]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
7b3c3a50 | 19 | #include <linux/proc_fs.h> |
81819f0f | 20 | #include <linux/seq_file.h> |
5a896d9e | 21 | #include <linux/kmemcheck.h> |
81819f0f CL |
22 | #include <linux/cpu.h> |
23 | #include <linux/cpuset.h> | |
24 | #include <linux/mempolicy.h> | |
25 | #include <linux/ctype.h> | |
3ac7fe5a | 26 | #include <linux/debugobjects.h> |
81819f0f | 27 | #include <linux/kallsyms.h> |
b9049e23 | 28 | #include <linux/memory.h> |
f8bd2258 | 29 | #include <linux/math64.h> |
773ff60e | 30 | #include <linux/fault-inject.h> |
bfa71457 | 31 | #include <linux/stacktrace.h> |
4de900b4 | 32 | #include <linux/prefetch.h> |
81819f0f | 33 | |
4a92379b RK |
34 | #include <trace/events/kmem.h> |
35 | ||
81819f0f CL |
36 | /* |
37 | * Lock order: | |
881db7fb CL |
38 | * 1. slub_lock (Global Semaphore) |
39 | * 2. node->list_lock | |
40 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 41 | * |
881db7fb CL |
42 | * slub_lock |
43 | * | |
44 | * The role of the slub_lock is to protect the list of all the slabs | |
45 | * and to synchronize major metadata changes to slab cache structures. | |
46 | * | |
47 | * The slab_lock is only used for debugging and on arches that do not | |
48 | * have the ability to do a cmpxchg_double. It only protects the second | |
49 | * double word in the page struct. Meaning | |
50 | * A. page->freelist -> List of object free in a page | |
51 | * B. page->counters -> Counters of objects | |
52 | * C. page->frozen -> frozen state | |
53 | * | |
54 | * If a slab is frozen then it is exempt from list management. It is not | |
55 | * on any list. The processor that froze the slab is the one who can | |
56 | * perform list operations on the page. Other processors may put objects | |
57 | * onto the freelist but the processor that froze the slab is the only | |
58 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
59 | * |
60 | * The list_lock protects the partial and full list on each node and | |
61 | * the partial slab counter. If taken then no new slabs may be added or | |
62 | * removed from the lists nor make the number of partial slabs be modified. | |
63 | * (Note that the total number of slabs is an atomic value that may be | |
64 | * modified without taking the list lock). | |
65 | * | |
66 | * The list_lock is a centralized lock and thus we avoid taking it as | |
67 | * much as possible. As long as SLUB does not have to handle partial | |
68 | * slabs, operations can continue without any centralized lock. F.e. | |
69 | * allocating a long series of objects that fill up slabs does not require | |
70 | * the list lock. | |
81819f0f CL |
71 | * Interrupts are disabled during allocation and deallocation in order to |
72 | * make the slab allocator safe to use in the context of an irq. In addition | |
73 | * interrupts are disabled to ensure that the processor does not change | |
74 | * while handling per_cpu slabs, due to kernel preemption. | |
75 | * | |
76 | * SLUB assigns one slab for allocation to each processor. | |
77 | * Allocations only occur from these slabs called cpu slabs. | |
78 | * | |
672bba3a CL |
79 | * Slabs with free elements are kept on a partial list and during regular |
80 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 81 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
82 | * We track full slabs for debugging purposes though because otherwise we |
83 | * cannot scan all objects. | |
81819f0f CL |
84 | * |
85 | * Slabs are freed when they become empty. Teardown and setup is | |
86 | * minimal so we rely on the page allocators per cpu caches for | |
87 | * fast frees and allocs. | |
88 | * | |
89 | * Overloading of page flags that are otherwise used for LRU management. | |
90 | * | |
4b6f0750 CL |
91 | * PageActive The slab is frozen and exempt from list processing. |
92 | * This means that the slab is dedicated to a purpose | |
93 | * such as satisfying allocations for a specific | |
94 | * processor. Objects may be freed in the slab while | |
95 | * it is frozen but slab_free will then skip the usual | |
96 | * list operations. It is up to the processor holding | |
97 | * the slab to integrate the slab into the slab lists | |
98 | * when the slab is no longer needed. | |
99 | * | |
100 | * One use of this flag is to mark slabs that are | |
101 | * used for allocations. Then such a slab becomes a cpu | |
102 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 103 | * freelist that allows lockless access to |
894b8788 CL |
104 | * free objects in addition to the regular freelist |
105 | * that requires the slab lock. | |
81819f0f CL |
106 | * |
107 | * PageError Slab requires special handling due to debug | |
108 | * options set. This moves slab handling out of | |
894b8788 | 109 | * the fast path and disables lockless freelists. |
81819f0f CL |
110 | */ |
111 | ||
af537b0a CL |
112 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ |
113 | SLAB_TRACE | SLAB_DEBUG_FREE) | |
114 | ||
115 | static inline int kmem_cache_debug(struct kmem_cache *s) | |
116 | { | |
5577bd8a | 117 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 118 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 119 | #else |
af537b0a | 120 | return 0; |
5577bd8a | 121 | #endif |
af537b0a | 122 | } |
5577bd8a | 123 | |
81819f0f CL |
124 | /* |
125 | * Issues still to be resolved: | |
126 | * | |
81819f0f CL |
127 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
128 | * | |
81819f0f CL |
129 | * - Variable sizing of the per node arrays |
130 | */ | |
131 | ||
132 | /* Enable to test recovery from slab corruption on boot */ | |
133 | #undef SLUB_RESILIENCY_TEST | |
134 | ||
b789ef51 CL |
135 | /* Enable to log cmpxchg failures */ |
136 | #undef SLUB_DEBUG_CMPXCHG | |
137 | ||
2086d26a CL |
138 | /* |
139 | * Mininum number of partial slabs. These will be left on the partial | |
140 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
141 | */ | |
76be8950 | 142 | #define MIN_PARTIAL 5 |
e95eed57 | 143 | |
2086d26a CL |
144 | /* |
145 | * Maximum number of desirable partial slabs. | |
146 | * The existence of more partial slabs makes kmem_cache_shrink | |
147 | * sort the partial list by the number of objects in the. | |
148 | */ | |
149 | #define MAX_PARTIAL 10 | |
150 | ||
81819f0f CL |
151 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
152 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 153 | |
fa5ec8a1 | 154 | /* |
3de47213 DR |
155 | * Debugging flags that require metadata to be stored in the slab. These get |
156 | * disabled when slub_debug=O is used and a cache's min order increases with | |
157 | * metadata. | |
fa5ec8a1 | 158 | */ |
3de47213 | 159 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 160 | |
81819f0f CL |
161 | /* |
162 | * Set of flags that will prevent slab merging | |
163 | */ | |
164 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
165 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
166 | SLAB_FAILSLAB) | |
81819f0f CL |
167 | |
168 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 169 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 170 | |
210b5c06 CG |
171 | #define OO_SHIFT 16 |
172 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 173 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 174 | |
81819f0f | 175 | /* Internal SLUB flags */ |
f90ec390 | 176 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 177 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f CL |
178 | |
179 | static int kmem_size = sizeof(struct kmem_cache); | |
180 | ||
181 | #ifdef CONFIG_SMP | |
182 | static struct notifier_block slab_notifier; | |
183 | #endif | |
184 | ||
185 | static enum { | |
186 | DOWN, /* No slab functionality available */ | |
51df1142 | 187 | PARTIAL, /* Kmem_cache_node works */ |
672bba3a | 188 | UP, /* Everything works but does not show up in sysfs */ |
81819f0f CL |
189 | SYSFS /* Sysfs up */ |
190 | } slab_state = DOWN; | |
191 | ||
192 | /* A list of all slab caches on the system */ | |
193 | static DECLARE_RWSEM(slub_lock); | |
5af328a5 | 194 | static LIST_HEAD(slab_caches); |
81819f0f | 195 | |
02cbc874 CL |
196 | /* |
197 | * Tracking user of a slab. | |
198 | */ | |
d6543e39 | 199 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 200 | struct track { |
ce71e27c | 201 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
202 | #ifdef CONFIG_STACKTRACE |
203 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
204 | #endif | |
02cbc874 CL |
205 | int cpu; /* Was running on cpu */ |
206 | int pid; /* Pid context */ | |
207 | unsigned long when; /* When did the operation occur */ | |
208 | }; | |
209 | ||
210 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
211 | ||
ab4d5ed5 | 212 | #ifdef CONFIG_SYSFS |
81819f0f CL |
213 | static int sysfs_slab_add(struct kmem_cache *); |
214 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
215 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 216 | |
81819f0f | 217 | #else |
0c710013 CL |
218 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
219 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
220 | { return 0; } | |
151c602f CL |
221 | static inline void sysfs_slab_remove(struct kmem_cache *s) |
222 | { | |
84c1cf62 | 223 | kfree(s->name); |
151c602f CL |
224 | kfree(s); |
225 | } | |
8ff12cfc | 226 | |
81819f0f CL |
227 | #endif |
228 | ||
4fdccdfb | 229 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
230 | { |
231 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 232 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
233 | #endif |
234 | } | |
235 | ||
81819f0f CL |
236 | /******************************************************************** |
237 | * Core slab cache functions | |
238 | *******************************************************************/ | |
239 | ||
240 | int slab_is_available(void) | |
241 | { | |
242 | return slab_state >= UP; | |
243 | } | |
244 | ||
245 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
246 | { | |
81819f0f | 247 | return s->node[node]; |
81819f0f CL |
248 | } |
249 | ||
6446faa2 | 250 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
251 | static inline int check_valid_pointer(struct kmem_cache *s, |
252 | struct page *page, const void *object) | |
253 | { | |
254 | void *base; | |
255 | ||
a973e9dd | 256 | if (!object) |
02cbc874 CL |
257 | return 1; |
258 | ||
a973e9dd | 259 | base = page_address(page); |
39b26464 | 260 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
261 | (object - base) % s->size) { |
262 | return 0; | |
263 | } | |
264 | ||
265 | return 1; | |
266 | } | |
267 | ||
7656c72b CL |
268 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
269 | { | |
270 | return *(void **)(object + s->offset); | |
271 | } | |
272 | ||
0ad9500e ED |
273 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
274 | { | |
275 | prefetch(object + s->offset); | |
276 | } | |
277 | ||
1393d9a1 CL |
278 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
279 | { | |
280 | void *p; | |
281 | ||
282 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
283 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
284 | #else | |
285 | p = get_freepointer(s, object); | |
286 | #endif | |
287 | return p; | |
288 | } | |
289 | ||
7656c72b CL |
290 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
291 | { | |
292 | *(void **)(object + s->offset) = fp; | |
293 | } | |
294 | ||
295 | /* Loop over all objects in a slab */ | |
224a88be CL |
296 | #define for_each_object(__p, __s, __addr, __objects) \ |
297 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
298 | __p += (__s)->size) |
299 | ||
7656c72b CL |
300 | /* Determine object index from a given position */ |
301 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
302 | { | |
303 | return (p - addr) / s->size; | |
304 | } | |
305 | ||
d71f606f MK |
306 | static inline size_t slab_ksize(const struct kmem_cache *s) |
307 | { | |
308 | #ifdef CONFIG_SLUB_DEBUG | |
309 | /* | |
310 | * Debugging requires use of the padding between object | |
311 | * and whatever may come after it. | |
312 | */ | |
313 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
314 | return s->objsize; | |
315 | ||
316 | #endif | |
317 | /* | |
318 | * If we have the need to store the freelist pointer | |
319 | * back there or track user information then we can | |
320 | * only use the space before that information. | |
321 | */ | |
322 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
323 | return s->inuse; | |
324 | /* | |
325 | * Else we can use all the padding etc for the allocation | |
326 | */ | |
327 | return s->size; | |
328 | } | |
329 | ||
ab9a0f19 LJ |
330 | static inline int order_objects(int order, unsigned long size, int reserved) |
331 | { | |
332 | return ((PAGE_SIZE << order) - reserved) / size; | |
333 | } | |
334 | ||
834f3d11 | 335 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 336 | unsigned long size, int reserved) |
834f3d11 CL |
337 | { |
338 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 339 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
340 | }; |
341 | ||
342 | return x; | |
343 | } | |
344 | ||
345 | static inline int oo_order(struct kmem_cache_order_objects x) | |
346 | { | |
210b5c06 | 347 | return x.x >> OO_SHIFT; |
834f3d11 CL |
348 | } |
349 | ||
350 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
351 | { | |
210b5c06 | 352 | return x.x & OO_MASK; |
834f3d11 CL |
353 | } |
354 | ||
881db7fb CL |
355 | /* |
356 | * Per slab locking using the pagelock | |
357 | */ | |
358 | static __always_inline void slab_lock(struct page *page) | |
359 | { | |
360 | bit_spin_lock(PG_locked, &page->flags); | |
361 | } | |
362 | ||
363 | static __always_inline void slab_unlock(struct page *page) | |
364 | { | |
365 | __bit_spin_unlock(PG_locked, &page->flags); | |
366 | } | |
367 | ||
1d07171c CL |
368 | /* Interrupts must be disabled (for the fallback code to work right) */ |
369 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
370 | void *freelist_old, unsigned long counters_old, | |
371 | void *freelist_new, unsigned long counters_new, | |
372 | const char *n) | |
373 | { | |
374 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
375 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
376 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 377 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 378 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
379 | freelist_old, counters_old, |
380 | freelist_new, counters_new)) | |
381 | return 1; | |
382 | } else | |
383 | #endif | |
384 | { | |
385 | slab_lock(page); | |
386 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
387 | page->freelist = freelist_new; | |
388 | page->counters = counters_new; | |
389 | slab_unlock(page); | |
390 | return 1; | |
391 | } | |
392 | slab_unlock(page); | |
393 | } | |
394 | ||
395 | cpu_relax(); | |
396 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
397 | ||
398 | #ifdef SLUB_DEBUG_CMPXCHG | |
399 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
400 | #endif | |
401 | ||
402 | return 0; | |
403 | } | |
404 | ||
b789ef51 CL |
405 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
406 | void *freelist_old, unsigned long counters_old, | |
407 | void *freelist_new, unsigned long counters_new, | |
408 | const char *n) | |
409 | { | |
2565409f HC |
410 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
411 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 412 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 413 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
414 | freelist_old, counters_old, |
415 | freelist_new, counters_new)) | |
416 | return 1; | |
417 | } else | |
418 | #endif | |
419 | { | |
1d07171c CL |
420 | unsigned long flags; |
421 | ||
422 | local_irq_save(flags); | |
881db7fb | 423 | slab_lock(page); |
b789ef51 CL |
424 | if (page->freelist == freelist_old && page->counters == counters_old) { |
425 | page->freelist = freelist_new; | |
426 | page->counters = counters_new; | |
881db7fb | 427 | slab_unlock(page); |
1d07171c | 428 | local_irq_restore(flags); |
b789ef51 CL |
429 | return 1; |
430 | } | |
881db7fb | 431 | slab_unlock(page); |
1d07171c | 432 | local_irq_restore(flags); |
b789ef51 CL |
433 | } |
434 | ||
435 | cpu_relax(); | |
436 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
437 | ||
438 | #ifdef SLUB_DEBUG_CMPXCHG | |
439 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
440 | #endif | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
41ecc55b | 445 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
446 | /* |
447 | * Determine a map of object in use on a page. | |
448 | * | |
881db7fb | 449 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
450 | * not vanish from under us. |
451 | */ | |
452 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
453 | { | |
454 | void *p; | |
455 | void *addr = page_address(page); | |
456 | ||
457 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
458 | set_bit(slab_index(p, s, addr), map); | |
459 | } | |
460 | ||
41ecc55b CL |
461 | /* |
462 | * Debug settings: | |
463 | */ | |
f0630fff CL |
464 | #ifdef CONFIG_SLUB_DEBUG_ON |
465 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
466 | #else | |
41ecc55b | 467 | static int slub_debug; |
f0630fff | 468 | #endif |
41ecc55b CL |
469 | |
470 | static char *slub_debug_slabs; | |
fa5ec8a1 | 471 | static int disable_higher_order_debug; |
41ecc55b | 472 | |
81819f0f CL |
473 | /* |
474 | * Object debugging | |
475 | */ | |
476 | static void print_section(char *text, u8 *addr, unsigned int length) | |
477 | { | |
ffc79d28 SAS |
478 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
479 | length, 1); | |
81819f0f CL |
480 | } |
481 | ||
81819f0f CL |
482 | static struct track *get_track(struct kmem_cache *s, void *object, |
483 | enum track_item alloc) | |
484 | { | |
485 | struct track *p; | |
486 | ||
487 | if (s->offset) | |
488 | p = object + s->offset + sizeof(void *); | |
489 | else | |
490 | p = object + s->inuse; | |
491 | ||
492 | return p + alloc; | |
493 | } | |
494 | ||
495 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 496 | enum track_item alloc, unsigned long addr) |
81819f0f | 497 | { |
1a00df4a | 498 | struct track *p = get_track(s, object, alloc); |
81819f0f | 499 | |
81819f0f | 500 | if (addr) { |
d6543e39 BG |
501 | #ifdef CONFIG_STACKTRACE |
502 | struct stack_trace trace; | |
503 | int i; | |
504 | ||
505 | trace.nr_entries = 0; | |
506 | trace.max_entries = TRACK_ADDRS_COUNT; | |
507 | trace.entries = p->addrs; | |
508 | trace.skip = 3; | |
509 | save_stack_trace(&trace); | |
510 | ||
511 | /* See rant in lockdep.c */ | |
512 | if (trace.nr_entries != 0 && | |
513 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
514 | trace.nr_entries--; | |
515 | ||
516 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
517 | p->addrs[i] = 0; | |
518 | #endif | |
81819f0f CL |
519 | p->addr = addr; |
520 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 521 | p->pid = current->pid; |
81819f0f CL |
522 | p->when = jiffies; |
523 | } else | |
524 | memset(p, 0, sizeof(struct track)); | |
525 | } | |
526 | ||
81819f0f CL |
527 | static void init_tracking(struct kmem_cache *s, void *object) |
528 | { | |
24922684 CL |
529 | if (!(s->flags & SLAB_STORE_USER)) |
530 | return; | |
531 | ||
ce71e27c EGM |
532 | set_track(s, object, TRACK_FREE, 0UL); |
533 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
534 | } |
535 | ||
536 | static void print_track(const char *s, struct track *t) | |
537 | { | |
538 | if (!t->addr) | |
539 | return; | |
540 | ||
7daf705f | 541 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 542 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
543 | #ifdef CONFIG_STACKTRACE |
544 | { | |
545 | int i; | |
546 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
547 | if (t->addrs[i]) | |
548 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
549 | else | |
550 | break; | |
551 | } | |
552 | #endif | |
24922684 CL |
553 | } |
554 | ||
555 | static void print_tracking(struct kmem_cache *s, void *object) | |
556 | { | |
557 | if (!(s->flags & SLAB_STORE_USER)) | |
558 | return; | |
559 | ||
560 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
561 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
562 | } | |
563 | ||
564 | static void print_page_info(struct page *page) | |
565 | { | |
39b26464 CL |
566 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
567 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
568 | |
569 | } | |
570 | ||
571 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
572 | { | |
573 | va_list args; | |
574 | char buf[100]; | |
575 | ||
576 | va_start(args, fmt); | |
577 | vsnprintf(buf, sizeof(buf), fmt, args); | |
578 | va_end(args); | |
579 | printk(KERN_ERR "========================================" | |
580 | "=====================================\n"); | |
265d47e7 | 581 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
582 | printk(KERN_ERR "----------------------------------------" |
583 | "-------------------------------------\n\n"); | |
81819f0f CL |
584 | } |
585 | ||
24922684 CL |
586 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
587 | { | |
588 | va_list args; | |
589 | char buf[100]; | |
590 | ||
591 | va_start(args, fmt); | |
592 | vsnprintf(buf, sizeof(buf), fmt, args); | |
593 | va_end(args); | |
594 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
595 | } | |
596 | ||
597 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
598 | { |
599 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 600 | u8 *addr = page_address(page); |
24922684 CL |
601 | |
602 | print_tracking(s, p); | |
603 | ||
604 | print_page_info(page); | |
605 | ||
606 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
607 | p, p - addr, get_freepointer(s, p)); | |
608 | ||
609 | if (p > addr + 16) | |
ffc79d28 | 610 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 611 | |
ffc79d28 SAS |
612 | print_section("Object ", p, min_t(unsigned long, s->objsize, |
613 | PAGE_SIZE)); | |
81819f0f | 614 | if (s->flags & SLAB_RED_ZONE) |
ffc79d28 | 615 | print_section("Redzone ", p + s->objsize, |
81819f0f CL |
616 | s->inuse - s->objsize); |
617 | ||
81819f0f CL |
618 | if (s->offset) |
619 | off = s->offset + sizeof(void *); | |
620 | else | |
621 | off = s->inuse; | |
622 | ||
24922684 | 623 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 624 | off += 2 * sizeof(struct track); |
81819f0f CL |
625 | |
626 | if (off != s->size) | |
627 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 628 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
629 | |
630 | dump_stack(); | |
81819f0f CL |
631 | } |
632 | ||
633 | static void object_err(struct kmem_cache *s, struct page *page, | |
634 | u8 *object, char *reason) | |
635 | { | |
3dc50637 | 636 | slab_bug(s, "%s", reason); |
24922684 | 637 | print_trailer(s, page, object); |
81819f0f CL |
638 | } |
639 | ||
24922684 | 640 | static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...) |
81819f0f CL |
641 | { |
642 | va_list args; | |
643 | char buf[100]; | |
644 | ||
24922684 CL |
645 | va_start(args, fmt); |
646 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 647 | va_end(args); |
3dc50637 | 648 | slab_bug(s, "%s", buf); |
24922684 | 649 | print_page_info(page); |
81819f0f CL |
650 | dump_stack(); |
651 | } | |
652 | ||
f7cb1933 | 653 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
654 | { |
655 | u8 *p = object; | |
656 | ||
657 | if (s->flags & __OBJECT_POISON) { | |
658 | memset(p, POISON_FREE, s->objsize - 1); | |
06428780 | 659 | p[s->objsize - 1] = POISON_END; |
81819f0f CL |
660 | } |
661 | ||
662 | if (s->flags & SLAB_RED_ZONE) | |
f7cb1933 | 663 | memset(p + s->objsize, val, s->inuse - s->objsize); |
81819f0f CL |
664 | } |
665 | ||
24922684 CL |
666 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
667 | void *from, void *to) | |
668 | { | |
669 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
670 | memset(from, data, to - from); | |
671 | } | |
672 | ||
673 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
674 | u8 *object, char *what, | |
06428780 | 675 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
676 | { |
677 | u8 *fault; | |
678 | u8 *end; | |
679 | ||
79824820 | 680 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
681 | if (!fault) |
