]>
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> |
36994e58 | 20 | #include <trace/kmemtrace.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 | ||
5577bd8a | 109 | #ifdef CONFIG_SLUB_DEBUG |
8a38082d | 110 | #define SLABDEBUG 1 |
5577bd8a CL |
111 | #else |
112 | #define SLABDEBUG 0 | |
113 | #endif | |
114 | ||
81819f0f CL |
115 | /* |
116 | * Issues still to be resolved: | |
117 | * | |
81819f0f CL |
118 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
119 | * | |
81819f0f CL |
120 | * - Variable sizing of the per node arrays |
121 | */ | |
122 | ||
123 | /* Enable to test recovery from slab corruption on boot */ | |
124 | #undef SLUB_RESILIENCY_TEST | |
125 | ||
2086d26a CL |
126 | /* |
127 | * Mininum number of partial slabs. These will be left on the partial | |
128 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
129 | */ | |
76be8950 | 130 | #define MIN_PARTIAL 5 |
e95eed57 | 131 | |
2086d26a CL |
132 | /* |
133 | * Maximum number of desirable partial slabs. | |
134 | * The existence of more partial slabs makes kmem_cache_shrink | |
135 | * sort the partial list by the number of objects in the. | |
136 | */ | |
137 | #define MAX_PARTIAL 10 | |
138 | ||
81819f0f CL |
139 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
140 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 141 | |
81819f0f CL |
142 | /* |
143 | * Set of flags that will prevent slab merging | |
144 | */ | |
145 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
146 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
147 | ||
148 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
149 | SLAB_CACHE_DMA) | |
150 | ||
151 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 152 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
153 | #endif |
154 | ||
155 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 156 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
157 | #endif |
158 | ||
210b5c06 CG |
159 | #define OO_SHIFT 16 |
160 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
161 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
162 | ||
81819f0f | 163 | /* Internal SLUB flags */ |
1ceef402 CL |
164 | #define __OBJECT_POISON 0x80000000 /* Poison object */ |
165 | #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */ | |
81819f0f CL |
166 | |
167 | static int kmem_size = sizeof(struct kmem_cache); | |
168 | ||
169 | #ifdef CONFIG_SMP | |
170 | static struct notifier_block slab_notifier; | |
171 | #endif | |
172 | ||
173 | static enum { | |
174 | DOWN, /* No slab functionality available */ | |
175 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
672bba3a | 176 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
177 | SYSFS /* Sysfs up */ |
178 | } slab_state = DOWN; | |
179 | ||
180 | /* A list of all slab caches on the system */ | |
181 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 182 | static LIST_HEAD(slab_caches); |
81819f0f | 183 | |
02cbc874 CL |
184 | /* |
185 | * Tracking user of a slab. | |
186 | */ | |
187 | struct track { | |
ce71e27c | 188 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
189 | int cpu; /* Was running on cpu */ |
190 | int pid; /* Pid context */ | |
191 | unsigned long when; /* When did the operation occur */ | |
192 | }; | |
193 | ||
194 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
195 | ||
f6acb635 | 196 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
197 | static int sysfs_slab_add(struct kmem_cache *); |
198 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
199 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 200 | |
81819f0f | 201 | #else |
0c710013 CL |
202 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
203 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
204 | { return 0; } | |
151c602f CL |
205 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
206 | { | |
207 | kfree(s); | |
208 | } | |
8ff12cfc | 209 | |
81819f0f CL |
210 | #endif |
211 | ||
8ff12cfc CL |
212 | static inline void stat(struct kmem_cache_cpu *c, enum stat_item si) |
213 | { | |
214 | #ifdef CONFIG_SLUB_STATS | |
215 | c->stat[si]++; | |
216 | #endif | |
217 | } | |
218 | ||
81819f0f CL |
219 | /******************************************************************** |
220 | * Core slab cache functions | |
221 | *******************************************************************/ | |
222 | ||
223 | int slab_is_available(void) | |
224 | { | |
225 | return slab_state >= UP; | |
226 | } | |
227 | ||
228 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
229 | { | |
230 | #ifdef CONFIG_NUMA | |
231 | return s->node[node]; | |
232 | #else | |
233 | return &s->local_node; | |
234 | #endif | |
235 | } | |
236 | ||
dfb4f096 CL |
237 | static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu) |
238 | { | |
4c93c355 CL |
239 | #ifdef CONFIG_SMP |
240 | return s->cpu_slab[cpu]; | |
241 | #else | |
242 | return &s->cpu_slab; | |
243 | #endif | |
dfb4f096 CL |
244 | } |
245 | ||
6446faa2 | 246 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
247 | static inline int check_valid_pointer(struct kmem_cache *s, |
248 | struct page *page, const void *object) | |
249 | { | |
250 | void *base; | |
251 | ||
a973e9dd | 252 | if (!object) |
02cbc874 CL |
253 | return 1; |
254 | ||
a973e9dd | 255 | base = page_address(page); |
39b26464 | 256 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
257 | (object - base) % s->size) { |
258 | return 0; | |
259 | } | |
260 | ||
261 | return 1; | |
262 | } | |
263 | ||
7656c72b CL |
264 | /* |
265 | * Slow version of get and set free pointer. | |
266 | * | |
267 | * This version requires touching the cache lines of kmem_cache which | |
268 | * we avoid to do in the fast alloc free paths. There we obtain the offset | |
269 | * from the page struct. | |
270 | */ | |
271 | static inline void *get_freepointer(struct kmem_cache *s, void *object) | |
272 | { | |
273 | return *(void **)(object + s->offset); | |
274 | } | |
275 | ||
276 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
277 | { | |
278 | *(void **)(object + s->offset) = fp; | |
279 | } | |
280 | ||
281 | /* Loop over all objects in a slab */ | |
224a88be CL |
282 | #define for_each_object(__p, __s, __addr, __objects) \ |
283 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
284 | __p += (__s)->size) |
285 | ||
286 | /* Scan freelist */ | |
287 | #define for_each_free_object(__p, __s, __free) \ | |
a973e9dd | 288 | for (__p = (__free); __p; __p = get_freepointer((__s), __p)) |
7656c72b CL |
289 | |
290 | /* Determine object index from a given position */ | |
291 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
292 | { | |
293 | return (p - addr) / s->size; | |
294 | } | |
295 | ||
834f3d11 CL |
296 | static inline struct kmem_cache_order_objects oo_make(int order, |
297 | unsigned long size) | |
298 | { | |
299 | struct kmem_cache_order_objects x = { | |
210b5c06 | 300 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size |
834f3d11 CL |
301 | }; |
302 | ||
303 | return x; | |
304 | } | |
305 | ||
306 | static inline int oo_order(struct kmem_cache_order_objects x) | |
307 | { | |
210b5c06 | 308 | return x.x >> OO_SHIFT; |
834f3d11 CL |
309 | } |
310 | ||
311 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
312 | { | |
210b5c06 | 313 | return x.x & OO_MASK; |
834f3d11 CL |
314 | } |
315 | ||
41ecc55b CL |
316 | #ifdef CONFIG_SLUB_DEBUG |
317 | /* | |
318 | * Debug settings: | |
319 | */ | |
f0630fff CL |
320 | #ifdef CONFIG_SLUB_DEBUG_ON |
321 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
322 | #else | |
41ecc55b | 323 | static int slub_debug; |
f0630fff | 324 | #endif |
41ecc55b CL |
325 | |
326 | static char *slub_debug_slabs; | |
327 | ||
81819f0f CL |
328 | /* |
329 | * Object debugging | |
330 | */ | |
331 | static void print_section(char *text, u8 *addr, unsigned int length) | |
332 | { | |
333 | int i, offset; | |
334 | int newline = 1; | |
335 | char ascii[17]; | |
336 | ||
337 | ascii[16] = 0; | |
338 | ||
339 | for (i = 0; i < length; i++) { | |
340 | if (newline) { | |
24922684 | 341 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
342 | newline = 0; |
343 | } | |
06428780 | 344 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
345 | offset = i % 16; |
346 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
347 | if (offset == 15) { | |
06428780 | 348 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
349 | newline = 1; |
350 | } | |
351 | } | |
352 | if (!newline) { | |
353 | i %= 16; | |
354 | while (i < 16) { | |
06428780 | 355 | printk(KERN_CONT " "); |
81819f0f CL |
356 | ascii[i] = ' '; |
357 | i++; | |
358 | } | |
06428780 | 359 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
360 | } |
361 | } | |
362 | ||
81819f0f CL |
363 | static struct track *get_track(struct kmem_cache *s, void *object, |
364 | enum track_item alloc) | |
365 | { | |
366 | struct track *p; | |
367 | ||
368 | if (s->offset) | |
369 | p = object + s->offset + sizeof(void *); | |
370 | else | |
371 | p = object + s->inuse; | |
372 | ||
373 | return p + alloc; | |
374 | } | |
375 | ||
376 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 377 | enum track_item alloc, unsigned long addr) |
81819f0f | 378 | { |
1a00df4a | 379 | struct track *p = get_track(s, object, alloc); |
81819f0f | 380 | |
81819f0f CL |
381 | if (addr) { |
382 | p->addr = addr; | |
383 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 384 | p->pid = current->pid; |
81819f0f CL |
385 | p->when = jiffies; |
386 | } else | |
387 | memset(p, 0, sizeof(struct track)); | |
388 | } | |
389 | ||
81819f0f CL |
390 | static void init_tracking(struct kmem_cache *s, void *object) |
391 | { | |
24922684 CL |
392 | if (!(s->flags & SLAB_STORE_USER)) |
393 | return; | |
394 | ||
ce71e27c EGM |
395 | set_track(s, object, TRACK_FREE, 0UL); |
396 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
397 | } |
398 | ||
399 | static void print_track(const char *s, struct track *t) | |
400 | { | |
401 | if (!t->addr) | |
402 | return; | |
403 | ||
7daf705f | 404 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 405 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
406 | } |
407 | ||
408 | static void print_tracking(struct kmem_cache *s, void *object) | |
409 | { | |
410 | if (!(s->flags & SLAB_STORE_USER)) | |
411 | return; | |
412 | ||
413 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
414 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
415 | } | |
416 | ||
417 | static void print_page_info(struct page *page) | |
418 | { | |
39b26464 CL |
419 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
420 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
421 | |
422 | } | |
423 | ||
424 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
425 | { | |
426 | va_list args; | |
427 | char buf[100]; | |
428 | ||
429 | va_start(args, fmt); | |
430 | vsnprintf(buf, sizeof(buf), fmt, args); | |
431 | va_end(args); | |
432 | printk(KERN_ERR "========================================" | |
433 | "=====================================\n"); | |
434 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
435 | printk(KERN_ERR "----------------------------------------" | |
436 | "-------------------------------------\n\n"); | |
81819f0f CL |
437 | } |
438 | ||
24922684 CL |
439 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
440 | { | |
441 | va_list args; | |
442 | char buf[100]; | |
443 | ||
444 | va_start(args, fmt); | |
445 | vsnprintf(buf, sizeof(buf), fmt, args); | |
446 | va_end(args); | |
447 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
448 | } | |
449 | ||
450 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
451 | { |
452 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 453 | u8 *addr = page_address(page); |
24922684 CL |
454 | |
455 | print_tracking(s, p); | |
456 | ||
457 | print_page_info(page); | |
458 | ||
459 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
460 | p, p - addr, get_freepointer(s, p)); | |
461 | ||
462 | if (p > addr + 16) | |
463 | print_section("Bytes b4", p - 16, 16); | |
464 | ||
0ebd652b | 465 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
466 | |
467 | if (s->flags & SLAB_RED_ZONE) | |
468 | print_section("Redzone", p + s->objsize, | |
469 | s->inuse - s->objsize); | |
470 | ||
81819f0f CL |
471 | if (s->offset) |
472 | off = s->offset + sizeof(void *); | |
473 | else | |
474 | off = s->inuse; | |
475 | ||
24922684 | 476 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 477 | off += 2 * sizeof(struct track); |
81819f0f CL |
478 | |
479 | if (off != s->size) | |
480 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
481 | print_section("Padding", p + off, s->size - off); |
482 | ||
483 | dump_stack(); | |
81819f0f CL |
484 | } |
485 | ||
486 | static void object_err(struct kmem_cache *s, struct page *page, | |
487 | u8 *object, char *reason) | |
488 | { | |
3dc50637 | 489 | slab_bug(s, "%s", reason); |
24922684 | 490 | print_trailer(s, page, object); |
81819f0f CL |
491 | } |
492 | ||
24922684 | 493 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
494 | { |
495 | va_list args; | |
496 | char buf[100]; | |
497 | ||
24922684 CL |
498 | va_start(args, fmt); |
499 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 500 | va_end(args); |
3dc50637 | 501 | slab_bug(s, "%s", buf); |
24922684 | 502 | print_page_info(page); |
81819f0f CL |
503 | dump_stack(); |
504 | } | |
505 | ||
506 | static void init_object(struct kmem_cache *s, void *object, int active) | |
507 | { | |
508 | u8 *p = object; | |
509 | ||
510 | if (s->flags & __OBJECT_POISON) { | |
511 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 512 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
513 | } |
514 | ||
515 | if (s->flags & SLAB_RED_ZONE) | |
516 | memset(p + s->objsize, | |
517 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
518 | s->inuse - s->objsize); | |
519 | } | |
520 | ||
24922684 | 521 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
522 | { |
523 | while (bytes) { | |
524 | if (*start != (u8)value) | |
24922684 | 525 | return start; |
81819f0f CL |
526 | start++; |
527 | bytes--; | |
528 | } | |
24922684 CL |
529 | return NULL; |
530 | } | |
531 | ||
532 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
533 | void *from, void *to) | |
534 | { | |
535 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
536 | memset(from, data, to - from); | |
537 | } | |
538 | ||
539 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
540 | u8 *object, char *what, | |
06428780 | 541 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
542 | { |
543 | u8 *fault; | |
544 | u8 *end; | |
545 | ||
546 | fault = check_bytes(start, value, bytes); | |
547 | if (!fault) | |
548 | return 1; | |
549 | ||
550 | end = start + bytes; | |
551 | while (end > fault && end[-1] == value) | |
552 | end--; | |
553 | ||
554 | slab_bug(s, "%s overwritten", what); | |
555 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
556 | fault, end - 1, fault[0], value); | |
557 | print_trailer(s, page, object); | |
558 | ||
559 | restore_bytes(s, what, value, fault, end); | |
560 | return 0; | |
81819f0f CL |
561 | } |
562 | ||
81819f0f CL |
563 | /* |
564 | * Object layout: | |
565 | * | |
566 | * object address | |
567 | * Bytes of the object to be managed. | |
568 | * If the freepointer may overlay the object then the free | |
569 | * pointer is the first word of the object. | |
672bba3a | 570 | * |
81819f0f CL |
571 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
572 | * 0xa5 (POISON_END) | |
573 | * | |
574 | * object + s->objsize | |
575 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
576 | * Padding is extended by another word if Redzoning is enabled and |
577 | * objsize == inuse. | |
578 | * | |
81819f0f CL |
579 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
580 | * 0xcc (RED_ACTIVE) for objects in use. | |
581 | * | |
582 | * object + s->inuse | |
672bba3a CL |
583 | * Meta data starts here. |
584 | * | |
81819f0f CL |
585 | * A. Free pointer (if we cannot overwrite object on free) |
586 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 587 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 588 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
589 | * before the word boundary. |
590 | * | |
591 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
592 | * |
593 | * object + s->size | |
672bba3a | 594 | * Nothing is used beyond s->size. |
81819f0f | 595 | * |
672bba3a CL |
596 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
597 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
598 | * may be used with merged slabcaches. |
599 | */ | |
600 | ||
81819f0f CL |
601 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
602 | { | |
603 | unsigned long off = s->inuse; /* The end of info */ | |
604 | ||
605 | if (s->offset) | |
606 | /* Freepointer is placed after the object. */ | |
607 | off += sizeof(void *); | |
608 | ||
609 | if (s->flags & SLAB_STORE_USER) | |
610 | /* We also have user information there */ | |
611 | off += 2 * sizeof(struct track); | |
612 | ||
613 | if (s->size == off) | |
614 | return 1; | |
615 | ||
24922684 CL |
616 | return check_bytes_and_report(s, page, p, "Object padding", |
617 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
618 | } |
619 | ||
39b26464 | 620 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
621 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
622 | { | |
24922684 CL |
623 | u8 *start; |
624 | u8 *fault; | |
625 | u8 *end; | |
626 | int length; | |
627 | int remainder; | |
81819f0f CL |
628 | |
629 | if (!(s->flags & SLAB_POISON)) | |
630 | return 1; | |
631 | ||
a973e9dd | 632 | start = page_address(page); |
834f3d11 | 633 | length = (PAGE_SIZE << compound_order(page)); |
39b26464 CL |
634 | end = start + length; |
635 | remainder = length % s->size; | |
81819f0f CL |
636 | if (!remainder) |
637 | return 1; | |
638 | ||
39b26464 | 639 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
640 | if (!fault) |
641 | return 1; | |
642 | while (end > fault && end[-1] == POISON_INUSE) | |
643 | end--; | |
644 | ||
645 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 646 | print_section("Padding", end - remainder, remainder); |
24922684 CL |
647 | |
648 | restore_bytes(s, "slab padding", POISON_INUSE, start, end); | |
649 | return 0; | |
81819f0f CL |
650 | } |
651 | ||
652 | static int check_object(struct kmem_cache *s, struct page *page, | |
653 | void *object, int active) | |
654 | { | |
655 | u8 *p = object; | |
656 | u8 *endobject = object + s->objsize; | |
657 | ||
658 | if (s->flags & SLAB_RED_ZONE) { | |
659 | unsigned int red = | |
660 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
661 | ||
24922684 CL |
662 | if (!check_bytes_and_report(s, page, object, "Redzone", |
663 | endobject, red, s->inuse - s->objsize)) | |
81819f0f | 664 | return 0; |
81819f0f | 665 | } else { |
3adbefee IM |
666 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
667 | check_bytes_and_report(s, page, p, "Alignment padding", | |
668 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
669 | } | |
81819f0f CL |
670 | } |
671 | ||
672 | if (s->flags & SLAB_POISON) { | |
673 | if (!active && (s->flags & __OBJECT_POISON) && | |
24922684 CL |
674 | (!check_bytes_and_report(s, page, p, "Poison", p, |
675 | POISON_FREE, s->objsize - 1) || | |
676 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 677 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 678 | return 0; |
81819f0f CL |
679 | /* |
680 | * check_pad_bytes cleans up on its own. | |
681 | */ | |
682 | check_pad_bytes(s, page, p); | |
683 | } | |
