]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
81819f0f | 21 | #include <linux/seq_file.h> |
5a896d9e | 22 | #include <linux/kmemcheck.h> |
81819f0f CL |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpuset.h> | |
25 | #include <linux/mempolicy.h> | |
26 | #include <linux/ctype.h> | |
3ac7fe5a | 27 | #include <linux/debugobjects.h> |
81819f0f | 28 | #include <linux/kallsyms.h> |
b9049e23 | 29 | #include <linux/memory.h> |
f8bd2258 | 30 | #include <linux/math64.h> |
773ff60e | 31 | #include <linux/fault-inject.h> |
bfa71457 | 32 | #include <linux/stacktrace.h> |
4de900b4 | 33 | #include <linux/prefetch.h> |
2633d7a0 | 34 | #include <linux/memcontrol.h> |
81819f0f | 35 | |
4a92379b RK |
36 | #include <trace/events/kmem.h> |
37 | ||
072bb0aa MG |
38 | #include "internal.h" |
39 | ||
81819f0f CL |
40 | /* |
41 | * Lock order: | |
18004c5d | 42 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
43 | * 2. node->list_lock |
44 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 45 | * |
18004c5d | 46 | * slab_mutex |
881db7fb | 47 | * |
18004c5d | 48 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
49 | * and to synchronize major metadata changes to slab cache structures. |
50 | * | |
51 | * The slab_lock is only used for debugging and on arches that do not | |
52 | * have the ability to do a cmpxchg_double. It only protects the second | |
53 | * double word in the page struct. Meaning | |
54 | * A. page->freelist -> List of object free in a page | |
55 | * B. page->counters -> Counters of objects | |
56 | * C. page->frozen -> frozen state | |
57 | * | |
58 | * If a slab is frozen then it is exempt from list management. It is not | |
59 | * on any list. The processor that froze the slab is the one who can | |
60 | * perform list operations on the page. Other processors may put objects | |
61 | * onto the freelist but the processor that froze the slab is the only | |
62 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
63 | * |
64 | * The list_lock protects the partial and full list on each node and | |
65 | * the partial slab counter. If taken then no new slabs may be added or | |
66 | * removed from the lists nor make the number of partial slabs be modified. | |
67 | * (Note that the total number of slabs is an atomic value that may be | |
68 | * modified without taking the list lock). | |
69 | * | |
70 | * The list_lock is a centralized lock and thus we avoid taking it as | |
71 | * much as possible. As long as SLUB does not have to handle partial | |
72 | * slabs, operations can continue without any centralized lock. F.e. | |
73 | * allocating a long series of objects that fill up slabs does not require | |
74 | * the list lock. | |
81819f0f CL |
75 | * Interrupts are disabled during allocation and deallocation in order to |
76 | * make the slab allocator safe to use in the context of an irq. In addition | |
77 | * interrupts are disabled to ensure that the processor does not change | |
78 | * while handling per_cpu slabs, due to kernel preemption. | |
79 | * | |
80 | * SLUB assigns one slab for allocation to each processor. | |
81 | * Allocations only occur from these slabs called cpu slabs. | |
82 | * | |
672bba3a CL |
83 | * Slabs with free elements are kept on a partial list and during regular |
84 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 85 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
86 | * We track full slabs for debugging purposes though because otherwise we |
87 | * cannot scan all objects. | |
81819f0f CL |
88 | * |
89 | * Slabs are freed when they become empty. Teardown and setup is | |
90 | * minimal so we rely on the page allocators per cpu caches for | |
91 | * fast frees and allocs. | |
92 | * | |
93 | * Overloading of page flags that are otherwise used for LRU management. | |
94 | * | |
4b6f0750 CL |
95 | * PageActive The slab is frozen and exempt from list processing. |
96 | * This means that the slab is dedicated to a purpose | |
97 | * such as satisfying allocations for a specific | |
98 | * processor. Objects may be freed in the slab while | |
99 | * it is frozen but slab_free will then skip the usual | |
100 | * list operations. It is up to the processor holding | |
101 | * the slab to integrate the slab into the slab lists | |
102 | * when the slab is no longer needed. | |
103 | * | |
104 | * One use of this flag is to mark slabs that are | |
105 | * used for allocations. Then such a slab becomes a cpu | |
106 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 107 | * freelist that allows lockless access to |
894b8788 CL |
108 | * free objects in addition to the regular freelist |
109 | * that requires the slab lock. | |
81819f0f CL |
110 | * |
111 | * PageError Slab requires special handling due to debug | |
112 | * options set. This moves slab handling out of | |
894b8788 | 113 | * the fast path and disables lockless freelists. |
81819f0f CL |
114 | */ |
115 | ||
af537b0a CL |
116 | static inline int kmem_cache_debug(struct kmem_cache *s) |
117 | { | |
5577bd8a | 118 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 119 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 120 | #else |
af537b0a | 121 | return 0; |
5577bd8a | 122 | #endif |
af537b0a | 123 | } |
5577bd8a | 124 | |
81819f0f CL |
125 | /* |
126 | * Issues still to be resolved: | |
127 | * | |
81819f0f CL |
128 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
129 | * | |
81819f0f CL |
130 | * - Variable sizing of the per node arrays |
131 | */ | |
132 | ||
133 | /* Enable to test recovery from slab corruption on boot */ | |
134 | #undef SLUB_RESILIENCY_TEST | |
135 | ||
b789ef51 CL |
136 | /* Enable to log cmpxchg failures */ |
137 | #undef SLUB_DEBUG_CMPXCHG | |
138 | ||
2086d26a CL |
139 | /* |
140 | * Mininum number of partial slabs. These will be left on the partial | |
141 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
142 | */ | |
76be8950 | 143 | #define MIN_PARTIAL 5 |
e95eed57 | 144 | |
2086d26a CL |
145 | /* |
146 | * Maximum number of desirable partial slabs. | |
147 | * The existence of more partial slabs makes kmem_cache_shrink | |
148 | * sort the partial list by the number of objects in the. | |
149 | */ | |
150 | #define MAX_PARTIAL 10 | |
151 | ||
81819f0f CL |
152 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
153 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 154 | |
fa5ec8a1 | 155 | /* |
3de47213 DR |
156 | * Debugging flags that require metadata to be stored in the slab. These get |
157 | * disabled when slub_debug=O is used and a cache's min order increases with | |
158 | * metadata. | |
fa5ec8a1 | 159 | */ |
3de47213 | 160 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 161 | |
81819f0f CL |
162 | /* |
163 | * Set of flags that will prevent slab merging | |
164 | */ | |
165 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
166 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
167 | SLAB_FAILSLAB) | |
81819f0f CL |
168 | |
169 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 170 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 171 | |
210b5c06 CG |
172 | #define OO_SHIFT 16 |
173 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 174 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 175 | |
81819f0f | 176 | /* Internal SLUB flags */ |
f90ec390 | 177 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 178 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 179 | |
81819f0f CL |
180 | #ifdef CONFIG_SMP |
181 | static struct notifier_block slab_notifier; | |
182 | #endif | |
183 | ||
02cbc874 CL |
184 | /* |
185 | * Tracking user of a slab. | |
186 | */ | |
d6543e39 | 187 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 188 | struct track { |
ce71e27c | 189 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
190 | #ifdef CONFIG_STACKTRACE |
191 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
192 | #endif | |
02cbc874 CL |
193 | int cpu; /* Was running on cpu */ |
194 | int pid; /* Pid context */ | |
195 | unsigned long when; /* When did the operation occur */ | |
196 | }; | |
197 | ||
198 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
199 | ||
ab4d5ed5 | 200 | #ifdef CONFIG_SYSFS |
81819f0f CL |
201 | static int sysfs_slab_add(struct kmem_cache *); |
202 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
203 | static void sysfs_slab_remove(struct kmem_cache *); | |
8ff12cfc | 204 | |
81819f0f | 205 | #else |
0c710013 CL |
206 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
207 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
208 | { return 0; } | |
db265eca | 209 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
8ff12cfc | 210 | |
81819f0f CL |
211 | #endif |
212 | ||
4fdccdfb | 213 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
214 | { |
215 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 216 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
217 | #endif |
218 | } | |
219 | ||
81819f0f CL |
220 | /******************************************************************** |
221 | * Core slab cache functions | |
222 | *******************************************************************/ | |
223 | ||
81819f0f CL |
224 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
225 | { | |
81819f0f | 226 | return s->node[node]; |
81819f0f CL |
227 | } |
228 | ||
6446faa2 | 229 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
230 | static inline int check_valid_pointer(struct kmem_cache *s, |
231 | struct page *page, const void *object) | |
232 | { | |
233 | void *base; | |
234 | ||
a973e9dd | 235 | if (!object) |
02cbc874 CL |
236 | return 1; |
237 | ||
a973e9dd | 238 | base = page_address(page); |
39b26464 | 239 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
240 | (object - base) % s->size) { |
241 | return 0; | |
242 | } | |
243 | ||
244 | return 1; | |
245 | } | |
246 | ||
7656c72b CL |
247 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
248 | { | |
249 | return *(void **)(object + s->offset); | |
250 | } | |
251 | ||
0ad9500e ED |
252 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
253 | { | |
254 | prefetch(object + s->offset); | |
255 | } | |
256 | ||
1393d9a1 CL |
257 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
258 | { | |
259 | void *p; | |
260 | ||
261 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
262 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
263 | #else | |
264 | p = get_freepointer(s, object); | |
265 | #endif | |
266 | return p; | |
267 | } | |
268 | ||
7656c72b CL |
269 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
270 | { | |
271 | *(void **)(object + s->offset) = fp; | |
272 | } | |
273 | ||
274 | /* Loop over all objects in a slab */ | |
224a88be CL |
275 | #define for_each_object(__p, __s, __addr, __objects) \ |
276 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
277 | __p += (__s)->size) |
278 | ||
7656c72b CL |
279 | /* Determine object index from a given position */ |
280 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
281 | { | |
282 | return (p - addr) / s->size; | |
283 | } | |
284 | ||
d71f606f MK |
285 | static inline size_t slab_ksize(const struct kmem_cache *s) |
286 | { | |
287 | #ifdef CONFIG_SLUB_DEBUG | |
288 | /* | |
289 | * Debugging requires use of the padding between object | |
290 | * and whatever may come after it. | |
291 | */ | |
292 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 293 | return s->object_size; |
d71f606f MK |
294 | |
295 | #endif | |
296 | /* | |
297 | * If we have the need to store the freelist pointer | |
298 | * back there or track user information then we can | |
299 | * only use the space before that information. | |
300 | */ | |
301 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
302 | return s->inuse; | |
303 | /* | |
304 | * Else we can use all the padding etc for the allocation | |
305 | */ | |
306 | return s->size; | |
307 | } | |
308 | ||
ab9a0f19 LJ |
309 | static inline int order_objects(int order, unsigned long size, int reserved) |
310 | { | |
311 | return ((PAGE_SIZE << order) - reserved) / size; | |
312 | } | |
313 | ||
834f3d11 | 314 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 315 | unsigned long size, int reserved) |
834f3d11 CL |
316 | { |
317 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 318 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
319 | }; |
320 | ||
321 | return x; | |
322 | } | |
323 | ||
324 | static inline int oo_order(struct kmem_cache_order_objects x) | |
325 | { | |
210b5c06 | 326 | return x.x >> OO_SHIFT; |
834f3d11 CL |
327 | } |
328 | ||
329 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
330 | { | |
210b5c06 | 331 | return x.x & OO_MASK; |
834f3d11 CL |
332 | } |
333 | ||
881db7fb CL |
334 | /* |
335 | * Per slab locking using the pagelock | |
336 | */ | |
337 | static __always_inline void slab_lock(struct page *page) | |
338 | { | |
339 | bit_spin_lock(PG_locked, &page->flags); | |
340 | } | |
341 | ||
342 | static __always_inline void slab_unlock(struct page *page) | |
343 | { | |
344 | __bit_spin_unlock(PG_locked, &page->flags); | |
345 | } | |
346 | ||
1d07171c CL |
347 | /* Interrupts must be disabled (for the fallback code to work right) */ |
348 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
349 | void *freelist_old, unsigned long counters_old, | |
350 | void *freelist_new, unsigned long counters_new, | |
351 | const char *n) | |
352 | { | |
353 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
354 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
355 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 356 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 357 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
358 | freelist_old, counters_old, |
359 | freelist_new, counters_new)) | |
360 | return 1; | |
361 | } else | |
362 | #endif | |
363 | { | |
364 | slab_lock(page); | |
365 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
366 | page->freelist = freelist_new; | |
367 | page->counters = counters_new; | |
368 | slab_unlock(page); | |
369 | return 1; | |
370 | } | |
371 | slab_unlock(page); | |
372 | } | |
373 | ||
374 | cpu_relax(); | |
375 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
376 | ||
377 | #ifdef SLUB_DEBUG_CMPXCHG | |
378 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
379 | #endif | |
380 | ||
381 | return 0; | |
382 | } | |
383 | ||
b789ef51 CL |
384 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
385 | void *freelist_old, unsigned long counters_old, | |
386 | void *freelist_new, unsigned long counters_new, | |
387 | const char *n) | |
388 | { | |
2565409f HC |
389 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
390 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 391 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 392 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
393 | freelist_old, counters_old, |
394 | freelist_new, counters_new)) | |
395 | return 1; | |
396 | } else | |
397 | #endif | |
398 | { | |
1d07171c CL |
399 | unsigned long flags; |
400 | ||
401 | local_irq_save(flags); | |
881db7fb | 402 | slab_lock(page); |
b789ef51 CL |
403 | if (page->freelist == freelist_old && page->counters == counters_old) { |
404 | page->freelist = freelist_new; | |
405 | page->counters = counters_new; | |
881db7fb | 406 | slab_unlock(page); |
1d07171c | 407 | local_irq_restore(flags); |
b789ef51 CL |
408 | return 1; |
409 | } | |
881db7fb | 410 | slab_unlock(page); |
1d07171c | 411 | local_irq_restore(flags); |
b789ef51 CL |
412 | } |
413 | ||
414 | cpu_relax(); | |
415 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
416 | ||
417 | #ifdef SLUB_DEBUG_CMPXCHG | |
418 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
419 | #endif | |
420 | ||
421 | return 0; | |
422 | } | |
423 | ||
41ecc55b | 424 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
425 | /* |
426 | * Determine a map of object in use on a page. | |
427 | * | |
881db7fb | 428 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
429 | * not vanish from under us. |
430 | */ | |
431 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
432 | { | |
433 | void *p; | |
434 | void *addr = page_address(page); | |
435 | ||
436 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
437 | set_bit(slab_index(p, s, addr), map); | |
438 | } | |
439 | ||
41ecc55b CL |
440 | /* |
441 | * Debug settings: | |
442 | */ | |
f0630fff CL |
443 | #ifdef CONFIG_SLUB_DEBUG_ON |
444 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
445 | #else | |
41ecc55b | 446 | static int slub_debug; |
f0630fff | 447 | #endif |
41ecc55b CL |
448 | |
449 | static char *slub_debug_slabs; | |
fa5ec8a1 | 450 | static int disable_higher_order_debug; |
41ecc55b | 451 | |
81819f0f CL |
452 | /* |
453 | * Object debugging | |
454 | */ | |
455 | static void print_section(char *text, u8 *addr, unsigned int length) | |
456 | { | |
ffc79d28 SAS |
457 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
458 | length, 1); | |
81819f0f CL |
459 | } |
460 | ||
81819f0f CL |
461 | static struct track *get_track(struct kmem_cache *s, void *object, |
462 | enum track_item alloc) | |
463 | { | |
464 | struct track *p; | |
465 | ||
466 | if (s->offset) | |
467 | p = object + s->offset + sizeof(void *); | |
468 | else | |
469 | p = object + s->inuse; | |
470 | ||
471 | return p + alloc; | |
472 | } | |
473 | ||
474 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 475 | enum track_item alloc, unsigned long addr) |
81819f0f | 476 | { |
1a00df4a | 477 | struct track *p = get_track(s, object, alloc); |
81819f0f | 478 | |
81819f0f | 479 | if (addr) { |
d6543e39 BG |
480 | #ifdef CONFIG_STACKTRACE |
481 | struct stack_trace trace; | |
482 | int i; | |
483 | ||
484 | trace.nr_entries = 0; | |
485 | trace.max_entries = TRACK_ADDRS_COUNT; | |
486 | trace.entries = p->addrs; | |
487 | trace.skip = 3; | |
488 | save_stack_trace(&trace); | |
489 | ||
490 | /* See rant in lockdep.c */ | |
491 | if (trace.nr_entries != 0 && | |
492 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
493 | trace.nr_entries--; | |
494 | ||
495 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
496 | p->addrs[i] = 0; | |
497 | #endif | |
81819f0f CL |
498 | p->addr = addr; |
499 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 500 | p->pid = current->pid; |
81819f0f CL |
501 | p->when = jiffies; |
502 | } else | |
503 | memset(p, 0, sizeof(struct track)); | |
504 | } | |
505 | ||
81819f0f CL |
506 | static void init_tracking(struct kmem_cache *s, void *object) |
507 | { | |
