]>
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; | |
ef62fb32 | 704 | value64 = (value64 & 0xffffffff) | value64 << 32; |
c4089f98 MS |
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; | |
e6e82ea1 | 1450 | page->inuse = page->objects; |
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 | * |
497b66f2 CL |
1557 | * Returns a list of objects or NULL if it fails. |
1558 | * | |
672bba3a | 1559 | * Must hold list_lock. |
81819f0f | 1560 | */ |
497b66f2 | 1561 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1562 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1563 | int mode) |
81819f0f | 1564 | { |
2cfb7455 CL |
1565 | void *freelist; |
1566 | unsigned long counters; | |
1567 | struct page new; | |
1568 | ||
2cfb7455 CL |
1569 | /* |
1570 | * Zap the freelist and set the frozen bit. | |
1571 | * The old freelist is the list of objects for the | |
1572 | * per cpu allocation list. | |
1573 | */ | |
1574 | do { | |
1575 | freelist = page->freelist; | |
1576 | counters = page->counters; | |
1577 | new.counters = counters; | |
49e22585 CL |
1578 | if (mode) |
1579 | new.inuse = page->objects; | |
2cfb7455 CL |
1580 | |
1581 | VM_BUG_ON(new.frozen); | |
1582 | new.frozen = 1; | |
1583 | ||
1d07171c | 1584 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1585 | freelist, counters, |
1586 | NULL, new.counters, | |
1587 | "lock and freeze")); | |
1588 | ||
1589 | remove_partial(n, page); | |
49e22585 | 1590 | return freelist; |
81819f0f CL |
1591 | } |
1592 | ||
49e22585 CL |
1593 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
1594 | ||
81819f0f | 1595 | /* |
672bba3a | 1596 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1597 | */ |
497b66f2 | 1598 | static void *get_partial_node(struct kmem_cache *s, |
acd19fd1 | 1599 | struct kmem_cache_node *n, struct kmem_cache_cpu *c) |
81819f0f | 1600 | { |
49e22585 CL |
1601 | struct page *page, *page2; |
1602 | void *object = NULL; | |
1603 | int count = 0; | |
81819f0f CL |
1604 | |
1605 | /* | |
1606 | * Racy check. If we mistakenly see no partial slabs then we | |
1607 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1608 | * partial slab and there is none available then get_partials() |
1609 | * will return NULL. | |
81819f0f CL |
1610 | */ |
1611 | if (!n || !n->nr_partial) | |
1612 | return NULL; | |
1613 | ||
1614 | spin_lock(&n->list_lock); | |
49e22585 CL |
1615 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
1616 | void *t = acquire_slab(s, n, page, count == 0); | |
1617 | int available; | |
1618 | ||
1619 | if (!t) | |
1620 | break; | |
1621 | ||
1622 | if (!count) { | |
1623 | c->page = page; | |
1624 | c->node = page_to_nid(page); | |
1625 | stat(s, ALLOC_FROM_PARTIAL); | |
1626 | count++; | |
1627 | object = t; | |
1628 | available = page->objects - page->inuse; | |
1629 | } else { | |
1630 | page->freelist = t; | |
1631 | available = put_cpu_partial(s, page, 0); | |
1632 | } | |
1633 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1634 | break; | |
1635 | ||
497b66f2 | 1636 | } |
81819f0f | 1637 | spin_unlock(&n->list_lock); |
497b66f2 | 1638 | return object; |
81819f0f CL |
1639 | } |
1640 | ||
1641 | /* | |
672bba3a | 1642 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1643 | */ |
acd19fd1 CL |
1644 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags, |
1645 | struct kmem_cache_cpu *c) | |
81819f0f CL |
1646 | { |
1647 | #ifdef CONFIG_NUMA | |
1648 | struct zonelist *zonelist; | |
dd1a239f | 1649 | struct zoneref *z; |
54a6eb5c MG |
1650 | struct zone *zone; |
1651 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1652 | void *object; |
81819f0f CL |
1653 | |
1654 | /* | |
672bba3a CL |
1655 | * The defrag ratio allows a configuration of the tradeoffs between |
1656 | * inter node defragmentation and node local allocations. A lower | |
1657 | * defrag_ratio increases the tendency to do local allocations | |
1658 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1659 | * |
672bba3a CL |
1660 | * If the defrag_ratio is set to 0 then kmalloc() always |
1661 | * returns node local objects. If the ratio is higher then kmalloc() | |
1662 | * may return off node objects because partial slabs are obtained | |
1663 | * from other nodes and filled up. | |
81819f0f | 1664 | * |
6446faa2 | 1665 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1666 | * defrag_ratio = 1000) then every (well almost) allocation will |
1667 | * first attempt to defrag slab caches on other nodes. This means | |
1668 | * scanning over all nodes to look for partial slabs which may be | |
1669 | * expensive if we do it every time we are trying to find a slab | |
1670 | * with available objects. | |
81819f0f | 1671 | */ |
9824601e CL |
1672 | if (!s->remote_node_defrag_ratio || |
1673 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1674 | return NULL; |
1675 | ||
c0ff7453 | 1676 | get_mems_allowed(); |
0e88460d | 1677 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
54a6eb5c | 1678 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
81819f0f CL |
1679 | struct kmem_cache_node *n; |
1680 | ||
54a6eb5c | 1681 | n = get_node(s, zone_to_nid(zone)); |
81819f0f | 1682 | |
54a6eb5c | 1683 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && |
3b89d7d8 | 1684 | n->nr_partial > s->min_partial) { |
497b66f2 CL |
1685 | object = get_partial_node(s, n, c); |
1686 | if (object) { | |
c0ff7453 | 1687 | put_mems_allowed(); |
497b66f2 | 1688 | return object; |
c0ff7453 | 1689 | } |
81819f0f CL |
1690 | } |
1691 | } | |
c0ff7453 | 1692 | put_mems_allowed(); |
81819f0f CL |
1693 | #endif |
1694 | return NULL; | |
1695 | } | |
1696 | ||
1697 | /* | |
1698 | * Get a partial page, lock it and return it. | |
1699 | */ | |
497b66f2 | 1700 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1701 | struct kmem_cache_cpu *c) |
81819f0f | 1702 | { |
497b66f2 | 1703 | void *object; |
2154a336 | 1704 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1705 | |
497b66f2 CL |
1706 | object = get_partial_node(s, get_node(s, searchnode), c); |
1707 | if (object || node != NUMA_NO_NODE) | |
1708 | return object; | |
81819f0f | 1709 | |
acd19fd1 | 1710 | return get_any_partial(s, flags, c); |
81819f0f CL |
1711 | } |
1712 | ||
8a5ec0ba CL |
1713 | #ifdef CONFIG_PREEMPT |
1714 | /* | |
1715 | * Calculate the next globally unique transaction for disambiguiation | |
1716 | * during cmpxchg. The transactions start with the cpu number and are then | |
1717 | * incremented by CONFIG_NR_CPUS. | |
1718 | */ | |
1719 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1720 | #else | |
1721 | /* | |
1722 | * No preemption supported therefore also no need to check for | |
1723 | * different cpus. | |
1724 | */ | |
1725 | #define TID_STEP 1 | |
1726 | #endif | |
1727 | ||
1728 | static inline unsigned long next_tid(unsigned long tid) | |
1729 | { | |
1730 | return tid + TID_STEP; | |
1731 | } | |
1732 | ||
1733 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1734 | { | |
1735 | return tid % TID_STEP; | |
1736 | } | |
1737 | ||
1738 | static inline unsigned long tid_to_event(unsigned long tid) | |
1739 | { | |
1740 | return tid / TID_STEP; | |
1741 | } | |
1742 | ||
1743 | static inline unsigned int init_tid(int cpu) | |
1744 | { | |
1745 | return cpu; | |
1746 | } | |
1747 | ||
1748 | static inline void note_cmpxchg_failure(const char *n, | |
1749 | const struct kmem_cache *s, unsigned long tid) | |
1750 | { | |
1751 | #ifdef SLUB_DEBUG_CMPXCHG | |
1752 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1753 | ||
1754 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1755 | ||
1756 | #ifdef CONFIG_PREEMPT | |
1757 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1758 | printk("due to cpu change %d -> %d\n", | |
1759 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1760 | else | |
1761 | #endif | |
1762 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1763 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1764 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1765 | else | |
1766 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1767 | actual_tid, tid, next_tid(tid)); | |
1768 | #endif | |
4fdccdfb | 1769 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1770 | } |
1771 | ||
8a5ec0ba CL |
1772 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1773 | { | |
8a5ec0ba CL |
1774 | int cpu; |
1775 | ||
1776 | for_each_possible_cpu(cpu) | |
1777 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1778 | } |
2cfb7455 | 1779 | |
81819f0f CL |
1780 | /* |
1781 | * Remove the cpu slab | |
1782 | */ | |
dfb4f096 | 1783 | static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1784 | { |
2cfb7455 | 1785 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
dfb4f096 | 1786 | struct page *page = c->page; |
2cfb7455 CL |
1787 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1788 | int lock = 0; | |
1789 | enum slab_modes l = M_NONE, m = M_NONE; | |
1790 | void *freelist; | |
1791 | void *nextfree; | |
1792 | int tail = 0; | |
1793 | struct page new; | |
1794 | struct page old; | |
1795 | ||
1796 | if (page->freelist) { | |
84e554e6 | 1797 | stat(s, DEACTIVATE_REMOTE_FREES); |
2cfb7455 CL |
1798 | tail = 1; |
1799 | } | |
1800 | ||
1801 | c->tid = next_tid(c->tid); | |
1802 | c->page = NULL; | |
1803 | freelist = c->freelist; | |
1804 | c->freelist = NULL; | |
1805 | ||
894b8788 | 1806 | /* |
2cfb7455 CL |
1807 | * Stage one: Free all available per cpu objects back |
1808 | * to the page freelist while it is still frozen. Leave the | |
1809 | * last one. | |
1810 | * | |
1811 | * There is no need to take the list->lock because the page | |
1812 | * is still frozen. | |
1813 | */ | |
1814 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1815 | void *prior; | |
1816 | unsigned long counters; | |
1817 | ||
1818 | do { | |
1819 | prior = page->freelist; | |
1820 | counters = page->counters; | |
1821 | set_freepointer(s, freelist, prior); | |
1822 | new.counters = counters; | |
1823 | new.inuse--; | |
1824 | VM_BUG_ON(!new.frozen); | |
1825 | ||
1d07171c | 1826 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1827 | prior, counters, |
1828 | freelist, new.counters, | |
1829 | "drain percpu freelist")); | |
1830 | ||
1831 | freelist = nextfree; | |
1832 | } | |
1833 | ||
894b8788 | 1834 | /* |
2cfb7455 CL |
1835 | * Stage two: Ensure that the page is unfrozen while the |
1836 | * list presence reflects the actual number of objects | |
1837 | * during unfreeze. | |
1838 | * | |
1839 | * We setup the list membership and then perform a cmpxchg | |
1840 | * with the count. If there is a mismatch then the page | |
1841 | * is not unfrozen but the page is on the wrong list. | |
1842 | * | |
1843 | * Then we restart the process which may have to remove | |
1844 | * the page from the list that we just put it on again | |
1845 | * because the number of objects in the slab may have | |
1846 | * changed. | |
894b8788 | 1847 | */ |
2cfb7455 | 1848 | redo: |
894b8788 | 1849 | |
2cfb7455 CL |
1850 | old.freelist = page->freelist; |
1851 | old.counters = page->counters; | |
1852 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1853 | |
2cfb7455 CL |
1854 | /* Determine target state of the slab */ |
1855 | new.counters = old.counters; | |
1856 | if (freelist) { | |
1857 | new.inuse--; | |
1858 | set_freepointer(s, freelist, old.freelist); | |
1859 | new.freelist = freelist; | |
1860 | } else | |
1861 | new.freelist = old.freelist; | |
1862 | ||
1863 | new.frozen = 0; | |
1864 | ||
81107188 | 1865 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1866 | m = M_FREE; |
1867 | else if (new.freelist) { | |
1868 | m = M_PARTIAL; | |
1869 | if (!lock) { | |
1870 | lock = 1; | |
1871 | /* | |
1872 | * Taking the spinlock removes the possiblity | |
1873 | * that acquire_slab() will see a slab page that | |
1874 | * is frozen | |
1875 | */ | |
1876 | spin_lock(&n->list_lock); | |
1877 | } | |
1878 | } else { | |
1879 | m = M_FULL; | |
1880 | if (kmem_cache_debug(s) && !lock) { | |
1881 | lock = 1; | |
1882 | /* | |
1883 | * This also ensures that the scanning of full | |
1884 | * slabs from diagnostic functions will not see | |
1885 | * any frozen slabs. | |
1886 | */ | |
1887 | spin_lock(&n->list_lock); | |
1888 | } | |
1889 | } | |
1890 | ||
1891 | if (l != m) { | |
1892 | ||
1893 | if (l == M_PARTIAL) | |
1894 | ||
1895 | remove_partial(n, page); | |
1896 | ||
1897 | else if (l == M_FULL) | |
894b8788 | 1898 | |
2cfb7455 CL |
1899 | remove_full(s, page); |
1900 | ||
1901 | if (m == M_PARTIAL) { | |
1902 | ||
1903 | add_partial(n, page, tail); | |
1904 | stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD); | |
1905 | ||
1906 | } else if (m == M_FULL) { | |
894b8788 | 1907 | |
2cfb7455 CL |
1908 | stat(s, DEACTIVATE_FULL); |
1909 | add_full(s, n, page); | |
1910 | ||
1911 | } | |
1912 | } | |
1913 | ||
1914 | l = m; | |
1d07171c | 1915 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1916 | old.freelist, old.counters, |
1917 | new.freelist, new.counters, | |
1918 | "unfreezing slab")) | |
1919 | goto redo; | |
1920 | ||
2cfb7455 CL |
1921 | if (lock) |
1922 | spin_unlock(&n->list_lock); | |
1923 | ||
1924 | if (m == M_FREE) { | |
1925 | stat(s, DEACTIVATE_EMPTY); | |
1926 | discard_slab(s, page); | |
1927 | stat(s, FREE_SLAB); | |
894b8788 | 1928 | } |
81819f0f CL |
1929 | } |
1930 | ||
49e22585 CL |
1931 | /* Unfreeze all the cpu partial slabs */ |
1932 | static void unfreeze_partials(struct kmem_cache *s) | |
1933 | { | |
1934 | struct kmem_cache_node *n = NULL; | |
1935 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); | |
1936 | struct page *page; | |
1937 | ||
1938 | while ((page = c->partial)) { | |
1939 | enum slab_modes { M_PARTIAL, M_FREE }; | |
1940 | enum slab_modes l, m; | |
1941 | struct page new; | |
1942 | struct page old; | |
1943 | ||
1944 | c->partial = page->next; | |
1945 | l = M_FREE; | |
1946 | ||
1947 | do { | |
1948 | ||
1949 | old.freelist = page->freelist; | |
1950 | old.counters = page->counters; | |
1951 | VM_BUG_ON(!old.frozen); | |
1952 | ||
1953 | new.counters = old.counters; | |
1954 | new.freelist = old.freelist; | |
1955 | ||
1956 | new.frozen = 0; | |
1957 | ||
1958 | if (!new.inuse && (!n || n->nr_partial < s->min_partial)) | |
1959 | m = M_FREE; | |
1960 | else { | |
1961 | struct kmem_cache_node *n2 = get_node(s, | |
1962 | page_to_nid(page)); | |
1963 | ||
1964 | m = M_PARTIAL; | |
1965 | if (n != n2) { | |
1966 | if (n) | |
1967 | spin_unlock(&n->list_lock); | |
1968 | ||
1969 | n = n2; | |
1970 | spin_lock(&n->list_lock); | |
1971 | } | |
1972 | } | |
1973 | ||
1974 | if (l != m) { | |
1975 | if (l == M_PARTIAL) | |
1976 | remove_partial(n, page); | |
1977 | else | |
1978 | add_partial(n, page, 1); | |
1979 | ||
1980 | l = m; | |
1981 | } | |
1982 | ||
1983 | } while (!cmpxchg_double_slab(s, page, | |
1984 | old.freelist, old.counters, | |
1985 | new.freelist, new.counters, | |
1986 | "unfreezing slab")); | |
1987 | ||
1988 | if (m == M_FREE) { | |
1989 | stat(s, DEACTIVATE_EMPTY); | |
1990 | discard_slab(s, page); | |
1991 | stat(s, FREE_SLAB); | |
1992 | } | |
1993 | } | |
1994 | ||
1995 | if (n) | |
1996 | spin_unlock(&n->list_lock); | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2001 | * slot if available. This is done without interrupts disabled and without | |
2002 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2003 | * onto a random cpus partial slot. | |
2004 | * | |
2005 | * If we did not find a slot then simply move all the partials to the | |
2006 | * per node partial list. | |
2007 | */ | |
2008 | int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
2009 | { | |
2010 | struct page *oldpage; | |
2011 | int pages; | |
2012 | int pobjects; | |
2013 | ||
2014 | do { | |
2015 | pages = 0; | |
2016 | pobjects = 0; | |
2017 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2018 | ||
2019 | if (oldpage) { | |
2020 | pobjects = oldpage->pobjects; | |
2021 | pages = oldpage->pages; | |
2022 | if (drain && pobjects > s->cpu_partial) { | |
2023 | unsigned long flags; | |
2024 | /* | |
2025 | * partial array is full. Move the existing | |
2026 | * set to the per node partial list. | |
2027 | */ | |
2028 | local_irq_save(flags); | |
2029 | unfreeze_partials(s); | |
2030 | local_irq_restore(flags); | |
2031 | pobjects = 0; | |
2032 | pages = 0; | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | pages++; | |
2037 | pobjects += page->objects - page->inuse; | |
2038 | ||
2039 | page->pages = pages; | |
2040 | page->pobjects = pobjects; | |
2041 | page->next = oldpage; | |
2042 | ||
2043 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); | |
2044 | stat(s, CPU_PARTIAL_FREE); | |
2045 | return pobjects; | |
2046 | } | |
2047 | ||
dfb4f096 | 2048 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2049 | { |
84e554e6 | 2050 | stat(s, CPUSLAB_FLUSH); |
dfb4f096 | 2051 | deactivate_slab(s, c); |
81819f0f CL |
2052 | } |
2053 | ||
2054 | /* | |
2055 | * Flush cpu slab. | |
6446faa2 | 2056 | * |
81819f0f CL |
2057 | * Called from IPI handler with interrupts disabled. |
2058 | */ | |
0c710013 | 2059 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2060 | { |
9dfc6e68 | 2061 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2062 | |
49e22585 CL |
2063 | if (likely(c)) { |
