ||
5e6d444e | 2024 | while (min_objects > 1) { |
c124f5b5 | 2025 | fraction = 16; |
5e6d444e CL |
2026 | while (fraction >= 4) { |
2027 | order = slab_order(size, min_objects, | |
2028 | slub_max_order, fraction); | |
2029 | if (order <= slub_max_order) | |
2030 | return order; | |
2031 | fraction /= 2; | |
2032 | } | |
5086c389 | 2033 | min_objects--; |
5e6d444e CL |
2034 | } |
2035 | ||
2036 | /* | |
2037 | * We were unable to place multiple objects in a slab. Now | |
2038 | * lets see if we can place a single object there. | |
2039 | */ | |
2040 | order = slab_order(size, 1, slub_max_order, 1); | |
2041 | if (order <= slub_max_order) | |
2042 | return order; | |
2043 | ||
2044 | /* | |
2045 | * Doh this slab cannot be placed using slub_max_order. | |
2046 | */ | |
2047 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2048 | if (order < MAX_ORDER) |
5e6d444e CL |
2049 | return order; |
2050 | return -ENOSYS; | |
2051 | } | |
2052 | ||
81819f0f | 2053 | /* |
672bba3a | 2054 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2055 | */ |
2056 | static unsigned long calculate_alignment(unsigned long flags, | |
2057 | unsigned long align, unsigned long size) | |
2058 | { | |
2059 | /* | |
6446faa2 CL |
2060 | * If the user wants hardware cache aligned objects then follow that |
2061 | * suggestion if the object is sufficiently large. | |
81819f0f | 2062 | * |
6446faa2 CL |
2063 | * The hardware cache alignment cannot override the specified |
2064 | * alignment though. If that is greater then use it. | |
81819f0f | 2065 | */ |
b6210386 NP |
2066 | if (flags & SLAB_HWCACHE_ALIGN) { |
2067 | unsigned long ralign = cache_line_size(); | |
2068 | while (size <= ralign / 2) | |
2069 | ralign /= 2; | |
2070 | align = max(align, ralign); | |
2071 | } | |
81819f0f CL |
2072 | |
2073 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2074 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2075 | |
2076 | return ALIGN(align, sizeof(void *)); | |
2077 | } | |
2078 | ||
5595cffc PE |
2079 | static void |
2080 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2081 | { |
2082 | n->nr_partial = 0; | |
81819f0f CL |
2083 | spin_lock_init(&n->list_lock); |
2084 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2085 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2086 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2087 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2088 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2089 | #endif |
81819f0f CL |
2090 | } |
2091 | ||
55136592 | 2092 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2093 | { |
6c182dc0 CL |
2094 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2095 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2096 | |
6c182dc0 | 2097 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); |
4c93c355 | 2098 | |
6c182dc0 | 2099 | return s->cpu_slab != NULL; |
4c93c355 | 2100 | } |
4c93c355 | 2101 | |
81819f0f | 2102 | #ifdef CONFIG_NUMA |
51df1142 CL |
2103 | static struct kmem_cache *kmem_cache_node; |
2104 | ||
81819f0f CL |
2105 | /* |
2106 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2107 | * slab on the node for this slabcache. There are no concurrent accesses | |
2108 | * possible. | |
2109 | * | |
2110 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2111 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2112 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2113 | */ |
55136592 | 2114 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2115 | { |
2116 | struct page *page; | |
2117 | struct kmem_cache_node *n; | |
ba84c73c | 2118 | unsigned long flags; |
81819f0f | 2119 | |
51df1142 | 2120 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2121 | |
51df1142 | 2122 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2123 | |
2124 | BUG_ON(!page); | |
a2f92ee7 CL |
2125 | if (page_to_nid(page) != node) { |
2126 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2127 | "node %d\n", node); | |
2128 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2129 | "in order to be able to continue\n"); | |
2130 | } | |
2131 | ||
81819f0f CL |
2132 | n = page->freelist; |
2133 | BUG_ON(!n); | |
51df1142 | 2134 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2135 | page->inuse++; |
51df1142 | 2136 | kmem_cache_node->node[node] = n; |
8ab1372f | 2137 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
2138 | init_object(kmem_cache_node, n, 1); |
2139 | init_tracking(kmem_cache_node, n); | |
8ab1372f | 2140 | #endif |
51df1142 CL |
2141 | init_kmem_cache_node(n, kmem_cache_node); |
2142 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2143 | |
ba84c73c | 2144 | /* |
2145 | * lockdep requires consistent irq usage for each lock | |
2146 | * so even though there cannot be a race this early in | |
2147 | * the boot sequence, we still disable irqs. | |
2148 | */ | |
2149 | local_irq_save(flags); | |
7c2e132c | 2150 | add_partial(n, page, 0); |
ba84c73c | 2151 | local_irq_restore(flags); |
81819f0f CL |
2152 | } |
2153 | ||
2154 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2155 | { | |
2156 | int node; | |
2157 | ||
f64dc58c | 2158 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2159 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2160 | |
73367bd8 | 2161 | if (n) |
51df1142 CL |
2162 | kmem_cache_free(kmem_cache_node, n); |
2163 | ||
81819f0f CL |
2164 | s->node[node] = NULL; |
2165 | } | |
2166 | } | |
2167 | ||
55136592 | 2168 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2169 | { |
2170 | int node; | |
81819f0f | 2171 | |
f64dc58c | 2172 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2173 | struct kmem_cache_node *n; |
2174 | ||
73367bd8 | 2175 | if (slab_state == DOWN) { |
55136592 | 2176 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2177 | continue; |
2178 | } | |
51df1142 | 2179 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2180 | GFP_KERNEL, node); |
81819f0f | 2181 | |
73367bd8 AD |
2182 | if (!n) { |
2183 | free_kmem_cache_nodes(s); | |
2184 | return 0; | |
81819f0f | 2185 | } |
73367bd8 | 2186 | |
81819f0f | 2187 | s->node[node] = n; |
5595cffc | 2188 | init_kmem_cache_node(n, s); |
81819f0f CL |
2189 | } |
2190 | return 1; | |
2191 | } | |
2192 | #else | |
2193 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2194 | { | |
2195 | } | |
2196 | ||
55136592 | 2197 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f | 2198 | { |
5595cffc | 2199 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2200 | return 1; |
2201 | } | |
2202 | #endif | |
2203 | ||
c0bdb232 | 2204 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2205 | { |
2206 | if (min < MIN_PARTIAL) | |
2207 | min = MIN_PARTIAL; | |
2208 | else if (min > MAX_PARTIAL) | |
2209 | min = MAX_PARTIAL; | |
2210 | s->min_partial = min; | |
2211 | } | |
2212 | ||
81819f0f CL |
2213 | /* |
2214 | * calculate_sizes() determines the order and the distribution of data within | |
2215 | * a slab object. | |
2216 | */ | |
06b285dc | 2217 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2218 | { |
2219 | unsigned long flags = s->flags; | |
2220 | unsigned long size = s->objsize; | |
2221 | unsigned long align = s->align; | |
834f3d11 | 2222 | int order; |
81819f0f | 2223 | |
d8b42bf5 CL |
2224 | /* |
2225 | * Round up object size to the next word boundary. We can only | |
2226 | * place the free pointer at word boundaries and this determines | |
2227 | * the possible location of the free pointer. | |
2228 | */ | |
2229 | size = ALIGN(size, sizeof(void *)); | |
2230 | ||
2231 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2232 | /* |
2233 | * Determine if we can poison the object itself. If the user of | |
2234 | * the slab may touch the object after free or before allocation | |
2235 | * then we should never poison the object itself. | |
2236 | */ | |
2237 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2238 | !s->ctor) |
81819f0f CL |
2239 | s->flags |= __OBJECT_POISON; |
2240 | else | |
2241 | s->flags &= ~__OBJECT_POISON; | |
2242 | ||
81819f0f CL |
2243 | |
2244 | /* | |
672bba3a | 2245 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2246 | * end of the object and the free pointer. If not then add an |
672bba3a | 2247 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2248 | */ |
2249 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2250 | size += sizeof(void *); | |
41ecc55b | 2251 | #endif |
81819f0f CL |
2252 | |
2253 | /* | |
672bba3a CL |
2254 | * With that we have determined the number of bytes in actual use |
2255 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2256 | */ |
2257 | s->inuse = size; | |
2258 | ||
2259 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2260 | s->ctor)) { |
81819f0f CL |
2261 | /* |
2262 | * Relocate free pointer after the object if it is not | |
2263 | * permitted to overwrite the first word of the object on | |
2264 | * kmem_cache_free. | |
2265 | * | |
2266 | * This is the case if we do RCU, have a constructor or | |
2267 | * destructor or are poisoning the objects. | |
2268 | */ | |
2269 | s->offset = size; | |
2270 | size += sizeof(void *); | |
2271 | } | |
2272 | ||
c12b3c62 | 2273 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2274 | if (flags & SLAB_STORE_USER) |
2275 | /* | |
2276 | * Need to store information about allocs and frees after | |
2277 | * the object. | |
2278 | */ | |
2279 | size += 2 * sizeof(struct track); | |
2280 | ||
be7b3fbc | 2281 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2282 | /* |
2283 | * Add some empty padding so that we can catch | |
2284 | * overwrites from earlier objects rather than let | |
2285 | * tracking information or the free pointer be | |
0211a9c8 | 2286 | * corrupted if a user writes before the start |
81819f0f CL |
2287 | * of the object. |
2288 | */ | |
2289 | size += sizeof(void *); | |
41ecc55b | 2290 | #endif |
672bba3a | 2291 | |
81819f0f CL |
2292 | /* |
2293 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2294 | * user specified and the dynamic determination of cache line size |
2295 | * on bootup. | |
81819f0f CL |
2296 | */ |
2297 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2298 | s->align = align; |
81819f0f CL |
2299 | |
2300 | /* | |
2301 | * SLUB stores one object immediately after another beginning from | |
2302 | * offset 0. In order to align the objects we have to simply size | |
2303 | * each object to conform to the alignment. | |
2304 | */ | |
2305 | size = ALIGN(size, align); | |
2306 | s->size = size; | |
06b285dc CL |
2307 | if (forced_order >= 0) |
2308 | order = forced_order; | |
2309 | else | |
2310 | order = calculate_order(size); | |
81819f0f | 2311 | |
834f3d11 | 2312 | if (order < 0) |
81819f0f CL |
2313 | return 0; |
2314 | ||
b7a49f0d | 2315 | s->allocflags = 0; |
834f3d11 | 2316 | if (order) |
b7a49f0d CL |
2317 | s->allocflags |= __GFP_COMP; |
2318 | ||
2319 | if (s->flags & SLAB_CACHE_DMA) | |
2320 | s->allocflags |= SLUB_DMA; | |
2321 | ||
2322 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2323 | s->allocflags |= __GFP_RECLAIMABLE; | |
2324 | ||
81819f0f CL |
2325 | /* |
2326 | * Determine the number of objects per slab | |
2327 | */ | |
834f3d11 | 2328 | s->oo = oo_make(order, size); |
65c3376a | 2329 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2330 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2331 | s->max = s->oo; | |
81819f0f | 2332 | |
834f3d11 | 2333 | return !!oo_objects(s->oo); |