684 | ||
685 | if (!s->offset && active) | |
686 | /* | |
687 | * Object and freepointer overlap. Cannot check | |
688 | * freepointer while object is allocated. | |
689 | */ | |
690 | return 1; | |
691 | ||
692 | /* Check free pointer validity */ | |
693 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
694 | object_err(s, page, p, "Freepointer corrupt"); | |
695 | /* | |
9f6c708e | 696 | * No choice but to zap it and thus lose the remainder |
81819f0f | 697 | * of the free objects in this slab. May cause |
672bba3a | 698 | * another error because the object count is now wrong. |
81819f0f | 699 | */ |
a973e9dd | 700 | set_freepointer(s, p, NULL); |
81819f0f CL |
701 | return 0; |
702 | } | |
703 | return 1; | |
704 | } | |
705 | ||
706 | static int check_slab(struct kmem_cache *s, struct page *page) | |
707 | { | |
39b26464 CL |
708 | int maxobj; |
709 | ||
81819f0f CL |
710 | VM_BUG_ON(!irqs_disabled()); |
711 | ||
712 | if (!PageSlab(page)) { | |
24922684 | 713 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
714 | return 0; |
715 | } | |
39b26464 CL |
716 | |
717 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | |
718 | if (page->objects > maxobj) { | |
719 | slab_err(s, page, "objects %u > max %u", | |
720 | s->name, page->objects, maxobj); | |
721 | return 0; | |
722 | } | |
723 | if (page->inuse > page->objects) { | |
24922684 | 724 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 725 | s->name, page->inuse, page->objects); |
81819f0f CL |
726 | return 0; |
727 | } | |
728 | /* Slab_pad_check fixes things up after itself */ | |
729 | slab_pad_check(s, page); | |
730 | return 1; | |
731 | } | |
732 | ||
733 | /* | |
672bba3a CL |
734 | * Determine if a certain object on a page is on the freelist. Must hold the |
735 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
736 | */ |
737 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
738 | { | |
739 | int nr = 0; | |
740 | void *fp = page->freelist; | |
741 | void *object = NULL; | |
224a88be | 742 | unsigned long max_objects; |
81819f0f | 743 | |
39b26464 | 744 | while (fp && nr <= page->objects) { |
81819f0f CL |
745 | if (fp == search) |
746 | return 1; | |
747 | if (!check_valid_pointer(s, page, fp)) { | |
748 | if (object) { | |
749 | object_err(s, page, object, | |
750 | "Freechain corrupt"); | |
a973e9dd | 751 | set_freepointer(s, object, NULL); |
81819f0f CL |
752 | break; |
753 | } else { | |
24922684 | 754 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 755 | page->freelist = NULL; |
39b26464 | 756 | page->inuse = page->objects; |
24922684 | 757 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
758 | return 0; |
759 | } | |
760 | break; | |
761 | } | |
762 | object = fp; | |
763 | fp = get_freepointer(s, object); | |
764 | nr++; | |
765 | } | |
766 | ||
224a88be | 767 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; |
210b5c06 CG |
768 | if (max_objects > MAX_OBJS_PER_PAGE) |
769 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
770 | |
771 | if (page->objects != max_objects) { | |
772 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
773 | "should be %d", page->objects, max_objects); | |
774 | page->objects = max_objects; | |
775 | slab_fix(s, "Number of objects adjusted."); | |
776 | } | |
39b26464 | 777 | if (page->inuse != page->objects - nr) { |
70d71228 | 778 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
779 | "counted were %d", page->inuse, page->objects - nr); |
780 | page->inuse = page->objects - nr; | |
24922684 | 781 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
782 | } |
783 | return search == NULL; | |
784 | } | |
785 | ||
0121c619 CL |
786 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
787 | int alloc) | |
3ec09742 CL |
788 | { |
789 | if (s->flags & SLAB_TRACE) { | |
790 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
791 | s->name, | |
792 | alloc ? "alloc" : "free", | |
793 | object, page->inuse, | |
794 | page->freelist); | |
795 | ||
796 | if (!alloc) | |
797 | print_section("Object", (void *)object, s->objsize); | |
798 | ||
799 | dump_stack(); | |
800 | } | |
801 | } | |
802 | ||
643b1138 | 803 | /* |
672bba3a | 804 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 805 | */ |
e95eed57 | 806 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 807 | { |
643b1138 CL |
808 | spin_lock(&n->list_lock); |
809 | list_add(&page->lru, &n->full); | |
810 | spin_unlock(&n->list_lock); | |
811 | } | |
812 | ||
813 | static void remove_full(struct kmem_cache *s, struct page *page) | |
814 | { | |
815 | struct kmem_cache_node *n; | |
816 | ||
817 | if (!(s->flags & SLAB_STORE_USER)) | |
818 | return; | |
819 | ||
820 | n = get_node(s, page_to_nid(page)); | |
821 | ||
822 | spin_lock(&n->list_lock); | |
823 | list_del(&page->lru); | |
824 | spin_unlock(&n->list_lock); | |
825 | } | |
826 | ||
0f389ec6 CL |
827 | /* Tracking of the number of slabs for debugging purposes */ |
828 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
829 | { | |
830 | struct kmem_cache_node *n = get_node(s, node); | |
831 | ||
832 | return atomic_long_read(&n->nr_slabs); | |
833 | } | |
834 | ||
205ab99d | 835 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
836 | { |
837 | struct kmem_cache_node *n = get_node(s, node); | |
838 | ||
839 | /* | |
840 | * May be called early in order to allocate a slab for the | |
841 | * kmem_cache_node structure. Solve the chicken-egg | |
842 | * dilemma by deferring the increment of the count during | |
843 | * bootstrap (see early_kmem_cache_node_alloc). | |
844 | */ | |
205ab99d | 845 | if (!NUMA_BUILD || n) { |
0f389ec6 | 846 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
847 | atomic_long_add(objects, &n->total_objects); |
848 | } | |
0f389ec6 | 849 | } |
205ab99d | 850 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
851 | { |
852 | struct kmem_cache_node *n = get_node(s, node); | |
853 | ||
854 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 855 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
856 | } |
857 | ||
858 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
859 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
860 | void *object) | |
861 | { | |
862 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
863 | return; | |
864 | ||
865 | init_object(s, object, 0); | |
866 | init_tracking(s, object); | |
867 | } | |
868 | ||
869 | static int alloc_debug_processing(struct kmem_cache *s, struct page *page, | |
ce71e27c | 870 | void *object, unsigned long addr) |
81819f0f CL |
871 | { |
872 | if (!check_slab(s, page)) | |
873 | goto bad; | |
874 | ||
d692ef6d | 875 | if (!on_freelist(s, page, object)) { |
24922684 | 876 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 877 | goto bad; |
81819f0f CL |
878 | } |
879 | ||
880 | if (!check_valid_pointer(s, page, object)) { | |
881 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 882 | goto bad; |
81819f0f CL |
883 | } |
884 | ||
d692ef6d | 885 | if (!check_object(s, page, object, 0)) |
81819f0f | 886 | goto bad; |
81819f0f | 887 | |
3ec09742 CL |
888 | /* Success perform special debug activities for allocs */ |
889 | if (s->flags & SLAB_STORE_USER) | |
890 | set_track(s, object, TRACK_ALLOC, addr); | |
891 | trace(s, page, object, 1); | |
892 | init_object(s, object, 1); | |
81819f0f | 893 | return 1; |
3ec09742 | 894 | |
81819f0f CL |
895 | bad: |
896 | if (PageSlab(page)) { | |
897 | /* | |
898 | * If this is a slab page then lets do the best we can | |
899 | * to avoid issues in the future. Marking all objects | |
672bba3a | 900 | * as used avoids touching the remaining objects. |
81819f0f | 901 | */ |
24922684 | 902 | slab_fix(s, "Marking all objects used"); |
39b26464 | 903 | page->inuse = page->objects; |
a973e9dd | 904 | page->freelist = NULL; |
81819f0f CL |
905 | } |
906 | return 0; | |
907 | } | |
908 | ||
3ec09742 | 909 | static int free_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 910 | void *object, unsigned long addr) |
81819f0f CL |
911 | { |
912 | if (!check_slab(s, page)) | |
913 | goto fail; | |
914 | ||
915 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 916 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
917 | goto fail; |
918 | } | |
919 | ||
920 | if (on_freelist(s, page, object)) { | |
24922684 | 921 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
922 | goto fail; |
923 | } | |
924 | ||
925 | if (!check_object(s, page, object, 1)) | |
926 | return 0; | |
927 | ||
928 | if (unlikely(s != page->slab)) { | |
3adbefee | 929 | if (!PageSlab(page)) { |
70d71228 CL |
930 | slab_err(s, page, "Attempt to free object(0x%p) " |
931 | "outside of slab", object); | |
3adbefee | 932 | } else if (!page->slab) { |
81819f0f | 933 | printk(KERN_ERR |
70d71228 | 934 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 935 | object); |
70d71228 | 936 | dump_stack(); |
06428780 | 937 | } else |
24922684 CL |
938 | object_err(s, page, object, |
939 | "page slab pointer corrupt."); | |
81819f0f CL |
940 | goto fail; |
941 | } | |
3ec09742 CL |
942 | |
943 | /* Special debug activities for freeing objects */ | |
8a38082d | 944 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
945 | remove_full(s, page); |
946 | if (s->flags & SLAB_STORE_USER) | |
947 | set_track(s, object, TRACK_FREE, addr); | |
948 | trace(s, page, object, 0); | |
949 | init_object(s, object, 0); | |
81819f0f | 950 | return 1; |
3ec09742 | 951 | |
81819f0f | 952 | fail: |
24922684 | 953 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
954 | return 0; |
955 | } | |
956 | ||
41ecc55b CL |
957 | static int __init setup_slub_debug(char *str) |
958 | { | |
f0630fff CL |
959 | slub_debug = DEBUG_DEFAULT_FLAGS; |
960 | if (*str++ != '=' || !*str) | |
961 | /* | |
962 | * No options specified. Switch on full debugging. | |
963 | */ | |
964 | goto out; | |
965 | ||
966 | if (*str == ',') | |
967 | /* | |
968 | * No options but restriction on slabs. This means full | |
969 | * debugging for slabs matching a pattern. | |
970 | */ | |
971 | goto check_slabs; | |
972 | ||
973 | slub_debug = 0; | |
974 | if (*str == '-') | |
975 | /* | |
976 | * Switch off all debugging measures. | |
977 | */ | |
978 | goto out; | |
979 | ||
980 | /* | |
981 | * Determine which debug features should be switched on | |
982 | */ | |
06428780 | 983 | for (; *str && *str != ','; str++) { |
f0630fff CL |
984 | switch (tolower(*str)) { |
985 | case 'f': | |
986 | slub_debug |= SLAB_DEBUG_FREE; | |
987 | break; | |
988 | case 'z': | |
989 | slub_debug |= SLAB_RED_ZONE; | |
990 | break; | |
991 | case 'p': | |
992 | slub_debug |= SLAB_POISON; | |
993 | break; | |
994 | case 'u': | |
995 | slub_debug |= SLAB_STORE_USER; | |
996 | break; | |
997 | case 't': | |
998 | slub_debug |= SLAB_TRACE; | |
999 | break; | |
1000 | default: | |
1001 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1002 | "unknown. skipped\n", *str); |
f0630fff | 1003 | } |
41ecc55b CL |
1004 | } |
1005 | ||
f0630fff | 1006 | check_slabs: |
41ecc55b CL |
1007 | if (*str == ',') |
1008 | slub_debug_slabs = str + 1; | |
f0630fff | 1009 | out: |
41ecc55b CL |
1010 | return 1; |
1011 | } | |
1012 | ||
1013 | __setup("slub_debug", setup_slub_debug); | |
1014 | ||
ba0268a8 CL |
1015 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1016 | unsigned long flags, const char *name, | |
51cc5068 | 1017 | void (*ctor)(void *)) |
41ecc55b CL |
1018 | { |
1019 | /* | |
e153362a | 1020 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1021 | */ |
e153362a CL |
1022 | if (slub_debug && (!slub_debug_slabs || |
1023 | strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0)) | |
1024 | flags |= slub_debug; | |
ba0268a8 CL |
1025 | |
1026 | return flags; | |
41ecc55b CL |
1027 | } |
1028 | #else | |
3ec09742 CL |
1029 | static inline void setup_object_debug(struct kmem_cache *s, |
1030 | struct page *page, void *object) {} | |
41ecc55b | 1031 | |
3ec09742 | 1032 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1033 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1034 | |
3ec09742 | 1035 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1036 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1037 | |
41ecc55b CL |
1038 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1039 | { return 1; } | |
1040 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
1041 | void *object, int active) { return 1; } | |
3ec09742 | 1042 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1043 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1044 | unsigned long flags, const char *name, | |
51cc5068 | 1045 | void (*ctor)(void *)) |
ba0268a8 CL |
1046 | { |
1047 | return flags; | |
1048 | } | |
41ecc55b | 1049 | #define slub_debug 0 |
0f389ec6 CL |
1050 | |
1051 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1052 | { return 0; } | |
205ab99d CL |
1053 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1054 | int objects) {} | |
1055 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1056 | int objects) {} | |
41ecc55b | 1057 | #endif |
205ab99d | 1058 | |
81819f0f CL |
1059 | /* |
1060 | * Slab allocation and freeing | |
1061 | */ | |
65c3376a CL |
1062 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1063 | struct kmem_cache_order_objects oo) | |
1064 | { | |
1065 | int order = oo_order(oo); | |
1066 | ||
1067 | if (node == -1) | |
1068 | return alloc_pages(flags, order); | |
1069 | else | |
1070 | return alloc_pages_node(node, flags, order); | |
1071 | } | |
1072 | ||
81819f0f CL |
1073 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1074 | { | |
06428780 | 1075 | struct page *page; |
834f3d11 | 1076 | struct kmem_cache_order_objects oo = s->oo; |
81819f0f | 1077 | |
b7a49f0d | 1078 | flags |= s->allocflags; |
e12ba74d | 1079 | |
65c3376a CL |
1080 | page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node, |
1081 | oo); | |
1082 | if (unlikely(!page)) { | |
1083 | oo = s->min; | |
1084 | /* | |
1085 | * Allocation may have failed due to fragmentation. | |
1086 | * Try a lower order alloc if possible | |
1087 | */ | |
1088 | page = alloc_slab_page(flags, node, oo); | |
1089 | if (!page) | |
1090 | return NULL; | |
81819f0f | 1091 | |
65c3376a CL |
1092 | stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK); |
1093 | } | |
834f3d11 | 1094 | page->objects = oo_objects(oo); |
81819f0f CL |
1095 | mod_zone_page_state(page_zone(page), |
1096 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1097 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1098 | 1 << oo_order(oo)); |
81819f0f CL |
1099 | |
1100 | return page; | |
1101 | } | |
1102 | ||
1103 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1104 | void *object) | |
1105 | { | |
3ec09742 | 1106 | setup_object_debug(s, page, object); |
4f104934 | 1107 | if (unlikely(s->ctor)) |
51cc5068 | 1108 | s->ctor(object); |
81819f0f CL |
1109 | } |
1110 | ||
1111 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1112 | { | |
1113 | struct page *page; | |
81819f0f | 1114 | void *start; |
81819f0f CL |
1115 | void *last; |
1116 | void *p; | |
1117 | ||
6cb06229 | 1118 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1119 | |
6cb06229 CL |
1120 | page = allocate_slab(s, |
1121 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1122 | if (!page) |
1123 | goto out; | |
1124 | ||
205ab99d | 1125 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1126 | page->slab = s; |
1127 | page->flags |= 1 << PG_slab; | |
1128 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
1129 | SLAB_STORE_USER | SLAB_TRACE)) | |
8a38082d | 1130 | __SetPageSlubDebug(page); |
81819f0f CL |
1131 | |
1132 | start = page_address(page); | |
81819f0f CL |
1133 | |
1134 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1135 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1136 | |
1137 | last = start; | |
224a88be | 1138 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1139 | setup_object(s, page, last); |
1140 | set_freepointer(s, last, p); | |
1141 | last = p; | |
1142 | } | |
1143 | setup_object(s, page, last); | |
a973e9dd | 1144 | set_freepointer(s, last, NULL); |
81819f0f CL |
1145 | |
1146 | page->freelist = start; | |
1147 | page->inuse = 0; | |
1148 | out: | |
81819f0f CL |
1149 | return page; |
1150 | } | |
1151 | ||
1152 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1153 | { | |
834f3d11 CL |
1154 | int order = compound_order(page); |
1155 | int pages = 1 << order; | |
81819f0f | 1156 | |
8a38082d | 1157 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) { |
81819f0f CL |
1158 | void *p; |
1159 | ||
1160 | slab_pad_check(s, page); | |
224a88be CL |
1161 | for_each_object(p, s, page_address(page), |
1162 | page->objects) | |
81819f0f | 1163 | check_object(s, page, p, 0); |
8a38082d | 1164 | __ClearPageSlubDebug(page); |
81819f0f CL |
1165 | } |
1166 | ||
1167 | mod_zone_page_state(page_zone(page), | |
1168 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1169 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1170 | -pages); |
81819f0f | 1171 | |
49bd5221 CL |
1172 | __ClearPageSlab(page); |
1173 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1174 | if (current->reclaim_state) |
1175 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1176 | __free_pages(page, order); |
81819f0f CL |
1177 | } |
1178 | ||
1179 | static void rcu_free_slab(struct rcu_head *h) | |
1180 | { | |
1181 | struct page *page; | |
1182 | ||
1183 | page = container_of((struct list_head *)h, struct page, lru); | |
1184 | __free_slab(page->slab, page); | |
1185 | } | |
1186 | ||
1187 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1188 | { | |
1189 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
1190 | /* | |
1191 | * RCU free overloads the RCU head over the LRU | |
1192 | */ | |
1193 | struct rcu_head *head = (void *)&page->lru; | |
1194 | ||
1195 | call_rcu(head, rcu_free_slab); | |
1196 | } else | |
1197 | __free_slab(s, page); | |
1198 | } | |
1199 | ||
1200 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1201 | { | |
205ab99d | 1202 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1203 | free_slab(s, page); |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * Per slab locking using the pagelock | |
1208 | */ | |
1209 | static __always_inline void slab_lock(struct page *page) | |
1210 | { | |
1211 | bit_spin_lock(PG_locked, &page->flags); | |
1212 | } | |
1213 | ||
1214 | static __always_inline void slab_unlock(struct page *page) | |
1215 | { | |
a76d3546 | 1216 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1217 | } |
1218 | ||
1219 | static __always_inline int slab_trylock(struct page *page) | |
1220 | { | |
1221 | int rc = 1; | |
1222 | ||
1223 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1224 | return rc; | |
1225 | } | |
1226 | ||
1227 | /* | |
1228 | * Management of partially allocated slabs | |
1229 | */ | |
7c2e132c CL |
1230 | static void add_partial(struct kmem_cache_node *n, |
1231 | struct page *page, int tail) | |
81819f0f | 1232 | { |
e95eed57 CL |
1233 | spin_lock(&n->list_lock); |
1234 | n->nr_partial++; | |
7c2e132c CL |
1235 | if (tail) |
1236 | list_add_tail(&page->lru, &n->partial); | |
1237 | else | |
1238 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1239 | spin_unlock(&n->list_lock); |
1240 | } | |
1241 | ||
0121c619 | 1242 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1243 | { |
1244 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1245 | ||
1246 | spin_lock(&n->list_lock); | |
1247 | list_del(&page->lru); | |
1248 | n->nr_partial--; | |
1249 | spin_unlock(&n->list_lock); | |
1250 | } | |
1251 | ||
1252 | /* | |
672bba3a | 1253 | * Lock slab and remove from the partial list. |
