]>
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> |
bfa71457 | 31 | #include <linux/stacktrace.h> |
81819f0f | 32 | |
4a92379b RK |
33 | #include <trace/events/kmem.h> |
34 | ||
81819f0f CL |
35 | /* |
36 | * Lock order: | |
881db7fb CL |
37 | * 1. slub_lock (Global Semaphore) |
38 | * 2. node->list_lock | |
39 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 40 | * |
881db7fb CL |
41 | * slub_lock |
42 | * | |
43 | * The role of the slub_lock is to protect the list of all the slabs | |
44 | * and to synchronize major metadata changes to slab cache structures. | |
45 | * | |
46 | * The slab_lock is only used for debugging and on arches that do not | |
47 | * have the ability to do a cmpxchg_double. It only protects the second | |
48 | * double word in the page struct. Meaning | |
49 | * A. page->freelist -> List of object free in a page | |
50 | * B. page->counters -> Counters of objects | |
51 | * C. page->frozen -> frozen state | |
52 | * | |
53 | * If a slab is frozen then it is exempt from list management. It is not | |
54 | * on any list. The processor that froze the slab is the one who can | |
55 | * perform list operations on the page. Other processors may put objects | |
56 | * onto the freelist but the processor that froze the slab is the only | |
57 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
58 | * |
59 | * The list_lock protects the partial and full list on each node and | |
60 | * the partial slab counter. If taken then no new slabs may be added or | |
61 | * removed from the lists nor make the number of partial slabs be modified. | |
62 | * (Note that the total number of slabs is an atomic value that may be | |
63 | * modified without taking the list lock). | |
64 | * | |
65 | * The list_lock is a centralized lock and thus we avoid taking it as | |
66 | * much as possible. As long as SLUB does not have to handle partial | |
67 | * slabs, operations can continue without any centralized lock. F.e. | |
68 | * allocating a long series of objects that fill up slabs does not require | |
69 | * the list lock. | |
81819f0f CL |
70 | * Interrupts are disabled during allocation and deallocation in order to |
71 | * make the slab allocator safe to use in the context of an irq. In addition | |
72 | * interrupts are disabled to ensure that the processor does not change | |
73 | * while handling per_cpu slabs, due to kernel preemption. | |
74 | * | |
75 | * SLUB assigns one slab for allocation to each processor. | |
76 | * Allocations only occur from these slabs called cpu slabs. | |
77 | * | |
672bba3a CL |
78 | * Slabs with free elements are kept on a partial list and during regular |
79 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 80 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
81 | * We track full slabs for debugging purposes though because otherwise we |
82 | * cannot scan all objects. | |
81819f0f CL |
83 | * |
84 | * Slabs are freed when they become empty. Teardown and setup is | |
85 | * minimal so we rely on the page allocators per cpu caches for | |
86 | * fast frees and allocs. | |
87 | * | |
88 | * Overloading of page flags that are otherwise used for LRU management. | |
89 | * | |
4b6f0750 CL |
90 | * PageActive The slab is frozen and exempt from list processing. |
91 | * This means that the slab is dedicated to a purpose | |
92 | * such as satisfying allocations for a specific | |
93 | * processor. Objects may be freed in the slab while | |
94 | * it is frozen but slab_free will then skip the usual | |
95 | * list operations. It is up to the processor holding | |
96 | * the slab to integrate the slab into the slab lists | |
97 | * when the slab is no longer needed. | |
98 | * | |
99 | * One use of this flag is to mark slabs that are | |
100 | * used for allocations. Then such a slab becomes a cpu | |
101 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 102 | * freelist that allows lockless access to |
894b8788 CL |
103 | * free objects in addition to the regular freelist |
104 | * that requires the slab lock. | |
81819f0f CL |
105 | * |
106 | * PageError Slab requires special handling due to debug | |
107 | * options set. This moves slab handling out of | |
894b8788 | 108 | * the fast path and disables lockless freelists. |
81819f0f CL |
109 | */ |
110 | ||
af537b0a CL |
111 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
112 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
113 | ||
114 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
115 | { | |
5577bd8a | 116 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 117 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 118 | #else |
af537b0a | 119 | return 0; |
5577bd8a | 120 | #endif |
af537b0a | 121 | } |
5577bd8a | 122 | |
81819f0f CL |
123 | /* |
124 | * Issues still to be resolved: | |
125 | * | |
81819f0f CL |
126 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
127 | * | |
81819f0f CL |
128 | * - Variable sizing of the per node arrays |
129 | */ | |
130 | ||
131 | /* Enable to test recovery from slab corruption on boot */ | |
132 | #undef SLUB_RESILIENCY_TEST | |
133 | ||
b789ef51 CL |
134 | /* Enable to log cmpxchg failures */ |
135 | #undef SLUB_DEBUG_CMPXCHG | |
136 | ||
2086d26a CL |
137 | /* |
138 | * Mininum number of partial slabs. These will be left on the partial | |
139 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
140 | */ | |
76be8950 | 141 | #define MIN_PARTIAL 5 |
e95eed57 | 142 | |
2086d26a CL |
143 | /* |
144 | * Maximum number of desirable partial slabs. | |
145 | * The existence of more partial slabs makes kmem_cache_shrink | |
146 | * sort the partial list by the number of objects in the. | |
147 | */ | |
148 | #define MAX_PARTIAL 10 | |
149 | ||
81819f0f CL |
150 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
151 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 152 | |
fa5ec8a1 | 153 | /* |
3de47213 DR |
154 | * Debugging flags that require metadata to be stored in the slab. These get |
155 | * disabled when slub_debug=O is used and a cache's min order increases with | |
156 | * metadata. | |
fa5ec8a1 | 157 | */ |
3de47213 | 158 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 159 | |
81819f0f CL |
160 | /* |
161 | * Set of flags that will prevent slab merging | |
162 | */ | |
163 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
164 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
165 | SLAB_FAILSLAB) | |
81819f0f CL |
166 | |
167 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 168 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 169 | |
210b5c06 CG |
170 | #define OO_SHIFT 16 |
171 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 172 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 173 | |
81819f0f | 174 | /* Internal SLUB flags */ |
f90ec390 | 175 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 176 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f CL |
177 | |
178 | static int kmem_size = sizeof(struct kmem_cache); | |
179 | ||
180 | #ifdef CONFIG_SMP | |
181 | static struct notifier_block slab_notifier; | |
182 | #endif | |
183 | ||
184 | static enum { | |
185 | DOWN, /* No slab functionality available */ | |
51df1142 | 186 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 187 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
188 | SYSFS /* Sysfs up */ |
189 | } slab_state = DOWN; | |
190 | ||
191 | /* A list of all slab caches on the system */ | |
192 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 193 | static LIST_HEAD(slab_caches); |
81819f0f | 194 | |
02cbc874 CL |
195 | /* |
196 | * Tracking user of a slab. | |
197 | */ | |
d6543e39 | 198 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 199 | struct track { |
ce71e27c | 200 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
201 | #ifdef CONFIG_STACKTRACE |
202 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
203 | #endif | |
02cbc874 CL |
204 | int cpu; /* Was running on cpu */ |
205 | int pid; /* Pid context */ | |
206 | unsigned long when; /* When did the operation occur */ | |
207 | }; | |
208 | ||
209 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
210 | ||
ab4d5ed5 | 211 | #ifdef CONFIG_SYSFS |
81819f0f CL |
212 | static int sysfs_slab_add(struct kmem_cache *); |
213 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
214 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 215 | |
81819f0f | 216 | #else |
0c710013 CL |
217 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
218 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
219 | { return 0; } | |
151c602f CL |
220 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
221 | { | |
84c1cf62 | 222 | kfree(s->name); |
151c602f CL |
223 | kfree(s); |
224 | } | |
8ff12cfc | 225 | |
81819f0f CL |
226 | #endif |
227 | ||
4fdccdfb | 228 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
229 | { |
230 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 231 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
232 | #endif |
233 | } | |
234 | ||
81819f0f CL |
235 | /******************************************************************** |
236 | * Core slab cache functions | |
237 | *******************************************************************/ | |
238 | ||
239 | int slab_is_available(void) | |
240 | { | |
241 | return slab_state >= UP; | |
242 | } | |
243 | ||
244 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
245 | { | |
81819f0f | 246 | return s->node[node]; |
81819f0f CL |
247 | } |
248 | ||
6446faa2 | 249 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
250 | static inline int check_valid_pointer(struct kmem_cache *s, |
251 | struct page *page, const void *object) | |
252 | { | |
253 | void *base; | |
254 | ||
a973e9dd | 255 | if (!object) |
02cbc874 CL |
256 | return 1; |
257 | ||
a973e9dd | 258 | base = page_address(page); |
39b26464 | 259 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
260 | (object - base) % s->size) { |
261 | return 0; | |
262 | } | |
263 | ||
264 | return 1; | |
265 | } | |
266 | ||
7656c72b CL |
267 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
268 | { | |
269 | return *(void **)(object + s->offset); | |
270 | } | |
271 | ||
1393d9a1 CL |
272 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
273 | { | |
274 | void *p; | |
275 | ||
276 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
277 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
278 | #else | |
279 | p = get_freepointer(s, object); | |
280 | #endif | |
281 | return p; | |
282 | } | |
283 | ||
7656c72b CL |
284 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
285 | { | |
286 | *(void **)(object + s->offset) = fp; | |
287 | } | |
288 | ||
289 | /* Loop over all objects in a slab */ | |
224a88be CL |
290 | #define for_each_object(__p, __s, __addr, __objects) \ |
291 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
292 | __p += (__s)->size) |
293 | ||
7656c72b CL |
294 | /* Determine object index from a given position */ |
295 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
296 | { | |
297 | return (p - addr) / s->size; | |
298 | } | |
299 | ||
d71f606f MK |
300 | static inline size_t slab_ksize(const struct kmem_cache *s) |
301 | { | |
302 | #ifdef CONFIG_SLUB_DEBUG | |
303 | /* | |
304 | * Debugging requires use of the padding between object | |
305 | * and whatever may come after it. | |
306 | */ | |
307 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
308 | return s->objsize; | |
309 | ||
310 | #endif | |
311 | /* | |
312 | * If we have the need to store the freelist pointer | |
313 | * back there or track user information then we can | |
314 | * only use the space before that information. | |
315 | */ | |
316 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
317 | return s->inuse; | |
318 | /* | |
319 | * Else we can use all the padding etc for the allocation | |
320 | */ | |
321 | return s->size; | |
322 | } | |
323 | ||
ab9a0f19 LJ |
324 | static inline int order_objects(int order, unsigned long size, int reserved) |
325 | { | |
326 | return ((PAGE_SIZE << order) - reserved) / size; | |
327 | } | |
328 | ||
834f3d11 | 329 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 330 | unsigned long size, int reserved) |
834f3d11 CL |
331 | { |
332 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 333 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
334 | }; |
335 | ||
336 | return x; | |
337 | } | |
338 | ||
339 | static inline int oo_order(struct kmem_cache_order_objects x) | |
340 | { | |
210b5c06 | 341 | return x.x >> OO_SHIFT; |
834f3d11 CL |
342 | } |
343 | ||
344 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
345 | { | |
210b5c06 | 346 | return x.x & OO_MASK; |
834f3d11 CL |
347 | } |
348 | ||
881db7fb CL |
349 | /* |
350 | * Per slab locking using the pagelock | |
351 | */ | |
352 | static __always_inline void slab_lock(struct page *page) | |
353 | { | |
354 | bit_spin_lock(PG_locked, &page->flags); | |
355 | } | |
356 | ||
357 | static __always_inline void slab_unlock(struct page *page) | |
358 | { | |
359 | __bit_spin_unlock(PG_locked, &page->flags); | |
360 | } | |
361 | ||
1d07171c CL |
362 | /* Interrupts must be disabled (for the fallback code to work right) */ |
363 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
364 | void *freelist_old, unsigned long counters_old, | |
365 | void *freelist_new, unsigned long counters_new, | |
366 | const char *n) | |
367 | { | |
368 | VM_BUG_ON(!irqs_disabled()); | |
369 | #ifdef CONFIG_CMPXCHG_DOUBLE | |
370 | if (s->flags & __CMPXCHG_DOUBLE) { | |
371 | if (cmpxchg_double(&page->freelist, | |
372 | freelist_old, counters_old, | |
373 | freelist_new, counters_new)) | |
374 | return 1; | |
375 | } else | |
376 | #endif | |
377 | { | |
378 | slab_lock(page); | |
379 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
380 | page->freelist = freelist_new; | |
381 | page->counters = counters_new; | |
382 | slab_unlock(page); | |
383 | return 1; | |
384 | } | |
385 | slab_unlock(page); | |
386 | } | |
387 | ||
388 | cpu_relax(); | |
389 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
390 | ||
391 | #ifdef SLUB_DEBUG_CMPXCHG | |
392 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
393 | #endif | |
394 | ||
395 | return 0; | |
396 | } | |
397 | ||
b789ef51 CL |
398 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
399 | void *freelist_old, unsigned long counters_old, | |
400 | void *freelist_new, unsigned long counters_new, | |
401 | const char *n) | |
402 | { | |
403 | #ifdef CONFIG_CMPXCHG_DOUBLE | |
404 | if (s->flags & __CMPXCHG_DOUBLE) { | |
405 | if (cmpxchg_double(&page->freelist, | |
406 | freelist_old, counters_old, | |
407 | freelist_new, counters_new)) | |
408 | return 1; | |
409 | } else | |
410 | #endif | |
411 | { | |
1d07171c CL |
412 | unsigned long flags; |
413 | ||
414 | local_irq_save(flags); | |
881db7fb | 415 | slab_lock(page); |
b789ef51 CL |
416 | if (page->freelist == freelist_old && page->counters == counters_old) { |
417 | page->freelist = freelist_new; | |
418 | page->counters = counters_new; | |
881db7fb | 419 | slab_unlock(page); |
1d07171c | 420 | local_irq_restore(flags); |
b789ef51 CL |
421 | return 1; |
422 | } | |
881db7fb | 423 | slab_unlock(page); |
1d07171c | 424 | local_irq_restore(flags); |
b789ef51 CL |
425 | } |
426 | ||
427 | cpu_relax(); | |
428 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
429 | ||
430 | #ifdef SLUB_DEBUG_CMPXCHG | |
431 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
432 | #endif | |
433 | ||
434 | return 0; | |
435 | } | |
436 | ||
41ecc55b | 437 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
438 | /* |
439 | * Determine a map of object in use on a page. | |
440 | * | |
881db7fb | 441 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
442 | * not vanish from under us. |
443 | */ | |
444 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
445 | { | |
446 | void *p; | |
447 | void *addr = page_address(page); | |
448 | ||
449 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
450 | set_bit(slab_index(p, s, addr), map); | |
451 | } | |
452 | ||
41ecc55b CL |
453 | /* |
454 | * Debug settings: | |
455 | */ | |
f0630fff CL |
456 | #ifdef CONFIG_SLUB_DEBUG_ON |
457 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
458 | #else | |
41ecc55b | 459 | static int slub_debug; |
f0630fff | 460 | #endif |
41ecc55b CL |
461 | |
462 | static char *slub_debug_slabs; | |
fa5ec8a1 | 463 | static int disable_higher_order_debug; |
41ecc55b | 464 | |
81819f0f CL |
465 | /* |
466 | * Object debugging | |
467 | */ | |
468 | static void print_section(char *text, u8 *addr, unsigned int length) | |
469 | { | |
470 | int i, offset; | |
471 | int newline = 1; | |
472 | char ascii[17]; | |
473 | ||
474 | ascii[16] = 0; | |
475 | ||
476 | for (i = 0; i < length; i++) { | |
477 | if (newline) { | |
24922684 | 478 | printk(KERN_ERR "%8s 0x%p: ", text, addr + i); |
81819f0f CL |
479 | newline = 0; |
480 | } | |
06428780 | 481 | printk(KERN_CONT " %02x", addr[i]); |
81819f0f CL |
482 | offset = i % 16; |
483 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
484 | if (offset == 15) { | |
06428780 | 485 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
486 | newline = 1; |
487 | } | |
488 | } | |
489 | if (!newline) { | |
490 | i %= 16; | |
491 | while (i < 16) { | |
06428780 | 492 | printk(KERN_CONT " "); |
81819f0f CL |
493 | ascii[i] = ' '; |
494 | i++; | |
495 | } | |
06428780 | 496 | printk(KERN_CONT " %s\n", ascii); |
81819f0f CL |
497 | } |
498 | } | |
499 | ||
81819f0f CL |
500 | static struct track *get_track(struct kmem_cache *s, void *object, |
501 | enum track_item alloc) | |
502 | { | |
503 | struct track *p; | |
504 | ||
505 | if (s->offset) | |
506 | p = object + s->offset + sizeof(void *); | |
507 | else | |
508 | p = object + s->inuse; | |
509 | ||
510 | return p + alloc; | |
511 | } | |
512 | ||
513 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 514 | enum track_item alloc, unsigned long addr) |
81819f0f | 515 | { |
1a00df4a | 516 | struct track *p = get_track(s, object, alloc); |
81819f0f | 517 | |
81819f0f | 518 | if (addr) { |
d6543e39 BG |
519 | #ifdef CONFIG_STACKTRACE |
520 | struct stack_trace trace; | |
521 | int i; | |
522 | ||
523 | trace.nr_entries = 0; | |
524 | trace.max_entries = TRACK_ADDRS_COUNT; | |
525 | trace.entries = p->addrs; | |
526 | trace.skip = 3; | |
527 | save_stack_trace(&trace); | |
528 | ||
529 | /* See rant in lockdep.c */ | |
530 | if (trace.nr_entries != 0 && | |
531 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
532 | trace.nr_entries--; | |
533 | ||
534 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
535 | p->addrs[i] = 0; | |
536 | #endif | |
81819f0f CL |
537 | p->addr = addr; |
538 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 539 | p->pid = current->pid; |
81819f0f CL |
540 | p->when = jiffies; |
541 | } else | |
542 | memset(p, 0, sizeof(struct track)); | |
543 | } | |
544 | ||
81819f0f CL |
545 | static void init_tracking(struct kmem_cache *s, void *object) |
546 | { | |
24922684 CL |
547 | if (!(s->flags & SLAB_STORE_USER)) |
548 | return; | |
549 | ||
ce71e27c EGM |
550 | set_track(s, object, TRACK_FREE, 0UL); |
551 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
552 | } |
553 | ||
554 | static void print_track(const char *s, struct track *t) | |
555 | { | |
556 | if (!t->addr) | |
557 | return; | |
558 | ||
7daf705f | 559 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 560 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
561 | #ifdef CONFIG_STACKTRACE |
562 | { | |
563 | int i; | |
564 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
565 | if (t->addrs[i]) | |
566 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
567 | else | |
568 | break; | |
569 | } | |
570 | #endif | |
24922684 CL |
571 | } |
572 | ||
573 | static void print_tracking(struct kmem_cache *s, void *object) | |
574 | { | |
575 | if (!(s->flags & SLAB_STORE_USER)) | |
576 | return; | |
577 | ||
578 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
579 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
580 | } | |
581 | ||
582 | static void print_page_info(struct page *page) | |
583 | { | |
39b26464 CL |
584 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
585 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
586 | |
587 | } | |
588 | ||
589 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
590 | { | |
591 | va_list args; | |
592 | char buf[100]; | |
593 | ||
594 | va_start(args, fmt); | |
595 | vsnprintf(buf, sizeof(buf), fmt, args); | |
596 | va_end(args); | |
597 | printk(KERN_ERR "========================================" | |
598 | "=====================================\n"); | |
599 | printk(KERN_ERR "BUG %s: %s\n", s->name, buf); | |
600 | printk(KERN_ERR "----------------------------------------" | |
601 | "-------------------------------------\n\n"); | |
81819f0f CL |
602 | } |
603 | ||
24922684 CL |
604 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
605 | { | |
606 | va_list args; | |
607 | char buf[100]; | |
608 | ||
609 | va_start(args, fmt); | |
610 | vsnprintf(buf, sizeof(buf), fmt, args); | |
611 | va_end(args); | |
612 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
613 | } | |
614 | ||
615 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
616 | { |
617 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 618 | u8 *addr = page_address(page); |
24922684 CL |
619 | |
620 | print_tracking(s, p); | |
621 | ||
622 | print_page_info(page); | |
623 | ||
624 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
625 | p, p - addr, get_freepointer(s, p)); | |
626 | ||
627 | if (p > addr + 16) | |
628 | print_section("Bytes b4", p - 16, 16); | |
629 | ||
0ebd652b | 630 | print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE)); |
81819f0f CL |
631 | |
632 | if (s->flags & SLAB_RED_ZONE) | |
633 | print_section("Redzone", p + s->objsize, | |
634 | s->inuse - s->objsize); | |
635 | ||
81819f0f CL |
636 | if (s->offset) |
637 | off = s->offset + sizeof(void *); | |
638 | else | |
639 | off = s->inuse; | |
640 | ||
24922684 | 641 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 642 | off += 2 * sizeof(struct track); |
81819f0f CL |
643 | |
644 | if (off != s->size) | |
645 | /* Beginning of the filler is the free pointer */ | |
24922684 CL |
646 | print_section("Padding", p + off, s->size - off); |
647 | ||
648 | dump_stack(); | |
81819f0f CL |
649 | } |
650 | ||
651 | static void object_err(struct kmem_cache *s, struct page *page, | |
652 | u8 *object, char *reason) | |
653 | { | |
3dc50637 | 654 | slab_bug(s, "%s", reason); |
24922684 | 655 | print_trailer(s, page, object); |
81819f0f CL |
656 | } |
657 | ||
24922684 | 658 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
659 | { |
660 | va_list args; | |
661 | char buf[100]; | |
662 | ||
24922684 CL |
663 | va_start(args, fmt); |
664 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 665 | va_end(args); |
3dc50637 | 666 | slab_bug(s, "%s", buf); |