81819f0f CL |
2334 | |
2335 | } | |
2336 | ||
55136592 | 2337 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2338 | const char *name, size_t size, |
2339 | size_t align, unsigned long flags, | |
51cc5068 | 2340 | void (*ctor)(void *)) |
81819f0f CL |
2341 | { |
2342 | memset(s, 0, kmem_size); | |
2343 | s->name = name; | |
2344 | s->ctor = ctor; | |
81819f0f | 2345 | s->objsize = size; |
81819f0f | 2346 | s->align = align; |
ba0268a8 | 2347 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2348 | |
06b285dc | 2349 | if (!calculate_sizes(s, -1)) |
81819f0f | 2350 | goto error; |
3de47213 DR |
2351 | if (disable_higher_order_debug) { |
2352 | /* | |
2353 | * Disable debugging flags that store metadata if the min slab | |
2354 | * order increased. | |
2355 | */ | |
2356 | if (get_order(s->size) > get_order(s->objsize)) { | |
2357 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2358 | s->offset = 0; | |
2359 | if (!calculate_sizes(s, -1)) | |
2360 | goto error; | |
2361 | } | |
2362 | } | |
81819f0f | 2363 | |
3b89d7d8 DR |
2364 | /* |
2365 | * The larger the object size is, the more pages we want on the partial | |
2366 | * list to avoid pounding the page allocator excessively. | |
2367 | */ | |
c0bdb232 | 2368 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2369 | s->refcount = 1; |
2370 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2371 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2372 | #endif |
55136592 | 2373 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2374 | goto error; |
81819f0f | 2375 | |
55136592 | 2376 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2377 | return 1; |
ff12059e | 2378 | |
4c93c355 | 2379 | free_kmem_cache_nodes(s); |
81819f0f CL |
2380 | error: |
2381 | if (flags & SLAB_PANIC) | |
2382 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2383 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2384 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2385 | s->offset, flags); |
2386 | return 0; | |
2387 | } | |
81819f0f CL |
2388 | |
2389 | /* | |
2390 | * Check if a given pointer is valid | |
2391 | */ | |
2392 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2393 | { | |
06428780 | 2394 | struct page *page; |
81819f0f | 2395 | |
d3e06e2b PE |
2396 | if (!kern_ptr_validate(object, s->size)) |
2397 | return 0; | |
2398 | ||
81819f0f CL |
2399 | page = get_object_page(object); |
2400 | ||
2401 | if (!page || s != page->slab) | |
2402 | /* No slab or wrong slab */ | |
2403 | return 0; | |
2404 | ||
abcd08a6 | 2405 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2406 | return 0; |
2407 | ||
2408 | /* | |
2409 | * We could also check if the object is on the slabs freelist. | |
2410 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2411 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2412 | * to a certain slab. |
2413 | */ | |
2414 | return 1; | |
2415 | } | |
2416 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2417 | ||
2418 | /* | |
2419 | * Determine the size of a slab object | |
2420 | */ | |
2421 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2422 | { | |
2423 | return s->objsize; | |
2424 | } | |
2425 | EXPORT_SYMBOL(kmem_cache_size); | |
2426 | ||
2427 | const char *kmem_cache_name(struct kmem_cache *s) | |
2428 | { | |
2429 | return s->name; | |
2430 | } | |
2431 | EXPORT_SYMBOL(kmem_cache_name); | |
2432 | ||
33b12c38 CL |
2433 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2434 | const char *text) | |
2435 | { | |
2436 | #ifdef CONFIG_SLUB_DEBUG | |
2437 | void *addr = page_address(page); | |
2438 | void *p; | |
bbd7d57b ED |
2439 | long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long), |
2440 | GFP_ATOMIC); | |
33b12c38 | 2441 | |
bbd7d57b ED |
2442 | if (!map) |
2443 | return; | |
33b12c38 CL |
2444 | slab_err(s, page, "%s", text); |
2445 | slab_lock(page); | |
2446 | for_each_free_object(p, s, page->freelist) | |
2447 | set_bit(slab_index(p, s, addr), map); | |
2448 | ||
2449 | for_each_object(p, s, addr, page->objects) { | |
2450 | ||
2451 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2452 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2453 | p, p - addr); | |
2454 | print_tracking(s, p); | |
2455 | } | |
2456 | } | |
2457 | slab_unlock(page); | |
bbd7d57b | 2458 | kfree(map); |
33b12c38 CL |
2459 | #endif |
2460 | } | |
2461 | ||
81819f0f | 2462 | /* |
599870b1 | 2463 | * Attempt to free all partial slabs on a node. |
81819f0f | 2464 | */ |
599870b1 | 2465 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2466 | { |
81819f0f CL |
2467 | unsigned long flags; |
2468 | struct page *page, *h; | |
2469 | ||
2470 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2471 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2472 | if (!page->inuse) { |
2473 | list_del(&page->lru); | |
2474 | discard_slab(s, page); | |
599870b1 | 2475 | n->nr_partial--; |
33b12c38 CL |
2476 | } else { |
2477 | list_slab_objects(s, page, | |
2478 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2479 | } |
33b12c38 | 2480 | } |
81819f0f | 2481 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2482 | } |
2483 | ||
2484 | /* | |
672bba3a | 2485 | * Release all resources used by a slab cache. |
81819f0f | 2486 | */ |
0c710013 | 2487 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2488 | { |
2489 | int node; | |
2490 | ||
2491 | flush_all(s); | |
9dfc6e68 | 2492 | free_percpu(s->cpu_slab); |
81819f0f | 2493 | /* Attempt to free all objects */ |
f64dc58c | 2494 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2495 | struct kmem_cache_node *n = get_node(s, node); |
2496 | ||
599870b1 CL |
2497 | free_partial(s, n); |
2498 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2499 | return 1; |
2500 | } | |
2501 | free_kmem_cache_nodes(s); | |
2502 | return 0; | |
2503 | } | |
2504 | ||
2505 | /* | |
2506 | * Close a cache and release the kmem_cache structure | |
2507 | * (must be used for caches created using kmem_cache_create) | |
2508 | */ | |
2509 | void kmem_cache_destroy(struct kmem_cache *s) | |
2510 | { | |
2511 | down_write(&slub_lock); | |
2512 | s->refcount--; | |
2513 | if (!s->refcount) { | |
2514 | list_del(&s->list); | |
d629d819 PE |
2515 | if (kmem_cache_close(s)) { |
2516 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2517 | "still has objects.\n", s->name, __func__); | |
2518 | dump_stack(); | |
2519 | } | |
d76b1590 ED |
2520 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2521 | rcu_barrier(); | |
81819f0f | 2522 | sysfs_slab_remove(s); |
2bce6485 CL |
2523 | } |
2524 | up_write(&slub_lock); | |
81819f0f CL |
2525 | } |
2526 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2527 | ||
2528 | /******************************************************************** | |
2529 | * Kmalloc subsystem | |
2530 | *******************************************************************/ | |
2531 | ||
51df1142 | 2532 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2533 | EXPORT_SYMBOL(kmalloc_caches); |
2534 | ||
51df1142 CL |
2535 | static struct kmem_cache *kmem_cache; |
2536 | ||
55136592 | 2537 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2538 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2539 | #endif |
2540 | ||
81819f0f CL |
2541 | static int __init setup_slub_min_order(char *str) |
2542 | { | |
06428780 | 2543 | get_option(&str, &slub_min_order); |
81819f0f CL |
2544 | |
2545 | return 1; | |
2546 | } | |
2547 | ||
2548 | __setup("slub_min_order=", setup_slub_min_order); | |
2549 | ||
2550 | static int __init setup_slub_max_order(char *str) | |
2551 | { | |
06428780 | 2552 | get_option(&str, &slub_max_order); |
818cf590 | 2553 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2554 | |
2555 | return 1; | |
2556 | } | |
2557 | ||
2558 | __setup("slub_max_order=", setup_slub_max_order); | |
2559 | ||
2560 | static int __init setup_slub_min_objects(char *str) | |
2561 | { | |
06428780 | 2562 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2563 | |
2564 | return 1; | |
2565 | } | |
2566 | ||
2567 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2568 | ||
2569 | static int __init setup_slub_nomerge(char *str) | |
2570 | { | |
2571 | slub_nomerge = 1; | |
2572 | return 1; | |
2573 | } | |
2574 | ||
2575 | __setup("slub_nomerge", setup_slub_nomerge); | |
2576 | ||
51df1142 CL |
2577 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2578 | int size, unsigned int flags) | |
81819f0f | 2579 | { |
51df1142 CL |
2580 | struct kmem_cache *s; |
2581 | ||
2582 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2583 | ||
83b519e8 PE |
2584 | /* |
2585 | * This function is called with IRQs disabled during early-boot on | |
2586 | * single CPU so there's no need to take slub_lock here. | |
2587 | */ | |
55136592 | 2588 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2589 | flags, NULL)) |
81819f0f CL |
2590 | goto panic; |
2591 | ||
2592 | list_add(&s->list, &slab_caches); | |
51df1142 | 2593 | return s; |
81819f0f CL |
2594 | |
2595 | panic: | |
2596 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2597 | return NULL; |
81819f0f CL |
2598 | } |
2599 | ||
f1b26339 CL |
2600 | /* |
2601 | * Conversion table for small slabs sizes / 8 to the index in the | |
2602 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2603 | * of two cache sizes there. The size of larger slabs can be determined using | |
2604 | * fls. | |
2605 | */ | |
2606 | static s8 size_index[24] = { | |
2607 | 3, /* 8 */ | |
2608 | 4, /* 16 */ | |
2609 | 5, /* 24 */ | |
2610 | 5, /* 32 */ | |
2611 | 6, /* 40 */ | |
2612 | 6, /* 48 */ | |
2613 | 6, /* 56 */ | |
2614 | 6, /* 64 */ | |
2615 | 1, /* 72 */ | |
2616 | 1, /* 80 */ | |
2617 | 1, /* 88 */ | |
2618 | 1, /* 96 */ | |
2619 | 7, /* 104 */ | |
2620 | 7, /* 112 */ | |
2621 | 7, /* 120 */ | |
2622 | 7, /* 128 */ | |
2623 | 2, /* 136 */ | |
2624 | 2, /* 144 */ | |
2625 | 2, /* 152 */ | |
2626 | 2, /* 160 */ | |
2627 | 2, /* 168 */ | |
2628 | 2, /* 176 */ | |
2629 | 2, /* 184 */ | |
2630 | 2 /* 192 */ | |
2631 | }; | |
2632 | ||
acdfcd04 AK |
2633 | static inline int size_index_elem(size_t bytes) |
2634 | { | |
2635 | return (bytes - 1) / 8; | |
2636 | } | |
2637 | ||
81819f0f CL |
2638 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2639 | { | |
f1b26339 | 2640 | int index; |
81819f0f | 2641 | |
f1b26339 CL |
2642 | if (size <= 192) { |
2643 | if (!size) | |
2644 | return ZERO_SIZE_PTR; | |
81819f0f | 2645 | |
acdfcd04 | 2646 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2647 | } else |
f1b26339 | 2648 | index = fls(size - 1); |
81819f0f CL |
2649 | |
2650 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2651 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2652 | return kmalloc_dma_caches[index]; |
f1b26339 | 2653 | |
81819f0f | 2654 | #endif |
51df1142 | 2655 | return kmalloc_caches[index]; |
81819f0f CL |
2656 | } |
2657 | ||
2658 | void *__kmalloc(size_t size, gfp_t flags) | |
2659 | { | |
aadb4bc4 | 2660 | struct kmem_cache *s; |
5b882be4 | 2661 | void *ret; |
81819f0f | 2662 | |
ffadd4d0 | 2663 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2664 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2665 | |
2666 | s = get_slab(size, flags); | |
2667 | ||
2668 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2669 | return s; |
2670 | ||
2154a336 | 2671 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2672 | |
ca2b84cb | 2673 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2674 | |
