]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
9 | */ |
10 | ||
11 | #include <linux/mm.h> | |
1eb5ac64 | 12 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
13 | #include <linux/module.h> |
14 | #include <linux/bit_spinlock.h> | |
15 | #include <linux/interrupt.h> | |
16 | #include <linux/bitops.h> | |
17 | #include <linux/slab.h> | |
7b3c3a50 | 18 | #include <linux/proc_fs.h> |
81819f0f | 19 | #include <linux/seq_file.h> |
5a896d9e | 20 | #include <linux/kmemcheck.h> |
81819f0f CL |
21 | #include <linux/cpu.h> |
22 | #include <linux/cpuset.h> | |
23 | #include <linux/mempolicy.h> | |
24 | #include <linux/ctype.h> | |
3ac7fe5a | 25 | #include <linux/debugobjects.h> |
81819f0f | 26 | #include <linux/kallsyms.h> |
b9049e23 | 27 | #include <linux/memory.h> |
f8bd2258 | 28 | #include <linux/math64.h> |
773ff60e | 29 | #include <linux/fault-inject.h> |
81819f0f | 30 | |
4a92379b RK |
31 | #include <trace/events/kmem.h> |
32 | ||
81819f0f CL |
33 | /* |
34 | * Lock order: | |
35 | * 1. slab_lock(page) | |
36 | * 2. slab->list_lock | |
37 | * | |
38 | * The slab_lock protects operations on the object of a particular | |
39 | * slab and its metadata in the page struct. If the slab lock | |
40 | * has been taken then no allocations nor frees can be performed | |
41 | * on the objects in the slab nor can the slab be added or removed | |
42 | * from the partial or full lists since this would mean modifying | |
43 | * the page_struct of the slab. | |
44 | * | |
45 | * The list_lock protects the partial and full list on each node and | |
46 | * the partial slab counter. If taken then no new slabs may be added or | |
47 | * removed from the lists nor make the number of partial slabs be modified. | |
48 | * (Note that the total number of slabs is an atomic value that may be | |
49 | * modified without taking the list lock). | |
50 | * | |
51 | * The list_lock is a centralized lock and thus we avoid taking it as | |
52 | * much as possible. As long as SLUB does not have to handle partial | |
53 | * slabs, operations can continue without any centralized lock. F.e. | |
54 | * allocating a long series of objects that fill up slabs does not require | |
55 | * the list lock. | |
56 | * | |
57 | * The lock order is sometimes inverted when we are trying to get a slab | |
58 | * off a list. We take the list_lock and then look for a page on the list | |
59 | * to use. While we do that objects in the slabs may be freed. We can | |
60 | * only operate on the slab if we have also taken the slab_lock. So we use | |
61 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
62 | * can occur anymore and we can use the slab for allocations etc. If the | |
63 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
64 | * we must stay away from it for a while since we may cause a bouncing | |
65 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
66 | * If all pages are busy then we may allocate a new slab instead of reusing | |
67 | * a partial slab. A new slab has noone operating on it and thus there is | |
68 | * no danger of cacheline contention. | |
69 | * | |
70 | * Interrupts are disabled during allocation and deallocation in order to | |
71 | * make the slab allocator safe to use in the context of an irq. In addition | |
72 | * interrupts are disabled to ensure that the processor does not change | |
73 | * while handling per_cpu slabs, due to kernel preemption. | |
74 | * | |
75 | * SLUB assigns one slab for allocation to each processor. | |
76 | * Allocations only occur from these slabs called cpu slabs. | |
77 | * | |
672bba3a CL |
78 | * Slabs with free elements are kept on a partial list and during regular |
79 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 80 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
81 | * We track full slabs for debugging purposes though because otherwise we |
82 | * cannot scan all objects. | |
81819f0f CL |
83 | * |
84 | * Slabs are freed when they become empty. Teardown and setup is | |
85 | * minimal so we rely on the page allocators per cpu caches for | |
86 | * fast frees and allocs. | |
87 | * | |
88 | * Overloading of page flags that are otherwise used for LRU management. | |
89 | * | |
4b6f0750 CL |
90 | * PageActive The slab is frozen and exempt from list processing. |
91 | * This means that the slab is dedicated to a purpose | |
92 | * such as satisfying allocations for a specific | |
93 | * processor. Objects may be freed in the slab while | |
94 | * it is frozen but slab_free will then skip the usual | |
95 | * list operations. It is up to the processor holding | |
96 | * the slab to integrate the slab into the slab lists | |
97 | * when the slab is no longer needed. | |
98 | * | |
99 | * One use of this flag is to mark slabs that are | |
100 | * used for allocations. Then such a slab becomes a cpu | |
101 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 102 | * freelist that allows lockless access to |
894b8788 CL |
103 | * free objects in addition to the regular freelist |
104 | * that requires the slab lock. | |
81819f0f CL |
105 | * |
106 | * PageError Slab requires special handling due to debug | |
107 | * options set. This moves slab handling out of | |
894b8788 | 108 | * the fast path and disables lockless freelists. |
81819f0f CL |
109 | */ |
110 | ||
af537b0a CL |
111 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
112 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
113 | ||
114 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
115 | { | |
5577bd8a | 116 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 117 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 118 | #else |
af537b0a | 119 | return 0; |
5577bd8a | 120 | #endif |
af537b0a | 121 | } |
5577bd8a | 122 | |
81819f0f CL |
123 | /* |
124 | * Issues still to be resolved: | |
125 | * | |
81819f0f CL |
126 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
127 | * | |
81819f0f CL |
128 | * - Variable sizing of the per node arrays |
129 | */ | |
130 | ||
131 | /* Enable to test recovery from slab corruption on boot */ | |
132 | #undef SLUB_RESILIENCY_TEST | |
133 | ||
2086d26a CL |
134 | /* |
135 | * Mininum number of partial slabs. These will be left on the partial | |
136 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
137 | */ | |
76be8950 | 138 | #define MIN_PARTIAL 5 |
e95eed57 | 139 | |
2086d26a CL |
140 | /* |
141 | * Maximum number of desirable partial slabs. | |
142 | * The existence of more partial slabs makes kmem_cache_shrink | |
143 | * sort the partial list by the number of objects in the. | |
144 | */ | |
145 | #define MAX_PARTIAL 10 | |
146 | ||
81819f0f CL |
147 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
148 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 149 | |
fa5ec8a1 | 150 | /* |
3de47213 DR |
151 | * Debugging flags that require metadata to be stored in the slab. These get |
152 | * disabled when slub_debug=O is used and a cache's min order increases with | |
153 | * metadata. | |
fa5ec8a1 | 154 | */ |
3de47213 | 155 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 156 | |
81819f0f CL |
157 | /* |
158 | * Set of flags that will prevent slab merging | |
159 | */ | |
160 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
161 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
162 | SLAB_FAILSLAB) | |
81819f0f CL |
163 | |
164 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 165 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 166 | |
210b5c06 CG |
167 | #define OO_SHIFT 16 |
168 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
169 | #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */ | |
170 | ||
81819f0f | 171 | /* Internal SLUB flags */ |
f90ec390 | 172 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
81819f0f CL |
173 | |
174 | static int kmem_size = sizeof(struct kmem_cache); | |
175 | ||
176 | #ifdef CONFIG_SMP | |
177 | static struct notifier_block slab_notifier; | |
178 | #endif | |
179 | ||
180 | static enum { | |
181 | DOWN, /* No slab functionality available */ | |
51df1142 | 182 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 183 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
184 | SYSFS /* Sysfs up */ |
185 | } slab_state = DOWN; | |
186 | ||
187 | /* A list of all slab caches on the system */ | |
188 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 189 | static LIST_HEAD(slab_caches); |
81819f0f | 190 | |
02cbc874 CL |
191 | /* |
192 | * Tracking user of a slab. | |
193 | */ | |
194 | struct track { | |
ce71e27c | 195 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
196 | int cpu; /* Was running on cpu */ |
197 | int pid; /* Pid context */ | |
198 | unsigned long when; /* When did the operation occur */ | |
199 | }; | |
200 | ||
201 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
202 | ||
ab4d5ed5 | 203 | #ifdef CONFIG_SYSFS |
81819f0f CL |
204 | static int sysfs_slab_add(struct kmem_cache *); |
205 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
206 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 207 | |
81819f0f | 208 | #else |
0c710013 CL |
209 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
210 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
211 | { return 0; } | |
151c602f CL |
212 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
213 | { | |
84c1cf62 | 214 | kfree(s->name); |
151c602f CL |
215 | kfree(s); |
216 | } | |
8ff12cfc | 217 | |
81819f0f CL |
218 | #endif |
219 | ||
4fdccdfb | 220 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
221 | { |
222 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 223 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
224 | #endif |
225 | } | |
226 | ||
81819f0f CL |
227 | /******************************************************************** |
228 | * Core slab cache functions | |
229 | *******************************************************************/ | |
230 | ||
231 | int slab_is_available(void) | |
232 | { | |
233 | return slab_state >= UP; | |
234 | } | |
235 | ||
236 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
237 | { | |
81819f0f | 238 | return s->node[node]; |
81819f0f CL |
239 | } |
240 | ||
6446faa2 | 241 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
242 | static inline int check_valid_pointer(struct kmem_cache *s, |
243 | struct page *page, const void *object) | |
244 | { | |
245 | void *base; | |
246 | ||
a973e9dd | 247 | if (!object) |
02cbc874 CL |
248 | return 1; |
249 | ||
a973e9dd | 250 | base = page_address(page); |
39b26464 | 251 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
252 | (object - base) % s->size) { |
253 | return 0; | |
254 | } | |
255 | ||
256 | return 1; | |
257 | } | |
258 | ||
7656c72b CL |
259 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
260 | { | |
261 | return *(void **)(object + s->offset); | |
262 | } | |
263 | ||
264 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
265 | { | |
266 | *(void **)(object + s->offset) = fp; | |
267 | } | |
268 | ||
269 | /* Loop over all objects in a slab */ | |
224a88be CL |
270 | #define for_each_object(__p, __s, __addr, __objects) \ |
271 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
272 | __p += (__s)->size) |
273 | ||
7656c72b CL |
274 | /* Determine object index from a given position */ |
275 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
276 | { | |
277 | return (p - addr) / s->size; | |
278 | } | |
279 | ||
d71f606f MK |
280 | static inline size_t slab_ksize(const struct kmem_cache *s) |
281 | { | |
282 | #ifdef CONFIG_SLUB_DEBUG | |
283 | /* | |
284 | * Debugging requires use of the padding between object | |
285 | * and whatever may come after it. | |
286 | */ | |
287 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
288 | return s->objsize; | |
289 | ||
290 | #endif | |
291 | /* | |
292 | * If we have the need to store the freelist pointer | |
293 | * back there or track user information then we can | |
294 | * only use the space before that information. | |
295 | */ | |
296 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
297 | return s->inuse; | |
298 | /* | |
299 | * Else we can use all the padding etc for the allocation | |
300 | */ | |
301 | return s->size; | |
302 | } | |
303 | ||
ab9a0f19 LJ |
304 | static inline int order_objects(int order, unsigned long size, int reserved) |
305 | { | |
306 | return ((PAGE_SIZE << order) - reserved) / size; | |
307 | } | |
308 | ||
834f3d11 | 309 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 310 | unsigned long size, int reserved) |
834f3d11 CL |
311 | { |
312 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 313 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
314 | }; |
315 | ||
316 | return x; | |
317 | } | |
318 | ||
319 | static inline int oo_order(struct kmem_cache_order_objects x) | |
320 | { | |
210b5c06 | 321 | return x.x >> OO_SHIFT; |
834f3d11 CL |
322 | } |
323 | ||
324 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
325 | { | |
210b5c06 | 326 | return x.x & OO_MASK; |
834f3d11 CL |
327 | } |
328 | ||
5f80b13a CL |
329 | /* |
330 | * Determine a map of object in use on a page. | |
331 | * | |
332 | * Slab lock or node listlock must be held to guarantee that the page does | |
333 | * not vanish from under us. | |
334 | */ | |
335 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
336 | { | |
337 | void *p; | |
338 | void *addr = page_address(page); | |
339 | ||
340 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
341 | set_bit(slab_index(p, s, addr), map); | |
342 | } | |
343 | ||
41ecc55b CL |
344 | #ifdef CONFIG_SLUB_DEBUG |
345 | /* | |
346 | * Debug settings: | |
347 | */ | |
f0630fff CL |
348 | #ifdef CONFIG_SLUB_DEBUG_ON |
349 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
350 | #else | |
41ecc55b | 351 | static int slub_debug; |
f0630fff | 352 | #endif |
41ecc55b CL |
353 | |
354 | static char *slub_debug_slabs; | |
fa5ec8a1 | 355 | static int disable_higher_order_debug; |
41ecc55b | 356 | |
81819f0f CL |
357 | /* |
358 | * Object debugging | |
359 | */ | |
360 | static void print_section(char *text, u8 *addr, unsigned int length) | |
361 | { | |
362 | int i, offset; | |
363 | int newline = 1; | |
364 | char ascii[17]; | |
365 | ||
366 | ascii[16] = 0; | |
367 | ||
368 | for (i = 0; i < length; i++) { | |
369 | if (newline) { | |
24922684 | 370 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
371 | newline = 0; |
372 | } | |
06428780 | 373 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
374 | offset = i % 16; |
375 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
376 | if (offset == 15) { | |
06428780 | 377 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
378 | newline = 1; |
379 | } | |
380 | } | |
381 | if (!newline) { | |
382 | i %= 16; | |
383 | while (i < 16) { | |
06428780 | 384 | printk(KERN_CONT " "); |
81819f0f CL |
385 | ascii[i] = ' '; |
386 | i++; | |
387 | } | |
06428780 | 388 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
389 | } |
390 | } | |
391 | ||
81819f0f CL |
392 | static struct track *get_track(struct kmem_cache *s, void *object, |
393 | enum track_item alloc) | |
394 | { | |
395 | struct track *p; | |
396 | ||
397 | if (s->offset) | |
398 | p = object + s->offset + sizeof(void *); | |
399 | else | |
400 | p = object + s->inuse; | |
401 | ||
402 | return p + alloc; | |
403 | } | |
404 | ||
405 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 406 | enum track_item alloc, unsigned long addr) |
81819f0f | 407 | { |
1a00df4a | 408 | struct track *p = get_track(s, object, alloc); |
81819f0f | 409 | |
81819f0f CL |
410 | if (addr) { |
411 | p->addr = addr; | |
412 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 413 | p->pid = current->pid; |
81819f0f CL |
414 | p->when = jiffies; |
415 | } else | |
416 | memset(p, 0, sizeof(struct track)); | |
417 | } | |
418 | ||
81819f0f CL |
419 | static void init_tracking(struct kmem_cache *s, void *object) |
420 | { | |
24922684 CL |
421 | if (!(s->flags & SLAB_STORE_USER)) |
422 | return; | |
423 | ||
ce71e27c EGM |
424 | set_track(s, object, TRACK_FREE, 0UL); |
425 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
426 | } |
427 | ||
428 | static void print_track(const char *s, struct track *t) | |
429 | { | |
430 | if (!t->addr) | |
431 | return; | |
432 | ||
7daf705f | 433 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 434 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
435 | } |
436 | ||
437 | static void print_tracking(struct kmem_cache *s, void *object) | |
438 | { | |
439 | if (!(s->flags & SLAB_STORE_USER)) | |
440 | return; | |
441 | ||
442 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
443 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
444 | } | |
445 | ||
446 | static void print_page_info(struct page *page) | |
447 | { | |
39b26464 CL |
448 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
449 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
450 | |
451 | } | |
452 | ||
453 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
454 | { | |
455 | va_list args; | |
456 | char buf[100]; | |
457 | ||
458 | va_start(args, fmt); | |
459 | vsnprintf(buf, sizeof(buf), fmt, args); | |
460 | va_end(args); | |
461 | printk(KERN_ERR "========================================" | |
462 | "=====================================\n"); | |
463 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
464 | printk(KERN_ERR "----------------------------------------" | |
465 | "-------------------------------------\n\n"); | |
81819f0f CL |
466 | } |
467 | ||
24922684 CL |
468 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
469 | { | |
470 | va_list args; | |
471 | char buf[100]; | |
472 | ||
473 | va_start(args, fmt); | |
474 | vsnprintf(buf, sizeof(buf), fmt, args); | |
475 | va_end(args); | |
476 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
477 | } | |
478 | ||
479 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
480 | { |
481 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 482 | u8 *addr = page_address(page); |
24922684 CL |
483 | |
484 | print_tracking(s, p); | |
485 | ||
486 | print_page_info(page); | |
487 | ||
488 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
489 | p, p - addr, get_freepointer(s, p)); | |
490 | ||
491 | if (p > addr + 16) | |
492 | print_section("Bytes b4", p - 16, 16); | |
493 | ||
0ebd652b | 494 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
495 | |
496 | if (s->flags & SLAB_RED_ZONE) | |
497 | print_section("Redzone", p + s->objsize, | |
498 | s->inuse - s->objsize); | |
499 | ||
81819f0f CL |
500 | if (s->offset) |
501 | off = s->offset + sizeof(void *); | |
502 | else | |
503 | off = s->inuse; | |
504 | ||
24922684 | 505 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 506 | off += 2 * sizeof(struct track); |
81819f0f CL |
507 | |
508 | if (off != s->size) | |
509 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
510 | print_section("Padding", p + off, s->size - off); |
511 | ||
512 | dump_stack(); | |
81819f0f CL |
513 | } |
514 | ||
515 | static void object_err(struct kmem_cache *s, struct page *page, | |
516 | u8 *object, char *reason) | |
517 | { | |
3dc50637 | 518 | slab_bug(s, "%s", reason); |
24922684 | 519 | print_trailer(s, page, object); |
81819f0f CL |
520 | } |
521 | ||
24922684 | 522 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
523 | { |
524 | va_list args; | |
525 | char buf[100]; | |
526 | ||
24922684 CL |
527 | va_start(args, fmt); |
528 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 529 | va_end(args); |
3dc50637 | 530 | slab_bug(s, "%s", buf); |
24922684 | 531 | print_page_info(page); |
81819f0f CL |
532 | dump_stack(); |
533 | } | |
534 | ||
f7cb1933 | 535 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
536 | { |
537 | u8 *p = object; | |
538 | ||
539 | if (s->flags & __OBJECT_POISON) { | |
540 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 541 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
542 | } |
543 | ||
544 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 545 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
546 | } |
547 | ||
24922684 | 548 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
549 | { |
550 | while (bytes) { | |
551 | if (*start != (u8)value) | |
24922684 | 552 | return start; |
81819f0f CL |
553 | start++; |
554 | bytes--; | |
555 | } | |
24922684 CL |
556 | return NULL; |
557 | } | |
558 | ||
559 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
560 | void *from, void *to) | |
561 | { | |
562 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
563 | memset(from, data, to - from); | |
564 | } | |
565 | ||
566 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
567 | u8 *object, char *what, | |
06428780 | 568 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
569 | { |
570 | u8 *fault; | |
571 | u8 *end; | |
572 | ||
573 | fault = check_bytes(start, value, bytes); | |
574 | if (!fault) | |
575 | return 1; | |
576 | ||
577 | end = start + bytes; | |
578 | while (end > fault && end[-1] == value) | |
579 | end--; | |
580 | ||
581 | slab_bug(s, "%s overwritten", what); | |
582 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
583 | fault, end - 1, fault[0], value); | |
584 | print_trailer(s, page, object); | |
585 | ||
586 | restore_bytes(s, what, value, fault, end); | |
587 | return 0; | |
81819f0f CL |
588 | } |
589 | ||
81819f0f CL |
590 | /* |
591 | * Object layout: | |
592 | * | |
593 | * object address | |
594 | * Bytes of the object to be managed. | |
595 | * If the freepointer may overlay the object then the free | |
596 | * pointer is the first word of the object. | |
672bba3a | 597 | * |
81819f0f CL |
598 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
599 | * 0xa5 (POISON_END) | |
600 | * | |
601 | * object + s->objsize | |
602 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
603 | * Padding is extended by another word if Redzoning is enabled and |