24922684 | 667 | print_page_info(page); |
81819f0f CL |
668 | dump_stack(); |
669 | } | |
670 | ||
f7cb1933 | 671 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
672 | { |
673 | u8 *p = object; | |
674 | ||
675 | if (s->flags & __OBJECT_POISON) { | |
676 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 677 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
678 | } |
679 | ||
680 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 681 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
682 | } |
683 | ||
c4089f98 | 684 | static u8 *check_bytes8(u8 *start, u8 value, unsigned int bytes) |
81819f0f CL |
685 | { |
686 | while (bytes) { | |
c4089f98 | 687 | if (*start != value) |
24922684 | 688 | return start; |
81819f0f CL |
689 | start++; |
690 | bytes--; | |
691 | } | |
24922684 CL |
692 | return NULL; |
693 | } | |
694 | ||
c4089f98 MS |
695 | static u8 *check_bytes(u8 *start, u8 value, unsigned int bytes) |
696 | { | |
697 | u64 value64; | |
698 | unsigned int words, prefix; | |
699 | ||
700 | if (bytes <= 16) | |
701 | return check_bytes8(start, value, bytes); | |
702 | ||
703 | value64 = value | value << 8 | value << 16 | value << 24; | |
704 | value64 = value64 | value64 << 32; | |
705 | prefix = 8 - ((unsigned long)start) % 8; | |
706 | ||
707 | if (prefix) { | |
708 | u8 *r = check_bytes8(start, value, prefix); | |
709 | if (r) | |
710 | return r; | |
711 | start += prefix; | |
712 | bytes -= prefix; | |
713 | } | |
714 | ||
715 | words = bytes / 8; | |
716 | ||
717 | while (words) { | |
718 | if (*(u64 *)start != value64) | |
719 | return check_bytes8(start, value, 8); | |
720 | start += 8; | |
721 | words--; | |
722 | } | |
723 | ||
724 | return check_bytes8(start, value, bytes % 8); | |
725 | } | |
726 | ||
24922684 CL |
727 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
728 | void *from, void *to) | |
729 | { | |
730 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
731 | memset(from, data, to - from); | |
732 | } | |
733 | ||
734 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
735 | u8 *object, char *what, | |
06428780 | 736 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
737 | { |
738 | u8 *fault; | |
739 | u8 *end; | |
740 | ||
741 | fault = check_bytes(start, value, bytes); | |
742 | if (!fault) | |
743 | return 1; | |
744 | ||
745 | end = start + bytes; | |
746 | while (end > fault && end[-1] == value) | |
747 | end--; | |
748 | ||
749 | slab_bug(s, "%s overwritten", what); | |
750 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
751 | fault, end - 1, fault[0], value); | |
752 | print_trailer(s, page, object); | |
753 | ||
754 | restore_bytes(s, what, value, fault, end); | |
755 | return 0; | |
81819f0f CL |
756 | } |
757 | ||
81819f0f CL |
758 | /* |
759 | * Object layout: | |
760 | * | |
761 | * object address | |
762 | * Bytes of the object to be managed. | |
763 | * If the freepointer may overlay the object then the free | |
764 | * pointer is the first word of the object. | |
672bba3a | 765 | * |
81819f0f CL |
766 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
767 | * 0xa5 (POISON_END) | |
768 | * | |
769 | * object + s->objsize | |
770 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
771 | * Padding is extended by another word if Redzoning is enabled and |
772 | * objsize == inuse. | |
773 | * | |
81819f0f CL |
774 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
775 | * 0xcc (RED_ACTIVE) for objects in use. | |
776 | * | |
777 | * object + s->inuse | |
672bba3a CL |
778 | * Meta data starts here. |
779 | * | |
81819f0f CL |
780 | * A. Free pointer (if we cannot overwrite object on free) |
781 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 782 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 783 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
784 | * before the word boundary. |
785 | * | |
786 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
787 | * |
788 | * object + s->size | |
672bba3a | 789 | * Nothing is used beyond s->size. |
81819f0f | 790 | * |
672bba3a CL |
791 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
792 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
793 | * may be used with merged slabcaches. |
794 | */ | |
795 | ||
81819f0f CL |
796 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
797 | { | |
798 | unsigned long off = s->inuse; /* The end of info */ | |
799 | ||
800 | if (s->offset) | |
801 | /* Freepointer is placed after the object. */ | |
802 | off += sizeof(void *); | |
803 | ||
804 | if (s->flags & SLAB_STORE_USER) | |
805 | /* We also have user information there */ | |
806 | off += 2 * sizeof(struct track); | |
807 | ||
808 | if (s->size == off) | |
809 | return 1; | |
810 | ||
24922684 CL |
811 | return check_bytes_and_report(s, page, p, "Object padding", |
812 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
813 | } |
814 | ||
39b26464 | 815 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
816 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
817 | { | |
24922684 CL |
818 | u8 *start; |
819 | u8 *fault; | |
820 | u8 *end; | |
821 | int length; | |
822 | int remainder; | |
81819f0f CL |
823 | |
824 | if (!(s->flags & SLAB_POISON)) | |
825 | return 1; | |
826 | ||
a973e9dd | 827 | start = page_address(page); |
ab9a0f19 | 828 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
829 | end = start + length; |
830 | remainder = length % s->size; | |
81819f0f CL |
831 | if (!remainder) |
832 | return 1; | |
833 | ||
39b26464 | 834 | fault = check_bytes(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
835 | if (!fault) |
836 | return 1; | |
837 | while (end > fault && end[-1] == POISON_INUSE) | |
838 | end--; | |
839 | ||
840 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
39b26464 | 841 | print_section("Padding", end - remainder, remainder); |
24922684 | 842 | |
8a3d271d | 843 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 844 | return 0; |
81819f0f CL |
845 | } |
846 | ||
847 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 848 | void *object, u8 val) |
81819f0f CL |
849 | { |
850 | u8 *p = object; | |
851 | u8 *endobject = object + s->objsize; | |
852 | ||
853 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 854 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 855 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 856 | return 0; |
81819f0f | 857 | } else { |
3adbefee IM |
858 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
859 | check_bytes_and_report(s, page, p, "Alignment padding", | |
860 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
861 | } | |
81819f0f CL |
862 | } |
863 | ||
864 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 865 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
866 | (!check_bytes_and_report(s, page, p, "Poison", p, |
867 | POISON_FREE, s->objsize - 1) || | |
868 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 869 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 870 | return 0; |
81819f0f CL |
871 | /* |
872 | * check_pad_bytes cleans up on its own. | |
873 | */ | |
874 | check_pad_bytes(s, page, p); | |
875 | } | |
876 | ||
f7cb1933 | 877 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
878 | /* |
879 | * Object and freepointer overlap. Cannot check | |
880 | * freepointer while object is allocated. | |
881 | */ | |
882 | return 1; | |
883 | ||
884 | /* Check free pointer validity */ | |
885 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
886 | object_err(s, page, p, "Freepointer corrupt"); | |
887 | /* | |
9f6c708e | 888 | * No choice but to zap it and thus lose the remainder |
81819f0f | 889 | * of the free objects in this slab. May cause |
672bba3a | 890 | * another error because the object count is now wrong. |
81819f0f | 891 | */ |
a973e9dd | 892 | set_freepointer(s, p, NULL); |
81819f0f CL |
893 | return 0; |
894 | } | |
895 | return 1; | |
896 | } | |
897 | ||
898 | static int check_slab(struct kmem_cache *s, struct page *page) | |
899 | { | |
39b26464 CL |
900 | int maxobj; |
901 | ||
81819f0f CL |
902 | VM_BUG_ON(!irqs_disabled()); |
903 | ||
904 | if (!PageSlab(page)) { | |
24922684 | 905 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
906 | return 0; |
907 | } | |
39b26464 | 908 | |
ab9a0f19 | 909 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
910 | if (page->objects > maxobj) { |
911 | slab_err(s, page, "objects %u > max %u", | |
912 | s->name, page->objects, maxobj); | |
913 | return 0; | |
914 | } | |
915 | if (page->inuse > page->objects) { | |
24922684 | 916 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 917 | s->name, page->inuse, page->objects); |
81819f0f CL |
918 | return 0; |
919 | } | |
920 | /* Slab_pad_check fixes things up after itself */ | |
921 | slab_pad_check(s, page); | |
922 | return 1; | |
923 | } | |
924 | ||
925 | /* | |
672bba3a CL |
926 | * Determine if a certain object on a page is on the freelist. Must hold the |
927 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
928 | */ |
929 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
930 | { | |
931 | int nr = 0; | |
881db7fb | 932 | void *fp; |
81819f0f | 933 | void *object = NULL; |
224a88be | 934 | unsigned long max_objects; |
81819f0f | 935 | |
881db7fb | 936 | fp = page->freelist; |
39b26464 | 937 | while (fp && nr <= page->objects) { |
81819f0f CL |
938 | if (fp == search) |
939 | return 1; | |
940 | if (!check_valid_pointer(s, page, fp)) { | |
941 | if (object) { | |
942 | object_err(s, page, object, | |
943 | "Freechain corrupt"); | |
a973e9dd | 944 | set_freepointer(s, object, NULL); |
81819f0f CL |
945 | break; |
946 | } else { | |
24922684 | 947 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 948 | page->freelist = NULL; |
39b26464 | 949 | page->inuse = page->objects; |
24922684 | 950 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
951 | return 0; |
952 | } | |
953 | break; | |
954 | } | |
955 | object = fp; | |
956 | fp = get_freepointer(s, object); | |
957 | nr++; | |
958 | } | |
959 | ||
ab9a0f19 | 960 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
961 | if (max_objects > MAX_OBJS_PER_PAGE) |
962 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
963 | |
964 | if (page->objects != max_objects) { | |
965 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
966 | "should be %d", page->objects, max_objects); | |
967 | page->objects = max_objects; | |
968 | slab_fix(s, "Number of objects adjusted."); | |
969 | } | |
39b26464 | 970 | if (page->inuse != page->objects - nr) { |
70d71228 | 971 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
972 | "counted were %d", page->inuse, page->objects - nr); |
973 | page->inuse = page->objects - nr; | |
24922684 | 974 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
975 | } |
976 | return search == NULL; | |
977 | } | |
978 | ||
0121c619 CL |
979 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
980 | int alloc) | |
3ec09742 CL |
981 | { |
982 | if (s->flags & SLAB_TRACE) { | |
983 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
984 | s->name, | |
985 | alloc ? "alloc" : "free", | |
986 | object, page->inuse, | |
987 | page->freelist); | |
988 | ||
989 | if (!alloc) | |
990 | print_section("Object", (void *)object, s->objsize); | |
991 | ||
992 | dump_stack(); | |
993 | } | |
994 | } | |
995 | ||
c016b0bd CL |
996 | /* |
997 | * Hooks for other subsystems that check memory allocations. In a typical | |
998 | * production configuration these hooks all should produce no code at all. | |
999 | */ | |
1000 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1001 | { | |
c1d50836 | 1002 | flags &= gfp_allowed_mask; |
c016b0bd CL |
1003 | lockdep_trace_alloc(flags); |
1004 | might_sleep_if(flags & __GFP_WAIT); | |
1005 | ||
1006 | return should_failslab(s->objsize, flags, s->flags); | |
1007 | } | |
1008 | ||
1009 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
1010 | { | |
c1d50836 | 1011 | flags &= gfp_allowed_mask; |
b3d41885 | 1012 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
c016b0bd CL |
1013 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
1014 | } | |
1015 | ||
1016 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
1017 | { | |
1018 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 1019 | |
d3f661d6 CL |
1020 | /* |
1021 | * Trouble is that we may no longer disable interupts in the fast path | |
1022 | * So in order to make the debug calls that expect irqs to be | |
1023 | * disabled we need to disable interrupts temporarily. | |
1024 | */ | |
1025 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1026 | { | |
1027 | unsigned long flags; | |
1028 | ||
1029 | local_irq_save(flags); | |
1030 | kmemcheck_slab_free(s, x, s->objsize); | |
1031 | debug_check_no_locks_freed(x, s->objsize); | |
d3f661d6 CL |
1032 | local_irq_restore(flags); |
1033 | } | |
1034 | #endif | |
f9b615de TG |
1035 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
1036 | debug_check_no_obj_freed(x, s->objsize); | |
c016b0bd CL |
1037 | } |
1038 | ||
643b1138 | 1039 | /* |
672bba3a | 1040 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
1041 | * |
1042 | * list_lock must be held. | |
643b1138 | 1043 | */ |
5cc6eee8 CL |
1044 | static void add_full(struct kmem_cache *s, |
1045 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1046 | { |
5cc6eee8 CL |
1047 | if (!(s->flags & SLAB_STORE_USER)) |
1048 | return; | |
1049 | ||
643b1138 | 1050 | list_add(&page->lru, &n->full); |
643b1138 CL |
1051 | } |
1052 | ||
5cc6eee8 CL |
1053 | /* |
1054 | * list_lock must be held. | |
1055 | */ | |
643b1138 CL |
1056 | static void remove_full(struct kmem_cache *s, struct page *page) |
1057 | { | |
643b1138 CL |
1058 | if (!(s->flags & SLAB_STORE_USER)) |
1059 | return; | |
1060 | ||
643b1138 | 1061 | list_del(&page->lru); |
643b1138 CL |
1062 | } |
1063 | ||
0f389ec6 CL |
1064 | /* Tracking of the number of slabs for debugging purposes */ |
1065 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1066 | { | |
1067 | struct kmem_cache_node *n = get_node(s, node); | |
1068 | ||
1069 | return atomic_long_read(&n->nr_slabs); | |
1070 | } | |
1071 | ||
26c02cf0 AB |
1072 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1073 | { | |
1074 | return atomic_long_read(&n->nr_slabs); | |
1075 | } | |
1076 | ||
205ab99d | 1077 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1078 | { |
1079 | struct kmem_cache_node *n = get_node(s, node); | |
1080 | ||
1081 | /* | |
1082 | * May be called early in order to allocate a slab for the | |
1083 | * kmem_cache_node structure. Solve the chicken-egg | |
1084 | * dilemma by deferring the increment of the count during | |
1085 | * bootstrap (see early_kmem_cache_node_alloc). | |
1086 | */ | |
7340cc84 | 1087 | if (n) { |
0f389ec6 | 1088 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1089 | atomic_long_add(objects, &n->total_objects); |
1090 | } | |
0f389ec6 | 1091 | } |
205ab99d | 1092 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1093 | { |
1094 | struct kmem_cache_node *n = get_node(s, node); | |
1095 | ||
1096 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1097 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1098 | } |
1099 | ||
1100 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1101 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1102 | void *object) | |
1103 | { | |
1104 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1105 | return; | |
1106 | ||
f7cb1933 | 1107 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1108 | init_tracking(s, object); |
1109 | } | |
1110 | ||
1537066c | 1111 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1112 | void *object, unsigned long addr) |
81819f0f CL |
1113 | { |
1114 | if (!check_slab(s, page)) | |
1115 | goto bad; | |
1116 | ||
81819f0f CL |
1117 | if (!check_valid_pointer(s, page, object)) { |
1118 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1119 | goto bad; |
81819f0f CL |
1120 | } |
1121 | ||
f7cb1933 | 1122 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1123 | goto bad; |
81819f0f | 1124 | |
3ec09742 CL |
1125 | /* Success perform special debug activities for allocs */ |
1126 | if (s->flags & SLAB_STORE_USER) | |
1127 | set_track(s, object, TRACK_ALLOC, addr); | |
1128 | trace(s, page, object, 1); | |
f7cb1933 | 1129 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1130 | return 1; |
3ec09742 | 1131 | |
81819f0f CL |
1132 | bad: |
1133 | if (PageSlab(page)) { | |
1134 | /* | |
1135 | * If this is a slab page then lets do the best we can | |
1136 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1137 | * as used avoids touching the remaining objects. |
81819f0f | 1138 | */ |
24922684 | 1139 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1140 | page->inuse = page->objects; |
a973e9dd | 1141 | page->freelist = NULL; |
81819f0f CL |
1142 | } |
1143 | return 0; | |
1144 | } | |
1145 | ||
1537066c CL |
1146 | static noinline int free_debug_processing(struct kmem_cache *s, |
1147 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1148 | { |
5c2e4bbb CL |
1149 | unsigned long flags; |
1150 | int rc = 0; | |
1151 | ||
1152 | local_irq_save(flags); | |
881db7fb CL |
1153 | slab_lock(page); |
1154 | ||
81819f0f CL |
1155 | if (!check_slab(s, page)) |
1156 | goto fail; | |
1157 | ||
1158 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1159 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1160 | goto fail; |
1161 | } | |
1162 | ||
1163 | if (on_freelist(s, page, object)) { | |
24922684 | 1164 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1165 | goto fail; |
1166 | } | |
1167 | ||
f7cb1933 | 1168 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1169 | goto out; |
81819f0f CL |
1170 | |
1171 | if (unlikely(s != page->slab)) { | |
3adbefee | 1172 | if (!PageSlab(page)) { |
70d71228 CL |
1173 | slab_err(s, page, "Attempt to free object(0x%p) " |
1174 | "outside of slab", object); | |
3adbefee | 1175 | } else if (!page->slab) { |
81819f0f | 1176 | printk(KERN_ERR |
70d71228 | 1177 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1178 | object); |
70d71228 | 1179 | dump_stack(); |
06428780 | 1180 | } else |
24922684 CL |
1181 | object_err(s, page, object, |
1182 | "page slab pointer corrupt."); | |
81819f0f CL |
1183 | goto fail; |
1184 | } | |
3ec09742 | 1185 | |
3ec09742 CL |
1186 | if (s->flags & SLAB_STORE_USER) |
1187 | set_track(s, object, TRACK_FREE, addr); | |
1188 | trace(s, page, object, 0); | |
f7cb1933 | 1189 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb CL |
1190 | rc = 1; |
1191 | out: | |
881db7fb | 1192 | slab_unlock(page); |
5c2e4bbb CL |
1193 | local_irq_restore(flags); |
1194 | return rc; | |
3ec09742 | 1195 | |
81819f0f | 1196 | fail: |
24922684 | 1197 | slab_fix(s, "Object at 0x%p not freed", object); |
5c2e4bbb | 1198 | goto out; |
81819f0f CL |
1199 | } |
1200 | ||
41ecc55b CL |
1201 | static int __init setup_slub_debug(char *str) |
1202 | { | |
f0630fff CL |
1203 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1204 | if (*str++ != '=' || !*str) | |
1205 | /* | |
1206 | * No options specified. Switch on full debugging. | |
1207 | */ | |
1208 | goto out; | |
1209 | ||
1210 | if (*str == ',') | |
1211 | /* | |
1212 | * No options but restriction on slabs. This means full | |
1213 | * debugging for slabs matching a pattern. | |
1214 | */ | |
1215 | goto check_slabs; | |
1216 | ||
fa5ec8a1 DR |
1217 | if (tolower(*str) == 'o') { |
1218 | /* | |
1219 | * Avoid enabling debugging on caches if its minimum order | |
1220 | * would increase as a result. | |
1221 | */ | |
1222 | disable_higher_order_debug = 1; | |
1223 | goto out; | |
1224 | } | |
1225 | ||
f0630fff CL |
1226 | slub_debug = 0; |
1227 | if (*str == '-') | |
1228 | /* | |
1229 | * Switch off all debugging measures. | |
1230 | */ | |
1231 | goto out; | |
1232 | ||
1233 | /* | |
1234 | * Determine which debug features should be switched on | |
1235 | */ | |
06428780 | 1236 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1237 | switch (tolower(*str)) { |
1238 | case 'f': | |
1239 | slub_debug |= SLAB_DEBUG_FREE; | |
1240 | break; | |
1241 | case 'z': | |
1242 | slub_debug |= SLAB_RED_ZONE; | |
1243 | break; | |
1244 | case 'p': | |
1245 | slub_debug |= SLAB_POISON; | |
1246 | break; | |
1247 | case 'u': | |
1248 | slub_debug |= SLAB_STORE_USER; | |
1249 | break; | |
1250 | case 't': | |
1251 | slub_debug |= SLAB_TRACE; | |
1252 | break; | |
4c13dd3b DM |
1253 | case 'a': |
1254 | slub_debug |= SLAB_FAILSLAB; | |
1255 | break; | |
f0630fff CL |
1256 | default: |
1257 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1258 | "unknown. skipped\n", *str); |
f0630fff | 1259 | } |
41ecc55b CL |
1260 | } |
1261 | ||
f0630fff | 1262 | check_slabs: |
41ecc55b CL |
1263 | if (*str == ',') |
1264 | slub_debug_slabs = str + 1; | |
f0630fff | 1265 | out: |
41ecc55b CL |
1266 | return 1; |
1267 | } | |
1268 | ||
1269 | __setup("slub_debug", setup_slub_debug); | |
1270 | ||
ba0268a8 CL |
1271 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1272 | unsigned long flags, const char *name, | |
51cc5068 | 1273 | void (*ctor)(void *)) |
41ecc55b CL |
1274 | { |
1275 | /* | |
e153362a | 1276 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1277 | */ |
e153362a | 1278 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1279 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1280 | flags |= slub_debug; | |
ba0268a8 CL |
1281 | |
1282 | return flags; | |
41ecc55b CL |
1283 | } |
1284 | #else | |
3ec09742 CL |
1285 | static inline void setup_object_debug(struct kmem_cache *s, |
1286 | struct page *page, void *object) {} | |
41ecc55b | 1287 | |
3ec09742 | 1288 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1289 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1290 | |
3ec09742 | 1291 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1292 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1293 | |
41ecc55b CL |
1294 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1295 | { return 1; } | |
1296 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1297 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1298 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1299 | struct page *page) {} | |
2cfb7455 | 1300 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
ba0268a8 CL |
1301 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1302 | unsigned long flags, const char *name, | |
51cc5068 | 1303 | void (*ctor)(void *)) |
ba0268a8 CL |
1304 | { |
1305 | return flags; | |
1306 | } | |
41ecc55b | 1307 | #define slub_debug 0 |
0f389ec6 | 1308 | |
fdaa45e9 IM |
1309 | #define disable_higher_order_debug 0 |
1310 | ||
0f389ec6 CL |
1311 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1312 | { return 0; } | |
26c02cf0 AB |
1313 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1314 | { return 0; } | |
205ab99d CL |
1315 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1316 | int objects) {} | |
1317 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1318 | int objects) {} | |
7d550c56 CL |
