]>
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"); | |
645df230 DJ |
571 | |
572 | add_taint(TAINT_BAD_PAGE); | |
81819f0f CL |
573 | } |
574 | ||
24922684 CL |
575 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
576 | { | |
577 | va_list args; | |
578 | char buf[100]; | |
579 | ||
580 | va_start(args, fmt); | |
581 | vsnprintf(buf, sizeof(buf), fmt, args); | |
582 | va_end(args); | |
583 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
584 | } | |
585 | ||
586 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
587 | { |
588 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 589 | u8 *addr = page_address(page); |
24922684 CL |
590 | |
591 | print_tracking(s, p); | |
592 | ||
593 | print_page_info(page); | |
594 | ||
595 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
596 | p, p - addr, get_freepointer(s, p)); | |
597 | ||
598 | if (p > addr + 16) | |
ffc79d28 | 599 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 600 | |
3b0efdfa | 601 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 602 | PAGE_SIZE)); |
81819f0f | 603 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
604 | print_section("Redzone ", p + s->object_size, |
605 | s->inuse - s->object_size); | |
81819f0f | 606 | |
81819f0f CL |
607 | if (s->offset) |
608 | off = s->offset + sizeof(void *); | |
609 | else | |
610 | off = s->inuse; | |
611 | ||
24922684 | 612 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 613 | off += 2 * sizeof(struct track); |
81819f0f CL |
614 | |
615 | if (off != s->size) | |
616 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 617 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
618 | |
619 | dump_stack(); | |
81819f0f CL |
620 | } |
621 | ||
622 | static void object_err(struct kmem_cache *s, struct page *page, | |
623 | u8 *object, char *reason) | |
624 | { | |
3dc50637 | 625 | slab_bug(s, "%s", reason); |
24922684 | 626 | print_trailer(s, page, object); |
81819f0f CL |
627 | } |
628 | ||
24922684 | 629 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
630 | { |
631 | va_list args; | |
632 | char buf[100]; | |
633 | ||
24922684 CL |
634 | va_start(args, fmt); |
635 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 636 | va_end(args); |
3dc50637 | 637 | slab_bug(s, "%s", buf); |
24922684 | 638 | print_page_info(page); |
81819f0f CL |
639 | dump_stack(); |
640 | } | |
641 | ||
f7cb1933 | 642 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
643 | { |
644 | u8 *p = object; | |
645 | ||
646 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
647 | memset(p, POISON_FREE, s->object_size - 1); |
648 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
649 | } |
650 | ||
651 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 652 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
653 | } |
654 | ||
24922684 CL |
655 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
656 | void *from, void *to) | |
657 | { | |
658 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
659 | memset(from, data, to - from); | |
660 | } | |
661 | ||
662 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
663 | u8 *object, char *what, | |
06428780 | 664 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
665 | { |
666 | u8 *fault; | |
667 | u8 *end; | |
668 | ||
79824820 | 669 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
670 | if (!fault) |
671 | return 1; | |
672 | ||
673 | end = start + bytes; | |
674 | while (end > fault && end[-1] == value) | |
675 | end--; | |
676 | ||
677 | slab_bug(s, "%s overwritten", what); | |
678 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
679 | fault, end - 1, fault[0], value); | |
680 | print_trailer(s, page, object); | |
681 | ||
682 | restore_bytes(s, what, value, fault, end); | |
683 | return 0; | |
81819f0f CL |
684 | } |
685 | ||
81819f0f CL |
686 | /* |
687 | * Object layout: | |
688 | * | |
689 | * object address | |
690 | * Bytes of the object to be managed. | |
691 | * If the freepointer may overlay the object then the free | |
692 | * pointer is the first word of the object. | |
672bba3a | 693 | * |
81819f0f CL |
694 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
695 | * 0xa5 (POISON_END) | |
696 | * | |
3b0efdfa | 697 | * object + s->object_size |
81819f0f | 698 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 699 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 700 | * object_size == inuse. |
672bba3a | 701 | * |
81819f0f CL |
702 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
703 | * 0xcc (RED_ACTIVE) for objects in use. | |
704 | * | |
705 | * object + s->inuse | |
672bba3a CL |
706 | * Meta data starts here. |
707 | * | |
81819f0f CL |
708 | * A. Free pointer (if we cannot overwrite object on free) |
709 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 710 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 711 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
712 | * before the word boundary. |
713 | * | |
714 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
715 | * |
716 | * object + s->size | |
672bba3a | 717 | * Nothing is used beyond s->size. |
81819f0f | 718 | * |
3b0efdfa | 719 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 720 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
721 | * may be used with merged slabcaches. |
722 | */ | |
723 | ||
81819f0f CL |
724 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
725 | { | |
726 | unsigned long off = s->inuse; /* The end of info */ | |
727 | ||
728 | if (s->offset) | |
729 | /* Freepointer is placed after the object. */ | |
730 | off += sizeof(void *); | |
731 | ||
732 | if (s->flags & SLAB_STORE_USER) | |
733 | /* We also have user information there */ | |
734 | off += 2 * sizeof(struct track); | |
735 | ||
736 | if (s->size == off) | |
737 | return 1; | |
738 | ||
24922684 CL |
739 | return check_bytes_and_report(s, page, p, "Object padding", |
740 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
741 | } |
742 | ||
39b26464 | 743 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
744 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
745 | { | |
24922684 CL |
746 | u8 *start; |
747 | u8 *fault; | |
748 | u8 *end; | |
749 | int length; | |
750 | int remainder; | |
81819f0f CL |
751 | |
752 | if (!(s->flags & SLAB_POISON)) | |
753 | return 1; | |
754 | ||
a973e9dd | 755 | start = page_address(page); |
ab9a0f19 | 756 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
757 | end = start + length; |
758 | remainder = length % s->size; | |
81819f0f CL |
759 | if (!remainder) |
760 | return 1; | |
761 | ||
79824820 | 762 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
763 | if (!fault) |
764 | return 1; | |
765 | while (end > fault && end[-1] == POISON_INUSE) | |
766 | end--; | |
767 | ||
768 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 769 | print_section("Padding ", end - remainder, remainder); |
24922684 | 770 | |
8a3d271d | 771 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 772 | return 0; |
81819f0f CL |
773 | } |
774 | ||
775 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 776 | void *object, u8 val) |
81819f0f CL |
777 | { |
778 | u8 *p = object; | |
3b0efdfa | 779 | u8 *endobject = object + s->object_size; |
81819f0f CL |
780 | |
781 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 782 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 783 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 784 | return 0; |
81819f0f | 785 | } else { |
3b0efdfa | 786 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 787 | check_bytes_and_report(s, page, p, "Alignment padding", |
3b0efdfa | 788 | endobject, POISON_INUSE, s->inuse - s->object_size); |
3adbefee | 789 | } |
81819f0f CL |
790 | } |
791 | ||
792 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 793 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 794 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 795 | POISON_FREE, s->object_size - 1) || |
24922684 | 796 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 797 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 798 | return 0; |
81819f0f CL |
799 | /* |
800 | * check_pad_bytes cleans up on its own. | |
801 | */ | |
802 | check_pad_bytes(s, page, p); | |
803 | } | |
804 | ||
f7cb1933 | 805 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
806 | /* |
807 | * Object and freepointer overlap. Cannot check | |
808 | * freepointer while object is allocated. | |
809 | */ | |
810 | return 1; | |
811 | ||
812 | /* Check free pointer validity */ | |
813 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
814 | object_err(s, page, p, "Freepointer corrupt"); | |
815 | /* | |
9f6c708e | 816 | * No choice but to zap it and thus lose the remainder |
81819f0f | 817 | * of the free objects in this slab. May cause |
672bba3a | 818 | * another error because the object count is now wrong. |
81819f0f | 819 | */ |
a973e9dd | 820 | set_freepointer(s, p, NULL); |
81819f0f CL |
821 | return 0; |
822 | } | |
823 | return 1; | |
824 | } | |
825 | ||
826 | static int check_slab(struct kmem_cache *s, struct page *page) | |
827 | { | |
39b26464 CL |
828 | int maxobj; |
829 | ||
81819f0f CL |
830 | VM_BUG_ON(!irqs_disabled()); |
831 | ||
832 | if (!PageSlab(page)) { | |
24922684 | 833 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
834 | return 0; |
835 | } | |
39b26464 | 836 | |
ab9a0f19 | 837 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
838 | if (page->objects > maxobj) { |
839 | slab_err(s, page, "objects %u > max %u", | |
840 | s->name, page->objects, maxobj); | |
841 | return 0; | |
842 | } | |
843 | if (page->inuse > page->objects) { | |
24922684 | 844 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 845 | s->name, page->inuse, page->objects); |
81819f0f CL |
846 | return 0; |
847 | } | |
848 | /* Slab_pad_check fixes things up after itself */ | |
849 | slab_pad_check(s, page); | |
850 | return 1; | |
851 | } | |
852 | ||
853 | /* | |
672bba3a CL |
854 | * Determine if a certain object on a page is on the freelist. Must hold the |
855 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
856 | */ |
857 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
858 | { | |
859 | int nr = 0; | |
881db7fb | 860 | void *fp; |
81819f0f | 861 | void *object = NULL; |
224a88be | 862 | unsigned long max_objects; |
81819f0f | 863 | |
881db7fb | 864 | fp = page->freelist; |
39b26464 | 865 | while (fp && nr <= page->objects) { |
81819f0f CL |
866 | if (fp == search) |
867 | return 1; | |
868 | if (!check_valid_pointer(s, page, fp)) { | |
869 | if (object) { | |
870 | object_err(s, page, object, | |
871 | "Freechain corrupt"); | |
a973e9dd | 872 | set_freepointer(s, object, NULL); |
81819f0f CL |
873 | break; |
874 | } else { | |
24922684 | 875 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 876 | page->freelist = NULL; |
39b26464 | 877 | page->inuse = page->objects; |
24922684 | 878 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
879 | return 0; |
880 | } | |
881 | break; | |
882 | } | |
883 | object = fp; | |
884 | fp = get_freepointer(s, object); | |
885 | nr++; | |
886 | } | |
887 | ||
ab9a0f19 | 888 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
889 | if (max_objects > MAX_OBJS_PER_PAGE) |
890 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
891 | |
892 | if (page->objects != max_objects) { | |
893 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
894 | "should be %d", page->objects, max_objects); | |
895 | page->objects = max_objects; | |
896 | slab_fix(s, "Number of objects adjusted."); | |
897 | } | |
39b26464 | 898 | if (page->inuse != page->objects - nr) { |
70d71228 | 899 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
900 | "counted were %d", page->inuse, page->objects - nr); |
901 | page->inuse = page->objects - nr; | |
24922684 | 902 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
903 | } |
904 | return search == NULL; | |
905 | } | |
906 | ||
0121c619 CL |
907 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
908 | int alloc) | |
3ec09742 CL |
909 | { |
910 | if (s->flags & SLAB_TRACE) { | |
911 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
912 | s->name, | |
913 | alloc ? "alloc" : "free", | |
914 | object, page->inuse, | |
915 | page->freelist); | |
916 | ||
917 | if (!alloc) | |
3b0efdfa | 918 | print_section("Object ", (void *)object, s->object_size); |
3ec09742 CL |
919 | |
920 | dump_stack(); | |
921 | } | |
922 | } | |
923 | ||
c016b0bd CL |
924 | /* |
925 | * Hooks for other subsystems that check memory allocations. In a typical | |
926 | * production configuration these hooks all should produce no code at all. | |
927 | */ | |
928 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
929 | { | |
c1d50836 | 930 | flags &= gfp_allowed_mask; |
c016b0bd CL |
931 | lockdep_trace_alloc(flags); |
932 | might_sleep_if(flags & __GFP_WAIT); | |
933 | ||
3b0efdfa | 934 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
935 | } |
936 | ||
937 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
938 | { | |
c1d50836 | 939 | flags &= gfp_allowed_mask; |
b3d41885 | 940 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 941 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
942 | } |
943 | ||
944 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
945 | { | |
946 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 947 | |
d3f661d6 CL |
948 | /* |
949 | * Trouble is that we may no longer disable interupts in the fast path | |
950 | * So in order to make the debug calls that expect irqs to be | |
951 | * disabled we need to disable interrupts temporarily. | |
952 | */ | |
953 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
954 | { | |
955 | unsigned long flags; | |
956 | ||
957 | local_irq_save(flags); | |
3b0efdfa CL |
958 | kmemcheck_slab_free(s, x, s->object_size); |
959 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
960 | local_irq_restore(flags); |
961 | } | |
962 | #endif | |
f9b615de | 963 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 964 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
965 | } |
966 | ||
643b1138 | 967 | /* |
672bba3a | 968 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
969 | * |
970 | * list_lock must be held. | |
643b1138 | 971 | */ |
5cc6eee8 CL |
972 | static void add_full(struct kmem_cache *s, |
973 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 974 | { |
5cc6eee8 CL |
975 | if (!(s->flags & SLAB_STORE_USER)) |
976 | return; | |
977 | ||
643b1138 | 978 | list_add(&page->lru, &n->full); |
643b1138 CL |
979 | } |
980 | ||
5cc6eee8 CL |
981 | /* |
982 | * list_lock must be held. | |
983 | */ | |
643b1138 CL |
984 | static void remove_full(struct kmem_cache *s, struct page *page) |
985 | { | |
643b1138 CL |
986 | if (!(s->flags & SLAB_STORE_USER)) |
987 | return; | |
988 | ||
643b1138 | 989 | list_del(&page->lru); |
643b1138 CL |
990 | } |
991 | ||
0f389ec6 CL |
992 | /* Tracking of the number of slabs for debugging purposes */ |
993 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
994 | { | |
995 | struct kmem_cache_node *n = get_node(s, node); | |
996 | ||
997 | return atomic_long_read(&n->nr_slabs); | |
998 | } | |
999 | ||
26c02cf0 AB |
1000 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1001 | { | |
1002 | return atomic_long_read(&n->nr_slabs); | |
1003 | } | |
1004 | ||
205ab99d | 1005 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1006 | { |
1007 | struct kmem_cache_node *n = get_node(s, node); | |
1008 | ||
1009 | /* | |
1010 | * May be called early in order to allocate a slab for the | |
1011 | * kmem_cache_node structure. Solve the chicken-egg | |
1012 | * dilemma by deferring the increment of the count during | |
1013 | * bootstrap (see early_kmem_cache_node_alloc). | |
1014 | */ | |
7340cc84 | 1015 | if (n) { |
0f389ec6 | 1016 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1017 | atomic_long_add(objects, &n->total_objects); |
1018 | } | |
0f389ec6 | 1019 | } |
205ab99d | 1020 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1021 | { |
1022 | struct kmem_cache_node *n = get_node(s, node); | |
1023 | ||
1024 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1025 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1026 | } |
1027 | ||
1028 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1029 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1030 | void *object) | |
1031 | { | |
1032 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1033 | return; | |
1034 | ||
f7cb1933 | 1035 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1036 | init_tracking(s, object); |
1037 | } | |
1038 | ||
1537066c | 1039 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1040 | void *object, unsigned long addr) |
81819f0f CL |
1041 | { |
1042 | if (!check_slab(s, page)) | |
1043 | goto bad; | |
1044 | ||
81819f0f CL |
1045 | if (!check_valid_pointer(s, page, object)) { |
1046 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1047 | goto bad; |
81819f0f CL |
1048 | } |
1049 | ||
f7cb1933 | 1050 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1051 | goto bad; |
81819f0f | 1052 | |
3ec09742 CL |
1053 | /* Success perform special debug activities for allocs */ |
1054 | if (s->flags & SLAB_STORE_USER) | |
1055 | set_track(s, object, TRACK_ALLOC, addr); | |
1056 | trace(s, page, object, 1); | |
f7cb1933 | 1057 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1058 | return 1; |
3ec09742 | 1059 | |
81819f0f CL |
1060 | bad: |
1061 | if (PageSlab(page)) { | |
1062 | /* | |
1063 | * If this is a slab page then lets do the best we can | |
1064 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1065 | * as used avoids touching the remaining objects. |
81819f0f | 1066 | */ |
24922684 | 1067 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1068 | page->inuse = page->objects; |
a973e9dd | 1069 | page->freelist = NULL; |
81819f0f CL |
1070 | } |
1071 | return 0; | |
1072 | } | |
1073 | ||
19c7ff9e CL |
1074 | static noinline struct kmem_cache_node *free_debug_processing( |
1075 | struct kmem_cache *s, struct page *page, void *object, | |
1076 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1077 | { |
19c7ff9e | 1078 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1079 | |
19c7ff9e | 1080 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1081 | slab_lock(page); |
1082 | ||
81819f0f CL |
1083 | if (!check_slab(s, page)) |
1084 | goto fail; | |
1085 | ||
1086 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1087 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1088 | goto fail; |
1089 | } | |
1090 | ||
1091 | if (on_freelist(s, page, object)) { | |
24922684 | 1092 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1093 | goto fail; |
1094 | } | |
1095 | ||
f7cb1933 | 1096 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1097 | goto out; |
81819f0f CL |
1098 | |
1099 | if (unlikely(s != page->slab)) { | |
3adbefee | 1100 | if (!PageSlab(page)) { |
70d71228 CL |
1101 | slab_err(s, page, "Attempt to free object(0x%p) " |
1102 | "outside of slab", object); | |
3adbefee | 1103 | } else if (!page->slab) { |
81819f0f | 1104 | printk(KERN_ERR |
70d71228 | 1105 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1106 | object); |
70d71228 | 1107 | dump_stack(); |
06428780 | 1108 | } else |
24922684 CL |
1109 | object_err(s, page, object, |
1110 | "page slab pointer corrupt."); | |
81819f0f CL |
1111 | goto fail; |
1112 | } | |
3ec09742 | 1113 | |
3ec09742 CL |
1114 | if (s->flags & SLAB_STORE_USER) |
1115 | set_track(s, object, TRACK_FREE, addr); | |
1116 | trace(s, page, object, 0); | |
f7cb1933 | 1117 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1118 | out: |
881db7fb | 1119 | slab_unlock(page); |
19c7ff9e CL |
1120 | /* |
1121 | * Keep node_lock to preserve integrity | |
1122 | * until the object is actually freed | |
1123 | */ | |
1124 | return n; | |
3ec09742 | 1125 | |
81819f0f | 1126 | fail: |
19c7ff9e CL |
1127 | slab_unlock(page); |
1128 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1129 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1130 | return NULL; |
81819f0f CL |
1131 | } |
1132 | ||
41ecc55b CL |
1133 | static int __init setup_slub_debug(char *str) |
1134 | { | |
f0630fff CL |
1135 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1136 | if (*str++ != '=' || !*str) | |
1137 | /* | |
1138 | * No options specified. Switch on full debugging. | |
1139 | */ | |
1140 | goto out; | |
1141 | ||
1142 | if (*str == ',') | |
1143 | /* | |
1144 | * No options but restriction on slabs. This means full | |
1145 | * debugging for slabs matching a pattern. | |
1146 | */ | |
1147 | goto check_slabs; | |
1148 | ||
fa5ec8a1 DR |
1149 | if (tolower(*str) == 'o') { |
1150 | /* | |
1151 | * Avoid enabling debugging on caches if its minimum order | |
1152 | * would increase as a result. | |
1153 | */ | |
1154 | disable_higher_order_debug = 1; | |
1155 | goto out; | |
1156 | } | |
1157 | ||
f0630fff CL |
1158 | slub_debug = 0; |
1159 | if (*str == '-') | |
1160 | /* | |
1161 | * Switch off all debugging measures. | |
1162 | */ | |
1163 | goto out; | |
1164 | ||
1165 | /* | |
1166 | * Determine which debug features should be switched on | |
1167 | */ | |
06428780 | 1168 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1169 | switch (tolower(*str)) { |
1170 | case 'f': | |
1171 | slub_debug |= SLAB_DEBUG_FREE; | |
1172 | break; | |
1173 | case 'z': | |
1174 | slub_debug |= SLAB_RED_ZONE; | |
1175 | break; | |
1176 | case 'p': | |
1177 | slub_debug |= SLAB_POISON; | |
1178 | break; | |
1179 | case 'u': | |
1180 | slub_debug |= SLAB_STORE_USER; | |
1181 | break; | |
1182 | case 't': | |
1183 | slub_debug |= SLAB_TRACE; | |
1184 | break; | |
4c13dd3b DM |
1185 | case 'a': |
1186 | slub_debug |= SLAB_FAILSLAB; | |
1187 | break; | |
f0630fff CL |
1188 | default: |