682 | return 1; | |
683 | ||
684 | end = start + bytes; | |
685 | while (end > fault && end[-1] == value) | |
686 | end--; | |
687 | ||
688 | slab_bug(s, "%s overwritten", what); | |
689 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
690 | fault, end - 1, fault[0], value); | |
691 | print_trailer(s, page, object); | |
692 | ||
693 | restore_bytes(s, what, value, fault, end); | |
694 | return 0; | |
81819f0f CL |
695 | } |
696 | ||
81819f0f CL |
697 | /* |
698 | * Object layout: | |
699 | * | |
700 | * object address | |
701 | * Bytes of the object to be managed. | |
702 | * If the freepointer may overlay the object then the free | |
703 | * pointer is the first word of the object. | |
672bba3a | 704 | * |
81819f0f CL |
705 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
706 | * 0xa5 (POISON_END) | |
707 | * | |
708 | * object + s->objsize | |
709 | * Padding to reach word boundary. This is also used for Redzoning. | |
672bba3a CL |
710 | * Padding is extended by another word if Redzoning is enabled and |
711 | * objsize == inuse. | |
712 | * | |
81819f0f CL |
713 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
714 | * 0xcc (RED_ACTIVE) for objects in use. | |
715 | * | |
716 | * object + s->inuse | |
672bba3a CL |
717 | * Meta data starts here. |
718 | * | |
81819f0f CL |
719 | * A. Free pointer (if we cannot overwrite object on free) |
720 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 721 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 722 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
723 | * before the word boundary. |
724 | * | |
725 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
726 | * |
727 | * object + s->size | |
672bba3a | 728 | * Nothing is used beyond s->size. |
81819f0f | 729 | * |
672bba3a CL |
730 | * If slabcaches are merged then the objsize and inuse boundaries are mostly |
731 | * ignored. And therefore no slab options that rely on these boundaries | |
81819f0f CL |
732 | * may be used with merged slabcaches. |
733 | */ | |
734 | ||
81819f0f CL |
735 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
736 | { | |
737 | unsigned long off = s->inuse; /* The end of info */ | |
738 | ||
739 | if (s->offset) | |
740 | /* Freepointer is placed after the object. */ | |
741 | off += sizeof(void *); | |
742 | ||
743 | if (s->flags & SLAB_STORE_USER) | |
744 | /* We also have user information there */ | |
745 | off += 2 * sizeof(struct track); | |
746 | ||
747 | if (s->size == off) | |
748 | return 1; | |
749 | ||
24922684 CL |
750 | return check_bytes_and_report(s, page, p, "Object padding", |
751 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
752 | } |
753 | ||
39b26464 | 754 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
755 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
756 | { | |
24922684 CL |
757 | u8 *start; |
758 | u8 *fault; | |
759 | u8 *end; | |
760 | int length; | |
761 | int remainder; | |
81819f0f CL |
762 | |
763 | if (!(s->flags & SLAB_POISON)) | |
764 | return 1; | |
765 | ||
a973e9dd | 766 | start = page_address(page); |
ab9a0f19 | 767 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
768 | end = start + length; |
769 | remainder = length % s->size; | |
81819f0f CL |
770 | if (!remainder) |
771 | return 1; | |
772 | ||
79824820 | 773 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
774 | if (!fault) |
775 | return 1; | |
776 | while (end > fault && end[-1] == POISON_INUSE) | |
777 | end--; | |
778 | ||
779 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 780 | print_section("Padding ", end - remainder, remainder); |
24922684 | 781 | |
8a3d271d | 782 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 783 | return 0; |
81819f0f CL |
784 | } |
785 | ||
786 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 787 | void *object, u8 val) |
81819f0f CL |
788 | { |
789 | u8 *p = object; | |
790 | u8 *endobject = object + s->objsize; | |
791 | ||
792 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 793 | if (!check_bytes_and_report(s, page, object, "Redzone", |
f7cb1933 | 794 | endobject, val, s->inuse - s->objsize)) |
81819f0f | 795 | return 0; |
81819f0f | 796 | } else { |
3adbefee IM |
797 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) { |
798 | check_bytes_and_report(s, page, p, "Alignment padding", | |
799 | endobject, POISON_INUSE, s->inuse - s->objsize); | |
800 | } | |
81819f0f CL |
801 | } |
802 | ||
803 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 804 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 CL |
805 | (!check_bytes_and_report(s, page, p, "Poison", p, |
806 | POISON_FREE, s->objsize - 1) || | |
807 | !check_bytes_and_report(s, page, p, "Poison", | |
06428780 | 808 | p + s->objsize - 1, POISON_END, 1))) |
81819f0f | 809 | return 0; |
81819f0f CL |
810 | /* |
811 | * check_pad_bytes cleans up on its own. | |
812 | */ | |
813 | check_pad_bytes(s, page, p); | |
814 | } | |
815 | ||
f7cb1933 | 816 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
817 | /* |
818 | * Object and freepointer overlap. Cannot check | |
819 | * freepointer while object is allocated. | |
820 | */ | |
821 | return 1; | |
822 | ||
823 | /* Check free pointer validity */ | |
824 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
825 | object_err(s, page, p, "Freepointer corrupt"); | |
826 | /* | |
9f6c708e | 827 | * No choice but to zap it and thus lose the remainder |
81819f0f | 828 | * of the free objects in this slab. May cause |
672bba3a | 829 | * another error because the object count is now wrong. |
81819f0f | 830 | */ |
a973e9dd | 831 | set_freepointer(s, p, NULL); |
81819f0f CL |
832 | return 0; |
833 | } | |
834 | return 1; | |
835 | } | |
836 | ||
837 | static int check_slab(struct kmem_cache *s, struct page *page) | |
838 | { | |
39b26464 CL |
839 | int maxobj; |
840 | ||
81819f0f CL |
841 | VM_BUG_ON(!irqs_disabled()); |
842 | ||
843 | if (!PageSlab(page)) { | |
24922684 | 844 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
845 | return 0; |
846 | } | |
39b26464 | 847 | |
ab9a0f19 | 848 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
849 | if (page->objects > maxobj) { |
850 | slab_err(s, page, "objects %u > max %u", | |
851 | s->name, page->objects, maxobj); | |
852 | return 0; | |
853 | } | |
854 | if (page->inuse > page->objects) { | |
24922684 | 855 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 856 | s->name, page->inuse, page->objects); |
81819f0f CL |
857 | return 0; |
858 | } | |
859 | /* Slab_pad_check fixes things up after itself */ | |
860 | slab_pad_check(s, page); | |
861 | return 1; | |
862 | } | |
863 | ||
864 | /* | |
672bba3a CL |
865 | * Determine if a certain object on a page is on the freelist. Must hold the |
866 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
867 | */ |
868 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
869 | { | |
870 | int nr = 0; | |
881db7fb | 871 | void *fp; |
81819f0f | 872 | void *object = NULL; |
224a88be | 873 | unsigned long max_objects; |
81819f0f | 874 | |
881db7fb | 875 | fp = page->freelist; |
39b26464 | 876 | while (fp && nr <= page->objects) { |
81819f0f CL |
877 | if (fp == search) |
878 | return 1; | |
879 | if (!check_valid_pointer(s, page, fp)) { | |
880 | if (object) { | |
881 | object_err(s, page, object, | |
882 | "Freechain corrupt"); | |
a973e9dd | 883 | set_freepointer(s, object, NULL); |
81819f0f CL |
884 | break; |
885 | } else { | |
24922684 | 886 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 887 | page->freelist = NULL; |
39b26464 | 888 | page->inuse = page->objects; |
24922684 | 889 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
890 | return 0; |
891 | } | |
892 | break; | |
893 | } | |
894 | object = fp; | |
895 | fp = get_freepointer(s, object); | |
896 | nr++; | |
897 | } | |
898 | ||
ab9a0f19 | 899 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
900 | if (max_objects > MAX_OBJS_PER_PAGE) |
901 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
902 | |
903 | if (page->objects != max_objects) { | |
904 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
905 | "should be %d", page->objects, max_objects); | |
906 | page->objects = max_objects; | |
907 | slab_fix(s, "Number of objects adjusted."); | |
908 | } | |
39b26464 | 909 | if (page->inuse != page->objects - nr) { |
70d71228 | 910 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
911 | "counted were %d", page->inuse, page->objects - nr); |
912 | page->inuse = page->objects - nr; | |
24922684 | 913 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
914 | } |
915 | return search == NULL; | |
916 | } | |
917 | ||
0121c619 CL |
918 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
919 | int alloc) | |
3ec09742 CL |
920 | { |
921 | if (s->flags & SLAB_TRACE) { | |
922 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
923 | s->name, | |
924 | alloc ? "alloc" : "free", | |
925 | object, page->inuse, | |
926 | page->freelist); | |
927 | ||
928 | if (!alloc) | |
ffc79d28 | 929 | print_section("Object ", (void *)object, s->objsize); |
3ec09742 CL |
930 | |
931 | dump_stack(); | |
932 | } | |
933 | } | |
934 | ||
c016b0bd CL |
935 | /* |
936 | * Hooks for other subsystems that check memory allocations. In a typical | |
937 | * production configuration these hooks all should produce no code at all. | |
938 | */ | |
939 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
940 | { | |
c1d50836 | 941 | flags &= gfp_allowed_mask; |
c016b0bd CL |
942 | lockdep_trace_alloc(flags); |
943 | might_sleep_if(flags & __GFP_WAIT); | |
944 | ||
945 | return should_failslab(s->objsize, flags, s->flags); | |
946 | } | |
947 | ||
948 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
949 | { | |
c1d50836 | 950 | flags &= gfp_allowed_mask; |
b3d41885 | 951 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
c016b0bd CL |
952 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
953 | } | |
954 | ||
955 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
956 | { | |
957 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 958 | |
d3f661d6 CL |
959 | /* |
960 | * Trouble is that we may no longer disable interupts in the fast path | |
961 | * So in order to make the debug calls that expect irqs to be | |
962 | * disabled we need to disable interrupts temporarily. | |
963 | */ | |
964 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
965 | { | |
966 | unsigned long flags; | |
967 | ||
968 | local_irq_save(flags); | |
969 | kmemcheck_slab_free(s, x, s->objsize); | |
970 | debug_check_no_locks_freed(x, s->objsize); | |
d3f661d6 CL |
971 | local_irq_restore(flags); |
972 | } | |
973 | #endif | |
f9b615de TG |
974 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
975 | debug_check_no_obj_freed(x, s->objsize); | |
c016b0bd CL |
976 | } |
977 | ||
643b1138 | 978 | /* |
672bba3a | 979 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
980 | * |
981 | * list_lock must be held. | |
643b1138 | 982 | */ |
5cc6eee8 CL |
983 | static void add_full(struct kmem_cache *s, |
984 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 985 | { |
5cc6eee8 CL |
986 | if (!(s->flags & SLAB_STORE_USER)) |
987 | return; | |
988 | ||
643b1138 | 989 | list_add(&page->lru, &n->full); |
643b1138 CL |
990 | } |
991 | ||
5cc6eee8 CL |
992 | /* |
993 | * list_lock must be held. | |
994 | */ | |
643b1138 CL |
995 | static void remove_full(struct kmem_cache *s, struct page *page) |
996 | { | |
643b1138 CL |
997 | if (!(s->flags & SLAB_STORE_USER)) |
998 | return; | |
999 | ||
643b1138 | 1000 | list_del(&page->lru); |
643b1138 CL |
1001 | } |
1002 | ||
0f389ec6 CL |
1003 | /* Tracking of the number of slabs for debugging purposes */ |
1004 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1005 | { | |
1006 | struct kmem_cache_node *n = get_node(s, node); | |
1007 | ||
1008 | return atomic_long_read(&n->nr_slabs); | |
1009 | } | |
1010 | ||
26c02cf0 AB |
1011 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1012 | { | |
1013 | return atomic_long_read(&n->nr_slabs); | |
1014 | } | |
1015 | ||
205ab99d | 1016 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1017 | { |
1018 | struct kmem_cache_node *n = get_node(s, node); | |
1019 | ||
1020 | /* | |
1021 | * May be called early in order to allocate a slab for the | |
1022 | * kmem_cache_node structure. Solve the chicken-egg | |
1023 | * dilemma by deferring the increment of the count during | |
1024 | * bootstrap (see early_kmem_cache_node_alloc). | |
1025 | */ | |
7340cc84 | 1026 | if (n) { |
0f389ec6 | 1027 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1028 | atomic_long_add(objects, &n->total_objects); |
1029 | } | |
0f389ec6 | 1030 | } |
205ab99d | 1031 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1032 | { |
1033 | struct kmem_cache_node *n = get_node(s, node); | |
1034 | ||
1035 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1036 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1037 | } |
1038 | ||
1039 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1040 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1041 | void *object) | |
1042 | { | |
1043 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1044 | return; | |
1045 | ||
f7cb1933 | 1046 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1047 | init_tracking(s, object); |
1048 | } | |
1049 | ||
1537066c | 1050 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1051 | void *object, unsigned long addr) |
81819f0f CL |
1052 | { |
1053 | if (!check_slab(s, page)) | |
1054 | goto bad; | |
1055 | ||
81819f0f CL |
1056 | if (!check_valid_pointer(s, page, object)) { |
1057 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1058 | goto bad; |
81819f0f CL |
1059 | } |
1060 | ||
f7cb1933 | 1061 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1062 | goto bad; |
81819f0f | 1063 | |
3ec09742 CL |
1064 | /* Success perform special debug activities for allocs */ |
1065 | if (s->flags & SLAB_STORE_USER) | |
1066 | set_track(s, object, TRACK_ALLOC, addr); | |
1067 | trace(s, page, object, 1); | |
f7cb1933 | 1068 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1069 | return 1; |
3ec09742 | 1070 | |
81819f0f CL |
1071 | bad: |
1072 | if (PageSlab(page)) { | |
1073 | /* | |
1074 | * If this is a slab page then lets do the best we can | |
1075 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1076 | * as used avoids touching the remaining objects. |
81819f0f | 1077 | */ |
24922684 | 1078 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1079 | page->inuse = page->objects; |
a973e9dd | 1080 | page->freelist = NULL; |
81819f0f CL |
1081 | } |
1082 | return 0; | |
1083 | } | |
1084 | ||
1537066c CL |
1085 | static noinline int free_debug_processing(struct kmem_cache *s, |
1086 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1087 | { |
5c2e4bbb CL |
1088 | unsigned long flags; |
1089 | int rc = 0; | |
1090 | ||
1091 | local_irq_save(flags); | |
881db7fb CL |
1092 | slab_lock(page); |
1093 | ||
81819f0f CL |
1094 | if (!check_slab(s, page)) |
1095 | goto fail; | |
1096 | ||
1097 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1098 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1099 | goto fail; |
1100 | } | |
1101 | ||
1102 | if (on_freelist(s, page, object)) { | |
24922684 | 1103 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1104 | goto fail; |
1105 | } | |
1106 | ||
f7cb1933 | 1107 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1108 | goto out; |
81819f0f CL |
1109 | |
1110 | if (unlikely(s != page->slab)) { | |
3adbefee | 1111 | if (!PageSlab(page)) { |
70d71228 CL |
1112 | slab_err(s, page, "Attempt to free object(0x%p) " |
1113 | "outside of slab", object); | |
3adbefee | 1114 | } else if (!page->slab) { |
81819f0f | 1115 | printk(KERN_ERR |
70d71228 | 1116 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1117 | object); |
70d71228 | 1118 | dump_stack(); |
06428780 | 1119 | } else |
24922684 CL |
1120 | object_err(s, page, object, |
1121 | "page slab pointer corrupt."); | |
81819f0f CL |
1122 | goto fail; |
1123 | } | |
3ec09742 | 1124 | |
3ec09742 CL |
1125 | if (s->flags & SLAB_STORE_USER) |
1126 | set_track(s, object, TRACK_FREE, addr); | |
1127 | trace(s, page, object, 0); | |
f7cb1933 | 1128 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb CL |
1129 | rc = 1; |
1130 | out: | |
881db7fb | 1131 | slab_unlock(page); |
5c2e4bbb CL |
1132 | local_irq_restore(flags); |
1133 | return rc; | |
3ec09742 | 1134 | |
81819f0f | 1135 | fail: |
24922684 | 1136 | slab_fix(s, "Object at 0x%p not freed", object); |
5c2e4bbb | 1137 | goto out; |
81819f0f CL |
1138 | } |
1139 | ||
41ecc55b CL |
1140 | static int __init setup_slub_debug(char *str) |
1141 | { | |
f0630fff CL |
1142 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1143 | if (*str++ != '=' || !*str) | |
1144 | /* | |
1145 | * No options specified. Switch on full debugging. | |
1146 | */ | |
1147 | goto out; | |
1148 | ||
1149 | if (*str == ',') | |
1150 | /* | |
1151 | * No options but restriction on slabs. This means full | |
1152 | * debugging for slabs matching a pattern. | |
1153 | */ | |
1154 | goto check_slabs; | |
1155 | ||
fa5ec8a1 DR |
1156 | if (tolower(*str) == 'o') { |
1157 | /* | |
1158 | * Avoid enabling debugging on caches if its minimum order | |
1159 | * would increase as a result. | |
1160 | */ | |
1161 | disable_higher_order_debug = 1; | |
1162 | goto out; | |
1163 | } | |
1164 | ||
f0630fff CL |
1165 | slub_debug = 0; |
1166 | if (*str == '-') | |
1167 | /* | |
1168 | * Switch off all debugging measures. | |
1169 | */ | |
1170 | goto out; | |
1171 | ||
1172 | /* | |
1173 | * Determine which debug features should be switched on | |
1174 | */ | |
06428780 | 1175 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1176 | switch (tolower(*str)) { |
1177 | case 'f': | |
1178 | slub_debug |= SLAB_DEBUG_FREE; | |
1179 | break; | |
1180 | case 'z': | |
1181 | slub_debug |= SLAB_RED_ZONE; | |
1182 | break; | |
1183 | case 'p': | |
1184 | slub_debug |= SLAB_POISON; | |
1185 | break; | |
1186 | case 'u': | |
1187 | slub_debug |= SLAB_STORE_USER; | |
1188 | break; | |
1189 | case 't': | |
1190 | slub_debug |= SLAB_TRACE; | |
1191 | break; | |
4c13dd3b DM |
1192 | case 'a': |
1193 | slub_debug |= SLAB_FAILSLAB; | |
1194 | break; | |
f0630fff CL |
1195 | default: |
1196 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1197 | "unknown. skipped\n", *str); |
f0630fff | 1198 | } |
41ecc55b CL |
1199 | } |
1200 | ||
f0630fff | 1201 | check_slabs: |
41ecc55b CL |
1202 | if (*str == ',') |
1203 | slub_debug_slabs = str + 1; | |
f0630fff | 1204 | out: |
41ecc55b CL |
1205 | return 1; |
1206 | } | |
1207 | ||
1208 | __setup("slub_debug", setup_slub_debug); | |
1209 | ||
ba0268a8 CL |
1210 | static unsigned long kmem_cache_flags(unsigned long objsize, |
1211 | unsigned long flags, const char *name, | |
51cc5068 | 1212 | void (*ctor)(void *)) |
41ecc55b CL |
1213 | { |
1214 | /* | |
e153362a | 1215 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1216 | */ |
e153362a | 1217 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1218 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1219 | flags |= slub_debug; | |
ba0268a8 CL |
1220 | |
1221 | return flags; | |
41ecc55b CL |
1222 | } |
1223 | #else | |
3ec09742 CL |
1224 | static inline void setup_object_debug(struct kmem_cache *s, |
1225 | struct page *page, void *object) {} | |
41ecc55b | 1226 | |
3ec09742 | 1227 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1228 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1229 | |
3ec09742 | 1230 | static inline int free_debug_processing(struct kmem_cache *s, |
ce71e27c | 1231 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1232 | |
41ecc55b CL |
1233 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1234 | { return 1; } | |
1235 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1236 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1237 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1238 | struct page *page) {} | |
2cfb7455 | 1239 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
ba0268a8 CL |
1240 | static inline unsigned long kmem_cache_flags(unsigned long objsize, |
1241 | unsigned long flags, const char *name, | |
51cc5068 | 1242 | void (*ctor)(void *)) |
ba0268a8 CL |
1243 | { |
1244 | return flags; | |
1245 | } | |
41ecc55b | 1246 | #define slub_debug 0 |
0f389ec6 | 1247 | |
fdaa45e9 IM |
1248 | #define disable_higher_order_debug 0 |
1249 | ||
0f389ec6 CL |
1250 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1251 | { return 0; } | |
26c02cf0 AB |
1252 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1253 | { return 0; } | |
205ab99d CL |
1254 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1255 | int objects) {} | |
1256 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1257 | int objects) {} | |
7d550c56 CL |
1258 | |
1259 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1260 | { return 0; } | |
1261 | ||
1262 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1263 | void *object) {} | |
1264 | ||
1265 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1266 | ||
ab4d5ed5 | 1267 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1268 | |
81819f0f CL |
1269 | /* |
1270 | * Slab allocation and freeing | |
1271 | */ | |
65c3376a CL |
1272 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1273 | struct kmem_cache_order_objects oo) | |
1274 | { | |
1275 | int order = oo_order(oo); | |
1276 | ||
b1eeab67 VN |
1277 | flags |= __GFP_NOTRACK; |
1278 | ||
2154a336 | 1279 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1280 | return alloc_pages(flags, order); |
1281 | else | |
6b65aaf3 | 1282 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1283 | } |
1284 | ||
81819f0f CL |
1285 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1286 | { | |
06428780 | 1287 | struct page *page; |
834f3d11 | 1288 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1289 | gfp_t alloc_gfp; |
81819f0f | 1290 | |
7e0528da CL |
1291 | flags &= gfp_allowed_mask; |
1292 | ||
1293 | if (flags & __GFP_WAIT) | |
1294 | local_irq_enable(); | |
1295 | ||
b7a49f0d | 1296 | flags |= s->allocflags; |
e12ba74d | 1297 | |
ba52270d PE |
1298 | /* |
1299 | * Let the initial higher-order allocation fail under memory pressure | |
1300 | * so we fall-back to the minimum order allocation. | |
1301 | */ | |
1302 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1303 | ||
1304 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1305 | if (unlikely(!page)) { |
1306 | oo = s->min; | |
1307 | /* | |
1308 | * Allocation may have failed due to fragmentation. | |
1309 | * Try a lower order alloc if possible | |
1310 | */ | |
1311 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1312 | |
7e0528da CL |
1313 | if (page) |
1314 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1315 | } |
5a896d9e | 1316 | |
7e0528da CL |
1317 | if (flags & __GFP_WAIT) |
1318 | local_irq_disable(); | |
1319 | ||
1320 | if (!page) | |
1321 | return NULL; | |
1322 | ||
5a896d9e | 1323 | if (kmemcheck_enabled |
5086c389 | 1324 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1325 | int pages = 1 << oo_order(oo); |
1326 | ||
1327 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1328 | ||
1329 | /* | |
1330 | * Objects from caches that have a constructor don't get | |
1331 | * cleared when they're allocated, so we need to do it here. | |
1332 | */ | |
1333 | if (s->ctor) | |
1334 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1335 | else | |
1336 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1337 | } |
1338 | ||
834f3d11 | 1339 | page->objects = oo_objects(oo); |
81819f0f CL |
1340 | mod_zone_page_state(page_zone(page), |
1341 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1342 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1343 | 1 << oo_order(oo)); |
81819f0f CL |
1344 | |
1345 | return page; | |
1346 | } | |
1347 | ||
1348 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1349 | void *object) | |
1350 | { | |
3ec09742 | 1351 | setup_object_debug(s, page, object); |
4f104934 | 1352 | if (unlikely(s->ctor)) |
51cc5068 | 1353 | s->ctor(object); |
81819f0f CL |
1354 | } |
1355 | ||
1356 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1357 | { | |
1358 | struct page *page; | |
81819f0f | 1359 | void *start; |
81819f0f CL |
1360 | void *last; |
1361 | void *p; | |