1319 | |
1320 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1321 | { return 0; } | |
1322 | ||
1323 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1324 | void *object) {} | |
1325 | ||
1326 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1327 | ||
ab4d5ed5 | 1328 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1329 | |
81819f0f CL |
1330 | /* |
1331 | * Slab allocation and freeing | |
1332 | */ | |
65c3376a CL |
1333 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1334 | struct kmem_cache_order_objects oo) | |
1335 | { | |
1336 | int order = oo_order(oo); | |
1337 | ||
b1eeab67 VN |
1338 | flags |= __GFP_NOTRACK; |
1339 | ||
2154a336 | 1340 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1341 | return alloc_pages(flags, order); |
1342 | else | |
6b65aaf3 | 1343 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1344 | } |
1345 | ||
81819f0f CL |
1346 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1347 | { | |
06428780 | 1348 | struct page *page; |
834f3d11 | 1349 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1350 | gfp_t alloc_gfp; |
81819f0f | 1351 | |
7e0528da CL |
1352 | flags &= gfp_allowed_mask; |
1353 | ||
1354 | if (flags & __GFP_WAIT) | |
1355 | local_irq_enable(); | |
1356 | ||
b7a49f0d | 1357 | flags |= s->allocflags; |
e12ba74d | 1358 | |
ba52270d PE |
1359 | /* |
1360 | * Let the initial higher-order allocation fail under memory pressure | |
1361 | * so we fall-back to the minimum order allocation. | |
1362 | */ | |
1363 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1364 | ||
1365 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1366 | if (unlikely(!page)) { |
1367 | oo = s->min; | |
1368 | /* | |
1369 | * Allocation may have failed due to fragmentation. | |
1370 | * Try a lower order alloc if possible | |
1371 | */ | |
1372 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1373 | |
7e0528da CL |
1374 | if (page) |
1375 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1376 | } |
5a896d9e | 1377 | |
7e0528da CL |
1378 | if (flags & __GFP_WAIT) |
1379 | local_irq_disable(); | |
1380 | ||
1381 | if (!page) | |
1382 | return NULL; | |
1383 | ||
5a896d9e | 1384 | if (kmemcheck_enabled |
5086c389 | 1385 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1386 | int pages = 1 << oo_order(oo); |
1387 | ||
1388 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1389 | ||
1390 | /* | |
1391 | * Objects from caches that have a constructor don't get | |
1392 | * cleared when they're allocated, so we need to do it here. | |
1393 | */ | |
1394 | if (s->ctor) | |
1395 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1396 | else | |
1397 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1398 | } |
1399 | ||
834f3d11 | 1400 | page->objects = oo_objects(oo); |
81819f0f CL |
1401 | mod_zone_page_state(page_zone(page), |
1402 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1403 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1404 | 1 << oo_order(oo)); |
81819f0f CL |
1405 | |
1406 | return page; | |
1407 | } | |
1408 | ||
1409 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1410 | void *object) | |
1411 | { | |
3ec09742 | 1412 | setup_object_debug(s, page, object); |
4f104934 | 1413 | if (unlikely(s->ctor)) |
51cc5068 | 1414 | s->ctor(object); |
81819f0f CL |
1415 | } |
1416 | ||
1417 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1418 | { | |
1419 | struct page *page; | |
81819f0f | 1420 | void *start; |
81819f0f CL |
1421 | void *last; |
1422 | void *p; | |
1423 | ||
6cb06229 | 1424 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1425 | |
6cb06229 CL |
1426 | page = allocate_slab(s, |
1427 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1428 | if (!page) |
1429 | goto out; | |
1430 | ||
205ab99d | 1431 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1432 | page->slab = s; |
1433 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1434 | |
1435 | start = page_address(page); | |
81819f0f CL |
1436 | |
1437 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1438 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1439 | |
1440 | last = start; | |
224a88be | 1441 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1442 | setup_object(s, page, last); |
1443 | set_freepointer(s, last, p); | |
1444 | last = p; | |
1445 | } | |
1446 | setup_object(s, page, last); | |
a973e9dd | 1447 | set_freepointer(s, last, NULL); |
81819f0f CL |
1448 | |
1449 | page->freelist = start; | |
1450 | page->inuse = 0; | |
8cb0a506 | 1451 | page->frozen = 1; |
81819f0f | 1452 | out: |
81819f0f CL |
1453 | return page; |
1454 | } | |
1455 | ||
1456 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1457 | { | |
834f3d11 CL |
1458 | int order = compound_order(page); |
1459 | int pages = 1 << order; | |
81819f0f | 1460 | |
af537b0a | 1461 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1462 | void *p; |
1463 | ||
1464 | slab_pad_check(s, page); | |
224a88be CL |
1465 | for_each_object(p, s, page_address(page), |
1466 | page->objects) | |
f7cb1933 | 1467 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1468 | } |
1469 | ||
b1eeab67 | 1470 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1471 | |
81819f0f CL |
1472 | mod_zone_page_state(page_zone(page), |
1473 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1474 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1475 | -pages); |
81819f0f | 1476 | |
49bd5221 CL |
1477 | __ClearPageSlab(page); |
1478 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1479 | if (current->reclaim_state) |
1480 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1481 | __free_pages(page, order); |
81819f0f CL |
1482 | } |
1483 | ||
da9a638c LJ |
1484 | #define need_reserve_slab_rcu \ |
1485 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1486 | ||
81819f0f CL |
1487 | static void rcu_free_slab(struct rcu_head *h) |
1488 | { | |
1489 | struct page *page; | |
1490 | ||
da9a638c LJ |
1491 | if (need_reserve_slab_rcu) |
1492 | page = virt_to_head_page(h); | |
1493 | else | |
1494 | page = container_of((struct list_head *)h, struct page, lru); | |
1495 | ||
81819f0f CL |
1496 | __free_slab(page->slab, page); |
1497 | } | |
1498 | ||
1499 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1500 | { | |
1501 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1502 | struct rcu_head *head; |
1503 | ||
1504 | if (need_reserve_slab_rcu) { | |
1505 | int order = compound_order(page); | |
1506 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1507 | ||
1508 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1509 | head = page_address(page) + offset; | |
1510 | } else { | |
1511 | /* | |
1512 | * RCU free overloads the RCU head over the LRU | |
1513 | */ | |
1514 | head = (void *)&page->lru; | |
1515 | } | |
81819f0f CL |
1516 | |
1517 | call_rcu(head, rcu_free_slab); | |
1518 | } else | |
1519 | __free_slab(s, page); | |
1520 | } | |
1521 | ||
1522 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1523 | { | |
205ab99d | 1524 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1525 | free_slab(s, page); |
1526 | } | |
1527 | ||
1528 | /* | |
5cc6eee8 CL |
1529 | * Management of partially allocated slabs. |
1530 | * | |
1531 | * list_lock must be held. | |
81819f0f | 1532 | */ |
5cc6eee8 | 1533 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1534 | struct page *page, int tail) |
81819f0f | 1535 | { |
e95eed57 | 1536 | n->nr_partial++; |
7c2e132c CL |
1537 | if (tail) |
1538 | list_add_tail(&page->lru, &n->partial); | |
1539 | else | |
1540 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1541 | } |
1542 | ||
5cc6eee8 CL |
1543 | /* |
1544 | * list_lock must be held. | |
1545 | */ | |
1546 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1547 | struct page *page) |
1548 | { | |
1549 | list_del(&page->lru); | |
1550 | n->nr_partial--; | |
1551 | } | |
1552 | ||
81819f0f | 1553 | /* |
5cc6eee8 CL |
1554 | * Lock slab, remove from the partial list and put the object into the |
1555 | * per cpu freelist. | |
81819f0f | 1556 | * |
672bba3a | 1557 | * Must hold list_lock. |
81819f0f | 1558 | */ |
881db7fb | 1559 | static inline int acquire_slab(struct kmem_cache *s, |
61728d1e | 1560 | struct kmem_cache_node *n, struct page *page) |
81819f0f | 1561 | { |
2cfb7455 CL |
1562 | void *freelist; |
1563 | unsigned long counters; | |
1564 | struct page new; | |
1565 | ||
2cfb7455 CL |
1566 | /* |
1567 | * Zap the freelist and set the frozen bit. | |
1568 | * The old freelist is the list of objects for the | |
1569 | * per cpu allocation list. | |
1570 | */ | |
1571 | do { | |
1572 | freelist = page->freelist; | |
1573 | counters = page->counters; | |
1574 | new.counters = counters; | |
1575 | new.inuse = page->objects; | |
1576 | ||
1577 | VM_BUG_ON(new.frozen); | |
1578 | new.frozen = 1; | |
1579 | ||
1d07171c | 1580 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1581 | freelist, counters, |
1582 | NULL, new.counters, | |
1583 | "lock and freeze")); | |
1584 | ||
1585 | remove_partial(n, page); | |
1586 | ||
1587 | if (freelist) { | |
1588 | /* Populate the per cpu freelist */ | |
1589 | this_cpu_write(s->cpu_slab->freelist, freelist); | |
1590 | this_cpu_write(s->cpu_slab->page, page); | |
1591 | this_cpu_write(s->cpu_slab->node, page_to_nid(page)); | |
81819f0f | 1592 | return 1; |
2cfb7455 CL |
1593 | } else { |
1594 | /* | |
1595 | * Slab page came from the wrong list. No object to allocate | |
1596 | * from. Put it onto the correct list and continue partial | |
1597 | * scan. | |
1598 | */ | |
1599 | printk(KERN_ERR "SLUB: %s : Page without available objects on" | |
1600 | " partial list\n", s->name); | |
2cfb7455 | 1601 | return 0; |
81819f0f | 1602 | } |
81819f0f CL |
1603 | } |
1604 | ||
1605 | /* | |
672bba3a | 1606 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1607 | */ |
61728d1e CL |
1608 | static struct page *get_partial_node(struct kmem_cache *s, |
1609 | struct kmem_cache_node *n) | |
81819f0f CL |
1610 | { |
1611 | struct page *page; | |
1612 | ||
1613 | /* | |
1614 | * Racy check. If we mistakenly see no partial slabs then we | |
1615 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1616 | * partial slab and there is none available then get_partials() |
1617 | * will return NULL. | |
81819f0f CL |
1618 | */ |
1619 | if (!n || !n->nr_partial) | |
1620 | return NULL; | |
1621 | ||
1622 | spin_lock(&n->list_lock); | |
1623 | list_for_each_entry(page, &n->partial, lru) | |
881db7fb | 1624 | if (acquire_slab(s, n, page)) |
81819f0f CL |
1625 | goto out; |
1626 | page = NULL; | |
1627 | out: | |
1628 | spin_unlock(&n->list_lock); | |
1629 | return page; | |
1630 | } | |
1631 | ||
1632 | /* | |
672bba3a | 1633 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f CL |
1634 | */ |
1635 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1636 | { | |
1637 | #ifdef CONFIG_NUMA | |
1638 | struct zonelist *zonelist; | |
dd1a239f | 1639 | struct zoneref *z; |
54a6eb5c MG |
1640 | struct zone *zone; |
1641 | enum zone_type high_zoneidx = gfp_zone(flags); | |
81819f0f CL |
1642 | struct page *page; |
1643 | ||
1644 | /* | |
672bba3a CL |
1645 | * The defrag ratio allows a configuration of the tradeoffs between |
1646 | * inter node defragmentation and node local allocations. A lower | |
1647 | * defrag_ratio increases the tendency to do local allocations | |
1648 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1649 | * |
672bba3a CL |
1650 | * If the defrag_ratio is set to 0 then kmalloc() always |
1651 | * returns node local objects. If the ratio is higher then kmalloc() | |
1652 | * may return off node objects because partial slabs are obtained | |
1653 | * from other nodes and filled up. | |
81819f0f | 1654 | * |
6446faa2 | 1655 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1656 | * defrag_ratio = 1000) then every (well almost) allocation will |
1657 | * first attempt to defrag slab caches on other nodes. This means | |
1658 | * scanning over all nodes to look for partial slabs which may be | |
1659 | * expensive if we do it every time we are trying to find a slab | |
1660 | * with available objects. | |
81819f0f | 1661 | */ |
9824601e CL |
1662 | if (!s->remote_node_defrag_ratio || |
1663 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1664 | return NULL; |
1665 | ||
c0ff7453 | 1666 | get_mems_allowed(); |
0e88460d | 1667 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1668 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1669 | struct kmem_cache_node *n; |
1670 | ||
54a6eb5c | 1671 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1672 | |
54a6eb5c | 1673 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1674 | n->nr_partial > s->min_partial) { |
61728d1e | 1675 | page = get_partial_node(s, n); |
c0ff7453 MX |
1676 | if (page) { |
1677 | put_mems_allowed(); | |
81819f0f | 1678 | return page; |
c0ff7453 | 1679 | } |
81819f0f CL |
1680 | } |
1681 | } | |
c0ff7453 | 1682 | put_mems_allowed(); |
81819f0f CL |
1683 | #endif |
1684 | return NULL; | |
1685 | } | |
1686 | ||
1687 | /* | |
1688 | * Get a partial page, lock it and return it. | |
1689 | */ | |
1690 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1691 | { | |
1692 | struct page *page; | |
2154a336 | 1693 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1694 | |
61728d1e | 1695 | page = get_partial_node(s, get_node(s, searchnode)); |
33de04ec | 1696 | if (page || node != NUMA_NO_NODE) |
81819f0f CL |
1697 | return page; |
1698 | ||
1699 | return get_any_partial(s, flags); | |
1700 | } | |
1701 | ||
8a5ec0ba CL |
1702 | #ifdef CONFIG_PREEMPT |
1703 | /* | |
1704 | * Calculate the next globally unique transaction for disambiguiation | |
1705 | * during cmpxchg. The transactions start with the cpu number and are then | |
1706 | * incremented by CONFIG_NR_CPUS. | |
1707 | */ | |
1708 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1709 | #else | |
1710 | /* | |
1711 | * No preemption supported therefore also no need to check for | |
1712 | * different cpus. | |
1713 | */ | |
1714 | #define TID_STEP 1 | |
1715 | #endif | |
1716 | ||
1717 | static inline unsigned long next_tid(unsigned long tid) | |
1718 | { | |
1719 | return tid + TID_STEP; | |
1720 | } | |
1721 | ||
1722 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1723 | { | |
1724 | return tid % TID_STEP; | |
1725 | } | |
1726 | ||
1727 | static inline unsigned long tid_to_event(unsigned long tid) | |
1728 | { | |
1729 | return tid / TID_STEP; | |
1730 | } | |
1731 | ||
1732 | static inline unsigned int init_tid(int cpu) | |
1733 | { | |
1734 | return cpu; | |
1735 | } | |
1736 | ||
1737 | static inline void note_cmpxchg_failure(const char *n, | |
1738 | const struct kmem_cache *s, unsigned long tid) | |
1739 | { | |
1740 | #ifdef SLUB_DEBUG_CMPXCHG | |
1741 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1742 | ||
1743 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1744 | ||
1745 | #ifdef CONFIG_PREEMPT | |
1746 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1747 | printk("due to cpu change %d -> %d\n", | |
1748 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1749 | else | |
1750 | #endif | |
1751 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1752 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1753 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1754 | else | |
1755 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1756 | actual_tid, tid, next_tid(tid)); | |
1757 | #endif | |
4fdccdfb | 1758 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1759 | } |
1760 | ||
8a5ec0ba CL |
1761 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1762 | { | |
8a5ec0ba CL |
1763 | int cpu; |
1764 | ||
1765 | for_each_possible_cpu(cpu) | |
1766 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1767 | } |
2cfb7455 CL |
1768 | /* |
1769 | * Remove the cpu slab | |
1770 | */ | |
1771 | ||
81819f0f CL |
1772 | /* |
1773 | * Remove the cpu slab | |
1774 | */ | |
dfb4f096 | 1775 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1776 | { |
2cfb7455 | 1777 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
dfb4f096 | 1778 | struct page *page = c->page; |
2cfb7455 CL |
1779 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1780 | int lock = 0; | |
1781 | enum slab_modes l = M_NONE, m = M_NONE; | |
1782 | void *freelist; | |
1783 | void *nextfree; | |
1784 | int tail = 0; | |
1785 | struct page new; | |
1786 | struct page old; | |
1787 | ||
1788 | if (page->freelist) { | |
84e554e6 | 1789 | stat(s, DEACTIVATE_REMOTE_FREES); |
2cfb7455 CL |
1790 | tail = 1; |
1791 | } | |
1792 | ||
1793 | c->tid = next_tid(c->tid); | |
1794 | c->page = NULL; | |
1795 | freelist = c->freelist; | |
1796 | c->freelist = NULL; | |
1797 | ||
894b8788 | 1798 | /* |
2cfb7455 CL |
1799 | * Stage one: Free all available per cpu objects back |
1800 | * to the page freelist while it is still frozen. Leave the | |
1801 | * last one. | |
1802 | * | |
1803 | * There is no need to take the list->lock because the page | |
1804 | * is still frozen. | |
1805 | */ | |
1806 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1807 | void *prior; | |
1808 | unsigned long counters; | |
1809 | ||
1810 | do { | |
1811 | prior = page->freelist; | |
1812 | counters = page->counters; | |
1813 | set_freepointer(s, freelist, prior); | |
1814 | new.counters = counters; | |
1815 | new.inuse--; | |
1816 | VM_BUG_ON(!new.frozen); | |
1817 | ||
1d07171c | 1818 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1819 | prior, counters, |
1820 | freelist, new.counters, | |
1821 | "drain percpu freelist")); | |
1822 | ||
1823 | freelist = nextfree; | |
1824 | } | |
1825 | ||
894b8788 | 1826 | /* |
2cfb7455 CL |
1827 | * Stage two: Ensure that the page is unfrozen while the |
1828 | * list presence reflects the actual number of objects | |
1829 | * during unfreeze. | |
1830 | * | |
1831 | * We setup the list membership and then perform a cmpxchg | |
1832 | * with the count. If there is a mismatch then the page | |
1833 | * is not unfrozen but the page is on the wrong list. | |
1834 | * | |
1835 | * Then we restart the process which may have to remove | |
1836 | * the page from the list that we just put it on again | |
1837 | * because the number of objects in the slab may have | |
1838 | * changed. | |
894b8788 | 1839 | */ |
2cfb7455 | 1840 | redo: |
894b8788 | 1841 | |
2cfb7455 CL |
1842 | old.freelist = page->freelist; |
1843 | old.counters = page->counters; | |
1844 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1845 | |
2cfb7455 CL |
1846 | /* Determine target state of the slab */ |
1847 | new.counters = old.counters; | |
1848 | if (freelist) { | |
1849 | new.inuse--; | |
1850 | set_freepointer(s, freelist, old.freelist); | |
1851 | new.freelist = freelist; | |
1852 | } else | |
1853 | new.freelist = old.freelist; | |
1854 | ||
1855 | new.frozen = 0; | |
1856 | ||
1857 | if (!new.inuse && n->nr_partial < s->min_partial) | |
1858 | m = M_FREE; | |
1859 | else if (new.freelist) { | |
1860 | m = M_PARTIAL; | |
1861 | if (!lock) { | |
1862 | lock = 1; | |
1863 | /* | |
1864 | * Taking the spinlock removes the possiblity | |
1865 | * that acquire_slab() will see a slab page that | |
1866 | * is frozen | |
1867 | */ | |
1868 | spin_lock(&n->list_lock); | |
1869 | } | |
1870 | } else { | |
1871 | m = M_FULL; | |
1872 | if (kmem_cache_debug(s) && !lock) { | |
1873 | lock = 1; | |
1874 | /* | |
1875 | * This also ensures that the scanning of full | |
1876 | * slabs from diagnostic functions will not see | |
1877 | * any frozen slabs. | |
1878 | */ | |
1879 | spin_lock(&n->list_lock); | |
1880 | } | |
1881 | } | |
1882 | ||
1883 | if (l != m) { | |
1884 | ||
1885 | if (l == M_PARTIAL) | |
1886 | ||
1887 | remove_partial(n, page); | |
1888 | ||
1889 | else if (l == M_FULL) | |
894b8788 | 1890 | |
2cfb7455 CL |
1891 | remove_full(s, page); |
1892 | ||
1893 | if (m == M_PARTIAL) { | |
1894 | ||
1895 | add_partial(n, page, tail); | |
1896 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); | |
1897 | ||
1898 | } else if (m == M_FULL) { | |
894b8788 | 1899 | |
2cfb7455 CL |
1900 | stat(s, DEACTIVATE_FULL); |
1901 | add_full(s, n, page); | |
1902 | ||
1903 | } | |
1904 | } | |
1905 | ||
1906 | l = m; | |
1d07171c | 1907 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1908 | old.freelist, old.counters, |
1909 | new.freelist, new.counters, | |
1910 | "unfreezing slab")) | |
1911 | goto redo; | |
1912 | ||
2cfb7455 CL |
1913 | if (lock) |
1914 | spin_unlock(&n->list_lock); | |
1915 | ||
1916 | if (m == M_FREE) { | |
1917 | stat(s, DEACTIVATE_EMPTY); | |
1918 | discard_slab(s, page); | |
1919 | stat(s, FREE_SLAB); | |
894b8788 | 1920 | } |
81819f0f CL |
1921 | } |
1922 | ||
dfb4f096 | 1923 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1924 | { |
84e554e6 | 1925 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 | 1926 | deactivate_slab(s, c); |
81819f0f CL |
1927 | } |
1928 | ||
1929 | /* | |
1930 | * Flush cpu slab. | |
6446faa2 | 1931 | * |
81819f0f CL |
1932 | * Called from IPI handler with interrupts disabled. |
1933 | */ | |
0c710013 | 1934 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1935 | { |
9dfc6e68 | 1936 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 1937 | |
dfb4f096 CL |
1938 | if (likely(c && c->page)) |
1939 | flush_slab(s, c); | |
81819f0f CL |
1940 | } |
1941 | ||
1942 | static void flush_cpu_slab(void *d) | |
1943 | { | |
1944 | struct kmem_cache *s = d; | |
81819f0f | 1945 | |
dfb4f096 | 1946 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
1947 | } |
1948 | ||
1949 | static void flush_all(struct kmem_cache *s) | |
1950 | { | |
15c8b6c1 | 1951 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
1952 | } |
1953 | ||
dfb4f096 CL |
1954 | /* |
1955 | * Check if the objects in a per cpu structure fit numa | |
1956 | * locality expectations. | |
1957 | */ | |
1958 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
1959 | { | |
1960 | #ifdef CONFIG_NUMA | |
2154a336 | 1961 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
1962 | return 0; |
1963 | #endif | |
1964 | return 1; | |
1965 | } | |
1966 | ||
781b2ba6 PE |
1967 | static int count_free(struct page *page) |
1968 | { | |
1969 | return page->objects - page->inuse; | |
1970 | } | |
1971 | ||
1972 | static unsigned long count_partial(struct kmem_cache_node *n, | |
1973 | int (*get_count)(struct page *)) | |
1974 | { | |
1975 | unsigned long flags; | |
1976 | unsigned long x = 0; | |
1977 | struct page *page; | |
1978 | ||
1979 | spin_lock_irqsave(&n->list_lock, flags); | |
1980 | list_for_each_entry(page, &n->partial, lru) | |
1981 | x += get_count(page); | |
1982 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1983 | return x; | |
1984 | } | |
1985 | ||
26c02cf0 AB |
1986 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
1987 | { | |
1988 | #ifdef CONFIG_SLUB_DEBUG | |
1989 | return atomic_long_read(&n->total_objects); | |
1990 | #else | |
1991 | return 0; | |
1992 | #endif | |
1993 | } | |
1994 | ||