1189 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1190 | "unknown. skipped\n", *str); |
f0630fff | 1191 | } |
41ecc55b CL |
1192 | } |
1193 | ||
f0630fff | 1194 | check_slabs: |
41ecc55b CL |
1195 | if (*str == ',') |
1196 | slub_debug_slabs = str + 1; | |
f0630fff | 1197 | out: |
41ecc55b CL |
1198 | return 1; |
1199 | } | |
1200 | ||
1201 | __setup("slub_debug", setup_slub_debug); | |
1202 | ||
3b0efdfa | 1203 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1204 | unsigned long flags, const char *name, |
51cc5068 | 1205 | void (*ctor)(void *)) |
41ecc55b CL |
1206 | { |
1207 | /* | |
e153362a | 1208 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1209 | */ |
e153362a | 1210 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1211 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1212 | flags |= slub_debug; | |
ba0268a8 CL |
1213 | |
1214 | return flags; | |
41ecc55b CL |
1215 | } |
1216 | #else | |
3ec09742 CL |
1217 | static inline void setup_object_debug(struct kmem_cache *s, |
1218 | struct page *page, void *object) {} | |
41ecc55b | 1219 | |
3ec09742 | 1220 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1221 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1222 | |
19c7ff9e CL |
1223 | static inline struct kmem_cache_node *free_debug_processing( |
1224 | struct kmem_cache *s, struct page *page, void *object, | |
1225 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1226 | |
41ecc55b CL |
1227 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1228 | { return 1; } | |
1229 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1230 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1231 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1232 | struct page *page) {} | |
2cfb7455 | 1233 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
3b0efdfa | 1234 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1235 | unsigned long flags, const char *name, |
51cc5068 | 1236 | void (*ctor)(void *)) |
ba0268a8 CL |
1237 | { |
1238 | return flags; | |
1239 | } | |
41ecc55b | 1240 | #define slub_debug 0 |
0f389ec6 | 1241 | |
fdaa45e9 IM |
1242 | #define disable_higher_order_debug 0 |
1243 | ||
0f389ec6 CL |
1244 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1245 | { return 0; } | |
26c02cf0 AB |
1246 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1247 | { return 0; } | |
205ab99d CL |
1248 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1249 | int objects) {} | |
1250 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1251 | int objects) {} | |
7d550c56 CL |
1252 | |
1253 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1254 | { return 0; } | |
1255 | ||
1256 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1257 | void *object) {} | |
1258 | ||
1259 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1260 | ||
ab4d5ed5 | 1261 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1262 | |
81819f0f CL |
1263 | /* |
1264 | * Slab allocation and freeing | |
1265 | */ | |
65c3376a CL |
1266 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1267 | struct kmem_cache_order_objects oo) | |
1268 | { | |
1269 | int order = oo_order(oo); | |
1270 | ||
b1eeab67 VN |
1271 | flags |= __GFP_NOTRACK; |
1272 | ||
2154a336 | 1273 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1274 | return alloc_pages(flags, order); |
1275 | else | |
6b65aaf3 | 1276 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1277 | } |
1278 | ||
81819f0f CL |
1279 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1280 | { | |
06428780 | 1281 | struct page *page; |
834f3d11 | 1282 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1283 | gfp_t alloc_gfp; |
81819f0f | 1284 | |
7e0528da CL |
1285 | flags &= gfp_allowed_mask; |
1286 | ||
1287 | if (flags & __GFP_WAIT) | |
1288 | local_irq_enable(); | |
1289 | ||
b7a49f0d | 1290 | flags |= s->allocflags; |
e12ba74d | 1291 | |
ba52270d PE |
1292 | /* |
1293 | * Let the initial higher-order allocation fail under memory pressure | |
1294 | * so we fall-back to the minimum order allocation. | |
1295 | */ | |
1296 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1297 | ||
1298 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1299 | if (unlikely(!page)) { |
1300 | oo = s->min; | |
1301 | /* | |
1302 | * Allocation may have failed due to fragmentation. | |
1303 | * Try a lower order alloc if possible | |
1304 | */ | |
1305 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1306 | |
7e0528da CL |
1307 | if (page) |
1308 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1309 | } |
5a896d9e | 1310 | |
737b719e | 1311 | if (kmemcheck_enabled && page |
5086c389 | 1312 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1313 | int pages = 1 << oo_order(oo); |
1314 | ||
1315 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1316 | ||
1317 | /* | |
1318 | * Objects from caches that have a constructor don't get | |
1319 | * cleared when they're allocated, so we need to do it here. | |
1320 | */ | |
1321 | if (s->ctor) | |
1322 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1323 | else | |
1324 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1325 | } |
1326 | ||
737b719e DR |
1327 | if (flags & __GFP_WAIT) |
1328 | local_irq_disable(); | |
1329 | if (!page) | |
1330 | return NULL; | |
1331 | ||
834f3d11 | 1332 | page->objects = oo_objects(oo); |
81819f0f CL |
1333 | mod_zone_page_state(page_zone(page), |
1334 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1335 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1336 | 1 << oo_order(oo)); |
81819f0f CL |
1337 | |
1338 | return page; | |
1339 | } | |
1340 | ||
1341 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1342 | void *object) | |
1343 | { | |
3ec09742 | 1344 | setup_object_debug(s, page, object); |
4f104934 | 1345 | if (unlikely(s->ctor)) |
51cc5068 | 1346 | s->ctor(object); |
81819f0f CL |
1347 | } |
1348 | ||
1349 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1350 | { | |
1351 | struct page *page; | |
81819f0f | 1352 | void *start; |
81819f0f CL |
1353 | void *last; |
1354 | void *p; | |
1355 | ||
6cb06229 | 1356 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1357 | |
6cb06229 CL |
1358 | page = allocate_slab(s, |
1359 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1360 | if (!page) |
1361 | goto out; | |
1362 | ||
205ab99d | 1363 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f | 1364 | page->slab = s; |
c03f94cc | 1365 | __SetPageSlab(page); |
072bb0aa MG |
1366 | if (page->pfmemalloc) |
1367 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1368 | |
1369 | start = page_address(page); | |
81819f0f CL |
1370 | |
1371 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1372 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1373 | |
1374 | last = start; | |
224a88be | 1375 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1376 | setup_object(s, page, last); |
1377 | set_freepointer(s, last, p); | |
1378 | last = p; | |
1379 | } | |
1380 | setup_object(s, page, last); | |
a973e9dd | 1381 | set_freepointer(s, last, NULL); |
81819f0f CL |
1382 | |
1383 | page->freelist = start; | |
e6e82ea1 | 1384 | page->inuse = page->objects; |
8cb0a506 | 1385 | page->frozen = 1; |
81819f0f | 1386 | out: |
81819f0f CL |
1387 | return page; |
1388 | } | |
1389 | ||
1390 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1391 | { | |
834f3d11 CL |
1392 | int order = compound_order(page); |
1393 | int pages = 1 << order; | |
81819f0f | 1394 | |
af537b0a | 1395 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1396 | void *p; |
1397 | ||
1398 | slab_pad_check(s, page); | |
224a88be CL |
1399 | for_each_object(p, s, page_address(page), |
1400 | page->objects) | |
f7cb1933 | 1401 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1402 | } |
1403 | ||
b1eeab67 | 1404 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1405 | |
81819f0f CL |
1406 | mod_zone_page_state(page_zone(page), |
1407 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1408 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1409 | -pages); |
81819f0f | 1410 | |
072bb0aa | 1411 | __ClearPageSlabPfmemalloc(page); |
49bd5221 CL |
1412 | __ClearPageSlab(page); |
1413 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1414 | if (current->reclaim_state) |
1415 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1416 | __free_pages(page, order); |
81819f0f CL |
1417 | } |
1418 | ||
da9a638c LJ |
1419 | #define need_reserve_slab_rcu \ |
1420 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1421 | ||
81819f0f CL |
1422 | static void rcu_free_slab(struct rcu_head *h) |
1423 | { | |
1424 | struct page *page; | |
1425 | ||
da9a638c LJ |
1426 | if (need_reserve_slab_rcu) |
1427 | page = virt_to_head_page(h); | |
1428 | else | |
1429 | page = container_of((struct list_head *)h, struct page, lru); | |
1430 | ||
81819f0f CL |
1431 | __free_slab(page->slab, page); |
1432 | } | |
1433 | ||
1434 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1435 | { | |
1436 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1437 | struct rcu_head *head; |
1438 | ||
1439 | if (need_reserve_slab_rcu) { | |
1440 | int order = compound_order(page); | |
1441 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1442 | ||
1443 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1444 | head = page_address(page) + offset; | |
1445 | } else { | |
1446 | /* | |
1447 | * RCU free overloads the RCU head over the LRU | |
1448 | */ | |
1449 | head = (void *)&page->lru; | |
1450 | } | |
81819f0f CL |
1451 | |
1452 | call_rcu(head, rcu_free_slab); | |
1453 | } else | |
1454 | __free_slab(s, page); | |
1455 | } | |
1456 | ||
1457 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1458 | { | |
205ab99d | 1459 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1460 | free_slab(s, page); |
1461 | } | |
1462 | ||
1463 | /* | |
5cc6eee8 CL |
1464 | * Management of partially allocated slabs. |
1465 | * | |
1466 | * list_lock must be held. | |
81819f0f | 1467 | */ |
5cc6eee8 | 1468 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1469 | struct page *page, int tail) |
81819f0f | 1470 | { |
e95eed57 | 1471 | n->nr_partial++; |
136333d1 | 1472 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1473 | list_add_tail(&page->lru, &n->partial); |
1474 | else | |
1475 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1476 | } |
1477 | ||
5cc6eee8 CL |
1478 | /* |
1479 | * list_lock must be held. | |
1480 | */ | |
1481 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1482 | struct page *page) |
1483 | { | |
1484 | list_del(&page->lru); | |
1485 | n->nr_partial--; | |
1486 | } | |
1487 | ||
81819f0f | 1488 | /* |
7ced3719 CL |
1489 | * Remove slab from the partial list, freeze it and |
1490 | * return the pointer to the freelist. | |
81819f0f | 1491 | * |
497b66f2 CL |
1492 | * Returns a list of objects or NULL if it fails. |
1493 | * | |
7ced3719 | 1494 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1495 | */ |
497b66f2 | 1496 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1497 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1498 | int mode) |
81819f0f | 1499 | { |
2cfb7455 CL |
1500 | void *freelist; |
1501 | unsigned long counters; | |
1502 | struct page new; | |
1503 | ||
2cfb7455 CL |
1504 | /* |
1505 | * Zap the freelist and set the frozen bit. | |
1506 | * The old freelist is the list of objects for the | |
1507 | * per cpu allocation list. | |
1508 | */ | |
7ced3719 CL |
1509 | freelist = page->freelist; |
1510 | counters = page->counters; | |
1511 | new.counters = counters; | |
23910c50 | 1512 | if (mode) { |
7ced3719 | 1513 | new.inuse = page->objects; |
23910c50 PE |
1514 | new.freelist = NULL; |
1515 | } else { | |
1516 | new.freelist = freelist; | |
1517 | } | |
2cfb7455 | 1518 | |
7ced3719 CL |
1519 | VM_BUG_ON(new.frozen); |
1520 | new.frozen = 1; | |
2cfb7455 | 1521 | |
7ced3719 | 1522 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1523 | freelist, counters, |
02d7633f | 1524 | new.freelist, new.counters, |
7ced3719 | 1525 | "acquire_slab")) |
7ced3719 | 1526 | return NULL; |
2cfb7455 CL |
1527 | |
1528 | remove_partial(n, page); | |
7ced3719 | 1529 | WARN_ON(!freelist); |
49e22585 | 1530 | return freelist; |
81819f0f CL |
1531 | } |
1532 | ||
49e22585 CL |
1533 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
1534 | ||
81819f0f | 1535 | /* |
672bba3a | 1536 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1537 | */ |
497b66f2 | 1538 | static void *get_partial_node(struct kmem_cache *s, |
acd19fd1 | 1539 | struct kmem_cache_node *n, struct kmem_cache_cpu *c) |
81819f0f | 1540 | { |
49e22585 CL |
1541 | struct page *page, *page2; |
1542 | void *object = NULL; | |
81819f0f CL |
1543 | |
1544 | /* | |
1545 | * Racy check. If we mistakenly see no partial slabs then we | |
1546 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1547 | * partial slab and there is none available then get_partials() |
1548 | * will return NULL. | |
81819f0f CL |
1549 | */ |
1550 | if (!n || !n->nr_partial) | |
1551 | return NULL; | |
1552 | ||
1553 | spin_lock(&n->list_lock); | |
49e22585 | 1554 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
12d79634 | 1555 | void *t = acquire_slab(s, n, page, object == NULL); |
49e22585 CL |
1556 | int available; |
1557 | ||
1558 | if (!t) | |
1559 | break; | |
1560 | ||
12d79634 | 1561 | if (!object) { |
49e22585 | 1562 | c->page = page; |
49e22585 | 1563 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 CL |
1564 | object = t; |
1565 | available = page->objects - page->inuse; | |
1566 | } else { | |
49e22585 | 1567 | available = put_cpu_partial(s, page, 0); |
8028dcea | 1568 | stat(s, CPU_PARTIAL_NODE); |
49e22585 CL |
1569 | } |
1570 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1571 | break; | |
1572 | ||
497b66f2 | 1573 | } |
81819f0f | 1574 | spin_unlock(&n->list_lock); |
497b66f2 | 1575 | return object; |
81819f0f CL |
1576 | } |
1577 | ||
1578 | /* | |
672bba3a | 1579 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1580 | */ |
de3ec035 | 1581 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1582 | struct kmem_cache_cpu *c) |
81819f0f CL |
1583 | { |
1584 | #ifdef CONFIG_NUMA | |
1585 | struct zonelist *zonelist; | |
dd1a239f | 1586 | struct zoneref *z; |
54a6eb5c MG |
1587 | struct zone *zone; |
1588 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1589 | void *object; |
cc9a6c87 | 1590 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1591 | |
1592 | /* | |
672bba3a CL |
1593 | * The defrag ratio allows a configuration of the tradeoffs between |
1594 | * inter node defragmentation and node local allocations. A lower | |
1595 | * defrag_ratio increases the tendency to do local allocations | |
1596 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1597 | * |
672bba3a CL |
1598 | * If the defrag_ratio is set to 0 then kmalloc() always |
1599 | * returns node local objects. If the ratio is higher then kmalloc() | |
1600 | * may return off node objects because partial slabs are obtained | |
1601 | * from other nodes and filled up. | |
81819f0f | 1602 | * |
6446faa2 | 1603 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1604 | * defrag_ratio = 1000) then every (well almost) allocation will |
1605 | * first attempt to defrag slab caches on other nodes. This means | |
1606 | * scanning over all nodes to look for partial slabs which may be | |
1607 | * expensive if we do it every time we are trying to find a slab | |
1608 | * with available objects. | |
81819f0f | 1609 | */ |
9824601e CL |
1610 | if (!s->remote_node_defrag_ratio || |
1611 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1612 | return NULL; |
1613 | ||
cc9a6c87 MG |
1614 | do { |
1615 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 1616 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
1617 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1618 | struct kmem_cache_node *n; | |
1619 | ||
1620 | n = get_node(s, zone_to_nid(zone)); | |
1621 | ||
1622 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1623 | n->nr_partial > s->min_partial) { | |
1624 | object = get_partial_node(s, n, c); | |
1625 | if (object) { | |
1626 | /* | |
1627 | * Return the object even if | |
1628 | * put_mems_allowed indicated that | |
1629 | * the cpuset mems_allowed was | |
1630 | * updated in parallel. It's a | |
1631 | * harmless race between the alloc | |
1632 | * and the cpuset update. | |
1633 | */ | |
1634 | put_mems_allowed(cpuset_mems_cookie); | |
1635 | return object; | |
1636 | } | |
c0ff7453 | 1637 | } |
81819f0f | 1638 | } |
cc9a6c87 | 1639 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1640 | #endif |
1641 | return NULL; | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * Get a partial page, lock it and return it. | |
1646 | */ | |
497b66f2 | 1647 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1648 | struct kmem_cache_cpu *c) |
81819f0f | 1649 | { |
497b66f2 | 1650 | void *object; |
2154a336 | 1651 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1652 | |
497b66f2 CL |
1653 | object = get_partial_node(s, get_node(s, searchnode), c); |
1654 | if (object || node != NUMA_NO_NODE) | |
1655 | return object; | |
81819f0f | 1656 | |
acd19fd1 | 1657 | return get_any_partial(s, flags, c); |
81819f0f CL |
1658 | } |
1659 | ||
8a5ec0ba CL |
1660 | #ifdef CONFIG_PREEMPT |
1661 | /* | |
1662 | * Calculate the next globally unique transaction for disambiguiation | |
1663 | * during cmpxchg. The transactions start with the cpu number and are then | |
1664 | * incremented by CONFIG_NR_CPUS. | |
1665 | */ | |
1666 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1667 | #else | |
1668 | /* | |
1669 | * No preemption supported therefore also no need to check for | |
1670 | * different cpus. | |
1671 | */ | |
1672 | #define TID_STEP 1 | |
1673 | #endif | |
1674 | ||
1675 | static inline unsigned long next_tid(unsigned long tid) | |
1676 | { | |
1677 | return tid + TID_STEP; | |
1678 | } | |
1679 | ||
1680 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1681 | { | |
1682 | return tid % TID_STEP; | |
1683 | } | |
1684 | ||
1685 | static inline unsigned long tid_to_event(unsigned long tid) | |
1686 | { | |
1687 | return tid / TID_STEP; | |
1688 | } | |
1689 | ||
1690 | static inline unsigned int init_tid(int cpu) | |
1691 | { | |
1692 | return cpu; | |
1693 | } | |
1694 | ||
1695 | static inline void note_cmpxchg_failure(const char *n, | |
1696 | const struct kmem_cache *s, unsigned long tid) | |
1697 | { | |
1698 | #ifdef SLUB_DEBUG_CMPXCHG | |
1699 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1700 | ||
1701 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1702 | ||
1703 | #ifdef CONFIG_PREEMPT | |
1704 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1705 | printk("due to cpu change %d -> %d\n", | |
1706 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1707 | else | |
1708 | #endif | |
1709 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1710 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1711 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1712 | else | |
1713 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1714 | actual_tid, tid, next_tid(tid)); | |
1715 | #endif | |
4fdccdfb | 1716 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1717 | } |
1718 | ||
8a5ec0ba CL |
1719 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1720 | { | |
8a5ec0ba CL |
1721 | int cpu; |
1722 | ||
1723 | for_each_possible_cpu(cpu) | |
1724 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1725 | } |
2cfb7455 | 1726 | |
81819f0f CL |
1727 | /* |
1728 | * Remove the cpu slab | |
1729 | */ | |
c17dda40 | 1730 | static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist) |
81819f0f | 1731 | { |
2cfb7455 | 1732 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1733 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1734 | int lock = 0; | |
1735 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1736 | void *nextfree; |
136333d1 | 1737 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1738 | struct page new; |
1739 | struct page old; | |
1740 | ||
1741 | if (page->freelist) { | |
84e554e6 | 1742 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1743 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1744 | } |
1745 | ||
894b8788 | 1746 | /* |
2cfb7455 CL |
1747 | * Stage one: Free all available per cpu objects back |
1748 | * to the page freelist while it is still frozen. Leave the | |
1749 | * last one. | |
1750 | * | |
1751 | * There is no need to take the list->lock because the page | |
1752 | * is still frozen. | |
1753 | */ | |
1754 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1755 | void *prior; | |
1756 | unsigned long counters; | |
1757 | ||
1758 | do { | |
1759 | prior = page->freelist; | |
1760 | counters = page->counters; | |
1761 | set_freepointer(s, freelist, prior); | |
1762 | new.counters = counters; | |
1763 | new.inuse--; | |
1764 | VM_BUG_ON(!new.frozen); | |
1765 | ||
1d07171c | 1766 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1767 | prior, counters, |
1768 | freelist, new.counters, | |
1769 | "drain percpu freelist")); | |
1770 | ||
1771 | freelist = nextfree; | |
1772 | } | |
1773 | ||
894b8788 | 1774 | /* |
2cfb7455 CL |
1775 | * Stage two: Ensure that the page is unfrozen while the |
1776 | * list presence reflects the actual number of objects | |
1777 | * during unfreeze. | |
1778 | * | |
1779 | * We setup the list membership and then perform a cmpxchg | |
1780 | * with the count. If there is a mismatch then the page | |
1781 | * is not unfrozen but the page is on the wrong list. | |
1782 | * | |
1783 | * Then we restart the process which may have to remove | |
1784 | * the page from the list that we just put it on again | |
1785 | * because the number of objects in the slab may have | |
1786 | * changed. | |
894b8788 | 1787 | */ |