1362 | ||
6cb06229 | 1363 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1364 | |
6cb06229 CL |
1365 | page = allocate_slab(s, |
1366 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1367 | if (!page) |
1368 | goto out; | |
1369 | ||
205ab99d | 1370 | inc_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1371 | page->slab = s; |
1372 | page->flags |= 1 << PG_slab; | |
81819f0f CL |
1373 | |
1374 | start = page_address(page); | |
81819f0f CL |
1375 | |
1376 | if (unlikely(s->flags & SLAB_POISON)) | |
834f3d11 | 1377 | memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page)); |
81819f0f CL |
1378 | |
1379 | last = start; | |
224a88be | 1380 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1381 | setup_object(s, page, last); |
1382 | set_freepointer(s, last, p); | |
1383 | last = p; | |
1384 | } | |
1385 | setup_object(s, page, last); | |
a973e9dd | 1386 | set_freepointer(s, last, NULL); |
81819f0f CL |
1387 | |
1388 | page->freelist = start; | |
e6e82ea1 | 1389 | page->inuse = page->objects; |
8cb0a506 | 1390 | page->frozen = 1; |
81819f0f | 1391 | out: |
81819f0f CL |
1392 | return page; |
1393 | } | |
1394 | ||
1395 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1396 | { | |
834f3d11 CL |
1397 | int order = compound_order(page); |
1398 | int pages = 1 << order; | |
81819f0f | 1399 | |
af537b0a | 1400 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1401 | void *p; |
1402 | ||
1403 | slab_pad_check(s, page); | |
224a88be CL |
1404 | for_each_object(p, s, page_address(page), |
1405 | page->objects) | |
f7cb1933 | 1406 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1407 | } |
1408 | ||
b1eeab67 | 1409 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1410 | |
81819f0f CL |
1411 | mod_zone_page_state(page_zone(page), |
1412 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1413 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1414 | -pages); |
81819f0f | 1415 | |
49bd5221 CL |
1416 | __ClearPageSlab(page); |
1417 | reset_page_mapcount(page); | |
1eb5ac64 NP |
1418 | if (current->reclaim_state) |
1419 | current->reclaim_state->reclaimed_slab += pages; | |
834f3d11 | 1420 | __free_pages(page, order); |
81819f0f CL |
1421 | } |
1422 | ||
da9a638c LJ |
1423 | #define need_reserve_slab_rcu \ |
1424 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1425 | ||
81819f0f CL |
1426 | static void rcu_free_slab(struct rcu_head *h) |
1427 | { | |
1428 | struct page *page; | |
1429 | ||
da9a638c LJ |
1430 | if (need_reserve_slab_rcu) |
1431 | page = virt_to_head_page(h); | |
1432 | else | |
1433 | page = container_of((struct list_head *)h, struct page, lru); | |
1434 | ||
81819f0f CL |
1435 | __free_slab(page->slab, page); |
1436 | } | |
1437 | ||
1438 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1439 | { | |
1440 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1441 | struct rcu_head *head; |
1442 | ||
1443 | if (need_reserve_slab_rcu) { | |
1444 | int order = compound_order(page); | |
1445 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1446 | ||
1447 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1448 | head = page_address(page) + offset; | |
1449 | } else { | |
1450 | /* | |
1451 | * RCU free overloads the RCU head over the LRU | |
1452 | */ | |
1453 | head = (void *)&page->lru; | |
1454 | } | |
81819f0f CL |
1455 | |
1456 | call_rcu(head, rcu_free_slab); | |
1457 | } else | |
1458 | __free_slab(s, page); | |
1459 | } | |
1460 | ||
1461 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1462 | { | |
205ab99d | 1463 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1464 | free_slab(s, page); |
1465 | } | |
1466 | ||
1467 | /* | |
5cc6eee8 CL |
1468 | * Management of partially allocated slabs. |
1469 | * | |
1470 | * list_lock must be held. | |
81819f0f | 1471 | */ |
5cc6eee8 | 1472 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1473 | struct page *page, int tail) |
81819f0f | 1474 | { |
e95eed57 | 1475 | n->nr_partial++; |
136333d1 | 1476 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1477 | list_add_tail(&page->lru, &n->partial); |
1478 | else | |
1479 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1480 | } |
1481 | ||
5cc6eee8 CL |
1482 | /* |
1483 | * list_lock must be held. | |
1484 | */ | |
1485 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1486 | struct page *page) |
1487 | { | |
1488 | list_del(&page->lru); | |
1489 | n->nr_partial--; | |
1490 | } | |
1491 | ||
81819f0f | 1492 | /* |
7ced3719 CL |
1493 | * Remove slab from the partial list, freeze it and |
1494 | * return the pointer to the freelist. | |
81819f0f | 1495 | * |
497b66f2 CL |
1496 | * Returns a list of objects or NULL if it fails. |
1497 | * | |
7ced3719 | 1498 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1499 | */ |
497b66f2 | 1500 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1501 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1502 | int mode) |
81819f0f | 1503 | { |
2cfb7455 CL |
1504 | void *freelist; |
1505 | unsigned long counters; | |
1506 | struct page new; | |
1507 | ||
2cfb7455 CL |
1508 | /* |
1509 | * Zap the freelist and set the frozen bit. | |
1510 | * The old freelist is the list of objects for the | |
1511 | * per cpu allocation list. | |
1512 | */ | |
7ced3719 CL |
1513 | freelist = page->freelist; |
1514 | counters = page->counters; | |
1515 | new.counters = counters; | |
1516 | if (mode) | |
1517 | new.inuse = page->objects; | |
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 CL |
1523 | freelist, counters, |
1524 | NULL, new.counters, | |
7ced3719 CL |
1525 | "acquire_slab")) |
1526 | ||
1527 | return NULL; | |
2cfb7455 CL |
1528 | |
1529 | remove_partial(n, page); | |
7ced3719 | 1530 | WARN_ON(!freelist); |
49e22585 | 1531 | return freelist; |
81819f0f CL |
1532 | } |
1533 | ||
49e22585 CL |
1534 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
1535 | ||
81819f0f | 1536 | /* |
672bba3a | 1537 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1538 | */ |
497b66f2 | 1539 | static void *get_partial_node(struct kmem_cache *s, |
acd19fd1 | 1540 | struct kmem_cache_node *n, struct kmem_cache_cpu *c) |
81819f0f | 1541 | { |
49e22585 CL |
1542 | struct page *page, *page2; |
1543 | void *object = NULL; | |
81819f0f CL |
1544 | |
1545 | /* | |
1546 | * Racy check. If we mistakenly see no partial slabs then we | |
1547 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1548 | * partial slab and there is none available then get_partials() |
1549 | * will return NULL. | |
81819f0f CL |
1550 | */ |
1551 | if (!n || !n->nr_partial) | |
1552 | return NULL; | |
1553 | ||
1554 | spin_lock(&n->list_lock); | |
49e22585 | 1555 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
12d79634 | 1556 | void *t = acquire_slab(s, n, page, object == NULL); |
49e22585 CL |
1557 | int available; |
1558 | ||
1559 | if (!t) | |
1560 | break; | |
1561 | ||
12d79634 | 1562 | if (!object) { |
49e22585 | 1563 | c->page = page; |
49e22585 | 1564 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 CL |
1565 | object = t; |
1566 | available = page->objects - page->inuse; | |
1567 | } else { | |
1568 | page->freelist = t; | |
1569 | available = put_cpu_partial(s, page, 0); | |
8028dcea | 1570 | stat(s, CPU_PARTIAL_NODE); |
49e22585 CL |
1571 | } |
1572 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1573 | break; | |
1574 | ||
497b66f2 | 1575 | } |
81819f0f | 1576 | spin_unlock(&n->list_lock); |
497b66f2 | 1577 | return object; |
81819f0f CL |
1578 | } |
1579 | ||
1580 | /* | |
672bba3a | 1581 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1582 | */ |
acd19fd1 CL |
1583 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags, |
1584 | struct kmem_cache_cpu *c) | |
81819f0f CL |
1585 | { |
1586 | #ifdef CONFIG_NUMA | |
1587 | struct zonelist *zonelist; | |
dd1a239f | 1588 | struct zoneref *z; |
54a6eb5c MG |
1589 | struct zone *zone; |
1590 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1591 | void *object; |
cc9a6c87 | 1592 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1593 | |
1594 | /* | |
672bba3a CL |
1595 | * The defrag ratio allows a configuration of the tradeoffs between |
1596 | * inter node defragmentation and node local allocations. A lower | |
1597 | * defrag_ratio increases the tendency to do local allocations | |
1598 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1599 | * |
672bba3a CL |
1600 | * If the defrag_ratio is set to 0 then kmalloc() always |
1601 | * returns node local objects. If the ratio is higher then kmalloc() | |
1602 | * may return off node objects because partial slabs are obtained | |
1603 | * from other nodes and filled up. | |
81819f0f | 1604 | * |
6446faa2 | 1605 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1606 | * defrag_ratio = 1000) then every (well almost) allocation will |
1607 | * first attempt to defrag slab caches on other nodes. This means | |
1608 | * scanning over all nodes to look for partial slabs which may be | |
1609 | * expensive if we do it every time we are trying to find a slab | |
1610 | * with available objects. | |
81819f0f | 1611 | */ |
9824601e CL |
1612 | if (!s->remote_node_defrag_ratio || |
1613 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1614 | return NULL; |
1615 | ||
cc9a6c87 MG |
1616 | do { |
1617 | cpuset_mems_cookie = get_mems_allowed(); | |
1618 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); | |
1619 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { | |
1620 | struct kmem_cache_node *n; | |
1621 | ||
1622 | n = get_node(s, zone_to_nid(zone)); | |
1623 | ||
1624 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1625 | n->nr_partial > s->min_partial) { | |
1626 | object = get_partial_node(s, n, c); | |
1627 | if (object) { | |
1628 | /* | |
1629 | * Return the object even if | |
1630 | * put_mems_allowed indicated that | |
1631 | * the cpuset mems_allowed was | |
1632 | * updated in parallel. It's a | |
1633 | * harmless race between the alloc | |
1634 | * and the cpuset update. | |
1635 | */ | |
1636 | put_mems_allowed(cpuset_mems_cookie); | |
1637 | return object; | |
1638 | } | |
c0ff7453 | 1639 | } |
81819f0f | 1640 | } |
cc9a6c87 | 1641 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1642 | #endif |
1643 | return NULL; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * Get a partial page, lock it and return it. | |
1648 | */ | |
497b66f2 | 1649 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1650 | struct kmem_cache_cpu *c) |
81819f0f | 1651 | { |
497b66f2 | 1652 | void *object; |
2154a336 | 1653 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1654 | |
497b66f2 CL |
1655 | object = get_partial_node(s, get_node(s, searchnode), c); |
1656 | if (object || node != NUMA_NO_NODE) | |
1657 | return object; | |
81819f0f | 1658 | |
acd19fd1 | 1659 | return get_any_partial(s, flags, c); |
81819f0f CL |
1660 | } |
1661 | ||
8a5ec0ba CL |
1662 | #ifdef CONFIG_PREEMPT |
1663 | /* | |
1664 | * Calculate the next globally unique transaction for disambiguiation | |
1665 | * during cmpxchg. The transactions start with the cpu number and are then | |
1666 | * incremented by CONFIG_NR_CPUS. | |
1667 | */ | |
1668 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1669 | #else | |
1670 | /* | |
1671 | * No preemption supported therefore also no need to check for | |
1672 | * different cpus. | |
1673 | */ | |
1674 | #define TID_STEP 1 | |
1675 | #endif | |
1676 | ||
1677 | static inline unsigned long next_tid(unsigned long tid) | |
1678 | { | |
1679 | return tid + TID_STEP; | |
1680 | } | |
1681 | ||
1682 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1683 | { | |
1684 | return tid % TID_STEP; | |
1685 | } | |
1686 | ||
1687 | static inline unsigned long tid_to_event(unsigned long tid) | |
1688 | { | |
1689 | return tid / TID_STEP; | |
1690 | } | |
1691 | ||
1692 | static inline unsigned int init_tid(int cpu) | |
1693 | { | |
1694 | return cpu; | |
1695 | } | |
1696 | ||
1697 | static inline void note_cmpxchg_failure(const char *n, | |
1698 | const struct kmem_cache *s, unsigned long tid) | |
1699 | { | |
1700 | #ifdef SLUB_DEBUG_CMPXCHG | |
1701 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1702 | ||
1703 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1704 | ||
1705 | #ifdef CONFIG_PREEMPT | |
1706 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1707 | printk("due to cpu change %d -> %d\n", | |
1708 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1709 | else | |
1710 | #endif | |
1711 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1712 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1713 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1714 | else | |
1715 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1716 | actual_tid, tid, next_tid(tid)); | |
1717 | #endif | |
4fdccdfb | 1718 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1719 | } |
1720 | ||
8a5ec0ba CL |
1721 | void init_kmem_cache_cpus(struct kmem_cache *s) |
1722 | { | |
8a5ec0ba CL |
1723 | int cpu; |
1724 | ||
1725 | for_each_possible_cpu(cpu) | |
1726 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1727 | } |
2cfb7455 | 1728 | |
81819f0f CL |
1729 | /* |
1730 | * Remove the cpu slab | |
1731 | */ | |
c17dda40 | 1732 | static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist) |
81819f0f | 1733 | { |
2cfb7455 | 1734 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1735 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1736 | int lock = 0; | |
1737 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1738 | void *nextfree; |
136333d1 | 1739 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1740 | struct page new; |
1741 | struct page old; | |
1742 | ||
1743 | if (page->freelist) { | |
84e554e6 | 1744 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1745 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1746 | } |
1747 | ||
894b8788 | 1748 | /* |
2cfb7455 CL |
1749 | * Stage one: Free all available per cpu objects back |
1750 | * to the page freelist while it is still frozen. Leave the | |
1751 | * last one. | |
1752 | * | |
1753 | * There is no need to take the list->lock because the page | |
1754 | * is still frozen. | |
1755 | */ | |
1756 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1757 | void *prior; | |
1758 | unsigned long counters; | |
1759 | ||
1760 | do { | |
1761 | prior = page->freelist; | |
1762 | counters = page->counters; | |
1763 | set_freepointer(s, freelist, prior); | |
1764 | new.counters = counters; | |
1765 | new.inuse--; | |
1766 | VM_BUG_ON(!new.frozen); | |
1767 | ||
1d07171c | 1768 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1769 | prior, counters, |
1770 | freelist, new.counters, | |
1771 | "drain percpu freelist")); | |
1772 | ||
1773 | freelist = nextfree; | |
1774 | } | |
1775 | ||
894b8788 | 1776 | /* |
2cfb7455 CL |
1777 | * Stage two: Ensure that the page is unfrozen while the |
1778 | * list presence reflects the actual number of objects | |
1779 | * during unfreeze. | |
1780 | * | |
1781 | * We setup the list membership and then perform a cmpxchg | |
1782 | * with the count. If there is a mismatch then the page | |
1783 | * is not unfrozen but the page is on the wrong list. | |
1784 | * | |
1785 | * Then we restart the process which may have to remove | |
1786 | * the page from the list that we just put it on again | |
1787 | * because the number of objects in the slab may have | |
1788 | * changed. | |
894b8788 | 1789 | */ |
2cfb7455 | 1790 | redo: |
894b8788 | 1791 | |
2cfb7455 CL |
1792 | old.freelist = page->freelist; |
1793 | old.counters = page->counters; | |
1794 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1795 | |
2cfb7455 CL |
1796 | /* Determine target state of the slab */ |
1797 | new.counters = old.counters; | |
1798 | if (freelist) { | |
1799 | new.inuse--; | |
1800 | set_freepointer(s, freelist, old.freelist); | |
1801 | new.freelist = freelist; | |
1802 | } else | |
1803 | new.freelist = old.freelist; | |
1804 | ||
1805 | new.frozen = 0; | |
1806 | ||
81107188 | 1807 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1808 | m = M_FREE; |
1809 | else if (new.freelist) { | |
1810 | m = M_PARTIAL; | |
1811 | if (!lock) { | |
1812 | lock = 1; | |
1813 | /* | |
1814 | * Taking the spinlock removes the possiblity | |
1815 | * that acquire_slab() will see a slab page that | |
1816 | * is frozen | |
1817 | */ | |
1818 | spin_lock(&n->list_lock); | |
1819 | } | |
1820 | } else { | |
1821 | m = M_FULL; | |
1822 | if (kmem_cache_debug(s) && !lock) { | |
1823 | lock = 1; | |
1824 | /* | |
1825 | * This also ensures that the scanning of full | |
1826 | * slabs from diagnostic functions will not see | |
1827 | * any frozen slabs. | |
1828 | */ | |
1829 | spin_lock(&n->list_lock); | |
1830 | } | |
1831 | } | |
1832 | ||
1833 | if (l != m) { | |
1834 | ||
1835 | if (l == M_PARTIAL) | |
1836 | ||
1837 | remove_partial(n, page); | |
1838 | ||
1839 | else if (l == M_FULL) | |
894b8788 | 1840 | |
2cfb7455 CL |
1841 | remove_full(s, page); |
1842 | ||
1843 | if (m == M_PARTIAL) { | |
1844 | ||
1845 | add_partial(n, page, tail); | |
136333d1 | 1846 | stat(s, tail); |
2cfb7455 CL |
1847 | |
1848 | } else if (m == M_FULL) { | |
894b8788 | 1849 | |
2cfb7455 CL |
1850 | stat(s, DEACTIVATE_FULL); |
1851 | add_full(s, n, page); | |
1852 | ||
1853 | } | |
1854 | } | |
1855 | ||
1856 | l = m; | |
1d07171c | 1857 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1858 | old.freelist, old.counters, |
1859 | new.freelist, new.counters, | |
1860 | "unfreezing slab")) | |
1861 | goto redo; | |
1862 | ||
2cfb7455 CL |
1863 | if (lock) |
1864 | spin_unlock(&n->list_lock); | |
1865 | ||
1866 | if (m == M_FREE) { | |
1867 | stat(s, DEACTIVATE_EMPTY); | |
1868 | discard_slab(s, page); | |
1869 | stat(s, FREE_SLAB); | |
894b8788 | 1870 | } |
81819f0f CL |
1871 | } |
1872 | ||
49e22585 CL |
1873 | /* Unfreeze all the cpu partial slabs */ |
1874 | static void unfreeze_partials(struct kmem_cache *s) | |
1875 | { | |
1876 | struct kmem_cache_node *n = NULL; | |
1877 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); | |
9ada1934 | 1878 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1879 | |
1880 | while ((page = c->partial)) { | |
1881 | enum slab_modes { M_PARTIAL, M_FREE }; | |
1882 | enum slab_modes l, m; | |
1883 | struct page new; | |
1884 | struct page old; | |
1885 | ||
1886 | c->partial = page->next; | |
1887 | l = M_FREE; | |
1888 | ||
1889 | do { | |
1890 | ||
1891 | old.freelist = page->freelist; | |
1892 | old.counters = page->counters; | |
1893 | VM_BUG_ON(!old.frozen); | |
1894 | ||
1895 | new.counters = old.counters; | |
1896 | new.freelist = old.freelist; | |
1897 | ||
1898 | new.frozen = 0; | |
1899 | ||
dcc3be6a | 1900 | if (!new.inuse && (!n || n->nr_partial > s->min_partial)) |
49e22585 CL |
1901 | m = M_FREE; |
1902 | else { | |
1903 | struct kmem_cache_node *n2 = get_node(s, | |
1904 | page_to_nid(page)); | |
1905 | ||
1906 | m = M_PARTIAL; | |
1907 | if (n != n2) { | |
1908 | if (n) | |
1909 | spin_unlock(&n->list_lock); | |
1910 | ||
1911 | n = n2; | |
1912 | spin_lock(&n->list_lock); | |
1913 | } | |
1914 | } | |
1915 | ||
1916 | if (l != m) { | |
4c493a5a | 1917 | if (l == M_PARTIAL) { |
49e22585 | 1918 | remove_partial(n, page); |
4c493a5a SL |
1919 | stat(s, FREE_REMOVE_PARTIAL); |
1920 | } else { | |
f64ae042 SL |
1921 | add_partial(n, page, |
1922 | DEACTIVATE_TO_TAIL); | |
4c493a5a SL |
1923 | stat(s, FREE_ADD_PARTIAL); |
1924 | } | |
49e22585 CL |
1925 | |
1926 | l = m; | |
1927 | } | |
1928 | ||
1929 | } while (!cmpxchg_double_slab(s, page, | |
1930 | old.freelist, old.counters, | |
1931 | new.freelist, new.counters, | |
1932 | "unfreezing slab")); | |
1933 | ||
1934 | if (m == M_FREE) { | |
9ada1934 SL |
1935 | page->next = discard_page; |
1936 | discard_page = page; | |
49e22585 CL |
1937 | } |
1938 | } | |
1939 | ||
1940 | if (n) | |
1941 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1942 | |
1943 | while (discard_page) { | |
1944 | page = discard_page; | |
1945 | discard_page = discard_page->next; | |
1946 | ||
1947 | stat(s, DEACTIVATE_EMPTY); | |
1948 | discard_slab(s, page); | |
1949 | stat(s, FREE_SLAB); | |
1950 | } | |
49e22585 CL |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1955 | * slot if available. This is done without interrupts disabled and without | |
1956 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1957 | * onto a random cpus partial slot. | |
1958 | * | |
1959 | * If we did not find a slot then simply move all the partials to the | |
1960 | * per node partial list. | |
1961 | */ | |
1962 | int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) | |
1963 | { | |
1964 | struct page *oldpage; | |
1965 | int pages; | |
1966 | int pobjects; | |
1967 | ||
1968 | do { | |
1969 | pages = 0; | |
1970 | pobjects = 0; | |
1971 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
1972 | ||
1973 | if (oldpage) { | |
1974 | pobjects = oldpage->pobjects; | |
1975 | pages = oldpage->pages; | |
1976 | if (drain && pobjects > s->cpu_partial) { | |
1977 | unsigned long flags; | |
1978 | /* | |
1979 | * partial array is full. Move the existing | |
1980 | * set to the per node partial list. | |
1981 | */ | |
1982 | local_irq_save(flags); | |
1983 | unfreeze_partials(s); | |
1984 | local_irq_restore(flags); | |
1985 | pobjects = 0; | |
1986 | pages = 0; | |
8028dcea | 1987 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
1988 | } |
1989 | } | |
1990 | ||
1991 | pages++; | |
1992 | pobjects += page->objects - page->inuse; | |
1993 | ||
1994 | page->pages = pages; | |
1995 | page->pobjects = pobjects; | |
1996 | page->next = oldpage; | |
1997 | ||
933393f5 | 1998 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
1999 | return pobjects; |
2000 | } | |
2001 | ||
dfb4f096 | 2002 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2003 | { |
84e554e6 | 2004 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2005 | deactivate_slab(s, c->page, c->freelist); |
2006 | ||
2007 | c->tid = next_tid(c->tid); | |
2008 | c->page = NULL; | |
2009 | c->freelist = NULL; | |
81819f0f CL |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * Flush cpu slab. | |
6446faa2 | 2014 | * |
81819f0f CL |
2015 | * Called from IPI handler with interrupts disabled. |
2016 | */ | |
0c710013 | 2017 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2018 | { |
9dfc6e68 | 2019 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2020 | |
49e22585 CL |
2021 | if (likely(c)) { |
2022 | if (c->page) | |
2023 | flush_slab(s, c); | |
2024 | ||
2025 | unfreeze_partials(s); | |
2026 | } | |
81819f0f CL |
2027 | } |
2028 | ||
2029 | static void flush_cpu_slab(void *d) | |
2030 | { | |
2031 | struct kmem_cache *s = d; | |
81819f0f | 2032 | |
dfb4f096 | 2033 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2034 | } |
2035 | ||
a8364d55 GBY |
2036 | static bool has_cpu_slab(int cpu, void *info) |
2037 | { | |
2038 | struct kmem_cache *s = info; | |
2039 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2040 | ||
02e1a9cd | 2041 | return c->page || c->partial; |
a8364d55 GBY |
2042 | } |
2043 | ||
81819f0f CL |
2044 | static void flush_all(struct kmem_cache *s) |
2045 | { | |
a8364d55 | 2046 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2047 | } |
2048 | ||
dfb4f096 CL |
2049 | /* |
2050 | * Check if the objects in a per cpu structure fit numa | |
2051 | * locality expectations. | |
2052 | */ | |
2053 | static inline int node_match(struct kmem_cache_cpu *c, int node) | |
2054 | { | |
2055 | #ifdef CONFIG_NUMA | |
ec3ab083 | 2056 | if (node != NUMA_NO_NODE && page_to_nid(c->page) != node) |
dfb4f096 CL |
2057 | return 0; |
2058 | #endif | |
2059 | return 1; | |
2060 | } | |
2061 | ||
781b2ba6 PE |
2062 | static int count_free(struct page *page) |
2063 | { | |
2064 | return page->objects - page->inuse; | |
2065 | } | |
2066 | ||
2067 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2068 | int (*get_count)(struct page *)) | |
2069 | { | |
2070 | unsigned long flags; | |
2071 | unsigned long x = 0; | |
2072 | struct page *page; | |
2073 | ||
2074 | spin_lock_irqsave(&n->list_lock, flags); | |
2075 | list_for_each_entry(page, &n->partial, lru) | |
2076 | x += get_count(page); | |
2077 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2078 | return x; | |
2079 | } | |
2080 | ||
26c02cf0 AB |