81819f0f | 1254 | * |
672bba3a | 1255 | * Must hold list_lock. |
81819f0f | 1256 | */ |
0121c619 CL |
1257 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1258 | struct page *page) | |
81819f0f CL |
1259 | { |
1260 | if (slab_trylock(page)) { | |
1261 | list_del(&page->lru); | |
1262 | n->nr_partial--; | |
8a38082d | 1263 | __SetPageSlubFrozen(page); |
81819f0f CL |
1264 | return 1; |
1265 | } | |
1266 | return 0; | |
1267 | } | |
1268 | ||
1269 | /* | |
672bba3a | 1270 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1271 | */ |
1272 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1273 | { | |
1274 | struct page *page; | |
1275 | ||
1276 | /* | |
1277 | * Racy check. If we mistakenly see no partial slabs then we | |
1278 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1279 | * partial slab and there is none available then get_partials() |
1280 | * will return NULL. | |
81819f0f CL |
1281 | */ |
1282 | if (!n || !n->nr_partial) | |
1283 | return NULL; | |
1284 | ||
1285 | spin_lock(&n->list_lock); | |
1286 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1287 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1288 | goto out; |
1289 | page = NULL; | |
1290 | out: | |
1291 | spin_unlock(&n->list_lock); | |
1292 | return page; | |
1293 | } | |
1294 | ||
1295 | /* | |
672bba3a | 1296 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1297 | */ |
1298 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1299 | { | |
1300 | #ifdef CONFIG_NUMA | |
1301 | struct zonelist *zonelist; | |
dd1a239f | 1302 | struct zoneref *z; |
54a6eb5c MG |
1303 | struct zone *zone; |
1304 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1305 | struct page *page; |
1306 | ||
1307 | /* | |
672bba3a CL |
1308 | * The defrag ratio allows a configuration of the tradeoffs between |
1309 | * inter node defragmentation and node local allocations. A lower | |
1310 | * defrag_ratio increases the tendency to do local allocations | |
1311 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1312 | * |
672bba3a CL |
1313 | * If the defrag_ratio is set to 0 then kmalloc() always |
1314 | * returns node local objects. If the ratio is higher then kmalloc() | |
1315 | * may return off node objects because partial slabs are obtained | |
1316 | * from other nodes and filled up. | |
81819f0f | 1317 | * |
6446faa2 | 1318 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1319 | * defrag_ratio = 1000) then every (well almost) allocation will |
1320 | * first attempt to defrag slab caches on other nodes. This means | |
1321 | * scanning over all nodes to look for partial slabs which may be | |
1322 | * expensive if we do it every time we are trying to find a slab | |
1323 | * with available objects. | |
81819f0f | 1324 | */ |
9824601e CL |
1325 | if (!s->remote_node_defrag_ratio || |
1326 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1327 | return NULL; |
1328 | ||
0e88460d | 1329 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1330 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1331 | struct kmem_cache_node *n; |
1332 | ||
54a6eb5c | 1333 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1334 | |
54a6eb5c | 1335 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1336 | n->nr_partial > s->min_partial) { |
81819f0f CL |
1337 | page = get_partial_node(n); |
1338 | if (page) | |
1339 | return page; | |
1340 | } | |
1341 | } | |
1342 | #endif | |
1343 | return NULL; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Get a partial page, lock it and return it. | |
1348 | */ | |
1349 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1350 | { | |
1351 | struct page *page; | |
1352 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1353 | ||
1354 | page = get_partial_node(get_node(s, searchnode)); | |
1355 | if (page || (flags & __GFP_THISNODE)) | |
1356 | return page; | |
1357 | ||
1358 | return get_any_partial(s, flags); | |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * Move a page back to the lists. | |
1363 | * | |
1364 | * Must be called with the slab lock held. | |
1365 | * | |
1366 | * On exit the slab lock will have been dropped. | |
1367 | */ | |
7c2e132c | 1368 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
81819f0f | 1369 | { |
e95eed57 | 1370 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
8ff12cfc | 1371 | struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id()); |
e95eed57 | 1372 | |
8a38082d | 1373 | __ClearPageSlubFrozen(page); |
81819f0f | 1374 | if (page->inuse) { |
e95eed57 | 1375 | |
a973e9dd | 1376 | if (page->freelist) { |
7c2e132c | 1377 | add_partial(n, page, tail); |
8ff12cfc CL |
1378 | stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
1379 | } else { | |
1380 | stat(c, DEACTIVATE_FULL); | |
8a38082d AW |
1381 | if (SLABDEBUG && PageSlubDebug(page) && |
1382 | (s->flags & SLAB_STORE_USER)) | |
8ff12cfc CL |
1383 | add_full(n, page); |
1384 | } | |
81819f0f CL |
1385 | slab_unlock(page); |
1386 | } else { | |
8ff12cfc | 1387 | stat(c, DEACTIVATE_EMPTY); |
3b89d7d8 | 1388 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1389 | /* |
672bba3a CL |
1390 | * Adding an empty slab to the partial slabs in order |
1391 | * to avoid page allocator overhead. This slab needs | |
1392 | * to come after the other slabs with objects in | |
6446faa2 CL |
1393 | * so that the others get filled first. That way the |
1394 | * size of the partial list stays small. | |
1395 | * | |
0121c619 CL |
1396 | * kmem_cache_shrink can reclaim any empty slabs from |
1397 | * the partial list. | |
e95eed57 | 1398 | */ |
7c2e132c | 1399 | add_partial(n, page, 1); |
e95eed57 CL |
1400 | slab_unlock(page); |
1401 | } else { | |
1402 | slab_unlock(page); | |
8ff12cfc | 1403 | stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB); |
e95eed57 CL |
1404 | discard_slab(s, page); |
1405 | } | |
81819f0f CL |
1406 | } |
1407 | } | |
1408 | ||
1409 | /* | |
1410 | * Remove the cpu slab | |
1411 | */ | |
dfb4f096 | 1412 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1413 | { |
dfb4f096 | 1414 | struct page *page = c->page; |
7c2e132c | 1415 | int tail = 1; |
8ff12cfc | 1416 | |
b773ad73 | 1417 | if (page->freelist) |
8ff12cfc | 1418 | stat(c, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1419 | /* |
6446faa2 | 1420 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1421 | * because both freelists are empty. So this is unlikely |
1422 | * to occur. | |
1423 | */ | |
a973e9dd | 1424 | while (unlikely(c->freelist)) { |
894b8788 CL |
1425 | void **object; |
1426 | ||
7c2e132c CL |
1427 | tail = 0; /* Hot objects. Put the slab first */ |
1428 | ||
894b8788 | 1429 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1430 | object = c->freelist; |
b3fba8da | 1431 | c->freelist = c->freelist[c->offset]; |
894b8788 CL |
1432 | |
1433 | /* And put onto the regular freelist */ | |
b3fba8da | 1434 | object[c->offset] = page->freelist; |
894b8788 CL |
1435 | page->freelist = object; |
1436 | page->inuse--; | |
1437 | } | |
dfb4f096 | 1438 | c->page = NULL; |
7c2e132c | 1439 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1440 | } |
1441 | ||
dfb4f096 | 1442 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1443 | { |
8ff12cfc | 1444 | stat(c, CPUSLAB_FLUSH); |
dfb4f096 CL |
1445 | slab_lock(c->page); |
1446 | deactivate_slab(s, c); | |
81819f0f CL |
1447 | } |
1448 | ||
1449 | /* | |
1450 | * Flush cpu slab. | |
6446faa2 | 1451 | * |
81819f0f CL |
1452 | * Called from IPI handler with interrupts disabled. |
1453 | */ | |
0c710013 | 1454 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1455 | { |
dfb4f096 | 1456 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
81819f0f | 1457 | |
dfb4f096 CL |
1458 | if (likely(c && c->page)) |
1459 | flush_slab(s, c); | |
81819f0f CL |
1460 | } |
1461 | ||
1462 | static void flush_cpu_slab(void *d) | |
1463 | { | |
1464 | struct kmem_cache *s = d; | |
81819f0f | 1465 | |
dfb4f096 | 1466 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1467 | } |
1468 | ||
1469 | static void flush_all(struct kmem_cache *s) | |
1470 | { | |
15c8b6c1 | 1471 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1472 | } |
1473 | ||
dfb4f096 CL |
1474 | /* |
1475 | * Check if the objects in a per cpu structure fit numa | |
1476 | * locality expectations. | |
1477 | */ | |
1478 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1479 | { | |
1480 | #ifdef CONFIG_NUMA | |
1481 | if (node != -1 && c->node != node) | |
1482 | return 0; | |
1483 | #endif | |
1484 | return 1; | |
1485 | } | |
1486 | ||
81819f0f | 1487 | /* |
894b8788 CL |
1488 | * Slow path. The lockless freelist is empty or we need to perform |
1489 | * debugging duties. | |
1490 | * | |
1491 | * Interrupts are disabled. | |
81819f0f | 1492 | * |
894b8788 CL |
1493 | * Processing is still very fast if new objects have been freed to the |
1494 | * regular freelist. In that case we simply take over the regular freelist | |
1495 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1496 | * |
894b8788 CL |
1497 | * If that is not working then we fall back to the partial lists. We take the |
1498 | * first element of the freelist as the object to allocate now and move the | |
1499 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1500 | * |
894b8788 | 1501 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1502 | * we need to allocate a new slab. This is the slowest path since it involves |
1503 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1504 | */ |
ce71e27c EGM |
1505 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1506 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1507 | { |
81819f0f | 1508 | void **object; |
dfb4f096 | 1509 | struct page *new; |
81819f0f | 1510 | |
e72e9c23 LT |
1511 | /* We handle __GFP_ZERO in the caller */ |
1512 | gfpflags &= ~__GFP_ZERO; | |
1513 | ||
dfb4f096 | 1514 | if (!c->page) |
81819f0f CL |
1515 | goto new_slab; |
1516 | ||
dfb4f096 CL |
1517 | slab_lock(c->page); |
1518 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1519 | goto another_slab; |
6446faa2 | 1520 | |
8ff12cfc | 1521 | stat(c, ALLOC_REFILL); |
6446faa2 | 1522 | |
894b8788 | 1523 | load_freelist: |
dfb4f096 | 1524 | object = c->page->freelist; |
a973e9dd | 1525 | if (unlikely(!object)) |
81819f0f | 1526 | goto another_slab; |
8a38082d | 1527 | if (unlikely(SLABDEBUG && PageSlubDebug(c->page))) |
81819f0f CL |
1528 | goto debug; |
1529 | ||
b3fba8da | 1530 | c->freelist = object[c->offset]; |
39b26464 | 1531 | c->page->inuse = c->page->objects; |
a973e9dd | 1532 | c->page->freelist = NULL; |
dfb4f096 | 1533 | c->node = page_to_nid(c->page); |
1f84260c | 1534 | unlock_out: |
dfb4f096 | 1535 | slab_unlock(c->page); |
8ff12cfc | 1536 | stat(c, ALLOC_SLOWPATH); |
81819f0f CL |
1537 | return object; |
1538 | ||
1539 | another_slab: | |
dfb4f096 | 1540 | deactivate_slab(s, c); |
81819f0f CL |
1541 | |
1542 | new_slab: | |
dfb4f096 CL |
1543 | new = get_partial(s, gfpflags, node); |
1544 | if (new) { | |
1545 | c->page = new; | |
8ff12cfc | 1546 | stat(c, ALLOC_FROM_PARTIAL); |
894b8788 | 1547 | goto load_freelist; |
81819f0f CL |
1548 | } |
1549 | ||
b811c202 CL |
1550 | if (gfpflags & __GFP_WAIT) |
1551 | local_irq_enable(); | |
1552 | ||
dfb4f096 | 1553 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1554 | |
1555 | if (gfpflags & __GFP_WAIT) | |
1556 | local_irq_disable(); | |
1557 | ||
dfb4f096 CL |
1558 | if (new) { |
1559 | c = get_cpu_slab(s, smp_processor_id()); | |
8ff12cfc | 1560 | stat(c, ALLOC_SLAB); |
05aa3450 | 1561 | if (c->page) |
dfb4f096 | 1562 | flush_slab(s, c); |
dfb4f096 | 1563 | slab_lock(new); |
8a38082d | 1564 | __SetPageSlubFrozen(new); |
dfb4f096 | 1565 | c->page = new; |
4b6f0750 | 1566 | goto load_freelist; |
81819f0f | 1567 | } |
71c7a06f | 1568 | return NULL; |
81819f0f | 1569 | debug: |
dfb4f096 | 1570 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1571 | goto another_slab; |
894b8788 | 1572 | |
dfb4f096 | 1573 | c->page->inuse++; |
b3fba8da | 1574 | c->page->freelist = object[c->offset]; |
ee3c72a1 | 1575 | c->node = -1; |
1f84260c | 1576 | goto unlock_out; |
894b8788 CL |
1577 | } |
1578 | ||
1579 | /* | |
1580 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1581 | * have the fastpath folded into their functions. So no function call | |
1582 | * overhead for requests that can be satisfied on the fastpath. | |
1583 | * | |
1584 | * The fastpath works by first checking if the lockless freelist can be used. | |
1585 | * If not then __slab_alloc is called for slow processing. | |
1586 | * | |
1587 | * Otherwise we can simply pick the next object from the lockless free list. | |
1588 | */ | |
06428780 | 1589 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1590 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1591 | { |
894b8788 | 1592 | void **object; |
dfb4f096 | 1593 | struct kmem_cache_cpu *c; |
1f84260c | 1594 | unsigned long flags; |
bdb21928 | 1595 | unsigned int objsize; |
1f84260c | 1596 | |
cf40bd16 | 1597 | lockdep_trace_alloc(gfpflags); |
89124d70 | 1598 | might_sleep_if(gfpflags & __GFP_WAIT); |
3c506efd | 1599 | |
773ff60e AM |
1600 | if (should_failslab(s->objsize, gfpflags)) |
1601 | return NULL; | |
1f84260c | 1602 | |
894b8788 | 1603 | local_irq_save(flags); |
dfb4f096 | 1604 | c = get_cpu_slab(s, smp_processor_id()); |
bdb21928 | 1605 | objsize = c->objsize; |
a973e9dd | 1606 | if (unlikely(!c->freelist || !node_match(c, node))) |
894b8788 | 1607 | |
dfb4f096 | 1608 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1609 | |
1610 | else { | |
dfb4f096 | 1611 | object = c->freelist; |
b3fba8da | 1612 | c->freelist = object[c->offset]; |
8ff12cfc | 1613 | stat(c, ALLOC_FASTPATH); |
894b8788 CL |
1614 | } |
1615 | local_irq_restore(flags); | |
d07dbea4 CL |
1616 | |
1617 | if (unlikely((gfpflags & __GFP_ZERO) && object)) | |
bdb21928 | 1618 | memset(object, 0, objsize); |
d07dbea4 | 1619 | |
894b8788 | 1620 | return object; |
81819f0f CL |
1621 | } |
1622 | ||
1623 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1624 | { | |
5b882be4 EGM |
1625 | void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_); |
1626 | ||
ca2b84cb | 1627 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1628 | |
1629 | return ret; | |
81819f0f CL |
1630 | } |
1631 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1632 | ||
5b882be4 EGM |
1633 | #ifdef CONFIG_KMEMTRACE |
1634 | void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags) | |
1635 | { | |
1636 | return slab_alloc(s, gfpflags, -1, _RET_IP_); | |
1637 | } | |
1638 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
1639 | #endif | |
1640 | ||
81819f0f CL |
1641 | #ifdef CONFIG_NUMA |
1642 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1643 | { | |
5b882be4 EGM |
1644 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
1645 | ||
ca2b84cb EGM |
1646 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
1647 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
1648 | |
1649 | return ret; | |
81819f0f CL |
1650 | } |
1651 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1652 | #endif | |
1653 | ||
5b882be4 EGM |
1654 | #ifdef CONFIG_KMEMTRACE |
1655 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *s, | |
1656 | gfp_t gfpflags, | |
1657 | int node) | |
1658 | { | |
1659 | return slab_alloc(s, gfpflags, node, _RET_IP_); | |
1660 | } | |
1661 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
1662 | #endif | |
1663 | ||
81819f0f | 1664 | /* |
894b8788 CL |
1665 | * Slow patch handling. This may still be called frequently since objects |
1666 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 1667 | * |
894b8788 CL |
1668 | * So we still attempt to reduce cache line usage. Just take the slab |
1669 | * lock and free the item. If there is no additional partial page | |
1670 | * handling required then we can return immediately. | |
81819f0f | 1671 | */ |
894b8788 | 1672 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ce71e27c | 1673 | void *x, unsigned long addr, unsigned int offset) |
81819f0f CL |
1674 | { |
1675 | void *prior; | |
1676 | void **object = (void *)x; | |
8ff12cfc | 1677 | struct kmem_cache_cpu *c; |
81819f0f | 1678 | |
8ff12cfc CL |
1679 | c = get_cpu_slab(s, raw_smp_processor_id()); |
1680 | stat(c, FREE_SLOWPATH); | |
81819f0f CL |
1681 | slab_lock(page); |
1682 | ||
8a38082d | 1683 | if (unlikely(SLABDEBUG && PageSlubDebug(page))) |
81819f0f | 1684 | goto debug; |
6446faa2 | 1685 | |
81819f0f | 1686 | checks_ok: |
b3fba8da | 1687 | prior = object[offset] = page->freelist; |
81819f0f CL |
1688 | page->freelist = object; |
1689 | page->inuse--; | |
1690 | ||
8a38082d | 1691 | if (unlikely(PageSlubFrozen(page))) { |
8ff12cfc | 1692 | stat(c, FREE_FROZEN); |
81819f0f | 1693 | goto out_unlock; |
8ff12cfc | 1694 | } |
81819f0f CL |
1695 | |
1696 | if (unlikely(!page->inuse)) | |
1697 | goto slab_empty; | |
1698 | ||
1699 | /* | |
6446faa2 | 1700 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
1701 | * then add it. |
1702 | */ | |
a973e9dd | 1703 | if (unlikely(!prior)) { |
7c2e132c | 1704 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
8ff12cfc CL |
1705 | stat(c, FREE_ADD_PARTIAL); |
1706 | } | |
81819f0f CL |
1707 | |
1708 | out_unlock: | |
1709 | slab_unlock(page); | |
81819f0f CL |
1710 | return; |
1711 | ||
1712 | slab_empty: | |
a973e9dd | 1713 | if (prior) { |
81819f0f | 1714 | /* |
672bba3a | 1715 | * Slab still on the partial list. |
81819f0f CL |
1716 | */ |
1717 | remove_partial(s, page); | |
8ff12cfc CL |
1718 | stat(c, FREE_REMOVE_PARTIAL); |
1719 | } | |
81819f0f | 1720 | slab_unlock(page); |
8ff12cfc | 1721 | stat(c, FREE_SLAB); |
81819f0f | 1722 | discard_slab(s, page); |
81819f0f CL |
1723 | return; |
1724 | ||
1725 | debug: | |
3ec09742 | 1726 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 1727 | goto out_unlock; |
77c5e2d0 | 1728 | goto checks_ok; |
81819f0f CL |
1729 | } |
1730 | ||
894b8788 CL |
1731 | /* |
1732 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
1733 | * can perform fastpath freeing without additional function calls. | |
1734 | * | |
1735 | * The fastpath is only possible if we are freeing to the current cpu slab | |
1736 | * of this processor. This typically the case if we have just allocated | |
1737 | * the item before. | |
1738 | * | |
1739 | * If fastpath is not possible then fall back to __slab_free where we deal | |
1740 | * with all sorts of special processing. | |
1741 | */ | |
06428780 | 1742 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 1743 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
1744 | { |
1745 | void **object = (void *)x; | |
dfb4f096 | 1746 | struct kmem_cache_cpu *c; |
1f84260c CL |
1747 | unsigned long flags; |
1748 | ||
894b8788 | 1749 | local_irq_save(flags); |
dfb4f096 | 1750 | c = get_cpu_slab(s, smp_processor_id()); |
27d9e4e9 | 1751 | debug_check_no_locks_freed(object, c->objsize); |
3ac7fe5a | 1752 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
6047a007 | 1753 | debug_check_no_obj_freed(object, c->objsize); |
ee3c72a1 | 1754 | if (likely(page == c->page && c->node >= 0)) { |
b3fba8da | 1755 | object[c->offset] = c->freelist; |
dfb4f096 | 1756 | c->freelist = object; |
8ff12cfc | 1757 | stat(c, FREE_FASTPATH); |
894b8788 | 1758 | } else |
b3fba8da | 1759 | __slab_free(s, page, x, addr, c->offset); |
894b8788 CL |
1760 | |
1761 | local_irq_restore(flags); | |
1762 | } | |
1763 | ||
81819f0f CL |
1764 | void kmem_cache_free(struct kmem_cache *s, void *x) |
1765 | { | |
77c5e2d0 | 1766 | struct page *page; |
81819f0f | 1767 | |
b49af68f | 1768 | page = virt_to_head_page(x); |