604 | * objsize == inuse. | |
605 | * | |
81819f0f CL |
606 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
607 | * 0xcc (RED_ACTIVE) for objects in use. | |
608 | * | |
609 | * object + s->inuse | |
672bba3a CL |
610 | * Meta data starts here. |
611 | * | |
81819f0f CL |
612 | * A. Free pointer (if we cannot overwrite object on free) |
613 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 614 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 615 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
616 | * before the word boundary. |
617 | * | |
618 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
619 | * |
620 | * object + s->size | |
672bba3a | 621 | * Nothing is used beyond s->size. |
81819f0f | 622 | * |
672bba3a CL |
623 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
624 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
625 | * may be used with merged slabcaches. |
626 | */ | |
627 | ||
81819f0f CL |
628 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
629 | { | |
630 | unsigned long off = s->inuse; /* The end of info */ | |
631 | ||
632 | if (s->offset) | |
633 | /* Freepointer is placed after the object. */ | |
634 | off += sizeof(void *); | |
635 | ||
636 | if (s->flags & SLAB_STORE_USER) | |
637 | /* We also have user information there */ | |
638 | off += 2 * sizeof(struct track); | |
639 | ||
640 | if (s->size == off) | |
641 | return 1; | |
642 | ||
24922684 CL |
643 | return check_bytes_and_report(s, page, p, "Object padding", |
644 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
645 | } |
646 | ||
39b26464 | 647 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
648 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
649 | { | |
24922684 CL |
650 | u8 *start; |
651 | u8 *fault; | |
652 | u8 *end; | |
653 | int length; | |
654 | int remainder; | |
81819f0f CL |
655 | |
656 | if (!(s->flags & SLAB_POISON)) | |
657 | return 1; | |
658 | ||
a973e9dd | 659 | start = page_address(page); |
ab9a0f19 | 660 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
661 | end = start + length; |
662 | remainder = length % s->size; | |
81819f0f CL |
663 | if (!remainder) |
664 | return 1; | |
665 | ||
39b26464 | 666 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
667 | if (!fault) |
668 | return 1; | |
669 | while (end > fault && end[-1] == POISON_INUSE) | |
670 | end--; | |
671 | ||
672 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 673 | print_section("Padding", end - remainder, remainder); |
24922684 | 674 | |
8a3d271d | 675 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 676 | return 0; |
81819f0f CL |
677 | } |
678 | ||
679 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 680 | void *object, u8 val) |
81819f0f CL |
681 | { |
682 | u8 *p = object; | |
683 | u8 *endobject = object + s->objsize; | |
684 | ||
685 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 686 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 687 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 688 | return 0; |
81819f0f | 689 | } else { |
3adbefee IM |
690 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
691 | check_bytes_and_report(s, page, p, "Alignment padding", | |
692 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
693 | } | |
81819f0f CL |
694 | } |
695 | ||
696 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 697 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
698 | (!check_bytes_and_report(s, page, p, "Poison", p, |
699 | POISON_FREE, s->objsize - 1) || | |
700 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 701 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 702 | return 0; |
81819f0f CL |
703 | /* |
704 | * check_pad_bytes cleans up on its own. | |
705 | */ | |
706 | check_pad_bytes(s, page, p); | |
707 | } | |
708 | ||
f7cb1933 | 709 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
710 | /* |
711 | * Object and freepointer overlap. Cannot check | |
712 | * freepointer while object is allocated. | |
713 | */ | |
714 | return 1; | |
715 | ||
716 | /* Check free pointer validity */ | |
717 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
718 | object_err(s, page, p, "Freepointer corrupt"); | |
719 | /* | |
9f6c708e | 720 | * No choice but to zap it and thus lose the remainder |
81819f0f | 721 | * of the free objects in this slab. May cause |
672bba3a | 722 | * another error because the object count is now wrong. |
81819f0f | 723 | */ |
a973e9dd | 724 | set_freepointer(s, p, NULL); |
81819f0f CL |
725 | return 0; |
726 | } | |
727 | return 1; | |
728 | } | |
729 | ||
730 | static int check_slab(struct kmem_cache *s, struct page *page) | |
731 | { | |
39b26464 CL |
732 | int maxobj; |
733 | ||
81819f0f CL |
734 | VM_BUG_ON(!irqs_disabled()); |
735 | ||
736 | if (!PageSlab(page)) { | |
24922684 | 737 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
738 | return 0; |
739 | } | |
39b26464 | 740 | |
ab9a0f19 | 741 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
742 | if (page->objects > maxobj) { |
743 | slab_err(s, page, "objects %u > max %u", | |
744 | s->name, page->objects, maxobj); | |
745 | return 0; | |
746 | } | |
747 | if (page->inuse > page->objects) { | |
24922684 | 748 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 749 | s->name, page->inuse, page->objects); |
81819f0f CL |
750 | return 0; |
751 | } | |
752 | /* Slab_pad_check fixes things up after itself */ | |
753 | slab_pad_check(s, page); | |
754 | return 1; | |
755 | } | |
756 | ||
757 | /* | |
672bba3a CL |
758 | * Determine if a certain object on a page is on the freelist. Must hold the |
759 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
760 | */ |
761 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
762 | { | |
763 | int nr = 0; | |
764 | void *fp = page->freelist; | |
765 | void *object = NULL; | |
224a88be | 766 | unsigned long max_objects; |
81819f0f | 767 | |
39b26464 | 768 | while (fp && nr <= page->objects) { |
81819f0f CL |
769 | if (fp == search) |
770 | return 1; | |
771 | if (!check_valid_pointer(s, page, fp)) { | |
772 | if (object) { | |
773 | object_err(s, page, object, | |
774 | "Freechain corrupt"); | |
a973e9dd | 775 | set_freepointer(s, object, NULL); |
81819f0f CL |
776 | break; |
777 | } else { | |
24922684 | 778 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 779 | page->freelist = NULL; |
39b26464 | 780 | page->inuse = page->objects; |
24922684 | 781 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
782 | return 0; |
783 | } | |
784 | break; | |
785 | } | |
786 | object = fp; | |
787 | fp = get_freepointer(s, object); | |
788 | nr++; | |
789 | } | |
790 | ||
ab9a0f19 | 791 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
792 | if (max_objects > MAX_OBJS_PER_PAGE) |
793 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
794 | |
795 | if (page->objects != max_objects) { | |
796 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
797 | "should be %d", page->objects, max_objects); | |
798 | page->objects = max_objects; | |
799 | slab_fix(s, "Number of objects adjusted."); | |
800 | } | |
39b26464 | 801 | if (page->inuse != page->objects - nr) { |
70d71228 | 802 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
803 | "counted were %d", page->inuse, page->objects - nr); |
804 | page->inuse = page->objects - nr; | |
24922684 | 805 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
806 | } |
807 | return search == NULL; | |
808 | } | |
809 | ||
0121c619 CL |
810 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
811 | int alloc) | |
3ec09742 CL |
812 | { |
813 | if (s->flags & SLAB_TRACE) { | |
814 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
815 | s->name, | |
816 | alloc ? "alloc" : "free", | |
817 | object, page->inuse, | |
818 | page->freelist); | |
819 | ||
820 | if (!alloc) | |
821 | print_section("Object", (void *)object, s->objsize); | |
822 | ||
823 | dump_stack(); | |
824 | } | |
825 | } | |
826 | ||
c016b0bd CL |
827 | /* |
828 | * Hooks for other subsystems that check memory allocations. In a typical | |
829 | * production configuration these hooks all should produce no code at all. | |
830 | */ | |
831 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
832 | { | |
c1d50836 | 833 | flags &= gfp_allowed_mask; |
c016b0bd CL |
834 | lockdep_trace_alloc(flags); |
835 | might_sleep_if(flags & __GFP_WAIT); | |
836 | ||
837 | return should_failslab(s->objsize, flags, s->flags); | |
838 | } | |
839 | ||
840 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
841 | { | |
c1d50836 | 842 | flags &= gfp_allowed_mask; |
b3d41885 | 843 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
c016b0bd CL |
844 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
845 | } | |
846 | ||
847 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
848 | { | |
849 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 850 | |
d3f661d6 CL |
851 | /* |
852 | * Trouble is that we may no longer disable interupts in the fast path | |
853 | * So in order to make the debug calls that expect irqs to be | |
854 | * disabled we need to disable interrupts temporarily. | |
855 | */ | |
856 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
857 | { | |
858 | unsigned long flags; | |
859 | ||
860 | local_irq_save(flags); | |
861 | kmemcheck_slab_free(s, x, s->objsize); | |
862 | debug_check_no_locks_freed(x, s->objsize); | |
d3f661d6 CL |
863 | local_irq_restore(flags); |
864 | } | |
865 | #endif | |
f9b615de TG |
866 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
867 | debug_check_no_obj_freed(x, s->objsize); | |
c016b0bd CL |
868 | } |
869 | ||
643b1138 | 870 | /* |
672bba3a | 871 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 872 | */ |
e95eed57 | 873 | static void add_full(struct kmem_cache_node *n, struct page *page) |
643b1138 | 874 | { |
643b1138 CL |
875 | spin_lock(&n->list_lock); |
876 | list_add(&page->lru, &n->full); | |
877 | spin_unlock(&n->list_lock); | |
878 | } | |
879 | ||
880 | static void remove_full(struct kmem_cache *s, struct page *page) | |
881 | { | |
882 | struct kmem_cache_node *n; | |
883 | ||
884 | if (!(s->flags & SLAB_STORE_USER)) | |
885 | return; | |
886 | ||
887 | n = get_node(s, page_to_nid(page)); | |
888 | ||
889 | spin_lock(&n->list_lock); | |
890 | list_del(&page->lru); | |
891 | spin_unlock(&n->list_lock); | |
892 | } | |
893 | ||
0f389ec6 CL |
894 | /* Tracking of the number of slabs for debugging purposes */ |
895 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
896 | { | |
897 | struct kmem_cache_node *n = get_node(s, node); | |
898 | ||
899 | return atomic_long_read(&n->nr_slabs); | |
900 | } | |
901 | ||
26c02cf0 AB |
902 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
903 | { | |
904 | return atomic_long_read(&n->nr_slabs); | |
905 | } | |
906 | ||
205ab99d | 907 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
908 | { |
909 | struct kmem_cache_node *n = get_node(s, node); | |
910 | ||
911 | /* | |
912 | * May be called early in order to allocate a slab for the | |
913 | * kmem_cache_node structure. Solve the chicken-egg | |
914 | * dilemma by deferring the increment of the count during | |
915 | * bootstrap (see early_kmem_cache_node_alloc). | |
916 | */ | |
7340cc84 | 917 | if (n) { |
0f389ec6 | 918 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
919 | atomic_long_add(objects, &n->total_objects); |
920 | } | |
0f389ec6 | 921 | } |
205ab99d | 922 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
923 | { |
924 | struct kmem_cache_node *n = get_node(s, node); | |
925 | ||
926 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 927 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
928 | } |
929 | ||
930 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
931 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
932 | void *object) | |
933 | { | |
934 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
935 | return; | |
936 | ||
f7cb1933 | 937 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
938 | init_tracking(s, object); |
939 | } | |
940 | ||
1537066c | 941 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 942 | void *object, unsigned long addr) |
81819f0f CL |
943 | { |
944 | if (!check_slab(s, page)) | |
945 | goto bad; | |
946 | ||
d692ef6d | 947 | if (!on_freelist(s, page, object)) { |
24922684 | 948 | object_err(s, page, object, "Object already allocated"); |
70d71228 | 949 | goto bad; |
81819f0f CL |
950 | } |
951 | ||
952 | if (!check_valid_pointer(s, page, object)) { | |
953 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 954 | goto bad; |
81819f0f CL |
955 | } |
956 | ||
f7cb1933 | 957 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 958 | goto bad; |
81819f0f | 959 | |
3ec09742 CL |
960 | /* Success perform special debug activities for allocs */ |
961 | if (s->flags & SLAB_STORE_USER) | |
962 | set_track(s, object, TRACK_ALLOC, addr); | |
963 | trace(s, page, object, 1); | |
f7cb1933 | 964 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 965 | return 1; |
3ec09742 | 966 | |
81819f0f CL |
967 | bad: |
968 | if (PageSlab(page)) { | |
969 | /* | |
970 | * If this is a slab page then lets do the best we can | |
971 | * to avoid issues in the future. Marking all objects | |
672bba3a | 972 | * as used avoids touching the remaining objects. |
81819f0f | 973 | */ |
24922684 | 974 | slab_fix(s, "Marking all objects used"); |
39b26464 | 975 | page->inuse = page->objects; |
a973e9dd | 976 | page->freelist = NULL; |
81819f0f CL |
977 | } |
978 | return 0; | |
979 | } | |
980 | ||
1537066c CL |
981 | static noinline int free_debug_processing(struct kmem_cache *s, |
982 | struct page *page, void *object, unsigned long addr) | |
81819f0f CL |
983 | { |
984 | if (!check_slab(s, page)) | |
985 | goto fail; | |
986 | ||
987 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 988 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
989 | goto fail; |
990 | } | |
991 | ||
992 | if (on_freelist(s, page, object)) { | |
24922684 | 993 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
994 | goto fail; |
995 | } | |
996 | ||
f7cb1933 | 997 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
81819f0f CL |
998 | return 0; |
999 | ||
1000 | if (unlikely(s != page->slab)) { | |
3adbefee | 1001 | if (!PageSlab(page)) { |
70d71228 CL |
1002 | slab_err(s, page, "Attempt to free object(0x%p) " |
1003 | "outside of slab", object); | |
3adbefee | 1004 | } else if (!page->slab) { |
81819f0f | 1005 | printk(KERN_ERR |
70d71228 | 1006 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1007 | object); |
70d71228 | 1008 | dump_stack(); |
06428780 | 1009 | } else |
24922684 CL |
1010 | object_err(s, page, object, |
1011 | "page slab pointer corrupt."); | |
81819f0f CL |
1012 | goto fail; |
1013 | } | |
3ec09742 CL |
1014 | |
1015 | /* Special debug activities for freeing objects */ | |
8a38082d | 1016 | if (!PageSlubFrozen(page) && !page->freelist) |
3ec09742 CL |
1017 | remove_full(s, page); |
1018 | if (s->flags & SLAB_STORE_USER) | |
1019 | set_track(s, object, TRACK_FREE, addr); | |
1020 | trace(s, page, object, 0); | |
f7cb1933 | 1021 | init_object(s, object, SLUB_RED_INACTIVE); |
81819f0f | 1022 | return 1; |
3ec09742 | 1023 | |
81819f0f | 1024 | fail: |
24922684 | 1025 | slab_fix(s, "Object at 0x%p not freed", object); |
81819f0f CL |
1026 | return 0; |
1027 | } | |
1028 | ||
41ecc55b CL |
1029 | static int __init setup_slub_debug(char *str) |
1030 | { | |
f0630fff CL |
1031 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1032 | if (*str++ != '=' || !*str) | |
1033 | /* | |
1034 | * No options specified. Switch on full debugging. | |
1035 | */ | |
1036 | goto out; | |
1037 | ||
1038 | if (*str == ',') | |
1039 | /* | |
1040 | * No options but restriction on slabs. This means full | |
1041 | * debugging for slabs matching a pattern. | |
1042 | */ | |
1043 | goto check_slabs; | |
1044 | ||
fa5ec8a1 DR |
1045 | if (tolower(*str) == 'o') { |
1046 | /* | |
1047 | * Avoid enabling debugging on caches if its minimum order | |
1048 | * would increase as a result. | |
1049 | */ | |
1050 | disable_higher_order_debug = 1; | |
1051 | goto out; | |
1052 | } | |
1053 | ||
f0630fff CL |
1054 | slub_debug = 0; |
1055 | if (*str == '-') | |
1056 | /* | |
1057 | * Switch off all debugging measures. | |
1058 | */ | |
1059 | goto out; | |
1060 | ||
1061 | /* | |
1062 | * Determine which debug features should be switched on | |
1063 | */ | |
06428780 | 1064 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1065 | switch (tolower(*str)) { |
1066 | case 'f': | |
1067 | slub_debug |= SLAB_DEBUG_FREE; | |
1068 | break; | |
1069 | case 'z': | |
1070 | slub_debug |= SLAB_RED_ZONE; | |
1071 | break; | |
1072 | case 'p': | |
1073 | slub_debug |= SLAB_POISON; | |
1074 | break; | |
1075 | case 'u': | |
1076 | slub_debug |= SLAB_STORE_USER; | |
1077 | break; | |
1078 | case 't': | |
1079 | slub_debug |= SLAB_TRACE; | |
1080 | break; | |
4c13dd3b DM |
1081 | case 'a': |
1082 | slub_debug |= SLAB_FAILSLAB; | |
1083 | break; | |
f0630fff CL |
1084 | default: |
1085 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1086 | "unknown. skipped\n", *str); |
f0630fff | 1087 | } |
41ecc55b CL |
1088 | } |
1089 | ||
f0630fff | 1090 | check_slabs: |
41ecc55b CL |
1091 | if (*str == ',') |
1092 | slub_debug_slabs = str + 1; | |
f0630fff | 1093 | out: |
41ecc55b CL |
1094 | return 1; |
1095 | } | |
1096 | ||
1097 | __setup("slub_debug", setup_slub_debug); | |
1098 | ||
ba0268a8 CL |
1099 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1100 | unsigned long flags, const char *name, | |
51cc5068 | 1101 | void (*ctor)(void *)) |
41ecc55b CL |
1102 | { |
1103 | /* | |
e153362a | 1104 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1105 | */ |
e153362a | 1106 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1107 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1108 | flags |= slub_debug; | |
ba0268a8 CL |
1109 | |
1110 | return flags; | |
41ecc55b CL |
1111 | } |
1112 | #else | |
3ec09742 CL |
1113 | static inline void setup_object_debug(struct kmem_cache *s, |
1114 | struct page *page, void *object) {} | |
41ecc55b | 1115 | |
3ec09742 | 1116 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1117 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1118 | |
3ec09742 | 1119 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1120 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1121 | |
41ecc55b CL |
1122 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1123 | { return 1; } | |
1124 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1125 | void *object, u8 val) { return 1; } |
3ec09742 | 1126 | static inline void add_full(struct kmem_cache_node *n, struct page *page) {} |
ba0268a8 CL |
1127 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1128 | unsigned long flags, const char *name, | |
51cc5068 | 1129 | void (*ctor)(void *)) |
ba0268a8 CL |
1130 | { |
1131 | return flags; | |
1132 | } | |
41ecc55b | 1133 | #define slub_debug 0 |
0f389ec6 | 1134 | |
fdaa45e9 IM |
1135 | #define disable_higher_order_debug 0 |
1136 | ||
0f389ec6 CL |
1137 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1138 | { return 0; } | |
26c02cf0 AB |
1139 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1140 | { return 0; } | |
205ab99d CL |
1141 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1142 | int objects) {} | |
1143 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1144 | int objects) {} | |
7d550c56 CL |
1145 | |
1146 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1147 | { return 0; } | |
1148 | ||
1149 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1150 | void *object) {} | |
1151 | ||
1152 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1153 | ||
ab4d5ed5 | 1154 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1155 | |
81819f0f CL |
1156 | /* |
1157 | * Slab allocation and freeing | |
1158 | */ | |
65c3376a CL |
1159 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1160 | struct kmem_cache_order_objects oo) | |
1161 | { | |
1162 | int order = oo_order(oo); | |
1163 | ||
b1eeab67 VN |
1164 | flags |= __GFP_NOTRACK; |
1165 | ||
2154a336 | 1166 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1167 | return alloc_pages(flags, order); |
1168 | else | |
6b65aaf3 | 1169 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1170 | } |
1171 | ||
81819f0f CL |
1172 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1173 | { | |
06428780 | 1174 | struct page *page; |
834f3d11 | 1175 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1176 | gfp_t alloc_gfp; |
81819f0f | 1177 | |
b7a49f0d | 1178 | flags |= s->allocflags; |
e12ba74d | 1179 | |
ba52270d PE |
1180 | /* |
1181 | * Let the initial higher-order allocation fail under memory pressure | |
1182 | * so we fall-back to the minimum order allocation. | |
1183 | */ | |
1184 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1185 | ||
1186 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1187 | if (unlikely(!page)) { |
1188 | oo = s->min; | |
1189 | /* | |
1190 | * Allocation may have failed due to fragmentation. | |
1191 | * Try a lower order alloc if possible | |
1192 | */ | |
1193 | page = alloc_slab_page(flags, node, oo); | |
1194 | if (!page) | |
1195 | return NULL; | |
81819f0f | 1196 | |
84e554e6 | 1197 | stat(s, ORDER_FALLBACK); |
65c3376a | 1198 | } |
5a896d9e VN |
1199 | |
1200 | if (kmemcheck_enabled | |
5086c389 | 1201 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1202 | int pages = 1 << oo_order(oo); |
1203 | ||
1204 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1205 | ||
1206 | /* | |
1207 | * Objects from caches that have a constructor don't get | |
1208 | * cleared when they're allocated, so we need to do it here. | |
1209 | */ | |
1210 | if (s->ctor) | |
1211 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1212 | else | |
