]>
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> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
81819f0f | 21 | #include <linux/seq_file.h> |
5a896d9e | 22 | #include <linux/kmemcheck.h> |
81819f0f CL |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpuset.h> | |
25 | #include <linux/mempolicy.h> | |
26 | #include <linux/ctype.h> | |
3ac7fe5a | 27 | #include <linux/debugobjects.h> |
81819f0f | 28 | #include <linux/kallsyms.h> |
b9049e23 | 29 | #include <linux/memory.h> |
f8bd2258 | 30 | #include <linux/math64.h> |
773ff60e | 31 | #include <linux/fault-inject.h> |
bfa71457 | 32 | #include <linux/stacktrace.h> |
4de900b4 | 33 | #include <linux/prefetch.h> |
81819f0f | 34 | |
4a92379b RK |
35 | #include <trace/events/kmem.h> |
36 | ||
072bb0aa MG |
37 | #include "internal.h" |
38 | ||
81819f0f CL |
39 | /* |
40 | * Lock order: | |
18004c5d | 41 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
42 | * 2. node->list_lock |
43 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 44 | * |
18004c5d | 45 | * slab_mutex |
881db7fb | 46 | * |
18004c5d | 47 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
48 | * and to synchronize major metadata changes to slab cache structures. |
49 | * | |
50 | * The slab_lock is only used for debugging and on arches that do not | |
51 | * have the ability to do a cmpxchg_double. It only protects the second | |
52 | * double word in the page struct. Meaning | |
53 | * A. page->freelist -> List of object free in a page | |
54 | * B. page->counters -> Counters of objects | |
55 | * C. page->frozen -> frozen state | |
56 | * | |
57 | * If a slab is frozen then it is exempt from list management. It is not | |
58 | * on any list. The processor that froze the slab is the one who can | |
59 | * perform list operations on the page. Other processors may put objects | |
60 | * onto the freelist but the processor that froze the slab is the only | |
61 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
62 | * |
63 | * The list_lock protects the partial and full list on each node and | |
64 | * the partial slab counter. If taken then no new slabs may be added or | |
65 | * removed from the lists nor make the number of partial slabs be modified. | |
66 | * (Note that the total number of slabs is an atomic value that may be | |
67 | * modified without taking the list lock). | |
68 | * | |
69 | * The list_lock is a centralized lock and thus we avoid taking it as | |
70 | * much as possible. As long as SLUB does not have to handle partial | |
71 | * slabs, operations can continue without any centralized lock. F.e. | |
72 | * allocating a long series of objects that fill up slabs does not require | |
73 | * the list lock. | |
81819f0f CL |
74 | * Interrupts are disabled during allocation and deallocation in order to |
75 | * make the slab allocator safe to use in the context of an irq. In addition | |
76 | * interrupts are disabled to ensure that the processor does not change | |
77 | * while handling per_cpu slabs, due to kernel preemption. | |
78 | * | |
79 | * SLUB assigns one slab for allocation to each processor. | |
80 | * Allocations only occur from these slabs called cpu slabs. | |
81 | * | |
672bba3a CL |
82 | * Slabs with free elements are kept on a partial list and during regular |
83 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 84 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
85 | * We track full slabs for debugging purposes though because otherwise we |
86 | * cannot scan all objects. | |
81819f0f CL |
87 | * |
88 | * Slabs are freed when they become empty. Teardown and setup is | |
89 | * minimal so we rely on the page allocators per cpu caches for | |
90 | * fast frees and allocs. | |
91 | * | |
92 | * Overloading of page flags that are otherwise used for LRU management. | |
93 | * | |
4b6f0750 CL |
94 | * PageActive The slab is frozen and exempt from list processing. |
95 | * This means that the slab is dedicated to a purpose | |
96 | * such as satisfying allocations for a specific | |
97 | * processor. Objects may be freed in the slab while | |
98 | * it is frozen but slab_free will then skip the usual | |
99 | * list operations. It is up to the processor holding | |
100 | * the slab to integrate the slab into the slab lists | |
101 | * when the slab is no longer needed. | |
102 | * | |
103 | * One use of this flag is to mark slabs that are | |
104 | * used for allocations. Then such a slab becomes a cpu | |
105 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 106 | * freelist that allows lockless access to |
894b8788 CL |
107 | * free objects in addition to the regular freelist |
108 | * that requires the slab lock. | |
81819f0f CL |
109 | * |
110 | * PageError Slab requires special handling due to debug | |
111 | * options set. This moves slab handling out of | |
894b8788 | 112 | * the fast path and disables lockless freelists. |
81819f0f CL |
113 | */ |
114 | ||
af537b0a CL |
115 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
116 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
117 | ||
118 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
119 | { | |
5577bd8a | 120 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 121 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 122 | #else |
af537b0a | 123 | return 0; |
5577bd8a | 124 | #endif |
af537b0a | 125 | } |
5577bd8a | 126 | |
81819f0f CL |
127 | /* |
128 | * Issues still to be resolved: | |
129 | * | |
81819f0f CL |
130 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
131 | * | |
81819f0f CL |
132 | * - Variable sizing of the per node arrays |
133 | */ | |
134 | ||
135 | /* Enable to test recovery from slab corruption on boot */ | |
136 | #undef SLUB_RESILIENCY_TEST | |
137 | ||
b789ef51 CL |
138 | /* Enable to log cmpxchg failures */ |
139 | #undef SLUB_DEBUG_CMPXCHG | |
140 | ||
2086d26a CL |
141 | /* |
142 | * Mininum number of partial slabs. These will be left on the partial | |
143 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
144 | */ | |
76be8950 | 145 | #define MIN_PARTIAL 5 |
e95eed57 | 146 | |
2086d26a CL |
147 | /* |
148 | * Maximum number of desirable partial slabs. | |
149 | * The existence of more partial slabs makes kmem_cache_shrink | |
150 | * sort the partial list by the number of objects in the. | |
151 | */ | |
152 | #define MAX_PARTIAL 10 | |
153 | ||
81819f0f CL |
154 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
155 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 156 | |
fa5ec8a1 | 157 | /* |
3de47213 DR |
158 | * Debugging flags that require metadata to be stored in the slab. These get |
159 | * disabled when slub_debug=O is used and a cache's min order increases with | |
160 | * metadata. | |
fa5ec8a1 | 161 | */ |
3de47213 | 162 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 163 | |
81819f0f CL |
164 | /* |
165 | * Set of flags that will prevent slab merging | |
166 | */ | |
167 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
168 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
169 | SLAB_FAILSLAB) | |
81819f0f CL |
170 | |
171 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 172 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 173 | |
210b5c06 CG |
174 | #define OO_SHIFT 16 |
175 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 176 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 177 | |
81819f0f | 178 | /* Internal SLUB flags */ |
f90ec390 | 179 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 180 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f CL |
181 | |
182 | static int kmem_size = sizeof(struct kmem_cache); | |
183 | ||
184 | #ifdef CONFIG_SMP | |
185 | static struct notifier_block slab_notifier; | |
186 | #endif | |
187 | ||
02cbc874 CL |
188 | /* |
189 | * Tracking user of a slab. | |
190 | */ | |
d6543e39 | 191 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 192 | struct track { |
ce71e27c | 193 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
194 | #ifdef CONFIG_STACKTRACE |
195 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
196 | #endif | |
02cbc874 CL |
197 | int cpu; /* Was running on cpu */ |
198 | int pid; /* Pid context */ | |
199 | unsigned long when; /* When did the operation occur */ | |
200 | }; | |
201 | ||
202 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
203 | ||
ab4d5ed5 | 204 | #ifdef CONFIG_SYSFS |
81819f0f CL |
205 | static int sysfs_slab_add(struct kmem_cache *); |
206 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
207 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 208 | |
81819f0f | 209 | #else |
0c710013 CL |
210 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
211 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
212 | { return 0; } | |
151c602f CL |
213 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
214 | { | |
84c1cf62 | 215 | kfree(s->name); |
151c602f CL |
216 | kfree(s); |
217 | } | |
8ff12cfc | 218 | |
81819f0f CL |
219 | #endif |
220 | ||
4fdccdfb | 221 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
222 | { |
223 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 224 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
225 | #endif |
226 | } | |
227 | ||
81819f0f CL |
228 | /******************************************************************** |
229 | * Core slab cache functions | |
230 | *******************************************************************/ | |
231 | ||
81819f0f CL |
232 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
233 | { | |
81819f0f | 234 | return s->node[node]; |
81819f0f CL |
235 | } |
236 | ||
6446faa2 | 237 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
238 | static inline int check_valid_pointer(struct kmem_cache *s, |
239 | struct page *page, const void *object) | |
240 | { | |
241 | void *base; | |
242 | ||
a973e9dd | 243 | if (!object) |
02cbc874 CL |
244 | return 1; |
245 | ||
a973e9dd | 246 | base = page_address(page); |
39b26464 | 247 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
248 | (object - base) % s->size) { |
249 | return 0; | |
250 | } | |
251 | ||
252 | return 1; | |
253 | } | |
254 | ||
7656c72b CL |
255 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
256 | { | |
257 | return *(void **)(object + s->offset); | |
258 | } | |
259 | ||
0ad9500e ED |
260 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
261 | { | |
262 | prefetch(object + s->offset); | |
263 | } | |
264 | ||
1393d9a1 CL |
265 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
266 | { | |
267 | void *p; | |
268 | ||
269 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
270 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
271 | #else | |
272 | p = get_freepointer(s, object); | |
273 | #endif | |
274 | return p; | |
275 | } | |
276 | ||
7656c72b CL |
277 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
278 | { | |
279 | *(void **)(object + s->offset) = fp; | |
280 | } | |
281 | ||
282 | /* Loop over all objects in a slab */ | |
224a88be CL |
283 | #define for_each_object(__p, __s, __addr, __objects) \ |
284 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
285 | __p += (__s)->size) |
286 | ||
7656c72b CL |
287 | /* Determine object index from a given position */ |
288 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
289 | { | |
290 | return (p - addr) / s->size; | |
291 | } | |
292 | ||
d71f606f MK |
293 | static inline size_t slab_ksize(const struct kmem_cache *s) |
294 | { | |
295 | #ifdef CONFIG_SLUB_DEBUG | |
296 | /* | |
297 | * Debugging requires use of the padding between object | |
298 | * and whatever may come after it. | |
299 | */ | |
300 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 301 | return s->object_size; |
d71f606f MK |
302 | |
303 | #endif | |
304 | /* | |
305 | * If we have the need to store the freelist pointer | |
306 | * back there or track user information then we can | |
307 | * only use the space before that information. | |
308 | */ | |
309 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
310 | return s->inuse; | |
311 | /* | |
312 | * Else we can use all the padding etc for the allocation | |
313 | */ | |
314 | return s->size; | |
315 | } | |
316 | ||
ab9a0f19 LJ |
317 | static inline int order_objects(int order, unsigned long size, int reserved) |
318 | { | |
319 | return ((PAGE_SIZE << order) - reserved) / size; | |
320 | } | |
321 | ||
834f3d11 | 322 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 323 | unsigned long size, int reserved) |
834f3d11 CL |
324 | { |
325 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 326 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
327 | }; |
328 | ||
329 | return x; | |
330 | } | |
331 | ||
332 | static inline int oo_order(struct kmem_cache_order_objects x) | |
333 | { | |
210b5c06 | 334 | return x.x >> OO_SHIFT; |
834f3d11 CL |
335 | } |
336 | ||
337 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
338 | { | |
210b5c06 | 339 | return x.x & OO_MASK; |
834f3d11 CL |
340 | } |
341 | ||
881db7fb CL |
342 | /* |
343 | * Per slab locking using the pagelock | |
344 | */ | |
345 | static __always_inline void slab_lock(struct page *page) | |
346 | { | |
347 | bit_spin_lock(PG_locked, &page->flags); | |
348 | } | |
349 | ||
350 | static __always_inline void slab_unlock(struct page *page) | |
351 | { | |
352 | __bit_spin_unlock(PG_locked, &page->flags); | |
353 | } | |
354 | ||
1d07171c CL |
355 | /* Interrupts must be disabled (for the fallback code to work right) */ |
356 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
357 | void *freelist_old, unsigned long counters_old, | |
358 | void *freelist_new, unsigned long counters_new, | |
359 | const char *n) | |
360 | { | |
361 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
362 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
363 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 364 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 365 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
366 | freelist_old, counters_old, |
367 | freelist_new, counters_new)) | |
368 | return 1; | |
369 | } else | |
370 | #endif | |
371 | { | |
372 | slab_lock(page); | |
373 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
374 | page->freelist = freelist_new; | |
375 | page->counters = counters_new; | |
376 | slab_unlock(page); | |
377 | return 1; | |
378 | } | |
379 | slab_unlock(page); | |
380 | } | |
381 | ||
382 | cpu_relax(); | |
383 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
384 | ||
385 | #ifdef SLUB_DEBUG_CMPXCHG | |
386 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
387 | #endif | |
388 | ||
389 | return 0; | |
390 | } | |
391 | ||
b789ef51 CL |
392 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
393 | void *freelist_old, unsigned long counters_old, | |
394 | void *freelist_new, unsigned long counters_new, | |
395 | const char *n) | |
396 | { | |
2565409f HC |
397 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
398 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 399 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 400 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
401 | freelist_old, counters_old, |
402 | freelist_new, counters_new)) | |
403 | return 1; | |
404 | } else | |
405 | #endif | |
406 | { | |
1d07171c CL |
407 | unsigned long flags; |
408 | ||
409 | local_irq_save(flags); | |
881db7fb | 410 | slab_lock(page); |
b789ef51 CL |
411 | if (page->freelist == freelist_old && page->counters == counters_old) { |
412 | page->freelist = freelist_new; | |
413 | page->counters = counters_new; | |
881db7fb | 414 | slab_unlock(page); |
1d07171c | 415 | local_irq_restore(flags); |
b789ef51 CL |
416 | return 1; |
417 | } | |
881db7fb | 418 | slab_unlock(page); |
1d07171c | 419 | local_irq_restore(flags); |
b789ef51 CL |
420 | } |
421 | ||
422 | cpu_relax(); | |
423 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
424 | ||
425 | #ifdef SLUB_DEBUG_CMPXCHG | |
426 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
427 | #endif | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
41ecc55b | 432 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
433 | /* |
434 | * Determine a map of object in use on a page. | |
435 | * | |
881db7fb | 436 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
437 | * not vanish from under us. |
438 | */ | |
439 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
440 | { | |
441 | void *p; | |
442 | void *addr = page_address(page); | |
443 | ||
444 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
445 | set_bit(slab_index(p, s, addr), map); | |
446 | } | |
447 | ||
41ecc55b CL |
448 | /* |
449 | * Debug settings: | |
450 | */ | |
f0630fff CL |
451 | #ifdef CONFIG_SLUB_DEBUG_ON |
452 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
453 | #else | |
41ecc55b | 454 | static int slub_debug; |
f0630fff | 455 | #endif |
41ecc55b CL |
456 | |
457 | static char *slub_debug_slabs; | |
fa5ec8a1 | 458 | static int disable_higher_order_debug; |
41ecc55b | 459 | |
81819f0f CL |
460 | /* |
461 | * Object debugging | |
462 | */ | |
463 | static void print_section(char *text, u8 *addr, unsigned int length) | |
464 | { | |
ffc79d28 SAS |
465 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
466 | length, 1); | |
81819f0f CL |
467 | } |
468 | ||
81819f0f CL |
469 | static struct track *get_track(struct kmem_cache *s, void *object, |
470 | enum track_item alloc) | |
471 | { | |
472 | struct track *p; | |
473 | ||
474 | if (s->offset) | |
475 | p = object + s->offset + sizeof(void *); | |
476 | else | |
477 | p = object + s->inuse; | |
478 | ||
479 | return p + alloc; | |
480 | } | |
481 | ||
482 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 483 | enum track_item alloc, unsigned long addr) |
81819f0f | 484 | { |
1a00df4a | 485 | struct track *p = get_track(s, object, alloc); |
81819f0f | 486 | |
81819f0f | 487 | if (addr) { |
d6543e39 BG |
488 | #ifdef CONFIG_STACKTRACE |
489 | struct stack_trace trace; | |
490 | int i; | |
491 | ||
492 | trace.nr_entries = 0; | |
493 | trace.max_entries = TRACK_ADDRS_COUNT; | |
494 | trace.entries = p->addrs; | |
495 | trace.skip = 3; | |
496 | save_stack_trace(&trace); | |
497 | ||
498 | /* See rant in lockdep.c */ | |
499 | if (trace.nr_entries != 0 && | |
500 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
501 | trace.nr_entries--; | |
502 | ||
503 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
504 | p->addrs[i] = 0; | |
505 | #endif | |
81819f0f CL |
506 | p->addr = addr; |
507 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 508 | p->pid = current->pid; |
81819f0f CL |
509 | p->when = jiffies; |
510 | } else | |
511 | memset(p, 0, sizeof(struct track)); | |
512 | } | |
513 | ||
81819f0f CL |
514 | static void init_tracking(struct kmem_cache *s, void *object) |
515 | { | |
24922684 CL |
516 | if (!(s->flags & SLAB_STORE_USER)) |
517 | return; | |
518 | ||
ce71e27c EGM |
519 | set_track(s, object, TRACK_FREE, 0UL); |
520 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
521 | } |
522 | ||
523 | static void print_track(const char *s, struct track *t) | |
524 | { | |
525 | if (!t->addr) | |
526 | return; | |
527 | ||
7daf705f | 528 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 529 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
530 | #ifdef CONFIG_STACKTRACE |
531 | { | |
532 | int i; | |
533 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
534 | if (t->addrs[i]) | |
535 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
536 | else | |
537 | break; | |
538 | } | |
539 | #endif | |
24922684 CL |
540 | } |
541 | ||
542 | static void print_tracking(struct kmem_cache *s, void *object) | |
543 | { | |
544 | if (!(s->flags & SLAB_STORE_USER)) | |
545 | return; | |
546 | ||
547 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
548 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
549 | } | |
550 | ||
551 | static void print_page_info(struct page *page) | |
552 | { | |
39b26464 CL |
553 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
554 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
555 | |
556 | } | |
557 | ||
558 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
559 | { | |
560 | va_list args; | |
561 | char buf[100]; | |
562 | ||
563 | va_start(args, fmt); | |
564 | vsnprintf(buf, sizeof(buf), fmt, args); | |
565 | va_end(args); | |
566 | printk(KERN_ERR "========================================" | |
567 | "=====================================\n"); | |
265d47e7 | 568 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
569 | printk(KERN_ERR "----------------------------------------" |
570 | "-------------------------------------\n\n"); | |
81819f0f CL |
571 | } |
572 | ||
24922684 CL |
573 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
574 | { | |
575 | va_list args; | |
576 | char buf[100]; | |
577 | ||
578 | va_start(args, fmt); | |
579 | vsnprintf(buf, sizeof(buf), fmt, args); | |
580 | va_end(args); | |
581 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
582 | } | |
583 | ||
584 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
585 | { |
586 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 587 | u8 *addr = page_address(page); |
24922684 CL |
588 | |
589 | print_tracking(s, p); | |
590 | ||
591 | print_page_info(page); | |
592 | ||
593 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
594 | p, p - addr, get_freepointer(s, p)); | |
595 | ||
596 | if (p > addr + 16) | |
ffc79d28 | 597 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 598 | |
3b0efdfa | 599 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 600 | PAGE_SIZE)); |
81819f0f | 601 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
602 | print_section("Redzone ", p + s->object_size, |
603 | s->inuse - s->object_size); | |
81819f0f | 604 | |
81819f0f CL |
605 | if (s->offset) |
606 | off = s->offset + sizeof(void *); | |
607 | else | |
608 | off = s->inuse; | |
609 | ||
24922684 | 610 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 611 | off += 2 * sizeof(struct track); |
81819f0f CL |
612 | |
613 | if (off != s->size) | |
614 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 615 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
616 | |
617 | dump_stack(); | |
81819f0f CL |
618 | } |
619 | ||
620 | static void object_err(struct kmem_cache *s, struct page *page, | |
621 | u8 *object, char *reason) | |
622 | { | |
3dc50637 | 623 | slab_bug(s, "%s", reason); |
24922684 | 624 | print_trailer(s, page, object); |
81819f0f CL |
625 | } |
626 | ||
24922684 | 627 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
628 | { |
629 | va_list args; | |
630 | char buf[100]; | |
631 | ||
24922684 CL |
632 | va_start(args, fmt); |
633 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 634 | va_end(args); |
3dc50637 | 635 | slab_bug(s, "%s", buf); |
24922684 | 636 | print_page_info(page); |
81819f0f CL |
637 | dump_stack(); |
638 | } | |
639 | ||
f7cb1933 | 640 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
641 | { |
642 | u8 *p = object; | |
643 | ||
644 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
645 | memset(p, POISON_FREE, s->object_size - 1); |
646 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
647 | } |
648 | ||
649 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 650 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
651 | } |
652 | ||
24922684 CL |
653 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
654 | void *from, void *to) | |
655 | { | |
656 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
657 | memset(from, data, to - from); | |
658 | } | |
659 | ||
660 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
661 | u8 *object, char *what, | |
06428780 | 662 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
663 | { |
664 | u8 *fault; | |
665 | u8 *end; | |
666 | ||
79824820 | 667 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
668 | if (!fault) |
669 | return 1; | |
670 | ||
671 | end = start + bytes; | |
672 | while (end > fault && end[-1] == value) | |
673 | end--; | |
674 | ||
675 | slab_bug(s, "%s overwritten", what); | |
676 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
677 | fault, end - 1, fault[0], value); | |
678 | print_trailer(s, page, object); | |
679 | ||
680 | restore_bytes(s, what, value, fault, end); | |
681 | return 0; | |
81819f0f CL |
682 | } |
683 | ||
81819f0f CL |
684 | /* |
685 | * Object layout: | |