781b2ba6 PE |
1995 | static noinline void |
1996 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
1997 | { | |
1998 | int node; | |
1999 | ||
2000 | printk(KERN_WARNING | |
2001 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2002 | nid, gfpflags); | |
2003 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
2004 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
2005 | s->size, oo_order(s->oo), oo_order(s->min)); | |
2006 | ||
fa5ec8a1 DR |
2007 | if (oo_order(s->min) > get_order(s->objsize)) |
2008 | printk(KERN_WARNING " %s debugging increased min order, use " | |
2009 | "slub_debug=O to disable.\n", s->name); | |
2010 | ||
781b2ba6 PE |
2011 | for_each_online_node(node) { |
2012 | struct kmem_cache_node *n = get_node(s, node); | |
2013 | unsigned long nr_slabs; | |
2014 | unsigned long nr_objs; | |
2015 | unsigned long nr_free; | |
2016 | ||
2017 | if (!n) | |
2018 | continue; | |
2019 | ||
26c02cf0 AB |
2020 | nr_free = count_partial(n, count_free); |
2021 | nr_slabs = node_nr_slabs(n); | |
2022 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2023 | |
2024 | printk(KERN_WARNING | |
2025 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2026 | node, nr_slabs, nr_objs, nr_free); | |
2027 | } | |
2028 | } | |
2029 | ||
81819f0f | 2030 | /* |
894b8788 CL |
2031 | * Slow path. The lockless freelist is empty or we need to perform |
2032 | * debugging duties. | |
2033 | * | |
2034 | * Interrupts are disabled. | |
81819f0f | 2035 | * |
894b8788 CL |
2036 | * Processing is still very fast if new objects have been freed to the |
2037 | * regular freelist. In that case we simply take over the regular freelist | |
2038 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2039 | * |
894b8788 CL |
2040 | * If that is not working then we fall back to the partial lists. We take the |
2041 | * first element of the freelist as the object to allocate now and move the | |
2042 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2043 | * |
894b8788 | 2044 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2045 | * we need to allocate a new slab. This is the slowest path since it involves |
2046 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2047 | */ |
ce71e27c EGM |
2048 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2049 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2050 | { |
81819f0f | 2051 | void **object; |
01ad8a7b | 2052 | struct page *page; |
8a5ec0ba | 2053 | unsigned long flags; |
2cfb7455 CL |
2054 | struct page new; |
2055 | unsigned long counters; | |
8a5ec0ba CL |
2056 | |
2057 | local_irq_save(flags); | |
2058 | #ifdef CONFIG_PREEMPT | |
2059 | /* | |
2060 | * We may have been preempted and rescheduled on a different | |
2061 | * cpu before disabling interrupts. Need to reload cpu area | |
2062 | * pointer. | |
2063 | */ | |
2064 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2065 | #endif |
81819f0f | 2066 | |
e72e9c23 LT |
2067 | /* We handle __GFP_ZERO in the caller */ |
2068 | gfpflags &= ~__GFP_ZERO; | |
2069 | ||
01ad8a7b CL |
2070 | page = c->page; |
2071 | if (!page) | |
81819f0f CL |
2072 | goto new_slab; |
2073 | ||
fc59c053 | 2074 | if (unlikely(!node_match(c, node))) { |
e36a2652 | 2075 | stat(s, ALLOC_NODE_MISMATCH); |
fc59c053 CL |
2076 | deactivate_slab(s, c); |
2077 | goto new_slab; | |
2078 | } | |
6446faa2 | 2079 | |
2cfb7455 CL |
2080 | stat(s, ALLOC_SLOWPATH); |
2081 | ||
2082 | do { | |
2083 | object = page->freelist; | |
2084 | counters = page->counters; | |
2085 | new.counters = counters; | |
2cfb7455 CL |
2086 | VM_BUG_ON(!new.frozen); |
2087 | ||
03e404af CL |
2088 | /* |
2089 | * If there is no object left then we use this loop to | |
2090 | * deactivate the slab which is simple since no objects | |
2091 | * are left in the slab and therefore we do not need to | |
2092 | * put the page back onto the partial list. | |
2093 | * | |
2094 | * If there are objects left then we retrieve them | |
2095 | * and use them to refill the per cpu queue. | |
2096 | */ | |
2097 | ||
2098 | new.inuse = page->objects; | |
2099 | new.frozen = object != NULL; | |
2100 | ||
1d07171c | 2101 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2102 | object, counters, |
2103 | NULL, new.counters, | |
2104 | "__slab_alloc")); | |
6446faa2 | 2105 | |
03e404af CL |
2106 | if (unlikely(!object)) { |
2107 | c->page = NULL; | |
2108 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2109 | goto new_slab; |
03e404af | 2110 | } |
6446faa2 | 2111 | |
84e554e6 | 2112 | stat(s, ALLOC_REFILL); |
6446faa2 | 2113 | |
894b8788 | 2114 | load_freelist: |
4eade540 | 2115 | VM_BUG_ON(!page->frozen); |
ff12059e | 2116 | c->freelist = get_freepointer(s, object); |
8a5ec0ba CL |
2117 | c->tid = next_tid(c->tid); |
2118 | local_irq_restore(flags); | |
81819f0f CL |
2119 | return object; |
2120 | ||
81819f0f | 2121 | new_slab: |
01ad8a7b CL |
2122 | page = get_partial(s, gfpflags, node); |
2123 | if (page) { | |
84e554e6 | 2124 | stat(s, ALLOC_FROM_PARTIAL); |
2cfb7455 CL |
2125 | object = c->freelist; |
2126 | ||
2127 | if (kmem_cache_debug(s)) | |
2128 | goto debug; | |
894b8788 | 2129 | goto load_freelist; |
81819f0f CL |
2130 | } |
2131 | ||
01ad8a7b | 2132 | page = new_slab(s, gfpflags, node); |
b811c202 | 2133 | |
01ad8a7b | 2134 | if (page) { |
9dfc6e68 | 2135 | c = __this_cpu_ptr(s->cpu_slab); |
05aa3450 | 2136 | if (c->page) |
dfb4f096 | 2137 | flush_slab(s, c); |
01ad8a7b | 2138 | |
2cfb7455 CL |
2139 | /* |
2140 | * No other reference to the page yet so we can | |
2141 | * muck around with it freely without cmpxchg | |
2142 | */ | |
2143 | object = page->freelist; | |
2144 | page->freelist = NULL; | |
2145 | page->inuse = page->objects; | |
2146 | ||
2147 | stat(s, ALLOC_SLAB); | |
bd07d87f DR |
2148 | c->node = page_to_nid(page); |
2149 | c->page = page; | |
9e577e8b CL |
2150 | |
2151 | if (kmem_cache_debug(s)) | |
2152 | goto debug; | |
4b6f0750 | 2153 | goto load_freelist; |
81819f0f | 2154 | } |
95f85989 PE |
2155 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) |
2156 | slab_out_of_memory(s, gfpflags, node); | |
2fd66c51 | 2157 | local_irq_restore(flags); |
71c7a06f | 2158 | return NULL; |
2cfb7455 | 2159 | |
81819f0f | 2160 | debug: |
2cfb7455 CL |
2161 | if (!object || !alloc_debug_processing(s, page, object, addr)) |
2162 | goto new_slab; | |
894b8788 | 2163 | |
2cfb7455 | 2164 | c->freelist = get_freepointer(s, object); |
442b06bc CL |
2165 | deactivate_slab(s, c); |
2166 | c->page = NULL; | |
15b7c514 | 2167 | c->node = NUMA_NO_NODE; |
a71ae47a CL |
2168 | local_irq_restore(flags); |
2169 | return object; | |
894b8788 CL |
2170 | } |
2171 | ||
2172 | /* | |
2173 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2174 | * have the fastpath folded into their functions. So no function call | |
2175 | * overhead for requests that can be satisfied on the fastpath. | |
2176 | * | |
2177 | * The fastpath works by first checking if the lockless freelist can be used. | |
2178 | * If not then __slab_alloc is called for slow processing. | |
2179 | * | |
2180 | * Otherwise we can simply pick the next object from the lockless free list. | |
2181 | */ | |
06428780 | 2182 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2183 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2184 | { |
894b8788 | 2185 | void **object; |
dfb4f096 | 2186 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2187 | unsigned long tid; |
1f84260c | 2188 | |
c016b0bd | 2189 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2190 | return NULL; |
1f84260c | 2191 | |
8a5ec0ba | 2192 | redo: |
8a5ec0ba CL |
2193 | |
2194 | /* | |
2195 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2196 | * enabled. We may switch back and forth between cpus while | |
2197 | * reading from one cpu area. That does not matter as long | |
2198 | * as we end up on the original cpu again when doing the cmpxchg. | |
2199 | */ | |
9dfc6e68 | 2200 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2201 | |
8a5ec0ba CL |
2202 | /* |
2203 | * The transaction ids are globally unique per cpu and per operation on | |
2204 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2205 | * occurs on the right processor and that there was no operation on the | |
2206 | * linked list in between. | |
2207 | */ | |
2208 | tid = c->tid; | |
2209 | barrier(); | |
8a5ec0ba | 2210 | |
9dfc6e68 | 2211 | object = c->freelist; |
9dfc6e68 | 2212 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 2213 | |
dfb4f096 | 2214 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2215 | |
2216 | else { | |
8a5ec0ba | 2217 | /* |
25985edc | 2218 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2219 | * operation and if we are on the right processor. |
2220 | * | |
2221 | * The cmpxchg does the following atomically (without lock semantics!) | |
2222 | * 1. Relocate first pointer to the current per cpu area. | |
2223 | * 2. Verify that tid and freelist have not been changed | |
2224 | * 3. If they were not changed replace tid and freelist | |
2225 | * | |
2226 | * Since this is without lock semantics the protection is only against | |
2227 | * code executing on this cpu *not* from access by other cpus. | |
2228 | */ | |
30106b8c | 2229 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2230 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2231 | object, tid, | |
1393d9a1 | 2232 | get_freepointer_safe(s, object), next_tid(tid)))) { |
8a5ec0ba CL |
2233 | |
2234 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2235 | goto redo; | |
2236 | } | |
84e554e6 | 2237 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2238 | } |
8a5ec0ba | 2239 | |
74e2134f | 2240 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 2241 | memset(object, 0, s->objsize); |
d07dbea4 | 2242 | |
c016b0bd | 2243 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2244 | |
894b8788 | 2245 | return object; |
81819f0f CL |
2246 | } |
2247 | ||
2248 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2249 | { | |
2154a336 | 2250 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2251 | |
ca2b84cb | 2252 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
2253 | |
2254 | return ret; | |
81819f0f CL |
2255 | } |
2256 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2257 | ||
0f24f128 | 2258 | #ifdef CONFIG_TRACING |
4a92379b RK |
2259 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2260 | { | |
2261 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2262 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2263 | return ret; | |
2264 | } | |
2265 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2266 | ||
2267 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2268 | { |
4a92379b RK |
2269 | void *ret = kmalloc_order(size, flags, order); |
2270 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2271 | return ret; | |
5b882be4 | 2272 | } |
4a92379b | 2273 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2274 | #endif |
2275 | ||
81819f0f CL |
2276 | #ifdef CONFIG_NUMA |
2277 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2278 | { | |
5b882be4 EGM |
2279 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2280 | ||
ca2b84cb EGM |
2281 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2282 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2283 | |
2284 | return ret; | |
81819f0f CL |
2285 | } |
2286 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2287 | |
0f24f128 | 2288 | #ifdef CONFIG_TRACING |
4a92379b | 2289 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2290 | gfp_t gfpflags, |
4a92379b | 2291 | int node, size_t size) |
5b882be4 | 2292 | { |
4a92379b RK |
2293 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2294 | ||
2295 | trace_kmalloc_node(_RET_IP_, ret, | |
2296 | size, s->size, gfpflags, node); | |
2297 | return ret; | |
5b882be4 | 2298 | } |
4a92379b | 2299 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2300 | #endif |
5d1f57e4 | 2301 | #endif |
5b882be4 | 2302 | |
81819f0f | 2303 | /* |
894b8788 CL |
2304 | * Slow patch handling. This may still be called frequently since objects |
2305 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2306 | * |
894b8788 CL |
2307 | * So we still attempt to reduce cache line usage. Just take the slab |
2308 | * lock and free the item. If there is no additional partial page | |
2309 | * handling required then we can return immediately. | |
81819f0f | 2310 | */ |
894b8788 | 2311 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2312 | void *x, unsigned long addr) |
81819f0f CL |
2313 | { |
2314 | void *prior; | |
2315 | void **object = (void *)x; | |
2cfb7455 CL |
2316 | int was_frozen; |
2317 | int inuse; | |
2318 | struct page new; | |
2319 | unsigned long counters; | |
2320 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2321 | unsigned long uninitialized_var(flags); |
81819f0f | 2322 | |
8a5ec0ba | 2323 | stat(s, FREE_SLOWPATH); |
81819f0f | 2324 | |
8dc16c6c | 2325 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
80f08c19 | 2326 | return; |
6446faa2 | 2327 | |
2cfb7455 CL |
2328 | do { |
2329 | prior = page->freelist; | |
2330 | counters = page->counters; | |
2331 | set_freepointer(s, object, prior); | |
2332 | new.counters = counters; | |
2333 | was_frozen = new.frozen; | |
2334 | new.inuse--; | |
2335 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
2336 | n = get_node(s, page_to_nid(page)); | |
2337 | /* | |
2338 | * Speculatively acquire the list_lock. | |
2339 | * If the cmpxchg does not succeed then we may | |
2340 | * drop the list_lock without any processing. | |
2341 | * | |
2342 | * Otherwise the list_lock will synchronize with | |
2343 | * other processors updating the list of slabs. | |
2344 | */ | |
80f08c19 | 2345 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2346 | } |
2347 | inuse = new.inuse; | |
81819f0f | 2348 | |
2cfb7455 CL |
2349 | } while (!cmpxchg_double_slab(s, page, |
2350 | prior, counters, | |
2351 | object, new.counters, | |
2352 | "__slab_free")); | |
81819f0f | 2353 | |
2cfb7455 CL |
2354 | if (likely(!n)) { |
2355 | /* | |
2356 | * The list lock was not taken therefore no list | |
2357 | * activity can be necessary. | |
2358 | */ | |
2359 | if (was_frozen) | |
2360 | stat(s, FREE_FROZEN); | |
80f08c19 | 2361 | return; |
2cfb7455 | 2362 | } |
81819f0f CL |
2363 | |
2364 | /* | |
2cfb7455 CL |
2365 | * was_frozen may have been set after we acquired the list_lock in |
2366 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2367 | */ |
2cfb7455 CL |
2368 | if (was_frozen) |
2369 | stat(s, FREE_FROZEN); | |
2370 | else { | |
2371 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2372 | goto slab_empty; | |
81819f0f | 2373 | |
2cfb7455 CL |
2374 | /* |
2375 | * Objects left in the slab. If it was not on the partial list before | |
2376 | * then add it. | |
2377 | */ | |
2378 | if (unlikely(!prior)) { | |
2379 | remove_full(s, page); | |
2380 | add_partial(n, page, 0); | |
2381 | stat(s, FREE_ADD_PARTIAL); | |
2382 | } | |
8ff12cfc | 2383 | } |
80f08c19 | 2384 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2385 | return; |
2386 | ||
2387 | slab_empty: | |
a973e9dd | 2388 | if (prior) { |
81819f0f | 2389 | /* |
672bba3a | 2390 | * Slab still on the partial list. |
81819f0f | 2391 | */ |
5cc6eee8 | 2392 | remove_partial(n, page); |
84e554e6 | 2393 | stat(s, FREE_REMOVE_PARTIAL); |
8ff12cfc | 2394 | } |
2cfb7455 | 2395 | |
80f08c19 | 2396 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2397 | stat(s, FREE_SLAB); |
81819f0f | 2398 | discard_slab(s, page); |
81819f0f CL |
2399 | } |
2400 | ||
894b8788 CL |
2401 | /* |
2402 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2403 | * can perform fastpath freeing without additional function calls. | |
2404 | * | |
2405 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2406 | * of this processor. This typically the case if we have just allocated | |
2407 | * the item before. | |
2408 | * | |
2409 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2410 | * with all sorts of special processing. | |
2411 | */ | |
06428780 | 2412 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2413 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2414 | { |
2415 | void **object = (void *)x; | |
dfb4f096 | 2416 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2417 | unsigned long tid; |
1f84260c | 2418 | |
c016b0bd CL |
2419 | slab_free_hook(s, x); |
2420 | ||
8a5ec0ba | 2421 | redo: |
a24c5a0e | 2422 | |
8a5ec0ba CL |
2423 | /* |
2424 | * Determine the currently cpus per cpu slab. | |
2425 | * The cpu may change afterward. However that does not matter since | |
2426 | * data is retrieved via this pointer. If we are on the same cpu | |
2427 | * during the cmpxchg then the free will succedd. | |
2428 | */ | |
9dfc6e68 | 2429 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2430 | |
8a5ec0ba CL |
2431 | tid = c->tid; |
2432 | barrier(); | |
c016b0bd | 2433 | |
442b06bc | 2434 | if (likely(page == c->page)) { |
ff12059e | 2435 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2436 | |
30106b8c | 2437 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2438 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2439 | c->freelist, tid, | |
2440 | object, next_tid(tid)))) { | |
2441 | ||
2442 | note_cmpxchg_failure("slab_free", s, tid); | |
2443 | goto redo; | |
2444 | } | |
84e554e6 | 2445 | stat(s, FREE_FASTPATH); |
894b8788 | 2446 | } else |
ff12059e | 2447 | __slab_free(s, page, x, addr); |
894b8788 | 2448 | |
894b8788 CL |
2449 | } |
2450 | ||
81819f0f CL |
2451 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2452 | { | |
77c5e2d0 | 2453 | struct page *page; |
81819f0f | 2454 | |
b49af68f | 2455 | page = virt_to_head_page(x); |
81819f0f | 2456 | |
ce71e27c | 2457 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2458 | |
ca2b84cb | 2459 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2460 | } |
2461 | EXPORT_SYMBOL(kmem_cache_free); | |
2462 | ||
81819f0f | 2463 | /* |
672bba3a CL |
2464 | * Object placement in a slab is made very easy because we always start at |
2465 | * offset 0. If we tune the size of the object to the alignment then we can | |
2466 | * get the required alignment by putting one properly sized object after | |
2467 | * another. | |
81819f0f CL |
2468 | * |
2469 | * Notice that the allocation order determines the sizes of the per cpu | |
2470 | * caches. Each processor has always one slab available for allocations. | |
2471 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2472 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2473 | * locking overhead. |
81819f0f CL |
2474 | */ |
2475 | ||
2476 | /* | |
2477 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2478 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2479 | * and increases the number of allocations possible without having to | |
2480 | * take the list_lock. | |
2481 | */ | |
2482 | static int slub_min_order; | |
114e9e89 | 2483 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2484 | static int slub_min_objects; |
81819f0f CL |
2485 | |
2486 | /* | |
2487 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2488 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2489 | */ |
2490 | static int slub_nomerge; | |
2491 | ||
81819f0f CL |
2492 | /* |
2493 | * Calculate the order of allocation given an slab object size. | |
2494 | * | |
672bba3a CL |
2495 | * The order of allocation has significant impact on performance and other |
2496 | * system components. Generally order 0 allocations should be preferred since | |
2497 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2498 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2499 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2500 | * would be wasted. |
2501 | * | |
2502 | * In order to reach satisfactory performance we must ensure that a minimum | |
2503 | * number of objects is in one slab. Otherwise we may generate too much | |
2504 | * activity on the partial lists which requires taking the list_lock. This is | |
2505 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2506 | * |
672bba3a CL |
2507 | * slub_max_order specifies the order where we begin to stop considering the |
2508 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2509 | * we try to keep the page order as low as possible. So we accept more waste | |
2510 | * of space in favor of a small page order. | |
81819f0f | 2511 | * |
672bba3a CL |
2512 | * Higher order allocations also allow the placement of more objects in a |
2513 | * slab and thereby reduce object handling overhead. If the user has | |
2514 | * requested a higher mininum order then we start with that one instead of | |
2515 | * the smallest order which will fit the object. | |
81819f0f | 2516 | */ |
5e6d444e | 2517 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2518 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2519 | { |
2520 | int order; | |
2521 | int rem; | |
6300ea75 | 2522 | int min_order = slub_min_order; |
81819f0f | 2523 | |
ab9a0f19 | 2524 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2525 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2526 | |
6300ea75 | 2527 | for (order = max(min_order, |
5e6d444e CL |
2528 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2529 | order <= max_order; order++) { | |
81819f0f | 2530 | |
5e6d444e | 2531 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2532 | |
ab9a0f19 | 2533 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2534 | continue; |
2535 | ||
ab9a0f19 | 2536 | rem = (slab_size - reserved) % size; |
81819f0f | 2537 | |
5e6d444e | 2538 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2539 | break; |
2540 | ||
2541 | } | |
672bba3a | 2542 | |
81819f0f CL |
2543 | return order; |
2544 | } | |
2545 | ||
ab9a0f19 | 2546 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2547 | { |
2548 | int order; | |
2549 | int min_objects; | |
2550 | int fraction; | |
e8120ff1 | 2551 | int max_objects; |
5e6d444e CL |
2552 | |
2553 | /* | |
2554 | * Attempt to find best configuration for a slab. This | |
2555 | * works by first attempting to generate a layout with | |
2556 | * the best configuration and backing off gradually. | |
2557 | * | |
2558 | * First we reduce the acceptable waste in a slab. Then | |
2559 | * we reduce the minimum objects required in a slab. | |
2560 | */ | |
2561 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2562 | if (!min_objects) |
2563 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2564 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2565 | min_objects = min(min_objects, max_objects); |
2566 | ||
5e6d444e | 2567 | while (min_objects > 1) { |
c124f5b5 | 2568 | fraction = 16; |
5e6d444e CL |
2569 | while (fraction >= 4) { |
2570 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2571 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2572 | if (order <= slub_max_order) |
2573 | return order; | |
2574 | fraction /= 2; | |
2575 | } | |
5086c389 | 2576 | min_objects--; |
5e6d444e CL |
2577 | } |
2578 | ||
2579 | /* | |