1213 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1214 | } |
1215 | ||
834f3d11 | 1216 | page->objects = oo_objects(oo); |
81819f0f CL |
1217 | mod_zone_page_state(page_zone(page), |
1218 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1219 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1220 | 1 << oo_order(oo)); |
81819f0f CL |
1221 | |
1222 | return page; | |
1223 | } | |
1224 | ||
1225 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1226 | void *object) | |
1227 | { | |
3ec09742 | 1228 | setup_object_debug(s, page, object); |
4f104934 | 1229 | if (unlikely(s->ctor)) |
51cc5068 | 1230 | s->ctor(object); |
81819f0f CL |
1231 | } |
1232 | ||
1233 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1234 | { | |
1235 | struct page *page; | |
81819f0f | 1236 | void *start; |
81819f0f CL |
1237 | void *last; |
1238 | void *p; | |
1239 | ||
6cb06229 | 1240 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1241 | |
6cb06229 CL |
1242 | page = allocate_slab(s, |
1243 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1244 | if (!page) |
1245 | goto out; | |
1246 | ||
205ab99d | 1247 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1248 | page->slab = s; |
1249 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1250 | |
1251 | start = page_address(page); | |
81819f0f CL |
1252 | |
1253 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1254 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1255 | |
1256 | last = start; | |
224a88be | 1257 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1258 | setup_object(s, page, last); |
1259 | set_freepointer(s, last, p); | |
1260 | last = p; | |
1261 | } | |
1262 | setup_object(s, page, last); | |
a973e9dd | 1263 | set_freepointer(s, last, NULL); |
81819f0f CL |
1264 | |
1265 | page->freelist = start; | |
1266 | page->inuse = 0; | |
1267 | out: | |
81819f0f CL |
1268 | return page; |
1269 | } | |
1270 | ||
1271 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1272 | { | |
834f3d11 CL |
1273 | int order = compound_order(page); |
1274 | int pages = 1 << order; | |
81819f0f | 1275 | |
af537b0a | 1276 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1277 | void *p; |
1278 | ||
1279 | slab_pad_check(s, page); | |
224a88be CL |
1280 | for_each_object(p, s, page_address(page), |
1281 | page->objects) | |
f7cb1933 | 1282 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1283 | } |
1284 | ||
b1eeab67 | 1285 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1286 | |
81819f0f CL |
1287 | mod_zone_page_state(page_zone(page), |
1288 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1289 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1290 | -pages); |
81819f0f | 1291 | |
49bd5221 CL |
1292 | __ClearPageSlab(page); |
1293 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1294 | if (current->reclaim_state) |
1295 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1296 | __free_pages(page, order); |
81819f0f CL |
1297 | } |
1298 | ||
da9a638c LJ |
1299 | #define need_reserve_slab_rcu \ |
1300 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1301 | ||
81819f0f CL |
1302 | static void rcu_free_slab(struct rcu_head *h) |
1303 | { | |
1304 | struct page *page; | |
1305 | ||
da9a638c LJ |
1306 | if (need_reserve_slab_rcu) |
1307 | page = virt_to_head_page(h); | |
1308 | else | |
1309 | page = container_of((struct list_head *)h, struct page, lru); | |
1310 | ||
81819f0f CL |
1311 | __free_slab(page->slab, page); |
1312 | } | |
1313 | ||
1314 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1315 | { | |
1316 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1317 | struct rcu_head *head; |
1318 | ||
1319 | if (need_reserve_slab_rcu) { | |
1320 | int order = compound_order(page); | |
1321 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1322 | ||
1323 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1324 | head = page_address(page) + offset; | |
1325 | } else { | |
1326 | /* | |
1327 | * RCU free overloads the RCU head over the LRU | |
1328 | */ | |
1329 | head = (void *)&page->lru; | |
1330 | } | |
81819f0f CL |
1331 | |
1332 | call_rcu(head, rcu_free_slab); | |
1333 | } else | |
1334 | __free_slab(s, page); | |
1335 | } | |
1336 | ||
1337 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1338 | { | |
205ab99d | 1339 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1340 | free_slab(s, page); |
1341 | } | |
1342 | ||
1343 | /* | |
1344 | * Per slab locking using the pagelock | |
1345 | */ | |
1346 | static __always_inline void slab_lock(struct page *page) | |
1347 | { | |
1348 | bit_spin_lock(PG_locked, &page->flags); | |
1349 | } | |
1350 | ||
1351 | static __always_inline void slab_unlock(struct page *page) | |
1352 | { | |
a76d3546 | 1353 | __bit_spin_unlock(PG_locked, &page->flags); |
81819f0f CL |
1354 | } |
1355 | ||
1356 | static __always_inline int slab_trylock(struct page *page) | |
1357 | { | |
1358 | int rc = 1; | |
1359 | ||
1360 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
1361 | return rc; | |
1362 | } | |
1363 | ||
1364 | /* | |
1365 | * Management of partially allocated slabs | |
1366 | */ | |
7c2e132c CL |
1367 | static void add_partial(struct kmem_cache_node *n, |
1368 | struct page *page, int tail) | |
81819f0f | 1369 | { |
e95eed57 CL |
1370 | spin_lock(&n->list_lock); |
1371 | n->nr_partial++; | |
7c2e132c CL |
1372 | if (tail) |
1373 | list_add_tail(&page->lru, &n->partial); | |
1374 | else | |
1375 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1376 | spin_unlock(&n->list_lock); |
1377 | } | |
1378 | ||
62e346a8 CL |
1379 | static inline void __remove_partial(struct kmem_cache_node *n, |
1380 | struct page *page) | |
1381 | { | |
1382 | list_del(&page->lru); | |
1383 | n->nr_partial--; | |
1384 | } | |
1385 | ||
0121c619 | 1386 | static void remove_partial(struct kmem_cache *s, struct page *page) |
81819f0f CL |
1387 | { |
1388 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1389 | ||
1390 | spin_lock(&n->list_lock); | |
62e346a8 | 1391 | __remove_partial(n, page); |
81819f0f CL |
1392 | spin_unlock(&n->list_lock); |
1393 | } | |
1394 | ||
1395 | /* | |
672bba3a | 1396 | * Lock slab and remove from the partial list. |
81819f0f | 1397 | * |
672bba3a | 1398 | * Must hold list_lock. |
81819f0f | 1399 | */ |
0121c619 CL |
1400 | static inline int lock_and_freeze_slab(struct kmem_cache_node *n, |
1401 | struct page *page) | |
81819f0f CL |
1402 | { |
1403 | if (slab_trylock(page)) { | |
62e346a8 | 1404 | __remove_partial(n, page); |
8a38082d | 1405 | __SetPageSlubFrozen(page); |
81819f0f CL |
1406 | return 1; |
1407 | } | |
1408 | return 0; | |
1409 | } | |
1410 | ||
1411 | /* | |
672bba3a | 1412 | * Try to allocate a partial slab from a specific node. |
81819f0f CL |
1413 | */ |
1414 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
1415 | { | |
1416 | struct page *page; | |
1417 | ||
1418 | /* | |
1419 | * Racy check. If we mistakenly see no partial slabs then we | |
1420 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1421 | * partial slab and there is none available then get_partials() |
1422 | * will return NULL. | |
81819f0f CL |
1423 | */ |
1424 | if (!n || !n->nr_partial) | |
1425 | return NULL; | |
1426 | ||
1427 | spin_lock(&n->list_lock); | |
1428 | list_for_each_entry(page, &n->partial, lru) | |
4b6f0750 | 1429 | if (lock_and_freeze_slab(n, page)) |
81819f0f CL |
1430 | goto out; |
1431 | page = NULL; | |
1432 | out: | |
1433 | spin_unlock(&n->list_lock); | |
1434 | return page; | |
1435 | } | |
1436 | ||
1437 | /* | |
672bba3a | 1438 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1439 | */ |
1440 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1441 | { | |
1442 | #ifdef CONFIG_NUMA | |
1443 | struct zonelist *zonelist; | |
dd1a239f | 1444 | struct zoneref *z; |
54a6eb5c MG |
1445 | struct zone *zone; |
1446 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1447 | struct page *page; |
1448 | ||
1449 | /* | |
672bba3a CL |
1450 | * The defrag ratio allows a configuration of the tradeoffs between |
1451 | * inter node defragmentation and node local allocations. A lower | |
1452 | * defrag_ratio increases the tendency to do local allocations | |
1453 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1454 | * |
672bba3a CL |
1455 | * If the defrag_ratio is set to 0 then kmalloc() always |
1456 | * returns node local objects. If the ratio is higher then kmalloc() | |
1457 | * may return off node objects because partial slabs are obtained | |
1458 | * from other nodes and filled up. | |
81819f0f | 1459 | * |
6446faa2 | 1460 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1461 | * defrag_ratio = 1000) then every (well almost) allocation will |
1462 | * first attempt to defrag slab caches on other nodes. This means | |
1463 | * scanning over all nodes to look for partial slabs which may be | |
1464 | * expensive if we do it every time we are trying to find a slab | |
1465 | * with available objects. | |
81819f0f | 1466 | */ |
9824601e CL |
1467 | if (!s->remote_node_defrag_ratio || |
1468 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1469 | return NULL; |
1470 | ||
c0ff7453 | 1471 | get_mems_allowed(); |
0e88460d | 1472 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1473 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1474 | struct kmem_cache_node *n; |
1475 | ||
54a6eb5c | 1476 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1477 | |
54a6eb5c | 1478 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1479 | n->nr_partial > s->min_partial) { |
81819f0f | 1480 | page = get_partial_node(n); |
c0ff7453 MX |
1481 | if (page) { |
1482 | put_mems_allowed(); | |
81819f0f | 1483 | return page; |
c0ff7453 | 1484 | } |
81819f0f CL |
1485 | } |
1486 | } | |
c0ff7453 | 1487 | put_mems_allowed(); |
81819f0f CL |
1488 | #endif |
1489 | return NULL; | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Get a partial page, lock it and return it. | |
1494 | */ | |
1495 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1496 | { | |
1497 | struct page *page; | |
2154a336 | 1498 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f CL |
1499 | |
1500 | page = get_partial_node(get_node(s, searchnode)); | |
33de04ec | 1501 | if (page || node != NUMA_NO_NODE) |
81819f0f CL |
1502 | return page; |
1503 | ||
1504 | return get_any_partial(s, flags); | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * Move a page back to the lists. | |
1509 | * | |
1510 | * Must be called with the slab lock held. | |
1511 | * | |
1512 | * On exit the slab lock will have been dropped. | |
1513 | */ | |
7c2e132c | 1514 | static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail) |
3478973d | 1515 | __releases(bitlock) |
81819f0f | 1516 | { |
e95eed57 CL |
1517 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1518 | ||
8a38082d | 1519 | __ClearPageSlubFrozen(page); |
81819f0f | 1520 | if (page->inuse) { |
e95eed57 | 1521 | |
a973e9dd | 1522 | if (page->freelist) { |
7c2e132c | 1523 | add_partial(n, page, tail); |
84e554e6 | 1524 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); |
8ff12cfc | 1525 | } else { |
84e554e6 | 1526 | stat(s, DEACTIVATE_FULL); |
af537b0a | 1527 | if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) |
8ff12cfc CL |
1528 | add_full(n, page); |
1529 | } | |
81819f0f CL |
1530 | slab_unlock(page); |
1531 | } else { | |
84e554e6 | 1532 | stat(s, DEACTIVATE_EMPTY); |
3b89d7d8 | 1533 | if (n->nr_partial < s->min_partial) { |
e95eed57 | 1534 | /* |
672bba3a CL |
1535 | * Adding an empty slab to the partial slabs in order |
1536 | * to avoid page allocator overhead. This slab needs | |
1537 | * to come after the other slabs with objects in | |
6446faa2 CL |
1538 | * so that the others get filled first. That way the |
1539 | * size of the partial list stays small. | |
1540 | * | |
0121c619 CL |
1541 | * kmem_cache_shrink can reclaim any empty slabs from |
1542 | * the partial list. | |
e95eed57 | 1543 | */ |
7c2e132c | 1544 | add_partial(n, page, 1); |
e95eed57 CL |
1545 | slab_unlock(page); |
1546 | } else { | |
1547 | slab_unlock(page); | |
84e554e6 | 1548 | stat(s, FREE_SLAB); |
e95eed57 CL |
1549 | discard_slab(s, page); |
1550 | } | |
81819f0f CL |
1551 | } |
1552 | } | |
1553 | ||
8a5ec0ba CL |
1554 | #ifdef CONFIG_CMPXCHG_LOCAL |
1555 | #ifdef CONFIG_PREEMPT | |
1556 | /* | |
1557 | * Calculate the next globally unique transaction for disambiguiation | |
1558 | * during cmpxchg. The transactions start with the cpu number and are then | |
1559 | * incremented by CONFIG_NR_CPUS. | |
1560 | */ | |
1561 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1562 | #else | |
1563 | /* | |
1564 | * No preemption supported therefore also no need to check for | |
1565 | * different cpus. | |
1566 | */ | |
1567 | #define TID_STEP 1 | |
1568 | #endif | |
1569 | ||
1570 | static inline unsigned long next_tid(unsigned long tid) | |
1571 | { | |
1572 | return tid + TID_STEP; | |
1573 | } | |
1574 | ||
1575 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1576 | { | |
1577 | return tid % TID_STEP; | |
1578 | } | |
1579 | ||
1580 | static inline unsigned long tid_to_event(unsigned long tid) | |
1581 | { | |
1582 | return tid / TID_STEP; | |
1583 | } | |
1584 | ||
1585 | static inline unsigned int init_tid(int cpu) | |
1586 | { | |
1587 | return cpu; | |
1588 | } | |
1589 | ||
1590 | static inline void note_cmpxchg_failure(const char *n, | |
1591 | const struct kmem_cache *s, unsigned long tid) | |
1592 | { | |
1593 | #ifdef SLUB_DEBUG_CMPXCHG | |
1594 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1595 | ||
1596 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1597 | ||
1598 | #ifdef CONFIG_PREEMPT | |
1599 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1600 | printk("due to cpu change %d -> %d\n", | |
1601 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1602 | else | |
1603 | #endif | |
1604 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1605 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1606 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1607 | else | |
1608 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1609 | actual_tid, tid, next_tid(tid)); | |
1610 | #endif | |
4fdccdfb | 1611 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1612 | } |
1613 | ||
1614 | #endif | |
1615 | ||
1616 | void init_kmem_cache_cpus(struct kmem_cache *s) | |
1617 | { | |
b8c4c96e | 1618 | #ifdef CONFIG_CMPXCHG_LOCAL |
8a5ec0ba CL |
1619 | int cpu; |
1620 | ||
1621 | for_each_possible_cpu(cpu) | |
1622 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
1623 | #endif | |
1624 | ||
1625 | } | |
81819f0f CL |
1626 | /* |
1627 | * Remove the cpu slab | |
1628 | */ | |
dfb4f096 | 1629 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
3478973d | 1630 | __releases(bitlock) |
81819f0f | 1631 | { |
dfb4f096 | 1632 | struct page *page = c->page; |
7c2e132c | 1633 | int tail = 1; |
8ff12cfc | 1634 | |
b773ad73 | 1635 | if (page->freelist) |
84e554e6 | 1636 | stat(s, DEACTIVATE_REMOTE_FREES); |
894b8788 | 1637 | /* |
6446faa2 | 1638 | * Merge cpu freelist into slab freelist. Typically we get here |
894b8788 CL |
1639 | * because both freelists are empty. So this is unlikely |
1640 | * to occur. | |
1641 | */ | |
a973e9dd | 1642 | while (unlikely(c->freelist)) { |
894b8788 CL |
1643 | void **object; |
1644 | ||
7c2e132c CL |
1645 | tail = 0; /* Hot objects. Put the slab first */ |
1646 | ||
894b8788 | 1647 | /* Retrieve object from cpu_freelist */ |
dfb4f096 | 1648 | object = c->freelist; |
ff12059e | 1649 | c->freelist = get_freepointer(s, c->freelist); |
894b8788 CL |
1650 | |
1651 | /* And put onto the regular freelist */ | |
ff12059e | 1652 | set_freepointer(s, object, page->freelist); |
894b8788 CL |
1653 | page->freelist = object; |
1654 | page->inuse--; | |
1655 | } | |
dfb4f096 | 1656 | c->page = NULL; |
8a5ec0ba CL |
1657 | #ifdef CONFIG_CMPXCHG_LOCAL |
1658 | c->tid = next_tid(c->tid); | |
1659 | #endif | |
7c2e132c | 1660 | unfreeze_slab(s, page, tail); |
81819f0f CL |
1661 | } |
1662 | ||
dfb4f096 | 1663 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1664 | { |
84e554e6 | 1665 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 CL |
1666 | slab_lock(c->page); |
1667 | deactivate_slab(s, c); | |
81819f0f CL |
1668 | } |
1669 | ||
1670 | /* | |
1671 | * Flush cpu slab. | |
6446faa2 | 1672 | * |
81819f0f CL |
1673 | * Called from IPI handler with interrupts disabled. |
1674 | */ | |
0c710013 | 1675 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1676 | { |
9dfc6e68 | 1677 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1678 | |
dfb4f096 CL |
1679 | if (likely(c && c->page)) |
1680 | flush_slab(s, c); | |
81819f0f CL |
1681 | } |
1682 | ||
1683 | static void flush_cpu_slab(void *d) | |
1684 | { | |
1685 | struct kmem_cache *s = d; | |
81819f0f | 1686 | |
dfb4f096 | 1687 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1688 | } |
1689 | ||
1690 | static void flush_all(struct kmem_cache *s) | |
1691 | { | |
15c8b6c1 | 1692 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1693 | } |
1694 | ||
dfb4f096 CL |
1695 | /* |
1696 | * Check if the objects in a per cpu structure fit numa | |
1697 | * locality expectations. | |
1698 | */ | |
1699 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1700 | { | |
1701 | #ifdef CONFIG_NUMA | |
2154a336 | 1702 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1703 | return 0; |
1704 | #endif | |
1705 | return 1; | |
1706 | } | |
1707 | ||
781b2ba6 PE |
1708 | static int count_free(struct page *page) |
1709 | { | |
1710 | return page->objects - page->inuse; | |
1711 | } | |
1712 | ||
1713 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1714 | int (*get_count)(struct page *)) | |
1715 | { | |
1716 | unsigned long flags; | |
1717 | unsigned long x = 0; | |
1718 | struct page *page; | |
1719 | ||
1720 | spin_lock_irqsave(&n->list_lock, flags); | |
1721 | list_for_each_entry(page, &n->partial, lru) | |
1722 | x += get_count(page); | |
1723 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1724 | return x; | |
1725 | } | |
1726 | ||
26c02cf0 AB |
1727 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1728 | { | |
1729 | #ifdef CONFIG_SLUB_DEBUG | |
1730 | return atomic_long_read(&n->total_objects); | |
1731 | #else | |
1732 | return 0; | |
1733 | #endif | |
1734 | } | |
1735 | ||
781b2ba6 PE |
1736 | static noinline void |
1737 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1738 | { | |
1739 | int node; | |
1740 | ||
1741 | printk(KERN_WARNING | |
1742 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1743 | nid, gfpflags); | |
1744 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1745 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1746 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1747 | ||
fa5ec8a1 DR |
1748 | if (oo_order(s->min) > get_order(s->objsize)) |
1749 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1750 | "slub_debug=O to disable.\n", s->name); | |
1751 | ||
781b2ba6 PE |
1752 | for_each_online_node(node) { |
1753 | struct kmem_cache_node *n = get_node(s, node); | |
1754 | unsigned long nr_slabs; | |
1755 | unsigned long nr_objs; | |
1756 | unsigned long nr_free; | |
1757 | ||
1758 | if (!n) | |
1759 | continue; | |
1760 | ||
26c02cf0 AB |
1761 | nr_free = count_partial(n, count_free); |
1762 | nr_slabs = node_nr_slabs(n); | |
1763 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1764 | |
1765 | printk(KERN_WARNING | |
1766 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1767 | node, nr_slabs, nr_objs, nr_free); | |
1768 | } | |
1769 | } | |
1770 | ||
81819f0f | 1771 | /* |
894b8788 CL |
1772 | * Slow path. The lockless freelist is empty or we need to perform |
1773 | * debugging duties. | |
1774 | * | |
1775 | * Interrupts are disabled. | |
81819f0f | 1776 | * |
894b8788 CL |
1777 | * Processing is still very fast if new objects have been freed to the |
1778 | * regular freelist. In that case we simply take over the regular freelist | |
1779 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1780 | * |
894b8788 CL |
1781 | * If that is not working then we fall back to the partial lists. We take the |
1782 | * first element of the freelist as the object to allocate now and move the | |
1783 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1784 | * |
894b8788 | 1785 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1786 | * we need to allocate a new slab. This is the slowest path since it involves |
1787 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1788 | */ |
ce71e27c EGM |
1789 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1790 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1791 | { |
81819f0f | 1792 | void **object; |
dfb4f096 | 1793 | struct page *new; |
8a5ec0ba CL |
1794 | #ifdef CONFIG_CMPXCHG_LOCAL |
1795 | unsigned long flags; | |
1796 | ||
1797 | local_irq_save(flags); | |
1798 | #ifdef CONFIG_PREEMPT | |
1799 | /* | |
1800 | * We may have been preempted and rescheduled on a different | |
1801 | * cpu before disabling interrupts. Need to reload cpu area | |
1802 | * pointer. | |
1803 | */ | |
1804 | c = this_cpu_ptr(s->cpu_slab); | |
1805 | #endif | |
1806 | #endif | |
81819f0f | 1807 | |
e72e9c23 LT |
1808 | /* We handle __GFP_ZERO in the caller */ |
1809 | gfpflags &= ~__GFP_ZERO; | |
1810 | ||
dfb4f096 | 1811 | if (!c->page) |
81819f0f CL |
1812 | goto new_slab; |
1813 | ||
dfb4f096 CL |
1814 | slab_lock(c->page); |
1815 | if (unlikely(!node_match(c, node))) | |
81819f0f | 1816 | goto another_slab; |
6446faa2 | 1817 | |
84e554e6 | 1818 | stat(s, ALLOC_REFILL); |
6446faa2 | 1819 | |
894b8788 | 1820 | load_freelist: |
dfb4f096 | 1821 | object = c->page->freelist; |
a973e9dd | 1822 | if (unlikely(!object)) |
81819f0f | 1823 | goto another_slab; |
af537b0a | 1824 | if (kmem_cache_debug(s)) |
81819f0f CL |
1825 | goto debug; |
1826 | ||
