]>
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 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
7b3c3a50 | 19 | #include <linux/proc_fs.h> |
81819f0f | 20 | #include <linux/seq_file.h> |
5a896d9e | 21 | #include <linux/kmemcheck.h> |
81819f0f CL |
22 | #include <linux/cpu.h> |
23 | #include <linux/cpuset.h> | |
24 | #include <linux/mempolicy.h> | |
25 | #include <linux/ctype.h> | |
3ac7fe5a | 26 | #include <linux/debugobjects.h> |
81819f0f | 27 | #include <linux/kallsyms.h> |
b9049e23 | 28 | #include <linux/memory.h> |
f8bd2258 | 29 | #include <linux/math64.h> |
773ff60e | 30 | #include <linux/fault-inject.h> |
81819f0f | 31 | |
4a92379b RK |
32 | #include <trace/events/kmem.h> |
33 | ||
81819f0f CL |
34 | /* |
35 | * Lock order: | |
881db7fb CL |
36 | * 1. slub_lock (Global Semaphore) |
37 | * 2. node->list_lock | |
38 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 39 | * |
881db7fb CL |
40 | * slub_lock |
41 | * | |
42 | * The role of the slub_lock is to protect the list of all the slabs | |
43 | * and to synchronize major metadata changes to slab cache structures. | |
44 | * | |
45 | * The slab_lock is only used for debugging and on arches that do not | |
46 | * have the ability to do a cmpxchg_double. It only protects the second | |
47 | * double word in the page struct. Meaning | |
48 | * A. page->freelist -> List of object free in a page | |
49 | * B. page->counters -> Counters of objects | |
50 | * C. page->frozen -> frozen state | |
51 | * | |
52 | * If a slab is frozen then it is exempt from list management. It is not | |
53 | * on any list. The processor that froze the slab is the one who can | |
54 | * perform list operations on the page. Other processors may put objects | |
55 | * onto the freelist but the processor that froze the slab is the only | |
56 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
57 | * |
58 | * The list_lock protects the partial and full list on each node and | |
59 | * the partial slab counter. If taken then no new slabs may be added or | |
60 | * removed from the lists nor make the number of partial slabs be modified. | |
61 | * (Note that the total number of slabs is an atomic value that may be | |
62 | * modified without taking the list lock). | |
63 | * | |
64 | * The list_lock is a centralized lock and thus we avoid taking it as | |
65 | * much as possible. As long as SLUB does not have to handle partial | |
66 | * slabs, operations can continue without any centralized lock. F.e. | |
67 | * allocating a long series of objects that fill up slabs does not require | |
68 | * the list lock. | |
81819f0f CL |
69 | * Interrupts are disabled during allocation and deallocation in order to |
70 | * make the slab allocator safe to use in the context of an irq. In addition | |
71 | * interrupts are disabled to ensure that the processor does not change | |
72 | * while handling per_cpu slabs, due to kernel preemption. | |
73 | * | |
74 | * SLUB assigns one slab for allocation to each processor. | |
75 | * Allocations only occur from these slabs called cpu slabs. | |
76 | * | |
672bba3a CL |
77 | * Slabs with free elements are kept on a partial list and during regular |
78 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 79 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
80 | * We track full slabs for debugging purposes though because otherwise we |
81 | * cannot scan all objects. | |
81819f0f CL |
82 | * |
83 | * Slabs are freed when they become empty. Teardown and setup is | |
84 | * minimal so we rely on the page allocators per cpu caches for | |
85 | * fast frees and allocs. | |
86 | * | |
87 | * Overloading of page flags that are otherwise used for LRU management. | |
88 | * | |
4b6f0750 CL |
89 | * PageActive The slab is frozen and exempt from list processing. |
90 | * This means that the slab is dedicated to a purpose | |
91 | * such as satisfying allocations for a specific | |
92 | * processor. Objects may be freed in the slab while | |
93 | * it is frozen but slab_free will then skip the usual | |
94 | * list operations. It is up to the processor holding | |
95 | * the slab to integrate the slab into the slab lists | |
96 | * when the slab is no longer needed. | |
97 | * | |
98 | * One use of this flag is to mark slabs that are | |
99 | * used for allocations. Then such a slab becomes a cpu | |
100 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 101 | * freelist that allows lockless access to |
894b8788 CL |
102 | * free objects in addition to the regular freelist |
103 | * that requires the slab lock. | |
81819f0f CL |
104 | * |
105 | * PageError Slab requires special handling due to debug | |
106 | * options set. This moves slab handling out of | |
894b8788 | 107 | * the fast path and disables lockless freelists. |
81819f0f CL |
108 | */ |
109 | ||
af537b0a CL |
110 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
111 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
112 | ||
113 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
114 | { | |
5577bd8a | 115 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 116 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 117 | #else |
af537b0a | 118 | return 0; |
5577bd8a | 119 | #endif |
af537b0a | 120 | } |
5577bd8a | 121 | |
81819f0f CL |
122 | /* |
123 | * Issues still to be resolved: | |
124 | * | |
81819f0f CL |
125 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
126 | * | |
81819f0f CL |
127 | * - Variable sizing of the per node arrays |
128 | */ | |
129 | ||
130 | /* Enable to test recovery from slab corruption on boot */ | |
131 | #undef SLUB_RESILIENCY_TEST | |
132 | ||
b789ef51 CL |
133 | /* Enable to log cmpxchg failures */ |
134 | #undef SLUB_DEBUG_CMPXCHG | |
135 | ||
2086d26a CL |
136 | /* |
137 | * Mininum number of partial slabs. These will be left on the partial | |
138 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
139 | */ | |
76be8950 | 140 | #define MIN_PARTIAL 5 |
e95eed57 | 141 | |
2086d26a CL |
142 | /* |
143 | * Maximum number of desirable partial slabs. | |
144 | * The existence of more partial slabs makes kmem_cache_shrink | |
145 | * sort the partial list by the number of objects in the. | |
146 | */ | |
147 | #define MAX_PARTIAL 10 | |
148 | ||
81819f0f CL |
149 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
150 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 151 | |
fa5ec8a1 | 152 | /* |
3de47213 DR |
153 | * Debugging flags that require metadata to be stored in the slab. These get |
154 | * disabled when slub_debug=O is used and a cache's min order increases with | |
155 | * metadata. | |
fa5ec8a1 | 156 | */ |
3de47213 | 157 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 158 | |
81819f0f CL |
159 | /* |
160 | * Set of flags that will prevent slab merging | |
161 | */ | |
162 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
163 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
164 | SLAB_FAILSLAB) | |
81819f0f CL |
165 | |
166 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 167 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 168 | |
210b5c06 CG |
169 | #define OO_SHIFT 16 |
170 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 171 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 172 | |
81819f0f | 173 | /* Internal SLUB flags */ |
f90ec390 | 174 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 175 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f CL |
176 | |
177 | static int kmem_size = sizeof(struct kmem_cache); | |
178 | ||
179 | #ifdef CONFIG_SMP | |
180 | static struct notifier_block slab_notifier; | |
181 | #endif | |
182 | ||
183 | static enum { | |
184 | DOWN, /* No slab functionality available */ | |
51df1142 | 185 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 186 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
187 | SYSFS /* Sysfs up */ |
188 | } slab_state = DOWN; | |
189 | ||
190 | /* A list of all slab caches on the system */ | |
191 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 192 | static LIST_HEAD(slab_caches); |
81819f0f | 193 | |
02cbc874 CL |
194 | /* |
195 | * Tracking user of a slab. | |
196 | */ | |
197 | struct track { | |
ce71e27c | 198 | unsigned long addr; /* Called from address */ |
02cbc874 CL |
199 | int cpu; /* Was running on cpu */ |
200 | int pid; /* Pid context */ | |
201 | unsigned long when; /* When did the operation occur */ | |
202 | }; | |
203 | ||
204 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
205 | ||
ab4d5ed5 | 206 | #ifdef CONFIG_SYSFS |
81819f0f CL |
207 | static int sysfs_slab_add(struct kmem_cache *); |
208 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
209 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 210 | |
81819f0f | 211 | #else |
0c710013 CL |
212 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
213 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
214 | { return 0; } | |
151c602f CL |
215 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
216 | { | |
84c1cf62 | 217 | kfree(s->name); |
151c602f CL |
218 | kfree(s); |
219 | } | |
8ff12cfc | 220 | |
81819f0f CL |
221 | #endif |
222 | ||
4fdccdfb | 223 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
224 | { |
225 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 226 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
227 | #endif |
228 | } | |
229 | ||
81819f0f CL |
230 | /******************************************************************** |
231 | * Core slab cache functions | |
232 | *******************************************************************/ | |
233 | ||
234 | int slab_is_available(void) | |
235 | { | |
236 | return slab_state >= UP; | |
237 | } | |
238 | ||
239 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
240 | { | |
81819f0f | 241 | return s->node[node]; |
81819f0f CL |
242 | } |
243 | ||
6446faa2 | 244 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
245 | static inline int check_valid_pointer(struct kmem_cache *s, |
246 | struct page *page, const void *object) | |
247 | { | |
248 | void *base; | |
249 | ||
a973e9dd | 250 | if (!object) |
02cbc874 CL |
251 | return 1; |
252 | ||
a973e9dd | 253 | base = page_address(page); |
39b26464 | 254 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
255 | (object - base) % s->size) { |
256 | return 0; | |
257 | } | |
258 | ||
259 | return 1; | |
260 | } | |
261 | ||
7656c72b CL |
262 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
263 | { | |
264 | return *(void **)(object + s->offset); | |
265 | } | |
266 | ||
1393d9a1 CL |
267 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
268 | { | |
269 | void *p; | |
270 | ||
271 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
272 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
273 | #else | |
274 | p = get_freepointer(s, object); | |
275 | #endif | |
276 | return p; | |
277 | } | |
278 | ||
7656c72b CL |
279 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
280 | { | |
281 | *(void **)(object + s->offset) = fp; | |
282 | } | |
283 | ||
284 | /* Loop over all objects in a slab */ | |
224a88be CL |
285 | #define for_each_object(__p, __s, __addr, __objects) \ |
286 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
287 | __p += (__s)->size) |
288 | ||
7656c72b CL |
289 | /* Determine object index from a given position */ |
290 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
291 | { | |
292 | return (p - addr) / s->size; | |
293 | } | |
294 | ||
d71f606f MK |
295 | static inline size_t slab_ksize(const struct kmem_cache *s) |
296 | { | |
297 | #ifdef CONFIG_SLUB_DEBUG | |
298 | /* | |
299 | * Debugging requires use of the padding between object | |
300 | * and whatever may come after it. | |
301 | */ | |
302 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
303 | return s->objsize; | |
304 | ||
305 | #endif | |
306 | /* | |
307 | * If we have the need to store the freelist pointer | |
308 | * back there or track user information then we can | |
309 | * only use the space before that information. | |
310 | */ | |
311 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
312 | return s->inuse; | |
313 | /* | |
314 | * Else we can use all the padding etc for the allocation | |
315 | */ | |
316 | return s->size; | |
317 | } | |
318 | ||
ab9a0f19 LJ |
319 | static inline int order_objects(int order, unsigned long size, int reserved) |
320 | { | |
321 | return ((PAGE_SIZE << order) - reserved) / size; | |
322 | } | |
323 | ||
834f3d11 | 324 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 325 | unsigned long size, int reserved) |
834f3d11 CL |
326 | { |
327 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 328 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
329 | }; |
330 | ||
331 | return x; | |
332 | } | |
333 | ||
334 | static inline int oo_order(struct kmem_cache_order_objects x) | |
335 | { | |
210b5c06 | 336 | return x.x >> OO_SHIFT; |
834f3d11 CL |
337 | } |
338 | ||
339 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
340 | { | |
210b5c06 | 341 | return x.x & OO_MASK; |
834f3d11 CL |
342 | } |
343 | ||
881db7fb CL |
344 | /* |
345 | * Per slab locking using the pagelock | |
346 | */ | |
347 | static __always_inline void slab_lock(struct page *page) | |
348 | { | |
349 | bit_spin_lock(PG_locked, &page->flags); | |
350 | } | |
351 | ||
352 | static __always_inline void slab_unlock(struct page *page) | |
353 | { | |
354 | __bit_spin_unlock(PG_locked, &page->flags); | |
355 | } | |
356 | ||
b789ef51 CL |
357 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
358 | void *freelist_old, unsigned long counters_old, | |
359 | void *freelist_new, unsigned long counters_new, | |
360 | const char *n) | |
361 | { | |
362 | #ifdef CONFIG_CMPXCHG_DOUBLE | |
363 | if (s->flags & __CMPXCHG_DOUBLE) { | |
364 | if (cmpxchg_double(&page->freelist, | |
365 | freelist_old, counters_old, | |
366 | freelist_new, counters_new)) | |
367 | return 1; | |
368 | } else | |
369 | #endif | |
370 | { | |
881db7fb | 371 | slab_lock(page); |
b789ef51 CL |
372 | if (page->freelist == freelist_old && page->counters == counters_old) { |
373 | page->freelist = freelist_new; | |
374 | page->counters = counters_new; | |
881db7fb | 375 | slab_unlock(page); |
b789ef51 CL |
376 | return 1; |
377 | } | |
881db7fb | 378 | slab_unlock(page); |
b789ef51 CL |
379 | } |
380 | ||
381 | cpu_relax(); | |
382 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
383 | ||
384 | #ifdef SLUB_DEBUG_CMPXCHG | |
385 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
386 | #endif | |
387 | ||
388 | return 0; | |
389 | } | |
390 | ||
41ecc55b | 391 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
392 | /* |
393 | * Determine a map of object in use on a page. | |
394 | * | |
881db7fb | 395 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
396 | * not vanish from under us. |
397 | */ | |
398 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
399 | { | |
400 | void *p; | |
401 | void *addr = page_address(page); | |
402 | ||
403 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
404 | set_bit(slab_index(p, s, addr), map); | |
405 | } | |
406 | ||
41ecc55b CL |
407 | /* |
408 | * Debug settings: | |
409 | */ | |
f0630fff CL |
410 | #ifdef CONFIG_SLUB_DEBUG_ON |
411 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
412 | #else | |
41ecc55b | 413 | static int slub_debug; |
f0630fff | 414 | #endif |
41ecc55b CL |
415 | |
416 | static char *slub_debug_slabs; | |
fa5ec8a1 | 417 | static int disable_higher_order_debug; |
41ecc55b | 418 | |
81819f0f CL |
419 | /* |
420 | * Object debugging | |
421 | */ | |
422 | static void print_section(char *text, u8 *addr, unsigned int length) | |
423 | { | |
424 | int i, offset; | |
425 | int newline = 1; | |
426 | char ascii[17]; | |
427 | ||
428 | ascii[16] = 0; | |
429 | ||
430 | for (i = 0; i < length; i++) { | |
431 | if (newline) { | |
24922684 | 432 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
433 | newline = 0; |
434 | } | |
06428780 | 435 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
436 | offset = i % 16; |
437 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
438 | if (offset == 15) { | |
06428780 | 439 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
440 | newline = 1; |
441 | } | |
442 | } | |
443 | if (!newline) { | |
444 | i %= 16; | |
445 | while (i < 16) { | |
06428780 | 446 | printk(KERN_CONT " "); |
81819f0f CL |
447 | ascii[i] = ' '; |
448 | i++; | |
449 | } | |
06428780 | 450 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
451 | } |
452 | } | |
453 | ||
81819f0f CL |
454 | static struct track *get_track(struct kmem_cache *s, void *object, |
455 | enum track_item alloc) | |
456 | { | |
457 | struct track *p; | |
458 | ||
459 | if (s->offset) | |
460 | p = object + s->offset + sizeof(void *); | |
461 | else | |
462 | p = object + s->inuse; | |
463 | ||
464 | return p + alloc; | |
465 | } | |
466 | ||
467 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 468 | enum track_item alloc, unsigned long addr) |
81819f0f | 469 | { |
1a00df4a | 470 | struct track *p = get_track(s, object, alloc); |
81819f0f | 471 | |
81819f0f CL |
472 | if (addr) { |
473 | p->addr = addr; | |
474 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 475 | p->pid = current->pid; |
81819f0f CL |
476 | p->when = jiffies; |
477 | } else | |
478 | memset(p, 0, sizeof(struct track)); | |
479 | } | |
480 | ||
81819f0f CL |
481 | static void init_tracking(struct kmem_cache *s, void *object) |
482 | { | |
24922684 CL |
483 | if (!(s->flags & SLAB_STORE_USER)) |
484 | return; | |
485 | ||
ce71e27c EGM |
486 | set_track(s, object, TRACK_FREE, 0UL); |
487 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
488 | } |
489 | ||
490 | static void print_track(const char *s, struct track *t) | |
491 | { | |
492 | if (!t->addr) | |
493 | return; | |
494 | ||
7daf705f | 495 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 496 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
24922684 CL |
497 | } |
498 | ||
499 | static void print_tracking(struct kmem_cache *s, void *object) | |
500 | { | |
501 | if (!(s->flags & SLAB_STORE_USER)) | |
502 | return; | |
503 | ||
504 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
505 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
506 | } | |
507 | ||
508 | static void print_page_info(struct page *page) | |
509 | { | |
39b26464 CL |
510 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
511 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
512 | |
513 | } | |
514 | ||
515 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
516 | { | |
517 | va_list args; | |
518 | char buf[100]; | |
519 | ||
520 | va_start(args, fmt); | |
521 | vsnprintf(buf, sizeof(buf), fmt, args); | |
522 | va_end(args); | |
523 | printk(KERN_ERR "========================================" | |
524 | "=====================================\n"); | |
525 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
526 | printk(KERN_ERR "----------------------------------------" | |
527 | "-------------------------------------\n\n"); | |
81819f0f CL |
528 | } |
529 | ||
24922684 CL |
530 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
531 | { | |
532 | va_list args; | |
533 | char buf[100]; | |
534 | ||
535 | va_start(args, fmt); | |
536 | vsnprintf(buf, sizeof(buf), fmt, args); | |
537 | va_end(args); | |
538 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
539 | } | |
540 | ||
541 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
542 | { |
543 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 544 | u8 *addr = page_address(page); |
24922684 CL |
545 | |
546 | print_tracking(s, p); | |
547 | ||
548 | print_page_info(page); | |
549 | ||
550 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
551 | p, p - addr, get_freepointer(s, p)); | |
552 | ||
553 | if (p > addr + 16) | |
554 | print_section("Bytes b4", p - 16, 16); | |
555 | ||
0ebd652b | 556 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
557 | |
558 | if (s->flags & SLAB_RED_ZONE) | |
559 | print_section("Redzone", p + s->objsize, | |
560 | s->inuse - s->objsize); | |
561 | ||
81819f0f CL |
562 | if (s->offset) |
563 | off = s->offset + sizeof(void *); | |
564 | else | |
565 | off = s->inuse; | |
566 | ||
24922684 | 567 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 568 | off += 2 * sizeof(struct track); |
81819f0f CL |
569 | |
570 | if (off != s->size) | |
571 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
572 | print_section("Padding", p + off, s->size - off); |
573 | ||
574 | dump_stack(); | |
81819f0f CL |
575 | } |
576 | ||
577 | static void object_err(struct kmem_cache *s, struct page *page, | |
578 | u8 *object, char *reason) | |
579 | { | |
3dc50637 | 580 | slab_bug(s, "%s", reason); |
24922684 | 581 | print_trailer(s, page, object); |
81819f0f CL |
582 | } |
583 | ||
24922684 | 584 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
585 | { |
586 | va_list args; | |
587 | char buf[100]; | |
588 | ||
24922684 CL |
589 | va_start(args, fmt); |
590 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 591 | va_end(args); |
3dc50637 | 592 | slab_bug(s, "%s", buf); |
24922684 | 593 | print_page_info(page); |
81819f0f CL |
594 | dump_stack(); |
595 | } | |
596 | ||
f7cb1933 | 597 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
598 | { |
599 | u8 *p = object; | |
600 | ||
601 | if (s->flags & __OBJECT_POISON) { | |
602 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 603 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
604 | } |
605 | ||
606 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 607 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
608 | } |
609 | ||
24922684 | 610 | static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes) |
81819f0f CL |
611 | { |
612 | while (bytes) { | |
613 | if (*start != (u8)value) | |
24922684 | 614 | return start; |
81819f0f CL |
615 | start++; |
616 | bytes--; | |
617 | } | |
24922684 CL |
618 | return NULL; |
619 | } | |
620 | ||
621 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
622 | void *from, void *to) | |
623 | { | |
624 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
625 | memset(from, data, to - from); | |
626 | } | |
627 | ||
628 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
629 | u8 *object, char *what, | |
06428780 | 630 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
631 | { |
632 | u8 *fault; | |
633 | u8 *end; | |
634 | ||
635 | fault = check_bytes(start, value, bytes); | |
636 | if (!fault) | |