2081 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2082 | { | |
2083 | #ifdef CONFIG_SLUB_DEBUG | |
2084 | return atomic_long_read(&n->total_objects); | |
2085 | #else | |
2086 | return 0; | |
2087 | #endif | |
2088 | } | |
2089 | ||
781b2ba6 PE |
2090 | static noinline void |
2091 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2092 | { | |
2093 | int node; | |
2094 | ||
2095 | printk(KERN_WARNING | |
2096 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2097 | nid, gfpflags); | |
2098 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
2099 | "default order: %d, min order: %d\n", s->name, s->objsize, | |
2100 | s->size, oo_order(s->oo), oo_order(s->min)); | |
2101 | ||
fa5ec8a1 DR |
2102 | if (oo_order(s->min) > get_order(s->objsize)) |
2103 | printk(KERN_WARNING " %s debugging increased min order, use " | |
2104 | "slub_debug=O to disable.\n", s->name); | |
2105 | ||
781b2ba6 PE |
2106 | for_each_online_node(node) { |
2107 | struct kmem_cache_node *n = get_node(s, node); | |
2108 | unsigned long nr_slabs; | |
2109 | unsigned long nr_objs; | |
2110 | unsigned long nr_free; | |
2111 | ||
2112 | if (!n) | |
2113 | continue; | |
2114 | ||
26c02cf0 AB |
2115 | nr_free = count_partial(n, count_free); |
2116 | nr_slabs = node_nr_slabs(n); | |
2117 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2118 | |
2119 | printk(KERN_WARNING | |
2120 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2121 | node, nr_slabs, nr_objs, nr_free); | |
2122 | } | |
2123 | } | |
2124 | ||
497b66f2 CL |
2125 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2126 | int node, struct kmem_cache_cpu **pc) | |
2127 | { | |
6faa6833 | 2128 | void *freelist; |
188fd063 CL |
2129 | struct kmem_cache_cpu *c = *pc; |
2130 | struct page *page; | |
2131 | ||
2132 | freelist = get_partial(s, flags, node, c); | |
497b66f2 | 2133 | |
188fd063 CL |
2134 | if (freelist) |
2135 | return freelist; | |
2136 | ||
2137 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2138 | if (page) { |
2139 | c = __this_cpu_ptr(s->cpu_slab); | |
2140 | if (c->page) | |
2141 | flush_slab(s, c); | |
2142 | ||
2143 | /* | |
2144 | * No other reference to the page yet so we can | |
2145 | * muck around with it freely without cmpxchg | |
2146 | */ | |
6faa6833 | 2147 | freelist = page->freelist; |
497b66f2 CL |
2148 | page->freelist = NULL; |
2149 | ||
2150 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2151 | c->page = page; |
2152 | *pc = c; | |
2153 | } else | |
6faa6833 | 2154 | freelist = NULL; |
497b66f2 | 2155 | |
6faa6833 | 2156 | return freelist; |
497b66f2 CL |
2157 | } |
2158 | ||
213eeb9f CL |
2159 | /* |
2160 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2161 | * or deactivate the page. | |
2162 | * | |
2163 | * The page is still frozen if the return value is not NULL. | |
2164 | * | |
2165 | * If this function returns NULL then the page has been unfrozen. | |
2166 | */ | |
2167 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2168 | { | |
2169 | struct page new; | |
2170 | unsigned long counters; | |
2171 | void *freelist; | |
2172 | ||
2173 | do { | |
2174 | freelist = page->freelist; | |
2175 | counters = page->counters; | |
6faa6833 | 2176 | |
213eeb9f CL |
2177 | new.counters = counters; |
2178 | VM_BUG_ON(!new.frozen); | |
2179 | ||
2180 | new.inuse = page->objects; | |
2181 | new.frozen = freelist != NULL; | |
2182 | ||
2183 | } while (!cmpxchg_double_slab(s, page, | |
2184 | freelist, counters, | |
2185 | NULL, new.counters, | |
2186 | "get_freelist")); | |
2187 | ||
2188 | return freelist; | |
2189 | } | |
2190 | ||
81819f0f | 2191 | /* |
894b8788 CL |
2192 | * Slow path. The lockless freelist is empty or we need to perform |
2193 | * debugging duties. | |
2194 | * | |
894b8788 CL |
2195 | * Processing is still very fast if new objects have been freed to the |
2196 | * regular freelist. In that case we simply take over the regular freelist | |
2197 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2198 | * |
894b8788 CL |
2199 | * If that is not working then we fall back to the partial lists. We take the |
2200 | * first element of the freelist as the object to allocate now and move the | |
2201 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2202 | * |
894b8788 | 2203 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2204 | * we need to allocate a new slab. This is the slowest path since it involves |
2205 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2206 | */ |
ce71e27c EGM |
2207 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2208 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2209 | { |
6faa6833 | 2210 | void *freelist; |
8a5ec0ba CL |
2211 | unsigned long flags; |
2212 | ||
2213 | local_irq_save(flags); | |
2214 | #ifdef CONFIG_PREEMPT | |
2215 | /* | |
2216 | * We may have been preempted and rescheduled on a different | |
2217 | * cpu before disabling interrupts. Need to reload cpu area | |
2218 | * pointer. | |
2219 | */ | |
2220 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2221 | #endif |
81819f0f | 2222 | |
497b66f2 | 2223 | if (!c->page) |
81819f0f | 2224 | goto new_slab; |
49e22585 | 2225 | redo: |
6faa6833 | 2226 | |
fc59c053 | 2227 | if (unlikely(!node_match(c, node))) { |
e36a2652 | 2228 | stat(s, ALLOC_NODE_MISMATCH); |
c17dda40 CL |
2229 | deactivate_slab(s, c->page, c->freelist); |
2230 | c->page = NULL; | |
2231 | c->freelist = NULL; | |
fc59c053 CL |
2232 | goto new_slab; |
2233 | } | |
6446faa2 | 2234 | |
73736e03 | 2235 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2236 | freelist = c->freelist; |
2237 | if (freelist) | |
73736e03 | 2238 | goto load_freelist; |
03e404af | 2239 | |
2cfb7455 | 2240 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2241 | |
6faa6833 | 2242 | freelist = get_freelist(s, c->page); |
6446faa2 | 2243 | |
6faa6833 | 2244 | if (!freelist) { |
03e404af CL |
2245 | c->page = NULL; |
2246 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2247 | goto new_slab; |
03e404af | 2248 | } |
6446faa2 | 2249 | |
84e554e6 | 2250 | stat(s, ALLOC_REFILL); |
6446faa2 | 2251 | |
894b8788 | 2252 | load_freelist: |
507effea CL |
2253 | /* |
2254 | * freelist is pointing to the list of objects to be used. | |
2255 | * page is pointing to the page from which the objects are obtained. | |
2256 | * That page must be frozen for per cpu allocations to work. | |
2257 | */ | |
2258 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2259 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2260 | c->tid = next_tid(c->tid); |
2261 | local_irq_restore(flags); | |
6faa6833 | 2262 | return freelist; |
81819f0f | 2263 | |
81819f0f | 2264 | new_slab: |
2cfb7455 | 2265 | |
49e22585 CL |
2266 | if (c->partial) { |
2267 | c->page = c->partial; | |
2268 | c->partial = c->page->next; | |
49e22585 CL |
2269 | stat(s, CPU_PARTIAL_ALLOC); |
2270 | c->freelist = NULL; | |
2271 | goto redo; | |
81819f0f CL |
2272 | } |
2273 | ||
188fd063 | 2274 | freelist = new_slab_objects(s, gfpflags, node, &c); |
2cfb7455 | 2275 | |
f4697436 CL |
2276 | if (unlikely(!freelist)) { |
2277 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2278 | slab_out_of_memory(s, gfpflags, node); | |
9e577e8b | 2279 | |
f4697436 CL |
2280 | local_irq_restore(flags); |
2281 | return NULL; | |
81819f0f | 2282 | } |
2cfb7455 | 2283 | |
497b66f2 | 2284 | if (likely(!kmem_cache_debug(s))) |
4b6f0750 | 2285 | goto load_freelist; |
2cfb7455 | 2286 | |
497b66f2 | 2287 | /* Only entered in the debug case */ |
6faa6833 | 2288 | if (!alloc_debug_processing(s, c->page, freelist, addr)) |
497b66f2 | 2289 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2290 | |
c17dda40 CL |
2291 | deactivate_slab(s, c->page, get_freepointer(s, freelist)); |
2292 | c->page = NULL; | |
2293 | c->freelist = NULL; | |
a71ae47a | 2294 | local_irq_restore(flags); |
6faa6833 | 2295 | return freelist; |
894b8788 CL |
2296 | } |
2297 | ||
2298 | /* | |
2299 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2300 | * have the fastpath folded into their functions. So no function call | |
2301 | * overhead for requests that can be satisfied on the fastpath. | |
2302 | * | |
2303 | * The fastpath works by first checking if the lockless freelist can be used. | |
2304 | * If not then __slab_alloc is called for slow processing. | |
2305 | * | |
2306 | * Otherwise we can simply pick the next object from the lockless free list. | |
2307 | */ | |
06428780 | 2308 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
ce71e27c | 2309 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2310 | { |
894b8788 | 2311 | void **object; |
dfb4f096 | 2312 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2313 | unsigned long tid; |
1f84260c | 2314 | |
c016b0bd | 2315 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2316 | return NULL; |
1f84260c | 2317 | |
8a5ec0ba | 2318 | redo: |
8a5ec0ba CL |
2319 | |
2320 | /* | |
2321 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2322 | * enabled. We may switch back and forth between cpus while | |
2323 | * reading from one cpu area. That does not matter as long | |
2324 | * as we end up on the original cpu again when doing the cmpxchg. | |
2325 | */ | |
9dfc6e68 | 2326 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2327 | |
8a5ec0ba CL |
2328 | /* |
2329 | * The transaction ids are globally unique per cpu and per operation on | |
2330 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2331 | * occurs on the right processor and that there was no operation on the | |
2332 | * linked list in between. | |
2333 | */ | |
2334 | tid = c->tid; | |
2335 | barrier(); | |
8a5ec0ba | 2336 | |
9dfc6e68 | 2337 | object = c->freelist; |
9dfc6e68 | 2338 | if (unlikely(!object || !node_match(c, node))) |
894b8788 | 2339 | |
dfb4f096 | 2340 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2341 | |
2342 | else { | |
0ad9500e ED |
2343 | void *next_object = get_freepointer_safe(s, object); |
2344 | ||
8a5ec0ba | 2345 | /* |
25985edc | 2346 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2347 | * operation and if we are on the right processor. |
2348 | * | |
2349 | * The cmpxchg does the following atomically (without lock semantics!) | |
2350 | * 1. Relocate first pointer to the current per cpu area. | |
2351 | * 2. Verify that tid and freelist have not been changed | |
2352 | * 3. If they were not changed replace tid and freelist | |
2353 | * | |
2354 | * Since this is without lock semantics the protection is only against | |
2355 | * code executing on this cpu *not* from access by other cpus. | |
2356 | */ | |
933393f5 | 2357 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2358 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2359 | object, tid, | |
0ad9500e | 2360 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2361 | |
2362 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2363 | goto redo; | |
2364 | } | |
0ad9500e | 2365 | prefetch_freepointer(s, next_object); |
84e554e6 | 2366 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2367 | } |
8a5ec0ba | 2368 | |
74e2134f | 2369 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
ff12059e | 2370 | memset(object, 0, s->objsize); |
d07dbea4 | 2371 | |
c016b0bd | 2372 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2373 | |
894b8788 | 2374 | return object; |
81819f0f CL |
2375 | } |
2376 | ||
2377 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
2378 | { | |
2154a336 | 2379 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 2380 | |
ca2b84cb | 2381 | trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags); |
5b882be4 EGM |
2382 | |
2383 | return ret; | |
81819f0f CL |
2384 | } |
2385 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2386 | ||
0f24f128 | 2387 | #ifdef CONFIG_TRACING |
4a92379b RK |
2388 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2389 | { | |
2390 | void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_); | |
2391 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); | |
2392 | return ret; | |
2393 | } | |
2394 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2395 | ||
2396 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2397 | { |
4a92379b RK |
2398 | void *ret = kmalloc_order(size, flags, order); |
2399 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2400 | return ret; | |
5b882be4 | 2401 | } |
4a92379b | 2402 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2403 | #endif |
2404 | ||
81819f0f CL |
2405 | #ifdef CONFIG_NUMA |
2406 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2407 | { | |
5b882be4 EGM |
2408 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2409 | ||
ca2b84cb EGM |
2410 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
2411 | s->objsize, s->size, gfpflags, node); | |
5b882be4 EGM |
2412 | |
2413 | return ret; | |
81819f0f CL |
2414 | } |
2415 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2416 | |
0f24f128 | 2417 | #ifdef CONFIG_TRACING |
4a92379b | 2418 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2419 | gfp_t gfpflags, |
4a92379b | 2420 | int node, size_t size) |
5b882be4 | 2421 | { |
4a92379b RK |
2422 | void *ret = slab_alloc(s, gfpflags, node, _RET_IP_); |
2423 | ||
2424 | trace_kmalloc_node(_RET_IP_, ret, | |
2425 | size, s->size, gfpflags, node); | |
2426 | return ret; | |
5b882be4 | 2427 | } |
4a92379b | 2428 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2429 | #endif |
5d1f57e4 | 2430 | #endif |
5b882be4 | 2431 | |
81819f0f | 2432 | /* |
894b8788 CL |
2433 | * Slow patch handling. This may still be called frequently since objects |
2434 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2435 | * |
894b8788 CL |
2436 | * So we still attempt to reduce cache line usage. Just take the slab |
2437 | * lock and free the item. If there is no additional partial page | |
2438 | * handling required then we can return immediately. | |
81819f0f | 2439 | */ |
894b8788 | 2440 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2441 | void *x, unsigned long addr) |
81819f0f CL |
2442 | { |
2443 | void *prior; | |
2444 | void **object = (void *)x; | |
2cfb7455 CL |
2445 | int was_frozen; |
2446 | int inuse; | |
2447 | struct page new; | |
2448 | unsigned long counters; | |
2449 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2450 | unsigned long uninitialized_var(flags); |
81819f0f | 2451 | |
8a5ec0ba | 2452 | stat(s, FREE_SLOWPATH); |
81819f0f | 2453 | |
8dc16c6c | 2454 | if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr)) |
80f08c19 | 2455 | return; |
6446faa2 | 2456 | |
2cfb7455 CL |
2457 | do { |
2458 | prior = page->freelist; | |
2459 | counters = page->counters; | |
2460 | set_freepointer(s, object, prior); | |
2461 | new.counters = counters; | |
2462 | was_frozen = new.frozen; | |
2463 | new.inuse--; | |
2464 | if ((!new.inuse || !prior) && !was_frozen && !n) { | |
49e22585 CL |
2465 | |
2466 | if (!kmem_cache_debug(s) && !prior) | |
2467 | ||
2468 | /* | |
2469 | * Slab was on no list before and will be partially empty | |
2470 | * We can defer the list move and instead freeze it. | |
2471 | */ | |
2472 | new.frozen = 1; | |
2473 | ||
2474 | else { /* Needs to be taken off a list */ | |
2475 | ||
2476 | n = get_node(s, page_to_nid(page)); | |
2477 | /* | |
2478 | * Speculatively acquire the list_lock. | |
2479 | * If the cmpxchg does not succeed then we may | |
2480 | * drop the list_lock without any processing. | |
2481 | * | |
2482 | * Otherwise the list_lock will synchronize with | |
2483 | * other processors updating the list of slabs. | |
2484 | */ | |
2485 | spin_lock_irqsave(&n->list_lock, flags); | |
2486 | ||
2487 | } | |
2cfb7455 CL |
2488 | } |
2489 | inuse = new.inuse; | |
81819f0f | 2490 | |
2cfb7455 CL |
2491 | } while (!cmpxchg_double_slab(s, page, |
2492 | prior, counters, | |
2493 | object, new.counters, | |
2494 | "__slab_free")); | |
81819f0f | 2495 | |
2cfb7455 | 2496 | if (likely(!n)) { |
49e22585 CL |
2497 | |
2498 | /* | |
2499 | * If we just froze the page then put it onto the | |
2500 | * per cpu partial list. | |
2501 | */ | |
8028dcea | 2502 | if (new.frozen && !was_frozen) { |
49e22585 | 2503 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2504 | stat(s, CPU_PARTIAL_FREE); |
2505 | } | |
49e22585 | 2506 | /* |
2cfb7455 CL |
2507 | * The list lock was not taken therefore no list |
2508 | * activity can be necessary. | |
2509 | */ | |
2510 | if (was_frozen) | |
2511 | stat(s, FREE_FROZEN); | |
80f08c19 | 2512 | return; |
2cfb7455 | 2513 | } |
81819f0f CL |
2514 | |
2515 | /* | |
2cfb7455 CL |
2516 | * was_frozen may have been set after we acquired the list_lock in |
2517 | * an earlier loop. So we need to check it here again. | |
81819f0f | 2518 | */ |
2cfb7455 CL |
2519 | if (was_frozen) |
2520 | stat(s, FREE_FROZEN); | |
2521 | else { | |
2522 | if (unlikely(!inuse && n->nr_partial > s->min_partial)) | |
2523 | goto slab_empty; | |
81819f0f | 2524 | |
2cfb7455 CL |
2525 | /* |
2526 | * Objects left in the slab. If it was not on the partial list before | |
2527 | * then add it. | |
2528 | */ | |
2529 | if (unlikely(!prior)) { | |
2530 | remove_full(s, page); | |
136333d1 | 2531 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2cfb7455 CL |
2532 | stat(s, FREE_ADD_PARTIAL); |
2533 | } | |
8ff12cfc | 2534 | } |
80f08c19 | 2535 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2536 | return; |
2537 | ||
2538 | slab_empty: | |
a973e9dd | 2539 | if (prior) { |
81819f0f | 2540 | /* |
6fbabb20 | 2541 | * Slab on the partial list. |
81819f0f | 2542 | */ |
5cc6eee8 | 2543 | remove_partial(n, page); |
84e554e6 | 2544 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2545 | } else |
2546 | /* Slab must be on the full list */ | |
2547 | remove_full(s, page); | |
2cfb7455 | 2548 | |
80f08c19 | 2549 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2550 | stat(s, FREE_SLAB); |
81819f0f | 2551 | discard_slab(s, page); |
81819f0f CL |
2552 | } |
2553 | ||
894b8788 CL |
2554 | /* |
2555 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2556 | * can perform fastpath freeing without additional function calls. | |
2557 | * | |
2558 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2559 | * of this processor. This typically the case if we have just allocated | |
2560 | * the item before. | |
2561 | * | |
2562 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2563 | * with all sorts of special processing. | |
2564 | */ | |
06428780 | 2565 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2566 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2567 | { |
2568 | void **object = (void *)x; | |
dfb4f096 | 2569 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2570 | unsigned long tid; |
1f84260c | 2571 | |
c016b0bd CL |
2572 | slab_free_hook(s, x); |
2573 | ||
8a5ec0ba CL |
2574 | redo: |
2575 | /* | |
2576 | * Determine the currently cpus per cpu slab. | |
2577 | * The cpu may change afterward. However that does not matter since | |
2578 | * data is retrieved via this pointer. If we are on the same cpu | |
2579 | * during the cmpxchg then the free will succedd. | |
2580 | */ | |
9dfc6e68 | 2581 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2582 | |
8a5ec0ba CL |
2583 | tid = c->tid; |
2584 | barrier(); | |
c016b0bd | 2585 | |
442b06bc | 2586 | if (likely(page == c->page)) { |
ff12059e | 2587 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2588 | |
933393f5 | 2589 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2590 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2591 | c->freelist, tid, | |
2592 | object, next_tid(tid)))) { | |
2593 | ||
2594 | note_cmpxchg_failure("slab_free", s, tid); | |
2595 | goto redo; | |
2596 | } | |
84e554e6 | 2597 | stat(s, FREE_FASTPATH); |
894b8788 | 2598 | } else |
ff12059e | 2599 | __slab_free(s, page, x, addr); |
894b8788 | 2600 | |
894b8788 CL |
2601 | } |
2602 | ||
81819f0f CL |
2603 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2604 | { | |
77c5e2d0 | 2605 | struct page *page; |
81819f0f | 2606 | |
b49af68f | 2607 | page = virt_to_head_page(x); |
81819f0f | 2608 | |
ce71e27c | 2609 | slab_free(s, page, x, _RET_IP_); |
5b882be4 | 2610 | |
ca2b84cb | 2611 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2612 | } |
2613 | EXPORT_SYMBOL(kmem_cache_free); | |
2614 | ||
81819f0f | 2615 | /* |
672bba3a CL |
2616 | * Object placement in a slab is made very easy because we always start at |
2617 | * offset 0. If we tune the size of the object to the alignment then we can | |
2618 | * get the required alignment by putting one properly sized object after | |
2619 | * another. | |
81819f0f CL |
2620 | * |
2621 | * Notice that the allocation order determines the sizes of the per cpu | |
2622 | * caches. Each processor has always one slab available for allocations. | |
2623 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2624 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2625 | * locking overhead. |
81819f0f CL |
2626 | */ |
2627 | ||
2628 | /* | |
2629 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2630 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2631 | * and increases the number of allocations possible without having to | |
2632 | * take the list_lock. | |
2633 | */ | |
2634 | static int slub_min_order; | |
114e9e89 | 2635 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2636 | static int slub_min_objects; |
81819f0f CL |
2637 | |
2638 | /* | |
2639 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2640 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2641 | */ |
2642 | static int slub_nomerge; | |
2643 | ||
81819f0f CL |
2644 | /* |
2645 | * Calculate the order of allocation given an slab object size. | |
2646 | * | |
672bba3a CL |
2647 | * The order of allocation has significant impact on performance and other |
2648 | * system components. Generally order 0 allocations should be preferred since | |
2649 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2650 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2651 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2652 | * would be wasted. |
2653 | * | |
2654 | * In order to reach satisfactory performance we must ensure that a minimum | |
2655 | * number of objects is in one slab. Otherwise we may generate too much | |
2656 | * activity on the partial lists which requires taking the list_lock. This is | |
2657 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2658 | * |
672bba3a CL |
2659 | * slub_max_order specifies the order where we begin to stop considering the |
2660 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2661 | * we try to keep the page order as low as possible. So we accept more waste | |
2662 | * of space in favor of a small page order. | |
81819f0f | 2663 | * |
672bba3a CL |
2664 | * Higher order allocations also allow the placement of more objects in a |
2665 | * slab and thereby reduce object handling overhead. If the user has | |
2666 | * requested a higher mininum order then we start with that one instead of | |
2667 | * the smallest order which will fit the object. | |
81819f0f | 2668 | */ |
5e6d444e | 2669 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2670 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2671 | { |
2672 | int order; | |
2673 | int rem; | |
6300ea75 | 2674 | int min_order = slub_min_order; |
81819f0f | 2675 | |
ab9a0f19 | 2676 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2677 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2678 | |
6300ea75 | 2679 | for (order = max(min_order, |
5e6d444e CL |
2680 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2681 | order <= max_order; order++) { | |
81819f0f | 2682 | |
5e6d444e | 2683 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2684 | |
ab9a0f19 | 2685 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2686 | continue; |
2687 | ||
ab9a0f19 | 2688 | rem = (slab_size - reserved) % size; |
81819f0f | 2689 | |
5e6d444e | 2690 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2691 | break; |
2692 | ||
2693 | } | |
672bba3a | 2694 | |
81819f0f CL |
2695 | return order; |
2696 | } | |
2697 | ||
ab9a0f19 | 2698 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2699 | { |
2700 | int order; | |
2701 | int min_objects; | |
2702 | int fraction; | |
e8120ff1 | 2703 | int max_objects; |
5e6d444e CL |
2704 | |
2705 | /* | |
2706 | * Attempt to find best configuration for a slab. This | |