ff12059e | 1827 | c->freelist = get_freepointer(s, object); |
39b26464 | 1828 | c->page->inuse = c->page->objects; |
a973e9dd | 1829 | c->page->freelist = NULL; |
dfb4f096 | 1830 | c->node = page_to_nid(c->page); |
1f84260c | 1831 | unlock_out: |
dfb4f096 | 1832 | slab_unlock(c->page); |
8a5ec0ba CL |
1833 | #ifdef CONFIG_CMPXCHG_LOCAL |
1834 | c->tid = next_tid(c->tid); | |
1835 | local_irq_restore(flags); | |
1836 | #endif | |
84e554e6 | 1837 | stat(s, ALLOC_SLOWPATH); |
81819f0f CL |
1838 | return object; |
1839 | ||
1840 | another_slab: | |
dfb4f096 | 1841 | deactivate_slab(s, c); |
81819f0f CL |
1842 | |
1843 | new_slab: | |
dfb4f096 CL |
1844 | new = get_partial(s, gfpflags, node); |
1845 | if (new) { | |
1846 | c->page = new; | |
84e554e6 | 1847 | stat(s, ALLOC_FROM_PARTIAL); |
894b8788 | 1848 | goto load_freelist; |
81819f0f CL |
1849 | } |
1850 | ||
c1d50836 | 1851 | gfpflags &= gfp_allowed_mask; |
b811c202 CL |
1852 | if (gfpflags & __GFP_WAIT) |
1853 | local_irq_enable(); | |
1854 | ||
dfb4f096 | 1855 | new = new_slab(s, gfpflags, node); |
b811c202 CL |
1856 | |
1857 | if (gfpflags & __GFP_WAIT) | |
1858 | local_irq_disable(); | |
1859 | ||
dfb4f096 | 1860 | if (new) { |
9dfc6e68 | 1861 | c = __this_cpu_ptr(s->cpu_slab); |
84e554e6 | 1862 | stat(s, ALLOC_SLAB); |
05aa3450 | 1863 | if (c->page) |
dfb4f096 | 1864 | flush_slab(s, c); |
dfb4f096 | 1865 | slab_lock(new); |
8a38082d | 1866 | __SetPageSlubFrozen(new); |
dfb4f096 | 1867 | c->page = new; |
4b6f0750 | 1868 | goto load_freelist; |
81819f0f | 1869 | } |
95f85989 PE |
1870 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
1871 | slab_out_of_memory(s, gfpflags, node); | |
2fd66c51 CL |
1872 | #ifdef CONFIG_CMPXCHG_LOCAL |
1873 | local_irq_restore(flags); | |
1874 | #endif | |
71c7a06f | 1875 | return NULL; |
81819f0f | 1876 | debug: |
dfb4f096 | 1877 | if (!alloc_debug_processing(s, c->page, object, addr)) |
81819f0f | 1878 | goto another_slab; |
894b8788 | 1879 | |
dfb4f096 | 1880 | c->page->inuse++; |
ff12059e | 1881 | c->page->freelist = get_freepointer(s, object); |
15b7c514 | 1882 | c->node = NUMA_NO_NODE; |
1f84260c | 1883 | goto unlock_out; |
894b8788 CL |
1884 | } |
1885 | ||
1886 | /* | |
1887 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
1888 | * have the fastpath folded into their functions. So no function call | |
1889 | * overhead for requests that can be satisfied on the fastpath. | |
1890 | * | |
1891 | * The fastpath works by first checking if the lockless freelist can be used. | |
1892 | * If not then __slab_alloc is called for slow processing. | |
1893 | * | |
1894 | * Otherwise we can simply pick the next object from the lockless free list. | |
1895 | */ | |
06428780 | 1896 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 1897 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 1898 | { |
894b8788 | 1899 | void **object; |
dfb4f096 | 1900 | struct kmem_cache_cpu *c; |
8a5ec0ba CL |
1901 | #ifdef CONFIG_CMPXCHG_LOCAL |
1902 | unsigned long tid; | |
1903 | #else | |
1f84260c | 1904 | unsigned long flags; |
8a5ec0ba | 1905 | #endif |
1f84260c | 1906 | |
c016b0bd | 1907 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 1908 | return NULL; |
1f84260c | 1909 | |
8a5ec0ba | 1910 | #ifndef CONFIG_CMPXCHG_LOCAL |
894b8788 | 1911 | local_irq_save(flags); |
8a5ec0ba CL |
1912 | #else |
1913 | redo: | |
1914 | #endif | |
1915 | ||
1916 | /* | |
1917 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
1918 | * enabled. We may switch back and forth between cpus while | |
1919 | * reading from one cpu area. That does not matter as long | |
1920 | * as we end up on the original cpu again when doing the cmpxchg. | |
1921 | */ | |
9dfc6e68 | 1922 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba CL |
1923 | |
1924 | #ifdef CONFIG_CMPXCHG_LOCAL | |
1925 | /* | |
1926 | * The transaction ids are globally unique per cpu and per operation on | |
1927 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
1928 | * occurs on the right processor and that there was no operation on the | |
1929 | * linked list in between. | |
1930 | */ | |
1931 | tid = c->tid; | |
1932 | barrier(); | |
1933 | #endif | |
1934 | ||
9dfc6e68 | 1935 | object = c->freelist; |
9dfc6e68 | 1936 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 1937 | |
dfb4f096 | 1938 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
1939 | |
1940 | else { | |
8a5ec0ba CL |
1941 | #ifdef CONFIG_CMPXCHG_LOCAL |
1942 | /* | |
1943 | * The cmpxchg will only match if there was no additonal | |
1944 | * operation and if we are on the right processor. | |
1945 | * | |
1946 | * The cmpxchg does the following atomically (without lock semantics!) | |
1947 | * 1. Relocate first pointer to the current per cpu area. | |
1948 | * 2. Verify that tid and freelist have not been changed | |
1949 | * 3. If they were not changed replace tid and freelist | |
1950 | * | |
1951 | * Since this is without lock semantics the protection is only against | |
1952 | * code executing on this cpu *not* from access by other cpus. | |
1953 | */ | |
1954 | if (unlikely(!this_cpu_cmpxchg_double( | |
1955 | s->cpu_slab->freelist, s->cpu_slab->tid, | |
1956 | object, tid, | |
1957 | get_freepointer(s, object), next_tid(tid)))) { | |
1958 | ||
1959 | note_cmpxchg_failure("slab_alloc", s, tid); | |
1960 | goto redo; | |
1961 | } | |
1962 | #else | |
ff12059e | 1963 | c->freelist = get_freepointer(s, object); |
8a5ec0ba | 1964 | #endif |
84e554e6 | 1965 | stat(s, ALLOC_FASTPATH); |
894b8788 | 1966 | } |
8a5ec0ba CL |
1967 | |
1968 | #ifndef CONFIG_CMPXCHG_LOCAL | |
894b8788 | 1969 | local_irq_restore(flags); |
8a5ec0ba | 1970 | #endif |
d07dbea4 | 1971 | |
74e2134f | 1972 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 1973 | memset(object, 0, s->objsize); |
d07dbea4 | 1974 | |
c016b0bd | 1975 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 1976 | |
894b8788 | 1977 | return object; |
81819f0f CL |
1978 | } |
1979 | ||
1980 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1981 | { | |
2154a336 | 1982 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 1983 | |
ca2b84cb | 1984 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
1985 | |
1986 | return ret; | |
81819f0f CL |
1987 | } |
1988 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1989 | ||
0f24f128 | 1990 | #ifdef CONFIG_TRACING |
4a92379b RK |
1991 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
1992 | { | |
1993 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
1994 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
1995 | return ret; | |
1996 | } | |
1997 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
1998 | ||
1999 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2000 | { |
4a92379b RK |
2001 | void *ret = kmalloc_order(size, flags, order); |
2002 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2003 | return ret; | |
5b882be4 | 2004 | } |
4a92379b | 2005 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2006 | #endif |
2007 | ||
81819f0f CL |
2008 | #ifdef CONFIG_NUMA |
2009 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2010 | { | |
5b882be4 EGM |
2011 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2012 | ||
ca2b84cb EGM |
2013 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2014 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2015 | |
2016 | return ret; | |
81819f0f CL |
2017 | } |
2018 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2019 | |
0f24f128 | 2020 | #ifdef CONFIG_TRACING |
4a92379b | 2021 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2022 | gfp_t gfpflags, |
4a92379b | 2023 | int node, size_t size) |
5b882be4 | 2024 | { |
4a92379b RK |
2025 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2026 | ||
2027 | trace_kmalloc_node(_RET_IP_, ret, | |
2028 | size, s->size, gfpflags, node); | |
2029 | return ret; | |
5b882be4 | 2030 | } |
4a92379b | 2031 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2032 | #endif |
5d1f57e4 | 2033 | #endif |
5b882be4 | 2034 | |
81819f0f | 2035 | /* |
894b8788 CL |
2036 | * Slow patch handling. This may still be called frequently since objects |
2037 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2038 | * |
894b8788 CL |
2039 | * So we still attempt to reduce cache line usage. Just take the slab |
2040 | * lock and free the item. If there is no additional partial page | |
2041 | * handling required then we can return immediately. | |
81819f0f | 2042 | */ |
894b8788 | 2043 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2044 | void *x, unsigned long addr) |
81819f0f CL |
2045 | { |
2046 | void *prior; | |
2047 | void **object = (void *)x; | |
8a5ec0ba CL |
2048 | #ifdef CONFIG_CMPXCHG_LOCAL |
2049 | unsigned long flags; | |
81819f0f | 2050 | |
8a5ec0ba CL |
2051 | local_irq_save(flags); |
2052 | #endif | |
81819f0f | 2053 | slab_lock(page); |
8a5ec0ba | 2054 | stat(s, FREE_SLOWPATH); |
81819f0f | 2055 | |
af537b0a | 2056 | if (kmem_cache_debug(s)) |
81819f0f | 2057 | goto debug; |
6446faa2 | 2058 | |
81819f0f | 2059 | checks_ok: |
ff12059e CL |
2060 | prior = page->freelist; |
2061 | set_freepointer(s, object, prior); | |
81819f0f CL |
2062 | page->freelist = object; |
2063 | page->inuse--; | |
2064 | ||
8a38082d | 2065 | if (unlikely(PageSlubFrozen(page))) { |
84e554e6 | 2066 | stat(s, FREE_FROZEN); |
81819f0f | 2067 | goto out_unlock; |
8ff12cfc | 2068 | } |
81819f0f CL |
2069 | |
2070 | if (unlikely(!page->inuse)) | |
2071 | goto slab_empty; | |
2072 | ||
2073 | /* | |
6446faa2 | 2074 | * Objects left in the slab. If it was not on the partial list before |
81819f0f CL |
2075 | * then add it. |
2076 | */ | |
a973e9dd | 2077 | if (unlikely(!prior)) { |
7c2e132c | 2078 | add_partial(get_node(s, page_to_nid(page)), page, 1); |
84e554e6 | 2079 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 2080 | } |
81819f0f CL |
2081 | |
2082 | out_unlock: | |
2083 | slab_unlock(page); | |
8a5ec0ba CL |
2084 | #ifdef CONFIG_CMPXCHG_LOCAL |
2085 | local_irq_restore(flags); | |
2086 | #endif | |
81819f0f CL |
2087 | return; |
2088 | ||
2089 | slab_empty: | |
a973e9dd | 2090 | if (prior) { |
81819f0f | 2091 | /* |
672bba3a | 2092 | * Slab still on the partial list. |
81819f0f CL |
2093 | */ |
2094 | remove_partial(s, page); | |
84e554e6 | 2095 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 2096 | } |
81819f0f | 2097 | slab_unlock(page); |
8a5ec0ba CL |
2098 | #ifdef CONFIG_CMPXCHG_LOCAL |
2099 | local_irq_restore(flags); | |
2100 | #endif | |
84e554e6 | 2101 | stat(s, FREE_SLAB); |
81819f0f | 2102 | discard_slab(s, page); |
81819f0f CL |
2103 | return; |
2104 | ||
2105 | debug: | |
3ec09742 | 2106 | if (!free_debug_processing(s, page, x, addr)) |
77c5e2d0 | 2107 | goto out_unlock; |
77c5e2d0 | 2108 | goto checks_ok; |
81819f0f CL |
2109 | } |
2110 | ||
894b8788 CL |
2111 | /* |
2112 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2113 | * can perform fastpath freeing without additional function calls. | |
2114 | * | |
2115 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2116 | * of this processor. This typically the case if we have just allocated | |
2117 | * the item before. | |
2118 | * | |
2119 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2120 | * with all sorts of special processing. | |
2121 | */ | |
06428780 | 2122 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2123 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2124 | { |
2125 | void **object = (void *)x; | |
dfb4f096 | 2126 | struct kmem_cache_cpu *c; |
8a5ec0ba CL |
2127 | #ifdef CONFIG_CMPXCHG_LOCAL |
2128 | unsigned long tid; | |
2129 | #else | |
1f84260c | 2130 | unsigned long flags; |
8a5ec0ba | 2131 | #endif |
1f84260c | 2132 | |
c016b0bd CL |
2133 | slab_free_hook(s, x); |
2134 | ||
8a5ec0ba | 2135 | #ifndef CONFIG_CMPXCHG_LOCAL |
894b8788 | 2136 | local_irq_save(flags); |
8a5ec0ba | 2137 | |
a24c5a0e | 2138 | #else |
8a5ec0ba | 2139 | redo: |
a24c5a0e CL |
2140 | #endif |
2141 | ||
8a5ec0ba CL |
2142 | /* |
2143 | * Determine the currently cpus per cpu slab. | |
2144 | * The cpu may change afterward. However that does not matter since | |
2145 | * data is retrieved via this pointer. If we are on the same cpu | |
2146 | * during the cmpxchg then the free will succedd. | |
2147 | */ | |
9dfc6e68 | 2148 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2149 | |
8a5ec0ba CL |
2150 | #ifdef CONFIG_CMPXCHG_LOCAL |
2151 | tid = c->tid; | |
2152 | barrier(); | |
2153 | #endif | |
c016b0bd | 2154 | |
15b7c514 | 2155 | if (likely(page == c->page && c->node != NUMA_NO_NODE)) { |
ff12059e | 2156 | set_freepointer(s, object, c->freelist); |
8a5ec0ba CL |
2157 | |
2158 | #ifdef CONFIG_CMPXCHG_LOCAL | |
2159 | if (unlikely(!this_cpu_cmpxchg_double( | |
2160 | s->cpu_slab->freelist, s->cpu_slab->tid, | |
2161 | c->freelist, tid, | |
2162 | object, next_tid(tid)))) { | |
2163 | ||
2164 | note_cmpxchg_failure("slab_free", s, tid); | |
2165 | goto redo; | |
2166 | } | |
2167 | #else | |
dfb4f096 | 2168 | c->freelist = object; |
8a5ec0ba | 2169 | #endif |
84e554e6 | 2170 | stat(s, FREE_FASTPATH); |
894b8788 | 2171 | } else |
ff12059e | 2172 | __slab_free(s, page, x, addr); |
894b8788 | 2173 | |
8a5ec0ba | 2174 | #ifndef CONFIG_CMPXCHG_LOCAL |
894b8788 | 2175 | local_irq_restore(flags); |
8a5ec0ba | 2176 | #endif |
894b8788 CL |
2177 | } |
2178 | ||
81819f0f CL |
2179 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2180 | { | |
77c5e2d0 | 2181 | struct page *page; |
81819f0f | 2182 | |
b49af68f | 2183 | page = virt_to_head_page(x); |
81819f0f | 2184 | |
ce71e27c | 2185 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2186 | |
ca2b84cb | 2187 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2188 | } |
2189 | EXPORT_SYMBOL(kmem_cache_free); | |
2190 | ||
81819f0f | 2191 | /* |
672bba3a CL |
2192 | * Object placement in a slab is made very easy because we always start at |
2193 | * offset 0. If we tune the size of the object to the alignment then we can | |
2194 | * get the required alignment by putting one properly sized object after | |
2195 | * another. | |
81819f0f CL |
2196 | * |
2197 | * Notice that the allocation order determines the sizes of the per cpu | |
2198 | * caches. Each processor has always one slab available for allocations. | |
2199 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2200 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2201 | * locking overhead. |
81819f0f CL |
2202 | */ |
2203 | ||
2204 | /* | |
2205 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2206 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2207 | * and increases the number of allocations possible without having to | |
2208 | * take the list_lock. | |
2209 | */ | |
2210 | static int slub_min_order; | |
114e9e89 | 2211 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2212 | static int slub_min_objects; |
81819f0f CL |
2213 | |
2214 | /* | |
2215 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2216 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2217 | */ |
2218 | static int slub_nomerge; | |
2219 | ||
81819f0f CL |
2220 | /* |
2221 | * Calculate the order of allocation given an slab object size. | |
2222 | * | |
672bba3a CL |
2223 | * The order of allocation has significant impact on performance and other |
2224 | * system components. Generally order 0 allocations should be preferred since | |
2225 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2226 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2227 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2228 | * would be wasted. |
2229 | * | |
2230 | * In order to reach satisfactory performance we must ensure that a minimum | |
2231 | * number of objects is in one slab. Otherwise we may generate too much | |
2232 | * activity on the partial lists which requires taking the list_lock. This is | |
2233 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2234 | * |
672bba3a CL |
2235 | * slub_max_order specifies the order where we begin to stop considering the |
2236 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2237 | * we try to keep the page order as low as possible. So we accept more waste | |
2238 | * of space in favor of a small page order. | |
81819f0f | 2239 | * |
672bba3a CL |
2240 | * Higher order allocations also allow the placement of more objects in a |
2241 | * slab and thereby reduce object handling overhead. If the user has | |
2242 | * requested a higher mininum order then we start with that one instead of | |
2243 | * the smallest order which will fit the object. | |
81819f0f | 2244 | */ |
5e6d444e | 2245 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2246 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2247 | { |
2248 | int order; | |
2249 | int rem; | |
6300ea75 | 2250 | int min_order = slub_min_order; |
81819f0f | 2251 | |
ab9a0f19 | 2252 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2253 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2254 | |
6300ea75 | 2255 | for (order = max(min_order, |
5e6d444e CL |
2256 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2257 | order <= max_order; order++) { | |
81819f0f | 2258 | |
5e6d444e | 2259 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2260 | |
ab9a0f19 | 2261 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2262 | continue; |
2263 | ||
ab9a0f19 | 2264 | rem = (slab_size - reserved) % size; |
81819f0f | 2265 | |
5e6d444e | 2266 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2267 | break; |
2268 | ||
2269 | } | |
672bba3a | 2270 | |
81819f0f CL |
2271 | return order; |
2272 | } | |
2273 | ||
ab9a0f19 | 2274 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2275 | { |
2276 | int order; | |
2277 | int min_objects; | |
2278 | int fraction; | |
e8120ff1 | 2279 | int max_objects; |
5e6d444e CL |
2280 | |
2281 | /* | |
2282 | * Attempt to find best configuration for a slab. This | |
2283 | * works by first attempting to generate a layout with | |
2284 | * the best configuration and backing off gradually. | |
2285 | * | |
2286 | * First we reduce the acceptable waste in a slab. Then | |
2287 | * we reduce the minimum objects required in a slab. | |
2288 | */ | |
2289 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2290 | if (!min_objects) |
2291 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2292 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2293 | min_objects = min(min_objects, max_objects); |
2294 | ||
5e6d444e | 2295 | while (min_objects > 1) { |
c124f5b5 | 2296 | fraction = 16; |
5e6d444e CL |
2297 | while (fraction >= 4) { |
2298 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2299 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2300 | if (order <= slub_max_order) |
2301 | return order; | |
2302 | fraction /= 2; | |
2303 | } | |
5086c389 | 2304 | min_objects--; |
5e6d444e CL |
2305 | } |
2306 | ||
2307 | /* | |
2308 | * We were unable to place multiple objects in a slab. Now | |
2309 | * lets see if we can place a single object there. | |
2310 | */ | |
ab9a0f19 | 2311 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2312 | if (order <= slub_max_order) |
2313 | return order; | |
2314 | ||
2315 | /* | |
2316 | * Doh this slab cannot be placed using slub_max_order. | |
2317 | */ | |
ab9a0f19 | 2318 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2319 | if (order < MAX_ORDER) |
5e6d444e CL |
2320 | return order; |
2321 | return -ENOSYS; | |
2322 | } | |
2323 | ||
81819f0f | 2324 | /* |
672bba3a | 2325 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2326 | */ |
2327 | static unsigned long calculate_alignment(unsigned long flags, | |
2328 | unsigned long align, unsigned long size) | |
2329 | { | |
2330 | /* | |
6446faa2 CL |
2331 | * If the user wants hardware cache aligned objects then follow that |
2332 | * suggestion if the object is sufficiently large. | |
81819f0f | 2333 | * |
6446faa2 CL |
2334 | * The hardware cache alignment cannot override the specified |
2335 | * alignment though. If that is greater then use it. | |
81819f0f | 2336 | */ |
b6210386 NP |
2337 | if (flags & SLAB_HWCACHE_ALIGN) { |
2338 | unsigned long ralign = cache_line_size(); | |
2339 | while (size <= ralign / 2) | |
2340 | ralign /= 2; | |
2341 | align = max(align, ralign); | |
2342 | } | |
81819f0f CL |
2343 | |
2344 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2345 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2346 | |
2347 | return ALIGN(align, sizeof(void *)); | |
2348 | } | |
2349 | ||
5595cffc PE |
2350 | static void |
2351 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2352 | { |
2353 | n->nr_partial = 0; | |
81819f0f CL |
2354 | spin_lock_init(&n->list_lock); |
2355 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2356 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2357 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2358 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2359 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2360 | #endif |
81819f0f CL |
2361 | } |
2362 | ||
55136592 | 2363 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2364 | { |
6c182dc0 CL |
2365 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2366 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2367 | |
8a5ec0ba CL |
2368 | #ifdef CONFIG_CMPXCHG_LOCAL |
2369 | /* | |
2370 | * Must align to double word boundary for the double cmpxchg instructions | |
2371 | * to work. | |
2372 | */ | |
2373 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *)); | |
2374 | #else | |
2375 | /* Regular alignment is sufficient */ | |
6c182dc0 | 2376 | s->cpu_slab = alloc_percpu(struct kmem_cache_cpu); |
8a5ec0ba CL |
2377 | #endif |
2378 | ||
2379 | if (!s->cpu_slab) | |
2380 | return 0; | |
2381 | ||
2382 | init_kmem_cache_cpus(s); | |
4c93c355 | 2383 | |
8a5ec0ba | 2384 | return 1; |
4c93c355 | 2385 | } |
4c93c355 | 2386 | |
51df1142 CL |
2387 | static struct kmem_cache *kmem_cache_node; |
2388 | ||