637 | return 1; | |
638 | ||
639 | end = start + bytes; | |
640 | while (end > fault && end[-1] == value) | |
641 | end--; | |
642 | ||
643 | slab_bug(s, "%s overwritten", what); | |
644 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
645 | fault, end - 1, fault[0], value); | |
646 | print_trailer(s, page, object); | |
647 | ||
648 | restore_bytes(s, what, value, fault, end); | |
649 | return 0; | |
81819f0f CL |
650 | } |
651 | ||
81819f0f CL |
652 | /* |
653 | * Object layout: | |
654 | * | |
655 | * object address | |
656 | * Bytes of the object to be managed. | |
657 | * If the freepointer may overlay the object then the free | |
658 | * pointer is the first word of the object. | |
672bba3a | 659 | * |
81819f0f CL |
660 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
661 | * 0xa5 (POISON_END) | |
662 | * | |
663 | * object + s->objsize | |
664 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
665 | * Padding is extended by another word if Redzoning is enabled and |
666 | * objsize == inuse. | |
667 | * | |
81819f0f CL |
668 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
669 | * 0xcc (RED_ACTIVE) for objects in use. | |
670 | * | |
671 | * object + s->inuse | |
672bba3a CL |
672 | * Meta data starts here. |
673 | * | |
81819f0f CL |
674 | * A. Free pointer (if we cannot overwrite object on free) |
675 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 676 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 677 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
678 | * before the word boundary. |
679 | * | |
680 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
681 | * |
682 | * object + s->size | |
672bba3a | 683 | * Nothing is used beyond s->size. |
81819f0f | 684 | * |
672bba3a CL |
685 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
686 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
687 | * may be used with merged slabcaches. |
688 | */ | |
689 | ||
81819f0f CL |
690 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
691 | { | |
692 | unsigned long off = s->inuse; /* The end of info */ | |
693 | ||
694 | if (s->offset) | |
695 | /* Freepointer is placed after the object. */ | |
696 | off += sizeof(void *); | |
697 | ||
698 | if (s->flags & SLAB_STORE_USER) | |
699 | /* We also have user information there */ | |
700 | off += 2 * sizeof(struct track); | |
701 | ||
702 | if (s->size == off) | |
703 | return 1; | |
704 | ||
24922684 CL |
705 | return check_bytes_and_report(s, page, p, "Object padding", |
706 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
707 | } |
708 | ||
39b26464 | 709 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
710 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
711 | { | |
24922684 CL |
712 | u8 *start; |
713 | u8 *fault; | |
714 | u8 *end; | |
715 | int length; | |
716 | int remainder; | |
81819f0f CL |
717 | |
718 | if (!(s->flags & SLAB_POISON)) | |
719 | return 1; | |
720 | ||
a973e9dd | 721 | start = page_address(page); |
ab9a0f19 | 722 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
723 | end = start + length; |
724 | remainder = length % s->size; | |
81819f0f CL |
725 | if (!remainder) |
726 | return 1; | |
727 | ||
39b26464 | 728 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
729 | if (!fault) |
730 | return 1; | |
731 | while (end > fault && end[-1] == POISON_INUSE) | |
732 | end--; | |
733 | ||
734 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 735 | print_section("Padding", end - remainder, remainder); |
24922684 | 736 | |
8a3d271d | 737 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 738 | return 0; |
81819f0f CL |
739 | } |
740 | ||
741 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 742 | void *object, u8 val) |
81819f0f CL |
743 | { |
744 | u8 *p = object; | |
745 | u8 *endobject = object + s->objsize; | |
746 | ||
747 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 748 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 749 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 750 | return 0; |
81819f0f | 751 | } else { |
3adbefee IM |
752 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
753 | check_bytes_and_report(s, page, p, "Alignment padding", | |
754 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
755 | } | |
81819f0f CL |
756 | } |
757 | ||
758 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 759 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
760 | (!check_bytes_and_report(s, page, p, "Poison", p, |
761 | POISON_FREE, s->objsize - 1) || | |
762 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 763 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 764 | return 0; |
81819f0f CL |
765 | /* |
766 | * check_pad_bytes cleans up on its own. | |
767 | */ | |
768 | check_pad_bytes(s, page, p); | |
769 | } | |
770 | ||
f7cb1933 | 771 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
772 | /* |
773 | * Object and freepointer overlap. Cannot check | |
774 | * freepointer while object is allocated. | |
775 | */ | |
776 | return 1; | |
777 | ||
778 | /* Check free pointer validity */ | |
779 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
780 | object_err(s, page, p, "Freepointer corrupt"); | |
781 | /* | |
9f6c708e | 782 | * No choice but to zap it and thus lose the remainder |
81819f0f | 783 | * of the free objects in this slab. May cause |
672bba3a | 784 | * another error because the object count is now wrong. |
81819f0f | 785 | */ |
a973e9dd | 786 | set_freepointer(s, p, NULL); |
81819f0f CL |
787 | return 0; |
788 | } | |
789 | return 1; | |
790 | } | |
791 | ||
792 | static int check_slab(struct kmem_cache *s, struct page *page) | |
793 | { | |
39b26464 CL |
794 | int maxobj; |
795 | ||
81819f0f CL |
796 | VM_BUG_ON(!irqs_disabled()); |
797 | ||
798 | if (!PageSlab(page)) { | |
24922684 | 799 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
800 | return 0; |
801 | } | |
39b26464 | 802 | |
ab9a0f19 | 803 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
804 | if (page->objects > maxobj) { |
805 | slab_err(s, page, "objects %u > max %u", | |
806 | s->name, page->objects, maxobj); | |
807 | return 0; | |
808 | } | |
809 | if (page->inuse > page->objects) { | |
24922684 | 810 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 811 | s->name, page->inuse, page->objects); |
81819f0f CL |
812 | return 0; |
813 | } | |
814 | /* Slab_pad_check fixes things up after itself */ | |
815 | slab_pad_check(s, page); | |
816 | return 1; | |
817 | } | |
818 | ||
819 | /* | |
672bba3a CL |
820 | * Determine if a certain object on a page is on the freelist. Must hold the |
821 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
822 | */ |
823 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
824 | { | |
825 | int nr = 0; | |
881db7fb | 826 | void *fp; |
81819f0f | 827 | void *object = NULL; |
224a88be | 828 | unsigned long max_objects; |
81819f0f | 829 | |
881db7fb | 830 | fp = page->freelist; |
39b26464 | 831 | while (fp && nr <= page->objects) { |
81819f0f CL |
832 | if (fp == search) |
833 | return 1; | |
834 | if (!check_valid_pointer(s, page, fp)) { | |
835 | if (object) { | |
836 | object_err(s, page, object, | |
837 | "Freechain corrupt"); | |
a973e9dd | 838 | set_freepointer(s, object, NULL); |
81819f0f CL |
839 | break; |
840 | } else { | |
24922684 | 841 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 842 | page->freelist = NULL; |
39b26464 | 843 | page->inuse = page->objects; |
24922684 | 844 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
845 | return 0; |
846 | } | |
847 | break; | |
848 | } | |
849 | object = fp; | |
850 | fp = get_freepointer(s, object); | |
851 | nr++; | |
852 | } | |
853 | ||
ab9a0f19 | 854 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
855 | if (max_objects > MAX_OBJS_PER_PAGE) |
856 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
857 | |
858 | if (page->objects != max_objects) { | |
859 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
860 | "should be %d", page->objects, max_objects); | |
861 | page->objects = max_objects; | |
862 | slab_fix(s, "Number of objects adjusted."); | |
863 | } | |
39b26464 | 864 | if (page->inuse != page->objects - nr) { |
70d71228 | 865 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
866 | "counted were %d", page->inuse, page->objects - nr); |
867 | page->inuse = page->objects - nr; | |
24922684 | 868 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
869 | } |
870 | return search == NULL; | |
871 | } | |
872 | ||
0121c619 CL |
873 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
874 | int alloc) | |
3ec09742 CL |
875 | { |
876 | if (s->flags & SLAB_TRACE) { | |
877 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
878 | s->name, | |
879 | alloc ? "alloc" : "free", | |
880 | object, page->inuse, | |
881 | page->freelist); | |
882 | ||
883 | if (!alloc) | |
884 | print_section("Object", (void *)object, s->objsize); | |
885 | ||
886 | dump_stack(); | |
887 | } | |
888 | } | |
889 | ||
c016b0bd CL |
890 | /* |
891 | * Hooks for other subsystems that check memory allocations. In a typical | |
892 | * production configuration these hooks all should produce no code at all. | |
893 | */ | |
894 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
895 | { | |
c1d50836 | 896 | flags &= gfp_allowed_mask; |
c016b0bd CL |
897 | lockdep_trace_alloc(flags); |
898 | might_sleep_if(flags & __GFP_WAIT); | |
899 | ||
900 | return should_failslab(s->objsize, flags, s->flags); | |
901 | } | |
902 | ||
903 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
904 | { | |
c1d50836 | 905 | flags &= gfp_allowed_mask; |
b3d41885 | 906 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
c016b0bd CL |
907 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
908 | } | |
909 | ||
910 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
911 | { | |
912 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 913 | |
d3f661d6 CL |
914 | /* |
915 | * Trouble is that we may no longer disable interupts in the fast path | |
916 | * So in order to make the debug calls that expect irqs to be | |
917 | * disabled we need to disable interrupts temporarily. | |
918 | */ | |
919 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
920 | { | |
921 | unsigned long flags; | |
922 | ||
923 | local_irq_save(flags); | |
924 | kmemcheck_slab_free(s, x, s->objsize); | |
925 | debug_check_no_locks_freed(x, s->objsize); | |
d3f661d6 CL |
926 | local_irq_restore(flags); |
927 | } | |
928 | #endif | |
f9b615de TG |
929 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
930 | debug_check_no_obj_freed(x, s->objsize); | |
c016b0bd CL |
931 | } |
932 | ||
643b1138 | 933 | /* |
672bba3a | 934 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
935 | * |
936 | * list_lock must be held. | |
643b1138 | 937 | */ |
5cc6eee8 CL |
938 | static void add_full(struct kmem_cache *s, |
939 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 940 | { |
5cc6eee8 CL |
941 | if (!(s->flags & SLAB_STORE_USER)) |
942 | return; | |
943 | ||
643b1138 | 944 | list_add(&page->lru, &n->full); |
643b1138 CL |
945 | } |
946 | ||
5cc6eee8 CL |
947 | /* |
948 | * list_lock must be held. | |
949 | */ | |
643b1138 CL |
950 | static void remove_full(struct kmem_cache *s, struct page *page) |
951 | { | |
643b1138 CL |
952 | if (!(s->flags & SLAB_STORE_USER)) |
953 | return; | |
954 | ||
643b1138 | 955 | list_del(&page->lru); |
643b1138 CL |
956 | } |
957 | ||
0f389ec6 CL |
958 | /* Tracking of the number of slabs for debugging purposes */ |
959 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
960 | { | |
961 | struct kmem_cache_node *n = get_node(s, node); | |
962 | ||
963 | return atomic_long_read(&n->nr_slabs); | |
964 | } | |
965 | ||
26c02cf0 AB |
966 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
967 | { | |
968 | return atomic_long_read(&n->nr_slabs); | |
969 | } | |
970 | ||
205ab99d | 971 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
972 | { |
973 | struct kmem_cache_node *n = get_node(s, node); | |
974 | ||
975 | /* | |
976 | * May be called early in order to allocate a slab for the | |
977 | * kmem_cache_node structure. Solve the chicken-egg | |
978 | * dilemma by deferring the increment of the count during | |
979 | * bootstrap (see early_kmem_cache_node_alloc). | |
980 | */ | |
7340cc84 | 981 | if (n) { |
0f389ec6 | 982 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
983 | atomic_long_add(objects, &n->total_objects); |
984 | } | |
0f389ec6 | 985 | } |
205ab99d | 986 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
987 | { |
988 | struct kmem_cache_node *n = get_node(s, node); | |
989 | ||
990 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 991 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
992 | } |
993 | ||
994 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
995 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
996 | void *object) | |
997 | { | |
998 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
999 | return; | |
1000 | ||
f7cb1933 | 1001 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1002 | init_tracking(s, object); |
1003 | } | |
1004 | ||
1537066c | 1005 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1006 | void *object, unsigned long addr) |
81819f0f CL |
1007 | { |
1008 | if (!check_slab(s, page)) | |
1009 | goto bad; | |
1010 | ||
81819f0f CL |
1011 | if (!check_valid_pointer(s, page, object)) { |
1012 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1013 | goto bad; |
81819f0f CL |
1014 | } |
1015 | ||
f7cb1933 | 1016 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1017 | goto bad; |
81819f0f | 1018 | |
3ec09742 CL |
1019 | /* Success perform special debug activities for allocs */ |
1020 | if (s->flags & SLAB_STORE_USER) | |
1021 | set_track(s, object, TRACK_ALLOC, addr); | |
1022 | trace(s, page, object, 1); | |
f7cb1933 | 1023 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1024 | return 1; |
3ec09742 | 1025 | |
81819f0f CL |
1026 | bad: |
1027 | if (PageSlab(page)) { | |
1028 | /* | |
1029 | * If this is a slab page then lets do the best we can | |
1030 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1031 | * as used avoids touching the remaining objects. |
81819f0f | 1032 | */ |
24922684 | 1033 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1034 | page->inuse = page->objects; |
a973e9dd | 1035 | page->freelist = NULL; |
81819f0f CL |
1036 | } |
1037 | return 0; | |
1038 | } | |
1039 | ||
1537066c CL |
1040 | static noinline int free_debug_processing(struct kmem_cache *s, |
1041 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1042 | { |
5c2e4bbb CL |
1043 | unsigned long flags; |
1044 | int rc = 0; | |
1045 | ||
1046 | local_irq_save(flags); | |
881db7fb CL |
1047 | slab_lock(page); |
1048 | ||
81819f0f CL |
1049 | if (!check_slab(s, page)) |
1050 | goto fail; | |
1051 | ||
1052 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1053 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1054 | goto fail; |
1055 | } | |
1056 | ||
1057 | if (on_freelist(s, page, object)) { | |
24922684 | 1058 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1059 | goto fail; |
1060 | } | |
1061 | ||
f7cb1933 | 1062 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1063 | goto out; |
81819f0f CL |
1064 | |
1065 | if (unlikely(s != page->slab)) { | |
3adbefee | 1066 | if (!PageSlab(page)) { |
70d71228 CL |
1067 | slab_err(s, page, "Attempt to free object(0x%p) " |
1068 | "outside of slab", object); | |
3adbefee | 1069 | } else if (!page->slab) { |
81819f0f | 1070 | printk(KERN_ERR |
70d71228 | 1071 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1072 | object); |
70d71228 | 1073 | dump_stack(); |
06428780 | 1074 | } else |
24922684 CL |
1075 | object_err(s, page, object, |
1076 | "page slab pointer corrupt."); | |
81819f0f CL |
1077 | goto fail; |
1078 | } | |
3ec09742 | 1079 | |
3ec09742 CL |
1080 | if (s->flags & SLAB_STORE_USER) |
1081 | set_track(s, object, TRACK_FREE, addr); | |
1082 | trace(s, page, object, 0); | |
f7cb1933 | 1083 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb CL |
1084 | rc = 1; |
1085 | out: | |
881db7fb | 1086 | slab_unlock(page); |
5c2e4bbb CL |
1087 | local_irq_restore(flags); |
1088 | return rc; | |
3ec09742 | 1089 | |
81819f0f | 1090 | fail: |
24922684 | 1091 | slab_fix(s, "Object at 0x%p not freed", object); |
5c2e4bbb | 1092 | goto out; |
81819f0f CL |
1093 | } |
1094 | ||
41ecc55b CL |
1095 | static int __init setup_slub_debug(char *str) |
1096 | { | |
f0630fff CL |
1097 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1098 | if (*str++ != '=' || !*str) | |
1099 | /* | |
1100 | * No options specified. Switch on full debugging. | |
1101 | */ | |
1102 | goto out; | |
1103 | ||
1104 | if (*str == ',') | |
1105 | /* | |
1106 | * No options but restriction on slabs. This means full | |
1107 | * debugging for slabs matching a pattern. | |
1108 | */ | |
1109 | goto check_slabs; | |
1110 | ||
fa5ec8a1 DR |
1111 | if (tolower(*str) == 'o') { |
1112 | /* | |
1113 | * Avoid enabling debugging on caches if its minimum order | |
1114 | * would increase as a result. | |
1115 | */ | |
1116 | disable_higher_order_debug = 1; | |
1117 | goto out; | |
1118 | } | |
1119 | ||
f0630fff CL |
1120 | slub_debug = 0; |
1121 | if (*str == '-') | |
1122 | /* | |
1123 | * Switch off all debugging measures. | |
1124 | */ | |
1125 | goto out; | |
1126 | ||
1127 | /* | |
1128 | * Determine which debug features should be switched on | |
1129 | */ | |
06428780 | 1130 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1131 | switch (tolower(*str)) { |
1132 | case 'f': | |
1133 | slub_debug |= SLAB_DEBUG_FREE; | |
1134 | break; | |
1135 | case 'z': | |
1136 | slub_debug |= SLAB_RED_ZONE; | |
1137 | break; | |
1138 | case 'p': | |
1139 | slub_debug |= SLAB_POISON; | |
1140 | break; | |
1141 | case 'u': | |
1142 | slub_debug |= SLAB_STORE_USER; | |
1143 | break; | |
1144 | case 't': | |
1145 | slub_debug |= SLAB_TRACE; | |
1146 | break; | |
4c13dd3b DM |
1147 | case 'a': |
1148 | slub_debug |= SLAB_FAILSLAB; | |
1149 | break; | |
f0630fff CL |
1150 | default: |
1151 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1152 | "unknown. skipped\n", *str); |
f0630fff | 1153 | } |
41ecc55b CL |
1154 | } |
1155 | ||
f0630fff | 1156 | check_slabs: |
41ecc55b CL |
1157 | if (*str == ',') |
1158 | slub_debug_slabs = str + 1; | |
f0630fff | 1159 | out: |
41ecc55b CL |
1160 | return 1; |
1161 | } | |
1162 | ||
1163 | __setup("slub_debug", setup_slub_debug); | |
1164 | ||
ba0268a8 CL |
1165 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1166 | unsigned long flags, const char *name, | |
51cc5068 | 1167 | void (*ctor)(void *)) |
41ecc55b CL |
1168 | { |
1169 | /* | |
e153362a | 1170 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1171 | */ |
e153362a | 1172 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1173 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1174 | flags |= slub_debug; | |
ba0268a8 CL |
1175 | |
1176 | return flags; | |
41ecc55b CL |
1177 | } |
1178 | #else | |
3ec09742 CL |
1179 | static inline void setup_object_debug(struct kmem_cache *s, |
1180 | struct page *page, void *object) {} | |
41ecc55b | 1181 | |
3ec09742 | 1182 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1183 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1184 | |
3ec09742 | 1185 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1186 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1187 | |
41ecc55b CL |
1188 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1189 | { return 1; } | |
1190 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1191 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1192 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1193 | struct page *page) {} | |
2cfb7455 | 1194 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
ba0268a8 CL |
1195 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1196 | unsigned long flags, const char *name, | |
51cc5068 | 1197 | void (*ctor)(void *)) |
ba0268a8 CL |
1198 | { |
1199 | return flags; | |
1200 | } | |
41ecc55b | 1201 | #define slub_debug 0 |
0f389ec6 | 1202 | |
fdaa45e9 IM |
1203 | #define disable_higher_order_debug 0 |
1204 | ||
0f389ec6 CL |
1205 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1206 | { return 0; } | |
26c02cf0 AB |
1207 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1208 | { return 0; } | |
205ab99d CL |
1209 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1210 | int objects) {} | |
1211 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1212 | int objects) {} | |
7d550c56 CL |
1213 | |
1214 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1215 | { return 0; } | |
1216 | ||
1217 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1218 | void *object) {} | |
1219 | ||
1220 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1221 | ||
ab4d5ed5 | 1222 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1223 | |
81819f0f CL |
1224 | /* |
1225 | * Slab allocation and freeing | |
1226 | */ | |
65c3376a CL |
1227 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1228 | struct kmem_cache_order_objects oo) | |
1229 | { | |
1230 | int order = oo_order(oo); | |
1231 | ||
b1eeab67 VN |
1232 | flags |= __GFP_NOTRACK; |
1233 | ||
2154a336 | 1234 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1235 | return alloc_pages(flags, order); |
1236 | else | |
6b65aaf3 | 1237 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1238 | } |
1239 | ||
81819f0f CL |
1240 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1241 | { | |
06428780 | 1242 | struct page *page; |
834f3d11 | 1243 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1244 | gfp_t alloc_gfp; |
81819f0f | 1245 | |
7e0528da CL |
1246 | flags &= gfp_allowed_mask; |
1247 | ||
1248 | if (flags & __GFP_WAIT) | |
1249 | local_irq_enable(); | |
1250 | ||
b7a49f0d | 1251 | flags |= s->allocflags; |
e12ba74d | 1252 | |
ba52270d PE |
1253 | /* |
1254 | * Let the initial higher-order allocation fail under memory pressure | |
1255 | * so we fall-back to the minimum order allocation. | |
1256 | */ | |
1257 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1258 | ||
1259 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1260 | if (unlikely(!page)) { |
1261 | oo = s->min; | |
1262 | /* | |
1263 | * Allocation may have failed due to fragmentation. | |
1264 | * Try a lower order alloc if possible | |
1265 | */ | |
1266 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1267 | |