2580 | * We were unable to place multiple objects in a slab. Now | |
2581 | * lets see if we can place a single object there. | |
2582 | */ | |
ab9a0f19 | 2583 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2584 | if (order <= slub_max_order) |
2585 | return order; | |
2586 | ||
2587 | /* | |
2588 | * Doh this slab cannot be placed using slub_max_order. | |
2589 | */ | |
ab9a0f19 | 2590 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2591 | if (order < MAX_ORDER) |
5e6d444e CL |
2592 | return order; |
2593 | return -ENOSYS; | |
2594 | } | |
2595 | ||
81819f0f | 2596 | /* |
672bba3a | 2597 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2598 | */ |
2599 | static unsigned long calculate_alignment(unsigned long flags, | |
2600 | unsigned long align, unsigned long size) | |
2601 | { | |
2602 | /* | |
6446faa2 CL |
2603 | * If the user wants hardware cache aligned objects then follow that |
2604 | * suggestion if the object is sufficiently large. | |
81819f0f | 2605 | * |
6446faa2 CL |
2606 | * The hardware cache alignment cannot override the specified |
2607 | * alignment though. If that is greater then use it. | |
81819f0f | 2608 | */ |
b6210386 NP |
2609 | if (flags & SLAB_HWCACHE_ALIGN) { |
2610 | unsigned long ralign = cache_line_size(); | |
2611 | while (size <= ralign / 2) | |
2612 | ralign /= 2; | |
2613 | align = max(align, ralign); | |
2614 | } | |
81819f0f CL |
2615 | |
2616 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2617 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2618 | |
2619 | return ALIGN(align, sizeof(void *)); | |
2620 | } | |
2621 | ||
5595cffc PE |
2622 | static void |
2623 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2624 | { |
2625 | n->nr_partial = 0; | |
81819f0f CL |
2626 | spin_lock_init(&n->list_lock); |
2627 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2628 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2629 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2630 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2631 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2632 | #endif |
81819f0f CL |
2633 | } |
2634 | ||
55136592 | 2635 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2636 | { |
6c182dc0 CL |
2637 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2638 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2639 | |
8a5ec0ba | 2640 | /* |
d4d84fef CM |
2641 | * Must align to double word boundary for the double cmpxchg |
2642 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2643 | */ |
d4d84fef CM |
2644 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2645 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2646 | |
2647 | if (!s->cpu_slab) | |
2648 | return 0; | |
2649 | ||
2650 | init_kmem_cache_cpus(s); | |
4c93c355 | 2651 | |
8a5ec0ba | 2652 | return 1; |
4c93c355 | 2653 | } |
4c93c355 | 2654 | |
51df1142 CL |
2655 | static struct kmem_cache *kmem_cache_node; |
2656 | ||
81819f0f CL |
2657 | /* |
2658 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2659 | * slab on the node for this slabcache. There are no concurrent accesses | |
2660 | * possible. | |
2661 | * | |
2662 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2663 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2664 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2665 | */ |
55136592 | 2666 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2667 | { |
2668 | struct page *page; | |
2669 | struct kmem_cache_node *n; | |
2670 | ||
51df1142 | 2671 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2672 | |
51df1142 | 2673 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2674 | |
2675 | BUG_ON(!page); | |
a2f92ee7 CL |
2676 | if (page_to_nid(page) != node) { |
2677 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2678 | "node %d\n", node); | |
2679 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2680 | "in order to be able to continue\n"); | |
2681 | } | |
2682 | ||
81819f0f CL |
2683 | n = page->freelist; |
2684 | BUG_ON(!n); | |
51df1142 | 2685 | page->freelist = get_freepointer(kmem_cache_node, n); |
81819f0f | 2686 | page->inuse++; |
8cb0a506 | 2687 | page->frozen = 0; |
51df1142 | 2688 | kmem_cache_node->node[node] = n; |
8ab1372f | 2689 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2690 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2691 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2692 | #endif |
51df1142 CL |
2693 | init_kmem_cache_node(n, kmem_cache_node); |
2694 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2695 | |
7c2e132c | 2696 | add_partial(n, page, 0); |
81819f0f CL |
2697 | } |
2698 | ||
2699 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2700 | { | |
2701 | int node; | |
2702 | ||
f64dc58c | 2703 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2704 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2705 | |
73367bd8 | 2706 | if (n) |
51df1142 CL |
2707 | kmem_cache_free(kmem_cache_node, n); |
2708 | ||
81819f0f CL |
2709 | s->node[node] = NULL; |
2710 | } | |
2711 | } | |
2712 | ||
55136592 | 2713 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2714 | { |
2715 | int node; | |
81819f0f | 2716 | |
f64dc58c | 2717 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2718 | struct kmem_cache_node *n; |
2719 | ||
73367bd8 | 2720 | if (slab_state == DOWN) { |
55136592 | 2721 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2722 | continue; |
2723 | } | |
51df1142 | 2724 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2725 | GFP_KERNEL, node); |
81819f0f | 2726 | |
73367bd8 AD |
2727 | if (!n) { |
2728 | free_kmem_cache_nodes(s); | |
2729 | return 0; | |
81819f0f | 2730 | } |
73367bd8 | 2731 | |
81819f0f | 2732 | s->node[node] = n; |
5595cffc | 2733 | init_kmem_cache_node(n, s); |
81819f0f CL |
2734 | } |
2735 | return 1; | |
2736 | } | |
81819f0f | 2737 | |
c0bdb232 | 2738 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2739 | { |
2740 | if (min < MIN_PARTIAL) | |
2741 | min = MIN_PARTIAL; | |
2742 | else if (min > MAX_PARTIAL) | |
2743 | min = MAX_PARTIAL; | |
2744 | s->min_partial = min; | |
2745 | } | |
2746 | ||
81819f0f CL |
2747 | /* |
2748 | * calculate_sizes() determines the order and the distribution of data within | |
2749 | * a slab object. | |
2750 | */ | |
06b285dc | 2751 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2752 | { |
2753 | unsigned long flags = s->flags; | |
2754 | unsigned long size = s->objsize; | |
2755 | unsigned long align = s->align; | |
834f3d11 | 2756 | int order; |
81819f0f | 2757 | |
d8b42bf5 CL |
2758 | /* |
2759 | * Round up object size to the next word boundary. We can only | |
2760 | * place the free pointer at word boundaries and this determines | |
2761 | * the possible location of the free pointer. | |
2762 | */ | |
2763 | size = ALIGN(size, sizeof(void *)); | |
2764 | ||
2765 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2766 | /* |
2767 | * Determine if we can poison the object itself. If the user of | |
2768 | * the slab may touch the object after free or before allocation | |
2769 | * then we should never poison the object itself. | |
2770 | */ | |
2771 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2772 | !s->ctor) |
81819f0f CL |
2773 | s->flags |= __OBJECT_POISON; |
2774 | else | |
2775 | s->flags &= ~__OBJECT_POISON; | |
2776 | ||
81819f0f CL |
2777 | |
2778 | /* | |
672bba3a | 2779 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2780 | * end of the object and the free pointer. If not then add an |
672bba3a | 2781 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2782 | */ |
2783 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2784 | size += sizeof(void *); | |
41ecc55b | 2785 | #endif |
81819f0f CL |
2786 | |
2787 | /* | |
672bba3a CL |
2788 | * With that we have determined the number of bytes in actual use |
2789 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2790 | */ |
2791 | s->inuse = size; | |
2792 | ||
2793 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2794 | s->ctor)) { |
81819f0f CL |
2795 | /* |
2796 | * Relocate free pointer after the object if it is not | |
2797 | * permitted to overwrite the first word of the object on | |
2798 | * kmem_cache_free. | |
2799 | * | |
2800 | * This is the case if we do RCU, have a constructor or | |
2801 | * destructor or are poisoning the objects. | |
2802 | */ | |
2803 | s->offset = size; | |
2804 | size += sizeof(void *); | |
2805 | } | |
2806 | ||
c12b3c62 | 2807 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2808 | if (flags & SLAB_STORE_USER) |
2809 | /* | |
2810 | * Need to store information about allocs and frees after | |
2811 | * the object. | |
2812 | */ | |
2813 | size += 2 * sizeof(struct track); | |
2814 | ||
be7b3fbc | 2815 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2816 | /* |
2817 | * Add some empty padding so that we can catch | |
2818 | * overwrites from earlier objects rather than let | |
2819 | * tracking information or the free pointer be | |
0211a9c8 | 2820 | * corrupted if a user writes before the start |
81819f0f CL |
2821 | * of the object. |
2822 | */ | |
2823 | size += sizeof(void *); | |
41ecc55b | 2824 | #endif |
672bba3a | 2825 | |
81819f0f CL |
2826 | /* |
2827 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2828 | * user specified and the dynamic determination of cache line size |
2829 | * on bootup. | |
81819f0f CL |
2830 | */ |
2831 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2832 | s->align = align; |
81819f0f CL |
2833 | |
2834 | /* | |
2835 | * SLUB stores one object immediately after another beginning from | |
2836 | * offset 0. In order to align the objects we have to simply size | |
2837 | * each object to conform to the alignment. | |
2838 | */ | |
2839 | size = ALIGN(size, align); | |
2840 | s->size = size; | |
06b285dc CL |
2841 | if (forced_order >= 0) |
2842 | order = forced_order; | |
2843 | else | |
ab9a0f19 | 2844 | order = calculate_order(size, s->reserved); |
81819f0f | 2845 | |
834f3d11 | 2846 | if (order < 0) |
81819f0f CL |
2847 | return 0; |
2848 | ||
b7a49f0d | 2849 | s->allocflags = 0; |
834f3d11 | 2850 | if (order) |
b7a49f0d CL |
2851 | s->allocflags |= __GFP_COMP; |
2852 | ||
2853 | if (s->flags & SLAB_CACHE_DMA) | |
2854 | s->allocflags |= SLUB_DMA; | |
2855 | ||
2856 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2857 | s->allocflags |= __GFP_RECLAIMABLE; | |
2858 | ||
81819f0f CL |
2859 | /* |
2860 | * Determine the number of objects per slab | |
2861 | */ | |
ab9a0f19 LJ |
2862 | s->oo = oo_make(order, size, s->reserved); |
2863 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2864 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2865 | s->max = s->oo; | |
81819f0f | 2866 | |
834f3d11 | 2867 | return !!oo_objects(s->oo); |
81819f0f CL |
2868 | |
2869 | } | |
2870 | ||
55136592 | 2871 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
2872 | const char *name, size_t size, |
2873 | size_t align, unsigned long flags, | |
51cc5068 | 2874 | void (*ctor)(void *)) |
81819f0f CL |
2875 | { |
2876 | memset(s, 0, kmem_size); | |
2877 | s->name = name; | |
2878 | s->ctor = ctor; | |
81819f0f | 2879 | s->objsize = size; |
81819f0f | 2880 | s->align = align; |
ba0268a8 | 2881 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 2882 | s->reserved = 0; |
81819f0f | 2883 | |
da9a638c LJ |
2884 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
2885 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 2886 | |
06b285dc | 2887 | if (!calculate_sizes(s, -1)) |
81819f0f | 2888 | goto error; |
3de47213 DR |
2889 | if (disable_higher_order_debug) { |
2890 | /* | |
2891 | * Disable debugging flags that store metadata if the min slab | |
2892 | * order increased. | |
2893 | */ | |
2894 | if (get_order(s->size) > get_order(s->objsize)) { | |
2895 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
2896 | s->offset = 0; | |
2897 | if (!calculate_sizes(s, -1)) | |
2898 | goto error; | |
2899 | } | |
2900 | } | |
81819f0f | 2901 | |
b789ef51 CL |
2902 | #ifdef CONFIG_CMPXCHG_DOUBLE |
2903 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) | |
2904 | /* Enable fast mode */ | |
2905 | s->flags |= __CMPXCHG_DOUBLE; | |
2906 | #endif | |
2907 | ||
3b89d7d8 DR |
2908 | /* |
2909 | * The larger the object size is, the more pages we want on the partial | |
2910 | * list to avoid pounding the page allocator excessively. | |
2911 | */ | |
c0bdb232 | 2912 | set_min_partial(s, ilog2(s->size)); |
81819f0f CL |
2913 | s->refcount = 1; |
2914 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 2915 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 2916 | #endif |
55136592 | 2917 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 2918 | goto error; |
81819f0f | 2919 | |
55136592 | 2920 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 2921 | return 1; |
ff12059e | 2922 | |
4c93c355 | 2923 | free_kmem_cache_nodes(s); |
81819f0f CL |
2924 | error: |
2925 | if (flags & SLAB_PANIC) | |
2926 | panic("Cannot create slab %s size=%lu realsize=%u " | |
2927 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 2928 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
2929 | s->offset, flags); |
2930 | return 0; | |
2931 | } | |
81819f0f | 2932 | |
81819f0f CL |
2933 | /* |
2934 | * Determine the size of a slab object | |
2935 | */ | |
2936 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
2937 | { | |
2938 | return s->objsize; | |
2939 | } | |
2940 | EXPORT_SYMBOL(kmem_cache_size); | |
2941 | ||
33b12c38 CL |
2942 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2943 | const char *text) | |
2944 | { | |
2945 | #ifdef CONFIG_SLUB_DEBUG | |
2946 | void *addr = page_address(page); | |
2947 | void *p; | |
a5dd5c11 NK |
2948 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
2949 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
2950 | if (!map) |
2951 | return; | |
33b12c38 CL |
2952 | slab_err(s, page, "%s", text); |
2953 | slab_lock(page); | |
33b12c38 | 2954 | |
5f80b13a | 2955 | get_map(s, page, map); |
33b12c38 CL |
2956 | for_each_object(p, s, addr, page->objects) { |
2957 | ||
2958 | if (!test_bit(slab_index(p, s, addr), map)) { | |
2959 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
2960 | p, p - addr); | |
2961 | print_tracking(s, p); | |
2962 | } | |
2963 | } | |
2964 | slab_unlock(page); | |
bbd7d57b | 2965 | kfree(map); |
33b12c38 CL |
2966 | #endif |
2967 | } | |
2968 | ||
81819f0f | 2969 | /* |
599870b1 | 2970 | * Attempt to free all partial slabs on a node. |
81819f0f | 2971 | */ |
599870b1 | 2972 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 2973 | { |
81819f0f CL |
2974 | unsigned long flags; |
2975 | struct page *page, *h; | |
2976 | ||
2977 | spin_lock_irqsave(&n->list_lock, flags); | |
33b12c38 | 2978 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 2979 | if (!page->inuse) { |
5cc6eee8 | 2980 | remove_partial(n, page); |
81819f0f | 2981 | discard_slab(s, page); |
33b12c38 CL |
2982 | } else { |
2983 | list_slab_objects(s, page, | |
2984 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 2985 | } |
33b12c38 | 2986 | } |
81819f0f | 2987 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2988 | } |
2989 | ||
2990 | /* | |
672bba3a | 2991 | * Release all resources used by a slab cache. |
81819f0f | 2992 | */ |
0c710013 | 2993 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
2994 | { |
2995 | int node; | |
2996 | ||
2997 | flush_all(s); | |
9dfc6e68 | 2998 | free_percpu(s->cpu_slab); |
81819f0f | 2999 | /* Attempt to free all objects */ |
f64dc58c | 3000 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3001 | struct kmem_cache_node *n = get_node(s, node); |
3002 | ||
599870b1 CL |
3003 | free_partial(s, n); |
3004 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3005 | return 1; |
3006 | } | |
3007 | free_kmem_cache_nodes(s); | |
3008 | return 0; | |
3009 | } | |
3010 | ||
3011 | /* | |
3012 | * Close a cache and release the kmem_cache structure | |
3013 | * (must be used for caches created using kmem_cache_create) | |
3014 | */ | |
3015 | void kmem_cache_destroy(struct kmem_cache *s) | |
3016 | { | |
3017 | down_write(&slub_lock); | |
3018 | s->refcount--; | |
3019 | if (!s->refcount) { | |
3020 | list_del(&s->list); | |
d629d819 PE |
3021 | if (kmem_cache_close(s)) { |
3022 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3023 | "still has objects.\n", s->name, __func__); | |
3024 | dump_stack(); | |
3025 | } | |
d76b1590 ED |
3026 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3027 | rcu_barrier(); | |
81819f0f | 3028 | sysfs_slab_remove(s); |
2bce6485 CL |
3029 | } |
3030 | up_write(&slub_lock); | |
81819f0f CL |
3031 | } |
3032 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3033 | ||
3034 | /******************************************************************** | |
3035 | * Kmalloc subsystem | |
3036 | *******************************************************************/ | |
3037 | ||
51df1142 | 3038 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3039 | EXPORT_SYMBOL(kmalloc_caches); |
3040 | ||
51df1142 CL |
3041 | static struct kmem_cache *kmem_cache; |
3042 | ||
55136592 | 3043 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3044 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3045 | #endif |
3046 | ||
81819f0f CL |
3047 | static int __init setup_slub_min_order(char *str) |
3048 | { | |
06428780 | 3049 | get_option(&str, &slub_min_order); |
81819f0f CL |
3050 | |
3051 | return 1; | |
3052 | } | |
3053 | ||
3054 | __setup("slub_min_order=", setup_slub_min_order); | |
3055 | ||
3056 | static int __init setup_slub_max_order(char *str) | |
3057 | { | |
06428780 | 3058 | get_option(&str, &slub_max_order); |
818cf590 | 3059 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3060 | |
3061 | return 1; | |
3062 | } | |
3063 | ||
3064 | __setup("slub_max_order=", setup_slub_max_order); | |
3065 | ||
3066 | static int __init setup_slub_min_objects(char *str) | |
3067 | { | |
06428780 | 3068 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3069 | |
3070 | return 1; | |
3071 | } | |
3072 | ||
3073 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3074 | ||
3075 | static int __init setup_slub_nomerge(char *str) | |
3076 | { | |
3077 | slub_nomerge = 1; | |
3078 | return 1; | |
3079 | } | |
3080 | ||
3081 | __setup("slub_nomerge", setup_slub_nomerge); | |
3082 | ||
51df1142 CL |
3083 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3084 | int size, unsigned int flags) | |
81819f0f | 3085 | { |
51df1142 CL |
3086 | struct kmem_cache *s; |
3087 | ||
3088 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3089 | ||
83b519e8 PE |
3090 | /* |
3091 | * This function is called with IRQs disabled during early-boot on | |
3092 | * single CPU so there's no need to take slub_lock here. | |
3093 | */ | |
55136592 | 3094 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3095 | flags, NULL)) |
81819f0f CL |
3096 | goto panic; |
3097 | ||
3098 | list_add(&s->list, &slab_caches); | |
51df1142 | 3099 | return s; |
81819f0f CL |
3100 | |
3101 | panic: | |
3102 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3103 | return NULL; |
81819f0f CL |
3104 | } |
3105 | ||
f1b26339 CL |
3106 | /* |
3107 | * Conversion table for small slabs sizes / 8 to the index in the | |
3108 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3109 | * of two cache sizes there. The size of larger slabs can be determined using | |
3110 | * fls. | |
3111 | */ | |
3112 | static s8 size_index[24] = { | |
3113 | 3, /* 8 */ | |
3114 | 4, /* 16 */ | |
3115 | 5, /* 24 */ | |
3116 | 5, /* 32 */ | |
3117 | 6, /* 40 */ | |
3118 | 6, /* 48 */ | |
3119 | 6, /* 56 */ | |
3120 | 6, /* 64 */ | |
3121 | 1, /* 72 */ | |
3122 | 1, /* 80 */ | |
3123 | 1, /* 88 */ | |
3124 | 1, /* 96 */ | |
3125 | 7, /* 104 */ | |
3126 | 7, /* 112 */ | |
3127 | 7, /* 120 */ | |
3128 | 7, /* 128 */ | |
3129 | 2, /* 136 */ | |
3130 | 2, /* 144 */ | |
3131 | 2, /* 152 */ | |
3132 | 2, /* 160 */ | |
3133 | 2, /* 168 */ | |
3134 | 2, /* 176 */ | |
3135 | 2, /* 184 */ | |
3136 | 2 /* 192 */ | |
3137 | }; | |
3138 | ||
acdfcd04 AK |
3139 | static inline int size_index_elem(size_t bytes) |
3140 | { | |
3141 | return (bytes - 1) / 8; | |
3142 | } | |
3143 | ||
81819f0f CL |
3144 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3145 | { | |
f1b26339 | 3146 | int index; |
81819f0f | 3147 | |
f1b26339 CL |
3148 | if (size <= 192) { |
3149 | if (!size) | |
3150 | return ZERO_SIZE_PTR; | |
81819f0f | 3151 | |
acdfcd04 | 3152 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3153 | } else |
f1b26339 | 3154 | index = fls(size - 1); |
81819f0f CL |
3155 | |
3156 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3157 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3158 | return kmalloc_dma_caches[index]; |
f1b26339 | 3159 | |
81819f0f | 3160 | #endif |
51df1142 | 3161 | return kmalloc_caches[index]; |
81819f0f CL |
3162 | } |
3163 | ||
3164 | void *__kmalloc(size_t size, gfp_t flags) | |
3165 | { | |
aadb4bc4 | 3166 | struct kmem_cache *s; |
5b882be4 | 3167 | void *ret; |
81819f0f | 3168 | |
ffadd4d0 | 3169 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3170 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3171 | |
3172 | s = get_slab(size, flags); | |
3173 | ||
3174 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3175 | return s; |
3176 | ||
2154a336 | 3177 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3178 | |
ca2b84cb | 3179 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3180 | |
3181 | return ret; | |
81819f0f CL |
3182 | } |
3183 | EXPORT_SYMBOL(__kmalloc); | |
3184 | ||
5d1f57e4 | 3185 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3186 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3187 | { | |
b1eeab67 | 3188 | struct page *page; |
e4f7c0b4 | 3189 | void *ptr = NULL; |
f619cfe1 | 3190 | |
b1eeab67 VN |
3191 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3192 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3193 | if (page) |
e4f7c0b4 CM |
3194 | ptr = page_address(page); |
3195 | ||
3196 | kmemleak_alloc(ptr, size, 1, flags); | |
3197 | return ptr; | |
f619cfe1 CL |
3198 | } |
3199 | ||
81819f0f CL |
3200 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3201 | { | |
aadb4bc4 | 3202 | struct kmem_cache *s; |
5b882be4 | 3203 | void *ret; |
81819f0f | 3204 | |
057685cf | 3205 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3206 | ret = kmalloc_large_node(size, flags, node); |
3207 | ||
ca2b84cb EGM |
3208 | trace_kmalloc_node(_RET_IP_, ret, |
3209 | size, PAGE_SIZE << get_order(size), | |
3210 | flags, node); | |
5b882be4 EGM |
3211 | |
3212 | return ret; | |
3213 | } | |
aadb4bc4 CL |
3214 | |
3215 | s = get_slab(size, flags); | |
3216 | ||
3217 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3218 | return s; |
3219 | ||
5b882be4 EGM |
3220 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3221 | ||
ca2b84cb | 3222 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3223 | |
3224 | return ret; | |
81819f0f CL |