2064 | if (c->page) | |
2065 | flush_slab(s, c); | |
2066 | ||
2067 | unfreeze_partials(s); | |
2068 | } | |
81819f0f CL |
2069 | } |
2070 | ||
2071 | static void flush_cpu_slab(void *d) | |
2072 | { | |
2073 | struct kmem_cache *s = d; | |
81819f0f | 2074 | |
dfb4f096 | 2075 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2076 | } |
2077 | ||
2078 | static void flush_all(struct kmem_cache *s) | |
2079 | { | |
15c8b6c1 | 2080 | on_each_cpu(flush_cpu_slab, s, 1); |
81819f0f CL |
2081 | } |
2082 | ||
dfb4f096 CL |
2083 | /* |
2084 | * Check if the objects in a per cpu structure fit numa | |
2085 | * locality expectations. | |
2086 | */ | |
2087 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
2088 | { | |
2089 | #ifdef CONFIG_NUMA | |
2154a336 | 2090 | if (node != NUMA_NO_NODE && c->node != node) |
dfb4f096 CL |
2091 | return 0; |
2092 | #endif | |
2093 | return 1; | |
2094 | } | |
2095 | ||
781b2ba6 PE |
2096 | static int count_free(struct page *page) |
2097 | { | |
2098 | return page->objects - page->inuse; | |
2099 | } | |
2100 | ||
2101 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2102 | int (*get_count)(struct page *)) | |
2103 | { | |
2104 | unsigned long flags; | |
2105 | unsigned long x = 0; | |
2106 | struct page *page; | |
2107 | ||
2108 | spin_lock_irqsave(&n->list_lock, flags); | |
2109 | list_for_each_entry(page, &n->partial, lru) | |
2110 | x += get_count(page); | |
2111 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2112 | return x; | |
2113 | } | |
2114 | ||
26c02cf0 AB |
2115 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2116 | { | |
2117 | #ifdef CONFIG_SLUB_DEBUG | |
2118 | return atomic_long_read(&n->total_objects); | |
2119 | #else | |
2120 | return 0; | |
2121 | #endif | |
2122 | } | |
2123 | ||
781b2ba6 PE |
2124 | static noinline void |
2125 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2126 | { | |
2127 | int node; | |
2128 | ||
2129 | printk(KERN_WARNING | |
2130 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2131 | nid, gfpflags); | |
2132 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
2133 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
2134 | s->size, oo_order(s->oo), oo_order(s->min)); | |
2135 | ||
fa5ec8a1 DR |
2136 | if (oo_order(s->min) > get_order(s->objsize)) |
2137 | printk(KERN_WARNING " %s debugging increased min order, use " | |
2138 | "slub_debug=O to disable.\n", s->name); | |
2139 | ||
781b2ba6 PE |
2140 | for_each_online_node(node) { |
2141 | struct kmem_cache_node *n = get_node(s, node); | |
2142 | unsigned long nr_slabs; | |
2143 | unsigned long nr_objs; | |
2144 | unsigned long nr_free; | |
2145 | ||
2146 | if (!n) | |
2147 | continue; | |
2148 | ||
26c02cf0 AB |
2149 | nr_free = count_partial(n, count_free); |
2150 | nr_slabs = node_nr_slabs(n); | |
2151 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2152 | |
2153 | printk(KERN_WARNING | |
2154 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2155 | node, nr_slabs, nr_objs, nr_free); | |
2156 | } | |
2157 | } | |
2158 | ||
497b66f2 CL |
2159 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2160 | int node, struct kmem_cache_cpu **pc) | |
2161 | { | |
2162 | void *object; | |
2163 | struct kmem_cache_cpu *c; | |
2164 | struct page *page = new_slab(s, flags, node); | |
2165 | ||
2166 | if (page) { | |
2167 | c = __this_cpu_ptr(s->cpu_slab); | |
2168 | if (c->page) | |
2169 | flush_slab(s, c); | |
2170 | ||
2171 | /* | |
2172 | * No other reference to the page yet so we can | |
2173 | * muck around with it freely without cmpxchg | |
2174 | */ | |
2175 | object = page->freelist; | |
2176 | page->freelist = NULL; | |
2177 | ||
2178 | stat(s, ALLOC_SLAB); | |
2179 | c->node = page_to_nid(page); | |
2180 | c->page = page; | |
2181 | *pc = c; | |
2182 | } else | |
2183 | object = NULL; | |
2184 | ||
2185 | return object; | |
2186 | } | |
2187 | ||
81819f0f | 2188 | /* |
894b8788 CL |
2189 | * Slow path. The lockless freelist is empty or we need to perform |
2190 | * debugging duties. | |
2191 | * | |
894b8788 CL |
2192 | * Processing is still very fast if new objects have been freed to the |
2193 | * regular freelist. In that case we simply take over the regular freelist | |
2194 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2195 | * |
894b8788 CL |
2196 | * If that is not working then we fall back to the partial lists. We take the |
2197 | * first element of the freelist as the object to allocate now and move the | |
2198 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2199 | * |
894b8788 | 2200 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2201 | * we need to allocate a new slab. This is the slowest path since it involves |
2202 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2203 | */ |
ce71e27c EGM |
2204 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2205 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2206 | { |
81819f0f | 2207 | void **object; |
8a5ec0ba | 2208 | unsigned long flags; |
2cfb7455 CL |
2209 | struct page new; |
2210 | unsigned long counters; | |
8a5ec0ba CL |
2211 | |
2212 | local_irq_save(flags); | |
2213 | #ifdef CONFIG_PREEMPT | |
2214 | /* | |
2215 | * We may have been preempted and rescheduled on a different | |
2216 | * cpu before disabling interrupts. Need to reload cpu area | |
2217 | * pointer. | |
2218 | */ | |
2219 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2220 | #endif |
81819f0f | 2221 | |
497b66f2 | 2222 | if (!c->page) |
81819f0f | 2223 | goto new_slab; |
49e22585 | 2224 | redo: |
fc59c053 | 2225 | if (unlikely(!node_match(c, node))) { |
e36a2652 | 2226 | stat(s, ALLOC_NODE_MISMATCH); |
fc59c053 CL |
2227 | deactivate_slab(s, c); |
2228 | goto new_slab; | |
2229 | } | |
6446faa2 | 2230 | |
2cfb7455 CL |
2231 | stat(s, ALLOC_SLOWPATH); |
2232 | ||
2233 | do { | |
497b66f2 CL |
2234 | object = c->page->freelist; |
2235 | counters = c->page->counters; | |
2cfb7455 | 2236 | new.counters = counters; |
2cfb7455 CL |
2237 | VM_BUG_ON(!new.frozen); |
2238 | ||
03e404af CL |
2239 | /* |
2240 | * If there is no object left then we use this loop to | |
2241 | * deactivate the slab which is simple since no objects | |
2242 | * are left in the slab and therefore we do not need to | |
2243 | * put the page back onto the partial list. | |
2244 | * | |
2245 | * If there are objects left then we retrieve them | |
2246 | * and use them to refill the per cpu queue. | |
497b66f2 | 2247 | */ |
03e404af | 2248 | |
497b66f2 | 2249 | new.inuse = c->page->objects; |
03e404af CL |
2250 | new.frozen = object != NULL; |
2251 | ||
497b66f2 | 2252 | } while (!__cmpxchg_double_slab(s, c->page, |
2cfb7455 CL |
2253 | object, counters, |
2254 | NULL, new.counters, | |
2255 | "__slab_alloc")); | |
6446faa2 | 2256 | |
49e22585 | 2257 | if (!object) { |
03e404af CL |
2258 | c->page = NULL; |
2259 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2260 | goto new_slab; |
03e404af | 2261 | } |
6446faa2 | 2262 | |
84e554e6 | 2263 | stat(s, ALLOC_REFILL); |
6446faa2 | 2264 | |
894b8788 | 2265 | load_freelist: |
ff12059e | 2266 | c->freelist = get_freepointer(s, object); |
8a5ec0ba CL |
2267 | c->tid = next_tid(c->tid); |
2268 | local_irq_restore(flags); | |
81819f0f CL |
2269 | return object; |
2270 | ||
81819f0f | 2271 | new_slab: |
49e22585 CL |
2272 | |
2273 | if (c->partial) { | |
2274 | c->page = c->partial; | |
2275 | c->partial = c->page->next; | |
2276 | c->node = page_to_nid(c->page); | |
2277 | stat(s, CPU_PARTIAL_ALLOC); | |
2278 | c->freelist = NULL; | |
2279 | goto redo; | |
2280 | } | |
2281 | ||
2282 | /* Then do expensive stuff like retrieving pages from the partial lists */ | |
497b66f2 | 2283 | object = get_partial(s, gfpflags, node, c); |
81819f0f | 2284 | |
497b66f2 | 2285 | if (unlikely(!object)) { |
b811c202 | 2286 | |
497b66f2 | 2287 | object = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2288 | |
497b66f2 CL |
2289 | if (unlikely(!object)) { |
2290 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2291 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2292 | |
497b66f2 CL |
2293 | local_irq_restore(flags); |
2294 | return NULL; | |
2295 | } | |
2296 | } | |
9e577e8b | 2297 | |
497b66f2 | 2298 | if (likely(!kmem_cache_debug(s))) |
4b6f0750 | 2299 | goto load_freelist; |
2cfb7455 | 2300 | |
497b66f2 CL |
2301 | /* Only entered in the debug case */ |
2302 | if (!alloc_debug_processing(s, c->page, object, addr)) | |
2303 | goto new_slab; /* Slab failed checks. Next slab needed */ | |
894b8788 | 2304 | |
2cfb7455 | 2305 | c->freelist = get_freepointer(s, object); |
442b06bc | 2306 | deactivate_slab(s, c); |
15b7c514 | 2307 | c->node = NUMA_NO_NODE; |
a71ae47a CL |
2308 | local_irq_restore(flags); |
2309 | return object; | |
894b8788 CL |
2310 | } |
2311 | ||
2312 | /* | |
2313 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2314 | * have the fastpath folded into their functions. So no function call | |
2315 | * overhead for requests that can be satisfied on the fastpath. | |
2316 | * | |
2317 | * The fastpath works by first checking if the lockless freelist can be used. | |
2318 | * If not then __slab_alloc is called for slow processing. | |
2319 | * | |
2320 | * Otherwise we can simply pick the next object from the lockless free list. | |
2321 | */ | |
06428780 | 2322 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2323 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2324 | { |
894b8788 | 2325 | void **object; |
dfb4f096 | 2326 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2327 | unsigned long tid; |
1f84260c | 2328 | |
c016b0bd | 2329 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2330 | return NULL; |
1f84260c | 2331 | |
8a5ec0ba | 2332 | redo: |
8a5ec0ba CL |
2333 | |
2334 | /* | |
2335 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2336 | * enabled. We may switch back and forth between cpus while | |
2337 | * reading from one cpu area. That does not matter as long | |
2338 | * as we end up on the original cpu again when doing the cmpxchg. | |
2339 | */ | |
9dfc6e68 | 2340 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2341 | |
8a5ec0ba CL |
2342 | /* |
2343 | * The transaction ids are globally unique per cpu and per operation on | |
2344 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2345 | * occurs on the right processor and that there was no operation on the | |
2346 | * linked list in between. | |
2347 | */ | |
2348 | tid = c->tid; | |
2349 | barrier(); | |
8a5ec0ba | 2350 | |
9dfc6e68 | 2351 | object = c->freelist; |
9dfc6e68 | 2352 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 2353 | |
dfb4f096 | 2354 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2355 | |
2356 | else { | |
8a5ec0ba | 2357 | /* |
25985edc | 2358 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2359 | * operation and if we are on the right processor. |
2360 | * | |
2361 | * The cmpxchg does the following atomically (without lock semantics!) | |
2362 | * 1. Relocate first pointer to the current per cpu area. | |
2363 | * 2. Verify that tid and freelist have not been changed | |
2364 | * 3. If they were not changed replace tid and freelist | |
2365 | * | |
2366 | * Since this is without lock semantics the protection is only against | |
2367 | * code executing on this cpu *not* from access by other cpus. | |
2368 | */ | |
30106b8c | 2369 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2370 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2371 | object, tid, | |
1393d9a1 | 2372 | get_freepointer_safe(s, object), next_tid(tid)))) { |
8a5ec0ba CL |
2373 | |
2374 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2375 | goto redo; | |
2376 | } | |
84e554e6 | 2377 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2378 | } |
8a5ec0ba | 2379 | |
74e2134f | 2380 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 2381 | memset(object, 0, s->objsize); |
d07dbea4 | 2382 | |
c016b0bd | 2383 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2384 | |
894b8788 | 2385 | return object; |
81819f0f CL |
2386 | } |
2387 | ||
2388 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2389 | { | |
2154a336 | 2390 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2391 | |
ca2b84cb | 2392 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
2393 | |
2394 | return ret; | |
81819f0f CL |
2395 | } |
2396 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2397 | ||
0f24f128 | 2398 | #ifdef CONFIG_TRACING |
4a92379b RK |
2399 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2400 | { | |
2401 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2402 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2403 | return ret; | |
2404 | } | |
2405 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2406 | ||
2407 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2408 | { |
4a92379b RK |
2409 | void *ret = kmalloc_order(size, flags, order); |
2410 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2411 | return ret; | |
5b882be4 | 2412 | } |
4a92379b | 2413 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2414 | #endif |
2415 | ||
81819f0f CL |
2416 | #ifdef CONFIG_NUMA |
2417 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2418 | { | |
5b882be4 EGM |
2419 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2420 | ||
ca2b84cb EGM |
2421 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2422 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2423 | |
2424 | return ret; | |
81819f0f CL |
2425 | } |
2426 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2427 | |
0f24f128 | 2428 | #ifdef CONFIG_TRACING |
4a92379b | 2429 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2430 | gfp_t gfpflags, |
4a92379b | 2431 | int node, size_t size) |
5b882be4 | 2432 | { |
4a92379b RK |
2433 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2434 | ||
2435 | trace_kmalloc_node(_RET_IP_, ret, | |
2436 | size, s->size, gfpflags, node); | |
2437 | return ret; | |
5b882be4 | 2438 | } |
4a92379b | 2439 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2440 | #endif |
5d1f57e4 | 2441 | #endif |
5b882be4 | 2442 | |
81819f0f | 2443 | /* |
894b8788 CL |
2444 | * Slow patch handling. This may still be called frequently since objects |
2445 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2446 | * |
894b8788 CL |
2447 | * So we still attempt to reduce cache line usage. Just take the slab |
2448 | * lock and free the item. If there is no additional partial page | |
2449 | * handling required then we can return immediately. | |
81819f0f | 2450 | */ |
894b8788 | 2451 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2452 | void *x, unsigned long addr) |
81819f0f CL |
2453 | { |
2454 | void *prior; | |
2455 | void **object = (void *)x; | |
2cfb7455 CL |
2456 | int was_frozen; |
2457 | int inuse; | |
2458 | struct page new; | |
2459 | unsigned long counters; | |
2460 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2461 | unsigned long uninitialized_var(flags); |
81819f0f | 2462 | |
8a5ec0ba | 2463 | stat(s, FREE_SLOWPATH); |
81819f0f | 2464 | |
8dc16c6c | 2465 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
80f08c19 | 2466 | return; |
6446faa2 | 2467 | |
2cfb7455 CL |
2468 | do { |
2469 | prior = page->freelist; | |
2470 | counters = page->counters; | |
2471 | set_freepointer(s, object, prior); | |
2472 | new.counters = counters; | |
2473 | was_frozen = new.frozen; | |
2474 | new.inuse--; | |
2475 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
49e22585 CL |
2476 | |
2477 | if (!kmem_cache_debug(s) && !prior) | |
2478 | ||
2479 | /* | |
2480 | * Slab was on no list before and will be partially empty | |
2481 | * We can defer the list move and instead freeze it. | |
2482 | */ | |
2483 | new.frozen = 1; | |
2484 | ||
2485 | else { /* Needs to be taken off a list */ | |
2486 | ||
2487 | n = get_node(s, page_to_nid(page)); | |
2488 | /* | |
2489 | * Speculatively acquire the list_lock. | |
2490 | * If the cmpxchg does not succeed then we may | |
2491 | * drop the list_lock without any processing. | |
2492 | * | |
2493 | * Otherwise the list_lock will synchronize with | |
2494 | * other processors updating the list of slabs. | |
2495 | */ | |
2496 | spin_lock_irqsave(&n->list_lock, flags); | |
2497 | ||
2498 | } | |
2cfb7455 CL |
2499 | } |
2500 | inuse = new.inuse; | |
81819f0f | 2501 | |
2cfb7455 CL |
2502 | } while (!cmpxchg_double_slab(s, page, |
2503 | prior, counters, | |
2504 | object, new.counters, | |
2505 | "__slab_free")); | |
81819f0f | 2506 | |
2cfb7455 | 2507 | if (likely(!n)) { |
49e22585 CL |
2508 | |
2509 | /* | |
2510 | * If we just froze the page then put it onto the | |
2511 | * per cpu partial list. | |
2512 | */ | |
2513 | if (new.frozen && !was_frozen) | |
2514 | put_cpu_partial(s, page, 1); | |
2515 | ||
2516 | /* | |
2cfb7455 CL |
2517 | * The list lock was not taken therefore no list |
2518 | * activity can be necessary. | |
2519 | */ | |
2520 | if (was_frozen) | |
2521 | stat(s, FREE_FROZEN); | |
80f08c19 | 2522 | return; |
2cfb7455 | 2523 | } |
81819f0f CL |
2524 | |
2525 | /* | |
2cfb7455 CL |
2526 | * was_frozen may have been set after we acquired the list_lock in |
2527 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2528 | */ |
2cfb7455 CL |
2529 | if (was_frozen) |
2530 | stat(s, FREE_FROZEN); | |