686 | * | |
687 | * object address | |
688 | * Bytes of the object to be managed. | |
689 | * If the freepointer may overlay the object then the free | |
690 | * pointer is the first word of the object. | |
672bba3a | 691 | * |
81819f0f CL |
692 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
693 | * 0xa5 (POISON_END) | |
694 | * | |
3b0efdfa | 695 | * object + s->object_size |
81819f0f | 696 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 697 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 698 | * object_size == inuse. |
672bba3a | 699 | * |
81819f0f CL |
700 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
701 | * 0xcc (RED_ACTIVE) for objects in use. | |
702 | * | |
703 | * object + s->inuse | |
672bba3a CL |
704 | * Meta data starts here. |
705 | * | |
81819f0f CL |
706 | * A. Free pointer (if we cannot overwrite object on free) |
707 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 708 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 709 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
710 | * before the word boundary. |
711 | * | |
712 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
713 | * |
714 | * object + s->size | |
672bba3a | 715 | * Nothing is used beyond s->size. |
81819f0f | 716 | * |
3b0efdfa | 717 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 718 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
719 | * may be used with merged slabcaches. |
720 | */ | |
721 | ||
81819f0f CL |
722 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
723 | { | |
724 | unsigned long off = s->inuse; /* The end of info */ | |
725 | ||
726 | if (s->offset) | |
727 | /* Freepointer is placed after the object. */ | |
728 | off += sizeof(void *); | |
729 | ||
730 | if (s->flags & SLAB_STORE_USER) | |
731 | /* We also have user information there */ | |
732 | off += 2 * sizeof(struct track); | |
733 | ||
734 | if (s->size == off) | |
735 | return 1; | |
736 | ||
24922684 CL |
737 | return check_bytes_and_report(s, page, p, "Object padding", |
738 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
739 | } |
740 | ||
39b26464 | 741 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
742 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
743 | { | |
24922684 CL |
744 | u8 *start; |
745 | u8 *fault; | |
746 | u8 *end; | |
747 | int length; | |
748 | int remainder; | |
81819f0f CL |
749 | |
750 | if (!(s->flags & SLAB_POISON)) | |
751 | return 1; | |
752 | ||
a973e9dd | 753 | start = page_address(page); |
ab9a0f19 | 754 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
755 | end = start + length; |
756 | remainder = length % s->size; | |
81819f0f CL |
757 | if (!remainder) |
758 | return 1; | |
759 | ||
79824820 | 760 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
761 | if (!fault) |
762 | return 1; | |
763 | while (end > fault && end[-1] == POISON_INUSE) | |
764 | end--; | |
765 | ||
766 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 767 | print_section("Padding ", end - remainder, remainder); |
24922684 | 768 | |
8a3d271d | 769 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 770 | return 0; |
81819f0f CL |
771 | } |
772 | ||
773 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 774 | void *object, u8 val) |
81819f0f CL |
775 | { |
776 | u8 *p = object; | |
3b0efdfa | 777 | u8 *endobject = object + s->object_size; |
81819f0f CL |
778 | |
779 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 780 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 781 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 782 | return 0; |
81819f0f | 783 | } else { |
3b0efdfa | 784 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 785 | check_bytes_and_report(s, page, p, "Alignment padding", |
3b0efdfa | 786 | endobject, POISON_INUSE, s->inuse - s->object_size); |
3adbefee | 787 | } |
81819f0f CL |
788 | } |
789 | ||
790 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 791 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 792 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 793 | POISON_FREE, s->object_size - 1) || |
24922684 | 794 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 795 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 796 | return 0; |
81819f0f CL |
797 | /* |
798 | * check_pad_bytes cleans up on its own. | |
799 | */ | |
800 | check_pad_bytes(s, page, p); | |
801 | } | |
802 | ||
f7cb1933 | 803 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
804 | /* |
805 | * Object and freepointer overlap. Cannot check | |
806 | * freepointer while object is allocated. | |
807 | */ | |
808 | return 1; | |
809 | ||
810 | /* Check free pointer validity */ | |
811 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
812 | object_err(s, page, p, "Freepointer corrupt"); | |
813 | /* | |
9f6c708e | 814 | * No choice but to zap it and thus lose the remainder |
81819f0f | 815 | * of the free objects in this slab. May cause |
672bba3a | 816 | * another error because the object count is now wrong. |
81819f0f | 817 | */ |
a973e9dd | 818 | set_freepointer(s, p, NULL); |
81819f0f CL |
819 | return 0; |
820 | } | |
821 | return 1; | |
822 | } | |
823 | ||
824 | static int check_slab(struct kmem_cache *s, struct page *page) | |
825 | { | |
39b26464 CL |
826 | int maxobj; |
827 | ||
81819f0f CL |
828 | VM_BUG_ON(!irqs_disabled()); |
829 | ||
830 | if (!PageSlab(page)) { | |
24922684 | 831 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
832 | return 0; |
833 | } | |
39b26464 | 834 | |
ab9a0f19 | 835 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
836 | if (page->objects > maxobj) { |
837 | slab_err(s, page, "objects %u > max %u", | |
838 | s->name, page->objects, maxobj); | |
839 | return 0; | |
840 | } | |
841 | if (page->inuse > page->objects) { | |
24922684 | 842 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 843 | s->name, page->inuse, page->objects); |
81819f0f CL |
844 | return 0; |
845 | } | |
846 | /* Slab_pad_check fixes things up after itself */ | |
847 | slab_pad_check(s, page); | |
848 | return 1; | |
849 | } | |
850 | ||
851 | /* | |
672bba3a CL |
852 | * Determine if a certain object on a page is on the freelist. Must hold the |
853 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
854 | */ |
855 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
856 | { | |
857 | int nr = 0; | |
881db7fb | 858 | void *fp; |
81819f0f | 859 | void *object = NULL; |
224a88be | 860 | unsigned long max_objects; |
81819f0f | 861 | |
881db7fb | 862 | fp = page->freelist; |
39b26464 | 863 | while (fp && nr <= page->objects) { |
81819f0f CL |
864 | if (fp == search) |
865 | return 1; | |
866 | if (!check_valid_pointer(s, page, fp)) { | |
867 | if (object) { | |
868 | object_err(s, page, object, | |
869 | "Freechain corrupt"); | |
a973e9dd | 870 | set_freepointer(s, object, NULL); |
81819f0f CL |
871 | break; |
872 | } else { | |
24922684 | 873 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 874 | page->freelist = NULL; |
39b26464 | 875 | page->inuse = page->objects; |
24922684 | 876 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
877 | return 0; |
878 | } | |
879 | break; | |
880 | } | |
881 | object = fp; | |
882 | fp = get_freepointer(s, object); | |
883 | nr++; | |
884 | } | |
885 | ||
ab9a0f19 | 886 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
887 | if (max_objects > MAX_OBJS_PER_PAGE) |
888 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
889 | |
890 | if (page->objects != max_objects) { | |
891 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
892 | "should be %d", page->objects, max_objects); | |
893 | page->objects = max_objects; | |
894 | slab_fix(s, "Number of objects adjusted."); | |
895 | } | |
39b26464 | 896 | if (page->inuse != page->objects - nr) { |
70d71228 | 897 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
898 | "counted were %d", page->inuse, page->objects - nr); |
899 | page->inuse = page->objects - nr; | |
24922684 | 900 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
901 | } |
902 | return search == NULL; | |
903 | } | |
904 | ||
0121c619 CL |
905 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
906 | int alloc) | |
3ec09742 CL |
907 | { |
908 | if (s->flags & SLAB_TRACE) { | |
909 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
910 | s->name, | |
911 | alloc ? "alloc" : "free", | |
912 | object, page->inuse, | |
913 | page->freelist); | |
914 | ||
915 | if (!alloc) | |
3b0efdfa | 916 | print_section("Object ", (void *)object, s->object_size); |
3ec09742 CL |
917 | |
918 | dump_stack(); | |
919 | } | |
920 | } | |
921 | ||
c016b0bd CL |
922 | /* |
923 | * Hooks for other subsystems that check memory allocations. In a typical | |
924 | * production configuration these hooks all should produce no code at all. | |
925 | */ | |
926 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
927 | { | |
c1d50836 | 928 | flags &= gfp_allowed_mask; |
c016b0bd CL |
929 | lockdep_trace_alloc(flags); |
930 | might_sleep_if(flags & __GFP_WAIT); | |
931 | ||
3b0efdfa | 932 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
933 | } |
934 | ||
935 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
936 | { | |
c1d50836 | 937 | flags &= gfp_allowed_mask; |
b3d41885 | 938 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 939 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
940 | } |
941 | ||
942 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
943 | { | |
944 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 945 | |
d3f661d6 CL |
946 | /* |
947 | * Trouble is that we may no longer disable interupts in the fast path | |
948 | * So in order to make the debug calls that expect irqs to be | |
949 | * disabled we need to disable interrupts temporarily. | |
950 | */ | |
951 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
952 | { | |
953 | unsigned long flags; | |
954 | ||
955 | local_irq_save(flags); | |
3b0efdfa CL |
956 | kmemcheck_slab_free(s, x, s->object_size); |
957 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
958 | local_irq_restore(flags); |
959 | } | |
960 | #endif | |
f9b615de | 961 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 962 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
963 | } |
964 | ||
643b1138 | 965 | /* |
672bba3a | 966 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
967 | * |
968 | * list_lock must be held. | |
643b1138 | 969 | */ |
5cc6eee8 CL |
970 | static void add_full(struct kmem_cache *s, |
971 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 972 | { |
5cc6eee8 CL |
973 | if (!(s->flags & SLAB_STORE_USER)) |
974 | return; | |
975 | ||
643b1138 | 976 | list_add(&page->lru, &n->full); |
643b1138 CL |
977 | } |
978 | ||
5cc6eee8 CL |
979 | /* |
980 | * list_lock must be held. | |
981 | */ | |
643b1138 CL |
982 | static void remove_full(struct kmem_cache *s, struct page *page) |
983 | { | |
643b1138 CL |
984 | if (!(s->flags & SLAB_STORE_USER)) |
985 | return; | |
986 | ||
643b1138 | 987 | list_del(&page->lru); |
643b1138 CL |
988 | } |
989 | ||
0f389ec6 CL |
990 | /* Tracking of the number of slabs for debugging purposes */ |
991 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
992 | { | |
993 | struct kmem_cache_node *n = get_node(s, node); | |
994 | ||
995 | return atomic_long_read(&n->nr_slabs); | |
996 | } | |
997 | ||
26c02cf0 AB |
998 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
999 | { | |
1000 | return atomic_long_read(&n->nr_slabs); | |
1001 | } | |
1002 | ||
205ab99d | 1003 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1004 | { |
1005 | struct kmem_cache_node *n = get_node(s, node); | |
1006 | ||
1007 | /* | |
1008 | * May be called early in order to allocate a slab for the | |
1009 | * kmem_cache_node structure. Solve the chicken-egg | |
1010 | * dilemma by deferring the increment of the count during | |
1011 | * bootstrap (see early_kmem_cache_node_alloc). | |
1012 | */ | |
7340cc84 | 1013 | if (n) { |
0f389ec6 | 1014 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1015 | atomic_long_add(objects, &n->total_objects); |
1016 | } | |
0f389ec6 | 1017 | } |
205ab99d | 1018 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1019 | { |
1020 | struct kmem_cache_node *n = get_node(s, node); | |
1021 | ||
1022 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1023 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1024 | } |
1025 | ||
1026 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1027 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1028 | void *object) | |
1029 | { | |
1030 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1031 | return; | |
1032 | ||
f7cb1933 | 1033 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1034 | init_tracking(s, object); |
1035 | } | |
1036 | ||
1537066c | 1037 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1038 | void *object, unsigned long addr) |
81819f0f CL |
1039 | { |
1040 | if (!check_slab(s, page)) | |
1041 | goto bad; | |
1042 | ||
81819f0f CL |
1043 | if (!check_valid_pointer(s, page, object)) { |
1044 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1045 | goto bad; |
81819f0f CL |
1046 | } |
1047 | ||
f7cb1933 | 1048 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1049 | goto bad; |
81819f0f | 1050 | |
3ec09742 CL |
1051 | /* Success perform special debug activities for allocs */ |
1052 | if (s->flags & SLAB_STORE_USER) | |
1053 | set_track(s, object, TRACK_ALLOC, addr); | |
1054 | trace(s, page, object, 1); | |
f7cb1933 | 1055 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1056 | return 1; |
3ec09742 | 1057 | |
81819f0f CL |
1058 | bad: |
1059 | if (PageSlab(page)) { | |
1060 | /* | |
1061 | * If this is a slab page then lets do the best we can | |
1062 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1063 | * as used avoids touching the remaining objects. |
81819f0f | 1064 | */ |
24922684 | 1065 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1066 | page->inuse = page->objects; |
a973e9dd | 1067 | page->freelist = NULL; |
81819f0f CL |
1068 | } |
1069 | return 0; | |
1070 | } | |
1071 | ||
19c7ff9e CL |
1072 | static noinline struct kmem_cache_node *free_debug_processing( |
1073 | struct kmem_cache *s, struct page *page, void *object, | |
1074 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1075 | { |
19c7ff9e | 1076 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1077 | |
19c7ff9e | 1078 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1079 | slab_lock(page); |
1080 | ||
81819f0f CL |
1081 | if (!check_slab(s, page)) |
1082 | goto fail; | |
1083 | ||
1084 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1085 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1086 | goto fail; |
1087 | } | |
1088 | ||
1089 | if (on_freelist(s, page, object)) { | |
24922684 | 1090 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1091 | goto fail; |
1092 | } | |
1093 | ||
f7cb1933 | 1094 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1095 | goto out; |
81819f0f CL |
1096 | |
1097 | if (unlikely(s != page->slab)) { | |
3adbefee | 1098 | if (!PageSlab(page)) { |
70d71228 CL |
1099 | slab_err(s, page, "Attempt to free object(0x%p) " |
1100 | "outside of slab", object); | |
3adbefee | 1101 | } else if (!page->slab) { |
81819f0f | 1102 | printk(KERN_ERR |
70d71228 | 1103 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1104 | object); |
70d71228 | 1105 | dump_stack(); |
06428780 | 1106 | } else |
24922684 CL |
1107 | object_err(s, page, object, |
1108 | "page slab pointer corrupt."); | |
81819f0f CL |
1109 | goto fail; |
1110 | } | |
3ec09742 | 1111 | |
3ec09742 CL |
1112 | if (s->flags & SLAB_STORE_USER) |
1113 | set_track(s, object, TRACK_FREE, addr); | |
1114 | trace(s, page, object, 0); | |
f7cb1933 | 1115 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1116 | out: |
881db7fb | 1117 | slab_unlock(page); |
19c7ff9e CL |
1118 | /* |
1119 | * Keep node_lock to preserve integrity | |
1120 | * until the object is actually freed | |
1121 | */ | |
1122 | return n; | |
3ec09742 | 1123 | |
81819f0f | 1124 | fail: |
19c7ff9e CL |
1125 | slab_unlock(page); |
1126 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1127 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1128 | return NULL; |
81819f0f CL |
1129 | } |
1130 | ||
41ecc55b CL |
1131 | static int __init setup_slub_debug(char *str) |
1132 | { | |
f0630fff CL |
1133 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1134 | if (*str++ != '=' || !*str) | |
1135 | /* | |
1136 | * No options specified. Switch on full debugging. | |
1137 | */ | |
1138 | goto out; | |
1139 | ||
1140 | if (*str == ',') | |
1141 | /* | |
1142 | * No options but restriction on slabs. This means full | |
1143 | * debugging for slabs matching a pattern. | |
1144 | */ | |
1145 | goto check_slabs; | |
1146 | ||
fa5ec8a1 DR |
1147 | if (tolower(*str) == 'o') { |
1148 | /* | |
1149 | * Avoid enabling debugging on caches if its minimum order | |
1150 | * would increase as a result. | |
1151 | */ | |
1152 | disable_higher_order_debug = 1; | |
1153 | goto out; | |
1154 | } | |
1155 | ||
f0630fff CL |
1156 | slub_debug = 0; |
1157 | if (*str == '-') | |
1158 | /* | |
1159 | * Switch off all debugging measures. | |
1160 | */ | |
1161 | goto out; | |
1162 | ||
1163 | /* | |
1164 | * Determine which debug features should be switched on | |
1165 | */ | |
06428780 | 1166 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1167 | switch (tolower(*str)) { |
1168 | case 'f': | |
1169 | slub_debug |= SLAB_DEBUG_FREE; | |
1170 | break; | |
1171 | case 'z': | |
1172 | slub_debug |= SLAB_RED_ZONE; | |
1173 | break; | |
1174 | case 'p': | |
1175 | slub_debug |= SLAB_POISON; | |
1176 | break; | |
1177 | case 'u': | |
1178 | slub_debug |= SLAB_STORE_USER; | |
1179 | break; | |
1180 | case 't': | |
1181 | slub_debug |= SLAB_TRACE; | |
1182 | break; | |
4c13dd3b DM |
1183 | case 'a': |
1184 | slub_debug |= SLAB_FAILSLAB; | |
1185 | break; | |
f0630fff CL |
1186 | default: |
1187 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1188 | "unknown. skipped\n", *str); |
f0630fff | 1189 | } |
41ecc55b CL |
1190 | } |
1191 | ||
f0630fff | 1192 | check_slabs: |
41ecc55b CL |
1193 | if (*str == ',') |
1194 | slub_debug_slabs = str + 1; | |
f0630fff | 1195 | out: |
41ecc55b CL |
1196 | return 1; |
1197 | } | |
1198 | ||
1199 | __setup("slub_debug", setup_slub_debug); | |
1200 | ||
3b0efdfa | 1201 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1202 | unsigned long flags, const char *name, |
51cc5068 | 1203 | void (*ctor)(void *)) |
41ecc55b CL |
1204 | { |
1205 | /* | |
e153362a | 1206 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1207 | */ |
e153362a | 1208 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1209 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1210 | flags |= slub_debug; | |
ba0268a8 CL |
1211 | |
1212 | return flags; | |
41ecc55b CL |
1213 | } |
1214 | #else | |
3ec09742 CL |
1215 | static inline void setup_object_debug(struct kmem_cache *s, |
1216 | struct page *page, void *object) {} | |
41ecc55b | 1217 | |
3ec09742 | 1218 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1219 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1220 | |
19c7ff9e CL |
1221 | static inline struct kmem_cache_node *free_debug_processing( |
1222 | struct kmem_cache *s, struct page *page, void *object, | |
1223 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1224 | |
41ecc55b CL |
1225 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1226 | { return 1; } | |
1227 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1228 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1229 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1230 | struct page *page) {} | |
2cfb7455 | 1231 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
3b0efdfa | 1232 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1233 | unsigned long flags, const char *name, |
51cc5068 | 1234 | void (*ctor)(void *)) |
ba0268a8 CL |
1235 | { |
1236 | return flags; | |
1237 | } | |
41ecc55b | 1238 | #define slub_debug 0 |
0f389ec6 | 1239 | |
fdaa45e9 IM |
1240 | #define disable_higher_order_debug 0 |
1241 | ||
0f389ec6 CL |
1242 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1243 | { return 0; } | |
26c02cf0 AB |
1244 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1245 | { return 0; } | |
205ab99d CL |
1246 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1247 | int objects) {} | |
1248 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1249 | int objects) {} | |
7d550c56 CL |
1250 | |
1251 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1252 | { return 0; } | |
1253 | ||
1254 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1255 | void *object) {} | |
1256 | ||
1257 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1258 | ||
ab4d5ed5 | 1259 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1260 | |
81819f0f CL |
1261 | /* |
1262 | * Slab allocation and freeing | |
1263 | */ | |
65c3376a CL |
1264 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1265 | struct kmem_cache_order_objects oo) | |
1266 | { | |
1267 | int order = oo_order(oo); | |
1268 | ||
b1eeab67 VN |
1269 | flags |= __GFP_NOTRACK; |
1270 | ||
2154a336 | 1271 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1272 | return alloc_pages(flags, order); |
1273 | else | |
6b65aaf3 | 1274 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1275 | } |
1276 | ||
81819f0f CL |
1277 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1278 | { | |
06428780 | 1279 | struct page *page; |
834f3d11 | 1280 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1281 | gfp_t alloc_gfp; |
81819f0f | 1282 | |
7e0528da CL |
1283 | flags &= gfp_allowed_mask; |
1284 | ||
1285 | if (flags & __GFP_WAIT) | |
1286 | local_irq_enable(); | |
1287 | ||
b7a49f0d | 1288 | flags |= s->allocflags; |
e12ba74d | 1289 | |
ba52270d PE |
1290 | /* |
1291 | * Let the initial higher-order allocation fail under memory pressure | |
1292 | * so we fall-back to the minimum order allocation. | |
1293 | */ | |
1294 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1295 | ||
1296 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1297 | if (unlikely(!page)) { |
1298 | oo = s->min; | |
1299 | /* | |
1300 | * Allocation may have failed due to fragmentation. | |
1301 | * Try a lower order alloc if possible | |
1302 | */ | |
1303 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1304 | |
7e0528da CL |
1305 | if (page) |
1306 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1307 | } |
5a896d9e | 1308 | |
737b719e | 1309 | if (kmemcheck_enabled && page |
5086c389 | 1310 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1311 | int pages = 1 << oo_order(oo); |
1312 | ||
1313 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1314 | ||
1315 | /* | |
1316 | * Objects from caches that have a constructor don't get | |
1317 | * cleared when they're allocated, so we need to do it here. | |
1318 | */ | |
1319 | if (s->ctor) | |
1320 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1321 | else | |
1322 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1323 | } |
1324 | ||
737b719e DR |
1325 | if (flags & __GFP_WAIT) |
1326 | local_irq_disable(); | |
1327 | if (!page) | |
1328 | return NULL; | |
1329 | ||
834f3d11 | 1330 | page->objects = oo_objects(oo); |
81819f0f CL |
1331 | mod_zone_page_state(page_zone(page), |
1332 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1333 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1334 | 1 << oo_order(oo)); |
81819f0f CL |
1335 | |
1336 | return page; | |
1337 | } | |
1338 | ||
1339 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1340 | void *object) | |
1341 | { | |
3ec09742 | 1342 | setup_object_debug(s, page, object); |
4f104934 | 1343 | if (unlikely(s->ctor)) |
51cc5068 | 1344 | s->ctor(object); |
81819f0f CL |
1345 | } |
1346 | ||
1347 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1348 | { | |
1349 | struct page *page; | |
81819f0f | 1350 | void *start; |
81819f0f CL |
1351 | void *last; |
1352 | void *p; | |
1353 | ||
6cb06229 | 1354 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1355 | |
6cb06229 CL |
1356 | page = allocate_slab(s, |
1357 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1358 | if (!page) |
1359 | goto out; | |
1360 | ||
205ab99d | 1361 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f | 1362 | page->slab = s; |