2675 | return ret; | |
81819f0f CL |
2676 | } |
2677 | EXPORT_SYMBOL(__kmalloc); | |
2678 | ||
f619cfe1 CL |
2679 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2680 | { | |
b1eeab67 | 2681 | struct page *page; |
e4f7c0b4 | 2682 | void *ptr = NULL; |
f619cfe1 | 2683 | |
b1eeab67 VN |
2684 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2685 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2686 | if (page) |
e4f7c0b4 CM |
2687 | ptr = page_address(page); |
2688 | ||
2689 | kmemleak_alloc(ptr, size, 1, flags); | |
2690 | return ptr; | |
f619cfe1 CL |
2691 | } |
2692 | ||
81819f0f CL |
2693 | #ifdef CONFIG_NUMA |
2694 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2695 | { | |
aadb4bc4 | 2696 | struct kmem_cache *s; |
5b882be4 | 2697 | void *ret; |
81819f0f | 2698 | |
057685cf | 2699 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2700 | ret = kmalloc_large_node(size, flags, node); |
2701 | ||
ca2b84cb EGM |
2702 | trace_kmalloc_node(_RET_IP_, ret, |
2703 | size, PAGE_SIZE << get_order(size), | |
2704 | flags, node); | |
5b882be4 EGM |
2705 | |
2706 | return ret; | |
2707 | } | |
aadb4bc4 CL |
2708 | |
2709 | s = get_slab(size, flags); | |
2710 | ||
2711 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2712 | return s; |
2713 | ||
5b882be4 EGM |
2714 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2715 | ||
ca2b84cb | 2716 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2717 | |
2718 | return ret; | |
81819f0f CL |
2719 | } |
2720 | EXPORT_SYMBOL(__kmalloc_node); | |
2721 | #endif | |
2722 | ||
2723 | size_t ksize(const void *object) | |
2724 | { | |
272c1d21 | 2725 | struct page *page; |
81819f0f CL |
2726 | struct kmem_cache *s; |
2727 | ||
ef8b4520 | 2728 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2729 | return 0; |
2730 | ||
294a80a8 | 2731 | page = virt_to_head_page(object); |
294a80a8 | 2732 | |
76994412 PE |
2733 | if (unlikely(!PageSlab(page))) { |
2734 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2735 | return PAGE_SIZE << compound_order(page); |
76994412 | 2736 | } |
81819f0f | 2737 | s = page->slab; |
81819f0f | 2738 | |
ae20bfda | 2739 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2740 | /* |
2741 | * Debugging requires use of the padding between object | |
2742 | * and whatever may come after it. | |
2743 | */ | |
2744 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2745 | return s->objsize; | |
2746 | ||
ae20bfda | 2747 | #endif |
81819f0f CL |
2748 | /* |
2749 | * If we have the need to store the freelist pointer | |
2750 | * back there or track user information then we can | |
2751 | * only use the space before that information. | |
2752 | */ | |
2753 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2754 | return s->inuse; | |
81819f0f CL |
2755 | /* |
2756 | * Else we can use all the padding etc for the allocation | |
2757 | */ | |
2758 | return s->size; | |
2759 | } | |
b1aabecd | 2760 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2761 | |
2762 | void kfree(const void *x) | |
2763 | { | |
81819f0f | 2764 | struct page *page; |
5bb983b0 | 2765 | void *object = (void *)x; |
81819f0f | 2766 | |
2121db74 PE |
2767 | trace_kfree(_RET_IP_, x); |
2768 | ||
2408c550 | 2769 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2770 | return; |
2771 | ||
b49af68f | 2772 | page = virt_to_head_page(x); |
aadb4bc4 | 2773 | if (unlikely(!PageSlab(page))) { |
0937502a | 2774 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2775 | kmemleak_free(x); |
aadb4bc4 CL |
2776 | put_page(page); |
2777 | return; | |
2778 | } | |
ce71e27c | 2779 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2780 | } |
2781 | EXPORT_SYMBOL(kfree); | |
2782 | ||
2086d26a | 2783 | /* |
672bba3a CL |
2784 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2785 | * the remaining slabs by the number of items in use. The slabs with the | |
2786 | * most items in use come first. New allocations will then fill those up | |
2787 | * and thus they can be removed from the partial lists. | |
2788 | * | |
2789 | * The slabs with the least items are placed last. This results in them | |
2790 | * being allocated from last increasing the chance that the last objects | |
2791 | * are freed in them. | |
2086d26a CL |
2792 | */ |
2793 | int kmem_cache_shrink(struct kmem_cache *s) | |
2794 | { | |
2795 | int node; | |
2796 | int i; | |
2797 | struct kmem_cache_node *n; | |
2798 | struct page *page; | |
2799 | struct page *t; | |
205ab99d | 2800 | int objects = oo_objects(s->max); |
2086d26a | 2801 | struct list_head *slabs_by_inuse = |
834f3d11 | 2802 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2803 | unsigned long flags; |
2804 | ||
2805 | if (!slabs_by_inuse) | |
2806 | return -ENOMEM; | |
2807 | ||
2808 | flush_all(s); | |
f64dc58c | 2809 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2810 | n = get_node(s, node); |
2811 | ||
2812 | if (!n->nr_partial) | |
2813 | continue; | |
2814 | ||
834f3d11 | 2815 | for (i = 0; i < objects; i++) |
2086d26a CL |
2816 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2817 | ||
2818 | spin_lock_irqsave(&n->list_lock, flags); | |
2819 | ||
2820 | /* | |
672bba3a | 2821 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2822 | * |
672bba3a CL |
2823 | * Note that concurrent frees may occur while we hold the |
2824 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2825 | */ |
2826 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2827 | if (!page->inuse && slab_trylock(page)) { | |
2828 | /* | |
2829 | * Must hold slab lock here because slab_free | |
2830 | * may have freed the last object and be | |
2831 | * waiting to release the slab. | |
2832 | */ | |
2833 | list_del(&page->lru); | |
2834 | n->nr_partial--; | |
2835 | slab_unlock(page); | |
2836 | discard_slab(s, page); | |
2837 | } else { | |
fcda3d89 CL |
2838 | list_move(&page->lru, |
2839 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2840 | } |
2841 | } | |
2842 | ||
2086d26a | 2843 | /* |
672bba3a CL |
2844 | * Rebuild the partial list with the slabs filled up most |
2845 | * first and the least used slabs at the end. | |
2086d26a | 2846 | */ |
834f3d11 | 2847 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2848 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2849 | ||
2086d26a CL |
2850 | spin_unlock_irqrestore(&n->list_lock, flags); |
2851 | } | |
2852 | ||
2853 | kfree(slabs_by_inuse); | |
2854 | return 0; | |
2855 | } | |
2856 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2857 | ||
b9049e23 YG |
2858 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2859 | static int slab_mem_going_offline_callback(void *arg) | |
2860 | { | |
2861 | struct kmem_cache *s; | |
2862 | ||
2863 | down_read(&slub_lock); | |
2864 | list_for_each_entry(s, &slab_caches, list) | |
2865 | kmem_cache_shrink(s); | |
2866 | up_read(&slub_lock); | |
2867 | ||
2868 | return 0; | |
2869 | } | |
2870 | ||
2871 | static void slab_mem_offline_callback(void *arg) | |
2872 | { | |
2873 | struct kmem_cache_node *n; | |
2874 | struct kmem_cache *s; | |
2875 | struct memory_notify *marg = arg; | |
2876 | int offline_node; | |
2877 | ||
2878 | offline_node = marg->status_change_nid; | |
2879 | ||
2880 | /* | |
2881 | * If the node still has available memory. we need kmem_cache_node | |
2882 | * for it yet. | |
2883 | */ | |
2884 | if (offline_node < 0) | |
2885 | return; | |
2886 | ||
2887 | down_read(&slub_lock); | |
2888 | list_for_each_entry(s, &slab_caches, list) { | |
2889 | n = get_node(s, offline_node); | |
2890 | if (n) { | |
2891 | /* | |
2892 | * if n->nr_slabs > 0, slabs still exist on the node | |
2893 | * that is going down. We were unable to free them, | |
c9404c9c | 2894 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2895 | * callback. So, we must fail. |
2896 | */ | |
0f389ec6 | 2897 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2898 | |
2899 | s->node[offline_node] = NULL; | |
2900 | kmem_cache_free(kmalloc_caches, n); | |
2901 | } | |
2902 | } | |
2903 | up_read(&slub_lock); | |
2904 | } | |
2905 | ||
2906 | static int slab_mem_going_online_callback(void *arg) | |
2907 | { | |
2908 | struct kmem_cache_node *n; | |
2909 | struct kmem_cache *s; | |
2910 | struct memory_notify *marg = arg; | |
2911 | int nid = marg->status_change_nid; | |
2912 | int ret = 0; | |
2913 | ||
2914 | /* | |
2915 | * If the node's memory is already available, then kmem_cache_node is | |
2916 | * already created. Nothing to do. | |
2917 | */ | |
2918 | if (nid < 0) | |
2919 | return 0; | |
2920 | ||
2921 | /* | |
0121c619 | 2922 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2923 | * allocate a kmem_cache_node structure in order to bring the node |
2924 | * online. | |
2925 | */ | |
2926 | down_read(&slub_lock); | |
2927 | list_for_each_entry(s, &slab_caches, list) { | |
2928 | /* | |
2929 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2930 | * since memory is not yet available from the node that | |
2931 | * is brought up. | |
2932 | */ | |
2933 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2934 | if (!n) { | |
2935 | ret = -ENOMEM; | |
2936 | goto out; | |
2937 | } | |
5595cffc | 2938 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2939 | s->node[nid] = n; |
2940 | } | |
2941 | out: | |
2942 | up_read(&slub_lock); | |
2943 | return ret; | |
2944 | } | |
2945 | ||
2946 | static int slab_memory_callback(struct notifier_block *self, | |
2947 | unsigned long action, void *arg) | |
2948 | { | |
2949 | int ret = 0; | |
2950 | ||
2951 | switch (action) { | |
2952 | case MEM_GOING_ONLINE: | |
2953 | ret = slab_mem_going_online_callback(arg); | |
2954 | break; | |
2955 | case MEM_GOING_OFFLINE: | |
2956 | ret = slab_mem_going_offline_callback(arg); | |
2957 | break; | |
2958 | case MEM_OFFLINE: | |
2959 | case MEM_CANCEL_ONLINE: | |
2960 | slab_mem_offline_callback(arg); | |
2961 | break; | |
2962 | case MEM_ONLINE: | |
2963 | case MEM_CANCEL_OFFLINE: | |
2964 | break; | |
2965 | } | |
dc19f9db KH |
2966 | if (ret) |
2967 | ret = notifier_from_errno(ret); | |
2968 | else | |
2969 | ret = NOTIFY_OK; | |
b9049e23 YG |
2970 | return ret; |
2971 | } | |
2972 | ||
2973 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2974 | ||
81819f0f CL |
2975 | /******************************************************************** |
2976 | * Basic setup of slabs | |
2977 | *******************************************************************/ | |
2978 | ||
51df1142 CL |
2979 | /* |
2980 | * Used for early kmem_cache structures that were allocated using | |
2981 | * the page allocator | |
2982 | */ | |
2983 | ||
2984 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
2985 | { | |
2986 | int node; | |
2987 | ||
2988 | list_add(&s->list, &slab_caches); | |
2989 | s->refcount = -1; | |
2990 | ||
2991 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
2992 | struct kmem_cache_node *n = get_node(s, node); | |
2993 | struct page *p; | |
2994 | ||
2995 | if (n) { | |
2996 | list_for_each_entry(p, &n->partial, lru) | |
2997 | p->slab = s; | |
2998 | ||
2999 | #ifdef CONFIG_SLAB_DEBUG | |
3000 | list_for_each_entry(p, &n->full, lru) | |
3001 | p->slab = s; | |
3002 | #endif | |
3003 | } | |
3004 | } | |
3005 | } | |
3006 | ||
81819f0f CL |
3007 | void __init kmem_cache_init(void) |
3008 | { | |
3009 | int i; | |
4b356be0 | 3010 | int caches = 0; |
51df1142 CL |
3011 | struct kmem_cache *temp_kmem_cache; |
3012 | int order; | |
81819f0f CL |