81819f0f CL |
2389 | /* |
2390 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2391 | * slab on the node for this slabcache. There are no concurrent accesses | |
2392 | * possible. | |
2393 | * | |
2394 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2395 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2396 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2397 | */ |
55136592 | 2398 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2399 | { |
2400 | struct page *page; | |
2401 | struct kmem_cache_node *n; | |
ba84c73c | 2402 | unsigned long flags; |
81819f0f | 2403 | |
51df1142 | 2404 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2405 | |
51df1142 | 2406 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2407 | |
2408 | BUG_ON(!page); | |
a2f92ee7 CL |
2409 | if (page_to_nid(page) != node) { |
2410 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2411 | "node %d\n", node); | |
2412 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2413 | "in order to be able to continue\n"); | |
2414 | } | |
2415 | ||
81819f0f CL |
2416 | n = page->freelist; |
2417 | BUG_ON(!n); | |
51df1142 | 2418 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2419 | page->inuse++; |
51df1142 | 2420 | kmem_cache_node->node[node] = n; |
8ab1372f | 2421 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2422 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2423 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2424 | #endif |
51df1142 CL |
2425 | init_kmem_cache_node(n, kmem_cache_node); |
2426 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2427 | |
ba84c73c | 2428 | /* |
2429 | * lockdep requires consistent irq usage for each lock | |
2430 | * so even though there cannot be a race this early in | |
2431 | * the boot sequence, we still disable irqs. | |
2432 | */ | |
2433 | local_irq_save(flags); | |
7c2e132c | 2434 | add_partial(n, page, 0); |
ba84c73c | 2435 | local_irq_restore(flags); |
81819f0f CL |
2436 | } |
2437 | ||
2438 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2439 | { | |
2440 | int node; | |
2441 | ||
f64dc58c | 2442 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2443 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2444 | |
73367bd8 | 2445 | if (n) |
51df1142 CL |
2446 | kmem_cache_free(kmem_cache_node, n); |
2447 | ||
81819f0f CL |
2448 | s->node[node] = NULL; |
2449 | } | |
2450 | } | |
2451 | ||
55136592 | 2452 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2453 | { |
2454 | int node; | |
81819f0f | 2455 | |
f64dc58c | 2456 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2457 | struct kmem_cache_node *n; |
2458 | ||
73367bd8 | 2459 | if (slab_state == DOWN) { |
55136592 | 2460 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2461 | continue; |
2462 | } | |
51df1142 | 2463 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2464 | GFP_KERNEL, node); |
81819f0f | 2465 | |
73367bd8 AD |
2466 | if (!n) { |
2467 | free_kmem_cache_nodes(s); | |
2468 | return 0; | |
81819f0f | 2469 | } |
73367bd8 | 2470 | |
81819f0f | 2471 | s->node[node] = n; |
5595cffc | 2472 | init_kmem_cache_node(n, s); |
81819f0f CL |
2473 | } |
2474 | return 1; | |
2475 | } | |
81819f0f | 2476 | |
c0bdb232 | 2477 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2478 | { |
2479 | if (min < MIN_PARTIAL) | |
2480 | min = MIN_PARTIAL; | |
2481 | else if (min > MAX_PARTIAL) | |
2482 | min = MAX_PARTIAL; | |
2483 | s->min_partial = min; | |
2484 | } | |
2485 | ||
81819f0f CL |
2486 | /* |
2487 | * calculate_sizes() determines the order and the distribution of data within | |
2488 | * a slab object. | |
2489 | */ | |
06b285dc | 2490 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2491 | { |
2492 | unsigned long flags = s->flags; | |
2493 | unsigned long size = s->objsize; | |
2494 | unsigned long align = s->align; | |
834f3d11 | 2495 | int order; |
81819f0f | 2496 | |
d8b42bf5 CL |
2497 | /* |
2498 | * Round up object size to the next word boundary. We can only | |
2499 | * place the free pointer at word boundaries and this determines | |
2500 | * the possible location of the free pointer. | |
2501 | */ | |
2502 | size = ALIGN(size, sizeof(void *)); | |
2503 | ||
2504 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2505 | /* |
2506 | * Determine if we can poison the object itself. If the user of | |
2507 | * the slab may touch the object after free or before allocation | |
2508 | * then we should never poison the object itself. | |
2509 | */ | |
2510 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2511 | !s->ctor) |
81819f0f CL |
2512 | s->flags |= __OBJECT_POISON; |
2513 | else | |
2514 | s->flags &= ~__OBJECT_POISON; | |
2515 | ||
81819f0f CL |
2516 | |
2517 | /* | |
672bba3a | 2518 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2519 | * end of the object and the free pointer. If not then add an |
672bba3a | 2520 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2521 | */ |
2522 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2523 | size += sizeof(void *); | |
41ecc55b | 2524 | #endif |
81819f0f CL |
2525 | |
2526 | /* | |
672bba3a CL |
2527 | * With that we have determined the number of bytes in actual use |
2528 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2529 | */ |
2530 | s->inuse = size; | |
2531 | ||
2532 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2533 | s->ctor)) { |
81819f0f CL |
2534 | /* |
2535 | * Relocate free pointer after the object if it is not | |
2536 | * permitted to overwrite the first word of the object on | |
2537 | * kmem_cache_free. | |
2538 | * | |
2539 | * This is the case if we do RCU, have a constructor or | |
2540 | * destructor or are poisoning the objects. | |
2541 | */ | |
2542 | s->offset = size; | |
2543 | size += sizeof(void *); | |
2544 | } | |
2545 | ||
c12b3c62 | 2546 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2547 | if (flags & SLAB_STORE_USER) |
2548 | /* | |
2549 | * Need to store information about allocs and frees after | |
2550 | * the object. | |
2551 | */ | |
2552 | size += 2 * sizeof(struct track); | |
2553 | ||
be7b3fbc | 2554 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2555 | /* |
2556 | * Add some empty padding so that we can catch | |
2557 | * overwrites from earlier objects rather than let | |
2558 | * tracking information or the free pointer be | |
0211a9c8 | 2559 | * corrupted if a user writes before the start |
81819f0f CL |
2560 | * of the object. |
2561 | */ | |
2562 | size += sizeof(void *); | |
41ecc55b | 2563 | #endif |
672bba3a | 2564 | |
81819f0f CL |
2565 | /* |
2566 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2567 | * user specified and the dynamic determination of cache line size |
2568 | * on bootup. | |
81819f0f CL |
2569 | */ |
2570 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2571 | s->align = align; |
81819f0f CL |
2572 | |
2573 | /* | |
2574 | * SLUB stores one object immediately after another beginning from | |
2575 | * offset 0. In order to align the objects we have to simply size | |
2576 | * each object to conform to the alignment. | |
2577 | */ | |
2578 | size = ALIGN(size, align); | |
2579 | s->size = size; | |
06b285dc CL |
2580 | if (forced_order >= 0) |
2581 | order = forced_order; | |
2582 | else | |
ab9a0f19 | 2583 | order = calculate_order(size, s->reserved); |
81819f0f | 2584 | |
834f3d11 | 2585 | if (order < 0) |
81819f0f CL |
2586 | return 0; |
2587 | ||
b7a49f0d | 2588 | s->allocflags = 0; |
834f3d11 | 2589 | if (order) |
b7a49f0d CL |
2590 | s->allocflags |= __GFP_COMP; |
2591 | ||
2592 | if (s->flags & SLAB_CACHE_DMA) | |
2593 | s->allocflags |= SLUB_DMA; | |
2594 | ||
2595 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2596 | s->allocflags |= __GFP_RECLAIMABLE; | |
2597 | ||
81819f0f CL |
2598 | /* |
2599 | * Determine the number of objects per slab | |
2600 | */ | |
ab9a0f19 LJ |
2601 | s->oo = oo_make(order, size, s->reserved); |
2602 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2603 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2604 | s->max = s->oo; | |
81819f0f | 2605 | |
834f3d11 | 2606 | return !!oo_objects(s->oo); |
81819f0f CL |
2607 | |
2608 | } | |
2609 | ||
55136592 | 2610 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2611 | const char *name, size_t size, |
2612 | size_t align, unsigned long flags, | |
51cc5068 | 2613 | void (*ctor)(void *)) |
81819f0f CL |
2614 | { |
2615 | memset(s, 0, kmem_size); | |
2616 | s->name = name; | |
2617 | s->ctor = ctor; | |
81819f0f | 2618 | s->objsize = size; |
81819f0f | 2619 | s->align = align; |
ba0268a8 | 2620 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 2621 | s->reserved = 0; |
81819f0f | 2622 | |
da9a638c LJ |
2623 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
2624 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 2625 | |
06b285dc | 2626 | if (!calculate_sizes(s, -1)) |
81819f0f | 2627 | goto error; |
3de47213 DR |
2628 | if (disable_higher_order_debug) { |
2629 | /* | |
2630 | * Disable debugging flags that store metadata if the min slab | |
2631 | * order increased. | |
2632 | */ | |
2633 | if (get_order(s->size) > get_order(s->objsize)) { | |
2634 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2635 | s->offset = 0; | |
2636 | if (!calculate_sizes(s, -1)) | |
2637 | goto error; | |
2638 | } | |
2639 | } | |
81819f0f | 2640 | |
3b89d7d8 DR |
2641 | /* |
2642 | * The larger the object size is, the more pages we want on the partial | |
2643 | * list to avoid pounding the page allocator excessively. | |
2644 | */ | |
c0bdb232 | 2645 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2646 | s->refcount = 1; |
2647 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2648 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2649 | #endif |
55136592 | 2650 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2651 | goto error; |
81819f0f | 2652 | |
55136592 | 2653 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2654 | return 1; |
ff12059e | 2655 | |
4c93c355 | 2656 | free_kmem_cache_nodes(s); |
81819f0f CL |
2657 | error: |
2658 | if (flags & SLAB_PANIC) | |
2659 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2660 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2661 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2662 | s->offset, flags); |
2663 | return 0; | |
2664 | } | |
81819f0f | 2665 | |
81819f0f CL |
2666 | /* |
2667 | * Determine the size of a slab object | |
2668 | */ | |
2669 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2670 | { | |
2671 | return s->objsize; | |
2672 | } | |
2673 | EXPORT_SYMBOL(kmem_cache_size); | |
2674 | ||
33b12c38 CL |
2675 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2676 | const char *text) | |
2677 | { | |
2678 | #ifdef CONFIG_SLUB_DEBUG | |
2679 | void *addr = page_address(page); | |
2680 | void *p; | |
a5dd5c11 NK |
2681 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2682 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2683 | if (!map) |
2684 | return; | |
33b12c38 CL |
2685 | slab_err(s, page, "%s", text); |
2686 | slab_lock(page); | |
33b12c38 | 2687 | |
5f80b13a | 2688 | get_map(s, page, map); |
33b12c38 CL |
2689 | for_each_object(p, s, addr, page->objects) { |
2690 | ||
2691 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2692 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2693 | p, p - addr); | |
2694 | print_tracking(s, p); | |
2695 | } | |
2696 | } | |
2697 | slab_unlock(page); | |
bbd7d57b | 2698 | kfree(map); |
33b12c38 CL |
2699 | #endif |
2700 | } | |
2701 | ||
81819f0f | 2702 | /* |
599870b1 | 2703 | * Attempt to free all partial slabs on a node. |
81819f0f | 2704 | */ |
599870b1 | 2705 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2706 | { |
81819f0f CL |
2707 | unsigned long flags; |
2708 | struct page *page, *h; | |
2709 | ||
2710 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2711 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2712 | if (!page->inuse) { |
62e346a8 | 2713 | __remove_partial(n, page); |
81819f0f | 2714 | discard_slab(s, page); |
33b12c38 CL |
2715 | } else { |
2716 | list_slab_objects(s, page, | |
2717 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2718 | } |
33b12c38 | 2719 | } |
81819f0f | 2720 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2721 | } |
2722 | ||
2723 | /* | |
672bba3a | 2724 | * Release all resources used by a slab cache. |
81819f0f | 2725 | */ |
0c710013 | 2726 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2727 | { |
2728 | int node; | |
2729 | ||
2730 | flush_all(s); | |
9dfc6e68 | 2731 | free_percpu(s->cpu_slab); |
81819f0f | 2732 | /* Attempt to free all objects */ |
f64dc58c | 2733 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2734 | struct kmem_cache_node *n = get_node(s, node); |
2735 | ||
599870b1 CL |
2736 | free_partial(s, n); |
2737 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2738 | return 1; |
2739 | } | |
2740 | free_kmem_cache_nodes(s); | |
2741 | return 0; | |
2742 | } | |
2743 | ||
2744 | /* | |
2745 | * Close a cache and release the kmem_cache structure | |
2746 | * (must be used for caches created using kmem_cache_create) | |
2747 | */ | |
2748 | void kmem_cache_destroy(struct kmem_cache *s) | |
2749 | { | |
2750 | down_write(&slub_lock); | |
2751 | s->refcount--; | |
2752 | if (!s->refcount) { | |
2753 | list_del(&s->list); | |
d629d819 PE |
2754 | if (kmem_cache_close(s)) { |
2755 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2756 | "still has objects.\n", s->name, __func__); | |
2757 | dump_stack(); | |
2758 | } | |
d76b1590 ED |
2759 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2760 | rcu_barrier(); | |
81819f0f | 2761 | sysfs_slab_remove(s); |
2bce6485 CL |
2762 | } |
2763 | up_write(&slub_lock); | |
81819f0f CL |
2764 | } |
2765 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2766 | ||
2767 | /******************************************************************** | |
2768 | * Kmalloc subsystem | |
2769 | *******************************************************************/ | |
2770 | ||
51df1142 | 2771 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2772 | EXPORT_SYMBOL(kmalloc_caches); |
2773 | ||
51df1142 CL |
2774 | static struct kmem_cache *kmem_cache; |
2775 | ||
55136592 | 2776 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2777 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2778 | #endif |
2779 | ||
81819f0f CL |
2780 | static int __init setup_slub_min_order(char *str) |
2781 | { | |
06428780 | 2782 | get_option(&str, &slub_min_order); |
81819f0f CL |
2783 | |
2784 | return 1; | |
2785 | } | |
2786 | ||
2787 | __setup("slub_min_order=", setup_slub_min_order); | |
2788 | ||
2789 | static int __init setup_slub_max_order(char *str) | |
2790 | { | |
06428780 | 2791 | get_option(&str, &slub_max_order); |
818cf590 | 2792 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2793 | |
2794 | return 1; | |
2795 | } | |
2796 | ||
2797 | __setup("slub_max_order=", setup_slub_max_order); | |
2798 | ||
2799 | static int __init setup_slub_min_objects(char *str) | |
2800 | { | |
06428780 | 2801 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2802 | |
2803 | return 1; | |
2804 | } | |
2805 | ||
2806 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2807 | ||
2808 | static int __init setup_slub_nomerge(char *str) | |
2809 | { | |
2810 | slub_nomerge = 1; | |
2811 | return 1; | |
2812 | } | |
2813 | ||
2814 | __setup("slub_nomerge", setup_slub_nomerge); | |
2815 | ||
51df1142 CL |
2816 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2817 | int size, unsigned int flags) | |
81819f0f | 2818 | { |
51df1142 CL |
2819 | struct kmem_cache *s; |
2820 | ||
2821 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2822 | ||
83b519e8 PE |
2823 | /* |
2824 | * This function is called with IRQs disabled during early-boot on | |
2825 | * single CPU so there's no need to take slub_lock here. | |
2826 | */ | |
55136592 | 2827 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2828 | flags, NULL)) |
81819f0f CL |
2829 | goto panic; |
2830 | ||
2831 | list_add(&s->list, &slab_caches); | |
51df1142 | 2832 | return s; |
81819f0f CL |
2833 | |
2834 | panic: | |
2835 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2836 | return NULL; |
81819f0f CL |
2837 | } |
2838 | ||
f1b26339 CL |
2839 | /* |
2840 | * Conversion table for small slabs sizes / 8 to the index in the | |
2841 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
2842 | * of two cache sizes there. The size of larger slabs can be determined using | |
2843 | * fls. | |
2844 | */ | |
2845 | static s8 size_index[24] = { | |
2846 | 3, /* 8 */ | |
2847 | 4, /* 16 */ | |
2848 | 5, /* 24 */ | |
2849 | 5, /* 32 */ | |
2850 | 6, /* 40 */ | |
2851 | 6, /* 48 */ | |
2852 | 6, /* 56 */ | |
2853 | 6, /* 64 */ | |
2854 | 1, /* 72 */ | |
2855 | 1, /* 80 */ | |
2856 | 1, /* 88 */ | |
2857 | 1, /* 96 */ | |
2858 | 7, /* 104 */ | |
2859 | 7, /* 112 */ | |
2860 | 7, /* 120 */ | |
2861 | 7, /* 128 */ | |
2862 | 2, /* 136 */ | |
2863 | 2, /* 144 */ | |
2864 | 2, /* 152 */ | |
2865 | 2, /* 160 */ | |
2866 | 2, /* 168 */ | |
2867 | 2, /* 176 */ | |
2868 | 2, /* 184 */ | |
2869 | 2 /* 192 */ | |
2870 | }; | |
2871 | ||
acdfcd04 AK |
2872 | static inline int size_index_elem(size_t bytes) |
2873 | { | |
2874 | return (bytes - 1) / 8; | |
2875 | } | |
2876 | ||
81819f0f CL |
2877 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
2878 | { | |
f1b26339 | 2879 | int index; |
81819f0f | 2880 | |
f1b26339 CL |
2881 | if (size <= 192) { |
2882 | if (!size) | |
2883 | return ZERO_SIZE_PTR; | |
81819f0f | 2884 | |
acdfcd04 | 2885 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 2886 | } else |
f1b26339 | 2887 | index = fls(size - 1); |
81819f0f CL |
2888 | |
2889 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 2890 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 2891 | return kmalloc_dma_caches[index]; |
f1b26339 | 2892 | |
81819f0f | 2893 | #endif |
51df1142 | 2894 | return kmalloc_caches[index]; |
81819f0f CL |
2895 | } |
2896 | ||
2897 | void *__kmalloc(size_t size, gfp_t flags) | |
2898 | { | |
aadb4bc4 | 2899 | struct kmem_cache *s; |
5b882be4 | 2900 | void *ret; |
81819f0f | 2901 | |
ffadd4d0 | 2902 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 2903 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
2904 | |
2905 | s = get_slab(size, flags); | |
2906 | ||
2907 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2908 | return s; |
2909 | ||
2154a336 | 2910 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2911 | |
ca2b84cb | 2912 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
2913 | |
2914 | return ret; | |
81819f0f CL |
2915 | } |
2916 | EXPORT_SYMBOL(__kmalloc); | |
2917 | ||
5d1f57e4 | 2918 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
2919 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
2920 | { | |
b1eeab67 | 2921 | struct page *page; |
e4f7c0b4 | 2922 | void *ptr = NULL; |
f619cfe1 | 2923 | |
b1eeab67 VN |
2924 | flags |= __GFP_COMP | __GFP_NOTRACK; |
2925 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 2926 | if (page) |
e4f7c0b4 CM |
2927 | ptr = page_address(page); |
2928 | ||
2929 | kmemleak_alloc(ptr, size, 1, flags); | |
2930 | return ptr; | |
f619cfe1 CL |
2931 | } |
2932 | ||
81819f0f CL |
2933 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
2934 | { | |
aadb4bc4 | 2935 | struct kmem_cache *s; |
5b882be4 | 2936 | void *ret; |
81819f0f | 2937 | |
057685cf | 2938 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
2939 | ret = kmalloc_large_node(size, flags, node); |
2940 | ||
ca2b84cb EGM |
2941 | trace_kmalloc_node(_RET_IP_, ret, |
2942 | size, PAGE_SIZE << get_order(size), | |
2943 | flags, node); | |
5b882be4 EGM |
2944 | |
2945 | return ret; | |
2946 | } | |
aadb4bc4 CL |
2947 | |
2948 | s = get_slab(size, flags); | |
2949 | ||
2950 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
2951 | return s; |
2952 | ||
5b882be4 EGM |
2953 | ret = slab_alloc(s, flags, node, _RET_IP_); |
2954 | ||
ca2b84cb | 2955 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
2956 | |
2957 | return ret; | |
81819f0f CL |
2958 | } |
2959 | EXPORT_SYMBOL(__kmalloc_node); | |
2960 | #endif | |
2961 | ||
2962 | size_t ksize(const void *object) | |
2963 | { | |
272c1d21 | 2964 | struct page *page; |
81819f0f | 2965 | |
ef8b4520 | 2966 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
2967 | return 0; |
2968 | ||
294a80a8 | 2969 | page = virt_to_head_page(object); |
294a80a8 | 2970 | |
76994412 PE |
2971 | if (unlikely(!PageSlab(page))) { |
2972 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 2973 | return PAGE_SIZE << compound_order(page); |
76994412 | 2974 | } |
81819f0f | 2975 | |
b3d41885 | 2976 | return slab_ksize(page->slab); |
81819f0f | 2977 | } |
b1aabecd | 2978 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
2979 | |
2980 | void kfree(const void *x) | |
2981 | { | |
81819f0f | 2982 | struct page *page; |
5bb983b0 | 2983 | void *object = (void *)x; |
81819f0f | 2984 | |
2121db74 PE |
2985 | trace_kfree(_RET_IP_, x); |
2986 | ||
2408c550 | 2987 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
2988 | return; |
2989 | ||
b49af68f | 2990 | page = virt_to_head_page(x); |
aadb4bc4 | 2991 | if (unlikely(!PageSlab(page))) { |
0937502a | 2992 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 2993 | kmemleak_free(x); |
aadb4bc4 CL |
2994 | put_page(page); |
2995 | return; | |
2996 | } | |
ce71e27c | 2997 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
2998 | } |
2999 | EXPORT_SYMBOL(kfree); | |
3000 | ||
2086d26a | 3001 | /* |
672bba3a CL |
3002 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3003 | * the remaining slabs by the number of items in use. The slabs with the | |
3004 | * most items in use come first. New allocations will then fill those up | |
3005 | * and thus they can be removed from the partial lists. | |