2cfb7455 | 1788 | redo: |
894b8788 | 1789 | |
2cfb7455 CL |
1790 | old.freelist = page->freelist; |
1791 | old.counters = page->counters; | |
1792 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1793 | |
2cfb7455 CL |
1794 | /* Determine target state of the slab */ |
1795 | new.counters = old.counters; | |
1796 | if (freelist) { | |
1797 | new.inuse--; | |
1798 | set_freepointer(s, freelist, old.freelist); | |
1799 | new.freelist = freelist; | |
1800 | } else | |
1801 | new.freelist = old.freelist; | |
1802 | ||
1803 | new.frozen = 0; | |
1804 | ||
81107188 | 1805 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1806 | m = M_FREE; |
1807 | else if (new.freelist) { | |
1808 | m = M_PARTIAL; | |
1809 | if (!lock) { | |
1810 | lock = 1; | |
1811 | /* | |
1812 | * Taking the spinlock removes the possiblity | |
1813 | * that acquire_slab() will see a slab page that | |
1814 | * is frozen | |
1815 | */ | |
1816 | spin_lock(&n->list_lock); | |
1817 | } | |
1818 | } else { | |
1819 | m = M_FULL; | |
1820 | if (kmem_cache_debug(s) && !lock) { | |
1821 | lock = 1; | |
1822 | /* | |
1823 | * This also ensures that the scanning of full | |
1824 | * slabs from diagnostic functions will not see | |
1825 | * any frozen slabs. | |
1826 | */ | |
1827 | spin_lock(&n->list_lock); | |
1828 | } | |
1829 | } | |
1830 | ||
1831 | if (l != m) { | |
1832 | ||
1833 | if (l == M_PARTIAL) | |
1834 | ||
1835 | remove_partial(n, page); | |
1836 | ||
1837 | else if (l == M_FULL) | |
894b8788 | 1838 | |
2cfb7455 CL |
1839 | remove_full(s, page); |
1840 | ||
1841 | if (m == M_PARTIAL) { | |
1842 | ||
1843 | add_partial(n, page, tail); | |
136333d1 | 1844 | stat(s, tail); |
2cfb7455 CL |
1845 | |
1846 | } else if (m == M_FULL) { | |
894b8788 | 1847 | |
2cfb7455 CL |
1848 | stat(s, DEACTIVATE_FULL); |
1849 | add_full(s, n, page); | |
1850 | ||
1851 | } | |
1852 | } | |
1853 | ||
1854 | l = m; | |
1d07171c | 1855 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1856 | old.freelist, old.counters, |
1857 | new.freelist, new.counters, | |
1858 | "unfreezing slab")) | |
1859 | goto redo; | |
1860 | ||
2cfb7455 CL |
1861 | if (lock) |
1862 | spin_unlock(&n->list_lock); | |
1863 | ||
1864 | if (m == M_FREE) { | |
1865 | stat(s, DEACTIVATE_EMPTY); | |
1866 | discard_slab(s, page); | |
1867 | stat(s, FREE_SLAB); | |
894b8788 | 1868 | } |
81819f0f CL |
1869 | } |
1870 | ||
d24ac77f JK |
1871 | /* |
1872 | * Unfreeze all the cpu partial slabs. | |
1873 | * | |
1874 | * This function must be called with interrupt disabled. | |
1875 | */ | |
49e22585 CL |
1876 | static void unfreeze_partials(struct kmem_cache *s) |
1877 | { | |
43d77867 | 1878 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
49e22585 | 1879 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); |
9ada1934 | 1880 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1881 | |
1882 | while ((page = c->partial)) { | |
49e22585 CL |
1883 | struct page new; |
1884 | struct page old; | |
1885 | ||
1886 | c->partial = page->next; | |
43d77867 JK |
1887 | |
1888 | n2 = get_node(s, page_to_nid(page)); | |
1889 | if (n != n2) { | |
1890 | if (n) | |
1891 | spin_unlock(&n->list_lock); | |
1892 | ||
1893 | n = n2; | |
1894 | spin_lock(&n->list_lock); | |
1895 | } | |
49e22585 CL |
1896 | |
1897 | do { | |
1898 | ||
1899 | old.freelist = page->freelist; | |
1900 | old.counters = page->counters; | |
1901 | VM_BUG_ON(!old.frozen); | |
1902 | ||
1903 | new.counters = old.counters; | |
1904 | new.freelist = old.freelist; | |
1905 | ||
1906 | new.frozen = 0; | |
1907 | ||
d24ac77f | 1908 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1909 | old.freelist, old.counters, |
1910 | new.freelist, new.counters, | |
1911 | "unfreezing slab")); | |
1912 | ||
43d77867 | 1913 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1914 | page->next = discard_page; |
1915 | discard_page = page; | |
43d77867 JK |
1916 | } else { |
1917 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1918 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1919 | } |
1920 | } | |
1921 | ||
1922 | if (n) | |
1923 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1924 | |
1925 | while (discard_page) { | |
1926 | page = discard_page; | |
1927 | discard_page = discard_page->next; | |
1928 | ||
1929 | stat(s, DEACTIVATE_EMPTY); | |
1930 | discard_slab(s, page); | |
1931 | stat(s, FREE_SLAB); | |
1932 | } | |
49e22585 CL |
1933 | } |
1934 | ||
1935 | /* | |
1936 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1937 | * slot if available. This is done without interrupts disabled and without | |
1938 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1939 | * onto a random cpus partial slot. | |
1940 | * | |
1941 | * If we did not find a slot then simply move all the partials to the | |
1942 | * per node partial list. | |
1943 | */ | |
1944 | int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
1945 | { | |
1946 | struct page *oldpage; | |
1947 | int pages; | |
1948 | int pobjects; | |
1949 | ||
1950 | do { | |
1951 | pages = 0; | |
1952 | pobjects = 0; | |
1953 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
1954 | ||
1955 | if (oldpage) { | |
1956 | pobjects = oldpage->pobjects; | |
1957 | pages = oldpage->pages; | |
1958 | if (drain && pobjects > s->cpu_partial) { | |
1959 | unsigned long flags; | |
1960 | /* | |
1961 | * partial array is full. Move the existing | |
1962 | * set to the per node partial list. | |
1963 | */ | |
1964 | local_irq_save(flags); | |
1965 | unfreeze_partials(s); | |
1966 | local_irq_restore(flags); | |
e24fc410 | 1967 | oldpage = NULL; |
49e22585 CL |
1968 | pobjects = 0; |
1969 | pages = 0; | |
8028dcea | 1970 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
1971 | } |
1972 | } | |
1973 | ||
1974 | pages++; | |
1975 | pobjects += page->objects - page->inuse; | |
1976 | ||
1977 | page->pages = pages; | |
1978 | page->pobjects = pobjects; | |
1979 | page->next = oldpage; | |
1980 | ||
933393f5 | 1981 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
1982 | return pobjects; |
1983 | } | |
1984 | ||
dfb4f096 | 1985 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1986 | { |
84e554e6 | 1987 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
1988 | deactivate_slab(s, c->page, c->freelist); |
1989 | ||
1990 | c->tid = next_tid(c->tid); | |
1991 | c->page = NULL; | |
1992 | c->freelist = NULL; | |
81819f0f CL |
1993 | } |
1994 | ||
1995 | /* | |
1996 | * Flush cpu slab. | |
6446faa2 | 1997 | * |
81819f0f CL |
1998 | * Called from IPI handler with interrupts disabled. |
1999 | */ | |
0c710013 | 2000 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2001 | { |
9dfc6e68 | 2002 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2003 | |
49e22585 CL |
2004 | if (likely(c)) { |
2005 | if (c->page) | |
2006 | flush_slab(s, c); | |
2007 | ||
2008 | unfreeze_partials(s); | |
2009 | } | |
81819f0f CL |
2010 | } |
2011 | ||
2012 | static void flush_cpu_slab(void *d) | |
2013 | { | |
2014 | struct kmem_cache *s = d; | |
81819f0f | 2015 | |
dfb4f096 | 2016 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2017 | } |
2018 | ||
a8364d55 GBY |
2019 | static bool has_cpu_slab(int cpu, void *info) |
2020 | { | |
2021 | struct kmem_cache *s = info; | |
2022 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2023 | ||
02e1a9cd | 2024 | return c->page || c->partial; |
a8364d55 GBY |
2025 | } |
2026 | ||
81819f0f CL |
2027 | static void flush_all(struct kmem_cache *s) |
2028 | { | |
a8364d55 | 2029 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2030 | } |
2031 | ||
dfb4f096 CL |
2032 | /* |
2033 | * Check if the objects in a per cpu structure fit numa | |
2034 | * locality expectations. | |
2035 | */ | |
57d437d2 | 2036 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2037 | { |
2038 | #ifdef CONFIG_NUMA | |
57d437d2 | 2039 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2040 | return 0; |
2041 | #endif | |
2042 | return 1; | |
2043 | } | |
2044 | ||
781b2ba6 PE |
2045 | static int count_free(struct page *page) |
2046 | { | |
2047 | return page->objects - page->inuse; | |
2048 | } | |
2049 | ||
2050 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2051 | int (*get_count)(struct page *)) | |
2052 | { | |
2053 | unsigned long flags; | |
2054 | unsigned long x = 0; | |
2055 | struct page *page; | |
2056 | ||
2057 | spin_lock_irqsave(&n->list_lock, flags); | |
2058 | list_for_each_entry(page, &n->partial, lru) | |
2059 | x += get_count(page); | |
2060 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2061 | return x; | |
2062 | } | |
2063 | ||
26c02cf0 AB |
2064 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2065 | { | |
2066 | #ifdef CONFIG_SLUB_DEBUG | |
2067 | return atomic_long_read(&n->total_objects); | |
2068 | #else | |
2069 | return 0; | |
2070 | #endif | |
2071 | } | |
2072 | ||
781b2ba6 PE |
2073 | static noinline void |
2074 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2075 | { | |
2076 | int node; | |
2077 | ||
2078 | printk(KERN_WARNING | |
2079 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2080 | nid, gfpflags); | |
2081 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2082 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2083 | s->size, oo_order(s->oo), oo_order(s->min)); |
2084 | ||
3b0efdfa | 2085 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2086 | printk(KERN_WARNING " %s debugging increased min order, use " |
2087 | "slub_debug=O to disable.\n", s->name); | |
2088 | ||
781b2ba6 PE |
2089 | for_each_online_node(node) { |
2090 | struct kmem_cache_node *n = get_node(s, node); | |
2091 | unsigned long nr_slabs; | |
2092 | unsigned long nr_objs; | |
2093 | unsigned long nr_free; | |
2094 | ||
2095 | if (!n) | |
2096 | continue; | |
2097 | ||
26c02cf0 AB |
2098 | nr_free = count_partial(n, count_free); |
2099 | nr_slabs = node_nr_slabs(n); | |
2100 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2101 | |
2102 | printk(KERN_WARNING | |
2103 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2104 | node, nr_slabs, nr_objs, nr_free); | |
2105 | } | |
2106 | } | |
2107 | ||
497b66f2 CL |
2108 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2109 | int node, struct kmem_cache_cpu **pc) | |
2110 | { | |
6faa6833 | 2111 | void *freelist; |
188fd063 CL |
2112 | struct kmem_cache_cpu *c = *pc; |
2113 | struct page *page; | |
497b66f2 | 2114 | |
188fd063 | 2115 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2116 | |
188fd063 CL |
2117 | if (freelist) |
2118 | return freelist; | |
2119 | ||
2120 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2121 | if (page) { |
2122 | c = __this_cpu_ptr(s->cpu_slab); | |
2123 | if (c->page) | |
2124 | flush_slab(s, c); | |
2125 | ||
2126 | /* | |
2127 | * No other reference to the page yet so we can | |
2128 | * muck around with it freely without cmpxchg | |
2129 | */ | |
6faa6833 | 2130 | freelist = page->freelist; |
497b66f2 CL |
2131 | page->freelist = NULL; |
2132 | ||
2133 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2134 | c->page = page; |
2135 | *pc = c; | |
2136 | } else | |
6faa6833 | 2137 | freelist = NULL; |
497b66f2 | 2138 | |
6faa6833 | 2139 | return freelist; |
497b66f2 CL |
2140 | } |
2141 | ||
072bb0aa MG |
2142 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2143 | { | |
2144 | if (unlikely(PageSlabPfmemalloc(page))) | |
2145 | return gfp_pfmemalloc_allowed(gfpflags); | |
2146 | ||
2147 | return true; | |
2148 | } | |
2149 | ||
213eeb9f CL |
2150 | /* |
2151 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2152 | * or deactivate the page. | |
2153 | * | |
2154 | * The page is still frozen if the return value is not NULL. | |
2155 | * | |
2156 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2157 | * |
2158 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2159 | */ |
2160 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2161 | { | |
2162 | struct page new; | |
2163 | unsigned long counters; | |
2164 | void *freelist; | |
2165 | ||
2166 | do { | |
2167 | freelist = page->freelist; | |
2168 | counters = page->counters; | |
6faa6833 | 2169 | |
213eeb9f CL |
2170 | new.counters = counters; |
2171 | VM_BUG_ON(!new.frozen); | |
2172 | ||
2173 | new.inuse = page->objects; | |
2174 | new.frozen = freelist != NULL; | |
2175 | ||
d24ac77f | 2176 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2177 | freelist, counters, |
2178 | NULL, new.counters, | |
2179 | "get_freelist")); | |
2180 | ||
2181 | return freelist; | |
2182 | } | |
2183 | ||
81819f0f | 2184 | /* |
894b8788 CL |
2185 | * Slow path. The lockless freelist is empty or we need to perform |
2186 | * debugging duties. | |
2187 | * | |
894b8788 CL |
2188 | * Processing is still very fast if new objects have been freed to the |
2189 | * regular freelist. In that case we simply take over the regular freelist | |
2190 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2191 | * |
894b8788 CL |
2192 | * If that is not working then we fall back to the partial lists. We take the |
2193 | * first element of the freelist as the object to allocate now and move the | |
2194 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2195 | * |
894b8788 | 2196 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2197 | * we need to allocate a new slab. This is the slowest path since it involves |
2198 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2199 | */ |
ce71e27c EGM |
2200 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2201 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2202 | { |
6faa6833 | 2203 | void *freelist; |
f6e7def7 | 2204 | struct page *page; |
8a5ec0ba CL |
2205 | unsigned long flags; |
2206 | ||
2207 | local_irq_save(flags); | |
2208 | #ifdef CONFIG_PREEMPT | |
2209 | /* | |
2210 | * We may have been preempted and rescheduled on a different | |
2211 | * cpu before disabling interrupts. Need to reload cpu area | |
2212 | * pointer. | |
2213 | */ | |
2214 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2215 | #endif |
81819f0f | 2216 | |
f6e7def7 CL |
2217 | page = c->page; |
2218 | if (!page) | |
81819f0f | 2219 | goto new_slab; |
49e22585 | 2220 | redo: |
6faa6833 | 2221 | |
57d437d2 | 2222 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2223 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2224 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2225 | c->page = NULL; |
2226 | c->freelist = NULL; | |
fc59c053 CL |
2227 | goto new_slab; |
2228 | } | |
6446faa2 | 2229 | |
072bb0aa MG |
2230 | /* |
2231 | * By rights, we should be searching for a slab page that was | |
2232 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2233 | * information when the page leaves the per-cpu allocator | |
2234 | */ | |
2235 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2236 | deactivate_slab(s, page, c->freelist); | |
2237 | c->page = NULL; | |
2238 | c->freelist = NULL; | |
2239 | goto new_slab; | |
2240 | } | |
2241 | ||
73736e03 | 2242 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2243 | freelist = c->freelist; |
2244 | if (freelist) | |
73736e03 | 2245 | goto load_freelist; |
03e404af | 2246 | |
2cfb7455 | 2247 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2248 | |
f6e7def7 | 2249 | freelist = get_freelist(s, page); |
6446faa2 | 2250 | |
6faa6833 | 2251 | if (!freelist) { |
03e404af CL |
2252 | c->page = NULL; |
2253 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2254 | goto new_slab; |
03e404af | 2255 | } |
6446faa2 | 2256 | |
84e554e6 | 2257 | stat(s, ALLOC_REFILL); |
6446faa2 | 2258 | |
894b8788 | 2259 | load_freelist: |
507effea CL |
2260 | /* |
2261 | * freelist is pointing to the list of objects to be used. | |
2262 | * page is pointing to the page from which the objects are obtained. | |
2263 | * That page must be frozen for per cpu allocations to work. | |
2264 | */ | |
2265 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2266 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2267 | c->tid = next_tid(c->tid); |
2268 | local_irq_restore(flags); | |
6faa6833 | 2269 | return freelist; |
81819f0f | 2270 | |
81819f0f | 2271 | new_slab: |
2cfb7455 | 2272 | |
49e22585 | 2273 | if (c->partial) { |
f6e7def7 CL |
2274 | page = c->page = c->partial; |
2275 | c->partial = page->next; | |
49e22585 CL |
2276 | stat(s, CPU_PARTIAL_ALLOC); |
2277 | c->freelist = NULL; | |
2278 | goto redo; | |
81819f0f CL |
2279 | } |
2280 | ||
188fd063 | 2281 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2282 | |
f4697436 CL |
2283 | if (unlikely(!freelist)) { |
2284 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2285 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2286 | |
f4697436 CL |
2287 | local_irq_restore(flags); |
2288 | return NULL; | |
81819f0f | 2289 | } |
2cfb7455 | 2290 | |
f6e7def7 | 2291 | page = c->page; |
5091b74a | 2292 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2293 | goto load_freelist; |
2cfb7455 | 2294 | |
497b66f2 | 2295 | /* Only entered in the debug case */ |
5091b74a | 2296 | if (kmem_cache_debug(s) && !alloc_debug_processing(s, page, freelist, addr)) |
497b66f2 | 2297 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2298 | |
f6e7def7 | 2299 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2300 | c->page = NULL; |
2301 | c->freelist = NULL; | |
a71ae47a | 2302 | local_irq_restore(flags); |
6faa6833 | 2303 | return freelist; |
894b8788 CL |
2304 | } |
2305 | ||
2306 | /* | |
2307 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2308 | * have the fastpath folded into their functions. So no function call | |
2309 | * overhead for requests that can be satisfied on the fastpath. | |
2310 | * | |
2311 | * The fastpath works by first checking if the lockless freelist can be used. | |
2312 | * If not then __slab_alloc is called for slow processing. | |
2313 | * | |
2314 | * Otherwise we can simply pick the next object from the lockless free list. | |
2315 | */ | |
06428780 | 2316 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2317 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2318 | { |
894b8788 | 2319 | void **object; |
dfb4f096 | 2320 | struct kmem_cache_cpu *c; |
57d437d2 | 2321 | struct page *page; |
8a5ec0ba | 2322 | unsigned long tid; |
1f84260c | 2323 | |
c016b0bd | 2324 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2325 | return NULL; |
1f84260c | 2326 | |
8a5ec0ba | 2327 | redo: |
8a5ec0ba CL |
2328 | |
2329 | /* | |
2330 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2331 | * enabled. We may switch back and forth between cpus while | |
2332 | * reading from one cpu area. That does not matter as long | |
2333 | * as we end up on the original cpu again when doing the cmpxchg. | |
2334 | */ | |
9dfc6e68 | 2335 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2336 | |
8a5ec0ba CL |
2337 | /* |
2338 | * The transaction ids are globally unique per cpu and per operation on | |
2339 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2340 | * occurs on the right processor and that there was no operation on the | |
2341 | * linked list in between. | |
2342 | */ | |
2343 | tid = c->tid; | |
2344 | barrier(); | |
8a5ec0ba | 2345 | |
9dfc6e68 | 2346 | object = c->freelist; |
57d437d2 | 2347 | page = c->page; |
5091b74a | 2348 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2349 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2350 | |
2351 | else { | |
0ad9500e ED |
2352 | void *next_object = get_freepointer_safe(s, object); |
2353 | ||
8a5ec0ba | 2354 | /* |
25985edc | 2355 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2356 | * operation and if we are on the right processor. |
2357 | * | |
2358 | * The cmpxchg does the following atomically (without lock semantics!) | |
2359 | * 1. Relocate first pointer to the current per cpu area. | |
2360 | * 2. Verify that tid and freelist have not been changed | |
2361 | * 3. If they were not changed replace tid and freelist | |
2362 | * | |
2363 | * Since this is without lock semantics the protection is only against | |
2364 | * code executing on this cpu *not* from access by other cpus. | |
2365 | */ | |
933393f5 | 2366 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2367 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2368 | object, tid, | |
0ad9500e | 2369 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2370 | |
2371 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2372 | goto redo; | |
2373 | } | |
0ad9500e | 2374 | prefetch_freepointer(s, next_object); |
84e554e6 | 2375 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2376 | } |
8a5ec0ba | 2377 | |
74e2134f | 2378 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2379 | memset(object, 0, s->object_size); |
d07dbea4 | 2380 | |
c016b0bd | 2381 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2382 | |
894b8788 | 2383 | return object; |
81819f0f CL |
2384 | } |
2385 | ||
2386 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2387 | { | |
2154a336 | 2388 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2389 | |
3b0efdfa | 2390 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags); |
5b882be4 EGM |
2391 | |
2392 | return ret; | |
81819f0f CL |
2393 | } |
2394 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2395 | ||
0f24f128 | 2396 | #ifdef CONFIG_TRACING |
4a92379b RK |
2397 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2398 | { | |
2399 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2400 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2401 | return ret; | |
2402 | } | |
2403 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2404 | ||
2405 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2406 | { |
4a92379b RK |
2407 | void *ret = kmalloc_order(size, flags, order); |
2408 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2409 | return ret; | |
5b882be4 | 2410 | } |
4a92379b | 2411 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2412 | #endif |