7e0528da CL |
1268 | if (page) |
1269 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1270 | } |
5a896d9e | 1271 | |
7e0528da CL |
1272 | if (flags & __GFP_WAIT) |
1273 | local_irq_disable(); | |
1274 | ||
1275 | if (!page) | |
1276 | return NULL; | |
1277 | ||
5a896d9e | 1278 | if (kmemcheck_enabled |
5086c389 | 1279 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1280 | int pages = 1 << oo_order(oo); |
1281 | ||
1282 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1283 | ||
1284 | /* | |
1285 | * Objects from caches that have a constructor don't get | |
1286 | * cleared when they're allocated, so we need to do it here. | |
1287 | */ | |
1288 | if (s->ctor) | |
1289 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1290 | else | |
1291 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1292 | } |
1293 | ||
834f3d11 | 1294 | page->objects = oo_objects(oo); |
81819f0f CL |
1295 | mod_zone_page_state(page_zone(page), |
1296 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1297 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1298 | 1 << oo_order(oo)); |
81819f0f CL |
1299 | |
1300 | return page; | |
1301 | } | |
1302 | ||
1303 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1304 | void *object) | |
1305 | { | |
3ec09742 | 1306 | setup_object_debug(s, page, object); |
4f104934 | 1307 | if (unlikely(s->ctor)) |
51cc5068 | 1308 | s->ctor(object); |
81819f0f CL |
1309 | } |
1310 | ||
1311 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1312 | { | |
1313 | struct page *page; | |
81819f0f | 1314 | void *start; |
81819f0f CL |
1315 | void *last; |
1316 | void *p; | |
1317 | ||
6cb06229 | 1318 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1319 | |
6cb06229 CL |
1320 | page = allocate_slab(s, |
1321 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1322 | if (!page) |
1323 | goto out; | |
1324 | ||
205ab99d | 1325 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1326 | page->slab = s; |
1327 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1328 | |
1329 | start = page_address(page); | |
81819f0f CL |
1330 | |
1331 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1332 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1333 | |
1334 | last = start; | |
224a88be | 1335 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1336 | setup_object(s, page, last); |
1337 | set_freepointer(s, last, p); | |
1338 | last = p; | |
1339 | } | |
1340 | setup_object(s, page, last); | |
a973e9dd | 1341 | set_freepointer(s, last, NULL); |
81819f0f CL |
1342 | |
1343 | page->freelist = start; | |
1344 | page->inuse = 0; | |
8cb0a506 | 1345 | page->frozen = 1; |
81819f0f | 1346 | out: |
81819f0f CL |
1347 | return page; |
1348 | } | |
1349 | ||
1350 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1351 | { | |
834f3d11 CL |
1352 | int order = compound_order(page); |
1353 | int pages = 1 << order; | |
81819f0f | 1354 | |
af537b0a | 1355 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1356 | void *p; |
1357 | ||
1358 | slab_pad_check(s, page); | |
224a88be CL |
1359 | for_each_object(p, s, page_address(page), |
1360 | page->objects) | |
f7cb1933 | 1361 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1362 | } |
1363 | ||
b1eeab67 | 1364 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1365 | |
81819f0f CL |
1366 | mod_zone_page_state(page_zone(page), |
1367 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1368 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1369 | -pages); |
81819f0f | 1370 | |
49bd5221 CL |
1371 | __ClearPageSlab(page); |
1372 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1373 | if (current->reclaim_state) |
1374 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1375 | __free_pages(page, order); |
81819f0f CL |
1376 | } |
1377 | ||
da9a638c LJ |
1378 | #define need_reserve_slab_rcu \ |
1379 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1380 | ||
81819f0f CL |
1381 | static void rcu_free_slab(struct rcu_head *h) |
1382 | { | |
1383 | struct page *page; | |
1384 | ||
da9a638c LJ |
1385 | if (need_reserve_slab_rcu) |
1386 | page = virt_to_head_page(h); | |
1387 | else | |
1388 | page = container_of((struct list_head *)h, struct page, lru); | |
1389 | ||
81819f0f CL |
1390 | __free_slab(page->slab, page); |
1391 | } | |
1392 | ||
1393 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1394 | { | |
1395 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1396 | struct rcu_head *head; |
1397 | ||
1398 | if (need_reserve_slab_rcu) { | |
1399 | int order = compound_order(page); | |
1400 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1401 | ||
1402 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1403 | head = page_address(page) + offset; | |
1404 | } else { | |
1405 | /* | |
1406 | * RCU free overloads the RCU head over the LRU | |
1407 | */ | |
1408 | head = (void *)&page->lru; | |
1409 | } | |
81819f0f CL |
1410 | |
1411 | call_rcu(head, rcu_free_slab); | |
1412 | } else | |
1413 | __free_slab(s, page); | |
1414 | } | |
1415 | ||
1416 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1417 | { | |
205ab99d | 1418 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1419 | free_slab(s, page); |
1420 | } | |
1421 | ||
81819f0f | 1422 | /* |
5cc6eee8 CL |
1423 | * Management of partially allocated slabs. |
1424 | * | |
1425 | * list_lock must be held. | |
81819f0f | 1426 | */ |
5cc6eee8 | 1427 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1428 | struct page *page, int tail) |
81819f0f | 1429 | { |
e95eed57 | 1430 | n->nr_partial++; |
7c2e132c CL |
1431 | if (tail) |
1432 | list_add_tail(&page->lru, &n->partial); | |
1433 | else | |
1434 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1435 | } |
1436 | ||
5cc6eee8 CL |
1437 | /* |
1438 | * list_lock must be held. | |
1439 | */ | |
1440 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1441 | struct page *page) |
1442 | { | |
1443 | list_del(&page->lru); | |
1444 | n->nr_partial--; | |
1445 | } | |
1446 | ||
81819f0f | 1447 | /* |
5cc6eee8 CL |
1448 | * Lock slab, remove from the partial list and put the object into the |
1449 | * per cpu freelist. | |
81819f0f | 1450 | * |
672bba3a | 1451 | * Must hold list_lock. |
81819f0f | 1452 | */ |
881db7fb | 1453 | static inline int acquire_slab(struct kmem_cache *s, |
61728d1e | 1454 | struct kmem_cache_node *n, struct page *page) |
81819f0f | 1455 | { |
2cfb7455 CL |
1456 | void *freelist; |
1457 | unsigned long counters; | |
1458 | struct page new; | |
1459 | ||
2cfb7455 CL |
1460 | /* |
1461 | * Zap the freelist and set the frozen bit. | |
1462 | * The old freelist is the list of objects for the | |
1463 | * per cpu allocation list. | |
1464 | */ | |
1465 | do { | |
1466 | freelist = page->freelist; | |
1467 | counters = page->counters; | |
1468 | new.counters = counters; | |
1469 | new.inuse = page->objects; | |
1470 | ||
1471 | VM_BUG_ON(new.frozen); | |
1472 | new.frozen = 1; | |
1473 | ||
1474 | } while (!cmpxchg_double_slab(s, page, | |
1475 | freelist, counters, | |
1476 | NULL, new.counters, | |
1477 | "lock and freeze")); | |
1478 | ||
1479 | remove_partial(n, page); | |
1480 | ||
1481 | if (freelist) { | |
1482 | /* Populate the per cpu freelist */ | |
1483 | this_cpu_write(s->cpu_slab->freelist, freelist); | |
1484 | this_cpu_write(s->cpu_slab->page, page); | |
1485 | this_cpu_write(s->cpu_slab->node, page_to_nid(page)); | |
81819f0f | 1486 | return 1; |
2cfb7455 CL |
1487 | } else { |
1488 | /* | |
1489 | * Slab page came from the wrong list. No object to allocate | |
1490 | * from. Put it onto the correct list and continue partial | |
1491 | * scan. | |
1492 | */ | |
1493 | printk(KERN_ERR "SLUB: %s : Page without available objects on" | |
1494 | " partial list\n", s->name); | |
2cfb7455 | 1495 | return 0; |
81819f0f | 1496 | } |
81819f0f CL |
1497 | } |
1498 | ||
1499 | /* | |
672bba3a | 1500 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1501 | */ |
61728d1e CL |
1502 | static struct page *get_partial_node(struct kmem_cache *s, |
1503 | struct kmem_cache_node *n) | |
81819f0f CL |
1504 | { |
1505 | struct page *page; | |
1506 | ||
1507 | /* | |
1508 | * Racy check. If we mistakenly see no partial slabs then we | |
1509 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1510 | * partial slab and there is none available then get_partials() |
1511 | * will return NULL. | |
81819f0f CL |
1512 | */ |
1513 | if (!n || !n->nr_partial) | |
1514 | return NULL; | |
1515 | ||
1516 | spin_lock(&n->list_lock); | |
1517 | list_for_each_entry(page, &n->partial, lru) | |
881db7fb | 1518 | if (acquire_slab(s, n, page)) |
81819f0f CL |
1519 | goto out; |
1520 | page = NULL; | |
1521 | out: | |
1522 | spin_unlock(&n->list_lock); | |
1523 | return page; | |
1524 | } | |
1525 | ||
1526 | /* | |
672bba3a | 1527 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1528 | */ |
1529 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1530 | { | |
1531 | #ifdef CONFIG_NUMA | |
1532 | struct zonelist *zonelist; | |
dd1a239f | 1533 | struct zoneref *z; |
54a6eb5c MG |
1534 | struct zone *zone; |
1535 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1536 | struct page *page; |
1537 | ||
1538 | /* | |
672bba3a CL |
1539 | * The defrag ratio allows a configuration of the tradeoffs between |
1540 | * inter node defragmentation and node local allocations. A lower | |
1541 | * defrag_ratio increases the tendency to do local allocations | |
1542 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1543 | * |
672bba3a CL |
1544 | * If the defrag_ratio is set to 0 then kmalloc() always |
1545 | * returns node local objects. If the ratio is higher then kmalloc() | |
1546 | * may return off node objects because partial slabs are obtained | |
1547 | * from other nodes and filled up. | |
81819f0f | 1548 | * |
6446faa2 | 1549 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1550 | * defrag_ratio = 1000) then every (well almost) allocation will |
1551 | * first attempt to defrag slab caches on other nodes. This means | |
1552 | * scanning over all nodes to look for partial slabs which may be | |
1553 | * expensive if we do it every time we are trying to find a slab | |
1554 | * with available objects. | |
81819f0f | 1555 | */ |
9824601e CL |
1556 | if (!s->remote_node_defrag_ratio || |
1557 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1558 | return NULL; |
1559 | ||
c0ff7453 | 1560 | get_mems_allowed(); |
0e88460d | 1561 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1562 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1563 | struct kmem_cache_node *n; |
1564 | ||
54a6eb5c | 1565 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1566 | |
54a6eb5c | 1567 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1568 | n->nr_partial > s->min_partial) { |
61728d1e | 1569 | page = get_partial_node(s, n); |
c0ff7453 MX |
1570 | if (page) { |
1571 | put_mems_allowed(); | |
81819f0f | 1572 | return page; |
c0ff7453 | 1573 | } |
81819f0f CL |
1574 | } |
1575 | } | |
c0ff7453 | 1576 | put_mems_allowed(); |
81819f0f CL |
1577 | #endif |
1578 | return NULL; | |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * Get a partial page, lock it and return it. | |
1583 | */ | |
1584 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1585 | { | |
1586 | struct page *page; | |
2154a336 | 1587 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1588 | |
61728d1e | 1589 | page = get_partial_node(s, get_node(s, searchnode)); |
33de04ec | 1590 | if (page || node != NUMA_NO_NODE) |
81819f0f CL |
1591 | return page; |
1592 | ||
1593 | return get_any_partial(s, flags); | |
1594 | } | |
1595 | ||
8a5ec0ba CL |
1596 | #ifdef CONFIG_PREEMPT |
1597 | /* | |
1598 | * Calculate the next globally unique transaction for disambiguiation | |
1599 | * during cmpxchg. The transactions start with the cpu number and are then | |
1600 | * incremented by CONFIG_NR_CPUS. | |
1601 | */ | |
1602 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1603 | #else | |
1604 | /* | |
1605 | * No preemption supported therefore also no need to check for | |
1606 | * different cpus. | |
1607 | */ | |
1608 | #define TID_STEP 1 | |
1609 | #endif | |
1610 | ||
1611 | static inline unsigned long next_tid(unsigned long tid) | |
1612 | { | |
1613 | return tid + TID_STEP; | |
1614 | } | |
1615 | ||
1616 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1617 | { | |
1618 | return tid % TID_STEP; | |
1619 | } | |
1620 | ||
1621 | static inline unsigned long tid_to_event(unsigned long tid) | |
1622 | { | |
1623 | return tid / TID_STEP; | |
1624 | } | |
1625 | ||
1626 | static inline unsigned int init_tid(int cpu) | |
1627 | { | |
1628 | return cpu; | |
1629 | } | |
1630 | ||
1631 | static inline void note_cmpxchg_failure(const char *n, | |
1632 | const struct kmem_cache *s, unsigned long tid) | |
1633 | { | |
1634 | #ifdef SLUB_DEBUG_CMPXCHG | |
1635 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1636 | ||
1637 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1638 | ||
1639 | #ifdef CONFIG_PREEMPT | |
1640 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1641 | printk("due to cpu change %d -> %d\n", | |
1642 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1643 | else | |
1644 | #endif | |
1645 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1646 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1647 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1648 | else | |
1649 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1650 | actual_tid, tid, next_tid(tid)); | |
1651 | #endif | |
4fdccdfb | 1652 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1653 | } |
1654 | ||
8a5ec0ba CL |
1655 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1656 | { | |
8a5ec0ba CL |
1657 | int cpu; |
1658 | ||
1659 | for_each_possible_cpu(cpu) | |
1660 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1661 | } |
2cfb7455 CL |
1662 | /* |
1663 | * Remove the cpu slab | |
1664 | */ | |
1665 | ||
81819f0f CL |
1666 | /* |
1667 | * Remove the cpu slab | |
1668 | */ | |
dfb4f096 | 1669 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1670 | { |
2cfb7455 | 1671 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
dfb4f096 | 1672 | struct page *page = c->page; |
2cfb7455 CL |
1673 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1674 | int lock = 0; | |
1675 | enum slab_modes l = M_NONE, m = M_NONE; | |
1676 | void *freelist; | |
1677 | void *nextfree; | |
1678 | int tail = 0; | |
1679 | struct page new; | |
1680 | struct page old; | |
1681 | ||
1682 | if (page->freelist) { | |
84e554e6 | 1683 | stat(s, DEACTIVATE_REMOTE_FREES); |
2cfb7455 CL |
1684 | tail = 1; |
1685 | } | |
1686 | ||
1687 | c->tid = next_tid(c->tid); | |
1688 | c->page = NULL; | |
1689 | freelist = c->freelist; | |
1690 | c->freelist = NULL; | |
1691 | ||
894b8788 | 1692 | /* |
2cfb7455 CL |
1693 | * Stage one: Free all available per cpu objects back |
1694 | * to the page freelist while it is still frozen. Leave the | |
1695 | * last one. | |
1696 | * | |
1697 | * There is no need to take the list->lock because the page | |
1698 | * is still frozen. | |
1699 | */ | |
1700 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1701 | void *prior; | |
1702 | unsigned long counters; | |
1703 | ||
1704 | do { | |
1705 | prior = page->freelist; | |
1706 | counters = page->counters; | |
1707 | set_freepointer(s, freelist, prior); | |
1708 | new.counters = counters; | |
1709 | new.inuse--; | |
1710 | VM_BUG_ON(!new.frozen); | |
1711 | ||
1712 | } while (!cmpxchg_double_slab(s, page, | |
1713 | prior, counters, | |
1714 | freelist, new.counters, | |
1715 | "drain percpu freelist")); | |
1716 | ||
1717 | freelist = nextfree; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Stage two: Ensure that the page is unfrozen while the | |
1722 | * list presence reflects the actual number of objects | |
1723 | * during unfreeze. | |
1724 | * | |
1725 | * We setup the list membership and then perform a cmpxchg | |
1726 | * with the count. If there is a mismatch then the page | |
1727 | * is not unfrozen but the page is on the wrong list. | |
1728 | * | |
1729 | * Then we restart the process which may have to remove | |
1730 | * the page from the list that we just put it on again | |
1731 | * because the number of objects in the slab may have | |
1732 | * changed. | |
894b8788 | 1733 | */ |
2cfb7455 | 1734 | redo: |
894b8788 | 1735 | |
2cfb7455 CL |
1736 | old.freelist = page->freelist; |
1737 | old.counters = page->counters; | |
1738 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1739 | |
2cfb7455 CL |
1740 | /* Determine target state of the slab */ |
1741 | new.counters = old.counters; | |
1742 | if (freelist) { | |
1743 | new.inuse--; | |
1744 | set_freepointer(s, freelist, old.freelist); | |
1745 | new.freelist = freelist; | |
1746 | } else | |
1747 | new.freelist = old.freelist; | |
1748 | ||
1749 | new.frozen = 0; | |
1750 | ||
1751 | if (!new.inuse && n->nr_partial < s->min_partial) | |
1752 | m = M_FREE; | |
1753 | else if (new.freelist) { | |
1754 | m = M_PARTIAL; | |
1755 | if (!lock) { | |
1756 | lock = 1; | |
1757 | /* | |
1758 | * Taking the spinlock removes the possiblity | |
1759 | * that acquire_slab() will see a slab page that | |
1760 | * is frozen | |
1761 | */ | |
1762 | spin_lock(&n->list_lock); | |
1763 | } | |
1764 | } else { | |
1765 | m = M_FULL; | |
1766 | if (kmem_cache_debug(s) && !lock) { | |
1767 | lock = 1; | |
1768 | /* | |
1769 | * This also ensures that the scanning of full | |
1770 | * slabs from diagnostic functions will not see | |
1771 | * any frozen slabs. | |
1772 | */ | |
1773 | spin_lock(&n->list_lock); | |
1774 | } | |
1775 | } | |
1776 | ||
1777 | if (l != m) { | |
1778 | ||
1779 | if (l == M_PARTIAL) | |
1780 | ||
1781 | remove_partial(n, page); | |
1782 | ||
1783 | else if (l == M_FULL) | |
894b8788 | 1784 | |
2cfb7455 CL |
1785 | remove_full(s, page); |
1786 | ||
1787 | if (m == M_PARTIAL) { | |
1788 | ||
1789 | add_partial(n, page, tail); | |
1790 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); | |
1791 | ||
1792 | } else if (m == M_FULL) { | |
1793 | ||
1794 | stat(s, DEACTIVATE_FULL); | |
1795 | add_full(s, n, page); | |
1796 | ||
1797 | } | |
1798 | } | |
1799 | ||
1800 | l = m; | |
1801 | if (!cmpxchg_double_slab(s, page, | |
1802 | old.freelist, old.counters, | |
1803 | new.freelist, new.counters, | |
1804 | "unfreezing slab")) | |
1805 | goto redo; | |
1806 | ||
2cfb7455 CL |
1807 | if (lock) |
1808 | spin_unlock(&n->list_lock); | |
1809 | ||
1810 | if (m == M_FREE) { | |
1811 | stat(s, DEACTIVATE_EMPTY); | |
1812 | discard_slab(s, page); | |
1813 | stat(s, FREE_SLAB); | |
894b8788 | 1814 | } |
81819f0f CL |
1815 | } |
1816 | ||
dfb4f096 | 1817 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1818 | { |
84e554e6 | 1819 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 | 1820 | deactivate_slab(s, c); |
81819f0f CL |
1821 | } |
1822 | ||
1823 | /* | |
1824 | * Flush cpu slab. | |
6446faa2 | 1825 | * |
81819f0f CL |
1826 | * Called from IPI handler with interrupts disabled. |
1827 | */ | |
0c710013 | 1828 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1829 | { |
9dfc6e68 | 1830 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1831 | |
dfb4f096 CL |
1832 | if (likely(c && c->page)) |
1833 | flush_slab(s, c); | |
81819f0f CL |
1834 | } |
1835 | ||
1836 | static void flush_cpu_slab(void *d) | |
1837 | { | |
1838 | struct kmem_cache *s = d; | |
81819f0f | 1839 | |
dfb4f096 | 1840 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1841 | } |
1842 | ||
1843 | static void flush_all(struct kmem_cache *s) | |
1844 | { | |
15c8b6c1 | 1845 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1846 | } |
1847 | ||
dfb4f096 CL |
1848 | /* |
1849 | * Check if the objects in a per cpu structure fit numa | |
1850 | * locality expectations. | |
1851 | */ | |
1852 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1853 | { | |
1854 | #ifdef CONFIG_NUMA | |
2154a336 | 1855 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1856 | return 0; |
1857 | #endif | |
1858 | return 1; | |
1859 | } | |
1860 | ||
781b2ba6 PE |
1861 | static int count_free(struct page *page) |
1862 | { | |
1863 | return page->objects - page->inuse; | |
1864 | } | |
1865 | ||
1866 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1867 | int (*get_count)(struct page *)) | |
1868 | { | |
1869 | unsigned long flags; | |
1870 | unsigned long x = 0; | |
1871 | struct page *page; | |
1872 | ||
1873 | spin_lock_irqsave(&n->list_lock, flags); | |
1874 | list_for_each_entry(page, &n->partial, lru) | |
1875 | x += get_count(page); | |
1876 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1877 | return x; | |
1878 | } | |
1879 | ||
26c02cf0 AB |
1880 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1881 | { | |
1882 | #ifdef CONFIG_SLUB_DEBUG | |
1883 | return atomic_long_read(&n->total_objects); | |
1884 | #else | |
1885 | return 0; | |
1886 | #endif | |
1887 | } | |
1888 | ||
781b2ba6 PE |
1889 | static noinline void |
1890 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1891 | { | |
1892 | int node; | |
1893 | ||
1894 | printk(KERN_WARNING | |
1895 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1896 | nid, gfpflags); | |
1897 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
1898 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
1899 | s->size, oo_order(s->oo), oo_order(s->min)); | |
1900 | ||
fa5ec8a1 DR |
1901 | if (oo_order(s->min) > get_order(s->objsize)) |
1902 | printk(KERN_WARNING " %s debugging increased min order, use " | |
1903 | "slub_debug=O to disable.\n", s->name); | |
1904 | ||
781b2ba6 PE |
1905 | for_each_online_node(node) { |
1906 | struct kmem_cache_node *n = get_node(s, node); | |
1907 | unsigned long nr_slabs; | |
1908 | unsigned long nr_objs; | |
1909 | unsigned long nr_free; | |
1910 | ||
1911 | if (!n) | |
1912 | continue; | |
1913 | ||
26c02cf0 AB |
1914 | nr_free = count_partial(n, count_free); |
1915 | nr_slabs = node_nr_slabs(n); | |
1916 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
1917 | |
1918 | printk(KERN_WARNING | |
1919 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
1920 | node, nr_slabs, nr_objs, nr_free); | |
1921 | } | |
1922 | } | |
1923 | ||
81819f0f | 1924 | /* |
894b8788 CL |
1925 | * Slow path. The lockless freelist is empty or we need to perform |
1926 | * debugging duties. | |
1927 | * | |
1928 | * Interrupts are disabled. | |
81819f0f | 1929 | * |