3013 | |
3014 | #ifdef CONFIG_NUMA | |
51df1142 CL |
3015 | struct kmem_cache *temp_kmem_cache_node; |
3016 | unsigned long kmalloc_size; | |
3017 | ||
3018 | kmem_size = offsetof(struct kmem_cache, node) + | |
3019 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3020 | ||
3021 | /* Allocate two kmem_caches from the page allocator */ | |
3022 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3023 | order = get_order(2 * kmalloc_size); | |
3024 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3025 | ||
81819f0f CL |
3026 | /* |
3027 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3028 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3029 | * kmem_cache_open for slab_state == DOWN. |
3030 | */ | |
51df1142 CL |
3031 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3032 | ||
3033 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3034 | sizeof(struct kmem_cache_node), | |
3035 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3036 | |
0c40ba4f | 3037 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
51df1142 CL |
3038 | #else |
3039 | /* Allocate a single kmem_cache from the page allocator */ | |
3040 | kmem_size = sizeof(struct kmem_cache); | |
3041 | order = get_order(kmem_size); | |
3042 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
81819f0f CL |
3043 | #endif |
3044 | ||
3045 | /* Able to allocate the per node structures */ | |
3046 | slab_state = PARTIAL; | |
3047 | ||
51df1142 CL |
3048 | temp_kmem_cache = kmem_cache; |
3049 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3050 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3051 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3052 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3053 | |
51df1142 CL |
3054 | #ifdef CONFIG_NUMA |
3055 | /* | |
3056 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3057 | * kmem_cache_node is separately allocated so no need to | |
3058 | * update any list pointers. | |
3059 | */ | |
3060 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3061 | |
51df1142 CL |
3062 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3063 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3064 | ||
3065 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3066 | ||
3067 | caches++; | |
3068 | #else | |
3069 | /* | |
3070 | * kmem_cache has kmem_cache_node embedded and we moved it! | |
3071 | * Update the list heads | |
3072 | */ | |
3073 | INIT_LIST_HEAD(&kmem_cache->local_node.partial); | |
3074 | list_splice(&temp_kmem_cache->local_node.partial, &kmem_cache->local_node.partial); | |
3075 | #ifdef CONFIG_SLUB_DEBUG | |
3076 | INIT_LIST_HEAD(&kmem_cache->local_node.full); | |
3077 | list_splice(&temp_kmem_cache->local_node.full, &kmem_cache->local_node.full); | |
3078 | #endif | |
3079 | #endif | |
3080 | kmem_cache_bootstrap_fixup(kmem_cache); | |
3081 | caches++; | |
3082 | /* Free temporary boot structure */ | |
3083 | free_pages((unsigned long)temp_kmem_cache, order); | |
3084 | ||
3085 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3086 | |
3087 | /* | |
3088 | * Patch up the size_index table if we have strange large alignment | |
3089 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3090 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3091 | * |
3092 | * Largest permitted alignment is 256 bytes due to the way we | |
3093 | * handle the index determination for the smaller caches. | |
3094 | * | |
3095 | * Make sure that nothing crazy happens if someone starts tinkering | |
3096 | * around with ARCH_KMALLOC_MINALIGN | |
3097 | */ | |
3098 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3099 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3100 | ||
acdfcd04 AK |
3101 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3102 | int elem = size_index_elem(i); | |
3103 | if (elem >= ARRAY_SIZE(size_index)) | |
3104 | break; | |
3105 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3106 | } | |
f1b26339 | 3107 | |
acdfcd04 AK |
3108 | if (KMALLOC_MIN_SIZE == 64) { |
3109 | /* | |
3110 | * The 96 byte size cache is not used if the alignment | |
3111 | * is 64 byte. | |
3112 | */ | |
3113 | for (i = 64 + 8; i <= 96; i += 8) | |
3114 | size_index[size_index_elem(i)] = 7; | |
3115 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3116 | /* |
3117 | * The 192 byte sized cache is not used if the alignment | |
3118 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3119 | * instead. | |
3120 | */ | |
3121 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3122 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3123 | } |
3124 | ||
51df1142 CL |
3125 | /* Caches that are not of the two-to-the-power-of size */ |
3126 | if (KMALLOC_MIN_SIZE <= 32) { | |
3127 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3128 | caches++; | |
3129 | } | |
3130 | ||
3131 | if (KMALLOC_MIN_SIZE <= 64) { | |
3132 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3133 | caches++; | |
3134 | } | |
3135 | ||
3136 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3137 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3138 | caches++; | |
3139 | } | |
3140 | ||
81819f0f CL |
3141 | slab_state = UP; |
3142 | ||
3143 | /* Provide the correct kmalloc names now that the caches are up */ | |
d7278bd7 CL |
3144 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3145 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3146 | ||
3147 | BUG_ON(!s); | |
51df1142 | 3148 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3149 | } |
81819f0f CL |
3150 | |
3151 | #ifdef CONFIG_SMP | |
3152 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3153 | #endif |
81819f0f | 3154 | |
55136592 | 3155 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3156 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3157 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3158 | |
51df1142 | 3159 | if (s && s->size) { |
55136592 CL |
3160 | char *name = kasprintf(GFP_NOWAIT, |
3161 | "dma-kmalloc-%d", s->objsize); | |
3162 | ||
3163 | BUG_ON(!name); | |
51df1142 CL |
3164 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3165 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3166 | } |
3167 | } | |
3168 | #endif | |
3adbefee IM |
3169 | printk(KERN_INFO |
3170 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3171 | " CPUs=%d, Nodes=%d\n", |
3172 | caches, cache_line_size(), | |
81819f0f CL |
3173 | slub_min_order, slub_max_order, slub_min_objects, |
3174 | nr_cpu_ids, nr_node_ids); | |
3175 | } | |
3176 | ||
7e85ee0c PE |
3177 | void __init kmem_cache_init_late(void) |
3178 | { | |
7e85ee0c PE |
3179 | } |
3180 | ||
81819f0f CL |
3181 | /* |
3182 | * Find a mergeable slab cache | |
3183 | */ | |
3184 | static int slab_unmergeable(struct kmem_cache *s) | |
3185 | { | |
3186 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3187 | return 1; | |
3188 | ||
c59def9f | 3189 | if (s->ctor) |
81819f0f CL |
3190 | return 1; |
3191 | ||
8ffa6875 CL |
3192 | /* |
3193 | * We may have set a slab to be unmergeable during bootstrap. | |
3194 | */ | |
3195 | if (s->refcount < 0) | |
3196 | return 1; | |
3197 | ||
81819f0f CL |
3198 | return 0; |
3199 | } | |
3200 | ||
3201 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3202 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3203 | void (*ctor)(void *)) |
81819f0f | 3204 | { |
5b95a4ac | 3205 | struct kmem_cache *s; |
81819f0f CL |
3206 | |
3207 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3208 | return NULL; | |
3209 | ||
c59def9f | 3210 | if (ctor) |
81819f0f CL |
3211 | return NULL; |
3212 | ||
3213 | size = ALIGN(size, sizeof(void *)); | |
3214 | align = calculate_alignment(flags, align, size); | |
3215 | size = ALIGN(size, align); | |
ba0268a8 | 3216 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3217 | |
5b95a4ac | 3218 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3219 | if (slab_unmergeable(s)) |
3220 | continue; | |
3221 | ||
3222 | if (size > s->size) | |
3223 | continue; | |
3224 | ||
ba0268a8 | 3225 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3226 | continue; |
3227 | /* | |
3228 | * Check if alignment is compatible. | |
3229 | * Courtesy of Adrian Drzewiecki | |
3230 | */ | |
06428780 | 3231 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3232 | continue; |
3233 | ||
3234 | if (s->size - size >= sizeof(void *)) | |
3235 | continue; | |
3236 | ||
3237 | return s; | |
3238 | } | |
3239 | return NULL; | |
3240 | } | |
3241 | ||
3242 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3243 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3244 | { |
3245 | struct kmem_cache *s; | |
3246 | ||
fe1ff49d BH |
3247 | if (WARN_ON(!name)) |
3248 | return NULL; | |
3249 | ||
81819f0f | 3250 | down_write(&slub_lock); |
ba0268a8 | 3251 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3252 | if (s) { |
3253 | s->refcount++; | |
3254 | /* | |
3255 | * Adjust the object sizes so that we clear | |
3256 | * the complete object on kzalloc. | |
3257 | */ | |
3258 | s->objsize = max(s->objsize, (int)size); | |
3259 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3260 | |
7b8f3b66 | 3261 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3262 | s->refcount--; |
81819f0f | 3263 | goto err; |
7b8f3b66 | 3264 | } |
2bce6485 | 3265 | up_write(&slub_lock); |
a0e1d1be CL |
3266 | return s; |
3267 | } | |
6446faa2 | 3268 | |
a0e1d1be CL |
3269 | s = kmalloc(kmem_size, GFP_KERNEL); |
3270 | if (s) { | |
55136592 | 3271 | if (kmem_cache_open(s, name, |
c59def9f | 3272 | size, align, flags, ctor)) { |
81819f0f | 3273 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3274 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3275 | list_del(&s->list); |
7b8f3b66 | 3276 | kfree(s); |
a0e1d1be | 3277 | goto err; |
7b8f3b66 | 3278 | } |
2bce6485 | 3279 | up_write(&slub_lock); |
a0e1d1be CL |
3280 | return s; |
3281 | } | |
3282 | kfree(s); | |
81819f0f CL |
3283 | } |
3284 | up_write(&slub_lock); | |
81819f0f CL |
3285 | |
3286 | err: | |
81819f0f CL |
3287 | if (flags & SLAB_PANIC) |
3288 | panic("Cannot create slabcache %s\n", name); | |
3289 | else | |
3290 | s = NULL; | |
3291 | return s; | |
3292 | } | |
3293 | EXPORT_SYMBOL(kmem_cache_create); | |
3294 | ||
81819f0f | 3295 | #ifdef CONFIG_SMP |
81819f0f | 3296 | /* |
672bba3a CL |
3297 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3298 | * necessary. | |
81819f0f CL |
3299 | */ |
3300 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3301 | unsigned long action, void *hcpu) | |
3302 | { | |
3303 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3304 | struct kmem_cache *s; |
3305 | unsigned long flags; | |
81819f0f CL |
3306 | |
3307 | switch (action) { | |
3308 | case CPU_UP_CANCELED: | |
8bb78442 | 3309 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3310 | case CPU_DEAD: |
8bb78442 | 3311 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3312 | down_read(&slub_lock); |
3313 | list_for_each_entry(s, &slab_caches, list) { | |
3314 | local_irq_save(flags); | |
3315 | __flush_cpu_slab(s, cpu); | |
3316 | local_irq_restore(flags); | |
3317 | } | |
3318 | up_read(&slub_lock); | |
81819f0f CL |
3319 | break; |
3320 | default: | |
3321 | break; | |
3322 | } | |
3323 | return NOTIFY_OK; | |
3324 | } | |
3325 | ||
06428780 | 3326 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3327 | .notifier_call = slab_cpuup_callback |
06428780 | 3328 | }; |
81819f0f CL |
3329 | |
3330 | #endif | |
3331 | ||
