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 | { | |
84c1cf62 | 212 | kfree(s->name); |
151c602f CL |
213 | kfree(s); |
214 | } | |
8ff12cfc | 215 | |
81819f0f CL |
216 | #endif |
217 | ||
84e554e6 | 218 | static inline void stat(struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
219 | { |
220 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 221 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
222 | #endif |
223 | } | |
224 | ||
81819f0f CL |
225 | /******************************************************************** |
226 | * Core slab cache functions | |
227 | *******************************************************************/ | |
228 | ||
229 | int slab_is_available(void) | |
230 | { | |
231 | return slab_state >= UP; | |
232 | } | |
233 | ||
234 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
235 | { | |
236 | #ifdef CONFIG_NUMA | |
237 | return s->node[node]; | |
238 | #else | |
239 | return &s->local_node; | |
240 | #endif | |
241 | } | |
242 | ||
6446faa2 | 243 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
244 | static inline int check_valid_pointer(struct kmem_cache *s, |
245 | struct page *page, const void *object) | |
246 | { | |
247 | void *base; | |
248 | ||
a973e9dd | 249 | if (!object) |
02cbc874 CL |
250 | return 1; |
251 | ||
a973e9dd | 252 | base = page_address(page); |
39b26464 | 253 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
254 | (object - base) % s->size) { |
255 | return 0; | |
256 | } | |
257 | ||
258 | return 1; | |
259 | } | |
260 | ||
7656c72b CL |
261 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
262 | { | |
263 | return *(void **)(object + s->offset); | |
264 | } | |
265 | ||
266 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
267 | { | |
268 | *(void **)(object + s->offset) = fp; | |
269 | } | |
270 | ||
271 | /* Loop over all objects in a slab */ | |
224a88be CL |
272 | #define for_each_object(__p, __s, __addr, __objects) \ |
273 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
274 | __p += (__s)->size) |
275 | ||
276 | /* Scan freelist */ | |
277 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 278 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
279 | |
280 | /* Determine object index from a given position */ | |
281 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
282 | { | |
283 | return (p - addr) / s->size; | |
284 | } | |
285 | ||
834f3d11 CL |
286 | static inline struct kmem_cache_order_objects oo_make(int order, |
287 | unsigned long size) | |
288 | { | |
289 | struct kmem_cache_order_objects x = { | |
210b5c06 | 290 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
291 | }; |
292 | ||
293 | return x; | |
294 | } | |
295 | ||
296 | static inline int oo_order(struct kmem_cache_order_objects x) | |
297 | { | |
210b5c06 | 298 | return x.x >> OO_SHIFT; |
834f3d11 CL |
299 | } |
300 | ||
301 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
302 | { | |
210b5c06 | 303 | return x.x & OO_MASK; |
834f3d11 CL |
304 | } |
305 | ||
41ecc55b CL |
306 | #ifdef CONFIG_SLUB_DEBUG |
307 | /* | |
308 | * Debug settings: | |
309 | */ | |
f0630fff CL |
310 | #ifdef CONFIG_SLUB_DEBUG_ON |
311 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
312 | #else | |
41ecc55b | 313 | static int slub_debug; |
f0630fff | 314 | #endif |
41ecc55b CL |
315 | |
316 | static char *slub_debug_slabs; | |
fa5ec8a1 | 317 | static int disable_higher_order_debug; |
41ecc55b | 318 | |
81819f0f CL |
319 | /* |
320 | * Object debugging | |
321 | */ | |
322 | static void print_section(char *text, u8 *addr, unsigned int length) | |
323 | { | |
324 | int i, offset; | |
325 | int newline = 1; | |
326 | char ascii[17]; | |
327 | ||
328 | ascii[16] = 0; | |
329 | ||
330 | for (i = 0; i < length; i++) { | |
331 | if (newline) { | |
24922684 | 332 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
333 | newline = 0; |
334 | } | |
06428780 | 335 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
336 | offset = i % 16; |
337 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
338 | if (offset == 15) { | |
06428780 | 339 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
340 | newline = 1; |
341 | } | |
342 | } | |
343 | if (!newline) { | |
344 | i %= 16; | |
345 | while (i < 16) { | |
06428780 | 346 | printk(KERN_CONT " "); |
81819f0f CL |
347 | ascii[i] = ' '; |
348 | i++; | |
349 | } | |
06428780 | 350 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
351 | } |
352 | } | |
353 | ||
81819f0f CL |
354 | static struct track *get_track(struct kmem_cache *s, void *object, |
355 | enum track_item alloc) | |
356 | { | |
357 | struct track *p; | |
358 | ||
359 | if (s->offset) | |
360 | p = object + s->offset + sizeof(void *); | |
361 | else | |
362 | p = object + s->inuse; | |
363 | ||
364 | return p + alloc; | |
365 | } | |
366 | ||
367 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 368 | enum track_item alloc, unsigned long addr) |
81819f0f | 369 | { |
1a00df4a | 370 | struct track *p = get_track(s, object, alloc); |
81819f0f | 371 | |
81819f0f CL |
372 | if (addr) { |
373 | p->addr = addr; | |
374 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 375 | p->pid = current->pid; |
81819f0f CL |
376 | p->when = jiffies; |
377 | } else | |
378 | memset(p, 0, sizeof(struct track)); | |
379 | } | |
380 | ||
81819f0f CL |
381 | static void init_tracking(struct kmem_cache *s, void *object) |
382 | { | |
24922684 CL |
383 | if (!(s->flags & SLAB_STORE_USER)) |
384 | return; | |
385 | ||
ce71e27c EGM |
386 | set_track(s, object, TRACK_FREE, 0UL); |
387 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
388 | } |
389 | ||
390 | static void print_track(const char *s, struct track *t) | |
391 | { | |
392 | if (!t->addr) | |
393 | return; | |
394 | ||
7daf705f | 395 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 396 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
397 | } |
398 | ||
399 | static void print_tracking(struct kmem_cache *s, void *object) | |
400 | { | |
401 | if (!(s->flags & SLAB_STORE_USER)) | |
402 | return; | |
403 | ||
404 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
405 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
406 | } | |
407 | ||
408 | static void print_page_info(struct page *page) | |
409 | { | |
39b26464 CL |
410 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
411 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
412 | |
413 | } | |
414 | ||
415 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
416 | { | |
417 | va_list args; | |
418 | char buf[100]; | |
419 | ||
420 | va_start(args, fmt); | |
421 | vsnprintf(buf, sizeof(buf), fmt, args); | |
422 | va_end(args); | |
423 | printk(KERN_ERR "========================================" | |
424 | "=====================================\n"); | |
425 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
426 | printk(KERN_ERR "----------------------------------------" | |
427 | "-------------------------------------\n\n"); | |
81819f0f CL |
428 | } |
429 | ||
24922684 CL |
430 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
431 | { | |
432 | va_list args; | |
433 | char buf[100]; | |
434 | ||
435 | va_start(args, fmt); | |
436 | vsnprintf(buf, sizeof(buf), fmt, args); | |
437 | va_end(args); | |
438 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
439 | } | |
440 | ||
441 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
442 | { |
443 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 444 | u8 *addr = page_address(page); |
24922684 CL |
445 | |
446 | print_tracking(s, p); | |
447 | ||
448 | print_page_info(page); | |
449 | ||
450 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
451 | p, p - addr, get_freepointer(s, p)); | |
452 | ||
453 | if (p > addr + 16) | |
454 | print_section("Bytes b4", p - 16, 16); | |
455 | ||
0ebd652b | 456 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
457 | |
458 | if (s->flags & SLAB_RED_ZONE) | |
459 | print_section("Redzone", p + s->objsize, | |
460 | s->inuse - s->objsize); | |
461 | ||
81819f0f CL |
462 | if (s->offset) |
463 | off = s->offset + sizeof(void *); | |
464 | else | |
465 | off = s->inuse; | |
466 | ||
24922684 | 467 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 468 | off += 2 * sizeof(struct track); |
81819f0f CL |
469 | |
470 | if (off != s->size) | |
471 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
472 | print_section("Padding", p + off, s->size - off); |
473 | ||
474 | dump_stack(); | |
81819f0f CL |
475 | } |
476 | ||
477 | static void object_err(struct kmem_cache *s, struct page *page, | |
478 | u8 *object, char *reason) | |
479 | { | |
3dc50637 | 480 | slab_bug(s, "%s", reason); |
24922684 | 481 | print_trailer(s, page, object); |
81819f0f CL |
482 | } |
483 | ||
24922684 | 484 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
485 | { |
486 | va_list args; | |
487 | char buf[100]; | |
488 | ||
24922684 CL |
489 | va_start(args, fmt); |
490 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 491 | va_end(args); |
3dc50637 | 492 | slab_bug(s, "%s", buf); |
24922684 | 493 | print_page_info(page); |
81819f0f CL |
494 | dump_stack(); |
495 | } | |
496 | ||
497 | static void init_object(struct kmem_cache *s, void *object, int active) | |
498 | { | |
499 | u8 *p = object; | |
500 | ||
501 | if (s->flags & __OBJECT_POISON) { | |
502 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 503 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
504 | } |
505 | ||
506 | if (s->flags & SLAB_RED_ZONE) | |
507 | memset(p + s->objsize, | |
508 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
509 | s->inuse - s->objsize); | |
510 | } | |
511 | ||
24922684 | 512 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
513 | { |
514 | while (bytes) { | |
515 | if (*start != (u8)value) | |
24922684 | 516 | return start; |
81819f0f CL |
517 | start++; |
518 | bytes--; | |
519 | } | |
24922684 CL |
520 | return NULL; |
521 | } | |
522 | ||
523 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
524 | void *from, void *to) | |
525 | { | |
526 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
527 | memset(from, data, to - from); | |
528 | } | |
529 | ||
530 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
531 | u8 *object, char *what, | |
06428780 | 532 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
533 | { |
534 | u8 *fault; | |
535 | u8 *end; | |
536 | ||
537 | fault = check_bytes(start, value, bytes); | |
538 | if (!fault) | |
539 | return 1; | |
540 | ||
541 | end = start + bytes; | |
542 | while (end > fault && end[-1] == value) | |
543 | end--; | |
544 | ||
545 | slab_bug(s, "%s overwritten", what); | |
546 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
547 | fault, end - 1, fault[0], value); | |
548 | print_trailer(s, page, object); | |
549 | ||
550 | restore_bytes(s, what, value, fault, end); | |
551 | return 0; | |
81819f0f CL |
552 | } |
553 | ||
81819f0f CL |
554 | /* |
555 | * Object layout: | |
556 | * | |
557 | * object address | |
558 | * Bytes of the object to be managed. | |
559 | * If the freepointer may overlay the object then the free | |
560 | * pointer is the first word of the object. | |
672bba3a | 561 | * |
81819f0f CL |
562 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
563 | * 0xa5 (POISON_END) | |
564 | * | |
565 | * object + s->objsize | |
566 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
567 | * Padding is extended by another word if Redzoning is enabled and |
568 | * objsize == inuse. | |
569 | * | |
81819f0f CL |
570 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
571 | * 0xcc (RED_ACTIVE) for objects in use. | |
572 | * | |
573 | * object + s->inuse | |
672bba3a CL |
574 | * Meta data starts here. |
575 | * | |
81819f0f CL |
576 | * A. Free pointer (if we cannot overwrite object on free) |
577 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 578 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 579 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
580 | * before the word boundary. |
581 | * | |
582 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
583 | * |
584 | * object + s->size | |
672bba3a | 585 | * Nothing is used beyond s->size. |
81819f0f | 586 | * |
672bba3a CL |
587 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
588 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
589 | * may be used with merged slabcaches. |
590 | */ | |
591 | ||
81819f0f CL |
592 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
593 | { | |
594 | unsigned long off = s->inuse; /* The end of info */ | |
595 | ||
596 | if (s->offset) | |
597 | /* Freepointer is placed after the object. */ | |
598 | off += sizeof(void *); | |
599 | ||
600 | if (s->flags & SLAB_STORE_USER) | |
601 | /* We also have user information there */ | |
602 | off += 2 * sizeof(struct track); | |
603 | ||
604 | if (s->size == off) | |
605 | return 1; | |
606 | ||
24922684 CL |
607 | return check_bytes_and_report(s, page, p, "Object padding", |
608 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
609 | } |
610 | ||
39b26464 | 611 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
612 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
613 | { | |
24922684 CL |
614 | u8 *start; |
615 | u8 *fault; | |
616 | u8 *end; | |
617 | int length; | |
618 | int remainder; | |
81819f0f CL |
619 | |
620 | if (!(s->flags & SLAB_POISON)) | |
621 | return 1; | |
622 | ||
a973e9dd | 623 | start = page_address(page); |
834f3d11 | 624 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
625 | end = start + length; |
626 | remainder = length % s->size; | |
81819f0f CL |
627 | if (!remainder) |
628 | return 1; | |
629 | ||
39b26464 | 630 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
631 | if (!fault) |
632 | return 1; | |
633 | while (end > fault && end[-1] == POISON_INUSE) | |
634 | end--; | |
635 | ||
636 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 637 | print_section("Padding", end - remainder, remainder); |
24922684 | 638 | |
8a3d271d | 639 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 640 | return 0; |
81819f0f CL |
641 | } |
642 | ||
643 | static int check_object(struct kmem_cache *s, struct page *page, | |
644 | void *object, int active) | |
645 | { | |
646 | u8 *p = object; | |
647 | u8 *endobject = object + s->objsize; | |
648 | ||
649 | if (s->flags & SLAB_RED_ZONE) { | |
650 | unsigned int red = | |
651 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
652 | ||
24922684 CL |
653 | if (!check_bytes_and_report(s, page, object, "Redzone", |
654 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 655 | return 0; |
81819f0f | 656 | } else { |
3adbefee IM |
657 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
658 | check_bytes_and_report(s, page, p, "Alignment padding", | |
659 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
660 | } | |
81819f0f CL |
661 | } |
662 | ||
663 | if (s->flags & SLAB_POISON) { | |
664 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
665 | (!check_bytes_and_report(s, page, p, "Poison", p, |
666 | POISON_FREE, s->objsize - 1) || | |
667 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 668 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 669 | return 0; |
81819f0f CL |
670 | /* |
671 | * check_pad_bytes cleans up on its own. | |
672 | */ | |
673 | check_pad_bytes(s, page, p); | |
674 | } | |
675 | ||
676 | if (!s->offset && active) | |
677 | /* | |
678 | * Object and freepointer overlap. Cannot check | |
679 | * freepointer while object is allocated. | |
680 | */ | |
681 | return 1; | |
682 | ||
683 | /* Check free pointer validity */ | |
684 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
685 | object_err(s, page, p, "Freepointer corrupt"); | |
686 | /* | |
9f6c708e | 687 | * No choice but to zap it and thus lose the remainder |
81819f0f | 688 | * of the free objects in this slab. May cause |
672bba3a | 689 | * another error because the object count is now wrong. |
81819f0f | 690 | */ |
a973e9dd | 691 | set_freepointer(s, p, NULL); |
81819f0f CL |
692 | return 0; |
693 | } | |
694 | return 1; | |
695 | } | |
696 | ||
697 | static int check_slab(struct kmem_cache *s, struct page *page) | |
698 | { | |
39b26464 CL |
699 | int maxobj; |
700 | ||
81819f0f CL |
701 | VM_BUG_ON(!irqs_disabled()); |
702 | ||
703 | if (!PageSlab(page)) { | |
24922684 | 704 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
705 | return 0; |
706 | } | |
39b26464 CL |
707 | |
708 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
709 | if (page->objects > maxobj) { | |
710 | slab_err(s, page, "objects %u > max %u", | |
711 | s->name, page->objects, maxobj); | |
712 | return 0; | |
713 | } | |
714 | if (page->inuse > page->objects) { | |
24922684 | 715 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 716 | s->name, page->inuse, page->objects); |
81819f0f CL |
717 | return 0; |
718 | } | |
719 | /* Slab_pad_check fixes things up after itself */ | |
720 | slab_pad_check(s, page); | |
721 | return 1; | |
722 | } | |
723 | ||
724 | /* | |
672bba3a CL |
725 | * Determine if a certain object on a page is on the freelist. Must hold the |
726 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
727 | */ |
728 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
729 | { | |
730 | int nr = 0; | |
731 | void *fp = page->freelist; | |
732 | void *object = NULL; | |
224a88be | 733 | unsigned long max_objects; |
81819f0f | 734 | |
39b26464 | 735 | while (fp && nr <= page->objects) { |
81819f0f CL |
736 | if (fp == search) |
737 | return 1; | |
738 | if (!check_valid_pointer(s, page, fp)) { | |
739 | if (object) { | |
740 | object_err(s, page, object, | |
741 | "Freechain corrupt"); | |
a973e9dd | 742 | set_freepointer(s, object, NULL); |
81819f0f CL |
743 | break; |
744 | } else { | |
24922684 | 745 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 746 | page->freelist = NULL; |
39b26464 | 747 | page->inuse = page->objects; |
24922684 | 748 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
749 | return 0; |
750 | } | |
751 | break; | |
752 | } | |
753 | object = fp; | |
754 | fp = get_freepointer(s, object); | |
755 | nr++; | |
756 | } | |
757 | ||
224a88be | 758 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
759 | if (max_objects > MAX_OBJS_PER_PAGE) |
760 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
761 | |
762 | if (page->objects != max_objects) { | |
763 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
764 | "should be %d", page->objects, max_objects); | |
765 | page->objects = max_objects; | |
766 | slab_fix(s, "Number of objects adjusted."); | |
767 | } | |
39b26464 | 768 | if (page->inuse != page->objects - nr) { |
70d71228 | 769 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
770 | "counted were %d", page->inuse, page->objects - nr); |
771 | page->inuse = page->objects - nr; | |
24922684 | 772 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
773 | } |
774 | return search == NULL; | |
775 | } | |
776 | ||
0121c619 CL |
777 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
778 | int alloc) | |
3ec09742 CL |
779 | { |
780 | if (s->flags & SLAB_TRACE) { | |
781 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
782 | s->name, | |
783 | alloc ? "alloc" : "free", | |
784 | object, page->inuse, | |
785 | page->freelist); | |
786 | ||
787 | if (!alloc) | |
788 | print_section("Object", (void *)object, s->objsize); | |
789 | ||
790 | dump_stack(); | |
791 | } | |
792 | } | |
793 | ||
c016b0bd CL |
794 | /* |
795 | * Hooks for other subsystems that check memory allocations. In a typical | |
796 | * production configuration these hooks all should produce no code at all. | |
797 | */ | |