894b8788 CL |
1930 | * Processing is still very fast if new objects have been freed to the |
1931 | * regular freelist. In that case we simply take over the regular freelist | |
1932 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 1933 | * |
894b8788 CL |
1934 | * If that is not working then we fall back to the partial lists. We take the |
1935 | * first element of the freelist as the object to allocate now and move the | |
1936 | * rest of the freelist to the lockless freelist. | |
81819f0f | 1937 | * |
894b8788 | 1938 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
1939 | * we need to allocate a new slab. This is the slowest path since it involves |
1940 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 1941 | */ |
ce71e27c EGM |
1942 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
1943 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 1944 | { |
81819f0f | 1945 | void **object; |
01ad8a7b | 1946 | struct page *page; |
8a5ec0ba | 1947 | unsigned long flags; |
2cfb7455 CL |
1948 | struct page new; |
1949 | unsigned long counters; | |
8a5ec0ba CL |
1950 | |
1951 | local_irq_save(flags); | |
1952 | #ifdef CONFIG_PREEMPT | |
1953 | /* | |
1954 | * We may have been preempted and rescheduled on a different | |
1955 | * cpu before disabling interrupts. Need to reload cpu area | |
1956 | * pointer. | |
1957 | */ | |
1958 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 1959 | #endif |
81819f0f | 1960 | |
e72e9c23 LT |
1961 | /* We handle __GFP_ZERO in the caller */ |
1962 | gfpflags &= ~__GFP_ZERO; | |
1963 | ||
01ad8a7b CL |
1964 | page = c->page; |
1965 | if (!page) | |
81819f0f CL |
1966 | goto new_slab; |
1967 | ||
fc59c053 | 1968 | if (unlikely(!node_match(c, node))) { |
e36a2652 | 1969 | stat(s, ALLOC_NODE_MISMATCH); |
fc59c053 CL |
1970 | deactivate_slab(s, c); |
1971 | goto new_slab; | |
1972 | } | |
6446faa2 | 1973 | |
2cfb7455 CL |
1974 | stat(s, ALLOC_SLOWPATH); |
1975 | ||
1976 | do { | |
1977 | object = page->freelist; | |
1978 | counters = page->counters; | |
1979 | new.counters = counters; | |
2cfb7455 CL |
1980 | VM_BUG_ON(!new.frozen); |
1981 | ||
03e404af CL |
1982 | /* |
1983 | * If there is no object left then we use this loop to | |
1984 | * deactivate the slab which is simple since no objects | |
1985 | * are left in the slab and therefore we do not need to | |
1986 | * put the page back onto the partial list. | |
1987 | * | |
1988 | * If there are objects left then we retrieve them | |
1989 | * and use them to refill the per cpu queue. | |
1990 | */ | |
1991 | ||
1992 | new.inuse = page->objects; | |
1993 | new.frozen = object != NULL; | |
1994 | ||
2cfb7455 CL |
1995 | } while (!cmpxchg_double_slab(s, page, |
1996 | object, counters, | |
1997 | NULL, new.counters, | |
1998 | "__slab_alloc")); | |
6446faa2 | 1999 | |
03e404af CL |
2000 | if (unlikely(!object)) { |
2001 | c->page = NULL; | |
2002 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2003 | goto new_slab; |
03e404af | 2004 | } |
81819f0f | 2005 | |
2cfb7455 | 2006 | stat(s, ALLOC_REFILL); |
01ad8a7b | 2007 | |
4eade540 CL |
2008 | load_freelist: |
2009 | VM_BUG_ON(!page->frozen); | |
2cfb7455 | 2010 | c->freelist = get_freepointer(s, object); |
8a5ec0ba CL |
2011 | c->tid = next_tid(c->tid); |
2012 | local_irq_restore(flags); | |
81819f0f CL |
2013 | return object; |
2014 | ||
81819f0f | 2015 | new_slab: |
01ad8a7b CL |
2016 | page = get_partial(s, gfpflags, node); |
2017 | if (page) { | |
84e554e6 | 2018 | stat(s, ALLOC_FROM_PARTIAL); |
2cfb7455 CL |
2019 | object = c->freelist; |
2020 | ||
2021 | if (kmem_cache_debug(s)) | |
2022 | goto debug; | |
894b8788 | 2023 | goto load_freelist; |
81819f0f CL |
2024 | } |
2025 | ||
01ad8a7b | 2026 | page = new_slab(s, gfpflags, node); |
b811c202 | 2027 | |
01ad8a7b | 2028 | if (page) { |
9dfc6e68 | 2029 | c = __this_cpu_ptr(s->cpu_slab); |
05aa3450 | 2030 | if (c->page) |
dfb4f096 | 2031 | flush_slab(s, c); |
01ad8a7b | 2032 | |
2cfb7455 CL |
2033 | /* |
2034 | * No other reference to the page yet so we can | |
2035 | * muck around with it freely without cmpxchg | |
2036 | */ | |
2037 | object = page->freelist; | |
2038 | page->freelist = NULL; | |
2039 | page->inuse = page->objects; | |
2040 | ||
2041 | stat(s, ALLOC_SLAB); | |
bd07d87f DR |
2042 | c->node = page_to_nid(page); |
2043 | c->page = page; | |
4b6f0750 | 2044 | goto load_freelist; |
81819f0f | 2045 | } |
95f85989 PE |
2046 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
2047 | slab_out_of_memory(s, gfpflags, node); | |
2fd66c51 | 2048 | local_irq_restore(flags); |
71c7a06f | 2049 | return NULL; |
2cfb7455 | 2050 | |
81819f0f | 2051 | debug: |
2cfb7455 CL |
2052 | if (!object || !alloc_debug_processing(s, page, object, addr)) |
2053 | goto new_slab; | |
894b8788 | 2054 | |
2cfb7455 | 2055 | c->freelist = get_freepointer(s, object); |
442b06bc CL |
2056 | deactivate_slab(s, c); |
2057 | c->page = NULL; | |
15b7c514 | 2058 | c->node = NUMA_NO_NODE; |
a71ae47a CL |
2059 | local_irq_restore(flags); |
2060 | return object; | |
894b8788 CL |
2061 | } |
2062 | ||
2063 | /* | |
2064 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2065 | * have the fastpath folded into their functions. So no function call | |
2066 | * overhead for requests that can be satisfied on the fastpath. | |
2067 | * | |
2068 | * The fastpath works by first checking if the lockless freelist can be used. | |
2069 | * If not then __slab_alloc is called for slow processing. | |
2070 | * | |
2071 | * Otherwise we can simply pick the next object from the lockless free list. | |
2072 | */ | |
06428780 | 2073 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2074 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2075 | { |
894b8788 | 2076 | void **object; |
dfb4f096 | 2077 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2078 | unsigned long tid; |
1f84260c | 2079 | |
c016b0bd | 2080 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2081 | return NULL; |
1f84260c | 2082 | |
8a5ec0ba | 2083 | redo: |
8a5ec0ba CL |
2084 | |
2085 | /* | |
2086 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2087 | * enabled. We may switch back and forth between cpus while | |
2088 | * reading from one cpu area. That does not matter as long | |
2089 | * as we end up on the original cpu again when doing the cmpxchg. | |
2090 | */ | |
9dfc6e68 | 2091 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2092 | |
8a5ec0ba CL |
2093 | /* |
2094 | * The transaction ids are globally unique per cpu and per operation on | |
2095 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2096 | * occurs on the right processor and that there was no operation on the | |
2097 | * linked list in between. | |
2098 | */ | |
2099 | tid = c->tid; | |
2100 | barrier(); | |
8a5ec0ba | 2101 | |
9dfc6e68 | 2102 | object = c->freelist; |
9dfc6e68 | 2103 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 2104 | |
dfb4f096 | 2105 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2106 | |
2107 | else { | |
8a5ec0ba | 2108 | /* |
25985edc | 2109 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2110 | * operation and if we are on the right processor. |
2111 | * | |
2112 | * The cmpxchg does the following atomically (without lock semantics!) | |
2113 | * 1. Relocate first pointer to the current per cpu area. | |
2114 | * 2. Verify that tid and freelist have not been changed | |
2115 | * 3. If they were not changed replace tid and freelist | |
2116 | * | |
2117 | * Since this is without lock semantics the protection is only against | |
2118 | * code executing on this cpu *not* from access by other cpus. | |
2119 | */ | |
30106b8c | 2120 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2121 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2122 | object, tid, | |
1393d9a1 | 2123 | get_freepointer_safe(s, object), next_tid(tid)))) { |
8a5ec0ba CL |
2124 | |
2125 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2126 | goto redo; | |
2127 | } | |
84e554e6 | 2128 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2129 | } |
8a5ec0ba | 2130 | |
74e2134f | 2131 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 2132 | memset(object, 0, s->objsize); |
d07dbea4 | 2133 | |
c016b0bd | 2134 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2135 | |
894b8788 | 2136 | return object; |
81819f0f CL |
2137 | } |
2138 | ||
2139 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2140 | { | |
2154a336 | 2141 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2142 | |
ca2b84cb | 2143 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
2144 | |
2145 | return ret; | |
81819f0f CL |
2146 | } |
2147 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2148 | ||
0f24f128 | 2149 | #ifdef CONFIG_TRACING |
4a92379b RK |
2150 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2151 | { | |
2152 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2153 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2154 | return ret; | |
2155 | } | |
2156 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2157 | ||
2158 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2159 | { |
4a92379b RK |
2160 | void *ret = kmalloc_order(size, flags, order); |
2161 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2162 | return ret; | |
5b882be4 | 2163 | } |
4a92379b | 2164 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2165 | #endif |
2166 | ||
81819f0f CL |
2167 | #ifdef CONFIG_NUMA |
2168 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2169 | { | |
5b882be4 EGM |
2170 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2171 | ||
ca2b84cb EGM |
2172 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2173 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2174 | |
2175 | return ret; | |
81819f0f CL |
2176 | } |
2177 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2178 | |
0f24f128 | 2179 | #ifdef CONFIG_TRACING |
4a92379b | 2180 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2181 | gfp_t gfpflags, |
4a92379b | 2182 | int node, size_t size) |
5b882be4 | 2183 | { |
4a92379b RK |
2184 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2185 | ||
2186 | trace_kmalloc_node(_RET_IP_, ret, | |
2187 | size, s->size, gfpflags, node); | |
2188 | return ret; | |
5b882be4 | 2189 | } |
4a92379b | 2190 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2191 | #endif |
5d1f57e4 | 2192 | #endif |
5b882be4 | 2193 | |
81819f0f | 2194 | /* |
894b8788 CL |
2195 | * Slow patch handling. This may still be called frequently since objects |
2196 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2197 | * |
894b8788 CL |
2198 | * So we still attempt to reduce cache line usage. Just take the slab |
2199 | * lock and free the item. If there is no additional partial page | |
2200 | * handling required then we can return immediately. | |
81819f0f | 2201 | */ |
894b8788 | 2202 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2203 | void *x, unsigned long addr) |
81819f0f CL |
2204 | { |
2205 | void *prior; | |
2206 | void **object = (void *)x; | |
2cfb7455 CL |
2207 | int was_frozen; |
2208 | int inuse; | |
2209 | struct page new; | |
2210 | unsigned long counters; | |
2211 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2212 | unsigned long uninitialized_var(flags); |
81819f0f | 2213 | |
8a5ec0ba | 2214 | stat(s, FREE_SLOWPATH); |
81819f0f | 2215 | |
8dc16c6c | 2216 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
80f08c19 | 2217 | return; |
6446faa2 | 2218 | |
2cfb7455 CL |
2219 | do { |
2220 | prior = page->freelist; | |
2221 | counters = page->counters; | |
2222 | set_freepointer(s, object, prior); | |
2223 | new.counters = counters; | |
2224 | was_frozen = new.frozen; | |
2225 | new.inuse--; | |
2226 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
2227 | n = get_node(s, page_to_nid(page)); | |
2228 | /* | |
2229 | * Speculatively acquire the list_lock. | |
2230 | * If the cmpxchg does not succeed then we may | |
2231 | * drop the list_lock without any processing. | |
2232 | * | |
2233 | * Otherwise the list_lock will synchronize with | |
2234 | * other processors updating the list of slabs. | |
2235 | */ | |
80f08c19 | 2236 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2237 | } |
2238 | inuse = new.inuse; | |
81819f0f | 2239 | |
2cfb7455 CL |
2240 | } while (!cmpxchg_double_slab(s, page, |
2241 | prior, counters, | |
2242 | object, new.counters, | |
2243 | "__slab_free")); | |
81819f0f | 2244 | |
2cfb7455 CL |
2245 | if (likely(!n)) { |
2246 | /* | |
2247 | * The list lock was not taken therefore no list | |
2248 | * activity can be necessary. | |
2249 | */ | |
2250 | if (was_frozen) | |
2251 | stat(s, FREE_FROZEN); | |
80f08c19 | 2252 | return; |
2cfb7455 | 2253 | } |
81819f0f CL |
2254 | |
2255 | /* | |
2cfb7455 CL |
2256 | * was_frozen may have been set after we acquired the list_lock in |
2257 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2258 | */ |
2cfb7455 CL |
2259 | if (was_frozen) |
2260 | stat(s, FREE_FROZEN); | |
2261 | else { | |
2262 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2263 | goto slab_empty; | |
5cc6eee8 | 2264 | |
2cfb7455 CL |
2265 | /* |
2266 | * Objects left in the slab. If it was not on the partial list before | |
2267 | * then add it. | |
2268 | */ | |
2269 | if (unlikely(!prior)) { | |
2270 | remove_full(s, page); | |
2271 | add_partial(n, page, 0); | |
2272 | stat(s, FREE_ADD_PARTIAL); | |
2273 | } | |
8ff12cfc | 2274 | } |
80f08c19 | 2275 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2276 | return; |
2277 | ||
2278 | slab_empty: | |
a973e9dd | 2279 | if (prior) { |
81819f0f | 2280 | /* |
672bba3a | 2281 | * Slab still on the partial list. |
81819f0f | 2282 | */ |
5cc6eee8 | 2283 | remove_partial(n, page); |
84e554e6 | 2284 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 2285 | } |
2cfb7455 | 2286 | |
80f08c19 | 2287 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2288 | stat(s, FREE_SLAB); |
81819f0f | 2289 | discard_slab(s, page); |
81819f0f CL |
2290 | } |
2291 | ||
894b8788 CL |
2292 | /* |
2293 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2294 | * can perform fastpath freeing without additional function calls. | |
2295 | * | |
2296 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2297 | * of this processor. This typically the case if we have just allocated | |
2298 | * the item before. | |
2299 | * | |
2300 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2301 | * with all sorts of special processing. | |
2302 | */ | |
06428780 | 2303 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2304 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2305 | { |
2306 | void **object = (void *)x; | |
dfb4f096 | 2307 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2308 | unsigned long tid; |
1f84260c | 2309 | |
c016b0bd CL |
2310 | slab_free_hook(s, x); |
2311 | ||
8a5ec0ba | 2312 | redo: |
a24c5a0e | 2313 | |
8a5ec0ba CL |
2314 | /* |
2315 | * Determine the currently cpus per cpu slab. | |
2316 | * The cpu may change afterward. However that does not matter since | |
2317 | * data is retrieved via this pointer. If we are on the same cpu | |
2318 | * during the cmpxchg then the free will succedd. | |
2319 | */ | |
9dfc6e68 | 2320 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2321 | |
8a5ec0ba CL |
2322 | tid = c->tid; |
2323 | barrier(); | |
c016b0bd | 2324 | |
442b06bc | 2325 | if (likely(page == c->page)) { |
ff12059e | 2326 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2327 | |
30106b8c | 2328 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2329 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2330 | c->freelist, tid, | |
2331 | object, next_tid(tid)))) { | |
2332 | ||
2333 | note_cmpxchg_failure("slab_free", s, tid); | |
2334 | goto redo; | |
2335 | } | |
84e554e6 | 2336 | stat(s, FREE_FASTPATH); |
894b8788 | 2337 | } else |
ff12059e | 2338 | __slab_free(s, page, x, addr); |
894b8788 | 2339 | |
894b8788 CL |
2340 | } |
2341 | ||
81819f0f CL |
2342 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2343 | { | |
77c5e2d0 | 2344 | struct page *page; |
81819f0f | 2345 | |
b49af68f | 2346 | page = virt_to_head_page(x); |
81819f0f | 2347 | |
ce71e27c | 2348 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2349 | |
ca2b84cb | 2350 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2351 | } |
2352 | EXPORT_SYMBOL(kmem_cache_free); | |
2353 | ||
81819f0f | 2354 | /* |
672bba3a CL |
2355 | * Object placement in a slab is made very easy because we always start at |
2356 | * offset 0. If we tune the size of the object to the alignment then we can | |
2357 | * get the required alignment by putting one properly sized object after | |
2358 | * another. | |
81819f0f CL |
2359 | * |
2360 | * Notice that the allocation order determines the sizes of the per cpu | |
2361 | * caches. Each processor has always one slab available for allocations. | |
2362 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2363 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2364 | * locking overhead. |
81819f0f CL |
2365 | */ |
2366 | ||
2367 | /* | |
2368 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2369 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2370 | * and increases the number of allocations possible without having to | |
2371 | * take the list_lock. | |
2372 | */ | |
2373 | static int slub_min_order; | |
114e9e89 | 2374 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2375 | static int slub_min_objects; |
81819f0f CL |
2376 | |
2377 | /* | |
2378 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2379 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2380 | */ |
2381 | static int slub_nomerge; | |
2382 | ||
81819f0f CL |
2383 | /* |
2384 | * Calculate the order of allocation given an slab object size. | |
2385 | * | |
672bba3a CL |
2386 | * The order of allocation has significant impact on performance and other |
2387 | * system components. Generally order 0 allocations should be preferred since | |
2388 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2389 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2390 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2391 | * would be wasted. |
2392 | * | |
2393 | * In order to reach satisfactory performance we must ensure that a minimum | |
2394 | * number of objects is in one slab. Otherwise we may generate too much | |
2395 | * activity on the partial lists which requires taking the list_lock. This is | |
2396 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2397 | * |
672bba3a CL |
2398 | * slub_max_order specifies the order where we begin to stop considering the |
2399 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2400 | * we try to keep the page order as low as possible. So we accept more waste | |
2401 | * of space in favor of a small page order. | |
81819f0f | 2402 | * |
672bba3a CL |
2403 | * Higher order allocations also allow the placement of more objects in a |
2404 | * slab and thereby reduce object handling overhead. If the user has | |
2405 | * requested a higher mininum order then we start with that one instead of | |
2406 | * the smallest order which will fit the object. | |
81819f0f | 2407 | */ |
5e6d444e | 2408 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2409 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2410 | { |
2411 | int order; | |
2412 | int rem; | |
6300ea75 | 2413 | int min_order = slub_min_order; |
81819f0f | 2414 | |
ab9a0f19 | 2415 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2416 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2417 | |
6300ea75 | 2418 | for (order = max(min_order, |
5e6d444e CL |
2419 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2420 | order <= max_order; order++) { | |
81819f0f | 2421 | |
5e6d444e | 2422 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2423 | |
ab9a0f19 | 2424 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2425 | continue; |
2426 | ||
ab9a0f19 | 2427 | rem = (slab_size - reserved) % size; |
81819f0f | 2428 | |
5e6d444e | 2429 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2430 | break; |
2431 | ||
2432 | } | |
672bba3a | 2433 | |
81819f0f CL |
2434 | return order; |
2435 | } | |
2436 | ||
ab9a0f19 | 2437 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2438 | { |
2439 | int order; | |
2440 | int min_objects; | |
2441 | int fraction; | |
e8120ff1 | 2442 | int max_objects; |
5e6d444e CL |
2443 | |
2444 | /* | |
2445 | * Attempt to find best configuration for a slab. This | |
2446 | * works by first attempting to generate a layout with | |
2447 | * the best configuration and backing off gradually. | |
2448 | * | |
2449 | * First we reduce the acceptable waste in a slab. Then | |
2450 | * we reduce the minimum objects required in a slab. | |
2451 | */ | |
2452 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2453 | if (!min_objects) |
2454 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2455 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2456 | min_objects = min(min_objects, max_objects); |
2457 | ||
5e6d444e | 2458 | while (min_objects > 1) { |
c124f5b5 | 2459 | fraction = 16; |
5e6d444e CL |
2460 | while (fraction >= 4) { |
2461 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2462 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2463 | if (order <= slub_max_order) |
2464 | return order; | |
2465 | fraction /= 2; | |
2466 | } | |
5086c389 | 2467 | min_objects--; |
5e6d444e CL |
2468 | } |
2469 | ||
2470 | /* | |
2471 | * We were unable to place multiple objects in a slab. Now | |
2472 | * lets see if we can place a single object there. | |
2473 | */ | |
ab9a0f19 | 2474 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2475 | if (order <= slub_max_order) |
2476 | return order; | |
2477 | ||
2478 | /* | |
2479 | * Doh this slab cannot be placed using slub_max_order. | |
2480 | */ | |
ab9a0f19 | 2481 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2482 | if (order < MAX_ORDER) |
5e6d444e CL |
2483 | return order; |
2484 | return -ENOSYS; | |
2485 | } | |
2486 | ||
81819f0f | 2487 | /* |
672bba3a | 2488 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2489 | */ |
2490 | static unsigned long calculate_alignment(unsigned long flags, | |
2491 | unsigned long align, unsigned long size) | |
2492 | { | |
2493 | /* | |
6446faa2 CL |
2494 | * If the user wants hardware cache aligned objects then follow that |
2495 | * suggestion if the object is sufficiently large. | |
81819f0f | 2496 | * |
6446faa2 CL |