2531 | else { | |
2532 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2533 | goto slab_empty; | |
81819f0f | 2534 | |
2cfb7455 CL |
2535 | /* |
2536 | * Objects left in the slab. If it was not on the partial list before | |
2537 | * then add it. | |
2538 | */ | |
2539 | if (unlikely(!prior)) { | |
2540 | remove_full(s, page); | |
2541 | add_partial(n, page, 0); | |
2542 | stat(s, FREE_ADD_PARTIAL); | |
2543 | } | |
8ff12cfc | 2544 | } |
80f08c19 | 2545 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2546 | return; |
2547 | ||
2548 | slab_empty: | |
a973e9dd | 2549 | if (prior) { |
81819f0f | 2550 | /* |
6fbabb20 | 2551 | * Slab on the partial list. |
81819f0f | 2552 | */ |
5cc6eee8 | 2553 | remove_partial(n, page); |
84e554e6 | 2554 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2555 | } else |
2556 | /* Slab must be on the full list */ | |
2557 | remove_full(s, page); | |
2cfb7455 | 2558 | |
80f08c19 | 2559 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2560 | stat(s, FREE_SLAB); |
81819f0f | 2561 | discard_slab(s, page); |
81819f0f CL |
2562 | } |
2563 | ||
894b8788 CL |
2564 | /* |
2565 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2566 | * can perform fastpath freeing without additional function calls. | |
2567 | * | |
2568 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2569 | * of this processor. This typically the case if we have just allocated | |
2570 | * the item before. | |
2571 | * | |
2572 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2573 | * with all sorts of special processing. | |
2574 | */ | |
06428780 | 2575 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2576 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2577 | { |
2578 | void **object = (void *)x; | |
dfb4f096 | 2579 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2580 | unsigned long tid; |
1f84260c | 2581 | |
c016b0bd CL |
2582 | slab_free_hook(s, x); |
2583 | ||
8a5ec0ba CL |
2584 | redo: |
2585 | /* | |
2586 | * Determine the currently cpus per cpu slab. | |
2587 | * The cpu may change afterward. However that does not matter since | |
2588 | * data is retrieved via this pointer. If we are on the same cpu | |
2589 | * during the cmpxchg then the free will succedd. | |
2590 | */ | |
9dfc6e68 | 2591 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2592 | |
8a5ec0ba CL |
2593 | tid = c->tid; |
2594 | barrier(); | |
c016b0bd | 2595 | |
442b06bc | 2596 | if (likely(page == c->page)) { |
ff12059e | 2597 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2598 | |
30106b8c | 2599 | if (unlikely(!irqsafe_cpu_cmpxchg_double( |
8a5ec0ba CL |
2600 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2601 | c->freelist, tid, | |
2602 | object, next_tid(tid)))) { | |
2603 | ||
2604 | note_cmpxchg_failure("slab_free", s, tid); | |
2605 | goto redo; | |
2606 | } | |
84e554e6 | 2607 | stat(s, FREE_FASTPATH); |
894b8788 | 2608 | } else |
ff12059e | 2609 | __slab_free(s, page, x, addr); |
894b8788 | 2610 | |
894b8788 CL |
2611 | } |
2612 | ||
81819f0f CL |
2613 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2614 | { | |
77c5e2d0 | 2615 | struct page *page; |
81819f0f | 2616 | |
b49af68f | 2617 | page = virt_to_head_page(x); |
81819f0f | 2618 | |
ce71e27c | 2619 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2620 | |
ca2b84cb | 2621 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2622 | } |
2623 | EXPORT_SYMBOL(kmem_cache_free); | |
2624 | ||
81819f0f | 2625 | /* |
672bba3a CL |
2626 | * Object placement in a slab is made very easy because we always start at |
2627 | * offset 0. If we tune the size of the object to the alignment then we can | |
2628 | * get the required alignment by putting one properly sized object after | |
2629 | * another. | |
81819f0f CL |
2630 | * |
2631 | * Notice that the allocation order determines the sizes of the per cpu | |
2632 | * caches. Each processor has always one slab available for allocations. | |
2633 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2634 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2635 | * locking overhead. |
81819f0f CL |
2636 | */ |
2637 | ||
2638 | /* | |
2639 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2640 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2641 | * and increases the number of allocations possible without having to | |
2642 | * take the list_lock. | |
2643 | */ | |
2644 | static int slub_min_order; | |
114e9e89 | 2645 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2646 | static int slub_min_objects; |
81819f0f CL |
2647 | |
2648 | /* | |
2649 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2650 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2651 | */ |
2652 | static int slub_nomerge; | |
2653 | ||
81819f0f CL |
2654 | /* |
2655 | * Calculate the order of allocation given an slab object size. | |
2656 | * | |
672bba3a CL |
2657 | * The order of allocation has significant impact on performance and other |
2658 | * system components. Generally order 0 allocations should be preferred since | |
2659 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2660 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2661 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2662 | * would be wasted. |
2663 | * | |
2664 | * In order to reach satisfactory performance we must ensure that a minimum | |
2665 | * number of objects is in one slab. Otherwise we may generate too much | |
2666 | * activity on the partial lists which requires taking the list_lock. This is | |
2667 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2668 | * |
672bba3a CL |
2669 | * slub_max_order specifies the order where we begin to stop considering the |
2670 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2671 | * we try to keep the page order as low as possible. So we accept more waste | |
2672 | * of space in favor of a small page order. | |
81819f0f | 2673 | * |
672bba3a CL |
2674 | * Higher order allocations also allow the placement of more objects in a |
2675 | * slab and thereby reduce object handling overhead. If the user has | |
2676 | * requested a higher mininum order then we start with that one instead of | |
2677 | * the smallest order which will fit the object. | |
81819f0f | 2678 | */ |
5e6d444e | 2679 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2680 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2681 | { |
2682 | int order; | |
2683 | int rem; | |
6300ea75 | 2684 | int min_order = slub_min_order; |
81819f0f | 2685 | |
ab9a0f19 | 2686 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2687 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2688 | |
6300ea75 | 2689 | for (order = max(min_order, |
5e6d444e CL |
2690 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2691 | order <= max_order; order++) { | |
81819f0f | 2692 | |
5e6d444e | 2693 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2694 | |
ab9a0f19 | 2695 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2696 | continue; |
2697 | ||
ab9a0f19 | 2698 | rem = (slab_size - reserved) % size; |
81819f0f | 2699 | |
5e6d444e | 2700 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2701 | break; |
2702 | ||
2703 | } | |
672bba3a | 2704 | |
81819f0f CL |
2705 | return order; |
2706 | } | |
2707 | ||
ab9a0f19 | 2708 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2709 | { |
2710 | int order; | |
2711 | int min_objects; | |
2712 | int fraction; | |
e8120ff1 | 2713 | int max_objects; |
5e6d444e CL |
2714 | |
2715 | /* | |
2716 | * Attempt to find best configuration for a slab. This | |
2717 | * works by first attempting to generate a layout with | |
2718 | * the best configuration and backing off gradually. | |
2719 | * | |
2720 | * First we reduce the acceptable waste in a slab. Then | |
2721 | * we reduce the minimum objects required in a slab. | |
2722 | */ | |
2723 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2724 | if (!min_objects) |
2725 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2726 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2727 | min_objects = min(min_objects, max_objects); |
2728 | ||
5e6d444e | 2729 | while (min_objects > 1) { |
c124f5b5 | 2730 | fraction = 16; |
5e6d444e CL |
2731 | while (fraction >= 4) { |
2732 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2733 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2734 | if (order <= slub_max_order) |
2735 | return order; | |
2736 | fraction /= 2; | |
2737 | } | |
5086c389 | 2738 | min_objects--; |
5e6d444e CL |
2739 | } |
2740 | ||
2741 | /* | |
2742 | * We were unable to place multiple objects in a slab. Now | |
2743 | * lets see if we can place a single object there. | |
2744 | */ | |
ab9a0f19 | 2745 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2746 | if (order <= slub_max_order) |
2747 | return order; | |
2748 | ||
2749 | /* | |
2750 | * Doh this slab cannot be placed using slub_max_order. | |
2751 | */ | |
ab9a0f19 | 2752 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2753 | if (order < MAX_ORDER) |
5e6d444e CL |
2754 | return order; |
2755 | return -ENOSYS; | |
2756 | } | |
2757 | ||
81819f0f | 2758 | /* |
672bba3a | 2759 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2760 | */ |
2761 | static unsigned long calculate_alignment(unsigned long flags, | |
2762 | unsigned long align, unsigned long size) | |
2763 | { | |
2764 | /* | |
6446faa2 CL |
2765 | * If the user wants hardware cache aligned objects then follow that |
2766 | * suggestion if the object is sufficiently large. | |
81819f0f | 2767 | * |
6446faa2 CL |
2768 | * The hardware cache alignment cannot override the specified |
2769 | * alignment though. If that is greater then use it. | |
81819f0f | 2770 | */ |
b6210386 NP |
2771 | if (flags & SLAB_HWCACHE_ALIGN) { |
2772 | unsigned long ralign = cache_line_size(); | |
2773 | while (size <= ralign / 2) | |
2774 | ralign /= 2; | |
2775 | align = max(align, ralign); | |
2776 | } | |
81819f0f CL |
2777 | |
2778 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2779 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2780 | |
2781 | return ALIGN(align, sizeof(void *)); | |
2782 | } | |
2783 | ||
5595cffc PE |
2784 | static void |
2785 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2786 | { |
2787 | n->nr_partial = 0; | |
81819f0f CL |
2788 | spin_lock_init(&n->list_lock); |
2789 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2790 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2791 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2792 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2793 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2794 | #endif |
81819f0f CL |
2795 | } |
2796 | ||
55136592 | 2797 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2798 | { |
6c182dc0 CL |
2799 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2800 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2801 | |
8a5ec0ba | 2802 | /* |
d4d84fef CM |
2803 | * Must align to double word boundary for the double cmpxchg |
2804 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2805 | */ |
d4d84fef CM |
2806 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2807 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2808 | |
2809 | if (!s->cpu_slab) | |
2810 | return 0; | |
2811 | ||
2812 | init_kmem_cache_cpus(s); | |
4c93c355 | 2813 | |
8a5ec0ba | 2814 | return 1; |
4c93c355 | 2815 | } |
4c93c355 | 2816 | |
51df1142 CL |
2817 | static struct kmem_cache *kmem_cache_node; |
2818 | ||
81819f0f CL |
2819 | /* |
2820 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2821 | * slab on the node for this slabcache. There are no concurrent accesses | |
2822 | * possible. | |
2823 | * | |
2824 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2825 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2826 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2827 | */ |
55136592 | 2828 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2829 | { |
2830 | struct page *page; | |
2831 | struct kmem_cache_node *n; | |
2832 | ||
51df1142 | 2833 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2834 | |
51df1142 | 2835 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2836 | |
2837 | BUG_ON(!page); | |
a2f92ee7 CL |
2838 | if (page_to_nid(page) != node) { |
2839 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2840 | "node %d\n", node); | |
2841 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2842 | "in order to be able to continue\n"); | |
2843 | } | |
2844 | ||
81819f0f CL |
2845 | n = page->freelist; |
2846 | BUG_ON(!n); | |
51df1142 | 2847 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2848 | page->inuse = 1; |
8cb0a506 | 2849 | page->frozen = 0; |
51df1142 | 2850 | kmem_cache_node->node[node] = n; |
8ab1372f | 2851 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2852 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2853 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2854 | #endif |
51df1142 CL |
2855 | init_kmem_cache_node(n, kmem_cache_node); |
2856 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2857 | |
7c2e132c | 2858 | add_partial(n, page, 0); |
81819f0f CL |
2859 | } |
2860 | ||
2861 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2862 | { | |
2863 | int node; | |
2864 | ||
f64dc58c | 2865 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2866 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2867 | |
73367bd8 | 2868 | if (n) |
51df1142 CL |
2869 | kmem_cache_free(kmem_cache_node, n); |
2870 | ||
81819f0f CL |
2871 | s->node[node] = NULL; |
2872 | } | |
2873 | } | |
2874 | ||
55136592 | 2875 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2876 | { |
2877 | int node; | |
81819f0f | 2878 | |
f64dc58c | 2879 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2880 | struct kmem_cache_node *n; |
2881 | ||
73367bd8 | 2882 | if (slab_state == DOWN) { |
55136592 | 2883 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2884 | continue; |
2885 | } | |
51df1142 | 2886 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2887 | GFP_KERNEL, node); |
81819f0f | 2888 | |
73367bd8 AD |
2889 | if (!n) { |
2890 | free_kmem_cache_nodes(s); | |
2891 | return 0; | |
81819f0f | 2892 | } |
73367bd8 | 2893 | |
81819f0f | 2894 | s->node[node] = n; |
5595cffc | 2895 | init_kmem_cache_node(n, s); |
81819f0f CL |
2896 | } |
2897 | return 1; | |
2898 | } | |
81819f0f | 2899 | |
c0bdb232 | 2900 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2901 | { |
2902 | if (min < MIN_PARTIAL) | |
2903 | min = MIN_PARTIAL; | |
2904 | else if (min > MAX_PARTIAL) | |
2905 | min = MAX_PARTIAL; | |
2906 | s->min_partial = min; | |
2907 | } | |
2908 | ||
81819f0f CL |
2909 | /* |
2910 | * calculate_sizes() determines the order and the distribution of data within | |
2911 | * a slab object. | |
2912 | */ | |
06b285dc | 2913 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2914 | { |
2915 | unsigned long flags = s->flags; | |
2916 | unsigned long size = s->objsize; | |
2917 | unsigned long align = s->align; | |
834f3d11 | 2918 | int order; |
81819f0f | 2919 | |
d8b42bf5 CL |
2920 | /* |
2921 | * Round up object size to the next word boundary. We can only | |
2922 | * place the free pointer at word boundaries and this determines | |
2923 | * the possible location of the free pointer. | |
2924 | */ | |
2925 | size = ALIGN(size, sizeof(void *)); | |
2926 | ||
2927 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2928 | /* |
2929 | * Determine if we can poison the object itself. If the user of | |
2930 | * the slab may touch the object after free or before allocation | |
2931 | * then we should never poison the object itself. | |
2932 | */ | |
2933 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2934 | !s->ctor) |
81819f0f CL |
2935 | s->flags |= __OBJECT_POISON; |
2936 | else | |
2937 | s->flags &= ~__OBJECT_POISON; | |
2938 | ||
81819f0f CL |
2939 | |
2940 | /* | |
672bba3a | 2941 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2942 | * end of the object and the free pointer. If not then add an |
672bba3a | 2943 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2944 | */ |
2945 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2946 | size += sizeof(void *); | |
41ecc55b | 2947 | #endif |
81819f0f CL |
2948 | |
2949 | /* | |
672bba3a CL |
2950 | * With that we have determined the number of bytes in actual use |
2951 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2952 | */ |
2953 | s->inuse = size; | |
2954 | ||
2955 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2956 | s->ctor)) { |
81819f0f CL |
2957 | /* |
2958 | * Relocate free pointer after the object if it is not | |
2959 | * permitted to overwrite the first word of the object on | |
2960 | * kmem_cache_free. | |
2961 | * | |
2962 | * This is the case if we do RCU, have a constructor or | |
2963 | * destructor or are poisoning the objects. | |
2964 | */ | |
2965 | s->offset = size; | |
2966 | size += sizeof(void *); | |
2967 | } | |
2968 | ||
c12b3c62 | 2969 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2970 | if (flags & SLAB_STORE_USER) |
2971 | /* | |
2972 | * Need to store information about allocs and frees after | |
2973 | * the object. | |
2974 | */ | |
2975 | size += 2 * sizeof(struct track); | |
2976 | ||
be7b3fbc | 2977 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2978 | /* |