3225 | } |
3226 | EXPORT_SYMBOL(__kmalloc_node); | |
3227 | #endif | |
3228 | ||
3229 | size_t ksize(const void *object) | |
3230 | { | |
272c1d21 | 3231 | struct page *page; |
81819f0f | 3232 | |
ef8b4520 | 3233 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3234 | return 0; |
3235 | ||
294a80a8 | 3236 | page = virt_to_head_page(object); |
294a80a8 | 3237 | |
76994412 PE |
3238 | if (unlikely(!PageSlab(page))) { |
3239 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3240 | return PAGE_SIZE << compound_order(page); |
76994412 | 3241 | } |
81819f0f | 3242 | |
b3d41885 | 3243 | return slab_ksize(page->slab); |
81819f0f | 3244 | } |
b1aabecd | 3245 | EXPORT_SYMBOL(ksize); |
81819f0f | 3246 | |
d18a90dd BG |
3247 | #ifdef CONFIG_SLUB_DEBUG |
3248 | bool verify_mem_not_deleted(const void *x) | |
3249 | { | |
3250 | struct page *page; | |
3251 | void *object = (void *)x; | |
3252 | unsigned long flags; | |
3253 | bool rv; | |
3254 | ||
3255 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3256 | return false; | |
3257 | ||
3258 | local_irq_save(flags); | |
3259 | ||
3260 | page = virt_to_head_page(x); | |
3261 | if (unlikely(!PageSlab(page))) { | |
3262 | /* maybe it was from stack? */ | |
3263 | rv = true; | |
3264 | goto out_unlock; | |
3265 | } | |
3266 | ||
3267 | slab_lock(page); | |
3268 | if (on_freelist(page->slab, page, object)) { | |
3269 | object_err(page->slab, page, object, "Object is on free-list"); | |
3270 | rv = false; | |
3271 | } else { | |
3272 | rv = true; | |
3273 | } | |
3274 | slab_unlock(page); | |
3275 | ||
3276 | out_unlock: | |
3277 | local_irq_restore(flags); | |
3278 | return rv; | |
3279 | } | |
3280 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3281 | #endif | |
3282 | ||
81819f0f CL |
3283 | void kfree(const void *x) |
3284 | { | |
81819f0f | 3285 | struct page *page; |
5bb983b0 | 3286 | void *object = (void *)x; |
81819f0f | 3287 | |
2121db74 PE |
3288 | trace_kfree(_RET_IP_, x); |
3289 | ||
2408c550 | 3290 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3291 | return; |
3292 | ||
b49af68f | 3293 | page = virt_to_head_page(x); |
aadb4bc4 | 3294 | if (unlikely(!PageSlab(page))) { |
0937502a | 3295 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3296 | kmemleak_free(x); |
aadb4bc4 CL |
3297 | put_page(page); |
3298 | return; | |
3299 | } | |
ce71e27c | 3300 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3301 | } |
3302 | EXPORT_SYMBOL(kfree); | |
3303 | ||
2086d26a | 3304 | /* |
672bba3a CL |
3305 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3306 | * the remaining slabs by the number of items in use. The slabs with the | |
3307 | * most items in use come first. New allocations will then fill those up | |
3308 | * and thus they can be removed from the partial lists. | |
3309 | * | |
3310 | * The slabs with the least items are placed last. This results in them | |
3311 | * being allocated from last increasing the chance that the last objects | |
3312 | * are freed in them. | |
2086d26a CL |
3313 | */ |
3314 | int kmem_cache_shrink(struct kmem_cache *s) | |
3315 | { | |
3316 | int node; | |
3317 | int i; | |
3318 | struct kmem_cache_node *n; | |
3319 | struct page *page; | |
3320 | struct page *t; | |
205ab99d | 3321 | int objects = oo_objects(s->max); |
2086d26a | 3322 | struct list_head *slabs_by_inuse = |
834f3d11 | 3323 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3324 | unsigned long flags; |
3325 | ||
3326 | if (!slabs_by_inuse) | |
3327 | return -ENOMEM; | |
3328 | ||
3329 | flush_all(s); | |
f64dc58c | 3330 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3331 | n = get_node(s, node); |
3332 | ||
3333 | if (!n->nr_partial) | |
3334 | continue; | |
3335 | ||
834f3d11 | 3336 | for (i = 0; i < objects; i++) |
2086d26a CL |
3337 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3338 | ||
3339 | spin_lock_irqsave(&n->list_lock, flags); | |
3340 | ||
3341 | /* | |
672bba3a | 3342 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3343 | * |
672bba3a CL |
3344 | * Note that concurrent frees may occur while we hold the |
3345 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3346 | */ |
3347 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
881db7fb | 3348 | if (!page->inuse) { |
5cc6eee8 | 3349 | remove_partial(n, page); |
2086d26a CL |
3350 | discard_slab(s, page); |
3351 | } else { | |
fcda3d89 CL |
3352 | list_move(&page->lru, |
3353 | slabs_by_inuse + page->inuse); | |
2086d26a CL |
3354 | } |
3355 | } | |
3356 | ||
2086d26a | 3357 | /* |
672bba3a CL |
3358 | * Rebuild the partial list with the slabs filled up most |
3359 | * first and the least used slabs at the end. | |
2086d26a | 3360 | */ |
834f3d11 | 3361 | for (i = objects - 1; i >= 0; i--) |
2086d26a CL |
3362 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3363 | ||
2086d26a CL |
3364 | spin_unlock_irqrestore(&n->list_lock, flags); |
3365 | } | |
3366 | ||
3367 | kfree(slabs_by_inuse); | |
3368 | return 0; | |
3369 | } | |
3370 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3371 | ||
92a5bbc1 | 3372 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3373 | static int slab_mem_going_offline_callback(void *arg) |
3374 | { | |
3375 | struct kmem_cache *s; | |
3376 | ||
3377 | down_read(&slub_lock); | |
3378 | list_for_each_entry(s, &slab_caches, list) | |
3379 | kmem_cache_shrink(s); | |
3380 | up_read(&slub_lock); | |
3381 | ||
3382 | return 0; | |
3383 | } | |
3384 | ||
3385 | static void slab_mem_offline_callback(void *arg) | |
3386 | { | |
3387 | struct kmem_cache_node *n; | |
3388 | struct kmem_cache *s; | |
3389 | struct memory_notify *marg = arg; | |
3390 | int offline_node; | |
3391 | ||
3392 | offline_node = marg->status_change_nid; | |
3393 | ||
3394 | /* | |
3395 | * If the node still has available memory. we need kmem_cache_node | |
3396 | * for it yet. | |
3397 | */ | |
3398 | if (offline_node < 0) | |
3399 | return; | |
3400 | ||
3401 | down_read(&slub_lock); | |
3402 | list_for_each_entry(s, &slab_caches, list) { | |
3403 | n = get_node(s, offline_node); | |
3404 | if (n) { | |
3405 | /* | |
3406 | * if n->nr_slabs > 0, slabs still exist on the node | |
3407 | * that is going down. We were unable to free them, | |
c9404c9c | 3408 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3409 | * callback. So, we must fail. |
3410 | */ | |
0f389ec6 | 3411 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3412 | |
3413 | s->node[offline_node] = NULL; | |
8de66a0c | 3414 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3415 | } |
3416 | } | |
3417 | up_read(&slub_lock); | |
3418 | } | |
3419 | ||
3420 | static int slab_mem_going_online_callback(void *arg) | |
3421 | { | |
3422 | struct kmem_cache_node *n; | |
3423 | struct kmem_cache *s; | |
3424 | struct memory_notify *marg = arg; | |
3425 | int nid = marg->status_change_nid; | |
3426 | int ret = 0; | |
3427 | ||
3428 | /* | |
3429 | * If the node's memory is already available, then kmem_cache_node is | |
3430 | * already created. Nothing to do. | |
3431 | */ | |
3432 | if (nid < 0) | |
3433 | return 0; | |
3434 | ||
3435 | /* | |
0121c619 | 3436 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3437 | * allocate a kmem_cache_node structure in order to bring the node |
3438 | * online. | |
3439 | */ | |
3440 | down_read(&slub_lock); | |
3441 | list_for_each_entry(s, &slab_caches, list) { | |
3442 | /* | |
3443 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3444 | * since memory is not yet available from the node that | |
3445 | * is brought up. | |
3446 | */ | |
8de66a0c | 3447 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3448 | if (!n) { |
3449 | ret = -ENOMEM; | |
3450 | goto out; | |
3451 | } | |
5595cffc | 3452 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3453 | s->node[nid] = n; |
3454 | } | |
3455 | out: | |
3456 | up_read(&slub_lock); | |
3457 | return ret; | |
3458 | } | |
3459 | ||
3460 | static int slab_memory_callback(struct notifier_block *self, | |
3461 | unsigned long action, void *arg) | |
3462 | { | |
3463 | int ret = 0; | |
3464 | ||
3465 | switch (action) { | |
3466 | case MEM_GOING_ONLINE: | |
3467 | ret = slab_mem_going_online_callback(arg); | |
3468 | break; | |
3469 | case MEM_GOING_OFFLINE: | |
3470 | ret = slab_mem_going_offline_callback(arg); | |
3471 | break; | |
3472 | case MEM_OFFLINE: | |
3473 | case MEM_CANCEL_ONLINE: | |
3474 | slab_mem_offline_callback(arg); | |
3475 | break; | |
3476 | case MEM_ONLINE: | |
3477 | case MEM_CANCEL_OFFLINE: | |
3478 | break; | |
3479 | } | |
dc19f9db KH |
3480 | if (ret) |
3481 | ret = notifier_from_errno(ret); | |
3482 | else | |
3483 | ret = NOTIFY_OK; | |
b9049e23 YG |
3484 | return ret; |
3485 | } | |
3486 | ||
3487 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3488 | ||
81819f0f CL |
3489 | /******************************************************************** |
3490 | * Basic setup of slabs | |
3491 | *******************************************************************/ | |
3492 | ||
51df1142 CL |
3493 | /* |
3494 | * Used for early kmem_cache structures that were allocated using | |
3495 | * the page allocator | |
3496 | */ | |
3497 | ||
3498 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3499 | { | |
3500 | int node; | |
3501 | ||
3502 | list_add(&s->list, &slab_caches); | |
3503 | s->refcount = -1; | |
3504 | ||
3505 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3506 | struct kmem_cache_node *n = get_node(s, node); | |
3507 | struct page *p; | |
3508 | ||
3509 | if (n) { | |
3510 | list_for_each_entry(p, &n->partial, lru) | |
3511 | p->slab = s; | |
3512 | ||
607bf324 | 3513 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3514 | list_for_each_entry(p, &n->full, lru) |
3515 | p->slab = s; | |
3516 | #endif | |
3517 | } | |
3518 | } | |
3519 | } | |
3520 | ||
81819f0f CL |
3521 | void __init kmem_cache_init(void) |
3522 | { | |
3523 | int i; | |
4b356be0 | 3524 | int caches = 0; |
51df1142 CL |
3525 | struct kmem_cache *temp_kmem_cache; |
3526 | int order; | |
51df1142 CL |
3527 | struct kmem_cache *temp_kmem_cache_node; |
3528 | unsigned long kmalloc_size; | |
3529 | ||
3530 | kmem_size = offsetof(struct kmem_cache, node) + | |
3531 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3532 | ||
3533 | /* Allocate two kmem_caches from the page allocator */ | |
3534 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3535 | order = get_order(2 * kmalloc_size); | |
3536 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3537 | ||
81819f0f CL |
3538 | /* |
3539 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3540 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3541 | * kmem_cache_open for slab_state == DOWN. |
3542 | */ | |
51df1142 CL |
3543 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3544 | ||
3545 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3546 | sizeof(struct kmem_cache_node), | |
3547 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3548 | |
0c40ba4f | 3549 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3550 | |
3551 | /* Able to allocate the per node structures */ | |
3552 | slab_state = PARTIAL; | |
3553 | ||
51df1142 CL |
3554 | temp_kmem_cache = kmem_cache; |
3555 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3556 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3557 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3558 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3559 | |
51df1142 CL |
3560 | /* |
3561 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3562 | * kmem_cache_node is separately allocated so no need to | |
3563 | * update any list pointers. | |
3564 | */ | |
3565 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3566 | |
51df1142 CL |
3567 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3568 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3569 | ||
3570 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3571 | ||
3572 | caches++; | |
51df1142 CL |
3573 | kmem_cache_bootstrap_fixup(kmem_cache); |
3574 | caches++; | |
3575 | /* Free temporary boot structure */ | |
3576 | free_pages((unsigned long)temp_kmem_cache, order); | |
3577 | ||
3578 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3579 | |
3580 | /* | |
3581 | * Patch up the size_index table if we have strange large alignment | |
3582 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3583 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3584 | * |
3585 | * Largest permitted alignment is 256 bytes due to the way we | |
3586 | * handle the index determination for the smaller caches. | |
3587 | * | |
3588 | * Make sure that nothing crazy happens if someone starts tinkering | |
3589 | * around with ARCH_KMALLOC_MINALIGN | |
3590 | */ | |
3591 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3592 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3593 | ||
acdfcd04 AK |
3594 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3595 | int elem = size_index_elem(i); | |
3596 | if (elem >= ARRAY_SIZE(size_index)) | |
3597 | break; | |
3598 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3599 | } | |
f1b26339 | 3600 | |
acdfcd04 AK |
3601 | if (KMALLOC_MIN_SIZE == 64) { |
3602 | /* | |
3603 | * The 96 byte size cache is not used if the alignment | |
3604 | * is 64 byte. | |
3605 | */ | |
3606 | for (i = 64 + 8; i <= 96; i += 8) | |
3607 | size_index[size_index_elem(i)] = 7; | |
3608 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3609 | /* |
3610 | * The 192 byte sized cache is not used if the alignment | |
3611 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3612 | * instead. | |
3613 | */ | |
3614 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3615 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3616 | } |
3617 | ||
51df1142 CL |
3618 | /* Caches that are not of the two-to-the-power-of size */ |
3619 | if (KMALLOC_MIN_SIZE <= 32) { | |
3620 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3621 | caches++; | |
3622 | } | |
3623 | ||
3624 | if (KMALLOC_MIN_SIZE <= 64) { | |
3625 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3626 | caches++; | |
3627 | } | |
3628 | ||
3629 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3630 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3631 | caches++; | |
3632 | } | |
3633 | ||
81819f0f CL |
3634 | slab_state = UP; |
3635 | ||
3636 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3637 | if (KMALLOC_MIN_SIZE <= 32) { |
3638 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3639 | BUG_ON(!kmalloc_caches[1]->name); | |
3640 | } | |
3641 | ||
3642 | if (KMALLOC_MIN_SIZE <= 64) { | |
3643 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3644 | BUG_ON(!kmalloc_caches[2]->name); | |
3645 | } | |
3646 | ||
d7278bd7 CL |
3647 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3648 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3649 | ||
3650 | BUG_ON(!s); | |
51df1142 | 3651 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3652 | } |
81819f0f CL |
3653 | |
3654 | #ifdef CONFIG_SMP | |
3655 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3656 | #endif |
81819f0f | 3657 | |
55136592 | 3658 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3659 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3660 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3661 | |
51df1142 | 3662 | if (s && s->size) { |
55136592 CL |
3663 | char *name = kasprintf(GFP_NOWAIT, |
3664 | "dma-kmalloc-%d", s->objsize); | |
3665 | ||
3666 | BUG_ON(!name); | |
51df1142 CL |
3667 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3668 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3669 | } |
3670 | } | |
3671 | #endif | |
3adbefee IM |
3672 | printk(KERN_INFO |
3673 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3674 | " CPUs=%d, Nodes=%d\n", |
3675 | caches, cache_line_size(), | |
81819f0f CL |
3676 | slub_min_order, slub_max_order, slub_min_objects, |
3677 | nr_cpu_ids, nr_node_ids); | |
3678 | } | |
3679 | ||
7e85ee0c PE |
3680 | void __init kmem_cache_init_late(void) |
3681 | { | |
7e85ee0c PE |
3682 | } |
3683 | ||
81819f0f CL |
3684 | /* |
3685 | * Find a mergeable slab cache | |
3686 | */ | |
3687 | static int slab_unmergeable(struct kmem_cache *s) | |
3688 | { | |
3689 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3690 | return 1; | |
3691 | ||
c59def9f | 3692 | if (s->ctor) |
81819f0f CL |
3693 | return 1; |
3694 | ||
8ffa6875 CL |
3695 | /* |
3696 | * We may have set a slab to be unmergeable during bootstrap. | |
3697 | */ | |
3698 | if (s->refcount < 0) | |
3699 | return 1; | |
3700 | ||
81819f0f CL |
3701 | return 0; |
3702 | } | |
3703 | ||
3704 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3705 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3706 | void (*ctor)(void *)) |
81819f0f | 3707 | { |
5b95a4ac | 3708 | struct kmem_cache *s; |
81819f0f CL |
3709 | |
3710 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3711 | return NULL; | |
3712 | ||
c59def9f | 3713 | if (ctor) |
81819f0f CL |
3714 | return NULL; |
3715 | ||
3716 | size = ALIGN(size, sizeof(void *)); | |
3717 | align = calculate_alignment(flags, align, size); | |
3718 | size = ALIGN(size, align); | |
ba0268a8 | 3719 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3720 | |
5b95a4ac | 3721 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3722 | if (slab_unmergeable(s)) |
3723 | continue; | |
3724 | ||
3725 | if (size > s->size) | |
3726 | continue; | |
3727 | ||
ba0268a8 | 3728 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3729 | continue; |
3730 | /* | |
3731 | * Check if alignment is compatible. | |
3732 | * Courtesy of Adrian Drzewiecki | |
3733 | */ | |
06428780 | 3734 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3735 | continue; |
3736 | ||
3737 | if (s->size - size >= sizeof(void *)) | |
3738 | continue; | |
3739 | ||
3740 | return s; | |
3741 | } | |
3742 | return NULL; | |
3743 | } | |
3744 | ||
3745 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3746 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3747 | { |
3748 | struct kmem_cache *s; | |
84c1cf62 | 3749 | char *n; |
81819f0f | 3750 | |
fe1ff49d BH |
3751 | if (WARN_ON(!name)) |
3752 | return NULL; | |
3753 | ||
81819f0f | 3754 | down_write(&slub_lock); |
ba0268a8 | 3755 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3756 | if (s) { |
3757 | s->refcount++; | |
3758 | /* | |
3759 | * Adjust the object sizes so that we clear | |
3760 | * the complete object on kzalloc. | |
3761 | */ | |
3762 | s->objsize = max(s->objsize, (int)size); | |
3763 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3764 | |
7b8f3b66 | 3765 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3766 | s->refcount--; |
81819f0f | 3767 | goto err; |
7b8f3b66 | 3768 | } |
2bce6485 | 3769 | up_write(&slub_lock); |
a0e1d1be CL |
3770 | return s; |
3771 | } | |
6446faa2 | 3772 | |
84c1cf62 PE |
3773 | n = kstrdup(name, GFP_KERNEL); |
3774 | if (!n) | |
3775 | goto err; | |
3776 | ||
a0e1d1be CL |
3777 | s = kmalloc(kmem_size, GFP_KERNEL); |
3778 | if (s) { | |
84c1cf62 | 3779 | if (kmem_cache_open(s, n, |
c59def9f | 3780 | size, align, flags, ctor)) { |
81819f0f | 3781 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3782 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3783 | list_del(&s->list); |
84c1cf62 | 3784 | kfree(n); |
7b8f3b66 | 3785 | kfree(s); |
a0e1d1be | 3786 | goto err; |
7b8f3b66 | 3787 | } |
2bce6485 | 3788 | up_write(&slub_lock); |
a0e1d1be CL |
3789 | return s; |
3790 | } | |
84c1cf62 | 3791 | kfree(n); |
a0e1d1be | 3792 | kfree(s); |
81819f0f | 3793 | } |
68cee4f1 | 3794 | err: |
81819f0f | 3795 | up_write(&slub_lock); |
81819f0f | 3796 | |
81819f0f CL |
3797 | if (flags & SLAB_PANIC) |
3798 | panic("Cannot create slabcache %s\n", name); | |
3799 | else | |
3800 | s = NULL; | |
3801 | return s; | |
3802 | } | |
3803 | EXPORT_SYMBOL(kmem_cache_create); | |
3804 | ||
81819f0f | 3805 | #ifdef CONFIG_SMP |
81819f0f | 3806 | /* |
672bba3a CL |
3807 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3808 | * necessary. | |
81819f0f CL |
3809 | */ |
3810 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3811 | unsigned long action, void *hcpu) | |
3812 | { | |
3813 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3814 | struct kmem_cache *s; |
3815 | unsigned long flags; | |
81819f0f CL |
3816 | |
3817 | switch (action) { | |
3818 | case CPU_UP_CANCELED: | |
8bb78442 | 3819 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3820 | case CPU_DEAD: |
8bb78442 | 3821 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
3822 | down_read(&slub_lock); |
3823 | list_for_each_entry(s, &slab_caches, list) { | |
3824 | local_irq_save(flags); | |
3825 | __flush_cpu_slab(s, cpu); | |
3826 | local_irq_restore(flags); | |
3827 | } | |
3828 | up_read(&slub_lock); | |
81819f0f CL |
3829 | break; |
3830 | default: | |
3831 | break; | |
3832 | } | |
3833 | return NOTIFY_OK; | |
3834 | } | |
3835 | ||
06428780 | 3836 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3837 | .notifier_call = slab_cpuup_callback |
06428780 | 3838 | }; |
81819f0f CL |
3839 | |
3840 | #endif | |
3841 | ||
ce71e27c | 3842 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3843 | { |
aadb4bc4 | 3844 | struct kmem_cache *s; |
94b528d0 | 3845 | void *ret; |
aadb4bc4 | 3846 | |
ffadd4d0 | 3847 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3848 | return kmalloc_large(size, gfpflags); |
3849 | ||
aadb4bc4 | 3850 | s = get_slab(size, gfpflags); |
81819f0f | 3851 | |
2408c550 | 3852 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3853 | return s; |
81819f0f | 3854 | |
2154a336 | 3855 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 3856 | |
25985edc | 3857 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3858 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3859 | |
3860 | return ret; | |
81819f0f CL |
3861 | } |
3862 | ||
5d1f57e4 | 3863 | #ifdef CONFIG_NUMA |
81819f0f | 3864 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3865 | int node, unsigned long caller) |
81819f0f | 3866 | { |
aadb4bc4 | 3867 | struct kmem_cache *s; |
94b528d0 | 3868 | void *ret; |
aadb4bc4 | 3869 | |
d3e14aa3 XF |
3870 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3871 | ret = kmalloc_large_node(size, gfpflags, node); | |
3872 | ||
3873 | trace_kmalloc_node(caller, ret, | |
3874 | size, PAGE_SIZE << get_order(size), | |
3875 | gfpflags, node); | |
3876 | ||
3877 | return ret; | |
3878 | } | |
eada35ef | 3879 | |
aadb4bc4 | 3880 | s = get_slab(size, gfpflags); |
81819f0f | 3881 | |
2408c550 | 3882 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3883 | return s; |
81819f0f | 3884 | |
94b528d0 EGM |
3885 | ret = slab_alloc(s, gfpflags, node, caller); |
3886 | ||
25985edc | 3887 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3888 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3889 | |
3890 | return ret; | |
81819f0f | 3891 | } |
5d1f57e4 | 3892 | #endif |
81819f0f | 3893 | |
ab4d5ed5 | 3894 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3895 | static int count_inuse(struct page *page) |