ce71e27c | 3332 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3333 | { |
aadb4bc4 | 3334 | struct kmem_cache *s; |
94b528d0 | 3335 | void *ret; |
aadb4bc4 | 3336 | |
ffadd4d0 | 3337 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3338 | return kmalloc_large(size, gfpflags); |
3339 | ||
aadb4bc4 | 3340 | s = get_slab(size, gfpflags); |
81819f0f | 3341 | |
2408c550 | 3342 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3343 | return s; |
81819f0f | 3344 | |
2154a336 | 3345 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3346 | |
3347 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3348 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3349 | |
3350 | return ret; | |
81819f0f CL |
3351 | } |
3352 | ||
3353 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3354 | int node, unsigned long caller) |
81819f0f | 3355 | { |
aadb4bc4 | 3356 | struct kmem_cache *s; |
94b528d0 | 3357 | void *ret; |
aadb4bc4 | 3358 | |
d3e14aa3 XF |
3359 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3360 | ret = kmalloc_large_node(size, gfpflags, node); | |
3361 | ||
3362 | trace_kmalloc_node(caller, ret, | |
3363 | size, PAGE_SIZE << get_order(size), | |
3364 | gfpflags, node); | |
3365 | ||
3366 | return ret; | |
3367 | } | |
eada35ef | 3368 | |
aadb4bc4 | 3369 | s = get_slab(size, gfpflags); |
81819f0f | 3370 | |
2408c550 | 3371 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3372 | return s; |
81819f0f | 3373 | |
94b528d0 EGM |
3374 | ret = slab_alloc(s, gfpflags, node, caller); |
3375 | ||
3376 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3377 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3378 | |
3379 | return ret; | |
81819f0f CL |
3380 | } |
3381 | ||
f6acb635 | 3382 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3383 | static int count_inuse(struct page *page) |
3384 | { | |
3385 | return page->inuse; | |
3386 | } | |
3387 | ||
3388 | static int count_total(struct page *page) | |
3389 | { | |
3390 | return page->objects; | |
3391 | } | |
3392 | ||
434e245d CL |
3393 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3394 | unsigned long *map) | |
53e15af0 CL |
3395 | { |
3396 | void *p; | |
a973e9dd | 3397 | void *addr = page_address(page); |
53e15af0 CL |
3398 | |
3399 | if (!check_slab(s, page) || | |
3400 | !on_freelist(s, page, NULL)) | |
3401 | return 0; | |
3402 | ||
3403 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3404 | bitmap_zero(map, page->objects); |
53e15af0 | 3405 | |
7656c72b CL |
3406 | for_each_free_object(p, s, page->freelist) { |
3407 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3408 | if (!check_object(s, page, p, 0)) |
3409 | return 0; | |
3410 | } | |
3411 | ||
224a88be | 3412 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3413 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3414 | if (!check_object(s, page, p, 1)) |
3415 | return 0; | |
3416 | return 1; | |
3417 | } | |
3418 | ||
434e245d CL |
3419 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3420 | unsigned long *map) | |
53e15af0 CL |
3421 | { |
3422 | if (slab_trylock(page)) { | |
434e245d | 3423 | validate_slab(s, page, map); |
53e15af0 CL |
3424 | slab_unlock(page); |
3425 | } else | |
3426 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3427 | s->name, page); | |
53e15af0 CL |
3428 | } |
3429 | ||
434e245d CL |
3430 | static int validate_slab_node(struct kmem_cache *s, |
3431 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3432 | { |
3433 | unsigned long count = 0; | |
3434 | struct page *page; | |
3435 | unsigned long flags; | |
3436 | ||
3437 | spin_lock_irqsave(&n->list_lock, flags); | |
3438 | ||
3439 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3440 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3441 | count++; |
3442 | } | |
3443 | if (count != n->nr_partial) | |
3444 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3445 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3446 | ||
3447 | if (!(s->flags & SLAB_STORE_USER)) | |
3448 | goto out; | |
3449 | ||
3450 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3451 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3452 | count++; |
3453 | } | |
3454 | if (count != atomic_long_read(&n->nr_slabs)) | |
3455 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3456 | "counter=%ld\n", s->name, count, | |
3457 | atomic_long_read(&n->nr_slabs)); | |
3458 | ||
3459 | out: | |
3460 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3461 | return count; | |
3462 | } | |
3463 | ||
434e245d | 3464 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3465 | { |
3466 | int node; | |
3467 | unsigned long count = 0; | |
205ab99d | 3468 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3469 | sizeof(unsigned long), GFP_KERNEL); |
3470 | ||
3471 | if (!map) | |
3472 | return -ENOMEM; | |
53e15af0 CL |
3473 | |
3474 | flush_all(s); | |
f64dc58c | 3475 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3476 | struct kmem_cache_node *n = get_node(s, node); |
3477 | ||
434e245d | 3478 | count += validate_slab_node(s, n, map); |
53e15af0 | 3479 | } |
434e245d | 3480 | kfree(map); |
53e15af0 CL |
3481 | return count; |
3482 | } | |
3483 | ||
b3459709 CL |
3484 | #ifdef SLUB_RESILIENCY_TEST |
3485 | static void resiliency_test(void) | |
3486 | { | |
3487 | u8 *p; | |
3488 | ||
3489 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3490 | printk(KERN_ERR "-----------------------\n"); | |
3491 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3492 | ||
3493 | p = kzalloc(16, GFP_KERNEL); | |
3494 | p[16] = 0x12; | |
3495 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3496 | " 0x12->0x%p\n\n", p + 16); | |
3497 | ||
3498 | validate_slab_cache(kmalloc_caches + 4); | |
3499 | ||
3500 | /* Hmmm... The next two are dangerous */ | |
3501 | p = kzalloc(32, GFP_KERNEL); | |
3502 | p[32 + sizeof(void *)] = 0x34; | |
3503 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3504 | " 0x34 -> -0x%p\n", p); |
3505 | printk(KERN_ERR | |
3506 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3507 | |
3508 | validate_slab_cache(kmalloc_caches + 5); | |
3509 | p = kzalloc(64, GFP_KERNEL); | |
3510 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3511 | *p = 0x56; | |
3512 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3513 | p); | |
3adbefee IM |
3514 | printk(KERN_ERR |
3515 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3516 | validate_slab_cache(kmalloc_caches + 6); |
3517 | ||
3518 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3519 | p = kzalloc(128, GFP_KERNEL); | |
3520 | kfree(p); | |
3521 | *p = 0x78; | |
3522 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3523 | validate_slab_cache(kmalloc_caches + 7); | |
3524 | ||
3525 | p = kzalloc(256, GFP_KERNEL); | |
3526 | kfree(p); | |
3527 | p[50] = 0x9a; | |
3adbefee IM |
3528 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3529 | p); | |
b3459709 CL |
3530 | validate_slab_cache(kmalloc_caches + 8); |
3531 | ||
3532 | p = kzalloc(512, GFP_KERNEL); | |
3533 | kfree(p); | |
3534 | p[512] = 0xab; | |
3535 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3536 | validate_slab_cache(kmalloc_caches + 9); | |
3537 | } | |
3538 | #else | |
3539 | static void resiliency_test(void) {}; | |
3540 | #endif | |
3541 | ||
88a420e4 | 3542 | /* |
672bba3a | 3543 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3544 | * and freed. |
3545 | */ | |
3546 | ||
3547 | struct location { | |
3548 | unsigned long count; | |
ce71e27c | 3549 | unsigned long addr; |
45edfa58 CL |
3550 | long long sum_time; |
3551 | long min_time; | |
3552 | long max_time; | |
3553 | long min_pid; | |
3554 | long max_pid; | |
174596a0 | 3555 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3556 | nodemask_t nodes; |
88a420e4 CL |
3557 | }; |
3558 | ||
3559 | struct loc_track { | |
3560 | unsigned long max; | |
3561 | unsigned long count; | |
3562 | struct location *loc; | |
3563 | }; | |
3564 | ||
3565 | static void free_loc_track(struct loc_track *t) | |
3566 | { | |
3567 | if (t->max) | |
3568 | free_pages((unsigned long)t->loc, | |
3569 | get_order(sizeof(struct location) * t->max)); | |
3570 | } | |
3571 | ||
68dff6a9 | 3572 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3573 | { |
3574 | struct location *l; | |
3575 | int order; | |
3576 | ||
88a420e4 CL |
3577 | order = get_order(sizeof(struct location) * max); |
3578 | ||
68dff6a9 | 3579 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3580 | if (!l) |
3581 | return 0; | |
3582 | ||
3583 | if (t->count) { | |
3584 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3585 | free_loc_track(t); | |
3586 | } | |
3587 | t->max = max; | |
3588 | t->loc = l; | |
3589 | return 1; | |
3590 | } | |
3591 | ||
3592 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3593 | const struct track *track) |
88a420e4 CL |
3594 | { |
3595 | long start, end, pos; | |
3596 | struct location *l; | |
ce71e27c | 3597 | unsigned long caddr; |
45edfa58 | 3598 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3599 | |
3600 | start = -1; | |
3601 | end = t->count; | |
3602 | ||
3603 | for ( ; ; ) { | |
3604 | pos = start + (end - start + 1) / 2; | |
3605 | ||
3606 | /* | |
3607 | * There is nothing at "end". If we end up there | |
3608 | * we need to add something to before end. | |
3609 | */ | |
3610 | if (pos == end) | |
3611 | break; | |
3612 | ||
3613 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3614 | if (track->addr == caddr) { |
3615 | ||
3616 | l = &t->loc[pos]; | |
3617 | l->count++; | |
3618 | if (track->when) { | |
3619 | l->sum_time += age; | |
3620 | if (age < l->min_time) | |
3621 | l->min_time = age; | |
3622 | if (age > l->max_time) | |
3623 | l->max_time = age; | |
3624 | ||
3625 | if (track->pid < l->min_pid) | |
3626 | l->min_pid = track->pid; | |
3627 | if (track->pid > l->max_pid) | |
3628 | l->max_pid = track->pid; | |
3629 | ||
174596a0 RR |
3630 | cpumask_set_cpu(track->cpu, |
3631 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3632 | } |
3633 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3634 | return 1; |
3635 | } | |
3636 | ||
45edfa58 | 3637 | if (track->addr < caddr) |
88a420e4 CL |
3638 | end = pos; |
3639 | else | |
3640 | start = pos; | |
3641 | } | |
3642 | ||
3643 | /* | |
672bba3a | 3644 | * Not found. Insert new tracking element. |
88a420e4 | 3645 | */ |
68dff6a9 | 3646 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3647 | return 0; |
3648 | ||
3649 | l = t->loc + pos; | |
3650 | if (pos < t->count) | |
3651 | memmove(l + 1, l, | |
3652 | (t->count - pos) * sizeof(struct location)); | |
3653 | t->count++; | |
3654 | l->count = 1; | |
45edfa58 CL |
3655 | l->addr = track->addr; |
3656 | l->sum_time = age; | |
3657 | l->min_time = age; | |
3658 | l->max_time = age; | |
3659 | l->min_pid = track->pid; | |
3660 | l->max_pid = track->pid; | |
174596a0 RR |
3661 | cpumask_clear(to_cpumask(l->cpus)); |
3662 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3663 | nodes_clear(l->nodes); |
3664 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3665 | return 1; |
3666 | } | |
3667 | ||
3668 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b ED |
3669 | struct page *page, enum track_item alloc, |
3670 | long *map) | |
88a420e4 | 3671 | { |
a973e9dd | 3672 | void *addr = page_address(page); |
88a420e4 CL |
3673 | void *p; |
3674 | ||
39b26464 | 3675 | bitmap_zero(map, page->objects); |
7656c72b CL |
3676 | for_each_free_object(p, s, page->freelist) |