24922684 CL |
508 | if (!(s->flags & SLAB_STORE_USER)) |
509 | return; | |
510 | ||
ce71e27c EGM |
511 | set_track(s, object, TRACK_FREE, 0UL); |
512 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
513 | } |
514 | ||
515 | static void print_track(const char *s, struct track *t) | |
516 | { | |
517 | if (!t->addr) | |
518 | return; | |
519 | ||
7daf705f | 520 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 521 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
522 | #ifdef CONFIG_STACKTRACE |
523 | { | |
524 | int i; | |
525 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
526 | if (t->addrs[i]) | |
527 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
528 | else | |
529 | break; | |
530 | } | |
531 | #endif | |
24922684 CL |
532 | } |
533 | ||
534 | static void print_tracking(struct kmem_cache *s, void *object) | |
535 | { | |
536 | if (!(s->flags & SLAB_STORE_USER)) | |
537 | return; | |
538 | ||
539 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
540 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
541 | } | |
542 | ||
543 | static void print_page_info(struct page *page) | |
544 | { | |
39b26464 CL |
545 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
546 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
547 | |
548 | } | |
549 | ||
550 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
551 | { | |
552 | va_list args; | |
553 | char buf[100]; | |
554 | ||
555 | va_start(args, fmt); | |
556 | vsnprintf(buf, sizeof(buf), fmt, args); | |
557 | va_end(args); | |
558 | printk(KERN_ERR "========================================" | |
559 | "=====================================\n"); | |
265d47e7 | 560 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
561 | printk(KERN_ERR "----------------------------------------" |
562 | "-------------------------------------\n\n"); | |
645df230 DJ |
563 | |
564 | add_taint(TAINT_BAD_PAGE); | |
81819f0f CL |
565 | } |
566 | ||
24922684 CL |
567 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
568 | { | |
569 | va_list args; | |
570 | char buf[100]; | |
571 | ||
572 | va_start(args, fmt); | |
573 | vsnprintf(buf, sizeof(buf), fmt, args); | |
574 | va_end(args); | |
575 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
576 | } | |
577 | ||
578 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
579 | { |
580 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 581 | u8 *addr = page_address(page); |
24922684 CL |
582 | |
583 | print_tracking(s, p); | |
584 | ||
585 | print_page_info(page); | |
586 | ||
587 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
588 | p, p - addr, get_freepointer(s, p)); | |
589 | ||
590 | if (p > addr + 16) | |
ffc79d28 | 591 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 592 | |
3b0efdfa | 593 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 594 | PAGE_SIZE)); |
81819f0f | 595 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
596 | print_section("Redzone ", p + s->object_size, |
597 | s->inuse - s->object_size); | |
81819f0f | 598 | |
81819f0f CL |
599 | if (s->offset) |
600 | off = s->offset + sizeof(void *); | |
601 | else | |
602 | off = s->inuse; | |
603 | ||
24922684 | 604 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 605 | off += 2 * sizeof(struct track); |
81819f0f CL |
606 | |
607 | if (off != s->size) | |
608 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 609 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
610 | |
611 | dump_stack(); | |
81819f0f CL |
612 | } |
613 | ||
614 | static void object_err(struct kmem_cache *s, struct page *page, | |
615 | u8 *object, char *reason) | |
616 | { | |
3dc50637 | 617 | slab_bug(s, "%s", reason); |
24922684 | 618 | print_trailer(s, page, object); |
81819f0f CL |
619 | } |
620 | ||
945cf2b6 | 621 | static void slab_err(struct kmem_cache *s, struct page *page, const char *fmt, ...) |
81819f0f CL |
622 | { |
623 | va_list args; | |
624 | char buf[100]; | |
625 | ||
24922684 CL |
626 | va_start(args, fmt); |
627 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 628 | va_end(args); |
3dc50637 | 629 | slab_bug(s, "%s", buf); |
24922684 | 630 | print_page_info(page); |
81819f0f CL |
631 | dump_stack(); |
632 | } | |
633 | ||
f7cb1933 | 634 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
635 | { |
636 | u8 *p = object; | |
637 | ||
638 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
639 | memset(p, POISON_FREE, s->object_size - 1); |
640 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
641 | } |
642 | ||
643 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 644 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
645 | } |
646 | ||
24922684 CL |
647 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
648 | void *from, void *to) | |
649 | { | |
650 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
651 | memset(from, data, to - from); | |
652 | } | |
653 | ||
654 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
655 | u8 *object, char *what, | |
06428780 | 656 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
657 | { |
658 | u8 *fault; | |
659 | u8 *end; | |
660 | ||
79824820 | 661 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
662 | if (!fault) |
663 | return 1; | |
664 | ||
665 | end = start + bytes; | |
666 | while (end > fault && end[-1] == value) | |
667 | end--; | |
668 | ||
669 | slab_bug(s, "%s overwritten", what); | |
670 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
671 | fault, end - 1, fault[0], value); | |
672 | print_trailer(s, page, object); | |
673 | ||
674 | restore_bytes(s, what, value, fault, end); | |
675 | return 0; | |
81819f0f CL |
676 | } |
677 | ||
81819f0f CL |
678 | /* |
679 | * Object layout: | |
680 | * | |
681 | * object address | |
682 | * Bytes of the object to be managed. | |
683 | * If the freepointer may overlay the object then the free | |
684 | * pointer is the first word of the object. | |
672bba3a | 685 | * |
81819f0f CL |
686 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
687 | * 0xa5 (POISON_END) | |
688 | * | |
3b0efdfa | 689 | * object + s->object_size |
81819f0f | 690 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 691 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 692 | * object_size == inuse. |
672bba3a | 693 | * |
81819f0f CL |
694 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
695 | * 0xcc (RED_ACTIVE) for objects in use. | |
696 | * | |
697 | * object + s->inuse | |
672bba3a CL |
698 | * Meta data starts here. |
699 | * | |
81819f0f CL |
700 | * A. Free pointer (if we cannot overwrite object on free) |
701 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 702 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 703 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
704 | * before the word boundary. |
705 | * | |
706 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
707 | * |
708 | * object + s->size | |
672bba3a | 709 | * Nothing is used beyond s->size. |
81819f0f | 710 | * |
3b0efdfa | 711 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 712 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
713 | * may be used with merged slabcaches. |
714 | */ | |
715 | ||
81819f0f CL |
716 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
717 | { | |
718 | unsigned long off = s->inuse; /* The end of info */ | |
719 | ||
720 | if (s->offset) | |
721 | /* Freepointer is placed after the object. */ | |
722 | off += sizeof(void *); | |
723 | ||
724 | if (s->flags & SLAB_STORE_USER) | |
725 | /* We also have user information there */ | |
726 | off += 2 * sizeof(struct track); | |
727 | ||
728 | if (s->size == off) | |
729 | return 1; | |
730 | ||
24922684 CL |
731 | return check_bytes_and_report(s, page, p, "Object padding", |
732 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
733 | } |
734 | ||
39b26464 | 735 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
736 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
737 | { | |
24922684 CL |
738 | u8 *start; |
739 | u8 *fault; | |
740 | u8 *end; | |
741 | int length; | |
742 | int remainder; | |
81819f0f CL |
743 | |
744 | if (!(s->flags & SLAB_POISON)) | |
745 | return 1; | |
746 | ||
a973e9dd | 747 | start = page_address(page); |
ab9a0f19 | 748 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
749 | end = start + length; |
750 | remainder = length % s->size; | |
81819f0f CL |
751 | if (!remainder) |
752 | return 1; | |
753 | ||
79824820 | 754 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
755 | if (!fault) |
756 | return 1; | |
757 | while (end > fault && end[-1] == POISON_INUSE) | |
758 | end--; | |
759 | ||
760 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 761 | print_section("Padding ", end - remainder, remainder); |
24922684 | 762 | |
8a3d271d | 763 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 764 | return 0; |
81819f0f CL |
765 | } |
766 | ||
767 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 768 | void *object, u8 val) |
81819f0f CL |
769 | { |
770 | u8 *p = object; | |
3b0efdfa | 771 | u8 *endobject = object + s->object_size; |
81819f0f CL |
772 | |
773 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 774 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 775 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 776 | return 0; |
81819f0f | 777 | } else { |
3b0efdfa | 778 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 779 | check_bytes_and_report(s, page, p, "Alignment padding", |
3b0efdfa | 780 | endobject, POISON_INUSE, s->inuse - s->object_size); |
3adbefee | 781 | } |
81819f0f CL |
782 | } |
783 | ||
784 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 785 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 786 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 787 | POISON_FREE, s->object_size - 1) || |
24922684 | 788 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 789 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 790 | return 0; |
81819f0f CL |
791 | /* |
792 | * check_pad_bytes cleans up on its own. | |
793 | */ | |
794 | check_pad_bytes(s, page, p); | |
795 | } | |
796 | ||
f7cb1933 | 797 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
798 | /* |
799 | * Object and freepointer overlap. Cannot check | |
800 | * freepointer while object is allocated. | |
801 | */ | |
802 | return 1; | |
803 | ||
804 | /* Check free pointer validity */ | |
805 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
806 | object_err(s, page, p, "Freepointer corrupt"); | |
807 | /* | |
9f6c708e | 808 | * No choice but to zap it and thus lose the remainder |
81819f0f | 809 | * of the free objects in this slab. May cause |
672bba3a | 810 | * another error because the object count is now wrong. |
81819f0f | 811 | */ |
a973e9dd | 812 | set_freepointer(s, p, NULL); |
81819f0f CL |
813 | return 0; |
814 | } | |
815 | return 1; | |
816 | } | |
817 | ||
818 | static int check_slab(struct kmem_cache *s, struct page *page) | |
819 | { | |
39b26464 CL |
820 | int maxobj; |
821 | ||
81819f0f CL |
822 | VM_BUG_ON(!irqs_disabled()); |
823 | ||
824 | if (!PageSlab(page)) { | |
24922684 | 825 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
826 | return 0; |
827 | } | |
39b26464 | 828 | |
ab9a0f19 | 829 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
830 | if (page->objects > maxobj) { |
831 | slab_err(s, page, "objects %u > max %u", | |
832 | s->name, page->objects, maxobj); | |
833 | return 0; | |
834 | } | |
835 | if (page->inuse > page->objects) { | |
24922684 | 836 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 837 | s->name, page->inuse, page->objects); |
81819f0f CL |
838 | return 0; |
839 | } | |
840 | /* Slab_pad_check fixes things up after itself */ | |
841 | slab_pad_check(s, page); | |
842 | return 1; | |
843 | } | |
844 | ||
845 | /* | |
672bba3a CL |
846 | * Determine if a certain object on a page is on the freelist. Must hold the |
847 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
848 | */ |
849 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
850 | { | |
851 | int nr = 0; | |
881db7fb | 852 | void *fp; |
81819f0f | 853 | void *object = NULL; |
224a88be | 854 | unsigned long max_objects; |
81819f0f | 855 | |
881db7fb | 856 | fp = page->freelist; |
39b26464 | 857 | while (fp && nr <= page->objects) { |
81819f0f CL |
858 | if (fp == search) |
859 | return 1; | |
860 | if (!check_valid_pointer(s, page, fp)) { | |
861 | if (object) { | |
862 | object_err(s, page, object, | |
863 | "Freechain corrupt"); | |
a973e9dd | 864 | set_freepointer(s, object, NULL); |
81819f0f CL |
865 | break; |
866 | } else { | |
24922684 | 867 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 868 | page->freelist = NULL; |
39b26464 | 869 | page->inuse = page->objects; |
24922684 | 870 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
871 | return 0; |
872 | } | |
873 | break; | |
874 | } | |
875 | object = fp; | |
876 | fp = get_freepointer(s, object); | |
877 | nr++; | |
878 | } | |
879 | ||
ab9a0f19 | 880 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
881 | if (max_objects > MAX_OBJS_PER_PAGE) |
882 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
883 | |
884 | if (page->objects != max_objects) { | |
885 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
886 | "should be %d", page->objects, max_objects); | |
887 | page->objects = max_objects; | |
888 | slab_fix(s, "Number of objects adjusted."); | |
889 | } | |
39b26464 | 890 | if (page->inuse != page->objects - nr) { |
70d71228 | 891 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
892 | "counted were %d", page->inuse, page->objects - nr); |
893 | page->inuse = page->objects - nr; | |
24922684 | 894 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
895 | } |
896 | return search == NULL; | |
897 | } | |
898 | ||
0121c619 CL |
899 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
900 | int alloc) | |
3ec09742 CL |
901 | { |
902 | if (s->flags & SLAB_TRACE) { | |
903 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
904 | s->name, | |
905 | alloc ? "alloc" : "free", | |
906 | object, page->inuse, | |
907 | page->freelist); | |
908 | ||
909 | if (!alloc) | |
3b0efdfa | 910 | print_section("Object ", (void *)object, s->object_size); |
3ec09742 CL |
911 | |
912 | dump_stack(); | |
913 | } | |
914 | } | |
915 | ||
c016b0bd CL |
916 | /* |
917 | * Hooks for other subsystems that check memory allocations. In a typical | |
918 | * production configuration these hooks all should produce no code at all. | |
919 | */ | |
920 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
921 | { | |
c1d50836 | 922 | flags &= gfp_allowed_mask; |
c016b0bd CL |
923 | lockdep_trace_alloc(flags); |
924 | might_sleep_if(flags & __GFP_WAIT); | |
925 | ||
3b0efdfa | 926 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
927 | } |
928 | ||
929 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
930 | { | |
c1d50836 | 931 | flags &= gfp_allowed_mask; |
b3d41885 | 932 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 933 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
934 | } |
935 | ||
936 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
937 | { | |
938 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 939 | |
d3f661d6 CL |
940 | /* |
941 | * Trouble is that we may no longer disable interupts in the fast path | |
942 | * So in order to make the debug calls that expect irqs to be | |
943 | * disabled we need to disable interrupts temporarily. | |
944 | */ | |
945 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
946 | { | |
947 | unsigned long flags; | |
948 | ||
949 | local_irq_save(flags); | |
3b0efdfa CL |
950 | kmemcheck_slab_free(s, x, s->object_size); |
951 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
952 | local_irq_restore(flags); |
953 | } | |
954 | #endif | |
f9b615de | 955 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 956 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
957 | } |
958 | ||
643b1138 | 959 | /* |
672bba3a | 960 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
961 | * |
962 | * list_lock must be held. | |
643b1138 | 963 | */ |
5cc6eee8 CL |
964 | static void add_full(struct kmem_cache *s, |
965 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 966 | { |
5cc6eee8 CL |
967 | if (!(s->flags & SLAB_STORE_USER)) |
968 | return; | |
969 | ||
643b1138 | 970 | list_add(&page->lru, &n->full); |
643b1138 CL |
971 | } |
972 | ||
5cc6eee8 CL |
973 | /* |
974 | * list_lock must be held. | |
975 | */ | |
643b1138 CL |
976 | static void remove_full(struct kmem_cache *s, struct page *page) |
977 | { | |
643b1138 CL |
978 | if (!(s->flags & SLAB_STORE_USER)) |
979 | return; | |
980 | ||
643b1138 | 981 | list_del(&page->lru); |
643b1138 CL |
982 | } |
983 | ||
0f389ec6 CL |
984 | /* Tracking of the number of slabs for debugging purposes */ |
985 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
986 | { | |
987 | struct kmem_cache_node *n = get_node(s, node); | |
988 | ||
989 | return atomic_long_read(&n->nr_slabs); | |
990 | } | |
991 | ||
26c02cf0 AB |
992 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
993 | { | |
994 | return atomic_long_read(&n->nr_slabs); | |
995 | } | |
996 | ||
205ab99d | 997 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
998 | { |
999 | struct kmem_cache_node *n = get_node(s, node); | |
1000 | ||
1001 | /* | |
1002 | * May be called early in order to allocate a slab for the | |
1003 | * kmem_cache_node structure. Solve the chicken-egg | |
1004 | * dilemma by deferring the increment of the count during | |
1005 | * bootstrap (see early_kmem_cache_node_alloc). | |
1006 | */ | |
7340cc84 | 1007 | if (n) { |
0f389ec6 | 1008 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1009 | atomic_long_add(objects, &n->total_objects); |
1010 | } | |
0f389ec6 | 1011 | } |
205ab99d | 1012 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1013 | { |
1014 | struct kmem_cache_node *n = get_node(s, node); | |
1015 | ||
1016 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1017 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1018 | } |
1019 | ||
1020 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1021 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1022 | void *object) | |
1023 | { | |
1024 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1025 | return; | |
1026 | ||
f7cb1933 | 1027 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1028 | init_tracking(s, object); |
1029 | } | |
1030 | ||
1537066c | 1031 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1032 | void *object, unsigned long addr) |