2497 | * The hardware cache alignment cannot override the specified |
2498 | * alignment though. If that is greater then use it. | |
81819f0f | 2499 | */ |
b6210386 NP |
2500 | if (flags & SLAB_HWCACHE_ALIGN) { |
2501 | unsigned long ralign = cache_line_size(); | |
2502 | while (size <= ralign / 2) | |
2503 | ralign /= 2; | |
2504 | align = max(align, ralign); | |
2505 | } | |
81819f0f CL |
2506 | |
2507 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2508 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2509 | |
2510 | return ALIGN(align, sizeof(void *)); | |
2511 | } | |
2512 | ||
5595cffc PE |
2513 | static void |
2514 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2515 | { |
2516 | n->nr_partial = 0; | |
81819f0f CL |
2517 | spin_lock_init(&n->list_lock); |
2518 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2519 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2520 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2521 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2522 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2523 | #endif |
81819f0f CL |
2524 | } |
2525 | ||
55136592 | 2526 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2527 | { |
6c182dc0 CL |
2528 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2529 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2530 | |
8a5ec0ba | 2531 | /* |
d4d84fef CM |
2532 | * Must align to double word boundary for the double cmpxchg |
2533 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2534 | */ |
d4d84fef CM |
2535 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2536 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2537 | |
2538 | if (!s->cpu_slab) | |
2539 | return 0; | |
2540 | ||
2541 | init_kmem_cache_cpus(s); | |
4c93c355 | 2542 | |
8a5ec0ba | 2543 | return 1; |
4c93c355 | 2544 | } |
4c93c355 | 2545 | |
51df1142 CL |
2546 | static struct kmem_cache *kmem_cache_node; |
2547 | ||
81819f0f CL |
2548 | /* |
2549 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2550 | * slab on the node for this slabcache. There are no concurrent accesses | |
2551 | * possible. | |
2552 | * | |
2553 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2554 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2555 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2556 | */ |
55136592 | 2557 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2558 | { |
2559 | struct page *page; | |
2560 | struct kmem_cache_node *n; | |
2561 | ||
51df1142 | 2562 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2563 | |
51df1142 | 2564 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2565 | |
2566 | BUG_ON(!page); | |
a2f92ee7 CL |
2567 | if (page_to_nid(page) != node) { |
2568 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2569 | "node %d\n", node); | |
2570 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2571 | "in order to be able to continue\n"); | |
2572 | } | |
2573 | ||
81819f0f CL |
2574 | n = page->freelist; |
2575 | BUG_ON(!n); | |
51df1142 | 2576 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2577 | page->inuse++; |
8cb0a506 | 2578 | page->frozen = 0; |
51df1142 | 2579 | kmem_cache_node->node[node] = n; |
8ab1372f | 2580 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2581 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2582 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2583 | #endif |
51df1142 CL |
2584 | init_kmem_cache_node(n, kmem_cache_node); |
2585 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2586 | |
7c2e132c | 2587 | add_partial(n, page, 0); |
81819f0f CL |
2588 | } |
2589 | ||
2590 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2591 | { | |
2592 | int node; | |
2593 | ||
f64dc58c | 2594 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2595 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2596 | |
73367bd8 | 2597 | if (n) |
51df1142 CL |
2598 | kmem_cache_free(kmem_cache_node, n); |
2599 | ||
81819f0f CL |
2600 | s->node[node] = NULL; |
2601 | } | |
2602 | } | |
2603 | ||
55136592 | 2604 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2605 | { |
2606 | int node; | |
81819f0f | 2607 | |
f64dc58c | 2608 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2609 | struct kmem_cache_node *n; |
2610 | ||
73367bd8 | 2611 | if (slab_state == DOWN) { |
55136592 | 2612 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2613 | continue; |
2614 | } | |
51df1142 | 2615 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2616 | GFP_KERNEL, node); |
81819f0f | 2617 | |
73367bd8 AD |
2618 | if (!n) { |
2619 | free_kmem_cache_nodes(s); | |
2620 | return 0; | |
81819f0f | 2621 | } |
73367bd8 | 2622 | |
81819f0f | 2623 | s->node[node] = n; |
5595cffc | 2624 | init_kmem_cache_node(n, s); |
81819f0f CL |
2625 | } |
2626 | return 1; | |
2627 | } | |
81819f0f | 2628 | |
c0bdb232 | 2629 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2630 | { |
2631 | if (min < MIN_PARTIAL) | |
2632 | min = MIN_PARTIAL; | |
2633 | else if (min > MAX_PARTIAL) | |
2634 | min = MAX_PARTIAL; | |
2635 | s->min_partial = min; | |
2636 | } | |
2637 | ||
81819f0f CL |
2638 | /* |
2639 | * calculate_sizes() determines the order and the distribution of data within | |
2640 | * a slab object. | |
2641 | */ | |
06b285dc | 2642 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2643 | { |
2644 | unsigned long flags = s->flags; | |
2645 | unsigned long size = s->objsize; | |
2646 | unsigned long align = s->align; | |
834f3d11 | 2647 | int order; |
81819f0f | 2648 | |
d8b42bf5 CL |
2649 | /* |
2650 | * Round up object size to the next word boundary. We can only | |
2651 | * place the free pointer at word boundaries and this determines | |
2652 | * the possible location of the free pointer. | |
2653 | */ | |
2654 | size = ALIGN(size, sizeof(void *)); | |
2655 | ||
2656 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2657 | /* |
2658 | * Determine if we can poison the object itself. If the user of | |
2659 | * the slab may touch the object after free or before allocation | |
2660 | * then we should never poison the object itself. | |
2661 | */ | |
2662 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2663 | !s->ctor) |
81819f0f CL |
2664 | s->flags |= __OBJECT_POISON; |
2665 | else | |
2666 | s->flags &= ~__OBJECT_POISON; | |
2667 | ||
81819f0f CL |
2668 | |
2669 | /* | |
672bba3a | 2670 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2671 | * end of the object and the free pointer. If not then add an |
672bba3a | 2672 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2673 | */ |
2674 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2675 | size += sizeof(void *); | |
41ecc55b | 2676 | #endif |
81819f0f CL |
2677 | |
2678 | /* | |
672bba3a CL |
2679 | * With that we have determined the number of bytes in actual use |
2680 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2681 | */ |
2682 | s->inuse = size; | |
2683 | ||
2684 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2685 | s->ctor)) { |
81819f0f CL |
2686 | /* |
2687 | * Relocate free pointer after the object if it is not | |
2688 | * permitted to overwrite the first word of the object on | |
2689 | * kmem_cache_free. | |
2690 | * | |
2691 | * This is the case if we do RCU, have a constructor or | |
2692 | * destructor or are poisoning the objects. | |
2693 | */ | |
2694 | s->offset = size; | |
2695 | size += sizeof(void *); | |
2696 | } | |
2697 | ||
c12b3c62 | 2698 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2699 | if (flags & SLAB_STORE_USER) |
2700 | /* | |
2701 | * Need to store information about allocs and frees after | |
2702 | * the object. | |
2703 | */ | |
2704 | size += 2 * sizeof(struct track); | |
2705 | ||
be7b3fbc | 2706 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2707 | /* |
2708 | * Add some empty padding so that we can catch | |
2709 | * overwrites from earlier objects rather than let | |
2710 | * tracking information or the free pointer be | |
0211a9c8 | 2711 | * corrupted if a user writes before the start |
81819f0f CL |
2712 | * of the object. |
2713 | */ | |
2714 | size += sizeof(void *); | |
41ecc55b | 2715 | #endif |
672bba3a | 2716 | |
81819f0f CL |
2717 | /* |
2718 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2719 | * user specified and the dynamic determination of cache line size |
2720 | * on bootup. | |
81819f0f CL |
2721 | */ |
2722 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2723 | s->align = align; |
81819f0f CL |
2724 | |
2725 | /* | |
2726 | * SLUB stores one object immediately after another beginning from | |
2727 | * offset 0. In order to align the objects we have to simply size | |
2728 | * each object to conform to the alignment. | |
2729 | */ | |
2730 | size = ALIGN(size, align); | |
2731 | s->size = size; | |
06b285dc CL |
2732 | if (forced_order >= 0) |
2733 | order = forced_order; | |
2734 | else | |
ab9a0f19 | 2735 | order = calculate_order(size, s->reserved); |
81819f0f | 2736 | |
834f3d11 | 2737 | if (order < 0) |
81819f0f CL |
2738 | return 0; |
2739 | ||
b7a49f0d | 2740 | s->allocflags = 0; |
834f3d11 | 2741 | if (order) |
b7a49f0d CL |
2742 | s->allocflags |= __GFP_COMP; |
2743 | ||
2744 | if (s->flags & SLAB_CACHE_DMA) | |
2745 | s->allocflags |= SLUB_DMA; | |
2746 | ||
2747 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2748 | s->allocflags |= __GFP_RECLAIMABLE; | |
2749 | ||
81819f0f CL |
2750 | /* |
2751 | * Determine the number of objects per slab | |
2752 | */ | |
ab9a0f19 LJ |
2753 | s->oo = oo_make(order, size, s->reserved); |
2754 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2755 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2756 | s->max = s->oo; | |
81819f0f | 2757 | |
834f3d11 | 2758 | return !!oo_objects(s->oo); |
81819f0f CL |
2759 | |
2760 | } | |
2761 | ||
55136592 | 2762 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2763 | const char *name, size_t size, |
2764 | size_t align, unsigned long flags, | |
51cc5068 | 2765 | void (*ctor)(void *)) |
81819f0f CL |
2766 | { |
2767 | memset(s, 0, kmem_size); | |
2768 | s->name = name; | |
2769 | s->ctor = ctor; | |
81819f0f | 2770 | s->objsize = size; |
81819f0f | 2771 | s->align = align; |
ba0268a8 | 2772 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 2773 | s->reserved = 0; |
81819f0f | 2774 | |
da9a638c LJ |
2775 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
2776 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 2777 | |
06b285dc | 2778 | if (!calculate_sizes(s, -1)) |
81819f0f | 2779 | goto error; |
3de47213 DR |
2780 | if (disable_higher_order_debug) { |
2781 | /* | |
2782 | * Disable debugging flags that store metadata if the min slab | |
2783 | * order increased. | |
2784 | */ | |
2785 | if (get_order(s->size) > get_order(s->objsize)) { | |
2786 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2787 | s->offset = 0; | |
2788 | if (!calculate_sizes(s, -1)) | |
2789 | goto error; | |
2790 | } | |
2791 | } | |
81819f0f | 2792 | |
b789ef51 CL |
2793 | #ifdef CONFIG_CMPXCHG_DOUBLE |
2794 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) | |
2795 | /* Enable fast mode */ | |
2796 | s->flags |= __CMPXCHG_DOUBLE; | |
2797 | #endif | |
2798 | ||
3b89d7d8 DR |
2799 | /* |
2800 | * The larger the object size is, the more pages we want on the partial | |
2801 | * list to avoid pounding the page allocator excessively. | |
2802 | */ | |
c0bdb232 | 2803 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2804 | s->refcount = 1; |
2805 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2806 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2807 | #endif |
55136592 | 2808 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2809 | goto error; |
81819f0f | 2810 | |
55136592 | 2811 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2812 | return 1; |
ff12059e | 2813 | |
4c93c355 | 2814 | free_kmem_cache_nodes(s); |
81819f0f CL |
2815 | error: |
2816 | if (flags & SLAB_PANIC) | |
2817 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2818 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2819 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2820 | s->offset, flags); |
2821 | return 0; | |
2822 | } | |
81819f0f | 2823 | |
81819f0f CL |
2824 | /* |
2825 | * Determine the size of a slab object | |
2826 | */ | |
2827 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2828 | { | |
2829 | return s->objsize; | |
2830 | } | |
2831 | EXPORT_SYMBOL(kmem_cache_size); | |
2832 | ||
33b12c38 CL |
2833 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2834 | const char *text) | |
2835 | { | |
2836 | #ifdef CONFIG_SLUB_DEBUG | |
2837 | void *addr = page_address(page); | |
2838 | void *p; | |
a5dd5c11 NK |
2839 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2840 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2841 | if (!map) |
2842 | return; | |
33b12c38 CL |
2843 | slab_err(s, page, "%s", text); |
2844 | slab_lock(page); | |
33b12c38 | 2845 | |
5f80b13a | 2846 | get_map(s, page, map); |
33b12c38 CL |
2847 | for_each_object(p, s, addr, page->objects) { |
2848 | ||
2849 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2850 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2851 | p, p - addr); | |
2852 | print_tracking(s, p); | |
2853 | } | |
2854 | } | |
2855 | slab_unlock(page); | |
bbd7d57b | 2856 | kfree(map); |
33b12c38 CL |
2857 | #endif |
2858 | } | |
2859 | ||
81819f0f | 2860 | /* |
599870b1 | 2861 | * Attempt to free all partial slabs on a node. |
81819f0f | 2862 | */ |
599870b1 | 2863 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2864 | { |
81819f0f CL |
2865 | unsigned long flags; |
2866 | struct page *page, *h; | |
2867 | ||
2868 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2869 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2870 | if (!page->inuse) { |
5cc6eee8 | 2871 | remove_partial(n, page); |
81819f0f | 2872 | discard_slab(s, page); |
33b12c38 CL |
2873 | } else { |
2874 | list_slab_objects(s, page, | |
2875 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2876 | } |
33b12c38 | 2877 | } |
81819f0f | 2878 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2879 | } |
2880 | ||
2881 | /* | |
672bba3a | 2882 | * Release all resources used by a slab cache. |
81819f0f | 2883 | */ |
0c710013 | 2884 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2885 | { |
2886 | int node; | |
2887 | ||
2888 | flush_all(s); | |
9dfc6e68 | 2889 | free_percpu(s->cpu_slab); |
81819f0f | 2890 | /* Attempt to free all objects */ |
f64dc58c | 2891 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2892 | struct kmem_cache_node *n = get_node(s, node); |
2893 | ||
599870b1 CL |
2894 | free_partial(s, n); |
2895 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
2896 | return 1; |
2897 | } | |
2898 | free_kmem_cache_nodes(s); | |
2899 | return 0; | |
2900 | } | |
2901 | ||
2902 | /* | |
2903 | * Close a cache and release the kmem_cache structure | |
2904 | * (must be used for caches created using kmem_cache_create) | |
2905 | */ | |
2906 | void kmem_cache_destroy(struct kmem_cache *s) | |
2907 | { | |
2908 | down_write(&slub_lock); | |
2909 | s->refcount--; | |
2910 | if (!s->refcount) { | |
2911 | list_del(&s->list); | |
d629d819 PE |
2912 | if (kmem_cache_close(s)) { |
2913 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
2914 | "still has objects.\n", s->name, __func__); | |
2915 | dump_stack(); | |
2916 | } | |
d76b1590 ED |
2917 | if (s->flags & SLAB_DESTROY_BY_RCU) |
2918 | rcu_barrier(); | |
81819f0f | 2919 | sysfs_slab_remove(s); |
2bce6485 CL |
2920 | } |
2921 | up_write(&slub_lock); | |
81819f0f CL |
2922 | } |
2923 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2924 | ||
2925 | /******************************************************************** | |
2926 | * Kmalloc subsystem | |
2927 | *******************************************************************/ | |
2928 | ||
51df1142 | 2929 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
2930 | EXPORT_SYMBOL(kmalloc_caches); |
2931 | ||
51df1142 CL |
2932 | static struct kmem_cache *kmem_cache; |
2933 | ||
55136592 | 2934 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 2935 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
2936 | #endif |
2937 | ||
81819f0f CL |
2938 | static int __init setup_slub_min_order(char *str) |
2939 | { | |
06428780 | 2940 | get_option(&str, &slub_min_order); |
81819f0f CL |
2941 | |
2942 | return 1; | |
2943 | } | |
2944 | ||
2945 | __setup("slub_min_order=", setup_slub_min_order); | |
2946 | ||
2947 | static int __init setup_slub_max_order(char *str) | |
2948 | { | |
06428780 | 2949 | get_option(&str, &slub_max_order); |
818cf590 | 2950 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
2951 | |
2952 | return 1; | |
2953 | } | |
2954 | ||
2955 | __setup("slub_max_order=", setup_slub_max_order); | |
2956 | ||
2957 | static int __init setup_slub_min_objects(char *str) | |
2958 | { | |
06428780 | 2959 | get_option(&str, &slub_min_objects); |
81819f0f CL |
2960 | |
2961 | return 1; | |
2962 | } | |
2963 | ||
2964 | __setup("slub_min_objects=", setup_slub_min_objects); | |
2965 | ||
2966 | static int __init setup_slub_nomerge(char *str) | |
2967 | { | |
2968 | slub_nomerge = 1; | |
2969 | return 1; | |
2970 | } | |
2971 | ||
2972 | __setup("slub_nomerge", setup_slub_nomerge); | |
2973 | ||
51df1142 CL |
2974 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
2975 | int size, unsigned int flags) | |
81819f0f | 2976 | { |
51df1142 CL |
2977 | struct kmem_cache *s; |
2978 | ||
2979 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
2980 | ||
83b519e8 PE |
2981 | /* |
2982 | * This function is called with IRQs disabled during early-boot on | |
2983 | * single CPU so there's no need to take slub_lock here. | |
2984 | */ | |
55136592 | 2985 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 2986 | flags, NULL)) |
81819f0f CL |
2987 | goto panic; |
2988 | ||
2989 | list_add(&s->list, &slab_caches); | |
51df1142 | 2990 | return s; |
81819f0f CL |
2991 | |
2992 | panic: | |
2993 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 2994 | return NULL; |
81819f0f CL |
2995 | } |
2996 | ||
f1b26339 CL |
2997 | /* |
2998 | * Conversion table for small slabs sizes / 8 to the index in the | |
2999 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3000 | * of two cache sizes there. The size of larger slabs can be determined using | |
3001 | * fls. | |
3002 | */ | |
3003 | static s8 size_index[24] = { | |
3004 | 3, /* 8 */ | |
3005 | 4, /* 16 */ | |
3006 | 5, /* 24 */ | |
3007 | 5, /* 32 */ | |
3008 | 6, /* 40 */ | |
3009 | 6, /* 48 */ | |
3010 | 6, /* 56 */ | |
3011 | 6, /* 64 */ | |
3012 | 1, /* 72 */ | |
3013 | 1, /* 80 */ | |
3014 | 1, /* 88 */ | |
3015 | 1, /* 96 */ | |
3016 | 7, /* 104 */ | |
3017 | 7, /* 112 */ | |
3018 | 7, /* 120 */ | |
3019 | 7, /* 128 */ | |
3020 | 2, /* 136 */ | |
3021 | 2, /* 144 */ | |
3022 | 2, /* 152 */ | |
3023 | 2, /* 160 */ | |
3024 | 2, /* 168 */ | |
3025 | 2, /* 176 */ | |
3026 | 2, /* 184 */ | |
3027 | 2 /* 192 */ | |
3028 | }; | |
3029 | ||
acdfcd04 AK |
3030 | static inline int size_index_elem(size_t bytes) |
3031 | { | |
3032 | return (bytes - 1) / 8; | |
3033 | } | |
3034 | ||
81819f0f CL |
3035 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3036 | { | |
f1b26339 | 3037 | int index; |
81819f0f | 3038 | |
f1b26339 CL |
3039 | if (size <= 192) { |
3040 | if (!size) | |
3041 | return ZERO_SIZE_PTR; | |
81819f0f | 3042 | |
acdfcd04 | 3043 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3044 | } else |
f1b26339 | 3045 | index = fls(size - 1); |
81819f0f CL |
3046 | |
3047 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3048 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3049 | return kmalloc_dma_caches[index]; |
f1b26339 | 3050 | |
81819f0f | 3051 | #endif |
51df1142 | 3052 | return kmalloc_caches[index]; |
81819f0f CL |
3053 | } |
3054 | ||
3055 | void *__kmalloc(size_t size, gfp_t flags) | |
3056 | { | |
aadb4bc4 | 3057 | struct kmem_cache *s; |
5b882be4 | 3058 | void *ret; |
81819f0f | 3059 | |
ffadd4d0 | 3060 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3061 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3062 | |
3063 | s = get_slab(size, flags); | |
3064 | ||
3065 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3066 | return s; |
3067 | ||
2154a336 | 3068 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3069 | |
ca2b84cb | 3070 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3071 | |
3072 | return ret; | |
81819f0f CL |
3073 | } |
3074 | EXPORT_SYMBOL(__kmalloc); | |
3075 | ||
5d1f57e4 | 3076 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3077 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3078 | { | |
b1eeab67 | 3079 | struct page *page; |
e4f7c0b4 | 3080 | void *ptr = NULL; |
f619cfe1 | 3081 | |
b1eeab67 VN |
3082 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3083 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3084 | if (page) |
e4f7c0b4 CM |
3085 | ptr = page_address(page); |
3086 | ||
3087 | kmemleak_alloc(ptr, size, 1, flags); | |
3088 | return ptr; | |
f619cfe1 CL |
3089 | } |
3090 | ||
81819f0f CL |
3091 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3092 | { | |
aadb4bc4 | 3093 | struct kmem_cache *s; |
5b882be4 | 3094 | void *ret; |
81819f0f | 3095 | |
057685cf | 3096 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3097 | ret = kmalloc_large_node(size, flags, node); |
3098 | ||
ca2b84cb EGM |
3099 | trace_kmalloc_node(_RET_IP_, ret, |
3100 | size, PAGE_SIZE << get_order(size), | |
3101 | flags, node); | |
5b882be4 EGM |
3102 | |
3103 | return ret; | |
3104 | } | |
aadb4bc4 CL |
3105 | |
3106 | s = get_slab(size, flags); | |
3107 | ||
3108 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3109 | return s; |
3110 | ||
5b882be4 EGM |
3111 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3112 | ||
ca2b84cb | 3113 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3114 | |
3115 | return ret; | |
81819f0f CL |
3116 | } |
3117 | EXPORT_SYMBOL(__kmalloc_node); | |
3118 | #endif | |
3119 | ||
3120 | size_t ksize(const void *object) | |
3121 | { | |
272c1d21 | 3122 | struct page *page; |
81819f0f | 3123 | |
ef8b4520 | 3124 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3125 | return 0; |
3126 | ||
294a80a8 | 3127 | page = virt_to_head_page(object); |
294a80a8 | 3128 | |
76994412 PE |
3129 | if (unlikely(!PageSlab(page))) { |
3130 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3131 | return PAGE_SIZE << compound_order(page); |