c03f94cc | 1363 | __SetPageSlab(page); |
072bb0aa MG |
1364 | if (page->pfmemalloc) |
1365 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1366 | |
1367 | start = page_address(page); | |
81819f0f CL |
1368 | |
1369 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1370 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1371 | |
1372 | last = start; | |
224a88be | 1373 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1374 | setup_object(s, page, last); |
1375 | set_freepointer(s, last, p); | |
1376 | last = p; | |
1377 | } | |
1378 | setup_object(s, page, last); | |
a973e9dd | 1379 | set_freepointer(s, last, NULL); |
81819f0f CL |
1380 | |
1381 | page->freelist = start; | |
e6e82ea1 | 1382 | page->inuse = page->objects; |
8cb0a506 | 1383 | page->frozen = 1; |
81819f0f | 1384 | out: |
81819f0f CL |
1385 | return page; |
1386 | } | |
1387 | ||
1388 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1389 | { | |
834f3d11 CL |
1390 | int order = compound_order(page); |
1391 | int pages = 1 << order; | |
81819f0f | 1392 | |
af537b0a | 1393 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1394 | void *p; |
1395 | ||
1396 | slab_pad_check(s, page); | |
224a88be CL |
1397 | for_each_object(p, s, page_address(page), |
1398 | page->objects) | |
f7cb1933 | 1399 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1400 | } |
1401 | ||
b1eeab67 | 1402 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1403 | |
81819f0f CL |
1404 | mod_zone_page_state(page_zone(page), |
1405 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1406 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1407 | -pages); |
81819f0f | 1408 | |
072bb0aa | 1409 | __ClearPageSlabPfmemalloc(page); |
49bd5221 CL |
1410 | __ClearPageSlab(page); |
1411 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1412 | if (current->reclaim_state) |
1413 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1414 | __free_pages(page, order); |
81819f0f CL |
1415 | } |
1416 | ||
da9a638c LJ |
1417 | #define need_reserve_slab_rcu \ |
1418 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1419 | ||
81819f0f CL |
1420 | static void rcu_free_slab(struct rcu_head *h) |
1421 | { | |
1422 | struct page *page; | |
1423 | ||
da9a638c LJ |
1424 | if (need_reserve_slab_rcu) |
1425 | page = virt_to_head_page(h); | |
1426 | else | |
1427 | page = container_of((struct list_head *)h, struct page, lru); | |
1428 | ||
81819f0f CL |
1429 | __free_slab(page->slab, page); |
1430 | } | |
1431 | ||
1432 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1433 | { | |
1434 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1435 | struct rcu_head *head; |
1436 | ||
1437 | if (need_reserve_slab_rcu) { | |
1438 | int order = compound_order(page); | |
1439 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1440 | ||
1441 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1442 | head = page_address(page) + offset; | |
1443 | } else { | |
1444 | /* | |
1445 | * RCU free overloads the RCU head over the LRU | |
1446 | */ | |
1447 | head = (void *)&page->lru; | |
1448 | } | |
81819f0f CL |
1449 | |
1450 | call_rcu(head, rcu_free_slab); | |
1451 | } else | |
1452 | __free_slab(s, page); | |
1453 | } | |
1454 | ||
1455 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1456 | { | |
205ab99d | 1457 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1458 | free_slab(s, page); |
1459 | } | |
1460 | ||
1461 | /* | |
5cc6eee8 CL |
1462 | * Management of partially allocated slabs. |
1463 | * | |
1464 | * list_lock must be held. | |
81819f0f | 1465 | */ |
5cc6eee8 | 1466 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1467 | struct page *page, int tail) |
81819f0f | 1468 | { |
e95eed57 | 1469 | n->nr_partial++; |
136333d1 | 1470 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1471 | list_add_tail(&page->lru, &n->partial); |
1472 | else | |
1473 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1474 | } |
1475 | ||
5cc6eee8 CL |
1476 | /* |
1477 | * list_lock must be held. | |
1478 | */ | |
1479 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1480 | struct page *page) |
1481 | { | |
1482 | list_del(&page->lru); | |
1483 | n->nr_partial--; | |
1484 | } | |
1485 | ||
81819f0f | 1486 | /* |
7ced3719 CL |
1487 | * Remove slab from the partial list, freeze it and |
1488 | * return the pointer to the freelist. | |
81819f0f | 1489 | * |
497b66f2 CL |
1490 | * Returns a list of objects or NULL if it fails. |
1491 | * | |
7ced3719 | 1492 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1493 | */ |
497b66f2 | 1494 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1495 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1496 | int mode) |
81819f0f | 1497 | { |
2cfb7455 CL |
1498 | void *freelist; |
1499 | unsigned long counters; | |
1500 | struct page new; | |
1501 | ||
2cfb7455 CL |
1502 | /* |
1503 | * Zap the freelist and set the frozen bit. | |
1504 | * The old freelist is the list of objects for the | |
1505 | * per cpu allocation list. | |
1506 | */ | |
7ced3719 CL |
1507 | freelist = page->freelist; |
1508 | counters = page->counters; | |
1509 | new.counters = counters; | |
23910c50 | 1510 | if (mode) { |
7ced3719 | 1511 | new.inuse = page->objects; |
23910c50 PE |
1512 | new.freelist = NULL; |
1513 | } else { | |
1514 | new.freelist = freelist; | |
1515 | } | |
2cfb7455 | 1516 | |
7ced3719 CL |
1517 | VM_BUG_ON(new.frozen); |
1518 | new.frozen = 1; | |
2cfb7455 | 1519 | |
7ced3719 | 1520 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1521 | freelist, counters, |
02d7633f | 1522 | new.freelist, new.counters, |
7ced3719 | 1523 | "acquire_slab")) |
7ced3719 | 1524 | return NULL; |
2cfb7455 CL |
1525 | |
1526 | remove_partial(n, page); | |
7ced3719 | 1527 | WARN_ON(!freelist); |
49e22585 | 1528 | return freelist; |
81819f0f CL |
1529 | } |
1530 | ||
49e22585 CL |
1531 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
1532 | ||
81819f0f | 1533 | /* |
672bba3a | 1534 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1535 | */ |
497b66f2 | 1536 | static void *get_partial_node(struct kmem_cache *s, |
acd19fd1 | 1537 | struct kmem_cache_node *n, struct kmem_cache_cpu *c) |
81819f0f | 1538 | { |
49e22585 CL |
1539 | struct page *page, *page2; |
1540 | void *object = NULL; | |
81819f0f CL |
1541 | |
1542 | /* | |
1543 | * Racy check. If we mistakenly see no partial slabs then we | |
1544 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1545 | * partial slab and there is none available then get_partials() |
1546 | * will return NULL. | |
81819f0f CL |
1547 | */ |
1548 | if (!n || !n->nr_partial) | |
1549 | return NULL; | |
1550 | ||
1551 | spin_lock(&n->list_lock); | |
49e22585 | 1552 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
12d79634 | 1553 | void *t = acquire_slab(s, n, page, object == NULL); |
49e22585 CL |
1554 | int available; |
1555 | ||
1556 | if (!t) | |
1557 | break; | |
1558 | ||
12d79634 | 1559 | if (!object) { |
49e22585 | 1560 | c->page = page; |
49e22585 | 1561 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 CL |
1562 | object = t; |
1563 | available = page->objects - page->inuse; | |
1564 | } else { | |
49e22585 | 1565 | available = put_cpu_partial(s, page, 0); |
8028dcea | 1566 | stat(s, CPU_PARTIAL_NODE); |
49e22585 CL |
1567 | } |
1568 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1569 | break; | |
1570 | ||
497b66f2 | 1571 | } |
81819f0f | 1572 | spin_unlock(&n->list_lock); |
497b66f2 | 1573 | return object; |
81819f0f CL |
1574 | } |
1575 | ||
1576 | /* | |
672bba3a | 1577 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1578 | */ |
de3ec035 | 1579 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1580 | struct kmem_cache_cpu *c) |
81819f0f CL |
1581 | { |
1582 | #ifdef CONFIG_NUMA | |
1583 | struct zonelist *zonelist; | |
dd1a239f | 1584 | struct zoneref *z; |
54a6eb5c MG |
1585 | struct zone *zone; |
1586 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1587 | void *object; |
cc9a6c87 | 1588 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1589 | |
1590 | /* | |
672bba3a CL |
1591 | * The defrag ratio allows a configuration of the tradeoffs between |
1592 | * inter node defragmentation and node local allocations. A lower | |
1593 | * defrag_ratio increases the tendency to do local allocations | |
1594 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1595 | * |
672bba3a CL |
1596 | * If the defrag_ratio is set to 0 then kmalloc() always |
1597 | * returns node local objects. If the ratio is higher then kmalloc() | |
1598 | * may return off node objects because partial slabs are obtained | |
1599 | * from other nodes and filled up. | |
81819f0f | 1600 | * |
6446faa2 | 1601 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1602 | * defrag_ratio = 1000) then every (well almost) allocation will |
1603 | * first attempt to defrag slab caches on other nodes. This means | |
1604 | * scanning over all nodes to look for partial slabs which may be | |
1605 | * expensive if we do it every time we are trying to find a slab | |
1606 | * with available objects. | |
81819f0f | 1607 | */ |
9824601e CL |
1608 | if (!s->remote_node_defrag_ratio || |
1609 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1610 | return NULL; |
1611 | ||
cc9a6c87 MG |
1612 | do { |
1613 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 1614 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
1615 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1616 | struct kmem_cache_node *n; | |
1617 | ||
1618 | n = get_node(s, zone_to_nid(zone)); | |
1619 | ||
1620 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1621 | n->nr_partial > s->min_partial) { | |
1622 | object = get_partial_node(s, n, c); | |
1623 | if (object) { | |
1624 | /* | |
1625 | * Return the object even if | |
1626 | * put_mems_allowed indicated that | |
1627 | * the cpuset mems_allowed was | |
1628 | * updated in parallel. It's a | |
1629 | * harmless race between the alloc | |
1630 | * and the cpuset update. | |
1631 | */ | |
1632 | put_mems_allowed(cpuset_mems_cookie); | |
1633 | return object; | |
1634 | } | |
c0ff7453 | 1635 | } |
81819f0f | 1636 | } |
cc9a6c87 | 1637 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1638 | #endif |
1639 | return NULL; | |
1640 | } | |
1641 | ||
1642 | /* | |
1643 | * Get a partial page, lock it and return it. | |
1644 | */ | |
497b66f2 | 1645 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1646 | struct kmem_cache_cpu *c) |
81819f0f | 1647 | { |
497b66f2 | 1648 | void *object; |
2154a336 | 1649 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1650 | |
497b66f2 CL |
1651 | object = get_partial_node(s, get_node(s, searchnode), c); |
1652 | if (object || node != NUMA_NO_NODE) | |
1653 | return object; | |
81819f0f | 1654 | |
acd19fd1 | 1655 | return get_any_partial(s, flags, c); |
81819f0f CL |
1656 | } |
1657 | ||
8a5ec0ba CL |
1658 | #ifdef CONFIG_PREEMPT |
1659 | /* | |
1660 | * Calculate the next globally unique transaction for disambiguiation | |
1661 | * during cmpxchg. The transactions start with the cpu number and are then | |
1662 | * incremented by CONFIG_NR_CPUS. | |
1663 | */ | |
1664 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1665 | #else | |
1666 | /* | |
1667 | * No preemption supported therefore also no need to check for | |
1668 | * different cpus. | |
1669 | */ | |
1670 | #define TID_STEP 1 | |
1671 | #endif | |
1672 | ||
1673 | static inline unsigned long next_tid(unsigned long tid) | |
1674 | { | |
1675 | return tid + TID_STEP; | |
1676 | } | |
1677 | ||
1678 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1679 | { | |
1680 | return tid % TID_STEP; | |
1681 | } | |
1682 | ||
1683 | static inline unsigned long tid_to_event(unsigned long tid) | |
1684 | { | |
1685 | return tid / TID_STEP; | |
1686 | } | |
1687 | ||
1688 | static inline unsigned int init_tid(int cpu) | |
1689 | { | |
1690 | return cpu; | |
1691 | } | |
1692 | ||
1693 | static inline void note_cmpxchg_failure(const char *n, | |
1694 | const struct kmem_cache *s, unsigned long tid) | |
1695 | { | |
1696 | #ifdef SLUB_DEBUG_CMPXCHG | |
1697 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1698 | ||
1699 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1700 | ||
1701 | #ifdef CONFIG_PREEMPT | |
1702 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1703 | printk("due to cpu change %d -> %d\n", | |
1704 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1705 | else | |
1706 | #endif | |
1707 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1708 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1709 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1710 | else | |
1711 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1712 | actual_tid, tid, next_tid(tid)); | |
1713 | #endif | |
4fdccdfb | 1714 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1715 | } |
1716 | ||
8a5ec0ba CL |
1717 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1718 | { | |
8a5ec0ba CL |
1719 | int cpu; |
1720 | ||
1721 | for_each_possible_cpu(cpu) | |
1722 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1723 | } |
2cfb7455 | 1724 | |
81819f0f CL |
1725 | /* |
1726 | * Remove the cpu slab | |
1727 | */ | |
c17dda40 | 1728 | static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist) |
81819f0f | 1729 | { |
2cfb7455 | 1730 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1731 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1732 | int lock = 0; | |
1733 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1734 | void *nextfree; |
136333d1 | 1735 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1736 | struct page new; |
1737 | struct page old; | |
1738 | ||
1739 | if (page->freelist) { | |
84e554e6 | 1740 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1741 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1742 | } |
1743 | ||
894b8788 | 1744 | /* |
2cfb7455 CL |
1745 | * Stage one: Free all available per cpu objects back |
1746 | * to the page freelist while it is still frozen. Leave the | |
1747 | * last one. | |
1748 | * | |
1749 | * There is no need to take the list->lock because the page | |
1750 | * is still frozen. | |
1751 | */ | |
1752 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1753 | void *prior; | |
1754 | unsigned long counters; | |
1755 | ||
1756 | do { | |
1757 | prior = page->freelist; | |
1758 | counters = page->counters; | |
1759 | set_freepointer(s, freelist, prior); | |
1760 | new.counters = counters; | |
1761 | new.inuse--; | |
1762 | VM_BUG_ON(!new.frozen); | |
1763 | ||
1d07171c | 1764 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1765 | prior, counters, |
1766 | freelist, new.counters, | |
1767 | "drain percpu freelist")); | |
1768 | ||
1769 | freelist = nextfree; | |
1770 | } | |
1771 | ||
894b8788 | 1772 | /* |
2cfb7455 CL |
1773 | * Stage two: Ensure that the page is unfrozen while the |
1774 | * list presence reflects the actual number of objects | |
1775 | * during unfreeze. | |
1776 | * | |
1777 | * We setup the list membership and then perform a cmpxchg | |
1778 | * with the count. If there is a mismatch then the page | |
1779 | * is not unfrozen but the page is on the wrong list. | |
1780 | * | |
1781 | * Then we restart the process which may have to remove | |
1782 | * the page from the list that we just put it on again | |
1783 | * because the number of objects in the slab may have | |
1784 | * changed. | |
894b8788 | 1785 | */ |
2cfb7455 | 1786 | redo: |
894b8788 | 1787 | |
2cfb7455 CL |
1788 | old.freelist = page->freelist; |
1789 | old.counters = page->counters; | |
1790 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1791 | |
2cfb7455 CL |
1792 | /* Determine target state of the slab */ |
1793 | new.counters = old.counters; | |
1794 | if (freelist) { | |
1795 | new.inuse--; | |
1796 | set_freepointer(s, freelist, old.freelist); | |
1797 | new.freelist = freelist; | |
1798 | } else | |
1799 | new.freelist = old.freelist; | |
1800 | ||
1801 | new.frozen = 0; | |
1802 | ||
81107188 | 1803 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1804 | m = M_FREE; |
1805 | else if (new.freelist) { | |
1806 | m = M_PARTIAL; | |
1807 | if (!lock) { | |
1808 | lock = 1; | |
1809 | /* | |
1810 | * Taking the spinlock removes the possiblity | |
1811 | * that acquire_slab() will see a slab page that | |
1812 | * is frozen | |
1813 | */ | |
1814 | spin_lock(&n->list_lock); | |
1815 | } | |
1816 | } else { | |
1817 | m = M_FULL; | |
1818 | if (kmem_cache_debug(s) && !lock) { | |
1819 | lock = 1; | |
1820 | /* | |
1821 | * This also ensures that the scanning of full | |
1822 | * slabs from diagnostic functions will not see | |
1823 | * any frozen slabs. | |
1824 | */ | |
1825 | spin_lock(&n->list_lock); | |
1826 | } | |
1827 | } | |
1828 | ||
1829 | if (l != m) { | |
1830 | ||
1831 | if (l == M_PARTIAL) | |
1832 | ||
1833 | remove_partial(n, page); | |
1834 | ||
1835 | else if (l == M_FULL) | |
894b8788 | 1836 | |
2cfb7455 CL |
1837 | remove_full(s, page); |
1838 | ||
1839 | if (m == M_PARTIAL) { | |
1840 | ||
1841 | add_partial(n, page, tail); | |
136333d1 | 1842 | stat(s, tail); |
2cfb7455 CL |
1843 | |
1844 | } else if (m == M_FULL) { | |
894b8788 | 1845 | |
2cfb7455 CL |
1846 | stat(s, DEACTIVATE_FULL); |
1847 | add_full(s, n, page); | |
1848 | ||
1849 | } | |
1850 | } | |
1851 | ||
1852 | l = m; | |
1d07171c | 1853 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1854 | old.freelist, old.counters, |
1855 | new.freelist, new.counters, | |
1856 | "unfreezing slab")) | |
1857 | goto redo; | |
1858 | ||
2cfb7455 CL |
1859 | if (lock) |
1860 | spin_unlock(&n->list_lock); | |
1861 | ||
1862 | if (m == M_FREE) { | |
1863 | stat(s, DEACTIVATE_EMPTY); | |
1864 | discard_slab(s, page); | |
1865 | stat(s, FREE_SLAB); | |
894b8788 | 1866 | } |
81819f0f CL |
1867 | } |
1868 | ||
d24ac77f JK |
1869 | /* |
1870 | * Unfreeze all the cpu partial slabs. | |
1871 | * | |
1872 | * This function must be called with interrupt disabled. | |
1873 | */ | |
49e22585 CL |
1874 | static void unfreeze_partials(struct kmem_cache *s) |
1875 | { | |
43d77867 | 1876 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
49e22585 | 1877 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); |
9ada1934 | 1878 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1879 | |
1880 | while ((page = c->partial)) { | |
49e22585 CL |
1881 | struct page new; |
1882 | struct page old; | |
1883 | ||
1884 | c->partial = page->next; | |
43d77867 JK |
1885 | |
1886 | n2 = get_node(s, page_to_nid(page)); | |
1887 | if (n != n2) { | |
1888 | if (n) | |
1889 | spin_unlock(&n->list_lock); | |
1890 | ||
1891 | n = n2; | |
1892 | spin_lock(&n->list_lock); | |
1893 | } | |
49e22585 CL |
1894 | |
1895 | do { | |
1896 | ||
1897 | old.freelist = page->freelist; | |
1898 | old.counters = page->counters; | |
1899 | VM_BUG_ON(!old.frozen); | |
1900 | ||
1901 | new.counters = old.counters; | |
1902 | new.freelist = old.freelist; | |
1903 | ||
1904 | new.frozen = 0; | |
1905 | ||
d24ac77f | 1906 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1907 | old.freelist, old.counters, |
1908 | new.freelist, new.counters, | |
1909 | "unfreezing slab")); | |
1910 | ||
43d77867 | 1911 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1912 | page->next = discard_page; |
1913 | discard_page = page; | |
43d77867 JK |
1914 | } else { |
1915 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1916 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1917 | } |
1918 | } | |
1919 | ||
1920 | if (n) | |
1921 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1922 | |
1923 | while (discard_page) { | |
1924 | page = discard_page; | |
1925 | discard_page = discard_page->next; | |
1926 | ||
1927 | stat(s, DEACTIVATE_EMPTY); | |
1928 | discard_slab(s, page); | |
1929 | stat(s, FREE_SLAB); | |
1930 | } | |
49e22585 CL |
1931 | } |
1932 | ||
1933 | /* | |
1934 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1935 | * slot if available. This is done without interrupts disabled and without | |
1936 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1937 | * onto a random cpus partial slot. | |
1938 | * | |
1939 | * If we did not find a slot then simply move all the partials to the | |
1940 | * per node partial list. | |
1941 | */ | |
1942 | int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
1943 | { | |
1944 | struct page *oldpage; | |
1945 | int pages; | |
1946 | int pobjects; | |
1947 | ||
1948 | do { | |
1949 | pages = 0; | |
1950 | pobjects = 0; | |
1951 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
1952 | ||
1953 | if (oldpage) { | |
1954 | pobjects = oldpage->pobjects; | |
1955 | pages = oldpage->pages; | |
1956 | if (drain && pobjects > s->cpu_partial) { | |
1957 | unsigned long flags; | |
1958 | /* | |
1959 | * partial array is full. Move the existing | |
1960 | * set to the per node partial list. | |
1961 | */ | |
1962 | local_irq_save(flags); | |
1963 | unfreeze_partials(s); | |
1964 | local_irq_restore(flags); | |
e24fc410 | 1965 | oldpage = NULL; |
49e22585 CL |
1966 | pobjects = 0; |
1967 | pages = 0; | |
8028dcea | 1968 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
1969 | } |
1970 | } | |
1971 | ||
1972 | pages++; | |
1973 | pobjects += page->objects - page->inuse; | |
1974 | ||
1975 | page->pages = pages; | |
1976 | page->pobjects = pobjects; | |
1977 | page->next = oldpage; | |
1978 | ||
933393f5 | 1979 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
1980 | return pobjects; |
1981 | } | |
1982 | ||
dfb4f096 | 1983 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1984 | { |
84e554e6 | 1985 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
1986 | deactivate_slab(s, c->page, c->freelist); |
1987 | ||
1988 | c->tid = next_tid(c->tid); | |
1989 | c->page = NULL; | |
1990 | c->freelist = NULL; | |
81819f0f CL |
1991 | } |
1992 | ||
1993 | /* | |
1994 | * Flush cpu slab. | |
6446faa2 | 1995 | * |
81819f0f CL |
1996 | * Called from IPI handler with interrupts disabled. |
1997 | */ | |
0c710013 | 1998 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 1999 | { |
9dfc6e68 | 2000 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2001 | |
49e22585 CL |
2002 | if (likely(c)) { |
2003 | if (c->page) | |
2004 | flush_slab(s, c); | |
2005 | ||
2006 | unfreeze_partials(s); | |
2007 | } | |
81819f0f CL |
2008 | } |
2009 | ||
2010 | static void flush_cpu_slab(void *d) | |
2011 | { | |
2012 | struct kmem_cache *s = d; | |
81819f0f | 2013 | |
dfb4f096 | 2014 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2015 | } |
2016 | ||
a8364d55 GBY |
2017 | static bool has_cpu_slab(int cpu, void *info) |
2018 | { | |
2019 | struct kmem_cache *s = info; | |
2020 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2021 | ||
02e1a9cd | 2022 | return c->page || c->partial; |
a8364d55 GBY |
2023 | } |
2024 | ||
81819f0f CL |
2025 | static void flush_all(struct kmem_cache *s) |
2026 | { | |
a8364d55 | 2027 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2028 | } |
2029 | ||
dfb4f096 CL |
2030 | /* |
2031 | * Check if the objects in a per cpu structure fit numa | |
2032 | * locality expectations. | |
2033 | */ | |
57d437d2 | 2034 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2035 | { |
2036 | #ifdef CONFIG_NUMA | |
57d437d2 | 2037 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2038 | return 0; |
2039 | #endif | |
2040 | return 1; | |
2041 | } | |
2042 | ||
781b2ba6 PE |
2043 | static int count_free(struct page *page) |
2044 | { | |
2045 | return page->objects - page->inuse; | |
2046 | } | |
2047 | ||
2048 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2049 | int (*get_count)(struct page *)) | |
2050 | { | |
2051 | unsigned long flags; | |
2052 | unsigned long x = 0; | |
2053 | struct page *page; | |
2054 | ||
2055 | spin_lock_irqsave(&n->list_lock, flags); | |
2056 | list_for_each_entry(page, &n->partial, lru) | |
2057 | x += get_count(page); | |
2058 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2059 | return x; | |
2060 | } | |
2061 | ||
26c02cf0 AB |
2062 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2063 | { | |
2064 | #ifdef CONFIG_SLUB_DEBUG | |
2065 | return atomic_long_read(&n->total_objects); | |
2066 | #else | |
2067 | return 0; | |
2068 | #endif | |
2069 | } | |
2070 | ||
781b2ba6 PE |
2071 | static noinline void |
2072 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2073 | { | |