81819f0f | 1769 | |
ce71e27c | 1770 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 1771 | |
ca2b84cb | 1772 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
1773 | } |
1774 | EXPORT_SYMBOL(kmem_cache_free); | |
1775 | ||
e9beef18 | 1776 | /* Figure out on which slab page the object resides */ |
81819f0f CL |
1777 | static struct page *get_object_page(const void *x) |
1778 | { | |
b49af68f | 1779 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1780 | |
1781 | if (!PageSlab(page)) | |
1782 | return NULL; | |
1783 | ||
1784 | return page; | |
1785 | } | |
1786 | ||
1787 | /* | |
672bba3a CL |
1788 | * Object placement in a slab is made very easy because we always start at |
1789 | * offset 0. If we tune the size of the object to the alignment then we can | |
1790 | * get the required alignment by putting one properly sized object after | |
1791 | * another. | |
81819f0f CL |
1792 | * |
1793 | * Notice that the allocation order determines the sizes of the per cpu | |
1794 | * caches. Each processor has always one slab available for allocations. | |
1795 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 1796 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 1797 | * locking overhead. |
81819f0f CL |
1798 | */ |
1799 | ||
1800 | /* | |
1801 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1802 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1803 | * and increases the number of allocations possible without having to | |
1804 | * take the list_lock. | |
1805 | */ | |
1806 | static int slub_min_order; | |
114e9e89 | 1807 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 1808 | static int slub_min_objects; |
81819f0f CL |
1809 | |
1810 | /* | |
1811 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 1812 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
1813 | */ |
1814 | static int slub_nomerge; | |
1815 | ||
81819f0f CL |
1816 | /* |
1817 | * Calculate the order of allocation given an slab object size. | |
1818 | * | |
672bba3a CL |
1819 | * The order of allocation has significant impact on performance and other |
1820 | * system components. Generally order 0 allocations should be preferred since | |
1821 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
1822 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 1823 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
1824 | * would be wasted. |
1825 | * | |
1826 | * In order to reach satisfactory performance we must ensure that a minimum | |
1827 | * number of objects is in one slab. Otherwise we may generate too much | |
1828 | * activity on the partial lists which requires taking the list_lock. This is | |
1829 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 1830 | * |
672bba3a CL |
1831 | * slub_max_order specifies the order where we begin to stop considering the |
1832 | * number of objects in a slab as critical. If we reach slub_max_order then | |
1833 | * we try to keep the page order as low as possible. So we accept more waste | |
1834 | * of space in favor of a small page order. | |
81819f0f | 1835 | * |
672bba3a CL |
1836 | * Higher order allocations also allow the placement of more objects in a |
1837 | * slab and thereby reduce object handling overhead. If the user has | |
1838 | * requested a higher mininum order then we start with that one instead of | |
1839 | * the smallest order which will fit the object. | |
81819f0f | 1840 | */ |
5e6d444e CL |
1841 | static inline int slab_order(int size, int min_objects, |
1842 | int max_order, int fract_leftover) | |
81819f0f CL |
1843 | { |
1844 | int order; | |
1845 | int rem; | |
6300ea75 | 1846 | int min_order = slub_min_order; |
81819f0f | 1847 | |
210b5c06 CG |
1848 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) |
1849 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
39b26464 | 1850 | |
6300ea75 | 1851 | for (order = max(min_order, |
5e6d444e CL |
1852 | fls(min_objects * size - 1) - PAGE_SHIFT); |
1853 | order <= max_order; order++) { | |
81819f0f | 1854 | |
5e6d444e | 1855 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 1856 | |
5e6d444e | 1857 | if (slab_size < min_objects * size) |
81819f0f CL |
1858 | continue; |
1859 | ||
1860 | rem = slab_size % size; | |
1861 | ||
5e6d444e | 1862 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
1863 | break; |
1864 | ||
1865 | } | |
672bba3a | 1866 | |
81819f0f CL |
1867 | return order; |
1868 | } | |
1869 | ||
5e6d444e CL |
1870 | static inline int calculate_order(int size) |
1871 | { | |
1872 | int order; | |
1873 | int min_objects; | |
1874 | int fraction; | |
e8120ff1 | 1875 | int max_objects; |
5e6d444e CL |
1876 | |
1877 | /* | |
1878 | * Attempt to find best configuration for a slab. This | |
1879 | * works by first attempting to generate a layout with | |
1880 | * the best configuration and backing off gradually. | |
1881 | * | |
1882 | * First we reduce the acceptable waste in a slab. Then | |
1883 | * we reduce the minimum objects required in a slab. | |
1884 | */ | |
1885 | min_objects = slub_min_objects; | |
9b2cd506 CL |
1886 | if (!min_objects) |
1887 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
e8120ff1 ZY |
1888 | max_objects = (PAGE_SIZE << slub_max_order)/size; |
1889 | min_objects = min(min_objects, max_objects); | |
1890 | ||
5e6d444e | 1891 | while (min_objects > 1) { |
c124f5b5 | 1892 | fraction = 16; |
5e6d444e CL |
1893 | while (fraction >= 4) { |
1894 | order = slab_order(size, min_objects, | |
1895 | slub_max_order, fraction); | |
1896 | if (order <= slub_max_order) | |
1897 | return order; | |
1898 | fraction /= 2; | |
1899 | } | |
e8120ff1 | 1900 | min_objects --; |
5e6d444e CL |
1901 | } |
1902 | ||
1903 | /* | |
1904 | * We were unable to place multiple objects in a slab. Now | |
1905 | * lets see if we can place a single object there. | |
1906 | */ | |
1907 | order = slab_order(size, 1, slub_max_order, 1); | |
1908 | if (order <= slub_max_order) | |
1909 | return order; | |
1910 | ||
1911 | /* | |
1912 | * Doh this slab cannot be placed using slub_max_order. | |
1913 | */ | |
1914 | order = slab_order(size, 1, MAX_ORDER, 1); | |
818cf590 | 1915 | if (order < MAX_ORDER) |
5e6d444e CL |
1916 | return order; |
1917 | return -ENOSYS; | |
1918 | } | |
1919 | ||
81819f0f | 1920 | /* |
672bba3a | 1921 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
1922 | */ |
1923 | static unsigned long calculate_alignment(unsigned long flags, | |
1924 | unsigned long align, unsigned long size) | |
1925 | { | |
1926 | /* | |
6446faa2 CL |
1927 | * If the user wants hardware cache aligned objects then follow that |
1928 | * suggestion if the object is sufficiently large. | |
81819f0f | 1929 | * |
6446faa2 CL |
1930 | * The hardware cache alignment cannot override the specified |
1931 | * alignment though. If that is greater then use it. | |
81819f0f | 1932 | */ |
b6210386 NP |
1933 | if (flags & SLAB_HWCACHE_ALIGN) { |
1934 | unsigned long ralign = cache_line_size(); | |
1935 | while (size <= ralign / 2) | |
1936 | ralign /= 2; | |
1937 | align = max(align, ralign); | |
1938 | } | |
81819f0f CL |
1939 | |
1940 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 1941 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
1942 | |
1943 | return ALIGN(align, sizeof(void *)); | |
1944 | } | |
1945 | ||
dfb4f096 CL |
1946 | static void init_kmem_cache_cpu(struct kmem_cache *s, |
1947 | struct kmem_cache_cpu *c) | |
1948 | { | |
1949 | c->page = NULL; | |
a973e9dd | 1950 | c->freelist = NULL; |
dfb4f096 | 1951 | c->node = 0; |
42a9fdbb CL |
1952 | c->offset = s->offset / sizeof(void *); |
1953 | c->objsize = s->objsize; | |
62f75532 PE |
1954 | #ifdef CONFIG_SLUB_STATS |
1955 | memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned)); | |
1956 | #endif | |
dfb4f096 CL |
1957 | } |
1958 | ||
5595cffc PE |
1959 | static void |
1960 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
1961 | { |
1962 | n->nr_partial = 0; | |
81819f0f CL |
1963 | spin_lock_init(&n->list_lock); |
1964 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 1965 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 1966 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 1967 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 1968 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 1969 | #endif |
81819f0f CL |
1970 | } |
1971 | ||
4c93c355 CL |
1972 | #ifdef CONFIG_SMP |
1973 | /* | |
1974 | * Per cpu array for per cpu structures. | |
1975 | * | |
1976 | * The per cpu array places all kmem_cache_cpu structures from one processor | |
1977 | * close together meaning that it becomes possible that multiple per cpu | |
1978 | * structures are contained in one cacheline. This may be particularly | |
1979 | * beneficial for the kmalloc caches. | |
1980 | * | |
1981 | * A desktop system typically has around 60-80 slabs. With 100 here we are | |
1982 | * likely able to get per cpu structures for all caches from the array defined | |
1983 | * here. We must be able to cover all kmalloc caches during bootstrap. | |
1984 | * | |
1985 | * If the per cpu array is exhausted then fall back to kmalloc | |
1986 | * of individual cachelines. No sharing is possible then. | |
1987 | */ | |
1988 | #define NR_KMEM_CACHE_CPU 100 | |
1989 | ||
1990 | static DEFINE_PER_CPU(struct kmem_cache_cpu, | |
1991 | kmem_cache_cpu)[NR_KMEM_CACHE_CPU]; | |
1992 | ||
1993 | static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free); | |
174596a0 | 1994 | static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS); |
4c93c355 CL |
1995 | |
1996 | static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s, | |
1997 | int cpu, gfp_t flags) | |
1998 | { | |
1999 | struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu); | |
2000 | ||
2001 | if (c) | |
2002 | per_cpu(kmem_cache_cpu_free, cpu) = | |
2003 | (void *)c->freelist; | |
2004 | else { | |
2005 | /* Table overflow: So allocate ourselves */ | |
2006 | c = kmalloc_node( | |
2007 | ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()), | |
2008 | flags, cpu_to_node(cpu)); | |
2009 | if (!c) | |
2010 | return NULL; | |
2011 | } | |
2012 | ||
2013 | init_kmem_cache_cpu(s, c); | |
2014 | return c; | |
2015 | } | |
2016 | ||
2017 | static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu) | |
2018 | { | |
2019 | if (c < per_cpu(kmem_cache_cpu, cpu) || | |
37189094 | 2020 | c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) { |
4c93c355 CL |
2021 | kfree(c); |
2022 | return; | |
2023 | } | |
2024 | c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu); | |
2025 | per_cpu(kmem_cache_cpu_free, cpu) = c; | |
2026 | } | |
2027 | ||
2028 | static void free_kmem_cache_cpus(struct kmem_cache *s) | |
2029 | { | |
2030 | int cpu; | |
2031 | ||
2032 | for_each_online_cpu(cpu) { | |
2033 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2034 | ||
2035 | if (c) { | |
2036 | s->cpu_slab[cpu] = NULL; | |
2037 | free_kmem_cache_cpu(c, cpu); | |
2038 | } | |
2039 | } | |
2040 | } | |
2041 | ||
2042 | static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2043 | { | |
2044 | int cpu; | |
2045 | ||
2046 | for_each_online_cpu(cpu) { | |
2047 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
2048 | ||
2049 | if (c) | |
2050 | continue; | |
2051 | ||
2052 | c = alloc_kmem_cache_cpu(s, cpu, flags); | |
2053 | if (!c) { | |
2054 | free_kmem_cache_cpus(s); | |
2055 | return 0; | |
2056 | } | |
2057 | s->cpu_slab[cpu] = c; | |
2058 | } | |
2059 | return 1; | |
2060 | } | |
2061 | ||
2062 | /* | |
2063 | * Initialize the per cpu array. | |
2064 | */ | |
2065 | static void init_alloc_cpu_cpu(int cpu) | |
2066 | { | |
2067 | int i; | |
2068 | ||
174596a0 | 2069 | if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once))) |
4c93c355 CL |
2070 | return; |
2071 | ||
2072 | for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--) | |
2073 | free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu); | |
2074 | ||
174596a0 | 2075 | cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)); |
4c93c355 CL |
2076 | } |
2077 | ||
2078 | static void __init init_alloc_cpu(void) | |
2079 | { | |
2080 | int cpu; | |
2081 | ||
2082 | for_each_online_cpu(cpu) | |
2083 | init_alloc_cpu_cpu(cpu); | |
2084 | } | |
2085 | ||
2086 | #else | |
2087 | static inline void free_kmem_cache_cpus(struct kmem_cache *s) {} | |
2088 | static inline void init_alloc_cpu(void) {} | |
2089 | ||
2090 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags) | |
2091 | { | |
2092 | init_kmem_cache_cpu(s, &s->cpu_slab); | |
2093 | return 1; | |
2094 | } | |
2095 | #endif | |
2096 | ||
81819f0f CL |
2097 | #ifdef CONFIG_NUMA |
2098 | /* | |
2099 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2100 | * slab on the node for this slabcache. There are no concurrent accesses | |
2101 | * possible. | |
2102 | * | |
2103 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2104 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2105 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2106 | */ |
0094de92 | 2107 | static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node) |
81819f0f CL |
2108 | { |
2109 | struct page *page; | |
2110 | struct kmem_cache_node *n; | |
ba84c73c | 2111 | unsigned long flags; |
81819f0f CL |
2112 | |
2113 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
2114 | ||
a2f92ee7 | 2115 | page = new_slab(kmalloc_caches, gfpflags, node); |
81819f0f CL |
2116 | |
2117 | BUG_ON(!page); | |
a2f92ee7 CL |
2118 | if (page_to_nid(page) != node) { |
2119 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2120 | "node %d\n", node); | |
2121 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2122 | "in order to be able to continue\n"); | |
2123 | } | |
2124 | ||
81819f0f CL |
2125 | n = page->freelist; |
2126 | BUG_ON(!n); | |
2127 | page->freelist = get_freepointer(kmalloc_caches, n); | |
2128 | page->inuse++; | |
2129 | kmalloc_caches->node[node] = n; | |
8ab1372f | 2130 | #ifdef CONFIG_SLUB_DEBUG |
d45f39cb CL |
2131 | init_object(kmalloc_caches, n, 1); |
2132 | init_tracking(kmalloc_caches, n); | |
8ab1372f | 2133 | #endif |
5595cffc | 2134 | init_kmem_cache_node(n, kmalloc_caches); |
205ab99d | 2135 | inc_slabs_node(kmalloc_caches, node, page->objects); |
6446faa2 | 2136 | |
ba84c73c | 2137 | /* |
2138 | * lockdep requires consistent irq usage for each lock | |
2139 | * so even though there cannot be a race this early in | |
2140 | * the boot sequence, we still disable irqs. | |
2141 | */ | |
2142 | local_irq_save(flags); | |
7c2e132c | 2143 | add_partial(n, page, 0); |
ba84c73c | 2144 | local_irq_restore(flags); |
81819f0f CL |
2145 | } |
2146 | ||
2147 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2148 | { | |
2149 | int node; | |
2150 | ||
f64dc58c | 2151 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2152 | struct kmem_cache_node *n = s->node[node]; |
2153 | if (n && n != &s->local_node) | |
2154 | kmem_cache_free(kmalloc_caches, n); | |
2155 | s->node[node] = NULL; | |
2156 | } | |
2157 | } | |
2158 | ||
2159 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2160 | { | |
2161 | int node; | |
2162 | int local_node; | |
2163 | ||
2164 | if (slab_state >= UP) | |
2165 | local_node = page_to_nid(virt_to_page(s)); | |
2166 | else | |
2167 | local_node = 0; | |
2168 | ||
f64dc58c | 2169 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2170 | struct kmem_cache_node *n; |
2171 | ||
2172 | if (local_node == node) | |
2173 | n = &s->local_node; | |
2174 | else { | |
2175 | if (slab_state == DOWN) { | |
0094de92 | 2176 | early_kmem_cache_node_alloc(gfpflags, node); |
81819f0f CL |
2177 | continue; |
2178 | } | |
2179 | n = kmem_cache_alloc_node(kmalloc_caches, | |
2180 | gfpflags, node); | |
2181 | ||
2182 | if (!n) { | |
2183 | free_kmem_cache_nodes(s); | |
2184 | return 0; | |
2185 | } | |
2186 | ||
2187 | } | |
2188 | s->node[node] = n; | |
5595cffc | 2189 | init_kmem_cache_node(n, s); |
81819f0f CL |
2190 | } |
2191 | return 1; | |
2192 | } | |
2193 | #else | |
2194 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2195 | { | |
2196 | } | |
2197 | ||
2198 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
2199 | { | |
5595cffc | 2200 | init_kmem_cache_node(&s->local_node, s); |
81819f0f CL |
2201 | return 1; |
2202 | } | |
2203 | #endif | |
2204 | ||
c0bdb232 | 2205 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2206 | { |
2207 | if (min < MIN_PARTIAL) | |
2208 | min = MIN_PARTIAL; | |
2209 | else if (min > MAX_PARTIAL) | |
2210 | min = MAX_PARTIAL; | |
2211 | s->min_partial = min; | |
2212 | } | |
2213 | ||
81819f0f CL |
2214 | /* |
2215 | * calculate_sizes() determines the order and the distribution of data within | |
2216 | * a slab object. | |
2217 | */ | |
06b285dc | 2218 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2219 | { |
2220 | unsigned long flags = s->flags; | |
2221 | unsigned long size = s->objsize; | |
2222 | unsigned long align = s->align; | |
834f3d11 | 2223 | int order; |
81819f0f | 2224 | |
d8b42bf5 CL |
2225 | /* |
2226 | * Round up object size to the next word boundary. We can only | |
2227 | * place the free pointer at word boundaries and this determines | |
2228 | * the possible location of the free pointer. | |
2229 | */ | |
2230 | size = ALIGN(size, sizeof(void *)); | |
2231 | ||
2232 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2233 | /* |
2234 | * Determine if we can poison the object itself. If the user of | |
2235 | * the slab may touch the object after free or before allocation | |
2236 | * then we should never poison the object itself. | |
2237 | */ | |
2238 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2239 | !s->ctor) |
81819f0f CL |
2240 | s->flags |= __OBJECT_POISON; |
2241 | else | |
2242 | s->flags &= ~__OBJECT_POISON; | |
2243 | ||
81819f0f CL |
2244 | |
2245 | /* | |
672bba3a | 2246 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2247 | * end of the object and the free pointer. If not then add an |
672bba3a | 2248 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2249 | */ |
2250 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2251 | size += sizeof(void *); | |
41ecc55b | 2252 | #endif |
81819f0f CL |
2253 | |
2254 | /* | |
672bba3a CL |
2255 | * With that we have determined the number of bytes in actual use |
2256 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2257 | */ |
2258 | s->inuse = size; | |
2259 | ||
2260 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2261 | s->ctor)) { |
81819f0f CL |
2262 | /* |
2263 | * Relocate free pointer after the object if it is not | |
2264 | * permitted to overwrite the first word of the object on | |
2265 | * kmem_cache_free. | |
2266 | * | |
2267 | * This is the case if we do RCU, have a constructor or | |
2268 | * destructor or are poisoning the objects. | |
2269 | */ | |
2270 | s->offset = size; | |
2271 | size += sizeof(void *); | |
2272 | } | |
2273 | ||
c12b3c62 | 2274 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2275 | if (flags & SLAB_STORE_USER) |
2276 | /* | |
2277 | * Need to store information about allocs and frees after | |
2278 | * the object. | |
2279 | */ | |
2280 | size += 2 * sizeof(struct track); | |
2281 | ||
be7b3fbc | 2282 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2283 | /* |
2284 | * Add some empty padding so that we can catch | |
2285 | * overwrites from earlier objects rather than let | |
2286 | * tracking information or the free pointer be | |
0211a9c8 | 2287 | * corrupted if a user writes before the start |
81819f0f CL |
2288 | * of the object. |
2289 | */ | |
2290 | size += sizeof(void *); | |
41ecc55b | 2291 | #endif |
672bba3a | 2292 | |
81819f0f CL |
2293 | /* |