81819f0f CL |
1033 | { |
1034 | if (!check_slab(s, page)) | |
1035 | goto bad; | |
1036 | ||
81819f0f CL |
1037 | if (!check_valid_pointer(s, page, object)) { |
1038 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1039 | goto bad; |
81819f0f CL |
1040 | } |
1041 | ||
f7cb1933 | 1042 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1043 | goto bad; |
81819f0f | 1044 | |
3ec09742 CL |
1045 | /* Success perform special debug activities for allocs */ |
1046 | if (s->flags & SLAB_STORE_USER) | |
1047 | set_track(s, object, TRACK_ALLOC, addr); | |
1048 | trace(s, page, object, 1); | |
f7cb1933 | 1049 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1050 | return 1; |
3ec09742 | 1051 | |
81819f0f CL |
1052 | bad: |
1053 | if (PageSlab(page)) { | |
1054 | /* | |
1055 | * If this is a slab page then lets do the best we can | |
1056 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1057 | * as used avoids touching the remaining objects. |
81819f0f | 1058 | */ |
24922684 | 1059 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1060 | page->inuse = page->objects; |
a973e9dd | 1061 | page->freelist = NULL; |
81819f0f CL |
1062 | } |
1063 | return 0; | |
1064 | } | |
1065 | ||
19c7ff9e CL |
1066 | static noinline struct kmem_cache_node *free_debug_processing( |
1067 | struct kmem_cache *s, struct page *page, void *object, | |
1068 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1069 | { |
19c7ff9e | 1070 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1071 | |
19c7ff9e | 1072 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1073 | slab_lock(page); |
1074 | ||
81819f0f CL |
1075 | if (!check_slab(s, page)) |
1076 | goto fail; | |
1077 | ||
1078 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1079 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1080 | goto fail; |
1081 | } | |
1082 | ||
1083 | if (on_freelist(s, page, object)) { | |
24922684 | 1084 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1085 | goto fail; |
1086 | } | |
1087 | ||
f7cb1933 | 1088 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1089 | goto out; |
81819f0f | 1090 | |
1b4f59e3 | 1091 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1092 | if (!PageSlab(page)) { |
70d71228 CL |
1093 | slab_err(s, page, "Attempt to free object(0x%p) " |
1094 | "outside of slab", object); | |
1b4f59e3 | 1095 | } else if (!page->slab_cache) { |
81819f0f | 1096 | printk(KERN_ERR |
70d71228 | 1097 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1098 | object); |
70d71228 | 1099 | dump_stack(); |
06428780 | 1100 | } else |
24922684 CL |
1101 | object_err(s, page, object, |
1102 | "page slab pointer corrupt."); | |
81819f0f CL |
1103 | goto fail; |
1104 | } | |
3ec09742 | 1105 | |
3ec09742 CL |
1106 | if (s->flags & SLAB_STORE_USER) |
1107 | set_track(s, object, TRACK_FREE, addr); | |
1108 | trace(s, page, object, 0); | |
f7cb1933 | 1109 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1110 | out: |
881db7fb | 1111 | slab_unlock(page); |
19c7ff9e CL |
1112 | /* |
1113 | * Keep node_lock to preserve integrity | |
1114 | * until the object is actually freed | |
1115 | */ | |
1116 | return n; | |
3ec09742 | 1117 | |
81819f0f | 1118 | fail: |
19c7ff9e CL |
1119 | slab_unlock(page); |
1120 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1121 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1122 | return NULL; |
81819f0f CL |
1123 | } |
1124 | ||
41ecc55b CL |
1125 | static int __init setup_slub_debug(char *str) |
1126 | { | |
f0630fff CL |
1127 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1128 | if (*str++ != '=' || !*str) | |
1129 | /* | |
1130 | * No options specified. Switch on full debugging. | |
1131 | */ | |
1132 | goto out; | |
1133 | ||
1134 | if (*str == ',') | |
1135 | /* | |
1136 | * No options but restriction on slabs. This means full | |
1137 | * debugging for slabs matching a pattern. | |
1138 | */ | |
1139 | goto check_slabs; | |
1140 | ||
fa5ec8a1 DR |
1141 | if (tolower(*str) == 'o') { |
1142 | /* | |
1143 | * Avoid enabling debugging on caches if its minimum order | |
1144 | * would increase as a result. | |
1145 | */ | |
1146 | disable_higher_order_debug = 1; | |
1147 | goto out; | |
1148 | } | |
1149 | ||
f0630fff CL |
1150 | slub_debug = 0; |
1151 | if (*str == '-') | |
1152 | /* | |
1153 | * Switch off all debugging measures. | |
1154 | */ | |
1155 | goto out; | |
1156 | ||
1157 | /* | |
1158 | * Determine which debug features should be switched on | |
1159 | */ | |
06428780 | 1160 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1161 | switch (tolower(*str)) { |
1162 | case 'f': | |
1163 | slub_debug |= SLAB_DEBUG_FREE; | |
1164 | break; | |
1165 | case 'z': | |
1166 | slub_debug |= SLAB_RED_ZONE; | |
1167 | break; | |
1168 | case 'p': | |
1169 | slub_debug |= SLAB_POISON; | |
1170 | break; | |
1171 | case 'u': | |
1172 | slub_debug |= SLAB_STORE_USER; | |
1173 | break; | |
1174 | case 't': | |
1175 | slub_debug |= SLAB_TRACE; | |
1176 | break; | |
4c13dd3b DM |
1177 | case 'a': |
1178 | slub_debug |= SLAB_FAILSLAB; | |
1179 | break; | |
f0630fff CL |
1180 | default: |
1181 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1182 | "unknown. skipped\n", *str); |
f0630fff | 1183 | } |
41ecc55b CL |
1184 | } |
1185 | ||
f0630fff | 1186 | check_slabs: |
41ecc55b CL |
1187 | if (*str == ',') |
1188 | slub_debug_slabs = str + 1; | |
f0630fff | 1189 | out: |
41ecc55b CL |
1190 | return 1; |
1191 | } | |
1192 | ||
1193 | __setup("slub_debug", setup_slub_debug); | |
1194 | ||
3b0efdfa | 1195 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1196 | unsigned long flags, const char *name, |
51cc5068 | 1197 | void (*ctor)(void *)) |
41ecc55b CL |
1198 | { |
1199 | /* | |
e153362a | 1200 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1201 | */ |
e153362a | 1202 | if (slub_debug && (!slub_debug_slabs || |
3de47213 DR |
1203 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))) |
1204 | flags |= slub_debug; | |
ba0268a8 CL |
1205 | |
1206 | return flags; | |
41ecc55b CL |
1207 | } |
1208 | #else | |
3ec09742 CL |
1209 | static inline void setup_object_debug(struct kmem_cache *s, |
1210 | struct page *page, void *object) {} | |
41ecc55b | 1211 | |
3ec09742 | 1212 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1213 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1214 | |
19c7ff9e CL |
1215 | static inline struct kmem_cache_node *free_debug_processing( |
1216 | struct kmem_cache *s, struct page *page, void *object, | |
1217 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1218 | |
41ecc55b CL |
1219 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1220 | { return 1; } | |
1221 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1222 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1223 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1224 | struct page *page) {} | |
2cfb7455 | 1225 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
3b0efdfa | 1226 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1227 | unsigned long flags, const char *name, |
51cc5068 | 1228 | void (*ctor)(void *)) |
ba0268a8 CL |
1229 | { |
1230 | return flags; | |
1231 | } | |
41ecc55b | 1232 | #define slub_debug 0 |
0f389ec6 | 1233 | |
fdaa45e9 IM |
1234 | #define disable_higher_order_debug 0 |
1235 | ||
0f389ec6 CL |
1236 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1237 | { return 0; } | |
26c02cf0 AB |
1238 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1239 | { return 0; } | |
205ab99d CL |
1240 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1241 | int objects) {} | |
1242 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1243 | int objects) {} | |
7d550c56 CL |
1244 | |
1245 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1246 | { return 0; } | |
1247 | ||
1248 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1249 | void *object) {} | |
1250 | ||
1251 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1252 | ||
ab4d5ed5 | 1253 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1254 | |
81819f0f CL |
1255 | /* |
1256 | * Slab allocation and freeing | |
1257 | */ | |
65c3376a CL |
1258 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1259 | struct kmem_cache_order_objects oo) | |
1260 | { | |
1261 | int order = oo_order(oo); | |
1262 | ||
b1eeab67 VN |
1263 | flags |= __GFP_NOTRACK; |
1264 | ||
2154a336 | 1265 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1266 | return alloc_pages(flags, order); |
1267 | else | |
6b65aaf3 | 1268 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1269 | } |
1270 | ||
81819f0f CL |
1271 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1272 | { | |
06428780 | 1273 | struct page *page; |
834f3d11 | 1274 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1275 | gfp_t alloc_gfp; |
81819f0f | 1276 | |
7e0528da CL |
1277 | flags &= gfp_allowed_mask; |
1278 | ||
1279 | if (flags & __GFP_WAIT) | |
1280 | local_irq_enable(); | |
1281 | ||
b7a49f0d | 1282 | flags |= s->allocflags; |
e12ba74d | 1283 | |
ba52270d PE |
1284 | /* |
1285 | * Let the initial higher-order allocation fail under memory pressure | |
1286 | * so we fall-back to the minimum order allocation. | |
1287 | */ | |
1288 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1289 | ||
1290 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1291 | if (unlikely(!page)) { |
1292 | oo = s->min; | |
1293 | /* | |
1294 | * Allocation may have failed due to fragmentation. | |
1295 | * Try a lower order alloc if possible | |
1296 | */ | |
1297 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1298 | |
7e0528da CL |
1299 | if (page) |
1300 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1301 | } |
5a896d9e | 1302 | |
737b719e | 1303 | if (kmemcheck_enabled && page |
5086c389 | 1304 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1305 | int pages = 1 << oo_order(oo); |
1306 | ||
1307 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1308 | ||
1309 | /* | |
1310 | * Objects from caches that have a constructor don't get | |
1311 | * cleared when they're allocated, so we need to do it here. | |
1312 | */ | |
1313 | if (s->ctor) | |
1314 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1315 | else | |
1316 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1317 | } |
1318 | ||
737b719e DR |
1319 | if (flags & __GFP_WAIT) |
1320 | local_irq_disable(); | |
1321 | if (!page) | |
1322 | return NULL; | |
1323 | ||
834f3d11 | 1324 | page->objects = oo_objects(oo); |
81819f0f CL |
1325 | mod_zone_page_state(page_zone(page), |
1326 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1327 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1328 | 1 << oo_order(oo)); |
81819f0f CL |
1329 | |
1330 | return page; | |
1331 | } | |
1332 | ||
1333 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1334 | void *object) | |
1335 | { | |
3ec09742 | 1336 | setup_object_debug(s, page, object); |
4f104934 | 1337 | if (unlikely(s->ctor)) |
51cc5068 | 1338 | s->ctor(object); |
81819f0f CL |
1339 | } |
1340 | ||
1341 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1342 | { | |
1343 | struct page *page; | |
81819f0f | 1344 | void *start; |
81819f0f CL |
1345 | void *last; |
1346 | void *p; | |
1f458cbf | 1347 | int order; |
81819f0f | 1348 | |
6cb06229 | 1349 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1350 | |
6cb06229 CL |
1351 | page = allocate_slab(s, |
1352 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1353 | if (!page) |
1354 | goto out; | |
1355 | ||
1f458cbf | 1356 | order = compound_order(page); |
205ab99d | 1357 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1f458cbf | 1358 | memcg_bind_pages(s, order); |
1b4f59e3 | 1359 | page->slab_cache = s; |
c03f94cc | 1360 | __SetPageSlab(page); |
072bb0aa MG |
1361 | if (page->pfmemalloc) |
1362 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1363 | |
1364 | start = page_address(page); | |
81819f0f CL |
1365 | |
1366 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1367 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f CL |
1368 | |
1369 | last = start; | |
224a88be | 1370 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1371 | setup_object(s, page, last); |
1372 | set_freepointer(s, last, p); | |
1373 | last = p; | |
1374 | } | |
1375 | setup_object(s, page, last); | |
a973e9dd | 1376 | set_freepointer(s, last, NULL); |
81819f0f CL |
1377 | |
1378 | page->freelist = start; | |
e6e82ea1 | 1379 | page->inuse = page->objects; |
8cb0a506 | 1380 | page->frozen = 1; |
81819f0f | 1381 | out: |
81819f0f CL |
1382 | return page; |
1383 | } | |
1384 | ||
1385 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1386 | { | |
834f3d11 CL |
1387 | int order = compound_order(page); |
1388 | int pages = 1 << order; | |
81819f0f | 1389 | |
af537b0a | 1390 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1391 | void *p; |
1392 | ||
1393 | slab_pad_check(s, page); | |
224a88be CL |
1394 | for_each_object(p, s, page_address(page), |
1395 | page->objects) | |
f7cb1933 | 1396 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1397 | } |
1398 | ||
b1eeab67 | 1399 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1400 | |
81819f0f CL |
1401 | mod_zone_page_state(page_zone(page), |
1402 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1403 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1404 | -pages); |
81819f0f | 1405 | |
072bb0aa | 1406 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1407 | __ClearPageSlab(page); |
1f458cbf GC |
1408 | |
1409 | memcg_release_pages(s, order); | |
49bd5221 | 1410 | reset_page_mapcount(page); |
1eb5ac64 NP |
1411 | if (current->reclaim_state) |
1412 | current->reclaim_state->reclaimed_slab += pages; | |
d79923fa | 1413 | __free_memcg_kmem_pages(page, order); |
81819f0f CL |
1414 | } |
1415 | ||
da9a638c LJ |
1416 | #define need_reserve_slab_rcu \ |
1417 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1418 | ||
81819f0f CL |
1419 | static void rcu_free_slab(struct rcu_head *h) |
1420 | { | |
1421 | struct page *page; | |
1422 | ||
da9a638c LJ |
1423 | if (need_reserve_slab_rcu) |
1424 | page = virt_to_head_page(h); | |
1425 | else | |
1426 | page = container_of((struct list_head *)h, struct page, lru); | |
1427 | ||
1b4f59e3 | 1428 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1429 | } |
1430 | ||
1431 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1432 | { | |
1433 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1434 | struct rcu_head *head; |
1435 | ||
1436 | if (need_reserve_slab_rcu) { | |
1437 | int order = compound_order(page); | |
1438 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1439 | ||
1440 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1441 | head = page_address(page) + offset; | |
1442 | } else { | |
1443 | /* | |
1444 | * RCU free overloads the RCU head over the LRU | |
1445 | */ | |
1446 | head = (void *)&page->lru; | |
1447 | } | |
81819f0f CL |
1448 | |
1449 | call_rcu(head, rcu_free_slab); | |
1450 | } else | |
1451 | __free_slab(s, page); | |
1452 | } | |
1453 | ||
1454 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1455 | { | |
205ab99d | 1456 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1457 | free_slab(s, page); |
1458 | } | |
1459 | ||
1460 | /* | |
5cc6eee8 CL |
1461 | * Management of partially allocated slabs. |
1462 | * | |
1463 | * list_lock must be held. | |
81819f0f | 1464 | */ |
5cc6eee8 | 1465 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1466 | struct page *page, int tail) |
81819f0f | 1467 | { |
e95eed57 | 1468 | n->nr_partial++; |
136333d1 | 1469 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1470 | list_add_tail(&page->lru, &n->partial); |
1471 | else | |
1472 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1473 | } |
1474 | ||
5cc6eee8 CL |
1475 | /* |
1476 | * list_lock must be held. | |
1477 | */ | |
1478 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1479 | struct page *page) |
1480 | { | |
1481 | list_del(&page->lru); | |
1482 | n->nr_partial--; | |
1483 | } | |
1484 | ||
81819f0f | 1485 | /* |
7ced3719 CL |
1486 | * Remove slab from the partial list, freeze it and |
1487 | * return the pointer to the freelist. | |
81819f0f | 1488 | * |
497b66f2 CL |
1489 | * Returns a list of objects or NULL if it fails. |
1490 | * | |
7ced3719 | 1491 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1492 | */ |
497b66f2 | 1493 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1494 | struct kmem_cache_node *n, struct page *page, |
49e22585 | 1495 | int mode) |
81819f0f | 1496 | { |
2cfb7455 CL |
1497 | void *freelist; |
1498 | unsigned long counters; | |
1499 | struct page new; | |
1500 | ||
2cfb7455 CL |
1501 | /* |
1502 | * Zap the freelist and set the frozen bit. | |
1503 | * The old freelist is the list of objects for the | |
1504 | * per cpu allocation list. | |
1505 | */ | |
7ced3719 CL |
1506 | freelist = page->freelist; |
1507 | counters = page->counters; | |
1508 | new.counters = counters; | |
23910c50 | 1509 | if (mode) { |
7ced3719 | 1510 | new.inuse = page->objects; |
23910c50 PE |
1511 | new.freelist = NULL; |
1512 | } else { | |
1513 | new.freelist = freelist; | |
1514 | } | |
2cfb7455 | 1515 | |
7ced3719 CL |
1516 | VM_BUG_ON(new.frozen); |
1517 | new.frozen = 1; | |
2cfb7455 | 1518 | |
7ced3719 | 1519 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1520 | freelist, counters, |
02d7633f | 1521 | new.freelist, new.counters, |
7ced3719 | 1522 | "acquire_slab")) |
7ced3719 | 1523 | return NULL; |
2cfb7455 CL |
1524 | |
1525 | remove_partial(n, page); | |
7ced3719 | 1526 | WARN_ON(!freelist); |
49e22585 | 1527 | return freelist; |
81819f0f CL |
1528 | } |
1529 | ||
49e22585 | 1530 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1531 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1532 | |
81819f0f | 1533 | /* |
672bba3a | 1534 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1535 | */ |
8ba00bb6 JK |
1536 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1537 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1538 | { |
49e22585 CL |
1539 | struct page *page, *page2; |
1540 | void *object = NULL; | |
81819f0f CL |
1541 | |
1542 | /* | |
1543 | * Racy check. If we mistakenly see no partial slabs then we | |
1544 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1545 | * partial slab and there is none available then get_partials() |
1546 | * will return NULL. | |
81819f0f CL |
1547 | */ |
1548 | if (!n || !n->nr_partial) | |
1549 | return NULL; | |
1550 | ||