2979 | * Add some empty padding so that we can catch | |
2980 | * overwrites from earlier objects rather than let | |
2981 | * tracking information or the free pointer be | |
0211a9c8 | 2982 | * corrupted if a user writes before the start |
81819f0f CL |
2983 | * of the object. |
2984 | */ | |
2985 | size += sizeof(void *); | |
41ecc55b | 2986 | #endif |
672bba3a | 2987 | |
81819f0f CL |
2988 | /* |
2989 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2990 | * user specified and the dynamic determination of cache line size |
2991 | * on bootup. | |
81819f0f CL |
2992 | */ |
2993 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2994 | s->align = align; |
81819f0f CL |
2995 | |
2996 | /* | |
2997 | * SLUB stores one object immediately after another beginning from | |
2998 | * offset 0. In order to align the objects we have to simply size | |
2999 | * each object to conform to the alignment. | |
3000 | */ | |
3001 | size = ALIGN(size, align); | |
3002 | s->size = size; | |
06b285dc CL |
3003 | if (forced_order >= 0) |
3004 | order = forced_order; | |
3005 | else | |
ab9a0f19 | 3006 | order = calculate_order(size, s->reserved); |
81819f0f | 3007 | |
834f3d11 | 3008 | if (order < 0) |
81819f0f CL |
3009 | return 0; |
3010 | ||
b7a49f0d | 3011 | s->allocflags = 0; |
834f3d11 | 3012 | if (order) |
b7a49f0d CL |
3013 | s->allocflags |= __GFP_COMP; |
3014 | ||
3015 | if (s->flags & SLAB_CACHE_DMA) | |
3016 | s->allocflags |= SLUB_DMA; | |
3017 | ||
3018 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3019 | s->allocflags |= __GFP_RECLAIMABLE; | |
3020 | ||
81819f0f CL |
3021 | /* |
3022 | * Determine the number of objects per slab | |
3023 | */ | |
ab9a0f19 LJ |
3024 | s->oo = oo_make(order, size, s->reserved); |
3025 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3026 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3027 | s->max = s->oo; | |
81819f0f | 3028 | |
834f3d11 | 3029 | return !!oo_objects(s->oo); |
81819f0f CL |
3030 | |
3031 | } | |
3032 | ||
55136592 | 3033 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
3034 | const char *name, size_t size, |
3035 | size_t align, unsigned long flags, | |
51cc5068 | 3036 | void (*ctor)(void *)) |
81819f0f CL |
3037 | { |
3038 | memset(s, 0, kmem_size); | |
3039 | s->name = name; | |
3040 | s->ctor = ctor; | |
81819f0f | 3041 | s->objsize = size; |
81819f0f | 3042 | s->align = align; |
ba0268a8 | 3043 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 3044 | s->reserved = 0; |
81819f0f | 3045 | |
da9a638c LJ |
3046 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3047 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3048 | |
06b285dc | 3049 | if (!calculate_sizes(s, -1)) |
81819f0f | 3050 | goto error; |
3de47213 DR |
3051 | if (disable_higher_order_debug) { |
3052 | /* | |
3053 | * Disable debugging flags that store metadata if the min slab | |
3054 | * order increased. | |
3055 | */ | |
3056 | if (get_order(s->size) > get_order(s->objsize)) { | |
3057 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
3058 | s->offset = 0; | |
3059 | if (!calculate_sizes(s, -1)) | |
3060 | goto error; | |
3061 | } | |
3062 | } | |
81819f0f | 3063 | |
b789ef51 CL |
3064 | #ifdef CONFIG_CMPXCHG_DOUBLE |
3065 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) | |
3066 | /* Enable fast mode */ | |
3067 | s->flags |= __CMPXCHG_DOUBLE; | |
3068 | #endif | |
3069 | ||
3b89d7d8 DR |
3070 | /* |
3071 | * The larger the object size is, the more pages we want on the partial | |
3072 | * list to avoid pounding the page allocator excessively. | |
3073 | */ | |
49e22585 CL |
3074 | set_min_partial(s, ilog2(s->size) / 2); |
3075 | ||
3076 | /* | |
3077 | * cpu_partial determined the maximum number of objects kept in the | |
3078 | * per cpu partial lists of a processor. | |
3079 | * | |
3080 | * Per cpu partial lists mainly contain slabs that just have one | |
3081 | * object freed. If they are used for allocation then they can be | |
3082 | * filled up again with minimal effort. The slab will never hit the | |
3083 | * per node partial lists and therefore no locking will be required. | |
3084 | * | |
3085 | * This setting also determines | |
3086 | * | |
3087 | * A) The number of objects from per cpu partial slabs dumped to the | |
3088 | * per node list when we reach the limit. | |
3089 | * B) The number of objects in partial partial slabs to extract from the | |
3090 | * per node list when we run out of per cpu objects. We only fetch 50% | |
3091 | * to keep some capacity around for frees. | |
3092 | */ | |
3093 | if (s->size >= PAGE_SIZE) | |
3094 | s->cpu_partial = 2; | |
3095 | else if (s->size >= 1024) | |
3096 | s->cpu_partial = 6; | |
3097 | else if (s->size >= 256) | |
3098 | s->cpu_partial = 13; | |
3099 | else | |
3100 | s->cpu_partial = 30; | |
3101 | ||
81819f0f CL |
3102 | s->refcount = 1; |
3103 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 3104 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3105 | #endif |
55136592 | 3106 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3107 | goto error; |
81819f0f | 3108 | |
55136592 | 3109 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 3110 | return 1; |
ff12059e | 3111 | |
4c93c355 | 3112 | free_kmem_cache_nodes(s); |
81819f0f CL |
3113 | error: |
3114 | if (flags & SLAB_PANIC) | |
3115 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3116 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 3117 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
3118 | s->offset, flags); |
3119 | return 0; | |
3120 | } | |
81819f0f | 3121 | |
81819f0f CL |
3122 | /* |
3123 | * Determine the size of a slab object | |
3124 | */ | |
3125 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
3126 | { | |
3127 | return s->objsize; | |
3128 | } | |
3129 | EXPORT_SYMBOL(kmem_cache_size); | |
3130 | ||
33b12c38 CL |
3131 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3132 | const char *text) | |
3133 | { | |
3134 | #ifdef CONFIG_SLUB_DEBUG | |
3135 | void *addr = page_address(page); | |
3136 | void *p; | |
a5dd5c11 NK |
3137 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3138 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3139 | if (!map) |
3140 | return; | |
33b12c38 CL |
3141 | slab_err(s, page, "%s", text); |
3142 | slab_lock(page); | |
33b12c38 | 3143 | |
5f80b13a | 3144 | get_map(s, page, map); |
33b12c38 CL |
3145 | for_each_object(p, s, addr, page->objects) { |
3146 | ||
3147 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3148 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3149 | p, p - addr); | |
3150 | print_tracking(s, p); | |
3151 | } | |
3152 | } | |
3153 | slab_unlock(page); | |
bbd7d57b | 3154 | kfree(map); |
33b12c38 CL |
3155 | #endif |
3156 | } | |
3157 | ||
81819f0f | 3158 | /* |
599870b1 | 3159 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3160 | * This is called from kmem_cache_close(). We must be the last thread |
3161 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3162 | */ |
599870b1 | 3163 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3164 | { |
81819f0f CL |
3165 | struct page *page, *h; |
3166 | ||
33b12c38 | 3167 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3168 | if (!page->inuse) { |
5cc6eee8 | 3169 | remove_partial(n, page); |
81819f0f | 3170 | discard_slab(s, page); |
33b12c38 CL |
3171 | } else { |
3172 | list_slab_objects(s, page, | |
3173 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 3174 | } |
33b12c38 | 3175 | } |
81819f0f CL |
3176 | } |
3177 | ||
3178 | /* | |
672bba3a | 3179 | * Release all resources used by a slab cache. |
81819f0f | 3180 | */ |
0c710013 | 3181 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3182 | { |
3183 | int node; | |
3184 | ||
3185 | flush_all(s); | |
9dfc6e68 | 3186 | free_percpu(s->cpu_slab); |
81819f0f | 3187 | /* Attempt to free all objects */ |
f64dc58c | 3188 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3189 | struct kmem_cache_node *n = get_node(s, node); |
3190 | ||
599870b1 CL |
3191 | free_partial(s, n); |
3192 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3193 | return 1; |
3194 | } | |
3195 | free_kmem_cache_nodes(s); | |
3196 | return 0; | |
3197 | } | |
3198 | ||
3199 | /* | |
3200 | * Close a cache and release the kmem_cache structure | |
3201 | * (must be used for caches created using kmem_cache_create) | |
3202 | */ | |
3203 | void kmem_cache_destroy(struct kmem_cache *s) | |
3204 | { | |
3205 | down_write(&slub_lock); | |
3206 | s->refcount--; | |
3207 | if (!s->refcount) { | |
3208 | list_del(&s->list); | |
69cb8e6b | 3209 | up_write(&slub_lock); |
d629d819 PE |
3210 | if (kmem_cache_close(s)) { |
3211 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3212 | "still has objects.\n", s->name, __func__); | |
3213 | dump_stack(); | |
3214 | } | |
d76b1590 ED |
3215 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3216 | rcu_barrier(); | |
81819f0f | 3217 | sysfs_slab_remove(s); |
69cb8e6b CL |
3218 | } else |
3219 | up_write(&slub_lock); | |
81819f0f CL |
3220 | } |
3221 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3222 | ||
3223 | /******************************************************************** | |
3224 | * Kmalloc subsystem | |
3225 | *******************************************************************/ | |
3226 | ||
51df1142 | 3227 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3228 | EXPORT_SYMBOL(kmalloc_caches); |
3229 | ||
51df1142 CL |
3230 | static struct kmem_cache *kmem_cache; |
3231 | ||
55136592 | 3232 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3233 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3234 | #endif |
3235 | ||
81819f0f CL |
3236 | static int __init setup_slub_min_order(char *str) |
3237 | { | |
06428780 | 3238 | get_option(&str, &slub_min_order); |
81819f0f CL |
3239 | |
3240 | return 1; | |
3241 | } | |
3242 | ||
3243 | __setup("slub_min_order=", setup_slub_min_order); | |
3244 | ||
3245 | static int __init setup_slub_max_order(char *str) | |
3246 | { | |
06428780 | 3247 | get_option(&str, &slub_max_order); |
818cf590 | 3248 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3249 | |
3250 | return 1; | |
3251 | } | |
3252 | ||
3253 | __setup("slub_max_order=", setup_slub_max_order); | |
3254 | ||
3255 | static int __init setup_slub_min_objects(char *str) | |
3256 | { | |
06428780 | 3257 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3258 | |
3259 | return 1; | |
3260 | } | |
3261 | ||
3262 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3263 | ||
3264 | static int __init setup_slub_nomerge(char *str) | |
3265 | { | |
3266 | slub_nomerge = 1; | |
3267 | return 1; | |
3268 | } | |
3269 | ||
3270 | __setup("slub_nomerge", setup_slub_nomerge); | |
3271 | ||
51df1142 CL |
3272 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3273 | int size, unsigned int flags) | |
81819f0f | 3274 | { |
51df1142 CL |
3275 | struct kmem_cache *s; |
3276 | ||
3277 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3278 | ||
83b519e8 PE |
3279 | /* |
3280 | * This function is called with IRQs disabled during early-boot on | |
3281 | * single CPU so there's no need to take slub_lock here. | |
3282 | */ | |
55136592 | 3283 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3284 | flags, NULL)) |
81819f0f CL |
3285 | goto panic; |
3286 | ||
3287 | list_add(&s->list, &slab_caches); | |
51df1142 | 3288 | return s; |
81819f0f CL |
3289 | |
3290 | panic: | |
3291 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3292 | return NULL; |
81819f0f CL |
3293 | } |
3294 | ||
f1b26339 CL |
3295 | /* |
3296 | * Conversion table for small slabs sizes / 8 to the index in the | |
3297 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3298 | * of two cache sizes there. The size of larger slabs can be determined using | |
3299 | * fls. | |
3300 | */ | |
3301 | static s8 size_index[24] = { | |
3302 | 3, /* 8 */ | |
3303 | 4, /* 16 */ | |
3304 | 5, /* 24 */ | |
3305 | 5, /* 32 */ | |
3306 | 6, /* 40 */ | |
3307 | 6, /* 48 */ | |
3308 | 6, /* 56 */ | |
3309 | 6, /* 64 */ | |
3310 | 1, /* 72 */ | |
3311 | 1, /* 80 */ | |
3312 | 1, /* 88 */ | |
3313 | 1, /* 96 */ | |
3314 | 7, /* 104 */ | |
3315 | 7, /* 112 */ | |
3316 | 7, /* 120 */ | |
3317 | 7, /* 128 */ | |
3318 | 2, /* 136 */ | |
3319 | 2, /* 144 */ | |
3320 | 2, /* 152 */ | |
3321 | 2, /* 160 */ | |
3322 | 2, /* 168 */ | |
3323 | 2, /* 176 */ | |
3324 | 2, /* 184 */ | |
3325 | 2 /* 192 */ | |
3326 | }; | |
3327 | ||
acdfcd04 AK |
3328 | static inline int size_index_elem(size_t bytes) |
3329 | { | |
3330 | return (bytes - 1) / 8; | |
3331 | } | |
3332 | ||
81819f0f CL |
3333 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3334 | { | |
f1b26339 | 3335 | int index; |
81819f0f | 3336 | |
f1b26339 CL |
3337 | if (size <= 192) { |
3338 | if (!size) | |
3339 | return ZERO_SIZE_PTR; | |
81819f0f | 3340 | |
acdfcd04 | 3341 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3342 | } else |
f1b26339 | 3343 | index = fls(size - 1); |
81819f0f CL |
3344 | |
3345 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3346 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3347 | return kmalloc_dma_caches[index]; |
f1b26339 | 3348 | |
81819f0f | 3349 | #endif |
51df1142 | 3350 | return kmalloc_caches[index]; |
81819f0f CL |
3351 | } |
3352 | ||
3353 | void *__kmalloc(size_t size, gfp_t flags) | |
3354 | { | |
aadb4bc4 | 3355 | struct kmem_cache *s; |
5b882be4 | 3356 | void *ret; |
81819f0f | 3357 | |
ffadd4d0 | 3358 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3359 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3360 | |
3361 | s = get_slab(size, flags); | |
3362 | ||
3363 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3364 | return s; |
3365 | ||
2154a336 | 3366 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3367 | |
ca2b84cb | 3368 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3369 | |
3370 | return ret; | |
81819f0f CL |
3371 | } |
3372 | EXPORT_SYMBOL(__kmalloc); | |
3373 | ||
5d1f57e4 | 3374 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3375 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3376 | { | |
b1eeab67 | 3377 | struct page *page; |
e4f7c0b4 | 3378 | void *ptr = NULL; |
f619cfe1 | 3379 | |
b1eeab67 VN |
3380 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3381 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3382 | if (page) |
e4f7c0b4 CM |
3383 | ptr = page_address(page); |
3384 | ||
3385 | kmemleak_alloc(ptr, size, 1, flags); | |
3386 | return ptr; | |
f619cfe1 CL |
3387 | } |
3388 | ||
81819f0f CL |
3389 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3390 | { | |
aadb4bc4 | 3391 | struct kmem_cache *s; |
5b882be4 | 3392 | void *ret; |
81819f0f | 3393 | |
057685cf | 3394 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3395 | ret = kmalloc_large_node(size, flags, node); |
3396 | ||
ca2b84cb EGM |
3397 | trace_kmalloc_node(_RET_IP_, ret, |
3398 | size, PAGE_SIZE << get_order(size), | |
3399 | flags, node); | |
5b882be4 EGM |
3400 | |
3401 | return ret; | |
3402 | } | |
aadb4bc4 CL |
3403 | |
3404 | s = get_slab(size, flags); | |
3405 | ||
3406 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3407 | return s; |
3408 | ||
5b882be4 EGM |
3409 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3410 | ||
ca2b84cb | 3411 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3412 | |
3413 | return ret; | |
81819f0f CL |
3414 | } |
3415 | EXPORT_SYMBOL(__kmalloc_node); | |
3416 | #endif | |
3417 | ||
3418 | size_t ksize(const void *object) | |
3419 | { | |
272c1d21 | 3420 | struct page *page; |
81819f0f | 3421 | |
ef8b4520 | 3422 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3423 | return 0; |
3424 | ||
294a80a8 | 3425 | page = virt_to_head_page(object); |
294a80a8 | 3426 | |
76994412 PE |
3427 | if (unlikely(!PageSlab(page))) { |
3428 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3429 | return PAGE_SIZE << compound_order(page); |
76994412 | 3430 | } |
81819f0f | 3431 | |
b3d41885 | 3432 | return slab_ksize(page->slab); |
81819f0f | 3433 | } |
b1aabecd | 3434 | EXPORT_SYMBOL(ksize); |
81819f0f | 3435 | |
d18a90dd BG |
3436 | #ifdef CONFIG_SLUB_DEBUG |
3437 | bool verify_mem_not_deleted(const void *x) | |
3438 | { | |
3439 | struct page *page; | |
3440 | void *object = (void *)x; | |
3441 | unsigned long flags; | |
3442 | bool rv; | |
3443 | ||
3444 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3445 | return false; | |
3446 | ||
3447 | local_irq_save(flags); | |
3448 | ||
3449 | page = virt_to_head_page(x); | |
3450 | if (unlikely(!PageSlab(page))) { | |
3451 | /* maybe it was from stack? */ | |
3452 | rv = true; | |
3453 | goto out_unlock; | |
3454 | } | |
3455 | ||
3456 | slab_lock(page); | |
3457 | if (on_freelist(page->slab, page, object)) { | |
3458 | object_err(page->slab, page, object, "Object is on free-list"); | |
3459 | rv = false; | |
3460 | } else { | |
3461 | rv = true; | |
3462 | } | |
3463 | slab_unlock(page); | |
3464 | ||
3465 | out_unlock: | |
3466 | local_irq_restore(flags); | |
3467 | return rv; | |
3468 | } | |
3469 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3470 | #endif | |
3471 | ||
81819f0f CL |
3472 | void kfree(const void *x) |
3473 | { | |
81819f0f | 3474 | struct page *page; |