798 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
799 | { | |
c1d50836 | 800 | flags &= gfp_allowed_mask; |
c016b0bd CL |
801 | lockdep_trace_alloc(flags); |
802 | might_sleep_if(flags & __GFP_WAIT); | |
803 | ||
804 | return should_failslab(s->objsize, flags, s->flags); | |
805 | } | |
806 | ||
807 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
808 | { | |
c1d50836 | 809 | flags &= gfp_allowed_mask; |
c016b0bd CL |
810 | kmemcheck_slab_alloc(s, flags, object, s->objsize); |
811 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | |
812 | } | |
813 | ||
814 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
815 | { | |
816 | kmemleak_free_recursive(x, s->flags); | |
817 | } | |
818 | ||
819 | static inline void slab_free_hook_irq(struct kmem_cache *s, void *object) | |
820 | { | |
821 | kmemcheck_slab_free(s, object, s->objsize); | |
822 | debug_check_no_locks_freed(object, s->objsize); | |
823 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
824 | debug_check_no_obj_freed(object, s->objsize); | |
825 | } | |
826 | ||
643b1138 | 827 | /* |
672bba3a | 828 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 829 | */ |
e95eed57 | 830 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 831 | { |
643b1138 CL |
832 | spin_lock(&n->list_lock); |
833 | list_add(&page->lru, &n->full); | |
834 | spin_unlock(&n->list_lock); | |
835 | } | |
836 | ||
837 | static void remove_full(struct kmem_cache *s, struct page *page) | |
838 | { | |
839 | struct kmem_cache_node *n; | |
840 | ||
841 | if (!(s->flags & SLAB_STORE_USER)) | |
842 | return; | |
843 | ||
844 | n = get_node(s, page_to_nid(page)); | |
845 | ||
846 | spin_lock(&n->list_lock); | |
847 | list_del(&page->lru); | |
848 | spin_unlock(&n->list_lock); | |
849 | } | |
850 | ||
0f389ec6 CL |
851 | /* Tracking of the number of slabs for debugging purposes */ |
852 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
853 | { | |
854 | struct kmem_cache_node *n = get_node(s, node); | |
855 | ||
856 | return atomic_long_read(&n->nr_slabs); | |
857 | } | |
858 | ||
26c02cf0 AB |
859 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
860 | { | |
861 | return atomic_long_read(&n->nr_slabs); | |
862 | } | |
863 | ||
205ab99d | 864 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
865 | { |
866 | struct kmem_cache_node *n = get_node(s, node); | |
867 | ||
868 | /* | |
869 | * May be called early in order to allocate a slab for the | |
870 | * kmem_cache_node structure. Solve the chicken-egg | |
871 | * dilemma by deferring the increment of the count during | |
872 | * bootstrap (see early_kmem_cache_node_alloc). | |
873 | */ | |
205ab99d | 874 | if (!NUMA_BUILD || n) { |
0f389ec6 | 875 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
876 | atomic_long_add(objects, &n->total_objects); |
877 | } | |
0f389ec6 | 878 | } |
205ab99d | 879 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
880 | { |
881 | struct kmem_cache_node *n = get_node(s, node); | |
882 | ||
883 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 884 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
885 | } |
886 | ||
887 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
888 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
889 | void *object) | |
890 | { | |
891 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
892 | return; | |
893 | ||
894 | init_object(s, object, 0); | |
895 | init_tracking(s, object); | |
896 | } | |
897 | ||
1537066c | 898 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 899 | void *object, unsigned long addr) |
81819f0f CL |
900 | { |
901 | if (!check_slab(s, page)) | |
902 | goto bad; | |
903 | ||
d692ef6d | 904 | if (!on_freelist(s, page, object)) { |
24922684 | 905 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 906 | goto bad; |
81819f0f CL |
907 | } |
908 | ||
909 | if (!check_valid_pointer(s, page, object)) { | |
910 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 911 | goto bad; |
81819f0f CL |
912 | } |
913 | ||
d692ef6d | 914 | if (!check_object(s, page, object, 0)) |
81819f0f | 915 | goto bad; |
81819f0f | 916 | |
3ec09742 CL |
917 | /* Success perform special debug activities for allocs */ |
918 | if (s->flags & SLAB_STORE_USER) | |
919 | set_track(s, object, TRACK_ALLOC, addr); | |
920 | trace(s, page, object, 1); | |
921 | init_object(s, object, 1); | |
81819f0f | 922 | return 1; |
3ec09742 | 923 | |
81819f0f CL |
924 | bad: |
925 | if (PageSlab(page)) { | |
926 | /* | |
927 | * If this is a slab page then lets do the best we can | |
928 | * to avoid issues in the future. Marking all objects | |
672bba3a | 929 | * as used avoids touching the remaining objects. |
81819f0f | 930 | */ |
24922684 | 931 | slab_fix(s, "Marking all objects used"); |
39b26464 | 932 | page->inuse = page->objects; |
a973e9dd | 933 | page->freelist = NULL; |
81819f0f CL |
934 | } |
935 | return 0; | |
936 | } | |
937 | ||
1537066c CL |
938 | static noinline int free_debug_processing(struct kmem_cache *s, |
939 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
940 | { |
941 | if (!check_slab(s, page)) | |
942 | goto fail; | |
943 | ||
944 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 945 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
946 | goto fail; |
947 | } | |
948 | ||
949 | if (on_freelist(s, page, object)) { | |
24922684 | 950 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
951 | goto fail; |
952 | } | |
953 | ||
954 | if (!check_object(s, page, object, 1)) | |
955 | return 0; | |
956 | ||
957 | if (unlikely(s != page->slab)) { | |
3adbefee | 958 | if (!PageSlab(page)) { |
70d71228 CL |
959 | slab_err(s, page, "Attempt to free object(0x%p) " |
960 | "outside of slab", object); | |
3adbefee | 961 | } else if (!page->slab) { |
81819f0f | 962 | printk(KERN_ERR |
70d71228 | 963 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 964 | object); |
70d71228 | 965 | dump_stack(); |
06428780 | 966 | } else |
24922684 CL |
967 | object_err(s, page, object, |
968 | "page slab pointer corrupt."); | |
81819f0f CL |
969 | goto fail; |
970 | } | |
3ec09742 CL |
971 | |
972 | /* Special debug activities for freeing objects */ | |
8a38082d | 973 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
974 | remove_full(s, page); |
975 | if (s->flags & SLAB_STORE_USER) | |
976 | set_track(s, object, TRACK_FREE, addr); | |
977 | trace(s, page, object, 0); | |
978 | init_object(s, object, 0); | |
81819f0f | 979 | return 1; |
3ec09742 | 980 | |
81819f0f | 981 | fail: |
24922684 | 982 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
983 | return 0; |
984 | } | |
985 | ||
41ecc55b CL |
986 | static int __init setup_slub_debug(char *str) |
987 | { | |
f0630fff CL |
988 | slub_debug = DEBUG_DEFAULT_FLAGS; |
989 | if (*str++ != '=' || !*str) | |
990 | /* | |
991 | * No options specified. Switch on full debugging. | |
992 | */ | |
993 | goto out; | |
994 | ||
995 | if (*str == ',') | |
996 | /* | |
997 | * No options but restriction on slabs. This means full | |
998 | * debugging for slabs matching a pattern. | |
999 | */ | |
1000 | goto check_slabs; | |
1001 | ||
fa5ec8a1 DR |
1002 | if (tolower(*str) == 'o') { |
1003 | /* | |
1004 | * Avoid enabling debugging on caches if its minimum order | |
1005 | * would increase as a result. | |
1006 | */ | |
1007 | disable_higher_order_debug = 1; | |
1008 | goto out; | |
1009 | } | |
1010 | ||
f0630fff CL |
1011 | slub_debug = 0; |
1012 | if (*str == '-') | |
1013 | /* | |
1014 | * Switch off all debugging measures. | |
1015 | */ | |
1016 | goto out; | |
1017 | ||
1018 | /* | |
1019 | * Determine which debug features should be switched on | |
1020 | */ | |
06428780 | 1021 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1022 | switch (tolower(*str)) { |
1023 | case 'f': | |
1024 | slub_debug |= SLAB_DEBUG_FREE; | |
1025 | break; | |
1026 | case 'z': | |
1027 | slub_debug |= SLAB_RED_ZONE; | |
1028 | break; | |
1029 | case 'p': | |
1030 | slub_debug |= SLAB_POISON; | |
1031 | break; | |
1032 | case 'u': | |
1033 | slub_debug |= SLAB_STORE_USER; | |
1034 | break; | |
1035 | case 't': | |
1036 | slub_debug |= SLAB_TRACE; | |
1037 | break; | |
4c13dd3b DM |
1038 | case 'a': |
1039 | slub_debug |= SLAB_FAILSLAB; | |
1040 | break; | |
f0630fff CL |
1041 | default: |
1042 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1043 | "unknown. skipped\n", *str); |
f0630fff | 1044 | } |
41ecc55b CL |
1045 | } |
1046 | ||
f0630fff | 1047 | check_slabs: |
41ecc55b CL |
1048 | if (*str == ',') |
1049 | slub_debug_slabs = str + 1; | |
f0630fff | 1050 | out: |
41ecc55b CL |
1051 | return 1; |
1052 | } | |
1053 | ||
1054 | __setup("slub_debug", setup_slub_debug); | |
1055 | ||
ba0268a8 CL |
1056 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1057 | unsigned long flags, const char *name, | |
51cc5068 | 1058 | void (*ctor)(void *)) |
41ecc55b CL |
1059 | { |
1060 | /* | |
e153362a | 1061 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1062 | */ |
e153362a | 1063 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1064 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1065 | flags |= slub_debug; | |
ba0268a8 CL |
1066 | |
1067 | return flags; | |
41ecc55b CL |
1068 | } |
1069 | #else | |
3ec09742 CL |
1070 | static inline void setup_object_debug(struct kmem_cache *s, |
1071 | struct page *page, void *object) {} | |
41ecc55b | 1072 | |
3ec09742 | 1073 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1074 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1075 | |
3ec09742 | 1076 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1077 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1078 | |
41ecc55b CL |
1079 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1080 | { return 1; } | |
1081 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1082 | void *object, int active) { return 1; } | |
3ec09742 | 1083 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1084 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1085 | unsigned long flags, const char *name, | |
51cc5068 | 1086 | void (*ctor)(void *)) |
ba0268a8 CL |
1087 | { |
1088 | return flags; | |
1089 | } | |
41ecc55b | 1090 | #define slub_debug 0 |
0f389ec6 | 1091 | |
fdaa45e9 IM |
1092 | #define disable_higher_order_debug 0 |
1093 | ||
0f389ec6 CL |
1094 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1095 | { return 0; } | |
26c02cf0 AB |
1096 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1097 | { return 0; } | |
205ab99d CL |
1098 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1099 | int objects) {} | |
1100 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1101 | int objects) {} | |
7d550c56 CL |
1102 | |
1103 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1104 | { return 0; } | |
1105 | ||
1106 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1107 | void *object) {} | |
1108 | ||
1109 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1110 | ||
1111 | static inline void slab_free_hook_irq(struct kmem_cache *s, | |
1112 | void *object) {} | |
1113 | ||
41ecc55b | 1114 | #endif |
205ab99d | 1115 | |
81819f0f CL |
1116 | /* |
1117 | * Slab allocation and freeing | |
1118 | */ | |
65c3376a CL |
1119 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1120 | struct kmem_cache_order_objects oo) | |
1121 | { | |
1122 | int order = oo_order(oo); | |
1123 | ||
b1eeab67 VN |
1124 | flags |= __GFP_NOTRACK; |
1125 | ||
2154a336 | 1126 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1127 | return alloc_pages(flags, order); |
1128 | else | |
6b65aaf3 | 1129 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1130 | } |
1131 | ||
81819f0f CL |
1132 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1133 | { | |
06428780 | 1134 | struct page *page; |
834f3d11 | 1135 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1136 | gfp_t alloc_gfp; |
81819f0f | 1137 | |
b7a49f0d | 1138 | flags |= s->allocflags; |
e12ba74d | 1139 | |
ba52270d PE |
1140 | /* |
1141 | * Let the initial higher-order allocation fail under memory pressure | |
1142 | * so we fall-back to the minimum order allocation. | |
1143 | */ | |
1144 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1145 | ||
1146 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1147 | if (unlikely(!page)) { |
1148 | oo = s->min; | |
1149 | /* | |
1150 | * Allocation may have failed due to fragmentation. | |
1151 | * Try a lower order alloc if possible | |
1152 | */ | |
1153 | page = alloc_slab_page(flags, node, oo); | |
1154 | if (!page) | |
1155 | return NULL; | |
81819f0f | 1156 | |
84e554e6 | 1157 | stat(s, ORDER_FALLBACK); |
65c3376a | 1158 | } |
5a896d9e VN |
1159 | |
1160 | if (kmemcheck_enabled | |
5086c389 | 1161 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1162 | int pages = 1 << oo_order(oo); |
1163 | ||
1164 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1165 | ||
1166 | /* | |
1167 | * Objects from caches that have a constructor don't get | |
1168 | * cleared when they're allocated, so we need to do it here. | |
1169 | */ | |
1170 | if (s->ctor) | |
1171 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1172 | else | |
1173 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1174 | } |
1175 | ||
834f3d11 | 1176 | page->objects = oo_objects(oo); |
81819f0f CL |
1177 | mod_zone_page_state(page_zone(page), |
1178 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1179 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1180 | 1 << oo_order(oo)); |
81819f0f CL |
1181 | |
1182 | return page; | |
1183 | } | |
1184 | ||
1185 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1186 | void *object) | |
1187 | { | |
3ec09742 | 1188 | setup_object_debug(s, page, object); |
4f104934 | 1189 | if (unlikely(s->ctor)) |
51cc5068 | 1190 | s->ctor(object); |
81819f0f CL |
1191 | } |
1192 | ||
1193 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1194 | { | |
1195 | struct page *page; | |
81819f0f | 1196 | void *start; |
81819f0f CL |
1197 | void *last; |
1198 | void *p; | |
1199 | ||
6cb06229 | 1200 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1201 | |
6cb06229 CL |
1202 | page = allocate_slab(s, |
1203 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1204 | if (!page) |
1205 | goto out; | |
1206 | ||
205ab99d | 1207 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1208 | page->slab = s; |
1209 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1210 | |
1211 | start = page_address(page); | |
81819f0f CL |
1212 | |
1213 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1214 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1215 | |
1216 | last = start; | |
224a88be | 1217 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1218 | setup_object(s, page, last); |
1219 | set_freepointer(s, last, p); | |
1220 | last = p; | |
1221 | } | |
1222 | setup_object(s, page, last); | |
a973e9dd | 1223 | set_freepointer(s, last, NULL); |
81819f0f CL |
1224 | |
1225 | page->freelist = start; | |
1226 | page->inuse = 0; | |
1227 | out: | |
81819f0f CL |
1228 | return page; |
1229 | } | |
1230 | ||
1231 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1232 | { | |
834f3d11 CL |
1233 | int order = compound_order(page); |
1234 | int pages = 1 << order; | |
81819f0f | 1235 | |
af537b0a | 1236 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1237 | void *p; |
1238 | ||
1239 | slab_pad_check(s, page); | |
224a88be CL |
1240 | for_each_object(p, s, page_address(page), |
1241 | page->objects) | |
81819f0f | 1242 | check_object(s, page, p, 0); |
81819f0f CL |
1243 | } |
1244 | ||
b1eeab67 | 1245 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1246 | |
81819f0f CL |
1247 | mod_zone_page_state(page_zone(page), |
1248 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1249 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1250 | -pages); |
81819f0f | 1251 | |
49bd5221 CL |
1252 | __ClearPageSlab(page); |
1253 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1254 | if (current->reclaim_state) |
1255 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1256 | __free_pages(page, order); |
81819f0f CL |
1257 | } |
1258 | ||
1259 | static void rcu_free_slab(struct rcu_head *h) | |
1260 | { | |
1261 | struct page *page; | |
1262 | ||
1263 | page = container_of((struct list_head *)h, struct page, lru); | |
1264 | __free_slab(page->slab, page); | |
1265 | } | |
1266 | ||
1267 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1268 | { | |
1269 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1270 | /* | |
1271 | * RCU free overloads the RCU head over the LRU | |
1272 | */ | |
1273 | struct rcu_head *head = (void *)&page->lru; | |
1274 | ||
1275 | call_rcu(head, rcu_free_slab); | |
1276 | } else | |
1277 | __free_slab(s, page); | |
1278 | } | |
1279 | ||
1280 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1281 | { | |
205ab99d | 1282 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1283 | free_slab(s, page); |
1284 | } | |
1285 | ||
1286 | /* | |
1287 | * Per slab locking using the pagelock | |
1288 | */ | |
1289 | static __always_inline void slab_lock(struct page *page) | |
1290 | { | |
1291 | bit_spin_lock(PG_locked, &page->flags); | |
1292 | } | |
1293 | ||
1294 | static __always_inline void slab_unlock(struct page *page) | |
1295 | { | |
a76d3546 | 1296 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1297 | } |
1298 | ||
1299 | static __always_inline int slab_trylock(struct page *page) | |
1300 | { | |
1301 | int rc = 1; | |
1302 | ||
1303 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1304 | return rc; | |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Management of partially allocated slabs | |
1309 | */ | |
7c2e132c CL |
1310 | static void add_partial(struct kmem_cache_node *n, |
1311 | struct page *page, int tail) | |
81819f0f | 1312 | { |
e95eed57 CL |
1313 | spin_lock(&n->list_lock); |
1314 | n->nr_partial++; | |
7c2e132c CL |
1315 | if (tail) |
1316 | list_add_tail(&page->lru, &n->partial); | |
1317 | else | |
1318 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1319 | spin_unlock(&n->list_lock); |
1320 | } | |
1321 | ||
0121c619 | 1322 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1323 | { |
1324 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1325 | ||
1326 | spin_lock(&n->list_lock); | |
1327 | list_del(&page->lru); | |
1328 | n->nr_partial--; | |
1329 | spin_unlock(&n->list_lock); | |
1330 | } | |
1331 | ||
1332 | /* | |
672bba3a | 1333 | * Lock slab and remove from the partial list. |
81819f0f | 1334 | * |
672bba3a | 1335 | * Must hold list_lock. |
81819f0f | 1336 | */ |
0121c619 CL |
1337 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1338 | struct page *page) | |
81819f0f CL |
1339 | { |
1340 | if (slab_trylock(page)) { | |
1341 | list_del(&page->lru); | |
1342 | n->nr_partial--; | |
8a38082d | 1343 | __SetPageSlubFrozen(page); |
81819f0f CL |
1344 | return 1; |
1345 | } | |
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | /* | |
672bba3a | 1350 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1351 | */ |
1352 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1353 | { | |
1354 | struct page *page; | |
1355 | ||
1356 | /* | |
1357 | * Racy check. If we mistakenly see no partial slabs then we | |
1358 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1359 | * partial slab and there is none available then get_partials() |
1360 | * will return NULL. | |
81819f0f CL |
1361 | */ |
1362 | if (!n || !n->nr_partial) | |
1363 | return NULL; | |
1364 | ||
1365 | spin_lock(&n->list_lock); | |
1366 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1367 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1368 | goto out; |
1369 | page = NULL; | |
1370 | out: | |