2074 | int node; | |
2075 | ||
2076 | printk(KERN_WARNING | |
2077 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2078 | nid, gfpflags); | |
2079 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2080 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2081 | s->size, oo_order(s->oo), oo_order(s->min)); |
2082 | ||
3b0efdfa | 2083 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2084 | printk(KERN_WARNING " %s debugging increased min order, use " |
2085 | "slub_debug=O to disable.\n", s->name); | |
2086 | ||
781b2ba6 PE |
2087 | for_each_online_node(node) { |
2088 | struct kmem_cache_node *n = get_node(s, node); | |
2089 | unsigned long nr_slabs; | |
2090 | unsigned long nr_objs; | |
2091 | unsigned long nr_free; | |
2092 | ||
2093 | if (!n) | |
2094 | continue; | |
2095 | ||
26c02cf0 AB |
2096 | nr_free = count_partial(n, count_free); |
2097 | nr_slabs = node_nr_slabs(n); | |
2098 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2099 | |
2100 | printk(KERN_WARNING | |
2101 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2102 | node, nr_slabs, nr_objs, nr_free); | |
2103 | } | |
2104 | } | |
2105 | ||
497b66f2 CL |
2106 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2107 | int node, struct kmem_cache_cpu **pc) | |
2108 | { | |
6faa6833 | 2109 | void *freelist; |
188fd063 CL |
2110 | struct kmem_cache_cpu *c = *pc; |
2111 | struct page *page; | |
497b66f2 | 2112 | |
188fd063 | 2113 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2114 | |
188fd063 CL |
2115 | if (freelist) |
2116 | return freelist; | |
2117 | ||
2118 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2119 | if (page) { |
2120 | c = __this_cpu_ptr(s->cpu_slab); | |
2121 | if (c->page) | |
2122 | flush_slab(s, c); | |
2123 | ||
2124 | /* | |
2125 | * No other reference to the page yet so we can | |
2126 | * muck around with it freely without cmpxchg | |
2127 | */ | |
6faa6833 | 2128 | freelist = page->freelist; |
497b66f2 CL |
2129 | page->freelist = NULL; |
2130 | ||
2131 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2132 | c->page = page; |
2133 | *pc = c; | |
2134 | } else | |
6faa6833 | 2135 | freelist = NULL; |
497b66f2 | 2136 | |
6faa6833 | 2137 | return freelist; |
497b66f2 CL |
2138 | } |
2139 | ||
072bb0aa MG |
2140 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2141 | { | |
2142 | if (unlikely(PageSlabPfmemalloc(page))) | |
2143 | return gfp_pfmemalloc_allowed(gfpflags); | |
2144 | ||
2145 | return true; | |
2146 | } | |
2147 | ||
213eeb9f CL |
2148 | /* |
2149 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2150 | * or deactivate the page. | |
2151 | * | |
2152 | * The page is still frozen if the return value is not NULL. | |
2153 | * | |
2154 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2155 | * |
2156 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2157 | */ |
2158 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2159 | { | |
2160 | struct page new; | |
2161 | unsigned long counters; | |
2162 | void *freelist; | |
2163 | ||
2164 | do { | |
2165 | freelist = page->freelist; | |
2166 | counters = page->counters; | |
6faa6833 | 2167 | |
213eeb9f CL |
2168 | new.counters = counters; |
2169 | VM_BUG_ON(!new.frozen); | |
2170 | ||
2171 | new.inuse = page->objects; | |
2172 | new.frozen = freelist != NULL; | |
2173 | ||
d24ac77f | 2174 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2175 | freelist, counters, |
2176 | NULL, new.counters, | |
2177 | "get_freelist")); | |
2178 | ||
2179 | return freelist; | |
2180 | } | |
2181 | ||
81819f0f | 2182 | /* |
894b8788 CL |
2183 | * Slow path. The lockless freelist is empty or we need to perform |
2184 | * debugging duties. | |
2185 | * | |
894b8788 CL |
2186 | * Processing is still very fast if new objects have been freed to the |
2187 | * regular freelist. In that case we simply take over the regular freelist | |
2188 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2189 | * |
894b8788 CL |
2190 | * If that is not working then we fall back to the partial lists. We take the |
2191 | * first element of the freelist as the object to allocate now and move the | |
2192 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2193 | * |
894b8788 | 2194 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2195 | * we need to allocate a new slab. This is the slowest path since it involves |
2196 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2197 | */ |
ce71e27c EGM |
2198 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2199 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2200 | { |
6faa6833 | 2201 | void *freelist; |
f6e7def7 | 2202 | struct page *page; |
8a5ec0ba CL |
2203 | unsigned long flags; |
2204 | ||
2205 | local_irq_save(flags); | |
2206 | #ifdef CONFIG_PREEMPT | |
2207 | /* | |
2208 | * We may have been preempted and rescheduled on a different | |
2209 | * cpu before disabling interrupts. Need to reload cpu area | |
2210 | * pointer. | |
2211 | */ | |
2212 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2213 | #endif |
81819f0f | 2214 | |
f6e7def7 CL |
2215 | page = c->page; |
2216 | if (!page) | |
81819f0f | 2217 | goto new_slab; |
49e22585 | 2218 | redo: |
6faa6833 | 2219 | |
57d437d2 | 2220 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2221 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2222 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2223 | c->page = NULL; |
2224 | c->freelist = NULL; | |
fc59c053 CL |
2225 | goto new_slab; |
2226 | } | |
6446faa2 | 2227 | |
072bb0aa MG |
2228 | /* |
2229 | * By rights, we should be searching for a slab page that was | |
2230 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2231 | * information when the page leaves the per-cpu allocator | |
2232 | */ | |
2233 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2234 | deactivate_slab(s, page, c->freelist); | |
2235 | c->page = NULL; | |
2236 | c->freelist = NULL; | |
2237 | goto new_slab; | |
2238 | } | |
2239 | ||
73736e03 | 2240 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2241 | freelist = c->freelist; |
2242 | if (freelist) | |
73736e03 | 2243 | goto load_freelist; |
03e404af | 2244 | |
2cfb7455 | 2245 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2246 | |
f6e7def7 | 2247 | freelist = get_freelist(s, page); |
6446faa2 | 2248 | |
6faa6833 | 2249 | if (!freelist) { |
03e404af CL |
2250 | c->page = NULL; |
2251 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2252 | goto new_slab; |
03e404af | 2253 | } |
6446faa2 | 2254 | |
84e554e6 | 2255 | stat(s, ALLOC_REFILL); |
6446faa2 | 2256 | |
894b8788 | 2257 | load_freelist: |
507effea CL |
2258 | /* |
2259 | * freelist is pointing to the list of objects to be used. | |
2260 | * page is pointing to the page from which the objects are obtained. | |
2261 | * That page must be frozen for per cpu allocations to work. | |
2262 | */ | |
2263 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2264 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2265 | c->tid = next_tid(c->tid); |
2266 | local_irq_restore(flags); | |
6faa6833 | 2267 | return freelist; |
81819f0f | 2268 | |
81819f0f | 2269 | new_slab: |
2cfb7455 | 2270 | |
49e22585 | 2271 | if (c->partial) { |
f6e7def7 CL |
2272 | page = c->page = c->partial; |
2273 | c->partial = page->next; | |
49e22585 CL |
2274 | stat(s, CPU_PARTIAL_ALLOC); |
2275 | c->freelist = NULL; | |
2276 | goto redo; | |
81819f0f CL |
2277 | } |
2278 | ||
188fd063 | 2279 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2280 | |
f4697436 CL |
2281 | if (unlikely(!freelist)) { |
2282 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2283 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2284 | |
f4697436 CL |
2285 | local_irq_restore(flags); |
2286 | return NULL; | |
81819f0f | 2287 | } |
2cfb7455 | 2288 | |
f6e7def7 | 2289 | page = c->page; |
5091b74a | 2290 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2291 | goto load_freelist; |
2cfb7455 | 2292 | |
497b66f2 | 2293 | /* Only entered in the debug case */ |
5091b74a | 2294 | if (kmem_cache_debug(s) && !alloc_debug_processing(s, page, freelist, addr)) |
497b66f2 | 2295 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2296 | |
f6e7def7 | 2297 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2298 | c->page = NULL; |
2299 | c->freelist = NULL; | |
a71ae47a | 2300 | local_irq_restore(flags); |
6faa6833 | 2301 | return freelist; |
894b8788 CL |
2302 | } |
2303 | ||
2304 | /* | |
2305 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2306 | * have the fastpath folded into their functions. So no function call | |
2307 | * overhead for requests that can be satisfied on the fastpath. | |
2308 | * | |
2309 | * The fastpath works by first checking if the lockless freelist can be used. | |
2310 | * If not then __slab_alloc is called for slow processing. | |
2311 | * | |
2312 | * Otherwise we can simply pick the next object from the lockless free list. | |
2313 | */ | |
06428780 | 2314 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2315 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2316 | { |
894b8788 | 2317 | void **object; |
dfb4f096 | 2318 | struct kmem_cache_cpu *c; |
57d437d2 | 2319 | struct page *page; |
8a5ec0ba | 2320 | unsigned long tid; |
1f84260c | 2321 | |
c016b0bd | 2322 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2323 | return NULL; |
1f84260c | 2324 | |
8a5ec0ba | 2325 | redo: |
8a5ec0ba CL |
2326 | |
2327 | /* | |
2328 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2329 | * enabled. We may switch back and forth between cpus while | |
2330 | * reading from one cpu area. That does not matter as long | |
2331 | * as we end up on the original cpu again when doing the cmpxchg. | |
2332 | */ | |
9dfc6e68 | 2333 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2334 | |
8a5ec0ba CL |
2335 | /* |
2336 | * The transaction ids are globally unique per cpu and per operation on | |
2337 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2338 | * occurs on the right processor and that there was no operation on the | |
2339 | * linked list in between. | |
2340 | */ | |
2341 | tid = c->tid; | |
2342 | barrier(); | |
8a5ec0ba | 2343 | |
9dfc6e68 | 2344 | object = c->freelist; |
57d437d2 | 2345 | page = c->page; |
5091b74a | 2346 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2347 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2348 | |
2349 | else { | |
0ad9500e ED |
2350 | void *next_object = get_freepointer_safe(s, object); |
2351 | ||
8a5ec0ba | 2352 | /* |
25985edc | 2353 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2354 | * operation and if we are on the right processor. |
2355 | * | |
2356 | * The cmpxchg does the following atomically (without lock semantics!) | |
2357 | * 1. Relocate first pointer to the current per cpu area. | |
2358 | * 2. Verify that tid and freelist have not been changed | |
2359 | * 3. If they were not changed replace tid and freelist | |
2360 | * | |
2361 | * Since this is without lock semantics the protection is only against | |
2362 | * code executing on this cpu *not* from access by other cpus. | |
2363 | */ | |
933393f5 | 2364 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2365 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2366 | object, tid, | |
0ad9500e | 2367 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2368 | |
2369 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2370 | goto redo; | |
2371 | } | |
0ad9500e | 2372 | prefetch_freepointer(s, next_object); |
84e554e6 | 2373 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2374 | } |
8a5ec0ba | 2375 | |
74e2134f | 2376 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2377 | memset(object, 0, s->object_size); |
d07dbea4 | 2378 | |
c016b0bd | 2379 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2380 | |
894b8788 | 2381 | return object; |
81819f0f CL |
2382 | } |
2383 | ||
2384 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2385 | { | |
2154a336 | 2386 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2387 | |
3b0efdfa | 2388 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags); |
5b882be4 EGM |
2389 | |
2390 | return ret; | |
81819f0f CL |
2391 | } |
2392 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2393 | ||
0f24f128 | 2394 | #ifdef CONFIG_TRACING |
4a92379b RK |
2395 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2396 | { | |
2397 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2398 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2399 | return ret; | |
2400 | } | |
2401 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2402 | ||
2403 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2404 | { |
4a92379b RK |
2405 | void *ret = kmalloc_order(size, flags, order); |
2406 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2407 | return ret; | |
5b882be4 | 2408 | } |
4a92379b | 2409 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2410 | #endif |
2411 | ||
81819f0f CL |
2412 | #ifdef CONFIG_NUMA |
2413 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2414 | { | |
5b882be4 EGM |
2415 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2416 | ||
ca2b84cb | 2417 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2418 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2419 | |
2420 | return ret; | |
81819f0f CL |
2421 | } |
2422 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2423 | |
0f24f128 | 2424 | #ifdef CONFIG_TRACING |
4a92379b | 2425 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2426 | gfp_t gfpflags, |
4a92379b | 2427 | int node, size_t size) |
5b882be4 | 2428 | { |
4a92379b RK |
2429 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2430 | ||
2431 | trace_kmalloc_node(_RET_IP_, ret, | |
2432 | size, s->size, gfpflags, node); | |
2433 | return ret; | |
5b882be4 | 2434 | } |
4a92379b | 2435 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2436 | #endif |
5d1f57e4 | 2437 | #endif |
5b882be4 | 2438 | |
81819f0f | 2439 | /* |
894b8788 CL |
2440 | * Slow patch handling. This may still be called frequently since objects |
2441 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2442 | * |
894b8788 CL |
2443 | * So we still attempt to reduce cache line usage. Just take the slab |
2444 | * lock and free the item. If there is no additional partial page | |
2445 | * handling required then we can return immediately. | |
81819f0f | 2446 | */ |
894b8788 | 2447 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2448 | void *x, unsigned long addr) |
81819f0f CL |
2449 | { |
2450 | void *prior; | |
2451 | void **object = (void *)x; | |
2cfb7455 CL |
2452 | int was_frozen; |
2453 | int inuse; | |
2454 | struct page new; | |
2455 | unsigned long counters; | |
2456 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2457 | unsigned long uninitialized_var(flags); |
81819f0f | 2458 | |
8a5ec0ba | 2459 | stat(s, FREE_SLOWPATH); |
81819f0f | 2460 | |
19c7ff9e CL |
2461 | if (kmem_cache_debug(s) && |
2462 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2463 | return; |
6446faa2 | 2464 | |
2cfb7455 CL |
2465 | do { |
2466 | prior = page->freelist; | |
2467 | counters = page->counters; | |
2468 | set_freepointer(s, object, prior); | |
2469 | new.counters = counters; | |
2470 | was_frozen = new.frozen; | |
2471 | new.inuse--; | |
2472 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
49e22585 CL |
2473 | |
2474 | if (!kmem_cache_debug(s) && !prior) | |
2475 | ||
2476 | /* | |
2477 | * Slab was on no list before and will be partially empty | |
2478 | * We can defer the list move and instead freeze it. | |
2479 | */ | |
2480 | new.frozen = 1; | |
2481 | ||
2482 | else { /* Needs to be taken off a list */ | |
2483 | ||
2484 | n = get_node(s, page_to_nid(page)); | |
2485 | /* | |
2486 | * Speculatively acquire the list_lock. | |
2487 | * If the cmpxchg does not succeed then we may | |
2488 | * drop the list_lock without any processing. | |
2489 | * | |
2490 | * Otherwise the list_lock will synchronize with | |
2491 | * other processors updating the list of slabs. | |
2492 | */ | |
2493 | spin_lock_irqsave(&n->list_lock, flags); | |
2494 | ||
2495 | } | |
2cfb7455 CL |
2496 | } |
2497 | inuse = new.inuse; | |
81819f0f | 2498 | |
2cfb7455 CL |
2499 | } while (!cmpxchg_double_slab(s, page, |
2500 | prior, counters, | |
2501 | object, new.counters, | |
2502 | "__slab_free")); | |
81819f0f | 2503 | |
2cfb7455 | 2504 | if (likely(!n)) { |
49e22585 CL |
2505 | |
2506 | /* | |
2507 | * If we just froze the page then put it onto the | |
2508 | * per cpu partial list. | |
2509 | */ | |
8028dcea | 2510 | if (new.frozen && !was_frozen) { |
49e22585 | 2511 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2512 | stat(s, CPU_PARTIAL_FREE); |
2513 | } | |
49e22585 | 2514 | /* |
2cfb7455 CL |
2515 | * The list lock was not taken therefore no list |
2516 | * activity can be necessary. | |
2517 | */ | |
2518 | if (was_frozen) | |
2519 | stat(s, FREE_FROZEN); | |
80f08c19 | 2520 | return; |
2cfb7455 | 2521 | } |
81819f0f CL |
2522 | |
2523 | /* | |
2cfb7455 CL |
2524 | * was_frozen may have been set after we acquired the list_lock in |
2525 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2526 | */ |
2cfb7455 CL |
2527 | if (was_frozen) |
2528 | stat(s, FREE_FROZEN); | |
2529 | else { | |
2530 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2531 | goto slab_empty; | |
81819f0f | 2532 | |
2cfb7455 CL |
2533 | /* |
2534 | * Objects left in the slab. If it was not on the partial list before | |
2535 | * then add it. | |
2536 | */ | |
2537 | if (unlikely(!prior)) { | |
2538 | remove_full(s, page); | |
136333d1 | 2539 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2cfb7455 CL |
2540 | stat(s, FREE_ADD_PARTIAL); |
2541 | } | |
8ff12cfc | 2542 | } |
80f08c19 | 2543 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2544 | return; |
2545 | ||
2546 | slab_empty: | |
a973e9dd | 2547 | if (prior) { |
81819f0f | 2548 | /* |
6fbabb20 | 2549 | * Slab on the partial list. |
81819f0f | 2550 | */ |
5cc6eee8 | 2551 | remove_partial(n, page); |
84e554e6 | 2552 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2553 | } else |
2554 | /* Slab must be on the full list */ | |
2555 | remove_full(s, page); | |
2cfb7455 | 2556 | |
80f08c19 | 2557 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2558 | stat(s, FREE_SLAB); |
81819f0f | 2559 | discard_slab(s, page); |
81819f0f CL |
2560 | } |
2561 | ||
894b8788 CL |
2562 | /* |
2563 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2564 | * can perform fastpath freeing without additional function calls. | |
2565 | * | |
2566 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2567 | * of this processor. This typically the case if we have just allocated | |
2568 | * the item before. | |
2569 | * | |
2570 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2571 | * with all sorts of special processing. | |
2572 | */ | |
06428780 | 2573 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2574 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2575 | { |
2576 | void **object = (void *)x; | |
dfb4f096 | 2577 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2578 | unsigned long tid; |
1f84260c | 2579 | |
c016b0bd CL |
2580 | slab_free_hook(s, x); |
2581 | ||
8a5ec0ba CL |
2582 | redo: |
2583 | /* | |
2584 | * Determine the currently cpus per cpu slab. | |
2585 | * The cpu may change afterward. However that does not matter since | |
2586 | * data is retrieved via this pointer. If we are on the same cpu | |
2587 | * during the cmpxchg then the free will succedd. | |
2588 | */ | |
9dfc6e68 | 2589 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2590 | |
8a5ec0ba CL |
2591 | tid = c->tid; |
2592 | barrier(); | |
c016b0bd | 2593 | |
442b06bc | 2594 | if (likely(page == c->page)) { |
ff12059e | 2595 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2596 | |
933393f5 | 2597 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2598 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2599 | c->freelist, tid, | |
2600 | object, next_tid(tid)))) { | |
2601 | ||
2602 | note_cmpxchg_failure("slab_free", s, tid); | |
2603 | goto redo; | |
2604 | } | |
84e554e6 | 2605 | stat(s, FREE_FASTPATH); |
894b8788 | 2606 | } else |
ff12059e | 2607 | __slab_free(s, page, x, addr); |
894b8788 | 2608 | |
894b8788 CL |
2609 | } |
2610 | ||
81819f0f CL |
2611 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2612 | { | |
77c5e2d0 | 2613 | struct page *page; |
81819f0f | 2614 | |
b49af68f | 2615 | page = virt_to_head_page(x); |
81819f0f | 2616 | |
ce71e27c | 2617 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2618 | |
ca2b84cb | 2619 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2620 | } |
2621 | EXPORT_SYMBOL(kmem_cache_free); | |
2622 | ||
81819f0f | 2623 | /* |
672bba3a CL |
2624 | * Object placement in a slab is made very easy because we always start at |
2625 | * offset 0. If we tune the size of the object to the alignment then we can | |
2626 | * get the required alignment by putting one properly sized object after | |
2627 | * another. | |
81819f0f CL |
2628 | * |
2629 | * Notice that the allocation order determines the sizes of the per cpu | |
2630 | * caches. Each processor has always one slab available for allocations. | |
2631 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2632 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2633 | * locking overhead. |
81819f0f CL |
2634 | */ |
2635 | ||
2636 | /* | |
2637 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2638 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2639 | * and increases the number of allocations possible without having to | |
2640 | * take the list_lock. | |
2641 | */ | |
2642 | static int slub_min_order; | |
114e9e89 | 2643 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2644 | static int slub_min_objects; |
81819f0f CL |
2645 | |
2646 | /* | |
2647 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2648 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2649 | */ |
2650 | static int slub_nomerge; | |
2651 | ||
81819f0f CL |
2652 | /* |
2653 | * Calculate the order of allocation given an slab object size. | |
2654 | * | |
672bba3a CL |
2655 | * The order of allocation has significant impact on performance and other |
2656 | * system components. Generally order 0 allocations should be preferred since | |
2657 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2658 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2659 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2660 | * would be wasted. |
2661 | * | |
2662 | * In order to reach satisfactory performance we must ensure that a minimum | |
2663 | * number of objects is in one slab. Otherwise we may generate too much | |
2664 | * activity on the partial lists which requires taking the list_lock. This is | |
2665 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2666 | * |
672bba3a CL |
2667 | * slub_max_order specifies the order where we begin to stop considering the |
2668 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2669 | * we try to keep the page order as low as possible. So we accept more waste | |
2670 | * of space in favor of a small page order. | |
81819f0f | 2671 | * |
672bba3a CL |
2672 | * Higher order allocations also allow the placement of more objects in a |
2673 | * slab and thereby reduce object handling overhead. If the user has | |
2674 | * requested a higher mininum order then we start with that one instead of | |
2675 | * the smallest order which will fit the object. | |
81819f0f | 2676 | */ |
5e6d444e | 2677 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2678 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2679 | { |
2680 | int order; | |
2681 | int rem; | |
6300ea75 | 2682 | int min_order = slub_min_order; |
81819f0f | 2683 | |
ab9a0f19 | 2684 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2685 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2686 | |
6300ea75 | 2687 | for (order = max(min_order, |
5e6d444e CL |
2688 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2689 | order <= max_order; order++) { | |
81819f0f | 2690 | |
5e6d444e | 2691 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2692 | |
ab9a0f19 | 2693 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2694 | continue; |
2695 | ||
ab9a0f19 | 2696 | rem = (slab_size - reserved) % size; |
81819f0f | 2697 | |