1551 | spin_lock(&n->list_lock); | |
49e22585 | 1552 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1553 | void *t; |
49e22585 CL |
1554 | int available; |
1555 | ||
8ba00bb6 JK |
1556 | if (!pfmemalloc_match(page, flags)) |
1557 | continue; | |
1558 | ||
1559 | t = acquire_slab(s, n, page, object == NULL); | |
49e22585 CL |
1560 | if (!t) |
1561 | break; | |
1562 | ||
12d79634 | 1563 | if (!object) { |
49e22585 | 1564 | c->page = page; |
49e22585 | 1565 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 CL |
1566 | object = t; |
1567 | available = page->objects - page->inuse; | |
1568 | } else { | |
49e22585 | 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 | */ |
de3ec035 | 1583 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 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(); | |
e7b691b0 | 1618 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
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) { | |
8ba00bb6 | 1626 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
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 | |
8ba00bb6 | 1655 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
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 | ||
788e1aad | 1721 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 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 | ||
d24ac77f JK |
1873 | /* |
1874 | * Unfreeze all the cpu partial slabs. | |
1875 | * | |
59a09917 CL |
1876 | * This function must be called with interrupts disabled |
1877 | * for the cpu using c (or some other guarantee must be there | |
1878 | * to guarantee no concurrent accesses). | |
d24ac77f | 1879 | */ |
59a09917 CL |
1880 | static void unfreeze_partials(struct kmem_cache *s, |
1881 | struct kmem_cache_cpu *c) | |
49e22585 | 1882 | { |
43d77867 | 1883 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1884 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1885 | |
1886 | while ((page = c->partial)) { | |
49e22585 CL |
1887 | struct page new; |
1888 | struct page old; | |
1889 | ||
1890 | c->partial = page->next; | |
43d77867 JK |
1891 | |
1892 | n2 = get_node(s, page_to_nid(page)); | |
1893 | if (n != n2) { | |
1894 | if (n) | |
1895 | spin_unlock(&n->list_lock); | |
1896 | ||
1897 | n = n2; | |
1898 | spin_lock(&n->list_lock); | |
1899 | } | |
49e22585 CL |
1900 | |
1901 | do { | |
1902 | ||
1903 | old.freelist = page->freelist; | |
1904 | old.counters = page->counters; | |
1905 | VM_BUG_ON(!old.frozen); | |
1906 | ||
1907 | new.counters = old.counters; | |
1908 | new.freelist = old.freelist; | |
1909 | ||
1910 | new.frozen = 0; | |
1911 | ||
d24ac77f | 1912 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1913 | old.freelist, old.counters, |
1914 | new.freelist, new.counters, | |
1915 | "unfreezing slab")); | |
1916 | ||
43d77867 | 1917 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1918 | page->next = discard_page; |
1919 | discard_page = page; | |
43d77867 JK |
1920 | } else { |
1921 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1922 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1923 | } |
1924 | } | |
1925 | ||
1926 | if (n) | |
1927 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1928 | |
1929 | while (discard_page) { | |
1930 | page = discard_page; | |
1931 | discard_page = discard_page->next; | |
1932 | ||
1933 | stat(s, DEACTIVATE_EMPTY); | |
1934 | discard_slab(s, page); | |
1935 | stat(s, FREE_SLAB); | |
1936 | } | |
49e22585 CL |
1937 | } |
1938 | ||
1939 | /* | |
1940 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1941 | * slot if available. This is done without interrupts disabled and without | |
1942 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1943 | * onto a random cpus partial slot. | |
1944 | * | |
1945 | * If we did not find a slot then simply move all the partials to the | |
1946 | * per node partial list. | |
1947 | */ | |
788e1aad | 1948 | static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 CL |
1949 | { |
1950 | struct page *oldpage; | |
1951 | int pages; | |
1952 | int pobjects; | |
1953 | ||
1954 | do { | |
1955 | pages = 0; | |
1956 | pobjects = 0; | |
1957 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
1958 | ||
1959 | if (oldpage) { | |
1960 | pobjects = oldpage->pobjects; | |
1961 | pages = oldpage->pages; | |
1962 | if (drain && pobjects > s->cpu_partial) { | |
1963 | unsigned long flags; | |
1964 | /* | |
1965 | * partial array is full. Move the existing | |
1966 | * set to the per node partial list. | |
1967 | */ | |
1968 | local_irq_save(flags); | |
59a09917 | 1969 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 1970 | local_irq_restore(flags); |
e24fc410 | 1971 | oldpage = NULL; |
49e22585 CL |
1972 | pobjects = 0; |
1973 | pages = 0; | |
8028dcea | 1974 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
1975 | } |
1976 | } | |
1977 | ||
1978 | pages++; | |
1979 | pobjects += page->objects - page->inuse; | |
1980 | ||
1981 | page->pages = pages; | |
1982 | page->pobjects = pobjects; | |
1983 | page->next = oldpage; | |
1984 | ||
933393f5 | 1985 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
1986 | return pobjects; |
1987 | } | |
1988 | ||
dfb4f096 | 1989 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 1990 | { |
84e554e6 | 1991 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
1992 | deactivate_slab(s, c->page, c->freelist); |
1993 | ||
1994 | c->tid = next_tid(c->tid); | |
1995 | c->page = NULL; | |
1996 | c->freelist = NULL; | |
81819f0f CL |
1997 | } |
1998 | ||
1999 | /* | |
2000 | * Flush cpu slab. | |
6446faa2 | 2001 | * |
81819f0f CL |
2002 | * Called from IPI handler with interrupts disabled. |
2003 | */ | |
0c710013 | 2004 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2005 | { |
9dfc6e68 | 2006 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2007 | |
49e22585 CL |
2008 | if (likely(c)) { |
2009 | if (c->page) | |
2010 | flush_slab(s, c); | |
2011 | ||
59a09917 | 2012 | unfreeze_partials(s, c); |
49e22585 | 2013 | } |
81819f0f CL |
2014 | } |
2015 | ||
2016 | static void flush_cpu_slab(void *d) | |
2017 | { | |
2018 | struct kmem_cache *s = d; | |
81819f0f | 2019 | |
dfb4f096 | 2020 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2021 | } |
2022 | ||
a8364d55 GBY |
2023 | static bool has_cpu_slab(int cpu, void *info) |
2024 | { | |
2025 | struct kmem_cache *s = info; | |
2026 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2027 | ||
02e1a9cd | 2028 | return c->page || c->partial; |
a8364d55 GBY |
2029 | } |
2030 | ||
81819f0f CL |
2031 | static void flush_all(struct kmem_cache *s) |
2032 | { | |
a8364d55 | 2033 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2034 | } |
2035 | ||
dfb4f096 CL |
2036 | /* |
2037 | * Check if the objects in a per cpu structure fit numa | |
2038 | * locality expectations. | |
2039 | */ | |
57d437d2 | 2040 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2041 | { |
2042 | #ifdef CONFIG_NUMA | |
57d437d2 | 2043 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2044 | return 0; |
2045 | #endif | |
2046 | return 1; | |
2047 | } | |
2048 | ||
781b2ba6 PE |
2049 | static int count_free(struct page *page) |
2050 | { | |
2051 | return page->objects - page->inuse; | |
2052 | } | |
2053 | ||
2054 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2055 | int (*get_count)(struct page *)) | |
2056 | { | |
2057 | unsigned long flags; | |
2058 | unsigned long x = 0; | |
2059 | struct page *page; | |
2060 | ||
2061 | spin_lock_irqsave(&n->list_lock, flags); | |
2062 | list_for_each_entry(page, &n->partial, lru) | |
2063 | x += get_count(page); | |
2064 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2065 | return x; | |
2066 | } | |
2067 | ||
26c02cf0 AB |
2068 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2069 | { | |
2070 | #ifdef CONFIG_SLUB_DEBUG | |
2071 | return atomic_long_read(&n->total_objects); | |
2072 | #else | |
2073 | return 0; | |
2074 | #endif | |
2075 | } | |
2076 | ||
781b2ba6 PE |
2077 | static noinline void |
2078 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2079 | { | |
2080 | int node; | |
2081 | ||
2082 | printk(KERN_WARNING | |
2083 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2084 | nid, gfpflags); | |
2085 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2086 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2087 | s->size, oo_order(s->oo), oo_order(s->min)); |
2088 | ||
3b0efdfa | 2089 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2090 | printk(KERN_WARNING " %s debugging increased min order, use " |
2091 | "slub_debug=O to disable.\n", s->name); | |
2092 | ||
781b2ba6 PE |
2093 | for_each_online_node(node) { |
2094 | struct kmem_cache_node *n = get_node(s, node); | |
2095 | unsigned long nr_slabs; | |
2096 | unsigned long nr_objs; | |
2097 | unsigned long nr_free; | |
2098 | ||
2099 | if (!n) | |
2100 | continue; | |
2101 | ||
26c02cf0 AB |
2102 | nr_free = count_partial(n, count_free); |
2103 | nr_slabs = node_nr_slabs(n); | |
2104 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2105 | |
2106 | printk(KERN_WARNING | |
2107 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2108 | node, nr_slabs, nr_objs, nr_free); | |
2109 | } | |
2110 | } | |
2111 | ||
497b66f2 CL |
2112 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2113 | int node, struct kmem_cache_cpu **pc) | |
2114 | { | |
6faa6833 | 2115 | void *freelist; |
188fd063 CL |
2116 | struct kmem_cache_cpu *c = *pc; |
2117 | struct page *page; | |
497b66f2 | 2118 | |
188fd063 | 2119 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2120 | |
188fd063 CL |
2121 | if (freelist) |
2122 | return freelist; | |
2123 | ||
2124 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2125 | if (page) { |
2126 | c = __this_cpu_ptr(s->cpu_slab); | |
2127 | if (c->page) | |
2128 | flush_slab(s, c); | |
2129 | ||
2130 | /* | |
2131 | * No other reference to the page yet so we can | |
2132 | * muck around with it freely without cmpxchg | |
2133 | */ | |
6faa6833 | 2134 | freelist = page->freelist; |
497b66f2 CL |
2135 | page->freelist = NULL; |
2136 | ||
2137 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2138 | c->page = page; |
2139 | *pc = c; | |
2140 | } else | |
6faa6833 | 2141 | freelist = NULL; |
497b66f2 | 2142 | |
6faa6833 | 2143 | return freelist; |
497b66f2 CL |
2144 | } |
2145 | ||
072bb0aa MG |
2146 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2147 | { | |
2148 | if (unlikely(PageSlabPfmemalloc(page))) | |
2149 | return gfp_pfmemalloc_allowed(gfpflags); | |
2150 | ||
2151 | return true; | |
2152 | } | |
2153 | ||
213eeb9f CL |
2154 | /* |
2155 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2156 | * or deactivate the page. | |
2157 | * | |
2158 | * The page is still frozen if the return value is not NULL. | |
2159 | * | |
2160 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2161 | * |
2162 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2163 | */ |
2164 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2165 | { | |
2166 | struct page new; | |
2167 | unsigned long counters; | |
2168 | void *freelist; | |
2169 | ||
2170 | do { | |
2171 | freelist = page->freelist; | |
2172 | counters = page->counters; | |
6faa6833 | 2173 | |
213eeb9f CL |
2174 | new.counters = counters; |
2175 | VM_BUG_ON(!new.frozen); | |
2176 | ||
2177 | new.inuse = page->objects; | |
2178 | new.frozen = freelist != NULL; | |
2179 | ||
d24ac77f | 2180 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2181 | freelist, counters, |
2182 | NULL, new.counters, | |
2183 | "get_freelist")); | |
2184 | ||
2185 | return freelist; | |
2186 | } | |
2187 | ||
81819f0f | 2188 | /* |
894b8788 CL |
2189 | * Slow path. The lockless freelist is empty or we need to perform |
2190 | * debugging duties. | |
2191 | * | |
894b8788 CL |
2192 | * Processing is still very fast if new objects have been freed to the |
2193 | * regular freelist. In that case we simply take over the regular freelist | |
2194 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2195 | * |
894b8788 CL |
2196 | * If that is not working then we fall back to the partial lists. We take the |
2197 | * first element of the freelist as the object to allocate now and move the | |
2198 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2199 | * |
894b8788 | 2200 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2201 | * we need to allocate a new slab. This is the slowest path since it involves |
2202 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2203 | */ |
ce71e27c EGM |
2204 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2205 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2206 | { |
6faa6833 | 2207 | void *freelist; |
f6e7def7 | 2208 | struct page *page; |
8a5ec0ba CL |
2209 | unsigned long flags; |
2210 | ||
2211 | local_irq_save(flags); | |
2212 | #ifdef CONFIG_PREEMPT | |
2213 | /* | |
2214 | * We may have been preempted and rescheduled on a different | |
2215 | * cpu before disabling interrupts. Need to reload cpu area | |
2216 | * pointer. | |
2217 | */ | |
2218 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2219 | #endif |
81819f0f | 2220 | |
f6e7def7 CL |
2221 | page = c->page; |
2222 | if (!page) | |
81819f0f | 2223 | goto new_slab; |
49e22585 | 2224 | redo: |
6faa6833 | 2225 | |
57d437d2 | 2226 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2227 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2228 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2229 | c->page = NULL; |
2230 | c->freelist = NULL; | |
fc59c053 CL |
2231 | goto new_slab; |
2232 | } | |
6446faa2 | 2233 | |
072bb0aa MG |
2234 | /* |
2235 | * By rights, we should be searching for a slab page that was | |
2236 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2237 | * information when the page leaves the per-cpu allocator | |
2238 | */ | |
2239 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2240 | deactivate_slab(s, page, c->freelist); | |
2241 | c->page = NULL; | |
2242 | c->freelist = NULL; | |
2243 | goto new_slab; | |
2244 | } | |
2245 | ||
73736e03 | 2246 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2247 | freelist = c->freelist; |
2248 | if (freelist) | |
73736e03 | 2249 | goto load_freelist; |
03e404af | 2250 | |
2cfb7455 | 2251 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2252 | |
f6e7def7 | 2253 | freelist = get_freelist(s, page); |
6446faa2 | 2254 | |
6faa6833 | 2255 | if (!freelist) { |
03e404af CL |
2256 | c->page = NULL; |
2257 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2258 | goto new_slab; |
03e404af | 2259 | } |
6446faa2 | 2260 | |
84e554e6 | 2261 | stat(s, ALLOC_REFILL); |
6446faa2 | 2262 | |
894b8788 | 2263 | load_freelist: |
507effea CL |
2264 | /* |
2265 | * freelist is pointing to the list of objects to be used. | |
2266 | * page is pointing to the page from which the objects are obtained. | |
2267 | * That page must be frozen for per cpu allocations to work. | |
2268 | */ | |
2269 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2270 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2271 | c->tid = next_tid(c->tid); |
2272 | local_irq_restore(flags); | |
6faa6833 | 2273 | return freelist; |
81819f0f | 2274 | |
81819f0f | 2275 | new_slab: |
2cfb7455 | 2276 | |
49e22585 | 2277 | if (c->partial) { |
f6e7def7 CL |
2278 | page = c->page = c->partial; |
2279 | c->partial = page->next; | |
49e22585 CL |
2280 | stat(s, CPU_PARTIAL_ALLOC); |
2281 | c->freelist = NULL; | |
2282 | goto redo; | |
81819f0f CL |
2283 | } |
2284 | ||
188fd063 | 2285 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2286 | |
f4697436 CL |
2287 | if (unlikely(!freelist)) { |
2288 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2289 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2290 | |
f4697436 CL |
2291 | local_irq_restore(flags); |
2292 | return NULL; | |
81819f0f | 2293 | } |
2cfb7455 | 2294 | |
f6e7def7 | 2295 | page = c->page; |
5091b74a | 2296 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2297 | goto load_freelist; |
2cfb7455 | 2298 | |
497b66f2 | 2299 | /* Only entered in the debug case */ |
5091b74a | 2300 | if (kmem_cache_debug(s) && !alloc_debug_processing(s, page, freelist, addr)) |
497b66f2 | 2301 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2302 | |
f6e7def7 | 2303 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2304 | c->page = NULL; |
2305 | c->freelist = NULL; | |
a71ae47a | 2306 | local_irq_restore(flags); |
6faa6833 | 2307 | return freelist; |
894b8788 CL |
2308 | } |
2309 | ||
2310 | /* | |
2311 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2312 | * have the fastpath folded into their functions. So no function call | |
2313 | * overhead for requests that can be satisfied on the fastpath. | |
2314 | * | |
2315 | * The fastpath works by first checking if the lockless freelist can be used. | |
2316 | * If not then __slab_alloc is called for slow processing. | |
2317 | * | |
2318 | * Otherwise we can simply pick the next object from the lockless free list. | |
2319 | */ | |
2b847c3c | 2320 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2321 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2322 | { |
894b8788 | 2323 | void **object; |
dfb4f096 | 2324 | struct kmem_cache_cpu *c; |
57d437d2 | 2325 | struct page *page; |
8a5ec0ba | 2326 | unsigned long tid; |
1f84260c | 2327 | |
c016b0bd | 2328 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2329 | return NULL; |
1f84260c | 2330 | |
d79923fa | 2331 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2332 | redo: |
8a5ec0ba CL |
2333 | |
2334 | /* | |
2335 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2336 | * enabled. We may switch back and forth between cpus while | |
2337 | * reading from one cpu area. That does not matter as long | |
2338 | * as we end up on the original cpu again when doing the cmpxchg. | |
2339 | */ | |
9dfc6e68 | 2340 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2341 | |
8a5ec0ba CL |
2342 | /* |
2343 | * The transaction ids are globally unique per cpu and per operation on | |
2344 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2345 | * occurs on the right processor and that there was no operation on the | |
2346 | * linked list in between. | |
2347 | */ | |
2348 | tid = c->tid; | |
2349 | barrier(); | |
8a5ec0ba | 2350 | |
9dfc6e68 | 2351 | object = c->freelist; |
57d437d2 | 2352 | page = c->page; |
5091b74a | 2353 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2354 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2355 | |
2356 | else { | |