2707 | * works by first attempting to generate a layout with | |
2708 | * the best configuration and backing off gradually. | |
2709 | * | |
2710 | * First we reduce the acceptable waste in a slab. Then | |
2711 | * we reduce the minimum objects required in a slab. | |
2712 | */ | |
2713 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2714 | if (!min_objects) |
2715 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2716 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2717 | min_objects = min(min_objects, max_objects); |
2718 | ||
5e6d444e | 2719 | while (min_objects > 1) { |
c124f5b5 | 2720 | fraction = 16; |
5e6d444e CL |
2721 | while (fraction >= 4) { |
2722 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2723 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2724 | if (order <= slub_max_order) |
2725 | return order; | |
2726 | fraction /= 2; | |
2727 | } | |
5086c389 | 2728 | min_objects--; |
5e6d444e CL |
2729 | } |
2730 | ||
2731 | /* | |
2732 | * We were unable to place multiple objects in a slab. Now | |
2733 | * lets see if we can place a single object there. | |
2734 | */ | |
ab9a0f19 | 2735 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2736 | if (order <= slub_max_order) |
2737 | return order; | |
2738 | ||
2739 | /* | |
2740 | * Doh this slab cannot be placed using slub_max_order. | |
2741 | */ | |
ab9a0f19 | 2742 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2743 | if (order < MAX_ORDER) |
5e6d444e CL |
2744 | return order; |
2745 | return -ENOSYS; | |
2746 | } | |
2747 | ||
81819f0f | 2748 | /* |
672bba3a | 2749 | * Figure out what the alignment of the objects will be. |
81819f0f CL |
2750 | */ |
2751 | static unsigned long calculate_alignment(unsigned long flags, | |
2752 | unsigned long align, unsigned long size) | |
2753 | { | |
2754 | /* | |
6446faa2 CL |
2755 | * If the user wants hardware cache aligned objects then follow that |
2756 | * suggestion if the object is sufficiently large. | |
81819f0f | 2757 | * |
6446faa2 CL |
2758 | * The hardware cache alignment cannot override the specified |
2759 | * alignment though. If that is greater then use it. | |
81819f0f | 2760 | */ |
b6210386 NP |
2761 | if (flags & SLAB_HWCACHE_ALIGN) { |
2762 | unsigned long ralign = cache_line_size(); | |
2763 | while (size <= ralign / 2) | |
2764 | ralign /= 2; | |
2765 | align = max(align, ralign); | |
2766 | } | |
81819f0f CL |
2767 | |
2768 | if (align < ARCH_SLAB_MINALIGN) | |
b6210386 | 2769 | align = ARCH_SLAB_MINALIGN; |
81819f0f CL |
2770 | |
2771 | return ALIGN(align, sizeof(void *)); | |
2772 | } | |
2773 | ||
5595cffc PE |
2774 | static void |
2775 | init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s) | |
81819f0f CL |
2776 | { |
2777 | n->nr_partial = 0; | |
81819f0f CL |
2778 | spin_lock_init(&n->list_lock); |
2779 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2780 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2781 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2782 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2783 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2784 | #endif |
81819f0f CL |
2785 | } |
2786 | ||
55136592 | 2787 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2788 | { |
6c182dc0 CL |
2789 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2790 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2791 | |
8a5ec0ba | 2792 | /* |
d4d84fef CM |
2793 | * Must align to double word boundary for the double cmpxchg |
2794 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2795 | */ |
d4d84fef CM |
2796 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2797 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2798 | |
2799 | if (!s->cpu_slab) | |
2800 | return 0; | |
2801 | ||
2802 | init_kmem_cache_cpus(s); | |
4c93c355 | 2803 | |
8a5ec0ba | 2804 | return 1; |
4c93c355 | 2805 | } |
4c93c355 | 2806 | |
51df1142 CL |
2807 | static struct kmem_cache *kmem_cache_node; |
2808 | ||
81819f0f CL |
2809 | /* |
2810 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2811 | * slab on the node for this slabcache. There are no concurrent accesses | |
2812 | * possible. | |
2813 | * | |
2814 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2815 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2816 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2817 | */ |
55136592 | 2818 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2819 | { |
2820 | struct page *page; | |
2821 | struct kmem_cache_node *n; | |
2822 | ||
51df1142 | 2823 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2824 | |
51df1142 | 2825 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2826 | |
2827 | BUG_ON(!page); | |
a2f92ee7 CL |
2828 | if (page_to_nid(page) != node) { |
2829 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2830 | "node %d\n", node); | |
2831 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2832 | "in order to be able to continue\n"); | |
2833 | } | |
2834 | ||
81819f0f CL |
2835 | n = page->freelist; |
2836 | BUG_ON(!n); | |
51df1142 | 2837 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2838 | page->inuse = 1; |
8cb0a506 | 2839 | page->frozen = 0; |
51df1142 | 2840 | kmem_cache_node->node[node] = n; |
8ab1372f | 2841 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2842 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2843 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2844 | #endif |
51df1142 CL |
2845 | init_kmem_cache_node(n, kmem_cache_node); |
2846 | inc_slabs_node(kmem_cache_node, node, page->objects); | |
6446faa2 | 2847 | |
136333d1 | 2848 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2849 | } |
2850 | ||
2851 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2852 | { | |
2853 | int node; | |
2854 | ||
f64dc58c | 2855 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2856 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2857 | |
73367bd8 | 2858 | if (n) |
51df1142 CL |
2859 | kmem_cache_free(kmem_cache_node, n); |
2860 | ||
81819f0f CL |
2861 | s->node[node] = NULL; |
2862 | } | |
2863 | } | |
2864 | ||
55136592 | 2865 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2866 | { |
2867 | int node; | |
81819f0f | 2868 | |
f64dc58c | 2869 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2870 | struct kmem_cache_node *n; |
2871 | ||
73367bd8 | 2872 | if (slab_state == DOWN) { |
55136592 | 2873 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2874 | continue; |
2875 | } | |
51df1142 | 2876 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2877 | GFP_KERNEL, node); |
81819f0f | 2878 | |
73367bd8 AD |
2879 | if (!n) { |
2880 | free_kmem_cache_nodes(s); | |
2881 | return 0; | |
81819f0f | 2882 | } |
73367bd8 | 2883 | |
81819f0f | 2884 | s->node[node] = n; |
5595cffc | 2885 | init_kmem_cache_node(n, s); |
81819f0f CL |
2886 | } |
2887 | return 1; | |
2888 | } | |
81819f0f | 2889 | |
c0bdb232 | 2890 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2891 | { |
2892 | if (min < MIN_PARTIAL) | |
2893 | min = MIN_PARTIAL; | |
2894 | else if (min > MAX_PARTIAL) | |
2895 | min = MAX_PARTIAL; | |
2896 | s->min_partial = min; | |
2897 | } | |
2898 | ||
81819f0f CL |
2899 | /* |
2900 | * calculate_sizes() determines the order and the distribution of data within | |
2901 | * a slab object. | |
2902 | */ | |
06b285dc | 2903 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2904 | { |
2905 | unsigned long flags = s->flags; | |
2906 | unsigned long size = s->objsize; | |
2907 | unsigned long align = s->align; | |
834f3d11 | 2908 | int order; |
81819f0f | 2909 | |
d8b42bf5 CL |
2910 | /* |
2911 | * Round up object size to the next word boundary. We can only | |
2912 | * place the free pointer at word boundaries and this determines | |
2913 | * the possible location of the free pointer. | |
2914 | */ | |
2915 | size = ALIGN(size, sizeof(void *)); | |
2916 | ||
2917 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2918 | /* |
2919 | * Determine if we can poison the object itself. If the user of | |
2920 | * the slab may touch the object after free or before allocation | |
2921 | * then we should never poison the object itself. | |
2922 | */ | |
2923 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2924 | !s->ctor) |
81819f0f CL |
2925 | s->flags |= __OBJECT_POISON; |
2926 | else | |
2927 | s->flags &= ~__OBJECT_POISON; | |
2928 | ||
81819f0f CL |
2929 | |
2930 | /* | |
672bba3a | 2931 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2932 | * end of the object and the free pointer. If not then add an |
672bba3a | 2933 | * additional word to have some bytes to store Redzone information. |
81819f0f CL |
2934 | */ |
2935 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
2936 | size += sizeof(void *); | |
41ecc55b | 2937 | #endif |
81819f0f CL |
2938 | |
2939 | /* | |
672bba3a CL |
2940 | * With that we have determined the number of bytes in actual use |
2941 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2942 | */ |
2943 | s->inuse = size; | |
2944 | ||
2945 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2946 | s->ctor)) { |
81819f0f CL |
2947 | /* |
2948 | * Relocate free pointer after the object if it is not | |
2949 | * permitted to overwrite the first word of the object on | |
2950 | * kmem_cache_free. | |
2951 | * | |
2952 | * This is the case if we do RCU, have a constructor or | |
2953 | * destructor or are poisoning the objects. | |
2954 | */ | |
2955 | s->offset = size; | |
2956 | size += sizeof(void *); | |
2957 | } | |
2958 | ||
c12b3c62 | 2959 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2960 | if (flags & SLAB_STORE_USER) |
2961 | /* | |
2962 | * Need to store information about allocs and frees after | |
2963 | * the object. | |
2964 | */ | |
2965 | size += 2 * sizeof(struct track); | |
2966 | ||
be7b3fbc | 2967 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2968 | /* |
2969 | * Add some empty padding so that we can catch | |
2970 | * overwrites from earlier objects rather than let | |
2971 | * tracking information or the free pointer be | |
0211a9c8 | 2972 | * corrupted if a user writes before the start |
81819f0f CL |
2973 | * of the object. |
2974 | */ | |
2975 | size += sizeof(void *); | |
41ecc55b | 2976 | #endif |
672bba3a | 2977 | |
81819f0f CL |
2978 | /* |
2979 | * Determine the alignment based on various parameters that the | |
65c02d4c CL |
2980 | * user specified and the dynamic determination of cache line size |
2981 | * on bootup. | |
81819f0f CL |
2982 | */ |
2983 | align = calculate_alignment(flags, align, s->objsize); | |
dcb0ce1b | 2984 | s->align = align; |
81819f0f CL |
2985 | |
2986 | /* | |
2987 | * SLUB stores one object immediately after another beginning from | |
2988 | * offset 0. In order to align the objects we have to simply size | |
2989 | * each object to conform to the alignment. | |
2990 | */ | |
2991 | size = ALIGN(size, align); | |
2992 | s->size = size; | |
06b285dc CL |
2993 | if (forced_order >= 0) |
2994 | order = forced_order; | |
2995 | else | |
ab9a0f19 | 2996 | order = calculate_order(size, s->reserved); |
81819f0f | 2997 | |
834f3d11 | 2998 | if (order < 0) |
81819f0f CL |
2999 | return 0; |
3000 | ||
b7a49f0d | 3001 | s->allocflags = 0; |
834f3d11 | 3002 | if (order) |
b7a49f0d CL |
3003 | s->allocflags |= __GFP_COMP; |
3004 | ||
3005 | if (s->flags & SLAB_CACHE_DMA) | |
3006 | s->allocflags |= SLUB_DMA; | |
3007 | ||
3008 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3009 | s->allocflags |= __GFP_RECLAIMABLE; | |
3010 | ||
81819f0f CL |
3011 | /* |
3012 | * Determine the number of objects per slab | |
3013 | */ | |
ab9a0f19 LJ |
3014 | s->oo = oo_make(order, size, s->reserved); |
3015 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3016 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3017 | s->max = s->oo; | |
81819f0f | 3018 | |
834f3d11 | 3019 | return !!oo_objects(s->oo); |
81819f0f CL |
3020 | |
3021 | } | |
3022 | ||
55136592 | 3023 | static int kmem_cache_open(struct kmem_cache *s, |
81819f0f CL |
3024 | const char *name, size_t size, |
3025 | size_t align, unsigned long flags, | |
51cc5068 | 3026 | void (*ctor)(void *)) |
81819f0f CL |
3027 | { |
3028 | memset(s, 0, kmem_size); | |
3029 | s->name = name; | |
3030 | s->ctor = ctor; | |
81819f0f | 3031 | s->objsize = size; |
81819f0f | 3032 | s->align = align; |
ba0268a8 | 3033 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
ab9a0f19 | 3034 | s->reserved = 0; |
81819f0f | 3035 | |
da9a638c LJ |
3036 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3037 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3038 | |
06b285dc | 3039 | if (!calculate_sizes(s, -1)) |
81819f0f | 3040 | goto error; |
3de47213 DR |
3041 | if (disable_higher_order_debug) { |
3042 | /* | |
3043 | * Disable debugging flags that store metadata if the min slab | |
3044 | * order increased. | |
3045 | */ | |
3046 | if (get_order(s->size) > get_order(s->objsize)) { | |
3047 | s->flags &= ~DEBUG_METADATA_FLAGS; | |
3048 | s->offset = 0; | |
3049 | if (!calculate_sizes(s, -1)) | |
3050 | goto error; | |
3051 | } | |
3052 | } | |
81819f0f | 3053 | |
2565409f HC |
3054 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3055 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3056 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3057 | /* Enable fast mode */ | |
3058 | s->flags |= __CMPXCHG_DOUBLE; | |
3059 | #endif | |
3060 | ||
3b89d7d8 DR |
3061 | /* |
3062 | * The larger the object size is, the more pages we want on the partial | |
3063 | * list to avoid pounding the page allocator excessively. | |
3064 | */ | |
49e22585 CL |
3065 | set_min_partial(s, ilog2(s->size) / 2); |
3066 | ||
3067 | /* | |
3068 | * cpu_partial determined the maximum number of objects kept in the | |
3069 | * per cpu partial lists of a processor. | |
3070 | * | |
3071 | * Per cpu partial lists mainly contain slabs that just have one | |
3072 | * object freed. If they are used for allocation then they can be | |
3073 | * filled up again with minimal effort. The slab will never hit the | |
3074 | * per node partial lists and therefore no locking will be required. | |
3075 | * | |
3076 | * This setting also determines | |
3077 | * | |
3078 | * A) The number of objects from per cpu partial slabs dumped to the | |
3079 | * per node list when we reach the limit. | |
9f264904 | 3080 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3081 | * per node list when we run out of per cpu objects. We only fetch 50% |
3082 | * to keep some capacity around for frees. | |
3083 | */ | |
8f1e33da CL |
3084 | if (kmem_cache_debug(s)) |
3085 | s->cpu_partial = 0; | |
3086 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3087 | s->cpu_partial = 2; |
3088 | else if (s->size >= 1024) | |
3089 | s->cpu_partial = 6; | |
3090 | else if (s->size >= 256) | |
3091 | s->cpu_partial = 13; | |
3092 | else | |
3093 | s->cpu_partial = 30; | |
3094 | ||
81819f0f CL |
3095 | s->refcount = 1; |
3096 | #ifdef CONFIG_NUMA | |
e2cb96b7 | 3097 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3098 | #endif |
55136592 | 3099 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3100 | goto error; |
81819f0f | 3101 | |
55136592 | 3102 | if (alloc_kmem_cache_cpus(s)) |
81819f0f | 3103 | return 1; |
ff12059e | 3104 | |
4c93c355 | 3105 | free_kmem_cache_nodes(s); |
81819f0f CL |
3106 | error: |
3107 | if (flags & SLAB_PANIC) | |
3108 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3109 | "order=%u offset=%u flags=%lx\n", | |
834f3d11 | 3110 | s->name, (unsigned long)size, s->size, oo_order(s->oo), |
81819f0f CL |
3111 | s->offset, flags); |
3112 | return 0; | |
3113 | } | |
81819f0f | 3114 | |
81819f0f CL |
3115 | /* |
3116 | * Determine the size of a slab object | |
3117 | */ | |
3118 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
3119 | { | |
3120 | return s->objsize; | |
3121 | } | |
3122 | EXPORT_SYMBOL(kmem_cache_size); | |
3123 | ||
33b12c38 CL |
3124 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3125 | const char *text) | |
3126 | { | |
3127 | #ifdef CONFIG_SLUB_DEBUG | |
3128 | void *addr = page_address(page); | |
3129 | void *p; | |
a5dd5c11 NK |
3130 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3131 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3132 | if (!map) |
3133 | return; | |
33b12c38 CL |
3134 | slab_err(s, page, "%s", text); |
3135 | slab_lock(page); | |
33b12c38 | 3136 | |
5f80b13a | 3137 | get_map(s, page, map); |
33b12c38 CL |
3138 | for_each_object(p, s, addr, page->objects) { |
3139 | ||
3140 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3141 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3142 | p, p - addr); | |
3143 | print_tracking(s, p); | |
3144 | } | |
3145 | } | |
3146 | slab_unlock(page); | |
bbd7d57b | 3147 | kfree(map); |
33b12c38 CL |
3148 | #endif |
3149 | } | |
3150 | ||
81819f0f | 3151 | /* |
599870b1 | 3152 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3153 | * This is called from kmem_cache_close(). We must be the last thread |
3154 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3155 | */ |
599870b1 | 3156 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3157 | { |
81819f0f CL |
3158 | struct page *page, *h; |
3159 | ||
33b12c38 | 3160 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3161 | if (!page->inuse) { |
5cc6eee8 | 3162 | remove_partial(n, page); |
81819f0f | 3163 | discard_slab(s, page); |
33b12c38 CL |
3164 | } else { |
3165 | list_slab_objects(s, page, | |
3166 | "Objects remaining on kmem_cache_close()"); | |
599870b1 | 3167 | } |
33b12c38 | 3168 | } |
81819f0f CL |
3169 | } |
3170 | ||
3171 | /* | |
672bba3a | 3172 | * Release all resources used by a slab cache. |
81819f0f | 3173 | */ |
0c710013 | 3174 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3175 | { |
3176 | int node; | |
3177 | ||
3178 | flush_all(s); | |
9dfc6e68 | 3179 | free_percpu(s->cpu_slab); |
81819f0f | 3180 | /* Attempt to free all objects */ |
f64dc58c | 3181 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3182 | struct kmem_cache_node *n = get_node(s, node); |
3183 | ||
599870b1 CL |
3184 | free_partial(s, n); |
3185 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3186 | return 1; |
3187 | } | |
3188 | free_kmem_cache_nodes(s); | |
3189 | return 0; | |
3190 | } | |
3191 | ||
3192 | /* | |
3193 | * Close a cache and release the kmem_cache structure | |
3194 | * (must be used for caches created using kmem_cache_create) | |
3195 | */ | |
3196 | void kmem_cache_destroy(struct kmem_cache *s) | |
3197 | { | |
3198 | down_write(&slub_lock); | |
3199 | s->refcount--; | |
3200 | if (!s->refcount) { | |
3201 | list_del(&s->list); | |
69cb8e6b | 3202 | up_write(&slub_lock); |
d629d819 PE |
3203 | if (kmem_cache_close(s)) { |
3204 | printk(KERN_ERR "SLUB %s: %s called for cache that " | |
3205 | "still has objects.\n", s->name, __func__); | |
3206 | dump_stack(); | |
3207 | } | |
d76b1590 ED |
3208 | if (s->flags & SLAB_DESTROY_BY_RCU) |
3209 | rcu_barrier(); | |
81819f0f | 3210 | sysfs_slab_remove(s); |
69cb8e6b CL |
3211 | } else |
3212 | up_write(&slub_lock); | |
81819f0f CL |
3213 | } |
3214 | EXPORT_SYMBOL(kmem_cache_destroy); | |
3215 | ||
3216 | /******************************************************************** | |
3217 | * Kmalloc subsystem | |
3218 | *******************************************************************/ | |
3219 | ||
51df1142 | 3220 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3221 | EXPORT_SYMBOL(kmalloc_caches); |
3222 | ||
51df1142 CL |
3223 | static struct kmem_cache *kmem_cache; |
3224 | ||
55136592 | 3225 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3226 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3227 | #endif |
3228 | ||
81819f0f CL |
3229 | static int __init setup_slub_min_order(char *str) |
3230 | { | |
06428780 | 3231 | get_option(&str, &slub_min_order); |
81819f0f CL |
3232 | |
3233 | return 1; | |
3234 | } | |
3235 | ||
3236 | __setup("slub_min_order=", setup_slub_min_order); | |
3237 | ||
3238 | static int __init setup_slub_max_order(char *str) | |
3239 | { | |
06428780 | 3240 | get_option(&str, &slub_max_order); |
818cf590 | 3241 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3242 | |
3243 | return 1; | |
3244 | } | |
3245 | ||
3246 | __setup("slub_max_order=", setup_slub_max_order); | |
3247 | ||
3248 | static int __init setup_slub_min_objects(char *str) | |
3249 | { | |
06428780 | 3250 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3251 | |
3252 | return 1; | |
3253 | } | |
3254 | ||
3255 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3256 | ||
3257 | static int __init setup_slub_nomerge(char *str) | |
3258 | { | |
3259 | slub_nomerge = 1; | |
3260 | return 1; | |
3261 | } | |
3262 | ||
3263 | __setup("slub_nomerge", setup_slub_nomerge); | |
3264 | ||
51df1142 CL |
3265 | static struct kmem_cache *__init create_kmalloc_cache(const char *name, |
3266 | int size, unsigned int flags) | |
81819f0f | 3267 | { |
51df1142 CL |
3268 | struct kmem_cache *s; |
3269 | ||
3270 | s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3271 | ||
83b519e8 PE |
3272 | /* |
3273 | * This function is called with IRQs disabled during early-boot on | |
3274 | * single CPU so there's no need to take slub_lock here. | |
3275 | */ | |
55136592 | 3276 | if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN, |
319d1e24 | 3277 | flags, NULL)) |
81819f0f CL |
3278 | goto panic; |
3279 | ||
3280 | list_add(&s->list, &slab_caches); | |
51df1142 | 3281 | return s; |
81819f0f CL |
3282 | |
3283 | panic: | |
3284 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
51df1142 | 3285 | return NULL; |
81819f0f CL |
3286 | } |
3287 | ||
f1b26339 CL |
3288 | /* |
3289 | * Conversion table for small slabs sizes / 8 to the index in the | |
3290 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3291 | * of two cache sizes there. The size of larger slabs can be determined using | |
3292 | * fls. | |
3293 | */ | |
3294 | static s8 size_index[24] = { | |
3295 | 3, /* 8 */ | |
3296 | 4, /* 16 */ | |
3297 | 5, /* 24 */ | |
3298 | 5, /* 32 */ | |
3299 | 6, /* 40 */ | |
3300 | 6, /* 48 */ | |
3301 | 6, /* 56 */ | |
3302 | 6, /* 64 */ | |
3303 | 1, /* 72 */ | |
3304 | 1, /* 80 */ | |
3305 | 1, /* 88 */ | |
3306 | 1, /* 96 */ | |
3307 | 7, /* 104 */ | |
3308 | 7, /* 112 */ | |
3309 | 7, /* 120 */ | |
3310 | 7, /* 128 */ | |
3311 | 2, /* 136 */ | |
3312 | 2, /* 144 */ | |
3313 | 2, /* 152 */ | |
3314 | 2, /* 160 */ | |
3315 | 2, /* 168 */ | |
3316 | 2, /* 176 */ | |
3317 | 2, /* 184 */ | |
3318 | 2 /* 192 */ | |
3319 | }; | |
3320 | ||
acdfcd04 AK |
3321 | static inline int size_index_elem(size_t bytes) |
3322 | { | |
3323 | return (bytes - 1) / 8; | |
3324 | } | |
3325 | ||
81819f0f CL |
3326 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3327 | { | |
f1b26339 | 3328 | int index; |
81819f0f | 3329 | |
f1b26339 CL |
3330 | if (size <= 192) { |
3331 | if (!size) | |
3332 | return ZERO_SIZE_PTR; | |
81819f0f | 3333 | |
acdfcd04 | 3334 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3335 | } else |
f1b26339 | 3336 | index = fls(size - 1); |
81819f0f CL |
3337 | |
3338 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3339 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3340 | return kmalloc_dma_caches[index]; |
f1b26339 | 3341 | |
81819f0f | 3342 | #endif |
51df1142 | 3343 | return kmalloc_caches[index]; |
81819f0f CL |
3344 | } |
3345 | ||
3346 | void *__kmalloc(size_t size, gfp_t flags) | |
3347 | { | |
aadb4bc4 | 3348 | struct kmem_cache *s; |
5b882be4 | 3349 | void *ret; |
81819f0f | 3350 | |
ffadd4d0 | 3351 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3352 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3353 | |
3354 | s = get_slab(size, flags); | |
3355 | ||
3356 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3357 | return s; |
3358 | ||
2154a336 | 3359 | ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_); |
5b882be4 | 3360 | |
ca2b84cb | 3361 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3362 | |
3363 | return ret; | |
81819f0f CL |
3364 | } |
3365 | EXPORT_SYMBOL(__kmalloc); | |
3366 | ||
5d1f57e4 | 3367 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3368 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3369 | { | |
b1eeab67 | 3370 | struct page *page; |
e4f7c0b4 | 3371 | void *ptr = NULL; |
f619cfe1 | 3372 | |
b1eeab67 VN |