3006 | * | |
3007 | * The slabs with the least items are placed last. This results in them | |
3008 | * being allocated from last increasing the chance that the last objects | |
3009 | * are freed in them. | |
2086d26a CL |
3010 | */ |
3011 | int kmem_cache_shrink(struct kmem_cache *s) | |
3012 | { | |
3013 | int node; | |
3014 | int i; | |
3015 | struct kmem_cache_node *n; | |
3016 | struct page *page; | |
3017 | struct page *t; | |
205ab99d | 3018 | int objects = oo_objects(s->max); |
2086d26a | 3019 | struct list_head *slabs_by_inuse = |
834f3d11 | 3020 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3021 | unsigned long flags; |
3022 | ||
3023 | if (!slabs_by_inuse) | |
3024 | return -ENOMEM; | |
3025 | ||
3026 | flush_all(s); | |
f64dc58c | 3027 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3028 | n = get_node(s, node); |
3029 | ||
3030 | if (!n->nr_partial) | |
3031 | continue; | |
3032 | ||
834f3d11 | 3033 | for (i = 0; i < objects; i++) |
2086d26a CL |
3034 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3035 | ||
3036 | spin_lock_irqsave(&n->list_lock, flags); | |
3037 | ||
3038 | /* | |
672bba3a | 3039 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3040 | * |
672bba3a CL |
3041 | * Note that concurrent frees may occur while we hold the |
3042 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3043 | */ |
3044 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
3045 | if (!page->inuse && slab_trylock(page)) { | |
3046 | /* | |
3047 | * Must hold slab lock here because slab_free | |
3048 | * may have freed the last object and be | |
3049 | * waiting to release the slab. | |
3050 | */ | |
62e346a8 | 3051 | __remove_partial(n, page); |
2086d26a CL |
3052 | slab_unlock(page); |
3053 | discard_slab(s, page); | |
3054 | } else { | |
fcda3d89 CL |
3055 | list_move(&page->lru, |
3056 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
3057 | } |
3058 | } | |
3059 | ||
2086d26a | 3060 | /* |
672bba3a CL |
3061 | * Rebuild the partial list with the slabs filled up most |
3062 | * first and the least used slabs at the end. | |
2086d26a | 3063 | */ |
834f3d11 | 3064 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3065 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3066 | ||
2086d26a CL |
3067 | spin_unlock_irqrestore(&n->list_lock, flags); |
3068 | } | |
3069 | ||
3070 | kfree(slabs_by_inuse); | |
3071 | return 0; | |
3072 | } | |
3073 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3074 | ||
92a5bbc1 | 3075 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3076 | static int slab_mem_going_offline_callback(void *arg) |
3077 | { | |
3078 | struct kmem_cache *s; | |
3079 | ||
3080 | down_read(&slub_lock); | |
3081 | list_for_each_entry(s, &slab_caches, list) | |
3082 | kmem_cache_shrink(s); | |
3083 | up_read(&slub_lock); | |
3084 | ||
3085 | return 0; | |
3086 | } | |
3087 | ||
3088 | static void slab_mem_offline_callback(void *arg) | |
3089 | { | |
3090 | struct kmem_cache_node *n; | |
3091 | struct kmem_cache *s; | |
3092 | struct memory_notify *marg = arg; | |
3093 | int offline_node; | |
3094 | ||
3095 | offline_node = marg->status_change_nid; | |
3096 | ||
3097 | /* | |
3098 | * If the node still has available memory. we need kmem_cache_node | |
3099 | * for it yet. | |
3100 | */ | |
3101 | if (offline_node < 0) | |
3102 | return; | |
3103 | ||
3104 | down_read(&slub_lock); | |
3105 | list_for_each_entry(s, &slab_caches, list) { | |
3106 | n = get_node(s, offline_node); | |
3107 | if (n) { | |
3108 | /* | |
3109 | * if n->nr_slabs > 0, slabs still exist on the node | |
3110 | * that is going down. We were unable to free them, | |
c9404c9c | 3111 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3112 | * callback. So, we must fail. |
3113 | */ | |
0f389ec6 | 3114 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3115 | |
3116 | s->node[offline_node] = NULL; | |
8de66a0c | 3117 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3118 | } |
3119 | } | |
3120 | up_read(&slub_lock); | |
3121 | } | |
3122 | ||
3123 | static int slab_mem_going_online_callback(void *arg) | |
3124 | { | |
3125 | struct kmem_cache_node *n; | |
3126 | struct kmem_cache *s; | |
3127 | struct memory_notify *marg = arg; | |
3128 | int nid = marg->status_change_nid; | |
3129 | int ret = 0; | |
3130 | ||
3131 | /* | |
3132 | * If the node's memory is already available, then kmem_cache_node is | |
3133 | * already created. Nothing to do. | |
3134 | */ | |
3135 | if (nid < 0) | |
3136 | return 0; | |
3137 | ||
3138 | /* | |
0121c619 | 3139 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3140 | * allocate a kmem_cache_node structure in order to bring the node |
3141 | * online. | |
3142 | */ | |
3143 | down_read(&slub_lock); | |
3144 | list_for_each_entry(s, &slab_caches, list) { | |
3145 | /* | |
3146 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3147 | * since memory is not yet available from the node that | |
3148 | * is brought up. | |
3149 | */ | |
8de66a0c | 3150 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3151 | if (!n) { |
3152 | ret = -ENOMEM; | |
3153 | goto out; | |
3154 | } | |
5595cffc | 3155 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3156 | s->node[nid] = n; |
3157 | } | |
3158 | out: | |
3159 | up_read(&slub_lock); | |
3160 | return ret; | |
3161 | } | |
3162 | ||
3163 | static int slab_memory_callback(struct notifier_block *self, | |
3164 | unsigned long action, void *arg) | |
3165 | { | |
3166 | int ret = 0; | |
3167 | ||
3168 | switch (action) { | |
3169 | case MEM_GOING_ONLINE: | |
3170 | ret = slab_mem_going_online_callback(arg); | |
3171 | break; | |
3172 | case MEM_GOING_OFFLINE: | |
3173 | ret = slab_mem_going_offline_callback(arg); | |
3174 | break; | |
3175 | case MEM_OFFLINE: | |
3176 | case MEM_CANCEL_ONLINE: | |
3177 | slab_mem_offline_callback(arg); | |
3178 | break; | |
3179 | case MEM_ONLINE: | |
3180 | case MEM_CANCEL_OFFLINE: | |
3181 | break; | |
3182 | } | |
dc19f9db KH |
3183 | if (ret) |
3184 | ret = notifier_from_errno(ret); | |
3185 | else | |
3186 | ret = NOTIFY_OK; | |
b9049e23 YG |
3187 | return ret; |
3188 | } | |
3189 | ||
3190 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3191 | ||
81819f0f CL |
3192 | /******************************************************************** |
3193 | * Basic setup of slabs | |
3194 | *******************************************************************/ | |
3195 | ||
51df1142 CL |
3196 | /* |
3197 | * Used for early kmem_cache structures that were allocated using | |
3198 | * the page allocator | |
3199 | */ | |
3200 | ||
3201 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3202 | { | |
3203 | int node; | |
3204 | ||
3205 | list_add(&s->list, &slab_caches); | |
3206 | s->refcount = -1; | |
3207 | ||
3208 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3209 | struct kmem_cache_node *n = get_node(s, node); | |
3210 | struct page *p; | |
3211 | ||
3212 | if (n) { | |
3213 | list_for_each_entry(p, &n->partial, lru) | |
3214 | p->slab = s; | |
3215 | ||
607bf324 | 3216 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3217 | list_for_each_entry(p, &n->full, lru) |
3218 | p->slab = s; | |
3219 | #endif | |
3220 | } | |
3221 | } | |
3222 | } | |
3223 | ||
81819f0f CL |
3224 | void __init kmem_cache_init(void) |
3225 | { | |
3226 | int i; | |
4b356be0 | 3227 | int caches = 0; |
51df1142 CL |
3228 | struct kmem_cache *temp_kmem_cache; |
3229 | int order; | |
51df1142 CL |
3230 | struct kmem_cache *temp_kmem_cache_node; |
3231 | unsigned long kmalloc_size; | |
3232 | ||
3233 | kmem_size = offsetof(struct kmem_cache, node) + | |
3234 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3235 | ||
3236 | /* Allocate two kmem_caches from the page allocator */ | |
3237 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3238 | order = get_order(2 * kmalloc_size); | |
3239 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3240 | ||
81819f0f CL |
3241 | /* |
3242 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3243 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3244 | * kmem_cache_open for slab_state == DOWN. |
3245 | */ | |
51df1142 CL |
3246 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3247 | ||
3248 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3249 | sizeof(struct kmem_cache_node), | |
3250 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3251 | |
0c40ba4f | 3252 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3253 | |
3254 | /* Able to allocate the per node structures */ | |
3255 | slab_state = PARTIAL; | |
3256 | ||
51df1142 CL |
3257 | temp_kmem_cache = kmem_cache; |
3258 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3259 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3260 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3261 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3262 | |
51df1142 CL |
3263 | /* |
3264 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3265 | * kmem_cache_node is separately allocated so no need to | |
3266 | * update any list pointers. | |
3267 | */ | |
3268 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3269 | |
51df1142 CL |
3270 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3271 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3272 | ||
3273 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3274 | ||
3275 | caches++; | |
51df1142 CL |
3276 | kmem_cache_bootstrap_fixup(kmem_cache); |
3277 | caches++; | |
3278 | /* Free temporary boot structure */ | |
3279 | free_pages((unsigned long)temp_kmem_cache, order); | |
3280 | ||
3281 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3282 | |
3283 | /* | |
3284 | * Patch up the size_index table if we have strange large alignment | |
3285 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3286 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3287 | * |
3288 | * Largest permitted alignment is 256 bytes due to the way we | |
3289 | * handle the index determination for the smaller caches. | |
3290 | * | |
3291 | * Make sure that nothing crazy happens if someone starts tinkering | |
3292 | * around with ARCH_KMALLOC_MINALIGN | |
3293 | */ | |
3294 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3295 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3296 | ||
acdfcd04 AK |
3297 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3298 | int elem = size_index_elem(i); | |
3299 | if (elem >= ARRAY_SIZE(size_index)) | |
3300 | break; | |
3301 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3302 | } | |
f1b26339 | 3303 | |
acdfcd04 AK |
3304 | if (KMALLOC_MIN_SIZE == 64) { |
3305 | /* | |
3306 | * The 96 byte size cache is not used if the alignment | |
3307 | * is 64 byte. | |
3308 | */ | |
3309 | for (i = 64 + 8; i <= 96; i += 8) | |
3310 | size_index[size_index_elem(i)] = 7; | |
3311 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3312 | /* |
3313 | * The 192 byte sized cache is not used if the alignment | |
3314 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3315 | * instead. | |
3316 | */ | |
3317 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3318 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3319 | } |
3320 | ||
51df1142 CL |
3321 | /* Caches that are not of the two-to-the-power-of size */ |
3322 | if (KMALLOC_MIN_SIZE <= 32) { | |
3323 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3324 | caches++; | |
3325 | } | |
3326 | ||
3327 | if (KMALLOC_MIN_SIZE <= 64) { | |
3328 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3329 | caches++; | |
3330 | } | |
3331 | ||
3332 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3333 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3334 | caches++; | |
3335 | } | |
3336 | ||
81819f0f CL |
3337 | slab_state = UP; |
3338 | ||
3339 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3340 | if (KMALLOC_MIN_SIZE <= 32) { |
3341 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3342 | BUG_ON(!kmalloc_caches[1]->name); | |
3343 | } | |
3344 | ||
3345 | if (KMALLOC_MIN_SIZE <= 64) { | |
3346 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3347 | BUG_ON(!kmalloc_caches[2]->name); | |
3348 | } | |
3349 | ||
d7278bd7 CL |
3350 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3351 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3352 | ||
3353 | BUG_ON(!s); | |
51df1142 | 3354 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3355 | } |
81819f0f CL |
3356 | |
3357 | #ifdef CONFIG_SMP | |
3358 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3359 | #endif |
81819f0f | 3360 | |
55136592 | 3361 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3362 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3363 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3364 | |
51df1142 | 3365 | if (s && s->size) { |
55136592 CL |
3366 | char *name = kasprintf(GFP_NOWAIT, |
3367 | "dma-kmalloc-%d", s->objsize); | |
3368 | ||
3369 | BUG_ON(!name); | |
51df1142 CL |
3370 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3371 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3372 | } |
3373 | } | |
3374 | #endif | |
3adbefee IM |
3375 | printk(KERN_INFO |
3376 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3377 | " CPUs=%d, Nodes=%d\n", |
3378 | caches, cache_line_size(), | |
81819f0f CL |
3379 | slub_min_order, slub_max_order, slub_min_objects, |
3380 | nr_cpu_ids, nr_node_ids); | |
3381 | } | |
3382 | ||
7e85ee0c PE |
3383 | void __init kmem_cache_init_late(void) |
3384 | { | |
7e85ee0c PE |
3385 | } |
3386 | ||
81819f0f CL |
3387 | /* |
3388 | * Find a mergeable slab cache | |
3389 | */ | |
3390 | static int slab_unmergeable(struct kmem_cache *s) | |
3391 | { | |
3392 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3393 | return 1; | |
3394 | ||
c59def9f | 3395 | if (s->ctor) |
81819f0f CL |
3396 | return 1; |
3397 | ||
8ffa6875 CL |
3398 | /* |
3399 | * We may have set a slab to be unmergeable during bootstrap. | |
3400 | */ | |
3401 | if (s->refcount < 0) | |
3402 | return 1; | |
3403 | ||
81819f0f CL |
3404 | return 0; |
3405 | } | |
3406 | ||
3407 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3408 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3409 | void (*ctor)(void *)) |
81819f0f | 3410 | { |
5b95a4ac | 3411 | struct kmem_cache *s; |
81819f0f CL |
3412 | |
3413 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3414 | return NULL; | |
3415 | ||
c59def9f | 3416 | if (ctor) |
81819f0f CL |
3417 | return NULL; |
3418 | ||
3419 | size = ALIGN(size, sizeof(void *)); | |
3420 | align = calculate_alignment(flags, align, size); | |
3421 | size = ALIGN(size, align); | |
ba0268a8 | 3422 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3423 | |
5b95a4ac | 3424 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3425 | if (slab_unmergeable(s)) |
3426 | continue; | |
3427 | ||
3428 | if (size > s->size) | |
3429 | continue; | |
3430 | ||
ba0268a8 | 3431 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3432 | continue; |
3433 | /* | |
3434 | * Check if alignment is compatible. | |
3435 | * Courtesy of Adrian Drzewiecki | |
3436 | */ | |
06428780 | 3437 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3438 | continue; |
3439 | ||
3440 | if (s->size - size >= sizeof(void *)) | |
3441 | continue; | |
3442 | ||
3443 | return s; | |
3444 | } | |
3445 | return NULL; | |
3446 | } | |
3447 | ||
3448 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3449 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3450 | { |
3451 | struct kmem_cache *s; | |
84c1cf62 | 3452 | char *n; |
81819f0f | 3453 | |
fe1ff49d BH |
3454 | if (WARN_ON(!name)) |
3455 | return NULL; | |
3456 | ||
81819f0f | 3457 | down_write(&slub_lock); |
ba0268a8 | 3458 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3459 | if (s) { |
3460 | s->refcount++; | |
3461 | /* | |
3462 | * Adjust the object sizes so that we clear | |
3463 | * the complete object on kzalloc. | |
3464 | */ | |
3465 | s->objsize = max(s->objsize, (int)size); | |
3466 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3467 | |
7b8f3b66 | 3468 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3469 | s->refcount--; |
81819f0f | 3470 | goto err; |
7b8f3b66 | 3471 | } |
2bce6485 | 3472 | up_write(&slub_lock); |
a0e1d1be CL |
3473 | return s; |
3474 | } | |
6446faa2 | 3475 | |
84c1cf62 PE |
3476 | n = kstrdup(name, GFP_KERNEL); |
3477 | if (!n) | |
3478 | goto err; | |
3479 | ||
a0e1d1be CL |
3480 | s = kmalloc(kmem_size, GFP_KERNEL); |
3481 | if (s) { | |
84c1cf62 | 3482 | if (kmem_cache_open(s, n, |
c59def9f | 3483 | size, align, flags, ctor)) { |
81819f0f | 3484 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3485 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3486 | list_del(&s->list); |
84c1cf62 | 3487 | kfree(n); |
7b8f3b66 | 3488 | kfree(s); |
a0e1d1be | 3489 | goto err; |
7b8f3b66 | 3490 | } |
2bce6485 | 3491 | up_write(&slub_lock); |
a0e1d1be CL |
3492 | return s; |
3493 | } | |
84c1cf62 | 3494 | kfree(n); |
a0e1d1be | 3495 | kfree(s); |
81819f0f | 3496 | } |
68cee4f1 | 3497 | err: |
81819f0f | 3498 | up_write(&slub_lock); |
81819f0f | 3499 | |
81819f0f CL |
3500 | if (flags & SLAB_PANIC) |
3501 | panic("Cannot create slabcache %s\n", name); | |
3502 | else | |
3503 | s = NULL; | |
3504 | return s; | |
3505 | } | |
3506 | EXPORT_SYMBOL(kmem_cache_create); | |
3507 | ||
81819f0f | 3508 | #ifdef CONFIG_SMP |
81819f0f | 3509 | /* |
672bba3a CL |
3510 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3511 | * necessary. | |
81819f0f CL |
3512 | */ |
3513 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3514 | unsigned long action, void *hcpu) | |
3515 | { | |
3516 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3517 | struct kmem_cache *s; |
3518 | unsigned long flags; | |
81819f0f CL |
3519 | |
3520 | switch (action) { | |
3521 | case CPU_UP_CANCELED: | |
8bb78442 | 3522 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3523 | case CPU_DEAD: |
8bb78442 | 3524 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3525 | down_read(&slub_lock); |
3526 | list_for_each_entry(s, &slab_caches, list) { | |
3527 | local_irq_save(flags); | |
3528 | __flush_cpu_slab(s, cpu); | |
3529 | local_irq_restore(flags); | |
3530 | } | |
3531 | up_read(&slub_lock); | |
81819f0f CL |
3532 | break; |
3533 | default: | |
3534 | break; | |
3535 | } | |
3536 | return NOTIFY_OK; | |
3537 | } | |
3538 | ||
06428780 | 3539 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3540 | .notifier_call = slab_cpuup_callback |
06428780 | 3541 | }; |
81819f0f CL |
3542 | |
3543 | #endif | |
3544 | ||
ce71e27c | 3545 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3546 | { |
aadb4bc4 | 3547 | struct kmem_cache *s; |
94b528d0 | 3548 | void *ret; |
aadb4bc4 | 3549 | |
ffadd4d0 | 3550 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3551 | return kmalloc_large(size, gfpflags); |
3552 | ||
aadb4bc4 | 3553 | s = get_slab(size, gfpflags); |
81819f0f | 3554 | |
2408c550 | 3555 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3556 | return s; |
81819f0f | 3557 | |
2154a336 | 3558 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 EGM |
3559 | |
3560 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3561 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3562 | |
3563 | return ret; | |
81819f0f CL |
3564 | } |
3565 | ||
5d1f57e4 | 3566 | #ifdef CONFIG_NUMA |
81819f0f | 3567 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3568 | int node, unsigned long caller) |
81819f0f | 3569 | { |
aadb4bc4 | 3570 | struct kmem_cache *s; |
94b528d0 | 3571 | void *ret; |
aadb4bc4 | 3572 | |
d3e14aa3 XF |
3573 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3574 | ret = kmalloc_large_node(size, gfpflags, node); | |
3575 | ||
3576 | trace_kmalloc_node(caller, ret, | |
3577 | size, PAGE_SIZE << get_order(size), | |
3578 | gfpflags, node); | |
3579 | ||
3580 | return ret; | |
3581 | } | |
eada35ef | 3582 | |
aadb4bc4 | 3583 | s = get_slab(size, gfpflags); |
81819f0f | 3584 | |
2408c550 | 3585 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3586 | return s; |
81819f0f | 3587 | |
94b528d0 EGM |
3588 | ret = slab_alloc(s, gfpflags, node, caller); |
3589 | ||
3590 | /* Honor the call site pointer we recieved. */ | |
ca2b84cb | 3591 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3592 | |
3593 | return ret; | |
81819f0f | 3594 | } |
5d1f57e4 | 3595 | #endif |
81819f0f | 3596 | |
ab4d5ed5 | 3597 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3598 | static int count_inuse(struct page *page) |
3599 | { | |
3600 | return page->inuse; | |
3601 | } | |
3602 | ||
3603 | static int count_total(struct page *page) | |
3604 | { | |
3605 | return page->objects; | |
3606 | } | |
ab4d5ed5 | 3607 | #endif |
205ab99d | 3608 | |
ab4d5ed5 | 3609 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3610 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3611 | unsigned long *map) | |
53e15af0 CL |
3612 | { |
3613 | void *p; | |
a973e9dd | 3614 | void *addr = page_address(page); |
53e15af0 CL |
3615 | |
3616 | if (!check_slab(s, page) || | |
3617 | !on_freelist(s, page, NULL)) | |
3618 | return 0; | |
3619 | ||
3620 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3621 | bitmap_zero(map, page->objects); |
53e15af0 | 3622 | |
5f80b13a CL |
3623 | get_map(s, page, map); |
3624 | for_each_object(p, s, addr, page->objects) { | |
3625 | if (test_bit(slab_index(p, s, addr), map)) | |
3626 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3627 | return 0; | |
53e15af0 CL |
3628 | } |
3629 | ||
224a88be | 3630 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3631 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3632 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3633 | return 0; |
3634 | return 1; | |
3635 | } | |
3636 | ||
434e245d CL |
3637 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3638 | unsigned long *map) | |
53e15af0 CL |
3639 | { |