0ad9500e ED |
2357 | void *next_object = get_freepointer_safe(s, object); |
2358 | ||
8a5ec0ba | 2359 | /* |
25985edc | 2360 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2361 | * operation and if we are on the right processor. |
2362 | * | |
2363 | * The cmpxchg does the following atomically (without lock semantics!) | |
2364 | * 1. Relocate first pointer to the current per cpu area. | |
2365 | * 2. Verify that tid and freelist have not been changed | |
2366 | * 3. If they were not changed replace tid and freelist | |
2367 | * | |
2368 | * Since this is without lock semantics the protection is only against | |
2369 | * code executing on this cpu *not* from access by other cpus. | |
2370 | */ | |
933393f5 | 2371 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2372 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2373 | object, tid, | |
0ad9500e | 2374 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2375 | |
2376 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2377 | goto redo; | |
2378 | } | |
0ad9500e | 2379 | prefetch_freepointer(s, next_object); |
84e554e6 | 2380 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2381 | } |
8a5ec0ba | 2382 | |
74e2134f | 2383 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2384 | memset(object, 0, s->object_size); |
d07dbea4 | 2385 | |
c016b0bd | 2386 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2387 | |
894b8788 | 2388 | return object; |
81819f0f CL |
2389 | } |
2390 | ||
2b847c3c EG |
2391 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2392 | gfp_t gfpflags, unsigned long addr) | |
2393 | { | |
2394 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2395 | } | |
2396 | ||
81819f0f CL |
2397 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2398 | { | |
2b847c3c | 2399 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2400 | |
3b0efdfa | 2401 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags); |
5b882be4 EGM |
2402 | |
2403 | return ret; | |
81819f0f CL |
2404 | } |
2405 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2406 | ||
0f24f128 | 2407 | #ifdef CONFIG_TRACING |
4a92379b RK |
2408 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2409 | { | |
2b847c3c | 2410 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2411 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2412 | return ret; | |
2413 | } | |
2414 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2415 | ||
2416 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2417 | { |
4a92379b RK |
2418 | void *ret = kmalloc_order(size, flags, order); |
2419 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2420 | return ret; | |
5b882be4 | 2421 | } |
4a92379b | 2422 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2423 | #endif |
2424 | ||
81819f0f CL |
2425 | #ifdef CONFIG_NUMA |
2426 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2427 | { | |
2b847c3c | 2428 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2429 | |
ca2b84cb | 2430 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2431 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2432 | |
2433 | return ret; | |
81819f0f CL |
2434 | } |
2435 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2436 | |
0f24f128 | 2437 | #ifdef CONFIG_TRACING |
4a92379b | 2438 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2439 | gfp_t gfpflags, |
4a92379b | 2440 | int node, size_t size) |
5b882be4 | 2441 | { |
2b847c3c | 2442 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2443 | |
2444 | trace_kmalloc_node(_RET_IP_, ret, | |
2445 | size, s->size, gfpflags, node); | |
2446 | return ret; | |
5b882be4 | 2447 | } |
4a92379b | 2448 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2449 | #endif |
5d1f57e4 | 2450 | #endif |
5b882be4 | 2451 | |
81819f0f | 2452 | /* |
894b8788 CL |
2453 | * Slow patch handling. This may still be called frequently since objects |
2454 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2455 | * |
894b8788 CL |
2456 | * So we still attempt to reduce cache line usage. Just take the slab |
2457 | * lock and free the item. If there is no additional partial page | |
2458 | * handling required then we can return immediately. | |
81819f0f | 2459 | */ |
894b8788 | 2460 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2461 | void *x, unsigned long addr) |
81819f0f CL |
2462 | { |
2463 | void *prior; | |
2464 | void **object = (void *)x; | |
2cfb7455 | 2465 | int was_frozen; |
2cfb7455 CL |
2466 | struct page new; |
2467 | unsigned long counters; | |
2468 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2469 | unsigned long uninitialized_var(flags); |
81819f0f | 2470 | |
8a5ec0ba | 2471 | stat(s, FREE_SLOWPATH); |
81819f0f | 2472 | |
19c7ff9e CL |
2473 | if (kmem_cache_debug(s) && |
2474 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2475 | return; |
6446faa2 | 2476 | |
2cfb7455 | 2477 | do { |
837d678d JK |
2478 | if (unlikely(n)) { |
2479 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2480 | n = NULL; | |
2481 | } | |
2cfb7455 CL |
2482 | prior = page->freelist; |
2483 | counters = page->counters; | |
2484 | set_freepointer(s, object, prior); | |
2485 | new.counters = counters; | |
2486 | was_frozen = new.frozen; | |
2487 | new.inuse--; | |
837d678d | 2488 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 CL |
2489 | |
2490 | if (!kmem_cache_debug(s) && !prior) | |
2491 | ||
2492 | /* | |
2493 | * Slab was on no list before and will be partially empty | |
2494 | * We can defer the list move and instead freeze it. | |
2495 | */ | |
2496 | new.frozen = 1; | |
2497 | ||
2498 | else { /* Needs to be taken off a list */ | |
2499 | ||
2500 | n = get_node(s, page_to_nid(page)); | |
2501 | /* | |
2502 | * Speculatively acquire the list_lock. | |
2503 | * If the cmpxchg does not succeed then we may | |
2504 | * drop the list_lock without any processing. | |
2505 | * | |
2506 | * Otherwise the list_lock will synchronize with | |
2507 | * other processors updating the list of slabs. | |
2508 | */ | |
2509 | spin_lock_irqsave(&n->list_lock, flags); | |
2510 | ||
2511 | } | |
2cfb7455 | 2512 | } |
81819f0f | 2513 | |
2cfb7455 CL |
2514 | } while (!cmpxchg_double_slab(s, page, |
2515 | prior, counters, | |
2516 | object, new.counters, | |
2517 | "__slab_free")); | |
81819f0f | 2518 | |
2cfb7455 | 2519 | if (likely(!n)) { |
49e22585 CL |
2520 | |
2521 | /* | |
2522 | * If we just froze the page then put it onto the | |
2523 | * per cpu partial list. | |
2524 | */ | |
8028dcea | 2525 | if (new.frozen && !was_frozen) { |
49e22585 | 2526 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2527 | stat(s, CPU_PARTIAL_FREE); |
2528 | } | |
49e22585 | 2529 | /* |
2cfb7455 CL |
2530 | * The list lock was not taken therefore no list |
2531 | * activity can be necessary. | |
2532 | */ | |
2533 | if (was_frozen) | |
2534 | stat(s, FREE_FROZEN); | |
80f08c19 | 2535 | return; |
2cfb7455 | 2536 | } |
81819f0f | 2537 | |
837d678d JK |
2538 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) |
2539 | goto slab_empty; | |
2540 | ||
81819f0f | 2541 | /* |
837d678d JK |
2542 | * Objects left in the slab. If it was not on the partial list before |
2543 | * then add it. | |
81819f0f | 2544 | */ |
837d678d JK |
2545 | if (kmem_cache_debug(s) && unlikely(!prior)) { |
2546 | remove_full(s, page); | |
2547 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2548 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2549 | } |
80f08c19 | 2550 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2551 | return; |
2552 | ||
2553 | slab_empty: | |
a973e9dd | 2554 | if (prior) { |
81819f0f | 2555 | /* |
6fbabb20 | 2556 | * Slab on the partial list. |
81819f0f | 2557 | */ |
5cc6eee8 | 2558 | remove_partial(n, page); |
84e554e6 | 2559 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2560 | } else |
2561 | /* Slab must be on the full list */ | |
2562 | remove_full(s, page); | |
2cfb7455 | 2563 | |
80f08c19 | 2564 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2565 | stat(s, FREE_SLAB); |
81819f0f | 2566 | discard_slab(s, page); |
81819f0f CL |
2567 | } |
2568 | ||
894b8788 CL |
2569 | /* |
2570 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2571 | * can perform fastpath freeing without additional function calls. | |
2572 | * | |
2573 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2574 | * of this processor. This typically the case if we have just allocated | |
2575 | * the item before. | |
2576 | * | |
2577 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2578 | * with all sorts of special processing. | |
2579 | */ | |
06428780 | 2580 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2581 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2582 | { |
2583 | void **object = (void *)x; | |
dfb4f096 | 2584 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2585 | unsigned long tid; |
1f84260c | 2586 | |
c016b0bd CL |
2587 | slab_free_hook(s, x); |
2588 | ||
8a5ec0ba CL |
2589 | redo: |
2590 | /* | |
2591 | * Determine the currently cpus per cpu slab. | |
2592 | * The cpu may change afterward. However that does not matter since | |
2593 | * data is retrieved via this pointer. If we are on the same cpu | |
2594 | * during the cmpxchg then the free will succedd. | |
2595 | */ | |
9dfc6e68 | 2596 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2597 | |
8a5ec0ba CL |
2598 | tid = c->tid; |
2599 | barrier(); | |
c016b0bd | 2600 | |
442b06bc | 2601 | if (likely(page == c->page)) { |
ff12059e | 2602 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2603 | |
933393f5 | 2604 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2605 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2606 | c->freelist, tid, | |
2607 | object, next_tid(tid)))) { | |
2608 | ||
2609 | note_cmpxchg_failure("slab_free", s, tid); | |
2610 | goto redo; | |
2611 | } | |
84e554e6 | 2612 | stat(s, FREE_FASTPATH); |
894b8788 | 2613 | } else |
ff12059e | 2614 | __slab_free(s, page, x, addr); |
894b8788 | 2615 | |
894b8788 CL |
2616 | } |
2617 | ||
81819f0f CL |
2618 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2619 | { | |
b9ce5ef4 GC |
2620 | s = cache_from_obj(s, x); |
2621 | if (!s) | |
79576102 | 2622 | return; |
b9ce5ef4 | 2623 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2624 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2625 | } |
2626 | EXPORT_SYMBOL(kmem_cache_free); | |
2627 | ||
81819f0f | 2628 | /* |
672bba3a CL |
2629 | * Object placement in a slab is made very easy because we always start at |
2630 | * offset 0. If we tune the size of the object to the alignment then we can | |
2631 | * get the required alignment by putting one properly sized object after | |
2632 | * another. | |
81819f0f CL |
2633 | * |
2634 | * Notice that the allocation order determines the sizes of the per cpu | |
2635 | * caches. Each processor has always one slab available for allocations. | |
2636 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2637 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2638 | * locking overhead. |
81819f0f CL |
2639 | */ |
2640 | ||
2641 | /* | |
2642 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2643 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2644 | * and increases the number of allocations possible without having to | |
2645 | * take the list_lock. | |
2646 | */ | |
2647 | static int slub_min_order; | |
114e9e89 | 2648 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2649 | static int slub_min_objects; |
81819f0f CL |
2650 | |
2651 | /* | |
2652 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2653 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2654 | */ |
2655 | static int slub_nomerge; | |
2656 | ||
81819f0f CL |
2657 | /* |
2658 | * Calculate the order of allocation given an slab object size. | |
2659 | * | |
672bba3a CL |
2660 | * The order of allocation has significant impact on performance and other |
2661 | * system components. Generally order 0 allocations should be preferred since | |
2662 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2663 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2664 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2665 | * would be wasted. |
2666 | * | |
2667 | * In order to reach satisfactory performance we must ensure that a minimum | |
2668 | * number of objects is in one slab. Otherwise we may generate too much | |
2669 | * activity on the partial lists which requires taking the list_lock. This is | |
2670 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2671 | * |
672bba3a CL |
2672 | * slub_max_order specifies the order where we begin to stop considering the |
2673 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2674 | * we try to keep the page order as low as possible. So we accept more waste | |
2675 | * of space in favor of a small page order. | |
81819f0f | 2676 | * |
672bba3a CL |
2677 | * Higher order allocations also allow the placement of more objects in a |
2678 | * slab and thereby reduce object handling overhead. If the user has | |
2679 | * requested a higher mininum order then we start with that one instead of | |
2680 | * the smallest order which will fit the object. | |
81819f0f | 2681 | */ |
5e6d444e | 2682 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2683 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2684 | { |
2685 | int order; | |
2686 | int rem; | |
6300ea75 | 2687 | int min_order = slub_min_order; |
81819f0f | 2688 | |
ab9a0f19 | 2689 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2690 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2691 | |
6300ea75 | 2692 | for (order = max(min_order, |
5e6d444e CL |
2693 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2694 | order <= max_order; order++) { | |
81819f0f | 2695 | |
5e6d444e | 2696 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2697 | |
ab9a0f19 | 2698 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2699 | continue; |
2700 | ||
ab9a0f19 | 2701 | rem = (slab_size - reserved) % size; |
81819f0f | 2702 | |
5e6d444e | 2703 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2704 | break; |
2705 | ||
2706 | } | |
672bba3a | 2707 | |
81819f0f CL |
2708 | return order; |
2709 | } | |
2710 | ||
ab9a0f19 | 2711 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2712 | { |
2713 | int order; | |
2714 | int min_objects; | |
2715 | int fraction; | |
e8120ff1 | 2716 | int max_objects; |
5e6d444e CL |
2717 | |
2718 | /* | |
2719 | * Attempt to find best configuration for a slab. This | |
2720 | * works by first attempting to generate a layout with | |
2721 | * the best configuration and backing off gradually. | |
2722 | * | |
2723 | * First we reduce the acceptable waste in a slab. Then | |
2724 | * we reduce the minimum objects required in a slab. | |
2725 | */ | |
2726 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2727 | if (!min_objects) |
2728 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2729 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2730 | min_objects = min(min_objects, max_objects); |
2731 | ||
5e6d444e | 2732 | while (min_objects > 1) { |
c124f5b5 | 2733 | fraction = 16; |
5e6d444e CL |
2734 | while (fraction >= 4) { |
2735 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2736 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2737 | if (order <= slub_max_order) |
2738 | return order; | |
2739 | fraction /= 2; | |
2740 | } | |
5086c389 | 2741 | min_objects--; |
5e6d444e CL |
2742 | } |
2743 | ||
2744 | /* | |
2745 | * We were unable to place multiple objects in a slab. Now | |
2746 | * lets see if we can place a single object there. | |
2747 | */ | |
ab9a0f19 | 2748 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2749 | if (order <= slub_max_order) |
2750 | return order; | |
2751 | ||
2752 | /* | |
2753 | * Doh this slab cannot be placed using slub_max_order. | |
2754 | */ | |
ab9a0f19 | 2755 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2756 | if (order < MAX_ORDER) |
5e6d444e CL |
2757 | return order; |
2758 | return -ENOSYS; | |
2759 | } | |
2760 | ||
5595cffc | 2761 | static void |
4053497d | 2762 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2763 | { |
2764 | n->nr_partial = 0; | |
81819f0f CL |
2765 | spin_lock_init(&n->list_lock); |
2766 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2767 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2768 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2769 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2770 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2771 | #endif |
81819f0f CL |
2772 | } |
2773 | ||
55136592 | 2774 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2775 | { |
6c182dc0 CL |
2776 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
2777 | SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 2778 | |
8a5ec0ba | 2779 | /* |
d4d84fef CM |
2780 | * Must align to double word boundary for the double cmpxchg |
2781 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2782 | */ |
d4d84fef CM |
2783 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2784 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2785 | |
2786 | if (!s->cpu_slab) | |
2787 | return 0; | |
2788 | ||
2789 | init_kmem_cache_cpus(s); | |
4c93c355 | 2790 | |
8a5ec0ba | 2791 | return 1; |
4c93c355 | 2792 | } |
4c93c355 | 2793 | |
51df1142 CL |
2794 | static struct kmem_cache *kmem_cache_node; |
2795 | ||
81819f0f CL |
2796 | /* |
2797 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2798 | * slab on the node for this slabcache. There are no concurrent accesses | |
2799 | * possible. | |
2800 | * | |
2801 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2802 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2803 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2804 | */ |
55136592 | 2805 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2806 | { |
2807 | struct page *page; | |
2808 | struct kmem_cache_node *n; | |
2809 | ||
51df1142 | 2810 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2811 | |
51df1142 | 2812 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2813 | |
2814 | BUG_ON(!page); | |
a2f92ee7 CL |
2815 | if (page_to_nid(page) != node) { |
2816 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2817 | "node %d\n", node); | |
2818 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2819 | "in order to be able to continue\n"); | |
2820 | } | |
2821 | ||
81819f0f CL |
2822 | n = page->freelist; |
2823 | BUG_ON(!n); | |
51df1142 | 2824 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2825 | page->inuse = 1; |