3373 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3374 | page = alloc_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3375 | if (page) |
e4f7c0b4 CM |
3376 | ptr = page_address(page); |
3377 | ||
3378 | kmemleak_alloc(ptr, size, 1, flags); | |
3379 | return ptr; | |
f619cfe1 CL |
3380 | } |
3381 | ||
81819f0f CL |
3382 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3383 | { | |
aadb4bc4 | 3384 | struct kmem_cache *s; |
5b882be4 | 3385 | void *ret; |
81819f0f | 3386 | |
057685cf | 3387 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3388 | ret = kmalloc_large_node(size, flags, node); |
3389 | ||
ca2b84cb EGM |
3390 | trace_kmalloc_node(_RET_IP_, ret, |
3391 | size, PAGE_SIZE << get_order(size), | |
3392 | flags, node); | |
5b882be4 EGM |
3393 | |
3394 | return ret; | |
3395 | } | |
aadb4bc4 CL |
3396 | |
3397 | s = get_slab(size, flags); | |
3398 | ||
3399 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3400 | return s; |
3401 | ||
5b882be4 EGM |
3402 | ret = slab_alloc(s, flags, node, _RET_IP_); |
3403 | ||
ca2b84cb | 3404 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3405 | |
3406 | return ret; | |
81819f0f CL |
3407 | } |
3408 | EXPORT_SYMBOL(__kmalloc_node); | |
3409 | #endif | |
3410 | ||
3411 | size_t ksize(const void *object) | |
3412 | { | |
272c1d21 | 3413 | struct page *page; |
81819f0f | 3414 | |
ef8b4520 | 3415 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3416 | return 0; |
3417 | ||
294a80a8 | 3418 | page = virt_to_head_page(object); |
294a80a8 | 3419 | |
76994412 PE |
3420 | if (unlikely(!PageSlab(page))) { |
3421 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3422 | return PAGE_SIZE << compound_order(page); |
76994412 | 3423 | } |
81819f0f | 3424 | |
b3d41885 | 3425 | return slab_ksize(page->slab); |
81819f0f | 3426 | } |
b1aabecd | 3427 | EXPORT_SYMBOL(ksize); |
81819f0f | 3428 | |
d18a90dd BG |
3429 | #ifdef CONFIG_SLUB_DEBUG |
3430 | bool verify_mem_not_deleted(const void *x) | |
3431 | { | |
3432 | struct page *page; | |
3433 | void *object = (void *)x; | |
3434 | unsigned long flags; | |
3435 | bool rv; | |
3436 | ||
3437 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3438 | return false; | |
3439 | ||
3440 | local_irq_save(flags); | |
3441 | ||
3442 | page = virt_to_head_page(x); | |
3443 | if (unlikely(!PageSlab(page))) { | |
3444 | /* maybe it was from stack? */ | |
3445 | rv = true; | |
3446 | goto out_unlock; | |
3447 | } | |
3448 | ||
3449 | slab_lock(page); | |
3450 | if (on_freelist(page->slab, page, object)) { | |
3451 | object_err(page->slab, page, object, "Object is on free-list"); | |
3452 | rv = false; | |
3453 | } else { | |
3454 | rv = true; | |
3455 | } | |
3456 | slab_unlock(page); | |
3457 | ||
3458 | out_unlock: | |
3459 | local_irq_restore(flags); | |
3460 | return rv; | |
3461 | } | |
3462 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3463 | #endif | |
3464 | ||
81819f0f CL |
3465 | void kfree(const void *x) |
3466 | { | |
81819f0f | 3467 | struct page *page; |
5bb983b0 | 3468 | void *object = (void *)x; |
81819f0f | 3469 | |
2121db74 PE |
3470 | trace_kfree(_RET_IP_, x); |
3471 | ||
2408c550 | 3472 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3473 | return; |
3474 | ||
b49af68f | 3475 | page = virt_to_head_page(x); |
aadb4bc4 | 3476 | if (unlikely(!PageSlab(page))) { |
0937502a | 3477 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3478 | kmemleak_free(x); |
aadb4bc4 CL |
3479 | put_page(page); |
3480 | return; | |
3481 | } | |
ce71e27c | 3482 | slab_free(page->slab, page, object, _RET_IP_); |
81819f0f CL |
3483 | } |
3484 | EXPORT_SYMBOL(kfree); | |
3485 | ||
2086d26a | 3486 | /* |
672bba3a CL |
3487 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3488 | * the remaining slabs by the number of items in use. The slabs with the | |
3489 | * most items in use come first. New allocations will then fill those up | |
3490 | * and thus they can be removed from the partial lists. | |
3491 | * | |
3492 | * The slabs with the least items are placed last. This results in them | |
3493 | * being allocated from last increasing the chance that the last objects | |
3494 | * are freed in them. | |
2086d26a CL |
3495 | */ |
3496 | int kmem_cache_shrink(struct kmem_cache *s) | |
3497 | { | |
3498 | int node; | |
3499 | int i; | |
3500 | struct kmem_cache_node *n; | |
3501 | struct page *page; | |
3502 | struct page *t; | |
205ab99d | 3503 | int objects = oo_objects(s->max); |
2086d26a | 3504 | struct list_head *slabs_by_inuse = |
834f3d11 | 3505 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3506 | unsigned long flags; |
3507 | ||
3508 | if (!slabs_by_inuse) | |
3509 | return -ENOMEM; | |
3510 | ||
3511 | flush_all(s); | |
f64dc58c | 3512 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3513 | n = get_node(s, node); |
3514 | ||
3515 | if (!n->nr_partial) | |
3516 | continue; | |
3517 | ||
834f3d11 | 3518 | for (i = 0; i < objects; i++) |
2086d26a CL |
3519 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3520 | ||
3521 | spin_lock_irqsave(&n->list_lock, flags); | |
3522 | ||
3523 | /* | |
672bba3a | 3524 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3525 | * |
672bba3a CL |
3526 | * Note that concurrent frees may occur while we hold the |
3527 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3528 | */ |
3529 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3530 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3531 | if (!page->inuse) | |
3532 | n->nr_partial--; | |
2086d26a CL |
3533 | } |
3534 | ||
2086d26a | 3535 | /* |
672bba3a CL |
3536 | * Rebuild the partial list with the slabs filled up most |
3537 | * first and the least used slabs at the end. | |
2086d26a | 3538 | */ |
69cb8e6b | 3539 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3540 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3541 | ||
2086d26a | 3542 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3543 | |
3544 | /* Release empty slabs */ | |
3545 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3546 | discard_slab(s, page); | |
2086d26a CL |
3547 | } |
3548 | ||
3549 | kfree(slabs_by_inuse); | |
3550 | return 0; | |
3551 | } | |
3552 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3553 | ||
92a5bbc1 | 3554 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3555 | static int slab_mem_going_offline_callback(void *arg) |
3556 | { | |
3557 | struct kmem_cache *s; | |
3558 | ||
3559 | down_read(&slub_lock); | |
3560 | list_for_each_entry(s, &slab_caches, list) | |
3561 | kmem_cache_shrink(s); | |
3562 | up_read(&slub_lock); | |
3563 | ||
3564 | return 0; | |
3565 | } | |
3566 | ||
3567 | static void slab_mem_offline_callback(void *arg) | |
3568 | { | |
3569 | struct kmem_cache_node *n; | |
3570 | struct kmem_cache *s; | |
3571 | struct memory_notify *marg = arg; | |
3572 | int offline_node; | |
3573 | ||
3574 | offline_node = marg->status_change_nid; | |
3575 | ||
3576 | /* | |
3577 | * If the node still has available memory. we need kmem_cache_node | |
3578 | * for it yet. | |
3579 | */ | |
3580 | if (offline_node < 0) | |
3581 | return; | |
3582 | ||
3583 | down_read(&slub_lock); | |
3584 | list_for_each_entry(s, &slab_caches, list) { | |
3585 | n = get_node(s, offline_node); | |
3586 | if (n) { | |
3587 | /* | |
3588 | * if n->nr_slabs > 0, slabs still exist on the node | |
3589 | * that is going down. We were unable to free them, | |
c9404c9c | 3590 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3591 | * callback. So, we must fail. |
3592 | */ | |
0f389ec6 | 3593 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3594 | |
3595 | s->node[offline_node] = NULL; | |
8de66a0c | 3596 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3597 | } |
3598 | } | |
3599 | up_read(&slub_lock); | |
3600 | } | |
3601 | ||
3602 | static int slab_mem_going_online_callback(void *arg) | |
3603 | { | |
3604 | struct kmem_cache_node *n; | |
3605 | struct kmem_cache *s; | |
3606 | struct memory_notify *marg = arg; | |
3607 | int nid = marg->status_change_nid; | |
3608 | int ret = 0; | |
3609 | ||
3610 | /* | |
3611 | * If the node's memory is already available, then kmem_cache_node is | |
3612 | * already created. Nothing to do. | |
3613 | */ | |
3614 | if (nid < 0) | |
3615 | return 0; | |
3616 | ||
3617 | /* | |
0121c619 | 3618 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3619 | * allocate a kmem_cache_node structure in order to bring the node |
3620 | * online. | |
3621 | */ | |
3622 | down_read(&slub_lock); | |
3623 | list_for_each_entry(s, &slab_caches, list) { | |
3624 | /* | |
3625 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3626 | * since memory is not yet available from the node that | |
3627 | * is brought up. | |
3628 | */ | |
8de66a0c | 3629 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3630 | if (!n) { |
3631 | ret = -ENOMEM; | |
3632 | goto out; | |
3633 | } | |
5595cffc | 3634 | init_kmem_cache_node(n, s); |
b9049e23 YG |
3635 | s->node[nid] = n; |
3636 | } | |
3637 | out: | |
3638 | up_read(&slub_lock); | |
3639 | return ret; | |
3640 | } | |
3641 | ||
3642 | static int slab_memory_callback(struct notifier_block *self, | |
3643 | unsigned long action, void *arg) | |
3644 | { | |
3645 | int ret = 0; | |
3646 | ||
3647 | switch (action) { | |
3648 | case MEM_GOING_ONLINE: | |
3649 | ret = slab_mem_going_online_callback(arg); | |
3650 | break; | |
3651 | case MEM_GOING_OFFLINE: | |
3652 | ret = slab_mem_going_offline_callback(arg); | |
3653 | break; | |
3654 | case MEM_OFFLINE: | |
3655 | case MEM_CANCEL_ONLINE: | |
3656 | slab_mem_offline_callback(arg); | |
3657 | break; | |
3658 | case MEM_ONLINE: | |
3659 | case MEM_CANCEL_OFFLINE: | |
3660 | break; | |
3661 | } | |
dc19f9db KH |
3662 | if (ret) |
3663 | ret = notifier_from_errno(ret); | |
3664 | else | |
3665 | ret = NOTIFY_OK; | |
b9049e23 YG |
3666 | return ret; |
3667 | } | |
3668 | ||
3669 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3670 | ||
81819f0f CL |
3671 | /******************************************************************** |
3672 | * Basic setup of slabs | |
3673 | *******************************************************************/ | |
3674 | ||
51df1142 CL |
3675 | /* |
3676 | * Used for early kmem_cache structures that were allocated using | |
3677 | * the page allocator | |
3678 | */ | |
3679 | ||
3680 | static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s) | |
3681 | { | |
3682 | int node; | |
3683 | ||
3684 | list_add(&s->list, &slab_caches); | |
3685 | s->refcount = -1; | |
3686 | ||
3687 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3688 | struct kmem_cache_node *n = get_node(s, node); | |
3689 | struct page *p; | |
3690 | ||
3691 | if (n) { | |
3692 | list_for_each_entry(p, &n->partial, lru) | |
3693 | p->slab = s; | |
3694 | ||
607bf324 | 3695 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 CL |
3696 | list_for_each_entry(p, &n->full, lru) |
3697 | p->slab = s; | |
3698 | #endif | |
3699 | } | |
3700 | } | |
3701 | } | |
3702 | ||
81819f0f CL |
3703 | void __init kmem_cache_init(void) |
3704 | { | |
3705 | int i; | |
4b356be0 | 3706 | int caches = 0; |
51df1142 CL |
3707 | struct kmem_cache *temp_kmem_cache; |
3708 | int order; | |
51df1142 CL |
3709 | struct kmem_cache *temp_kmem_cache_node; |
3710 | unsigned long kmalloc_size; | |
3711 | ||
fc8d8620 SG |
3712 | if (debug_guardpage_minorder()) |
3713 | slub_max_order = 0; | |
3714 | ||
51df1142 CL |
3715 | kmem_size = offsetof(struct kmem_cache, node) + |
3716 | nr_node_ids * sizeof(struct kmem_cache_node *); | |
3717 | ||
3718 | /* Allocate two kmem_caches from the page allocator */ | |
3719 | kmalloc_size = ALIGN(kmem_size, cache_line_size()); | |
3720 | order = get_order(2 * kmalloc_size); | |
3721 | kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order); | |
3722 | ||
81819f0f CL |
3723 | /* |
3724 | * Must first have the slab cache available for the allocations of the | |
672bba3a | 3725 | * struct kmem_cache_node's. There is special bootstrap code in |
81819f0f CL |
3726 | * kmem_cache_open for slab_state == DOWN. |
3727 | */ | |
51df1142 CL |
3728 | kmem_cache_node = (void *)kmem_cache + kmalloc_size; |
3729 | ||
3730 | kmem_cache_open(kmem_cache_node, "kmem_cache_node", | |
3731 | sizeof(struct kmem_cache_node), | |
3732 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
b9049e23 | 3733 | |
0c40ba4f | 3734 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3735 | |
3736 | /* Able to allocate the per node structures */ | |
3737 | slab_state = PARTIAL; | |
3738 | ||
51df1142 CL |
3739 | temp_kmem_cache = kmem_cache; |
3740 | kmem_cache_open(kmem_cache, "kmem_cache", kmem_size, | |
3741 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
3742 | kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); | |
3743 | memcpy(kmem_cache, temp_kmem_cache, kmem_size); | |
81819f0f | 3744 | |
51df1142 CL |
3745 | /* |
3746 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3747 | * kmem_cache_node is separately allocated so no need to | |
3748 | * update any list pointers. | |
3749 | */ | |
3750 | temp_kmem_cache_node = kmem_cache_node; | |
81819f0f | 3751 | |
51df1142 CL |
3752 | kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT); |
3753 | memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size); | |
3754 | ||
3755 | kmem_cache_bootstrap_fixup(kmem_cache_node); | |
3756 | ||
3757 | caches++; | |
51df1142 CL |
3758 | kmem_cache_bootstrap_fixup(kmem_cache); |
3759 | caches++; | |
3760 | /* Free temporary boot structure */ | |
3761 | free_pages((unsigned long)temp_kmem_cache, order); | |
3762 | ||
3763 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3764 | |
3765 | /* | |
3766 | * Patch up the size_index table if we have strange large alignment | |
3767 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3768 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3769 | * |
3770 | * Largest permitted alignment is 256 bytes due to the way we | |
3771 | * handle the index determination for the smaller caches. | |
3772 | * | |
3773 | * Make sure that nothing crazy happens if someone starts tinkering | |
3774 | * around with ARCH_KMALLOC_MINALIGN | |
3775 | */ | |
3776 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3777 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3778 | ||
acdfcd04 AK |
3779 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3780 | int elem = size_index_elem(i); | |
3781 | if (elem >= ARRAY_SIZE(size_index)) | |
3782 | break; | |
3783 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3784 | } | |
f1b26339 | 3785 | |
acdfcd04 AK |
3786 | if (KMALLOC_MIN_SIZE == 64) { |
3787 | /* | |
3788 | * The 96 byte size cache is not used if the alignment | |
3789 | * is 64 byte. | |
3790 | */ | |
3791 | for (i = 64 + 8; i <= 96; i += 8) | |
3792 | size_index[size_index_elem(i)] = 7; | |
3793 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3794 | /* |
3795 | * The 192 byte sized cache is not used if the alignment | |
3796 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3797 | * instead. | |
3798 | */ | |
3799 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3800 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3801 | } |
3802 | ||
51df1142 CL |
3803 | /* Caches that are not of the two-to-the-power-of size */ |
3804 | if (KMALLOC_MIN_SIZE <= 32) { | |
3805 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3806 | caches++; | |
3807 | } | |
3808 | ||
3809 | if (KMALLOC_MIN_SIZE <= 64) { | |
3810 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3811 | caches++; | |
3812 | } | |
3813 | ||
3814 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3815 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3816 | caches++; | |
3817 | } | |
3818 | ||
81819f0f CL |
3819 | slab_state = UP; |
3820 | ||
3821 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3822 | if (KMALLOC_MIN_SIZE <= 32) { |
3823 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3824 | BUG_ON(!kmalloc_caches[1]->name); | |
3825 | } | |
3826 | ||
3827 | if (KMALLOC_MIN_SIZE <= 64) { | |
3828 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3829 | BUG_ON(!kmalloc_caches[2]->name); | |
3830 | } | |
3831 | ||
d7278bd7 CL |
3832 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3833 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3834 | ||
3835 | BUG_ON(!s); | |
51df1142 | 3836 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3837 | } |
81819f0f CL |
3838 | |
3839 | #ifdef CONFIG_SMP | |
3840 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3841 | #endif |
81819f0f | 3842 | |
55136592 | 3843 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3844 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3845 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3846 | |
51df1142 | 3847 | if (s && s->size) { |
55136592 CL |
3848 | char *name = kasprintf(GFP_NOWAIT, |
3849 | "dma-kmalloc-%d", s->objsize); | |
3850 | ||
3851 | BUG_ON(!name); | |
51df1142 CL |
3852 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3853 | s->objsize, SLAB_CACHE_DMA); | |
55136592 CL |
3854 | } |
3855 | } | |
3856 | #endif | |
3adbefee IM |
3857 | printk(KERN_INFO |
3858 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3859 | " CPUs=%d, Nodes=%d\n", |
3860 | caches, cache_line_size(), | |
81819f0f CL |
3861 | slub_min_order, slub_max_order, slub_min_objects, |
3862 | nr_cpu_ids, nr_node_ids); | |
3863 | } | |
3864 | ||
7e85ee0c PE |
3865 | void __init kmem_cache_init_late(void) |
3866 | { | |
7e85ee0c PE |
3867 | } |
3868 | ||
81819f0f CL |
3869 | /* |
3870 | * Find a mergeable slab cache | |
3871 | */ | |
3872 | static int slab_unmergeable(struct kmem_cache *s) | |
3873 | { | |
3874 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3875 | return 1; | |
3876 | ||
c59def9f | 3877 | if (s->ctor) |
81819f0f CL |
3878 | return 1; |
3879 | ||
8ffa6875 CL |
3880 | /* |
3881 | * We may have set a slab to be unmergeable during bootstrap. | |
3882 | */ | |
3883 | if (s->refcount < 0) | |
3884 | return 1; | |
3885 | ||
81819f0f CL |
3886 | return 0; |
3887 | } | |
3888 | ||
3889 | static struct kmem_cache *find_mergeable(size_t size, | |
ba0268a8 | 3890 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3891 | void (*ctor)(void *)) |
81819f0f | 3892 | { |
5b95a4ac | 3893 | struct kmem_cache *s; |
81819f0f CL |
3894 | |
3895 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3896 | return NULL; | |
3897 | ||
c59def9f | 3898 | if (ctor) |
81819f0f CL |
3899 | return NULL; |
3900 | ||
3901 | size = ALIGN(size, sizeof(void *)); | |
3902 | align = calculate_alignment(flags, align, size); | |
3903 | size = ALIGN(size, align); | |
ba0268a8 | 3904 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3905 | |
5b95a4ac | 3906 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3907 | if (slab_unmergeable(s)) |
3908 | continue; | |
3909 | ||
3910 | if (size > s->size) | |
3911 | continue; | |
3912 | ||
ba0268a8 | 3913 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3914 | continue; |
3915 | /* | |
3916 | * Check if alignment is compatible. | |
3917 | * Courtesy of Adrian Drzewiecki | |
3918 | */ | |
06428780 | 3919 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3920 | continue; |
3921 | ||
3922 | if (s->size - size >= sizeof(void *)) | |
3923 | continue; | |
3924 | ||
3925 | return s; | |
3926 | } | |
3927 | return NULL; | |
3928 | } | |
3929 | ||
3930 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 3931 | size_t align, unsigned long flags, void (*ctor)(void *)) |
81819f0f CL |
3932 | { |
3933 | struct kmem_cache *s; | |
84c1cf62 | 3934 | char *n; |
81819f0f | 3935 | |
fe1ff49d BH |
3936 | if (WARN_ON(!name)) |
3937 | return NULL; | |
3938 | ||
81819f0f | 3939 | down_write(&slub_lock); |
ba0268a8 | 3940 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3941 | if (s) { |
3942 | s->refcount++; | |
3943 | /* | |
3944 | * Adjust the object sizes so that we clear | |
3945 | * the complete object on kzalloc. | |
3946 | */ | |
3947 | s->objsize = max(s->objsize, (int)size); | |
3948 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
6446faa2 | 3949 | |
7b8f3b66 | 3950 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3951 | s->refcount--; |
81819f0f | 3952 | goto err; |
7b8f3b66 | 3953 | } |
2bce6485 | 3954 | up_write(&slub_lock); |
a0e1d1be CL |
3955 | return s; |
3956 | } | |
6446faa2 | 3957 | |
84c1cf62 PE |
3958 | n = kstrdup(name, GFP_KERNEL); |
3959 | if (!n) | |
3960 | goto err; | |
3961 | ||
a0e1d1be CL |
3962 | s = kmalloc(kmem_size, GFP_KERNEL); |
3963 | if (s) { | |
84c1cf62 | 3964 | if (kmem_cache_open(s, n, |
c59def9f | 3965 | size, align, flags, ctor)) { |
81819f0f | 3966 | list_add(&s->list, &slab_caches); |
66c4c35c | 3967 | up_write(&slub_lock); |
7b8f3b66 | 3968 | if (sysfs_slab_add(s)) { |
66c4c35c | 3969 | down_write(&slub_lock); |
7b8f3b66 | 3970 | list_del(&s->list); |
84c1cf62 | 3971 | kfree(n); |
7b8f3b66 | 3972 | kfree(s); |
a0e1d1be | 3973 | goto err; |
7b8f3b66 | 3974 | } |
a0e1d1be CL |
3975 | return s; |
3976 | } | |
84c1cf62 | 3977 | kfree(n); |
a0e1d1be | 3978 | kfree(s); |
81819f0f | 3979 | } |
68cee4f1 | 3980 | err: |
81819f0f | 3981 | up_write(&slub_lock); |
81819f0f | 3982 | |
81819f0f CL |
3983 | if (flags & SLAB_PANIC) |
3984 | panic("Cannot create slabcache %s\n", name); | |
3985 | else | |
3986 | s = NULL; | |
3987 | return s; | |
3988 | } | |
3989 | EXPORT_SYMBOL(kmem_cache_create); | |
3990 | ||
81819f0f | 3991 | #ifdef CONFIG_SMP |
81819f0f | 3992 | /* |
672bba3a CL |
3993 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3994 | * necessary. | |
81819f0f CL |
3995 | */ |
3996 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3997 | unsigned long action, void *hcpu) | |
3998 | { | |
3999 | long cpu = (long)hcpu; | |
5b95a4ac CL |
4000 | struct kmem_cache *s; |
4001 | unsigned long flags; | |
81819f0f CL |
4002 | |
4003 | switch (action) { | |
4004 | case CPU_UP_CANCELED: | |
8bb78442 | 4005 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4006 | case CPU_DEAD: |
8bb78442 | 4007 | case CPU_DEAD_FROZEN: |
5b95a4ac CL |
4008 | down_read(&slub_lock); |
4009 | list_for_each_entry(s, &slab_caches, list) { | |
4010 | local_irq_save(flags); | |
4011 | __flush_cpu_slab(s, cpu); | |
4012 | local_irq_restore(flags); | |
4013 | } | |
4014 | up_read(&slub_lock); | |
81819f0f CL |
4015 | break; |
4016 | default: | |
4017 | break; | |
4018 | } | |
4019 | return NOTIFY_OK; | |
4020 | } | |
4021 | ||
06428780 | 4022 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 4023 | .notifier_call = slab_cpuup_callback |
06428780 | 4024 | }; |
81819f0f CL |
4025 | |
4026 | #endif | |
4027 | ||
ce71e27c | 4028 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4029 | { |
aadb4bc4 | 4030 | struct kmem_cache *s; |
94b528d0 | 4031 | void *ret; |
aadb4bc4 | 4032 | |
ffadd4d0 | 4033 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
4034 | return kmalloc_large(size, gfpflags); |
4035 | ||
aadb4bc4 | 4036 | s = get_slab(size, gfpflags); |
81819f0f | 4037 | |
2408c550 | 4038 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4039 | return s; |
81819f0f | 4040 | |
2154a336 | 4041 | ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller); |
94b528d0 | 4042 | |
25985edc | 4043 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4044 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4045 | |
4046 | return ret; | |
81819f0f CL |
4047 | } |
4048 | ||
5d1f57e4 | 4049 | #ifdef CONFIG_NUMA |
81819f0f | 4050 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4051 | int node, unsigned long caller) |
81819f0f | 4052 | { |
aadb4bc4 | 4053 | struct kmem_cache *s; |
94b528d0 | 4054 | void *ret; |
aadb4bc4 | 4055 | |
d3e14aa3 XF |
4056 | if (unlikely(size > SLUB_MAX_SIZE)) { |
4057 | ret = kmalloc_large_node(size, gfpflags, node); | |
4058 | ||
4059 | trace_kmalloc_node(caller, ret, | |
4060 | size, PAGE_SIZE << get_order(size), | |
4061 | gfpflags, node); | |
4062 | ||
4063 | return ret; | |
4064 | } | |
eada35ef | 4065 | |
aadb4bc4 | 4066 | s = get_slab(size, gfpflags); |
81819f0f | 4067 | |
2408c550 | 4068 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4069 | return s; |
81819f0f | 4070 | |
94b528d0 EGM |
4071 | ret = slab_alloc(s, gfpflags, node, caller); |
4072 | ||
25985edc | 4073 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4074 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4075 | |