3677 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3678 | |
224a88be | 3679 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3680 | if (!test_bit(slab_index(p, s, addr), map)) |
3681 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3682 | } |
3683 | ||
3684 | static int list_locations(struct kmem_cache *s, char *buf, | |
3685 | enum track_item alloc) | |
3686 | { | |
e374d483 | 3687 | int len = 0; |
88a420e4 | 3688 | unsigned long i; |
68dff6a9 | 3689 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3690 | int node; |
bbd7d57b ED |
3691 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3692 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3693 | |
bbd7d57b ED |
3694 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3695 | GFP_TEMPORARY)) { | |
3696 | kfree(map); | |
68dff6a9 | 3697 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3698 | } |
88a420e4 CL |
3699 | /* Push back cpu slabs */ |
3700 | flush_all(s); | |
3701 | ||
f64dc58c | 3702 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3703 | struct kmem_cache_node *n = get_node(s, node); |
3704 | unsigned long flags; | |
3705 | struct page *page; | |
3706 | ||
9e86943b | 3707 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3708 | continue; |
3709 | ||
3710 | spin_lock_irqsave(&n->list_lock, flags); | |
3711 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3712 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3713 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3714 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3715 | spin_unlock_irqrestore(&n->list_lock, flags); |
3716 | } | |
3717 | ||
3718 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3719 | struct location *l = &t.loc[i]; |
88a420e4 | 3720 | |
9c246247 | 3721 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3722 | break; |
e374d483 | 3723 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3724 | |
3725 | if (l->addr) | |
e374d483 | 3726 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3727 | else |
e374d483 | 3728 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3729 | |
3730 | if (l->sum_time != l->min_time) { | |
e374d483 | 3731 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3732 | l->min_time, |
3733 | (long)div_u64(l->sum_time, l->count), | |
3734 | l->max_time); | |
45edfa58 | 3735 | } else |
e374d483 | 3736 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3737 | l->min_time); |
3738 | ||
3739 | if (l->min_pid != l->max_pid) | |
e374d483 | 3740 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3741 | l->min_pid, l->max_pid); |
3742 | else | |
e374d483 | 3743 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3744 | l->min_pid); |
3745 | ||
174596a0 RR |
3746 | if (num_online_cpus() > 1 && |
3747 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3748 | len < PAGE_SIZE - 60) { |
3749 | len += sprintf(buf + len, " cpus="); | |
3750 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3751 | to_cpumask(l->cpus)); |
45edfa58 CL |
3752 | } |
3753 | ||
62bc62a8 | 3754 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3755 | len < PAGE_SIZE - 60) { |
3756 | len += sprintf(buf + len, " nodes="); | |
3757 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3758 | l->nodes); |
3759 | } | |
3760 | ||
e374d483 | 3761 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3762 | } |
3763 | ||
3764 | free_loc_track(&t); | |
bbd7d57b | 3765 | kfree(map); |
88a420e4 | 3766 | if (!t.count) |
e374d483 HH |
3767 | len += sprintf(buf, "No data\n"); |
3768 | return len; | |
88a420e4 CL |
3769 | } |
3770 | ||
81819f0f | 3771 | enum slab_stat_type { |
205ab99d CL |
3772 | SL_ALL, /* All slabs */ |
3773 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3774 | SL_CPU, /* Only slabs used for cpu caches */ | |
3775 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3776 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3777 | }; |
3778 | ||
205ab99d | 3779 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3780 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3781 | #define SO_CPU (1 << SL_CPU) | |
3782 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3783 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3784 | |
62e5c4b4 CG |
3785 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3786 | char *buf, unsigned long flags) | |
81819f0f CL |
3787 | { |
3788 | unsigned long total = 0; | |
81819f0f CL |
3789 | int node; |
3790 | int x; | |
3791 | unsigned long *nodes; | |
3792 | unsigned long *per_cpu; | |
3793 | ||
3794 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3795 | if (!nodes) |
3796 | return -ENOMEM; | |
81819f0f CL |
3797 | per_cpu = nodes + nr_node_ids; |
3798 | ||
205ab99d CL |
3799 | if (flags & SO_CPU) { |
3800 | int cpu; | |
81819f0f | 3801 | |
205ab99d | 3802 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3803 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3804 | |
205ab99d CL |
3805 | if (!c || c->node < 0) |
3806 | continue; | |
3807 | ||
3808 | if (c->page) { | |
3809 | if (flags & SO_TOTAL) | |
3810 | x = c->page->objects; | |
3811 | else if (flags & SO_OBJECTS) | |
3812 | x = c->page->inuse; | |
81819f0f CL |
3813 | else |
3814 | x = 1; | |
205ab99d | 3815 | |
81819f0f | 3816 | total += x; |
205ab99d | 3817 | nodes[c->node] += x; |
81819f0f | 3818 | } |
205ab99d | 3819 | per_cpu[c->node]++; |
81819f0f CL |
3820 | } |
3821 | } | |
3822 | ||
205ab99d CL |
3823 | if (flags & SO_ALL) { |
3824 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3825 | struct kmem_cache_node *n = get_node(s, node); | |
3826 | ||
3827 | if (flags & SO_TOTAL) | |
3828 | x = atomic_long_read(&n->total_objects); | |
3829 | else if (flags & SO_OBJECTS) | |
3830 | x = atomic_long_read(&n->total_objects) - | |
3831 | count_partial(n, count_free); | |
81819f0f | 3832 | |
81819f0f | 3833 | else |
205ab99d | 3834 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3835 | total += x; |
3836 | nodes[node] += x; | |
3837 | } | |
3838 | ||
205ab99d CL |
3839 | } else if (flags & SO_PARTIAL) { |
3840 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3841 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3842 | |
205ab99d CL |
3843 | if (flags & SO_TOTAL) |
3844 | x = count_partial(n, count_total); | |
3845 | else if (flags & SO_OBJECTS) | |
3846 | x = count_partial(n, count_inuse); | |
81819f0f | 3847 | else |
205ab99d | 3848 | x = n->nr_partial; |
81819f0f CL |
3849 | total += x; |
3850 | nodes[node] += x; | |
3851 | } | |
3852 | } | |
81819f0f CL |
3853 | x = sprintf(buf, "%lu", total); |
3854 | #ifdef CONFIG_NUMA | |
f64dc58c | 3855 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3856 | if (nodes[node]) |
3857 | x += sprintf(buf + x, " N%d=%lu", | |
3858 | node, nodes[node]); | |
3859 | #endif | |
3860 | kfree(nodes); | |
3861 | return x + sprintf(buf + x, "\n"); | |
3862 | } | |
3863 | ||
3864 | static int any_slab_objects(struct kmem_cache *s) | |
3865 | { | |
3866 | int node; | |
81819f0f | 3867 | |
dfb4f096 | 3868 | for_each_online_node(node) { |
81819f0f CL |
3869 | struct kmem_cache_node *n = get_node(s, node); |
3870 | ||
dfb4f096 CL |
3871 | if (!n) |
3872 | continue; | |
3873 | ||
4ea33e2d | 3874 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3875 | return 1; |
3876 | } | |
3877 | return 0; | |
3878 | } | |
3879 | ||
3880 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3881 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3882 | ||
3883 | struct slab_attribute { | |
3884 | struct attribute attr; | |
3885 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3886 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3887 | }; | |
3888 | ||
3889 | #define SLAB_ATTR_RO(_name) \ | |
3890 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3891 | ||
3892 | #define SLAB_ATTR(_name) \ | |
3893 | static struct slab_attribute _name##_attr = \ | |
3894 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3895 | ||
81819f0f CL |
3896 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3897 | { | |
3898 | return sprintf(buf, "%d\n", s->size); | |
3899 | } | |
3900 | SLAB_ATTR_RO(slab_size); | |
3901 | ||
3902 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3903 | { | |
3904 | return sprintf(buf, "%d\n", s->align); | |
3905 | } | |
3906 | SLAB_ATTR_RO(align); | |
3907 | ||
3908 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3909 | { | |
3910 | return sprintf(buf, "%d\n", s->objsize); | |
3911 | } | |
3912 | SLAB_ATTR_RO(object_size); | |
3913 | ||
3914 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3915 | { | |
834f3d11 | 3916 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3917 | } |
3918 | SLAB_ATTR_RO(objs_per_slab); | |
3919 | ||
06b285dc CL |
3920 | static ssize_t order_store(struct kmem_cache *s, |
3921 | const char *buf, size_t length) | |
3922 | { | |
0121c619 CL |
3923 | unsigned long order; |
3924 | int err; | |
3925 | ||
3926 | err = strict_strtoul(buf, 10, &order); | |
3927 | if (err) | |
3928 | return err; | |
06b285dc CL |
3929 | |
3930 | if (order > slub_max_order || order < slub_min_order) | |
3931 | return -EINVAL; | |
3932 | ||
3933 | calculate_sizes(s, order); | |
3934 | return length; | |
3935 | } | |
3936 | ||
81819f0f CL |
3937 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3938 | { | |
834f3d11 | 3939 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3940 | } |
06b285dc | 3941 | SLAB_ATTR(order); |
81819f0f | 3942 | |
73d342b1 DR |
3943 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3944 | { | |
3945 | return sprintf(buf, "%lu\n", s->min_partial); | |
3946 | } | |
3947 | ||
3948 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3949 | size_t length) | |
3950 | { | |
3951 | unsigned long min; | |
3952 | int err; | |
3953 | ||
3954 | err = strict_strtoul(buf, 10, &min); | |
3955 | if (err) | |
3956 | return err; | |
3957 | ||
c0bdb232 | 3958 | set_min_partial(s, min); |
73d342b1 DR |
3959 | return length; |
3960 | } | |
3961 | SLAB_ATTR(min_partial); | |
3962 | ||
81819f0f CL |
3963 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3964 | { | |
3965 | if (s->ctor) { | |
3966 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3967 | ||
3968 | return n + sprintf(buf + n, "\n"); | |
3969 | } | |
3970 | return 0; | |
3971 | } | |
3972 | SLAB_ATTR_RO(ctor); | |
3973 | ||
81819f0f CL |
3974 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3975 | { | |
3976 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3977 | } | |
3978 | SLAB_ATTR_RO(aliases); | |
3979 | ||
3980 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3981 | { | |
205ab99d | 3982 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3983 | } |
3984 | SLAB_ATTR_RO(slabs); | |
3985 | ||
3986 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3987 | { | |
d9acf4b7 | 3988 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3989 | } |
3990 | SLAB_ATTR_RO(partial); | |
3991 | ||
3992 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3993 | { | |
d9acf4b7 | 3994 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3995 | } |
3996 | SLAB_ATTR_RO(cpu_slabs); | |
3997 | ||
3998 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3999 | { | |
205ab99d | 4000 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4001 | } |
4002 | SLAB_ATTR_RO(objects); | |
4003 | ||
205ab99d CL |
4004 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4005 | { | |
4006 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4007 | } | |