5bb983b0 | 3475 | void *object = (void *)x; |
81819f0f | 3476 | |
2121db74 PE |
3477 | trace_kfree(_RET_IP_, x); |
3478 | ||
2408c550 | 3479 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3480 | return; |
3481 | ||
b49af68f | 3482 | page = virt_to_head_page(x); |
aadb4bc4 | 3483 | if (unlikely(!PageSlab(page))) { |
0937502a | 3484 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3485 | kmemleak_free(x); |
aadb4bc4 CL |
3486 | put_page(page); |
3487 | return; | |
3488 | } | |
ce71e27c | 3489 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3490 | } |
3491 | EXPORT_SYMBOL(kfree); | |
3492 | ||
2086d26a | 3493 | /* |
672bba3a CL |
3494 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3495 | * the remaining slabs by the number of items in use. The slabs with the | |
3496 | * most items in use come first. New allocations will then fill those up | |
3497 | * and thus they can be removed from the partial lists. | |
3498 | * | |
3499 | * The slabs with the least items are placed last. This results in them | |
3500 | * being allocated from last increasing the chance that the last objects | |
3501 | * are freed in them. | |
2086d26a CL |
3502 | */ |
3503 | int kmem_cache_shrink(struct kmem_cache *s) | |
3504 | { | |
3505 | int node; | |
3506 | int i; | |
3507 | struct kmem_cache_node *n; | |
3508 | struct page *page; | |
3509 | struct page *t; | |
205ab99d | 3510 | int objects = oo_objects(s->max); |
2086d26a | 3511 | struct list_head *slabs_by_inuse = |
834f3d11 | 3512 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3513 | unsigned long flags; |
3514 | ||
3515 | if (!slabs_by_inuse) | |
3516 | return -ENOMEM; | |
3517 | ||
3518 | flush_all(s); | |
f64dc58c | 3519 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3520 | n = get_node(s, node); |
3521 | ||
3522 | if (!n->nr_partial) | |
3523 | continue; | |
3524 | ||
834f3d11 | 3525 | for (i = 0; i < objects; i++) |
2086d26a CL |
3526 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3527 | ||
3528 | spin_lock_irqsave(&n->list_lock, flags); | |
3529 | ||
3530 | /* | |
672bba3a | 3531 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3532 | * |
672bba3a CL |
3533 | * Note that concurrent frees may occur while we hold the |
3534 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3535 | */ |
3536 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3537 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3538 | if (!page->inuse) | |
3539 | n->nr_partial--; | |
2086d26a CL |
3540 | } |
3541 | ||
2086d26a | 3542 | /* |
672bba3a CL |
3543 | * Rebuild the partial list with the slabs filled up most |
3544 | * first and the least used slabs at the end. | |
2086d26a | 3545 | */ |
69cb8e6b | 3546 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3547 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3548 | ||
2086d26a | 3549 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3550 | |
3551 | /* Release empty slabs */ | |
3552 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3553 | discard_slab(s, page); | |
2086d26a CL |
3554 | } |
3555 | ||
3556 | kfree(slabs_by_inuse); | |
3557 | return 0; | |
3558 | } | |
3559 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3560 | ||
92a5bbc1 | 3561 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3562 | static int slab_mem_going_offline_callback(void *arg) |
3563 | { | |
3564 | struct kmem_cache *s; | |
3565 | ||
3566 | down_read(&slub_lock); | |
3567 | list_for_each_entry(s, &slab_caches, list) | |
3568 | kmem_cache_shrink(s); | |
3569 | up_read(&slub_lock); | |
3570 | ||
3571 | return 0; | |
3572 | } | |
3573 | ||
3574 | static void slab_mem_offline_callback(void *arg) | |
3575 | { | |
3576 | struct kmem_cache_node *n; | |
3577 | struct kmem_cache *s; | |
3578 | struct memory_notify *marg = arg; | |
3579 | int offline_node; | |
3580 | ||
3581 | offline_node = marg->status_change_nid; | |
3582 | ||
3583 | /* | |
3584 | * If the node still has available memory. we need kmem_cache_node | |
3585 | * for it yet. | |
3586 | */ | |
3587 | if (offline_node < 0) | |
3588 | return; | |
3589 | ||
3590 | down_read(&slub_lock); | |
3591 | list_for_each_entry(s, &slab_caches, list) { | |
3592 | n = get_node(s, offline_node); | |
3593 | if (n) { | |
3594 | /* | |
3595 | * if n->nr_slabs > 0, slabs still exist on the node | |
3596 | * that is going down. We were unable to free them, | |
c9404c9c | 3597 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3598 | * callback. So, we must fail. |
3599 | */ | |
0f389ec6 | 3600 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3601 | |
3602 | s->node[offline_node] = NULL; | |
8de66a0c | 3603 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3604 | } |
3605 | } | |
3606 | up_read(&slub_lock); | |
3607 | } | |
3608 | ||
3609 | static int slab_mem_going_online_callback(void *arg) | |
3610 | { | |
3611 | struct kmem_cache_node *n; | |
3612 | struct kmem_cache *s; | |
3613 | struct memory_notify *marg = arg; | |
3614 | int nid = marg->status_change_nid; | |
3615 | int ret = 0; | |
3616 | ||
3617 | /* | |
3618 | * If the node's memory is already available, then kmem_cache_node is | |
3619 | * already created. Nothing to do. | |
3620 | */ | |
3621 | if (nid < 0) | |
3622 | return 0; | |
3623 | ||
3624 | /* | |
0121c619 | 3625 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3626 | * allocate a kmem_cache_node structure in order to bring the node |
3627 | * online. | |
3628 | */ | |
3629 | down_read(&slub_lock); | |
3630 | list_for_each_entry(s, &slab_caches, list) { | |
3631 | /* | |
3632 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3633 | * since memory is not yet available from the node that | |
3634 | * is brought up. | |
3635 | */ | |
8de66a0c | 3636 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3637 | if (!n) { |
3638 | ret = -ENOMEM; | |
3639 | goto out; | |
3640 | } | |
5595cffc | 3641 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3642 | s->node[nid] = n; |
3643 | } | |
3644 | out: | |
3645 | up_read(&slub_lock); | |
3646 | return ret; | |
3647 | } | |
3648 | ||
3649 | static int slab_memory_callback(struct notifier_block *self, | |
3650 | unsigned long action, void *arg) | |
3651 | { | |
3652 | int ret = 0; | |
3653 | ||
3654 | switch (action) { | |
3655 | case MEM_GOING_ONLINE: | |
3656 | ret = slab_mem_going_online_callback(arg); | |
3657 | break; | |
3658 | case MEM_GOING_OFFLINE: | |
3659 | ret = slab_mem_going_offline_callback(arg); | |
3660 | break; | |
3661 | case MEM_OFFLINE: | |
3662 | case MEM_CANCEL_ONLINE: | |
3663 | slab_mem_offline_callback(arg); | |
3664 | break; | |
3665 | case MEM_ONLINE: | |
3666 | case MEM_CANCEL_OFFLINE: | |
3667 | break; | |
3668 | } | |
dc19f9db KH |
3669 | if (ret) |
3670 | ret = notifier_from_errno(ret); | |
3671 | else | |
3672 | ret = NOTIFY_OK; | |
b9049e23 YG |
3673 | return ret; |
3674 | } | |
3675 | ||
3676 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3677 | ||
81819f0f CL |
3678 | /******************************************************************** |
3679 | * Basic setup of slabs | |
3680 | *******************************************************************/ | |
3681 | ||
51df1142 CL |
3682 | /* |
3683 | * Used for early kmem_cache structures that were allocated using | |
3684 | * the page allocator | |
3685 | */ | |
3686 | ||
3687 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3688 | { | |
3689 | int node; | |
3690 | ||
3691 | list_add(&s->list, &slab_caches); | |
3692 | s->refcount = -1; | |
3693 | ||
3694 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3695 | struct kmem_cache_node *n = get_node(s, node); | |
3696 | struct page *p; | |
3697 | ||
3698 | if (n) { | |
3699 | list_for_each_entry(p, &n->partial, lru) | |
3700 | p->slab = s; | |
3701 | ||
607bf324 | 3702 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3703 | list_for_each_entry(p, &n->full, lru) |
3704 | p->slab = s; | |
3705 | #endif | |
3706 | } | |
3707 | } | |
3708 | } | |
3709 | ||
81819f0f CL |
3710 | void __init kmem_cache_init(void) |
3711 | { | |
3712 | int i; | |
4b356be0 | 3713 | int caches = 0; |
51df1142 CL |
3714 | struct kmem_cache *temp_kmem_cache; |
3715 | int order; | |
51df1142 CL |
3716 | struct kmem_cache *temp_kmem_cache_node; |
3717 | unsigned long kmalloc_size; | |
3718 | ||
3719 | kmem_size = offsetof(struct kmem_cache, node) + | |
3720 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3721 | ||
3722 | /* Allocate two kmem_caches from the page allocator */ | |
3723 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3724 | order = get_order(2 * kmalloc_size); | |
3725 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3726 | ||
81819f0f CL |
3727 | /* |
3728 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3729 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3730 | * kmem_cache_open for slab_state == DOWN. |
3731 | */ | |
51df1142 CL |
3732 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3733 | ||
3734 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3735 | sizeof(struct kmem_cache_node), | |
3736 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3737 | |
0c40ba4f | 3738 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3739 | |
3740 | /* Able to allocate the per node structures */ | |
3741 | slab_state = PARTIAL; | |
3742 | ||
51df1142 CL |
3743 | temp_kmem_cache = kmem_cache; |
3744 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3745 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3746 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3747 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3748 | |
51df1142 CL |
3749 | /* |
3750 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3751 | * kmem_cache_node is separately allocated so no need to | |
3752 | * update any list pointers. | |
3753 | */ | |
3754 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3755 | |
51df1142 CL |
3756 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3757 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3758 | ||
3759 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3760 | ||
3761 | caches++; | |
51df1142 CL |
3762 | kmem_cache_bootstrap_fixup(kmem_cache); |
3763 | caches++; | |
3764 | /* Free temporary boot structure */ | |
3765 | free_pages((unsigned long)temp_kmem_cache, order); | |
3766 | ||
3767 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3768 | |
3769 | /* | |
3770 | * Patch up the size_index table if we have strange large alignment | |
3771 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3772 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3773 | * |
3774 | * Largest permitted alignment is 256 bytes due to the way we | |
3775 | * handle the index determination for the smaller caches. | |
3776 | * | |
3777 | * Make sure that nothing crazy happens if someone starts tinkering | |
3778 | * around with ARCH_KMALLOC_MINALIGN | |
3779 | */ | |
3780 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3781 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3782 | ||
acdfcd04 AK |
3783 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3784 | int elem = size_index_elem(i); | |
3785 | if (elem >= ARRAY_SIZE(size_index)) | |
3786 | break; | |
3787 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3788 | } | |
f1b26339 | 3789 | |
acdfcd04 AK |
3790 | if (KMALLOC_MIN_SIZE == 64) { |
3791 | /* | |
3792 | * The 96 byte size cache is not used if the alignment | |
3793 | * is 64 byte. | |
3794 | */ | |
3795 | for (i = 64 + 8; i <= 96; i += 8) | |
3796 | size_index[size_index_elem(i)] = 7; | |
3797 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3798 | /* |
3799 | * The 192 byte sized cache is not used if the alignment | |
3800 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3801 | * instead. | |
3802 | */ | |
3803 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3804 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3805 | } |
3806 | ||
51df1142 CL |
3807 | /* Caches that are not of the two-to-the-power-of size */ |
3808 | if (KMALLOC_MIN_SIZE <= 32) { | |
3809 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3810 | caches++; | |
3811 | } | |
3812 | ||
3813 | if (KMALLOC_MIN_SIZE <= 64) { | |
3814 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3815 | caches++; | |
3816 | } | |
3817 | ||
3818 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3819 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3820 | caches++; | |
3821 | } | |
3822 | ||
81819f0f CL |
3823 | slab_state = UP; |
3824 | ||
3825 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3826 | if (KMALLOC_MIN_SIZE <= 32) { |
3827 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3828 | BUG_ON(!kmalloc_caches[1]->name); | |
3829 | } | |
3830 | ||
3831 | if (KMALLOC_MIN_SIZE <= 64) { | |
3832 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3833 | BUG_ON(!kmalloc_caches[2]->name); | |
3834 | } | |
3835 | ||
d7278bd7 CL |
3836 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3837 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3838 | ||
3839 | BUG_ON(!s); | |
51df1142 | 3840 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3841 | } |
81819f0f CL |
3842 | |
3843 | #ifdef CONFIG_SMP | |
3844 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3845 | #endif |
81819f0f | 3846 | |
55136592 | 3847 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3848 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3849 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3850 | |
51df1142 | 3851 | if (s && s->size) { |
55136592 CL |
3852 | char *name = kasprintf(GFP_NOWAIT, |
3853 | "dma-kmalloc-%d", s->objsize); | |
3854 | ||
3855 | BUG_ON(!name); | |
51df1142 CL |
3856 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3857 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3858 | } |
3859 | } | |
3860 | #endif | |
3adbefee IM |
3861 | printk(KERN_INFO |
3862 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3863 | " CPUs=%d, Nodes=%d\n", |
3864 | caches, cache_line_size(), | |
81819f0f CL |
3865 | slub_min_order, slub_max_order, slub_min_objects, |
3866 | nr_cpu_ids, nr_node_ids); | |
3867 | } | |
3868 | ||
7e85ee0c PE |
3869 | void __init kmem_cache_init_late(void) |
3870 | { | |
7e85ee0c PE |
3871 | } |
3872 | ||
81819f0f CL |
3873 | /* |
3874 | * Find a mergeable slab cache | |
3875 | */ | |
3876 | static int slab_unmergeable(struct kmem_cache *s) | |
3877 | { | |
3878 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3879 | return 1; | |
3880 | ||
c59def9f | 3881 | if (s->ctor) |
81819f0f CL |
3882 | return 1; |
3883 | ||
8ffa6875 CL |
3884 | /* |
3885 | * We may have set a slab to be unmergeable during bootstrap. | |
3886 | */ | |
3887 | if (s->refcount < 0) | |
3888 | return 1; | |
3889 | ||
81819f0f CL |
3890 | return 0; |
3891 | } | |
3892 | ||
3893 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3894 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3895 | void (*ctor)(void *)) |
81819f0f | 3896 | { |
5b95a4ac | 3897 | struct kmem_cache *s; |
81819f0f CL |
3898 | |
3899 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3900 | return NULL; | |
3901 | ||
c59def9f | 3902 | if (ctor) |
81819f0f CL |
3903 | return NULL; |
3904 | ||
3905 | size = ALIGN(size, sizeof(void *)); | |
3906 | align = calculate_alignment(flags, align, size); | |
3907 | size = ALIGN(size, align); | |
ba0268a8 | 3908 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3909 | |
5b95a4ac | 3910 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3911 | if (slab_unmergeable(s)) |
3912 | continue; | |
3913 | ||
3914 | if (size > s->size) | |
3915 | continue; | |
3916 | ||
ba0268a8 | 3917 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3918 | continue; |
3919 | /* | |
3920 | * Check if alignment is compatible. | |
3921 | * Courtesy of Adrian Drzewiecki | |
3922 | */ | |
06428780 | 3923 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3924 | continue; |
3925 | ||
3926 | if (s->size - size >= sizeof(void *)) | |
3927 | continue; | |
3928 | ||
3929 | return s; | |
3930 | } | |
3931 | return NULL; | |
3932 | } | |
3933 | ||
3934 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3935 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3936 | { |
3937 | struct kmem_cache *s; | |
84c1cf62 | 3938 | char *n; |
81819f0f | 3939 | |
fe1ff49d BH |
3940 | if (WARN_ON(!name)) |
3941 | return NULL; | |
3942 | ||
81819f0f | 3943 | down_write(&slub_lock); |
ba0268a8 | 3944 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3945 | if (s) { |
3946 | s->refcount++; | |
3947 | /* | |
3948 | * Adjust the object sizes so that we clear | |
3949 | * the complete object on kzalloc. | |
3950 | */ | |
3951 | s->objsize = max(s->objsize, (int)size); | |
3952 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3953 | |
7b8f3b66 | 3954 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3955 | s->refcount--; |
81819f0f | 3956 | goto err; |
7b8f3b66 | 3957 | } |
2bce6485 | 3958 | up_write(&slub_lock); |
a0e1d1be CL |
3959 | return s; |
3960 | } | |
6446faa2 | 3961 | |
84c1cf62 PE |
3962 | n = kstrdup(name, GFP_KERNEL); |
3963 | if (!n) | |
3964 | goto err; | |
3965 | ||
a0e1d1be CL |
3966 | s = kmalloc(kmem_size, GFP_KERNEL); |
3967 | if (s) { | |
84c1cf62 | 3968 | if (kmem_cache_open(s, n, |
c59def9f | 3969 | size, align, flags, ctor)) { |