1371 | spin_unlock(&n->list_lock); | |
1372 | return page; | |
1373 | } | |
1374 | ||
1375 | /* | |
672bba3a | 1376 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1377 | */ |
1378 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1379 | { | |
1380 | #ifdef CONFIG_NUMA | |
1381 | struct zonelist *zonelist; | |
dd1a239f | 1382 | struct zoneref *z; |
54a6eb5c MG |
1383 | struct zone *zone; |
1384 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1385 | struct page *page; |
1386 | ||
1387 | /* | |
672bba3a CL |
1388 | * The defrag ratio allows a configuration of the tradeoffs between |
1389 | * inter node defragmentation and node local allocations. A lower | |
1390 | * defrag_ratio increases the tendency to do local allocations | |
1391 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1392 | * |
672bba3a CL |
1393 | * If the defrag_ratio is set to 0 then kmalloc() always |
1394 | * returns node local objects. If the ratio is higher then kmalloc() | |
1395 | * may return off node objects because partial slabs are obtained | |
1396 | * from other nodes and filled up. | |
81819f0f | 1397 | * |
6446faa2 | 1398 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1399 | * defrag_ratio = 1000) then every (well almost) allocation will |
1400 | * first attempt to defrag slab caches on other nodes. This means | |
1401 | * scanning over all nodes to look for partial slabs which may be | |
1402 | * expensive if we do it every time we are trying to find a slab | |
1403 | * with available objects. | |
81819f0f | 1404 | */ |
9824601e CL |
1405 | if (!s->remote_node_defrag_ratio || |
1406 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1407 | return NULL; |
1408 | ||
c0ff7453 | 1409 | get_mems_allowed(); |
0e88460d | 1410 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1411 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1412 | struct kmem_cache_node *n; |
1413 | ||
54a6eb5c | 1414 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1415 | |
54a6eb5c | 1416 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1417 | n->nr_partial > s->min_partial) { |
81819f0f | 1418 | page = get_partial_node(n); |
c0ff7453 MX |
1419 | if (page) { |
1420 | put_mems_allowed(); | |
81819f0f | 1421 | return page; |
c0ff7453 | 1422 | } |
81819f0f CL |
1423 | } |
1424 | } | |
c0ff7453 | 1425 | put_mems_allowed(); |
81819f0f CL |
1426 | #endif |
1427 | return NULL; | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * Get a partial page, lock it and return it. | |
1432 | */ | |
1433 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1434 | { | |
1435 | struct page *page; | |
2154a336 | 1436 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1437 | |
1438 | page = get_partial_node(get_node(s, searchnode)); | |
bc6488e9 | 1439 | if (page || node != -1) |
81819f0f CL |
1440 | return page; |
1441 | ||
1442 | return get_any_partial(s, flags); | |
1443 | } | |
1444 | ||
1445 | /* | |
1446 | * Move a page back to the lists. | |
1447 | * | |
1448 | * Must be called with the slab lock held. | |
1449 | * | |
1450 | * On exit the slab lock will have been dropped. | |
1451 | */ | |
7c2e132c | 1452 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1453 | { |
e95eed57 CL |
1454 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1455 | ||
8a38082d | 1456 | __ClearPageSlubFrozen(page); |
81819f0f | 1457 | if (page->inuse) { |
e95eed57 | 1458 | |
a973e9dd | 1459 | if (page->freelist) { |
7c2e132c | 1460 | add_partial(n, page, tail); |
84e554e6 | 1461 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1462 | } else { |
84e554e6 | 1463 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1464 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1465 | add_full(n, page); |
1466 | } | |
81819f0f CL |
1467 | slab_unlock(page); |
1468 | } else { | |
84e554e6 | 1469 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1470 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1471 | /* |
672bba3a CL |
1472 | * Adding an empty slab to the partial slabs in order |
1473 | * to avoid page allocator overhead. This slab needs | |
1474 | * to come after the other slabs with objects in | |
6446faa2 CL |
1475 | * so that the others get filled first. That way the |
1476 | * size of the partial list stays small. | |
1477 | * | |
0121c619 CL |
1478 | * kmem_cache_shrink can reclaim any empty slabs from |
1479 | * the partial list. | |
e95eed57 | 1480 | */ |
7c2e132c | 1481 | add_partial(n, page, 1); |
e95eed57 CL |
1482 | slab_unlock(page); |
1483 | } else { | |
1484 | slab_unlock(page); | |
84e554e6 | 1485 | stat(s, FREE_SLAB); |
e95eed57 CL |
1486 | discard_slab(s, page); |
1487 | } | |
81819f0f CL |
1488 | } |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * Remove the cpu slab | |
1493 | */ | |
dfb4f096 | 1494 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1495 | { |
dfb4f096 | 1496 | struct page *page = c->page; |
7c2e132c | 1497 | int tail = 1; |
8ff12cfc | 1498 | |
b773ad73 | 1499 | if (page->freelist) |
84e554e6 | 1500 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1501 | /* |
6446faa2 | 1502 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1503 | * because both freelists are empty. So this is unlikely |
1504 | * to occur. | |
1505 | */ | |
a973e9dd | 1506 | while (unlikely(c->freelist)) { |
894b8788 CL |
1507 | void **object; |
1508 | ||
7c2e132c CL |
1509 | tail = 0; /* Hot objects. Put the slab first */ |
1510 | ||
894b8788 | 1511 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1512 | object = c->freelist; |
ff12059e | 1513 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1514 | |
1515 | /* And put onto the regular freelist */ | |
ff12059e | 1516 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1517 | page->freelist = object; |
1518 | page->inuse--; | |
1519 | } | |
dfb4f096 | 1520 | c->page = NULL; |
7c2e132c | 1521 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1522 | } |
1523 | ||
dfb4f096 | 1524 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1525 | { |
84e554e6 | 1526 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1527 | slab_lock(c->page); |
1528 | deactivate_slab(s, c); | |
81819f0f CL |
1529 | } |
1530 | ||
1531 | /* | |
1532 | * Flush cpu slab. | |
6446faa2 | 1533 | * |
81819f0f CL |
1534 | * Called from IPI handler with interrupts disabled. |
1535 | */ | |
0c710013 | 1536 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1537 | { |
9dfc6e68 | 1538 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1539 | |
dfb4f096 CL |
1540 | if (likely(c && c->page)) |
1541 | flush_slab(s, c); | |
81819f0f CL |
1542 | } |
1543 | ||
1544 | static void flush_cpu_slab(void *d) | |
1545 | { | |
1546 | struct kmem_cache *s = d; | |
81819f0f | 1547 | |
dfb4f096 | 1548 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1549 | } |
1550 | ||
1551 | static void flush_all(struct kmem_cache *s) | |
1552 | { | |
15c8b6c1 | 1553 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1554 | } |
1555 | ||
dfb4f096 CL |
1556 | /* |
1557 | * Check if the objects in a per cpu structure fit numa | |
1558 | * locality expectations. | |
1559 | */ | |
1560 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1561 | { | |
1562 | #ifdef CONFIG_NUMA | |
2154a336 | 1563 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1564 | return 0; |
1565 | #endif | |
1566 | return 1; | |
1567 | } | |
1568 | ||
781b2ba6 PE |
1569 | static int count_free(struct page *page) |
1570 | { | |
1571 | return page->objects - page->inuse; | |
1572 | } | |
1573 | ||
1574 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1575 | int (*get_count)(struct page *)) | |
1576 | { | |
1577 | unsigned long flags; | |
1578 | unsigned long x = 0; | |
1579 | struct page *page; | |
1580 | ||
1581 | spin_lock_irqsave(&n->list_lock, flags); | |
1582 | list_for_each_entry(page, &n->partial, lru) | |
1583 | x += get_count(page); | |
1584 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1585 | return x; | |
1586 | } | |
1587 | ||
26c02cf0 AB |
1588 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1589 | { | |
1590 | #ifdef CONFIG_SLUB_DEBUG | |
1591 | return atomic_long_read(&n->total_objects); | |
1592 | #else | |
1593 | return 0; | |
1594 | #endif | |
1595 | } | |
1596 | ||
781b2ba6 PE |
1597 | static noinline void |
1598 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1599 | { | |
1600 | int node; | |
1601 | ||
1602 | printk(KERN_WARNING | |
1603 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1604 | nid, gfpflags); | |
1605 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1606 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1607 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1608 | ||
fa5ec8a1 DR |
1609 | if (oo_order(s->min) > get_order(s->objsize)) |
1610 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1611 | "slub_debug=O to disable.\n", s->name); | |
1612 | ||
781b2ba6 PE |
1613 | for_each_online_node(node) { |
1614 | struct kmem_cache_node *n = get_node(s, node); | |
1615 | unsigned long nr_slabs; | |
1616 | unsigned long nr_objs; | |
1617 | unsigned long nr_free; | |
1618 | ||
1619 | if (!n) | |
1620 | continue; | |
1621 | ||
26c02cf0 AB |
1622 | nr_free = count_partial(n, count_free); |
1623 | nr_slabs = node_nr_slabs(n); | |
1624 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1625 | |
1626 | printk(KERN_WARNING | |
1627 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1628 | node, nr_slabs, nr_objs, nr_free); | |
1629 | } | |
1630 | } | |
1631 | ||
81819f0f | 1632 | /* |
894b8788 CL |
1633 | * Slow path. The lockless freelist is empty or we need to perform |
1634 | * debugging duties. | |
1635 | * | |
1636 | * Interrupts are disabled. | |
81819f0f | 1637 | * |
894b8788 CL |
1638 | * Processing is still very fast if new objects have been freed to the |
1639 | * regular freelist. In that case we simply take over the regular freelist | |
1640 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1641 | * |
894b8788 CL |
1642 | * If that is not working then we fall back to the partial lists. We take the |
1643 | * first element of the freelist as the object to allocate now and move the | |
1644 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1645 | * |
894b8788 | 1646 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1647 | * we need to allocate a new slab. This is the slowest path since it involves |
1648 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1649 | */ |
ce71e27c EGM |
1650 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1651 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1652 | { |
81819f0f | 1653 | void **object; |
dfb4f096 | 1654 | struct page *new; |
81819f0f | 1655 | |
e72e9c23 LT |
1656 | /* We handle __GFP_ZERO in the caller */ |
1657 | gfpflags &= ~__GFP_ZERO; | |
1658 | ||
dfb4f096 | 1659 | if (!c->page) |
81819f0f CL |
1660 | goto new_slab; |
1661 | ||
dfb4f096 CL |
1662 | slab_lock(c->page); |
1663 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1664 | goto another_slab; |
6446faa2 | 1665 | |
84e554e6 | 1666 | stat(s, ALLOC_REFILL); |
6446faa2 | 1667 | |
894b8788 | 1668 | load_freelist: |
dfb4f096 | 1669 | object = c->page->freelist; |
a973e9dd | 1670 | if (unlikely(!object)) |
81819f0f | 1671 | goto another_slab; |
af537b0a | 1672 | if (kmem_cache_debug(s)) |
81819f0f CL |
1673 | goto debug; |
1674 | ||
ff12059e | 1675 | c->freelist = get_freepointer(s, object); |
39b26464 | 1676 | c->page->inuse = c->page->objects; |
a973e9dd | 1677 | c->page->freelist = NULL; |
dfb4f096 | 1678 | c->node = page_to_nid(c->page); |
1f84260c | 1679 | unlock_out: |
dfb4f096 | 1680 | slab_unlock(c->page); |
84e554e6 | 1681 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1682 | return object; |
1683 | ||
1684 | another_slab: | |
dfb4f096 | 1685 | deactivate_slab(s, c); |
81819f0f CL |
1686 | |
1687 | new_slab: | |
dfb4f096 CL |
1688 | new = get_partial(s, gfpflags, node); |
1689 | if (new) { | |
1690 | c->page = new; | |
84e554e6 | 1691 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1692 | goto load_freelist; |
81819f0f CL |
1693 | } |
1694 | ||
c1d50836 | 1695 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1696 | if (gfpflags & __GFP_WAIT) |
1697 | local_irq_enable(); | |
1698 | ||
dfb4f096 | 1699 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1700 | |
1701 | if (gfpflags & __GFP_WAIT) | |
1702 | local_irq_disable(); | |
1703 | ||
dfb4f096 | 1704 | if (new) { |
9dfc6e68 | 1705 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1706 | stat(s, ALLOC_SLAB); |
05aa3450 | 1707 | if (c->page) |
dfb4f096 | 1708 | flush_slab(s, c); |
dfb4f096 | 1709 | slab_lock(new); |
8a38082d | 1710 | __SetPageSlubFrozen(new); |
dfb4f096 | 1711 | c->page = new; |
4b6f0750 | 1712 | goto load_freelist; |
81819f0f | 1713 | } |
95f85989 PE |
1714 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1715 | slab_out_of_memory(s, gfpflags, node); | |
71c7a06f | 1716 | return NULL; |
81819f0f | 1717 | debug: |
dfb4f096 | 1718 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1719 | goto another_slab; |
894b8788 | 1720 | |
dfb4f096 | 1721 | c->page->inuse++; |
ff12059e | 1722 | c->page->freelist = get_freepointer(s, object); |
ee3c72a1 | 1723 | c->node = -1; |
1f84260c | 1724 | goto unlock_out; |
894b8788 CL |
1725 | } |
1726 | ||
1727 | /* | |
1728 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1729 | * have the fastpath folded into their functions. So no function call | |
1730 | * overhead for requests that can be satisfied on the fastpath. | |
1731 | * | |
1732 | * The fastpath works by first checking if the lockless freelist can be used. | |
1733 | * If not then __slab_alloc is called for slow processing. | |
1734 | * | |
1735 | * Otherwise we can simply pick the next object from the lockless free list. | |
1736 | */ | |
06428780 | 1737 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1738 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1739 | { |
894b8788 | 1740 | void **object; |
dfb4f096 | 1741 | struct kmem_cache_cpu *c; |
1f84260c CL |
1742 | unsigned long flags; |
1743 | ||
c016b0bd | 1744 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1745 | return NULL; |
1f84260c | 1746 | |
894b8788 | 1747 | local_irq_save(flags); |
9dfc6e68 CL |
1748 | c = __this_cpu_ptr(s->cpu_slab); |
1749 | object = c->freelist; | |
9dfc6e68 | 1750 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1751 | |
dfb4f096 | 1752 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1753 | |
1754 | else { | |
ff12059e | 1755 | c->freelist = get_freepointer(s, object); |
84e554e6 | 1756 | stat(s, ALLOC_FASTPATH); |
894b8788 CL |
1757 | } |
1758 | local_irq_restore(flags); | |
d07dbea4 | 1759 | |
74e2134f | 1760 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1761 | memset(object, 0, s->objsize); |
d07dbea4 | 1762 | |
c016b0bd | 1763 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1764 | |
894b8788 | 1765 | return object; |
81819f0f CL |
1766 | } |
1767 | ||
1768 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1769 | { | |
2154a336 | 1770 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1771 | |
ca2b84cb | 1772 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1773 | |
1774 | return ret; | |
81819f0f CL |
1775 | } |
1776 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1777 | ||
0f24f128 | 1778 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1779 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) |
1780 | { | |
2154a336 | 1781 | return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 EGM |
1782 | } |
1783 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1784 | #endif | |
1785 | ||
81819f0f CL |
1786 | #ifdef CONFIG_NUMA |
1787 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1788 | { | |
5b882be4 EGM |
1789 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1790 | ||
ca2b84cb EGM |
1791 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1792 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1793 | |
1794 | return ret; | |
81819f0f CL |
1795 | } |
1796 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1797 | #endif | |
1798 | ||
0f24f128 | 1799 | #ifdef CONFIG_TRACING |
5b882be4 EGM |
1800 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, |
1801 | gfp_t gfpflags, | |
1802 | int node) | |
1803 | { | |
1804 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1805 | } | |
1806 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1807 | #endif | |
1808 | ||
81819f0f | 1809 | /* |
894b8788 CL |
1810 | * Slow patch handling. This may still be called frequently since objects |
1811 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1812 | * |
894b8788 CL |
1813 | * So we still attempt to reduce cache line usage. Just take the slab |
1814 | * lock and free the item. If there is no additional partial page | |
1815 | * handling required then we can return immediately. | |
81819f0f | 1816 | */ |
894b8788 | 1817 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 1818 | void *x, unsigned long addr) |
81819f0f CL |
1819 | { |
1820 | void *prior; | |
1821 | void **object = (void *)x; | |
81819f0f | 1822 | |
84e554e6 | 1823 | stat(s, FREE_SLOWPATH); |
81819f0f CL |
1824 | slab_lock(page); |
1825 | ||
af537b0a | 1826 | if (kmem_cache_debug(s)) |
81819f0f | 1827 | goto debug; |
6446faa2 | 1828 | |
81819f0f | 1829 | checks_ok: |
ff12059e CL |
1830 | prior = page->freelist; |
1831 | set_freepointer(s, object, prior); | |
81819f0f CL |
1832 | page->freelist = object; |
1833 | page->inuse--; | |
1834 | ||
8a38082d | 1835 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 1836 | stat(s, FREE_FROZEN); |
81819f0f | 1837 | goto out_unlock; |
8ff12cfc | 1838 | } |
81819f0f CL |
1839 | |
1840 | if (unlikely(!page->inuse)) | |
1841 | goto slab_empty; | |
1842 | ||
1843 | /* | |
6446faa2 | 1844 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1845 | * then add it. |
1846 | */ | |
a973e9dd | 1847 | if (unlikely(!prior)) { |
7c2e132c | 1848 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 1849 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 1850 | } |
81819f0f CL |
1851 | |
1852 | out_unlock: | |
1853 | slab_unlock(page); | |
81819f0f CL |
1854 | return; |
1855 | ||
1856 | slab_empty: | |
a973e9dd | 1857 | if (prior) { |
81819f0f | 1858 | /* |
672bba3a | 1859 | * Slab still on the partial list. |
81819f0f CL |
1860 | */ |
1861 | remove_partial(s, page); | |
84e554e6 | 1862 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 1863 | } |
81819f0f | 1864 | slab_unlock(page); |
84e554e6 | 1865 | stat(s, FREE_SLAB); |
81819f0f | 1866 | discard_slab(s, page); |
81819f0f CL |
1867 | return; |
1868 | ||
1869 | debug: | |
3ec09742 | 1870 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1871 | goto out_unlock; |
77c5e2d0 | 1872 | goto checks_ok; |
81819f0f CL |
1873 | } |
1874 | ||
894b8788 CL |
1875 | /* |
1876 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1877 | * can perform fastpath freeing without additional function calls. | |
1878 | * | |
1879 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1880 | * of this processor. This typically the case if we have just allocated | |
1881 | * the item before. | |
1882 | * | |
1883 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1884 | * with all sorts of special processing. | |
1885 | */ | |
06428780 | 1886 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1887 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1888 | { |
1889 | void **object = (void *)x; | |
dfb4f096 | 1890 | struct kmem_cache_cpu *c; |
1f84260c CL |
1891 | unsigned long flags; |
1892 | ||
c016b0bd CL |
1893 | slab_free_hook(s, x); |
1894 | ||
894b8788 | 1895 | local_irq_save(flags); |