76994412 | 3132 | } |
81819f0f | 3133 | |
b3d41885 | 3134 | return slab_ksize(page->slab); |
81819f0f | 3135 | } |
b1aabecd | 3136 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3137 | |
3138 | void kfree(const void *x) | |
3139 | { | |
81819f0f | 3140 | struct page *page; |
5bb983b0 | 3141 | void *object = (void *)x; |
81819f0f | 3142 | |
2121db74 PE |
3143 | trace_kfree(_RET_IP_, x); |
3144 | ||
2408c550 | 3145 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3146 | return; |
3147 | ||
b49af68f | 3148 | page = virt_to_head_page(x); |
aadb4bc4 | 3149 | if (unlikely(!PageSlab(page))) { |
0937502a | 3150 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3151 | kmemleak_free(x); |
aadb4bc4 CL |
3152 | put_page(page); |
3153 | return; | |
3154 | } | |
ce71e27c | 3155 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3156 | } |
3157 | EXPORT_SYMBOL(kfree); | |
3158 | ||
2086d26a | 3159 | /* |
672bba3a CL |
3160 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3161 | * the remaining slabs by the number of items in use. The slabs with the | |
3162 | * most items in use come first. New allocations will then fill those up | |
3163 | * and thus they can be removed from the partial lists. | |
3164 | * | |
3165 | * The slabs with the least items are placed last. This results in them | |
3166 | * being allocated from last increasing the chance that the last objects | |
3167 | * are freed in them. | |
2086d26a CL |
3168 | */ |
3169 | int kmem_cache_shrink(struct kmem_cache *s) | |
3170 | { | |
3171 | int node; | |
3172 | int i; | |
3173 | struct kmem_cache_node *n; | |
3174 | struct page *page; | |
3175 | struct page *t; | |
205ab99d | 3176 | int objects = oo_objects(s->max); |
2086d26a | 3177 | struct list_head *slabs_by_inuse = |
834f3d11 | 3178 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3179 | unsigned long flags; |
3180 | ||
3181 | if (!slabs_by_inuse) | |
3182 | return -ENOMEM; | |
3183 | ||
3184 | flush_all(s); | |
f64dc58c | 3185 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3186 | n = get_node(s, node); |
3187 | ||
3188 | if (!n->nr_partial) | |
3189 | continue; | |
3190 | ||
834f3d11 | 3191 | for (i = 0; i < objects; i++) |
2086d26a CL |
3192 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3193 | ||
3194 | spin_lock_irqsave(&n->list_lock, flags); | |
3195 | ||
3196 | /* | |
672bba3a | 3197 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3198 | * |
672bba3a CL |
3199 | * Note that concurrent frees may occur while we hold the |
3200 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3201 | */ |
3202 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
881db7fb | 3203 | if (!page->inuse) { |
5cc6eee8 | 3204 | remove_partial(n, page); |
2086d26a CL |
3205 | discard_slab(s, page); |
3206 | } else { | |
fcda3d89 CL |
3207 | list_move(&page->lru, |
3208 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
3209 | } |
3210 | } | |
3211 | ||
2086d26a | 3212 | /* |
672bba3a CL |
3213 | * Rebuild the partial list with the slabs filled up most |
3214 | * first and the least used slabs at the end. | |
2086d26a | 3215 | */ |
834f3d11 | 3216 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3217 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3218 | ||
2086d26a CL |
3219 | spin_unlock_irqrestore(&n->list_lock, flags); |
3220 | } | |
3221 | ||
3222 | kfree(slabs_by_inuse); | |
3223 | return 0; | |
3224 | } | |
3225 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3226 | ||
92a5bbc1 | 3227 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3228 | static int slab_mem_going_offline_callback(void *arg) |
3229 | { | |
3230 | struct kmem_cache *s; | |
3231 | ||
3232 | down_read(&slub_lock); | |
3233 | list_for_each_entry(s, &slab_caches, list) | |
3234 | kmem_cache_shrink(s); | |
3235 | up_read(&slub_lock); | |
3236 | ||
3237 | return 0; | |
3238 | } | |
3239 | ||
3240 | static void slab_mem_offline_callback(void *arg) | |
3241 | { | |
3242 | struct kmem_cache_node *n; | |
3243 | struct kmem_cache *s; | |
3244 | struct memory_notify *marg = arg; | |
3245 | int offline_node; | |
3246 | ||
3247 | offline_node = marg->status_change_nid; | |
3248 | ||
3249 | /* | |
3250 | * If the node still has available memory. we need kmem_cache_node | |
3251 | * for it yet. | |
3252 | */ | |
3253 | if (offline_node < 0) | |
3254 | return; | |
3255 | ||
3256 | down_read(&slub_lock); | |
3257 | list_for_each_entry(s, &slab_caches, list) { | |
3258 | n = get_node(s, offline_node); | |
3259 | if (n) { | |
3260 | /* | |
3261 | * if n->nr_slabs > 0, slabs still exist on the node | |
3262 | * that is going down. We were unable to free them, | |
c9404c9c | 3263 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3264 | * callback. So, we must fail. |
3265 | */ | |
0f389ec6 | 3266 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3267 | |
3268 | s->node[offline_node] = NULL; | |
8de66a0c | 3269 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3270 | } |
3271 | } | |
3272 | up_read(&slub_lock); | |
3273 | } | |
3274 | ||
3275 | static int slab_mem_going_online_callback(void *arg) | |
3276 | { | |
3277 | struct kmem_cache_node *n; | |
3278 | struct kmem_cache *s; | |
3279 | struct memory_notify *marg = arg; | |
3280 | int nid = marg->status_change_nid; | |
3281 | int ret = 0; | |
3282 | ||
3283 | /* | |
3284 | * If the node's memory is already available, then kmem_cache_node is | |
3285 | * already created. Nothing to do. | |
3286 | */ | |
3287 | if (nid < 0) | |
3288 | return 0; | |
3289 | ||
3290 | /* | |
0121c619 | 3291 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3292 | * allocate a kmem_cache_node structure in order to bring the node |
3293 | * online. | |
3294 | */ | |
3295 | down_read(&slub_lock); | |
3296 | list_for_each_entry(s, &slab_caches, list) { | |
3297 | /* | |
3298 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3299 | * since memory is not yet available from the node that | |
3300 | * is brought up. | |
3301 | */ | |
8de66a0c | 3302 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3303 | if (!n) { |
3304 | ret = -ENOMEM; | |
3305 | goto out; | |
3306 | } | |
5595cffc | 3307 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3308 | s->node[nid] = n; |
3309 | } | |
3310 | out: | |
3311 | up_read(&slub_lock); | |
3312 | return ret; | |
3313 | } | |
3314 | ||
3315 | static int slab_memory_callback(struct notifier_block *self, | |
3316 | unsigned long action, void *arg) | |
3317 | { | |
3318 | int ret = 0; | |
3319 | ||
3320 | switch (action) { | |
3321 | case MEM_GOING_ONLINE: | |
3322 | ret = slab_mem_going_online_callback(arg); | |
3323 | break; | |
3324 | case MEM_GOING_OFFLINE: | |
3325 | ret = slab_mem_going_offline_callback(arg); | |
3326 | break; | |
3327 | case MEM_OFFLINE: | |
3328 | case MEM_CANCEL_ONLINE: | |
3329 | slab_mem_offline_callback(arg); | |
3330 | break; | |
3331 | case MEM_ONLINE: | |
3332 | case MEM_CANCEL_OFFLINE: | |
3333 | break; | |
3334 | } | |
dc19f9db KH |
3335 | if (ret) |
3336 | ret = notifier_from_errno(ret); | |
3337 | else | |
3338 | ret = NOTIFY_OK; | |
b9049e23 YG |
3339 | return ret; |
3340 | } | |
3341 | ||
3342 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3343 | ||
81819f0f CL |
3344 | /******************************************************************** |
3345 | * Basic setup of slabs | |
3346 | *******************************************************************/ | |
3347 | ||
51df1142 CL |
3348 | /* |
3349 | * Used for early kmem_cache structures that were allocated using | |
3350 | * the page allocator | |
3351 | */ | |
3352 | ||
3353 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3354 | { | |
3355 | int node; | |
3356 | ||
3357 | list_add(&s->list, &slab_caches); | |
3358 | s->refcount = -1; | |
3359 | ||
3360 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3361 | struct kmem_cache_node *n = get_node(s, node); | |
3362 | struct page *p; | |
3363 | ||
3364 | if (n) { | |
3365 | list_for_each_entry(p, &n->partial, lru) | |
3366 | p->slab = s; | |
3367 | ||
607bf324 | 3368 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3369 | list_for_each_entry(p, &n->full, lru) |
3370 | p->slab = s; | |
3371 | #endif | |
3372 | } | |
3373 | } | |
3374 | } | |
3375 | ||
81819f0f CL |
3376 | void __init kmem_cache_init(void) |
3377 | { | |
3378 | int i; | |
4b356be0 | 3379 | int caches = 0; |
51df1142 CL |
3380 | struct kmem_cache *temp_kmem_cache; |
3381 | int order; | |
51df1142 CL |
3382 | struct kmem_cache *temp_kmem_cache_node; |
3383 | unsigned long kmalloc_size; | |
3384 | ||
3385 | kmem_size = offsetof(struct kmem_cache, node) + | |
3386 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3387 | ||
3388 | /* Allocate two kmem_caches from the page allocator */ | |
3389 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3390 | order = get_order(2 * kmalloc_size); | |
3391 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3392 | ||
81819f0f CL |
3393 | /* |
3394 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3395 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3396 | * kmem_cache_open for slab_state == DOWN. |
3397 | */ | |
51df1142 CL |
3398 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3399 | ||
3400 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3401 | sizeof(struct kmem_cache_node), | |
3402 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3403 | |
0c40ba4f | 3404 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3405 | |
3406 | /* Able to allocate the per node structures */ | |
3407 | slab_state = PARTIAL; | |
3408 | ||
51df1142 CL |
3409 | temp_kmem_cache = kmem_cache; |
3410 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3411 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3412 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3413 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3414 | |
51df1142 CL |
3415 | /* |
3416 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3417 | * kmem_cache_node is separately allocated so no need to | |
3418 | * update any list pointers. | |
3419 | */ | |
3420 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3421 | |
51df1142 CL |
3422 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3423 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3424 | ||
3425 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3426 | ||
3427 | caches++; | |
51df1142 CL |
3428 | kmem_cache_bootstrap_fixup(kmem_cache); |
3429 | caches++; | |
3430 | /* Free temporary boot structure */ | |
3431 | free_pages((unsigned long)temp_kmem_cache, order); | |
3432 | ||
3433 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3434 | |
3435 | /* | |
3436 | * Patch up the size_index table if we have strange large alignment | |
3437 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3438 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3439 | * |
3440 | * Largest permitted alignment is 256 bytes due to the way we | |
3441 | * handle the index determination for the smaller caches. | |
3442 | * | |
3443 | * Make sure that nothing crazy happens if someone starts tinkering | |
3444 | * around with ARCH_KMALLOC_MINALIGN | |
3445 | */ | |
3446 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3447 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3448 | ||
acdfcd04 AK |
3449 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3450 | int elem = size_index_elem(i); | |
3451 | if (elem >= ARRAY_SIZE(size_index)) | |
3452 | break; | |
3453 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3454 | } | |
f1b26339 | 3455 | |
acdfcd04 AK |
3456 | if (KMALLOC_MIN_SIZE == 64) { |
3457 | /* | |
3458 | * The 96 byte size cache is not used if the alignment | |
3459 | * is 64 byte. | |
3460 | */ | |
3461 | for (i = 64 + 8; i <= 96; i += 8) | |
3462 | size_index[size_index_elem(i)] = 7; | |
3463 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3464 | /* |
3465 | * The 192 byte sized cache is not used if the alignment | |
3466 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3467 | * instead. | |
3468 | */ | |
3469 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3470 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3471 | } |
3472 | ||
51df1142 CL |
3473 | /* Caches that are not of the two-to-the-power-of size */ |
3474 | if (KMALLOC_MIN_SIZE <= 32) { | |
3475 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3476 | caches++; | |
3477 | } | |
3478 | ||
3479 | if (KMALLOC_MIN_SIZE <= 64) { | |
3480 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3481 | caches++; | |
3482 | } | |
3483 | ||
3484 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3485 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3486 | caches++; | |
3487 | } | |
3488 | ||
81819f0f CL |
3489 | slab_state = UP; |
3490 | ||
3491 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3492 | if (KMALLOC_MIN_SIZE <= 32) { |
3493 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3494 | BUG_ON(!kmalloc_caches[1]->name); | |
3495 | } | |
3496 | ||
3497 | if (KMALLOC_MIN_SIZE <= 64) { | |
3498 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3499 | BUG_ON(!kmalloc_caches[2]->name); | |
3500 | } | |
3501 | ||
d7278bd7 CL |
3502 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3503 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3504 | ||
3505 | BUG_ON(!s); | |
51df1142 | 3506 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3507 | } |
81819f0f CL |
3508 | |
3509 | #ifdef CONFIG_SMP | |
3510 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3511 | #endif |
81819f0f | 3512 | |
55136592 | 3513 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3514 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3515 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3516 | |
51df1142 | 3517 | if (s && s->size) { |
55136592 CL |
3518 | char *name = kasprintf(GFP_NOWAIT, |
3519 | "dma-kmalloc-%d", s->objsize); | |
3520 | ||
3521 | BUG_ON(!name); | |
51df1142 CL |
3522 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3523 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3524 | } |
3525 | } | |
3526 | #endif | |
3adbefee IM |
3527 | printk(KERN_INFO |
3528 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3529 | " CPUs=%d, Nodes=%d\n", |
3530 | caches, cache_line_size(), | |
81819f0f CL |
3531 | slub_min_order, slub_max_order, slub_min_objects, |
3532 | nr_cpu_ids, nr_node_ids); | |
3533 | } | |
3534 | ||
7e85ee0c PE |
3535 | void __init kmem_cache_init_late(void) |
3536 | { | |
7e85ee0c PE |
3537 | } |
3538 | ||
81819f0f CL |
3539 | /* |
3540 | * Find a mergeable slab cache | |
3541 | */ | |
3542 | static int slab_unmergeable(struct kmem_cache *s) | |
3543 | { | |
3544 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3545 | return 1; | |
3546 | ||
c59def9f | 3547 | if (s->ctor) |
81819f0f CL |
3548 | return 1; |
3549 | ||
8ffa6875 CL |
3550 | /* |
3551 | * We may have set a slab to be unmergeable during bootstrap. | |
3552 | */ | |
3553 | if (s->refcount < 0) | |
3554 | return 1; | |
3555 | ||
81819f0f CL |
3556 | return 0; |
3557 | } | |
3558 | ||
3559 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3560 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3561 | void (*ctor)(void *)) |
81819f0f | 3562 | { |
5b95a4ac | 3563 | struct kmem_cache *s; |
81819f0f CL |
3564 | |
3565 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3566 | return NULL; | |
3567 | ||
c59def9f | 3568 | if (ctor) |
81819f0f CL |
3569 | return NULL; |
3570 | ||
3571 | size = ALIGN(size, sizeof(void *)); | |
3572 | align = calculate_alignment(flags, align, size); | |
3573 | size = ALIGN(size, align); | |
ba0268a8 | 3574 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3575 | |
5b95a4ac | 3576 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3577 | if (slab_unmergeable(s)) |
3578 | continue; | |
3579 | ||
3580 | if (size > s->size) | |
3581 | continue; | |
3582 | ||
ba0268a8 | 3583 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3584 | continue; |
3585 | /* | |
3586 | * Check if alignment is compatible. | |
3587 | * Courtesy of Adrian Drzewiecki | |
3588 | */ | |
06428780 | 3589 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3590 | continue; |
3591 | ||
3592 | if (s->size - size >= sizeof(void *)) | |
3593 | continue; | |
3594 | ||
3595 | return s; | |
3596 | } | |
3597 | return NULL; | |
3598 | } | |
3599 | ||
3600 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3601 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3602 | { |
3603 | struct kmem_cache *s; | |
84c1cf62 | 3604 | char *n; |
81819f0f | 3605 | |
fe1ff49d BH |
3606 | if (WARN_ON(!name)) |
3607 | return NULL; | |
3608 | ||
81819f0f | 3609 | down_write(&slub_lock); |
ba0268a8 | 3610 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3611 | if (s) { |
3612 | s->refcount++; | |
3613 | /* | |
3614 | * Adjust the object sizes so that we clear | |
3615 | * the complete object on kzalloc. | |
3616 | */ | |
3617 | s->objsize = max(s->objsize, (int)size); | |
3618 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3619 | |
7b8f3b66 | 3620 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3621 | s->refcount--; |
81819f0f | 3622 | goto err; |
7b8f3b66 | 3623 | } |
2bce6485 | 3624 | up_write(&slub_lock); |
a0e1d1be CL |
3625 | return s; |
3626 | } | |
6446faa2 | 3627 | |
84c1cf62 PE |
3628 | n = kstrdup(name, GFP_KERNEL); |
3629 | if (!n) | |
3630 | goto err; | |
3631 | ||
a0e1d1be CL |
3632 | s = kmalloc(kmem_size, GFP_KERNEL); |
3633 | if (s) { | |
84c1cf62 | 3634 | if (kmem_cache_open(s, n, |
c59def9f | 3635 | size, align, flags, ctor)) { |
81819f0f | 3636 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3637 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3638 | list_del(&s->list); |
84c1cf62 | 3639 | kfree(n); |
7b8f3b66 | 3640 | kfree(s); |
a0e1d1be | 3641 | goto err; |
7b8f3b66 | 3642 | } |
2bce6485 | 3643 | up_write(&slub_lock); |
a0e1d1be CL |
3644 | return s; |
3645 | } | |
84c1cf62 | 3646 | kfree(n); |
a0e1d1be | 3647 | kfree(s); |
81819f0f | 3648 | } |
68cee4f1 | 3649 | err: |
81819f0f | 3650 | up_write(&slub_lock); |
81819f0f | 3651 | |
81819f0f CL |
3652 | if (flags & SLAB_PANIC) |
3653 | panic("Cannot create slabcache %s\n", name); | |
3654 | else | |
3655 | s = NULL; | |
3656 | return s; | |
3657 | } | |
3658 | EXPORT_SYMBOL(kmem_cache_create); | |
3659 | ||
81819f0f | 3660 | #ifdef CONFIG_SMP |
81819f0f | 3661 | /* |
672bba3a CL |
3662 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3663 | * necessary. | |
81819f0f CL |
3664 | */ |
3665 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3666 | unsigned long action, void *hcpu) | |
3667 | { | |
3668 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3669 | struct kmem_cache *s; |
3670 | unsigned long flags; | |
81819f0f CL |
3671 | |
3672 | switch (action) { | |
3673 | case CPU_UP_CANCELED: | |
8bb78442 | 3674 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3675 | case CPU_DEAD: |
8bb78442 | 3676 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3677 | down_read(&slub_lock); |
3678 | list_for_each_entry(s, &slab_caches, list) { | |
3679 | local_irq_save(flags); | |
3680 | __flush_cpu_slab(s, cpu); | |
3681 | local_irq_restore(flags); | |
3682 | } | |
3683 | up_read(&slub_lock); | |
81819f0f CL |
3684 | break; |
3685 | default: | |
3686 | break; | |
3687 | } | |
3688 | return NOTIFY_OK; | |
3689 | } | |
3690 | ||
06428780 | 3691 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3692 | .notifier_call = slab_cpuup_callback |
06428780 | 3693 | }; |
81819f0f CL |
3694 | |
3695 | #endif | |
3696 | ||
ce71e27c | 3697 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3698 | { |
aadb4bc4 | 3699 | struct kmem_cache *s; |
94b528d0 | 3700 | void *ret; |
aadb4bc4 | 3701 | |
ffadd4d0 | 3702 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3703 | return kmalloc_large(size, gfpflags); |
3704 | ||
aadb4bc4 | 3705 | s = get_slab(size, gfpflags); |
81819f0f | 3706 | |
2408c550 | 3707 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3708 | return s; |
81819f0f | 3709 | |
2154a336 | 3710 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 3711 | |
25985edc | 3712 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3713 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3714 | |
3715 | return ret; | |
81819f0f CL |
3716 | } |
3717 | ||
5d1f57e4 | 3718 | #ifdef CONFIG_NUMA |
81819f0f | 3719 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3720 | int node, unsigned long caller) |
81819f0f | 3721 | { |
aadb4bc4 | 3722 | struct kmem_cache *s; |
94b528d0 | 3723 | void *ret; |
aadb4bc4 | 3724 | |
d3e14aa3 XF |
3725 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3726 | ret = kmalloc_large_node(size, gfpflags, node); | |
3727 | ||
3728 | trace_kmalloc_node(caller, ret, | |
3729 | size, PAGE_SIZE << get_order(size), | |
3730 | gfpflags, node); | |
3731 | ||
3732 | return ret; | |
3733 | } | |
eada35ef | 3734 | |
aadb4bc4 | 3735 | s = get_slab(size, gfpflags); |
81819f0f | 3736 | |
2408c550 | 3737 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3738 | return s; |
81819f0f | 3739 | |
94b528d0 EGM |
3740 | ret = slab_alloc(s, gfpflags, node, caller); |
3741 | ||
25985edc | 3742 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3743 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3744 | |
3745 | return ret; | |
81819f0f | 3746 | } |
5d1f57e4 | 3747 | #endif |
81819f0f | 3748 | |
ab4d5ed5 | 3749 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3750 | static int count_inuse(struct page *page) |
3751 | { | |
3752 | return page->inuse; | |
3753 | } | |
3754 | ||
3755 | static int count_total(struct page *page) | |
3756 | { | |
3757 | return page->objects; | |
3758 | } | |
ab4d5ed5 | 3759 | #endif |
205ab99d | 3760 | |
ab4d5ed5 | 3761 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3762 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3763 | unsigned long *map) | |
53e15af0 CL |
3764 | { |
3765 | void *p; | |
a973e9dd | 3766 | void *addr = page_address(page); |
53e15af0 CL |
3767 | |
3768 | if (!check_slab(s, page) || | |
3769 | !on_freelist(s, page, NULL)) | |
3770 | return 0; | |
3771 | ||
3772 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3773 | bitmap_zero(map, page->objects); |
53e15af0 | 3774 | |
5f80b13a CL |
3775 | get_map(s, page, map); |
3776 | for_each_object(p, s, addr, page->objects) { | |
3777 | if (test_bit(slab_index(p, s, addr), map)) | |
3778 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3779 | return 0; | |