2413 | ||
81819f0f CL |
2414 | #ifdef CONFIG_NUMA |
2415 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2416 | { | |
5b882be4 EGM |
2417 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2418 | ||
ca2b84cb | 2419 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2420 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2421 | |
2422 | return ret; | |
81819f0f CL |
2423 | } |
2424 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2425 | |
0f24f128 | 2426 | #ifdef CONFIG_TRACING |
4a92379b | 2427 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2428 | gfp_t gfpflags, |
4a92379b | 2429 | int node, size_t size) |
5b882be4 | 2430 | { |
4a92379b RK |
2431 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2432 | ||
2433 | trace_kmalloc_node(_RET_IP_, ret, | |
2434 | size, s->size, gfpflags, node); | |
2435 | return ret; | |
5b882be4 | 2436 | } |
4a92379b | 2437 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2438 | #endif |
5d1f57e4 | 2439 | #endif |
5b882be4 | 2440 | |
81819f0f | 2441 | /* |
894b8788 CL |
2442 | * Slow patch handling. This may still be called frequently since objects |
2443 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2444 | * |
894b8788 CL |
2445 | * So we still attempt to reduce cache line usage. Just take the slab |
2446 | * lock and free the item. If there is no additional partial page | |
2447 | * handling required then we can return immediately. | |
81819f0f | 2448 | */ |
894b8788 | 2449 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2450 | void *x, unsigned long addr) |
81819f0f CL |
2451 | { |
2452 | void *prior; | |
2453 | void **object = (void *)x; | |
2cfb7455 CL |
2454 | int was_frozen; |
2455 | int inuse; | |
2456 | struct page new; | |
2457 | unsigned long counters; | |
2458 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2459 | unsigned long uninitialized_var(flags); |
81819f0f | 2460 | |
8a5ec0ba | 2461 | stat(s, FREE_SLOWPATH); |
81819f0f | 2462 | |
19c7ff9e CL |
2463 | if (kmem_cache_debug(s) && |
2464 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2465 | return; |
6446faa2 | 2466 | |
2cfb7455 CL |
2467 | do { |
2468 | prior = page->freelist; | |
2469 | counters = page->counters; | |
2470 | set_freepointer(s, object, prior); | |
2471 | new.counters = counters; | |
2472 | was_frozen = new.frozen; | |
2473 | new.inuse--; | |
2474 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
49e22585 CL |
2475 | |
2476 | if (!kmem_cache_debug(s) && !prior) | |
2477 | ||
2478 | /* | |
2479 | * Slab was on no list before and will be partially empty | |
2480 | * We can defer the list move and instead freeze it. | |
2481 | */ | |
2482 | new.frozen = 1; | |
2483 | ||
2484 | else { /* Needs to be taken off a list */ | |
2485 | ||
2486 | n = get_node(s, page_to_nid(page)); | |
2487 | /* | |
2488 | * Speculatively acquire the list_lock. | |
2489 | * If the cmpxchg does not succeed then we may | |
2490 | * drop the list_lock without any processing. | |
2491 | * | |
2492 | * Otherwise the list_lock will synchronize with | |
2493 | * other processors updating the list of slabs. | |
2494 | */ | |
2495 | spin_lock_irqsave(&n->list_lock, flags); | |
2496 | ||
2497 | } | |
2cfb7455 CL |
2498 | } |
2499 | inuse = new.inuse; | |
81819f0f | 2500 | |
2cfb7455 CL |
2501 | } while (!cmpxchg_double_slab(s, page, |
2502 | prior, counters, | |
2503 | object, new.counters, | |
2504 | "__slab_free")); | |
81819f0f | 2505 | |
2cfb7455 | 2506 | if (likely(!n)) { |
49e22585 CL |
2507 | |
2508 | /* | |
2509 | * If we just froze the page then put it onto the | |
2510 | * per cpu partial list. | |
2511 | */ | |
8028dcea | 2512 | if (new.frozen && !was_frozen) { |
49e22585 | 2513 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2514 | stat(s, CPU_PARTIAL_FREE); |
2515 | } | |
49e22585 | 2516 | /* |
2cfb7455 CL |
2517 | * The list lock was not taken therefore no list |
2518 | * activity can be necessary. | |
2519 | */ | |
2520 | if (was_frozen) | |
2521 | stat(s, FREE_FROZEN); | |
80f08c19 | 2522 | return; |
2cfb7455 | 2523 | } |
81819f0f CL |
2524 | |
2525 | /* | |
2cfb7455 CL |
2526 | * was_frozen may have been set after we acquired the list_lock in |
2527 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2528 | */ |
2cfb7455 CL |
2529 | if (was_frozen) |
2530 | stat(s, FREE_FROZEN); | |
2531 | else { | |
2532 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2533 | goto slab_empty; | |
81819f0f | 2534 | |
2cfb7455 CL |
2535 | /* |
2536 | * Objects left in the slab. If it was not on the partial list before | |
2537 | * then add it. | |
2538 | */ | |
2539 | if (unlikely(!prior)) { | |
2540 | remove_full(s, page); | |
136333d1 | 2541 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2cfb7455 CL |
2542 | stat(s, FREE_ADD_PARTIAL); |
2543 | } | |
8ff12cfc | 2544 | } |
80f08c19 | 2545 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2546 | return; |
2547 | ||
2548 | slab_empty: | |
a973e9dd | 2549 | if (prior) { |
81819f0f | 2550 | /* |
6fbabb20 | 2551 | * Slab on the partial list. |
81819f0f | 2552 | */ |
5cc6eee8 | 2553 | remove_partial(n, page); |
84e554e6 | 2554 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2555 | } else |
2556 | /* Slab must be on the full list */ | |
2557 | remove_full(s, page); | |
2cfb7455 | 2558 | |
80f08c19 | 2559 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2560 | stat(s, FREE_SLAB); |
81819f0f | 2561 | discard_slab(s, page); |
81819f0f CL |
2562 | } |
2563 | ||
894b8788 CL |
2564 | /* |
2565 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2566 | * can perform fastpath freeing without additional function calls. | |
2567 | * | |
2568 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2569 | * of this processor. This typically the case if we have just allocated | |
2570 | * the item before. | |
2571 | * | |
2572 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2573 | * with all sorts of special processing. | |
2574 | */ | |
06428780 | 2575 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2576 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2577 | { |
2578 | void **object = (void *)x; | |
dfb4f096 | 2579 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2580 | unsigned long tid; |
1f84260c | 2581 | |
c016b0bd CL |
2582 | slab_free_hook(s, x); |
2583 | ||
8a5ec0ba CL |
2584 | redo: |
2585 | /* | |
2586 | * Determine the currently cpus per cpu slab. | |
2587 | * The cpu may change afterward. However that does not matter since | |
2588 | * data is retrieved via this pointer. If we are on the same cpu | |
2589 | * during the cmpxchg then the free will succedd. | |
2590 | */ | |
9dfc6e68 | 2591 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2592 | |
8a5ec0ba CL |
2593 | tid = c->tid; |
2594 | barrier(); | |
c016b0bd | 2595 | |
442b06bc | 2596 | if (likely(page == c->page)) { |
ff12059e | 2597 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2598 | |
933393f5 | 2599 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2600 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2601 | c->freelist, tid, | |
2602 | object, next_tid(tid)))) { | |
2603 | ||
2604 | note_cmpxchg_failure("slab_free", s, tid); | |
2605 | goto redo; | |
2606 | } | |
84e554e6 | 2607 | stat(s, FREE_FASTPATH); |
894b8788 | 2608 | } else |
ff12059e | 2609 | __slab_free(s, page, x, addr); |
894b8788 | 2610 | |
894b8788 CL |
2611 | } |
2612 | ||
81819f0f CL |
2613 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2614 | { | |
77c5e2d0 | 2615 | struct page *page; |
81819f0f | 2616 | |
b49af68f | 2617 | page = virt_to_head_page(x); |
81819f0f | 2618 | |
ce71e27c | 2619 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2620 | |
ca2b84cb | 2621 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2622 | } |
2623 | EXPORT_SYMBOL(kmem_cache_free); | |
2624 | ||
81819f0f | 2625 | /* |
672bba3a CL |
2626 | * Object placement in a slab is made very easy because we always start at |
2627 | * offset 0. If we tune the size of the object to the alignment then we can | |
2628 | * get the required alignment by putting one properly sized object after | |
2629 | * another. | |
81819f0f CL |
2630 | * |
2631 | * Notice that the allocation order determines the sizes of the per cpu | |
2632 | * caches. Each processor has always one slab available for allocations. | |
2633 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2634 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2635 | * locking overhead. |
81819f0f CL |
2636 | */ |
2637 | ||
2638 | /* | |
2639 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2640 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2641 | * and increases the number of allocations possible without having to | |
2642 | * take the list_lock. | |
2643 | */ | |
2644 | static int slub_min_order; | |
114e9e89 | 2645 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2646 | static int slub_min_objects; |
81819f0f CL |
2647 | |
2648 | /* | |
2649 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2650 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2651 | */ |
2652 | static int slub_nomerge; | |
2653 | ||
81819f0f CL |
2654 | /* |
2655 | * Calculate the order of allocation given an slab object size. | |
2656 | * | |
672bba3a CL |
2657 | * The order of allocation has significant impact on performance and other |
2658 | * system components. Generally order 0 allocations should be preferred since | |
2659 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2660 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2661 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2662 | * would be wasted. |
2663 | * | |
2664 | * In order to reach satisfactory performance we must ensure that a minimum | |
2665 | * number of objects is in one slab. Otherwise we may generate too much | |
2666 | * activity on the partial lists which requires taking the list_lock. This is | |
2667 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2668 | * |
672bba3a CL |
2669 | * slub_max_order specifies the order where we begin to stop considering the |
2670 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2671 | * we try to keep the page order as low as possible. So we accept more waste | |
2672 | * of space in favor of a small page order. | |
81819f0f | 2673 | * |
672bba3a CL |
2674 | * Higher order allocations also allow the placement of more objects in a |
2675 | * slab and thereby reduce object handling overhead. If the user has | |
2676 | * requested a higher mininum order then we start with that one instead of | |
2677 | * the smallest order which will fit the object. | |
81819f0f | 2678 | */ |
5e6d444e | 2679 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2680 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2681 | { |
2682 | int order; | |
2683 | int rem; | |
6300ea75 | 2684 | int min_order = slub_min_order; |
81819f0f | 2685 | |
ab9a0f19 | 2686 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2687 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2688 | |
6300ea75 | 2689 | for (order = max(min_order, |
5e6d444e CL |
2690 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2691 | order <= max_order; order++) { | |
81819f0f | 2692 | |
5e6d444e | 2693 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2694 | |
ab9a0f19 | 2695 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2696 | continue; |
2697 | ||
ab9a0f19 | 2698 | rem = (slab_size - reserved) % size; |
81819f0f | 2699 | |
5e6d444e | 2700 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2701 | break; |
2702 | ||
2703 | } | |
672bba3a | 2704 | |
81819f0f CL |
2705 | return order; |
2706 | } | |
2707 | ||
ab9a0f19 | 2708 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2709 | { |
2710 | int order; | |
2711 | int min_objects; | |
2712 | int fraction; | |
e8120ff1 | 2713 | int max_objects; |
5e6d444e CL |
2714 | |
2715 | /* | |
2716 | * Attempt to find best configuration for a slab. This | |
2717 | * works by first attempting to generate a layout with | |
2718 | * the best configuration and backing off gradually. | |
2719 | * | |
2720 | * First we reduce the acceptable waste in a slab. Then | |
2721 | * we reduce the minimum objects required in a slab. | |
2722 | */ | |
2723 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2724 | if (!min_objects) |
2725 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2726 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2727 | min_objects = min(min_objects, max_objects); |
2728 | ||
5e6d444e | 2729 | while (min_objects > 1) { |
c124f5b5 | 2730 | fraction = 16; |
5e6d444e CL |
2731 | while (fraction >= 4) { |
2732 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2733 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2734 | if (order <= slub_max_order) |
2735 | return order; | |
2736 | fraction /= 2; | |
2737 | } | |
5086c389 | 2738 | min_objects--; |
5e6d444e CL |
2739 | } |
2740 | ||
2741 | /* | |
2742 | * We were unable to place multiple objects in a slab. Now | |
2743 | * lets see if we can place a single object there. | |
2744 | */ | |
ab9a0f19 | 2745 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2746 | if (order <= slub_max_order) |
2747 | return order; | |
2748 | ||
2749 | /* | |
2750 | * Doh this slab cannot be placed using slub_max_order. | |
2751 | */ | |
ab9a0f19 | 2752 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2753 | if (order < MAX_ORDER) |
5e6d444e CL |
2754 | return order; |
2755 | return -ENOSYS; | |
2756 | } | |
2757 | ||
81819f0f | 2758 | /* |
672bba3a | 2759 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2760 | */ |
2761 | static unsigned long calculate_alignment(unsigned long flags, | |
2762 | unsigned long align, unsigned long size) | |
2763 | { | |
2764 | /* | |
6446faa2 CL |
2765 | * If the user wants hardware cache aligned objects then follow that |
2766 | * suggestion if the object is sufficiently large. | |
81819f0f | 2767 | * |
6446faa2 CL |
2768 | * The hardware cache alignment cannot override the specified |
2769 | * alignment though. If that is greater then use it. | |
81819f0f | 2770 | */ |
b6210386 NP |
2771 | if (flags & SLAB_HWCACHE_ALIGN) { |
2772 | unsigned long ralign = cache_line_size(); | |
2773 | while (size <= ralign / 2) | |
2774 | ralign /= 2; | |
2775 | align = max(align, ralign); | |
2776 | } | |
81819f0f CL |
2777 | |
2778 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2779 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2780 | |
2781 | return ALIGN(align, sizeof(void *)); | |
2782 | } | |
2783 | ||
5595cffc | 2784 | static void |
4053497d | 2785 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2786 | { |
2787 | n->nr_partial = 0; | |
81819f0f CL |
2788 | spin_lock_init(&n->list_lock); |
2789 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2790 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2791 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2792 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2793 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2794 | #endif |
81819f0f CL |
2795 | } |
2796 | ||
55136592 | 2797 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2798 | { |
6c182dc0 CL |
2799 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2800 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2801 | |
8a5ec0ba | 2802 | /* |
d4d84fef CM |
2803 | * Must align to double word boundary for the double cmpxchg |
2804 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2805 | */ |
d4d84fef CM |
2806 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2807 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2808 | |
2809 | if (!s->cpu_slab) | |
2810 | return 0; | |
2811 | ||
2812 | init_kmem_cache_cpus(s); | |
4c93c355 | 2813 | |
8a5ec0ba | 2814 | return 1; |
4c93c355 | 2815 | } |
4c93c355 | 2816 | |
51df1142 CL |
2817 | static struct kmem_cache *kmem_cache_node; |
2818 | ||
81819f0f CL |
2819 | /* |
2820 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2821 | * slab on the node for this slabcache. There are no concurrent accesses | |
2822 | * possible. | |
2823 | * | |
2824 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2825 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2826 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2827 | */ |
55136592 | 2828 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2829 | { |
2830 | struct page *page; | |
2831 | struct kmem_cache_node *n; | |
2832 | ||
51df1142 | 2833 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2834 | |
51df1142 | 2835 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2836 | |
2837 | BUG_ON(!page); | |
a2f92ee7 CL |
2838 | if (page_to_nid(page) != node) { |
2839 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2840 | "node %d\n", node); | |
2841 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2842 | "in order to be able to continue\n"); | |
2843 | } | |
2844 | ||
81819f0f CL |
2845 | n = page->freelist; |
2846 | BUG_ON(!n); | |
51df1142 | 2847 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2848 | page->inuse = 1; |
8cb0a506 | 2849 | page->frozen = 0; |
51df1142 | 2850 | kmem_cache_node->node[node] = n; |
8ab1372f | 2851 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2852 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2853 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2854 | #endif |
4053497d | 2855 | init_kmem_cache_node(n); |
51df1142 | 2856 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2857 | |
136333d1 | 2858 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2859 | } |
2860 | ||
2861 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2862 | { | |
2863 | int node; | |
2864 | ||
f64dc58c | 2865 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2866 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2867 | |
73367bd8 | 2868 | if (n) |
51df1142 CL |
2869 | kmem_cache_free(kmem_cache_node, n); |
2870 | ||
81819f0f CL |
2871 | s->node[node] = NULL; |
2872 | } | |
2873 | } | |
2874 | ||
55136592 | 2875 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2876 | { |
2877 | int node; | |
81819f0f | 2878 | |
f64dc58c | 2879 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2880 | struct kmem_cache_node *n; |
2881 | ||
73367bd8 | 2882 | if (slab_state == DOWN) { |
55136592 | 2883 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2884 | continue; |
2885 | } | |
51df1142 | 2886 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2887 | GFP_KERNEL, node); |
81819f0f | 2888 | |
73367bd8 AD |
2889 | if (!n) { |
2890 | free_kmem_cache_nodes(s); | |
2891 | return 0; | |
81819f0f | 2892 | } |
73367bd8 | 2893 | |
81819f0f | 2894 | s->node[node] = n; |
4053497d | 2895 | init_kmem_cache_node(n); |
81819f0f CL |
2896 | } |
2897 | return 1; | |
2898 | } | |
81819f0f | 2899 | |
c0bdb232 | 2900 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2901 | { |
2902 | if (min < MIN_PARTIAL) | |
2903 | min = MIN_PARTIAL; | |
2904 | else if (min > MAX_PARTIAL) | |
2905 | min = MAX_PARTIAL; | |
2906 | s->min_partial = min; | |
2907 | } | |
2908 | ||
81819f0f CL |
2909 | /* |
2910 | * calculate_sizes() determines the order and the distribution of data within | |
2911 | * a slab object. | |
2912 | */ | |
06b285dc | 2913 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2914 | { |
2915 | unsigned long flags = s->flags; | |
3b0efdfa | 2916 | unsigned long size = s->object_size; |
81819f0f | 2917 | unsigned long align = s->align; |
834f3d11 | 2918 | int order; |
81819f0f | 2919 | |
d8b42bf5 CL |
2920 | /* |
2921 | * Round up object size to the next word boundary. We can only | |
2922 | * place the free pointer at word boundaries and this determines | |
2923 | * the possible location of the free pointer. | |
2924 | */ | |
2925 | size = ALIGN(size, sizeof(void *)); | |
2926 | ||
2927 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2928 | /* |
2929 | * Determine if we can poison the object itself. If the user of | |
2930 | * the slab may touch the object after free or before allocation | |
2931 | * then we should never poison the object itself. | |
2932 | */ | |
2933 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2934 | !s->ctor) |
81819f0f CL |
2935 | s->flags |= __OBJECT_POISON; |
2936 | else | |
2937 | s->flags &= ~__OBJECT_POISON; | |
2938 | ||
81819f0f CL |
2939 | |
2940 | /* | |
672bba3a | 2941 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2942 | * end of the object and the free pointer. If not then add an |
672bba3a | 2943 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2944 | */ |
3b0efdfa | 2945 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2946 | size += sizeof(void *); |
41ecc55b | 2947 | #endif |
81819f0f CL |
2948 | |
2949 | /* | |
672bba3a CL |
2950 | * With that we have determined the number of bytes in actual use |
2951 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2952 | */ |
2953 | s->inuse = size; | |
2954 | ||
2955 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2956 | s->ctor)) { |
81819f0f CL |
2957 | /* |
2958 | * Relocate free pointer after the object if it is not | |
2959 | * permitted to overwrite the first word of the object on | |
2960 | * kmem_cache_free. | |
2961 | * | |
2962 | * This is the case if we do RCU, have a constructor or | |
2963 | * destructor or are poisoning the objects. | |
2964 | */ | |
2965 | s->offset = size; | |
2966 | size += sizeof(void *); | |
2967 | } | |
2968 | ||
c12b3c62 | 2969 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2970 | if (flags & SLAB_STORE_USER) |
2971 | /* | |
2972 | * Need to store information about allocs and frees after | |
2973 | * the object. | |
2974 | */ | |
2975 | size += 2 * sizeof(struct track); | |
2976 | ||
be7b3fbc | 2977 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2978 | /* |
2979 | * Add some empty padding so that we can catch | |
2980 | * overwrites from earlier objects rather than let | |