2294 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2295 | * user specified and the dynamic determination of cache line size |
2296 | * on bootup. | |
81819f0f CL |
2297 | */ |
2298 | align = calculate_alignment(flags, align, s->objsize); | |
2299 | ||
2300 | /* | |
2301 | * SLUB stores one object immediately after another beginning from | |
2302 | * offset 0. In order to align the objects we have to simply size | |
2303 | * each object to conform to the alignment. | |
2304 | */ | |
2305 | size = ALIGN(size, align); | |
2306 | s->size = size; | |
06b285dc CL |
2307 | if (forced_order >= 0) |
2308 | order = forced_order; | |
2309 | else | |
2310 | order = calculate_order(size); | |
81819f0f | 2311 | |
834f3d11 | 2312 | if (order < 0) |
81819f0f CL |
2313 | return 0; |
2314 | ||
b7a49f0d | 2315 | s->allocflags = 0; |
834f3d11 | 2316 | if (order) |
b7a49f0d CL |
2317 | s->allocflags |= __GFP_COMP; |
2318 | ||
2319 | if (s->flags & SLAB_CACHE_DMA) | |
2320 | s->allocflags |= SLUB_DMA; | |
2321 | ||
2322 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2323 | s->allocflags |= __GFP_RECLAIMABLE; | |
2324 | ||
81819f0f CL |
2325 | /* |
2326 | * Determine the number of objects per slab | |
2327 | */ | |
834f3d11 | 2328 | s->oo = oo_make(order, size); |
65c3376a | 2329 | s->min = oo_make(get_order(size), size); |
205ab99d CL |
2330 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2331 | s->max = s->oo; | |
81819f0f | 2332 | |
834f3d11 | 2333 | return !!oo_objects(s->oo); |
81819f0f CL |
2334 | |
2335 | } | |
2336 | ||
81819f0f CL |
2337 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, |
2338 | const char *name, size_t size, | |
2339 | size_t align, unsigned long flags, | |
51cc5068 | 2340 | void (*ctor)(void *)) |
81819f0f CL |
2341 | { |
2342 | memset(s, 0, kmem_size); | |
2343 | s->name = name; | |
2344 | s->ctor = ctor; | |
81819f0f | 2345 | s->objsize = size; |
81819f0f | 2346 | s->align = align; |
ba0268a8 | 2347 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
81819f0f | 2348 | |
06b285dc | 2349 | if (!calculate_sizes(s, -1)) |
81819f0f CL |
2350 | goto error; |
2351 | ||
3b89d7d8 DR |
2352 | /* |
2353 | * The larger the object size is, the more pages we want on the partial | |
2354 | * list to avoid pounding the page allocator excessively. | |
2355 | */ | |
c0bdb232 | 2356 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2357 | s->refcount = 1; |
2358 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2359 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2360 | #endif |
dfb4f096 CL |
2361 | if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) |
2362 | goto error; | |
81819f0f | 2363 | |
dfb4f096 | 2364 | if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA)) |
81819f0f | 2365 | return 1; |
4c93c355 | 2366 | free_kmem_cache_nodes(s); |
81819f0f CL |
2367 | error: |
2368 | if (flags & SLAB_PANIC) | |
2369 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2370 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2371 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2372 | s->offset, flags); |
2373 | return 0; | |
2374 | } | |
81819f0f CL |
2375 | |
2376 | /* | |
2377 | * Check if a given pointer is valid | |
2378 | */ | |
2379 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
2380 | { | |
06428780 | 2381 | struct page *page; |
81819f0f CL |
2382 | |
2383 | page = get_object_page(object); | |
2384 | ||
2385 | if (!page || s != page->slab) | |
2386 | /* No slab or wrong slab */ | |
2387 | return 0; | |
2388 | ||
abcd08a6 | 2389 | if (!check_valid_pointer(s, page, object)) |
81819f0f CL |
2390 | return 0; |
2391 | ||
2392 | /* | |
2393 | * We could also check if the object is on the slabs freelist. | |
2394 | * But this would be too expensive and it seems that the main | |
6446faa2 | 2395 | * purpose of kmem_ptr_valid() is to check if the object belongs |
81819f0f CL |
2396 | * to a certain slab. |
2397 | */ | |
2398 | return 1; | |
2399 | } | |
2400 | EXPORT_SYMBOL(kmem_ptr_validate); | |
2401 | ||
2402 | /* | |
2403 | * Determine the size of a slab object | |
2404 | */ | |
2405 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2406 | { | |
2407 | return s->objsize; | |
2408 | } | |
2409 | EXPORT_SYMBOL(kmem_cache_size); | |
2410 | ||
2411 | const char *kmem_cache_name(struct kmem_cache *s) | |
2412 | { | |
2413 | return s->name; | |
2414 | } | |
2415 | EXPORT_SYMBOL(kmem_cache_name); | |
2416 | ||
33b12c38 CL |
2417 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2418 | const char *text) | |
2419 | { | |
2420 | #ifdef CONFIG_SLUB_DEBUG | |
2421 | void *addr = page_address(page); | |
2422 | void *p; | |
2423 | DECLARE_BITMAP(map, page->objects); | |
2424 | ||
2425 | bitmap_zero(map, page->objects); | |
2426 | slab_err(s, page, "%s", text); | |
2427 | slab_lock(page); | |
2428 | for_each_free_object(p, s, page->freelist) | |
2429 | set_bit(slab_index(p, s, addr), map); | |
2430 | ||
2431 | for_each_object(p, s, addr, page->objects) { | |
2432 | ||
2433 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2434 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2435 | p, p - addr); | |
2436 | print_tracking(s, p); | |
2437 | } | |
2438 | } | |
2439 | slab_unlock(page); | |
2440 | #endif | |
2441 | } | |
2442 | ||
81819f0f | 2443 | /* |
599870b1 | 2444 | * Attempt to free all partial slabs on a node. |
81819f0f | 2445 | */ |
599870b1 | 2446 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2447 | { |
81819f0f CL |
2448 | unsigned long flags; |
2449 | struct page *page, *h; | |
2450 | ||
2451 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2452 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f CL |
2453 | if (!page->inuse) { |
2454 | list_del(&page->lru); | |
2455 | discard_slab(s, page); | |
599870b1 | 2456 | n->nr_partial--; |
33b12c38 CL |
2457 | } else { |
2458 | list_slab_objects(s, page, | |
2459 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2460 | } |
33b12c38 | 2461 | } |
81819f0f | 2462 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2463 | } |
2464 | ||
2465 | /* | |
672bba3a | 2466 | * Release all resources used by a slab cache. |
81819f0f | 2467 | */ |
0c710013 | 2468 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2469 | { |
2470 | int node; | |
2471 | ||
2472 | flush_all(s); | |
2473 | ||
2474 | /* Attempt to free all objects */ | |
4c93c355 | 2475 | free_kmem_cache_cpus(s); |
f64dc58c | 2476 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2477 | struct kmem_cache_node *n = get_node(s, node); |
2478 | ||
599870b1 CL |
2479 | free_partial(s, n); |
2480 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2481 | return 1; |
2482 | } | |
2483 | free_kmem_cache_nodes(s); | |
2484 | return 0; | |
2485 | } | |
2486 | ||
2487 | /* | |
2488 | * Close a cache and release the kmem_cache structure | |
2489 | * (must be used for caches created using kmem_cache_create) | |
2490 | */ | |
2491 | void kmem_cache_destroy(struct kmem_cache *s) | |
2492 | { | |
2493 | down_write(&slub_lock); | |
2494 | s->refcount--; | |
2495 | if (!s->refcount) { | |
2496 | list_del(&s->list); | |
a0e1d1be | 2497 | up_write(&slub_lock); |
d629d819 PE |
2498 | if (kmem_cache_close(s)) { |
2499 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2500 | "still has objects.\n", s->name, __func__); | |
2501 | dump_stack(); | |
2502 | } | |
81819f0f | 2503 | sysfs_slab_remove(s); |
a0e1d1be CL |
2504 | } else |
2505 | up_write(&slub_lock); | |
81819f0f CL |
2506 | } |
2507 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2508 | ||
2509 | /******************************************************************** | |
2510 | * Kmalloc subsystem | |
2511 | *******************************************************************/ | |
2512 | ||
ffadd4d0 | 2513 | struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned; |
81819f0f CL |
2514 | EXPORT_SYMBOL(kmalloc_caches); |
2515 | ||
81819f0f CL |
2516 | static int __init setup_slub_min_order(char *str) |
2517 | { | |
06428780 | 2518 | get_option(&str, &slub_min_order); |
81819f0f CL |
2519 | |
2520 | return 1; | |
2521 | } | |
2522 | ||
2523 | __setup("slub_min_order=", setup_slub_min_order); | |
2524 | ||
2525 | static int __init setup_slub_max_order(char *str) | |
2526 | { | |
06428780 | 2527 | get_option(&str, &slub_max_order); |
818cf590 | 2528 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2529 | |
2530 | return 1; | |
2531 | } | |
2532 | ||
2533 | __setup("slub_max_order=", setup_slub_max_order); | |
2534 | ||
2535 | static int __init setup_slub_min_objects(char *str) | |
2536 | { | |
06428780 | 2537 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2538 | |
2539 | return 1; | |
2540 | } | |
2541 | ||
2542 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2543 | ||
2544 | static int __init setup_slub_nomerge(char *str) | |
2545 | { | |
2546 | slub_nomerge = 1; | |
2547 | return 1; | |
2548 | } | |
2549 | ||
2550 | __setup("slub_nomerge", setup_slub_nomerge); | |
2551 | ||
81819f0f CL |
2552 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, |
2553 | const char *name, int size, gfp_t gfp_flags) | |
2554 | { | |
2555 | unsigned int flags = 0; | |
2556 | ||
2557 | if (gfp_flags & SLUB_DMA) | |
2558 | flags = SLAB_CACHE_DMA; | |
2559 | ||
2560 | down_write(&slub_lock); | |
2561 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
319d1e24 | 2562 | flags, NULL)) |
81819f0f CL |
2563 | goto panic; |
2564 | ||
2565 | list_add(&s->list, &slab_caches); | |
2566 | up_write(&slub_lock); | |
2567 | if (sysfs_slab_add(s)) | |
2568 | goto panic; | |
2569 | return s; | |
2570 | ||
2571 | panic: | |
2572 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
2573 | } | |
2574 | ||
2e443fd0 | 2575 | #ifdef CONFIG_ZONE_DMA |
ffadd4d0 | 2576 | static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT]; |
1ceef402 CL |
2577 | |
2578 | static void sysfs_add_func(struct work_struct *w) | |
2579 | { | |
2580 | struct kmem_cache *s; | |
2581 | ||
2582 | down_write(&slub_lock); | |
2583 | list_for_each_entry(s, &slab_caches, list) { | |
2584 | if (s->flags & __SYSFS_ADD_DEFERRED) { | |
2585 | s->flags &= ~__SYSFS_ADD_DEFERRED; | |
2586 | sysfs_slab_add(s); | |
2587 | } | |
2588 | } | |
2589 | up_write(&slub_lock); | |
2590 | } | |
2591 | ||
2592 | static DECLARE_WORK(sysfs_add_work, sysfs_add_func); | |
2593 | ||
2e443fd0 CL |
2594 | static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags) |
2595 | { | |
2596 | struct kmem_cache *s; | |
2e443fd0 CL |
2597 | char *text; |
2598 | size_t realsize; | |
2599 | ||
2600 | s = kmalloc_caches_dma[index]; | |
2601 | if (s) | |
2602 | return s; | |
2603 | ||
2604 | /* Dynamically create dma cache */ | |
1ceef402 CL |
2605 | if (flags & __GFP_WAIT) |
2606 | down_write(&slub_lock); | |
2607 | else { | |
2608 | if (!down_write_trylock(&slub_lock)) | |
2609 | goto out; | |
2610 | } | |
2611 | ||
2612 | if (kmalloc_caches_dma[index]) | |
2613 | goto unlock_out; | |
2e443fd0 | 2614 | |
7b55f620 | 2615 | realsize = kmalloc_caches[index].objsize; |
3adbefee IM |
2616 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", |
2617 | (unsigned int)realsize); | |
1ceef402 CL |
2618 | s = kmalloc(kmem_size, flags & ~SLUB_DMA); |
2619 | ||
2620 | if (!s || !text || !kmem_cache_open(s, flags, text, | |
2621 | realsize, ARCH_KMALLOC_MINALIGN, | |
2622 | SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) { | |
2623 | kfree(s); | |
2624 | kfree(text); | |
2625 | goto unlock_out; | |
dfce8648 | 2626 | } |
1ceef402 CL |
2627 | |
2628 | list_add(&s->list, &slab_caches); | |
2629 | kmalloc_caches_dma[index] = s; | |
2630 | ||
2631 | schedule_work(&sysfs_add_work); | |
2632 | ||
2633 | unlock_out: | |
dfce8648 | 2634 | up_write(&slub_lock); |
1ceef402 | 2635 | out: |
dfce8648 | 2636 | return kmalloc_caches_dma[index]; |
2e443fd0 CL |
2637 | } |
2638 | #endif | |
2639 | ||
f1b26339 CL |
2640 | /* |
2641 | * Conversion table for small slabs sizes / 8 to the index in the | |
2642 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2643 | * of two cache sizes there. The size of larger slabs can be determined using | |
2644 | * fls. | |
2645 | */ | |
2646 | static s8 size_index[24] = { | |
2647 | 3, /* 8 */ | |
2648 | 4, /* 16 */ | |
2649 | 5, /* 24 */ | |
2650 | 5, /* 32 */ | |
2651 | 6, /* 40 */ | |
2652 | 6, /* 48 */ | |
2653 | 6, /* 56 */ | |
2654 | 6, /* 64 */ | |
2655 | 1, /* 72 */ | |
2656 | 1, /* 80 */ | |
2657 | 1, /* 88 */ | |
2658 | 1, /* 96 */ | |
2659 | 7, /* 104 */ | |
2660 | 7, /* 112 */ | |
2661 | 7, /* 120 */ | |
2662 | 7, /* 128 */ | |
2663 | 2, /* 136 */ | |
2664 | 2, /* 144 */ | |
2665 | 2, /* 152 */ | |
2666 | 2, /* 160 */ | |
2667 | 2, /* 168 */ | |
2668 | 2, /* 176 */ | |
2669 | 2, /* 184 */ | |
2670 | 2 /* 192 */ | |
2671 | }; | |
2672 | ||
81819f0f CL |
2673 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2674 | { | |
f1b26339 | 2675 | int index; |
81819f0f | 2676 | |
f1b26339 CL |
2677 | if (size <= 192) { |
2678 | if (!size) | |
2679 | return ZERO_SIZE_PTR; | |
81819f0f | 2680 | |
f1b26339 | 2681 | index = size_index[(size - 1) / 8]; |
aadb4bc4 | 2682 | } else |
f1b26339 | 2683 | index = fls(size - 1); |
81819f0f CL |
2684 | |
2685 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2686 | if (unlikely((flags & SLUB_DMA))) |
2e443fd0 | 2687 | return dma_kmalloc_cache(index, flags); |
f1b26339 | 2688 | |
81819f0f CL |
2689 | #endif |
2690 | return &kmalloc_caches[index]; | |
2691 | } | |
2692 | ||
2693 | void *__kmalloc(size_t size, gfp_t flags) | |
2694 | { | |
aadb4bc4 | 2695 | struct kmem_cache *s; |
5b882be4 | 2696 | void *ret; |
81819f0f | 2697 | |
ffadd4d0 | 2698 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2699 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2700 | |
2701 | s = get_slab(size, flags); | |
2702 | ||
2703 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2704 | return s; |
2705 | ||
5b882be4 EGM |
2706 | ret = slab_alloc(s, flags, -1, _RET_IP_); |
2707 | ||
ca2b84cb | 2708 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2709 | |
2710 | return ret; | |
81819f0f CL |
2711 | } |
2712 | EXPORT_SYMBOL(__kmalloc); | |
2713 | ||
f619cfe1 CL |
2714 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2715 | { | |
2716 | struct page *page = alloc_pages_node(node, flags | __GFP_COMP, | |
2717 | get_order(size)); | |
2718 | ||
2719 | if (page) | |
2720 | return page_address(page); | |
2721 | else | |
2722 | return NULL; | |
2723 | } | |
2724 | ||
81819f0f CL |
2725 | #ifdef CONFIG_NUMA |
2726 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2727 | { | |
aadb4bc4 | 2728 | struct kmem_cache *s; |
5b882be4 | 2729 | void *ret; |
81819f0f | 2730 | |
057685cf | 2731 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2732 | ret = kmalloc_large_node(size, flags, node); |
2733 | ||
ca2b84cb EGM |
2734 | trace_kmalloc_node(_RET_IP_, ret, |
2735 | size, PAGE_SIZE << get_order(size), | |
2736 | flags, node); | |
5b882be4 EGM |
2737 | |
2738 | return ret; | |
2739 | } | |
aadb4bc4 CL |
2740 | |
2741 | s = get_slab(size, flags); | |
2742 | ||
2743 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2744 | return s; |
2745 | ||
5b882be4 EGM |
2746 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2747 | ||
ca2b84cb | 2748 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2749 | |
2750 | return ret; | |
81819f0f CL |
2751 | } |
2752 | EXPORT_SYMBOL(__kmalloc_node); | |
2753 | #endif | |
2754 | ||
2755 | size_t ksize(const void *object) | |
2756 | { | |
272c1d21 | 2757 | struct page *page; |
81819f0f CL |
2758 | struct kmem_cache *s; |
2759 | ||
ef8b4520 | 2760 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2761 | return 0; |
2762 | ||
294a80a8 | 2763 | page = virt_to_head_page(object); |
294a80a8 | 2764 | |
76994412 PE |
2765 | if (unlikely(!PageSlab(page))) { |
2766 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2767 | return PAGE_SIZE << compound_order(page); |
76994412 | 2768 | } |
81819f0f | 2769 | s = page->slab; |
81819f0f | 2770 | |
ae20bfda | 2771 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2772 | /* |
2773 | * Debugging requires use of the padding between object | |
2774 | * and whatever may come after it. | |
2775 | */ | |
2776 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2777 | return s->objsize; | |
2778 | ||
ae20bfda | 2779 | #endif |
81819f0f CL |
2780 | /* |
2781 | * If we have the need to store the freelist pointer | |
2782 | * back there or track user information then we can | |
2783 | * only use the space before that information. | |
2784 | */ | |
2785 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2786 | return s->inuse; | |
81819f0f CL |
2787 | /* |
2788 | * Else we can use all the padding etc for the allocation | |
2789 | */ | |
2790 | return s->size; | |
2791 | } | |
b1aabecd | 2792 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2793 | |
2794 | void kfree(const void *x) | |
2795 | { | |
81819f0f | 2796 | struct page *page; |
5bb983b0 | 2797 | void *object = (void *)x; |
81819f0f | 2798 | |
2121db74 PE |
2799 | trace_kfree(_RET_IP_, x); |
2800 | ||
2408c550 | 2801 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2802 | return; |
2803 | ||
b49af68f | 2804 | page = virt_to_head_page(x); |
aadb4bc4 | 2805 | if (unlikely(!PageSlab(page))) { |
0937502a | 2806 | BUG_ON(!PageCompound(page)); |
aadb4bc4 CL |
2807 | put_page(page); |
2808 | return; | |
2809 | } | |
ce71e27c | 2810 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2811 | } |
2812 | EXPORT_SYMBOL(kfree); | |
2813 | ||
2086d26a | 2814 | /* |
672bba3a CL |
2815 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
2816 | * the remaining slabs by the number of items in use. The slabs with the | |
2817 | * most items in use come first. New allocations will then fill those up | |
2818 | * and thus they can be removed from the partial lists. | |
2819 | * | |
2820 | * The slabs with the least items are placed last. This results in them | |
2821 | * being allocated from last increasing the chance that the last objects | |
2822 | * are freed in them. | |
2086d26a CL |
2823 | */ |
2824 | int kmem_cache_shrink(struct kmem_cache *s) | |
2825 | { | |
2826 | int node; | |
2827 | int i; | |
2828 | struct kmem_cache_node *n; | |
2829 | struct page *page; | |
2830 | struct page *t; | |
205ab99d | 2831 | int objects = oo_objects(s->max); |
2086d26a | 2832 | struct list_head *slabs_by_inuse = |
834f3d11 | 2833 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
2834 | unsigned long flags; |
2835 | ||
2836 | if (!slabs_by_inuse) | |
2837 | return -ENOMEM; | |
2838 | ||
2839 | flush_all(s); | |
f64dc58c | 2840 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
2841 | n = get_node(s, node); |
2842 | ||
2843 | if (!n->nr_partial) | |
2844 | continue; | |
2845 | ||
834f3d11 | 2846 | for (i = 0; i < objects; i++) |
2086d26a CL |
2847 | INIT_LIST_HEAD(slabs_by_inuse + i); |
2848 | ||
2849 | spin_lock_irqsave(&n->list_lock, flags); | |
2850 | ||
2851 | /* | |
672bba3a | 2852 | * Build lists indexed by the items in use in each slab. |
2086d26a | 2853 | * |
672bba3a CL |
2854 | * Note that concurrent frees may occur while we hold the |
2855 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
2856 | */ |
2857 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
2858 | if (!page->inuse && slab_trylock(page)) { | |
2859 | /* | |
2860 | * Must hold slab lock here because slab_free | |
2861 | * may have freed the last object and be | |
2862 | * waiting to release the slab. | |
2863 | */ | |
2864 | list_del(&page->lru); | |
2865 | n->nr_partial--; | |
2866 | slab_unlock(page); | |
2867 | discard_slab(s, page); | |