3896 | { | |
3897 | return page->inuse; | |
3898 | } | |
3899 | ||
3900 | static int count_total(struct page *page) | |
3901 | { | |
3902 | return page->objects; | |
3903 | } | |
ab4d5ed5 | 3904 | #endif |
205ab99d | 3905 | |
ab4d5ed5 | 3906 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3907 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3908 | unsigned long *map) | |
53e15af0 CL |
3909 | { |
3910 | void *p; | |
a973e9dd | 3911 | void *addr = page_address(page); |
53e15af0 CL |
3912 | |
3913 | if (!check_slab(s, page) || | |
3914 | !on_freelist(s, page, NULL)) | |
3915 | return 0; | |
3916 | ||
3917 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3918 | bitmap_zero(map, page->objects); |
53e15af0 | 3919 | |
5f80b13a CL |
3920 | get_map(s, page, map); |
3921 | for_each_object(p, s, addr, page->objects) { | |
3922 | if (test_bit(slab_index(p, s, addr), map)) | |
3923 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3924 | return 0; | |
53e15af0 CL |
3925 | } |
3926 | ||
224a88be | 3927 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3928 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3929 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3930 | return 0; |
3931 | return 1; | |
3932 | } | |
3933 | ||
434e245d CL |
3934 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3935 | unsigned long *map) | |
53e15af0 | 3936 | { |
881db7fb CL |
3937 | slab_lock(page); |
3938 | validate_slab(s, page, map); | |
3939 | slab_unlock(page); | |
53e15af0 CL |
3940 | } |
3941 | ||
434e245d CL |
3942 | static int validate_slab_node(struct kmem_cache *s, |
3943 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3944 | { |
3945 | unsigned long count = 0; | |
3946 | struct page *page; | |
3947 | unsigned long flags; | |
3948 | ||
3949 | spin_lock_irqsave(&n->list_lock, flags); | |
3950 | ||
3951 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3952 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3953 | count++; |
3954 | } | |
3955 | if (count != n->nr_partial) | |
3956 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3957 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3958 | ||
3959 | if (!(s->flags & SLAB_STORE_USER)) | |
3960 | goto out; | |
3961 | ||
3962 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3963 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3964 | count++; |
3965 | } | |
3966 | if (count != atomic_long_read(&n->nr_slabs)) | |
3967 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3968 | "counter=%ld\n", s->name, count, | |
3969 | atomic_long_read(&n->nr_slabs)); | |
3970 | ||
3971 | out: | |
3972 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3973 | return count; | |
3974 | } | |
3975 | ||
434e245d | 3976 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3977 | { |
3978 | int node; | |
3979 | unsigned long count = 0; | |
205ab99d | 3980 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3981 | sizeof(unsigned long), GFP_KERNEL); |
3982 | ||
3983 | if (!map) | |
3984 | return -ENOMEM; | |
53e15af0 CL |
3985 | |
3986 | flush_all(s); | |
f64dc58c | 3987 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3988 | struct kmem_cache_node *n = get_node(s, node); |
3989 | ||
434e245d | 3990 | count += validate_slab_node(s, n, map); |
53e15af0 | 3991 | } |
434e245d | 3992 | kfree(map); |
53e15af0 CL |
3993 | return count; |
3994 | } | |
88a420e4 | 3995 | /* |
672bba3a | 3996 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3997 | * and freed. |
3998 | */ | |
3999 | ||
4000 | struct location { | |
4001 | unsigned long count; | |
ce71e27c | 4002 | unsigned long addr; |
45edfa58 CL |
4003 | long long sum_time; |
4004 | long min_time; | |
4005 | long max_time; | |
4006 | long min_pid; | |
4007 | long max_pid; | |
174596a0 | 4008 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4009 | nodemask_t nodes; |
88a420e4 CL |
4010 | }; |
4011 | ||
4012 | struct loc_track { | |
4013 | unsigned long max; | |
4014 | unsigned long count; | |
4015 | struct location *loc; | |
4016 | }; | |
4017 | ||
4018 | static void free_loc_track(struct loc_track *t) | |
4019 | { | |
4020 | if (t->max) | |
4021 | free_pages((unsigned long)t->loc, | |
4022 | get_order(sizeof(struct location) * t->max)); | |
4023 | } | |
4024 | ||
68dff6a9 | 4025 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4026 | { |
4027 | struct location *l; | |
4028 | int order; | |
4029 | ||
88a420e4 CL |
4030 | order = get_order(sizeof(struct location) * max); |
4031 | ||
68dff6a9 | 4032 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4033 | if (!l) |
4034 | return 0; | |
4035 | ||
4036 | if (t->count) { | |
4037 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4038 | free_loc_track(t); | |
4039 | } | |
4040 | t->max = max; | |
4041 | t->loc = l; | |
4042 | return 1; | |
4043 | } | |
4044 | ||
4045 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4046 | const struct track *track) |
88a420e4 CL |
4047 | { |
4048 | long start, end, pos; | |
4049 | struct location *l; | |
ce71e27c | 4050 | unsigned long caddr; |
45edfa58 | 4051 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4052 | |
4053 | start = -1; | |
4054 | end = t->count; | |
4055 | ||
4056 | for ( ; ; ) { | |
4057 | pos = start + (end - start + 1) / 2; | |
4058 | ||
4059 | /* | |
4060 | * There is nothing at "end". If we end up there | |
4061 | * we need to add something to before end. | |
4062 | */ | |
4063 | if (pos == end) | |
4064 | break; | |
4065 | ||
4066 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4067 | if (track->addr == caddr) { |
4068 | ||
4069 | l = &t->loc[pos]; | |
4070 | l->count++; | |
4071 | if (track->when) { | |
4072 | l->sum_time += age; | |
4073 | if (age < l->min_time) | |
4074 | l->min_time = age; | |
4075 | if (age > l->max_time) | |
4076 | l->max_time = age; | |
4077 | ||
4078 | if (track->pid < l->min_pid) | |
4079 | l->min_pid = track->pid; | |
4080 | if (track->pid > l->max_pid) | |
4081 | l->max_pid = track->pid; | |
4082 | ||
174596a0 RR |
4083 | cpumask_set_cpu(track->cpu, |
4084 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4085 | } |
4086 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4087 | return 1; |
4088 | } | |
4089 | ||
45edfa58 | 4090 | if (track->addr < caddr) |
88a420e4 CL |
4091 | end = pos; |
4092 | else | |
4093 | start = pos; | |
4094 | } | |
4095 | ||
4096 | /* | |
672bba3a | 4097 | * Not found. Insert new tracking element. |
88a420e4 | 4098 | */ |
68dff6a9 | 4099 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4100 | return 0; |
4101 | ||
4102 | l = t->loc + pos; | |
4103 | if (pos < t->count) | |
4104 | memmove(l + 1, l, | |
4105 | (t->count - pos) * sizeof(struct location)); | |
4106 | t->count++; | |
4107 | l->count = 1; | |
45edfa58 CL |
4108 | l->addr = track->addr; |
4109 | l->sum_time = age; | |
4110 | l->min_time = age; | |
4111 | l->max_time = age; | |
4112 | l->min_pid = track->pid; | |
4113 | l->max_pid = track->pid; | |
174596a0 RR |
4114 | cpumask_clear(to_cpumask(l->cpus)); |
4115 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4116 | nodes_clear(l->nodes); |
4117 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4118 | return 1; |
4119 | } | |
4120 | ||
4121 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4122 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4123 | unsigned long *map) |
88a420e4 | 4124 | { |
a973e9dd | 4125 | void *addr = page_address(page); |
88a420e4 CL |
4126 | void *p; |
4127 | ||
39b26464 | 4128 | bitmap_zero(map, page->objects); |
5f80b13a | 4129 | get_map(s, page, map); |
88a420e4 | 4130 | |
224a88be | 4131 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4132 | if (!test_bit(slab_index(p, s, addr), map)) |
4133 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4134 | } |
4135 | ||
4136 | static int list_locations(struct kmem_cache *s, char *buf, | |
4137 | enum track_item alloc) | |
4138 | { | |
e374d483 | 4139 | int len = 0; |
88a420e4 | 4140 | unsigned long i; |
68dff6a9 | 4141 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4142 | int node; |
bbd7d57b ED |
4143 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4144 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4145 | |
bbd7d57b ED |
4146 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4147 | GFP_TEMPORARY)) { | |
4148 | kfree(map); | |
68dff6a9 | 4149 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4150 | } |
88a420e4 CL |
4151 | /* Push back cpu slabs */ |
4152 | flush_all(s); | |
4153 | ||
f64dc58c | 4154 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4155 | struct kmem_cache_node *n = get_node(s, node); |
4156 | unsigned long flags; | |
4157 | struct page *page; | |
4158 | ||
9e86943b | 4159 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4160 | continue; |
4161 | ||
4162 | spin_lock_irqsave(&n->list_lock, flags); | |
4163 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4164 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4165 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4166 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4167 | spin_unlock_irqrestore(&n->list_lock, flags); |
4168 | } | |
4169 | ||
4170 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4171 | struct location *l = &t.loc[i]; |
88a420e4 | 4172 | |
9c246247 | 4173 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4174 | break; |
e374d483 | 4175 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4176 | |
4177 | if (l->addr) | |
62c70bce | 4178 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4179 | else |
e374d483 | 4180 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4181 | |
4182 | if (l->sum_time != l->min_time) { | |
e374d483 | 4183 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4184 | l->min_time, |
4185 | (long)div_u64(l->sum_time, l->count), | |
4186 | l->max_time); | |
45edfa58 | 4187 | } else |
e374d483 | 4188 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4189 | l->min_time); |
4190 | ||
4191 | if (l->min_pid != l->max_pid) | |
e374d483 | 4192 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4193 | l->min_pid, l->max_pid); |
4194 | else | |
e374d483 | 4195 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4196 | l->min_pid); |
4197 | ||
174596a0 RR |
4198 | if (num_online_cpus() > 1 && |
4199 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4200 | len < PAGE_SIZE - 60) { |
4201 | len += sprintf(buf + len, " cpus="); | |
4202 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4203 | to_cpumask(l->cpus)); |
45edfa58 CL |
4204 | } |
4205 | ||
62bc62a8 | 4206 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4207 | len < PAGE_SIZE - 60) { |
4208 | len += sprintf(buf + len, " nodes="); | |
4209 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4210 | l->nodes); |
4211 | } | |
4212 | ||
e374d483 | 4213 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4214 | } |
4215 | ||
4216 | free_loc_track(&t); | |
bbd7d57b | 4217 | kfree(map); |
88a420e4 | 4218 | if (!t.count) |
e374d483 HH |
4219 | len += sprintf(buf, "No data\n"); |
4220 | return len; | |
88a420e4 | 4221 | } |
ab4d5ed5 | 4222 | #endif |
88a420e4 | 4223 | |
a5a84755 CL |
4224 | #ifdef SLUB_RESILIENCY_TEST |
4225 | static void resiliency_test(void) | |
4226 | { | |
4227 | u8 *p; | |
4228 | ||
4229 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4230 | ||
4231 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4232 | printk(KERN_ERR "-----------------------\n"); | |
4233 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4234 | ||
4235 | p = kzalloc(16, GFP_KERNEL); | |
4236 | p[16] = 0x12; | |
4237 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4238 | " 0x12->0x%p\n\n", p + 16); | |
4239 | ||
4240 | validate_slab_cache(kmalloc_caches[4]); | |
4241 | ||
4242 | /* Hmmm... The next two are dangerous */ | |
4243 | p = kzalloc(32, GFP_KERNEL); | |
4244 | p[32 + sizeof(void *)] = 0x34; | |
4245 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4246 | " 0x34 -> -0x%p\n", p); | |
4247 | printk(KERN_ERR | |
4248 | "If allocated object is overwritten then not detectable\n\n"); | |
4249 | ||
4250 | validate_slab_cache(kmalloc_caches[5]); | |
4251 | p = kzalloc(64, GFP_KERNEL); | |
4252 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4253 | *p = 0x56; | |
4254 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4255 | p); | |
4256 | printk(KERN_ERR | |
4257 | "If allocated object is overwritten then not detectable\n\n"); | |
4258 | validate_slab_cache(kmalloc_caches[6]); | |
4259 | ||
4260 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4261 | p = kzalloc(128, GFP_KERNEL); | |
4262 | kfree(p); | |
4263 | *p = 0x78; | |
4264 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4265 | validate_slab_cache(kmalloc_caches[7]); | |
4266 | ||
4267 | p = kzalloc(256, GFP_KERNEL); | |
4268 | kfree(p); | |
4269 | p[50] = 0x9a; | |
4270 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4271 | p); | |
4272 | validate_slab_cache(kmalloc_caches[8]); | |
4273 | ||
4274 | p = kzalloc(512, GFP_KERNEL); | |
4275 | kfree(p); | |
4276 | p[512] = 0xab; | |
4277 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4278 | validate_slab_cache(kmalloc_caches[9]); | |
4279 | } | |
4280 | #else | |
4281 | #ifdef CONFIG_SYSFS | |
4282 | static void resiliency_test(void) {}; | |
4283 | #endif | |
4284 | #endif | |
4285 | ||
ab4d5ed5 | 4286 | #ifdef CONFIG_SYSFS |
81819f0f | 4287 | enum slab_stat_type { |
205ab99d CL |
4288 | SL_ALL, /* All slabs */ |
4289 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4290 | SL_CPU, /* Only slabs used for cpu caches */ | |
4291 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4292 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4293 | }; |
4294 | ||
205ab99d | 4295 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4296 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4297 | #define SO_CPU (1 << SL_CPU) | |
4298 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4299 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4300 | |
62e5c4b4 CG |
4301 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4302 | char *buf, unsigned long flags) | |
81819f0f CL |
4303 | { |
4304 | unsigned long total = 0; | |
81819f0f CL |
4305 | int node; |
4306 | int x; | |
4307 | unsigned long *nodes; | |
4308 | unsigned long *per_cpu; | |
4309 | ||
4310 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4311 | if (!nodes) |
4312 | return -ENOMEM; | |
81819f0f CL |
4313 | per_cpu = nodes + nr_node_ids; |
4314 | ||
205ab99d CL |
4315 | if (flags & SO_CPU) { |
4316 | int cpu; | |
81819f0f | 4317 | |
205ab99d | 4318 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4319 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
dfb4f096 | 4320 | |
205ab99d CL |
4321 | if (!c || c->node < 0) |
4322 | continue; | |
4323 | ||
4324 | if (c->page) { | |
4325 | if (flags & SO_TOTAL) | |
4326 | x = c->page->objects; | |
4327 | else if (flags & SO_OBJECTS) | |
4328 | x = c->page->inuse; | |
81819f0f CL |
4329 | else |
4330 | x = 1; | |
205ab99d | 4331 | |
81819f0f | 4332 | total += x; |
205ab99d | 4333 | nodes[c->node] += x; |
81819f0f | 4334 | } |
205ab99d | 4335 | per_cpu[c->node]++; |
81819f0f CL |
4336 | } |
4337 | } | |
4338 | ||
04d94879 | 4339 | lock_memory_hotplug(); |
ab4d5ed5 | 4340 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4341 | if (flags & SO_ALL) { |
4342 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4343 | struct kmem_cache_node *n = get_node(s, node); | |
4344 | ||
4345 | if (flags & SO_TOTAL) | |
4346 | x = atomic_long_read(&n->total_objects); | |
4347 | else if (flags & SO_OBJECTS) | |
4348 | x = atomic_long_read(&n->total_objects) - | |
4349 | count_partial(n, count_free); | |
81819f0f | 4350 | |
81819f0f | 4351 | else |
205ab99d | 4352 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4353 | total += x; |
4354 | nodes[node] += x; | |
4355 | } | |
4356 | ||
ab4d5ed5 CL |
4357 | } else |
4358 | #endif | |
4359 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4360 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4361 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4362 | |
205ab99d CL |
4363 | if (flags & SO_TOTAL) |
4364 | x = count_partial(n, count_total); | |
4365 | else if (flags & SO_OBJECTS) | |
4366 | x = count_partial(n, count_inuse); | |
81819f0f | 4367 | else |
205ab99d | 4368 | x = n->nr_partial; |
81819f0f CL |
4369 | total += x; |
4370 | nodes[node] += x; | |
4371 | } | |
4372 | } | |
81819f0f CL |
4373 | x = sprintf(buf, "%lu", total); |
4374 | #ifdef CONFIG_NUMA | |
f64dc58c | 4375 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4376 | if (nodes[node]) |
4377 | x += sprintf(buf + x, " N%d=%lu", | |
4378 | node, nodes[node]); | |
4379 | #endif | |
04d94879 | 4380 | unlock_memory_hotplug(); |
81819f0f CL |
4381 | kfree(nodes); |
4382 | return x + sprintf(buf + x, "\n"); | |
4383 | } | |
4384 | ||
ab4d5ed5 | 4385 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4386 | static int any_slab_objects(struct kmem_cache *s) |
4387 | { | |
4388 | int node; | |
81819f0f | 4389 | |
dfb4f096 | 4390 | for_each_online_node(node) { |
81819f0f CL |
4391 | struct kmem_cache_node *n = get_node(s, node); |
4392 | ||
dfb4f096 CL |
4393 | if (!n) |
4394 | continue; | |
4395 | ||
4ea33e2d | 4396 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4397 | return 1; |
4398 | } | |
4399 | return 0; | |
4400 | } | |
ab4d5ed5 | 4401 | #endif |
81819f0f CL |
4402 | |
4403 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4404 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4405 | |
4406 | struct slab_attribute { | |
4407 | struct attribute attr; | |
4408 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4409 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4410 | }; | |
4411 | ||
4412 | #define SLAB_ATTR_RO(_name) \ | |
4413 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4414 | ||
4415 | #define SLAB_ATTR(_name) \ | |
4416 | static struct slab_attribute _name##_attr = \ | |
4417 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4418 | ||
81819f0f CL |
4419 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4420 | { | |
4421 | return sprintf(buf, "%d\n", s->size); | |
4422 | } | |
4423 | SLAB_ATTR_RO(slab_size); | |
4424 | ||
4425 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4426 | { | |
4427 | return sprintf(buf, "%d\n", s->align); | |
4428 | } | |
4429 | SLAB_ATTR_RO(align); | |
4430 | ||
4431 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4432 | { | |
4433 | return sprintf(buf, "%d\n", s->objsize); | |
4434 | } | |
4435 | SLAB_ATTR_RO(object_size); | |
4436 | ||
4437 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4438 | { | |
834f3d11 | 4439 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4440 | } |
4441 | SLAB_ATTR_RO(objs_per_slab); | |
4442 | ||
06b285dc CL |
4443 | static ssize_t order_store(struct kmem_cache *s, |
4444 | const char *buf, size_t length) | |
4445 | { | |
0121c619 CL |
4446 | unsigned long order; |
4447 | int err; | |
4448 | ||
4449 | err = strict_strtoul(buf, 10, &order); | |
4450 | if (err) | |
4451 | return err; | |
06b285dc CL |
4452 | |
4453 | if (order > slub_max_order || order < slub_min_order) | |
4454 | return -EINVAL; | |
4455 | ||
4456 | calculate_sizes(s, order); | |
4457 | return length; | |
4458 | } | |
4459 | ||
81819f0f CL |
4460 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4461 | { | |
834f3d11 | 4462 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4463 | } |
06b285dc | 4464 | SLAB_ATTR(order); |
81819f0f | 4465 | |
73d342b1 DR |
4466 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4467 | { | |
4468 | return sprintf(buf, "%lu\n", s->min_partial); | |
4469 | } | |
4470 | ||
4471 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4472 | size_t length) | |
4473 | { | |
4474 | unsigned long min; | |
4475 | int err; | |
4476 | ||
4477 | err = strict_strtoul(buf, 10, &min); | |
4478 | if (err) | |
4479 | return err; | |
4480 | ||
c0bdb232 | 4481 | set_min_partial(s, min); |
73d342b1 DR |
4482 | return length; |
4483 | } | |
4484 | SLAB_ATTR(min_partial); | |
4485 | ||
81819f0f CL |
4486 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4487 | { | |
62c70bce JP |
4488 | if (!s->ctor) |
4489 | return 0; | |
4490 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4491 | } |
4492 | SLAB_ATTR_RO(ctor); | |
4493 | ||
81819f0f CL |
4494 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4495 | { | |
4496 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4497 | } | |
4498 | SLAB_ATTR_RO(aliases); | |
4499 | ||
81819f0f CL |
4500 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4501 | { | |
d9acf4b7 | 4502 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4503 | } |
4504 | SLAB_ATTR_RO(partial); | |
4505 | ||
4506 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4507 | { | |
d9acf4b7 | 4508 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4509 | } |
4510 | SLAB_ATTR_RO(cpu_slabs); | |
4511 | ||
4512 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4513 | { | |
205ab99d | 4514 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4515 | } |
4516 | SLAB_ATTR_RO(objects); | |
4517 | ||
205ab99d CL |
4518 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4519 | { | |
4520 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4521 | } | |
4522 | SLAB_ATTR_RO(objects_partial); | |
4523 | ||
a5a84755 CL |
4524 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4525 | { | |
4526 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4527 | } | |
4528 | ||
4529 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4530 | const char *buf, size_t length) | |
4531 | { | |
4532 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4533 | if (buf[0] == '1') | |
4534 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4535 | return length; | |
4536 | } | |
4537 | SLAB_ATTR(reclaim_account); | |
4538 | ||
4539 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4540 | { | |
4541 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4542 | } | |
4543 | SLAB_ATTR_RO(hwcache_align); | |
4544 | ||
4545 | #ifdef CONFIG_ZONE_DMA | |
4546 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4547 | { | |
4548 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4549 | } | |
4550 | SLAB_ATTR_RO(cache_dma); | |
4551 | #endif | |
4552 | ||
4553 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4554 | { | |
4555 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4556 | } | |
4557 | SLAB_ATTR_RO(destroy_by_rcu); | |
4558 | ||
ab9a0f19 LJ |
4559 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4560 | { | |
4561 | return sprintf(buf, "%d\n", s->reserved); | |
4562 | } | |
4563 | SLAB_ATTR_RO(reserved); | |
4564 | ||
ab4d5ed5 | 4565 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4566 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4567 | { | |
4568 | return show_slab_objects(s, buf, SO_ALL); | |
4569 | } | |
4570 | SLAB_ATTR_RO(slabs); | |
4571 | ||
205ab99d CL |