2981 | * tracking information or the free pointer be | |
0211a9c8 | 2982 | * corrupted if a user writes before the start |
81819f0f CL |
2983 | * of the object. |
2984 | */ | |
2985 | size += sizeof(void *); | |
41ecc55b | 2986 | #endif |
672bba3a | 2987 | |
81819f0f CL |
2988 | /* |
2989 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2990 | * user specified and the dynamic determination of cache line size |
2991 | * on bootup. | |
81819f0f | 2992 | */ |
3b0efdfa | 2993 | align = calculate_alignment(flags, align, s->object_size); |
dcb0ce1b | 2994 | s->align = align; |
81819f0f CL |
2995 | |
2996 | /* | |
2997 | * SLUB stores one object immediately after another beginning from | |
2998 | * offset 0. In order to align the objects we have to simply size | |
2999 | * each object to conform to the alignment. | |
3000 | */ | |
3001 | size = ALIGN(size, align); | |
3002 | s->size = size; | |
06b285dc CL |
3003 | if (forced_order >= 0) |
3004 | order = forced_order; | |
3005 | else | |
ab9a0f19 | 3006 | order = calculate_order(size, s->reserved); |
81819f0f | 3007 | |
834f3d11 | 3008 | if (order < 0) |
81819f0f CL |
3009 | return 0; |
3010 | ||
b7a49f0d | 3011 | s->allocflags = 0; |
834f3d11 | 3012 | if (order) |
b7a49f0d CL |
3013 | s->allocflags |= __GFP_COMP; |
3014 | ||
3015 | if (s->flags & SLAB_CACHE_DMA) | |
3016 | s->allocflags |= SLUB_DMA; | |
3017 | ||
3018 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3019 | s->allocflags |= __GFP_RECLAIMABLE; | |
3020 | ||
81819f0f CL |
3021 | /* |
3022 | * Determine the number of objects per slab | |
3023 | */ | |
ab9a0f19 LJ |
3024 | s->oo = oo_make(order, size, s->reserved); |
3025 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3026 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3027 | s->max = s->oo; | |
81819f0f | 3028 | |
834f3d11 | 3029 | return !!oo_objects(s->oo); |
81819f0f CL |
3030 | |
3031 | } | |
3032 | ||
55136592 | 3033 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
3034 | const char *name, size_t size, |
3035 | size_t align, unsigned long flags, | |
51cc5068 | 3036 | void (*ctor)(void *)) |
81819f0f CL |
3037 | { |
3038 | memset(s, 0, kmem_size); | |
3039 | s->name = name; | |
3040 | s->ctor = ctor; | |
3b0efdfa | 3041 | s->object_size = size; |
81819f0f | 3042 | s->align = align; |
ba0268a8 | 3043 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 3044 | s->reserved = 0; |
81819f0f | 3045 | |
da9a638c LJ |
3046 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3047 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3048 | |
06b285dc | 3049 | if (!calculate_sizes(s, -1)) |
81819f0f | 3050 | goto error; |
3de47213 DR |
3051 | if (disable_higher_order_debug) { |
3052 | /* | |
3053 | * Disable debugging flags that store metadata if the min slab | |
3054 | * order increased. | |
3055 | */ | |
3b0efdfa | 3056 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3057 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3058 | s->offset = 0; | |
3059 | if (!calculate_sizes(s, -1)) | |
3060 | goto error; | |
3061 | } | |
3062 | } | |
81819f0f | 3063 | |
2565409f HC |
3064 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3065 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3066 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3067 | /* Enable fast mode */ | |
3068 | s->flags |= __CMPXCHG_DOUBLE; | |
3069 | #endif | |
3070 | ||
3b89d7d8 DR |
3071 | /* |
3072 | * The larger the object size is, the more pages we want on the partial | |
3073 | * list to avoid pounding the page allocator excessively. | |
3074 | */ | |
49e22585 CL |
3075 | set_min_partial(s, ilog2(s->size) / 2); |
3076 | ||
3077 | /* | |
3078 | * cpu_partial determined the maximum number of objects kept in the | |
3079 | * per cpu partial lists of a processor. | |
3080 | * | |
3081 | * Per cpu partial lists mainly contain slabs that just have one | |
3082 | * object freed. If they are used for allocation then they can be | |
3083 | * filled up again with minimal effort. The slab will never hit the | |
3084 | * per node partial lists and therefore no locking will be required. | |
3085 | * | |
3086 | * This setting also determines | |
3087 | * | |
3088 | * A) The number of objects from per cpu partial slabs dumped to the | |
3089 | * per node list when we reach the limit. | |
9f264904 | 3090 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3091 | * per node list when we run out of per cpu objects. We only fetch 50% |
3092 | * to keep some capacity around for frees. | |
3093 | */ | |
8f1e33da CL |
3094 | if (kmem_cache_debug(s)) |
3095 | s->cpu_partial = 0; | |
3096 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3097 | s->cpu_partial = 2; |
3098 | else if (s->size >= 1024) | |
3099 | s->cpu_partial = 6; | |
3100 | else if (s->size >= 256) | |
3101 | s->cpu_partial = 13; | |
3102 | else | |
3103 | s->cpu_partial = 30; | |
3104 | ||
81819f0f CL |
3105 | s->refcount = 1; |
3106 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 3107 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3108 | #endif |
55136592 | 3109 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3110 | goto error; |
81819f0f | 3111 | |
55136592 | 3112 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 3113 | return 1; |
ff12059e | 3114 | |
4c93c355 | 3115 | free_kmem_cache_nodes(s); |
81819f0f CL |
3116 | error: |
3117 | if (flags & SLAB_PANIC) | |
3118 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3119 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 3120 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
3121 | s->offset, flags); |
3122 | return 0; | |
3123 | } | |
81819f0f | 3124 | |
81819f0f CL |
3125 | /* |
3126 | * Determine the size of a slab object | |
3127 | */ | |
3128 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
3129 | { | |
3b0efdfa | 3130 | return s->object_size; |
81819f0f CL |
3131 | } |
3132 | EXPORT_SYMBOL(kmem_cache_size); | |
3133 | ||
33b12c38 CL |
3134 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3135 | const char *text) | |
3136 | { | |
3137 | #ifdef CONFIG_SLUB_DEBUG | |
3138 | void *addr = page_address(page); | |
3139 | void *p; | |
a5dd5c11 NK |
3140 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3141 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3142 | if (!map) |
3143 | return; | |
33b12c38 CL |
3144 | slab_err(s, page, "%s", text); |
3145 | slab_lock(page); | |
33b12c38 | 3146 | |
5f80b13a | 3147 | get_map(s, page, map); |
33b12c38 CL |
3148 | for_each_object(p, s, addr, page->objects) { |
3149 | ||
3150 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3151 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3152 | p, p - addr); | |
3153 | print_tracking(s, p); | |
3154 | } | |
3155 | } | |
3156 | slab_unlock(page); | |
bbd7d57b | 3157 | kfree(map); |
33b12c38 CL |
3158 | #endif |
3159 | } | |
3160 | ||
81819f0f | 3161 | /* |
599870b1 | 3162 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3163 | * This is called from kmem_cache_close(). We must be the last thread |
3164 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3165 | */ |
599870b1 | 3166 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3167 | { |
81819f0f CL |
3168 | struct page *page, *h; |
3169 | ||
33b12c38 | 3170 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3171 | if (!page->inuse) { |
5cc6eee8 | 3172 | remove_partial(n, page); |
81819f0f | 3173 | discard_slab(s, page); |
33b12c38 CL |
3174 | } else { |
3175 | list_slab_objects(s, page, | |
3176 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 3177 | } |
33b12c38 | 3178 | } |
81819f0f CL |
3179 | } |
3180 | ||
3181 | /* | |
672bba3a | 3182 | * Release all resources used by a slab cache. |
81819f0f | 3183 | */ |
0c710013 | 3184 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3185 | { |
3186 | int node; | |
3187 | ||
3188 | flush_all(s); | |
9dfc6e68 | 3189 | free_percpu(s->cpu_slab); |
81819f0f | 3190 | /* Attempt to free all objects */ |
f64dc58c | 3191 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3192 | struct kmem_cache_node *n = get_node(s, node); |
3193 | ||
599870b1 CL |
3194 | free_partial(s, n); |
3195 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3196 | return 1; |
3197 | } | |
3198 | free_kmem_cache_nodes(s); | |
3199 | return 0; | |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * Close a cache and release the kmem_cache structure | |
3204 | * (must be used for caches created using kmem_cache_create) | |
3205 | */ | |
3206 | void kmem_cache_destroy(struct kmem_cache *s) | |
3207 | { | |
18004c5d | 3208 | mutex_lock(&slab_mutex); |
81819f0f CL |
3209 | s->refcount--; |
3210 | if (!s->refcount) { | |
3211 | list_del(&s->list); | |
18004c5d | 3212 | mutex_unlock(&slab_mutex); |
d629d819 PE |
3213 | if (kmem_cache_close(s)) { |
3214 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3215 | "still has objects.\n", s->name, __func__); | |
3216 | dump_stack(); | |
3217 | } | |
d76b1590 ED |
3218 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3219 | rcu_barrier(); | |
81819f0f | 3220 | sysfs_slab_remove(s); |
69cb8e6b | 3221 | } else |
18004c5d | 3222 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3223 | } |
3224 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3225 | ||
3226 | /******************************************************************** | |
3227 | * Kmalloc subsystem | |
3228 | *******************************************************************/ | |
3229 | ||
51df1142 | 3230 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3231 | EXPORT_SYMBOL(kmalloc_caches); |
3232 | ||
51df1142 CL |
3233 | static struct kmem_cache *kmem_cache; |
3234 | ||
55136592 | 3235 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3236 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3237 | #endif |
3238 | ||
81819f0f CL |
3239 | static int __init setup_slub_min_order(char *str) |
3240 | { | |
06428780 | 3241 | get_option(&str, &slub_min_order); |
81819f0f CL |
3242 | |
3243 | return 1; | |
3244 | } | |
3245 | ||
3246 | __setup("slub_min_order=", setup_slub_min_order); | |
3247 | ||
3248 | static int __init setup_slub_max_order(char *str) | |
3249 | { | |
06428780 | 3250 | get_option(&str, &slub_max_order); |
818cf590 | 3251 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3252 | |
3253 | return 1; | |
3254 | } | |
3255 | ||
3256 | __setup("slub_max_order=", setup_slub_max_order); | |
3257 | ||
3258 | static int __init setup_slub_min_objects(char *str) | |
3259 | { | |
06428780 | 3260 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3261 | |
3262 | return 1; | |
3263 | } | |
3264 | ||
3265 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3266 | ||
3267 | static int __init setup_slub_nomerge(char *str) | |
3268 | { | |
3269 | slub_nomerge = 1; | |
3270 | return 1; | |
3271 | } | |
3272 | ||
3273 | __setup("slub_nomerge", setup_slub_nomerge); | |
3274 | ||
51df1142 CL |
3275 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3276 | int size, unsigned int flags) | |
81819f0f | 3277 | { |
51df1142 CL |
3278 | struct kmem_cache *s; |
3279 | ||
3280 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3281 | ||
83b519e8 PE |
3282 | /* |
3283 | * This function is called with IRQs disabled during early-boot on | |
18004c5d | 3284 | * single CPU so there's no need to take slab_mutex here. |
83b519e8 | 3285 | */ |
55136592 | 3286 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3287 | flags, NULL)) |
81819f0f CL |
3288 | goto panic; |
3289 | ||
3290 | list_add(&s->list, &slab_caches); | |
51df1142 | 3291 | return s; |
81819f0f CL |
3292 | |
3293 | panic: | |
3294 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3295 | return NULL; |
81819f0f CL |
3296 | } |
3297 | ||
f1b26339 CL |
3298 | /* |
3299 | * Conversion table for small slabs sizes / 8 to the index in the | |
3300 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3301 | * of two cache sizes there. The size of larger slabs can be determined using | |
3302 | * fls. | |
3303 | */ | |
3304 | static s8 size_index[24] = { | |
3305 | 3, /* 8 */ | |
3306 | 4, /* 16 */ | |
3307 | 5, /* 24 */ | |
3308 | 5, /* 32 */ | |
3309 | 6, /* 40 */ | |
3310 | 6, /* 48 */ | |
3311 | 6, /* 56 */ | |
3312 | 6, /* 64 */ | |
3313 | 1, /* 72 */ | |
3314 | 1, /* 80 */ | |
3315 | 1, /* 88 */ | |
3316 | 1, /* 96 */ | |
3317 | 7, /* 104 */ | |
3318 | 7, /* 112 */ | |
3319 | 7, /* 120 */ | |
3320 | 7, /* 128 */ | |
3321 | 2, /* 136 */ | |
3322 | 2, /* 144 */ | |
3323 | 2, /* 152 */ | |
3324 | 2, /* 160 */ | |
3325 | 2, /* 168 */ | |
3326 | 2, /* 176 */ | |
3327 | 2, /* 184 */ | |
3328 | 2 /* 192 */ | |
3329 | }; | |
3330 | ||
acdfcd04 AK |
3331 | static inline int size_index_elem(size_t bytes) |
3332 | { | |
3333 | return (bytes - 1) / 8; | |
3334 | } | |
3335 | ||
81819f0f CL |
3336 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3337 | { | |
f1b26339 | 3338 | int index; |
81819f0f | 3339 | |
f1b26339 CL |
3340 | if (size <= 192) { |
3341 | if (!size) | |
3342 | return ZERO_SIZE_PTR; | |
81819f0f | 3343 | |
acdfcd04 | 3344 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3345 | } else |
f1b26339 | 3346 | index = fls(size - 1); |
81819f0f CL |
3347 | |
3348 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3349 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3350 | return kmalloc_dma_caches[index]; |
f1b26339 | 3351 | |
81819f0f | 3352 | #endif |
51df1142 | 3353 | return kmalloc_caches[index]; |
81819f0f CL |
3354 | } |
3355 | ||
3356 | void *__kmalloc(size_t size, gfp_t flags) | |
3357 | { | |
aadb4bc4 | 3358 | struct kmem_cache *s; |
5b882be4 | 3359 | void *ret; |
81819f0f | 3360 | |
ffadd4d0 | 3361 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3362 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3363 | |
3364 | s = get_slab(size, flags); | |
3365 | ||
3366 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3367 | return s; |
3368 | ||
2154a336 | 3369 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3370 | |
ca2b84cb | 3371 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3372 | |
3373 | return ret; | |
81819f0f CL |
3374 | } |
3375 | EXPORT_SYMBOL(__kmalloc); | |
3376 | ||
5d1f57e4 | 3377 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3378 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3379 | { | |
b1eeab67 | 3380 | struct page *page; |
e4f7c0b4 | 3381 | void *ptr = NULL; |
f619cfe1 | 3382 | |
b1eeab67 VN |
3383 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3384 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3385 | if (page) |
e4f7c0b4 CM |
3386 | ptr = page_address(page); |
3387 | ||
3388 | kmemleak_alloc(ptr, size, 1, flags); | |
3389 | return ptr; | |
f619cfe1 CL |
3390 | } |
3391 | ||
81819f0f CL |
3392 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3393 | { | |
aadb4bc4 | 3394 | struct kmem_cache *s; |
5b882be4 | 3395 | void *ret; |
81819f0f | 3396 | |
057685cf | 3397 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3398 | ret = kmalloc_large_node(size, flags, node); |
3399 | ||
ca2b84cb EGM |
3400 | trace_kmalloc_node(_RET_IP_, ret, |
3401 | size, PAGE_SIZE << get_order(size), | |
3402 | flags, node); | |
5b882be4 EGM |
3403 | |
3404 | return ret; | |
3405 | } | |
aadb4bc4 CL |
3406 | |
3407 | s = get_slab(size, flags); | |
3408 | ||
3409 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3410 | return s; |
3411 | ||
5b882be4 EGM |
3412 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3413 | ||
ca2b84cb | 3414 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3415 | |
3416 | return ret; | |
81819f0f CL |
3417 | } |
3418 | EXPORT_SYMBOL(__kmalloc_node); | |
3419 | #endif | |
3420 | ||
3421 | size_t ksize(const void *object) | |
3422 | { | |
272c1d21 | 3423 | struct page *page; |
81819f0f | 3424 | |
ef8b4520 | 3425 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3426 | return 0; |
3427 | ||
294a80a8 | 3428 | page = virt_to_head_page(object); |
294a80a8 | 3429 | |
76994412 PE |
3430 | if (unlikely(!PageSlab(page))) { |
3431 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3432 | return PAGE_SIZE << compound_order(page); |
76994412 | 3433 | } |
81819f0f | 3434 | |
b3d41885 | 3435 | return slab_ksize(page->slab); |
81819f0f | 3436 | } |
b1aabecd | 3437 | EXPORT_SYMBOL(ksize); |
81819f0f | 3438 | |
d18a90dd BG |
3439 | #ifdef CONFIG_SLUB_DEBUG |
3440 | bool verify_mem_not_deleted(const void *x) | |
3441 | { | |
3442 | struct page *page; | |
3443 | void *object = (void *)x; | |
3444 | unsigned long flags; | |
3445 | bool rv; | |
3446 | ||
3447 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3448 | return false; | |
3449 | ||
3450 | local_irq_save(flags); | |
3451 | ||
3452 | page = virt_to_head_page(x); | |
3453 | if (unlikely(!PageSlab(page))) { | |
3454 | /* maybe it was from stack? */ | |
3455 | rv = true; | |
3456 | goto out_unlock; | |
3457 | } | |
3458 | ||
3459 | slab_lock(page); | |
3460 | if (on_freelist(page->slab, page, object)) { | |
3461 | object_err(page->slab, page, object, "Object is on free-list"); | |
3462 | rv = false; | |
3463 | } else { | |
3464 | rv = true; | |
3465 | } | |
3466 | slab_unlock(page); | |
3467 | ||
3468 | out_unlock: | |
3469 | local_irq_restore(flags); | |
3470 | return rv; | |
3471 | } | |
3472 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3473 | #endif | |
3474 | ||
81819f0f CL |
3475 | void kfree(const void *x) |
3476 | { | |
81819f0f | 3477 | struct page *page; |
5bb983b0 | 3478 | void *object = (void *)x; |
81819f0f | 3479 | |
2121db74 PE |
3480 | trace_kfree(_RET_IP_, x); |
3481 | ||
2408c550 | 3482 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3483 | return; |
3484 | ||
b49af68f | 3485 | page = virt_to_head_page(x); |
aadb4bc4 | 3486 | if (unlikely(!PageSlab(page))) { |
0937502a | 3487 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3488 | kmemleak_free(x); |
d9b7f226 | 3489 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3490 | return; |
3491 | } | |
ce71e27c | 3492 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3493 | } |
3494 | EXPORT_SYMBOL(kfree); | |
3495 | ||
2086d26a | 3496 | /* |
672bba3a CL |
3497 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3498 | * the remaining slabs by the number of items in use. The slabs with the | |
3499 | * most items in use come first. New allocations will then fill those up | |
3500 | * and thus they can be removed from the partial lists. | |
3501 | * | |
3502 | * The slabs with the least items are placed last. This results in them | |
3503 | * being allocated from last increasing the chance that the last objects | |
3504 | * are freed in them. | |
2086d26a CL |
3505 | */ |
3506 | int kmem_cache_shrink(struct kmem_cache *s) | |
3507 | { | |
3508 | int node; | |
3509 | int i; | |
3510 | struct kmem_cache_node *n; | |
3511 | struct page *page; | |
3512 | struct page *t; | |
205ab99d | 3513 | int objects = oo_objects(s->max); |
2086d26a | 3514 | struct list_head *slabs_by_inuse = |
834f3d11 | 3515 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3516 | unsigned long flags; |
3517 | ||
3518 | if (!slabs_by_inuse) | |
3519 | return -ENOMEM; | |
3520 | ||
3521 | flush_all(s); | |
f64dc58c | 3522 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3523 | n = get_node(s, node); |
3524 | ||
3525 | if (!n->nr_partial) | |
3526 | continue; | |
3527 | ||
834f3d11 | 3528 | for (i = 0; i < objects; i++) |
2086d26a CL |
3529 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3530 | ||
3531 | spin_lock_irqsave(&n->list_lock, flags); | |
3532 | ||
3533 | /* | |
672bba3a | 3534 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3535 | * |
672bba3a CL |
3536 | * Note that concurrent frees may occur while we hold the |
3537 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3538 | */ |
3539 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3540 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3541 | if (!page->inuse) | |
3542 | n->nr_partial--; | |
2086d26a CL |
3543 | } |
3544 | ||
2086d26a | 3545 | /* |
672bba3a CL |
3546 | * Rebuild the partial list with the slabs filled up most |
3547 | * first and the least used slabs at the end. | |
2086d26a | 3548 | */ |
69cb8e6b | 3549 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3550 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3551 | ||
2086d26a | 3552 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3553 | |
3554 | /* Release empty slabs */ | |
3555 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3556 | discard_slab(s, page); | |
2086d26a CL |
3557 | } |
3558 | ||
3559 | kfree(slabs_by_inuse); | |
3560 | return 0; | |
3561 | } | |
3562 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3563 | ||
92a5bbc1 | 3564 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3565 | static int slab_mem_going_offline_callback(void *arg) |
3566 | { | |
3567 | struct kmem_cache *s; | |
3568 | ||
18004c5d | 3569 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3570 | list_for_each_entry(s, &slab_caches, list) |
3571 | kmem_cache_shrink(s); | |
18004c5d | 3572 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3573 | |
3574 | return 0; | |
3575 | } | |
3576 | ||
3577 | static void slab_mem_offline_callback(void *arg) | |
3578 | { | |
3579 | struct kmem_cache_node *n; | |
3580 | struct kmem_cache *s; | |
3581 | struct memory_notify *marg = arg; | |
3582 | int offline_node; | |
3583 | ||
3584 | offline_node = marg->status_change_nid; | |
3585 | ||
3586 | /* | |
3587 | * If the node still has available memory. we need kmem_cache_node | |
3588 | * for it yet. | |
3589 | */ | |
3590 | if (offline_node < 0) | |
3591 | return; | |
3592 | ||
18004c5d | 3593 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3594 | list_for_each_entry(s, &slab_caches, list) { |
3595 | n = get_node(s, offline_node); | |
3596 | if (n) { | |
3597 | /* | |