4076 | return ret; | |
81819f0f | 4077 | } |
5d1f57e4 | 4078 | #endif |
81819f0f | 4079 | |
ab4d5ed5 | 4080 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4081 | static int count_inuse(struct page *page) |
4082 | { | |
4083 | return page->inuse; | |
4084 | } | |
4085 | ||
4086 | static int count_total(struct page *page) | |
4087 | { | |
4088 | return page->objects; | |
4089 | } | |
ab4d5ed5 | 4090 | #endif |
205ab99d | 4091 | |
ab4d5ed5 | 4092 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4093 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4094 | unsigned long *map) | |
53e15af0 CL |
4095 | { |
4096 | void *p; | |
a973e9dd | 4097 | void *addr = page_address(page); |
53e15af0 CL |
4098 | |
4099 | if (!check_slab(s, page) || | |
4100 | !on_freelist(s, page, NULL)) | |
4101 | return 0; | |
4102 | ||
4103 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4104 | bitmap_zero(map, page->objects); |
53e15af0 | 4105 | |
5f80b13a CL |
4106 | get_map(s, page, map); |
4107 | for_each_object(p, s, addr, page->objects) { | |
4108 | if (test_bit(slab_index(p, s, addr), map)) | |
4109 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4110 | return 0; | |
53e15af0 CL |
4111 | } |
4112 | ||
224a88be | 4113 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4114 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4115 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4116 | return 0; |
4117 | return 1; | |
4118 | } | |
4119 | ||
434e245d CL |
4120 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4121 | unsigned long *map) | |
53e15af0 | 4122 | { |
881db7fb CL |
4123 | slab_lock(page); |
4124 | validate_slab(s, page, map); | |
4125 | slab_unlock(page); | |
53e15af0 CL |
4126 | } |
4127 | ||
434e245d CL |
4128 | static int validate_slab_node(struct kmem_cache *s, |
4129 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4130 | { |
4131 | unsigned long count = 0; | |
4132 | struct page *page; | |
4133 | unsigned long flags; | |
4134 | ||
4135 | spin_lock_irqsave(&n->list_lock, flags); | |
4136 | ||
4137 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4138 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4139 | count++; |
4140 | } | |
4141 | if (count != n->nr_partial) | |
4142 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4143 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4144 | ||
4145 | if (!(s->flags & SLAB_STORE_USER)) | |
4146 | goto out; | |
4147 | ||
4148 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4149 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4150 | count++; |
4151 | } | |
4152 | if (count != atomic_long_read(&n->nr_slabs)) | |
4153 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4154 | "counter=%ld\n", s->name, count, | |
4155 | atomic_long_read(&n->nr_slabs)); | |
4156 | ||
4157 | out: | |
4158 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4159 | return count; | |
4160 | } | |
4161 | ||
434e245d | 4162 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4163 | { |
4164 | int node; | |
4165 | unsigned long count = 0; | |
205ab99d | 4166 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4167 | sizeof(unsigned long), GFP_KERNEL); |
4168 | ||
4169 | if (!map) | |
4170 | return -ENOMEM; | |
53e15af0 CL |
4171 | |
4172 | flush_all(s); | |
f64dc58c | 4173 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4174 | struct kmem_cache_node *n = get_node(s, node); |
4175 | ||
434e245d | 4176 | count += validate_slab_node(s, n, map); |
53e15af0 | 4177 | } |
434e245d | 4178 | kfree(map); |
53e15af0 CL |
4179 | return count; |
4180 | } | |
88a420e4 | 4181 | /* |
672bba3a | 4182 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4183 | * and freed. |
4184 | */ | |
4185 | ||
4186 | struct location { | |
4187 | unsigned long count; | |
ce71e27c | 4188 | unsigned long addr; |
45edfa58 CL |
4189 | long long sum_time; |
4190 | long min_time; | |
4191 | long max_time; | |
4192 | long min_pid; | |
4193 | long max_pid; | |
174596a0 | 4194 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4195 | nodemask_t nodes; |
88a420e4 CL |
4196 | }; |
4197 | ||
4198 | struct loc_track { | |
4199 | unsigned long max; | |
4200 | unsigned long count; | |
4201 | struct location *loc; | |
4202 | }; | |
4203 | ||
4204 | static void free_loc_track(struct loc_track *t) | |
4205 | { | |
4206 | if (t->max) | |
4207 | free_pages((unsigned long)t->loc, | |
4208 | get_order(sizeof(struct location) * t->max)); | |
4209 | } | |
4210 | ||
68dff6a9 | 4211 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4212 | { |
4213 | struct location *l; | |
4214 | int order; | |
4215 | ||
88a420e4 CL |
4216 | order = get_order(sizeof(struct location) * max); |
4217 | ||
68dff6a9 | 4218 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4219 | if (!l) |
4220 | return 0; | |
4221 | ||
4222 | if (t->count) { | |
4223 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4224 | free_loc_track(t); | |
4225 | } | |
4226 | t->max = max; | |
4227 | t->loc = l; | |
4228 | return 1; | |
4229 | } | |
4230 | ||
4231 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4232 | const struct track *track) |
88a420e4 CL |
4233 | { |
4234 | long start, end, pos; | |
4235 | struct location *l; | |
ce71e27c | 4236 | unsigned long caddr; |
45edfa58 | 4237 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4238 | |
4239 | start = -1; | |
4240 | end = t->count; | |
4241 | ||
4242 | for ( ; ; ) { | |
4243 | pos = start + (end - start + 1) / 2; | |
4244 | ||
4245 | /* | |
4246 | * There is nothing at "end". If we end up there | |
4247 | * we need to add something to before end. | |
4248 | */ | |
4249 | if (pos == end) | |
4250 | break; | |
4251 | ||
4252 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4253 | if (track->addr == caddr) { |
4254 | ||
4255 | l = &t->loc[pos]; | |
4256 | l->count++; | |
4257 | if (track->when) { | |
4258 | l->sum_time += age; | |
4259 | if (age < l->min_time) | |
4260 | l->min_time = age; | |
4261 | if (age > l->max_time) | |
4262 | l->max_time = age; | |
4263 | ||
4264 | if (track->pid < l->min_pid) | |
4265 | l->min_pid = track->pid; | |
4266 | if (track->pid > l->max_pid) | |
4267 | l->max_pid = track->pid; | |
4268 | ||
174596a0 RR |
4269 | cpumask_set_cpu(track->cpu, |
4270 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4271 | } |
4272 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4273 | return 1; |
4274 | } | |
4275 | ||
45edfa58 | 4276 | if (track->addr < caddr) |
88a420e4 CL |
4277 | end = pos; |
4278 | else | |
4279 | start = pos; | |
4280 | } | |
4281 | ||
4282 | /* | |
672bba3a | 4283 | * Not found. Insert new tracking element. |
88a420e4 | 4284 | */ |
68dff6a9 | 4285 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4286 | return 0; |
4287 | ||
4288 | l = t->loc + pos; | |
4289 | if (pos < t->count) | |
4290 | memmove(l + 1, l, | |
4291 | (t->count - pos) * sizeof(struct location)); | |
4292 | t->count++; | |
4293 | l->count = 1; | |
45edfa58 CL |
4294 | l->addr = track->addr; |
4295 | l->sum_time = age; | |
4296 | l->min_time = age; | |
4297 | l->max_time = age; | |
4298 | l->min_pid = track->pid; | |
4299 | l->max_pid = track->pid; | |
174596a0 RR |
4300 | cpumask_clear(to_cpumask(l->cpus)); |
4301 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4302 | nodes_clear(l->nodes); |
4303 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4304 | return 1; |
4305 | } | |
4306 | ||
4307 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4308 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4309 | unsigned long *map) |
88a420e4 | 4310 | { |
a973e9dd | 4311 | void *addr = page_address(page); |
88a420e4 CL |
4312 | void *p; |
4313 | ||
39b26464 | 4314 | bitmap_zero(map, page->objects); |
5f80b13a | 4315 | get_map(s, page, map); |
88a420e4 | 4316 | |
224a88be | 4317 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4318 | if (!test_bit(slab_index(p, s, addr), map)) |
4319 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4320 | } |
4321 | ||
4322 | static int list_locations(struct kmem_cache *s, char *buf, | |
4323 | enum track_item alloc) | |
4324 | { | |
e374d483 | 4325 | int len = 0; |
88a420e4 | 4326 | unsigned long i; |
68dff6a9 | 4327 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4328 | int node; |
bbd7d57b ED |
4329 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4330 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4331 | |
bbd7d57b ED |
4332 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4333 | GFP_TEMPORARY)) { | |
4334 | kfree(map); | |
68dff6a9 | 4335 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4336 | } |
88a420e4 CL |
4337 | /* Push back cpu slabs */ |
4338 | flush_all(s); | |
4339 | ||
f64dc58c | 4340 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4341 | struct kmem_cache_node *n = get_node(s, node); |
4342 | unsigned long flags; | |
4343 | struct page *page; | |
4344 | ||
9e86943b | 4345 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4346 | continue; |
4347 | ||
4348 | spin_lock_irqsave(&n->list_lock, flags); | |
4349 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4350 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4351 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4352 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4353 | spin_unlock_irqrestore(&n->list_lock, flags); |
4354 | } | |
4355 | ||
4356 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4357 | struct location *l = &t.loc[i]; |
88a420e4 | 4358 | |
9c246247 | 4359 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4360 | break; |
e374d483 | 4361 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4362 | |
4363 | if (l->addr) | |
62c70bce | 4364 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4365 | else |
e374d483 | 4366 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4367 | |
4368 | if (l->sum_time != l->min_time) { | |
e374d483 | 4369 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4370 | l->min_time, |
4371 | (long)div_u64(l->sum_time, l->count), | |
4372 | l->max_time); | |
45edfa58 | 4373 | } else |
e374d483 | 4374 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4375 | l->min_time); |
4376 | ||
4377 | if (l->min_pid != l->max_pid) | |
e374d483 | 4378 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4379 | l->min_pid, l->max_pid); |
4380 | else | |
e374d483 | 4381 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4382 | l->min_pid); |
4383 | ||
174596a0 RR |
4384 | if (num_online_cpus() > 1 && |
4385 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4386 | len < PAGE_SIZE - 60) { |
4387 | len += sprintf(buf + len, " cpus="); | |
4388 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4389 | to_cpumask(l->cpus)); |
45edfa58 CL |
4390 | } |
4391 | ||
62bc62a8 | 4392 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4393 | len < PAGE_SIZE - 60) { |
4394 | len += sprintf(buf + len, " nodes="); | |
4395 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4396 | l->nodes); |
4397 | } | |
4398 | ||
e374d483 | 4399 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4400 | } |
4401 | ||
4402 | free_loc_track(&t); | |
bbd7d57b | 4403 | kfree(map); |
88a420e4 | 4404 | if (!t.count) |
e374d483 HH |
4405 | len += sprintf(buf, "No data\n"); |
4406 | return len; | |
88a420e4 | 4407 | } |
ab4d5ed5 | 4408 | #endif |
88a420e4 | 4409 | |
a5a84755 CL |
4410 | #ifdef SLUB_RESILIENCY_TEST |
4411 | static void resiliency_test(void) | |
4412 | { | |
4413 | u8 *p; | |
4414 | ||
4415 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4416 | ||
4417 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4418 | printk(KERN_ERR "-----------------------\n"); | |
4419 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4420 | ||
4421 | p = kzalloc(16, GFP_KERNEL); | |
4422 | p[16] = 0x12; | |
4423 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4424 | " 0x12->0x%p\n\n", p + 16); | |
4425 | ||
4426 | validate_slab_cache(kmalloc_caches[4]); | |
4427 | ||
4428 | /* Hmmm... The next two are dangerous */ | |
4429 | p = kzalloc(32, GFP_KERNEL); | |
4430 | p[32 + sizeof(void *)] = 0x34; | |
4431 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4432 | " 0x34 -> -0x%p\n", p); | |
4433 | printk(KERN_ERR | |
4434 | "If allocated object is overwritten then not detectable\n\n"); | |
4435 | ||
4436 | validate_slab_cache(kmalloc_caches[5]); | |
4437 | p = kzalloc(64, GFP_KERNEL); | |
4438 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4439 | *p = 0x56; | |
4440 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4441 | p); | |
4442 | printk(KERN_ERR | |
4443 | "If allocated object is overwritten then not detectable\n\n"); | |
4444 | validate_slab_cache(kmalloc_caches[6]); | |
4445 | ||
4446 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4447 | p = kzalloc(128, GFP_KERNEL); | |
4448 | kfree(p); | |
4449 | *p = 0x78; | |
4450 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4451 | validate_slab_cache(kmalloc_caches[7]); | |
4452 | ||
4453 | p = kzalloc(256, GFP_KERNEL); | |
4454 | kfree(p); | |
4455 | p[50] = 0x9a; | |
4456 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4457 | p); | |
4458 | validate_slab_cache(kmalloc_caches[8]); | |
4459 | ||
4460 | p = kzalloc(512, GFP_KERNEL); | |
4461 | kfree(p); | |
4462 | p[512] = 0xab; | |
4463 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4464 | validate_slab_cache(kmalloc_caches[9]); | |
4465 | } | |
4466 | #else | |
4467 | #ifdef CONFIG_SYSFS | |
4468 | static void resiliency_test(void) {}; | |
4469 | #endif | |
4470 | #endif | |
4471 | ||
ab4d5ed5 | 4472 | #ifdef CONFIG_SYSFS |
81819f0f | 4473 | enum slab_stat_type { |
205ab99d CL |
4474 | SL_ALL, /* All slabs */ |
4475 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4476 | SL_CPU, /* Only slabs used for cpu caches */ | |
4477 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4478 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4479 | }; |
4480 | ||
205ab99d | 4481 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4482 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4483 | #define SO_CPU (1 << SL_CPU) | |
4484 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4485 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4486 | |
62e5c4b4 CG |
4487 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4488 | char *buf, unsigned long flags) | |
81819f0f CL |
4489 | { |
4490 | unsigned long total = 0; | |
81819f0f CL |
4491 | int node; |
4492 | int x; | |
4493 | unsigned long *nodes; | |
4494 | unsigned long *per_cpu; | |
4495 | ||
4496 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4497 | if (!nodes) |
4498 | return -ENOMEM; | |
81819f0f CL |
4499 | per_cpu = nodes + nr_node_ids; |
4500 | ||
205ab99d CL |
4501 | if (flags & SO_CPU) { |
4502 | int cpu; | |
81819f0f | 4503 | |
205ab99d | 4504 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4505 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4506 | int node; |
49e22585 | 4507 | struct page *page; |
dfb4f096 | 4508 | |
bc6697d8 | 4509 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4510 | if (!page) |
4511 | continue; | |
205ab99d | 4512 | |
ec3ab083 CL |
4513 | node = page_to_nid(page); |
4514 | if (flags & SO_TOTAL) | |
4515 | x = page->objects; | |
4516 | else if (flags & SO_OBJECTS) | |
4517 | x = page->inuse; | |
4518 | else | |
4519 | x = 1; | |
49e22585 | 4520 | |
ec3ab083 CL |
4521 | total += x; |
4522 | nodes[node] += x; | |
4523 | ||
4524 | page = ACCESS_ONCE(c->partial); | |
49e22585 CL |
4525 | if (page) { |
4526 | x = page->pobjects; | |
bc6697d8 ED |
4527 | total += x; |
4528 | nodes[node] += x; | |
49e22585 | 4529 | } |
ec3ab083 | 4530 | |
bc6697d8 | 4531 | per_cpu[node]++; |
81819f0f CL |
4532 | } |
4533 | } | |
4534 | ||
04d94879 | 4535 | lock_memory_hotplug(); |
ab4d5ed5 | 4536 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4537 | if (flags & SO_ALL) { |
4538 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4539 | struct kmem_cache_node *n = get_node(s, node); | |
4540 | ||
4541 | if (flags & SO_TOTAL) | |
4542 | x = atomic_long_read(&n->total_objects); | |
4543 | else if (flags & SO_OBJECTS) | |
4544 | x = atomic_long_read(&n->total_objects) - | |
4545 | count_partial(n, count_free); | |
81819f0f | 4546 | |
81819f0f | 4547 | else |
205ab99d | 4548 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4549 | total += x; |
4550 | nodes[node] += x; | |
4551 | } | |
4552 | ||
ab4d5ed5 CL |
4553 | } else |
4554 | #endif | |
4555 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4556 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4557 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4558 | |
205ab99d CL |
4559 | if (flags & SO_TOTAL) |
4560 | x = count_partial(n, count_total); | |
4561 | else if (flags & SO_OBJECTS) | |
4562 | x = count_partial(n, count_inuse); | |
81819f0f | 4563 | else |
205ab99d | 4564 | x = n->nr_partial; |
81819f0f CL |
4565 | total += x; |
4566 | nodes[node] += x; | |
4567 | } | |
4568 | } | |
81819f0f CL |
4569 | x = sprintf(buf, "%lu", total); |
4570 | #ifdef CONFIG_NUMA | |
f64dc58c | 4571 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4572 | if (nodes[node]) |
4573 | x += sprintf(buf + x, " N%d=%lu", | |
4574 | node, nodes[node]); | |
4575 | #endif | |
04d94879 | 4576 | unlock_memory_hotplug(); |
81819f0f CL |
4577 | kfree(nodes); |
4578 | return x + sprintf(buf + x, "\n"); | |
4579 | } | |
4580 | ||
ab4d5ed5 | 4581 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4582 | static int any_slab_objects(struct kmem_cache *s) |
4583 | { | |
4584 | int node; | |
81819f0f | 4585 | |
dfb4f096 | 4586 | for_each_online_node(node) { |
81819f0f CL |
4587 | struct kmem_cache_node *n = get_node(s, node); |
4588 | ||
dfb4f096 CL |
4589 | if (!n) |
4590 | continue; | |
4591 | ||
4ea33e2d | 4592 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4593 | return 1; |
4594 | } | |
4595 | return 0; | |
4596 | } | |
ab4d5ed5 | 4597 | #endif |
81819f0f CL |
4598 | |
4599 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4600 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4601 | |
4602 | struct slab_attribute { | |
4603 | struct attribute attr; | |
4604 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4605 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4606 | }; | |
4607 | ||
4608 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4609 | static struct slab_attribute _name##_attr = \ |
4610 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4611 | |
4612 | #define SLAB_ATTR(_name) \ | |
4613 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4614 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4615 | |
81819f0f CL |
4616 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4617 | { | |
4618 | return sprintf(buf, "%d\n", s->size); | |
4619 | } | |
4620 | SLAB_ATTR_RO(slab_size); | |
4621 | ||
4622 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4623 | { | |
4624 | return sprintf(buf, "%d\n", s->align); | |
4625 | } | |
4626 | SLAB_ATTR_RO(align); | |
4627 | ||
4628 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4629 | { | |
4630 | return sprintf(buf, "%d\n", s->objsize); | |
4631 | } | |
4632 | SLAB_ATTR_RO(object_size); | |
4633 | ||
4634 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4635 | { | |
834f3d11 | 4636 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4637 | } |
4638 | SLAB_ATTR_RO(objs_per_slab); | |
4639 | ||
06b285dc CL |
4640 | static ssize_t order_store(struct kmem_cache *s, |
4641 | const char *buf, size_t length) | |
4642 | { | |
0121c619 CL |
4643 | unsigned long order; |
4644 | int err; | |
4645 | ||
4646 | err = strict_strtoul(buf, 10, &order); | |
4647 | if (err) | |
4648 | return err; | |
06b285dc CL |
4649 | |
4650 | if (order > slub_max_order || order < slub_min_order) | |
4651 | return -EINVAL; | |
4652 | ||
4653 | calculate_sizes(s, order); | |
4654 | return length; | |
4655 | } | |
4656 | ||
81819f0f CL |
4657 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4658 | { | |
834f3d11 | 4659 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4660 | } |
06b285dc | 4661 | SLAB_ATTR(order); |
81819f0f | 4662 | |
73d342b1 DR |
4663 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4664 | { | |
4665 | return sprintf(buf, "%lu\n", s->min_partial); | |
4666 | } | |
4667 | ||
4668 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4669 | size_t length) | |
4670 | { | |
4671 | unsigned long min; | |
4672 | int err; | |
4673 | ||
4674 | err = strict_strtoul(buf, 10, &min); | |
4675 | if (err) | |
4676 | return err; | |
4677 | ||
c0bdb232 | 4678 | set_min_partial(s, min); |
73d342b1 DR |
4679 | return length; |
4680 | } | |
4681 | SLAB_ATTR(min_partial); | |
4682 | ||
49e22585 CL |
4683 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4684 | { | |
4685 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4686 | } | |
4687 | ||
4688 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4689 | size_t length) | |
4690 | { | |
4691 | unsigned long objects; | |
4692 | int err; | |
4693 | ||
4694 | err = strict_strtoul(buf, 10, &objects); | |
4695 | if (err) | |
4696 | return err; | |
74ee4ef1 DR |
4697 | if (objects && kmem_cache_debug(s)) |
4698 | return -EINVAL; | |
49e22585 CL |
4699 | |
4700 | s->cpu_partial = objects; | |
4701 | flush_all(s); | |
4702 | return length; | |
4703 | } | |
4704 | SLAB_ATTR(cpu_partial); | |
4705 | ||
81819f0f CL |
4706 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4707 | { | |
62c70bce JP |
4708 | if (!s->ctor) |
4709 | return 0; | |
4710 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4711 | } |
4712 | SLAB_ATTR_RO(ctor); | |
4713 | ||
81819f0f CL |
4714 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4715 | { | |
4716 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4717 | } | |
4718 | SLAB_ATTR_RO(aliases); | |
4719 | ||
81819f0f CL |
4720 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4721 | { | |
d9acf4b7 | 4722 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4723 | } |
4724 | SLAB_ATTR_RO(partial); | |
4725 | ||
4726 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4727 | { | |
d9acf4b7 | 4728 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4729 | } |
4730 | SLAB_ATTR_RO(cpu_slabs); | |
4731 | ||
4732 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4733 | { | |
205ab99d | 4734 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4735 | } |
4736 | SLAB_ATTR_RO(objects); | |
4737 | ||
205ab99d CL |
4738 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4739 | { | |
4740 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4741 | } | |
4742 | SLAB_ATTR_RO(objects_partial); | |
4743 | ||
49e22585 CL |
4744 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4745 | { | |
4746 | int objects = 0; | |
4747 | int pages = 0; | |
4748 | int cpu; | |
4749 | int len; | |
4750 | ||
4751 | for_each_online_cpu(cpu) { | |
4752 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4753 | ||
4754 | if (page) { | |
4755 | pages += page->pages; | |
4756 | objects += page->pobjects; | |
4757 | } | |
4758 | } | |
4759 | ||
4760 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4761 | ||
4762 | #ifdef CONFIG_SMP | |
4763 | for_each_online_cpu(cpu) { | |
4764 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4765 | ||
4766 | if (page && len < PAGE_SIZE - 20) | |
4767 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4768 | page->pobjects, page->pages); | |
4769 | } | |
4770 | #endif | |
4771 | return len + sprintf(buf + len, "\n"); | |
4772 | } | |
4773 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4774 | ||
a5a84755 CL |
4775 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4776 | { | |
4777 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4778 | } | |
4779 | ||
4780 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4781 | const char *buf, size_t length) | |
4782 | { | |
4783 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4784 | if (buf[0] == '1') | |
4785 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4786 | return length; | |
4787 | } | |
4788 | SLAB_ATTR(reclaim_account); | |
4789 | ||
4790 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4791 | { | |
4792 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4793 | } | |
4794 | SLAB_ATTR_RO(hwcache_align); | |
4795 | ||
4796 | #ifdef CONFIG_ZONE_DMA | |
4797 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4798 | { | |