9dfc6e68 | 1896 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd CL |
1897 | |
1898 | slab_free_hook_irq(s, x); | |
1899 | ||
ee3c72a1 | 1900 | if (likely(page == c->page && c->node >= 0)) { |
ff12059e | 1901 | set_freepointer(s, object, c->freelist); |
dfb4f096 | 1902 | c->freelist = object; |
84e554e6 | 1903 | stat(s, FREE_FASTPATH); |
894b8788 | 1904 | } else |
ff12059e | 1905 | __slab_free(s, page, x, addr); |
894b8788 CL |
1906 | |
1907 | local_irq_restore(flags); | |
1908 | } | |
1909 | ||
81819f0f CL |
1910 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1911 | { | |
77c5e2d0 | 1912 | struct page *page; |
81819f0f | 1913 | |
b49af68f | 1914 | page = virt_to_head_page(x); |
81819f0f | 1915 | |
ce71e27c | 1916 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1917 | |
ca2b84cb | 1918 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1919 | } |
1920 | EXPORT_SYMBOL(kmem_cache_free); | |
1921 | ||
e9beef18 | 1922 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1923 | static struct page *get_object_page(const void *x) |
1924 | { | |
b49af68f | 1925 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1926 | |
1927 | if (!PageSlab(page)) | |
1928 | return NULL; | |
1929 | ||
1930 | return page; | |
1931 | } | |
1932 | ||
1933 | /* | |
672bba3a CL |
1934 | * Object placement in a slab is made very easy because we always start at |
1935 | * offset 0. If we tune the size of the object to the alignment then we can | |
1936 | * get the required alignment by putting one properly sized object after | |
1937 | * another. | |
81819f0f CL |
1938 | * |
1939 | * Notice that the allocation order determines the sizes of the per cpu | |
1940 | * caches. Each processor has always one slab available for allocations. | |
1941 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1942 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1943 | * locking overhead. |
81819f0f CL |
1944 | */ |
1945 | ||
1946 | /* | |
1947 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1948 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1949 | * and increases the number of allocations possible without having to | |
1950 | * take the list_lock. | |
1951 | */ | |
1952 | static int slub_min_order; | |
114e9e89 | 1953 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1954 | static int slub_min_objects; |
81819f0f CL |
1955 | |
1956 | /* | |
1957 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1958 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1959 | */ |
1960 | static int slub_nomerge; | |
1961 | ||
81819f0f CL |
1962 | /* |
1963 | * Calculate the order of allocation given an slab object size. | |
1964 | * | |
672bba3a CL |
1965 | * The order of allocation has significant impact on performance and other |
1966 | * system components. Generally order 0 allocations should be preferred since | |
1967 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1968 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1969 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1970 | * would be wasted. |
1971 | * | |
1972 | * In order to reach satisfactory performance we must ensure that a minimum | |
1973 | * number of objects is in one slab. Otherwise we may generate too much | |
1974 | * activity on the partial lists which requires taking the list_lock. This is | |
1975 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1976 | * |
672bba3a CL |
1977 | * slub_max_order specifies the order where we begin to stop considering the |
1978 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1979 | * we try to keep the page order as low as possible. So we accept more waste | |
1980 | * of space in favor of a small page order. | |
81819f0f | 1981 | * |
672bba3a CL |
1982 | * Higher order allocations also allow the placement of more objects in a |
1983 | * slab and thereby reduce object handling overhead. If the user has | |
1984 | * requested a higher mininum order then we start with that one instead of | |
1985 | * the smallest order which will fit the object. | |
81819f0f | 1986 | */ |
5e6d444e CL |
1987 | static inline int slab_order(int size, int min_objects, |
1988 | int max_order, int fract_leftover) | |
81819f0f CL |
1989 | { |
1990 | int order; | |
1991 | int rem; | |
6300ea75 | 1992 | int min_order = slub_min_order; |
81819f0f | 1993 | |
210b5c06 CG |
1994 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1995 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1996 | |
6300ea75 | 1997 | for (order = max(min_order, |
5e6d444e CL |
1998 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1999 | order <= max_order; order++) { | |
81819f0f | 2000 | |
5e6d444e | 2001 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2002 | |
5e6d444e | 2003 | if (slab_size < min_objects * size) |
81819f0f CL |
2004 | continue; |
2005 | ||
2006 | rem = slab_size % size; | |
2007 | ||
5e6d444e | 2008 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2009 | break; |
2010 | ||
2011 | } | |
672bba3a | 2012 | |
81819f0f CL |
2013 | return order; |
2014 | } | |
2015 | ||
5e6d444e CL |
2016 | static inline int calculate_order(int size) |
2017 | { | |
2018 | int order; | |
2019 | int min_objects; | |
2020 | int fraction; | |
e8120ff1 | 2021 | int max_objects; |
5e6d444e CL |
2022 | |
2023 | /* | |
2024 | * Attempt to find best configuration for a slab. This | |
2025 | * works by first attempting to generate a layout with | |
2026 | * the best configuration and backing off gradually. | |
2027 | * | |
2028 | * First we reduce the acceptable waste in a slab. Then | |
2029 | * we reduce the minimum objects required in a slab. | |
2030 | */ | |
2031 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2032 | if (!min_objects) |
2033 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
2034 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
2035 | min_objects = min(min_objects, max_objects); | |
2036 | ||
5e6d444e | 2037 | while (min_objects > 1) { |
c124f5b5 | 2038 | fraction = 16; |
5e6d444e CL |
2039 | while (fraction >= 4) { |
2040 | order = slab_order(size, min_objects, | |
2041 | slub_max_order, fraction); | |
2042 | if (order <= slub_max_order) | |
2043 | return order; | |
2044 | fraction /= 2; | |
2045 | } | |
5086c389 | 2046 | min_objects--; |
5e6d444e CL |
2047 | } |
2048 | ||
2049 | /* | |
2050 | * We were unable to place multiple objects in a slab. Now | |
2051 | * lets see if we can place a single object there. | |
2052 | */ | |
2053 | order = slab_order(size, 1, slub_max_order, 1); | |
2054 | if (order <= slub_max_order) | |
2055 | return order; | |
2056 | ||
2057 | /* | |
2058 | * Doh this slab cannot be placed using slub_max_order. | |
2059 | */ | |
2060 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 2061 | if (order < MAX_ORDER) |
5e6d444e CL |
2062 | return order; |
2063 | return -ENOSYS; | |
2064 | } | |
2065 | ||
81819f0f | 2066 | /* |
672bba3a | 2067 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2068 | */ |
2069 | static unsigned long calculate_alignment(unsigned long flags, | |
2070 | unsigned long align, unsigned long size) | |
2071 | { | |
2072 | /* | |
6446faa2 CL |
2073 | * If the user wants hardware cache aligned objects then follow that |
2074 | * suggestion if the object is sufficiently large. | |
81819f0f | 2075 | * |
6446faa2 CL |
2076 | * The hardware cache alignment cannot override the specified |
2077 | * alignment though. If that is greater then use it. | |
81819f0f | 2078 | */ |
b6210386 NP |
2079 | if (flags & SLAB_HWCACHE_ALIGN) { |
2080 | unsigned long ralign = cache_line_size(); | |
2081 | while (size <= ralign / 2) | |
2082 | ralign /= 2; | |
2083 | align = max(align, ralign); | |
2084 | } | |
81819f0f CL |
2085 | |
2086 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2087 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2088 | |
2089 | return ALIGN(align, sizeof(void *)); | |
2090 | } | |
2091 | ||
5595cffc PE |
2092 | static void |
2093 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2094 | { |
2095 | n->nr_partial = 0; | |
81819f0f CL |
2096 | spin_lock_init(&n->list_lock); |
2097 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2098 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2099 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2100 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2101 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2102 | #endif |
81819f0f CL |
2103 | } |
2104 | ||
55136592 | 2105 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2106 | { |
6c182dc0 CL |
2107 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2108 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2109 | |
6c182dc0 | 2110 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); |
4c93c355 | 2111 | |
6c182dc0 | 2112 | return s->cpu_slab != NULL; |
4c93c355 | 2113 | } |
4c93c355 | 2114 | |
81819f0f | 2115 | #ifdef CONFIG_NUMA |
51df1142 CL |
2116 | static struct kmem_cache *kmem_cache_node; |
2117 | ||
81819f0f CL |
2118 | /* |
2119 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2120 | * slab on the node for this slabcache. There are no concurrent accesses | |
2121 | * possible. | |
2122 | * | |
2123 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2124 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2125 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2126 | */ |
55136592 | 2127 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2128 | { |
2129 | struct page *page; | |
2130 | struct kmem_cache_node *n; | |
ba84c73c | 2131 | unsigned long flags; |
81819f0f | 2132 | |
51df1142 | 2133 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2134 | |
51df1142 | 2135 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2136 | |
2137 | BUG_ON(!page); | |
a2f92ee7 CL |
2138 | if (page_to_nid(page) != node) { |
2139 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2140 | "node %d\n", node); | |
2141 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2142 | "in order to be able to continue\n"); | |
2143 | } | |
2144 | ||
81819f0f CL |
2145 | n = page->freelist; |
2146 | BUG_ON(!n); | |
51df1142 | 2147 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2148 | page->inuse++; |
51df1142 | 2149 | kmem_cache_node->node[node] = n; |
8ab1372f | 2150 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
2151 | init_object(kmem_cache_node, n, 1); |
2152 | init_tracking(kmem_cache_node, n); | |
8ab1372f | 2153 | #endif |
51df1142 CL |
2154 | init_kmem_cache_node(n, kmem_cache_node); |
2155 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2156 | |
ba84c73c | 2157 | /* |
2158 | * lockdep requires consistent irq usage for each lock | |
2159 | * so even though there cannot be a race this early in | |
2160 | * the boot sequence, we still disable irqs. | |
2161 | */ | |
2162 | local_irq_save(flags); | |
7c2e132c | 2163 | add_partial(n, page, 0); |
ba84c73c | 2164 | local_irq_restore(flags); |
81819f0f CL |
2165 | } |
2166 | ||
2167 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2168 | { | |
2169 | int node; | |
2170 | ||
f64dc58c | 2171 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2172 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2173 | |
73367bd8 | 2174 | if (n) |
51df1142 CL |
2175 | kmem_cache_free(kmem_cache_node, n); |
2176 | ||
81819f0f CL |
2177 | s->node[node] = NULL; |
2178 | } | |
2179 | } | |
2180 | ||
55136592 | 2181 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2182 | { |
2183 | int node; | |
81819f0f | 2184 | |
f64dc58c | 2185 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2186 | struct kmem_cache_node *n; |
2187 | ||
73367bd8 | 2188 | if (slab_state == DOWN) { |
55136592 | 2189 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2190 | continue; |
2191 | } | |
51df1142 | 2192 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2193 | GFP_KERNEL, node); |
81819f0f | 2194 | |
73367bd8 AD |
2195 | if (!n) { |
2196 | free_kmem_cache_nodes(s); | |
2197 | return 0; | |
81819f0f | 2198 | } |
73367bd8 | 2199 | |
81819f0f | 2200 | s->node[node] = n; |
5595cffc | 2201 | init_kmem_cache_node(n, s); |
81819f0f CL |
2202 | } |
2203 | return 1; | |
2204 | } | |
2205 | #else | |
2206 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2207 | { | |
2208 | } | |
2209 | ||
55136592 | 2210 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f | 2211 | { |
5595cffc | 2212 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2213 | return 1; |
2214 | } | |
2215 | #endif | |
2216 | ||
c0bdb232 | 2217 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2218 | { |
2219 | if (min < MIN_PARTIAL) | |
2220 | min = MIN_PARTIAL; | |
2221 | else if (min > MAX_PARTIAL) | |
2222 | min = MAX_PARTIAL; | |
2223 | s->min_partial = min; | |
2224 | } | |
2225 | ||
81819f0f CL |
2226 | /* |
2227 | * calculate_sizes() determines the order and the distribution of data within | |
2228 | * a slab object. | |
2229 | */ | |
06b285dc | 2230 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2231 | { |
2232 | unsigned long flags = s->flags; | |
2233 | unsigned long size = s->objsize; | |
2234 | unsigned long align = s->align; | |
834f3d11 | 2235 | int order; |
81819f0f | 2236 | |
d8b42bf5 CL |
2237 | /* |
2238 | * Round up object size to the next word boundary. We can only | |
2239 | * place the free pointer at word boundaries and this determines | |
2240 | * the possible location of the free pointer. | |
2241 | */ | |
2242 | size = ALIGN(size, sizeof(void *)); | |
2243 | ||
2244 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2245 | /* |
2246 | * Determine if we can poison the object itself. If the user of | |
2247 | * the slab may touch the object after free or before allocation | |
2248 | * then we should never poison the object itself. | |
2249 | */ | |
2250 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2251 | !s->ctor) |
81819f0f CL |
2252 | s->flags |= __OBJECT_POISON; |
2253 | else | |
2254 | s->flags &= ~__OBJECT_POISON; | |
2255 | ||
81819f0f CL |
2256 | |
2257 | /* | |
672bba3a | 2258 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2259 | * end of the object and the free pointer. If not then add an |
672bba3a | 2260 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2261 | */ |
2262 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2263 | size += sizeof(void *); | |
41ecc55b | 2264 | #endif |
81819f0f CL |
2265 | |
2266 | /* | |
672bba3a CL |
2267 | * With that we have determined the number of bytes in actual use |
2268 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2269 | */ |
2270 | s->inuse = size; | |
2271 | ||
2272 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2273 | s->ctor)) { |
81819f0f CL |
2274 | /* |
2275 | * Relocate free pointer after the object if it is not | |
2276 | * permitted to overwrite the first word of the object on | |
2277 | * kmem_cache_free. | |
2278 | * | |
2279 | * This is the case if we do RCU, have a constructor or | |
2280 | * destructor or are poisoning the objects. | |
2281 | */ | |
2282 | s->offset = size; | |
2283 | size += sizeof(void *); | |
2284 | } | |
2285 | ||
c12b3c62 | 2286 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2287 | if (flags & SLAB_STORE_USER) |
2288 | /* | |
2289 | * Need to store information about allocs and frees after | |
2290 | * the object. | |
2291 | */ | |
2292 | size += 2 * sizeof(struct track); | |
2293 | ||
be7b3fbc | 2294 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2295 | /* |
2296 | * Add some empty padding so that we can catch | |
2297 | * overwrites from earlier objects rather than let | |
2298 | * tracking information or the free pointer be | |
0211a9c8 | 2299 | * corrupted if a user writes before the start |
81819f0f CL |
2300 | * of the object. |
2301 | */ | |
2302 | size += sizeof(void *); | |
41ecc55b | 2303 | #endif |
672bba3a | 2304 | |
81819f0f CL |
2305 | /* |
2306 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2307 | * user specified and the dynamic determination of cache line size |
2308 | * on bootup. | |
81819f0f CL |
2309 | */ |
2310 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2311 | s->align = align; |
81819f0f CL |
2312 | |
2313 | /* | |
2314 | * SLUB stores one object immediately after another beginning from | |
2315 | * offset 0. In order to align the objects we have to simply size | |
2316 | * each object to conform to the alignment. | |
2317 | */ | |
2318 | size = ALIGN(size, align); | |
2319 | s->size = size; | |
06b285dc CL |
2320 | if (forced_order >= 0) |
2321 | order = forced_order; | |
2322 | else | |
2323 | order = calculate_order(size); | |
81819f0f | 2324 | |
834f3d11 | 2325 | if (order < 0) |
81819f0f CL |
2326 | return 0; |
2327 | ||
b7a49f0d | 2328 | s->allocflags = 0; |
834f3d11 | 2329 | if (order) |
b7a49f0d CL |
2330 | s->allocflags |= __GFP_COMP; |
2331 | ||
2332 | if (s->flags & SLAB_CACHE_DMA) | |
2333 | s->allocflags |= SLUB_DMA; | |
2334 | ||
2335 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2336 | s->allocflags |= __GFP_RECLAIMABLE; | |
2337 | ||
81819f0f CL |
2338 | /* |
2339 | * Determine the number of objects per slab | |
2340 | */ | |
834f3d11 | 2341 | s->oo = oo_make(order, size); |
65c3376a | 2342 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2343 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2344 | s->max = s->oo; | |
81819f0f | 2345 | |
834f3d11 | 2346 | return !!oo_objects(s->oo); |
81819f0f CL |
2347 | |
2348 | } | |
2349 | ||
55136592 | 2350 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2351 | const char *name, size_t size, |
2352 | size_t align, unsigned long flags, | |
51cc5068 | 2353 | void (*ctor)(void *)) |
81819f0f CL |
2354 | { |
2355 | memset(s, 0, kmem_size); | |
2356 | s->name = name; | |
2357 | s->ctor = ctor; | |
81819f0f | 2358 | s->objsize = size; |
81819f0f | 2359 | s->align = align; |
ba0268a8 | 2360 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2361 | |
06b285dc | 2362 | if (!calculate_sizes(s, -1)) |
81819f0f | 2363 | goto error; |
3de47213 DR |
2364 | if (disable_higher_order_debug) { |
2365 | /* | |
2366 | * Disable debugging flags that store metadata if the min slab | |
2367 | * order increased. | |
2368 | */ | |
2369 | if (get_order(s->size) > get_order(s->objsize)) { | |
2370 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2371 | s->offset = 0; | |
2372 | if (!calculate_sizes(s, -1)) | |
2373 | goto error; | |
2374 | } | |
2375 | } | |
81819f0f | 2376 | |
3b89d7d8 DR |
2377 | /* |
2378 | * The larger the object size is, the more pages we want on the partial | |
2379 | * list to avoid pounding the page allocator excessively. | |
2380 | */ | |
c0bdb232 | 2381 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2382 | s->refcount = 1; |
2383 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2384 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2385 | #endif |
55136592 | 2386 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2387 | goto error; |
81819f0f | 2388 | |
55136592 | 2389 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2390 | return 1; |
ff12059e | 2391 | |
4c93c355 | 2392 | free_kmem_cache_nodes(s); |
81819f0f CL |
2393 | error: |
2394 | if (flags & SLAB_PANIC) | |
2395 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2396 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2397 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2398 | s->offset, flags); |
2399 | return 0; | |
2400 | } | |
81819f0f CL |
2401 | |
2402 | /* | |
2403 | * Check if a given pointer is valid | |
2404 | */ | |
2405 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2406 | { | |
06428780 | 2407 | struct page *page; |
81819f0f | 2408 | |
d3e06e2b PE |
2409 | if (!kern_ptr_validate(object, s->size)) |