8cb0a506 | 2826 | page->frozen = 0; |
51df1142 | 2827 | kmem_cache_node->node[node] = n; |
8ab1372f | 2828 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2829 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2830 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2831 | #endif |
4053497d | 2832 | init_kmem_cache_node(n); |
51df1142 | 2833 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2834 | |
136333d1 | 2835 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2836 | } |
2837 | ||
2838 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2839 | { | |
2840 | int node; | |
2841 | ||
f64dc58c | 2842 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2843 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2844 | |
73367bd8 | 2845 | if (n) |
51df1142 CL |
2846 | kmem_cache_free(kmem_cache_node, n); |
2847 | ||
81819f0f CL |
2848 | s->node[node] = NULL; |
2849 | } | |
2850 | } | |
2851 | ||
55136592 | 2852 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2853 | { |
2854 | int node; | |
81819f0f | 2855 | |
f64dc58c | 2856 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2857 | struct kmem_cache_node *n; |
2858 | ||
73367bd8 | 2859 | if (slab_state == DOWN) { |
55136592 | 2860 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2861 | continue; |
2862 | } | |
51df1142 | 2863 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2864 | GFP_KERNEL, node); |
81819f0f | 2865 | |
73367bd8 AD |
2866 | if (!n) { |
2867 | free_kmem_cache_nodes(s); | |
2868 | return 0; | |
81819f0f | 2869 | } |
73367bd8 | 2870 | |
81819f0f | 2871 | s->node[node] = n; |
4053497d | 2872 | init_kmem_cache_node(n); |
81819f0f CL |
2873 | } |
2874 | return 1; | |
2875 | } | |
81819f0f | 2876 | |
c0bdb232 | 2877 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2878 | { |
2879 | if (min < MIN_PARTIAL) | |
2880 | min = MIN_PARTIAL; | |
2881 | else if (min > MAX_PARTIAL) | |
2882 | min = MAX_PARTIAL; | |
2883 | s->min_partial = min; | |
2884 | } | |
2885 | ||
81819f0f CL |
2886 | /* |
2887 | * calculate_sizes() determines the order and the distribution of data within | |
2888 | * a slab object. | |
2889 | */ | |
06b285dc | 2890 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2891 | { |
2892 | unsigned long flags = s->flags; | |
3b0efdfa | 2893 | unsigned long size = s->object_size; |
834f3d11 | 2894 | int order; |
81819f0f | 2895 | |
d8b42bf5 CL |
2896 | /* |
2897 | * Round up object size to the next word boundary. We can only | |
2898 | * place the free pointer at word boundaries and this determines | |
2899 | * the possible location of the free pointer. | |
2900 | */ | |
2901 | size = ALIGN(size, sizeof(void *)); | |
2902 | ||
2903 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2904 | /* |
2905 | * Determine if we can poison the object itself. If the user of | |
2906 | * the slab may touch the object after free or before allocation | |
2907 | * then we should never poison the object itself. | |
2908 | */ | |
2909 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2910 | !s->ctor) |
81819f0f CL |
2911 | s->flags |= __OBJECT_POISON; |
2912 | else | |
2913 | s->flags &= ~__OBJECT_POISON; | |
2914 | ||
81819f0f CL |
2915 | |
2916 | /* | |
672bba3a | 2917 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2918 | * end of the object and the free pointer. If not then add an |
672bba3a | 2919 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2920 | */ |
3b0efdfa | 2921 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2922 | size += sizeof(void *); |
41ecc55b | 2923 | #endif |
81819f0f CL |
2924 | |
2925 | /* | |
672bba3a CL |
2926 | * With that we have determined the number of bytes in actual use |
2927 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2928 | */ |
2929 | s->inuse = size; | |
2930 | ||
2931 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2932 | s->ctor)) { |
81819f0f CL |
2933 | /* |
2934 | * Relocate free pointer after the object if it is not | |
2935 | * permitted to overwrite the first word of the object on | |
2936 | * kmem_cache_free. | |
2937 | * | |
2938 | * This is the case if we do RCU, have a constructor or | |
2939 | * destructor or are poisoning the objects. | |
2940 | */ | |
2941 | s->offset = size; | |
2942 | size += sizeof(void *); | |
2943 | } | |
2944 | ||
c12b3c62 | 2945 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2946 | if (flags & SLAB_STORE_USER) |
2947 | /* | |
2948 | * Need to store information about allocs and frees after | |
2949 | * the object. | |
2950 | */ | |
2951 | size += 2 * sizeof(struct track); | |
2952 | ||
be7b3fbc | 2953 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
2954 | /* |
2955 | * Add some empty padding so that we can catch | |
2956 | * overwrites from earlier objects rather than let | |
2957 | * tracking information or the free pointer be | |
0211a9c8 | 2958 | * corrupted if a user writes before the start |
81819f0f CL |
2959 | * of the object. |
2960 | */ | |
2961 | size += sizeof(void *); | |
41ecc55b | 2962 | #endif |
672bba3a | 2963 | |
81819f0f CL |
2964 | /* |
2965 | * SLUB stores one object immediately after another beginning from | |
2966 | * offset 0. In order to align the objects we have to simply size | |
2967 | * each object to conform to the alignment. | |
2968 | */ | |
45906855 | 2969 | size = ALIGN(size, s->align); |
81819f0f | 2970 | s->size = size; |
06b285dc CL |
2971 | if (forced_order >= 0) |
2972 | order = forced_order; | |
2973 | else | |
ab9a0f19 | 2974 | order = calculate_order(size, s->reserved); |
81819f0f | 2975 | |
834f3d11 | 2976 | if (order < 0) |
81819f0f CL |
2977 | return 0; |
2978 | ||
b7a49f0d | 2979 | s->allocflags = 0; |
834f3d11 | 2980 | if (order) |
b7a49f0d CL |
2981 | s->allocflags |= __GFP_COMP; |
2982 | ||
2983 | if (s->flags & SLAB_CACHE_DMA) | |
2984 | s->allocflags |= SLUB_DMA; | |
2985 | ||
2986 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2987 | s->allocflags |= __GFP_RECLAIMABLE; | |
2988 | ||
81819f0f CL |
2989 | /* |
2990 | * Determine the number of objects per slab | |
2991 | */ | |
ab9a0f19 LJ |
2992 | s->oo = oo_make(order, size, s->reserved); |
2993 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
2994 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2995 | s->max = s->oo; | |
81819f0f | 2996 | |
834f3d11 | 2997 | return !!oo_objects(s->oo); |
81819f0f CL |
2998 | } |
2999 | ||
8a13a4cc | 3000 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3001 | { |
8a13a4cc | 3002 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3003 | s->reserved = 0; |
81819f0f | 3004 | |
da9a638c LJ |
3005 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3006 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3007 | |
06b285dc | 3008 | if (!calculate_sizes(s, -1)) |
81819f0f | 3009 | goto error; |
3de47213 DR |
3010 | if (disable_higher_order_debug) { |
3011 | /* | |
3012 | * Disable debugging flags that store metadata if the min slab | |
3013 | * order increased. | |
3014 | */ | |
3b0efdfa | 3015 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3016 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3017 | s->offset = 0; | |
3018 | if (!calculate_sizes(s, -1)) | |
3019 | goto error; | |
3020 | } | |
3021 | } | |
81819f0f | 3022 | |
2565409f HC |
3023 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3024 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3025 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3026 | /* Enable fast mode */ | |
3027 | s->flags |= __CMPXCHG_DOUBLE; | |
3028 | #endif | |
3029 | ||
3b89d7d8 DR |
3030 | /* |
3031 | * The larger the object size is, the more pages we want on the partial | |
3032 | * list to avoid pounding the page allocator excessively. | |
3033 | */ | |
49e22585 CL |
3034 | set_min_partial(s, ilog2(s->size) / 2); |
3035 | ||
3036 | /* | |
3037 | * cpu_partial determined the maximum number of objects kept in the | |
3038 | * per cpu partial lists of a processor. | |
3039 | * | |
3040 | * Per cpu partial lists mainly contain slabs that just have one | |
3041 | * object freed. If they are used for allocation then they can be | |
3042 | * filled up again with minimal effort. The slab will never hit the | |
3043 | * per node partial lists and therefore no locking will be required. | |
3044 | * | |
3045 | * This setting also determines | |
3046 | * | |
3047 | * A) The number of objects from per cpu partial slabs dumped to the | |
3048 | * per node list when we reach the limit. | |
9f264904 | 3049 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3050 | * per node list when we run out of per cpu objects. We only fetch 50% |
3051 | * to keep some capacity around for frees. | |
3052 | */ | |
8f1e33da CL |
3053 | if (kmem_cache_debug(s)) |
3054 | s->cpu_partial = 0; | |
3055 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3056 | s->cpu_partial = 2; |
3057 | else if (s->size >= 1024) | |
3058 | s->cpu_partial = 6; | |
3059 | else if (s->size >= 256) | |
3060 | s->cpu_partial = 13; | |
3061 | else | |
3062 | s->cpu_partial = 30; | |
3063 | ||
81819f0f | 3064 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3065 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3066 | #endif |
55136592 | 3067 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3068 | goto error; |
81819f0f | 3069 | |
55136592 | 3070 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3071 | return 0; |
ff12059e | 3072 | |
4c93c355 | 3073 | free_kmem_cache_nodes(s); |
81819f0f CL |
3074 | error: |
3075 | if (flags & SLAB_PANIC) | |
3076 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3077 | "order=%u offset=%u flags=%lx\n", | |
8a13a4cc | 3078 | s->name, (unsigned long)s->size, s->size, oo_order(s->oo), |
81819f0f | 3079 | s->offset, flags); |
278b1bb1 | 3080 | return -EINVAL; |
81819f0f | 3081 | } |
81819f0f | 3082 | |
33b12c38 CL |
3083 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3084 | const char *text) | |
3085 | { | |
3086 | #ifdef CONFIG_SLUB_DEBUG | |
3087 | void *addr = page_address(page); | |
3088 | void *p; | |
a5dd5c11 NK |
3089 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3090 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3091 | if (!map) |
3092 | return; | |
945cf2b6 | 3093 | slab_err(s, page, text, s->name); |
33b12c38 | 3094 | slab_lock(page); |
33b12c38 | 3095 | |
5f80b13a | 3096 | get_map(s, page, map); |
33b12c38 CL |
3097 | for_each_object(p, s, addr, page->objects) { |
3098 | ||
3099 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3100 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3101 | p, p - addr); | |
3102 | print_tracking(s, p); | |
3103 | } | |
3104 | } | |
3105 | slab_unlock(page); | |
bbd7d57b | 3106 | kfree(map); |
33b12c38 CL |
3107 | #endif |
3108 | } | |
3109 | ||
81819f0f | 3110 | /* |
599870b1 | 3111 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3112 | * This is called from kmem_cache_close(). We must be the last thread |
3113 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3114 | */ |
599870b1 | 3115 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3116 | { |
81819f0f CL |
3117 | struct page *page, *h; |
3118 | ||
33b12c38 | 3119 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3120 | if (!page->inuse) { |
5cc6eee8 | 3121 | remove_partial(n, page); |
81819f0f | 3122 | discard_slab(s, page); |
33b12c38 CL |
3123 | } else { |
3124 | list_slab_objects(s, page, | |
945cf2b6 | 3125 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3126 | } |
33b12c38 | 3127 | } |
81819f0f CL |
3128 | } |
3129 | ||
3130 | /* | |
672bba3a | 3131 | * Release all resources used by a slab cache. |
81819f0f | 3132 | */ |
0c710013 | 3133 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3134 | { |
3135 | int node; | |
3136 | ||
3137 | flush_all(s); | |
81819f0f | 3138 | /* Attempt to free all objects */ |
f64dc58c | 3139 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3140 | struct kmem_cache_node *n = get_node(s, node); |
3141 | ||
599870b1 CL |
3142 | free_partial(s, n); |
3143 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3144 | return 1; |
3145 | } | |
945cf2b6 | 3146 | free_percpu(s->cpu_slab); |
81819f0f CL |
3147 | free_kmem_cache_nodes(s); |
3148 | return 0; | |
3149 | } | |
3150 | ||
945cf2b6 | 3151 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3152 | { |
12c3667f | 3153 | int rc = kmem_cache_close(s); |
945cf2b6 | 3154 | |
12c3667f | 3155 | if (!rc) |
81819f0f | 3156 | sysfs_slab_remove(s); |
12c3667f CL |
3157 | |
3158 | return rc; | |
81819f0f | 3159 | } |
81819f0f CL |
3160 | |
3161 | /******************************************************************** | |
3162 | * Kmalloc subsystem | |
3163 | *******************************************************************/ | |
3164 | ||
51df1142 | 3165 | struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
3166 | EXPORT_SYMBOL(kmalloc_caches); |
3167 | ||
55136592 | 3168 | #ifdef CONFIG_ZONE_DMA |
51df1142 | 3169 | static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT]; |
55136592 CL |
3170 | #endif |
3171 | ||
81819f0f CL |
3172 | static int __init setup_slub_min_order(char *str) |
3173 | { | |
06428780 | 3174 | get_option(&str, &slub_min_order); |
81819f0f CL |
3175 | |
3176 | return 1; | |
3177 | } | |
3178 | ||
3179 | __setup("slub_min_order=", setup_slub_min_order); | |
3180 | ||
3181 | static int __init setup_slub_max_order(char *str) | |
3182 | { | |
06428780 | 3183 | get_option(&str, &slub_max_order); |
818cf590 | 3184 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3185 | |
3186 | return 1; | |
3187 | } | |
3188 | ||
3189 | __setup("slub_max_order=", setup_slub_max_order); | |
3190 | ||
3191 | static int __init setup_slub_min_objects(char *str) | |
3192 | { | |
06428780 | 3193 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3194 | |
3195 | return 1; | |
3196 | } | |
3197 | ||
3198 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3199 | ||
3200 | static int __init setup_slub_nomerge(char *str) | |
3201 | { | |
3202 | slub_nomerge = 1; | |
3203 | return 1; | |
3204 | } | |
3205 | ||
3206 | __setup("slub_nomerge", setup_slub_nomerge); | |
3207 | ||
f1b26339 CL |
3208 | /* |
3209 | * Conversion table for small slabs sizes / 8 to the index in the | |
3210 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
3211 | * of two cache sizes there. The size of larger slabs can be determined using | |
3212 | * fls. | |
3213 | */ | |
3214 | static s8 size_index[24] = { | |
3215 | 3, /* 8 */ | |
3216 | 4, /* 16 */ | |
3217 | 5, /* 24 */ | |
3218 | 5, /* 32 */ | |
3219 | 6, /* 40 */ | |
3220 | 6, /* 48 */ | |
3221 | 6, /* 56 */ | |
3222 | 6, /* 64 */ | |
3223 | 1, /* 72 */ | |
3224 | 1, /* 80 */ | |
3225 | 1, /* 88 */ | |
3226 | 1, /* 96 */ | |
3227 | 7, /* 104 */ | |
3228 | 7, /* 112 */ | |
3229 | 7, /* 120 */ | |
3230 | 7, /* 128 */ | |
3231 | 2, /* 136 */ | |
3232 | 2, /* 144 */ | |
3233 | 2, /* 152 */ | |
3234 | 2, /* 160 */ | |
3235 | 2, /* 168 */ | |
3236 | 2, /* 176 */ | |
3237 | 2, /* 184 */ | |
3238 | 2 /* 192 */ | |
3239 | }; | |
3240 | ||
acdfcd04 AK |
3241 | static inline int size_index_elem(size_t bytes) |
3242 | { | |
3243 | return (bytes - 1) / 8; | |
3244 | } | |
3245 | ||
81819f0f CL |
3246 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) |
3247 | { | |
f1b26339 | 3248 | int index; |
81819f0f | 3249 | |
f1b26339 CL |
3250 | if (size <= 192) { |
3251 | if (!size) | |
3252 | return ZERO_SIZE_PTR; | |
81819f0f | 3253 | |
acdfcd04 | 3254 | index = size_index[size_index_elem(size)]; |
aadb4bc4 | 3255 | } else |
f1b26339 | 3256 | index = fls(size - 1); |
81819f0f CL |
3257 | |
3258 | #ifdef CONFIG_ZONE_DMA | |
f1b26339 | 3259 | if (unlikely((flags & SLUB_DMA))) |
51df1142 | 3260 | return kmalloc_dma_caches[index]; |
f1b26339 | 3261 | |
81819f0f | 3262 | #endif |
51df1142 | 3263 | return kmalloc_caches[index]; |
81819f0f CL |
3264 | } |
3265 | ||
3266 | void *__kmalloc(size_t size, gfp_t flags) | |
3267 | { | |
aadb4bc4 | 3268 | struct kmem_cache *s; |
5b882be4 | 3269 | void *ret; |
81819f0f | 3270 | |
ffadd4d0 | 3271 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef | 3272 | return kmalloc_large(size, flags); |
aadb4bc4 CL |
3273 | |
3274 | s = get_slab(size, flags); | |
3275 | ||
3276 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3277 | return s; |
3278 | ||
2b847c3c | 3279 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3280 | |
ca2b84cb | 3281 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3282 | |
3283 | return ret; | |
81819f0f CL |
3284 | } |
3285 | EXPORT_SYMBOL(__kmalloc); | |
3286 | ||
5d1f57e4 | 3287 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3288 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3289 | { | |
b1eeab67 | 3290 | struct page *page; |
e4f7c0b4 | 3291 | void *ptr = NULL; |
f619cfe1 | 3292 | |
d79923fa | 3293 | flags |= __GFP_COMP | __GFP_NOTRACK | __GFP_KMEMCG; |
b1eeab67 | 3294 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3295 | if (page) |
e4f7c0b4 CM |
3296 | ptr = page_address(page); |
3297 | ||
3298 | kmemleak_alloc(ptr, size, 1, flags); | |
3299 | return ptr; | |
f619cfe1 CL |
3300 | } |
3301 | ||
81819f0f CL |
3302 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3303 | { | |
aadb4bc4 | 3304 | struct kmem_cache *s; |
5b882be4 | 3305 | void *ret; |
81819f0f | 3306 | |
057685cf | 3307 | if (unlikely(size > SLUB_MAX_SIZE)) { |
5b882be4 EGM |
3308 | ret = kmalloc_large_node(size, flags, node); |
3309 | ||
ca2b84cb EGM |
3310 | trace_kmalloc_node(_RET_IP_, ret, |
3311 | size, PAGE_SIZE << get_order(size), | |
3312 | flags, node); | |
5b882be4 EGM |
3313 | |
3314 | return ret; | |
3315 | } | |
aadb4bc4 CL |
3316 | |
3317 | s = get_slab(size, flags); | |
3318 | ||
3319 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3320 | return s; |
3321 | ||
2b847c3c | 3322 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3323 | |
ca2b84cb | 3324 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3325 | |
3326 | return ret; | |
81819f0f CL |
3327 | } |
3328 | EXPORT_SYMBOL(__kmalloc_node); | |
3329 | #endif | |
3330 | ||
3331 | size_t ksize(const void *object) | |
3332 | { | |
272c1d21 | 3333 | struct page *page; |
81819f0f | 3334 | |
ef8b4520 | 3335 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3336 | return 0; |
3337 | ||
294a80a8 | 3338 | page = virt_to_head_page(object); |
294a80a8 | 3339 | |
76994412 PE |
3340 | if (unlikely(!PageSlab(page))) { |
3341 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3342 | return PAGE_SIZE << compound_order(page); |