4008 | SLAB_ATTR_RO(objects_partial); | |
4009 | ||
4010 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4011 | { | |
4012 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4013 | } | |
4014 | SLAB_ATTR_RO(total_objects); | |
4015 | ||
81819f0f CL |
4016 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4017 | { | |
4018 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4019 | } | |
4020 | ||
4021 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4022 | const char *buf, size_t length) | |
4023 | { | |
4024 | s->flags &= ~SLAB_DEBUG_FREE; | |
4025 | if (buf[0] == '1') | |
4026 | s->flags |= SLAB_DEBUG_FREE; | |
4027 | return length; | |
4028 | } | |
4029 | SLAB_ATTR(sanity_checks); | |
4030 | ||
4031 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4032 | { | |
4033 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4034 | } | |
4035 | ||
4036 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4037 | size_t length) | |
4038 | { | |
4039 | s->flags &= ~SLAB_TRACE; | |
4040 | if (buf[0] == '1') | |
4041 | s->flags |= SLAB_TRACE; | |
4042 | return length; | |
4043 | } | |
4044 | SLAB_ATTR(trace); | |
4045 | ||
4c13dd3b DM |
4046 | #ifdef CONFIG_FAILSLAB |
4047 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4048 | { | |
4049 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4050 | } | |
4051 | ||
4052 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4053 | size_t length) | |
4054 | { | |
4055 | s->flags &= ~SLAB_FAILSLAB; | |
4056 | if (buf[0] == '1') | |
4057 | s->flags |= SLAB_FAILSLAB; | |
4058 | return length; | |
4059 | } | |
4060 | SLAB_ATTR(failslab); | |
4061 | #endif | |
4062 | ||
81819f0f CL |
4063 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4064 | { | |
4065 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4066 | } | |
4067 | ||
4068 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4069 | const char *buf, size_t length) | |
4070 | { | |
4071 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4072 | if (buf[0] == '1') | |
4073 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4074 | return length; | |
4075 | } | |
4076 | SLAB_ATTR(reclaim_account); | |
4077 | ||
4078 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4079 | { | |
5af60839 | 4080 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4081 | } |
4082 | SLAB_ATTR_RO(hwcache_align); | |
4083 | ||
4084 | #ifdef CONFIG_ZONE_DMA | |
4085 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4086 | { | |
4087 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4088 | } | |
4089 | SLAB_ATTR_RO(cache_dma); | |
4090 | #endif | |
4091 | ||
4092 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4093 | { | |
4094 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4095 | } | |
4096 | SLAB_ATTR_RO(destroy_by_rcu); | |
4097 | ||
4098 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4099 | { | |
4100 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4101 | } | |
4102 | ||
4103 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4104 | const char *buf, size_t length) | |
4105 | { | |
4106 | if (any_slab_objects(s)) | |
4107 | return -EBUSY; | |
4108 | ||
4109 | s->flags &= ~SLAB_RED_ZONE; | |
4110 | if (buf[0] == '1') | |
4111 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4112 | calculate_sizes(s, -1); |
81819f0f CL |
4113 | return length; |
4114 | } | |
4115 | SLAB_ATTR(red_zone); | |
4116 | ||
4117 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4118 | { | |
4119 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4120 | } | |
4121 | ||
4122 | static ssize_t poison_store(struct kmem_cache *s, | |
4123 | const char *buf, size_t length) | |
4124 | { | |
4125 | if (any_slab_objects(s)) | |
4126 | return -EBUSY; | |
4127 | ||
4128 | s->flags &= ~SLAB_POISON; | |
4129 | if (buf[0] == '1') | |
4130 | s->flags |= SLAB_POISON; | |
06b285dc | 4131 | calculate_sizes(s, -1); |
81819f0f CL |
4132 | return length; |
4133 | } | |
4134 | SLAB_ATTR(poison); | |
4135 | ||
4136 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4137 | { | |
4138 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4139 | } | |
4140 | ||
4141 | static ssize_t store_user_store(struct kmem_cache *s, | |
4142 | const char *buf, size_t length) | |
4143 | { | |
4144 | if (any_slab_objects(s)) | |
4145 | return -EBUSY; | |
4146 | ||
4147 | s->flags &= ~SLAB_STORE_USER; | |
4148 | if (buf[0] == '1') | |
4149 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4150 | calculate_sizes(s, -1); |
81819f0f CL |
4151 | return length; |
4152 | } | |
4153 | SLAB_ATTR(store_user); | |
4154 | ||
53e15af0 CL |
4155 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4156 | { | |
4157 | return 0; | |
4158 | } | |
4159 | ||
4160 | static ssize_t validate_store(struct kmem_cache *s, | |
4161 | const char *buf, size_t length) | |
4162 | { | |
434e245d CL |
4163 | int ret = -EINVAL; |
4164 | ||
4165 | if (buf[0] == '1') { | |
4166 | ret = validate_slab_cache(s); | |
4167 | if (ret >= 0) | |
4168 | ret = length; | |
4169 | } | |
4170 | return ret; | |
53e15af0 CL |
4171 | } |
4172 | SLAB_ATTR(validate); | |
4173 | ||
2086d26a CL |
4174 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4175 | { | |
4176 | return 0; | |
4177 | } | |
4178 | ||
4179 | static ssize_t shrink_store(struct kmem_cache *s, | |
4180 | const char *buf, size_t length) | |
4181 | { | |
4182 | if (buf[0] == '1') { | |
4183 | int rc = kmem_cache_shrink(s); | |
4184 | ||
4185 | if (rc) | |
4186 | return rc; | |
4187 | } else | |
4188 | return -EINVAL; | |
4189 | return length; | |
4190 | } | |
4191 | SLAB_ATTR(shrink); | |
4192 | ||
88a420e4 CL |
4193 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4194 | { | |
4195 | if (!(s->flags & SLAB_STORE_USER)) | |
4196 | return -ENOSYS; | |
4197 | return list_locations(s, buf, TRACK_ALLOC); | |
4198 | } | |
4199 | SLAB_ATTR_RO(alloc_calls); | |
4200 | ||
4201 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4202 | { | |
4203 | if (!(s->flags & SLAB_STORE_USER)) | |
4204 | return -ENOSYS; | |
4205 | return list_locations(s, buf, TRACK_FREE); | |
4206 | } | |
4207 | SLAB_ATTR_RO(free_calls); | |
4208 | ||
81819f0f | 4209 | #ifdef CONFIG_NUMA |
9824601e | 4210 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4211 | { |
9824601e | 4212 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4213 | } |
4214 | ||
9824601e | 4215 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4216 | const char *buf, size_t length) |
4217 | { | |
0121c619 CL |
4218 | unsigned long ratio; |
4219 | int err; | |
4220 | ||
4221 | err = strict_strtoul(buf, 10, &ratio); | |
4222 | if (err) | |
4223 | return err; | |
4224 | ||
e2cb96b7 | 4225 | if (ratio <= 100) |
0121c619 | 4226 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4227 | |
81819f0f CL |
4228 | return length; |
4229 | } | |
9824601e | 4230 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4231 | #endif |
4232 | ||
8ff12cfc | 4233 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4234 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4235 | { | |
4236 | unsigned long sum = 0; | |
4237 | int cpu; | |
4238 | int len; | |
4239 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4240 | ||
4241 | if (!data) | |
4242 | return -ENOMEM; | |
4243 | ||
4244 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4245 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4246 | |
4247 | data[cpu] = x; | |
4248 | sum += x; | |
4249 | } | |
4250 | ||
4251 | len = sprintf(buf, "%lu", sum); | |
4252 | ||
50ef37b9 | 4253 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4254 | for_each_online_cpu(cpu) { |
4255 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4256 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4257 | } |
50ef37b9 | 4258 | #endif |
8ff12cfc CL |
4259 | kfree(data); |
4260 | return len + sprintf(buf + len, "\n"); | |
4261 | } | |
4262 | ||
78eb00cc DR |
4263 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4264 | { | |
4265 | int cpu; | |
4266 | ||
4267 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4268 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4269 | } |
4270 | ||
8ff12cfc CL |
4271 | #define STAT_ATTR(si, text) \ |
4272 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4273 | { \ | |
4274 | return show_stat(s, buf, si); \ | |
4275 | } \ | |
78eb00cc DR |
4276 | static ssize_t text##_store(struct kmem_cache *s, \ |
4277 | const char *buf, size_t length) \ | |
4278 | { \ | |
4279 | if (buf[0] != '0') \ | |
4280 | return -EINVAL; \ | |
4281 | clear_stat(s, si); \ | |
4282 | return length; \ | |
4283 | } \ | |
4284 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4285 | |
4286 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4287 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4288 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4289 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4290 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4291 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4292 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4293 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4294 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4295 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4296 | STAT_ATTR(FREE_SLAB, free_slab); | |
4297 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4298 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4299 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4300 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4301 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4302 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4303 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4304 | #endif |
4305 | ||
06428780 | 4306 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4307 | &slab_size_attr.attr, |
4308 | &object_size_attr.attr, | |
4309 | &objs_per_slab_attr.attr, | |
4310 | &order_attr.attr, | |
73d342b1 | 4311 | &min_partial_attr.attr, |
81819f0f | 4312 | &objects_attr.attr, |
205ab99d CL |
4313 | &objects_partial_attr.attr, |
4314 | &total_objects_attr.attr, | |
81819f0f CL |
4315 | &slabs_attr.attr, |
4316 | &partial_attr.attr, | |
4317 | &cpu_slabs_attr.attr, | |
4318 | &ctor_attr.attr, | |
81819f0f CL |
4319 | &aliases_attr.attr, |
4320 | &align_attr.attr, | |
4321 | &sanity_checks_attr.attr, | |
4322 | &trace_attr.attr, | |
4323 | &hwcache_align_attr.attr, | |
4324 | &reclaim_account_attr.attr, | |
4325 | &destroy_by_rcu_attr.attr, | |
4326 | &red_zone_attr.attr, | |
4327 | &poison_attr.attr, | |
4328 | &store_user_attr.attr, | |
53e15af0 | 4329 | &validate_attr.attr, |
2086d26a | 4330 | &shrink_attr.attr, |
88a420e4 CL |
4331 | &alloc_calls_attr.attr, |
4332 | &free_calls_attr.attr, | |
81819f0f CL |
4333 | #ifdef CONFIG_ZONE_DMA |
4334 | &cache_dma_attr.attr, | |
4335 | #endif | |
4336 | #ifdef CONFIG_NUMA | |
9824601e | 4337 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4338 | #endif |
4339 | #ifdef CONFIG_SLUB_STATS | |
4340 | &alloc_fastpath_attr.attr, | |
4341 | &alloc_slowpath_attr.attr, | |
4342 | &free_fastpath_attr.attr, | |
4343 | &free_slowpath_attr.attr, | |
4344 | &free_frozen_attr.attr, | |
4345 | &free_add_partial_attr.attr, | |
4346 | &free_remove_partial_attr.attr, | |
4347 | &alloc_from_partial_attr.attr, | |
4348 | &alloc_slab_attr.attr, | |
4349 | &alloc_refill_attr.attr, | |
4350 | &free_slab_attr.attr, | |
4351 | &cpuslab_flush_attr.attr, | |
4352 | &deactivate_full_attr.attr, | |