81819f0f | 3970 | list_add(&s->list, &slab_caches); |
7b8f3b66 | 3971 | if (sysfs_slab_add(s)) { |
7b8f3b66 | 3972 | list_del(&s->list); |
84c1cf62 | 3973 | kfree(n); |
7b8f3b66 | 3974 | kfree(s); |
a0e1d1be | 3975 | goto err; |
7b8f3b66 | 3976 | } |
2bce6485 | 3977 | up_write(&slub_lock); |
a0e1d1be CL |
3978 | return s; |
3979 | } | |
84c1cf62 | 3980 | kfree(n); |
a0e1d1be | 3981 | kfree(s); |
81819f0f | 3982 | } |
68cee4f1 | 3983 | err: |
81819f0f | 3984 | up_write(&slub_lock); |
81819f0f | 3985 | |
81819f0f CL |
3986 | if (flags & SLAB_PANIC) |
3987 | panic("Cannot create slabcache %s\n", name); | |
3988 | else | |
3989 | s = NULL; | |
3990 | return s; | |
3991 | } | |
3992 | EXPORT_SYMBOL(kmem_cache_create); | |
3993 | ||
81819f0f | 3994 | #ifdef CONFIG_SMP |
81819f0f | 3995 | /* |
672bba3a CL |
3996 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3997 | * necessary. | |
81819f0f CL |
3998 | */ |
3999 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
4000 | unsigned long action, void *hcpu) | |
4001 | { | |
4002 | long cpu = (long)hcpu; | |
5b95a4ac CL |
4003 | struct kmem_cache *s; |
4004 | unsigned long flags; | |
81819f0f CL |
4005 | |
4006 | switch (action) { | |
4007 | case CPU_UP_CANCELED: | |
8bb78442 | 4008 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4009 | case CPU_DEAD: |
8bb78442 | 4010 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
4011 | down_read(&slub_lock); |
4012 | list_for_each_entry(s, &slab_caches, list) { | |
4013 | local_irq_save(flags); | |
4014 | __flush_cpu_slab(s, cpu); | |
4015 | local_irq_restore(flags); | |
4016 | } | |
4017 | up_read(&slub_lock); | |
81819f0f CL |
4018 | break; |
4019 | default: | |
4020 | break; | |
4021 | } | |
4022 | return NOTIFY_OK; | |
4023 | } | |
4024 | ||
06428780 | 4025 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 4026 | .notifier_call = slab_cpuup_callback |
06428780 | 4027 | }; |
81819f0f CL |
4028 | |
4029 | #endif | |
4030 | ||
ce71e27c | 4031 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4032 | { |
aadb4bc4 | 4033 | struct kmem_cache *s; |
94b528d0 | 4034 | void *ret; |
aadb4bc4 | 4035 | |
ffadd4d0 | 4036 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
4037 | return kmalloc_large(size, gfpflags); |
4038 | ||
aadb4bc4 | 4039 | s = get_slab(size, gfpflags); |
81819f0f | 4040 | |
2408c550 | 4041 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4042 | return s; |
81819f0f | 4043 | |
2154a336 | 4044 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 4045 | |
25985edc | 4046 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4047 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4048 | |
4049 | return ret; | |
81819f0f CL |
4050 | } |
4051 | ||
5d1f57e4 | 4052 | #ifdef CONFIG_NUMA |
81819f0f | 4053 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4054 | int node, unsigned long caller) |
81819f0f | 4055 | { |
aadb4bc4 | 4056 | struct kmem_cache *s; |
94b528d0 | 4057 | void *ret; |
aadb4bc4 | 4058 | |
d3e14aa3 XF |
4059 | if (unlikely(size > SLUB_MAX_SIZE)) { |
4060 | ret = kmalloc_large_node(size, gfpflags, node); | |
4061 | ||
4062 | trace_kmalloc_node(caller, ret, | |
4063 | size, PAGE_SIZE << get_order(size), | |
4064 | gfpflags, node); | |
4065 | ||
4066 | return ret; | |
4067 | } | |
eada35ef | 4068 | |
aadb4bc4 | 4069 | s = get_slab(size, gfpflags); |
81819f0f | 4070 | |
2408c550 | 4071 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4072 | return s; |
81819f0f | 4073 | |
94b528d0 EGM |
4074 | ret = slab_alloc(s, gfpflags, node, caller); |
4075 | ||
25985edc | 4076 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4077 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4078 | |
4079 | return ret; | |
81819f0f | 4080 | } |
5d1f57e4 | 4081 | #endif |
81819f0f | 4082 | |
ab4d5ed5 | 4083 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4084 | static int count_inuse(struct page *page) |
4085 | { | |
4086 | return page->inuse; | |
4087 | } | |
4088 | ||
4089 | static int count_total(struct page *page) | |
4090 | { | |
4091 | return page->objects; | |
4092 | } | |
ab4d5ed5 | 4093 | #endif |
205ab99d | 4094 | |
ab4d5ed5 | 4095 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4096 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4097 | unsigned long *map) | |
53e15af0 CL |
4098 | { |
4099 | void *p; | |
a973e9dd | 4100 | void *addr = page_address(page); |
53e15af0 CL |
4101 | |
4102 | if (!check_slab(s, page) || | |
4103 | !on_freelist(s, page, NULL)) | |
4104 | return 0; | |
4105 | ||
4106 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4107 | bitmap_zero(map, page->objects); |
53e15af0 | 4108 | |
5f80b13a CL |
4109 | get_map(s, page, map); |
4110 | for_each_object(p, s, addr, page->objects) { | |
4111 | if (test_bit(slab_index(p, s, addr), map)) | |
4112 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4113 | return 0; | |
53e15af0 CL |
4114 | } |
4115 | ||
224a88be | 4116 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4117 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4118 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4119 | return 0; |
4120 | return 1; | |
4121 | } | |
4122 | ||
434e245d CL |
4123 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4124 | unsigned long *map) | |
53e15af0 | 4125 | { |
881db7fb CL |
4126 | slab_lock(page); |
4127 | validate_slab(s, page, map); | |
4128 | slab_unlock(page); | |
53e15af0 CL |
4129 | } |
4130 | ||
434e245d CL |
4131 | static int validate_slab_node(struct kmem_cache *s, |
4132 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4133 | { |
4134 | unsigned long count = 0; | |
4135 | struct page *page; | |
4136 | unsigned long flags; | |
4137 | ||
4138 | spin_lock_irqsave(&n->list_lock, flags); | |
4139 | ||
4140 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4141 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4142 | count++; |
4143 | } | |
4144 | if (count != n->nr_partial) | |
4145 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4146 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4147 | ||
4148 | if (!(s->flags & SLAB_STORE_USER)) | |
4149 | goto out; | |
4150 | ||
4151 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4152 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4153 | count++; |
4154 | } | |
4155 | if (count != atomic_long_read(&n->nr_slabs)) | |
4156 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4157 | "counter=%ld\n", s->name, count, | |
4158 | atomic_long_read(&n->nr_slabs)); | |
4159 | ||
4160 | out: | |
4161 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4162 | return count; | |
4163 | } | |
4164 | ||
434e245d | 4165 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4166 | { |
4167 | int node; | |
4168 | unsigned long count = 0; | |
205ab99d | 4169 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4170 | sizeof(unsigned long), GFP_KERNEL); |
4171 | ||
4172 | if (!map) | |
4173 | return -ENOMEM; | |
53e15af0 CL |
4174 | |
4175 | flush_all(s); | |
f64dc58c | 4176 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4177 | struct kmem_cache_node *n = get_node(s, node); |
4178 | ||
434e245d | 4179 | count += validate_slab_node(s, n, map); |
53e15af0 | 4180 | } |
434e245d | 4181 | kfree(map); |
53e15af0 CL |
4182 | return count; |
4183 | } | |
88a420e4 | 4184 | /* |
672bba3a | 4185 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4186 | * and freed. |
4187 | */ | |
4188 | ||
4189 | struct location { | |
4190 | unsigned long count; | |
ce71e27c | 4191 | unsigned long addr; |
45edfa58 CL |
4192 | long long sum_time; |
4193 | long min_time; | |
4194 | long max_time; | |
4195 | long min_pid; | |
4196 | long max_pid; | |
174596a0 | 4197 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4198 | nodemask_t nodes; |
88a420e4 CL |
4199 | }; |
4200 | ||
4201 | struct loc_track { | |
4202 | unsigned long max; | |
4203 | unsigned long count; | |
4204 | struct location *loc; | |
4205 | }; | |
4206 | ||
4207 | static void free_loc_track(struct loc_track *t) | |
4208 | { | |
4209 | if (t->max) | |
4210 | free_pages((unsigned long)t->loc, | |
4211 | get_order(sizeof(struct location) * t->max)); | |
4212 | } | |
4213 | ||
68dff6a9 | 4214 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4215 | { |
4216 | struct location *l; | |
4217 | int order; | |
4218 | ||
88a420e4 CL |
4219 | order = get_order(sizeof(struct location) * max); |
4220 | ||
68dff6a9 | 4221 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4222 | if (!l) |
4223 | return 0; | |
4224 | ||
4225 | if (t->count) { | |
4226 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4227 | free_loc_track(t); | |
4228 | } | |
4229 | t->max = max; | |
4230 | t->loc = l; | |
4231 | return 1; | |
4232 | } | |
4233 | ||
4234 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4235 | const struct track *track) |
88a420e4 CL |
4236 | { |
4237 | long start, end, pos; | |
4238 | struct location *l; | |
ce71e27c | 4239 | unsigned long caddr; |
45edfa58 | 4240 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4241 | |
4242 | start = -1; | |
4243 | end = t->count; | |
4244 | ||
4245 | for ( ; ; ) { | |
4246 | pos = start + (end - start + 1) / 2; | |
4247 | ||
4248 | /* | |
4249 | * There is nothing at "end". If we end up there | |
4250 | * we need to add something to before end. | |
4251 | */ | |
4252 | if (pos == end) | |
4253 | break; | |
4254 | ||
4255 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4256 | if (track->addr == caddr) { |
4257 | ||
4258 | l = &t->loc[pos]; | |
4259 | l->count++; | |
4260 | if (track->when) { | |
4261 | l->sum_time += age; | |
4262 | if (age < l->min_time) | |
4263 | l->min_time = age; | |
4264 | if (age > l->max_time) | |
4265 | l->max_time = age; | |
4266 | ||
4267 | if (track->pid < l->min_pid) | |
4268 | l->min_pid = track->pid; | |
4269 | if (track->pid > l->max_pid) | |
4270 | l->max_pid = track->pid; | |
4271 | ||
174596a0 RR |
4272 | cpumask_set_cpu(track->cpu, |
4273 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4274 | } |
4275 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4276 | return 1; |
4277 | } | |
4278 | ||
45edfa58 | 4279 | if (track->addr < caddr) |
88a420e4 CL |
4280 | end = pos; |
4281 | else | |
4282 | start = pos; | |
4283 | } | |
4284 | ||
4285 | /* | |
672bba3a | 4286 | * Not found. Insert new tracking element. |
88a420e4 | 4287 | */ |
68dff6a9 | 4288 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4289 | return 0; |
4290 | ||
4291 | l = t->loc + pos; | |
4292 | if (pos < t->count) | |
4293 | memmove(l + 1, l, | |
4294 | (t->count - pos) * sizeof(struct location)); | |
4295 | t->count++; | |
4296 | l->count = 1; | |
45edfa58 CL |
4297 | l->addr = track->addr; |
4298 | l->sum_time = age; | |
4299 | l->min_time = age; | |
4300 | l->max_time = age; | |
4301 | l->min_pid = track->pid; | |
4302 | l->max_pid = track->pid; | |
174596a0 RR |
4303 | cpumask_clear(to_cpumask(l->cpus)); |
4304 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4305 | nodes_clear(l->nodes); |
4306 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4307 | return 1; |
4308 | } | |
4309 | ||
4310 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4311 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4312 | unsigned long *map) |
88a420e4 | 4313 | { |
a973e9dd | 4314 | void *addr = page_address(page); |
88a420e4 CL |
4315 | void *p; |
4316 | ||
39b26464 | 4317 | bitmap_zero(map, page->objects); |
5f80b13a | 4318 | get_map(s, page, map); |
88a420e4 | 4319 | |
224a88be | 4320 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4321 | if (!test_bit(slab_index(p, s, addr), map)) |
4322 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4323 | } |
4324 | ||
4325 | static int list_locations(struct kmem_cache *s, char *buf, | |
4326 | enum track_item alloc) | |
4327 | { | |
e374d483 | 4328 | int len = 0; |
88a420e4 | 4329 | unsigned long i; |
68dff6a9 | 4330 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4331 | int node; |
bbd7d57b ED |
4332 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4333 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4334 | |
bbd7d57b ED |
4335 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4336 | GFP_TEMPORARY)) { | |
4337 | kfree(map); | |
68dff6a9 | 4338 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4339 | } |
88a420e4 CL |
4340 | /* Push back cpu slabs */ |
4341 | flush_all(s); | |
4342 | ||
f64dc58c | 4343 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4344 | struct kmem_cache_node *n = get_node(s, node); |
4345 | unsigned long flags; | |
4346 | struct page *page; | |
4347 | ||
9e86943b | 4348 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4349 | continue; |
4350 | ||
4351 | spin_lock_irqsave(&n->list_lock, flags); | |
4352 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4353 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4354 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4355 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4356 | spin_unlock_irqrestore(&n->list_lock, flags); |
4357 | } | |
4358 | ||
4359 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4360 | struct location *l = &t.loc[i]; |
88a420e4 | 4361 | |
9c246247 | 4362 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4363 | break; |
e374d483 | 4364 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4365 | |
4366 | if (l->addr) | |
62c70bce | 4367 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4368 | else |
e374d483 | 4369 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4370 | |
4371 | if (l->sum_time != l->min_time) { | |
e374d483 | 4372 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4373 | l->min_time, |
4374 | (long)div_u64(l->sum_time, l->count), | |
4375 | l->max_time); | |
45edfa58 | 4376 | } else |
e374d483 | 4377 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4378 | l->min_time); |
4379 | ||
4380 | if (l->min_pid != l->max_pid) | |
e374d483 | 4381 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4382 | l->min_pid, l->max_pid); |
4383 | else | |
e374d483 | 4384 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4385 | l->min_pid); |
4386 | ||
174596a0 RR |
4387 | if (num_online_cpus() > 1 && |
4388 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4389 | len < PAGE_SIZE - 60) { |
4390 | len += sprintf(buf + len, " cpus="); | |
4391 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4392 | to_cpumask(l->cpus)); |
45edfa58 CL |
4393 | } |
4394 | ||
62bc62a8 | 4395 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4396 | len < PAGE_SIZE - 60) { |
4397 | len += sprintf(buf + len, " nodes="); | |
4398 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4399 | l->nodes); |
4400 | } | |
4401 | ||
e374d483 | 4402 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4403 | } |
4404 | ||
4405 | free_loc_track(&t); | |
bbd7d57b | 4406 | kfree(map); |
88a420e4 | 4407 | if (!t.count) |
e374d483 HH |
4408 | len += sprintf(buf, "No data\n"); |
4409 | return len; | |
88a420e4 | 4410 | } |
ab4d5ed5 | 4411 | #endif |
88a420e4 | 4412 | |
a5a84755 CL |
4413 | #ifdef SLUB_RESILIENCY_TEST |
4414 | static void resiliency_test(void) | |
4415 | { | |
4416 | u8 *p; | |
4417 | ||
4418 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4419 | ||
4420 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4421 | printk(KERN_ERR "-----------------------\n"); | |
4422 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4423 | ||
4424 | p = kzalloc(16, GFP_KERNEL); | |
4425 | p[16] = 0x12; | |
4426 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4427 | " 0x12->0x%p\n\n", p + 16); | |
4428 | ||
4429 | validate_slab_cache(kmalloc_caches[4]); | |
4430 | ||
4431 | /* Hmmm... The next two are dangerous */ | |
4432 | p = kzalloc(32, GFP_KERNEL); | |
4433 | p[32 + sizeof(void *)] = 0x34; | |
4434 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4435 | " 0x34 -> -0x%p\n", p); | |
4436 | printk(KERN_ERR | |
4437 | "If allocated object is overwritten then not detectable\n\n"); | |
4438 | ||
4439 | validate_slab_cache(kmalloc_caches[5]); | |
4440 | p = kzalloc(64, GFP_KERNEL); | |
4441 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4442 | *p = 0x56; | |
4443 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4444 | p); | |
4445 | printk(KERN_ERR | |
4446 | "If allocated object is overwritten then not detectable\n\n"); | |
4447 | validate_slab_cache(kmalloc_caches[6]); | |
4448 | ||
4449 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4450 | p = kzalloc(128, GFP_KERNEL); | |
4451 | kfree(p); | |
4452 | *p = 0x78; | |
4453 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4454 | validate_slab_cache(kmalloc_caches[7]); | |
4455 | ||
4456 | p = kzalloc(256, GFP_KERNEL); | |
4457 | kfree(p); | |
4458 | p[50] = 0x9a; | |
4459 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4460 | p); | |
4461 | validate_slab_cache(kmalloc_caches[8]); | |
4462 | ||
4463 | p = kzalloc(512, GFP_KERNEL); | |
4464 | kfree(p); | |
4465 | p[512] = 0xab; | |
4466 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4467 | validate_slab_cache(kmalloc_caches[9]); | |
4468 | } | |
4469 | #else | |
4470 | #ifdef CONFIG_SYSFS | |
4471 | static void resiliency_test(void) {}; | |
4472 | #endif | |
4473 | #endif | |
4474 | ||
ab4d5ed5 | 4475 | #ifdef CONFIG_SYSFS |
81819f0f | 4476 | enum slab_stat_type { |
205ab99d CL |