5e6d444e | 2698 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2699 | break; |
2700 | ||
2701 | } | |
672bba3a | 2702 | |
81819f0f CL |
2703 | return order; |
2704 | } | |
2705 | ||
ab9a0f19 | 2706 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2707 | { |
2708 | int order; | |
2709 | int min_objects; | |
2710 | int fraction; | |
e8120ff1 | 2711 | int max_objects; |
5e6d444e CL |
2712 | |
2713 | /* | |
2714 | * Attempt to find best configuration for a slab. This | |
2715 | * works by first attempting to generate a layout with | |
2716 | * the best configuration and backing off gradually. | |
2717 | * | |
2718 | * First we reduce the acceptable waste in a slab. Then | |
2719 | * we reduce the minimum objects required in a slab. | |
2720 | */ | |
2721 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2722 | if (!min_objects) |
2723 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2724 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2725 | min_objects = min(min_objects, max_objects); |
2726 | ||
5e6d444e | 2727 | while (min_objects > 1) { |
c124f5b5 | 2728 | fraction = 16; |
5e6d444e CL |
2729 | while (fraction >= 4) { |
2730 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2731 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2732 | if (order <= slub_max_order) |
2733 | return order; | |
2734 | fraction /= 2; | |
2735 | } | |
5086c389 | 2736 | min_objects--; |
5e6d444e CL |
2737 | } |
2738 | ||
2739 | /* | |
2740 | * We were unable to place multiple objects in a slab. Now | |
2741 | * lets see if we can place a single object there. | |
2742 | */ | |
ab9a0f19 | 2743 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2744 | if (order <= slub_max_order) |
2745 | return order; | |
2746 | ||
2747 | /* | |
2748 | * Doh this slab cannot be placed using slub_max_order. | |
2749 | */ | |
ab9a0f19 | 2750 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2751 | if (order < MAX_ORDER) |
5e6d444e CL |
2752 | return order; |
2753 | return -ENOSYS; | |
2754 | } | |
2755 | ||
81819f0f | 2756 | /* |
672bba3a | 2757 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2758 | */ |
2759 | static unsigned long calculate_alignment(unsigned long flags, | |
2760 | unsigned long align, unsigned long size) | |
2761 | { | |
2762 | /* | |
6446faa2 CL |
2763 | * If the user wants hardware cache aligned objects then follow that |
2764 | * suggestion if the object is sufficiently large. | |
81819f0f | 2765 | * |
6446faa2 CL |
2766 | * The hardware cache alignment cannot override the specified |
2767 | * alignment though. If that is greater then use it. | |
81819f0f | 2768 | */ |
b6210386 NP |
2769 | if (flags & SLAB_HWCACHE_ALIGN) { |
2770 | unsigned long ralign = cache_line_size(); | |
2771 | while (size <= ralign / 2) | |
2772 | ralign /= 2; | |
2773 | align = max(align, ralign); | |
2774 | } | |
81819f0f CL |
2775 | |
2776 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2777 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2778 | |
2779 | return ALIGN(align, sizeof(void *)); | |
2780 | } | |
2781 | ||
5595cffc | 2782 | static void |
4053497d | 2783 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2784 | { |
2785 | n->nr_partial = 0; | |
81819f0f CL |
2786 | spin_lock_init(&n->list_lock); |
2787 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2788 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2789 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2790 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2791 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2792 | #endif |
81819f0f CL |
2793 | } |
2794 | ||
55136592 | 2795 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2796 | { |
6c182dc0 CL |
2797 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2798 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2799 | |
8a5ec0ba | 2800 | /* |
d4d84fef CM |
2801 | * Must align to double word boundary for the double cmpxchg |
2802 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2803 | */ |
d4d84fef CM |
2804 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2805 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2806 | |
2807 | if (!s->cpu_slab) | |
2808 | return 0; | |
2809 | ||
2810 | init_kmem_cache_cpus(s); | |
4c93c355 | 2811 | |
8a5ec0ba | 2812 | return 1; |
4c93c355 | 2813 | } |
4c93c355 | 2814 | |
51df1142 CL |
2815 | static struct kmem_cache *kmem_cache_node; |
2816 | ||
81819f0f CL |
2817 | /* |
2818 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2819 | * slab on the node for this slabcache. There are no concurrent accesses | |
2820 | * possible. | |
2821 | * | |
2822 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2823 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2824 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2825 | */ |
55136592 | 2826 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2827 | { |
2828 | struct page *page; | |
2829 | struct kmem_cache_node *n; | |
2830 | ||
51df1142 | 2831 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2832 | |
51df1142 | 2833 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2834 | |
2835 | BUG_ON(!page); | |
a2f92ee7 CL |
2836 | if (page_to_nid(page) != node) { |
2837 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2838 | "node %d\n", node); | |
2839 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2840 | "in order to be able to continue\n"); | |
2841 | } | |
2842 | ||
81819f0f CL |
2843 | n = page->freelist; |
2844 | BUG_ON(!n); | |
51df1142 | 2845 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2846 | page->inuse = 1; |
8cb0a506 | 2847 | page->frozen = 0; |
51df1142 | 2848 | kmem_cache_node->node[node] = n; |
8ab1372f | 2849 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2850 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2851 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2852 | #endif |
4053497d | 2853 | init_kmem_cache_node(n); |
51df1142 | 2854 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2855 | |
136333d1 | 2856 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2857 | } |
2858 | ||
2859 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2860 | { | |
2861 | int node; | |
2862 | ||
f64dc58c | 2863 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2864 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2865 | |
73367bd8 | 2866 | if (n) |
51df1142 CL |
2867 | kmem_cache_free(kmem_cache_node, n); |
2868 | ||
81819f0f CL |
2869 | s->node[node] = NULL; |
2870 | } | |
2871 | } | |
2872 | ||
55136592 | 2873 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2874 | { |
2875 | int node; | |
81819f0f | 2876 | |
f64dc58c | 2877 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2878 | struct kmem_cache_node *n; |
2879 | ||
73367bd8 | 2880 | if (slab_state == DOWN) { |
55136592 | 2881 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2882 | continue; |
2883 | } | |
51df1142 | 2884 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2885 | GFP_KERNEL, node); |
81819f0f | 2886 | |
73367bd8 AD |
2887 | if (!n) { |
2888 | free_kmem_cache_nodes(s); | |
2889 | return 0; | |
81819f0f | 2890 | } |
73367bd8 | 2891 | |
81819f0f | 2892 | s->node[node] = n; |
4053497d | 2893 | init_kmem_cache_node(n); |
81819f0f CL |
2894 | } |
2895 | return 1; | |
2896 | } | |
81819f0f | 2897 | |
c0bdb232 | 2898 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2899 | { |
2900 | if (min < MIN_PARTIAL) | |
2901 | min = MIN_PARTIAL; | |
2902 | else if (min > MAX_PARTIAL) | |
2903 | min = MAX_PARTIAL; | |
2904 | s->min_partial = min; | |
2905 | } | |
2906 | ||
81819f0f CL |
2907 | /* |
2908 | * calculate_sizes() determines the order and the distribution of data within | |
2909 | * a slab object. | |
2910 | */ | |
06b285dc | 2911 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2912 | { |
2913 | unsigned long flags = s->flags; | |
3b0efdfa | 2914 | unsigned long size = s->object_size; |
81819f0f | 2915 | unsigned long align = s->align; |
834f3d11 | 2916 | int order; |
81819f0f | 2917 | |
d8b42bf5 CL |
2918 | /* |
2919 | * Round up object size to the next word boundary. We can only | |
2920 | * place the free pointer at word boundaries and this determines | |
2921 | * the possible location of the free pointer. | |
2922 | */ | |
2923 | size = ALIGN(size, sizeof(void *)); | |
2924 | ||
2925 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2926 | /* |
2927 | * Determine if we can poison the object itself. If the user of | |
2928 | * the slab may touch the object after free or before allocation | |
2929 | * then we should never poison the object itself. | |
2930 | */ | |
2931 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2932 | !s->ctor) |
81819f0f CL |
2933 | s->flags |= __OBJECT_POISON; |
2934 | else | |
2935 | s->flags &= ~__OBJECT_POISON; | |
2936 | ||
81819f0f CL |
2937 | |
2938 | /* | |
672bba3a | 2939 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2940 | * end of the object and the free pointer. If not then add an |
672bba3a | 2941 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2942 | */ |
3b0efdfa | 2943 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2944 | size += sizeof(void *); |
41ecc55b | 2945 | #endif |
81819f0f CL |
2946 | |
2947 | /* | |
672bba3a CL |
2948 | * With that we have determined the number of bytes in actual use |
2949 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2950 | */ |
2951 | s->inuse = size; | |
2952 | ||
2953 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2954 | s->ctor)) { |
81819f0f CL |
2955 | /* |
2956 | * Relocate free pointer after the object if it is not | |
2957 | * permitted to overwrite the first word of the object on | |
2958 | * kmem_cache_free. | |
2959 | * | |
2960 | * This is the case if we do RCU, have a constructor or | |
2961 | * destructor or are poisoning the objects. | |
2962 | */ | |
2963 | s->offset = size; | |
2964 | size += sizeof(void *); | |
2965 | } | |
2966 | ||
c12b3c62 | 2967 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2968 | if (flags & SLAB_STORE_USER) |
2969 | /* | |
2970 | * Need to store information about allocs and frees after | |
2971 | * the object. | |
2972 | */ | |
2973 | size += 2 * sizeof(struct track); | |
2974 | ||
be7b3fbc | 2975 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2976 | /* |
2977 | * Add some empty padding so that we can catch | |
2978 | * overwrites from earlier objects rather than let | |
2979 | * tracking information or the free pointer be | |
0211a9c8 | 2980 | * corrupted if a user writes before the start |
81819f0f CL |
2981 | * of the object. |
2982 | */ | |
2983 | size += sizeof(void *); | |
41ecc55b | 2984 | #endif |
672bba3a | 2985 | |
81819f0f CL |
2986 | /* |
2987 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2988 | * user specified and the dynamic determination of cache line size |
2989 | * on bootup. | |
81819f0f | 2990 | */ |
3b0efdfa | 2991 | align = calculate_alignment(flags, align, s->object_size); |
dcb0ce1b | 2992 | s->align = align; |
81819f0f CL |
2993 | |
2994 | /* | |
2995 | * SLUB stores one object immediately after another beginning from | |
2996 | * offset 0. In order to align the objects we have to simply size | |
2997 | * each object to conform to the alignment. | |
2998 | */ | |
2999 | size = ALIGN(size, align); | |
3000 | s->size = size; | |
06b285dc CL |
3001 | if (forced_order >= 0) |
3002 | order = forced_order; | |
3003 | else | |
ab9a0f19 | 3004 | order = calculate_order(size, s->reserved); |
81819f0f | 3005 | |
834f3d11 | 3006 | if (order < 0) |
81819f0f CL |
3007 | return 0; |
3008 | ||
b7a49f0d | 3009 | s->allocflags = 0; |
834f3d11 | 3010 | if (order) |
b7a49f0d CL |
3011 | s->allocflags |= __GFP_COMP; |
3012 | ||
3013 | if (s->flags & SLAB_CACHE_DMA) | |
3014 | s->allocflags |= SLUB_DMA; | |
3015 | ||
3016 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3017 | s->allocflags |= __GFP_RECLAIMABLE; | |
3018 | ||
81819f0f CL |
3019 | /* |
3020 | * Determine the number of objects per slab | |
3021 | */ | |
ab9a0f19 LJ |
3022 | s->oo = oo_make(order, size, s->reserved); |
3023 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3024 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3025 | s->max = s->oo; | |
81819f0f | 3026 | |
834f3d11 | 3027 | return !!oo_objects(s->oo); |
81819f0f CL |
3028 | |
3029 | } | |
3030 | ||
55136592 | 3031 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
3032 | const char *name, size_t size, |
3033 | size_t align, unsigned long flags, | |
51cc5068 | 3034 | void (*ctor)(void *)) |
81819f0f CL |
3035 | { |
3036 | memset(s, 0, kmem_size); | |
3037 | s->name = name; | |
3038 | s->ctor = ctor; | |
3b0efdfa | 3039 | s->object_size = size; |
81819f0f | 3040 | s->align = align; |
ba0268a8 | 3041 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 3042 | s->reserved = 0; |
81819f0f | 3043 | |
da9a638c LJ |
3044 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3045 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3046 | |
06b285dc | 3047 | if (!calculate_sizes(s, -1)) |
81819f0f | 3048 | goto error; |
3de47213 DR |
3049 | if (disable_higher_order_debug) { |
3050 | /* | |
3051 | * Disable debugging flags that store metadata if the min slab | |
3052 | * order increased. | |
3053 | */ | |
3b0efdfa | 3054 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3055 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3056 | s->offset = 0; | |
3057 | if (!calculate_sizes(s, -1)) | |
3058 | goto error; | |
3059 | } | |
3060 | } | |
81819f0f | 3061 | |
2565409f HC |
3062 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3063 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3064 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3065 | /* Enable fast mode */ | |
3066 | s->flags |= __CMPXCHG_DOUBLE; | |
3067 | #endif | |
3068 | ||
3b89d7d8 DR |
3069 | /* |
3070 | * The larger the object size is, the more pages we want on the partial | |
3071 | * list to avoid pounding the page allocator excessively. | |
3072 | */ | |
49e22585 CL |
3073 | set_min_partial(s, ilog2(s->size) / 2); |
3074 | ||
3075 | /* | |
3076 | * cpu_partial determined the maximum number of objects kept in the | |
3077 | * per cpu partial lists of a processor. | |
3078 | * | |
3079 | * Per cpu partial lists mainly contain slabs that just have one | |
3080 | * object freed. If they are used for allocation then they can be | |
3081 | * filled up again with minimal effort. The slab will never hit the | |
3082 | * per node partial lists and therefore no locking will be required. | |
3083 | * | |
3084 | * This setting also determines | |
3085 | * | |
3086 | * A) The number of objects from per cpu partial slabs dumped to the | |
3087 | * per node list when we reach the limit. | |
9f264904 | 3088 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3089 | * per node list when we run out of per cpu objects. We only fetch 50% |
3090 | * to keep some capacity around for frees. | |
3091 | */ | |
8f1e33da CL |
3092 | if (kmem_cache_debug(s)) |
3093 | s->cpu_partial = 0; | |
3094 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3095 | s->cpu_partial = 2; |
3096 | else if (s->size >= 1024) | |
3097 | s->cpu_partial = 6; | |
3098 | else if (s->size >= 256) | |
3099 | s->cpu_partial = 13; | |
3100 | else | |
3101 | s->cpu_partial = 30; | |
3102 | ||
81819f0f CL |
3103 | s->refcount = 1; |
3104 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 3105 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3106 | #endif |
55136592 | 3107 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3108 | goto error; |
81819f0f | 3109 | |
55136592 | 3110 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 3111 | return 1; |
ff12059e | 3112 | |
4c93c355 | 3113 | free_kmem_cache_nodes(s); |
81819f0f CL |
3114 | error: |
3115 | if (flags & SLAB_PANIC) | |
3116 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3117 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 3118 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
3119 | s->offset, flags); |
3120 | return 0; | |
3121 | } | |
81819f0f | 3122 | |
81819f0f CL |
3123 | /* |
3124 | * Determine the size of a slab object | |
3125 | */ | |
3126 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
3127 | { | |
3b0efdfa | 3128 | return s->object_size; |
81819f0f CL |
3129 | } |
3130 | EXPORT_SYMBOL(kmem_cache_size); | |
3131 | ||
33b12c38 CL |
3132 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3133 | const char *text) | |
3134 | { | |
3135 | #ifdef CONFIG_SLUB_DEBUG | |
3136 | void *addr = page_address(page); | |
3137 | void *p; | |
a5dd5c11 NK |
3138 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3139 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3140 | if (!map) |
3141 | return; | |
33b12c38 CL |
3142 | slab_err(s, page, "%s", text); |
3143 | slab_lock(page); | |
33b12c38 | 3144 | |
5f80b13a | 3145 | get_map(s, page, map); |
33b12c38 CL |
3146 | for_each_object(p, s, addr, page->objects) { |
3147 | ||
3148 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3149 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3150 | p, p - addr); | |
3151 | print_tracking(s, p); | |
3152 | } | |
3153 | } | |
3154 | slab_unlock(page); | |
bbd7d57b | 3155 | kfree(map); |
33b12c38 CL |
3156 | #endif |
3157 | } | |
3158 | ||
81819f0f | 3159 | /* |
599870b1 | 3160 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3161 | * This is called from kmem_cache_close(). We must be the last thread |
3162 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3163 | */ |
599870b1 | 3164 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3165 | { |
81819f0f CL |
3166 | struct page *page, *h; |
3167 | ||
33b12c38 | 3168 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3169 | if (!page->inuse) { |
5cc6eee8 | 3170 | remove_partial(n, page); |
81819f0f | 3171 | discard_slab(s, page); |
33b12c38 CL |
3172 | } else { |
3173 | list_slab_objects(s, page, | |
3174 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 3175 | } |
33b12c38 | 3176 | } |
81819f0f CL |
3177 | } |
3178 | ||
3179 | /* | |
672bba3a | 3180 | * Release all resources used by a slab cache. |
81819f0f | 3181 | */ |
0c710013 | 3182 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3183 | { |
3184 | int node; | |
3185 | ||
3186 | flush_all(s); | |
9dfc6e68 | 3187 | free_percpu(s->cpu_slab); |
81819f0f | 3188 | /* Attempt to free all objects */ |
f64dc58c | 3189 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3190 | struct kmem_cache_node *n = get_node(s, node); |
3191 | ||
599870b1 CL |
3192 | free_partial(s, n); |
3193 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3194 | return 1; |
3195 | } | |
3196 | free_kmem_cache_nodes(s); | |
3197 | return 0; | |
3198 | } | |
3199 | ||
3200 | /* | |
3201 | * Close a cache and release the kmem_cache structure | |
3202 | * (must be used for caches created using kmem_cache_create) | |
3203 | */ | |
3204 | void kmem_cache_destroy(struct kmem_cache *s) | |
3205 | { | |
18004c5d | 3206 | mutex_lock(&slab_mutex); |
81819f0f CL |
3207 | s->refcount--; |
3208 | if (!s->refcount) { | |
3209 | list_del(&s->list); | |
18004c5d | 3210 | mutex_unlock(&slab_mutex); |
d629d819 PE |
3211 | if (kmem_cache_close(s)) { |
3212 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3213 | "still has objects.\n", s->name, __func__); | |
3214 | dump_stack(); | |
3215 | } | |
d76b1590 ED |
3216 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3217 | rcu_barrier(); | |
81819f0f | 3218 | sysfs_slab_remove(s); |
69cb8e6b | 3219 | } else |
18004c5d | 3220 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3221 | } |
3222 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3223 | ||
3224 | /******************************************************************** | |
3225 | * Kmalloc subsystem | |
3226 | *******************************************************************/ | |
3227 | ||
51df1142 | 3228 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3229 | EXPORT_SYMBOL(kmalloc_caches); |
3230 | ||
51df1142 CL |
3231 | static struct kmem_cache *kmem_cache; |
3232 | ||
55136592 | 3233 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3234 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3235 | #endif |
3236 | ||
81819f0f CL |
3237 | static int __init setup_slub_min_order(char *str) |
3238 | { | |
06428780 | 3239 | get_option(&str, &slub_min_order); |
81819f0f CL |
3240 | |
3241 | return 1; | |
3242 | } | |
3243 | ||
3244 | __setup("slub_min_order=", setup_slub_min_order); | |
3245 | ||
3246 | static int __init setup_slub_max_order(char *str) | |
3247 | { | |
06428780 | 3248 | get_option(&str, &slub_max_order); |
818cf590 | 3249 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3250 | |
3251 | return 1; | |
3252 | } | |
3253 | ||
3254 | __setup("slub_max_order=", setup_slub_max_order); | |
3255 | ||
3256 | static int __init setup_slub_min_objects(char *str) | |
3257 | { | |
06428780 | 3258 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3259 | |
3260 | return 1; | |
3261 | } | |
3262 | ||
3263 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3264 | ||
3265 | static int __init setup_slub_nomerge(char *str) | |
3266 | { | |
3267 | slub_nomerge = 1; | |
3268 | return 1; | |
3269 | } | |
3270 | ||
3271 | __setup("slub_nomerge", setup_slub_nomerge); | |
3272 | ||
51df1142 CL |
3273 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3274 | int size, unsigned int flags) | |
81819f0f | 3275 | { |
51df1142 CL |
3276 | struct kmem_cache *s; |
3277 | ||
3278 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3279 | ||
83b519e8 PE |
3280 | /* |
3281 | * This function is called with IRQs disabled during early-boot on | |
18004c5d | 3282 | * single CPU so there's no need to take slab_mutex here. |
83b519e8 | 3283 | */ |
55136592 | 3284 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3285 | flags, NULL)) |
81819f0f CL |
3286 | goto panic; |
3287 | ||
3288 | list_add(&s->list, &slab_caches); | |
51df1142 | 3289 | return s; |
81819f0f CL |
3290 | |
3291 | panic: | |
3292 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3293 | return NULL; |
81819f0f CL |
3294 | } |
3295 | ||
f1b26339 CL |
3296 | /* |
3297 | * Conversion table for small slabs sizes / 8 to the index in the | |
3298 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3299 | * of two cache sizes there. The size of larger slabs can be determined using | |
3300 | * fls. | |
3301 | */ | |
3302 | static s8 size_index[24] = { | |
3303 | 3, /* 8 */ | |
3304 | 4, /* 16 */ | |
3305 | 5, /* 24 */ | |
3306 | 5, /* 32 */ | |
3307 | 6, /* 40 */ | |
3308 | 6, /* 48 */ | |
3309 | 6, /* 56 */ | |
3310 | 6, /* 64 */ | |
3311 | 1, /* 72 */ | |
3312 | 1, /* 80 */ | |
3313 | 1, /* 88 */ | |
3314 | 1, /* 96 */ | |
3315 | 7, /* 104 */ | |
3316 | 7, /* 112 */ | |
3317 | 7, /* 120 */ | |
3318 | 7, /* 128 */ | |
3319 | 2, /* 136 */ | |
3320 | 2, /* 144 */ | |
3321 | 2, /* 152 */ | |
3322 | 2, /* 160 */ | |
3323 | 2, /* 168 */ | |
3324 | 2, /* 176 */ | |
3325 | 2, /* 184 */ | |
3326 | 2 /* 192 */ | |
3327 | }; | |
3328 | ||
acdfcd04 AK |
3329 | static inline int size_index_elem(size_t bytes) |
3330 | { | |
3331 | return (bytes - 1) / 8; | |
3332 | } | |
3333 | ||
81819f0f CL |
3334 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3335 | { | |
f1b26339 | 3336 | int index; |
81819f0f | 3337 | |
f1b26339 CL |
3338 | if (size <= 192) { |
3339 | if (!size) | |
3340 | return ZERO_SIZE_PTR; | |
81819f0f | 3341 | |
acdfcd04 | 3342 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3343 | } else |
f1b26339 | 3344 | index = fls(size - 1); |
81819f0f CL |
3345 | |
3346 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3347 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3348 | return kmalloc_dma_caches[index]; |
f1b26339 | 3349 | |
81819f0f | 3350 | #endif |
51df1142 | 3351 | return kmalloc_caches[index]; |
81819f0f CL |
3352 | } |
3353 | ||
3354 | void *__kmalloc(size_t size, gfp_t flags) | |
3355 | { | |
aadb4bc4 | 3356 | struct kmem_cache *s; |
5b882be4 | 3357 | void *ret; |
81819f0f | 3358 | |
ffadd4d0 | 3359 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3360 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3361 | |
3362 | s = get_slab(size, flags); | |
3363 | ||
3364 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3365 | return s; |
3366 | ||
2154a336 | 3367 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3368 | |