2410 | return 0; | |
2411 | ||
81819f0f CL |
2412 | page = get_object_page(object); |
2413 | ||
2414 | if (!page || s != page->slab) | |
2415 | /* No slab or wrong slab */ | |
2416 | return 0; | |
2417 | ||
abcd08a6 | 2418 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2419 | return 0; |
2420 | ||
2421 | /* | |
2422 | * We could also check if the object is on the slabs freelist. | |
2423 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2424 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2425 | * to a certain slab. |
2426 | */ | |
2427 | return 1; | |
2428 | } | |
2429 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2430 | ||
2431 | /* | |
2432 | * Determine the size of a slab object | |
2433 | */ | |
2434 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2435 | { | |
2436 | return s->objsize; | |
2437 | } | |
2438 | EXPORT_SYMBOL(kmem_cache_size); | |
2439 | ||
2440 | const char *kmem_cache_name(struct kmem_cache *s) | |
2441 | { | |
2442 | return s->name; | |
2443 | } | |
2444 | EXPORT_SYMBOL(kmem_cache_name); | |
2445 | ||
33b12c38 CL |
2446 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2447 | const char *text) | |
2448 | { | |
2449 | #ifdef CONFIG_SLUB_DEBUG | |
2450 | void *addr = page_address(page); | |
2451 | void *p; | |
bbd7d57b ED |
2452 | long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long), |
2453 | GFP_ATOMIC); | |
33b12c38 | 2454 | |
bbd7d57b ED |
2455 | if (!map) |
2456 | return; | |
33b12c38 CL |
2457 | slab_err(s, page, "%s", text); |
2458 | slab_lock(page); | |
2459 | for_each_free_object(p, s, page->freelist) | |
2460 | set_bit(slab_index(p, s, addr), map); | |
2461 | ||
2462 | for_each_object(p, s, addr, page->objects) { | |
2463 | ||
2464 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2465 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2466 | p, p - addr); | |
2467 | print_tracking(s, p); | |
2468 | } | |
2469 | } | |
2470 | slab_unlock(page); | |
bbd7d57b | 2471 | kfree(map); |
33b12c38 CL |
2472 | #endif |
2473 | } | |
2474 | ||
81819f0f | 2475 | /* |
599870b1 | 2476 | * Attempt to free all partial slabs on a node. |
81819f0f | 2477 | */ |
599870b1 | 2478 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2479 | { |
81819f0f CL |
2480 | unsigned long flags; |
2481 | struct page *page, *h; | |
2482 | ||
2483 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2484 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2485 | if (!page->inuse) { |
2486 | list_del(&page->lru); | |
2487 | discard_slab(s, page); | |
599870b1 | 2488 | n->nr_partial--; |
33b12c38 CL |
2489 | } else { |
2490 | list_slab_objects(s, page, | |
2491 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2492 | } |
33b12c38 | 2493 | } |
81819f0f | 2494 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2495 | } |
2496 | ||
2497 | /* | |
672bba3a | 2498 | * Release all resources used by a slab cache. |
81819f0f | 2499 | */ |
0c710013 | 2500 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2501 | { |
2502 | int node; | |
2503 | ||
2504 | flush_all(s); | |
9dfc6e68 | 2505 | free_percpu(s->cpu_slab); |
81819f0f | 2506 | /* Attempt to free all objects */ |
f64dc58c | 2507 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2508 | struct kmem_cache_node *n = get_node(s, node); |
2509 | ||
599870b1 CL |
2510 | free_partial(s, n); |
2511 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2512 | return 1; |
2513 | } | |
2514 | free_kmem_cache_nodes(s); | |
2515 | return 0; | |
2516 | } | |
2517 | ||
2518 | /* | |
2519 | * Close a cache and release the kmem_cache structure | |
2520 | * (must be used for caches created using kmem_cache_create) | |
2521 | */ | |
2522 | void kmem_cache_destroy(struct kmem_cache *s) | |
2523 | { | |
2524 | down_write(&slub_lock); | |
2525 | s->refcount--; | |
2526 | if (!s->refcount) { | |
2527 | list_del(&s->list); | |
d629d819 PE |
2528 | if (kmem_cache_close(s)) { |
2529 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2530 | "still has objects.\n", s->name, __func__); | |
2531 | dump_stack(); | |
2532 | } | |
d76b1590 ED |
2533 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2534 | rcu_barrier(); | |
81819f0f | 2535 | sysfs_slab_remove(s); |
2bce6485 CL |
2536 | } |
2537 | up_write(&slub_lock); | |
81819f0f CL |
2538 | } |
2539 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2540 | ||
2541 | /******************************************************************** | |
2542 | * Kmalloc subsystem | |
2543 | *******************************************************************/ | |
2544 | ||
51df1142 | 2545 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2546 | EXPORT_SYMBOL(kmalloc_caches); |
2547 | ||
51df1142 CL |
2548 | static struct kmem_cache *kmem_cache; |
2549 | ||
55136592 | 2550 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2551 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2552 | #endif |
2553 | ||
81819f0f CL |
2554 | static int __init setup_slub_min_order(char *str) |
2555 | { | |
06428780 | 2556 | get_option(&str, &slub_min_order); |
81819f0f CL |
2557 | |
2558 | return 1; | |
2559 | } | |
2560 | ||
2561 | __setup("slub_min_order=", setup_slub_min_order); | |
2562 | ||
2563 | static int __init setup_slub_max_order(char *str) | |
2564 | { | |
06428780 | 2565 | get_option(&str, &slub_max_order); |
818cf590 | 2566 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2567 | |
2568 | return 1; | |
2569 | } | |
2570 | ||
2571 | __setup("slub_max_order=", setup_slub_max_order); | |
2572 | ||
2573 | static int __init setup_slub_min_objects(char *str) | |
2574 | { | |
06428780 | 2575 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2576 | |
2577 | return 1; | |
2578 | } | |
2579 | ||
2580 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2581 | ||
2582 | static int __init setup_slub_nomerge(char *str) | |
2583 | { | |
2584 | slub_nomerge = 1; | |
2585 | return 1; | |
2586 | } | |
2587 | ||
2588 | __setup("slub_nomerge", setup_slub_nomerge); | |
2589 | ||
51df1142 CL |
2590 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2591 | int size, unsigned int flags) | |
81819f0f | 2592 | { |
51df1142 CL |
2593 | struct kmem_cache *s; |
2594 | ||
2595 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2596 | ||
83b519e8 PE |
2597 | /* |
2598 | * This function is called with IRQs disabled during early-boot on | |
2599 | * single CPU so there's no need to take slub_lock here. | |
2600 | */ | |
55136592 | 2601 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2602 | flags, NULL)) |
81819f0f CL |
2603 | goto panic; |
2604 | ||
2605 | list_add(&s->list, &slab_caches); | |
51df1142 | 2606 | return s; |
81819f0f CL |
2607 | |
2608 | panic: | |
2609 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2610 | return NULL; |
81819f0f CL |
2611 | } |
2612 | ||
f1b26339 CL |
2613 | /* |
2614 | * Conversion table for small slabs sizes / 8 to the index in the | |
2615 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2616 | * of two cache sizes there. The size of larger slabs can be determined using | |
2617 | * fls. | |
2618 | */ | |
2619 | static s8 size_index[24] = { | |
2620 | 3, /* 8 */ | |
2621 | 4, /* 16 */ | |
2622 | 5, /* 24 */ | |
2623 | 5, /* 32 */ | |
2624 | 6, /* 40 */ | |
2625 | 6, /* 48 */ | |
2626 | 6, /* 56 */ | |
2627 | 6, /* 64 */ | |
2628 | 1, /* 72 */ | |
2629 | 1, /* 80 */ | |
2630 | 1, /* 88 */ | |
2631 | 1, /* 96 */ | |
2632 | 7, /* 104 */ | |
2633 | 7, /* 112 */ | |
2634 | 7, /* 120 */ | |
2635 | 7, /* 128 */ | |
2636 | 2, /* 136 */ | |
2637 | 2, /* 144 */ | |
2638 | 2, /* 152 */ | |
2639 | 2, /* 160 */ | |
2640 | 2, /* 168 */ | |
2641 | 2, /* 176 */ | |
2642 | 2, /* 184 */ | |
2643 | 2 /* 192 */ | |
2644 | }; | |
2645 | ||
acdfcd04 AK |
2646 | static inline int size_index_elem(size_t bytes) |
2647 | { | |
2648 | return (bytes - 1) / 8; | |
2649 | } | |
2650 | ||
81819f0f CL |
2651 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2652 | { | |
f1b26339 | 2653 | int index; |
81819f0f | 2654 | |
f1b26339 CL |
2655 | if (size <= 192) { |
2656 | if (!size) | |
2657 | return ZERO_SIZE_PTR; | |
81819f0f | 2658 | |
acdfcd04 | 2659 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2660 | } else |
f1b26339 | 2661 | index = fls(size - 1); |
81819f0f CL |
2662 | |
2663 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2664 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2665 | return kmalloc_dma_caches[index]; |
f1b26339 | 2666 | |
81819f0f | 2667 | #endif |
51df1142 | 2668 | return kmalloc_caches[index]; |
81819f0f CL |
2669 | } |
2670 | ||
2671 | void *__kmalloc(size_t size, gfp_t flags) | |
2672 | { | |
aadb4bc4 | 2673 | struct kmem_cache *s; |
5b882be4 | 2674 | void *ret; |
81819f0f | 2675 | |
ffadd4d0 | 2676 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2677 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2678 | |
2679 | s = get_slab(size, flags); | |
2680 | ||
2681 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2682 | return s; |
2683 | ||
2154a336 | 2684 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2685 | |
ca2b84cb | 2686 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2687 | |
2688 | return ret; | |
81819f0f CL |
2689 | } |
2690 | EXPORT_SYMBOL(__kmalloc); | |
2691 | ||
f619cfe1 CL |
2692 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2693 | { | |
b1eeab67 | 2694 | struct page *page; |
e4f7c0b4 | 2695 | void *ptr = NULL; |
f619cfe1 | 2696 | |
b1eeab67 VN |
2697 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2698 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2699 | if (page) |
e4f7c0b4 CM |
2700 | ptr = page_address(page); |
2701 | ||
2702 | kmemleak_alloc(ptr, size, 1, flags); | |
2703 | return ptr; | |
f619cfe1 CL |
2704 | } |
2705 | ||
81819f0f CL |
2706 | #ifdef CONFIG_NUMA |
2707 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2708 | { | |
aadb4bc4 | 2709 | struct kmem_cache *s; |
5b882be4 | 2710 | void *ret; |
81819f0f | 2711 | |
057685cf | 2712 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2713 | ret = kmalloc_large_node(size, flags, node); |
2714 | ||
ca2b84cb EGM |
2715 | trace_kmalloc_node(_RET_IP_, ret, |
2716 | size, PAGE_SIZE << get_order(size), | |
2717 | flags, node); | |
5b882be4 EGM |
2718 | |
2719 | return ret; | |
2720 | } | |
aadb4bc4 CL |
2721 | |
2722 | s = get_slab(size, flags); | |
2723 | ||
2724 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2725 | return s; |
2726 | ||
5b882be4 EGM |
2727 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2728 | ||
ca2b84cb | 2729 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2730 | |
2731 | return ret; | |
81819f0f CL |
2732 | } |
2733 | EXPORT_SYMBOL(__kmalloc_node); | |
2734 | #endif | |
2735 | ||
2736 | size_t ksize(const void *object) | |
2737 | { | |
272c1d21 | 2738 | struct page *page; |
81819f0f CL |
2739 | struct kmem_cache *s; |
2740 | ||
ef8b4520 | 2741 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2742 | return 0; |
2743 | ||
294a80a8 | 2744 | page = virt_to_head_page(object); |
294a80a8 | 2745 | |
76994412 PE |
2746 | if (unlikely(!PageSlab(page))) { |
2747 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2748 | return PAGE_SIZE << compound_order(page); |
76994412 | 2749 | } |
81819f0f | 2750 | s = page->slab; |
81819f0f | 2751 | |
ae20bfda | 2752 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2753 | /* |
2754 | * Debugging requires use of the padding between object | |
2755 | * and whatever may come after it. | |
2756 | */ | |
2757 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2758 | return s->objsize; | |
2759 | ||
ae20bfda | 2760 | #endif |
81819f0f CL |
2761 | /* |
2762 | * If we have the need to store the freelist pointer | |
2763 | * back there or track user information then we can | |
2764 | * only use the space before that information. | |
2765 | */ | |
2766 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2767 | return s->inuse; | |
81819f0f CL |
2768 | /* |
2769 | * Else we can use all the padding etc for the allocation | |
2770 | */ | |
2771 | return s->size; | |
2772 | } | |
b1aabecd | 2773 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2774 | |
2775 | void kfree(const void *x) | |
2776 | { | |
81819f0f | 2777 | struct page *page; |
5bb983b0 | 2778 | void *object = (void *)x; |
81819f0f | 2779 | |
2121db74 PE |
2780 | trace_kfree(_RET_IP_, x); |
2781 | ||
2408c550 | 2782 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2783 | return; |
2784 | ||
b49af68f | 2785 | page = virt_to_head_page(x); |
aadb4bc4 | 2786 | if (unlikely(!PageSlab(page))) { |
0937502a | 2787 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2788 | kmemleak_free(x); |
aadb4bc4 CL |
2789 | put_page(page); |
2790 | return; | |
2791 | } | |
ce71e27c | 2792 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2793 | } |
2794 | EXPORT_SYMBOL(kfree); | |
2795 | ||
2086d26a | 2796 | /* |
672bba3a CL |
2797 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2798 | * the remaining slabs by the number of items in use. The slabs with the | |
2799 | * most items in use come first. New allocations will then fill those up | |
2800 | * and thus they can be removed from the partial lists. | |
2801 | * | |
2802 | * The slabs with the least items are placed last. This results in them | |
2803 | * being allocated from last increasing the chance that the last objects | |
2804 | * are freed in them. | |
2086d26a CL |
2805 | */ |
2806 | int kmem_cache_shrink(struct kmem_cache *s) | |
2807 | { | |
2808 | int node; | |
2809 | int i; | |
2810 | struct kmem_cache_node *n; | |
2811 | struct page *page; | |
2812 | struct page *t; | |
205ab99d | 2813 | int objects = oo_objects(s->max); |
2086d26a | 2814 | struct list_head *slabs_by_inuse = |
834f3d11 | 2815 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2816 | unsigned long flags; |
2817 | ||
2818 | if (!slabs_by_inuse) | |
2819 | return -ENOMEM; | |
2820 | ||
2821 | flush_all(s); | |
f64dc58c | 2822 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2823 | n = get_node(s, node); |
2824 | ||
2825 | if (!n->nr_partial) | |
2826 | continue; | |
2827 | ||
834f3d11 | 2828 | for (i = 0; i < objects; i++) |
2086d26a CL |
2829 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2830 | ||
2831 | spin_lock_irqsave(&n->list_lock, flags); | |
2832 | ||
2833 | /* | |
672bba3a | 2834 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2835 | * |
672bba3a CL |
2836 | * Note that concurrent frees may occur while we hold the |
2837 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2838 | */ |
2839 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2840 | if (!page->inuse && slab_trylock(page)) { | |
2841 | /* | |
2842 | * Must hold slab lock here because slab_free | |
2843 | * may have freed the last object and be | |
2844 | * waiting to release the slab. | |
2845 | */ | |
2846 | list_del(&page->lru); | |
2847 | n->nr_partial--; | |
2848 | slab_unlock(page); | |
2849 | discard_slab(s, page); | |
2850 | } else { | |
fcda3d89 CL |
2851 | list_move(&page->lru, |
2852 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2853 | } |
2854 | } | |
2855 | ||
2086d26a | 2856 | /* |
672bba3a CL |
2857 | * Rebuild the partial list with the slabs filled up most |
2858 | * first and the least used slabs at the end. | |
2086d26a | 2859 | */ |
834f3d11 | 2860 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2861 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2862 | ||
2086d26a CL |
2863 | spin_unlock_irqrestore(&n->list_lock, flags); |
2864 | } | |
2865 | ||
2866 | kfree(slabs_by_inuse); | |
2867 | return 0; | |
2868 | } | |
2869 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2870 | ||
b9049e23 YG |
2871 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2872 | static int slab_mem_going_offline_callback(void *arg) | |
2873 | { | |
2874 | struct kmem_cache *s; | |
2875 | ||
2876 | down_read(&slub_lock); | |
2877 | list_for_each_entry(s, &slab_caches, list) | |
2878 | kmem_cache_shrink(s); | |
2879 | up_read(&slub_lock); | |
2880 | ||
2881 | return 0; | |
2882 | } | |
2883 | ||
2884 | static void slab_mem_offline_callback(void *arg) | |
2885 | { | |
2886 | struct kmem_cache_node *n; | |
2887 | struct kmem_cache *s; | |
2888 | struct memory_notify *marg = arg; | |
2889 | int offline_node; | |
2890 | ||
2891 | offline_node = marg->status_change_nid; | |
2892 | ||
2893 | /* | |
2894 | * If the node still has available memory. we need kmem_cache_node | |
2895 | * for it yet. | |
2896 | */ | |
2897 | if (offline_node < 0) | |
2898 | return; | |
2899 | ||
2900 | down_read(&slub_lock); | |
2901 | list_for_each_entry(s, &slab_caches, list) { | |
2902 | n = get_node(s, offline_node); | |
2903 | if (n) { | |
2904 | /* | |
2905 | * if n->nr_slabs > 0, slabs still exist on the node | |
2906 | * that is going down. We were unable to free them, | |
c9404c9c | 2907 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
2908 | * callback. So, we must fail. |
2909 | */ | |
0f389ec6 | 2910 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2911 | |
2912 | s->node[offline_node] = NULL; | |
8de66a0c | 2913 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
2914 | } |
2915 | } | |
2916 | up_read(&slub_lock); | |
2917 | } | |
2918 | ||
2919 | static int slab_mem_going_online_callback(void *arg) | |
2920 | { | |
2921 | struct kmem_cache_node *n; | |
2922 | struct kmem_cache *s; | |
2923 | struct memory_notify *marg = arg; | |
2924 | int nid = marg->status_change_nid; | |
2925 | int ret = 0; | |
2926 | ||
2927 | /* | |
2928 | * If the node's memory is already available, then kmem_cache_node is | |
2929 | * already created. Nothing to do. | |
2930 | */ | |
2931 | if (nid < 0) | |
2932 | return 0; | |
2933 | ||
2934 | /* | |
0121c619 | 2935 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2936 | * allocate a kmem_cache_node structure in order to bring the node |
2937 | * online. | |
2938 | */ | |
2939 | down_read(&slub_lock); | |
2940 | list_for_each_entry(s, &slab_caches, list) { | |
2941 | /* | |
2942 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2943 | * since memory is not yet available from the node that | |
2944 | * is brought up. | |
2945 | */ | |
8de66a0c | 2946 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
2947 | if (!n) { |
2948 | ret = -ENOMEM; | |
2949 | goto out; | |
2950 | } | |
5595cffc | 2951 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2952 | s->node[nid] = n; |
2953 | } | |
2954 | out: | |
2955 | up_read(&slub_lock); | |
2956 | return ret; | |
2957 | } | |
2958 | ||
2959 | static int slab_memory_callback(struct notifier_block *self, | |
2960 | unsigned long action, void *arg) | |
2961 | { | |
2962 | int ret = 0; | |
2963 | ||
2964 | switch (action) { | |
2965 | case MEM_GOING_ONLINE: | |
2966 | ret = slab_mem_going_online_callback(arg); | |
2967 | break; | |
2968 | case MEM_GOING_OFFLINE: | |
2969 | ret = slab_mem_going_offline_callback(arg); | |
2970 | break; | |
2971 | case MEM_OFFLINE: | |
2972 | case MEM_CANCEL_ONLINE: | |
2973 | slab_mem_offline_callback(arg); | |
2974 | break; | |
2975 | case MEM_ONLINE: | |
2976 | case MEM_CANCEL_OFFLINE: | |
2977 | break; | |
2978 | } | |
dc19f9db KH |
2979 | if (ret) |
2980 | ret = notifier_from_errno(ret); | |
2981 | else | |
2982 | ret = NOTIFY_OK; | |
b9049e23 YG |
2983 | return ret; |
2984 | } | |
2985 | ||
2986 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
2987 | ||
81819f0f CL |
2988 | /******************************************************************** |
2989 | * Basic setup of slabs | |
2990 | *******************************************************************/ | |
2991 | ||
51df1142 CL |
2992 | /* |