4477 | SL_ALL, /* All slabs */ |
4478 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4479 | SL_CPU, /* Only slabs used for cpu caches */ | |
4480 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4481 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4482 | }; |
4483 | ||
205ab99d | 4484 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4485 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4486 | #define SO_CPU (1 << SL_CPU) | |
4487 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4488 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4489 | |
62e5c4b4 CG |
4490 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4491 | char *buf, unsigned long flags) | |
81819f0f CL |
4492 | { |
4493 | unsigned long total = 0; | |
81819f0f CL |
4494 | int node; |
4495 | int x; | |
4496 | unsigned long *nodes; | |
4497 | unsigned long *per_cpu; | |
4498 | ||
4499 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4500 | if (!nodes) |
4501 | return -ENOMEM; | |
81819f0f CL |
4502 | per_cpu = nodes + nr_node_ids; |
4503 | ||
205ab99d CL |
4504 | if (flags & SO_CPU) { |
4505 | int cpu; | |
81819f0f | 4506 | |
205ab99d | 4507 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4508 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
49e22585 | 4509 | struct page *page; |
dfb4f096 | 4510 | |
205ab99d CL |
4511 | if (!c || c->node < 0) |
4512 | continue; | |
4513 | ||
4514 | if (c->page) { | |
4515 | if (flags & SO_TOTAL) | |
4516 | x = c->page->objects; | |
4517 | else if (flags & SO_OBJECTS) | |
4518 | x = c->page->inuse; | |
81819f0f CL |
4519 | else |
4520 | x = 1; | |
205ab99d | 4521 | |
81819f0f | 4522 | total += x; |
205ab99d | 4523 | nodes[c->node] += x; |
81819f0f | 4524 | } |
49e22585 CL |
4525 | page = c->partial; |
4526 | ||
4527 | if (page) { | |
4528 | x = page->pobjects; | |
4529 | total += x; | |
4530 | nodes[c->node] += x; | |
4531 | } | |
205ab99d | 4532 | per_cpu[c->node]++; |
81819f0f CL |
4533 | } |
4534 | } | |
4535 | ||
04d94879 | 4536 | lock_memory_hotplug(); |
ab4d5ed5 | 4537 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4538 | if (flags & SO_ALL) { |
4539 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4540 | struct kmem_cache_node *n = get_node(s, node); | |
4541 | ||
4542 | if (flags & SO_TOTAL) | |
4543 | x = atomic_long_read(&n->total_objects); | |
4544 | else if (flags & SO_OBJECTS) | |
4545 | x = atomic_long_read(&n->total_objects) - | |
4546 | count_partial(n, count_free); | |
81819f0f | 4547 | |
81819f0f | 4548 | else |
205ab99d | 4549 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4550 | total += x; |
4551 | nodes[node] += x; | |
4552 | } | |
4553 | ||
ab4d5ed5 CL |
4554 | } else |
4555 | #endif | |
4556 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4557 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4558 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4559 | |
205ab99d CL |
4560 | if (flags & SO_TOTAL) |
4561 | x = count_partial(n, count_total); | |
4562 | else if (flags & SO_OBJECTS) | |
4563 | x = count_partial(n, count_inuse); | |
81819f0f | 4564 | else |
205ab99d | 4565 | x = n->nr_partial; |
81819f0f CL |
4566 | total += x; |
4567 | nodes[node] += x; | |
4568 | } | |
4569 | } | |
81819f0f CL |
4570 | x = sprintf(buf, "%lu", total); |
4571 | #ifdef CONFIG_NUMA | |
f64dc58c | 4572 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4573 | if (nodes[node]) |
4574 | x += sprintf(buf + x, " N%d=%lu", | |
4575 | node, nodes[node]); | |
4576 | #endif | |
04d94879 | 4577 | unlock_memory_hotplug(); |
81819f0f CL |
4578 | kfree(nodes); |
4579 | return x + sprintf(buf + x, "\n"); | |
4580 | } | |
4581 | ||
ab4d5ed5 | 4582 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4583 | static int any_slab_objects(struct kmem_cache *s) |
4584 | { | |
4585 | int node; | |
81819f0f | 4586 | |
dfb4f096 | 4587 | for_each_online_node(node) { |
81819f0f CL |
4588 | struct kmem_cache_node *n = get_node(s, node); |
4589 | ||
dfb4f096 CL |
4590 | if (!n) |
4591 | continue; | |
4592 | ||
4ea33e2d | 4593 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4594 | return 1; |
4595 | } | |
4596 | return 0; | |
4597 | } | |
ab4d5ed5 | 4598 | #endif |
81819f0f CL |
4599 | |
4600 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4601 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4602 | |
4603 | struct slab_attribute { | |
4604 | struct attribute attr; | |
4605 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4606 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4607 | }; | |
4608 | ||
4609 | #define SLAB_ATTR_RO(_name) \ | |
4610 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
4611 | ||
4612 | #define SLAB_ATTR(_name) \ | |
4613 | static struct slab_attribute _name##_attr = \ | |
4614 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
4615 | ||
81819f0f CL |
4616 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4617 | { | |
4618 | return sprintf(buf, "%d\n", s->size); | |
4619 | } | |
4620 | SLAB_ATTR_RO(slab_size); | |
4621 | ||
4622 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4623 | { | |
4624 | return sprintf(buf, "%d\n", s->align); | |
4625 | } | |
4626 | SLAB_ATTR_RO(align); | |
4627 | ||
4628 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4629 | { | |
4630 | return sprintf(buf, "%d\n", s->objsize); | |
4631 | } | |
4632 | SLAB_ATTR_RO(object_size); | |
4633 | ||
4634 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4635 | { | |
834f3d11 | 4636 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4637 | } |
4638 | SLAB_ATTR_RO(objs_per_slab); | |
4639 | ||
06b285dc CL |
4640 | static ssize_t order_store(struct kmem_cache *s, |
4641 | const char *buf, size_t length) | |
4642 | { | |
0121c619 CL |
4643 | unsigned long order; |
4644 | int err; | |
4645 | ||
4646 | err = strict_strtoul(buf, 10, &order); | |
4647 | if (err) | |
4648 | return err; | |
06b285dc CL |
4649 | |
4650 | if (order > slub_max_order || order < slub_min_order) | |
4651 | return -EINVAL; | |
4652 | ||
4653 | calculate_sizes(s, order); | |
4654 | return length; | |
4655 | } | |
4656 | ||
81819f0f CL |
4657 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4658 | { | |
834f3d11 | 4659 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4660 | } |
06b285dc | 4661 | SLAB_ATTR(order); |
81819f0f | 4662 | |
73d342b1 DR |
4663 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4664 | { | |
4665 | return sprintf(buf, "%lu\n", s->min_partial); | |
4666 | } | |
4667 | ||
4668 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4669 | size_t length) | |
4670 | { | |
4671 | unsigned long min; | |
4672 | int err; | |
4673 | ||
4674 | err = strict_strtoul(buf, 10, &min); | |
4675 | if (err) | |
4676 | return err; | |
4677 | ||
c0bdb232 | 4678 | set_min_partial(s, min); |
73d342b1 DR |
4679 | return length; |
4680 | } | |
4681 | SLAB_ATTR(min_partial); | |
4682 | ||
49e22585 CL |
4683 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4684 | { | |
4685 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4686 | } | |
4687 | ||
4688 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4689 | size_t length) | |
4690 | { | |
4691 | unsigned long objects; | |
4692 | int err; | |
4693 | ||
4694 | err = strict_strtoul(buf, 10, &objects); | |
4695 | if (err) | |
4696 | return err; | |
4697 | ||
4698 | s->cpu_partial = objects; | |
4699 | flush_all(s); | |
4700 | return length; | |
4701 | } | |
4702 | SLAB_ATTR(cpu_partial); | |
4703 | ||
81819f0f CL |
4704 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4705 | { | |
62c70bce JP |
4706 | if (!s->ctor) |
4707 | return 0; | |
4708 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4709 | } |
4710 | SLAB_ATTR_RO(ctor); | |
4711 | ||
81819f0f CL |
4712 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4713 | { | |
4714 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4715 | } | |
4716 | SLAB_ATTR_RO(aliases); | |
4717 | ||
81819f0f CL |
4718 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4719 | { | |
d9acf4b7 | 4720 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4721 | } |
4722 | SLAB_ATTR_RO(partial); | |
4723 | ||
4724 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4725 | { | |
d9acf4b7 | 4726 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4727 | } |
4728 | SLAB_ATTR_RO(cpu_slabs); | |
4729 | ||
4730 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4731 | { | |
205ab99d | 4732 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4733 | } |
4734 | SLAB_ATTR_RO(objects); | |
4735 | ||
205ab99d CL |
4736 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4737 | { | |
4738 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4739 | } | |
4740 | SLAB_ATTR_RO(objects_partial); | |
4741 | ||
49e22585 CL |
4742 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4743 | { | |
4744 | int objects = 0; | |
4745 | int pages = 0; | |
4746 | int cpu; | |
4747 | int len; | |
4748 | ||
4749 | for_each_online_cpu(cpu) { | |
4750 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4751 | ||
4752 | if (page) { | |
4753 | pages += page->pages; | |
4754 | objects += page->pobjects; | |
4755 | } | |
4756 | } | |
4757 | ||
4758 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4759 | ||
4760 | #ifdef CONFIG_SMP | |
4761 | for_each_online_cpu(cpu) { | |
4762 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4763 | ||
4764 | if (page && len < PAGE_SIZE - 20) | |
4765 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4766 | page->pobjects, page->pages); | |
4767 | } | |
4768 | #endif | |
4769 | return len + sprintf(buf + len, "\n"); | |
4770 | } | |
4771 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4772 | ||
a5a84755 CL |
4773 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4774 | { | |
4775 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4776 | } | |
4777 | ||
4778 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4779 | const char *buf, size_t length) | |
4780 | { | |
4781 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4782 | if (buf[0] == '1') | |
4783 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4784 | return length; | |
4785 | } | |
4786 | SLAB_ATTR(reclaim_account); | |
4787 | ||
4788 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4789 | { | |
4790 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4791 | } | |
4792 | SLAB_ATTR_RO(hwcache_align); | |
4793 | ||
4794 | #ifdef CONFIG_ZONE_DMA | |
4795 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4796 | { | |
4797 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4798 | } | |
4799 | SLAB_ATTR_RO(cache_dma); | |
4800 | #endif | |
4801 | ||
4802 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4803 | { | |
4804 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4805 | } | |
4806 | SLAB_ATTR_RO(destroy_by_rcu); | |
4807 | ||
ab9a0f19 LJ |
4808 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4809 | { | |
4810 | return sprintf(buf, "%d\n", s->reserved); | |
4811 | } | |
4812 | SLAB_ATTR_RO(reserved); | |
4813 | ||
ab4d5ed5 | 4814 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4815 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4816 | { | |
4817 | return show_slab_objects(s, buf, SO_ALL); | |
4818 | } | |
4819 | SLAB_ATTR_RO(slabs); | |
4820 | ||
205ab99d CL |
4821 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4822 | { | |
4823 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4824 | } | |
4825 | SLAB_ATTR_RO(total_objects); | |
4826 | ||
81819f0f CL |
4827 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4828 | { | |
4829 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4830 | } | |
4831 | ||
4832 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4833 | const char *buf, size_t length) | |
4834 | { | |
4835 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4836 | if (buf[0] == '1') { |
4837 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4838 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4839 | } |
81819f0f CL |
4840 | return length; |
4841 | } | |
4842 | SLAB_ATTR(sanity_checks); | |
4843 | ||
4844 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4845 | { | |
4846 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4847 | } | |
4848 | ||
4849 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4850 | size_t length) | |
4851 | { | |
4852 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4853 | if (buf[0] == '1') { |
4854 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4855 | s->flags |= SLAB_TRACE; |
b789ef51 | 4856 | } |
81819f0f CL |
4857 | return length; |
4858 | } | |
4859 | SLAB_ATTR(trace); | |
4860 | ||
81819f0f CL |
4861 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4862 | { | |
4863 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4864 | } | |
4865 | ||
4866 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4867 | const char *buf, size_t length) | |
4868 | { | |
4869 | if (any_slab_objects(s)) | |
4870 | return -EBUSY; | |
4871 | ||
4872 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4873 | if (buf[0] == '1') { |
4874 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4875 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4876 | } |
06b285dc | 4877 | calculate_sizes(s, -1); |
81819f0f CL |
4878 | return length; |
4879 | } | |
4880 | SLAB_ATTR(red_zone); | |
4881 | ||
4882 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4883 | { | |
4884 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4885 | } | |
4886 | ||
4887 | static ssize_t poison_store(struct kmem_cache *s, | |
4888 | const char *buf, size_t length) | |
4889 | { | |
4890 | if (any_slab_objects(s)) | |
4891 | return -EBUSY; | |
4892 | ||
4893 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4894 | if (buf[0] == '1') { |
4895 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4896 | s->flags |= SLAB_POISON; |
b789ef51 | 4897 | } |
06b285dc | 4898 | calculate_sizes(s, -1); |
81819f0f CL |
4899 | return length; |
4900 | } | |
4901 | SLAB_ATTR(poison); | |
4902 | ||
4903 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4904 | { | |
4905 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4906 | } | |
4907 | ||
4908 | static ssize_t store_user_store(struct kmem_cache *s, | |
4909 | const char *buf, size_t length) | |
4910 | { | |
4911 | if (any_slab_objects(s)) | |
4912 | return -EBUSY; | |
4913 | ||
4914 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4915 | if (buf[0] == '1') { |
4916 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4917 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4918 | } |
06b285dc | 4919 | calculate_sizes(s, -1); |
81819f0f CL |
4920 | return length; |
4921 | } | |
4922 | SLAB_ATTR(store_user); | |
4923 | ||
53e15af0 CL |
4924 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4925 | { | |
4926 | return 0; | |
4927 | } | |
4928 | ||
4929 | static ssize_t validate_store(struct kmem_cache *s, | |
4930 | const char *buf, size_t length) | |
4931 | { | |
434e245d CL |
4932 | int ret = -EINVAL; |
4933 | ||
4934 | if (buf[0] == '1') { | |
4935 | ret = validate_slab_cache(s); | |
4936 | if (ret >= 0) | |
4937 | ret = length; | |
4938 | } | |
4939 | return ret; | |
53e15af0 CL |
4940 | } |
4941 | SLAB_ATTR(validate); | |
a5a84755 CL |
4942 | |
4943 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4944 | { | |
4945 | if (!(s->flags & SLAB_STORE_USER)) | |
4946 | return -ENOSYS; | |
4947 | return list_locations(s, buf, TRACK_ALLOC); | |
4948 | } | |
4949 | SLAB_ATTR_RO(alloc_calls); | |
4950 | ||
4951 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4952 | { | |
4953 | if (!(s->flags & SLAB_STORE_USER)) | |
4954 | return -ENOSYS; | |
4955 | return list_locations(s, buf, TRACK_FREE); | |
4956 | } | |
4957 | SLAB_ATTR_RO(free_calls); | |
4958 | #endif /* CONFIG_SLUB_DEBUG */ | |
4959 | ||
4960 | #ifdef CONFIG_FAILSLAB | |
4961 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4962 | { | |
4963 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4964 | } | |
4965 | ||
4966 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4967 | size_t length) | |
4968 | { | |
4969 | s->flags &= ~SLAB_FAILSLAB; | |
4970 | if (buf[0] == '1') | |
4971 | s->flags |= SLAB_FAILSLAB; | |
4972 | return length; | |
4973 | } | |
4974 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4975 | #endif |
53e15af0 | 4976 | |
2086d26a CL |
4977 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4978 | { | |
4979 | return 0; | |
4980 | } | |
4981 | ||
4982 | static ssize_t shrink_store(struct kmem_cache *s, | |
4983 | const char *buf, size_t length) | |
4984 | { | |
4985 | if (buf[0] == '1') { | |
4986 | int rc = kmem_cache_shrink(s); | |
4987 | ||
4988 | if (rc) | |
4989 | return rc; | |
4990 | } else | |
4991 | return -EINVAL; | |
4992 | return length; | |
4993 | } | |
4994 | SLAB_ATTR(shrink); | |
4995 | ||
81819f0f | 4996 | #ifdef CONFIG_NUMA |
9824601e | 4997 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4998 | { |
9824601e | 4999 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5000 | } |
5001 | ||
9824601e | 5002 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5003 | const char *buf, size_t length) |
5004 | { | |
0121c619 CL |
5005 | unsigned long ratio; |
5006 | int err; | |
5007 | ||
5008 | err = strict_strtoul(buf, 10, &ratio); | |
5009 | if (err) | |
5010 | return err; | |