ca2b84cb | 3369 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3370 | |
3371 | return ret; | |
81819f0f CL |
3372 | } |
3373 | EXPORT_SYMBOL(__kmalloc); | |
3374 | ||
5d1f57e4 | 3375 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3376 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3377 | { | |
b1eeab67 | 3378 | struct page *page; |
e4f7c0b4 | 3379 | void *ptr = NULL; |
f619cfe1 | 3380 | |
b1eeab67 VN |
3381 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3382 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3383 | if (page) |
e4f7c0b4 CM |
3384 | ptr = page_address(page); |
3385 | ||
3386 | kmemleak_alloc(ptr, size, 1, flags); | |
3387 | return ptr; | |
f619cfe1 CL |
3388 | } |
3389 | ||
81819f0f CL |
3390 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3391 | { | |
aadb4bc4 | 3392 | struct kmem_cache *s; |
5b882be4 | 3393 | void *ret; |
81819f0f | 3394 | |
057685cf | 3395 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3396 | ret = kmalloc_large_node(size, flags, node); |
3397 | ||
ca2b84cb EGM |
3398 | trace_kmalloc_node(_RET_IP_, ret, |
3399 | size, PAGE_SIZE << get_order(size), | |
3400 | flags, node); | |
5b882be4 EGM |
3401 | |
3402 | return ret; | |
3403 | } | |
aadb4bc4 CL |
3404 | |
3405 | s = get_slab(size, flags); | |
3406 | ||
3407 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3408 | return s; |
3409 | ||
5b882be4 EGM |
3410 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3411 | ||
ca2b84cb | 3412 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3413 | |
3414 | return ret; | |
81819f0f CL |
3415 | } |
3416 | EXPORT_SYMBOL(__kmalloc_node); | |
3417 | #endif | |
3418 | ||
3419 | size_t ksize(const void *object) | |
3420 | { | |
272c1d21 | 3421 | struct page *page; |
81819f0f | 3422 | |
ef8b4520 | 3423 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3424 | return 0; |
3425 | ||
294a80a8 | 3426 | page = virt_to_head_page(object); |
294a80a8 | 3427 | |
76994412 PE |
3428 | if (unlikely(!PageSlab(page))) { |
3429 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3430 | return PAGE_SIZE << compound_order(page); |
76994412 | 3431 | } |
81819f0f | 3432 | |
b3d41885 | 3433 | return slab_ksize(page->slab); |
81819f0f | 3434 | } |
b1aabecd | 3435 | EXPORT_SYMBOL(ksize); |
81819f0f | 3436 | |
d18a90dd BG |
3437 | #ifdef CONFIG_SLUB_DEBUG |
3438 | bool verify_mem_not_deleted(const void *x) | |
3439 | { | |
3440 | struct page *page; | |
3441 | void *object = (void *)x; | |
3442 | unsigned long flags; | |
3443 | bool rv; | |
3444 | ||
3445 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3446 | return false; | |
3447 | ||
3448 | local_irq_save(flags); | |
3449 | ||
3450 | page = virt_to_head_page(x); | |
3451 | if (unlikely(!PageSlab(page))) { | |
3452 | /* maybe it was from stack? */ | |
3453 | rv = true; | |
3454 | goto out_unlock; | |
3455 | } | |
3456 | ||
3457 | slab_lock(page); | |
3458 | if (on_freelist(page->slab, page, object)) { | |
3459 | object_err(page->slab, page, object, "Object is on free-list"); | |
3460 | rv = false; | |
3461 | } else { | |
3462 | rv = true; | |
3463 | } | |
3464 | slab_unlock(page); | |
3465 | ||
3466 | out_unlock: | |
3467 | local_irq_restore(flags); | |
3468 | return rv; | |
3469 | } | |
3470 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3471 | #endif | |
3472 | ||
81819f0f CL |
3473 | void kfree(const void *x) |
3474 | { | |
81819f0f | 3475 | struct page *page; |
5bb983b0 | 3476 | void *object = (void *)x; |
81819f0f | 3477 | |
2121db74 PE |
3478 | trace_kfree(_RET_IP_, x); |
3479 | ||
2408c550 | 3480 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3481 | return; |
3482 | ||
b49af68f | 3483 | page = virt_to_head_page(x); |
aadb4bc4 | 3484 | if (unlikely(!PageSlab(page))) { |
0937502a | 3485 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3486 | kmemleak_free(x); |
d9b7f226 | 3487 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3488 | return; |
3489 | } | |
ce71e27c | 3490 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3491 | } |
3492 | EXPORT_SYMBOL(kfree); | |
3493 | ||
2086d26a | 3494 | /* |
672bba3a CL |
3495 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3496 | * the remaining slabs by the number of items in use. The slabs with the | |
3497 | * most items in use come first. New allocations will then fill those up | |
3498 | * and thus they can be removed from the partial lists. | |
3499 | * | |
3500 | * The slabs with the least items are placed last. This results in them | |
3501 | * being allocated from last increasing the chance that the last objects | |
3502 | * are freed in them. | |
2086d26a CL |
3503 | */ |
3504 | int kmem_cache_shrink(struct kmem_cache *s) | |
3505 | { | |
3506 | int node; | |
3507 | int i; | |
3508 | struct kmem_cache_node *n; | |
3509 | struct page *page; | |
3510 | struct page *t; | |
205ab99d | 3511 | int objects = oo_objects(s->max); |
2086d26a | 3512 | struct list_head *slabs_by_inuse = |
834f3d11 | 3513 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3514 | unsigned long flags; |
3515 | ||
3516 | if (!slabs_by_inuse) | |
3517 | return -ENOMEM; | |
3518 | ||
3519 | flush_all(s); | |
f64dc58c | 3520 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3521 | n = get_node(s, node); |
3522 | ||
3523 | if (!n->nr_partial) | |
3524 | continue; | |
3525 | ||
834f3d11 | 3526 | for (i = 0; i < objects; i++) |
2086d26a CL |
3527 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3528 | ||
3529 | spin_lock_irqsave(&n->list_lock, flags); | |
3530 | ||
3531 | /* | |
672bba3a | 3532 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3533 | * |
672bba3a CL |
3534 | * Note that concurrent frees may occur while we hold the |
3535 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3536 | */ |
3537 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3538 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3539 | if (!page->inuse) | |
3540 | n->nr_partial--; | |
2086d26a CL |
3541 | } |
3542 | ||
2086d26a | 3543 | /* |
672bba3a CL |
3544 | * Rebuild the partial list with the slabs filled up most |
3545 | * first and the least used slabs at the end. | |
2086d26a | 3546 | */ |
69cb8e6b | 3547 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3548 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3549 | ||
2086d26a | 3550 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3551 | |
3552 | /* Release empty slabs */ | |
3553 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3554 | discard_slab(s, page); | |
2086d26a CL |
3555 | } |
3556 | ||
3557 | kfree(slabs_by_inuse); | |
3558 | return 0; | |
3559 | } | |
3560 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3561 | ||
92a5bbc1 | 3562 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3563 | static int slab_mem_going_offline_callback(void *arg) |
3564 | { | |
3565 | struct kmem_cache *s; | |
3566 | ||
18004c5d | 3567 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3568 | list_for_each_entry(s, &slab_caches, list) |
3569 | kmem_cache_shrink(s); | |
18004c5d | 3570 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3571 | |
3572 | return 0; | |
3573 | } | |
3574 | ||
3575 | static void slab_mem_offline_callback(void *arg) | |
3576 | { | |
3577 | struct kmem_cache_node *n; | |
3578 | struct kmem_cache *s; | |
3579 | struct memory_notify *marg = arg; | |
3580 | int offline_node; | |
3581 | ||
3582 | offline_node = marg->status_change_nid; | |
3583 | ||
3584 | /* | |
3585 | * If the node still has available memory. we need kmem_cache_node | |
3586 | * for it yet. | |
3587 | */ | |
3588 | if (offline_node < 0) | |
3589 | return; | |
3590 | ||
18004c5d | 3591 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3592 | list_for_each_entry(s, &slab_caches, list) { |
3593 | n = get_node(s, offline_node); | |
3594 | if (n) { | |
3595 | /* | |
3596 | * if n->nr_slabs > 0, slabs still exist on the node | |
3597 | * that is going down. We were unable to free them, | |
c9404c9c | 3598 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3599 | * callback. So, we must fail. |
3600 | */ | |
0f389ec6 | 3601 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3602 | |
3603 | s->node[offline_node] = NULL; | |
8de66a0c | 3604 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3605 | } |
3606 | } | |
18004c5d | 3607 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3608 | } |
3609 | ||
3610 | static int slab_mem_going_online_callback(void *arg) | |
3611 | { | |
3612 | struct kmem_cache_node *n; | |
3613 | struct kmem_cache *s; | |
3614 | struct memory_notify *marg = arg; | |
3615 | int nid = marg->status_change_nid; | |
3616 | int ret = 0; | |
3617 | ||
3618 | /* | |
3619 | * If the node's memory is already available, then kmem_cache_node is | |
3620 | * already created. Nothing to do. | |
3621 | */ | |
3622 | if (nid < 0) | |
3623 | return 0; | |
3624 | ||
3625 | /* | |
0121c619 | 3626 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3627 | * allocate a kmem_cache_node structure in order to bring the node |
3628 | * online. | |
3629 | */ | |
18004c5d | 3630 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3631 | list_for_each_entry(s, &slab_caches, list) { |
3632 | /* | |
3633 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3634 | * since memory is not yet available from the node that | |
3635 | * is brought up. | |
3636 | */ | |
8de66a0c | 3637 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3638 | if (!n) { |
3639 | ret = -ENOMEM; | |
3640 | goto out; | |
3641 | } | |
4053497d | 3642 | init_kmem_cache_node(n); |
b9049e23 YG |
3643 | s->node[nid] = n; |
3644 | } | |
3645 | out: | |
18004c5d | 3646 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3647 | return ret; |
3648 | } | |
3649 | ||
3650 | static int slab_memory_callback(struct notifier_block *self, | |
3651 | unsigned long action, void *arg) | |
3652 | { | |
3653 | int ret = 0; | |
3654 | ||
3655 | switch (action) { | |
3656 | case MEM_GOING_ONLINE: | |
3657 | ret = slab_mem_going_online_callback(arg); | |
3658 | break; | |
3659 | case MEM_GOING_OFFLINE: | |
3660 | ret = slab_mem_going_offline_callback(arg); | |
3661 | break; | |
3662 | case MEM_OFFLINE: | |
3663 | case MEM_CANCEL_ONLINE: | |
3664 | slab_mem_offline_callback(arg); | |
3665 | break; | |
3666 | case MEM_ONLINE: | |
3667 | case MEM_CANCEL_OFFLINE: | |
3668 | break; | |
3669 | } | |
dc19f9db KH |
3670 | if (ret) |
3671 | ret = notifier_from_errno(ret); | |
3672 | else | |
3673 | ret = NOTIFY_OK; | |
b9049e23 YG |
3674 | return ret; |
3675 | } | |
3676 | ||
3677 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3678 | ||
81819f0f CL |
3679 | /******************************************************************** |
3680 | * Basic setup of slabs | |
3681 | *******************************************************************/ | |
3682 | ||
51df1142 CL |
3683 | /* |
3684 | * Used for early kmem_cache structures that were allocated using | |
3685 | * the page allocator | |
3686 | */ | |
3687 | ||
3688 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3689 | { | |
3690 | int node; | |
3691 | ||
3692 | list_add(&s->list, &slab_caches); | |
3693 | s->refcount = -1; | |
3694 | ||
3695 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3696 | struct kmem_cache_node *n = get_node(s, node); | |
3697 | struct page *p; | |
3698 | ||
3699 | if (n) { | |
3700 | list_for_each_entry(p, &n->partial, lru) | |
3701 | p->slab = s; | |
3702 | ||
607bf324 | 3703 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3704 | list_for_each_entry(p, &n->full, lru) |
3705 | p->slab = s; | |
3706 | #endif | |
3707 | } | |
3708 | } | |
3709 | } | |
3710 | ||
81819f0f CL |
3711 | void __init kmem_cache_init(void) |
3712 | { | |
3713 | int i; | |
4b356be0 | 3714 | int caches = 0; |
51df1142 CL |
3715 | struct kmem_cache *temp_kmem_cache; |
3716 | int order; | |
51df1142 CL |
3717 | struct kmem_cache *temp_kmem_cache_node; |
3718 | unsigned long kmalloc_size; | |
3719 | ||
fc8d8620 SG |
3720 | if (debug_guardpage_minorder()) |
3721 | slub_max_order = 0; | |
3722 | ||
51df1142 CL |
3723 | kmem_size = offsetof(struct kmem_cache, node) + |
3724 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3725 | ||
3726 | /* Allocate two kmem_caches from the page allocator */ | |
3727 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3728 | order = get_order(2 * kmalloc_size); | |
3729 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3730 | ||
81819f0f CL |
3731 | /* |
3732 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3733 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3734 | * kmem_cache_open for slab_state == DOWN. |
3735 | */ | |
51df1142 CL |
3736 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3737 | ||
3738 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3739 | sizeof(struct kmem_cache_node), | |
3740 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3741 | |
0c40ba4f | 3742 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3743 | |
3744 | /* Able to allocate the per node structures */ | |
3745 | slab_state = PARTIAL; | |
3746 | ||
51df1142 CL |
3747 | temp_kmem_cache = kmem_cache; |
3748 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3749 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3750 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3751 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3752 | |
51df1142 CL |
3753 | /* |
3754 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3755 | * kmem_cache_node is separately allocated so no need to | |
3756 | * update any list pointers. | |
3757 | */ | |
3758 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3759 | |
51df1142 CL |
3760 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3761 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3762 | ||
3763 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3764 | ||
3765 | caches++; | |
51df1142 CL |
3766 | kmem_cache_bootstrap_fixup(kmem_cache); |
3767 | caches++; | |
3768 | /* Free temporary boot structure */ | |
3769 | free_pages((unsigned long)temp_kmem_cache, order); | |
3770 | ||
3771 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3772 | |
3773 | /* | |
3774 | * Patch up the size_index table if we have strange large alignment | |
3775 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3776 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3777 | * |
3778 | * Largest permitted alignment is 256 bytes due to the way we | |
3779 | * handle the index determination for the smaller caches. | |
3780 | * | |
3781 | * Make sure that nothing crazy happens if someone starts tinkering | |
3782 | * around with ARCH_KMALLOC_MINALIGN | |
3783 | */ | |
3784 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3785 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3786 | ||
acdfcd04 AK |
3787 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3788 | int elem = size_index_elem(i); | |
3789 | if (elem >= ARRAY_SIZE(size_index)) | |
3790 | break; | |
3791 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3792 | } | |
f1b26339 | 3793 | |
acdfcd04 AK |
3794 | if (KMALLOC_MIN_SIZE == 64) { |
3795 | /* | |
3796 | * The 96 byte size cache is not used if the alignment | |
3797 | * is 64 byte. | |
3798 | */ | |
3799 | for (i = 64 + 8; i <= 96; i += 8) | |
3800 | size_index[size_index_elem(i)] = 7; | |
3801 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3802 | /* |
3803 | * The 192 byte sized cache is not used if the alignment | |
3804 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3805 | * instead. | |
3806 | */ | |
3807 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3808 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3809 | } |
3810 | ||
51df1142 CL |
3811 | /* Caches that are not of the two-to-the-power-of size */ |
3812 | if (KMALLOC_MIN_SIZE <= 32) { | |
3813 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3814 | caches++; | |
3815 | } | |
3816 | ||
3817 | if (KMALLOC_MIN_SIZE <= 64) { | |
3818 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3819 | caches++; | |
3820 | } | |
3821 | ||
3822 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3823 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3824 | caches++; | |
3825 | } | |
3826 | ||
81819f0f CL |
3827 | slab_state = UP; |
3828 | ||
3829 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3830 | if (KMALLOC_MIN_SIZE <= 32) { |
3831 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3832 | BUG_ON(!kmalloc_caches[1]->name); | |
3833 | } | |
3834 | ||
3835 | if (KMALLOC_MIN_SIZE <= 64) { | |
3836 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3837 | BUG_ON(!kmalloc_caches[2]->name); | |
3838 | } | |
3839 | ||
d7278bd7 CL |
3840 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3841 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3842 | ||
3843 | BUG_ON(!s); | |
51df1142 | 3844 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3845 | } |
81819f0f CL |
3846 | |
3847 | #ifdef CONFIG_SMP | |
3848 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3849 | #endif |
81819f0f | 3850 | |
55136592 | 3851 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3852 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3853 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3854 | |
51df1142 | 3855 | if (s && s->size) { |
55136592 | 3856 | char *name = kasprintf(GFP_NOWAIT, |
3b0efdfa | 3857 | "dma-kmalloc-%d", s->object_size); |
55136592 CL |
3858 | |
3859 | BUG_ON(!name); | |
51df1142 | 3860 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3b0efdfa | 3861 | s->object_size, SLAB_CACHE_DMA); |
55136592 CL |
3862 | } |
3863 | } | |
3864 | #endif | |
3adbefee IM |
3865 | printk(KERN_INFO |
3866 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3867 | " CPUs=%d, Nodes=%d\n", |
3868 | caches, cache_line_size(), | |
81819f0f CL |
3869 | slub_min_order, slub_max_order, slub_min_objects, |
3870 | nr_cpu_ids, nr_node_ids); | |
3871 | } | |
3872 | ||
7e85ee0c PE |
3873 | void __init kmem_cache_init_late(void) |
3874 | { | |
7e85ee0c PE |
3875 | } |
3876 | ||
81819f0f CL |
3877 | /* |
3878 | * Find a mergeable slab cache | |
3879 | */ | |
3880 | static int slab_unmergeable(struct kmem_cache *s) | |
3881 | { | |
3882 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3883 | return 1; | |
3884 | ||
c59def9f | 3885 | if (s->ctor) |
81819f0f CL |
3886 | return 1; |
3887 | ||
8ffa6875 CL |
3888 | /* |
3889 | * We may have set a slab to be unmergeable during bootstrap. | |
3890 | */ | |
3891 | if (s->refcount < 0) | |
3892 | return 1; | |
3893 | ||
81819f0f CL |
3894 | return 0; |
3895 | } | |
3896 | ||
3897 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3898 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3899 | void (*ctor)(void *)) |
81819f0f | 3900 | { |
5b95a4ac | 3901 | struct kmem_cache *s; |
81819f0f CL |
3902 | |
3903 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3904 | return NULL; | |
3905 | ||
c59def9f | 3906 | if (ctor) |
81819f0f CL |
3907 | return NULL; |
3908 | ||
3909 | size = ALIGN(size, sizeof(void *)); | |
3910 | align = calculate_alignment(flags, align, size); | |
3911 | size = ALIGN(size, align); | |
ba0268a8 | 3912 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3913 | |
5b95a4ac | 3914 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3915 | if (slab_unmergeable(s)) |
3916 | continue; | |
3917 | ||
3918 | if (size > s->size) | |
3919 | continue; | |
3920 | ||
ba0268a8 | 3921 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3922 | continue; |
3923 | /* | |
3924 | * Check if alignment is compatible. | |
3925 | * Courtesy of Adrian Drzewiecki | |
3926 | */ | |
06428780 | 3927 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3928 | continue; |
3929 | ||
3930 | if (s->size - size >= sizeof(void *)) | |
3931 | continue; | |
3932 | ||
3933 | return s; | |
3934 | } | |
3935 | return NULL; | |
3936 | } | |
3937 | ||
039363f3 | 3938 | struct kmem_cache *__kmem_cache_create(const char *name, size_t size, |
51cc5068 | 3939 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3940 | { |
3941 | struct kmem_cache *s; | |
84c1cf62 | 3942 | char *n; |
81819f0f | 3943 | |
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 | */ | |
3b0efdfa | 3951 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 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--; |
20cea968 | 3956 | return NULL; |
7b8f3b66 | 3957 | } |
a0e1d1be CL |
3958 | return s; |
3959 | } | |
6446faa2 | 3960 | |
84c1cf62 PE |
3961 | n = kstrdup(name, GFP_KERNEL); |
3962 | if (!n) | |
20cea968 | 3963 | return NULL; |
84c1cf62 | 3964 | |
a0e1d1be CL |
3965 | s = kmalloc(kmem_size, GFP_KERNEL); |
3966 | if (s) { | |
84c1cf62 | 3967 | if (kmem_cache_open(s, n, |
c59def9f | 3968 | size, align, flags, ctor)) { |
20cea968 CL |
3969 | int r; |
3970 | ||
81819f0f | 3971 | list_add(&s->list, &slab_caches); |
18004c5d | 3972 | mutex_unlock(&slab_mutex); |
20cea968 CL |
3973 | r = sysfs_slab_add(s); |
3974 | mutex_lock(&slab_mutex); | |
3975 | ||
3976 | if (!r) | |
3977 | return s; | |
3978 | ||
3979 | list_del(&s->list); | |
3980 | kmem_cache_close(s); | |
a0e1d1be CL |
3981 | } |
3982 | kfree(s); | |
81819f0f | 3983 | } |
601d39d0 | 3984 | kfree(n); |
20cea968 | 3985 | return NULL; |
81819f0f | 3986 | } |
81819f0f | 3987 | |
81819f0f | 3988 | #ifdef CONFIG_SMP |
81819f0f | 3989 | /* |
672bba3a CL |
3990 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3991 | * necessary. | |
81819f0f CL |
3992 | */ |
3993 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3994 | unsigned long action, void *hcpu) | |
3995 | { | |
3996 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3997 | struct kmem_cache *s; |
3998 | unsigned long flags; | |
81819f0f CL |
3999 | |
4000 | switch (action) { | |
4001 | case CPU_UP_CANCELED: | |
8bb78442 | 4002 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4003 | case CPU_DEAD: |
8bb78442 | 4004 | case CPU_DEAD_FROZEN: |
18004c5d | 4005 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4006 | list_for_each_entry(s, &slab_caches, list) { |
4007 | local_irq_save(flags); | |
4008 | __flush_cpu_slab(s, cpu); | |
4009 | local_irq_restore(flags); | |
4010 | } | |
18004c5d | 4011 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4012 | break; |
4013 | default: | |
4014 | break; | |
4015 | } | |
4016 | return NOTIFY_OK; | |
4017 | } | |
4018 | ||
06428780 | 4019 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 4020 | .notifier_call = slab_cpuup_callback |
06428780 | 4021 | }; |
81819f0f CL |
4022 | |
4023 | #endif | |
4024 | ||
ce71e27c | 4025 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4026 | { |
aadb4bc4 | 4027 | struct kmem_cache *s; |
94b528d0 | 4028 | void *ret; |
aadb4bc4 | 4029 | |
ffadd4d0 | 4030 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
4031 | return kmalloc_large(size, gfpflags); |
4032 | ||
aadb4bc4 | 4033 | s = get_slab(size, gfpflags); |
81819f0f | 4034 | |
2408c550 | 4035 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4036 | return s; |
81819f0f | 4037 | |
2154a336 | 4038 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 4039 | |
25985edc | 4040 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4041 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4042 | |
4043 | return ret; | |
81819f0f CL |
4044 | } |
4045 | ||
5d1f57e4 | 4046 | #ifdef CONFIG_NUMA |
81819f0f | 4047 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4048 | int node, unsigned long caller) |
81819f0f | 4049 | { |
aadb4bc4 | 4050 | struct kmem_cache *s; |
94b528d0 | 4051 | void *ret; |
aadb4bc4 | 4052 | |
d3e14aa3 XF |
4053 | if (unlikely(size > SLUB_MAX_SIZE)) { |
4054 | ret = kmalloc_large_node(size, gfpflags, node); | |
4055 | ||
4056 | trace_kmalloc_node(caller, ret, | |
4057 | size, PAGE_SIZE << get_order(size), | |
4058 | gfpflags, node); | |
4059 | ||
4060 | return ret; | |
4061 | } | |
eada35ef | 4062 | |
aadb4bc4 | 4063 | s = get_slab(size, gfpflags); |
81819f0f | 4064 | |
2408c550 | 4065 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4066 | return s; |
81819f0f | 4067 | |
94b528d0 EGM |
4068 | ret = slab_alloc(s, gfpflags, node, caller); |
4069 | ||
25985edc | 4070 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4071 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4072 | |
4073 | return ret; | |
81819f0f | 4074 | } |
5d1f57e4 | 4075 | #endif |
81819f0f | 4076 | |