4572 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4573 | { | |
4574 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4575 | } | |
4576 | SLAB_ATTR_RO(total_objects); | |
4577 | ||
81819f0f CL |
4578 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4579 | { | |
4580 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4581 | } | |
4582 | ||
4583 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4584 | const char *buf, size_t length) | |
4585 | { | |
4586 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4587 | if (buf[0] == '1') { |
4588 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4589 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4590 | } |
81819f0f CL |
4591 | return length; |
4592 | } | |
4593 | SLAB_ATTR(sanity_checks); | |
4594 | ||
4595 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4596 | { | |
4597 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4598 | } | |
4599 | ||
4600 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4601 | size_t length) | |
4602 | { | |
4603 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4604 | if (buf[0] == '1') { |
4605 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4606 | s->flags |= SLAB_TRACE; |
b789ef51 | 4607 | } |
81819f0f CL |
4608 | return length; |
4609 | } | |
4610 | SLAB_ATTR(trace); | |
4611 | ||
81819f0f CL |
4612 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4613 | { | |
4614 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4615 | } | |
4616 | ||
4617 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4618 | const char *buf, size_t length) | |
4619 | { | |
4620 | if (any_slab_objects(s)) | |
4621 | return -EBUSY; | |
4622 | ||
4623 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4624 | if (buf[0] == '1') { |
4625 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4626 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4627 | } |
06b285dc | 4628 | calculate_sizes(s, -1); |
81819f0f CL |
4629 | return length; |
4630 | } | |
4631 | SLAB_ATTR(red_zone); | |
4632 | ||
4633 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4634 | { | |
4635 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4636 | } | |
4637 | ||
4638 | static ssize_t poison_store(struct kmem_cache *s, | |
4639 | const char *buf, size_t length) | |
4640 | { | |
4641 | if (any_slab_objects(s)) | |
4642 | return -EBUSY; | |
4643 | ||
4644 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4645 | if (buf[0] == '1') { |
4646 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4647 | s->flags |= SLAB_POISON; |
b789ef51 | 4648 | } |
06b285dc | 4649 | calculate_sizes(s, -1); |
81819f0f CL |
4650 | return length; |
4651 | } | |
4652 | SLAB_ATTR(poison); | |
4653 | ||
4654 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4655 | { | |
4656 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4657 | } | |
4658 | ||
4659 | static ssize_t store_user_store(struct kmem_cache *s, | |
4660 | const char *buf, size_t length) | |
4661 | { | |
4662 | if (any_slab_objects(s)) | |
4663 | return -EBUSY; | |
4664 | ||
4665 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4666 | if (buf[0] == '1') { |
4667 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4668 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4669 | } |
06b285dc | 4670 | calculate_sizes(s, -1); |
81819f0f CL |
4671 | return length; |
4672 | } | |
4673 | SLAB_ATTR(store_user); | |
4674 | ||
53e15af0 CL |
4675 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4676 | { | |
4677 | return 0; | |
4678 | } | |
4679 | ||
4680 | static ssize_t validate_store(struct kmem_cache *s, | |
4681 | const char *buf, size_t length) | |
4682 | { | |
434e245d CL |
4683 | int ret = -EINVAL; |
4684 | ||
4685 | if (buf[0] == '1') { | |
4686 | ret = validate_slab_cache(s); | |
4687 | if (ret >= 0) | |
4688 | ret = length; | |
4689 | } | |
4690 | return ret; | |
53e15af0 CL |
4691 | } |
4692 | SLAB_ATTR(validate); | |
a5a84755 CL |
4693 | |
4694 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4695 | { | |
4696 | if (!(s->flags & SLAB_STORE_USER)) | |
4697 | return -ENOSYS; | |
4698 | return list_locations(s, buf, TRACK_ALLOC); | |
4699 | } | |
4700 | SLAB_ATTR_RO(alloc_calls); | |
4701 | ||
4702 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4703 | { | |
4704 | if (!(s->flags & SLAB_STORE_USER)) | |
4705 | return -ENOSYS; | |
4706 | return list_locations(s, buf, TRACK_FREE); | |
4707 | } | |
4708 | SLAB_ATTR_RO(free_calls); | |
4709 | #endif /* CONFIG_SLUB_DEBUG */ | |
4710 | ||
4711 | #ifdef CONFIG_FAILSLAB | |
4712 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4713 | { | |
4714 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4715 | } | |
4716 | ||
4717 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4718 | size_t length) | |
4719 | { | |
4720 | s->flags &= ~SLAB_FAILSLAB; | |
4721 | if (buf[0] == '1') | |
4722 | s->flags |= SLAB_FAILSLAB; | |
4723 | return length; | |
4724 | } | |
4725 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4726 | #endif |
53e15af0 | 4727 | |
2086d26a CL |
4728 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4729 | { | |
4730 | return 0; | |
4731 | } | |
4732 | ||
4733 | static ssize_t shrink_store(struct kmem_cache *s, | |
4734 | const char *buf, size_t length) | |
4735 | { | |
4736 | if (buf[0] == '1') { | |
4737 | int rc = kmem_cache_shrink(s); | |
4738 | ||
4739 | if (rc) | |
4740 | return rc; | |
4741 | } else | |
4742 | return -EINVAL; | |
4743 | return length; | |
4744 | } | |
4745 | SLAB_ATTR(shrink); | |
4746 | ||
81819f0f | 4747 | #ifdef CONFIG_NUMA |
9824601e | 4748 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4749 | { |
9824601e | 4750 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4751 | } |
4752 | ||
9824601e | 4753 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4754 | const char *buf, size_t length) |
4755 | { | |
0121c619 CL |
4756 | unsigned long ratio; |
4757 | int err; | |
4758 | ||
4759 | err = strict_strtoul(buf, 10, &ratio); | |
4760 | if (err) | |
4761 | return err; | |
4762 | ||
e2cb96b7 | 4763 | if (ratio <= 100) |
0121c619 | 4764 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4765 | |
81819f0f CL |
4766 | return length; |
4767 | } | |
9824601e | 4768 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4769 | #endif |
4770 | ||
8ff12cfc | 4771 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4772 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4773 | { | |
4774 | unsigned long sum = 0; | |
4775 | int cpu; | |
4776 | int len; | |
4777 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4778 | ||
4779 | if (!data) | |
4780 | return -ENOMEM; | |
4781 | ||
4782 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4783 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4784 | |
4785 | data[cpu] = x; | |
4786 | sum += x; | |
4787 | } | |
4788 | ||
4789 | len = sprintf(buf, "%lu", sum); | |
4790 | ||
50ef37b9 | 4791 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4792 | for_each_online_cpu(cpu) { |
4793 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4794 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4795 | } |
50ef37b9 | 4796 | #endif |
8ff12cfc CL |
4797 | kfree(data); |
4798 | return len + sprintf(buf + len, "\n"); | |
4799 | } | |
4800 | ||
78eb00cc DR |
4801 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4802 | { | |
4803 | int cpu; | |
4804 | ||
4805 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4806 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4807 | } |
4808 | ||
8ff12cfc CL |
4809 | #define STAT_ATTR(si, text) \ |
4810 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4811 | { \ | |
4812 | return show_stat(s, buf, si); \ | |
4813 | } \ | |
78eb00cc DR |
4814 | static ssize_t text##_store(struct kmem_cache *s, \ |
4815 | const char *buf, size_t length) \ | |
4816 | { \ | |
4817 | if (buf[0] != '0') \ | |
4818 | return -EINVAL; \ | |
4819 | clear_stat(s, si); \ | |
4820 | return length; \ | |
4821 | } \ | |
4822 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4823 | |
4824 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4825 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4826 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4827 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4828 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4829 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4830 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4831 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4832 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4833 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4834 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4835 | STAT_ATTR(FREE_SLAB, free_slab); |
4836 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4837 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4838 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4839 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4840 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4841 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4842 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4843 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4844 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4845 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
8ff12cfc CL |
4846 | #endif |
4847 | ||
06428780 | 4848 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4849 | &slab_size_attr.attr, |
4850 | &object_size_attr.attr, | |
4851 | &objs_per_slab_attr.attr, | |
4852 | &order_attr.attr, | |
73d342b1 | 4853 | &min_partial_attr.attr, |
81819f0f | 4854 | &objects_attr.attr, |
205ab99d | 4855 | &objects_partial_attr.attr, |
81819f0f CL |
4856 | &partial_attr.attr, |
4857 | &cpu_slabs_attr.attr, | |
4858 | &ctor_attr.attr, | |
81819f0f CL |
4859 | &aliases_attr.attr, |
4860 | &align_attr.attr, | |
81819f0f CL |
4861 | &hwcache_align_attr.attr, |
4862 | &reclaim_account_attr.attr, | |
4863 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4864 | &shrink_attr.attr, |
ab9a0f19 | 4865 | &reserved_attr.attr, |
ab4d5ed5 | 4866 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4867 | &total_objects_attr.attr, |
4868 | &slabs_attr.attr, | |
4869 | &sanity_checks_attr.attr, | |
4870 | &trace_attr.attr, | |
81819f0f CL |
4871 | &red_zone_attr.attr, |
4872 | &poison_attr.attr, | |
4873 | &store_user_attr.attr, | |
53e15af0 | 4874 | &validate_attr.attr, |
88a420e4 CL |
4875 | &alloc_calls_attr.attr, |
4876 | &free_calls_attr.attr, | |
ab4d5ed5 | 4877 | #endif |
81819f0f CL |
4878 | #ifdef CONFIG_ZONE_DMA |
4879 | &cache_dma_attr.attr, | |
4880 | #endif | |
4881 | #ifdef CONFIG_NUMA | |
9824601e | 4882 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4883 | #endif |
4884 | #ifdef CONFIG_SLUB_STATS | |
4885 | &alloc_fastpath_attr.attr, | |
4886 | &alloc_slowpath_attr.attr, | |
4887 | &free_fastpath_attr.attr, | |
4888 | &free_slowpath_attr.attr, | |
4889 | &free_frozen_attr.attr, | |
4890 | &free_add_partial_attr.attr, | |
4891 | &free_remove_partial_attr.attr, | |
4892 | &alloc_from_partial_attr.attr, | |
4893 | &alloc_slab_attr.attr, | |
4894 | &alloc_refill_attr.attr, | |
e36a2652 | 4895 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4896 | &free_slab_attr.attr, |
4897 | &cpuslab_flush_attr.attr, | |
4898 | &deactivate_full_attr.attr, | |
4899 | &deactivate_empty_attr.attr, | |
4900 | &deactivate_to_head_attr.attr, | |
4901 | &deactivate_to_tail_attr.attr, | |
4902 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4903 | &deactivate_bypass_attr.attr, |
65c3376a | 4904 | &order_fallback_attr.attr, |
b789ef51 CL |
4905 | &cmpxchg_double_fail_attr.attr, |
4906 | &cmpxchg_double_cpu_fail_attr.attr, | |
81819f0f | 4907 | #endif |
4c13dd3b DM |
4908 | #ifdef CONFIG_FAILSLAB |
4909 | &failslab_attr.attr, | |
4910 | #endif | |
4911 | ||
81819f0f CL |
4912 | NULL |
4913 | }; | |
4914 | ||
4915 | static struct attribute_group slab_attr_group = { | |
4916 | .attrs = slab_attrs, | |
4917 | }; | |
4918 | ||
4919 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4920 | struct attribute *attr, | |
4921 | char *buf) | |
4922 | { | |
4923 | struct slab_attribute *attribute; | |
4924 | struct kmem_cache *s; | |
4925 | int err; | |
4926 | ||
4927 | attribute = to_slab_attr(attr); | |
4928 | s = to_slab(kobj); | |
4929 | ||
4930 | if (!attribute->show) | |
4931 | return -EIO; | |
4932 | ||
4933 | err = attribute->show(s, buf); | |
4934 | ||
4935 | return err; | |
4936 | } | |
4937 | ||
4938 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4939 | struct attribute *attr, | |
4940 | const char *buf, size_t len) | |
4941 | { | |
4942 | struct slab_attribute *attribute; | |
4943 | struct kmem_cache *s; | |
4944 | int err; | |
4945 | ||
4946 | attribute = to_slab_attr(attr); | |
4947 | s = to_slab(kobj); | |
4948 | ||
4949 | if (!attribute->store) | |
4950 | return -EIO; | |
4951 | ||
4952 | err = attribute->store(s, buf, len); | |
4953 | ||
4954 | return err; | |
4955 | } | |
4956 | ||
151c602f CL |
4957 | static void kmem_cache_release(struct kobject *kobj) |
4958 | { | |
4959 | struct kmem_cache *s = to_slab(kobj); | |
4960 | ||
84c1cf62 | 4961 | kfree(s->name); |
151c602f CL |
4962 | kfree(s); |
4963 | } | |
4964 | ||
52cf25d0 | 4965 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
4966 | .show = slab_attr_show, |
4967 | .store = slab_attr_store, | |
4968 | }; | |
4969 | ||
4970 | static struct kobj_type slab_ktype = { | |
4971 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 4972 | .release = kmem_cache_release |
81819f0f CL |
4973 | }; |
4974 | ||
4975 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
4976 | { | |
4977 | struct kobj_type *ktype = get_ktype(kobj); | |
4978 | ||
4979 | if (ktype == &slab_ktype) | |
4980 | return 1; | |
4981 | return 0; | |
4982 | } | |
4983 | ||
9cd43611 | 4984 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
4985 | .filter = uevent_filter, |
4986 | }; | |
4987 | ||
27c3a314 | 4988 | static struct kset *slab_kset; |
81819f0f CL |
4989 | |
4990 | #define ID_STR_LENGTH 64 | |
4991 | ||
4992 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
4993 | * |
4994 | * Format :[flags-]size | |
81819f0f CL |
4995 | */ |
4996 | static char *create_unique_id(struct kmem_cache *s) | |
4997 | { | |
4998 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
4999 | char *p = name; | |
5000 | ||
5001 | BUG_ON(!name); | |
5002 | ||
5003 | *p++ = ':'; | |
5004 | /* | |
5005 | * First flags affecting slabcache operations. We will only | |
5006 | * get here for aliasable slabs so we do not need to support | |
5007 | * too many flags. The flags here must cover all flags that | |
5008 | * are matched during merging to guarantee that the id is | |
5009 | * unique. | |
5010 | */ | |
5011 | if (s->flags & SLAB_CACHE_DMA) | |
5012 | *p++ = 'd'; | |
5013 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5014 | *p++ = 'a'; | |
5015 | if (s->flags & SLAB_DEBUG_FREE) | |
5016 | *p++ = 'F'; | |
5a896d9e VN |
5017 | if (!(s->flags & SLAB_NOTRACK)) |
5018 | *p++ = 't'; | |
81819f0f CL |
5019 | if (p != name + 1) |
5020 | *p++ = '-'; | |
5021 | p += sprintf(p, "%07d", s->size); | |
5022 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5023 | return name; | |
5024 | } | |
5025 | ||
5026 | static int sysfs_slab_add(struct kmem_cache *s) | |
5027 | { | |
5028 | int err; | |
5029 | const char *name; | |
5030 | int unmergeable; | |
5031 | ||
5032 | if (slab_state < SYSFS) | |
5033 | /* Defer until later */ | |
5034 | return 0; | |
5035 | ||
5036 | unmergeable = slab_unmergeable(s); | |
5037 | if (unmergeable) { | |
5038 | /* | |
5039 | * Slabcache can never be merged so we can use the name proper. | |
5040 | * This is typically the case for debug situations. In that | |
5041 | * case we can catch duplicate names easily. | |
5042 | */ | |
27c3a314 | 5043 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5044 | name = s->name; |
5045 | } else { | |
5046 | /* | |
5047 | * Create a unique name for the slab as a target | |
5048 | * for the symlinks. | |
5049 | */ | |
5050 | name = create_unique_id(s); | |
5051 | } | |
5052 | ||
27c3a314 | 5053 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5054 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5055 | if (err) { | |
5056 | kobject_put(&s->kobj); | |
81819f0f | 5057 | return err; |
1eada11c | 5058 | } |
81819f0f CL |
5059 | |
5060 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5061 | if (err) { |
5062 | kobject_del(&s->kobj); | |
5063 | kobject_put(&s->kobj); | |
81819f0f | 5064 | return err; |
5788d8ad | 5065 | } |
81819f0f CL |
5066 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5067 | if (!unmergeable) { | |
5068 | /* Setup first alias */ | |
5069 | sysfs_slab_alias(s, s->name); | |
5070 | kfree(name); | |
5071 | } | |
5072 | return 0; | |
5073 | } | |
5074 | ||
5075 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5076 | { | |
2bce6485 CL |
5077 | if (slab_state < SYSFS) |
5078 | /* | |
5079 | * Sysfs has not been setup yet so no need to remove the | |
5080 | * cache from sysfs. | |
5081 | */ | |
5082 | return; | |
5083 | ||
81819f0f CL |
5084 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5085 | kobject_del(&s->kobj); | |
151c602f | 5086 | kobject_put(&s->kobj); |
81819f0f CL |
5087 | } |
5088 | ||
5089 | /* | |
5090 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5091 | * available lest we lose that information. |
81819f0f CL |
5092 | */ |
5093 | struct saved_alias { | |
5094 | struct kmem_cache *s; | |
5095 | const char *name; | |
5096 | struct saved_alias *next; | |
5097 | }; | |
5098 | ||
5af328a5 | 5099 | static struct saved_alias *alias_list; |
81819f0f CL |
5100 | |
5101 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5102 | { | |
5103 | struct saved_alias *al; | |
5104 | ||
5105 | if (slab_state == SYSFS) { | |
5106 | /* | |
5107 | * If we have a leftover link then remove it. | |
5108 | */ | |
27c3a314 GKH |
5109 | sysfs_remove_link(&slab_kset->kobj, name); |
5110 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5111 | } |
5112 | ||
5113 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5114 | if (!al) | |
5115 | return -ENOMEM; | |
5116 | ||
5117 | al->s = s; | |
5118 | al->name = name; | |
5119 | al->next = alias_list; | |
5120 | alias_list = al; | |
5121 | return 0; | |
5122 | } | |
5123 | ||
5124 | static int __init slab_sysfs_init(void) | |
5125 | { | |
5b95a4ac | 5126 | struct kmem_cache *s; |
81819f0f CL |
5127 | int err; |
5128 | ||
2bce6485 CL |
5129 | down_write(&slub_lock); |
5130 | ||
0ff21e46 | 5131 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5132 | if (!slab_kset) { |
2bce6485 | 5133 | up_write(&slub_lock); |
81819f0f CL |
5134 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5135 | return -ENOSYS; | |
5136 | } | |
5137 | ||
26a7bd03 CL |
5138 | slab_state = SYSFS; |
5139 | ||
5b95a4ac | 5140 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5141 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5142 | if (err) |
5143 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5144 | " to sysfs\n", s->name); | |
26a7bd03 | 5145 | } |
81819f0f CL |
5146 | |
5147 | while (alias_list) { | |
5148 | struct saved_alias *al = alias_list; | |
5149 | ||
5150 | alias_list = alias_list->next; | |
5151 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5152 | if (err) |
5153 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
5154 | " %s to sysfs\n", s->name); | |
81819f0f CL |
5155 | kfree(al); |
5156 | } | |
5157 | ||
2bce6485 | 5158 | up_write(&slub_lock); |
81819f0f CL |
5159 | resiliency_test(); |
5160 | return 0; | |
5161 | } | |
5162 | ||
5163 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5164 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5165 | |
5166 | /* | |
5167 | * The /proc/slabinfo ABI | |
5168 | */ | |
158a9624 | 5169 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5170 | static void print_slabinfo_header(struct seq_file *m) |
5171 | { | |
5172 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
5173 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
5174 | "<objperslab> <pagesperslab>"); | |
5175 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5176 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5177 | seq_putc(m, '\n'); | |
5178 | } | |
5179 | ||
5180 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5181 | { | |
5182 | loff_t n = *pos; | |
5183 | ||
5184 | down_read(&slub_lock); | |
5185 | if (!n) | |
5186 | print_slabinfo_header(m); | |
5187 | ||
5188 | return seq_list_start(&slab_caches, *pos); | |
5189 | } | |
5190 | ||
5191 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5192 | { | |
5193 | return seq_list_next(p, &slab_caches, pos); | |
5194 | } | |
5195 | ||
5196 | static void s_stop(struct seq_file *m, void *p) | |
5197 | { | |
5198 | up_read(&slub_lock); | |
5199 | } | |
5200 | ||
5201 | static int s_show(struct seq_file *m, void *p) | |
5202 | { | |
5203 | unsigned long nr_partials = 0; | |
5204 | unsigned long nr_slabs = 0; | |
5205 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5206 | unsigned long nr_objs = 0; |
5207 | unsigned long nr_free = 0; | |
57ed3eda PE |
5208 | struct kmem_cache *s; |
5209 | int node; | |
5210 | ||
5211 | s = list_entry(p, struct kmem_cache, list); | |
5212 | ||
5213 | for_each_online_node(node) { | |
5214 | struct kmem_cache_node *n = get_node(s, node); | |
5215 | ||
5216 | if (!n) | |
5217 | continue; | |
5218 | ||
5219 | nr_partials += n->nr_partial; | |
5220 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5221 | nr_objs += atomic_long_read(&n->total_objects); |
5222 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5223 | } |
5224 | ||
205ab99d | 5225 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5226 | |
5227 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5228 | nr_objs, s->size, oo_objects(s->oo), |
5229 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5230 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5231 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5232 | 0UL); | |
5233 | seq_putc(m, '\n'); | |
5234 | return 0; | |
5235 | } | |
5236 | ||
7b3c3a50 | 5237 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5238 | .start = s_start, |
5239 | .next = s_next, | |
5240 | .stop = s_stop, | |
5241 | .show = s_show, | |
5242 | }; | |
5243 | ||
7b3c3a50 AD |
5244 | static int slabinfo_open(struct inode *inode, struct file *file) |
5245 | { | |
5246 | return seq_open(file, &slabinfo_op); | |
5247 | } | |
5248 | ||
5249 | static const struct file_operations proc_slabinfo_operations = { | |
5250 | .open = slabinfo_open, | |
5251 | .read = seq_read, | |
5252 | .llseek = seq_lseek, | |
5253 | .release = seq_release, | |
5254 | }; | |
5255 | ||
5256 | static int __init slab_proc_init(void) | |
5257 | { | |
cf5d1131 | 5258 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5259 | return 0; |
5260 | } | |
5261 | module_init(slab_proc_init); | |
158a9624 | 5262 | #endif /* CONFIG_SLABINFO */ |