53e15af0 CL |
3780 | } |
3781 | ||
224a88be | 3782 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3783 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3784 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3785 | return 0; |
3786 | return 1; | |
3787 | } | |
3788 | ||
434e245d CL |
3789 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3790 | unsigned long *map) | |
53e15af0 | 3791 | { |
881db7fb CL |
3792 | slab_lock(page); |
3793 | validate_slab(s, page, map); | |
3794 | slab_unlock(page); | |
53e15af0 CL |
3795 | } |
3796 | ||
434e245d CL |
3797 | static int validate_slab_node(struct kmem_cache *s, |
3798 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3799 | { |
3800 | unsigned long count = 0; | |
3801 | struct page *page; | |
3802 | unsigned long flags; | |
3803 | ||
3804 | spin_lock_irqsave(&n->list_lock, flags); | |
3805 | ||
3806 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3807 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3808 | count++; |
3809 | } | |
3810 | if (count != n->nr_partial) | |
3811 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3812 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3813 | ||
3814 | if (!(s->flags & SLAB_STORE_USER)) | |
3815 | goto out; | |
3816 | ||
3817 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3818 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3819 | count++; |
3820 | } | |
3821 | if (count != atomic_long_read(&n->nr_slabs)) | |
3822 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3823 | "counter=%ld\n", s->name, count, | |
3824 | atomic_long_read(&n->nr_slabs)); | |
3825 | ||
3826 | out: | |
3827 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3828 | return count; | |
3829 | } | |
3830 | ||
434e245d | 3831 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3832 | { |
3833 | int node; | |
3834 | unsigned long count = 0; | |
205ab99d | 3835 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3836 | sizeof(unsigned long), GFP_KERNEL); |
3837 | ||
3838 | if (!map) | |
3839 | return -ENOMEM; | |
53e15af0 CL |
3840 | |
3841 | flush_all(s); | |
f64dc58c | 3842 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3843 | struct kmem_cache_node *n = get_node(s, node); |
3844 | ||
434e245d | 3845 | count += validate_slab_node(s, n, map); |
53e15af0 | 3846 | } |
434e245d | 3847 | kfree(map); |
53e15af0 CL |
3848 | return count; |
3849 | } | |
88a420e4 | 3850 | /* |
672bba3a | 3851 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3852 | * and freed. |
3853 | */ | |
3854 | ||
3855 | struct location { | |
3856 | unsigned long count; | |
ce71e27c | 3857 | unsigned long addr; |
45edfa58 CL |
3858 | long long sum_time; |
3859 | long min_time; | |
3860 | long max_time; | |
3861 | long min_pid; | |
3862 | long max_pid; | |
174596a0 | 3863 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3864 | nodemask_t nodes; |
88a420e4 CL |
3865 | }; |
3866 | ||
3867 | struct loc_track { | |
3868 | unsigned long max; | |
3869 | unsigned long count; | |
3870 | struct location *loc; | |
3871 | }; | |
3872 | ||
3873 | static void free_loc_track(struct loc_track *t) | |
3874 | { | |
3875 | if (t->max) | |
3876 | free_pages((unsigned long)t->loc, | |
3877 | get_order(sizeof(struct location) * t->max)); | |
3878 | } | |
3879 | ||
68dff6a9 | 3880 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3881 | { |
3882 | struct location *l; | |
3883 | int order; | |
3884 | ||
88a420e4 CL |
3885 | order = get_order(sizeof(struct location) * max); |
3886 | ||
68dff6a9 | 3887 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3888 | if (!l) |
3889 | return 0; | |
3890 | ||
3891 | if (t->count) { | |
3892 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3893 | free_loc_track(t); | |
3894 | } | |
3895 | t->max = max; | |
3896 | t->loc = l; | |
3897 | return 1; | |
3898 | } | |
3899 | ||
3900 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3901 | const struct track *track) |
88a420e4 CL |
3902 | { |
3903 | long start, end, pos; | |
3904 | struct location *l; | |
ce71e27c | 3905 | unsigned long caddr; |
45edfa58 | 3906 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
3907 | |
3908 | start = -1; | |
3909 | end = t->count; | |
3910 | ||
3911 | for ( ; ; ) { | |
3912 | pos = start + (end - start + 1) / 2; | |
3913 | ||
3914 | /* | |
3915 | * There is nothing at "end". If we end up there | |
3916 | * we need to add something to before end. | |
3917 | */ | |
3918 | if (pos == end) | |
3919 | break; | |
3920 | ||
3921 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
3922 | if (track->addr == caddr) { |
3923 | ||
3924 | l = &t->loc[pos]; | |
3925 | l->count++; | |
3926 | if (track->when) { | |
3927 | l->sum_time += age; | |
3928 | if (age < l->min_time) | |
3929 | l->min_time = age; | |
3930 | if (age > l->max_time) | |
3931 | l->max_time = age; | |
3932 | ||
3933 | if (track->pid < l->min_pid) | |
3934 | l->min_pid = track->pid; | |
3935 | if (track->pid > l->max_pid) | |
3936 | l->max_pid = track->pid; | |
3937 | ||
174596a0 RR |
3938 | cpumask_set_cpu(track->cpu, |
3939 | to_cpumask(l->cpus)); | |
45edfa58 CL |
3940 | } |
3941 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3942 | return 1; |
3943 | } | |
3944 | ||
45edfa58 | 3945 | if (track->addr < caddr) |
88a420e4 CL |
3946 | end = pos; |
3947 | else | |
3948 | start = pos; | |
3949 | } | |
3950 | ||
3951 | /* | |
672bba3a | 3952 | * Not found. Insert new tracking element. |
88a420e4 | 3953 | */ |
68dff6a9 | 3954 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
3955 | return 0; |
3956 | ||
3957 | l = t->loc + pos; | |
3958 | if (pos < t->count) | |
3959 | memmove(l + 1, l, | |
3960 | (t->count - pos) * sizeof(struct location)); | |
3961 | t->count++; | |
3962 | l->count = 1; | |
45edfa58 CL |
3963 | l->addr = track->addr; |
3964 | l->sum_time = age; | |
3965 | l->min_time = age; | |
3966 | l->max_time = age; | |
3967 | l->min_pid = track->pid; | |
3968 | l->max_pid = track->pid; | |
174596a0 RR |
3969 | cpumask_clear(to_cpumask(l->cpus)); |
3970 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
3971 | nodes_clear(l->nodes); |
3972 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
3973 | return 1; |
3974 | } | |
3975 | ||
3976 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 3977 | struct page *page, enum track_item alloc, |
a5dd5c11 | 3978 | unsigned long *map) |
88a420e4 | 3979 | { |
a973e9dd | 3980 | void *addr = page_address(page); |
88a420e4 CL |
3981 | void *p; |
3982 | ||
39b26464 | 3983 | bitmap_zero(map, page->objects); |
5f80b13a | 3984 | get_map(s, page, map); |
88a420e4 | 3985 | |
224a88be | 3986 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
3987 | if (!test_bit(slab_index(p, s, addr), map)) |
3988 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
3989 | } |
3990 | ||
3991 | static int list_locations(struct kmem_cache *s, char *buf, | |
3992 | enum track_item alloc) | |
3993 | { | |
e374d483 | 3994 | int len = 0; |
88a420e4 | 3995 | unsigned long i; |
68dff6a9 | 3996 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 3997 | int node; |
bbd7d57b ED |
3998 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
3999 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4000 | |
bbd7d57b ED |
4001 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4002 | GFP_TEMPORARY)) { | |
4003 | kfree(map); | |
68dff6a9 | 4004 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4005 | } |
88a420e4 CL |
4006 | /* Push back cpu slabs */ |
4007 | flush_all(s); | |
4008 | ||
f64dc58c | 4009 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4010 | struct kmem_cache_node *n = get_node(s, node); |
4011 | unsigned long flags; | |
4012 | struct page *page; | |
4013 | ||
9e86943b | 4014 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4015 | continue; |
4016 | ||
4017 | spin_lock_irqsave(&n->list_lock, flags); | |
4018 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4019 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4020 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4021 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4022 | spin_unlock_irqrestore(&n->list_lock, flags); |
4023 | } | |
4024 | ||
4025 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4026 | struct location *l = &t.loc[i]; |
88a420e4 | 4027 | |
9c246247 | 4028 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4029 | break; |
e374d483 | 4030 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4031 | |
4032 | if (l->addr) | |
62c70bce | 4033 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4034 | else |
e374d483 | 4035 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4036 | |
4037 | if (l->sum_time != l->min_time) { | |
e374d483 | 4038 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4039 | l->min_time, |
4040 | (long)div_u64(l->sum_time, l->count), | |
4041 | l->max_time); | |
45edfa58 | 4042 | } else |
e374d483 | 4043 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4044 | l->min_time); |
4045 | ||
4046 | if (l->min_pid != l->max_pid) | |
e374d483 | 4047 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4048 | l->min_pid, l->max_pid); |
4049 | else | |
e374d483 | 4050 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4051 | l->min_pid); |
4052 | ||
174596a0 RR |
4053 | if (num_online_cpus() > 1 && |
4054 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4055 | len < PAGE_SIZE - 60) { |
4056 | len += sprintf(buf + len, " cpus="); | |
4057 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4058 | to_cpumask(l->cpus)); |
45edfa58 CL |
4059 | } |
4060 | ||
62bc62a8 | 4061 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4062 | len < PAGE_SIZE - 60) { |
4063 | len += sprintf(buf + len, " nodes="); | |
4064 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4065 | l->nodes); |
4066 | } | |
4067 | ||
e374d483 | 4068 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4069 | } |
4070 | ||
4071 | free_loc_track(&t); | |
bbd7d57b | 4072 | kfree(map); |
88a420e4 | 4073 | if (!t.count) |
e374d483 HH |
4074 | len += sprintf(buf, "No data\n"); |
4075 | return len; | |
88a420e4 | 4076 | } |
ab4d5ed5 | 4077 | #endif |
88a420e4 | 4078 | |
a5a84755 CL |
4079 | #ifdef SLUB_RESILIENCY_TEST |
4080 | static void resiliency_test(void) | |
4081 | { | |
4082 | u8 *p; | |
4083 | ||
4084 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4085 | ||
4086 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4087 | printk(KERN_ERR "-----------------------\n"); | |
4088 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4089 | ||
4090 | p = kzalloc(16, GFP_KERNEL); | |
4091 | p[16] = 0x12; | |
4092 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4093 | " 0x12->0x%p\n\n", p + 16); | |
4094 | ||
4095 | validate_slab_cache(kmalloc_caches[4]); | |
4096 | ||
4097 | /* Hmmm... The next two are dangerous */ | |
4098 | p = kzalloc(32, GFP_KERNEL); | |
4099 | p[32 + sizeof(void *)] = 0x34; | |
4100 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4101 | " 0x34 -> -0x%p\n", p); | |
4102 | printk(KERN_ERR | |
4103 | "If allocated object is overwritten then not detectable\n\n"); | |
4104 | ||
4105 | validate_slab_cache(kmalloc_caches[5]); | |
4106 | p = kzalloc(64, GFP_KERNEL); | |
4107 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4108 | *p = 0x56; | |
4109 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4110 | p); | |
4111 | printk(KERN_ERR | |
4112 | "If allocated object is overwritten then not detectable\n\n"); | |
4113 | validate_slab_cache(kmalloc_caches[6]); | |
4114 | ||
4115 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4116 | p = kzalloc(128, GFP_KERNEL); | |
4117 | kfree(p); | |
4118 | *p = 0x78; | |
4119 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4120 | validate_slab_cache(kmalloc_caches[7]); | |
4121 | ||
4122 | p = kzalloc(256, GFP_KERNEL); | |
4123 | kfree(p); | |
4124 | p[50] = 0x9a; | |
4125 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4126 | p); | |
4127 | validate_slab_cache(kmalloc_caches[8]); | |
4128 | ||
4129 | p = kzalloc(512, GFP_KERNEL); | |
4130 | kfree(p); | |
4131 | p[512] = 0xab; | |
4132 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4133 | validate_slab_cache(kmalloc_caches[9]); | |
4134 | } | |
4135 | #else | |
4136 | #ifdef CONFIG_SYSFS | |
4137 | static void resiliency_test(void) {}; | |
4138 | #endif | |
4139 | #endif | |
4140 | ||
ab4d5ed5 | 4141 | #ifdef CONFIG_SYSFS |
81819f0f | 4142 | enum slab_stat_type { |
205ab99d CL |
4143 | SL_ALL, /* All slabs */ |
4144 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4145 | SL_CPU, /* Only slabs used for cpu caches */ | |
4146 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4147 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4148 | }; |
4149 | ||
205ab99d | 4150 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4151 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4152 | #define SO_CPU (1 << SL_CPU) | |
4153 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4154 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4155 | |
62e5c4b4 CG |
4156 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4157 | char *buf, unsigned long flags) | |
81819f0f CL |
4158 | { |
4159 | unsigned long total = 0; | |
81819f0f CL |
4160 | int node; |
4161 | int x; | |
4162 | unsigned long *nodes; | |
4163 | unsigned long *per_cpu; | |
4164 | ||
4165 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4166 | if (!nodes) |
4167 | return -ENOMEM; | |
81819f0f CL |
4168 | per_cpu = nodes + nr_node_ids; |
4169 | ||
205ab99d CL |
4170 | if (flags & SO_CPU) { |
4171 | int cpu; | |
81819f0f | 4172 | |
205ab99d | 4173 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4174 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 4175 | |
205ab99d CL |
4176 | if (!c || c->node < 0) |
4177 | continue; | |
4178 | ||
4179 | if (c->page) { | |
4180 | if (flags & SO_TOTAL) | |
4181 | x = c->page->objects; | |
4182 | else if (flags & SO_OBJECTS) | |
4183 | x = c->page->inuse; | |
81819f0f CL |
4184 | else |
4185 | x = 1; | |
205ab99d | 4186 | |
81819f0f | 4187 | total += x; |
205ab99d | 4188 | nodes[c->node] += x; |
81819f0f | 4189 | } |
205ab99d | 4190 | per_cpu[c->node]++; |
81819f0f CL |
4191 | } |
4192 | } | |
4193 | ||
04d94879 | 4194 | lock_memory_hotplug(); |
ab4d5ed5 | 4195 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4196 | if (flags & SO_ALL) { |
4197 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4198 | struct kmem_cache_node *n = get_node(s, node); | |
4199 | ||
4200 | if (flags & SO_TOTAL) | |
4201 | x = atomic_long_read(&n->total_objects); | |
4202 | else if (flags & SO_OBJECTS) | |
4203 | x = atomic_long_read(&n->total_objects) - | |
4204 | count_partial(n, count_free); | |
81819f0f | 4205 | |
81819f0f | 4206 | else |
205ab99d | 4207 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4208 | total += x; |
4209 | nodes[node] += x; | |
4210 | } | |
4211 | ||
ab4d5ed5 CL |
4212 | } else |
4213 | #endif | |
4214 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4215 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4216 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4217 | |
205ab99d CL |
4218 | if (flags & SO_TOTAL) |
4219 | x = count_partial(n, count_total); | |
4220 | else if (flags & SO_OBJECTS) | |
4221 | x = count_partial(n, count_inuse); | |
81819f0f | 4222 | else |
205ab99d | 4223 | x = n->nr_partial; |
81819f0f CL |
4224 | total += x; |
4225 | nodes[node] += x; | |
4226 | } | |
4227 | } | |
81819f0f CL |
4228 | x = sprintf(buf, "%lu", total); |
4229 | #ifdef CONFIG_NUMA | |
f64dc58c | 4230 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4231 | if (nodes[node]) |
4232 | x += sprintf(buf + x, " N%d=%lu", | |
4233 | node, nodes[node]); | |
4234 | #endif | |
04d94879 | 4235 | unlock_memory_hotplug(); |
81819f0f CL |
4236 | kfree(nodes); |
4237 | return x + sprintf(buf + x, "\n"); | |
4238 | } | |
4239 | ||
ab4d5ed5 | 4240 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4241 | static int any_slab_objects(struct kmem_cache *s) |
4242 | { | |
4243 | int node; | |
81819f0f | 4244 | |
dfb4f096 | 4245 | for_each_online_node(node) { |
81819f0f CL |
4246 | struct kmem_cache_node *n = get_node(s, node); |
4247 | ||
dfb4f096 CL |
4248 | if (!n) |
4249 | continue; | |
4250 | ||
4ea33e2d | 4251 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4252 | return 1; |
4253 | } | |
4254 | return 0; | |
4255 | } | |
ab4d5ed5 | 4256 | #endif |
81819f0f CL |
4257 | |
4258 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
4259 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
4260 | ||
4261 | struct slab_attribute { | |
4262 | struct attribute attr; | |
4263 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4264 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4265 | }; | |
4266 | ||
4267 | #define SLAB_ATTR_RO(_name) \ | |
4268 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4269 | ||
4270 | #define SLAB_ATTR(_name) \ | |
4271 | static struct slab_attribute _name##_attr = \ | |
4272 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4273 | ||
81819f0f CL |
4274 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4275 | { | |
4276 | return sprintf(buf, "%d\n", s->size); | |
4277 | } | |
4278 | SLAB_ATTR_RO(slab_size); | |
4279 | ||
4280 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4281 | { | |
4282 | return sprintf(buf, "%d\n", s->align); | |
4283 | } | |
4284 | SLAB_ATTR_RO(align); | |
4285 | ||
4286 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4287 | { | |
4288 | return sprintf(buf, "%d\n", s->objsize); | |
4289 | } | |
4290 | SLAB_ATTR_RO(object_size); | |
4291 | ||
4292 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4293 | { | |
834f3d11 | 4294 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4295 | } |
4296 | SLAB_ATTR_RO(objs_per_slab); | |
4297 | ||
06b285dc CL |
4298 | static ssize_t order_store(struct kmem_cache *s, |
4299 | const char *buf, size_t length) | |
4300 | { | |
0121c619 CL |
4301 | unsigned long order; |
4302 | int err; | |
4303 | ||
4304 | err = strict_strtoul(buf, 10, &order); | |
4305 | if (err) | |
4306 | return err; | |
06b285dc CL |
4307 | |
4308 | if (order > slub_max_order || order < slub_min_order) | |
4309 | return -EINVAL; | |
4310 | ||
4311 | calculate_sizes(s, order); | |
4312 | return length; | |
4313 | } | |
4314 | ||
81819f0f CL |
4315 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4316 | { | |
834f3d11 | 4317 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4318 | } |
06b285dc | 4319 | SLAB_ATTR(order); |
81819f0f | 4320 | |
73d342b1 DR |
4321 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4322 | { | |
4323 | return sprintf(buf, "%lu\n", s->min_partial); | |
4324 | } | |
4325 | ||
4326 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4327 | size_t length) | |
4328 | { | |
4329 | unsigned long min; | |
4330 | int err; | |
4331 | ||
4332 | err = strict_strtoul(buf, 10, &min); | |
4333 | if (err) | |
4334 | return err; | |
4335 | ||
c0bdb232 | 4336 | set_min_partial(s, min); |
73d342b1 DR |
4337 | return length; |
4338 | } | |
4339 | SLAB_ATTR(min_partial); | |
4340 | ||
81819f0f CL |
4341 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4342 | { | |
62c70bce JP |
4343 | if (!s->ctor) |
4344 | return 0; | |
4345 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4346 | } |
4347 | SLAB_ATTR_RO(ctor); | |
4348 | ||
81819f0f CL |
4349 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4350 | { | |
4351 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4352 | } | |
4353 | SLAB_ATTR_RO(aliases); | |
4354 | ||
81819f0f CL |
4355 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4356 | { | |
d9acf4b7 | 4357 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4358 | } |
4359 | SLAB_ATTR_RO(partial); | |
4360 | ||
4361 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4362 | { | |
d9acf4b7 | 4363 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4364 | } |
4365 | SLAB_ATTR_RO(cpu_slabs); | |
4366 | ||
4367 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4368 | { | |
205ab99d | 4369 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4370 | } |
4371 | SLAB_ATTR_RO(objects); | |
4372 | ||
205ab99d CL |
4373 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4374 | { | |
4375 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4376 | } | |
4377 | SLAB_ATTR_RO(objects_partial); | |
4378 | ||
a5a84755 CL |
4379 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4380 | { | |
4381 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4382 | } | |
4383 | ||
4384 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4385 | const char *buf, size_t length) | |
4386 | { | |
4387 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4388 | if (buf[0] == '1') | |
4389 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4390 | return length; | |
4391 | } | |
4392 | SLAB_ATTR(reclaim_account); | |
4393 | ||
4394 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4395 | { | |
4396 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4397 | } | |
4398 | SLAB_ATTR_RO(hwcache_align); | |
4399 | ||
4400 | #ifdef CONFIG_ZONE_DMA | |
4401 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4402 | { | |
4403 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4404 | } | |
4405 | SLAB_ATTR_RO(cache_dma); | |
4406 | #endif | |
4407 | ||
4408 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4409 | { | |
4410 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4411 | } | |
4412 | SLAB_ATTR_RO(destroy_by_rcu); | |
4413 | ||
ab9a0f19 LJ |
4414 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4415 | { | |
4416 | return sprintf(buf, "%d\n", s->reserved); | |
4417 | } | |
4418 | SLAB_ATTR_RO(reserved); | |
4419 | ||
ab4d5ed5 | 4420 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4421 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4422 | { | |
4423 | return show_slab_objects(s, buf, SO_ALL); | |
4424 | } | |
4425 | SLAB_ATTR_RO(slabs); | |
4426 | ||
205ab99d CL |
4427 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4428 | { | |
4429 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4430 | } | |
4431 | SLAB_ATTR_RO(total_objects); | |
4432 | ||
81819f0f CL |
4433 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4434 | { | |
4435 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4436 | } | |
4437 | ||
4438 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4439 | const char *buf, size_t length) | |
4440 | { | |