ab4d5ed5 | 4077 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4078 | static int count_inuse(struct page *page) |
4079 | { | |
4080 | return page->inuse; | |
4081 | } | |
4082 | ||
4083 | static int count_total(struct page *page) | |
4084 | { | |
4085 | return page->objects; | |
4086 | } | |
ab4d5ed5 | 4087 | #endif |
205ab99d | 4088 | |
ab4d5ed5 | 4089 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4090 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4091 | unsigned long *map) | |
53e15af0 CL |
4092 | { |
4093 | void *p; | |
a973e9dd | 4094 | void *addr = page_address(page); |
53e15af0 CL |
4095 | |
4096 | if (!check_slab(s, page) || | |
4097 | !on_freelist(s, page, NULL)) | |
4098 | return 0; | |
4099 | ||
4100 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4101 | bitmap_zero(map, page->objects); |
53e15af0 | 4102 | |
5f80b13a CL |
4103 | get_map(s, page, map); |
4104 | for_each_object(p, s, addr, page->objects) { | |
4105 | if (test_bit(slab_index(p, s, addr), map)) | |
4106 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4107 | return 0; | |
53e15af0 CL |
4108 | } |
4109 | ||
224a88be | 4110 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4111 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4112 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4113 | return 0; |
4114 | return 1; | |
4115 | } | |
4116 | ||
434e245d CL |
4117 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4118 | unsigned long *map) | |
53e15af0 | 4119 | { |
881db7fb CL |
4120 | slab_lock(page); |
4121 | validate_slab(s, page, map); | |
4122 | slab_unlock(page); | |
53e15af0 CL |
4123 | } |
4124 | ||
434e245d CL |
4125 | static int validate_slab_node(struct kmem_cache *s, |
4126 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4127 | { |
4128 | unsigned long count = 0; | |
4129 | struct page *page; | |
4130 | unsigned long flags; | |
4131 | ||
4132 | spin_lock_irqsave(&n->list_lock, flags); | |
4133 | ||
4134 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4135 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4136 | count++; |
4137 | } | |
4138 | if (count != n->nr_partial) | |
4139 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4140 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4141 | ||
4142 | if (!(s->flags & SLAB_STORE_USER)) | |
4143 | goto out; | |
4144 | ||
4145 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4146 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4147 | count++; |
4148 | } | |
4149 | if (count != atomic_long_read(&n->nr_slabs)) | |
4150 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4151 | "counter=%ld\n", s->name, count, | |
4152 | atomic_long_read(&n->nr_slabs)); | |
4153 | ||
4154 | out: | |
4155 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4156 | return count; | |
4157 | } | |
4158 | ||
434e245d | 4159 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4160 | { |
4161 | int node; | |
4162 | unsigned long count = 0; | |
205ab99d | 4163 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4164 | sizeof(unsigned long), GFP_KERNEL); |
4165 | ||
4166 | if (!map) | |
4167 | return -ENOMEM; | |
53e15af0 CL |
4168 | |
4169 | flush_all(s); | |
f64dc58c | 4170 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4171 | struct kmem_cache_node *n = get_node(s, node); |
4172 | ||
434e245d | 4173 | count += validate_slab_node(s, n, map); |
53e15af0 | 4174 | } |
434e245d | 4175 | kfree(map); |
53e15af0 CL |
4176 | return count; |
4177 | } | |
88a420e4 | 4178 | /* |
672bba3a | 4179 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4180 | * and freed. |
4181 | */ | |
4182 | ||
4183 | struct location { | |
4184 | unsigned long count; | |
ce71e27c | 4185 | unsigned long addr; |
45edfa58 CL |
4186 | long long sum_time; |
4187 | long min_time; | |
4188 | long max_time; | |
4189 | long min_pid; | |
4190 | long max_pid; | |
174596a0 | 4191 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4192 | nodemask_t nodes; |
88a420e4 CL |
4193 | }; |
4194 | ||
4195 | struct loc_track { | |
4196 | unsigned long max; | |
4197 | unsigned long count; | |
4198 | struct location *loc; | |
4199 | }; | |
4200 | ||
4201 | static void free_loc_track(struct loc_track *t) | |
4202 | { | |
4203 | if (t->max) | |
4204 | free_pages((unsigned long)t->loc, | |
4205 | get_order(sizeof(struct location) * t->max)); | |
4206 | } | |
4207 | ||
68dff6a9 | 4208 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4209 | { |
4210 | struct location *l; | |
4211 | int order; | |
4212 | ||
88a420e4 CL |
4213 | order = get_order(sizeof(struct location) * max); |
4214 | ||
68dff6a9 | 4215 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4216 | if (!l) |
4217 | return 0; | |
4218 | ||
4219 | if (t->count) { | |
4220 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4221 | free_loc_track(t); | |
4222 | } | |
4223 | t->max = max; | |
4224 | t->loc = l; | |
4225 | return 1; | |
4226 | } | |
4227 | ||
4228 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4229 | const struct track *track) |
88a420e4 CL |
4230 | { |
4231 | long start, end, pos; | |
4232 | struct location *l; | |
ce71e27c | 4233 | unsigned long caddr; |
45edfa58 | 4234 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4235 | |
4236 | start = -1; | |
4237 | end = t->count; | |
4238 | ||
4239 | for ( ; ; ) { | |
4240 | pos = start + (end - start + 1) / 2; | |
4241 | ||
4242 | /* | |
4243 | * There is nothing at "end". If we end up there | |
4244 | * we need to add something to before end. | |
4245 | */ | |
4246 | if (pos == end) | |
4247 | break; | |
4248 | ||
4249 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4250 | if (track->addr == caddr) { |
4251 | ||
4252 | l = &t->loc[pos]; | |
4253 | l->count++; | |
4254 | if (track->when) { | |
4255 | l->sum_time += age; | |
4256 | if (age < l->min_time) | |
4257 | l->min_time = age; | |
4258 | if (age > l->max_time) | |
4259 | l->max_time = age; | |
4260 | ||
4261 | if (track->pid < l->min_pid) | |
4262 | l->min_pid = track->pid; | |
4263 | if (track->pid > l->max_pid) | |
4264 | l->max_pid = track->pid; | |
4265 | ||
174596a0 RR |
4266 | cpumask_set_cpu(track->cpu, |
4267 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4268 | } |
4269 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4270 | return 1; |
4271 | } | |
4272 | ||
45edfa58 | 4273 | if (track->addr < caddr) |
88a420e4 CL |
4274 | end = pos; |
4275 | else | |
4276 | start = pos; | |
4277 | } | |
4278 | ||
4279 | /* | |
672bba3a | 4280 | * Not found. Insert new tracking element. |
88a420e4 | 4281 | */ |
68dff6a9 | 4282 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4283 | return 0; |
4284 | ||
4285 | l = t->loc + pos; | |
4286 | if (pos < t->count) | |
4287 | memmove(l + 1, l, | |
4288 | (t->count - pos) * sizeof(struct location)); | |
4289 | t->count++; | |
4290 | l->count = 1; | |
45edfa58 CL |
4291 | l->addr = track->addr; |
4292 | l->sum_time = age; | |
4293 | l->min_time = age; | |
4294 | l->max_time = age; | |
4295 | l->min_pid = track->pid; | |
4296 | l->max_pid = track->pid; | |
174596a0 RR |
4297 | cpumask_clear(to_cpumask(l->cpus)); |
4298 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4299 | nodes_clear(l->nodes); |
4300 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4301 | return 1; |
4302 | } | |
4303 | ||
4304 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4305 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4306 | unsigned long *map) |
88a420e4 | 4307 | { |
a973e9dd | 4308 | void *addr = page_address(page); |
88a420e4 CL |
4309 | void *p; |
4310 | ||
39b26464 | 4311 | bitmap_zero(map, page->objects); |
5f80b13a | 4312 | get_map(s, page, map); |
88a420e4 | 4313 | |
224a88be | 4314 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4315 | if (!test_bit(slab_index(p, s, addr), map)) |
4316 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4317 | } |
4318 | ||
4319 | static int list_locations(struct kmem_cache *s, char *buf, | |
4320 | enum track_item alloc) | |
4321 | { | |
e374d483 | 4322 | int len = 0; |
88a420e4 | 4323 | unsigned long i; |
68dff6a9 | 4324 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4325 | int node; |
bbd7d57b ED |
4326 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4327 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4328 | |
bbd7d57b ED |
4329 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4330 | GFP_TEMPORARY)) { | |
4331 | kfree(map); | |
68dff6a9 | 4332 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4333 | } |
88a420e4 CL |
4334 | /* Push back cpu slabs */ |
4335 | flush_all(s); | |
4336 | ||
f64dc58c | 4337 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4338 | struct kmem_cache_node *n = get_node(s, node); |
4339 | unsigned long flags; | |
4340 | struct page *page; | |
4341 | ||
9e86943b | 4342 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4343 | continue; |
4344 | ||
4345 | spin_lock_irqsave(&n->list_lock, flags); | |
4346 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4347 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4348 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4349 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4350 | spin_unlock_irqrestore(&n->list_lock, flags); |
4351 | } | |
4352 | ||
4353 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4354 | struct location *l = &t.loc[i]; |
88a420e4 | 4355 | |
9c246247 | 4356 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4357 | break; |
e374d483 | 4358 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4359 | |
4360 | if (l->addr) | |
62c70bce | 4361 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4362 | else |
e374d483 | 4363 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4364 | |
4365 | if (l->sum_time != l->min_time) { | |
e374d483 | 4366 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4367 | l->min_time, |
4368 | (long)div_u64(l->sum_time, l->count), | |
4369 | l->max_time); | |
45edfa58 | 4370 | } else |
e374d483 | 4371 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4372 | l->min_time); |
4373 | ||
4374 | if (l->min_pid != l->max_pid) | |
e374d483 | 4375 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4376 | l->min_pid, l->max_pid); |
4377 | else | |
e374d483 | 4378 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4379 | l->min_pid); |
4380 | ||
174596a0 RR |
4381 | if (num_online_cpus() > 1 && |
4382 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4383 | len < PAGE_SIZE - 60) { |
4384 | len += sprintf(buf + len, " cpus="); | |
4385 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4386 | to_cpumask(l->cpus)); |
45edfa58 CL |
4387 | } |
4388 | ||
62bc62a8 | 4389 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4390 | len < PAGE_SIZE - 60) { |
4391 | len += sprintf(buf + len, " nodes="); | |
4392 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4393 | l->nodes); |
4394 | } | |
4395 | ||
e374d483 | 4396 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4397 | } |
4398 | ||
4399 | free_loc_track(&t); | |
bbd7d57b | 4400 | kfree(map); |
88a420e4 | 4401 | if (!t.count) |
e374d483 HH |
4402 | len += sprintf(buf, "No data\n"); |
4403 | return len; | |
88a420e4 | 4404 | } |
ab4d5ed5 | 4405 | #endif |
88a420e4 | 4406 | |
a5a84755 CL |
4407 | #ifdef SLUB_RESILIENCY_TEST |
4408 | static void resiliency_test(void) | |
4409 | { | |
4410 | u8 *p; | |
4411 | ||
4412 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4413 | ||
4414 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4415 | printk(KERN_ERR "-----------------------\n"); | |
4416 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4417 | ||
4418 | p = kzalloc(16, GFP_KERNEL); | |
4419 | p[16] = 0x12; | |
4420 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4421 | " 0x12->0x%p\n\n", p + 16); | |
4422 | ||
4423 | validate_slab_cache(kmalloc_caches[4]); | |
4424 | ||
4425 | /* Hmmm... The next two are dangerous */ | |
4426 | p = kzalloc(32, GFP_KERNEL); | |
4427 | p[32 + sizeof(void *)] = 0x34; | |
4428 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4429 | " 0x34 -> -0x%p\n", p); | |
4430 | printk(KERN_ERR | |
4431 | "If allocated object is overwritten then not detectable\n\n"); | |
4432 | ||
4433 | validate_slab_cache(kmalloc_caches[5]); | |
4434 | p = kzalloc(64, GFP_KERNEL); | |
4435 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4436 | *p = 0x56; | |
4437 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4438 | p); | |
4439 | printk(KERN_ERR | |
4440 | "If allocated object is overwritten then not detectable\n\n"); | |
4441 | validate_slab_cache(kmalloc_caches[6]); | |
4442 | ||
4443 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4444 | p = kzalloc(128, GFP_KERNEL); | |
4445 | kfree(p); | |
4446 | *p = 0x78; | |
4447 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4448 | validate_slab_cache(kmalloc_caches[7]); | |
4449 | ||
4450 | p = kzalloc(256, GFP_KERNEL); | |
4451 | kfree(p); | |
4452 | p[50] = 0x9a; | |
4453 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4454 | p); | |
4455 | validate_slab_cache(kmalloc_caches[8]); | |
4456 | ||
4457 | p = kzalloc(512, GFP_KERNEL); | |
4458 | kfree(p); | |
4459 | p[512] = 0xab; | |
4460 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4461 | validate_slab_cache(kmalloc_caches[9]); | |
4462 | } | |
4463 | #else | |
4464 | #ifdef CONFIG_SYSFS | |
4465 | static void resiliency_test(void) {}; | |
4466 | #endif | |
4467 | #endif | |
4468 | ||
ab4d5ed5 | 4469 | #ifdef CONFIG_SYSFS |
81819f0f | 4470 | enum slab_stat_type { |
205ab99d CL |
4471 | SL_ALL, /* All slabs */ |
4472 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4473 | SL_CPU, /* Only slabs used for cpu caches */ | |
4474 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4475 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4476 | }; |
4477 | ||
205ab99d | 4478 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4479 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4480 | #define SO_CPU (1 << SL_CPU) | |
4481 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4482 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4483 | |
62e5c4b4 CG |
4484 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4485 | char *buf, unsigned long flags) | |
81819f0f CL |
4486 | { |
4487 | unsigned long total = 0; | |
81819f0f CL |
4488 | int node; |
4489 | int x; | |
4490 | unsigned long *nodes; | |
4491 | unsigned long *per_cpu; | |
4492 | ||
4493 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4494 | if (!nodes) |
4495 | return -ENOMEM; | |
81819f0f CL |
4496 | per_cpu = nodes + nr_node_ids; |
4497 | ||
205ab99d CL |
4498 | if (flags & SO_CPU) { |
4499 | int cpu; | |
81819f0f | 4500 | |
205ab99d | 4501 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4502 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4503 | int node; |
49e22585 | 4504 | struct page *page; |
dfb4f096 | 4505 | |
bc6697d8 | 4506 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4507 | if (!page) |
4508 | continue; | |
205ab99d | 4509 | |
ec3ab083 CL |
4510 | node = page_to_nid(page); |
4511 | if (flags & SO_TOTAL) | |
4512 | x = page->objects; | |
4513 | else if (flags & SO_OBJECTS) | |
4514 | x = page->inuse; | |
4515 | else | |
4516 | x = 1; | |
49e22585 | 4517 | |
ec3ab083 CL |
4518 | total += x; |
4519 | nodes[node] += x; | |
4520 | ||
4521 | page = ACCESS_ONCE(c->partial); | |
49e22585 CL |
4522 | if (page) { |
4523 | x = page->pobjects; | |
bc6697d8 ED |
4524 | total += x; |
4525 | nodes[node] += x; | |
49e22585 | 4526 | } |
ec3ab083 | 4527 | |
bc6697d8 | 4528 | per_cpu[node]++; |
81819f0f CL |
4529 | } |
4530 | } | |
4531 | ||
04d94879 | 4532 | lock_memory_hotplug(); |
ab4d5ed5 | 4533 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4534 | if (flags & SO_ALL) { |
4535 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4536 | struct kmem_cache_node *n = get_node(s, node); | |
4537 | ||
4538 | if (flags & SO_TOTAL) | |
4539 | x = atomic_long_read(&n->total_objects); | |
4540 | else if (flags & SO_OBJECTS) | |
4541 | x = atomic_long_read(&n->total_objects) - | |
4542 | count_partial(n, count_free); | |
81819f0f | 4543 | |
81819f0f | 4544 | else |
205ab99d | 4545 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4546 | total += x; |
4547 | nodes[node] += x; | |
4548 | } | |
4549 | ||
ab4d5ed5 CL |
4550 | } else |
4551 | #endif | |
4552 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4553 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4554 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4555 | |
205ab99d CL |
4556 | if (flags & SO_TOTAL) |
4557 | x = count_partial(n, count_total); | |
4558 | else if (flags & SO_OBJECTS) | |
4559 | x = count_partial(n, count_inuse); | |
81819f0f | 4560 | else |
205ab99d | 4561 | x = n->nr_partial; |
81819f0f CL |
4562 | total += x; |
4563 | nodes[node] += x; | |
4564 | } | |
4565 | } | |
81819f0f CL |
4566 | x = sprintf(buf, "%lu", total); |
4567 | #ifdef CONFIG_NUMA | |
f64dc58c | 4568 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4569 | if (nodes[node]) |
4570 | x += sprintf(buf + x, " N%d=%lu", | |
4571 | node, nodes[node]); | |
4572 | #endif | |
04d94879 | 4573 | unlock_memory_hotplug(); |
81819f0f CL |
4574 | kfree(nodes); |
4575 | return x + sprintf(buf + x, "\n"); | |
4576 | } | |
4577 | ||
ab4d5ed5 | 4578 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4579 | static int any_slab_objects(struct kmem_cache *s) |
4580 | { | |
4581 | int node; | |
81819f0f | 4582 | |
dfb4f096 | 4583 | for_each_online_node(node) { |
81819f0f CL |
4584 | struct kmem_cache_node *n = get_node(s, node); |
4585 | ||
dfb4f096 CL |
4586 | if (!n) |
4587 | continue; | |
4588 | ||
4ea33e2d | 4589 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4590 | return 1; |
4591 | } | |
4592 | return 0; | |
4593 | } | |
ab4d5ed5 | 4594 | #endif |
81819f0f CL |
4595 | |
4596 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4597 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4598 | |
4599 | struct slab_attribute { | |
4600 | struct attribute attr; | |
4601 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4602 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4603 | }; | |
4604 | ||
4605 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4606 | static struct slab_attribute _name##_attr = \ |
4607 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4608 | |
4609 | #define SLAB_ATTR(_name) \ | |
4610 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4611 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4612 | |
81819f0f CL |
4613 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4614 | { | |
4615 | return sprintf(buf, "%d\n", s->size); | |
4616 | } | |
4617 | SLAB_ATTR_RO(slab_size); | |
4618 | ||
4619 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4620 | { | |
4621 | return sprintf(buf, "%d\n", s->align); | |
4622 | } | |
4623 | SLAB_ATTR_RO(align); | |
4624 | ||
4625 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4626 | { | |
3b0efdfa | 4627 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4628 | } |
4629 | SLAB_ATTR_RO(object_size); | |
4630 | ||
4631 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4632 | { | |
834f3d11 | 4633 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4634 | } |
4635 | SLAB_ATTR_RO(objs_per_slab); | |
4636 | ||
06b285dc CL |
4637 | static ssize_t order_store(struct kmem_cache *s, |
4638 | const char *buf, size_t length) | |
4639 | { | |
0121c619 CL |
4640 | unsigned long order; |
4641 | int err; | |
4642 | ||
4643 | err = strict_strtoul(buf, 10, &order); | |
4644 | if (err) | |
4645 | return err; | |
06b285dc CL |
4646 | |
4647 | if (order > slub_max_order || order < slub_min_order) | |
4648 | return -EINVAL; | |
4649 | ||
4650 | calculate_sizes(s, order); | |
4651 | return length; | |
4652 | } | |
4653 | ||
81819f0f CL |
4654 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4655 | { | |
834f3d11 | 4656 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4657 | } |
06b285dc | 4658 | SLAB_ATTR(order); |
81819f0f | 4659 | |
73d342b1 DR |
4660 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4661 | { | |
4662 | return sprintf(buf, "%lu\n", s->min_partial); | |
4663 | } | |
4664 | ||
4665 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4666 | size_t length) | |
4667 | { | |
4668 | unsigned long min; | |
4669 | int err; | |
4670 | ||
4671 | err = strict_strtoul(buf, 10, &min); | |
4672 | if (err) | |
4673 | return err; | |
4674 | ||
c0bdb232 | 4675 | set_min_partial(s, min); |
73d342b1 DR |
4676 | return length; |
4677 | } | |
4678 | SLAB_ATTR(min_partial); | |
4679 | ||
49e22585 CL |
4680 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4681 | { | |
4682 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4683 | } | |
4684 | ||
4685 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4686 | size_t length) | |
4687 | { | |
4688 | unsigned long objects; | |
4689 | int err; | |
4690 | ||
4691 | err = strict_strtoul(buf, 10, &objects); | |
4692 | if (err) | |
4693 | return err; | |
74ee4ef1 DR |
4694 | if (objects && kmem_cache_debug(s)) |
4695 | return -EINVAL; | |
49e22585 CL |
4696 | |
4697 | s->cpu_partial = objects; | |
4698 | flush_all(s); | |
4699 | return length; | |
4700 | } | |
4701 | SLAB_ATTR(cpu_partial); | |
4702 | ||
81819f0f CL |
4703 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4704 | { | |
62c70bce JP |
4705 | if (!s->ctor) |
4706 | return 0; | |
4707 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4708 | } |
4709 | SLAB_ATTR_RO(ctor); | |
4710 | ||
81819f0f CL |
4711 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4712 | { | |
4713 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4714 | } | |
4715 | SLAB_ATTR_RO(aliases); | |
4716 | ||
81819f0f CL |
4717 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4718 | { | |
d9acf4b7 | 4719 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4720 | } |
4721 | SLAB_ATTR_RO(partial); | |
4722 | ||
4723 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4724 | { | |
d9acf4b7 | 4725 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4726 | } |
4727 | SLAB_ATTR_RO(cpu_slabs); | |
4728 | ||
4729 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4730 | { | |
205ab99d | 4731 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4732 | } |
4733 | SLAB_ATTR_RO(objects); | |
4734 | ||
205ab99d CL |
4735 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4736 | { | |
4737 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4738 | } | |
4739 | SLAB_ATTR_RO(objects_partial); | |
4740 | ||
49e22585 CL |
4741 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4742 | { | |
4743 | int objects = 0; | |
4744 | int pages = 0; | |
4745 | int cpu; | |
4746 | int len; | |
4747 | ||
4748 | for_each_online_cpu(cpu) { | |
4749 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4750 | ||
4751 | if (page) { | |
4752 | pages += page->pages; | |
4753 | objects += page->pobjects; | |
4754 | } | |
4755 | } | |
4756 | ||
4757 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4758 | ||
4759 | #ifdef CONFIG_SMP | |
4760 | for_each_online_cpu(cpu) { | |
4761 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4762 | ||
4763 | if (page && len < PAGE_SIZE - 20) | |
4764 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4765 | page->pobjects, page->pages); | |
4766 | } | |
4767 | #endif | |
4768 | return len + sprintf(buf + len, "\n"); | |
4769 | } | |
4770 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4771 | ||
a5a84755 CL |
4772 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4773 | { | |
4774 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4775 | } | |
4776 | ||
4777 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4778 | const char *buf, size_t length) | |
4779 | { | |
4780 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4781 | if (buf[0] == '1') | |
4782 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4783 | return length; | |
4784 | } | |
4785 | SLAB_ATTR(reclaim_account); | |
4786 | ||
4787 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4788 | { | |
4789 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4790 | } | |
4791 | SLAB_ATTR_RO(hwcache_align); | |
4792 | ||
4793 | #ifdef CONFIG_ZONE_DMA | |
4794 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4795 | { | |
4796 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4797 | } | |
4798 | SLAB_ATTR_RO(cache_dma); | |