3598 | * if n->nr_slabs > 0, slabs still exist on the node | |
3599 | * that is going down. We were unable to free them, | |
c9404c9c | 3600 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3601 | * callback. So, we must fail. |
3602 | */ | |
0f389ec6 | 3603 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3604 | |
3605 | s->node[offline_node] = NULL; | |
8de66a0c | 3606 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3607 | } |
3608 | } | |
18004c5d | 3609 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3610 | } |
3611 | ||
3612 | static int slab_mem_going_online_callback(void *arg) | |
3613 | { | |
3614 | struct kmem_cache_node *n; | |
3615 | struct kmem_cache *s; | |
3616 | struct memory_notify *marg = arg; | |
3617 | int nid = marg->status_change_nid; | |
3618 | int ret = 0; | |
3619 | ||
3620 | /* | |
3621 | * If the node's memory is already available, then kmem_cache_node is | |
3622 | * already created. Nothing to do. | |
3623 | */ | |
3624 | if (nid < 0) | |
3625 | return 0; | |
3626 | ||
3627 | /* | |
0121c619 | 3628 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3629 | * allocate a kmem_cache_node structure in order to bring the node |
3630 | * online. | |
3631 | */ | |
18004c5d | 3632 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3633 | list_for_each_entry(s, &slab_caches, list) { |
3634 | /* | |
3635 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3636 | * since memory is not yet available from the node that | |
3637 | * is brought up. | |
3638 | */ | |
8de66a0c | 3639 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3640 | if (!n) { |
3641 | ret = -ENOMEM; | |
3642 | goto out; | |
3643 | } | |
4053497d | 3644 | init_kmem_cache_node(n); |
b9049e23 YG |
3645 | s->node[nid] = n; |
3646 | } | |
3647 | out: | |
18004c5d | 3648 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3649 | return ret; |
3650 | } | |
3651 | ||
3652 | static int slab_memory_callback(struct notifier_block *self, | |
3653 | unsigned long action, void *arg) | |
3654 | { | |
3655 | int ret = 0; | |
3656 | ||
3657 | switch (action) { | |
3658 | case MEM_GOING_ONLINE: | |
3659 | ret = slab_mem_going_online_callback(arg); | |
3660 | break; | |
3661 | case MEM_GOING_OFFLINE: | |
3662 | ret = slab_mem_going_offline_callback(arg); | |
3663 | break; | |
3664 | case MEM_OFFLINE: | |
3665 | case MEM_CANCEL_ONLINE: | |
3666 | slab_mem_offline_callback(arg); | |
3667 | break; | |
3668 | case MEM_ONLINE: | |
3669 | case MEM_CANCEL_OFFLINE: | |
3670 | break; | |
3671 | } | |
dc19f9db KH |
3672 | if (ret) |
3673 | ret = notifier_from_errno(ret); | |
3674 | else | |
3675 | ret = NOTIFY_OK; | |
b9049e23 YG |
3676 | return ret; |
3677 | } | |
3678 | ||
3679 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3680 | ||
81819f0f CL |
3681 | /******************************************************************** |
3682 | * Basic setup of slabs | |
3683 | *******************************************************************/ | |
3684 | ||
51df1142 CL |
3685 | /* |
3686 | * Used for early kmem_cache structures that were allocated using | |
3687 | * the page allocator | |
3688 | */ | |
3689 | ||
3690 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3691 | { | |
3692 | int node; | |
3693 | ||
3694 | list_add(&s->list, &slab_caches); | |
3695 | s->refcount = -1; | |
3696 | ||
3697 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3698 | struct kmem_cache_node *n = get_node(s, node); | |
3699 | struct page *p; | |
3700 | ||
3701 | if (n) { | |
3702 | list_for_each_entry(p, &n->partial, lru) | |
3703 | p->slab = s; | |
3704 | ||
607bf324 | 3705 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3706 | list_for_each_entry(p, &n->full, lru) |
3707 | p->slab = s; | |
3708 | #endif | |
3709 | } | |
3710 | } | |
3711 | } | |
3712 | ||
81819f0f CL |
3713 | void __init kmem_cache_init(void) |
3714 | { | |
3715 | int i; | |
4b356be0 | 3716 | int caches = 0; |
51df1142 CL |
3717 | struct kmem_cache *temp_kmem_cache; |
3718 | int order; | |
51df1142 CL |
3719 | struct kmem_cache *temp_kmem_cache_node; |
3720 | unsigned long kmalloc_size; | |
3721 | ||
fc8d8620 SG |
3722 | if (debug_guardpage_minorder()) |
3723 | slub_max_order = 0; | |
3724 | ||
51df1142 CL |
3725 | kmem_size = offsetof(struct kmem_cache, node) + |
3726 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3727 | ||
3728 | /* Allocate two kmem_caches from the page allocator */ | |
3729 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3730 | order = get_order(2 * kmalloc_size); | |
3731 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3732 | ||
81819f0f CL |
3733 | /* |
3734 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3735 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3736 | * kmem_cache_open for slab_state == DOWN. |
3737 | */ | |
51df1142 CL |
3738 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3739 | ||
3740 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3741 | sizeof(struct kmem_cache_node), | |
3742 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3743 | |
0c40ba4f | 3744 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3745 | |
3746 | /* Able to allocate the per node structures */ | |
3747 | slab_state = PARTIAL; | |
3748 | ||
51df1142 CL |
3749 | temp_kmem_cache = kmem_cache; |
3750 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3751 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3752 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3753 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3754 | |
51df1142 CL |
3755 | /* |
3756 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3757 | * kmem_cache_node is separately allocated so no need to | |
3758 | * update any list pointers. | |
3759 | */ | |
3760 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3761 | |
51df1142 CL |
3762 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3763 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3764 | ||
3765 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3766 | ||
3767 | caches++; | |
51df1142 CL |
3768 | kmem_cache_bootstrap_fixup(kmem_cache); |
3769 | caches++; | |
3770 | /* Free temporary boot structure */ | |
3771 | free_pages((unsigned long)temp_kmem_cache, order); | |
3772 | ||
3773 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3774 | |
3775 | /* | |
3776 | * Patch up the size_index table if we have strange large alignment | |
3777 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3778 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3779 | * |
3780 | * Largest permitted alignment is 256 bytes due to the way we | |
3781 | * handle the index determination for the smaller caches. | |
3782 | * | |
3783 | * Make sure that nothing crazy happens if someone starts tinkering | |
3784 | * around with ARCH_KMALLOC_MINALIGN | |
3785 | */ | |
3786 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3787 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3788 | ||
acdfcd04 AK |
3789 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3790 | int elem = size_index_elem(i); | |
3791 | if (elem >= ARRAY_SIZE(size_index)) | |
3792 | break; | |
3793 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3794 | } | |
f1b26339 | 3795 | |
acdfcd04 AK |
3796 | if (KMALLOC_MIN_SIZE == 64) { |
3797 | /* | |
3798 | * The 96 byte size cache is not used if the alignment | |
3799 | * is 64 byte. | |
3800 | */ | |
3801 | for (i = 64 + 8; i <= 96; i += 8) | |
3802 | size_index[size_index_elem(i)] = 7; | |
3803 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3804 | /* |
3805 | * The 192 byte sized cache is not used if the alignment | |
3806 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3807 | * instead. | |
3808 | */ | |
3809 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3810 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3811 | } |
3812 | ||
51df1142 CL |
3813 | /* Caches that are not of the two-to-the-power-of size */ |
3814 | if (KMALLOC_MIN_SIZE <= 32) { | |
3815 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3816 | caches++; | |
3817 | } | |
3818 | ||
3819 | if (KMALLOC_MIN_SIZE <= 64) { | |
3820 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3821 | caches++; | |
3822 | } | |
3823 | ||
3824 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3825 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3826 | caches++; | |
3827 | } | |
3828 | ||
81819f0f CL |
3829 | slab_state = UP; |
3830 | ||
3831 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3832 | if (KMALLOC_MIN_SIZE <= 32) { |
3833 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3834 | BUG_ON(!kmalloc_caches[1]->name); | |
3835 | } | |
3836 | ||
3837 | if (KMALLOC_MIN_SIZE <= 64) { | |
3838 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3839 | BUG_ON(!kmalloc_caches[2]->name); | |
3840 | } | |
3841 | ||
d7278bd7 CL |
3842 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3843 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3844 | ||
3845 | BUG_ON(!s); | |
51df1142 | 3846 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3847 | } |
81819f0f CL |
3848 | |
3849 | #ifdef CONFIG_SMP | |
3850 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3851 | #endif |
81819f0f | 3852 | |
55136592 | 3853 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3854 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3855 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3856 | |
51df1142 | 3857 | if (s && s->size) { |
55136592 | 3858 | char *name = kasprintf(GFP_NOWAIT, |
3b0efdfa | 3859 | "dma-kmalloc-%d", s->object_size); |
55136592 CL |
3860 | |
3861 | BUG_ON(!name); | |
51df1142 | 3862 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3b0efdfa | 3863 | s->object_size, SLAB_CACHE_DMA); |
55136592 CL |
3864 | } |
3865 | } | |
3866 | #endif | |
3adbefee IM |
3867 | printk(KERN_INFO |
3868 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3869 | " CPUs=%d, Nodes=%d\n", |
3870 | caches, cache_line_size(), | |
81819f0f CL |
3871 | slub_min_order, slub_max_order, slub_min_objects, |
3872 | nr_cpu_ids, nr_node_ids); | |
3873 | } | |
3874 | ||
7e85ee0c PE |
3875 | void __init kmem_cache_init_late(void) |
3876 | { | |
7e85ee0c PE |
3877 | } |
3878 | ||
81819f0f CL |
3879 | /* |
3880 | * Find a mergeable slab cache | |
3881 | */ | |
3882 | static int slab_unmergeable(struct kmem_cache *s) | |
3883 | { | |
3884 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3885 | return 1; | |
3886 | ||
c59def9f | 3887 | if (s->ctor) |
81819f0f CL |
3888 | return 1; |
3889 | ||
8ffa6875 CL |
3890 | /* |
3891 | * We may have set a slab to be unmergeable during bootstrap. | |
3892 | */ | |
3893 | if (s->refcount < 0) | |
3894 | return 1; | |
3895 | ||
81819f0f CL |
3896 | return 0; |
3897 | } | |
3898 | ||
3899 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3900 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3901 | void (*ctor)(void *)) |
81819f0f | 3902 | { |
5b95a4ac | 3903 | struct kmem_cache *s; |
81819f0f CL |
3904 | |
3905 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3906 | return NULL; | |
3907 | ||
c59def9f | 3908 | if (ctor) |
81819f0f CL |
3909 | return NULL; |
3910 | ||
3911 | size = ALIGN(size, sizeof(void *)); | |
3912 | align = calculate_alignment(flags, align, size); | |
3913 | size = ALIGN(size, align); | |
ba0268a8 | 3914 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3915 | |
5b95a4ac | 3916 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3917 | if (slab_unmergeable(s)) |
3918 | continue; | |
3919 | ||
3920 | if (size > s->size) | |
3921 | continue; | |
3922 | ||
ba0268a8 | 3923 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3924 | continue; |
3925 | /* | |
3926 | * Check if alignment is compatible. | |
3927 | * Courtesy of Adrian Drzewiecki | |
3928 | */ | |
06428780 | 3929 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3930 | continue; |
3931 | ||
3932 | if (s->size - size >= sizeof(void *)) | |
3933 | continue; | |
3934 | ||
3935 | return s; | |
3936 | } | |
3937 | return NULL; | |
3938 | } | |
3939 | ||
039363f3 | 3940 | struct kmem_cache *__kmem_cache_create(const char *name, size_t size, |
51cc5068 | 3941 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3942 | { |
3943 | struct kmem_cache *s; | |
84c1cf62 | 3944 | char *n; |
81819f0f | 3945 | |
ba0268a8 | 3946 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3947 | if (s) { |
3948 | s->refcount++; | |
3949 | /* | |
3950 | * Adjust the object sizes so that we clear | |
3951 | * the complete object on kzalloc. | |
3952 | */ | |
3b0efdfa | 3953 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3954 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3955 | |
7b8f3b66 | 3956 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3957 | s->refcount--; |
20cea968 | 3958 | return NULL; |
7b8f3b66 | 3959 | } |
a0e1d1be CL |
3960 | return s; |
3961 | } | |
6446faa2 | 3962 | |
84c1cf62 PE |
3963 | n = kstrdup(name, GFP_KERNEL); |
3964 | if (!n) | |
20cea968 | 3965 | return NULL; |
84c1cf62 | 3966 | |
a0e1d1be CL |
3967 | s = kmalloc(kmem_size, GFP_KERNEL); |
3968 | if (s) { | |
84c1cf62 | 3969 | if (kmem_cache_open(s, n, |
c59def9f | 3970 | size, align, flags, ctor)) { |
20cea968 CL |
3971 | int r; |
3972 | ||
81819f0f | 3973 | list_add(&s->list, &slab_caches); |
18004c5d | 3974 | mutex_unlock(&slab_mutex); |
20cea968 CL |
3975 | r = sysfs_slab_add(s); |
3976 | mutex_lock(&slab_mutex); | |
3977 | ||
3978 | if (!r) | |
3979 | return s; | |
3980 | ||
3981 | list_del(&s->list); | |
3982 | kmem_cache_close(s); | |
a0e1d1be CL |
3983 | } |
3984 | kfree(s); | |
81819f0f | 3985 | } |
601d39d0 | 3986 | kfree(n); |
20cea968 | 3987 | return NULL; |
81819f0f | 3988 | } |
81819f0f | 3989 | |
81819f0f | 3990 | #ifdef CONFIG_SMP |
81819f0f | 3991 | /* |
672bba3a CL |
3992 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3993 | * necessary. | |
81819f0f CL |
3994 | */ |
3995 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3996 | unsigned long action, void *hcpu) | |
3997 | { | |
3998 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3999 | struct kmem_cache *s; |
4000 | unsigned long flags; | |
81819f0f CL |
4001 | |
4002 | switch (action) { | |
4003 | case CPU_UP_CANCELED: | |
8bb78442 | 4004 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4005 | case CPU_DEAD: |
8bb78442 | 4006 | case CPU_DEAD_FROZEN: |
18004c5d | 4007 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4008 | list_for_each_entry(s, &slab_caches, list) { |
4009 | local_irq_save(flags); | |
4010 | __flush_cpu_slab(s, cpu); | |
4011 | local_irq_restore(flags); | |
4012 | } | |
18004c5d | 4013 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4014 | break; |
4015 | default: | |
4016 | break; | |
4017 | } | |
4018 | return NOTIFY_OK; | |
4019 | } | |
4020 | ||
06428780 | 4021 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 4022 | .notifier_call = slab_cpuup_callback |
06428780 | 4023 | }; |
81819f0f CL |
4024 | |
4025 | #endif | |
4026 | ||
ce71e27c | 4027 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4028 | { |
aadb4bc4 | 4029 | struct kmem_cache *s; |
94b528d0 | 4030 | void *ret; |
aadb4bc4 | 4031 | |
ffadd4d0 | 4032 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
4033 | return kmalloc_large(size, gfpflags); |
4034 | ||
aadb4bc4 | 4035 | s = get_slab(size, gfpflags); |
81819f0f | 4036 | |
2408c550 | 4037 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4038 | return s; |
81819f0f | 4039 | |
2154a336 | 4040 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 4041 | |
25985edc | 4042 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4043 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4044 | |
4045 | return ret; | |
81819f0f CL |
4046 | } |
4047 | ||
5d1f57e4 | 4048 | #ifdef CONFIG_NUMA |
81819f0f | 4049 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4050 | int node, unsigned long caller) |
81819f0f | 4051 | { |
aadb4bc4 | 4052 | struct kmem_cache *s; |
94b528d0 | 4053 | void *ret; |
aadb4bc4 | 4054 | |
d3e14aa3 XF |
4055 | if (unlikely(size > SLUB_MAX_SIZE)) { |
4056 | ret = kmalloc_large_node(size, gfpflags, node); | |
4057 | ||
4058 | trace_kmalloc_node(caller, ret, | |
4059 | size, PAGE_SIZE << get_order(size), | |
4060 | gfpflags, node); | |
4061 | ||
4062 | return ret; | |
4063 | } | |
eada35ef | 4064 | |
aadb4bc4 | 4065 | s = get_slab(size, gfpflags); |
81819f0f | 4066 | |
2408c550 | 4067 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4068 | return s; |
81819f0f | 4069 | |
94b528d0 EGM |
4070 | ret = slab_alloc(s, gfpflags, node, caller); |
4071 | ||
25985edc | 4072 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4073 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4074 | |
4075 | return ret; | |
81819f0f | 4076 | } |
5d1f57e4 | 4077 | #endif |
81819f0f | 4078 | |
ab4d5ed5 | 4079 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4080 | static int count_inuse(struct page *page) |
4081 | { | |
4082 | return page->inuse; | |
4083 | } | |
4084 | ||
4085 | static int count_total(struct page *page) | |
4086 | { | |
4087 | return page->objects; | |
4088 | } | |
ab4d5ed5 | 4089 | #endif |
205ab99d | 4090 | |
ab4d5ed5 | 4091 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4092 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4093 | unsigned long *map) | |
53e15af0 CL |
4094 | { |
4095 | void *p; | |
a973e9dd | 4096 | void *addr = page_address(page); |
53e15af0 CL |
4097 | |
4098 | if (!check_slab(s, page) || | |
4099 | !on_freelist(s, page, NULL)) | |
4100 | return 0; | |
4101 | ||
4102 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4103 | bitmap_zero(map, page->objects); |
53e15af0 | 4104 | |
5f80b13a CL |
4105 | get_map(s, page, map); |
4106 | for_each_object(p, s, addr, page->objects) { | |
4107 | if (test_bit(slab_index(p, s, addr), map)) | |
4108 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4109 | return 0; | |
53e15af0 CL |
4110 | } |
4111 | ||
224a88be | 4112 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4113 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4114 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4115 | return 0; |
4116 | return 1; | |
4117 | } | |
4118 | ||
434e245d CL |
4119 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4120 | unsigned long *map) | |
53e15af0 | 4121 | { |
881db7fb CL |
4122 | slab_lock(page); |
4123 | validate_slab(s, page, map); | |
4124 | slab_unlock(page); | |
53e15af0 CL |
4125 | } |
4126 | ||
434e245d CL |
4127 | static int validate_slab_node(struct kmem_cache *s, |
4128 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4129 | { |
4130 | unsigned long count = 0; | |
4131 | struct page *page; | |
4132 | unsigned long flags; | |
4133 | ||
4134 | spin_lock_irqsave(&n->list_lock, flags); | |
4135 | ||
4136 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4137 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4138 | count++; |
4139 | } | |
4140 | if (count != n->nr_partial) | |
4141 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4142 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4143 | ||
4144 | if (!(s->flags & SLAB_STORE_USER)) | |
4145 | goto out; | |
4146 | ||
4147 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4148 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4149 | count++; |
4150 | } | |
4151 | if (count != atomic_long_read(&n->nr_slabs)) | |
4152 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4153 | "counter=%ld\n", s->name, count, | |
4154 | atomic_long_read(&n->nr_slabs)); | |
4155 | ||
4156 | out: | |
4157 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4158 | return count; | |
4159 | } | |
4160 | ||
434e245d | 4161 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4162 | { |
4163 | int node; | |
4164 | unsigned long count = 0; | |
205ab99d | 4165 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4166 | sizeof(unsigned long), GFP_KERNEL); |
4167 | ||
4168 | if (!map) | |
4169 | return -ENOMEM; | |
53e15af0 CL |
4170 | |
4171 | flush_all(s); | |
f64dc58c | 4172 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4173 | struct kmem_cache_node *n = get_node(s, node); |
4174 | ||
434e245d | 4175 | count += validate_slab_node(s, n, map); |
53e15af0 | 4176 | } |
434e245d | 4177 | kfree(map); |
53e15af0 CL |
4178 | return count; |
4179 | } | |
88a420e4 | 4180 | /* |
672bba3a | 4181 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4182 | * and freed. |
4183 | */ | |
4184 | ||
4185 | struct location { | |
4186 | unsigned long count; | |
ce71e27c | 4187 | unsigned long addr; |
45edfa58 CL |
4188 | long long sum_time; |
4189 | long min_time; | |
4190 | long max_time; | |
4191 | long min_pid; | |
4192 | long max_pid; | |
174596a0 | 4193 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4194 | nodemask_t nodes; |
88a420e4 CL |
4195 | }; |
4196 | ||
4197 | struct loc_track { | |
4198 | unsigned long max; | |
4199 | unsigned long count; | |
4200 | struct location *loc; | |
4201 | }; | |
4202 | ||
4203 | static void free_loc_track(struct loc_track *t) | |
4204 | { | |
4205 | if (t->max) | |
4206 | free_pages((unsigned long)t->loc, | |
4207 | get_order(sizeof(struct location) * t->max)); | |
4208 | } | |
4209 | ||
68dff6a9 | 4210 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4211 | { |
4212 | struct location *l; | |
4213 | int order; | |
4214 | ||
88a420e4 CL |
4215 | order = get_order(sizeof(struct location) * max); |
4216 | ||
68dff6a9 | 4217 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4218 | if (!l) |
4219 | return 0; | |
4220 | ||
4221 | if (t->count) { | |
4222 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4223 | free_loc_track(t); | |
4224 | } | |
4225 | t->max = max; | |
4226 | t->loc = l; | |
4227 | return 1; | |
4228 | } | |
4229 | ||
4230 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4231 | const struct track *track) |
88a420e4 CL |
4232 | { |