2868 | } else { | |
fcda3d89 CL |
2869 | list_move(&page->lru, |
2870 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
2871 | } |
2872 | } | |
2873 | ||
2086d26a | 2874 | /* |
672bba3a CL |
2875 | * Rebuild the partial list with the slabs filled up most |
2876 | * first and the least used slabs at the end. | |
2086d26a | 2877 | */ |
834f3d11 | 2878 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
2879 | list_splice(slabs_by_inuse + i, n->partial.prev); |
2880 | ||
2086d26a CL |
2881 | spin_unlock_irqrestore(&n->list_lock, flags); |
2882 | } | |
2883 | ||
2884 | kfree(slabs_by_inuse); | |
2885 | return 0; | |
2886 | } | |
2887 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2888 | ||
b9049e23 YG |
2889 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
2890 | static int slab_mem_going_offline_callback(void *arg) | |
2891 | { | |
2892 | struct kmem_cache *s; | |
2893 | ||
2894 | down_read(&slub_lock); | |
2895 | list_for_each_entry(s, &slab_caches, list) | |
2896 | kmem_cache_shrink(s); | |
2897 | up_read(&slub_lock); | |
2898 | ||
2899 | return 0; | |
2900 | } | |
2901 | ||
2902 | static void slab_mem_offline_callback(void *arg) | |
2903 | { | |
2904 | struct kmem_cache_node *n; | |
2905 | struct kmem_cache *s; | |
2906 | struct memory_notify *marg = arg; | |
2907 | int offline_node; | |
2908 | ||
2909 | offline_node = marg->status_change_nid; | |
2910 | ||
2911 | /* | |
2912 | * If the node still has available memory. we need kmem_cache_node | |
2913 | * for it yet. | |
2914 | */ | |
2915 | if (offline_node < 0) | |
2916 | return; | |
2917 | ||
2918 | down_read(&slub_lock); | |
2919 | list_for_each_entry(s, &slab_caches, list) { | |
2920 | n = get_node(s, offline_node); | |
2921 | if (n) { | |
2922 | /* | |
2923 | * if n->nr_slabs > 0, slabs still exist on the node | |
2924 | * that is going down. We were unable to free them, | |
2925 | * and offline_pages() function shoudn't call this | |
2926 | * callback. So, we must fail. | |
2927 | */ | |
0f389ec6 | 2928 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
2929 | |
2930 | s->node[offline_node] = NULL; | |
2931 | kmem_cache_free(kmalloc_caches, n); | |
2932 | } | |
2933 | } | |
2934 | up_read(&slub_lock); | |
2935 | } | |
2936 | ||
2937 | static int slab_mem_going_online_callback(void *arg) | |
2938 | { | |
2939 | struct kmem_cache_node *n; | |
2940 | struct kmem_cache *s; | |
2941 | struct memory_notify *marg = arg; | |
2942 | int nid = marg->status_change_nid; | |
2943 | int ret = 0; | |
2944 | ||
2945 | /* | |
2946 | * If the node's memory is already available, then kmem_cache_node is | |
2947 | * already created. Nothing to do. | |
2948 | */ | |
2949 | if (nid < 0) | |
2950 | return 0; | |
2951 | ||
2952 | /* | |
0121c619 | 2953 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
2954 | * allocate a kmem_cache_node structure in order to bring the node |
2955 | * online. | |
2956 | */ | |
2957 | down_read(&slub_lock); | |
2958 | list_for_each_entry(s, &slab_caches, list) { | |
2959 | /* | |
2960 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
2961 | * since memory is not yet available from the node that | |
2962 | * is brought up. | |
2963 | */ | |
2964 | n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL); | |
2965 | if (!n) { | |
2966 | ret = -ENOMEM; | |
2967 | goto out; | |
2968 | } | |
5595cffc | 2969 | init_kmem_cache_node(n, s); |
b9049e23 YG |
2970 | s->node[nid] = n; |
2971 | } | |
2972 | out: | |
2973 | up_read(&slub_lock); | |
2974 | return ret; | |
2975 | } | |
2976 | ||
2977 | static int slab_memory_callback(struct notifier_block *self, | |
2978 | unsigned long action, void *arg) | |
2979 | { | |
2980 | int ret = 0; | |
2981 | ||
2982 | switch (action) { | |
2983 | case MEM_GOING_ONLINE: | |
2984 | ret = slab_mem_going_online_callback(arg); | |
2985 | break; | |
2986 | case MEM_GOING_OFFLINE: | |
2987 | ret = slab_mem_going_offline_callback(arg); | |
2988 | break; | |
2989 | case MEM_OFFLINE: | |
2990 | case MEM_CANCEL_ONLINE: | |
2991 | slab_mem_offline_callback(arg); | |
2992 | break; | |
2993 | case MEM_ONLINE: | |
2994 | case MEM_CANCEL_OFFLINE: | |
2995 | break; | |
2996 | } | |
dc19f9db KH |
2997 | if (ret) |
2998 | ret = notifier_from_errno(ret); | |
2999 | else | |
3000 | ret = NOTIFY_OK; | |
b9049e23 YG |
3001 | return ret; |
3002 | } | |
3003 | ||
3004 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3005 | ||
81819f0f CL |
3006 | /******************************************************************** |
3007 | * Basic setup of slabs | |
3008 | *******************************************************************/ | |
3009 | ||
3010 | void __init kmem_cache_init(void) | |
3011 | { | |
3012 | int i; | |
4b356be0 | 3013 | int caches = 0; |
81819f0f | 3014 | |
4c93c355 CL |
3015 | init_alloc_cpu(); |
3016 | ||
81819f0f CL |
3017 | #ifdef CONFIG_NUMA |
3018 | /* | |
3019 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3020 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3021 | * kmem_cache_open for slab_state == DOWN. |
3022 | */ | |
3023 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
3024 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
8ffa6875 | 3025 | kmalloc_caches[0].refcount = -1; |
4b356be0 | 3026 | caches++; |
b9049e23 | 3027 | |
0c40ba4f | 3028 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3029 | #endif |
3030 | ||
3031 | /* Able to allocate the per node structures */ | |
3032 | slab_state = PARTIAL; | |
3033 | ||
3034 | /* Caches that are not of the two-to-the-power-of size */ | |
4b356be0 CL |
3035 | if (KMALLOC_MIN_SIZE <= 64) { |
3036 | create_kmalloc_cache(&kmalloc_caches[1], | |
81819f0f | 3037 | "kmalloc-96", 96, GFP_KERNEL); |
4b356be0 | 3038 | caches++; |
4b356be0 | 3039 | create_kmalloc_cache(&kmalloc_caches[2], |
81819f0f | 3040 | "kmalloc-192", 192, GFP_KERNEL); |
4b356be0 CL |
3041 | caches++; |
3042 | } | |
81819f0f | 3043 | |
ffadd4d0 | 3044 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
81819f0f CL |
3045 | create_kmalloc_cache(&kmalloc_caches[i], |
3046 | "kmalloc", 1 << i, GFP_KERNEL); | |
4b356be0 CL |
3047 | caches++; |
3048 | } | |
81819f0f | 3049 | |
f1b26339 CL |
3050 | |
3051 | /* | |
3052 | * Patch up the size_index table if we have strange large alignment | |
3053 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3054 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3055 | * |
3056 | * Largest permitted alignment is 256 bytes due to the way we | |
3057 | * handle the index determination for the smaller caches. | |
3058 | * | |
3059 | * Make sure that nothing crazy happens if someone starts tinkering | |
3060 | * around with ARCH_KMALLOC_MINALIGN | |
3061 | */ | |
3062 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3063 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3064 | ||
12ad6843 | 3065 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) |
f1b26339 CL |
3066 | size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW; |
3067 | ||
41d54d3b CL |
3068 | if (KMALLOC_MIN_SIZE == 128) { |
3069 | /* | |
3070 | * The 192 byte sized cache is not used if the alignment | |
3071 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3072 | * instead. | |
3073 | */ | |
3074 | for (i = 128 + 8; i <= 192; i += 8) | |
3075 | size_index[(i - 1) / 8] = 8; | |
3076 | } | |
3077 | ||
81819f0f CL |
3078 | slab_state = UP; |
3079 | ||
3080 | /* Provide the correct kmalloc names now that the caches are up */ | |
ffadd4d0 | 3081 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) |
81819f0f CL |
3082 | kmalloc_caches[i]. name = |
3083 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
3084 | ||
3085 | #ifdef CONFIG_SMP | |
3086 | register_cpu_notifier(&slab_notifier); | |
4c93c355 CL |
3087 | kmem_size = offsetof(struct kmem_cache, cpu_slab) + |
3088 | nr_cpu_ids * sizeof(struct kmem_cache_cpu *); | |
3089 | #else | |
3090 | kmem_size = sizeof(struct kmem_cache); | |
81819f0f CL |
3091 | #endif |
3092 | ||
3adbefee IM |
3093 | printk(KERN_INFO |
3094 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3095 | " CPUs=%d, Nodes=%d\n", |
3096 | caches, cache_line_size(), | |
81819f0f CL |
3097 | slub_min_order, slub_max_order, slub_min_objects, |
3098 | nr_cpu_ids, nr_node_ids); | |
3099 | } | |
3100 | ||
3101 | /* | |
3102 | * Find a mergeable slab cache | |
3103 | */ | |
3104 | static int slab_unmergeable(struct kmem_cache *s) | |
3105 | { | |
3106 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3107 | return 1; | |
3108 | ||
c59def9f | 3109 | if (s->ctor) |
81819f0f CL |
3110 | return 1; |
3111 | ||
8ffa6875 CL |
3112 | /* |
3113 | * We may have set a slab to be unmergeable during bootstrap. | |
3114 | */ | |
3115 | if (s->refcount < 0) | |
3116 | return 1; | |
3117 | ||
81819f0f CL |
3118 | return 0; |
3119 | } | |
3120 | ||
3121 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3122 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3123 | void (*ctor)(void *)) |
81819f0f | 3124 | { |
5b95a4ac | 3125 | struct kmem_cache *s; |
81819f0f CL |
3126 | |
3127 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3128 | return NULL; | |
3129 | ||
c59def9f | 3130 | if (ctor) |
81819f0f CL |
3131 | return NULL; |
3132 | ||
3133 | size = ALIGN(size, sizeof(void *)); | |
3134 | align = calculate_alignment(flags, align, size); | |
3135 | size = ALIGN(size, align); | |
ba0268a8 | 3136 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3137 | |
5b95a4ac | 3138 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3139 | if (slab_unmergeable(s)) |
3140 | continue; | |
3141 | ||
3142 | if (size > s->size) | |
3143 | continue; | |
3144 | ||
ba0268a8 | 3145 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3146 | continue; |
3147 | /* | |
3148 | * Check if alignment is compatible. | |
3149 | * Courtesy of Adrian Drzewiecki | |
3150 | */ | |
06428780 | 3151 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3152 | continue; |
3153 | ||
3154 | if (s->size - size >= sizeof(void *)) | |
3155 | continue; | |
3156 | ||
3157 | return s; | |
3158 | } | |
3159 | return NULL; | |
3160 | } | |
3161 | ||
3162 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3163 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3164 | { |
3165 | struct kmem_cache *s; | |
3166 | ||
3167 | down_write(&slub_lock); | |
ba0268a8 | 3168 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3169 | if (s) { |
42a9fdbb CL |
3170 | int cpu; |
3171 | ||
81819f0f CL |
3172 | s->refcount++; |
3173 | /* | |
3174 | * Adjust the object sizes so that we clear | |
3175 | * the complete object on kzalloc. | |
3176 | */ | |
3177 | s->objsize = max(s->objsize, (int)size); | |
42a9fdbb CL |
3178 | |
3179 | /* | |
3180 | * And then we need to update the object size in the | |
3181 | * per cpu structures | |
3182 | */ | |
3183 | for_each_online_cpu(cpu) | |
3184 | get_cpu_slab(s, cpu)->objsize = s->objsize; | |
6446faa2 | 3185 | |
81819f0f | 3186 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 3187 | up_write(&slub_lock); |
6446faa2 | 3188 | |
7b8f3b66 DR |
3189 | if (sysfs_slab_alias(s, name)) { |
3190 | down_write(&slub_lock); | |
3191 | s->refcount--; | |
3192 | up_write(&slub_lock); | |
81819f0f | 3193 | goto err; |
7b8f3b66 | 3194 | } |
a0e1d1be CL |
3195 | return s; |
3196 | } | |
6446faa2 | 3197 | |
a0e1d1be CL |
3198 | s = kmalloc(kmem_size, GFP_KERNEL); |
3199 | if (s) { | |
3200 | if (kmem_cache_open(s, GFP_KERNEL, name, | |
c59def9f | 3201 | size, align, flags, ctor)) { |
81819f0f | 3202 | list_add(&s->list, &slab_caches); |
a0e1d1be | 3203 | up_write(&slub_lock); |
7b8f3b66 DR |
3204 | if (sysfs_slab_add(s)) { |
3205 | down_write(&slub_lock); | |
3206 | list_del(&s->list); | |
3207 | up_write(&slub_lock); | |
3208 | kfree(s); | |
a0e1d1be | 3209 | goto err; |
7b8f3b66 | 3210 | } |
a0e1d1be CL |
3211 | return s; |
3212 | } | |
3213 | kfree(s); | |
81819f0f CL |
3214 | } |
3215 | up_write(&slub_lock); | |
81819f0f CL |
3216 | |
3217 | err: | |
81819f0f CL |
3218 | if (flags & SLAB_PANIC) |
3219 | panic("Cannot create slabcache %s\n", name); | |
3220 | else | |
3221 | s = NULL; | |
3222 | return s; | |
3223 | } | |
3224 | EXPORT_SYMBOL(kmem_cache_create); | |
3225 | ||
81819f0f | 3226 | #ifdef CONFIG_SMP |
81819f0f | 3227 | /* |
672bba3a CL |
3228 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3229 | * necessary. | |
81819f0f CL |
3230 | */ |
3231 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3232 | unsigned long action, void *hcpu) | |
3233 | { | |
3234 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3235 | struct kmem_cache *s; |
3236 | unsigned long flags; | |
81819f0f CL |
3237 | |
3238 | switch (action) { | |
4c93c355 CL |
3239 | case CPU_UP_PREPARE: |
3240 | case CPU_UP_PREPARE_FROZEN: | |
3241 | init_alloc_cpu_cpu(cpu); | |
3242 | down_read(&slub_lock); | |
3243 | list_for_each_entry(s, &slab_caches, list) | |
3244 | s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu, | |
3245 | GFP_KERNEL); | |
3246 | up_read(&slub_lock); | |
3247 | break; | |
3248 | ||
81819f0f | 3249 | case CPU_UP_CANCELED: |
8bb78442 | 3250 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3251 | case CPU_DEAD: |
8bb78442 | 3252 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3253 | down_read(&slub_lock); |
3254 | list_for_each_entry(s, &slab_caches, list) { | |
4c93c355 CL |
3255 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); |
3256 | ||
5b95a4ac CL |
3257 | local_irq_save(flags); |
3258 | __flush_cpu_slab(s, cpu); | |
3259 | local_irq_restore(flags); | |
4c93c355 CL |
3260 | free_kmem_cache_cpu(c, cpu); |
3261 | s->cpu_slab[cpu] = NULL; | |
5b95a4ac CL |
3262 | } |
3263 | up_read(&slub_lock); | |
81819f0f CL |
3264 | break; |
3265 | default: | |
3266 | break; | |
3267 | } | |
3268 | return NOTIFY_OK; | |
3269 | } | |
3270 | ||
06428780 | 3271 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3272 | .notifier_call = slab_cpuup_callback |
06428780 | 3273 | }; |
81819f0f CL |
3274 | |
3275 | #endif | |
3276 | ||
ce71e27c | 3277 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3278 | { |
aadb4bc4 | 3279 | struct kmem_cache *s; |
94b528d0 | 3280 | void *ret; |
aadb4bc4 | 3281 | |
ffadd4d0 | 3282 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3283 | return kmalloc_large(size, gfpflags); |
3284 | ||
aadb4bc4 | 3285 | s = get_slab(size, gfpflags); |
81819f0f | 3286 | |
2408c550 | 3287 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3288 | return s; |
81819f0f | 3289 | |
94b528d0 EGM |
3290 | ret = slab_alloc(s, gfpflags, -1, caller); |
3291 | ||
3292 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3293 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3294 | |
3295 | return ret; | |
81819f0f CL |
3296 | } |
3297 | ||
3298 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
ce71e27c | 3299 | int node, unsigned long caller) |
81819f0f | 3300 | { |
aadb4bc4 | 3301 | struct kmem_cache *s; |
94b528d0 | 3302 | void *ret; |
aadb4bc4 | 3303 | |
ffadd4d0 | 3304 | if (unlikely(size > SLUB_MAX_SIZE)) |
f619cfe1 | 3305 | return kmalloc_large_node(size, gfpflags, node); |
eada35ef | 3306 | |
aadb4bc4 | 3307 | s = get_slab(size, gfpflags); |
81819f0f | 3308 | |
2408c550 | 3309 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3310 | return s; |
81819f0f | 3311 | |
94b528d0 EGM |
3312 | ret = slab_alloc(s, gfpflags, node, caller); |
3313 | ||
3314 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3315 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3316 | |
3317 | return ret; | |
81819f0f CL |
3318 | } |
3319 | ||
f6acb635 | 3320 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
3321 | static unsigned long count_partial(struct kmem_cache_node *n, |
3322 | int (*get_count)(struct page *)) | |
5b06c853 CL |
3323 | { |
3324 | unsigned long flags; | |
3325 | unsigned long x = 0; | |
3326 | struct page *page; | |
3327 | ||
3328 | spin_lock_irqsave(&n->list_lock, flags); | |
3329 | list_for_each_entry(page, &n->partial, lru) | |
205ab99d | 3330 | x += get_count(page); |
5b06c853 CL |
3331 | spin_unlock_irqrestore(&n->list_lock, flags); |
3332 | return x; | |
3333 | } | |
205ab99d CL |
3334 | |
3335 | static int count_inuse(struct page *page) | |
3336 | { | |
3337 | return page->inuse; | |
3338 | } | |
3339 | ||
3340 | static int count_total(struct page *page) | |
3341 | { | |
3342 | return page->objects; | |
3343 | } | |
3344 | ||
3345 | static int count_free(struct page *page) | |
3346 | { | |
3347 | return page->objects - page->inuse; | |
3348 | } | |
5b06c853 | 3349 | |
434e245d CL |
3350 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3351 | unsigned long *map) | |
53e15af0 CL |
3352 | { |
3353 | void *p; | |
a973e9dd | 3354 | void *addr = page_address(page); |
53e15af0 CL |
3355 | |
3356 | if (!check_slab(s, page) || | |
3357 | !on_freelist(s, page, NULL)) | |
3358 | return 0; | |
3359 | ||
3360 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3361 | bitmap_zero(map, page->objects); |
53e15af0 | 3362 | |
7656c72b CL |
3363 | for_each_free_object(p, s, page->freelist) { |
3364 | set_bit(slab_index(p, s, addr), map); | |
53e15af0 CL |
3365 | if (!check_object(s, page, p, 0)) |
3366 | return 0; | |
3367 | } | |
3368 | ||
224a88be | 3369 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3370 | if (!test_bit(slab_index(p, s, addr), map)) |
53e15af0 CL |
3371 | if (!check_object(s, page, p, 1)) |
3372 | return 0; | |
3373 | return 1; | |
3374 | } | |
3375 | ||
434e245d CL |
3376 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3377 | unsigned long *map) | |
53e15af0 CL |
3378 | { |
3379 | if (slab_trylock(page)) { | |
434e245d | 3380 | validate_slab(s, page, map); |
53e15af0 CL |
3381 | slab_unlock(page); |
3382 | } else | |
3383 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3384 | s->name, page); | |
3385 | ||
3386 | if (s->flags & DEBUG_DEFAULT_FLAGS) { | |
8a38082d AW |
3387 | if (!PageSlubDebug(page)) |
3388 | printk(KERN_ERR "SLUB %s: SlubDebug not set " | |
53e15af0 CL |
3389 | "on slab 0x%p\n", s->name, page); |
3390 | } else { | |
8a38082d AW |
3391 | if (PageSlubDebug(page)) |
3392 | printk(KERN_ERR "SLUB %s: SlubDebug set on " | |
53e15af0 CL |
3393 | "slab 0x%p\n", s->name, page); |
3394 | } | |
3395 | } | |
3396 | ||
434e245d CL |
3397 | static int validate_slab_node(struct kmem_cache *s, |
3398 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3399 | { |
3400 | unsigned long count = 0; | |
3401 | struct page *page; | |
3402 | unsigned long flags; | |
3403 | ||
3404 | spin_lock_irqsave(&n->list_lock, flags); | |
3405 | ||
3406 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3407 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3408 | count++; |
3409 | } | |
3410 | if (count != n->nr_partial) | |
3411 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3412 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3413 | ||
3414 | if (!(s->flags & SLAB_STORE_USER)) | |
3415 | goto out; | |
3416 | ||
3417 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3418 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3419 | count++; |
3420 | } | |
3421 | if (count != atomic_long_read(&n->nr_slabs)) | |
3422 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3423 | "counter=%ld\n", s->name, count, | |
3424 | atomic_long_read(&n->nr_slabs)); | |
3425 | ||
3426 | out: | |
3427 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3428 | return count; | |
3429 | } | |
3430 | ||
434e245d | 3431 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3432 | { |
3433 | int node; | |
3434 | unsigned long count = 0; | |
205ab99d | 3435 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3436 | sizeof(unsigned long), GFP_KERNEL); |