5011 | ||
e2cb96b7 | 5012 | if (ratio <= 100) |
0121c619 | 5013 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5014 | |
81819f0f CL |
5015 | return length; |
5016 | } | |
9824601e | 5017 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5018 | #endif |
5019 | ||
8ff12cfc | 5020 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5021 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5022 | { | |
5023 | unsigned long sum = 0; | |
5024 | int cpu; | |
5025 | int len; | |
5026 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5027 | ||
5028 | if (!data) | |
5029 | return -ENOMEM; | |
5030 | ||
5031 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5032 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5033 | |
5034 | data[cpu] = x; | |
5035 | sum += x; | |
5036 | } | |
5037 | ||
5038 | len = sprintf(buf, "%lu", sum); | |
5039 | ||
50ef37b9 | 5040 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5041 | for_each_online_cpu(cpu) { |
5042 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5043 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5044 | } |
50ef37b9 | 5045 | #endif |
8ff12cfc CL |
5046 | kfree(data); |
5047 | return len + sprintf(buf + len, "\n"); | |
5048 | } | |
5049 | ||
78eb00cc DR |
5050 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5051 | { | |
5052 | int cpu; | |
5053 | ||
5054 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5055 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5056 | } |
5057 | ||
8ff12cfc CL |
5058 | #define STAT_ATTR(si, text) \ |
5059 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5060 | { \ | |
5061 | return show_stat(s, buf, si); \ | |
5062 | } \ | |
78eb00cc DR |
5063 | static ssize_t text##_store(struct kmem_cache *s, \ |
5064 | const char *buf, size_t length) \ | |
5065 | { \ | |
5066 | if (buf[0] != '0') \ | |
5067 | return -EINVAL; \ | |
5068 | clear_stat(s, si); \ | |
5069 | return length; \ | |
5070 | } \ | |
5071 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5072 | |
5073 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5074 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5075 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5076 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5077 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5078 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5079 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5080 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5081 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5082 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5083 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5084 | STAT_ATTR(FREE_SLAB, free_slab); |
5085 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5086 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5087 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5088 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5089 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5090 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5091 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5092 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5093 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5094 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5095 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5096 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8ff12cfc CL |
5097 | #endif |
5098 | ||
06428780 | 5099 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5100 | &slab_size_attr.attr, |
5101 | &object_size_attr.attr, | |
5102 | &objs_per_slab_attr.attr, | |
5103 | &order_attr.attr, | |
73d342b1 | 5104 | &min_partial_attr.attr, |
49e22585 | 5105 | &cpu_partial_attr.attr, |
81819f0f | 5106 | &objects_attr.attr, |
205ab99d | 5107 | &objects_partial_attr.attr, |
81819f0f CL |
5108 | &partial_attr.attr, |
5109 | &cpu_slabs_attr.attr, | |
5110 | &ctor_attr.attr, | |
81819f0f CL |
5111 | &aliases_attr.attr, |
5112 | &align_attr.attr, | |
81819f0f CL |
5113 | &hwcache_align_attr.attr, |
5114 | &reclaim_account_attr.attr, | |
5115 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5116 | &shrink_attr.attr, |
ab9a0f19 | 5117 | &reserved_attr.attr, |
49e22585 | 5118 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5119 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5120 | &total_objects_attr.attr, |
5121 | &slabs_attr.attr, | |
5122 | &sanity_checks_attr.attr, | |
5123 | &trace_attr.attr, | |
81819f0f CL |
5124 | &red_zone_attr.attr, |
5125 | &poison_attr.attr, | |
5126 | &store_user_attr.attr, | |
53e15af0 | 5127 | &validate_attr.attr, |
88a420e4 CL |
5128 | &alloc_calls_attr.attr, |
5129 | &free_calls_attr.attr, | |
ab4d5ed5 | 5130 | #endif |
81819f0f CL |
5131 | #ifdef CONFIG_ZONE_DMA |
5132 | &cache_dma_attr.attr, | |
5133 | #endif | |
5134 | #ifdef CONFIG_NUMA | |
9824601e | 5135 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5136 | #endif |
5137 | #ifdef CONFIG_SLUB_STATS | |
5138 | &alloc_fastpath_attr.attr, | |
5139 | &alloc_slowpath_attr.attr, | |
5140 | &free_fastpath_attr.attr, | |
5141 | &free_slowpath_attr.attr, | |
5142 | &free_frozen_attr.attr, | |
5143 | &free_add_partial_attr.attr, | |
5144 | &free_remove_partial_attr.attr, | |
5145 | &alloc_from_partial_attr.attr, | |
5146 | &alloc_slab_attr.attr, | |
5147 | &alloc_refill_attr.attr, | |
e36a2652 | 5148 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5149 | &free_slab_attr.attr, |
5150 | &cpuslab_flush_attr.attr, | |
5151 | &deactivate_full_attr.attr, | |
5152 | &deactivate_empty_attr.attr, | |
5153 | &deactivate_to_head_attr.attr, | |
5154 | &deactivate_to_tail_attr.attr, | |
5155 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5156 | &deactivate_bypass_attr.attr, |
65c3376a | 5157 | &order_fallback_attr.attr, |
b789ef51 CL |
5158 | &cmpxchg_double_fail_attr.attr, |
5159 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5160 | &cpu_partial_alloc_attr.attr, |
5161 | &cpu_partial_free_attr.attr, | |
81819f0f | 5162 | #endif |
4c13dd3b DM |
5163 | #ifdef CONFIG_FAILSLAB |
5164 | &failslab_attr.attr, | |
5165 | #endif | |
5166 | ||
81819f0f CL |
5167 | NULL |
5168 | }; | |
5169 | ||
5170 | static struct attribute_group slab_attr_group = { | |
5171 | .attrs = slab_attrs, | |
5172 | }; | |
5173 | ||
5174 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5175 | struct attribute *attr, | |
5176 | char *buf) | |
5177 | { | |
5178 | struct slab_attribute *attribute; | |
5179 | struct kmem_cache *s; | |
5180 | int err; | |
5181 | ||
5182 | attribute = to_slab_attr(attr); | |
5183 | s = to_slab(kobj); | |
5184 | ||
5185 | if (!attribute->show) | |
5186 | return -EIO; | |
5187 | ||
5188 | err = attribute->show(s, buf); | |
5189 | ||
5190 | return err; | |
5191 | } | |
5192 | ||
5193 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5194 | struct attribute *attr, | |
5195 | const char *buf, size_t len) | |
5196 | { | |
5197 | struct slab_attribute *attribute; | |
5198 | struct kmem_cache *s; | |
5199 | int err; | |
5200 | ||
5201 | attribute = to_slab_attr(attr); | |
5202 | s = to_slab(kobj); | |
5203 | ||
5204 | if (!attribute->store) | |
5205 | return -EIO; | |
5206 | ||
5207 | err = attribute->store(s, buf, len); | |
5208 | ||
5209 | return err; | |
5210 | } | |
5211 | ||
151c602f CL |
5212 | static void kmem_cache_release(struct kobject *kobj) |
5213 | { | |
5214 | struct kmem_cache *s = to_slab(kobj); | |
5215 | ||
84c1cf62 | 5216 | kfree(s->name); |
151c602f CL |
5217 | kfree(s); |
5218 | } | |
5219 | ||
52cf25d0 | 5220 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5221 | .show = slab_attr_show, |
5222 | .store = slab_attr_store, | |
5223 | }; | |
5224 | ||
5225 | static struct kobj_type slab_ktype = { | |
5226 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 5227 | .release = kmem_cache_release |
81819f0f CL |
5228 | }; |
5229 | ||
5230 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5231 | { | |
5232 | struct kobj_type *ktype = get_ktype(kobj); | |
5233 | ||
5234 | if (ktype == &slab_ktype) | |
5235 | return 1; | |
5236 | return 0; | |
5237 | } | |
5238 | ||
9cd43611 | 5239 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5240 | .filter = uevent_filter, |
5241 | }; | |
5242 | ||
27c3a314 | 5243 | static struct kset *slab_kset; |
81819f0f CL |
5244 | |
5245 | #define ID_STR_LENGTH 64 | |
5246 | ||
5247 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5248 | * |
5249 | * Format :[flags-]size | |
81819f0f CL |
5250 | */ |
5251 | static char *create_unique_id(struct kmem_cache *s) | |
5252 | { | |
5253 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5254 | char *p = name; | |
5255 | ||
5256 | BUG_ON(!name); | |
5257 | ||
5258 | *p++ = ':'; | |
5259 | /* | |
5260 | * First flags affecting slabcache operations. We will only | |
5261 | * get here for aliasable slabs so we do not need to support | |
5262 | * too many flags. The flags here must cover all flags that | |
5263 | * are matched during merging to guarantee that the id is | |
5264 | * unique. | |
5265 | */ | |
5266 | if (s->flags & SLAB_CACHE_DMA) | |
5267 | *p++ = 'd'; | |
5268 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5269 | *p++ = 'a'; | |
5270 | if (s->flags & SLAB_DEBUG_FREE) | |
5271 | *p++ = 'F'; | |
5a896d9e VN |
5272 | if (!(s->flags & SLAB_NOTRACK)) |
5273 | *p++ = 't'; | |
81819f0f CL |
5274 | if (p != name + 1) |
5275 | *p++ = '-'; | |
5276 | p += sprintf(p, "%07d", s->size); | |
5277 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5278 | return name; | |
5279 | } | |
5280 | ||
5281 | static int sysfs_slab_add(struct kmem_cache *s) | |
5282 | { | |
5283 | int err; | |
5284 | const char *name; | |
5285 | int unmergeable; | |
5286 | ||
5287 | if (slab_state < SYSFS) | |
5288 | /* Defer until later */ | |
5289 | return 0; | |
5290 | ||
5291 | unmergeable = slab_unmergeable(s); | |
5292 | if (unmergeable) { | |
5293 | /* | |
5294 | * Slabcache can never be merged so we can use the name proper. | |
5295 | * This is typically the case for debug situations. In that | |
5296 | * case we can catch duplicate names easily. | |
5297 | */ | |
27c3a314 | 5298 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5299 | name = s->name; |
5300 | } else { | |
5301 | /* | |
5302 | * Create a unique name for the slab as a target | |
5303 | * for the symlinks. | |
5304 | */ | |
5305 | name = create_unique_id(s); | |
5306 | } | |
5307 | ||
27c3a314 | 5308 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5309 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5310 | if (err) { | |
5311 | kobject_put(&s->kobj); | |
81819f0f | 5312 | return err; |
1eada11c | 5313 | } |
81819f0f CL |
5314 | |
5315 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5316 | if (err) { |
5317 | kobject_del(&s->kobj); | |
5318 | kobject_put(&s->kobj); | |
81819f0f | 5319 | return err; |
5788d8ad | 5320 | } |
81819f0f CL |
5321 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5322 | if (!unmergeable) { | |
5323 | /* Setup first alias */ | |
5324 | sysfs_slab_alias(s, s->name); | |
5325 | kfree(name); | |
5326 | } | |
5327 | return 0; | |
5328 | } | |
5329 | ||
5330 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5331 | { | |
2bce6485 CL |
5332 | if (slab_state < SYSFS) |
5333 | /* | |
5334 | * Sysfs has not been setup yet so no need to remove the | |
5335 | * cache from sysfs. | |
5336 | */ | |
5337 | return; | |
5338 | ||
81819f0f CL |
5339 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5340 | kobject_del(&s->kobj); | |
151c602f | 5341 | kobject_put(&s->kobj); |
81819f0f CL |
5342 | } |
5343 | ||
5344 | /* | |
5345 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5346 | * available lest we lose that information. |
81819f0f CL |
5347 | */ |
5348 | struct saved_alias { | |
5349 | struct kmem_cache *s; | |
5350 | const char *name; | |
5351 | struct saved_alias *next; | |
5352 | }; | |
5353 | ||
5af328a5 | 5354 | static struct saved_alias *alias_list; |
81819f0f CL |
5355 | |
5356 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5357 | { | |
5358 | struct saved_alias *al; | |
5359 | ||
5360 | if (slab_state == SYSFS) { | |
5361 | /* | |
5362 | * If we have a leftover link then remove it. | |
5363 | */ | |
27c3a314 GKH |
5364 | sysfs_remove_link(&slab_kset->kobj, name); |
5365 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5366 | } |
5367 | ||
5368 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5369 | if (!al) | |
5370 | return -ENOMEM; | |
5371 | ||
5372 | al->s = s; | |
5373 | al->name = name; | |
5374 | al->next = alias_list; | |
5375 | alias_list = al; | |
5376 | return 0; | |
5377 | } | |
5378 | ||
5379 | static int __init slab_sysfs_init(void) | |
5380 | { | |
5b95a4ac | 5381 | struct kmem_cache *s; |
81819f0f CL |
5382 | int err; |
5383 | ||
2bce6485 CL |
5384 | down_write(&slub_lock); |
5385 | ||
0ff21e46 | 5386 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5387 | if (!slab_kset) { |
2bce6485 | 5388 | up_write(&slub_lock); |
81819f0f CL |
5389 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5390 | return -ENOSYS; | |
5391 | } | |
5392 | ||
26a7bd03 CL |
5393 | slab_state = SYSFS; |
5394 | ||
5b95a4ac | 5395 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5396 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5397 | if (err) |
5398 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5399 | " to sysfs\n", s->name); | |
26a7bd03 | 5400 | } |
81819f0f CL |
5401 | |
5402 | while (alias_list) { | |
5403 | struct saved_alias *al = alias_list; | |
5404 | ||
5405 | alias_list = alias_list->next; | |
5406 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5407 | if (err) |
5408 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
5409 | " %s to sysfs\n", s->name); | |
81819f0f CL |
5410 | kfree(al); |
5411 | } | |
5412 | ||
2bce6485 | 5413 | up_write(&slub_lock); |
81819f0f CL |
5414 | resiliency_test(); |
5415 | return 0; | |
5416 | } | |
5417 | ||
5418 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5419 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5420 | |
5421 | /* | |
5422 | * The /proc/slabinfo ABI | |
5423 | */ | |
158a9624 | 5424 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5425 | static void print_slabinfo_header(struct seq_file *m) |
5426 | { | |
5427 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
5428 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
5429 | "<objperslab> <pagesperslab>"); | |
5430 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5431 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5432 | seq_putc(m, '\n'); | |
5433 | } | |
5434 | ||
5435 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5436 | { | |
5437 | loff_t n = *pos; | |
5438 | ||
5439 | down_read(&slub_lock); | |
5440 | if (!n) | |
5441 | print_slabinfo_header(m); | |
5442 | ||
5443 | return seq_list_start(&slab_caches, *pos); | |
5444 | } | |
5445 | ||
5446 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5447 | { | |
5448 | return seq_list_next(p, &slab_caches, pos); | |
5449 | } | |
5450 | ||
5451 | static void s_stop(struct seq_file *m, void *p) | |
5452 | { | |
5453 | up_read(&slub_lock); | |
5454 | } | |
5455 | ||
5456 | static int s_show(struct seq_file *m, void *p) | |
5457 | { | |
5458 | unsigned long nr_partials = 0; | |
5459 | unsigned long nr_slabs = 0; | |
5460 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5461 | unsigned long nr_objs = 0; |
5462 | unsigned long nr_free = 0; | |
57ed3eda PE |
5463 | struct kmem_cache *s; |
5464 | int node; | |
5465 | ||
5466 | s = list_entry(p, struct kmem_cache, list); | |
5467 | ||
5468 | for_each_online_node(node) { | |
5469 | struct kmem_cache_node *n = get_node(s, node); | |
5470 | ||
5471 | if (!n) | |
5472 | continue; | |
5473 | ||
5474 | nr_partials += n->nr_partial; | |
5475 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5476 | nr_objs += atomic_long_read(&n->total_objects); |
5477 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5478 | } |
5479 | ||
205ab99d | 5480 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5481 | |
5482 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5483 | nr_objs, s->size, oo_objects(s->oo), |
5484 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5485 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5486 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5487 | 0UL); | |
5488 | seq_putc(m, '\n'); | |
5489 | return 0; | |
5490 | } | |
5491 | ||
7b3c3a50 | 5492 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5493 | .start = s_start, |
5494 | .next = s_next, | |
5495 | .stop = s_stop, | |
5496 | .show = s_show, | |
5497 | }; | |
5498 | ||
7b3c3a50 AD |
5499 | static int slabinfo_open(struct inode *inode, struct file *file) |
5500 | { | |
5501 | return seq_open(file, &slabinfo_op); | |
5502 | } | |
5503 | ||
5504 | static const struct file_operations proc_slabinfo_operations = { | |
5505 | .open = slabinfo_open, | |
5506 | .read = seq_read, | |
5507 | .llseek = seq_lseek, | |
5508 | .release = seq_release, | |
5509 | }; | |
5510 | ||
5511 | static int __init slab_proc_init(void) | |
5512 | { | |
cf5d1131 | 5513 | proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5514 | return 0; |
5515 | } | |
5516 | module_init(slab_proc_init); | |
158a9624 | 5517 | #endif /* CONFIG_SLABINFO */ |