4233 | long start, end, pos; | |
4234 | struct location *l; | |
ce71e27c | 4235 | unsigned long caddr; |
45edfa58 | 4236 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4237 | |
4238 | start = -1; | |
4239 | end = t->count; | |
4240 | ||
4241 | for ( ; ; ) { | |
4242 | pos = start + (end - start + 1) / 2; | |
4243 | ||
4244 | /* | |
4245 | * There is nothing at "end". If we end up there | |
4246 | * we need to add something to before end. | |
4247 | */ | |
4248 | if (pos == end) | |
4249 | break; | |
4250 | ||
4251 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4252 | if (track->addr == caddr) { |
4253 | ||
4254 | l = &t->loc[pos]; | |
4255 | l->count++; | |
4256 | if (track->when) { | |
4257 | l->sum_time += age; | |
4258 | if (age < l->min_time) | |
4259 | l->min_time = age; | |
4260 | if (age > l->max_time) | |
4261 | l->max_time = age; | |
4262 | ||
4263 | if (track->pid < l->min_pid) | |
4264 | l->min_pid = track->pid; | |
4265 | if (track->pid > l->max_pid) | |
4266 | l->max_pid = track->pid; | |
4267 | ||
174596a0 RR |
4268 | cpumask_set_cpu(track->cpu, |
4269 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4270 | } |
4271 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4272 | return 1; |
4273 | } | |
4274 | ||
45edfa58 | 4275 | if (track->addr < caddr) |
88a420e4 CL |
4276 | end = pos; |
4277 | else | |
4278 | start = pos; | |
4279 | } | |
4280 | ||
4281 | /* | |
672bba3a | 4282 | * Not found. Insert new tracking element. |
88a420e4 | 4283 | */ |
68dff6a9 | 4284 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4285 | return 0; |
4286 | ||
4287 | l = t->loc + pos; | |
4288 | if (pos < t->count) | |
4289 | memmove(l + 1, l, | |
4290 | (t->count - pos) * sizeof(struct location)); | |
4291 | t->count++; | |
4292 | l->count = 1; | |
45edfa58 CL |
4293 | l->addr = track->addr; |
4294 | l->sum_time = age; | |
4295 | l->min_time = age; | |
4296 | l->max_time = age; | |
4297 | l->min_pid = track->pid; | |
4298 | l->max_pid = track->pid; | |
174596a0 RR |
4299 | cpumask_clear(to_cpumask(l->cpus)); |
4300 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4301 | nodes_clear(l->nodes); |
4302 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4303 | return 1; |
4304 | } | |
4305 | ||
4306 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4307 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4308 | unsigned long *map) |
88a420e4 | 4309 | { |
a973e9dd | 4310 | void *addr = page_address(page); |
88a420e4 CL |
4311 | void *p; |
4312 | ||
39b26464 | 4313 | bitmap_zero(map, page->objects); |
5f80b13a | 4314 | get_map(s, page, map); |
88a420e4 | 4315 | |
224a88be | 4316 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4317 | if (!test_bit(slab_index(p, s, addr), map)) |
4318 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4319 | } |
4320 | ||
4321 | static int list_locations(struct kmem_cache *s, char *buf, | |
4322 | enum track_item alloc) | |
4323 | { | |
e374d483 | 4324 | int len = 0; |
88a420e4 | 4325 | unsigned long i; |
68dff6a9 | 4326 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4327 | int node; |
bbd7d57b ED |
4328 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4329 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4330 | |
bbd7d57b ED |
4331 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4332 | GFP_TEMPORARY)) { | |
4333 | kfree(map); | |
68dff6a9 | 4334 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4335 | } |
88a420e4 CL |
4336 | /* Push back cpu slabs */ |
4337 | flush_all(s); | |
4338 | ||
f64dc58c | 4339 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4340 | struct kmem_cache_node *n = get_node(s, node); |
4341 | unsigned long flags; | |
4342 | struct page *page; | |
4343 | ||
9e86943b | 4344 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4345 | continue; |
4346 | ||
4347 | spin_lock_irqsave(&n->list_lock, flags); | |
4348 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4349 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4350 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4351 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4352 | spin_unlock_irqrestore(&n->list_lock, flags); |
4353 | } | |
4354 | ||
4355 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4356 | struct location *l = &t.loc[i]; |
88a420e4 | 4357 | |
9c246247 | 4358 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4359 | break; |
e374d483 | 4360 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4361 | |
4362 | if (l->addr) | |
62c70bce | 4363 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4364 | else |
e374d483 | 4365 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4366 | |
4367 | if (l->sum_time != l->min_time) { | |
e374d483 | 4368 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4369 | l->min_time, |
4370 | (long)div_u64(l->sum_time, l->count), | |
4371 | l->max_time); | |
45edfa58 | 4372 | } else |
e374d483 | 4373 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4374 | l->min_time); |
4375 | ||
4376 | if (l->min_pid != l->max_pid) | |
e374d483 | 4377 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4378 | l->min_pid, l->max_pid); |
4379 | else | |
e374d483 | 4380 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4381 | l->min_pid); |
4382 | ||
174596a0 RR |
4383 | if (num_online_cpus() > 1 && |
4384 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4385 | len < PAGE_SIZE - 60) { |
4386 | len += sprintf(buf + len, " cpus="); | |
4387 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4388 | to_cpumask(l->cpus)); |
45edfa58 CL |
4389 | } |
4390 | ||
62bc62a8 | 4391 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4392 | len < PAGE_SIZE - 60) { |
4393 | len += sprintf(buf + len, " nodes="); | |
4394 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4395 | l->nodes); |
4396 | } | |
4397 | ||
e374d483 | 4398 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4399 | } |
4400 | ||
4401 | free_loc_track(&t); | |
bbd7d57b | 4402 | kfree(map); |
88a420e4 | 4403 | if (!t.count) |
e374d483 HH |
4404 | len += sprintf(buf, "No data\n"); |
4405 | return len; | |
88a420e4 | 4406 | } |
ab4d5ed5 | 4407 | #endif |
88a420e4 | 4408 | |
a5a84755 CL |
4409 | #ifdef SLUB_RESILIENCY_TEST |
4410 | static void resiliency_test(void) | |
4411 | { | |
4412 | u8 *p; | |
4413 | ||
4414 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4415 | ||
4416 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4417 | printk(KERN_ERR "-----------------------\n"); | |
4418 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4419 | ||
4420 | p = kzalloc(16, GFP_KERNEL); | |
4421 | p[16] = 0x12; | |
4422 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4423 | " 0x12->0x%p\n\n", p + 16); | |
4424 | ||
4425 | validate_slab_cache(kmalloc_caches[4]); | |
4426 | ||
4427 | /* Hmmm... The next two are dangerous */ | |
4428 | p = kzalloc(32, GFP_KERNEL); | |
4429 | p[32 + sizeof(void *)] = 0x34; | |
4430 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4431 | " 0x34 -> -0x%p\n", p); | |
4432 | printk(KERN_ERR | |
4433 | "If allocated object is overwritten then not detectable\n\n"); | |
4434 | ||
4435 | validate_slab_cache(kmalloc_caches[5]); | |
4436 | p = kzalloc(64, GFP_KERNEL); | |
4437 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4438 | *p = 0x56; | |
4439 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4440 | p); | |
4441 | printk(KERN_ERR | |
4442 | "If allocated object is overwritten then not detectable\n\n"); | |
4443 | validate_slab_cache(kmalloc_caches[6]); | |
4444 | ||
4445 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4446 | p = kzalloc(128, GFP_KERNEL); | |
4447 | kfree(p); | |
4448 | *p = 0x78; | |
4449 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4450 | validate_slab_cache(kmalloc_caches[7]); | |
4451 | ||
4452 | p = kzalloc(256, GFP_KERNEL); | |
4453 | kfree(p); | |
4454 | p[50] = 0x9a; | |
4455 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4456 | p); | |
4457 | validate_slab_cache(kmalloc_caches[8]); | |
4458 | ||
4459 | p = kzalloc(512, GFP_KERNEL); | |
4460 | kfree(p); | |
4461 | p[512] = 0xab; | |
4462 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4463 | validate_slab_cache(kmalloc_caches[9]); | |
4464 | } | |
4465 | #else | |
4466 | #ifdef CONFIG_SYSFS | |
4467 | static void resiliency_test(void) {}; | |
4468 | #endif | |
4469 | #endif | |
4470 | ||
ab4d5ed5 | 4471 | #ifdef CONFIG_SYSFS |
81819f0f | 4472 | enum slab_stat_type { |
205ab99d CL |
4473 | SL_ALL, /* All slabs */ |
4474 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4475 | SL_CPU, /* Only slabs used for cpu caches */ | |
4476 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4477 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4478 | }; |
4479 | ||
205ab99d | 4480 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4481 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4482 | #define SO_CPU (1 << SL_CPU) | |
4483 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4484 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4485 | |
62e5c4b4 CG |
4486 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4487 | char *buf, unsigned long flags) | |
81819f0f CL |
4488 | { |
4489 | unsigned long total = 0; | |
81819f0f CL |
4490 | int node; |
4491 | int x; | |
4492 | unsigned long *nodes; | |
4493 | unsigned long *per_cpu; | |
4494 | ||
4495 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4496 | if (!nodes) |
4497 | return -ENOMEM; | |
81819f0f CL |
4498 | per_cpu = nodes + nr_node_ids; |
4499 | ||
205ab99d CL |
4500 | if (flags & SO_CPU) { |
4501 | int cpu; | |
81819f0f | 4502 | |
205ab99d | 4503 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4504 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4505 | int node; |
49e22585 | 4506 | struct page *page; |
dfb4f096 | 4507 | |
bc6697d8 | 4508 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4509 | if (!page) |
4510 | continue; | |
205ab99d | 4511 | |
ec3ab083 CL |
4512 | node = page_to_nid(page); |
4513 | if (flags & SO_TOTAL) | |
4514 | x = page->objects; | |
4515 | else if (flags & SO_OBJECTS) | |
4516 | x = page->inuse; | |
4517 | else | |
4518 | x = 1; | |
49e22585 | 4519 | |
ec3ab083 CL |
4520 | total += x; |
4521 | nodes[node] += x; | |
4522 | ||
4523 | page = ACCESS_ONCE(c->partial); | |
49e22585 CL |
4524 | if (page) { |
4525 | x = page->pobjects; | |
bc6697d8 ED |
4526 | total += x; |
4527 | nodes[node] += x; | |
49e22585 | 4528 | } |
ec3ab083 | 4529 | |
bc6697d8 | 4530 | per_cpu[node]++; |
81819f0f CL |
4531 | } |
4532 | } | |
4533 | ||
04d94879 | 4534 | lock_memory_hotplug(); |
ab4d5ed5 | 4535 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4536 | if (flags & SO_ALL) { |
4537 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4538 | struct kmem_cache_node *n = get_node(s, node); | |
4539 | ||
4540 | if (flags & SO_TOTAL) | |
4541 | x = atomic_long_read(&n->total_objects); | |
4542 | else if (flags & SO_OBJECTS) | |
4543 | x = atomic_long_read(&n->total_objects) - | |
4544 | count_partial(n, count_free); | |
81819f0f | 4545 | |
81819f0f | 4546 | else |
205ab99d | 4547 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4548 | total += x; |
4549 | nodes[node] += x; | |
4550 | } | |
4551 | ||
ab4d5ed5 CL |
4552 | } else |
4553 | #endif | |
4554 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4555 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4556 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4557 | |
205ab99d CL |
4558 | if (flags & SO_TOTAL) |
4559 | x = count_partial(n, count_total); | |
4560 | else if (flags & SO_OBJECTS) | |
4561 | x = count_partial(n, count_inuse); | |
81819f0f | 4562 | else |
205ab99d | 4563 | x = n->nr_partial; |
81819f0f CL |
4564 | total += x; |
4565 | nodes[node] += x; | |
4566 | } | |
4567 | } | |
81819f0f CL |
4568 | x = sprintf(buf, "%lu", total); |
4569 | #ifdef CONFIG_NUMA | |
f64dc58c | 4570 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4571 | if (nodes[node]) |
4572 | x += sprintf(buf + x, " N%d=%lu", | |
4573 | node, nodes[node]); | |
4574 | #endif | |
04d94879 | 4575 | unlock_memory_hotplug(); |
81819f0f CL |
4576 | kfree(nodes); |
4577 | return x + sprintf(buf + x, "\n"); | |
4578 | } | |
4579 | ||
ab4d5ed5 | 4580 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4581 | static int any_slab_objects(struct kmem_cache *s) |
4582 | { | |
4583 | int node; | |
81819f0f | 4584 | |
dfb4f096 | 4585 | for_each_online_node(node) { |
81819f0f CL |
4586 | struct kmem_cache_node *n = get_node(s, node); |
4587 | ||
dfb4f096 CL |
4588 | if (!n) |
4589 | continue; | |
4590 | ||
4ea33e2d | 4591 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4592 | return 1; |
4593 | } | |
4594 | return 0; | |
4595 | } | |
ab4d5ed5 | 4596 | #endif |
81819f0f CL |
4597 | |
4598 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4599 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4600 | |
4601 | struct slab_attribute { | |
4602 | struct attribute attr; | |
4603 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4604 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4605 | }; | |
4606 | ||
4607 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4608 | static struct slab_attribute _name##_attr = \ |
4609 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4610 | |
4611 | #define SLAB_ATTR(_name) \ | |
4612 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4613 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4614 | |
81819f0f CL |
4615 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4616 | { | |
4617 | return sprintf(buf, "%d\n", s->size); | |
4618 | } | |
4619 | SLAB_ATTR_RO(slab_size); | |
4620 | ||
4621 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4622 | { | |
4623 | return sprintf(buf, "%d\n", s->align); | |
4624 | } | |
4625 | SLAB_ATTR_RO(align); | |
4626 | ||
4627 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4628 | { | |
3b0efdfa | 4629 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4630 | } |
4631 | SLAB_ATTR_RO(object_size); | |
4632 | ||
4633 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4634 | { | |
834f3d11 | 4635 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4636 | } |
4637 | SLAB_ATTR_RO(objs_per_slab); | |
4638 | ||
06b285dc CL |
4639 | static ssize_t order_store(struct kmem_cache *s, |
4640 | const char *buf, size_t length) | |
4641 | { | |
0121c619 CL |
4642 | unsigned long order; |
4643 | int err; | |
4644 | ||
4645 | err = strict_strtoul(buf, 10, &order); | |
4646 | if (err) | |
4647 | return err; | |
06b285dc CL |
4648 | |
4649 | if (order > slub_max_order || order < slub_min_order) | |
4650 | return -EINVAL; | |
4651 | ||
4652 | calculate_sizes(s, order); | |
4653 | return length; | |
4654 | } | |
4655 | ||
81819f0f CL |
4656 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4657 | { | |
834f3d11 | 4658 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4659 | } |
06b285dc | 4660 | SLAB_ATTR(order); |
81819f0f | 4661 | |
73d342b1 DR |
4662 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4663 | { | |
4664 | return sprintf(buf, "%lu\n", s->min_partial); | |
4665 | } | |
4666 | ||
4667 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4668 | size_t length) | |
4669 | { | |
4670 | unsigned long min; | |
4671 | int err; | |
4672 | ||
4673 | err = strict_strtoul(buf, 10, &min); | |
4674 | if (err) | |
4675 | return err; | |
4676 | ||
c0bdb232 | 4677 | set_min_partial(s, min); |
73d342b1 DR |
4678 | return length; |
4679 | } | |
4680 | SLAB_ATTR(min_partial); | |
4681 | ||
49e22585 CL |
4682 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4683 | { | |
4684 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4685 | } | |
4686 | ||
4687 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4688 | size_t length) | |
4689 | { | |
4690 | unsigned long objects; | |
4691 | int err; | |
4692 | ||
4693 | err = strict_strtoul(buf, 10, &objects); | |
4694 | if (err) | |
4695 | return err; | |
74ee4ef1 DR |
4696 | if (objects && kmem_cache_debug(s)) |
4697 | return -EINVAL; | |
49e22585 CL |
4698 | |
4699 | s->cpu_partial = objects; | |
4700 | flush_all(s); | |
4701 | return length; | |
4702 | } | |
4703 | SLAB_ATTR(cpu_partial); | |
4704 | ||
81819f0f CL |
4705 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4706 | { | |
62c70bce JP |
4707 | if (!s->ctor) |
4708 | return 0; | |
4709 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4710 | } |
4711 | SLAB_ATTR_RO(ctor); | |
4712 | ||
81819f0f CL |
4713 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4714 | { | |
4715 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4716 | } | |
4717 | SLAB_ATTR_RO(aliases); | |
4718 | ||
81819f0f CL |
4719 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4720 | { | |
d9acf4b7 | 4721 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4722 | } |
4723 | SLAB_ATTR_RO(partial); | |
4724 | ||
4725 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4726 | { | |
d9acf4b7 | 4727 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4728 | } |
4729 | SLAB_ATTR_RO(cpu_slabs); | |
4730 | ||
4731 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4732 | { | |
205ab99d | 4733 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4734 | } |
4735 | SLAB_ATTR_RO(objects); | |
4736 | ||
205ab99d CL |
4737 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4738 | { | |
4739 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4740 | } | |
4741 | SLAB_ATTR_RO(objects_partial); | |
4742 | ||
49e22585 CL |
4743 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4744 | { | |
4745 | int objects = 0; | |
4746 | int pages = 0; | |
4747 | int cpu; | |
4748 | int len; | |
4749 | ||
4750 | for_each_online_cpu(cpu) { | |
4751 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4752 | ||
4753 | if (page) { | |
4754 | pages += page->pages; | |
4755 | objects += page->pobjects; | |
4756 | } | |
4757 | } | |
4758 | ||
4759 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4760 | ||
4761 | #ifdef CONFIG_SMP | |
4762 | for_each_online_cpu(cpu) { | |
4763 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4764 | ||
4765 | if (page && len < PAGE_SIZE - 20) | |
4766 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4767 | page->pobjects, page->pages); | |
4768 | } | |
4769 | #endif | |
4770 | return len + sprintf(buf + len, "\n"); | |
4771 | } | |
4772 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4773 | ||
a5a84755 CL |
4774 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4775 | { | |
4776 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4777 | } | |
4778 | ||
4779 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4780 | const char *buf, size_t length) | |
4781 | { | |
4782 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4783 | if (buf[0] == '1') | |
4784 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4785 | return length; | |
4786 | } | |
4787 | SLAB_ATTR(reclaim_account); | |
4788 | ||
4789 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4790 | { | |
4791 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4792 | } | |
4793 | SLAB_ATTR_RO(hwcache_align); | |
4794 | ||
4795 | #ifdef CONFIG_ZONE_DMA | |
4796 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4797 | { | |
4798 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4799 | } | |
4800 | SLAB_ATTR_RO(cache_dma); | |
4801 | #endif | |
4802 | ||
4803 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4804 | { | |
4805 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4806 | } | |
4807 | SLAB_ATTR_RO(destroy_by_rcu); | |
4808 | ||
ab9a0f19 LJ |
4809 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4810 | { | |
4811 | return sprintf(buf, "%d\n", s->reserved); | |
4812 | } | |
4813 | SLAB_ATTR_RO(reserved); | |
4814 | ||
ab4d5ed5 | 4815 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4816 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4817 | { | |
4818 | return show_slab_objects(s, buf, SO_ALL); | |
4819 | } | |
4820 | SLAB_ATTR_RO(slabs); | |
4821 | ||
205ab99d CL |
4822 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4823 | { | |
4824 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4825 | } | |
4826 | SLAB_ATTR_RO(total_objects); | |
4827 | ||
81819f0f CL |
4828 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4829 | { | |
4830 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4831 | } | |
4832 | ||
4833 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4834 | const char *buf, size_t length) | |
4835 | { | |
4836 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4837 | if (buf[0] == '1') { |
4838 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4839 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4840 | } |
81819f0f CL |
4841 | return length; |
4842 | } | |
4843 | SLAB_ATTR(sanity_checks); | |
4844 | ||
4845 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4846 | { | |
4847 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4848 | } | |
4849 | ||
4850 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4851 | size_t length) | |
4852 | { | |
4853 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4854 | if (buf[0] == '1') { |
4855 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4856 | s->flags |= SLAB_TRACE; |
b789ef51 | 4857 | } |
81819f0f CL |
4858 | return length; |
4859 | } | |
4860 | SLAB_ATTR(trace); | |
4861 | ||
81819f0f CL |
4862 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4863 | { | |
4864 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4865 | } | |
4866 | ||
4867 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4868 | const char *buf, size_t length) | |
4869 | { | |
4870 | if (any_slab_objects(s)) | |
4871 | return -EBUSY; | |
4872 | ||
4873 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4874 | if (buf[0] == '1') { |
4875 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4876 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4877 | } |