3437 | ||
3438 | if (!map) | |
3439 | return -ENOMEM; | |
53e15af0 CL |
3440 | |
3441 | flush_all(s); | |
f64dc58c | 3442 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3443 | struct kmem_cache_node *n = get_node(s, node); |
3444 | ||
434e245d | 3445 | count += validate_slab_node(s, n, map); |
53e15af0 | 3446 | } |
434e245d | 3447 | kfree(map); |
53e15af0 CL |
3448 | return count; |
3449 | } | |
3450 | ||
b3459709 CL |
3451 | #ifdef SLUB_RESILIENCY_TEST |
3452 | static void resiliency_test(void) | |
3453 | { | |
3454 | u8 *p; | |
3455 | ||
3456 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3457 | printk(KERN_ERR "-----------------------\n"); | |
3458 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3459 | ||
3460 | p = kzalloc(16, GFP_KERNEL); | |
3461 | p[16] = 0x12; | |
3462 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3463 | " 0x12->0x%p\n\n", p + 16); | |
3464 | ||
3465 | validate_slab_cache(kmalloc_caches + 4); | |
3466 | ||
3467 | /* Hmmm... The next two are dangerous */ | |
3468 | p = kzalloc(32, GFP_KERNEL); | |
3469 | p[32 + sizeof(void *)] = 0x34; | |
3470 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3adbefee IM |
3471 | " 0x34 -> -0x%p\n", p); |
3472 | printk(KERN_ERR | |
3473 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3474 | |
3475 | validate_slab_cache(kmalloc_caches + 5); | |
3476 | p = kzalloc(64, GFP_KERNEL); | |
3477 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3478 | *p = 0x56; | |
3479 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3480 | p); | |
3adbefee IM |
3481 | printk(KERN_ERR |
3482 | "If allocated object is overwritten then not detectable\n\n"); | |
b3459709 CL |
3483 | validate_slab_cache(kmalloc_caches + 6); |
3484 | ||
3485 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3486 | p = kzalloc(128, GFP_KERNEL); | |
3487 | kfree(p); | |
3488 | *p = 0x78; | |
3489 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3490 | validate_slab_cache(kmalloc_caches + 7); | |
3491 | ||
3492 | p = kzalloc(256, GFP_KERNEL); | |
3493 | kfree(p); | |
3494 | p[50] = 0x9a; | |
3adbefee IM |
3495 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", |
3496 | p); | |
b3459709 CL |
3497 | validate_slab_cache(kmalloc_caches + 8); |
3498 | ||
3499 | p = kzalloc(512, GFP_KERNEL); | |
3500 | kfree(p); | |
3501 | p[512] = 0xab; | |
3502 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3503 | validate_slab_cache(kmalloc_caches + 9); | |
3504 | } | |
3505 | #else | |
3506 | static void resiliency_test(void) {}; | |
3507 | #endif | |
3508 | ||
88a420e4 | 3509 | /* |
672bba3a | 3510 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3511 | * and freed. |
3512 | */ | |
3513 | ||
3514 | struct location { | |
3515 | unsigned long count; | |
ce71e27c | 3516 | unsigned long addr; |
45edfa58 CL |
3517 | long long sum_time; |
3518 | long min_time; | |
3519 | long max_time; | |
3520 | long min_pid; | |
3521 | long max_pid; | |
174596a0 | 3522 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3523 | nodemask_t nodes; |
88a420e4 CL |
3524 | }; |
3525 | ||
3526 | struct loc_track { | |
3527 | unsigned long max; | |
3528 | unsigned long count; | |
3529 | struct location *loc; | |
3530 | }; | |
3531 | ||
3532 | static void free_loc_track(struct loc_track *t) | |
3533 | { | |
3534 | if (t->max) | |
3535 | free_pages((unsigned long)t->loc, | |
3536 | get_order(sizeof(struct location) * t->max)); | |
3537 | } | |
3538 | ||
68dff6a9 | 3539 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3540 | { |
3541 | struct location *l; | |
3542 | int order; | |
3543 | ||
88a420e4 CL |
3544 | order = get_order(sizeof(struct location) * max); |
3545 | ||
68dff6a9 | 3546 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3547 | if (!l) |
3548 | return 0; | |
3549 | ||
3550 | if (t->count) { | |
3551 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3552 | free_loc_track(t); | |
3553 | } | |
3554 | t->max = max; | |
3555 | t->loc = l; | |
3556 | return 1; | |
3557 | } | |
3558 | ||
3559 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3560 | const struct track *track) |
88a420e4 CL |
3561 | { |
3562 | long start, end, pos; | |
3563 | struct location *l; | |
ce71e27c | 3564 | unsigned long caddr; |
45edfa58 | 3565 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3566 | |
3567 | start = -1; | |
3568 | end = t->count; | |
3569 | ||
3570 | for ( ; ; ) { | |
3571 | pos = start + (end - start + 1) / 2; | |
3572 | ||
3573 | /* | |
3574 | * There is nothing at "end". If we end up there | |
3575 | * we need to add something to before end. | |
3576 | */ | |
3577 | if (pos == end) | |
3578 | break; | |
3579 | ||
3580 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3581 | if (track->addr == caddr) { |
3582 | ||
3583 | l = &t->loc[pos]; | |
3584 | l->count++; | |
3585 | if (track->when) { | |
3586 | l->sum_time += age; | |
3587 | if (age < l->min_time) | |
3588 | l->min_time = age; | |
3589 | if (age > l->max_time) | |
3590 | l->max_time = age; | |
3591 | ||
3592 | if (track->pid < l->min_pid) | |
3593 | l->min_pid = track->pid; | |
3594 | if (track->pid > l->max_pid) | |
3595 | l->max_pid = track->pid; | |
3596 | ||
174596a0 RR |
3597 | cpumask_set_cpu(track->cpu, |
3598 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3599 | } |
3600 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3601 | return 1; |
3602 | } | |
3603 | ||
45edfa58 | 3604 | if (track->addr < caddr) |
88a420e4 CL |
3605 | end = pos; |
3606 | else | |
3607 | start = pos; | |
3608 | } | |
3609 | ||
3610 | /* | |
672bba3a | 3611 | * Not found. Insert new tracking element. |
88a420e4 | 3612 | */ |
68dff6a9 | 3613 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3614 | return 0; |
3615 | ||
3616 | l = t->loc + pos; | |
3617 | if (pos < t->count) | |
3618 | memmove(l + 1, l, | |
3619 | (t->count - pos) * sizeof(struct location)); | |
3620 | t->count++; | |
3621 | l->count = 1; | |
45edfa58 CL |
3622 | l->addr = track->addr; |
3623 | l->sum_time = age; | |
3624 | l->min_time = age; | |
3625 | l->max_time = age; | |
3626 | l->min_pid = track->pid; | |
3627 | l->max_pid = track->pid; | |
174596a0 RR |
3628 | cpumask_clear(to_cpumask(l->cpus)); |
3629 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3630 | nodes_clear(l->nodes); |
3631 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3632 | return 1; |
3633 | } | |
3634 | ||
3635 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
3636 | struct page *page, enum track_item alloc) | |
3637 | { | |
a973e9dd | 3638 | void *addr = page_address(page); |
39b26464 | 3639 | DECLARE_BITMAP(map, page->objects); |
88a420e4 CL |
3640 | void *p; |
3641 | ||
39b26464 | 3642 | bitmap_zero(map, page->objects); |
7656c72b CL |
3643 | for_each_free_object(p, s, page->freelist) |
3644 | set_bit(slab_index(p, s, addr), map); | |
88a420e4 | 3645 | |
224a88be | 3646 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3647 | if (!test_bit(slab_index(p, s, addr), map)) |
3648 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3649 | } |
3650 | ||
3651 | static int list_locations(struct kmem_cache *s, char *buf, | |
3652 | enum track_item alloc) | |
3653 | { | |
e374d483 | 3654 | int len = 0; |
88a420e4 | 3655 | unsigned long i; |
68dff6a9 | 3656 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 CL |
3657 | int node; |
3658 | ||
68dff6a9 | 3659 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
ea3061d2 | 3660 | GFP_TEMPORARY)) |
68dff6a9 | 3661 | return sprintf(buf, "Out of memory\n"); |
88a420e4 CL |
3662 | |
3663 | /* Push back cpu slabs */ | |
3664 | flush_all(s); | |
3665 | ||
f64dc58c | 3666 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3667 | struct kmem_cache_node *n = get_node(s, node); |
3668 | unsigned long flags; | |
3669 | struct page *page; | |
3670 | ||
9e86943b | 3671 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3672 | continue; |
3673 | ||
3674 | spin_lock_irqsave(&n->list_lock, flags); | |
3675 | list_for_each_entry(page, &n->partial, lru) | |
3676 | process_slab(&t, s, page, alloc); | |
3677 | list_for_each_entry(page, &n->full, lru) | |
3678 | process_slab(&t, s, page, alloc); | |
3679 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3680 | } | |
3681 | ||
3682 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3683 | struct location *l = &t.loc[i]; |
88a420e4 | 3684 | |
9c246247 | 3685 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3686 | break; |
e374d483 | 3687 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3688 | |
3689 | if (l->addr) | |
e374d483 | 3690 | len += sprint_symbol(buf + len, (unsigned long)l->addr); |
88a420e4 | 3691 | else |
e374d483 | 3692 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3693 | |
3694 | if (l->sum_time != l->min_time) { | |
e374d483 | 3695 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3696 | l->min_time, |
3697 | (long)div_u64(l->sum_time, l->count), | |
3698 | l->max_time); | |
45edfa58 | 3699 | } else |
e374d483 | 3700 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3701 | l->min_time); |
3702 | ||
3703 | if (l->min_pid != l->max_pid) | |
e374d483 | 3704 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3705 | l->min_pid, l->max_pid); |
3706 | else | |
e374d483 | 3707 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3708 | l->min_pid); |
3709 | ||
174596a0 RR |
3710 | if (num_online_cpus() > 1 && |
3711 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3712 | len < PAGE_SIZE - 60) { |
3713 | len += sprintf(buf + len, " cpus="); | |
3714 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3715 | to_cpumask(l->cpus)); |
45edfa58 CL |
3716 | } |
3717 | ||
84966343 | 3718 | if (num_online_nodes() > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3719 | len < PAGE_SIZE - 60) { |
3720 | len += sprintf(buf + len, " nodes="); | |
3721 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3722 | l->nodes); |
3723 | } | |
3724 | ||
e374d483 | 3725 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3726 | } |
3727 | ||
3728 | free_loc_track(&t); | |
3729 | if (!t.count) | |
e374d483 HH |
3730 | len += sprintf(buf, "No data\n"); |
3731 | return len; | |
88a420e4 CL |
3732 | } |
3733 | ||
81819f0f | 3734 | enum slab_stat_type { |
205ab99d CL |
3735 | SL_ALL, /* All slabs */ |
3736 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3737 | SL_CPU, /* Only slabs used for cpu caches */ | |
3738 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3739 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3740 | }; |
3741 | ||
205ab99d | 3742 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
3743 | #define SO_PARTIAL (1 << SL_PARTIAL) |
3744 | #define SO_CPU (1 << SL_CPU) | |
3745 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 3746 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 3747 | |
62e5c4b4 CG |
3748 | static ssize_t show_slab_objects(struct kmem_cache *s, |
3749 | char *buf, unsigned long flags) | |
81819f0f CL |
3750 | { |
3751 | unsigned long total = 0; | |
81819f0f CL |
3752 | int node; |
3753 | int x; | |
3754 | unsigned long *nodes; | |
3755 | unsigned long *per_cpu; | |
3756 | ||
3757 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
3758 | if (!nodes) |
3759 | return -ENOMEM; | |
81819f0f CL |
3760 | per_cpu = nodes + nr_node_ids; |
3761 | ||
205ab99d CL |
3762 | if (flags & SO_CPU) { |
3763 | int cpu; | |
81819f0f | 3764 | |
205ab99d CL |
3765 | for_each_possible_cpu(cpu) { |
3766 | struct kmem_cache_cpu *c = get_cpu_slab(s, cpu); | |
dfb4f096 | 3767 | |
205ab99d CL |
3768 | if (!c || c->node < 0) |
3769 | continue; | |
3770 | ||
3771 | if (c->page) { | |
3772 | if (flags & SO_TOTAL) | |
3773 | x = c->page->objects; | |
3774 | else if (flags & SO_OBJECTS) | |
3775 | x = c->page->inuse; | |
81819f0f CL |
3776 | else |
3777 | x = 1; | |
205ab99d | 3778 | |
81819f0f | 3779 | total += x; |
205ab99d | 3780 | nodes[c->node] += x; |
81819f0f | 3781 | } |
205ab99d | 3782 | per_cpu[c->node]++; |
81819f0f CL |
3783 | } |
3784 | } | |
3785 | ||
205ab99d CL |
3786 | if (flags & SO_ALL) { |
3787 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3788 | struct kmem_cache_node *n = get_node(s, node); | |
3789 | ||
3790 | if (flags & SO_TOTAL) | |
3791 | x = atomic_long_read(&n->total_objects); | |
3792 | else if (flags & SO_OBJECTS) | |
3793 | x = atomic_long_read(&n->total_objects) - | |
3794 | count_partial(n, count_free); | |
81819f0f | 3795 | |
81819f0f | 3796 | else |
205ab99d | 3797 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
3798 | total += x; |
3799 | nodes[node] += x; | |
3800 | } | |
3801 | ||
205ab99d CL |
3802 | } else if (flags & SO_PARTIAL) { |
3803 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3804 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 3805 | |
205ab99d CL |
3806 | if (flags & SO_TOTAL) |
3807 | x = count_partial(n, count_total); | |
3808 | else if (flags & SO_OBJECTS) | |
3809 | x = count_partial(n, count_inuse); | |
81819f0f | 3810 | else |
205ab99d | 3811 | x = n->nr_partial; |
81819f0f CL |
3812 | total += x; |
3813 | nodes[node] += x; | |
3814 | } | |
3815 | } | |
81819f0f CL |
3816 | x = sprintf(buf, "%lu", total); |
3817 | #ifdef CONFIG_NUMA | |
f64dc58c | 3818 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
3819 | if (nodes[node]) |
3820 | x += sprintf(buf + x, " N%d=%lu", | |
3821 | node, nodes[node]); | |
3822 | #endif | |
3823 | kfree(nodes); | |
3824 | return x + sprintf(buf + x, "\n"); | |
3825 | } | |
3826 | ||
3827 | static int any_slab_objects(struct kmem_cache *s) | |
3828 | { | |
3829 | int node; | |
81819f0f | 3830 | |
dfb4f096 | 3831 | for_each_online_node(node) { |
81819f0f CL |
3832 | struct kmem_cache_node *n = get_node(s, node); |
3833 | ||
dfb4f096 CL |
3834 | if (!n) |
3835 | continue; | |
3836 | ||
4ea33e2d | 3837 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
3838 | return 1; |
3839 | } | |
3840 | return 0; | |
3841 | } | |
3842 | ||
3843 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
3844 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
3845 | ||
3846 | struct slab_attribute { | |
3847 | struct attribute attr; | |
3848 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
3849 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
3850 | }; | |
3851 | ||
3852 | #define SLAB_ATTR_RO(_name) \ | |
3853 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
3854 | ||
3855 | #define SLAB_ATTR(_name) \ | |
3856 | static struct slab_attribute _name##_attr = \ | |
3857 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
3858 | ||
81819f0f CL |
3859 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
3860 | { | |
3861 | return sprintf(buf, "%d\n", s->size); | |
3862 | } | |
3863 | SLAB_ATTR_RO(slab_size); | |
3864 | ||
3865 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
3866 | { | |
3867 | return sprintf(buf, "%d\n", s->align); | |
3868 | } | |
3869 | SLAB_ATTR_RO(align); | |
3870 | ||
3871 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
3872 | { | |
3873 | return sprintf(buf, "%d\n", s->objsize); | |
3874 | } | |
3875 | SLAB_ATTR_RO(object_size); | |
3876 | ||
3877 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
3878 | { | |
834f3d11 | 3879 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
3880 | } |
3881 | SLAB_ATTR_RO(objs_per_slab); | |
3882 | ||
06b285dc CL |
3883 | static ssize_t order_store(struct kmem_cache *s, |
3884 | const char *buf, size_t length) | |
3885 | { | |
0121c619 CL |
3886 | unsigned long order; |
3887 | int err; | |
3888 | ||
3889 | err = strict_strtoul(buf, 10, &order); | |
3890 | if (err) | |
3891 | return err; | |
06b285dc CL |
3892 | |
3893 | if (order > slub_max_order || order < slub_min_order) | |
3894 | return -EINVAL; | |
3895 | ||
3896 | calculate_sizes(s, order); | |
3897 | return length; | |
3898 | } | |
3899 | ||
81819f0f CL |
3900 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
3901 | { | |
834f3d11 | 3902 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 3903 | } |
06b285dc | 3904 | SLAB_ATTR(order); |
81819f0f | 3905 | |
73d342b1 DR |
3906 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
3907 | { | |
3908 | return sprintf(buf, "%lu\n", s->min_partial); | |
3909 | } | |
3910 | ||
3911 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
3912 | size_t length) | |
3913 | { | |
3914 | unsigned long min; | |
3915 | int err; | |
3916 | ||
3917 | err = strict_strtoul(buf, 10, &min); | |
3918 | if (err) | |
3919 | return err; | |
3920 | ||
c0bdb232 | 3921 | set_min_partial(s, min); |
73d342b1 DR |
3922 | return length; |
3923 | } | |
3924 | SLAB_ATTR(min_partial); | |
3925 | ||
81819f0f CL |
3926 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
3927 | { | |
3928 | if (s->ctor) { | |
3929 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
3930 | ||
3931 | return n + sprintf(buf + n, "\n"); | |
3932 | } | |
3933 | return 0; | |
3934 | } | |
3935 | SLAB_ATTR_RO(ctor); | |
3936 | ||
81819f0f CL |
3937 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
3938 | { | |
3939 | return sprintf(buf, "%d\n", s->refcount - 1); | |
3940 | } | |
3941 | SLAB_ATTR_RO(aliases); | |
3942 | ||
3943 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
3944 | { | |
205ab99d | 3945 | return show_slab_objects(s, buf, SO_ALL); |
81819f0f CL |
3946 | } |
3947 | SLAB_ATTR_RO(slabs); | |
3948 | ||
3949 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
3950 | { | |
d9acf4b7 | 3951 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
3952 | } |
3953 | SLAB_ATTR_RO(partial); | |
3954 | ||
3955 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
3956 | { | |
d9acf4b7 | 3957 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
3958 | } |
3959 | SLAB_ATTR_RO(cpu_slabs); | |
3960 | ||
3961 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
3962 | { | |
205ab99d | 3963 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
3964 | } |
3965 | SLAB_ATTR_RO(objects); | |
3966 | ||
205ab99d CL |
3967 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
3968 | { | |
3969 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
3970 | } | |
3971 | SLAB_ATTR_RO(objects_partial); | |
3972 | ||
3973 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) | |
3974 | { | |
3975 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
3976 | } | |
3977 | SLAB_ATTR_RO(total_objects); | |
3978 | ||
81819f0f CL |
3979 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
3980 | { | |
3981 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
3982 | } | |
3983 | ||
3984 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
3985 | const char *buf, size_t length) | |
3986 | { | |
3987 | s->flags &= ~SLAB_DEBUG_FREE; | |
3988 | if (buf[0] == '1') | |
3989 | s->flags |= SLAB_DEBUG_FREE; | |
3990 | return length; | |
3991 | } | |
3992 | SLAB_ATTR(sanity_checks); | |
3993 | ||
3994 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
3995 | { | |
3996 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
3997 | } | |
3998 | ||
3999 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4000 | size_t length) | |
4001 | { | |
4002 | s->flags &= ~SLAB_TRACE; | |
4003 | if (buf[0] == '1') | |
4004 | s->flags |= SLAB_TRACE; | |
4005 | return length; | |
4006 | } | |
4007 | SLAB_ATTR(trace); | |
4008 | ||
4009 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
4010 | { | |
4011 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4012 | } | |
4013 | ||
4014 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4015 | const char *buf, size_t length) | |
4016 | { | |
4017 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4018 | if (buf[0] == '1') | |