4353 | &deactivate_empty_attr.attr, | |
4354 | &deactivate_to_head_attr.attr, | |
4355 | &deactivate_to_tail_attr.attr, | |
4356 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4357 | &order_fallback_attr.attr, |
81819f0f | 4358 | #endif |
4c13dd3b DM |
4359 | #ifdef CONFIG_FAILSLAB |
4360 | &failslab_attr.attr, | |
4361 | #endif | |
4362 | ||
81819f0f CL |
4363 | NULL |
4364 | }; | |
4365 | ||
4366 | static struct attribute_group slab_attr_group = { | |
4367 | .attrs = slab_attrs, | |
4368 | }; | |
4369 | ||
4370 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4371 | struct attribute *attr, | |
4372 | char *buf) | |
4373 | { | |
4374 | struct slab_attribute *attribute; | |
4375 | struct kmem_cache *s; | |
4376 | int err; | |
4377 | ||
4378 | attribute = to_slab_attr(attr); | |
4379 | s = to_slab(kobj); | |
4380 | ||
4381 | if (!attribute->show) | |
4382 | return -EIO; | |
4383 | ||
4384 | err = attribute->show(s, buf); | |
4385 | ||
4386 | return err; | |
4387 | } | |
4388 | ||
4389 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4390 | struct attribute *attr, | |
4391 | const char *buf, size_t len) | |
4392 | { | |
4393 | struct slab_attribute *attribute; | |
4394 | struct kmem_cache *s; | |
4395 | int err; | |
4396 | ||
4397 | attribute = to_slab_attr(attr); | |
4398 | s = to_slab(kobj); | |
4399 | ||
4400 | if (!attribute->store) | |
4401 | return -EIO; | |
4402 | ||
4403 | err = attribute->store(s, buf, len); | |
4404 | ||
4405 | return err; | |
4406 | } | |
4407 | ||
151c602f CL |
4408 | static void kmem_cache_release(struct kobject *kobj) |
4409 | { | |
4410 | struct kmem_cache *s = to_slab(kobj); | |
4411 | ||
4412 | kfree(s); | |
4413 | } | |
4414 | ||
52cf25d0 | 4415 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4416 | .show = slab_attr_show, |
4417 | .store = slab_attr_store, | |
4418 | }; | |
4419 | ||
4420 | static struct kobj_type slab_ktype = { | |
4421 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4422 | .release = kmem_cache_release |
81819f0f CL |
4423 | }; |
4424 | ||
4425 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4426 | { | |
4427 | struct kobj_type *ktype = get_ktype(kobj); | |
4428 | ||
4429 | if (ktype == &slab_ktype) | |
4430 | return 1; | |
4431 | return 0; | |
4432 | } | |
4433 | ||
9cd43611 | 4434 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4435 | .filter = uevent_filter, |
4436 | }; | |
4437 | ||
27c3a314 | 4438 | static struct kset *slab_kset; |
81819f0f CL |
4439 | |
4440 | #define ID_STR_LENGTH 64 | |
4441 | ||
4442 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4443 | * |
4444 | * Format :[flags-]size | |
81819f0f CL |
4445 | */ |
4446 | static char *create_unique_id(struct kmem_cache *s) | |
4447 | { | |
4448 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4449 | char *p = name; | |
4450 | ||
4451 | BUG_ON(!name); | |
4452 | ||
4453 | *p++ = ':'; | |
4454 | /* | |
4455 | * First flags affecting slabcache operations. We will only | |
4456 | * get here for aliasable slabs so we do not need to support | |
4457 | * too many flags. The flags here must cover all flags that | |
4458 | * are matched during merging to guarantee that the id is | |
4459 | * unique. | |
4460 | */ | |
4461 | if (s->flags & SLAB_CACHE_DMA) | |
4462 | *p++ = 'd'; | |
4463 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4464 | *p++ = 'a'; | |
4465 | if (s->flags & SLAB_DEBUG_FREE) | |
4466 | *p++ = 'F'; | |
5a896d9e VN |
4467 | if (!(s->flags & SLAB_NOTRACK)) |
4468 | *p++ = 't'; | |
81819f0f CL |
4469 | if (p != name + 1) |
4470 | *p++ = '-'; | |
4471 | p += sprintf(p, "%07d", s->size); | |
4472 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4473 | return name; | |
4474 | } | |
4475 | ||
4476 | static int sysfs_slab_add(struct kmem_cache *s) | |
4477 | { | |
4478 | int err; | |
4479 | const char *name; | |
4480 | int unmergeable; | |
4481 | ||
4482 | if (slab_state < SYSFS) | |
4483 | /* Defer until later */ | |
4484 | return 0; | |
4485 | ||
4486 | unmergeable = slab_unmergeable(s); | |
4487 | if (unmergeable) { | |
4488 | /* | |
4489 | * Slabcache can never be merged so we can use the name proper. | |
4490 | * This is typically the case for debug situations. In that | |
4491 | * case we can catch duplicate names easily. | |
4492 | */ | |
27c3a314 | 4493 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4494 | name = s->name; |
4495 | } else { | |
4496 | /* | |
4497 | * Create a unique name for the slab as a target | |
4498 | * for the symlinks. | |
4499 | */ | |
4500 | name = create_unique_id(s); | |
4501 | } | |
4502 | ||
27c3a314 | 4503 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4504 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4505 | if (err) { | |
4506 | kobject_put(&s->kobj); | |
81819f0f | 4507 | return err; |
1eada11c | 4508 | } |
81819f0f CL |
4509 | |
4510 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4511 | if (err) { |
4512 | kobject_del(&s->kobj); | |
4513 | kobject_put(&s->kobj); | |
81819f0f | 4514 | return err; |
5788d8ad | 4515 | } |
81819f0f CL |
4516 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4517 | if (!unmergeable) { | |
4518 | /* Setup first alias */ | |
4519 | sysfs_slab_alias(s, s->name); | |
4520 | kfree(name); | |
4521 | } | |
4522 | return 0; | |
4523 | } | |
4524 | ||
4525 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4526 | { | |
2bce6485 CL |
4527 | if (slab_state < SYSFS) |
4528 | /* | |
4529 | * Sysfs has not been setup yet so no need to remove the | |
4530 | * cache from sysfs. | |
4531 | */ | |
4532 | return; | |
4533 | ||
81819f0f CL |
4534 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4535 | kobject_del(&s->kobj); | |
151c602f | 4536 | kobject_put(&s->kobj); |
81819f0f CL |
4537 | } |
4538 | ||
4539 | /* | |
4540 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4541 | * available lest we lose that information. |
81819f0f CL |
4542 | */ |
4543 | struct saved_alias { | |
4544 | struct kmem_cache *s; | |
4545 | const char *name; | |
4546 | struct saved_alias *next; | |
4547 | }; | |
4548 | ||
5af328a5 | 4549 | static struct saved_alias *alias_list; |
81819f0f CL |
4550 | |
4551 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4552 | { | |
4553 | struct saved_alias *al; | |
4554 | ||
4555 | if (slab_state == SYSFS) { | |
4556 | /* | |
4557 | * If we have a leftover link then remove it. | |
4558 | */ | |
27c3a314 GKH |
4559 | sysfs_remove_link(&slab_kset->kobj, name); |
4560 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4561 | } |
4562 | ||
4563 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4564 | if (!al) | |
4565 | return -ENOMEM; | |
4566 | ||
4567 | al->s = s; | |
4568 | al->name = name; | |
4569 | al->next = alias_list; | |
4570 | alias_list = al; | |
4571 | return 0; | |
4572 | } | |
4573 | ||
4574 | static int __init slab_sysfs_init(void) | |
4575 | { | |
5b95a4ac | 4576 | struct kmem_cache *s; |
81819f0f CL |
4577 | int err; |
4578 | ||
2bce6485 CL |
4579 | down_write(&slub_lock); |
4580 | ||
0ff21e46 | 4581 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4582 | if (!slab_kset) { |
2bce6485 | 4583 | up_write(&slub_lock); |
81819f0f CL |
4584 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4585 | return -ENOSYS; | |
4586 | } | |
4587 | ||
26a7bd03 CL |
4588 | slab_state = SYSFS; |
4589 | ||
5b95a4ac | 4590 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4591 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4592 | if (err) |
4593 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4594 | " to sysfs\n", s->name); | |
26a7bd03 | 4595 | } |
81819f0f CL |
4596 | |
4597 | while (alias_list) { | |
4598 | struct saved_alias *al = alias_list; | |
4599 | ||
4600 | alias_list = alias_list->next; | |
4601 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4602 | if (err) |
4603 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4604 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4605 | kfree(al); |
4606 | } | |
4607 | ||
2bce6485 | 4608 | up_write(&slub_lock); |
81819f0f CL |
4609 | resiliency_test(); |
4610 | return 0; | |
4611 | } | |
4612 | ||
4613 | __initcall(slab_sysfs_init); | |
81819f0f | 4614 | #endif |
57ed3eda PE |
4615 | |
4616 | /* | |
4617 | * The /proc/slabinfo ABI | |
4618 | */ | |
158a9624 | 4619 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4620 | static void print_slabinfo_header(struct seq_file *m) |
4621 | { | |
4622 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4623 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4624 | "<objperslab> <pagesperslab>"); | |
4625 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4626 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4627 | seq_putc(m, '\n'); | |
4628 | } | |
4629 | ||
4630 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4631 | { | |
4632 | loff_t n = *pos; | |
4633 | ||
4634 | down_read(&slub_lock); | |
4635 | if (!n) | |
4636 | print_slabinfo_header(m); | |
4637 | ||
4638 | return seq_list_start(&slab_caches, *pos); | |
4639 | } | |
4640 | ||
4641 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4642 | { | |
4643 | return seq_list_next(p, &slab_caches, pos); | |
4644 | } | |
4645 | ||
4646 | static void s_stop(struct seq_file *m, void *p) | |
4647 | { | |
4648 | up_read(&slub_lock); | |
4649 | } | |
4650 | ||
4651 | static int s_show(struct seq_file *m, void *p) | |
4652 | { | |
4653 | unsigned long nr_partials = 0; | |
4654 | unsigned long nr_slabs = 0; | |
4655 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4656 | unsigned long nr_objs = 0; |
4657 | unsigned long nr_free = 0; | |
57ed3eda PE |
4658 | struct kmem_cache *s; |
4659 | int node; | |
4660 | ||
4661 | s = list_entry(p, struct kmem_cache, list); | |
4662 | ||
4663 | for_each_online_node(node) { | |
4664 | struct kmem_cache_node *n = get_node(s, node); | |
4665 | ||
4666 | if (!n) | |
4667 | continue; | |
4668 | ||
4669 | nr_partials += n->nr_partial; | |
4670 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4671 | nr_objs += atomic_long_read(&n->total_objects); |
4672 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4673 | } |
4674 | ||
205ab99d | 4675 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4676 | |
4677 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4678 | nr_objs, s->size, oo_objects(s->oo), |
4679 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4680 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4681 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4682 | 0UL); | |
4683 | seq_putc(m, '\n'); | |
4684 | return 0; | |
4685 | } | |
4686 | ||
7b3c3a50 | 4687 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4688 | .start = s_start, |
4689 | .next = s_next, | |
4690 | .stop = s_stop, | |
4691 | .show = s_show, | |
4692 | }; | |
4693 | ||
7b3c3a50 AD |
4694 | static int slabinfo_open(struct inode *inode, struct file *file) |
4695 | { | |
4696 | return seq_open(file, &slabinfo_op); | |
4697 | } | |
4698 | ||
4699 | static const struct file_operations proc_slabinfo_operations = { | |
4700 | .open = slabinfo_open, | |
4701 | .read = seq_read, | |
4702 | .llseek = seq_lseek, | |
4703 | .release = seq_release, | |
4704 | }; | |
4705 | ||
4706 | static int __init slab_proc_init(void) | |
4707 | { | |
cf5d1131 | 4708 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4709 | return 0; |
4710 | } | |
4711 | module_init(slab_proc_init); | |
158a9624 | 4712 | #endif /* CONFIG_SLABINFO */ |