76994412 | 3343 | } |
81819f0f | 3344 | |
1b4f59e3 | 3345 | return slab_ksize(page->slab_cache); |
81819f0f | 3346 | } |
b1aabecd | 3347 | EXPORT_SYMBOL(ksize); |
81819f0f | 3348 | |
d18a90dd BG |
3349 | #ifdef CONFIG_SLUB_DEBUG |
3350 | bool verify_mem_not_deleted(const void *x) | |
3351 | { | |
3352 | struct page *page; | |
3353 | void *object = (void *)x; | |
3354 | unsigned long flags; | |
3355 | bool rv; | |
3356 | ||
3357 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3358 | return false; | |
3359 | ||
3360 | local_irq_save(flags); | |
3361 | ||
3362 | page = virt_to_head_page(x); | |
3363 | if (unlikely(!PageSlab(page))) { | |
3364 | /* maybe it was from stack? */ | |
3365 | rv = true; | |
3366 | goto out_unlock; | |
3367 | } | |
3368 | ||
3369 | slab_lock(page); | |
1b4f59e3 GC |
3370 | if (on_freelist(page->slab_cache, page, object)) { |
3371 | object_err(page->slab_cache, page, object, "Object is on free-list"); | |
d18a90dd BG |
3372 | rv = false; |
3373 | } else { | |
3374 | rv = true; | |
3375 | } | |
3376 | slab_unlock(page); | |
3377 | ||
3378 | out_unlock: | |
3379 | local_irq_restore(flags); | |
3380 | return rv; | |
3381 | } | |
3382 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3383 | #endif | |
3384 | ||
81819f0f CL |
3385 | void kfree(const void *x) |
3386 | { | |
81819f0f | 3387 | struct page *page; |
5bb983b0 | 3388 | void *object = (void *)x; |
81819f0f | 3389 | |
2121db74 PE |
3390 | trace_kfree(_RET_IP_, x); |
3391 | ||
2408c550 | 3392 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3393 | return; |
3394 | ||
b49af68f | 3395 | page = virt_to_head_page(x); |
aadb4bc4 | 3396 | if (unlikely(!PageSlab(page))) { |
0937502a | 3397 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3398 | kmemleak_free(x); |
d79923fa | 3399 | __free_memcg_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3400 | return; |
3401 | } | |
1b4f59e3 | 3402 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3403 | } |
3404 | EXPORT_SYMBOL(kfree); | |
3405 | ||
2086d26a | 3406 | /* |
672bba3a CL |
3407 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3408 | * the remaining slabs by the number of items in use. The slabs with the | |
3409 | * most items in use come first. New allocations will then fill those up | |
3410 | * and thus they can be removed from the partial lists. | |
3411 | * | |
3412 | * The slabs with the least items are placed last. This results in them | |
3413 | * being allocated from last increasing the chance that the last objects | |
3414 | * are freed in them. | |
2086d26a CL |
3415 | */ |
3416 | int kmem_cache_shrink(struct kmem_cache *s) | |
3417 | { | |
3418 | int node; | |
3419 | int i; | |
3420 | struct kmem_cache_node *n; | |
3421 | struct page *page; | |
3422 | struct page *t; | |
205ab99d | 3423 | int objects = oo_objects(s->max); |
2086d26a | 3424 | struct list_head *slabs_by_inuse = |
834f3d11 | 3425 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3426 | unsigned long flags; |
3427 | ||
3428 | if (!slabs_by_inuse) | |
3429 | return -ENOMEM; | |
3430 | ||
3431 | flush_all(s); | |
f64dc58c | 3432 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3433 | n = get_node(s, node); |
3434 | ||
3435 | if (!n->nr_partial) | |
3436 | continue; | |
3437 | ||
834f3d11 | 3438 | for (i = 0; i < objects; i++) |
2086d26a CL |
3439 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3440 | ||
3441 | spin_lock_irqsave(&n->list_lock, flags); | |
3442 | ||
3443 | /* | |
672bba3a | 3444 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3445 | * |
672bba3a CL |
3446 | * Note that concurrent frees may occur while we hold the |
3447 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3448 | */ |
3449 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3450 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3451 | if (!page->inuse) | |
3452 | n->nr_partial--; | |
2086d26a CL |
3453 | } |
3454 | ||
2086d26a | 3455 | /* |
672bba3a CL |
3456 | * Rebuild the partial list with the slabs filled up most |
3457 | * first and the least used slabs at the end. | |
2086d26a | 3458 | */ |
69cb8e6b | 3459 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3460 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3461 | ||
2086d26a | 3462 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3463 | |
3464 | /* Release empty slabs */ | |
3465 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3466 | discard_slab(s, page); | |
2086d26a CL |
3467 | } |
3468 | ||
3469 | kfree(slabs_by_inuse); | |
3470 | return 0; | |
3471 | } | |
3472 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3473 | ||
92a5bbc1 | 3474 | #if defined(CONFIG_MEMORY_HOTPLUG) |
b9049e23 YG |
3475 | static int slab_mem_going_offline_callback(void *arg) |
3476 | { | |
3477 | struct kmem_cache *s; | |
3478 | ||
18004c5d | 3479 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3480 | list_for_each_entry(s, &slab_caches, list) |
3481 | kmem_cache_shrink(s); | |
18004c5d | 3482 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3483 | |
3484 | return 0; | |
3485 | } | |
3486 | ||
3487 | static void slab_mem_offline_callback(void *arg) | |
3488 | { | |
3489 | struct kmem_cache_node *n; | |
3490 | struct kmem_cache *s; | |
3491 | struct memory_notify *marg = arg; | |
3492 | int offline_node; | |
3493 | ||
b9d5ab25 | 3494 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3495 | |
3496 | /* | |
3497 | * If the node still has available memory. we need kmem_cache_node | |
3498 | * for it yet. | |
3499 | */ | |
3500 | if (offline_node < 0) | |
3501 | return; | |
3502 | ||
18004c5d | 3503 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3504 | list_for_each_entry(s, &slab_caches, list) { |
3505 | n = get_node(s, offline_node); | |
3506 | if (n) { | |
3507 | /* | |
3508 | * if n->nr_slabs > 0, slabs still exist on the node | |
3509 | * that is going down. We were unable to free them, | |
c9404c9c | 3510 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3511 | * callback. So, we must fail. |
3512 | */ | |
0f389ec6 | 3513 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3514 | |
3515 | s->node[offline_node] = NULL; | |
8de66a0c | 3516 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3517 | } |
3518 | } | |
18004c5d | 3519 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3520 | } |
3521 | ||
3522 | static int slab_mem_going_online_callback(void *arg) | |
3523 | { | |
3524 | struct kmem_cache_node *n; | |
3525 | struct kmem_cache *s; | |
3526 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3527 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3528 | int ret = 0; |
3529 | ||
3530 | /* | |
3531 | * If the node's memory is already available, then kmem_cache_node is | |
3532 | * already created. Nothing to do. | |
3533 | */ | |
3534 | if (nid < 0) | |
3535 | return 0; | |
3536 | ||
3537 | /* | |
0121c619 | 3538 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3539 | * allocate a kmem_cache_node structure in order to bring the node |
3540 | * online. | |
3541 | */ | |
18004c5d | 3542 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3543 | list_for_each_entry(s, &slab_caches, list) { |
3544 | /* | |
3545 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3546 | * since memory is not yet available from the node that | |
3547 | * is brought up. | |
3548 | */ | |
8de66a0c | 3549 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3550 | if (!n) { |
3551 | ret = -ENOMEM; | |
3552 | goto out; | |
3553 | } | |
4053497d | 3554 | init_kmem_cache_node(n); |
b9049e23 YG |
3555 | s->node[nid] = n; |
3556 | } | |
3557 | out: | |
18004c5d | 3558 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3559 | return ret; |
3560 | } | |
3561 | ||
3562 | static int slab_memory_callback(struct notifier_block *self, | |
3563 | unsigned long action, void *arg) | |
3564 | { | |
3565 | int ret = 0; | |
3566 | ||
3567 | switch (action) { | |
3568 | case MEM_GOING_ONLINE: | |
3569 | ret = slab_mem_going_online_callback(arg); | |
3570 | break; | |
3571 | case MEM_GOING_OFFLINE: | |
3572 | ret = slab_mem_going_offline_callback(arg); | |
3573 | break; | |
3574 | case MEM_OFFLINE: | |
3575 | case MEM_CANCEL_ONLINE: | |
3576 | slab_mem_offline_callback(arg); | |
3577 | break; | |
3578 | case MEM_ONLINE: | |
3579 | case MEM_CANCEL_OFFLINE: | |
3580 | break; | |
3581 | } | |
dc19f9db KH |
3582 | if (ret) |
3583 | ret = notifier_from_errno(ret); | |
3584 | else | |
3585 | ret = NOTIFY_OK; | |
b9049e23 YG |
3586 | return ret; |
3587 | } | |
3588 | ||
3589 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
3590 | ||
81819f0f CL |
3591 | /******************************************************************** |
3592 | * Basic setup of slabs | |
3593 | *******************************************************************/ | |
3594 | ||
51df1142 CL |
3595 | /* |
3596 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3597 | * the page allocator. Allocate them properly then fix up the pointers |
3598 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3599 | */ |
3600 | ||
dffb4d60 | 3601 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3602 | { |
3603 | int node; | |
dffb4d60 | 3604 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
51df1142 | 3605 | |
dffb4d60 | 3606 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 CL |
3607 | |
3608 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
3609 | struct kmem_cache_node *n = get_node(s, node); | |
3610 | struct page *p; | |
3611 | ||
3612 | if (n) { | |
3613 | list_for_each_entry(p, &n->partial, lru) | |
1b4f59e3 | 3614 | p->slab_cache = s; |
51df1142 | 3615 | |
607bf324 | 3616 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 | 3617 | list_for_each_entry(p, &n->full, lru) |
1b4f59e3 | 3618 | p->slab_cache = s; |
51df1142 CL |
3619 | #endif |
3620 | } | |
3621 | } | |
dffb4d60 CL |
3622 | list_add(&s->list, &slab_caches); |
3623 | return s; | |
51df1142 CL |
3624 | } |
3625 | ||
81819f0f CL |
3626 | void __init kmem_cache_init(void) |
3627 | { | |
dffb4d60 CL |
3628 | static __initdata struct kmem_cache boot_kmem_cache, |
3629 | boot_kmem_cache_node; | |
81819f0f | 3630 | int i; |
dffb4d60 | 3631 | int caches = 2; |
51df1142 | 3632 | |
fc8d8620 SG |
3633 | if (debug_guardpage_minorder()) |
3634 | slub_max_order = 0; | |
3635 | ||
dffb4d60 CL |
3636 | kmem_cache_node = &boot_kmem_cache_node; |
3637 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3638 | |
dffb4d60 CL |
3639 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3640 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3641 | |
0c40ba4f | 3642 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
3643 | |
3644 | /* Able to allocate the per node structures */ | |
3645 | slab_state = PARTIAL; | |
3646 | ||
dffb4d60 CL |
3647 | create_boot_cache(kmem_cache, "kmem_cache", |
3648 | offsetof(struct kmem_cache, node) + | |
3649 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3650 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3651 | |
dffb4d60 | 3652 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3653 | |
51df1142 CL |
3654 | /* |
3655 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3656 | * kmem_cache_node is separately allocated so no need to | |
3657 | * update any list pointers. | |
3658 | */ | |
dffb4d60 | 3659 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3660 | |
3661 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f1b26339 CL |
3662 | |
3663 | /* | |
3664 | * Patch up the size_index table if we have strange large alignment | |
3665 | * requirements for the kmalloc array. This is only the case for | |
6446faa2 | 3666 | * MIPS it seems. The standard arches will not generate any code here. |
f1b26339 CL |
3667 | * |
3668 | * Largest permitted alignment is 256 bytes due to the way we | |
3669 | * handle the index determination for the smaller caches. | |
3670 | * | |
3671 | * Make sure that nothing crazy happens if someone starts tinkering | |
3672 | * around with ARCH_KMALLOC_MINALIGN | |
3673 | */ | |
3674 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
3675 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
3676 | ||
acdfcd04 AK |
3677 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { |
3678 | int elem = size_index_elem(i); | |
3679 | if (elem >= ARRAY_SIZE(size_index)) | |
3680 | break; | |
3681 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
3682 | } | |
f1b26339 | 3683 | |
acdfcd04 AK |
3684 | if (KMALLOC_MIN_SIZE == 64) { |
3685 | /* | |
3686 | * The 96 byte size cache is not used if the alignment | |
3687 | * is 64 byte. | |
3688 | */ | |
3689 | for (i = 64 + 8; i <= 96; i += 8) | |
3690 | size_index[size_index_elem(i)] = 7; | |
3691 | } else if (KMALLOC_MIN_SIZE == 128) { | |
41d54d3b CL |
3692 | /* |
3693 | * The 192 byte sized cache is not used if the alignment | |
3694 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
3695 | * instead. | |
3696 | */ | |
3697 | for (i = 128 + 8; i <= 192; i += 8) | |
acdfcd04 | 3698 | size_index[size_index_elem(i)] = 8; |
41d54d3b CL |
3699 | } |
3700 | ||
51df1142 CL |
3701 | /* Caches that are not of the two-to-the-power-of size */ |
3702 | if (KMALLOC_MIN_SIZE <= 32) { | |
3703 | kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0); | |
3704 | caches++; | |
3705 | } | |
3706 | ||
3707 | if (KMALLOC_MIN_SIZE <= 64) { | |
3708 | kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0); | |
3709 | caches++; | |
3710 | } | |
3711 | ||
3712 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { | |
3713 | kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0); | |
3714 | caches++; | |
3715 | } | |
3716 | ||
81819f0f CL |
3717 | slab_state = UP; |
3718 | ||
3719 | /* Provide the correct kmalloc names now that the caches are up */ | |
84c1cf62 PE |
3720 | if (KMALLOC_MIN_SIZE <= 32) { |
3721 | kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT); | |
3722 | BUG_ON(!kmalloc_caches[1]->name); | |
3723 | } | |
3724 | ||
3725 | if (KMALLOC_MIN_SIZE <= 64) { | |
3726 | kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT); | |
3727 | BUG_ON(!kmalloc_caches[2]->name); | |
3728 | } | |
3729 | ||
d7278bd7 CL |
3730 | for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) { |
3731 | char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i); | |
3732 | ||
3733 | BUG_ON(!s); | |
51df1142 | 3734 | kmalloc_caches[i]->name = s; |
d7278bd7 | 3735 | } |
81819f0f CL |
3736 | |
3737 | #ifdef CONFIG_SMP | |
3738 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3739 | #endif |
81819f0f | 3740 | |
55136592 | 3741 | #ifdef CONFIG_ZONE_DMA |
51df1142 CL |
3742 | for (i = 0; i < SLUB_PAGE_SHIFT; i++) { |
3743 | struct kmem_cache *s = kmalloc_caches[i]; | |
55136592 | 3744 | |
51df1142 | 3745 | if (s && s->size) { |
55136592 | 3746 | char *name = kasprintf(GFP_NOWAIT, |
3b0efdfa | 3747 | "dma-kmalloc-%d", s->object_size); |
55136592 CL |
3748 | |
3749 | BUG_ON(!name); | |
51df1142 | 3750 | kmalloc_dma_caches[i] = create_kmalloc_cache(name, |
3b0efdfa | 3751 | s->object_size, SLAB_CACHE_DMA); |
55136592 CL |
3752 | } |
3753 | } | |
3754 | #endif | |
3adbefee IM |
3755 | printk(KERN_INFO |
3756 | "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
4b356be0 CL |
3757 | " CPUs=%d, Nodes=%d\n", |
3758 | caches, cache_line_size(), | |
81819f0f CL |
3759 | slub_min_order, slub_max_order, slub_min_objects, |
3760 | nr_cpu_ids, nr_node_ids); | |
3761 | } | |
3762 | ||
7e85ee0c PE |
3763 | void __init kmem_cache_init_late(void) |
3764 | { | |
7e85ee0c PE |
3765 | } |
3766 | ||
81819f0f CL |
3767 | /* |
3768 | * Find a mergeable slab cache | |
3769 | */ | |
3770 | static int slab_unmergeable(struct kmem_cache *s) | |
3771 | { | |
3772 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3773 | return 1; | |
3774 | ||
c59def9f | 3775 | if (s->ctor) |
81819f0f CL |
3776 | return 1; |
3777 | ||
8ffa6875 CL |
3778 | /* |
3779 | * We may have set a slab to be unmergeable during bootstrap. | |
3780 | */ | |
3781 | if (s->refcount < 0) | |
3782 | return 1; | |
3783 | ||
81819f0f CL |
3784 | return 0; |
3785 | } | |
3786 | ||
2633d7a0 | 3787 | static struct kmem_cache *find_mergeable(struct mem_cgroup *memcg, size_t size, |
ba0268a8 | 3788 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3789 | void (*ctor)(void *)) |
81819f0f | 3790 | { |
5b95a4ac | 3791 | struct kmem_cache *s; |
81819f0f CL |
3792 | |
3793 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3794 | return NULL; | |
3795 | ||
c59def9f | 3796 | if (ctor) |
81819f0f CL |
3797 | return NULL; |
3798 | ||
3799 | size = ALIGN(size, sizeof(void *)); | |
3800 | align = calculate_alignment(flags, align, size); | |
3801 | size = ALIGN(size, align); | |
ba0268a8 | 3802 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3803 | |
5b95a4ac | 3804 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3805 | if (slab_unmergeable(s)) |
3806 | continue; | |
3807 | ||
3808 | if (size > s->size) | |
3809 | continue; | |
3810 | ||
ba0268a8 | 3811 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3812 | continue; |
3813 | /* | |
3814 | * Check if alignment is compatible. | |
3815 | * Courtesy of Adrian Drzewiecki | |
3816 | */ | |
06428780 | 3817 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3818 | continue; |
3819 | ||
3820 | if (s->size - size >= sizeof(void *)) | |
3821 | continue; | |
3822 | ||
2633d7a0 GC |
3823 | if (!cache_match_memcg(s, memcg)) |
3824 | continue; | |
3825 | ||
81819f0f CL |
3826 | return s; |
3827 | } | |
3828 | return NULL; | |
3829 | } | |
3830 | ||
2633d7a0 GC |
3831 | struct kmem_cache * |
3832 | __kmem_cache_alias(struct mem_cgroup *memcg, const char *name, size_t size, | |
3833 | size_t align, unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3834 | { |
3835 | struct kmem_cache *s; | |
3836 | ||
2633d7a0 | 3837 | s = find_mergeable(memcg, size, align, flags, name, ctor); |
81819f0f CL |
3838 | if (s) { |
3839 | s->refcount++; | |
3840 | /* | |
3841 | * Adjust the object sizes so that we clear | |
3842 | * the complete object on kzalloc. | |
3843 | */ | |
3b0efdfa | 3844 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3845 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3846 | |
7b8f3b66 | 3847 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3848 | s->refcount--; |
cbb79694 | 3849 | s = NULL; |
7b8f3b66 | 3850 | } |
a0e1d1be | 3851 | } |
6446faa2 | 3852 | |
cbb79694 CL |
3853 | return s; |
3854 | } | |
84c1cf62 | 3855 | |
8a13a4cc | 3856 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3857 | { |
aac3a166 PE |