4019 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4020 | return length; | |
4021 | } | |
4022 | SLAB_ATTR(reclaim_account); | |
4023 | ||
4024 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4025 | { | |
5af60839 | 4026 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
81819f0f CL |
4027 | } |
4028 | SLAB_ATTR_RO(hwcache_align); | |
4029 | ||
4030 | #ifdef CONFIG_ZONE_DMA | |
4031 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4032 | { | |
4033 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4034 | } | |
4035 | SLAB_ATTR_RO(cache_dma); | |
4036 | #endif | |
4037 | ||
4038 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4039 | { | |
4040 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4041 | } | |
4042 | SLAB_ATTR_RO(destroy_by_rcu); | |
4043 | ||
4044 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
4045 | { | |
4046 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4047 | } | |
4048 | ||
4049 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4050 | const char *buf, size_t length) | |
4051 | { | |
4052 | if (any_slab_objects(s)) | |
4053 | return -EBUSY; | |
4054 | ||
4055 | s->flags &= ~SLAB_RED_ZONE; | |
4056 | if (buf[0] == '1') | |
4057 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4058 | calculate_sizes(s, -1); |
81819f0f CL |
4059 | return length; |
4060 | } | |
4061 | SLAB_ATTR(red_zone); | |
4062 | ||
4063 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4064 | { | |
4065 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4066 | } | |
4067 | ||
4068 | static ssize_t poison_store(struct kmem_cache *s, | |
4069 | const char *buf, size_t length) | |
4070 | { | |
4071 | if (any_slab_objects(s)) | |
4072 | return -EBUSY; | |
4073 | ||
4074 | s->flags &= ~SLAB_POISON; | |
4075 | if (buf[0] == '1') | |
4076 | s->flags |= SLAB_POISON; | |
06b285dc | 4077 | calculate_sizes(s, -1); |
81819f0f CL |
4078 | return length; |
4079 | } | |
4080 | SLAB_ATTR(poison); | |
4081 | ||
4082 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4083 | { | |
4084 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4085 | } | |
4086 | ||
4087 | static ssize_t store_user_store(struct kmem_cache *s, | |
4088 | const char *buf, size_t length) | |
4089 | { | |
4090 | if (any_slab_objects(s)) | |
4091 | return -EBUSY; | |
4092 | ||
4093 | s->flags &= ~SLAB_STORE_USER; | |
4094 | if (buf[0] == '1') | |
4095 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4096 | calculate_sizes(s, -1); |
81819f0f CL |
4097 | return length; |
4098 | } | |
4099 | SLAB_ATTR(store_user); | |
4100 | ||
53e15af0 CL |
4101 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4102 | { | |
4103 | return 0; | |
4104 | } | |
4105 | ||
4106 | static ssize_t validate_store(struct kmem_cache *s, | |
4107 | const char *buf, size_t length) | |
4108 | { | |
434e245d CL |
4109 | int ret = -EINVAL; |
4110 | ||
4111 | if (buf[0] == '1') { | |
4112 | ret = validate_slab_cache(s); | |
4113 | if (ret >= 0) | |
4114 | ret = length; | |
4115 | } | |
4116 | return ret; | |
53e15af0 CL |
4117 | } |
4118 | SLAB_ATTR(validate); | |
4119 | ||
2086d26a CL |
4120 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4121 | { | |
4122 | return 0; | |
4123 | } | |
4124 | ||
4125 | static ssize_t shrink_store(struct kmem_cache *s, | |
4126 | const char *buf, size_t length) | |
4127 | { | |
4128 | if (buf[0] == '1') { | |
4129 | int rc = kmem_cache_shrink(s); | |
4130 | ||
4131 | if (rc) | |
4132 | return rc; | |
4133 | } else | |
4134 | return -EINVAL; | |
4135 | return length; | |
4136 | } | |
4137 | SLAB_ATTR(shrink); | |
4138 | ||
88a420e4 CL |
4139 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) |
4140 | { | |
4141 | if (!(s->flags & SLAB_STORE_USER)) | |
4142 | return -ENOSYS; | |
4143 | return list_locations(s, buf, TRACK_ALLOC); | |
4144 | } | |
4145 | SLAB_ATTR_RO(alloc_calls); | |
4146 | ||
4147 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4148 | { | |
4149 | if (!(s->flags & SLAB_STORE_USER)) | |
4150 | return -ENOSYS; | |
4151 | return list_locations(s, buf, TRACK_FREE); | |
4152 | } | |
4153 | SLAB_ATTR_RO(free_calls); | |
4154 | ||
81819f0f | 4155 | #ifdef CONFIG_NUMA |
9824601e | 4156 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4157 | { |
9824601e | 4158 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4159 | } |
4160 | ||
9824601e | 4161 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4162 | const char *buf, size_t length) |
4163 | { | |
0121c619 CL |
4164 | unsigned long ratio; |
4165 | int err; | |
4166 | ||
4167 | err = strict_strtoul(buf, 10, &ratio); | |
4168 | if (err) | |
4169 | return err; | |
4170 | ||
e2cb96b7 | 4171 | if (ratio <= 100) |
0121c619 | 4172 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4173 | |
81819f0f CL |
4174 | return length; |
4175 | } | |
9824601e | 4176 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4177 | #endif |
4178 | ||
8ff12cfc | 4179 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4180 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4181 | { | |
4182 | unsigned long sum = 0; | |
4183 | int cpu; | |
4184 | int len; | |
4185 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4186 | ||
4187 | if (!data) | |
4188 | return -ENOMEM; | |
4189 | ||
4190 | for_each_online_cpu(cpu) { | |
4191 | unsigned x = get_cpu_slab(s, cpu)->stat[si]; | |
4192 | ||
4193 | data[cpu] = x; | |
4194 | sum += x; | |
4195 | } | |
4196 | ||
4197 | len = sprintf(buf, "%lu", sum); | |
4198 | ||
50ef37b9 | 4199 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4200 | for_each_online_cpu(cpu) { |
4201 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4202 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4203 | } |
50ef37b9 | 4204 | #endif |
8ff12cfc CL |
4205 | kfree(data); |
4206 | return len + sprintf(buf + len, "\n"); | |
4207 | } | |
4208 | ||
4209 | #define STAT_ATTR(si, text) \ | |
4210 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4211 | { \ | |
4212 | return show_stat(s, buf, si); \ | |
4213 | } \ | |
4214 | SLAB_ATTR_RO(text); \ | |
4215 | ||
4216 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4217 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4218 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4219 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4220 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4221 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4222 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4223 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4224 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4225 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4226 | STAT_ATTR(FREE_SLAB, free_slab); | |
4227 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4228 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4229 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4230 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4231 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4232 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4233 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4234 | #endif |
4235 | ||
06428780 | 4236 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4237 | &slab_size_attr.attr, |
4238 | &object_size_attr.attr, | |
4239 | &objs_per_slab_attr.attr, | |
4240 | &order_attr.attr, | |
73d342b1 | 4241 | &min_partial_attr.attr, |
81819f0f | 4242 | &objects_attr.attr, |
205ab99d CL |
4243 | &objects_partial_attr.attr, |
4244 | &total_objects_attr.attr, | |
81819f0f CL |
4245 | &slabs_attr.attr, |
4246 | &partial_attr.attr, | |
4247 | &cpu_slabs_attr.attr, | |
4248 | &ctor_attr.attr, | |
81819f0f CL |
4249 | &aliases_attr.attr, |
4250 | &align_attr.attr, | |
4251 | &sanity_checks_attr.attr, | |
4252 | &trace_attr.attr, | |
4253 | &hwcache_align_attr.attr, | |
4254 | &reclaim_account_attr.attr, | |
4255 | &destroy_by_rcu_attr.attr, | |
4256 | &red_zone_attr.attr, | |
4257 | &poison_attr.attr, | |
4258 | &store_user_attr.attr, | |
53e15af0 | 4259 | &validate_attr.attr, |
2086d26a | 4260 | &shrink_attr.attr, |
88a420e4 CL |
4261 | &alloc_calls_attr.attr, |
4262 | &free_calls_attr.attr, | |
81819f0f CL |
4263 | #ifdef CONFIG_ZONE_DMA |
4264 | &cache_dma_attr.attr, | |
4265 | #endif | |
4266 | #ifdef CONFIG_NUMA | |
9824601e | 4267 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4268 | #endif |
4269 | #ifdef CONFIG_SLUB_STATS | |
4270 | &alloc_fastpath_attr.attr, | |
4271 | &alloc_slowpath_attr.attr, | |
4272 | &free_fastpath_attr.attr, | |
4273 | &free_slowpath_attr.attr, | |
4274 | &free_frozen_attr.attr, | |
4275 | &free_add_partial_attr.attr, | |
4276 | &free_remove_partial_attr.attr, | |
4277 | &alloc_from_partial_attr.attr, | |
4278 | &alloc_slab_attr.attr, | |
4279 | &alloc_refill_attr.attr, | |
4280 | &free_slab_attr.attr, | |
4281 | &cpuslab_flush_attr.attr, | |
4282 | &deactivate_full_attr.attr, | |
4283 | &deactivate_empty_attr.attr, | |
4284 | &deactivate_to_head_attr.attr, | |
4285 | &deactivate_to_tail_attr.attr, | |
4286 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4287 | &order_fallback_attr.attr, |
81819f0f CL |
4288 | #endif |
4289 | NULL | |
4290 | }; | |
4291 | ||
4292 | static struct attribute_group slab_attr_group = { | |
4293 | .attrs = slab_attrs, | |
4294 | }; | |
4295 | ||
4296 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4297 | struct attribute *attr, | |
4298 | char *buf) | |
4299 | { | |
4300 | struct slab_attribute *attribute; | |
4301 | struct kmem_cache *s; | |
4302 | int err; | |
4303 | ||
4304 | attribute = to_slab_attr(attr); | |
4305 | s = to_slab(kobj); | |
4306 | ||
4307 | if (!attribute->show) | |
4308 | return -EIO; | |
4309 | ||
4310 | err = attribute->show(s, buf); | |
4311 | ||
4312 | return err; | |
4313 | } | |
4314 | ||
4315 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4316 | struct attribute *attr, | |
4317 | const char *buf, size_t len) | |
4318 | { | |
4319 | struct slab_attribute *attribute; | |
4320 | struct kmem_cache *s; | |
4321 | int err; | |
4322 | ||
4323 | attribute = to_slab_attr(attr); | |
4324 | s = to_slab(kobj); | |
4325 | ||
4326 | if (!attribute->store) | |
4327 | return -EIO; | |
4328 | ||
4329 | err = attribute->store(s, buf, len); | |
4330 | ||
4331 | return err; | |
4332 | } | |
4333 | ||
151c602f CL |
4334 | static void kmem_cache_release(struct kobject *kobj) |
4335 | { | |
4336 | struct kmem_cache *s = to_slab(kobj); | |
4337 | ||
4338 | kfree(s); | |
4339 | } | |
4340 | ||
81819f0f CL |
4341 | static struct sysfs_ops slab_sysfs_ops = { |
4342 | .show = slab_attr_show, | |
4343 | .store = slab_attr_store, | |
4344 | }; | |
4345 | ||
4346 | static struct kobj_type slab_ktype = { | |
4347 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4348 | .release = kmem_cache_release |
81819f0f CL |
4349 | }; |
4350 | ||
4351 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4352 | { | |
4353 | struct kobj_type *ktype = get_ktype(kobj); | |
4354 | ||
4355 | if (ktype == &slab_ktype) | |
4356 | return 1; | |
4357 | return 0; | |
4358 | } | |
4359 | ||
4360 | static struct kset_uevent_ops slab_uevent_ops = { | |
4361 | .filter = uevent_filter, | |
4362 | }; | |
4363 | ||
27c3a314 | 4364 | static struct kset *slab_kset; |
81819f0f CL |
4365 | |
4366 | #define ID_STR_LENGTH 64 | |
4367 | ||
4368 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4369 | * |
4370 | * Format :[flags-]size | |
81819f0f CL |
4371 | */ |
4372 | static char *create_unique_id(struct kmem_cache *s) | |
4373 | { | |
4374 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4375 | char *p = name; | |
4376 | ||
4377 | BUG_ON(!name); | |
4378 | ||
4379 | *p++ = ':'; | |
4380 | /* | |
4381 | * First flags affecting slabcache operations. We will only | |
4382 | * get here for aliasable slabs so we do not need to support | |
4383 | * too many flags. The flags here must cover all flags that | |
4384 | * are matched during merging to guarantee that the id is | |
4385 | * unique. | |
4386 | */ | |
4387 | if (s->flags & SLAB_CACHE_DMA) | |
4388 | *p++ = 'd'; | |
4389 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4390 | *p++ = 'a'; | |
4391 | if (s->flags & SLAB_DEBUG_FREE) | |
4392 | *p++ = 'F'; | |
4393 | if (p != name + 1) | |
4394 | *p++ = '-'; | |
4395 | p += sprintf(p, "%07d", s->size); | |
4396 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4397 | return name; | |
4398 | } | |
4399 | ||
4400 | static int sysfs_slab_add(struct kmem_cache *s) | |
4401 | { | |
4402 | int err; | |
4403 | const char *name; | |
4404 | int unmergeable; | |
4405 | ||
4406 | if (slab_state < SYSFS) | |
4407 | /* Defer until later */ | |
4408 | return 0; | |
4409 | ||
4410 | unmergeable = slab_unmergeable(s); | |
4411 | if (unmergeable) { | |
4412 | /* | |
4413 | * Slabcache can never be merged so we can use the name proper. | |
4414 | * This is typically the case for debug situations. In that | |
4415 | * case we can catch duplicate names easily. | |
4416 | */ | |
27c3a314 | 4417 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4418 | name = s->name; |
4419 | } else { | |
4420 | /* | |
4421 | * Create a unique name for the slab as a target | |
4422 | * for the symlinks. | |
4423 | */ | |
4424 | name = create_unique_id(s); | |
4425 | } | |
4426 | ||
27c3a314 | 4427 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4428 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4429 | if (err) { | |
4430 | kobject_put(&s->kobj); | |
81819f0f | 4431 | return err; |
1eada11c | 4432 | } |
81819f0f CL |
4433 | |
4434 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
4435 | if (err) | |
4436 | return err; | |
4437 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
4438 | if (!unmergeable) { | |
4439 | /* Setup first alias */ | |
4440 | sysfs_slab_alias(s, s->name); | |
4441 | kfree(name); | |
4442 | } | |
4443 | return 0; | |
4444 | } | |
4445 | ||
4446 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4447 | { | |
4448 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
4449 | kobject_del(&s->kobj); | |
151c602f | 4450 | kobject_put(&s->kobj); |
81819f0f CL |
4451 | } |
4452 | ||
4453 | /* | |
4454 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4455 | * available lest we lose that information. |
81819f0f CL |
4456 | */ |
4457 | struct saved_alias { | |
4458 | struct kmem_cache *s; | |
4459 | const char *name; | |
4460 | struct saved_alias *next; | |
4461 | }; | |
4462 | ||
5af328a5 | 4463 | static struct saved_alias *alias_list; |
81819f0f CL |
4464 | |
4465 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4466 | { | |
4467 | struct saved_alias *al; | |
4468 | ||
4469 | if (slab_state == SYSFS) { | |
4470 | /* | |
4471 | * If we have a leftover link then remove it. | |
4472 | */ | |
27c3a314 GKH |
4473 | sysfs_remove_link(&slab_kset->kobj, name); |
4474 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4475 | } |
4476 | ||
4477 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4478 | if (!al) | |
4479 | return -ENOMEM; | |
4480 | ||
4481 | al->s = s; | |
4482 | al->name = name; | |
4483 | al->next = alias_list; | |
4484 | alias_list = al; | |
4485 | return 0; | |
4486 | } | |
4487 | ||
4488 | static int __init slab_sysfs_init(void) | |
4489 | { | |
5b95a4ac | 4490 | struct kmem_cache *s; |
81819f0f CL |
4491 | int err; |
4492 | ||
0ff21e46 | 4493 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4494 | if (!slab_kset) { |
81819f0f CL |
4495 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4496 | return -ENOSYS; | |
4497 | } | |
4498 | ||
26a7bd03 CL |
4499 | slab_state = SYSFS; |
4500 | ||
5b95a4ac | 4501 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4502 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4503 | if (err) |
4504 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4505 | " to sysfs\n", s->name); | |
26a7bd03 | 4506 | } |
81819f0f CL |
4507 | |
4508 | while (alias_list) { | |
4509 | struct saved_alias *al = alias_list; | |
4510 | ||
4511 | alias_list = alias_list->next; | |
4512 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4513 | if (err) |
4514 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4515 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4516 | kfree(al); |
4517 | } | |
4518 | ||
4519 | resiliency_test(); | |
4520 | return 0; | |
4521 | } | |
4522 | ||
4523 | __initcall(slab_sysfs_init); | |
81819f0f | 4524 | #endif |
57ed3eda PE |
4525 | |
4526 | /* | |
4527 | * The /proc/slabinfo ABI | |
4528 | */ | |
158a9624 | 4529 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4530 | static void print_slabinfo_header(struct seq_file *m) |
4531 | { | |
4532 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4533 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4534 | "<objperslab> <pagesperslab>"); | |
4535 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4536 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4537 | seq_putc(m, '\n'); | |
4538 | } | |
4539 | ||
4540 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4541 | { | |
4542 | loff_t n = *pos; | |
4543 | ||
4544 | down_read(&slub_lock); | |
4545 | if (!n) | |
4546 | print_slabinfo_header(m); | |
4547 | ||
4548 | return seq_list_start(&slab_caches, *pos); | |
4549 | } | |
4550 | ||
4551 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4552 | { | |
4553 | return seq_list_next(p, &slab_caches, pos); | |
4554 | } | |
4555 | ||
4556 | static void s_stop(struct seq_file *m, void *p) | |
4557 | { | |
4558 | up_read(&slub_lock); | |
4559 | } | |
4560 | ||
4561 | static int s_show(struct seq_file *m, void *p) | |
4562 | { | |
4563 | unsigned long nr_partials = 0; | |
4564 | unsigned long nr_slabs = 0; | |
4565 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4566 | unsigned long nr_objs = 0; |
4567 | unsigned long nr_free = 0; | |
57ed3eda PE |
4568 | struct kmem_cache *s; |
4569 | int node; | |
4570 | ||
4571 | s = list_entry(p, struct kmem_cache, list); | |
4572 | ||
4573 | for_each_online_node(node) { | |
4574 | struct kmem_cache_node *n = get_node(s, node); | |
4575 | ||
4576 | if (!n) | |
4577 | continue; | |
4578 | ||
4579 | nr_partials += n->nr_partial; | |
4580 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4581 | nr_objs += atomic_long_read(&n->total_objects); |
4582 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4583 | } |
4584 | ||
205ab99d | 4585 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4586 | |
4587 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4588 | nr_objs, s->size, oo_objects(s->oo), |
4589 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4590 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4591 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4592 | 0UL); | |
4593 | seq_putc(m, '\n'); | |
4594 | return 0; | |
4595 | } | |
4596 | ||
7b3c3a50 | 4597 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4598 | .start = s_start, |
4599 | .next = s_next, | |
4600 | .stop = s_stop, | |
4601 | .show = s_show, | |
4602 | }; | |
4603 | ||
7b3c3a50 AD |
4604 | static int slabinfo_open(struct inode *inode, struct file *file) |
4605 | { | |
4606 | return seq_open(file, &slabinfo_op); | |
4607 | } | |
4608 | ||
4609 | static const struct file_operations proc_slabinfo_operations = { | |
4610 | .open = slabinfo_open, | |
4611 | .read = seq_read, | |
4612 | .llseek = seq_lseek, | |
4613 | .release = seq_release, | |
4614 | }; | |
4615 | ||
4616 | static int __init slab_proc_init(void) | |
4617 | { | |
4618 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); | |
4619 | return 0; | |
4620 | } | |
4621 | module_init(slab_proc_init); | |
158a9624 | 4622 | #endif /* CONFIG_SLABINFO */ |