2993 | * Used for early kmem_cache structures that were allocated using | |
2994 | * the page allocator | |
2995 | */ | |
2996 | ||
2997 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
2998 | { | |
2999 | int node; | |
3000 | ||
3001 | list_add(&s->list, &slab_caches); | |
3002 | s->refcount = -1; | |
3003 | ||
3004 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3005 | struct kmem_cache_node *n = get_node(s, node); | |
3006 | struct page *p; | |
3007 | ||
3008 | if (n) { | |
3009 | list_for_each_entry(p, &n->partial, lru) | |
3010 | p->slab = s; | |
3011 | ||
3012 | #ifdef CONFIG_SLAB_DEBUG | |
3013 | list_for_each_entry(p, &n->full, lru) | |
3014 | p->slab = s; | |
3015 | #endif | |
3016 | } | |
3017 | } | |
3018 | } | |
3019 | ||
81819f0f CL |
3020 | void __init kmem_cache_init(void) |
3021 | { | |
3022 | int i; | |
4b356be0 | 3023 | int caches = 0; |
51df1142 CL |
3024 | struct kmem_cache *temp_kmem_cache; |
3025 | int order; | |
81819f0f CL |
3026 | |
3027 | #ifdef CONFIG_NUMA | |
51df1142 CL |
3028 | struct kmem_cache *temp_kmem_cache_node; |
3029 | unsigned long kmalloc_size; | |
3030 | ||
3031 | kmem_size = offsetof(struct kmem_cache, node) + | |
3032 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3033 | ||
3034 | /* Allocate two kmem_caches from the page allocator */ | |
3035 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3036 | order = get_order(2 * kmalloc_size); | |
3037 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3038 | ||
81819f0f CL |
3039 | /* |
3040 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3041 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3042 | * kmem_cache_open for slab_state == DOWN. |
3043 | */ | |
51df1142 CL |
3044 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3045 | ||
3046 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3047 | sizeof(struct kmem_cache_node), | |
3048 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3049 | |
0c40ba4f | 3050 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
51df1142 CL |
3051 | #else |
3052 | /* Allocate a single kmem_cache from the page allocator */ | |
3053 | kmem_size = sizeof(struct kmem_cache); | |
3054 | order = get_order(kmem_size); | |
3055 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
81819f0f CL |
3056 | #endif |
3057 | ||
3058 | /* Able to allocate the per node structures */ | |
3059 | slab_state = PARTIAL; | |
3060 | ||
51df1142 CL |
3061 | temp_kmem_cache = kmem_cache; |
3062 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3063 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3064 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3065 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3066 | |
51df1142 CL |
3067 | #ifdef CONFIG_NUMA |
3068 | /* | |
3069 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3070 | * kmem_cache_node is separately allocated so no need to | |
3071 | * update any list pointers. | |
3072 | */ | |
3073 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3074 | |
51df1142 CL |
3075 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3076 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3077 | ||
3078 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3079 | ||
3080 | caches++; | |
3081 | #else | |
3082 | /* | |
3083 | * kmem_cache has kmem_cache_node embedded and we moved it! | |
3084 | * Update the list heads | |
3085 | */ | |
3086 | INIT_LIST_HEAD(&kmem_cache->local_node.partial); | |
3087 | list_splice(&temp_kmem_cache->local_node.partial, &kmem_cache->local_node.partial); | |
3088 | #ifdef CONFIG_SLUB_DEBUG | |
3089 | INIT_LIST_HEAD(&kmem_cache->local_node.full); | |
3090 | list_splice(&temp_kmem_cache->local_node.full, &kmem_cache->local_node.full); | |
3091 | #endif | |
3092 | #endif | |
3093 | kmem_cache_bootstrap_fixup(kmem_cache); | |
3094 | caches++; | |
3095 | /* Free temporary boot structure */ | |
3096 | free_pages((unsigned long)temp_kmem_cache, order); | |
3097 | ||
3098 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3099 | |
3100 | /* | |
3101 | * Patch up the size_index table if we have strange large alignment | |
3102 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3103 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3104 | * |
3105 | * Largest permitted alignment is 256 bytes due to the way we | |
3106 | * handle the index determination for the smaller caches. | |
3107 | * | |
3108 | * Make sure that nothing crazy happens if someone starts tinkering | |
3109 | * around with ARCH_KMALLOC_MINALIGN | |
3110 | */ | |
3111 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3112 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3113 | ||
acdfcd04 AK |
3114 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3115 | int elem = size_index_elem(i); | |
3116 | if (elem >= ARRAY_SIZE(size_index)) | |
3117 | break; | |
3118 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3119 | } | |
f1b26339 | 3120 | |
acdfcd04 AK |
3121 | if (KMALLOC_MIN_SIZE == 64) { |
3122 | /* | |
3123 | * The 96 byte size cache is not used if the alignment | |
3124 | * is 64 byte. | |
3125 | */ | |
3126 | for (i = 64 + 8; i <= 96; i += 8) | |
3127 | size_index[size_index_elem(i)] = 7; | |
3128 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3129 | /* |
3130 | * The 192 byte sized cache is not used if the alignment | |
3131 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3132 | * instead. | |
3133 | */ | |
3134 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3135 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3136 | } |
3137 | ||
51df1142 CL |
3138 | /* Caches that are not of the two-to-the-power-of size */ |
3139 | if (KMALLOC_MIN_SIZE <= 32) { | |
3140 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3141 | caches++; | |
3142 | } | |
3143 | ||
3144 | if (KMALLOC_MIN_SIZE <= 64) { | |
3145 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3146 | caches++; | |
3147 | } | |
3148 | ||
3149 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3150 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3151 | caches++; | |
3152 | } | |
3153 | ||
81819f0f CL |
3154 | slab_state = UP; |
3155 | ||
3156 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3157 | if (KMALLOC_MIN_SIZE <= 32) { |
3158 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3159 | BUG_ON(!kmalloc_caches[1]->name); | |
3160 | } | |
3161 | ||
3162 | if (KMALLOC_MIN_SIZE <= 64) { | |
3163 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3164 | BUG_ON(!kmalloc_caches[2]->name); | |
3165 | } | |
3166 | ||
d7278bd7 CL |
3167 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3168 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3169 | ||
3170 | BUG_ON(!s); | |
51df1142 | 3171 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3172 | } |
81819f0f CL |
3173 | |
3174 | #ifdef CONFIG_SMP | |
3175 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3176 | #endif |
81819f0f | 3177 | |
55136592 | 3178 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3179 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3180 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3181 | |
51df1142 | 3182 | if (s && s->size) { |
55136592 CL |
3183 | char *name = kasprintf(GFP_NOWAIT, |
3184 | "dma-kmalloc-%d", s->objsize); | |
3185 | ||
3186 | BUG_ON(!name); | |
51df1142 CL |
3187 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3188 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3189 | } |
3190 | } | |
3191 | #endif | |
3adbefee IM |
3192 | printk(KERN_INFO |
3193 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3194 | " CPUs=%d, Nodes=%d\n", |
3195 | caches, cache_line_size(), | |
81819f0f CL |
3196 | slub_min_order, slub_max_order, slub_min_objects, |
3197 | nr_cpu_ids, nr_node_ids); | |
3198 | } | |
3199 | ||
7e85ee0c PE |
3200 | void __init kmem_cache_init_late(void) |
3201 | { | |
7e85ee0c PE |
3202 | } |
3203 | ||
81819f0f CL |
3204 | /* |
3205 | * Find a mergeable slab cache | |
3206 | */ | |
3207 | static int slab_unmergeable(struct kmem_cache *s) | |
3208 | { | |
3209 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3210 | return 1; | |
3211 | ||
c59def9f | 3212 | if (s->ctor) |
81819f0f CL |
3213 | return 1; |
3214 | ||
8ffa6875 CL |
3215 | /* |
3216 | * We may have set a slab to be unmergeable during bootstrap. | |
3217 | */ | |
3218 | if (s->refcount < 0) | |
3219 | return 1; | |
3220 | ||
81819f0f CL |
3221 | return 0; |
3222 | } | |
3223 | ||
3224 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3225 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3226 | void (*ctor)(void *)) |
81819f0f | 3227 | { |
5b95a4ac | 3228 | struct kmem_cache *s; |
81819f0f CL |
3229 | |
3230 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3231 | return NULL; | |
3232 | ||
c59def9f | 3233 | if (ctor) |
81819f0f CL |
3234 | return NULL; |
3235 | ||
3236 | size = ALIGN(size, sizeof(void *)); | |
3237 | align = calculate_alignment(flags, align, size); | |
3238 | size = ALIGN(size, align); | |
ba0268a8 | 3239 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3240 | |
5b95a4ac | 3241 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3242 | if (slab_unmergeable(s)) |
3243 | continue; | |
3244 | ||
3245 | if (size > s->size) | |
3246 | continue; | |
3247 | ||
ba0268a8 | 3248 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3249 | continue; |
3250 | /* | |
3251 | * Check if alignment is compatible. | |
3252 | * Courtesy of Adrian Drzewiecki | |
3253 | */ | |
06428780 | 3254 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3255 | continue; |
3256 | ||
3257 | if (s->size - size >= sizeof(void *)) | |
3258 | continue; | |
3259 | ||
3260 | return s; | |
3261 | } | |
3262 | return NULL; | |
3263 | } | |
3264 | ||
3265 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3266 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3267 | { |
3268 | struct kmem_cache *s; | |
84c1cf62 | 3269 | char *n; |
81819f0f | 3270 | |
fe1ff49d BH |
3271 | if (WARN_ON(!name)) |
3272 | return NULL; | |
3273 | ||
81819f0f | 3274 | down_write(&slub_lock); |
ba0268a8 | 3275 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3276 | if (s) { |
3277 | s->refcount++; | |
3278 | /* | |
3279 | * Adjust the object sizes so that we clear | |
3280 | * the complete object on kzalloc. | |
3281 | */ | |
3282 | s->objsize = max(s->objsize, (int)size); | |
3283 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3284 | |
7b8f3b66 | 3285 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3286 | s->refcount--; |
81819f0f | 3287 | goto err; |
7b8f3b66 | 3288 | } |
2bce6485 | 3289 | up_write(&slub_lock); |
a0e1d1be CL |
3290 | return s; |
3291 | } | |
6446faa2 | 3292 | |
84c1cf62 PE |
3293 | n = kstrdup(name, GFP_KERNEL); |
3294 | if (!n) | |
3295 | goto err; | |
3296 | ||
a0e1d1be CL |
3297 | s = kmalloc(kmem_size, GFP_KERNEL); |
3298 | if (s) { | |
84c1cf62 | 3299 | if (kmem_cache_open(s, n, |
c59def9f | 3300 | size, align, flags, ctor)) { |
81819f0f | 3301 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3302 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3303 | list_del(&s->list); |
84c1cf62 | 3304 | kfree(n); |
7b8f3b66 | 3305 | kfree(s); |
a0e1d1be | 3306 | goto err; |
7b8f3b66 | 3307 | } |
2bce6485 | 3308 | up_write(&slub_lock); |
a0e1d1be CL |
3309 | return s; |
3310 | } | |
84c1cf62 | 3311 | kfree(n); |
a0e1d1be | 3312 | kfree(s); |
81819f0f CL |
3313 | } |
3314 | up_write(&slub_lock); | |
81819f0f CL |
3315 | |
3316 | err: | |
81819f0f CL |
3317 | if (flags & SLAB_PANIC) |
3318 | panic("Cannot create slabcache %s\n", name); | |
3319 | else | |
3320 | s = NULL; | |
3321 | return s; | |
3322 | } | |
3323 | EXPORT_SYMBOL(kmem_cache_create); | |
3324 | ||
81819f0f | 3325 | #ifdef CONFIG_SMP |
81819f0f | 3326 | /* |
672bba3a CL |
3327 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3328 | * necessary. | |
81819f0f CL |
3329 | */ |
3330 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3331 | unsigned long action, void *hcpu) | |
3332 | { | |
3333 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3334 | struct kmem_cache *s; |
3335 | unsigned long flags; | |
81819f0f CL |
3336 | |
3337 | switch (action) { | |
3338 | case CPU_UP_CANCELED: | |
8bb78442 | 3339 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3340 | case CPU_DEAD: |
8bb78442 | 3341 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3342 | down_read(&slub_lock); |
3343 | list_for_each_entry(s, &slab_caches, list) { | |
3344 | local_irq_save(flags); | |
3345 | __flush_cpu_slab(s, cpu); | |
3346 | local_irq_restore(flags); | |
3347 | } | |
3348 | up_read(&slub_lock); | |
81819f0f CL |
3349 | break; |
3350 | default: | |
3351 | break; | |
3352 | } | |
3353 | return NOTIFY_OK; | |
3354 | } | |
3355 | ||
06428780 | 3356 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3357 | .notifier_call = slab_cpuup_callback |
06428780 | 3358 | }; |
81819f0f CL |
3359 | |
3360 | #endif | |
3361 | ||
ce71e27c | 3362 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3363 | { |
aadb4bc4 | 3364 | struct kmem_cache *s; |
94b528d0 | 3365 | void *ret; |
aadb4bc4 | 3366 | |
ffadd4d0 | 3367 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3368 | return kmalloc_large(size, gfpflags); |
3369 | ||
aadb4bc4 | 3370 | s = get_slab(size, gfpflags); |
81819f0f | 3371 | |
2408c550 | 3372 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3373 | return s; |
81819f0f | 3374 | |
2154a336 | 3375 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3376 | |
3377 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3378 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3379 | |
3380 | return ret; | |
81819f0f CL |
3381 | } |
3382 | ||
3383 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3384 | int node, unsigned long caller) |
81819f0f | 3385 | { |
aadb4bc4 | 3386 | struct kmem_cache *s; |
94b528d0 | 3387 | void *ret; |
aadb4bc4 | 3388 | |
d3e14aa3 XF |
3389 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3390 | ret = kmalloc_large_node(size, gfpflags, node); | |
3391 | ||
3392 | trace_kmalloc_node(caller, ret, | |
3393 | size, PAGE_SIZE << get_order(size), | |
3394 | gfpflags, node); | |
3395 | ||
3396 | return ret; | |
3397 | } | |
eada35ef | 3398 | |
aadb4bc4 | 3399 | s = get_slab(size, gfpflags); |
81819f0f | 3400 | |
2408c550 | 3401 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3402 | return s; |
81819f0f | 3403 | |
94b528d0 EGM |
3404 | ret = slab_alloc(s, gfpflags, node, caller); |
3405 | ||
3406 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3407 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3408 | |
3409 | return ret; | |
81819f0f CL |
3410 | } |
3411 | ||
f6acb635 | 3412 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3413 | static int count_inuse(struct page *page) |
3414 | { | |
3415 | return page->inuse; | |
3416 | } | |
3417 | ||
3418 | static int count_total(struct page *page) | |
3419 | { | |
3420 | return page->objects; | |
3421 | } | |
3422 | ||
434e245d CL |
3423 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3424 | unsigned long *map) | |
53e15af0 CL |
3425 | { |
3426 | void *p; | |
a973e9dd | 3427 | void *addr = page_address(page); |
53e15af0 CL |
3428 | |
3429 | if (!check_slab(s, page) || | |
3430 | !on_freelist(s, page, NULL)) | |
3431 | return 0; | |
3432 | ||
3433 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3434 | bitmap_zero(map, page->objects); |
53e15af0 | 3435 | |
7656c72b CL |
3436 | for_each_free_object(p, s, page->freelist) { |
3437 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3438 | if (!check_object(s, page, p, 0)) |
3439 | return 0; | |
3440 | } | |
3441 | ||
224a88be | 3442 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3443 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3444 | if (!check_object(s, page, p, 1)) |
3445 | return 0; | |
3446 | return 1; | |
3447 | } | |
3448 | ||
434e245d CL |
3449 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3450 | unsigned long *map) | |
53e15af0 CL |
3451 | { |
3452 | if (slab_trylock(page)) { | |
434e245d | 3453 | validate_slab(s, page, map); |
53e15af0 CL |
3454 | slab_unlock(page); |
3455 | } else | |
3456 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3457 | s->name, page); | |
53e15af0 CL |
3458 | } |
3459 | ||
434e245d CL |
3460 | static int validate_slab_node(struct kmem_cache *s, |
3461 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3462 | { |
3463 | unsigned long count = 0; | |
3464 | struct page *page; | |
3465 | unsigned long flags; | |
3466 | ||
3467 | spin_lock_irqsave(&n->list_lock, flags); | |
3468 | ||
3469 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3470 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3471 | count++; |
3472 | } | |
3473 | if (count != n->nr_partial) | |
3474 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3475 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3476 | ||
3477 | if (!(s->flags & SLAB_STORE_USER)) | |
3478 | goto out; | |
3479 | ||
3480 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3481 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3482 | count++; |
3483 | } | |
3484 | if (count != atomic_long_read(&n->nr_slabs)) | |
3485 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3486 | "counter=%ld\n", s->name, count, | |
3487 | atomic_long_read(&n->nr_slabs)); | |
3488 | ||
3489 | out: | |
3490 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3491 | return count; | |
3492 | } | |
3493 | ||
434e245d | 3494 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3495 | { |
3496 | int node; | |
3497 | unsigned long count = 0; | |
205ab99d | 3498 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3499 | sizeof(unsigned long), GFP_KERNEL); |
3500 | ||
3501 | if (!map) | |
3502 | return -ENOMEM; | |
53e15af0 CL |
3503 | |
3504 | flush_all(s); | |
f64dc58c | 3505 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3506 | struct kmem_cache_node *n = get_node(s, node); |
3507 | ||
434e245d | 3508 | count += validate_slab_node(s, n, map); |
53e15af0 | 3509 | } |
434e245d | 3510 | kfree(map); |
53e15af0 CL |
3511 | return count; |
3512 | } | |
3513 | ||
b3459709 CL |
3514 | #ifdef SLUB_RESILIENCY_TEST |
3515 | static void resiliency_test(void) | |
3516 | { | |
3517 | u8 *p; | |
3518 | ||
a016471a DR |
3519 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); |
3520 | ||
b3459709 CL |
3521 | printk(KERN_ERR "SLUB resiliency testing\n"); |
3522 | printk(KERN_ERR "-----------------------\n"); | |
3523 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3524 | ||
3525 | p = kzalloc(16, GFP_KERNEL); | |
3526 | p[16] = 0x12; | |
3527 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3528 | " 0x12->0x%p\n\n", p + 16); | |
3529 | ||
a016471a | 3530 | validate_slab_cache(kmalloc_caches[4]); |
b3459709 CL |
3531 | |
3532 | /* Hmmm... The next two are dangerous */ | |
3533 | p = kzalloc(32, GFP_KERNEL); | |
3534 | p[32 + sizeof(void *)] = 0x34; | |
3535 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3536 | " 0x34 -> -0x%p\n", p); |
3537 | printk(KERN_ERR | |
3538 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 | 3539 | |
a016471a | 3540 | validate_slab_cache(kmalloc_caches[5]); |
b3459709 CL |
3541 | p = kzalloc(64, GFP_KERNEL); |
3542 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3543 | *p = 0x56; | |
3544 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3545 | p); | |
3adbefee IM |
3546 | printk(KERN_ERR |
3547 | "If allocated object is overwritten then not detectable\n\n"); | |
a016471a | 3548 | validate_slab_cache(kmalloc_caches[6]); |
b3459709 CL |
3549 | |
3550 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3551 | p = kzalloc(128, GFP_KERNEL); | |
3552 | kfree(p); | |