06b285dc | 4878 | calculate_sizes(s, -1); |
81819f0f CL |
4879 | return length; |
4880 | } | |
4881 | SLAB_ATTR(red_zone); | |
4882 | ||
4883 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4884 | { | |
4885 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4886 | } | |
4887 | ||
4888 | static ssize_t poison_store(struct kmem_cache *s, | |
4889 | const char *buf, size_t length) | |
4890 | { | |
4891 | if (any_slab_objects(s)) | |
4892 | return -EBUSY; | |
4893 | ||
4894 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4895 | if (buf[0] == '1') { |
4896 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4897 | s->flags |= SLAB_POISON; |
b789ef51 | 4898 | } |
06b285dc | 4899 | calculate_sizes(s, -1); |
81819f0f CL |
4900 | return length; |
4901 | } | |
4902 | SLAB_ATTR(poison); | |
4903 | ||
4904 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4905 | { | |
4906 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4907 | } | |
4908 | ||
4909 | static ssize_t store_user_store(struct kmem_cache *s, | |
4910 | const char *buf, size_t length) | |
4911 | { | |
4912 | if (any_slab_objects(s)) | |
4913 | return -EBUSY; | |
4914 | ||
4915 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4916 | if (buf[0] == '1') { |
4917 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4918 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4919 | } |
06b285dc | 4920 | calculate_sizes(s, -1); |
81819f0f CL |
4921 | return length; |
4922 | } | |
4923 | SLAB_ATTR(store_user); | |
4924 | ||
53e15af0 CL |
4925 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4926 | { | |
4927 | return 0; | |
4928 | } | |
4929 | ||
4930 | static ssize_t validate_store(struct kmem_cache *s, | |
4931 | const char *buf, size_t length) | |
4932 | { | |
434e245d CL |
4933 | int ret = -EINVAL; |
4934 | ||
4935 | if (buf[0] == '1') { | |
4936 | ret = validate_slab_cache(s); | |
4937 | if (ret >= 0) | |
4938 | ret = length; | |
4939 | } | |
4940 | return ret; | |
53e15af0 CL |
4941 | } |
4942 | SLAB_ATTR(validate); | |
a5a84755 CL |
4943 | |
4944 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4945 | { | |
4946 | if (!(s->flags & SLAB_STORE_USER)) | |
4947 | return -ENOSYS; | |
4948 | return list_locations(s, buf, TRACK_ALLOC); | |
4949 | } | |
4950 | SLAB_ATTR_RO(alloc_calls); | |
4951 | ||
4952 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4953 | { | |
4954 | if (!(s->flags & SLAB_STORE_USER)) | |
4955 | return -ENOSYS; | |
4956 | return list_locations(s, buf, TRACK_FREE); | |
4957 | } | |
4958 | SLAB_ATTR_RO(free_calls); | |
4959 | #endif /* CONFIG_SLUB_DEBUG */ | |
4960 | ||
4961 | #ifdef CONFIG_FAILSLAB | |
4962 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4963 | { | |
4964 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4965 | } | |
4966 | ||
4967 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4968 | size_t length) | |
4969 | { | |
4970 | s->flags &= ~SLAB_FAILSLAB; | |
4971 | if (buf[0] == '1') | |
4972 | s->flags |= SLAB_FAILSLAB; | |
4973 | return length; | |
4974 | } | |
4975 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4976 | #endif |
53e15af0 | 4977 | |
2086d26a CL |
4978 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4979 | { | |
4980 | return 0; | |
4981 | } | |
4982 | ||
4983 | static ssize_t shrink_store(struct kmem_cache *s, | |
4984 | const char *buf, size_t length) | |
4985 | { | |
4986 | if (buf[0] == '1') { | |
4987 | int rc = kmem_cache_shrink(s); | |
4988 | ||
4989 | if (rc) | |
4990 | return rc; | |
4991 | } else | |
4992 | return -EINVAL; | |
4993 | return length; | |
4994 | } | |
4995 | SLAB_ATTR(shrink); | |
4996 | ||
81819f0f | 4997 | #ifdef CONFIG_NUMA |
9824601e | 4998 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4999 | { |
9824601e | 5000 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5001 | } |
5002 | ||
9824601e | 5003 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5004 | const char *buf, size_t length) |
5005 | { | |
0121c619 CL |
5006 | unsigned long ratio; |
5007 | int err; | |
5008 | ||
5009 | err = strict_strtoul(buf, 10, &ratio); | |
5010 | if (err) | |
5011 | return err; | |
5012 | ||
e2cb96b7 | 5013 | if (ratio <= 100) |
0121c619 | 5014 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5015 | |
81819f0f CL |
5016 | return length; |
5017 | } | |
9824601e | 5018 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5019 | #endif |
5020 | ||
8ff12cfc | 5021 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5022 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5023 | { | |
5024 | unsigned long sum = 0; | |
5025 | int cpu; | |
5026 | int len; | |
5027 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5028 | ||
5029 | if (!data) | |
5030 | return -ENOMEM; | |
5031 | ||
5032 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5033 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5034 | |
5035 | data[cpu] = x; | |
5036 | sum += x; | |
5037 | } | |
5038 | ||
5039 | len = sprintf(buf, "%lu", sum); | |
5040 | ||
50ef37b9 | 5041 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5042 | for_each_online_cpu(cpu) { |
5043 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5044 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5045 | } |
50ef37b9 | 5046 | #endif |
8ff12cfc CL |
5047 | kfree(data); |
5048 | return len + sprintf(buf + len, "\n"); | |
5049 | } | |
5050 | ||
78eb00cc DR |
5051 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5052 | { | |
5053 | int cpu; | |
5054 | ||
5055 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5056 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5057 | } |
5058 | ||
8ff12cfc CL |
5059 | #define STAT_ATTR(si, text) \ |
5060 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5061 | { \ | |
5062 | return show_stat(s, buf, si); \ | |
5063 | } \ | |
78eb00cc DR |
5064 | static ssize_t text##_store(struct kmem_cache *s, \ |
5065 | const char *buf, size_t length) \ | |
5066 | { \ | |
5067 | if (buf[0] != '0') \ | |
5068 | return -EINVAL; \ | |
5069 | clear_stat(s, si); \ | |
5070 | return length; \ | |
5071 | } \ | |
5072 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5073 | |
5074 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5075 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5076 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5077 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5078 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5079 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5080 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5081 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5082 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5083 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5084 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5085 | STAT_ATTR(FREE_SLAB, free_slab); |
5086 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5087 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5088 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5089 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5090 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5091 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5092 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5093 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5094 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5095 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5096 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5097 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5098 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5099 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5100 | #endif |
5101 | ||
06428780 | 5102 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5103 | &slab_size_attr.attr, |
5104 | &object_size_attr.attr, | |
5105 | &objs_per_slab_attr.attr, | |
5106 | &order_attr.attr, | |
73d342b1 | 5107 | &min_partial_attr.attr, |
49e22585 | 5108 | &cpu_partial_attr.attr, |
81819f0f | 5109 | &objects_attr.attr, |
205ab99d | 5110 | &objects_partial_attr.attr, |
81819f0f CL |
5111 | &partial_attr.attr, |
5112 | &cpu_slabs_attr.attr, | |
5113 | &ctor_attr.attr, | |
81819f0f CL |
5114 | &aliases_attr.attr, |
5115 | &align_attr.attr, | |
81819f0f CL |
5116 | &hwcache_align_attr.attr, |
5117 | &reclaim_account_attr.attr, | |
5118 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5119 | &shrink_attr.attr, |
ab9a0f19 | 5120 | &reserved_attr.attr, |
49e22585 | 5121 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5122 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5123 | &total_objects_attr.attr, |
5124 | &slabs_attr.attr, | |
5125 | &sanity_checks_attr.attr, | |
5126 | &trace_attr.attr, | |
81819f0f CL |
5127 | &red_zone_attr.attr, |
5128 | &poison_attr.attr, | |
5129 | &store_user_attr.attr, | |
53e15af0 | 5130 | &validate_attr.attr, |
88a420e4 CL |
5131 | &alloc_calls_attr.attr, |
5132 | &free_calls_attr.attr, | |
ab4d5ed5 | 5133 | #endif |
81819f0f CL |
5134 | #ifdef CONFIG_ZONE_DMA |
5135 | &cache_dma_attr.attr, | |
5136 | #endif | |
5137 | #ifdef CONFIG_NUMA | |
9824601e | 5138 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5139 | #endif |
5140 | #ifdef CONFIG_SLUB_STATS | |
5141 | &alloc_fastpath_attr.attr, | |
5142 | &alloc_slowpath_attr.attr, | |
5143 | &free_fastpath_attr.attr, | |
5144 | &free_slowpath_attr.attr, | |
5145 | &free_frozen_attr.attr, | |
5146 | &free_add_partial_attr.attr, | |
5147 | &free_remove_partial_attr.attr, | |
5148 | &alloc_from_partial_attr.attr, | |
5149 | &alloc_slab_attr.attr, | |
5150 | &alloc_refill_attr.attr, | |
e36a2652 | 5151 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5152 | &free_slab_attr.attr, |
5153 | &cpuslab_flush_attr.attr, | |
5154 | &deactivate_full_attr.attr, | |
5155 | &deactivate_empty_attr.attr, | |
5156 | &deactivate_to_head_attr.attr, | |
5157 | &deactivate_to_tail_attr.attr, | |
5158 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5159 | &deactivate_bypass_attr.attr, |
65c3376a | 5160 | &order_fallback_attr.attr, |
b789ef51 CL |
5161 | &cmpxchg_double_fail_attr.attr, |
5162 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5163 | &cpu_partial_alloc_attr.attr, |
5164 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5165 | &cpu_partial_node_attr.attr, |
5166 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5167 | #endif |
4c13dd3b DM |
5168 | #ifdef CONFIG_FAILSLAB |
5169 | &failslab_attr.attr, | |
5170 | #endif | |
5171 | ||
81819f0f CL |
5172 | NULL |
5173 | }; | |
5174 | ||
5175 | static struct attribute_group slab_attr_group = { | |
5176 | .attrs = slab_attrs, | |
5177 | }; | |
5178 | ||
5179 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5180 | struct attribute *attr, | |
5181 | char *buf) | |
5182 | { | |
5183 | struct slab_attribute *attribute; | |
5184 | struct kmem_cache *s; | |
5185 | int err; | |
5186 | ||
5187 | attribute = to_slab_attr(attr); | |
5188 | s = to_slab(kobj); | |
5189 | ||
5190 | if (!attribute->show) | |
5191 | return -EIO; | |
5192 | ||
5193 | err = attribute->show(s, buf); | |
5194 | ||
5195 | return err; | |
5196 | } | |
5197 | ||
5198 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5199 | struct attribute *attr, | |
5200 | const char *buf, size_t len) | |
5201 | { | |
5202 | struct slab_attribute *attribute; | |
5203 | struct kmem_cache *s; | |
5204 | int err; | |
5205 | ||
5206 | attribute = to_slab_attr(attr); | |
5207 | s = to_slab(kobj); | |
5208 | ||
5209 | if (!attribute->store) | |
5210 | return -EIO; | |
5211 | ||
5212 | err = attribute->store(s, buf, len); | |
5213 | ||
5214 | return err; | |
5215 | } | |
5216 | ||
151c602f CL |
5217 | static void kmem_cache_release(struct kobject *kobj) |
5218 | { | |
5219 | struct kmem_cache *s = to_slab(kobj); | |
5220 | ||
84c1cf62 | 5221 | kfree(s->name); |
151c602f CL |
5222 | kfree(s); |
5223 | } | |
5224 | ||
52cf25d0 | 5225 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5226 | .show = slab_attr_show, |
5227 | .store = slab_attr_store, | |
5228 | }; | |
5229 | ||
5230 | static struct kobj_type slab_ktype = { | |
5231 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 5232 | .release = kmem_cache_release |
81819f0f CL |
5233 | }; |
5234 | ||
5235 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5236 | { | |
5237 | struct kobj_type *ktype = get_ktype(kobj); | |
5238 | ||
5239 | if (ktype == &slab_ktype) | |
5240 | return 1; | |
5241 | return 0; | |
5242 | } | |
5243 | ||
9cd43611 | 5244 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5245 | .filter = uevent_filter, |
5246 | }; | |
5247 | ||
27c3a314 | 5248 | static struct kset *slab_kset; |
81819f0f CL |
5249 | |
5250 | #define ID_STR_LENGTH 64 | |
5251 | ||
5252 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5253 | * |
5254 | * Format :[flags-]size | |
81819f0f CL |
5255 | */ |
5256 | static char *create_unique_id(struct kmem_cache *s) | |
5257 | { | |
5258 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5259 | char *p = name; | |
5260 | ||
5261 | BUG_ON(!name); | |
5262 | ||
5263 | *p++ = ':'; | |
5264 | /* | |
5265 | * First flags affecting slabcache operations. We will only | |
5266 | * get here for aliasable slabs so we do not need to support | |
5267 | * too many flags. The flags here must cover all flags that | |
5268 | * are matched during merging to guarantee that the id is | |
5269 | * unique. | |
5270 | */ | |
5271 | if (s->flags & SLAB_CACHE_DMA) | |
5272 | *p++ = 'd'; | |
5273 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5274 | *p++ = 'a'; | |
5275 | if (s->flags & SLAB_DEBUG_FREE) | |
5276 | *p++ = 'F'; | |
5a896d9e VN |
5277 | if (!(s->flags & SLAB_NOTRACK)) |
5278 | *p++ = 't'; | |
81819f0f CL |
5279 | if (p != name + 1) |
5280 | *p++ = '-'; | |
5281 | p += sprintf(p, "%07d", s->size); | |
5282 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5283 | return name; | |
5284 | } | |
5285 | ||
5286 | static int sysfs_slab_add(struct kmem_cache *s) | |
5287 | { | |
5288 | int err; | |
5289 | const char *name; | |
5290 | int unmergeable; | |
5291 | ||
97d06609 | 5292 | if (slab_state < FULL) |
81819f0f CL |
5293 | /* Defer until later */ |
5294 | return 0; | |
5295 | ||
5296 | unmergeable = slab_unmergeable(s); | |
5297 | if (unmergeable) { | |
5298 | /* | |
5299 | * Slabcache can never be merged so we can use the name proper. | |
5300 | * This is typically the case for debug situations. In that | |
5301 | * case we can catch duplicate names easily. | |
5302 | */ | |
27c3a314 | 5303 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5304 | name = s->name; |
5305 | } else { | |
5306 | /* | |
5307 | * Create a unique name for the slab as a target | |
5308 | * for the symlinks. | |
5309 | */ | |
5310 | name = create_unique_id(s); | |
5311 | } | |
5312 | ||
27c3a314 | 5313 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5314 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5315 | if (err) { | |
5316 | kobject_put(&s->kobj); | |
81819f0f | 5317 | return err; |
1eada11c | 5318 | } |
81819f0f CL |
5319 | |
5320 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5321 | if (err) { |
5322 | kobject_del(&s->kobj); | |
5323 | kobject_put(&s->kobj); | |
81819f0f | 5324 | return err; |
5788d8ad | 5325 | } |
81819f0f CL |
5326 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5327 | if (!unmergeable) { | |
5328 | /* Setup first alias */ | |
5329 | sysfs_slab_alias(s, s->name); | |
5330 | kfree(name); | |
5331 | } | |
5332 | return 0; | |
5333 | } | |
5334 | ||
5335 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5336 | { | |
97d06609 | 5337 | if (slab_state < FULL) |
2bce6485 CL |
5338 | /* |
5339 | * Sysfs has not been setup yet so no need to remove the | |
5340 | * cache from sysfs. | |
5341 | */ | |
5342 | return; | |
5343 | ||
81819f0f CL |
5344 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5345 | kobject_del(&s->kobj); | |
151c602f | 5346 | kobject_put(&s->kobj); |
81819f0f CL |
5347 | } |
5348 | ||
5349 | /* | |
5350 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5351 | * available lest we lose that information. |
81819f0f CL |
5352 | */ |
5353 | struct saved_alias { | |
5354 | struct kmem_cache *s; | |
5355 | const char *name; | |
5356 | struct saved_alias *next; | |
5357 | }; | |
5358 | ||
5af328a5 | 5359 | static struct saved_alias *alias_list; |
81819f0f CL |
5360 | |
5361 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5362 | { | |
5363 | struct saved_alias *al; | |
5364 | ||
97d06609 | 5365 | if (slab_state == FULL) { |
81819f0f CL |
5366 | /* |
5367 | * If we have a leftover link then remove it. | |
5368 | */ | |
27c3a314 GKH |
5369 | sysfs_remove_link(&slab_kset->kobj, name); |
5370 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5371 | } |
5372 | ||
5373 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5374 | if (!al) | |
5375 | return -ENOMEM; | |
5376 | ||
5377 | al->s = s; | |
5378 | al->name = name; | |
5379 | al->next = alias_list; | |
5380 | alias_list = al; | |
5381 | return 0; | |
5382 | } | |
5383 | ||
5384 | static int __init slab_sysfs_init(void) | |
5385 | { | |
5b95a4ac | 5386 | struct kmem_cache *s; |
81819f0f CL |
5387 | int err; |
5388 | ||
18004c5d | 5389 | mutex_lock(&slab_mutex); |
2bce6485 | 5390 | |
0ff21e46 | 5391 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5392 | if (!slab_kset) { |
18004c5d | 5393 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5394 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5395 | return -ENOSYS; | |
5396 | } | |
5397 | ||
97d06609 | 5398 | slab_state = FULL; |
26a7bd03 | 5399 | |
5b95a4ac | 5400 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5401 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5402 | if (err) |
5403 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5404 | " to sysfs\n", s->name); | |
26a7bd03 | 5405 | } |
81819f0f CL |
5406 | |
5407 | while (alias_list) { | |
5408 | struct saved_alias *al = alias_list; | |
5409 | ||
5410 | alias_list = alias_list->next; | |
5411 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5412 | if (err) |
5413 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5414 | " %s to sysfs\n", al->name); |
81819f0f CL |
5415 | kfree(al); |
5416 | } | |
5417 | ||
18004c5d | 5418 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5419 | resiliency_test(); |
5420 | return 0; | |
5421 | } | |
5422 | ||
5423 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5424 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5425 | |
5426 | /* | |
5427 | * The /proc/slabinfo ABI | |
5428 | */ | |
158a9624 | 5429 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5430 | static void print_slabinfo_header(struct seq_file *m) |
5431 | { | |
5432 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
3b0efdfa | 5433 | seq_puts(m, "# name <active_objs> <num_objs> <object_size> " |
57ed3eda PE |
5434 | "<objperslab> <pagesperslab>"); |
5435 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5436 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5437 | seq_putc(m, '\n'); | |
5438 | } | |
5439 | ||
5440 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5441 | { | |
5442 | loff_t n = *pos; | |
5443 | ||
18004c5d | 5444 | mutex_lock(&slab_mutex); |
57ed3eda PE |
5445 | if (!n) |
5446 | print_slabinfo_header(m); | |
5447 | ||
5448 | return seq_list_start(&slab_caches, *pos); | |
5449 | } | |
5450 | ||
5451 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5452 | { | |
5453 | return seq_list_next(p, &slab_caches, pos); | |
5454 | } | |
5455 | ||
5456 | static void s_stop(struct seq_file *m, void *p) | |
5457 | { | |
18004c5d | 5458 | mutex_unlock(&slab_mutex); |
57ed3eda PE |
5459 | } |
5460 | ||
5461 | static int s_show(struct seq_file *m, void *p) | |
5462 | { | |
5463 | unsigned long nr_partials = 0; | |
5464 | unsigned long nr_slabs = 0; | |
5465 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5466 | unsigned long nr_objs = 0; |
5467 | unsigned long nr_free = 0; | |
57ed3eda PE |
5468 | struct kmem_cache *s; |
5469 | int node; | |
5470 | ||
5471 | s = list_entry(p, struct kmem_cache, list); | |
5472 | ||
5473 | for_each_online_node(node) { | |
5474 | struct kmem_cache_node *n = get_node(s, node); | |
5475 | ||
5476 | if (!n) | |
5477 | continue; | |
5478 | ||
5479 | nr_partials += n->nr_partial; | |
5480 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5481 | nr_objs += atomic_long_read(&n->total_objects); |
5482 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5483 | } |
5484 | ||
205ab99d | 5485 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5486 | |
5487 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5488 | nr_objs, s->size, oo_objects(s->oo), |
5489 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5490 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5491 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5492 | 0UL); | |
5493 | seq_putc(m, '\n'); | |
5494 | return 0; | |
5495 | } | |
5496 | ||
7b3c3a50 | 5497 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5498 | .start = s_start, |
5499 | .next = s_next, | |
5500 | .stop = s_stop, | |
5501 | .show = s_show, | |
5502 | }; | |
5503 | ||
7b3c3a50 AD |
5504 | static int slabinfo_open(struct inode *inode, struct file *file) |
5505 | { | |
5506 | return seq_open(file, &slabinfo_op); | |
5507 | } | |
5508 | ||
5509 | static const struct file_operations proc_slabinfo_operations = { | |
5510 | .open = slabinfo_open, | |
5511 | .read = seq_read, | |
5512 | .llseek = seq_lseek, | |
5513 | .release = seq_release, | |
5514 | }; | |
5515 | ||
5516 | static int __init slab_proc_init(void) | |
5517 | { | |
ab067e99 | 5518 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5519 | return 0; |
5520 | } | |
5521 | module_init(slab_proc_init); | |
158a9624 | 5522 | #endif /* CONFIG_SLABINFO */ |