3640 | if (slab_trylock(page)) { | |
434e245d | 3641 | validate_slab(s, page, map); |
53e15af0 CL |
3642 | slab_unlock(page); |
3643 | } else | |
3644 | printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n", | |
3645 | s->name, page); | |
53e15af0 CL |
3646 | } |
3647 | ||
434e245d CL |
3648 | static int validate_slab_node(struct kmem_cache *s, |
3649 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3650 | { |
3651 | unsigned long count = 0; | |
3652 | struct page *page; | |
3653 | unsigned long flags; | |
3654 | ||
3655 | spin_lock_irqsave(&n->list_lock, flags); | |
3656 | ||
3657 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3658 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3659 | count++; |
3660 | } | |
3661 | if (count != n->nr_partial) | |
3662 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3663 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3664 | ||
3665 | if (!(s->flags & SLAB_STORE_USER)) | |
3666 | goto out; | |
3667 | ||
3668 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3669 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3670 | count++; |
3671 | } | |
3672 | if (count != atomic_long_read(&n->nr_slabs)) | |
3673 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3674 | "counter=%ld\n", s->name, count, | |
3675 | atomic_long_read(&n->nr_slabs)); | |
3676 | ||
3677 | out: | |
3678 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3679 | return count; | |
3680 | } | |
3681 | ||
434e245d | 3682 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3683 | { |
3684 | int node; | |
3685 | unsigned long count = 0; | |
205ab99d | 3686 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3687 | sizeof(unsigned long), GFP_KERNEL); |
3688 | ||
3689 | if (!map) | |
3690 | return -ENOMEM; | |
53e15af0 CL |
3691 | |
3692 | flush_all(s); | |
f64dc58c | 3693 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3694 | struct kmem_cache_node *n = get_node(s, node); |
3695 | ||
434e245d | 3696 | count += validate_slab_node(s, n, map); |
53e15af0 | 3697 | } |
434e245d | 3698 | kfree(map); |
53e15af0 CL |
3699 | return count; |
3700 | } | |
88a420e4 | 3701 | /* |
672bba3a | 3702 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3703 | * and freed. |
3704 | */ | |
3705 | ||
3706 | struct location { | |
3707 | unsigned long count; | |
ce71e27c | 3708 | unsigned long addr; |
45edfa58 CL |
3709 | long long sum_time; |
3710 | long min_time; | |
3711 | long max_time; | |
3712 | long min_pid; | |
3713 | long max_pid; | |
174596a0 | 3714 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3715 | nodemask_t nodes; |
88a420e4 CL |
3716 | }; |
3717 | ||
3718 | struct loc_track { | |
3719 | unsigned long max; | |
3720 | unsigned long count; | |
3721 | struct location *loc; | |
3722 | }; | |
3723 | ||
3724 | static void free_loc_track(struct loc_track *t) | |
3725 | { | |
3726 | if (t->max) | |
3727 | free_pages((unsigned long)t->loc, | |
3728 | get_order(sizeof(struct location) * t->max)); | |
3729 | } | |
3730 | ||
68dff6a9 | 3731 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3732 | { |
3733 | struct location *l; | |
3734 | int order; | |
3735 | ||
88a420e4 CL |
3736 | order = get_order(sizeof(struct location) * max); |
3737 | ||
68dff6a9 | 3738 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3739 | if (!l) |
3740 | return 0; | |
3741 | ||
3742 | if (t->count) { | |
3743 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3744 | free_loc_track(t); | |
3745 | } | |
3746 | t->max = max; | |
3747 | t->loc = l; | |
3748 | return 1; | |
3749 | } | |
3750 | ||
3751 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3752 | const struct track *track) |
88a420e4 CL |
3753 | { |
3754 | long start, end, pos; | |
3755 | struct location *l; | |
ce71e27c | 3756 | unsigned long caddr; |
45edfa58 | 3757 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3758 | |
3759 | start = -1; | |
3760 | end = t->count; | |
3761 | ||
3762 | for ( ; ; ) { | |
3763 | pos = start + (end - start + 1) / 2; | |
3764 | ||
3765 | /* | |
3766 | * There is nothing at "end". If we end up there | |
3767 | * we need to add something to before end. | |
3768 | */ | |
3769 | if (pos == end) | |
3770 | break; | |
3771 | ||
3772 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3773 | if (track->addr == caddr) { |
3774 | ||
3775 | l = &t->loc[pos]; | |
3776 | l->count++; | |
3777 | if (track->when) { | |
3778 | l->sum_time += age; | |
3779 | if (age < l->min_time) | |
3780 | l->min_time = age; | |
3781 | if (age > l->max_time) | |
3782 | l->max_time = age; | |
3783 | ||
3784 | if (track->pid < l->min_pid) | |
3785 | l->min_pid = track->pid; | |
3786 | if (track->pid > l->max_pid) | |
3787 | l->max_pid = track->pid; | |
3788 | ||
174596a0 RR |
3789 | cpumask_set_cpu(track->cpu, |
3790 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3791 | } |
3792 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3793 | return 1; |
3794 | } | |
3795 | ||
45edfa58 | 3796 | if (track->addr < caddr) |
88a420e4 CL |
3797 | end = pos; |
3798 | else | |
3799 | start = pos; | |
3800 | } | |
3801 | ||
3802 | /* | |
672bba3a | 3803 | * Not found. Insert new tracking element. |
88a420e4 | 3804 | */ |
68dff6a9 | 3805 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3806 | return 0; |
3807 | ||
3808 | l = t->loc + pos; | |
3809 | if (pos < t->count) | |
3810 | memmove(l + 1, l, | |
3811 | (t->count - pos) * sizeof(struct location)); | |
3812 | t->count++; | |
3813 | l->count = 1; | |
45edfa58 CL |
3814 | l->addr = track->addr; |
3815 | l->sum_time = age; | |
3816 | l->min_time = age; | |
3817 | l->max_time = age; | |
3818 | l->min_pid = track->pid; | |
3819 | l->max_pid = track->pid; | |
174596a0 RR |
3820 | cpumask_clear(to_cpumask(l->cpus)); |
3821 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3822 | nodes_clear(l->nodes); |
3823 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3824 | return 1; |
3825 | } | |
3826 | ||
3827 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3828 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3829 | unsigned long *map) |
88a420e4 | 3830 | { |
a973e9dd | 3831 | void *addr = page_address(page); |
88a420e4 CL |
3832 | void *p; |
3833 | ||
39b26464 | 3834 | bitmap_zero(map, page->objects); |
5f80b13a | 3835 | get_map(s, page, map); |
88a420e4 | 3836 | |
224a88be | 3837 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3838 | if (!test_bit(slab_index(p, s, addr), map)) |
3839 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3840 | } |
3841 | ||
3842 | static int list_locations(struct kmem_cache *s, char *buf, | |
3843 | enum track_item alloc) | |
3844 | { | |
e374d483 | 3845 | int len = 0; |
88a420e4 | 3846 | unsigned long i; |
68dff6a9 | 3847 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3848 | int node; |
bbd7d57b ED |
3849 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3850 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 3851 | |
bbd7d57b ED |
3852 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
3853 | GFP_TEMPORARY)) { | |
3854 | kfree(map); | |
68dff6a9 | 3855 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 3856 | } |
88a420e4 CL |
3857 | /* Push back cpu slabs */ |
3858 | flush_all(s); | |
3859 | ||
f64dc58c | 3860 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
3861 | struct kmem_cache_node *n = get_node(s, node); |
3862 | unsigned long flags; | |
3863 | struct page *page; | |
3864 | ||
9e86943b | 3865 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
3866 | continue; |
3867 | ||
3868 | spin_lock_irqsave(&n->list_lock, flags); | |
3869 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 3870 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 3871 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 3872 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
3873 | spin_unlock_irqrestore(&n->list_lock, flags); |
3874 | } | |
3875 | ||
3876 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 3877 | struct location *l = &t.loc[i]; |
88a420e4 | 3878 | |
9c246247 | 3879 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 3880 | break; |
e374d483 | 3881 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
3882 | |
3883 | if (l->addr) | |
62c70bce | 3884 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 3885 | else |
e374d483 | 3886 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
3887 | |
3888 | if (l->sum_time != l->min_time) { | |
e374d483 | 3889 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
3890 | l->min_time, |
3891 | (long)div_u64(l->sum_time, l->count), | |
3892 | l->max_time); | |
45edfa58 | 3893 | } else |
e374d483 | 3894 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
3895 | l->min_time); |
3896 | ||
3897 | if (l->min_pid != l->max_pid) | |
e374d483 | 3898 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
3899 | l->min_pid, l->max_pid); |
3900 | else | |
e374d483 | 3901 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
3902 | l->min_pid); |
3903 | ||
174596a0 RR |
3904 | if (num_online_cpus() > 1 && |
3905 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
3906 | len < PAGE_SIZE - 60) { |
3907 | len += sprintf(buf + len, " cpus="); | |
3908 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 3909 | to_cpumask(l->cpus)); |
45edfa58 CL |
3910 | } |
3911 | ||
62bc62a8 | 3912 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
3913 | len < PAGE_SIZE - 60) { |
3914 | len += sprintf(buf + len, " nodes="); | |
3915 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
3916 | l->nodes); |
3917 | } | |
3918 | ||
e374d483 | 3919 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
3920 | } |
3921 | ||
3922 | free_loc_track(&t); | |
bbd7d57b | 3923 | kfree(map); |
88a420e4 | 3924 | if (!t.count) |
e374d483 HH |
3925 | len += sprintf(buf, "No data\n"); |
3926 | return len; | |
88a420e4 | 3927 | } |
ab4d5ed5 | 3928 | #endif |
88a420e4 | 3929 | |
a5a84755 CL |
3930 | #ifdef SLUB_RESILIENCY_TEST |
3931 | static void resiliency_test(void) | |
3932 | { | |
3933 | u8 *p; | |
3934 | ||
3935 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
3936 | ||
3937 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
3938 | printk(KERN_ERR "-----------------------\n"); | |
3939 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
3940 | ||
3941 | p = kzalloc(16, GFP_KERNEL); | |
3942 | p[16] = 0x12; | |
3943 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
3944 | " 0x12->0x%p\n\n", p + 16); | |
3945 | ||
3946 | validate_slab_cache(kmalloc_caches[4]); | |
3947 | ||
3948 | /* Hmmm... The next two are dangerous */ | |
3949 | p = kzalloc(32, GFP_KERNEL); | |
3950 | p[32 + sizeof(void *)] = 0x34; | |
3951 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
3952 | " 0x34 -> -0x%p\n", p); | |
3953 | printk(KERN_ERR | |
3954 | "If allocated object is overwritten then not detectable\n\n"); | |
3955 | ||
3956 | validate_slab_cache(kmalloc_caches[5]); | |
3957 | p = kzalloc(64, GFP_KERNEL); | |
3958 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
3959 | *p = 0x56; | |
3960 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
3961 | p); | |
3962 | printk(KERN_ERR | |
3963 | "If allocated object is overwritten then not detectable\n\n"); | |
3964 | validate_slab_cache(kmalloc_caches[6]); | |
3965 | ||
3966 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
3967 | p = kzalloc(128, GFP_KERNEL); | |
3968 | kfree(p); | |
3969 | *p = 0x78; | |
3970 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
3971 | validate_slab_cache(kmalloc_caches[7]); | |
3972 | ||
3973 | p = kzalloc(256, GFP_KERNEL); | |
3974 | kfree(p); | |
3975 | p[50] = 0x9a; | |
3976 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
3977 | p); | |
3978 | validate_slab_cache(kmalloc_caches[8]); | |
3979 | ||
3980 | p = kzalloc(512, GFP_KERNEL); | |
3981 | kfree(p); | |
3982 | p[512] = 0xab; | |
3983 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
3984 | validate_slab_cache(kmalloc_caches[9]); | |
3985 | } | |
3986 | #else | |
3987 | #ifdef CONFIG_SYSFS | |
3988 | static void resiliency_test(void) {}; | |
3989 | #endif | |
3990 | #endif | |
3991 | ||
ab4d5ed5 | 3992 | #ifdef CONFIG_SYSFS |
81819f0f | 3993 | enum slab_stat_type { |
205ab99d CL |
3994 | SL_ALL, /* All slabs */ |
3995 | SL_PARTIAL, /* Only partially allocated slabs */ | |
3996 | SL_CPU, /* Only slabs used for cpu caches */ | |
3997 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
3998 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
3999 | }; |
4000 | ||
205ab99d | 4001 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4002 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4003 | #define SO_CPU (1 << SL_CPU) | |
4004 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4005 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4006 | |
62e5c4b4 CG |
4007 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4008 | char *buf, unsigned long flags) | |
81819f0f CL |
4009 | { |
4010 | unsigned long total = 0; | |
81819f0f CL |
4011 | int node; |
4012 | int x; | |
4013 | unsigned long *nodes; | |
4014 | unsigned long *per_cpu; | |
4015 | ||
4016 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4017 | if (!nodes) |
4018 | return -ENOMEM; | |
81819f0f CL |
4019 | per_cpu = nodes + nr_node_ids; |
4020 | ||
205ab99d CL |
4021 | if (flags & SO_CPU) { |
4022 | int cpu; | |
81819f0f | 4023 | |
205ab99d | 4024 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4025 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 4026 | |
205ab99d CL |
4027 | if (!c || c->node < 0) |
4028 | continue; | |
4029 | ||
4030 | if (c->page) { | |
4031 | if (flags & SO_TOTAL) | |
4032 | x = c->page->objects; | |
4033 | else if (flags & SO_OBJECTS) | |
4034 | x = c->page->inuse; | |
81819f0f CL |
4035 | else |
4036 | x = 1; | |
205ab99d | 4037 | |
81819f0f | 4038 | total += x; |
205ab99d | 4039 | nodes[c->node] += x; |
81819f0f | 4040 | } |
205ab99d | 4041 | per_cpu[c->node]++; |
81819f0f CL |
4042 | } |
4043 | } | |
4044 | ||
04d94879 | 4045 | lock_memory_hotplug(); |
ab4d5ed5 | 4046 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4047 | if (flags & SO_ALL) { |
4048 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4049 | struct kmem_cache_node *n = get_node(s, node); | |
4050 | ||
4051 | if (flags & SO_TOTAL) | |
4052 | x = atomic_long_read(&n->total_objects); | |
4053 | else if (flags & SO_OBJECTS) | |
4054 | x = atomic_long_read(&n->total_objects) - | |
4055 | count_partial(n, count_free); | |
81819f0f | 4056 | |
81819f0f | 4057 | else |
205ab99d | 4058 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4059 | total += x; |
4060 | nodes[node] += x; | |
4061 | } | |
4062 | ||
ab4d5ed5 CL |
4063 | } else |
4064 | #endif | |
4065 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4066 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4067 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4068 | |
205ab99d CL |
4069 | if (flags & SO_TOTAL) |
4070 | x = count_partial(n, count_total); | |
4071 | else if (flags & SO_OBJECTS) | |
4072 | x = count_partial(n, count_inuse); | |
81819f0f | 4073 | else |
205ab99d | 4074 | x = n->nr_partial; |
81819f0f CL |
4075 | total += x; |
4076 | nodes[node] += x; | |
4077 | } | |
4078 | } | |
81819f0f CL |
4079 | x = sprintf(buf, "%lu", total); |
4080 | #ifdef CONFIG_NUMA | |
f64dc58c | 4081 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4082 | if (nodes[node]) |
4083 | x += sprintf(buf + x, " N%d=%lu", | |
4084 | node, nodes[node]); | |
4085 | #endif | |
04d94879 | 4086 | unlock_memory_hotplug(); |
81819f0f CL |
4087 | kfree(nodes); |
4088 | return x + sprintf(buf + x, "\n"); | |
4089 | } | |
4090 | ||
ab4d5ed5 | 4091 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4092 | static int any_slab_objects(struct kmem_cache *s) |
4093 | { | |
4094 | int node; | |
81819f0f | 4095 | |
dfb4f096 | 4096 | for_each_online_node(node) { |
81819f0f CL |
4097 | struct kmem_cache_node *n = get_node(s, node); |
4098 | ||
dfb4f096 CL |
4099 | if (!n) |
4100 | continue; | |
4101 | ||
4ea33e2d | 4102 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4103 | return 1; |
4104 | } | |
4105 | return 0; | |
4106 | } | |
ab4d5ed5 | 4107 | #endif |
81819f0f CL |
4108 | |
4109 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
4110 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
4111 | ||
4112 | struct slab_attribute { | |
4113 | struct attribute attr; | |
4114 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4115 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4116 | }; | |
4117 | ||
4118 | #define SLAB_ATTR_RO(_name) \ | |
4119 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4120 | ||
4121 | #define SLAB_ATTR(_name) \ | |
4122 | static struct slab_attribute _name##_attr = \ | |
4123 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4124 | ||
81819f0f CL |
4125 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4126 | { | |
4127 | return sprintf(buf, "%d\n", s->size); | |
4128 | } | |
4129 | SLAB_ATTR_RO(slab_size); | |
4130 | ||
4131 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4132 | { | |
4133 | return sprintf(buf, "%d\n", s->align); | |
4134 | } | |
4135 | SLAB_ATTR_RO(align); | |
4136 | ||
4137 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4138 | { | |
4139 | return sprintf(buf, "%d\n", s->objsize); | |
4140 | } | |
4141 | SLAB_ATTR_RO(object_size); | |
4142 | ||
4143 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4144 | { | |
834f3d11 | 4145 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4146 | } |
4147 | SLAB_ATTR_RO(objs_per_slab); | |
4148 | ||
06b285dc CL |
4149 | static ssize_t order_store(struct kmem_cache *s, |
4150 | const char *buf, size_t length) | |
4151 | { | |
0121c619 CL |
4152 | unsigned long order; |
4153 | int err; | |
4154 | ||
4155 | err = strict_strtoul(buf, 10, &order); | |
4156 | if (err) | |
4157 | return err; | |
06b285dc CL |
4158 | |
4159 | if (order > slub_max_order || order < slub_min_order) | |
4160 | return -EINVAL; | |
4161 | ||
4162 | calculate_sizes(s, order); | |
4163 | return length; | |
4164 | } | |
4165 | ||
81819f0f CL |
4166 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4167 | { | |
834f3d11 | 4168 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4169 | } |
06b285dc | 4170 | SLAB_ATTR(order); |
81819f0f | 4171 | |
73d342b1 DR |
4172 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4173 | { | |
4174 | return sprintf(buf, "%lu\n", s->min_partial); | |
4175 | } | |
4176 | ||
4177 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4178 | size_t length) | |
4179 | { | |
4180 | unsigned long min; | |
4181 | int err; | |
4182 | ||
4183 | err = strict_strtoul(buf, 10, &min); | |
4184 | if (err) | |
4185 | return err; | |
4186 | ||
c0bdb232 | 4187 | set_min_partial(s, min); |
73d342b1 DR |
4188 | return length; |
4189 | } | |
4190 | SLAB_ATTR(min_partial); | |
4191 | ||
81819f0f CL |
4192 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4193 | { | |
62c70bce JP |
4194 | if (!s->ctor) |
4195 | return 0; | |
4196 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4197 | } |
4198 | SLAB_ATTR_RO(ctor); | |
4199 | ||
81819f0f CL |
4200 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4201 | { | |
4202 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4203 | } | |
4204 | SLAB_ATTR_RO(aliases); | |
4205 | ||
81819f0f CL |
4206 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4207 | { | |
d9acf4b7 | 4208 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4209 | } |
4210 | SLAB_ATTR_RO(partial); | |
4211 | ||
4212 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4213 | { | |
d9acf4b7 | 4214 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4215 | } |
4216 | SLAB_ATTR_RO(cpu_slabs); | |
4217 | ||
4218 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4219 | { | |
205ab99d | 4220 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4221 | } |
4222 | SLAB_ATTR_RO(objects); | |
4223 | ||
205ab99d CL |
4224 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4225 | { | |
4226 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4227 | } | |
4228 | SLAB_ATTR_RO(objects_partial); | |
4229 | ||
a5a84755 CL |
4230 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4231 | { | |
4232 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4233 | } | |
4234 | ||
4235 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4236 | const char *buf, size_t length) | |
4237 | { | |
4238 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4239 | if (buf[0] == '1') | |
4240 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4241 | return length; | |
4242 | } | |
4243 | SLAB_ATTR(reclaim_account); | |
4244 | ||
4245 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4246 | { | |
4247 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4248 | } | |
4249 | SLAB_ATTR_RO(hwcache_align); | |
4250 | ||
4251 | #ifdef CONFIG_ZONE_DMA | |
4252 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4253 | { | |
4254 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4255 | } | |
4256 | SLAB_ATTR_RO(cache_dma); | |
4257 | #endif | |
4258 | ||
4259 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4260 | { | |
4261 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4262 | } | |
4263 | SLAB_ATTR_RO(destroy_by_rcu); | |
4264 | ||
ab9a0f19 LJ |
4265 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4266 | { | |
4267 | return sprintf(buf, "%d\n", s->reserved); | |
4268 | } | |
4269 | SLAB_ATTR_RO(reserved); | |
4270 | ||
ab4d5ed5 | 4271 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4272 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4273 | { | |
4274 | return show_slab_objects(s, buf, SO_ALL); | |
4275 | } | |
4276 | SLAB_ATTR_RO(slabs); | |
4277 | ||
205ab99d CL |
4278 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4279 | { | |
4280 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4281 | } | |