3553 | *p = 0x78; | |
3554 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
a016471a | 3555 | validate_slab_cache(kmalloc_caches[7]); |
b3459709 CL |
3556 | |
3557 | p = kzalloc(256, GFP_KERNEL); | |
3558 | kfree(p); | |
3559 | p[50] = 0x9a; | |
3adbefee IM |
3560 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3561 | p); | |
a016471a | 3562 | validate_slab_cache(kmalloc_caches[8]); |
b3459709 CL |
3563 | |
3564 | p = kzalloc(512, GFP_KERNEL); | |
3565 | kfree(p); | |
3566 | p[512] = 0xab; | |
3567 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
a016471a | 3568 | validate_slab_cache(kmalloc_caches[9]); |
b3459709 CL |
3569 | } |
3570 | #else | |
3571 | static void resiliency_test(void) {}; | |
3572 | #endif | |
3573 | ||
88a420e4 | 3574 | /* |
672bba3a | 3575 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3576 | * and freed. |
3577 | */ | |
3578 | ||
3579 | struct location { | |
3580 | unsigned long count; | |
ce71e27c | 3581 | unsigned long addr; |
45edfa58 CL |
3582 | long long sum_time; |
3583 | long min_time; | |
3584 | long max_time; | |
3585 | long min_pid; | |
3586 | long max_pid; | |
174596a0 | 3587 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3588 | nodemask_t nodes; |
88a420e4 CL |
3589 | }; |
3590 | ||
3591 | struct loc_track { | |
3592 | unsigned long max; | |
3593 | unsigned long count; | |
3594 | struct location *loc; | |
3595 | }; | |
3596 | ||
3597 | static void free_loc_track(struct loc_track *t) | |
3598 | { | |
3599 | if (t->max) | |
3600 | free_pages((unsigned long)t->loc, | |
3601 | get_order(sizeof(struct location) * t->max)); | |
3602 | } | |
3603 | ||
68dff6a9 | 3604 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3605 | { |
3606 | struct location *l; | |
3607 | int order; | |
3608 | ||
88a420e4 CL |
3609 | order = get_order(sizeof(struct location) * max); |
3610 | ||
68dff6a9 | 3611 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3612 | if (!l) |
3613 | return 0; | |
3614 | ||
3615 | if (t->count) { | |
3616 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3617 | free_loc_track(t); | |
3618 | } | |
3619 | t->max = max; | |
3620 | t->loc = l; | |
3621 | return 1; | |
3622 | } | |
3623 | ||
3624 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3625 | const struct track *track) |
88a420e4 CL |
3626 | { |
3627 | long start, end, pos; | |
3628 | struct location *l; | |
ce71e27c | 3629 | unsigned long caddr; |
45edfa58 | 3630 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3631 | |
3632 | start = -1; | |
3633 | end = t->count; | |
3634 | ||
3635 | for ( ; ; ) { | |
3636 | pos = start + (end - start + 1) / 2; | |
3637 | ||
3638 | /* | |
3639 | * There is nothing at "end". If we end up there | |
3640 | * we need to add something to before end. | |
3641 | */ | |
3642 | if (pos == end) | |
3643 | break; | |
3644 | ||
3645 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3646 | if (track->addr == caddr) { |
3647 | ||
3648 | l = &t->loc[pos]; | |
3649 | l->count++; | |
3650 | if (track->when) { | |
3651 | l->sum_time += age; | |
3652 | if (age < l->min_time) | |
3653 | l->min_time = age; | |
3654 | if (age > l->max_time) | |
3655 | l->max_time = age; | |
3656 | ||
3657 | if (track->pid < l->min_pid) | |
3658 | l->min_pid = track->pid; | |
3659 | if (track->pid > l->max_pid) | |
3660 | l->max_pid = track->pid; | |
3661 | ||
174596a0 RR |
3662 | cpumask_set_cpu(track->cpu, |
3663 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3664 | } |
3665 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3666 | return 1; |
3667 | } | |
3668 | ||
45edfa58 | 3669 | if (track->addr < caddr) |
88a420e4 CL |
3670 | end = pos; |
3671 | else | |
3672 | start = pos; | |
3673 | } | |
3674 | ||
3675 | /* | |
672bba3a | 3676 | * Not found. Insert new tracking element. |
88a420e4 | 3677 | */ |
68dff6a9 | 3678 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3679 | return 0; |
3680 | ||
3681 | l = t->loc + pos; | |
3682 | if (pos < t->count) | |
3683 | memmove(l + 1, l, | |
3684 | (t->count - pos) * sizeof(struct location)); | |
3685 | t->count++; | |
3686 | l->count = 1; | |
45edfa58 CL |
3687 | l->addr = track->addr; |
3688 | l->sum_time = age; | |
3689 | l->min_time = age; | |
3690 | l->max_time = age; | |
3691 | l->min_pid = track->pid; | |
3692 | l->max_pid = track->pid; | |
174596a0 RR |
3693 | cpumask_clear(to_cpumask(l->cpus)); |
3694 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3695 | nodes_clear(l->nodes); |
3696 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3697 | return 1; |
3698 | } | |
3699 | ||
3700 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b ED |
3701 | struct page *page, enum track_item alloc, |
3702 | long *map) | |
88a420e4 | 3703 | { |
a973e9dd | 3704 | void *addr = page_address(page); |
88a420e4 CL |
3705 | void *p; |
3706 | ||
39b26464 | 3707 | bitmap_zero(map, page->objects); |
7656c72b CL |
3708 | for_each_free_object(p, s, page->freelist) |
3709 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3710 | |
224a88be | 3711 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3712 | if (!test_bit(slab_index(p, s, addr), map)) |
3713 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3714 | } |
3715 | ||
3716 | static int list_locations(struct kmem_cache *s, char *buf, | |
3717 | enum track_item alloc) | |
3718 | { | |
e374d483 | 3719 | int len = 0; |
88a420e4 | 3720 | unsigned long i; |
68dff6a9 | 3721 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3722 | int node; |
bbd7d57b ED |
3723 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3724 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3725 | |
bbd7d57b ED |
3726 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3727 | GFP_TEMPORARY)) { | |
3728 | kfree(map); | |
68dff6a9 | 3729 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3730 | } |
88a420e4 CL |
3731 | /* Push back cpu slabs */ |
3732 | flush_all(s); | |
3733 | ||
f64dc58c | 3734 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3735 | struct kmem_cache_node *n = get_node(s, node); |
3736 | unsigned long flags; | |
3737 | struct page *page; | |
3738 | ||
9e86943b | 3739 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3740 | continue; |
3741 | ||
3742 | spin_lock_irqsave(&n->list_lock, flags); | |
3743 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3744 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3745 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3746 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3747 | spin_unlock_irqrestore(&n->list_lock, flags); |
3748 | } | |
3749 | ||
3750 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3751 | struct location *l = &t.loc[i]; |
88a420e4 | 3752 | |
9c246247 | 3753 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3754 | break; |
e374d483 | 3755 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3756 | |
3757 | if (l->addr) | |
e374d483 | 3758 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3759 | else |
e374d483 | 3760 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3761 | |
3762 | if (l->sum_time != l->min_time) { | |
e374d483 | 3763 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3764 | l->min_time, |
3765 | (long)div_u64(l->sum_time, l->count), | |
3766 | l->max_time); | |
45edfa58 | 3767 | } else |
e374d483 | 3768 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3769 | l->min_time); |
3770 | ||
3771 | if (l->min_pid != l->max_pid) | |
e374d483 | 3772 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3773 | l->min_pid, l->max_pid); |
3774 | else | |
e374d483 | 3775 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3776 | l->min_pid); |
3777 | ||
174596a0 RR |
3778 | if (num_online_cpus() > 1 && |
3779 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3780 | len < PAGE_SIZE - 60) { |
3781 | len += sprintf(buf + len, " cpus="); | |
3782 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3783 | to_cpumask(l->cpus)); |
45edfa58 CL |
3784 | } |
3785 | ||
62bc62a8 | 3786 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3787 | len < PAGE_SIZE - 60) { |
3788 | len += sprintf(buf + len, " nodes="); | |
3789 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3790 | l->nodes); |
3791 | } | |
3792 | ||
e374d483 | 3793 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3794 | } |
3795 | ||
3796 | free_loc_track(&t); | |
bbd7d57b | 3797 | kfree(map); |
88a420e4 | 3798 | if (!t.count) |
e374d483 HH |
3799 | len += sprintf(buf, "No data\n"); |
3800 | return len; | |
88a420e4 CL |
3801 | } |
3802 | ||
81819f0f | 3803 | enum slab_stat_type { |
205ab99d CL |
3804 | SL_ALL, /* All slabs */ |
3805 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3806 | SL_CPU, /* Only slabs used for cpu caches */ | |
3807 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3808 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3809 | }; |
3810 | ||
205ab99d | 3811 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3812 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3813 | #define SO_CPU (1 << SL_CPU) | |
3814 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3815 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3816 | |
62e5c4b4 CG |
3817 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3818 | char *buf, unsigned long flags) | |
81819f0f CL |
3819 | { |
3820 | unsigned long total = 0; | |
81819f0f CL |
3821 | int node; |
3822 | int x; | |
3823 | unsigned long *nodes; | |
3824 | unsigned long *per_cpu; | |
3825 | ||
3826 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3827 | if (!nodes) |
3828 | return -ENOMEM; | |
81819f0f CL |
3829 | per_cpu = nodes + nr_node_ids; |
3830 | ||
205ab99d CL |
3831 | if (flags & SO_CPU) { |
3832 | int cpu; | |
81819f0f | 3833 | |
205ab99d | 3834 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 3835 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 3836 | |
205ab99d CL |
3837 | if (!c || c->node < 0) |
3838 | continue; | |
3839 | ||
3840 | if (c->page) { | |
3841 | if (flags & SO_TOTAL) | |
3842 | x = c->page->objects; | |
3843 | else if (flags & SO_OBJECTS) | |
3844 | x = c->page->inuse; | |
81819f0f CL |
3845 | else |
3846 | x = 1; | |
205ab99d | 3847 | |
81819f0f | 3848 | total += x; |
205ab99d | 3849 | nodes[c->node] += x; |
81819f0f | 3850 | } |
205ab99d | 3851 | per_cpu[c->node]++; |
81819f0f CL |
3852 | } |
3853 | } | |
3854 | ||
205ab99d CL |
3855 | if (flags & SO_ALL) { |
3856 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3857 | struct kmem_cache_node *n = get_node(s, node); | |
3858 | ||
3859 | if (flags & SO_TOTAL) | |
3860 | x = atomic_long_read(&n->total_objects); | |
3861 | else if (flags & SO_OBJECTS) | |
3862 | x = atomic_long_read(&n->total_objects) - | |
3863 | count_partial(n, count_free); | |
81819f0f | 3864 | |
81819f0f | 3865 | else |
205ab99d | 3866 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3867 | total += x; |
3868 | nodes[node] += x; | |
3869 | } | |
3870 | ||
205ab99d CL |
3871 | } else if (flags & SO_PARTIAL) { |
3872 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3873 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3874 | |
205ab99d CL |
3875 | if (flags & SO_TOTAL) |
3876 | x = count_partial(n, count_total); | |
3877 | else if (flags & SO_OBJECTS) | |
3878 | x = count_partial(n, count_inuse); | |
81819f0f | 3879 | else |
205ab99d | 3880 | x = n->nr_partial; |
81819f0f CL |
3881 | total += x; |
3882 | nodes[node] += x; | |
3883 | } | |
3884 | } | |
81819f0f CL |
3885 | x = sprintf(buf, "%lu", total); |
3886 | #ifdef CONFIG_NUMA | |
f64dc58c | 3887 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3888 | if (nodes[node]) |
3889 | x += sprintf(buf + x, " N%d=%lu", | |
3890 | node, nodes[node]); | |
3891 | #endif | |
3892 | kfree(nodes); | |
3893 | return x + sprintf(buf + x, "\n"); | |
3894 | } | |
3895 | ||
3896 | static int any_slab_objects(struct kmem_cache *s) | |
3897 | { | |
3898 | int node; | |
81819f0f | 3899 | |
dfb4f096 | 3900 | for_each_online_node(node) { |
81819f0f CL |
3901 | struct kmem_cache_node *n = get_node(s, node); |
3902 | ||
dfb4f096 CL |
3903 | if (!n) |
3904 | continue; | |
3905 | ||
4ea33e2d | 3906 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3907 | return 1; |
3908 | } | |
3909 | return 0; | |
3910 | } | |
3911 | ||
3912 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3913 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3914 | ||
3915 | struct slab_attribute { | |
3916 | struct attribute attr; | |
3917 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3918 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3919 | }; | |
3920 | ||
3921 | #define SLAB_ATTR_RO(_name) \ | |
3922 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3923 | ||
3924 | #define SLAB_ATTR(_name) \ | |
3925 | static struct slab_attribute _name##_attr = \ | |
3926 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3927 | ||
81819f0f CL |
3928 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3929 | { | |
3930 | return sprintf(buf, "%d\n", s->size); | |
3931 | } | |
3932 | SLAB_ATTR_RO(slab_size); | |
3933 | ||
3934 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3935 | { | |
3936 | return sprintf(buf, "%d\n", s->align); | |
3937 | } | |
3938 | SLAB_ATTR_RO(align); | |
3939 | ||
3940 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3941 | { | |
3942 | return sprintf(buf, "%d\n", s->objsize); | |
3943 | } | |
3944 | SLAB_ATTR_RO(object_size); | |
3945 | ||
3946 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3947 | { | |
834f3d11 | 3948 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3949 | } |
3950 | SLAB_ATTR_RO(objs_per_slab); | |
3951 | ||
06b285dc CL |
3952 | static ssize_t order_store(struct kmem_cache *s, |
3953 | const char *buf, size_t length) | |
3954 | { | |
0121c619 CL |
3955 | unsigned long order; |
3956 | int err; | |
3957 | ||
3958 | err = strict_strtoul(buf, 10, &order); | |
3959 | if (err) | |
3960 | return err; | |
06b285dc CL |
3961 | |
3962 | if (order > slub_max_order || order < slub_min_order) | |
3963 | return -EINVAL; | |
3964 | ||
3965 | calculate_sizes(s, order); | |
3966 | return length; | |
3967 | } | |
3968 | ||
81819f0f CL |
3969 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3970 | { | |
834f3d11 | 3971 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3972 | } |
06b285dc | 3973 | SLAB_ATTR(order); |
81819f0f | 3974 | |
73d342b1 DR |
3975 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3976 | { | |
3977 | return sprintf(buf, "%lu\n", s->min_partial); | |
3978 | } | |
3979 | ||
3980 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3981 | size_t length) | |
3982 | { | |
3983 | unsigned long min; | |
3984 | int err; | |
3985 | ||
3986 | err = strict_strtoul(buf, 10, &min); | |
3987 | if (err) | |
3988 | return err; | |
3989 | ||
c0bdb232 | 3990 | set_min_partial(s, min); |
73d342b1 DR |
3991 | return length; |
3992 | } | |
3993 | SLAB_ATTR(min_partial); | |
3994 | ||
81819f0f CL |
3995 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3996 | { | |
3997 | if (s->ctor) { | |
3998 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3999 | ||
4000 | return n + sprintf(buf + n, "\n"); | |
4001 | } | |
4002 | return 0; | |
4003 | } | |
4004 | SLAB_ATTR_RO(ctor); | |
4005 | ||
81819f0f CL |
4006 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4007 | { | |
4008 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4009 | } | |
4010 | SLAB_ATTR_RO(aliases); | |
4011 | ||
4012 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
4013 | { | |
205ab99d | 4014 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
4015 | } |
4016 | SLAB_ATTR_RO(slabs); | |
4017 | ||
4018 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
4019 | { | |
d9acf4b7 | 4020 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4021 | } |
4022 | SLAB_ATTR_RO(partial); | |
4023 | ||
4024 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4025 | { | |
d9acf4b7 | 4026 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4027 | } |
4028 | SLAB_ATTR_RO(cpu_slabs); | |
4029 | ||
4030 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4031 | { | |
205ab99d | 4032 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4033 | } |
4034 | SLAB_ATTR_RO(objects); | |
4035 | ||
205ab99d CL |
4036 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4037 | { | |
4038 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4039 | } | |
4040 | SLAB_ATTR_RO(objects_partial); | |
4041 | ||
4042 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
4043 | { | |
4044 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4045 | } | |
4046 | SLAB_ATTR_RO(total_objects); | |
4047 | ||
81819f0f CL |
4048 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4049 | { | |
4050 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4051 | } | |
4052 | ||
4053 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4054 | const char *buf, size_t length) | |
4055 | { | |
4056 | s->flags &= ~SLAB_DEBUG_FREE; | |
4057 | if (buf[0] == '1') | |
4058 | s->flags |= SLAB_DEBUG_FREE; | |
4059 | return length; | |
4060 | } | |
4061 | SLAB_ATTR(sanity_checks); | |
4062 | ||
4063 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4064 | { | |
4065 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4066 | } | |
4067 | ||
4068 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4069 | size_t length) | |
4070 | { | |
4071 | s->flags &= ~SLAB_TRACE; | |
4072 | if (buf[0] == '1') | |
4073 | s->flags |= SLAB_TRACE; | |
4074 | return length; | |
4075 | } | |
4076 | SLAB_ATTR(trace); | |
4077 | ||
4c13dd3b DM |
4078 | #ifdef CONFIG_FAILSLAB |
4079 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4080 | { | |
4081 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4082 | } | |
4083 | ||
4084 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4085 | size_t length) | |
4086 | { | |
4087 | s->flags &= ~SLAB_FAILSLAB; | |
4088 | if (buf[0] == '1') | |
4089 | s->flags |= SLAB_FAILSLAB; | |
4090 | return length; | |
4091 | } | |
4092 | SLAB_ATTR(failslab); | |
4093 | #endif | |
4094 | ||
81819f0f CL |
4095 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4096 | { | |
4097 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4098 | } | |
4099 | ||
4100 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4101 | const char *buf, size_t length) | |
4102 | { | |
4103 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4104 | if (buf[0] == '1') | |
4105 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4106 | return length; | |
4107 | } | |
4108 | SLAB_ATTR(reclaim_account); | |
4109 | ||
4110 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4111 | { | |
5af60839 | 4112 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4113 | } |
4114 | SLAB_ATTR_RO(hwcache_align); | |
4115 | ||
4116 | #ifdef CONFIG_ZONE_DMA | |
4117 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4118 | { | |
4119 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4120 | } | |
4121 | SLAB_ATTR_RO(cache_dma); | |
4122 | #endif | |
4123 | ||
4124 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4125 | { | |
4126 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4127 | } | |
4128 | SLAB_ATTR_RO(destroy_by_rcu); | |
4129 | ||
4130 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4131 | { | |
4132 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4133 | } | |
4134 | ||
4135 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4136 | const char *buf, size_t length) | |
4137 | { | |
4138 | if (any_slab_objects(s)) | |
4139 | return -EBUSY; | |
4140 | ||
4141 | s->flags &= ~SLAB_RED_ZONE; | |
4142 | if (buf[0] == '1') | |
4143 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4144 | calculate_sizes(s, -1); |