4799 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4800 | } | |
4801 | SLAB_ATTR_RO(cache_dma); | |
4802 | #endif | |
4803 | ||
4804 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4805 | { | |
4806 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4807 | } | |
4808 | SLAB_ATTR_RO(destroy_by_rcu); | |
4809 | ||
ab9a0f19 LJ |
4810 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4811 | { | |
4812 | return sprintf(buf, "%d\n", s->reserved); | |
4813 | } | |
4814 | SLAB_ATTR_RO(reserved); | |
4815 | ||
ab4d5ed5 | 4816 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4817 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4818 | { | |
4819 | return show_slab_objects(s, buf, SO_ALL); | |
4820 | } | |
4821 | SLAB_ATTR_RO(slabs); | |
4822 | ||
205ab99d CL |
4823 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4824 | { | |
4825 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4826 | } | |
4827 | SLAB_ATTR_RO(total_objects); | |
4828 | ||
81819f0f CL |
4829 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4830 | { | |
4831 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4832 | } | |
4833 | ||
4834 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4835 | const char *buf, size_t length) | |
4836 | { | |
4837 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4838 | if (buf[0] == '1') { |
4839 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4840 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4841 | } |
81819f0f CL |
4842 | return length; |
4843 | } | |
4844 | SLAB_ATTR(sanity_checks); | |
4845 | ||
4846 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4847 | { | |
4848 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4849 | } | |
4850 | ||
4851 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4852 | size_t length) | |
4853 | { | |
4854 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4855 | if (buf[0] == '1') { |
4856 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4857 | s->flags |= SLAB_TRACE; |
b789ef51 | 4858 | } |
81819f0f CL |
4859 | return length; |
4860 | } | |
4861 | SLAB_ATTR(trace); | |
4862 | ||
81819f0f CL |
4863 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4864 | { | |
4865 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4866 | } | |
4867 | ||
4868 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4869 | const char *buf, size_t length) | |
4870 | { | |
4871 | if (any_slab_objects(s)) | |
4872 | return -EBUSY; | |
4873 | ||
4874 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4875 | if (buf[0] == '1') { |
4876 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4877 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4878 | } |
06b285dc | 4879 | calculate_sizes(s, -1); |
81819f0f CL |
4880 | return length; |
4881 | } | |
4882 | SLAB_ATTR(red_zone); | |
4883 | ||
4884 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4885 | { | |
4886 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4887 | } | |
4888 | ||
4889 | static ssize_t poison_store(struct kmem_cache *s, | |
4890 | const char *buf, size_t length) | |
4891 | { | |
4892 | if (any_slab_objects(s)) | |
4893 | return -EBUSY; | |
4894 | ||
4895 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4896 | if (buf[0] == '1') { |
4897 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4898 | s->flags |= SLAB_POISON; |
b789ef51 | 4899 | } |
06b285dc | 4900 | calculate_sizes(s, -1); |
81819f0f CL |
4901 | return length; |
4902 | } | |
4903 | SLAB_ATTR(poison); | |
4904 | ||
4905 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4906 | { | |
4907 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4908 | } | |
4909 | ||
4910 | static ssize_t store_user_store(struct kmem_cache *s, | |
4911 | const char *buf, size_t length) | |
4912 | { | |
4913 | if (any_slab_objects(s)) | |
4914 | return -EBUSY; | |
4915 | ||
4916 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4917 | if (buf[0] == '1') { |
4918 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4919 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4920 | } |
06b285dc | 4921 | calculate_sizes(s, -1); |
81819f0f CL |
4922 | return length; |
4923 | } | |
4924 | SLAB_ATTR(store_user); | |
4925 | ||
53e15af0 CL |
4926 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4927 | { | |
4928 | return 0; | |
4929 | } | |
4930 | ||
4931 | static ssize_t validate_store(struct kmem_cache *s, | |
4932 | const char *buf, size_t length) | |
4933 | { | |
434e245d CL |
4934 | int ret = -EINVAL; |
4935 | ||
4936 | if (buf[0] == '1') { | |
4937 | ret = validate_slab_cache(s); | |
4938 | if (ret >= 0) | |
4939 | ret = length; | |
4940 | } | |
4941 | return ret; | |
53e15af0 CL |
4942 | } |
4943 | SLAB_ATTR(validate); | |
a5a84755 CL |
4944 | |
4945 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4946 | { | |
4947 | if (!(s->flags & SLAB_STORE_USER)) | |
4948 | return -ENOSYS; | |
4949 | return list_locations(s, buf, TRACK_ALLOC); | |
4950 | } | |
4951 | SLAB_ATTR_RO(alloc_calls); | |
4952 | ||
4953 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4954 | { | |
4955 | if (!(s->flags & SLAB_STORE_USER)) | |
4956 | return -ENOSYS; | |
4957 | return list_locations(s, buf, TRACK_FREE); | |
4958 | } | |
4959 | SLAB_ATTR_RO(free_calls); | |
4960 | #endif /* CONFIG_SLUB_DEBUG */ | |
4961 | ||
4962 | #ifdef CONFIG_FAILSLAB | |
4963 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4964 | { | |
4965 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4966 | } | |
4967 | ||
4968 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4969 | size_t length) | |
4970 | { | |
4971 | s->flags &= ~SLAB_FAILSLAB; | |
4972 | if (buf[0] == '1') | |
4973 | s->flags |= SLAB_FAILSLAB; | |
4974 | return length; | |
4975 | } | |
4976 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4977 | #endif |
53e15af0 | 4978 | |
2086d26a CL |
4979 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4980 | { | |
4981 | return 0; | |
4982 | } | |
4983 | ||
4984 | static ssize_t shrink_store(struct kmem_cache *s, | |
4985 | const char *buf, size_t length) | |
4986 | { | |
4987 | if (buf[0] == '1') { | |
4988 | int rc = kmem_cache_shrink(s); | |
4989 | ||
4990 | if (rc) | |
4991 | return rc; | |
4992 | } else | |
4993 | return -EINVAL; | |
4994 | return length; | |
4995 | } | |
4996 | SLAB_ATTR(shrink); | |
4997 | ||
81819f0f | 4998 | #ifdef CONFIG_NUMA |
9824601e | 4999 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5000 | { |
9824601e | 5001 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5002 | } |
5003 | ||
9824601e | 5004 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5005 | const char *buf, size_t length) |
5006 | { | |
0121c619 CL |
5007 | unsigned long ratio; |
5008 | int err; | |
5009 | ||
5010 | err = strict_strtoul(buf, 10, &ratio); | |
5011 | if (err) | |
5012 | return err; | |
5013 | ||
e2cb96b7 | 5014 | if (ratio <= 100) |
0121c619 | 5015 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5016 | |
81819f0f CL |
5017 | return length; |
5018 | } | |
9824601e | 5019 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5020 | #endif |
5021 | ||
8ff12cfc | 5022 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5023 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5024 | { | |
5025 | unsigned long sum = 0; | |
5026 | int cpu; | |
5027 | int len; | |
5028 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5029 | ||
5030 | if (!data) | |
5031 | return -ENOMEM; | |
5032 | ||
5033 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5034 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5035 | |
5036 | data[cpu] = x; | |
5037 | sum += x; | |
5038 | } | |
5039 | ||
5040 | len = sprintf(buf, "%lu", sum); | |
5041 | ||
50ef37b9 | 5042 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5043 | for_each_online_cpu(cpu) { |
5044 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5045 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5046 | } |
50ef37b9 | 5047 | #endif |
8ff12cfc CL |
5048 | kfree(data); |
5049 | return len + sprintf(buf + len, "\n"); | |
5050 | } | |
5051 | ||
78eb00cc DR |
5052 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5053 | { | |
5054 | int cpu; | |
5055 | ||
5056 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5057 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5058 | } |
5059 | ||
8ff12cfc CL |
5060 | #define STAT_ATTR(si, text) \ |
5061 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5062 | { \ | |
5063 | return show_stat(s, buf, si); \ | |
5064 | } \ | |
78eb00cc DR |
5065 | static ssize_t text##_store(struct kmem_cache *s, \ |
5066 | const char *buf, size_t length) \ | |
5067 | { \ | |
5068 | if (buf[0] != '0') \ | |
5069 | return -EINVAL; \ | |
5070 | clear_stat(s, si); \ | |
5071 | return length; \ | |
5072 | } \ | |
5073 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5074 | |
5075 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5076 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5077 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5078 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5079 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5080 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5081 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5082 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5083 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5084 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5085 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5086 | STAT_ATTR(FREE_SLAB, free_slab); |
5087 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5088 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5089 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5090 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5091 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5092 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5093 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5094 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5095 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5096 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5097 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5098 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5099 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5100 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5101 | #endif |
5102 | ||
06428780 | 5103 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5104 | &slab_size_attr.attr, |
5105 | &object_size_attr.attr, | |
5106 | &objs_per_slab_attr.attr, | |
5107 | &order_attr.attr, | |
73d342b1 | 5108 | &min_partial_attr.attr, |
49e22585 | 5109 | &cpu_partial_attr.attr, |
81819f0f | 5110 | &objects_attr.attr, |
205ab99d | 5111 | &objects_partial_attr.attr, |
81819f0f CL |
5112 | &partial_attr.attr, |
5113 | &cpu_slabs_attr.attr, | |
5114 | &ctor_attr.attr, | |
81819f0f CL |
5115 | &aliases_attr.attr, |
5116 | &align_attr.attr, | |
81819f0f CL |
5117 | &hwcache_align_attr.attr, |
5118 | &reclaim_account_attr.attr, | |
5119 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5120 | &shrink_attr.attr, |
ab9a0f19 | 5121 | &reserved_attr.attr, |
49e22585 | 5122 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5123 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5124 | &total_objects_attr.attr, |
5125 | &slabs_attr.attr, | |
5126 | &sanity_checks_attr.attr, | |
5127 | &trace_attr.attr, | |
81819f0f CL |
5128 | &red_zone_attr.attr, |
5129 | &poison_attr.attr, | |
5130 | &store_user_attr.attr, | |
53e15af0 | 5131 | &validate_attr.attr, |
88a420e4 CL |
5132 | &alloc_calls_attr.attr, |
5133 | &free_calls_attr.attr, | |
ab4d5ed5 | 5134 | #endif |
81819f0f CL |
5135 | #ifdef CONFIG_ZONE_DMA |
5136 | &cache_dma_attr.attr, | |
5137 | #endif | |
5138 | #ifdef CONFIG_NUMA | |
9824601e | 5139 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5140 | #endif |
5141 | #ifdef CONFIG_SLUB_STATS | |
5142 | &alloc_fastpath_attr.attr, | |
5143 | &alloc_slowpath_attr.attr, | |
5144 | &free_fastpath_attr.attr, | |
5145 | &free_slowpath_attr.attr, | |
5146 | &free_frozen_attr.attr, | |
5147 | &free_add_partial_attr.attr, | |
5148 | &free_remove_partial_attr.attr, | |
5149 | &alloc_from_partial_attr.attr, | |
5150 | &alloc_slab_attr.attr, | |
5151 | &alloc_refill_attr.attr, | |
e36a2652 | 5152 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5153 | &free_slab_attr.attr, |
5154 | &cpuslab_flush_attr.attr, | |
5155 | &deactivate_full_attr.attr, | |
5156 | &deactivate_empty_attr.attr, | |
5157 | &deactivate_to_head_attr.attr, | |
5158 | &deactivate_to_tail_attr.attr, | |
5159 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5160 | &deactivate_bypass_attr.attr, |
65c3376a | 5161 | &order_fallback_attr.attr, |
b789ef51 CL |
5162 | &cmpxchg_double_fail_attr.attr, |
5163 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5164 | &cpu_partial_alloc_attr.attr, |
5165 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5166 | &cpu_partial_node_attr.attr, |
5167 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5168 | #endif |
4c13dd3b DM |
5169 | #ifdef CONFIG_FAILSLAB |
5170 | &failslab_attr.attr, | |
5171 | #endif | |
5172 | ||
81819f0f CL |
5173 | NULL |
5174 | }; | |
5175 | ||
5176 | static struct attribute_group slab_attr_group = { | |
5177 | .attrs = slab_attrs, | |
5178 | }; | |
5179 | ||
5180 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5181 | struct attribute *attr, | |
5182 | char *buf) | |
5183 | { | |
5184 | struct slab_attribute *attribute; | |
5185 | struct kmem_cache *s; | |
5186 | int err; | |
5187 | ||
5188 | attribute = to_slab_attr(attr); | |
5189 | s = to_slab(kobj); | |
5190 | ||
5191 | if (!attribute->show) | |
5192 | return -EIO; | |
5193 | ||
5194 | err = attribute->show(s, buf); | |
5195 | ||
5196 | return err; | |
5197 | } | |
5198 | ||
5199 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5200 | struct attribute *attr, | |
5201 | const char *buf, size_t len) | |
5202 | { | |
5203 | struct slab_attribute *attribute; | |
5204 | struct kmem_cache *s; | |
5205 | int err; | |
5206 | ||
5207 | attribute = to_slab_attr(attr); | |
5208 | s = to_slab(kobj); | |
5209 | ||
5210 | if (!attribute->store) | |
5211 | return -EIO; | |
5212 | ||
5213 | err = attribute->store(s, buf, len); | |
5214 | ||
5215 | return err; | |
5216 | } | |
5217 | ||
151c602f CL |
5218 | static void kmem_cache_release(struct kobject *kobj) |
5219 | { | |
5220 | struct kmem_cache *s = to_slab(kobj); | |
5221 | ||
84c1cf62 | 5222 | kfree(s->name); |
151c602f CL |
5223 | kfree(s); |
5224 | } | |
5225 | ||
52cf25d0 | 5226 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5227 | .show = slab_attr_show, |
5228 | .store = slab_attr_store, | |
5229 | }; | |
5230 | ||
5231 | static struct kobj_type slab_ktype = { | |
5232 | .sysfs_ops = &slab_sysfs_ops, | |
151c602f | 5233 | .release = kmem_cache_release |
81819f0f CL |
5234 | }; |
5235 | ||
5236 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5237 | { | |
5238 | struct kobj_type *ktype = get_ktype(kobj); | |
5239 | ||
5240 | if (ktype == &slab_ktype) | |
5241 | return 1; | |
5242 | return 0; | |
5243 | } | |
5244 | ||
9cd43611 | 5245 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5246 | .filter = uevent_filter, |
5247 | }; | |
5248 | ||
27c3a314 | 5249 | static struct kset *slab_kset; |
81819f0f CL |
5250 | |
5251 | #define ID_STR_LENGTH 64 | |
5252 | ||
5253 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5254 | * |
5255 | * Format :[flags-]size | |
81819f0f CL |
5256 | */ |
5257 | static char *create_unique_id(struct kmem_cache *s) | |
5258 | { | |
5259 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5260 | char *p = name; | |
5261 | ||
5262 | BUG_ON(!name); | |
5263 | ||
5264 | *p++ = ':'; | |
5265 | /* | |
5266 | * First flags affecting slabcache operations. We will only | |
5267 | * get here for aliasable slabs so we do not need to support | |
5268 | * too many flags. The flags here must cover all flags that | |
5269 | * are matched during merging to guarantee that the id is | |
5270 | * unique. | |
5271 | */ | |
5272 | if (s->flags & SLAB_CACHE_DMA) | |
5273 | *p++ = 'd'; | |
5274 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5275 | *p++ = 'a'; | |
5276 | if (s->flags & SLAB_DEBUG_FREE) | |
5277 | *p++ = 'F'; | |
5a896d9e VN |
5278 | if (!(s->flags & SLAB_NOTRACK)) |
5279 | *p++ = 't'; | |
81819f0f CL |
5280 | if (p != name + 1) |
5281 | *p++ = '-'; | |
5282 | p += sprintf(p, "%07d", s->size); | |
5283 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
5284 | return name; | |
5285 | } | |
5286 | ||
5287 | static int sysfs_slab_add(struct kmem_cache *s) | |
5288 | { | |
5289 | int err; | |
5290 | const char *name; | |
5291 | int unmergeable; | |
5292 | ||
5293 | if (slab_state < SYSFS) | |
5294 | /* Defer until later */ | |
5295 | return 0; | |
5296 | ||
5297 | unmergeable = slab_unmergeable(s); | |
5298 | if (unmergeable) { | |
5299 | /* | |
5300 | * Slabcache can never be merged so we can use the name proper. | |
5301 | * This is typically the case for debug situations. In that | |
5302 | * case we can catch duplicate names easily. | |
5303 | */ | |
27c3a314 | 5304 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5305 | name = s->name; |
5306 | } else { | |
5307 | /* | |
5308 | * Create a unique name for the slab as a target | |
5309 | * for the symlinks. | |
5310 | */ | |
5311 | name = create_unique_id(s); | |
5312 | } | |
5313 | ||
27c3a314 | 5314 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5315 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5316 | if (err) { | |
5317 | kobject_put(&s->kobj); | |
81819f0f | 5318 | return err; |
1eada11c | 5319 | } |
81819f0f CL |
5320 | |
5321 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5322 | if (err) { |
5323 | kobject_del(&s->kobj); | |
5324 | kobject_put(&s->kobj); | |
81819f0f | 5325 | return err; |
5788d8ad | 5326 | } |
81819f0f CL |
5327 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5328 | if (!unmergeable) { | |
5329 | /* Setup first alias */ | |
5330 | sysfs_slab_alias(s, s->name); | |
5331 | kfree(name); | |
5332 | } | |
5333 | return 0; | |
5334 | } | |
5335 | ||
5336 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5337 | { | |
2bce6485 CL |
5338 | if (slab_state < SYSFS) |
5339 | /* | |
5340 | * Sysfs has not been setup yet so no need to remove the | |
5341 | * cache from sysfs. | |
5342 | */ | |
5343 | return; | |
5344 | ||
81819f0f CL |
5345 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5346 | kobject_del(&s->kobj); | |
151c602f | 5347 | kobject_put(&s->kobj); |
81819f0f CL |
5348 | } |
5349 | ||
5350 | /* | |
5351 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5352 | * available lest we lose that information. |
81819f0f CL |
5353 | */ |
5354 | struct saved_alias { | |
5355 | struct kmem_cache *s; | |
5356 | const char *name; | |
5357 | struct saved_alias *next; | |
5358 | }; | |
5359 | ||
5af328a5 | 5360 | static struct saved_alias *alias_list; |
81819f0f CL |
5361 | |
5362 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5363 | { | |
5364 | struct saved_alias *al; | |
5365 | ||
5366 | if (slab_state == SYSFS) { | |
5367 | /* | |
5368 | * If we have a leftover link then remove it. | |
5369 | */ | |
27c3a314 GKH |
5370 | sysfs_remove_link(&slab_kset->kobj, name); |
5371 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5372 | } |
5373 | ||
5374 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5375 | if (!al) | |
5376 | return -ENOMEM; | |
5377 | ||
5378 | al->s = s; | |
5379 | al->name = name; | |
5380 | al->next = alias_list; | |
5381 | alias_list = al; | |
5382 | return 0; | |
5383 | } | |
5384 | ||
5385 | static int __init slab_sysfs_init(void) | |
5386 | { | |
5b95a4ac | 5387 | struct kmem_cache *s; |
81819f0f CL |
5388 | int err; |
5389 | ||
2bce6485 CL |
5390 | down_write(&slub_lock); |
5391 | ||
0ff21e46 | 5392 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5393 | if (!slab_kset) { |
2bce6485 | 5394 | up_write(&slub_lock); |
81819f0f CL |
5395 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5396 | return -ENOSYS; | |
5397 | } | |
5398 | ||
26a7bd03 CL |
5399 | slab_state = SYSFS; |
5400 | ||
5b95a4ac | 5401 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5402 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5403 | if (err) |
5404 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5405 | " to sysfs\n", s->name); | |
26a7bd03 | 5406 | } |
81819f0f CL |
5407 | |
5408 | while (alias_list) { | |
5409 | struct saved_alias *al = alias_list; | |
5410 | ||
5411 | alias_list = alias_list->next; | |
5412 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5413 | if (err) |
5414 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
5415 | " %s to sysfs\n", s->name); | |
81819f0f CL |
5416 | kfree(al); |
5417 | } | |
5418 | ||
2bce6485 | 5419 | up_write(&slub_lock); |
81819f0f CL |
5420 | resiliency_test(); |
5421 | return 0; | |
5422 | } | |
5423 | ||
5424 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5425 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5426 | |
5427 | /* | |
5428 | * The /proc/slabinfo ABI | |
5429 | */ | |
158a9624 | 5430 | #ifdef CONFIG_SLABINFO |
57ed3eda PE |
5431 | static void print_slabinfo_header(struct seq_file *m) |
5432 | { | |
5433 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
5434 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
5435 | "<objperslab> <pagesperslab>"); | |
5436 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
5437 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
5438 | seq_putc(m, '\n'); | |
5439 | } | |
5440 | ||
5441 | static void *s_start(struct seq_file *m, loff_t *pos) | |
5442 | { | |
5443 | loff_t n = *pos; | |
5444 | ||
5445 | down_read(&slub_lock); | |
5446 | if (!n) | |
5447 | print_slabinfo_header(m); | |
5448 | ||
5449 | return seq_list_start(&slab_caches, *pos); | |
5450 | } | |
5451 | ||
5452 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
5453 | { | |
5454 | return seq_list_next(p, &slab_caches, pos); | |
5455 | } | |
5456 | ||
5457 | static void s_stop(struct seq_file *m, void *p) | |
5458 | { | |
5459 | up_read(&slub_lock); | |
5460 | } | |
5461 | ||
5462 | static int s_show(struct seq_file *m, void *p) | |
5463 | { | |
5464 | unsigned long nr_partials = 0; | |
5465 | unsigned long nr_slabs = 0; | |
5466 | unsigned long nr_inuse = 0; | |
205ab99d CL |
5467 | unsigned long nr_objs = 0; |
5468 | unsigned long nr_free = 0; | |
57ed3eda PE |
5469 | struct kmem_cache *s; |
5470 | int node; | |
5471 | ||
5472 | s = list_entry(p, struct kmem_cache, list); | |
5473 | ||
5474 | for_each_online_node(node) { | |
5475 | struct kmem_cache_node *n = get_node(s, node); | |
5476 | ||
5477 | if (!n) | |
5478 | continue; | |
5479 | ||
5480 | nr_partials += n->nr_partial; | |
5481 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5482 | nr_objs += atomic_long_read(&n->total_objects); |
5483 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5484 | } |
5485 | ||
205ab99d | 5486 | nr_inuse = nr_objs - nr_free; |
57ed3eda PE |
5487 | |
5488 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse, | |
834f3d11 CL |
5489 | nr_objs, s->size, oo_objects(s->oo), |
5490 | (1 << oo_order(s->oo))); | |
57ed3eda PE |
5491 | seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0); |
5492 | seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs, | |
5493 | 0UL); | |
5494 | seq_putc(m, '\n'); | |
5495 | return 0; | |
5496 | } | |
5497 | ||
7b3c3a50 | 5498 | static const struct seq_operations slabinfo_op = { |
57ed3eda PE |
5499 | .start = s_start, |
5500 | .next = s_next, | |
5501 | .stop = s_stop, | |
5502 | .show = s_show, | |
5503 | }; | |
5504 | ||
7b3c3a50 AD |
5505 | static int slabinfo_open(struct inode *inode, struct file *file) |
5506 | { | |
5507 | return seq_open(file, &slabinfo_op); | |
5508 | } | |
5509 | ||
5510 | static const struct file_operations proc_slabinfo_operations = { | |
5511 | .open = slabinfo_open, | |
5512 | .read = seq_read, | |
5513 | .llseek = seq_lseek, | |
5514 | .release = seq_release, | |
5515 | }; | |
5516 | ||
5517 | static int __init slab_proc_init(void) | |
5518 | { | |
ab067e99 | 5519 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); |
7b3c3a50 AD |
5520 | return 0; |
5521 | } | |
5522 | module_init(slab_proc_init); | |
158a9624 | 5523 | #endif /* CONFIG_SLABINFO */ |