4282 | SLAB_ATTR_RO(total_objects); | |
4283 | ||
81819f0f CL |
4284 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4285 | { | |
4286 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4287 | } | |
4288 | ||
4289 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4290 | const char *buf, size_t length) | |
4291 | { | |
4292 | s->flags &= ~SLAB_DEBUG_FREE; | |
4293 | if (buf[0] == '1') | |
4294 | s->flags |= SLAB_DEBUG_FREE; | |
4295 | return length; | |
4296 | } | |
4297 | SLAB_ATTR(sanity_checks); | |
4298 | ||
4299 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4300 | { | |
4301 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4302 | } | |
4303 | ||
4304 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4305 | size_t length) | |
4306 | { | |
4307 | s->flags &= ~SLAB_TRACE; | |
4308 | if (buf[0] == '1') | |
4309 | s->flags |= SLAB_TRACE; | |
4310 | return length; | |
4311 | } | |
4312 | SLAB_ATTR(trace); | |
4313 | ||
81819f0f CL |
4314 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4315 | { | |
4316 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4317 | } | |
4318 | ||
4319 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4320 | const char *buf, size_t length) | |
4321 | { | |
4322 | if (any_slab_objects(s)) | |
4323 | return -EBUSY; | |
4324 | ||
4325 | s->flags &= ~SLAB_RED_ZONE; | |
4326 | if (buf[0] == '1') | |
4327 | s->flags |= SLAB_RED_ZONE; | |
06b285dc | 4328 | calculate_sizes(s, -1); |
81819f0f CL |
4329 | return length; |
4330 | } | |
4331 | SLAB_ATTR(red_zone); | |
4332 | ||
4333 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4334 | { | |
4335 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4336 | } | |
4337 | ||
4338 | static ssize_t poison_store(struct kmem_cache *s, | |
4339 | const char *buf, size_t length) | |
4340 | { | |
4341 | if (any_slab_objects(s)) | |
4342 | return -EBUSY; | |
4343 | ||
4344 | s->flags &= ~SLAB_POISON; | |
4345 | if (buf[0] == '1') | |
4346 | s->flags |= SLAB_POISON; | |
06b285dc | 4347 | calculate_sizes(s, -1); |
81819f0f CL |
4348 | return length; |
4349 | } | |
4350 | SLAB_ATTR(poison); | |
4351 | ||
4352 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4353 | { | |
4354 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4355 | } | |
4356 | ||
4357 | static ssize_t store_user_store(struct kmem_cache *s, | |
4358 | const char *buf, size_t length) | |
4359 | { | |
4360 | if (any_slab_objects(s)) | |
4361 | return -EBUSY; | |
4362 | ||
4363 | s->flags &= ~SLAB_STORE_USER; | |
4364 | if (buf[0] == '1') | |
4365 | s->flags |= SLAB_STORE_USER; | |
06b285dc | 4366 | calculate_sizes(s, -1); |
81819f0f CL |
4367 | return length; |
4368 | } | |
4369 | SLAB_ATTR(store_user); | |
4370 | ||
53e15af0 CL |
4371 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4372 | { | |
4373 | return 0; | |
4374 | } | |
4375 | ||
4376 | static ssize_t validate_store(struct kmem_cache *s, | |
4377 | const char *buf, size_t length) | |
4378 | { | |
434e245d CL |
4379 | int ret = -EINVAL; |
4380 | ||
4381 | if (buf[0] == '1') { | |
4382 | ret = validate_slab_cache(s); | |
4383 | if (ret >= 0) | |
4384 | ret = length; | |
4385 | } | |
4386 | return ret; | |
53e15af0 CL |
4387 | } |
4388 | SLAB_ATTR(validate); | |
a5a84755 CL |
4389 | |
4390 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4391 | { | |
4392 | if (!(s->flags & SLAB_STORE_USER)) | |
4393 | return -ENOSYS; | |
4394 | return list_locations(s, buf, TRACK_ALLOC); | |
4395 | } | |
4396 | SLAB_ATTR_RO(alloc_calls); | |
4397 | ||
4398 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4399 | { | |
4400 | if (!(s->flags & SLAB_STORE_USER)) | |
4401 | return -ENOSYS; | |
4402 | return list_locations(s, buf, TRACK_FREE); | |
4403 | } | |
4404 | SLAB_ATTR_RO(free_calls); | |
4405 | #endif /* CONFIG_SLUB_DEBUG */ | |
4406 | ||
4407 | #ifdef CONFIG_FAILSLAB | |
4408 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4409 | { | |
4410 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4411 | } | |
4412 | ||
4413 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4414 | size_t length) | |
4415 | { | |
4416 | s->flags &= ~SLAB_FAILSLAB; | |
4417 | if (buf[0] == '1') | |
4418 | s->flags |= SLAB_FAILSLAB; | |
4419 | return length; | |
4420 | } | |
4421 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4422 | #endif |
53e15af0 | 4423 | |
2086d26a CL |
4424 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4425 | { | |
4426 | return 0; | |
4427 | } | |
4428 | ||
4429 | static ssize_t shrink_store(struct kmem_cache *s, | |
4430 | const char *buf, size_t length) | |
4431 | { | |
4432 | if (buf[0] == '1') { | |
4433 | int rc = kmem_cache_shrink(s); | |
4434 | ||
4435 | if (rc) | |
4436 | return rc; | |
4437 | } else | |
4438 | return -EINVAL; | |
4439 | return length; | |
4440 | } | |
4441 | SLAB_ATTR(shrink); | |
4442 | ||
81819f0f | 4443 | #ifdef CONFIG_NUMA |
9824601e | 4444 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4445 | { |
9824601e | 4446 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4447 | } |
4448 | ||
9824601e | 4449 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4450 | const char *buf, size_t length) |
4451 | { | |
0121c619 CL |
4452 | unsigned long ratio; |
4453 | int err; | |
4454 | ||
4455 | err = strict_strtoul(buf, 10, &ratio); | |
4456 | if (err) | |
4457 | return err; | |
4458 | ||
e2cb96b7 | 4459 | if (ratio <= 100) |
0121c619 | 4460 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4461 | |
81819f0f CL |
4462 | return length; |
4463 | } | |
9824601e | 4464 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4465 | #endif |
4466 | ||
8ff12cfc | 4467 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4468 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4469 | { | |
4470 | unsigned long sum = 0; | |
4471 | int cpu; | |
4472 | int len; | |
4473 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4474 | ||
4475 | if (!data) | |
4476 | return -ENOMEM; | |
4477 | ||
4478 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4479 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4480 | |
4481 | data[cpu] = x; | |
4482 | sum += x; | |
4483 | } | |
4484 | ||
4485 | len = sprintf(buf, "%lu", sum); | |
4486 | ||
50ef37b9 | 4487 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4488 | for_each_online_cpu(cpu) { |
4489 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4490 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4491 | } |
50ef37b9 | 4492 | #endif |
8ff12cfc CL |
4493 | kfree(data); |
4494 | return len + sprintf(buf + len, "\n"); | |
4495 | } | |
4496 | ||
78eb00cc DR |
4497 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4498 | { | |
4499 | int cpu; | |
4500 | ||
4501 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4502 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4503 | } |
4504 | ||
8ff12cfc CL |
4505 | #define STAT_ATTR(si, text) \ |
4506 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4507 | { \ | |
4508 | return show_stat(s, buf, si); \ | |
4509 | } \ | |
78eb00cc DR |
4510 | static ssize_t text##_store(struct kmem_cache *s, \ |
4511 | const char *buf, size_t length) \ | |
4512 | { \ | |
4513 | if (buf[0] != '0') \ | |
4514 | return -EINVAL; \ | |
4515 | clear_stat(s, si); \ | |
4516 | return length; \ | |
4517 | } \ | |
4518 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4519 | |
4520 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4521 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4522 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4523 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4524 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4525 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4526 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4527 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4528 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4529 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
4530 | STAT_ATTR(FREE_SLAB, free_slab); | |
4531 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4532 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4533 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4534 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4535 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4536 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
65c3376a | 4537 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
8ff12cfc CL |
4538 | #endif |
4539 | ||
06428780 | 4540 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4541 | &slab_size_attr.attr, |
4542 | &object_size_attr.attr, | |
4543 | &objs_per_slab_attr.attr, | |
4544 | &order_attr.attr, | |
73d342b1 | 4545 | &min_partial_attr.attr, |
81819f0f | 4546 | &objects_attr.attr, |
205ab99d | 4547 | &objects_partial_attr.attr, |
81819f0f CL |
4548 | &partial_attr.attr, |
4549 | &cpu_slabs_attr.attr, | |
4550 | &ctor_attr.attr, | |
81819f0f CL |
4551 | &aliases_attr.attr, |
4552 | &align_attr.attr, | |
81819f0f CL |
4553 | &hwcache_align_attr.attr, |
4554 | &reclaim_account_attr.attr, | |
4555 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4556 | &shrink_attr.attr, |
ab9a0f19 | 4557 | &reserved_attr.attr, |
ab4d5ed5 | 4558 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4559 | &total_objects_attr.attr, |
4560 | &slabs_attr.attr, | |
4561 | &sanity_checks_attr.attr, | |
4562 | &trace_attr.attr, | |
81819f0f CL |
4563 | &red_zone_attr.attr, |
4564 | &poison_attr.attr, | |
4565 | &store_user_attr.attr, | |
53e15af0 | 4566 | &validate_attr.attr, |
88a420e4 CL |
4567 | &alloc_calls_attr.attr, |
4568 | &free_calls_attr.attr, | |
ab4d5ed5 | 4569 | #endif |
81819f0f CL |
4570 | #ifdef CONFIG_ZONE_DMA |
4571 | &cache_dma_attr.attr, | |
4572 | #endif | |
4573 | #ifdef CONFIG_NUMA | |
9824601e | 4574 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4575 | #endif |
4576 | #ifdef CONFIG_SLUB_STATS | |
4577 | &alloc_fastpath_attr.attr, | |
4578 | &alloc_slowpath_attr.attr, | |
4579 | &free_fastpath_attr.attr, | |
4580 | &free_slowpath_attr.attr, | |
4581 | &free_frozen_attr.attr, | |
4582 | &free_add_partial_attr.attr, | |
4583 | &free_remove_partial_attr.attr, | |
4584 | &alloc_from_partial_attr.attr, | |
4585 | &alloc_slab_attr.attr, | |
4586 | &alloc_refill_attr.attr, | |
4587 | &free_slab_attr.attr, | |
4588 | &cpuslab_flush_attr.attr, | |
4589 | &deactivate_full_attr.attr, | |
4590 | &deactivate_empty_attr.attr, | |
4591 | &deactivate_to_head_attr.attr, | |
4592 | &deactivate_to_tail_attr.attr, | |
4593 | &deactivate_remote_frees_attr.attr, | |
65c3376a | 4594 | &order_fallback_attr.attr, |
81819f0f | 4595 | #endif |
4c13dd3b DM |
4596 | #ifdef CONFIG_FAILSLAB |
4597 | &failslab_attr.attr, | |
4598 | #endif | |
4599 | ||
81819f0f CL |
4600 | NULL |
4601 | }; | |
4602 | ||
4603 | static struct attribute_group slab_attr_group = { | |
4604 | .attrs = slab_attrs, | |
4605 | }; | |
4606 | ||
4607 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4608 | struct attribute *attr, | |
4609 | char *buf) | |
4610 | { | |
4611 | struct slab_attribute *attribute; | |
4612 | struct kmem_cache *s; | |
4613 | int err; | |
4614 | ||
4615 | attribute = to_slab_attr(attr); | |
4616 | s = to_slab(kobj); | |
4617 | ||
4618 | if (!attribute->show) | |
4619 | return -EIO; | |
4620 | ||
4621 | err = attribute->show(s, buf); | |
4622 | ||
4623 | return err; | |
4624 | } | |
4625 | ||
4626 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4627 | struct attribute *attr, | |
4628 | const char *buf, size_t len) | |
4629 | { | |
4630 | struct slab_attribute *attribute; | |
4631 | struct kmem_cache *s; | |
4632 | int err; | |
4633 | ||
4634 | attribute = to_slab_attr(attr); | |
4635 | s = to_slab(kobj); | |
4636 | ||
4637 | if (!attribute->store) | |
4638 | return -EIO; | |
4639 | ||
4640 | err = attribute->store(s, buf, len); | |
4641 | ||
4642 | return err; | |
4643 | } | |
4644 | ||
151c602f CL |
4645 | static void kmem_cache_release(struct kobject *kobj) |
4646 | { | |
4647 | struct kmem_cache *s = to_slab(kobj); | |
4648 | ||
84c1cf62 | 4649 | kfree(s->name); |
151c602f CL |
4650 | kfree(s); |
4651 | } | |
4652 | ||
52cf25d0 | 4653 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4654 | .show = slab_attr_show, |
4655 | .store = slab_attr_store, | |
4656 | }; | |
4657 | ||
4658 | static struct kobj_type slab_ktype = { | |
4659 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4660 | .release = kmem_cache_release |
81819f0f CL |
4661 | }; |
4662 | ||
4663 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4664 | { | |
4665 | struct kobj_type *ktype = get_ktype(kobj); | |
4666 | ||
4667 | if (ktype == &slab_ktype) | |
4668 | return 1; | |
4669 | return 0; | |
4670 | } | |
4671 | ||
9cd43611 | 4672 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4673 | .filter = uevent_filter, |
4674 | }; | |
4675 | ||
27c3a314 | 4676 | static struct kset *slab_kset; |
81819f0f CL |
4677 | |
4678 | #define ID_STR_LENGTH 64 | |
4679 | ||
4680 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4681 | * |
4682 | * Format :[flags-]size | |
81819f0f CL |
4683 | */ |
4684 | static char *create_unique_id(struct kmem_cache *s) | |
4685 | { | |
4686 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4687 | char *p = name; | |
4688 | ||
4689 | BUG_ON(!name); | |
4690 | ||
4691 | *p++ = ':'; | |
4692 | /* | |
4693 | * First flags affecting slabcache operations. We will only | |
4694 | * get here for aliasable slabs so we do not need to support | |
4695 | * too many flags. The flags here must cover all flags that | |
4696 | * are matched during merging to guarantee that the id is | |
4697 | * unique. | |
4698 | */ | |
4699 | if (s->flags & SLAB_CACHE_DMA) | |
4700 | *p++ = 'd'; | |
4701 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4702 | *p++ = 'a'; | |
4703 | if (s->flags & SLAB_DEBUG_FREE) | |
4704 | *p++ = 'F'; | |
5a896d9e VN |
4705 | if (!(s->flags & SLAB_NOTRACK)) |
4706 | *p++ = 't'; | |
81819f0f CL |
4707 | if (p != name + 1) |
4708 | *p++ = '-'; | |
4709 | p += sprintf(p, "%07d", s->size); | |
4710 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4711 | return name; | |
4712 | } | |
4713 | ||
4714 | static int sysfs_slab_add(struct kmem_cache *s) | |
4715 | { | |
4716 | int err; | |
4717 | const char *name; | |
4718 | int unmergeable; | |
4719 | ||
4720 | if (slab_state < SYSFS) | |
4721 | /* Defer until later */ | |
4722 | return 0; | |
4723 | ||
4724 | unmergeable = slab_unmergeable(s); | |
4725 | if (unmergeable) { | |
4726 | /* | |
4727 | * Slabcache can never be merged so we can use the name proper. | |
4728 | * This is typically the case for debug situations. In that | |
4729 | * case we can catch duplicate names easily. | |
4730 | */ | |
27c3a314 | 4731 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4732 | name = s->name; |
4733 | } else { | |
4734 | /* | |
4735 | * Create a unique name for the slab as a target | |
4736 | * for the symlinks. | |
4737 | */ | |
4738 | name = create_unique_id(s); | |
4739 | } | |
4740 | ||
27c3a314 | 4741 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4742 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4743 | if (err) { | |
4744 | kobject_put(&s->kobj); | |
81819f0f | 4745 | return err; |
1eada11c | 4746 | } |
81819f0f CL |
4747 | |
4748 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4749 | if (err) { |
4750 | kobject_del(&s->kobj); | |
4751 | kobject_put(&s->kobj); | |
81819f0f | 4752 | return err; |
5788d8ad | 4753 | } |
81819f0f CL |
4754 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4755 | if (!unmergeable) { | |
4756 | /* Setup first alias */ | |
4757 | sysfs_slab_alias(s, s->name); | |
4758 | kfree(name); | |
4759 | } | |
4760 | return 0; | |
4761 | } | |
4762 | ||
4763 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4764 | { | |
2bce6485 CL |
4765 | if (slab_state < SYSFS) |
4766 | /* | |
4767 | * Sysfs has not been setup yet so no need to remove the | |
4768 | * cache from sysfs. | |
4769 | */ | |
4770 | return; | |
4771 | ||
81819f0f CL |
4772 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4773 | kobject_del(&s->kobj); | |
151c602f | 4774 | kobject_put(&s->kobj); |
81819f0f CL |
4775 | } |
4776 | ||
4777 | /* | |
4778 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4779 | * available lest we lose that information. |
81819f0f CL |
4780 | */ |
4781 | struct saved_alias { | |
4782 | struct kmem_cache *s; | |
4783 | const char *name; | |
4784 | struct saved_alias *next; | |
4785 | }; | |
4786 | ||
5af328a5 | 4787 | static struct saved_alias *alias_list; |
81819f0f CL |
4788 | |
4789 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4790 | { | |
4791 | struct saved_alias *al; | |
4792 | ||
4793 | if (slab_state == SYSFS) { | |
4794 | /* | |
4795 | * If we have a leftover link then remove it. | |
4796 | */ | |
27c3a314 GKH |
4797 | sysfs_remove_link(&slab_kset->kobj, name); |
4798 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4799 | } |
4800 | ||
4801 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4802 | if (!al) | |
4803 | return -ENOMEM; | |
4804 | ||
4805 | al->s = s; | |
4806 | al->name = name; | |
4807 | al->next = alias_list; | |
4808 | alias_list = al; | |
4809 | return 0; | |
4810 | } | |
4811 | ||
4812 | static int __init slab_sysfs_init(void) | |
4813 | { | |
5b95a4ac | 4814 | struct kmem_cache *s; |
81819f0f CL |
4815 | int err; |
4816 | ||
2bce6485 CL |
4817 | down_write(&slub_lock); |
4818 | ||
0ff21e46 | 4819 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4820 | if (!slab_kset) { |
2bce6485 | 4821 | up_write(&slub_lock); |
81819f0f CL |
4822 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4823 | return -ENOSYS; | |
4824 | } | |
4825 | ||
26a7bd03 CL |
4826 | slab_state = SYSFS; |
4827 | ||
5b95a4ac | 4828 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4829 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4830 | if (err) |
4831 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4832 | " to sysfs\n", s->name); | |
26a7bd03 | 4833 | } |
81819f0f CL |
4834 | |
4835 | while (alias_list) { | |
4836 | struct saved_alias *al = alias_list; | |
4837 | ||
4838 | alias_list = alias_list->next; | |
4839 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
4840 | if (err) |
4841 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
4842 | " %s to sysfs\n", s->name); | |
81819f0f CL |
4843 | kfree(al); |
4844 | } | |
4845 | ||
2bce6485 | 4846 | up_write(&slub_lock); |
81819f0f CL |
4847 | resiliency_test(); |
4848 | return 0; | |
4849 | } | |
4850 | ||
4851 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 4852 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
4853 | |
4854 | /* | |
4855 | * The /proc/slabinfo ABI | |
4856 | */ | |
158a9624 | 4857 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
4858 | static void print_slabinfo_header(struct seq_file *m) |
4859 | { | |
4860 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
4861 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
4862 | "<objperslab> <pagesperslab>"); | |
4863 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4864 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
4865 | seq_putc(m, '\n'); | |
4866 | } | |
4867 | ||
4868 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4869 | { | |
4870 | loff_t n = *pos; | |
4871 | ||
4872 | down_read(&slub_lock); | |
4873 | if (!n) | |
4874 | print_slabinfo_header(m); | |
4875 | ||
4876 | return seq_list_start(&slab_caches, *pos); | |
4877 | } | |
4878 | ||
4879 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4880 | { | |
4881 | return seq_list_next(p, &slab_caches, pos); | |
4882 | } | |
4883 | ||
4884 | static void s_stop(struct seq_file *m, void *p) | |
4885 | { | |
4886 | up_read(&slub_lock); | |
4887 | } | |
4888 | ||
4889 | static int s_show(struct seq_file *m, void *p) | |
4890 | { | |
4891 | unsigned long nr_partials = 0; | |
4892 | unsigned long nr_slabs = 0; | |
4893 | unsigned long nr_inuse = 0; | |
205ab99d CL |
4894 | unsigned long nr_objs = 0; |
4895 | unsigned long nr_free = 0; | |
57ed3eda PE |
4896 | struct kmem_cache *s; |
4897 | int node; | |
4898 | ||
4899 | s = list_entry(p, struct kmem_cache, list); | |
4900 | ||
4901 | for_each_online_node(node) { | |
4902 | struct kmem_cache_node *n = get_node(s, node); | |
4903 | ||
4904 | if (!n) | |
4905 | continue; | |
4906 | ||
4907 | nr_partials += n->nr_partial; | |
4908 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
4909 | nr_objs += atomic_long_read(&n->total_objects); |
4910 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
4911 | } |
4912 | ||
205ab99d | 4913 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
4914 | |
4915 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
4916 | nr_objs, s->size, oo_objects(s->oo), |
4917 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
4918 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
4919 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
4920 | 0UL); | |
4921 | seq_putc(m, '\n'); | |
4922 | return 0; | |
4923 | } | |
4924 | ||
7b3c3a50 | 4925 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
4926 | .start = s_start, |
4927 | .next = s_next, | |
4928 | .stop = s_stop, | |
4929 | .show = s_show, | |
4930 | }; | |
4931 | ||
7b3c3a50 AD |
4932 | static int slabinfo_open(struct inode *inode, struct file *file) |
4933 | { | |
4934 | return seq_open(file, &slabinfo_op); | |
4935 | } | |
4936 | ||
4937 | static const struct file_operations proc_slabinfo_operations = { | |
4938 | .open = slabinfo_open, | |
4939 | .read = seq_read, | |
4940 | .llseek = seq_lseek, | |
4941 | .release = seq_release, | |
4942 | }; | |
4943 | ||
4944 | static int __init slab_proc_init(void) | |
4945 | { | |
cf5d1131 | 4946 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
4947 | return 0; |
4948 | } | |
4949 | module_init(slab_proc_init); | |
158a9624 | 4950 | #endif /* CONFIG_SLABINFO */ |