4441 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4442 | if (buf[0] == '1') { |
4443 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4444 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4445 | } |
81819f0f CL |
4446 | return length; |
4447 | } | |
4448 | SLAB_ATTR(sanity_checks); | |
4449 | ||
4450 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4451 | { | |
4452 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4453 | } | |
4454 | ||
4455 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4456 | size_t length) | |
4457 | { | |
4458 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4459 | if (buf[0] == '1') { |
4460 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4461 | s->flags |= SLAB_TRACE; |
b789ef51 | 4462 | } |
81819f0f CL |
4463 | return length; |
4464 | } | |
4465 | SLAB_ATTR(trace); | |
4466 | ||
81819f0f CL |
4467 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4468 | { | |
4469 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4470 | } | |
4471 | ||
4472 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4473 | const char *buf, size_t length) | |
4474 | { | |
4475 | if (any_slab_objects(s)) | |
4476 | return -EBUSY; | |
4477 | ||
4478 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4479 | if (buf[0] == '1') { |
4480 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4481 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4482 | } |
06b285dc | 4483 | calculate_sizes(s, -1); |
81819f0f CL |
4484 | return length; |
4485 | } | |
4486 | SLAB_ATTR(red_zone); | |
4487 | ||
4488 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4489 | { | |
4490 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4491 | } | |
4492 | ||
4493 | static ssize_t poison_store(struct kmem_cache *s, | |
4494 | const char *buf, size_t length) | |
4495 | { | |
4496 | if (any_slab_objects(s)) | |
4497 | return -EBUSY; | |
4498 | ||
4499 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4500 | if (buf[0] == '1') { |
4501 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4502 | s->flags |= SLAB_POISON; |
b789ef51 | 4503 | } |
06b285dc | 4504 | calculate_sizes(s, -1); |
81819f0f CL |
4505 | return length; |
4506 | } | |
4507 | SLAB_ATTR(poison); | |
4508 | ||
4509 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4510 | { | |
4511 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4512 | } | |
4513 | ||
4514 | static ssize_t store_user_store(struct kmem_cache *s, | |
4515 | const char *buf, size_t length) | |
4516 | { | |
4517 | if (any_slab_objects(s)) | |
4518 | return -EBUSY; | |
4519 | ||
4520 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4521 | if (buf[0] == '1') { |
4522 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4523 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4524 | } |
06b285dc | 4525 | calculate_sizes(s, -1); |
81819f0f CL |
4526 | return length; |
4527 | } | |
4528 | SLAB_ATTR(store_user); | |
4529 | ||
53e15af0 CL |
4530 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4531 | { | |
4532 | return 0; | |
4533 | } | |
4534 | ||
4535 | static ssize_t validate_store(struct kmem_cache *s, | |
4536 | const char *buf, size_t length) | |
4537 | { | |
434e245d CL |
4538 | int ret = -EINVAL; |
4539 | ||
4540 | if (buf[0] == '1') { | |
4541 | ret = validate_slab_cache(s); | |
4542 | if (ret >= 0) | |
4543 | ret = length; | |
4544 | } | |
4545 | return ret; | |
53e15af0 CL |
4546 | } |
4547 | SLAB_ATTR(validate); | |
a5a84755 CL |
4548 | |
4549 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4550 | { | |
4551 | if (!(s->flags & SLAB_STORE_USER)) | |
4552 | return -ENOSYS; | |
4553 | return list_locations(s, buf, TRACK_ALLOC); | |
4554 | } | |
4555 | SLAB_ATTR_RO(alloc_calls); | |
4556 | ||
4557 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4558 | { | |
4559 | if (!(s->flags & SLAB_STORE_USER)) | |
4560 | return -ENOSYS; | |
4561 | return list_locations(s, buf, TRACK_FREE); | |
4562 | } | |
4563 | SLAB_ATTR_RO(free_calls); | |
4564 | #endif /* CONFIG_SLUB_DEBUG */ | |
4565 | ||
4566 | #ifdef CONFIG_FAILSLAB | |
4567 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4568 | { | |
4569 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4570 | } | |
4571 | ||
4572 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4573 | size_t length) | |
4574 | { | |
4575 | s->flags &= ~SLAB_FAILSLAB; | |
4576 | if (buf[0] == '1') | |
4577 | s->flags |= SLAB_FAILSLAB; | |
4578 | return length; | |
4579 | } | |
4580 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4581 | #endif |
53e15af0 | 4582 | |
2086d26a CL |
4583 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4584 | { | |
4585 | return 0; | |
4586 | } | |
4587 | ||
4588 | static ssize_t shrink_store(struct kmem_cache *s, | |
4589 | const char *buf, size_t length) | |
4590 | { | |
4591 | if (buf[0] == '1') { | |
4592 | int rc = kmem_cache_shrink(s); | |
4593 | ||
4594 | if (rc) | |
4595 | return rc; | |
4596 | } else | |
4597 | return -EINVAL; | |
4598 | return length; | |
4599 | } | |
4600 | SLAB_ATTR(shrink); | |
4601 | ||
81819f0f | 4602 | #ifdef CONFIG_NUMA |
9824601e | 4603 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4604 | { |
9824601e | 4605 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4606 | } |
4607 | ||
9824601e | 4608 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4609 | const char *buf, size_t length) |
4610 | { | |
0121c619 CL |
4611 | unsigned long ratio; |
4612 | int err; | |
4613 | ||
4614 | err = strict_strtoul(buf, 10, &ratio); | |
4615 | if (err) | |
4616 | return err; | |
4617 | ||
e2cb96b7 | 4618 | if (ratio <= 100) |
0121c619 | 4619 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4620 | |
81819f0f CL |
4621 | return length; |
4622 | } | |
9824601e | 4623 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4624 | #endif |
4625 | ||
8ff12cfc | 4626 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4627 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4628 | { | |
4629 | unsigned long sum = 0; | |
4630 | int cpu; | |
4631 | int len; | |
4632 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4633 | ||
4634 | if (!data) | |
4635 | return -ENOMEM; | |
4636 | ||
4637 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4638 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4639 | |
4640 | data[cpu] = x; | |
4641 | sum += x; | |
4642 | } | |
4643 | ||
4644 | len = sprintf(buf, "%lu", sum); | |
4645 | ||
50ef37b9 | 4646 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4647 | for_each_online_cpu(cpu) { |
4648 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4649 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4650 | } |
50ef37b9 | 4651 | #endif |
8ff12cfc CL |
4652 | kfree(data); |
4653 | return len + sprintf(buf + len, "\n"); | |
4654 | } | |
4655 | ||
78eb00cc DR |
4656 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4657 | { | |
4658 | int cpu; | |
4659 | ||
4660 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4661 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4662 | } |
4663 | ||
8ff12cfc CL |
4664 | #define STAT_ATTR(si, text) \ |
4665 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4666 | { \ | |
4667 | return show_stat(s, buf, si); \ | |
4668 | } \ | |
78eb00cc DR |
4669 | static ssize_t text##_store(struct kmem_cache *s, \ |
4670 | const char *buf, size_t length) \ | |
4671 | { \ | |
4672 | if (buf[0] != '0') \ | |
4673 | return -EINVAL; \ | |
4674 | clear_stat(s, si); \ | |
4675 | return length; \ | |
4676 | } \ | |
4677 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4678 | |
4679 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4680 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4681 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4682 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4683 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4684 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4685 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4686 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4687 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4688 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4689 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4690 | STAT_ATTR(FREE_SLAB, free_slab); |
4691 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4692 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4693 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4694 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4695 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4696 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4697 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4698 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4699 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4700 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
8ff12cfc CL |
4701 | #endif |
4702 | ||
06428780 | 4703 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4704 | &slab_size_attr.attr, |
4705 | &object_size_attr.attr, | |
4706 | &objs_per_slab_attr.attr, | |
4707 | &order_attr.attr, | |
73d342b1 | 4708 | &min_partial_attr.attr, |
81819f0f | 4709 | &objects_attr.attr, |
205ab99d | 4710 | &objects_partial_attr.attr, |
81819f0f CL |
4711 | &partial_attr.attr, |
4712 | &cpu_slabs_attr.attr, | |
4713 | &ctor_attr.attr, | |
81819f0f CL |
4714 | &aliases_attr.attr, |
4715 | &align_attr.attr, | |
81819f0f CL |
4716 | &hwcache_align_attr.attr, |
4717 | &reclaim_account_attr.attr, | |
4718 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4719 | &shrink_attr.attr, |
ab9a0f19 | 4720 | &reserved_attr.attr, |
ab4d5ed5 | 4721 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4722 | &total_objects_attr.attr, |
4723 | &slabs_attr.attr, | |
4724 | &sanity_checks_attr.attr, | |
4725 | &trace_attr.attr, | |
81819f0f CL |
4726 | &red_zone_attr.attr, |
4727 | &poison_attr.attr, | |
4728 | &store_user_attr.attr, | |
53e15af0 | 4729 | &validate_attr.attr, |
88a420e4 CL |
4730 | &alloc_calls_attr.attr, |
4731 | &free_calls_attr.attr, | |
ab4d5ed5 | 4732 | #endif |
81819f0f CL |
4733 | #ifdef CONFIG_ZONE_DMA |
4734 | &cache_dma_attr.attr, | |
4735 | #endif | |
4736 | #ifdef CONFIG_NUMA | |
9824601e | 4737 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4738 | #endif |
4739 | #ifdef CONFIG_SLUB_STATS | |
4740 | &alloc_fastpath_attr.attr, | |
4741 | &alloc_slowpath_attr.attr, | |
4742 | &free_fastpath_attr.attr, | |
4743 | &free_slowpath_attr.attr, | |
4744 | &free_frozen_attr.attr, | |
4745 | &free_add_partial_attr.attr, | |
4746 | &free_remove_partial_attr.attr, | |
4747 | &alloc_from_partial_attr.attr, | |
4748 | &alloc_slab_attr.attr, | |
4749 | &alloc_refill_attr.attr, | |
e36a2652 | 4750 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4751 | &free_slab_attr.attr, |
4752 | &cpuslab_flush_attr.attr, | |
4753 | &deactivate_full_attr.attr, | |
4754 | &deactivate_empty_attr.attr, | |
4755 | &deactivate_to_head_attr.attr, | |
4756 | &deactivate_to_tail_attr.attr, | |
4757 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4758 | &deactivate_bypass_attr.attr, |
65c3376a | 4759 | &order_fallback_attr.attr, |
b789ef51 CL |
4760 | &cmpxchg_double_fail_attr.attr, |
4761 | &cmpxchg_double_cpu_fail_attr.attr, | |
81819f0f | 4762 | #endif |
4c13dd3b DM |
4763 | #ifdef CONFIG_FAILSLAB |
4764 | &failslab_attr.attr, | |
4765 | #endif | |
4766 | ||
81819f0f CL |
4767 | NULL |
4768 | }; | |
4769 | ||
4770 | static struct attribute_group slab_attr_group = { | |
4771 | .attrs = slab_attrs, | |
4772 | }; | |
4773 | ||
4774 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4775 | struct attribute *attr, | |
4776 | char *buf) | |
4777 | { | |
4778 | struct slab_attribute *attribute; | |
4779 | struct kmem_cache *s; | |
4780 | int err; | |
4781 | ||
4782 | attribute = to_slab_attr(attr); | |
4783 | s = to_slab(kobj); | |
4784 | ||
4785 | if (!attribute->show) | |
4786 | return -EIO; | |
4787 | ||
4788 | err = attribute->show(s, buf); | |
4789 | ||
4790 | return err; | |
4791 | } | |
4792 | ||
4793 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4794 | struct attribute *attr, | |
4795 | const char *buf, size_t len) | |
4796 | { | |
4797 | struct slab_attribute *attribute; | |
4798 | struct kmem_cache *s; | |
4799 | int err; | |
4800 | ||
4801 | attribute = to_slab_attr(attr); | |
4802 | s = to_slab(kobj); | |
4803 | ||
4804 | if (!attribute->store) | |
4805 | return -EIO; | |
4806 | ||
4807 | err = attribute->store(s, buf, len); | |
4808 | ||
4809 | return err; | |
4810 | } | |
4811 | ||
151c602f CL |
4812 | static void kmem_cache_release(struct kobject *kobj) |
4813 | { | |
4814 | struct kmem_cache *s = to_slab(kobj); | |
4815 | ||
84c1cf62 | 4816 | kfree(s->name); |
151c602f CL |
4817 | kfree(s); |
4818 | } | |
4819 | ||
52cf25d0 | 4820 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4821 | .show = slab_attr_show, |
4822 | .store = slab_attr_store, | |
4823 | }; | |
4824 | ||
4825 | static struct kobj_type slab_ktype = { | |
4826 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4827 | .release = kmem_cache_release |
81819f0f CL |
4828 | }; |
4829 | ||
4830 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4831 | { | |
4832 | struct kobj_type *ktype = get_ktype(kobj); | |
4833 | ||
4834 | if (ktype == &slab_ktype) | |
4835 | return 1; | |
4836 | return 0; | |
4837 | } | |
4838 | ||
9cd43611 | 4839 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4840 | .filter = uevent_filter, |
4841 | }; | |
4842 | ||
27c3a314 | 4843 | static struct kset *slab_kset; |
81819f0f CL |
4844 | |
4845 | #define ID_STR_LENGTH 64 | |
4846 | ||
4847 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4848 | * |
4849 | * Format :[flags-]size | |
81819f0f CL |
4850 | */ |
4851 | static char *create_unique_id(struct kmem_cache *s) | |
4852 | { | |
4853 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4854 | char *p = name; | |
4855 | ||
4856 | BUG_ON(!name); | |
4857 | ||
4858 | *p++ = ':'; | |
4859 | /* | |
4860 | * First flags affecting slabcache operations. We will only | |
4861 | * get here for aliasable slabs so we do not need to support | |
4862 | * too many flags. The flags here must cover all flags that | |
4863 | * are matched during merging to guarantee that the id is | |
4864 | * unique. | |
4865 | */ | |
4866 | if (s->flags & SLAB_CACHE_DMA) | |
4867 | *p++ = 'd'; | |
4868 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
4869 | *p++ = 'a'; | |
4870 | if (s->flags & SLAB_DEBUG_FREE) | |
4871 | *p++ = 'F'; | |
5a896d9e VN |
4872 | if (!(s->flags & SLAB_NOTRACK)) |
4873 | *p++ = 't'; | |
81819f0f CL |
4874 | if (p != name + 1) |
4875 | *p++ = '-'; | |
4876 | p += sprintf(p, "%07d", s->size); | |
4877 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
4878 | return name; | |
4879 | } | |
4880 | ||
4881 | static int sysfs_slab_add(struct kmem_cache *s) | |
4882 | { | |
4883 | int err; | |
4884 | const char *name; | |
4885 | int unmergeable; | |
4886 | ||
4887 | if (slab_state < SYSFS) | |
4888 | /* Defer until later */ | |
4889 | return 0; | |
4890 | ||
4891 | unmergeable = slab_unmergeable(s); | |
4892 | if (unmergeable) { | |
4893 | /* | |
4894 | * Slabcache can never be merged so we can use the name proper. | |
4895 | * This is typically the case for debug situations. In that | |
4896 | * case we can catch duplicate names easily. | |
4897 | */ | |
27c3a314 | 4898 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
4899 | name = s->name; |
4900 | } else { | |
4901 | /* | |
4902 | * Create a unique name for the slab as a target | |
4903 | * for the symlinks. | |
4904 | */ | |
4905 | name = create_unique_id(s); | |
4906 | } | |
4907 | ||
27c3a314 | 4908 | s->kobj.kset = slab_kset; |
1eada11c GKH |
4909 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
4910 | if (err) { | |
4911 | kobject_put(&s->kobj); | |
81819f0f | 4912 | return err; |
1eada11c | 4913 | } |
81819f0f CL |
4914 | |
4915 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
4916 | if (err) { |
4917 | kobject_del(&s->kobj); | |
4918 | kobject_put(&s->kobj); | |
81819f0f | 4919 | return err; |
5788d8ad | 4920 | } |
81819f0f CL |
4921 | kobject_uevent(&s->kobj, KOBJ_ADD); |
4922 | if (!unmergeable) { | |
4923 | /* Setup first alias */ | |
4924 | sysfs_slab_alias(s, s->name); | |
4925 | kfree(name); | |
4926 | } | |
4927 | return 0; | |
4928 | } | |
4929 | ||
4930 | static void sysfs_slab_remove(struct kmem_cache *s) | |
4931 | { | |
2bce6485 CL |
4932 | if (slab_state < SYSFS) |
4933 | /* | |
4934 | * Sysfs has not been setup yet so no need to remove the | |
4935 | * cache from sysfs. | |
4936 | */ | |
4937 | return; | |
4938 | ||
81819f0f CL |
4939 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
4940 | kobject_del(&s->kobj); | |
151c602f | 4941 | kobject_put(&s->kobj); |
81819f0f CL |
4942 | } |
4943 | ||
4944 | /* | |
4945 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 4946 | * available lest we lose that information. |
81819f0f CL |
4947 | */ |
4948 | struct saved_alias { | |
4949 | struct kmem_cache *s; | |
4950 | const char *name; | |
4951 | struct saved_alias *next; | |
4952 | }; | |
4953 | ||
5af328a5 | 4954 | static struct saved_alias *alias_list; |
81819f0f CL |
4955 | |
4956 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
4957 | { | |
4958 | struct saved_alias *al; | |
4959 | ||
4960 | if (slab_state == SYSFS) { | |
4961 | /* | |
4962 | * If we have a leftover link then remove it. | |
4963 | */ | |
27c3a314 GKH |
4964 | sysfs_remove_link(&slab_kset->kobj, name); |
4965 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
4966 | } |
4967 | ||
4968 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
4969 | if (!al) | |
4970 | return -ENOMEM; | |
4971 | ||
4972 | al->s = s; | |
4973 | al->name = name; | |
4974 | al->next = alias_list; | |
4975 | alias_list = al; | |
4976 | return 0; | |
4977 | } | |
4978 | ||
4979 | static int __init slab_sysfs_init(void) | |
4980 | { | |
5b95a4ac | 4981 | struct kmem_cache *s; |
81819f0f CL |
4982 | int err; |
4983 | ||
2bce6485 CL |
4984 | down_write(&slub_lock); |
4985 | ||
0ff21e46 | 4986 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 4987 | if (!slab_kset) { |
2bce6485 | 4988 | up_write(&slub_lock); |
81819f0f CL |
4989 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
4990 | return -ENOSYS; | |
4991 | } | |
4992 | ||
26a7bd03 CL |
4993 | slab_state = SYSFS; |
4994 | ||
5b95a4ac | 4995 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 4996 | err = sysfs_slab_add(s); |
5d540fb7 CL |
4997 | if (err) |
4998 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
4999 | " to sysfs\n", s->name); | |
26a7bd03 | 5000 | } |
81819f0f CL |
5001 | |
5002 | while (alias_list) { | |
5003 | struct saved_alias *al = alias_list; | |
5004 | ||
5005 | alias_list = alias_list->next; | |
5006 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5007 | if (err) |
5008 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
5009 | " %s to sysfs\n", s->name); | |
81819f0f CL |
5010 | kfree(al); |
5011 | } | |
5012 | ||
2bce6485 | 5013 | up_write(&slub_lock); |
81819f0f CL |
5014 | resiliency_test(); |
5015 | return 0; | |
5016 | } | |
5017 | ||
5018 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5019 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5020 | |
5021 | /* | |
5022 | * The /proc/slabinfo ABI | |
5023 | */ | |
158a9624 | 5024 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5025 | static void print_slabinfo_header(struct seq_file *m) |
5026 | { | |
5027 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
5028 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
5029 | "<objperslab> <pagesperslab>"); | |
5030 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5031 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5032 | seq_putc(m, '\n'); | |
5033 | } | |
5034 | ||
5035 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5036 | { | |
5037 | loff_t n = *pos; | |
5038 | ||
5039 | down_read(&slub_lock); | |
5040 | if (!n) | |
5041 | print_slabinfo_header(m); | |
5042 | ||
5043 | return seq_list_start(&slab_caches, *pos); | |
5044 | } | |
5045 | ||
5046 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5047 | { | |
5048 | return seq_list_next(p, &slab_caches, pos); | |
5049 | } | |
5050 | ||
5051 | static void s_stop(struct seq_file *m, void *p) | |
5052 | { | |
5053 | up_read(&slub_lock); | |
5054 | } | |
5055 | ||
5056 | static int s_show(struct seq_file *m, void *p) | |
5057 | { | |
5058 | unsigned long nr_partials = 0; | |
5059 | unsigned long nr_slabs = 0; | |
5060 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5061 | unsigned long nr_objs = 0; |
5062 | unsigned long nr_free = 0; | |
57ed3eda PE |
5063 | struct kmem_cache *s; |
5064 | int node; | |
5065 | ||
5066 | s = list_entry(p, struct kmem_cache, list); | |
5067 | ||
5068 | for_each_online_node(node) { | |
5069 | struct kmem_cache_node *n = get_node(s, node); | |
5070 | ||
5071 | if (!n) | |
5072 | continue; | |
5073 | ||
5074 | nr_partials += n->nr_partial; | |
5075 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5076 | nr_objs += atomic_long_read(&n->total_objects); |
5077 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5078 | } |
5079 | ||
205ab99d | 5080 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5081 | |
5082 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5083 | nr_objs, s->size, oo_objects(s->oo), |
5084 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5085 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5086 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5087 | 0UL); | |
5088 | seq_putc(m, '\n'); | |
5089 | return 0; | |
5090 | } | |
5091 | ||
7b3c3a50 | 5092 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5093 | .start = s_start, |
5094 | .next = s_next, | |
5095 | .stop = s_stop, | |
5096 | .show = s_show, | |
5097 | }; | |
5098 | ||
7b3c3a50 AD |
5099 | static int slabinfo_open(struct inode *inode, struct file *file) |
5100 | { | |
5101 | return seq_open(file, &slabinfo_op); | |
5102 | } | |
5103 | ||
5104 | static const struct file_operations proc_slabinfo_operations = { | |
5105 | .open = slabinfo_open, | |
5106 | .read = seq_read, | |
5107 | .llseek = seq_lseek, | |
5108 | .release = seq_release, | |
5109 | }; | |
5110 | ||
5111 | static int __init slab_proc_init(void) | |
5112 | { | |
cf5d1131 | 5113 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5114 | return 0; |
5115 | } | |
5116 | module_init(slab_proc_init); | |
158a9624 | 5117 | #endif /* CONFIG_SLABINFO */ |