4799 | #endif | |
4800 | ||
4801 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4802 | { | |
4803 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4804 | } | |
4805 | SLAB_ATTR_RO(destroy_by_rcu); | |
4806 | ||
ab9a0f19 LJ |
4807 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4808 | { | |
4809 | return sprintf(buf, "%d\n", s->reserved); | |
4810 | } | |
4811 | SLAB_ATTR_RO(reserved); | |
4812 | ||
ab4d5ed5 | 4813 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4814 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4815 | { | |
4816 | return show_slab_objects(s, buf, SO_ALL); | |
4817 | } | |
4818 | SLAB_ATTR_RO(slabs); | |
4819 | ||
205ab99d CL |
4820 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4821 | { | |
4822 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4823 | } | |
4824 | SLAB_ATTR_RO(total_objects); | |
4825 | ||
81819f0f CL |
4826 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4827 | { | |
4828 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4829 | } | |
4830 | ||
4831 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4832 | const char *buf, size_t length) | |
4833 | { | |
4834 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4835 | if (buf[0] == '1') { |
4836 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4837 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4838 | } |
81819f0f CL |
4839 | return length; |
4840 | } | |
4841 | SLAB_ATTR(sanity_checks); | |
4842 | ||
4843 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4844 | { | |
4845 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4846 | } | |
4847 | ||
4848 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4849 | size_t length) | |
4850 | { | |
4851 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4852 | if (buf[0] == '1') { |
4853 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4854 | s->flags |= SLAB_TRACE; |
b789ef51 | 4855 | } |
81819f0f CL |
4856 | return length; |
4857 | } | |
4858 | SLAB_ATTR(trace); | |
4859 | ||
81819f0f CL |
4860 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4861 | { | |
4862 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4863 | } | |
4864 | ||
4865 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4866 | const char *buf, size_t length) | |
4867 | { | |
4868 | if (any_slab_objects(s)) | |
4869 | return -EBUSY; | |
4870 | ||
4871 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4872 | if (buf[0] == '1') { |
4873 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4874 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4875 | } |
06b285dc | 4876 | calculate_sizes(s, -1); |
81819f0f CL |
4877 | return length; |
4878 | } | |
4879 | SLAB_ATTR(red_zone); | |
4880 | ||
4881 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4882 | { | |
4883 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4884 | } | |
4885 | ||
4886 | static ssize_t poison_store(struct kmem_cache *s, | |
4887 | const char *buf, size_t length) | |
4888 | { | |
4889 | if (any_slab_objects(s)) | |
4890 | return -EBUSY; | |
4891 | ||
4892 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4893 | if (buf[0] == '1') { |
4894 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4895 | s->flags |= SLAB_POISON; |
b789ef51 | 4896 | } |
06b285dc | 4897 | calculate_sizes(s, -1); |
81819f0f CL |
4898 | return length; |
4899 | } | |
4900 | SLAB_ATTR(poison); | |
4901 | ||
4902 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4903 | { | |
4904 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4905 | } | |
4906 | ||
4907 | static ssize_t store_user_store(struct kmem_cache *s, | |
4908 | const char *buf, size_t length) | |
4909 | { | |
4910 | if (any_slab_objects(s)) | |
4911 | return -EBUSY; | |
4912 | ||
4913 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4914 | if (buf[0] == '1') { |
4915 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4916 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4917 | } |
06b285dc | 4918 | calculate_sizes(s, -1); |
81819f0f CL |
4919 | return length; |
4920 | } | |
4921 | SLAB_ATTR(store_user); | |
4922 | ||
53e15af0 CL |
4923 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4924 | { | |
4925 | return 0; | |
4926 | } | |
4927 | ||
4928 | static ssize_t validate_store(struct kmem_cache *s, | |
4929 | const char *buf, size_t length) | |
4930 | { | |
434e245d CL |
4931 | int ret = -EINVAL; |
4932 | ||
4933 | if (buf[0] == '1') { | |
4934 | ret = validate_slab_cache(s); | |
4935 | if (ret >= 0) | |
4936 | ret = length; | |
4937 | } | |
4938 | return ret; | |
53e15af0 CL |
4939 | } |
4940 | SLAB_ATTR(validate); | |
a5a84755 CL |
4941 | |
4942 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4943 | { | |
4944 | if (!(s->flags & SLAB_STORE_USER)) | |
4945 | return -ENOSYS; | |
4946 | return list_locations(s, buf, TRACK_ALLOC); | |
4947 | } | |
4948 | SLAB_ATTR_RO(alloc_calls); | |
4949 | ||
4950 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4951 | { | |
4952 | if (!(s->flags & SLAB_STORE_USER)) | |
4953 | return -ENOSYS; | |
4954 | return list_locations(s, buf, TRACK_FREE); | |
4955 | } | |
4956 | SLAB_ATTR_RO(free_calls); | |
4957 | #endif /* CONFIG_SLUB_DEBUG */ | |
4958 | ||
4959 | #ifdef CONFIG_FAILSLAB | |
4960 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4961 | { | |
4962 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4963 | } | |
4964 | ||
4965 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4966 | size_t length) | |
4967 | { | |
4968 | s->flags &= ~SLAB_FAILSLAB; | |
4969 | if (buf[0] == '1') | |
4970 | s->flags |= SLAB_FAILSLAB; | |
4971 | return length; | |
4972 | } | |
4973 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4974 | #endif |
53e15af0 | 4975 | |
2086d26a CL |
4976 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4977 | { | |
4978 | return 0; | |
4979 | } | |
4980 | ||
4981 | static ssize_t shrink_store(struct kmem_cache *s, | |
4982 | const char *buf, size_t length) | |
4983 | { | |
4984 | if (buf[0] == '1') { | |
4985 | int rc = kmem_cache_shrink(s); | |
4986 | ||
4987 | if (rc) | |
4988 | return rc; | |
4989 | } else | |
4990 | return -EINVAL; | |
4991 | return length; | |
4992 | } | |
4993 | SLAB_ATTR(shrink); | |
4994 | ||
81819f0f | 4995 | #ifdef CONFIG_NUMA |
9824601e | 4996 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4997 | { |
9824601e | 4998 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4999 | } |
5000 | ||
9824601e | 5001 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5002 | const char *buf, size_t length) |
5003 | { | |
0121c619 CL |
5004 | unsigned long ratio; |
5005 | int err; | |
5006 | ||
5007 | err = strict_strtoul(buf, 10, &ratio); | |
5008 | if (err) | |
5009 | return err; | |
5010 | ||
e2cb96b7 | 5011 | if (ratio <= 100) |
0121c619 | 5012 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5013 | |
81819f0f CL |
5014 | return length; |
5015 | } | |
9824601e | 5016 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5017 | #endif |
5018 | ||
8ff12cfc | 5019 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5020 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5021 | { | |
5022 | unsigned long sum = 0; | |
5023 | int cpu; | |
5024 | int len; | |
5025 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5026 | ||
5027 | if (!data) | |
5028 | return -ENOMEM; | |
5029 | ||
5030 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5031 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5032 | |
5033 | data[cpu] = x; | |
5034 | sum += x; | |
5035 | } | |
5036 | ||
5037 | len = sprintf(buf, "%lu", sum); | |
5038 | ||
50ef37b9 | 5039 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5040 | for_each_online_cpu(cpu) { |
5041 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5042 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5043 | } |
50ef37b9 | 5044 | #endif |
8ff12cfc CL |
5045 | kfree(data); |
5046 | return len + sprintf(buf + len, "\n"); | |
5047 | } | |
5048 | ||
78eb00cc DR |
5049 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5050 | { | |
5051 | int cpu; | |
5052 | ||
5053 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5054 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5055 | } |
5056 | ||
8ff12cfc CL |
5057 | #define STAT_ATTR(si, text) \ |
5058 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5059 | { \ | |
5060 | return show_stat(s, buf, si); \ | |
5061 | } \ | |
78eb00cc DR |
5062 | static ssize_t text##_store(struct kmem_cache *s, \ |
5063 | const char *buf, size_t length) \ | |
5064 | { \ | |
5065 | if (buf[0] != '0') \ | |
5066 | return -EINVAL; \ | |
5067 | clear_stat(s, si); \ | |
5068 | return length; \ | |
5069 | } \ | |
5070 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5071 | |
5072 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5073 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5074 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5075 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5076 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5077 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5078 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5079 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5080 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5081 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5082 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5083 | STAT_ATTR(FREE_SLAB, free_slab); |
5084 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5085 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5086 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5087 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5088 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5089 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5090 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5091 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5092 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5093 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5094 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5095 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5096 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5097 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5098 | #endif |
5099 | ||
06428780 | 5100 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5101 | &slab_size_attr.attr, |
5102 | &object_size_attr.attr, | |
5103 | &objs_per_slab_attr.attr, | |
5104 | &order_attr.attr, | |
73d342b1 | 5105 | &min_partial_attr.attr, |
49e22585 | 5106 | &cpu_partial_attr.attr, |
81819f0f | 5107 | &objects_attr.attr, |
205ab99d | 5108 | &objects_partial_attr.attr, |
81819f0f CL |
5109 | &partial_attr.attr, |
5110 | &cpu_slabs_attr.attr, | |
5111 | &ctor_attr.attr, | |
81819f0f CL |
5112 | &aliases_attr.attr, |
5113 | &align_attr.attr, | |
81819f0f CL |
5114 | &hwcache_align_attr.attr, |
5115 | &reclaim_account_attr.attr, | |
5116 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5117 | &shrink_attr.attr, |
ab9a0f19 | 5118 | &reserved_attr.attr, |
49e22585 | 5119 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5120 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5121 | &total_objects_attr.attr, |
5122 | &slabs_attr.attr, | |
5123 | &sanity_checks_attr.attr, | |
5124 | &trace_attr.attr, | |
81819f0f CL |
5125 | &red_zone_attr.attr, |
5126 | &poison_attr.attr, | |
5127 | &store_user_attr.attr, | |
53e15af0 | 5128 | &validate_attr.attr, |
88a420e4 CL |
5129 | &alloc_calls_attr.attr, |
5130 | &free_calls_attr.attr, | |
ab4d5ed5 | 5131 | #endif |
81819f0f CL |
5132 | #ifdef CONFIG_ZONE_DMA |
5133 | &cache_dma_attr.attr, | |
5134 | #endif | |
5135 | #ifdef CONFIG_NUMA | |
9824601e | 5136 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5137 | #endif |
5138 | #ifdef CONFIG_SLUB_STATS | |
5139 | &alloc_fastpath_attr.attr, | |
5140 | &alloc_slowpath_attr.attr, | |
5141 | &free_fastpath_attr.attr, | |
5142 | &free_slowpath_attr.attr, | |
5143 | &free_frozen_attr.attr, | |
5144 | &free_add_partial_attr.attr, | |
5145 | &free_remove_partial_attr.attr, | |
5146 | &alloc_from_partial_attr.attr, | |
5147 | &alloc_slab_attr.attr, | |
5148 | &alloc_refill_attr.attr, | |
e36a2652 | 5149 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5150 | &free_slab_attr.attr, |
5151 | &cpuslab_flush_attr.attr, | |
5152 | &deactivate_full_attr.attr, | |
5153 | &deactivate_empty_attr.attr, | |
5154 | &deactivate_to_head_attr.attr, | |
5155 | &deactivate_to_tail_attr.attr, | |
5156 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5157 | &deactivate_bypass_attr.attr, |
65c3376a | 5158 | &order_fallback_attr.attr, |
b789ef51 CL |
5159 | &cmpxchg_double_fail_attr.attr, |
5160 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5161 | &cpu_partial_alloc_attr.attr, |
5162 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5163 | &cpu_partial_node_attr.attr, |
5164 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5165 | #endif |
4c13dd3b DM |
5166 | #ifdef CONFIG_FAILSLAB |
5167 | &failslab_attr.attr, | |
5168 | #endif | |
5169 | ||
81819f0f CL |
5170 | NULL |
5171 | }; | |
5172 | ||
5173 | static struct attribute_group slab_attr_group = { | |
5174 | .attrs = slab_attrs, | |
5175 | }; | |
5176 | ||
5177 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5178 | struct attribute *attr, | |
5179 | char *buf) | |
5180 | { | |
5181 | struct slab_attribute *attribute; | |
5182 | struct kmem_cache *s; | |
5183 | int err; | |
5184 | ||
5185 | attribute = to_slab_attr(attr); | |
5186 | s = to_slab(kobj); | |
5187 | ||
5188 | if (!attribute->show) | |
5189 | return -EIO; | |
5190 | ||
5191 | err = attribute->show(s, buf); | |
5192 | ||
5193 | return err; | |
5194 | } | |
5195 | ||
5196 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5197 | struct attribute *attr, | |
5198 | const char *buf, size_t len) | |
5199 | { | |
5200 | struct slab_attribute *attribute; | |
5201 | struct kmem_cache *s; | |
5202 | int err; | |
5203 | ||
5204 | attribute = to_slab_attr(attr); | |
5205 | s = to_slab(kobj); | |
5206 | ||
5207 | if (!attribute->store) | |
5208 | return -EIO; | |
5209 | ||
5210 | err = attribute->store(s, buf, len); | |
5211 | ||
5212 | return err; | |
5213 | } | |
5214 | ||
151c602f CL |
5215 | static void kmem_cache_release(struct kobject *kobj) |
5216 | { | |
5217 | struct kmem_cache *s = to_slab(kobj); | |
5218 | ||
84c1cf62 | 5219 | kfree(s->name); |
151c602f CL |
5220 | kfree(s); |
5221 | } | |
5222 | ||
52cf25d0 | 5223 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5224 | .show = slab_attr_show, |
5225 | .store = slab_attr_store, | |
5226 | }; | |
5227 | ||
5228 | static struct kobj_type slab_ktype = { | |
5229 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 5230 | .release = kmem_cache_release |
81819f0f CL |
5231 | }; |
5232 | ||
5233 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5234 | { | |
5235 | struct kobj_type *ktype = get_ktype(kobj); | |
5236 | ||
5237 | if (ktype == &slab_ktype) | |
5238 | return 1; | |
5239 | return 0; | |
5240 | } | |
5241 | ||
9cd43611 | 5242 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5243 | .filter = uevent_filter, |
5244 | }; | |
5245 | ||
27c3a314 | 5246 | static struct kset *slab_kset; |
81819f0f CL |
5247 | |
5248 | #define ID_STR_LENGTH 64 | |
5249 | ||
5250 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5251 | * |
5252 | * Format :[flags-]size | |
81819f0f CL |
5253 | */ |
5254 | static char *create_unique_id(struct kmem_cache *s) | |
5255 | { | |
5256 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5257 | char *p = name; | |
5258 | ||
5259 | BUG_ON(!name); | |
5260 | ||
5261 | *p++ = ':'; | |
5262 | /* | |
5263 | * First flags affecting slabcache operations. We will only | |
5264 | * get here for aliasable slabs so we do not need to support | |
5265 | * too many flags. The flags here must cover all flags that | |
5266 | * are matched during merging to guarantee that the id is | |
5267 | * unique. | |
5268 | */ | |
5269 | if (s->flags & SLAB_CACHE_DMA) | |
5270 | *p++ = 'd'; | |
5271 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5272 | *p++ = 'a'; | |
5273 | if (s->flags & SLAB_DEBUG_FREE) | |
5274 | *p++ = 'F'; | |
5a896d9e VN |
5275 | if (!(s->flags & SLAB_NOTRACK)) |
5276 | *p++ = 't'; | |
81819f0f CL |
5277 | if (p != name + 1) |
5278 | *p++ = '-'; | |
5279 | p += sprintf(p, "%07d", s->size); | |
5280 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5281 | return name; | |
5282 | } | |
5283 | ||
5284 | static int sysfs_slab_add(struct kmem_cache *s) | |
5285 | { | |
5286 | int err; | |
5287 | const char *name; | |
5288 | int unmergeable; | |
5289 | ||
97d06609 | 5290 | if (slab_state < FULL) |
81819f0f CL |
5291 | /* Defer until later */ |
5292 | return 0; | |
5293 | ||
5294 | unmergeable = slab_unmergeable(s); | |
5295 | if (unmergeable) { | |
5296 | /* | |
5297 | * Slabcache can never be merged so we can use the name proper. | |
5298 | * This is typically the case for debug situations. In that | |
5299 | * case we can catch duplicate names easily. | |
5300 | */ | |
27c3a314 | 5301 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5302 | name = s->name; |
5303 | } else { | |
5304 | /* | |
5305 | * Create a unique name for the slab as a target | |
5306 | * for the symlinks. | |
5307 | */ | |
5308 | name = create_unique_id(s); | |
5309 | } | |
5310 | ||
27c3a314 | 5311 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5312 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5313 | if (err) { | |
5314 | kobject_put(&s->kobj); | |
81819f0f | 5315 | return err; |
1eada11c | 5316 | } |
81819f0f CL |
5317 | |
5318 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5319 | if (err) { |
5320 | kobject_del(&s->kobj); | |
5321 | kobject_put(&s->kobj); | |
81819f0f | 5322 | return err; |
5788d8ad | 5323 | } |
81819f0f CL |
5324 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5325 | if (!unmergeable) { | |
5326 | /* Setup first alias */ | |
5327 | sysfs_slab_alias(s, s->name); | |
5328 | kfree(name); | |
5329 | } | |
5330 | return 0; | |
5331 | } | |
5332 | ||
5333 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5334 | { | |
97d06609 | 5335 | if (slab_state < FULL) |
2bce6485 CL |
5336 | /* |
5337 | * Sysfs has not been setup yet so no need to remove the | |
5338 | * cache from sysfs. | |
5339 | */ | |
5340 | return; | |
5341 | ||
81819f0f CL |
5342 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5343 | kobject_del(&s->kobj); | |
151c602f | 5344 | kobject_put(&s->kobj); |
81819f0f CL |
5345 | } |
5346 | ||
5347 | /* | |
5348 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5349 | * available lest we lose that information. |
81819f0f CL |
5350 | */ |
5351 | struct saved_alias { | |
5352 | struct kmem_cache *s; | |
5353 | const char *name; | |
5354 | struct saved_alias *next; | |
5355 | }; | |
5356 | ||
5af328a5 | 5357 | static struct saved_alias *alias_list; |
81819f0f CL |
5358 | |
5359 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5360 | { | |
5361 | struct saved_alias *al; | |
5362 | ||
97d06609 | 5363 | if (slab_state == FULL) { |
81819f0f CL |
5364 | /* |
5365 | * If we have a leftover link then remove it. | |
5366 | */ | |
27c3a314 GKH |
5367 | sysfs_remove_link(&slab_kset->kobj, name); |
5368 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5369 | } |
5370 | ||
5371 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5372 | if (!al) | |
5373 | return -ENOMEM; | |
5374 | ||
5375 | al->s = s; | |
5376 | al->name = name; | |
5377 | al->next = alias_list; | |
5378 | alias_list = al; | |
5379 | return 0; | |
5380 | } | |
5381 | ||
5382 | static int __init slab_sysfs_init(void) | |
5383 | { | |
5b95a4ac | 5384 | struct kmem_cache *s; |
81819f0f CL |
5385 | int err; |
5386 | ||
18004c5d | 5387 | mutex_lock(&slab_mutex); |
2bce6485 | 5388 | |
0ff21e46 | 5389 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5390 | if (!slab_kset) { |
18004c5d | 5391 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5392 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5393 | return -ENOSYS; | |
5394 | } | |
5395 | ||
97d06609 | 5396 | slab_state = FULL; |
26a7bd03 | 5397 | |
5b95a4ac | 5398 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5399 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5400 | if (err) |
5401 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5402 | " to sysfs\n", s->name); | |
26a7bd03 | 5403 | } |
81819f0f CL |
5404 | |
5405 | while (alias_list) { | |
5406 | struct saved_alias *al = alias_list; | |
5407 | ||
5408 | alias_list = alias_list->next; | |
5409 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5410 | if (err) |
5411 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5412 | " %s to sysfs\n", al->name); |
81819f0f CL |
5413 | kfree(al); |
5414 | } | |
5415 | ||
18004c5d | 5416 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5417 | resiliency_test(); |
5418 | return 0; | |
5419 | } | |
5420 | ||
5421 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5422 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5423 | |
5424 | /* | |
5425 | * The /proc/slabinfo ABI | |
5426 | */ | |
158a9624 | 5427 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5428 | static void print_slabinfo_header(struct seq_file *m) |
5429 | { | |
5430 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
3b0efdfa | 5431 | seq_puts(m, "# name <active_objs> <num_objs> <object_size> " |
57ed3eda PE |
5432 | "<objperslab> <pagesperslab>"); |
5433 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5434 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5435 | seq_putc(m, '\n'); | |
5436 | } | |
5437 | ||
5438 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5439 | { | |
5440 | loff_t n = *pos; | |
5441 | ||
18004c5d | 5442 | mutex_lock(&slab_mutex); |
57ed3eda PE |
5443 | if (!n) |
5444 | print_slabinfo_header(m); | |
5445 | ||
5446 | return seq_list_start(&slab_caches, *pos); | |
5447 | } | |
5448 | ||
5449 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5450 | { | |
5451 | return seq_list_next(p, &slab_caches, pos); | |
5452 | } | |
5453 | ||
5454 | static void s_stop(struct seq_file *m, void *p) | |
5455 | { | |
18004c5d | 5456 | mutex_unlock(&slab_mutex); |
57ed3eda PE |
5457 | } |
5458 | ||
5459 | static int s_show(struct seq_file *m, void *p) | |
5460 | { | |
5461 | unsigned long nr_partials = 0; | |
5462 | unsigned long nr_slabs = 0; | |
5463 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5464 | unsigned long nr_objs = 0; |
5465 | unsigned long nr_free = 0; | |
57ed3eda PE |
5466 | struct kmem_cache *s; |
5467 | int node; | |
5468 | ||
5469 | s = list_entry(p, struct kmem_cache, list); | |
5470 | ||
5471 | for_each_online_node(node) { | |
5472 | struct kmem_cache_node *n = get_node(s, node); | |
5473 | ||
5474 | if (!n) | |
5475 | continue; | |
5476 | ||
5477 | nr_partials += n->nr_partial; | |
5478 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5479 | nr_objs += atomic_long_read(&n->total_objects); |
5480 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5481 | } |
5482 | ||
205ab99d | 5483 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5484 | |
5485 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5486 | nr_objs, s->size, oo_objects(s->oo), |
5487 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5488 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5489 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5490 | 0UL); | |
5491 | seq_putc(m, '\n'); | |
5492 | return 0; | |
5493 | } | |
5494 | ||
7b3c3a50 | 5495 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5496 | .start = s_start, |
5497 | .next = s_next, | |
5498 | .stop = s_stop, | |
5499 | .show = s_show, | |
5500 | }; | |
5501 | ||
7b3c3a50 AD |
5502 | static int slabinfo_open(struct inode *inode, struct file *file) |
5503 | { | |
5504 | return seq_open(file, &slabinfo_op); | |
5505 | } | |
5506 | ||
5507 | static const struct file_operations proc_slabinfo_operations = { | |
5508 | .open = slabinfo_open, | |
5509 | .read = seq_read, | |
5510 | .llseek = seq_lseek, | |
5511 | .release = seq_release, | |
5512 | }; | |
5513 | ||
5514 | static int __init slab_proc_init(void) | |
5515 | { | |
ab067e99 | 5516 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5517 | return 0; |
5518 | } | |
5519 | module_init(slab_proc_init); | |
158a9624 | 5520 | #endif /* CONFIG_SLABINFO */ |