81819f0f CL |
4145 | return length; |
4146 | } | |
4147 | SLAB_ATTR(red_zone); | |
4148 | ||
4149 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4150 | { | |
4151 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4152 | } | |
4153 | ||
4154 | static ssize_t poison_store(struct kmem_cache *s, | |
4155 | const char *buf, size_t length) | |
4156 | { | |
4157 | if (any_slab_objects(s)) | |
4158 | return -EBUSY; | |
4159 | ||
4160 | s->flags &= ~SLAB_POISON; | |
4161 | if (buf[0] == '1') | |
4162 | s->flags |= SLAB_POISON; | |
06b285dc | 4163 | calculate_sizes(s, -1); |
81819f0f CL |
4164 | return length; |
4165 | } | |
4166 | SLAB_ATTR(poison); | |
4167 | ||
4168 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4169 | { | |
4170 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4171 | } | |
4172 | ||
4173 | static ssize_t store_user_store(struct kmem_cache *s, | |
4174 | const char *buf, size_t length) | |
4175 | { | |
4176 | if (any_slab_objects(s)) | |
4177 | return -EBUSY; | |
4178 | ||
4179 | s->flags &= ~SLAB_STORE_USER; | |
4180 | if (buf[0] == '1') | |
4181 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4182 | calculate_sizes(s, -1); |
81819f0f CL |
4183 | return length; |
4184 | } | |
4185 | SLAB_ATTR(store_user); | |
4186 | ||
53e15af0 CL |
4187 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4188 | { | |
4189 | return 0; | |
4190 | } | |
4191 | ||
4192 | static ssize_t validate_store(struct kmem_cache *s, | |
4193 | const char *buf, size_t length) | |
4194 | { | |
434e245d CL |
4195 | int ret = -EINVAL; |
4196 | ||
4197 | if (buf[0] == '1') { | |
4198 | ret = validate_slab_cache(s); | |
4199 | if (ret >= 0) | |
4200 | ret = length; | |
4201 | } | |
4202 | return ret; | |
53e15af0 CL |
4203 | } |
4204 | SLAB_ATTR(validate); | |
4205 | ||
2086d26a CL |
4206 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4207 | { | |
4208 | return 0; | |
4209 | } | |
4210 | ||
4211 | static ssize_t shrink_store(struct kmem_cache *s, | |
4212 | const char *buf, size_t length) | |
4213 | { | |
4214 | if (buf[0] == '1') { | |
4215 | int rc = kmem_cache_shrink(s); | |
4216 | ||
4217 | if (rc) | |
4218 | return rc; | |
4219 | } else | |
4220 | return -EINVAL; | |
4221 | return length; | |
4222 | } | |
4223 | SLAB_ATTR(shrink); | |
4224 | ||
88a420e4 CL |
4225 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4226 | { | |
4227 | if (!(s->flags & SLAB_STORE_USER)) | |
4228 | return -ENOSYS; | |
4229 | return list_locations(s, buf, TRACK_ALLOC); | |
4230 | } | |
4231 | SLAB_ATTR_RO(alloc_calls); | |
4232 | ||
4233 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4234 | { | |
4235 | if (!(s->flags & SLAB_STORE_USER)) | |
4236 | return -ENOSYS; | |
4237 | return list_locations(s, buf, TRACK_FREE); | |
4238 | } | |
4239 | SLAB_ATTR_RO(free_calls); | |
4240 | ||
81819f0f | 4241 | #ifdef CONFIG_NUMA |
9824601e | 4242 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4243 | { |
9824601e | 4244 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4245 | } |
4246 | ||
9824601e | 4247 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4248 | const char *buf, size_t length) |
4249 | { | |
0121c619 CL |
4250 | unsigned long ratio; |
4251 | int err; | |
4252 | ||
4253 | err = strict_strtoul(buf, 10, &ratio); | |
4254 | if (err) | |
4255 | return err; | |
4256 | ||
e2cb96b7 | 4257 | if (ratio <= 100) |
0121c619 | 4258 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4259 | |
81819f0f CL |
4260 | return length; |
4261 | } | |
9824601e | 4262 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4263 | #endif |
4264 | ||
8ff12cfc | 4265 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4266 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4267 | { | |
4268 | unsigned long sum = 0; | |
4269 | int cpu; | |
4270 | int len; | |
4271 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4272 | ||
4273 | if (!data) | |
4274 | return -ENOMEM; | |
4275 | ||
4276 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4277 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4278 | |
4279 | data[cpu] = x; | |
4280 | sum += x; | |
4281 | } | |
4282 | ||
4283 | len = sprintf(buf, "%lu", sum); | |
4284 | ||
50ef37b9 | 4285 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4286 | for_each_online_cpu(cpu) { |
4287 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4288 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4289 | } |
50ef37b9 | 4290 | #endif |
8ff12cfc CL |
4291 | kfree(data); |
4292 | return len + sprintf(buf + len, "\n"); | |
4293 | } | |
4294 | ||
78eb00cc DR |
4295 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4296 | { | |
4297 | int cpu; | |
4298 | ||
4299 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4300 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4301 | } |
4302 | ||
8ff12cfc CL |
4303 | #define STAT_ATTR(si, text) \ |
4304 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4305 | { \ | |
4306 | return show_stat(s, buf, si); \ | |
4307 | } \ | |
78eb00cc DR |
4308 | static ssize_t text##_store(struct kmem_cache *s, \ |
4309 | const char *buf, size_t length) \ | |
4310 | { \ | |
4311 | if (buf[0] != '0') \ | |
4312 | return -EINVAL; \ | |
4313 | clear_stat(s, si); \ | |
4314 | return length; \ | |
4315 | } \ | |
4316 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4317 | |
4318 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4319 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4320 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4321 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4322 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4323 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4324 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4325 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4326 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4327 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4328 | STAT_ATTR(FREE_SLAB, free_slab); | |
4329 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4330 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4331 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4332 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4333 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4334 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4335 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4336 | #endif |
4337 | ||
06428780 | 4338 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4339 | &slab_size_attr.attr, |
4340 | &object_size_attr.attr, | |
4341 | &objs_per_slab_attr.attr, | |
4342 | &order_attr.attr, | |
73d342b1 | 4343 | &min_partial_attr.attr, |
81819f0f | 4344 | &objects_attr.attr, |
205ab99d CL |
4345 | &objects_partial_attr.attr, |
4346 | &total_objects_attr.attr, | |
81819f0f CL |
4347 | &slabs_attr.attr, |
4348 | &partial_attr.attr, | |
4349 | &cpu_slabs_attr.attr, | |
4350 | &ctor_attr.attr, | |
81819f0f CL |
4351 | &aliases_attr.attr, |
4352 | &align_attr.attr, | |
4353 | &sanity_checks_attr.attr, | |
4354 | &trace_attr.attr, | |
4355 | &hwcache_align_attr.attr, | |
4356 | &reclaim_account_attr.attr, | |
4357 | &destroy_by_rcu_attr.attr, | |
4358 | &red_zone_attr.attr, | |
4359 | &poison_attr.attr, | |
4360 | &store_user_attr.attr, | |
53e15af0 | 4361 | &validate_attr.attr, |
2086d26a | 4362 | &shrink_attr.attr, |
88a420e4 CL |
4363 | &alloc_calls_attr.attr, |
4364 | &free_calls_attr.attr, | |
81819f0f CL |
4365 | #ifdef CONFIG_ZONE_DMA |
4366 | &cache_dma_attr.attr, | |
4367 | #endif | |
4368 | #ifdef CONFIG_NUMA | |
9824601e | 4369 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4370 | #endif |
4371 | #ifdef CONFIG_SLUB_STATS | |
4372 | &alloc_fastpath_attr.attr, | |
4373 | &alloc_slowpath_attr.attr, | |
4374 | &free_fastpath_attr.attr, | |
4375 | &free_slowpath_attr.attr, | |
4376 | &free_frozen_attr.attr, | |
4377 | &free_add_partial_attr.attr, | |
4378 | &free_remove_partial_attr.attr, | |
4379 | &alloc_from_partial_attr.attr, | |
4380 | &alloc_slab_attr.attr, | |
4381 | &alloc_refill_attr.attr, | |
4382 | &free_slab_attr.attr, | |
4383 | &cpuslab_flush_attr.attr, | |
4384 | &deactivate_full_attr.attr, | |
4385 | &deactivate_empty_attr.attr, | |
4386 | &deactivate_to_head_attr.attr, | |
4387 | &deactivate_to_tail_attr.attr, | |
4388 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4389 | &order_fallback_attr.attr, |
81819f0f | 4390 | #endif |
4c13dd3b DM |
4391 | #ifdef CONFIG_FAILSLAB |
4392 | &failslab_attr.attr, | |
4393 | #endif | |
4394 | ||
81819f0f CL |
4395 | NULL |
4396 | }; | |
4397 | ||
4398 | static struct attribute_group slab_attr_group = { | |
4399 | .attrs = slab_attrs, | |
4400 | }; | |
4401 | ||
4402 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4403 | struct attribute *attr, | |
4404 | char *buf) | |
4405 | { | |
4406 | struct slab_attribute *attribute; | |
4407 | struct kmem_cache *s; | |
4408 | int err; | |
4409 | ||
4410 | attribute = to_slab_attr(attr); | |
4411 | s = to_slab(kobj); | |
4412 | ||
4413 | if (!attribute->show) | |
4414 | return -EIO; | |
4415 | ||
4416 | err = attribute->show(s, buf); | |
4417 | ||
4418 | return err; | |
4419 | } | |
4420 | ||
4421 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4422 | struct attribute *attr, | |
4423 | const char *buf, size_t len) | |
4424 | { | |
4425 | struct slab_attribute *attribute; | |
4426 | struct kmem_cache *s; | |
4427 | int err; | |
4428 | ||
4429 | attribute = to_slab_attr(attr); | |
4430 | s = to_slab(kobj); | |
4431 | ||
4432 | if (!attribute->store) | |
4433 | return -EIO; | |
4434 | ||
4435 | err = attribute->store(s, buf, len); | |
4436 | ||
4437 | return err; | |
4438 | } | |
4439 | ||
151c602f CL |
4440 | static void kmem_cache_release(struct kobject *kobj) |
4441 | { | |
4442 | struct kmem_cache *s = to_slab(kobj); | |
4443 | ||
84c1cf62 | 4444 | kfree(s->name); |
151c602f CL |
4445 | kfree(s); |
4446 | } | |
4447 | ||
52cf25d0 | 4448 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4449 | .show = slab_attr_show, |
4450 | .store = slab_attr_store, | |
4451 | }; | |
4452 | ||
4453 | static struct kobj_type slab_ktype = { | |
4454 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4455 | .release = kmem_cache_release |
81819f0f CL |
4456 | }; |
4457 | ||
4458 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4459 | { | |
4460 | struct kobj_type *ktype = get_ktype(kobj); | |
4461 | ||
4462 | if (ktype == &slab_ktype) | |
4463 | return 1; | |
4464 | return 0; | |
4465 | } | |
4466 | ||
9cd43611 | 4467 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4468 | .filter = uevent_filter, |
4469 | }; | |
4470 | ||
27c3a314 | 4471 | static struct kset *slab_kset; |
81819f0f CL |
4472 | |
4473 | #define ID_STR_LENGTH 64 | |
4474 | ||
4475 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4476 | * |
4477 | * Format :[flags-]size | |
81819f0f CL |
4478 | */ |
4479 | static char *create_unique_id(struct kmem_cache *s) | |
4480 | { | |
4481 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4482 | char *p = name; | |
4483 | ||
4484 | BUG_ON(!name); | |
4485 | ||
4486 | *p++ = ':'; | |
4487 | /* | |
4488 | * First flags affecting slabcache operations. We will only | |
4489 | * get here for aliasable slabs so we do not need to support | |
4490 | * too many flags. The flags here must cover all flags that | |
4491 | * are matched during merging to guarantee that the id is | |
4492 | * unique. | |
4493 | */ | |
4494 | if (s->flags & SLAB_CACHE_DMA) | |
4495 | *p++ = 'd'; | |
4496 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4497 | *p++ = 'a'; | |
4498 | if (s->flags & SLAB_DEBUG_FREE) | |
4499 | *p++ = 'F'; | |
5a896d9e VN |
4500 | if (!(s->flags & SLAB_NOTRACK)) |
4501 | *p++ = 't'; | |
81819f0f CL |
4502 | if (p != name + 1) |
4503 | *p++ = '-'; | |
4504 | p += sprintf(p, "%07d", s->size); | |
4505 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4506 | return name; | |
4507 | } | |
4508 | ||
4509 | static int sysfs_slab_add(struct kmem_cache *s) | |
4510 | { | |
4511 | int err; | |
4512 | const char *name; | |
4513 | int unmergeable; | |
4514 | ||
4515 | if (slab_state < SYSFS) | |
4516 | /* Defer until later */ | |
4517 | return 0; | |
4518 | ||
4519 | unmergeable = slab_unmergeable(s); | |
4520 | if (unmergeable) { | |
4521 | /* | |
4522 | * Slabcache can never be merged so we can use the name proper. | |
4523 | * This is typically the case for debug situations. In that | |
4524 | * case we can catch duplicate names easily. | |
4525 | */ | |
27c3a314 | 4526 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4527 | name = s->name; |
4528 | } else { | |
4529 | /* | |
4530 | * Create a unique name for the slab as a target | |
4531 | * for the symlinks. | |
4532 | */ | |
4533 | name = create_unique_id(s); | |
4534 | } | |
4535 | ||
27c3a314 | 4536 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4537 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4538 | if (err) { | |
4539 | kobject_put(&s->kobj); | |
81819f0f | 4540 | return err; |
1eada11c | 4541 | } |
81819f0f CL |
4542 | |
4543 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4544 | if (err) { |
4545 | kobject_del(&s->kobj); | |
4546 | kobject_put(&s->kobj); | |
81819f0f | 4547 | return err; |
5788d8ad | 4548 | } |
81819f0f CL |
4549 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4550 | if (!unmergeable) { | |
4551 | /* Setup first alias */ | |
4552 | sysfs_slab_alias(s, s->name); | |
4553 | kfree(name); | |
4554 | } | |
4555 | return 0; | |
4556 | } | |
4557 | ||
4558 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4559 | { | |
2bce6485 CL |
4560 | if (slab_state < SYSFS) |
4561 | /* | |
4562 | * Sysfs has not been setup yet so no need to remove the | |
4563 | * cache from sysfs. | |
4564 | */ | |
4565 | return; | |
4566 | ||
81819f0f CL |
4567 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4568 | kobject_del(&s->kobj); | |
151c602f | 4569 | kobject_put(&s->kobj); |
81819f0f CL |
4570 | } |
4571 | ||
4572 | /* | |
4573 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4574 | * available lest we lose that information. |
81819f0f CL |
4575 | */ |
4576 | struct saved_alias { | |
4577 | struct kmem_cache *s; | |
4578 | const char *name; | |
4579 | struct saved_alias *next; | |
4580 | }; | |
4581 | ||
5af328a5 | 4582 | static struct saved_alias *alias_list; |
81819f0f CL |
4583 | |
4584 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4585 | { | |
4586 | struct saved_alias *al; | |
4587 | ||
4588 | if (slab_state == SYSFS) { | |
4589 | /* | |
4590 | * If we have a leftover link then remove it. | |
4591 | */ | |
27c3a314 GKH |
4592 | sysfs_remove_link(&slab_kset->kobj, name); |
4593 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4594 | } |
4595 | ||
4596 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4597 | if (!al) | |
4598 | return -ENOMEM; | |
4599 | ||
4600 | al->s = s; | |
4601 | al->name = name; | |
4602 | al->next = alias_list; | |
4603 | alias_list = al; | |
4604 | return 0; | |
4605 | } | |
4606 | ||
4607 | static int __init slab_sysfs_init(void) | |
4608 | { | |
5b95a4ac | 4609 | struct kmem_cache *s; |
81819f0f CL |
4610 | int err; |
4611 | ||
2bce6485 CL |
4612 | down_write(&slub_lock); |
4613 | ||
0ff21e46 | 4614 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4615 | if (!slab_kset) { |
2bce6485 | 4616 | up_write(&slub_lock); |
81819f0f CL |
4617 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4618 | return -ENOSYS; | |
4619 | } | |
4620 | ||
26a7bd03 CL |
4621 | slab_state = SYSFS; |
4622 | ||
5b95a4ac | 4623 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4624 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4625 | if (err) |
4626 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4627 | " to sysfs\n", s->name); | |
26a7bd03 | 4628 | } |
81819f0f CL |
4629 | |
4630 | while (alias_list) { | |
4631 | struct saved_alias *al = alias_list; | |
4632 | ||
4633 | alias_list = alias_list->next; | |
4634 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4635 | if (err) |
4636 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4637 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4638 | kfree(al); |
4639 | } | |
4640 | ||
2bce6485 | 4641 | up_write(&slub_lock); |
81819f0f CL |
4642 | resiliency_test(); |
4643 | return 0; | |
4644 | } | |
4645 | ||
4646 | __initcall(slab_sysfs_init); | |
81819f0f | 4647 | #endif |
57ed3eda PE |
4648 | |
4649 | /* | |
4650 | * The /proc/slabinfo ABI | |
4651 | */ | |
158a9624 | 4652 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4653 | static void print_slabinfo_header(struct seq_file *m) |
4654 | { | |
4655 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4656 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4657 | "<objperslab> <pagesperslab>"); | |
4658 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4659 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4660 | seq_putc(m, '\n'); | |
4661 | } | |
4662 | ||
4663 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4664 | { | |
4665 | loff_t n = *pos; | |
4666 | ||
4667 | down_read(&slub_lock); | |
4668 | if (!n) | |
4669 | print_slabinfo_header(m); | |
4670 | ||
4671 | return seq_list_start(&slab_caches, *pos); | |
4672 | } | |
4673 | ||
4674 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4675 | { | |
4676 | return seq_list_next(p, &slab_caches, pos); | |
4677 | } | |
4678 | ||
4679 | static void s_stop(struct seq_file *m, void *p) | |
4680 | { | |
4681 | up_read(&slub_lock); | |
4682 | } | |
4683 | ||
4684 | static int s_show(struct seq_file *m, void *p) | |
4685 | { | |
4686 | unsigned long nr_partials = 0; | |
4687 | unsigned long nr_slabs = 0; | |
4688 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4689 | unsigned long nr_objs = 0; |
4690 | unsigned long nr_free = 0; | |
57ed3eda PE |
4691 | struct kmem_cache *s; |
4692 | int node; | |
4693 | ||
4694 | s = list_entry(p, struct kmem_cache, list); | |
4695 | ||
4696 | for_each_online_node(node) { | |
4697 | struct kmem_cache_node *n = get_node(s, node); | |
4698 | ||
4699 | if (!n) | |
4700 | continue; | |
4701 | ||
4702 | nr_partials += n->nr_partial; | |
4703 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4704 | nr_objs += atomic_long_read(&n->total_objects); |
4705 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4706 | } |
4707 | ||
205ab99d | 4708 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4709 | |
4710 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4711 | nr_objs, s->size, oo_objects(s->oo), |
4712 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4713 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4714 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4715 | 0UL); | |
4716 | seq_putc(m, '\n'); | |
4717 | return 0; | |
4718 | } | |
4719 | ||
7b3c3a50 | 4720 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4721 | .start = s_start, |
4722 | .next = s_next, | |
4723 | .stop = s_stop, | |
4724 | .show = s_show, | |
4725 | }; | |
4726 | ||
7b3c3a50 AD |
4727 | static int slabinfo_open(struct inode *inode, struct file *file) |
4728 | { | |
4729 | return seq_open(file, &slabinfo_op); | |
4730 | } | |
4731 | ||
4732 | static const struct file_operations proc_slabinfo_operations = { | |
4733 | .open = slabinfo_open, | |
4734 | .read = seq_read, | |
4735 | .llseek = seq_lseek, | |
4736 | .release = seq_release, | |
4737 | }; | |
4738 | ||
4739 | static int __init slab_proc_init(void) | |
4740 | { | |
cf5d1131 | 4741 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4742 | return 0; |
4743 | } | |
4744 | module_init(slab_proc_init); | |
158a9624 | 4745 | #endif /* CONFIG_SLABINFO */ |