3858 | int err; |
3859 | ||
3860 | err = kmem_cache_open(s, flags); | |
3861 | if (err) | |
3862 | return err; | |
20cea968 | 3863 | |
45530c44 CL |
3864 | /* Mutex is not taken during early boot */ |
3865 | if (slab_state <= UP) | |
3866 | return 0; | |
3867 | ||
aac3a166 PE |
3868 | mutex_unlock(&slab_mutex); |
3869 | err = sysfs_slab_add(s); | |
3870 | mutex_lock(&slab_mutex); | |
20cea968 | 3871 | |
aac3a166 PE |
3872 | if (err) |
3873 | kmem_cache_close(s); | |
20cea968 | 3874 | |
aac3a166 | 3875 | return err; |
81819f0f | 3876 | } |
81819f0f | 3877 | |
81819f0f | 3878 | #ifdef CONFIG_SMP |
81819f0f | 3879 | /* |
672bba3a CL |
3880 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3881 | * necessary. | |
81819f0f CL |
3882 | */ |
3883 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3884 | unsigned long action, void *hcpu) | |
3885 | { | |
3886 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3887 | struct kmem_cache *s; |
3888 | unsigned long flags; | |
81819f0f CL |
3889 | |
3890 | switch (action) { | |
3891 | case CPU_UP_CANCELED: | |
8bb78442 | 3892 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3893 | case CPU_DEAD: |
8bb78442 | 3894 | case CPU_DEAD_FROZEN: |
18004c5d | 3895 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3896 | list_for_each_entry(s, &slab_caches, list) { |
3897 | local_irq_save(flags); | |
3898 | __flush_cpu_slab(s, cpu); | |
3899 | local_irq_restore(flags); | |
3900 | } | |
18004c5d | 3901 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3902 | break; |
3903 | default: | |
3904 | break; | |
3905 | } | |
3906 | return NOTIFY_OK; | |
3907 | } | |
3908 | ||
06428780 | 3909 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3910 | .notifier_call = slab_cpuup_callback |
06428780 | 3911 | }; |
81819f0f CL |
3912 | |
3913 | #endif | |
3914 | ||
ce71e27c | 3915 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3916 | { |
aadb4bc4 | 3917 | struct kmem_cache *s; |
94b528d0 | 3918 | void *ret; |
aadb4bc4 | 3919 | |
ffadd4d0 | 3920 | if (unlikely(size > SLUB_MAX_SIZE)) |
eada35ef PE |
3921 | return kmalloc_large(size, gfpflags); |
3922 | ||
aadb4bc4 | 3923 | s = get_slab(size, gfpflags); |
81819f0f | 3924 | |
2408c550 | 3925 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3926 | return s; |
81819f0f | 3927 | |
2b847c3c | 3928 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3929 | |
25985edc | 3930 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3931 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3932 | |
3933 | return ret; | |
81819f0f CL |
3934 | } |
3935 | ||
5d1f57e4 | 3936 | #ifdef CONFIG_NUMA |
81819f0f | 3937 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3938 | int node, unsigned long caller) |
81819f0f | 3939 | { |
aadb4bc4 | 3940 | struct kmem_cache *s; |
94b528d0 | 3941 | void *ret; |
aadb4bc4 | 3942 | |
d3e14aa3 XF |
3943 | if (unlikely(size > SLUB_MAX_SIZE)) { |
3944 | ret = kmalloc_large_node(size, gfpflags, node); | |
3945 | ||
3946 | trace_kmalloc_node(caller, ret, | |
3947 | size, PAGE_SIZE << get_order(size), | |
3948 | gfpflags, node); | |
3949 | ||
3950 | return ret; | |
3951 | } | |
eada35ef | 3952 | |
aadb4bc4 | 3953 | s = get_slab(size, gfpflags); |
81819f0f | 3954 | |
2408c550 | 3955 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3956 | return s; |
81819f0f | 3957 | |
2b847c3c | 3958 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3959 | |
25985edc | 3960 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3961 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3962 | |
3963 | return ret; | |
81819f0f | 3964 | } |
5d1f57e4 | 3965 | #endif |
81819f0f | 3966 | |
ab4d5ed5 | 3967 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3968 | static int count_inuse(struct page *page) |
3969 | { | |
3970 | return page->inuse; | |
3971 | } | |
3972 | ||
3973 | static int count_total(struct page *page) | |
3974 | { | |
3975 | return page->objects; | |
3976 | } | |
ab4d5ed5 | 3977 | #endif |
205ab99d | 3978 | |
ab4d5ed5 | 3979 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3980 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3981 | unsigned long *map) | |
53e15af0 CL |
3982 | { |
3983 | void *p; | |
a973e9dd | 3984 | void *addr = page_address(page); |
53e15af0 CL |
3985 | |
3986 | if (!check_slab(s, page) || | |
3987 | !on_freelist(s, page, NULL)) | |
3988 | return 0; | |
3989 | ||
3990 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3991 | bitmap_zero(map, page->objects); |
53e15af0 | 3992 | |
5f80b13a CL |
3993 | get_map(s, page, map); |
3994 | for_each_object(p, s, addr, page->objects) { | |
3995 | if (test_bit(slab_index(p, s, addr), map)) | |
3996 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3997 | return 0; | |
53e15af0 CL |
3998 | } |
3999 | ||
224a88be | 4000 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4001 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4002 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4003 | return 0; |
4004 | return 1; | |
4005 | } | |
4006 | ||
434e245d CL |
4007 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4008 | unsigned long *map) | |
53e15af0 | 4009 | { |
881db7fb CL |
4010 | slab_lock(page); |
4011 | validate_slab(s, page, map); | |
4012 | slab_unlock(page); | |
53e15af0 CL |
4013 | } |
4014 | ||
434e245d CL |
4015 | static int validate_slab_node(struct kmem_cache *s, |
4016 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4017 | { |
4018 | unsigned long count = 0; | |
4019 | struct page *page; | |
4020 | unsigned long flags; | |
4021 | ||
4022 | spin_lock_irqsave(&n->list_lock, flags); | |
4023 | ||
4024 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4025 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4026 | count++; |
4027 | } | |
4028 | if (count != n->nr_partial) | |
4029 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
4030 | "counter=%ld\n", s->name, count, n->nr_partial); | |
4031 | ||
4032 | if (!(s->flags & SLAB_STORE_USER)) | |
4033 | goto out; | |
4034 | ||
4035 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4036 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4037 | count++; |
4038 | } | |
4039 | if (count != atomic_long_read(&n->nr_slabs)) | |
4040 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
4041 | "counter=%ld\n", s->name, count, | |
4042 | atomic_long_read(&n->nr_slabs)); | |
4043 | ||
4044 | out: | |
4045 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4046 | return count; | |
4047 | } | |
4048 | ||
434e245d | 4049 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4050 | { |
4051 | int node; | |
4052 | unsigned long count = 0; | |
205ab99d | 4053 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
4054 | sizeof(unsigned long), GFP_KERNEL); |
4055 | ||
4056 | if (!map) | |
4057 | return -ENOMEM; | |
53e15af0 CL |
4058 | |
4059 | flush_all(s); | |
f64dc58c | 4060 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
4061 | struct kmem_cache_node *n = get_node(s, node); |
4062 | ||
434e245d | 4063 | count += validate_slab_node(s, n, map); |
53e15af0 | 4064 | } |
434e245d | 4065 | kfree(map); |
53e15af0 CL |
4066 | return count; |
4067 | } | |
88a420e4 | 4068 | /* |
672bba3a | 4069 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4070 | * and freed. |
4071 | */ | |
4072 | ||
4073 | struct location { | |
4074 | unsigned long count; | |
ce71e27c | 4075 | unsigned long addr; |
45edfa58 CL |
4076 | long long sum_time; |
4077 | long min_time; | |
4078 | long max_time; | |
4079 | long min_pid; | |
4080 | long max_pid; | |
174596a0 | 4081 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4082 | nodemask_t nodes; |
88a420e4 CL |
4083 | }; |
4084 | ||
4085 | struct loc_track { | |
4086 | unsigned long max; | |
4087 | unsigned long count; | |
4088 | struct location *loc; | |
4089 | }; | |
4090 | ||
4091 | static void free_loc_track(struct loc_track *t) | |
4092 | { | |
4093 | if (t->max) | |
4094 | free_pages((unsigned long)t->loc, | |
4095 | get_order(sizeof(struct location) * t->max)); | |
4096 | } | |
4097 | ||
68dff6a9 | 4098 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4099 | { |
4100 | struct location *l; | |
4101 | int order; | |
4102 | ||
88a420e4 CL |
4103 | order = get_order(sizeof(struct location) * max); |
4104 | ||
68dff6a9 | 4105 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4106 | if (!l) |
4107 | return 0; | |
4108 | ||
4109 | if (t->count) { | |
4110 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4111 | free_loc_track(t); | |
4112 | } | |
4113 | t->max = max; | |
4114 | t->loc = l; | |
4115 | return 1; | |
4116 | } | |
4117 | ||
4118 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4119 | const struct track *track) |
88a420e4 CL |
4120 | { |
4121 | long start, end, pos; | |
4122 | struct location *l; | |
ce71e27c | 4123 | unsigned long caddr; |
45edfa58 | 4124 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4125 | |
4126 | start = -1; | |
4127 | end = t->count; | |
4128 | ||
4129 | for ( ; ; ) { | |
4130 | pos = start + (end - start + 1) / 2; | |
4131 | ||
4132 | /* | |
4133 | * There is nothing at "end". If we end up there | |
4134 | * we need to add something to before end. | |
4135 | */ | |
4136 | if (pos == end) | |
4137 | break; | |
4138 | ||
4139 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4140 | if (track->addr == caddr) { |
4141 | ||
4142 | l = &t->loc[pos]; | |
4143 | l->count++; | |
4144 | if (track->when) { | |
4145 | l->sum_time += age; | |
4146 | if (age < l->min_time) | |
4147 | l->min_time = age; | |
4148 | if (age > l->max_time) | |
4149 | l->max_time = age; | |
4150 | ||
4151 | if (track->pid < l->min_pid) | |
4152 | l->min_pid = track->pid; | |
4153 | if (track->pid > l->max_pid) | |
4154 | l->max_pid = track->pid; | |
4155 | ||
174596a0 RR |
4156 | cpumask_set_cpu(track->cpu, |
4157 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4158 | } |
4159 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4160 | return 1; |
4161 | } | |
4162 | ||
45edfa58 | 4163 | if (track->addr < caddr) |
88a420e4 CL |
4164 | end = pos; |
4165 | else | |
4166 | start = pos; | |
4167 | } | |
4168 | ||
4169 | /* | |
672bba3a | 4170 | * Not found. Insert new tracking element. |
88a420e4 | 4171 | */ |
68dff6a9 | 4172 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4173 | return 0; |
4174 | ||
4175 | l = t->loc + pos; | |
4176 | if (pos < t->count) | |
4177 | memmove(l + 1, l, | |
4178 | (t->count - pos) * sizeof(struct location)); | |
4179 | t->count++; | |
4180 | l->count = 1; | |
45edfa58 CL |
4181 | l->addr = track->addr; |
4182 | l->sum_time = age; | |
4183 | l->min_time = age; | |
4184 | l->max_time = age; | |
4185 | l->min_pid = track->pid; | |
4186 | l->max_pid = track->pid; | |
174596a0 RR |
4187 | cpumask_clear(to_cpumask(l->cpus)); |
4188 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4189 | nodes_clear(l->nodes); |
4190 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4191 | return 1; |
4192 | } | |
4193 | ||
4194 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4195 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4196 | unsigned long *map) |
88a420e4 | 4197 | { |
a973e9dd | 4198 | void *addr = page_address(page); |
88a420e4 CL |
4199 | void *p; |
4200 | ||
39b26464 | 4201 | bitmap_zero(map, page->objects); |
5f80b13a | 4202 | get_map(s, page, map); |
88a420e4 | 4203 | |
224a88be | 4204 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4205 | if (!test_bit(slab_index(p, s, addr), map)) |
4206 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4207 | } |
4208 | ||
4209 | static int list_locations(struct kmem_cache *s, char *buf, | |
4210 | enum track_item alloc) | |
4211 | { | |
e374d483 | 4212 | int len = 0; |
88a420e4 | 4213 | unsigned long i; |
68dff6a9 | 4214 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4215 | int node; |
bbd7d57b ED |
4216 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4217 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4218 | |
bbd7d57b ED |
4219 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4220 | GFP_TEMPORARY)) { | |
4221 | kfree(map); | |
68dff6a9 | 4222 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4223 | } |
88a420e4 CL |
4224 | /* Push back cpu slabs */ |
4225 | flush_all(s); | |
4226 | ||
f64dc58c | 4227 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4228 | struct kmem_cache_node *n = get_node(s, node); |
4229 | unsigned long flags; | |
4230 | struct page *page; | |
4231 | ||
9e86943b | 4232 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4233 | continue; |
4234 | ||
4235 | spin_lock_irqsave(&n->list_lock, flags); | |
4236 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4237 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4238 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4239 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4240 | spin_unlock_irqrestore(&n->list_lock, flags); |
4241 | } | |
4242 | ||
4243 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4244 | struct location *l = &t.loc[i]; |
88a420e4 | 4245 | |
9c246247 | 4246 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4247 | break; |
e374d483 | 4248 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4249 | |
4250 | if (l->addr) | |
62c70bce | 4251 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4252 | else |
e374d483 | 4253 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4254 | |
4255 | if (l->sum_time != l->min_time) { | |
e374d483 | 4256 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4257 | l->min_time, |
4258 | (long)div_u64(l->sum_time, l->count), | |
4259 | l->max_time); | |
45edfa58 | 4260 | } else |
e374d483 | 4261 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4262 | l->min_time); |
4263 | ||
4264 | if (l->min_pid != l->max_pid) | |
e374d483 | 4265 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4266 | l->min_pid, l->max_pid); |
4267 | else | |
e374d483 | 4268 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4269 | l->min_pid); |
4270 | ||
174596a0 RR |
4271 | if (num_online_cpus() > 1 && |
4272 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4273 | len < PAGE_SIZE - 60) { |
4274 | len += sprintf(buf + len, " cpus="); | |
4275 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4276 | to_cpumask(l->cpus)); |
45edfa58 CL |
4277 | } |
4278 | ||
62bc62a8 | 4279 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4280 | len < PAGE_SIZE - 60) { |
4281 | len += sprintf(buf + len, " nodes="); | |
4282 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4283 | l->nodes); |
4284 | } | |
4285 | ||
e374d483 | 4286 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4287 | } |
4288 | ||
4289 | free_loc_track(&t); | |
bbd7d57b | 4290 | kfree(map); |
88a420e4 | 4291 | if (!t.count) |
e374d483 HH |
4292 | len += sprintf(buf, "No data\n"); |
4293 | return len; | |
88a420e4 | 4294 | } |
ab4d5ed5 | 4295 | #endif |
88a420e4 | 4296 | |
a5a84755 CL |
4297 | #ifdef SLUB_RESILIENCY_TEST |
4298 | static void resiliency_test(void) | |
4299 | { | |
4300 | u8 *p; | |
4301 | ||
4302 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10); | |
4303 | ||
4304 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4305 | printk(KERN_ERR "-----------------------\n"); | |
4306 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4307 | ||
4308 | p = kzalloc(16, GFP_KERNEL); | |
4309 | p[16] = 0x12; | |
4310 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4311 | " 0x12->0x%p\n\n", p + 16); | |
4312 | ||
4313 | validate_slab_cache(kmalloc_caches[4]); | |
4314 | ||
4315 | /* Hmmm... The next two are dangerous */ | |
4316 | p = kzalloc(32, GFP_KERNEL); | |
4317 | p[32 + sizeof(void *)] = 0x34; | |
4318 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4319 | " 0x34 -> -0x%p\n", p); | |
4320 | printk(KERN_ERR | |
4321 | "If allocated object is overwritten then not detectable\n\n"); | |
4322 | ||
4323 | validate_slab_cache(kmalloc_caches[5]); | |
4324 | p = kzalloc(64, GFP_KERNEL); | |
4325 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4326 | *p = 0x56; | |
4327 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4328 | p); | |
4329 | printk(KERN_ERR | |
4330 | "If allocated object is overwritten then not detectable\n\n"); | |
4331 | validate_slab_cache(kmalloc_caches[6]); | |
4332 | ||
4333 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4334 | p = kzalloc(128, GFP_KERNEL); | |
4335 | kfree(p); | |
4336 | *p = 0x78; | |
4337 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4338 | validate_slab_cache(kmalloc_caches[7]); | |
4339 | ||
4340 | p = kzalloc(256, GFP_KERNEL); | |
4341 | kfree(p); | |
4342 | p[50] = 0x9a; | |
4343 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4344 | p); | |
4345 | validate_slab_cache(kmalloc_caches[8]); | |
4346 | ||
4347 | p = kzalloc(512, GFP_KERNEL); | |
4348 | kfree(p); | |
4349 | p[512] = 0xab; | |
4350 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4351 | validate_slab_cache(kmalloc_caches[9]); | |
4352 | } | |
4353 | #else | |
4354 | #ifdef CONFIG_SYSFS | |
4355 | static void resiliency_test(void) {}; | |
4356 | #endif | |
4357 | #endif | |
4358 | ||
ab4d5ed5 | 4359 | #ifdef CONFIG_SYSFS |
81819f0f | 4360 | enum slab_stat_type { |
205ab99d CL |
4361 | SL_ALL, /* All slabs */ |
4362 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4363 | SL_CPU, /* Only slabs used for cpu caches */ | |
4364 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4365 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4366 | }; |
4367 | ||
205ab99d | 4368 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4369 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4370 | #define SO_CPU (1 << SL_CPU) | |
4371 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4372 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4373 | |
62e5c4b4 CG |
4374 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4375 | char *buf, unsigned long flags) | |
81819f0f CL |
4376 | { |
4377 | unsigned long total = 0; | |
81819f0f CL |
4378 | int node; |
4379 | int x; | |
4380 | unsigned long *nodes; | |
4381 | unsigned long *per_cpu; | |
4382 | ||
4383 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4384 | if (!nodes) |
4385 | return -ENOMEM; | |
81819f0f CL |
4386 | per_cpu = nodes + nr_node_ids; |
4387 | ||
205ab99d CL |
4388 | if (flags & SO_CPU) { |
4389 | int cpu; | |
81819f0f | 4390 | |
205ab99d | 4391 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4392 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4393 | int node; |
49e22585 | 4394 | struct page *page; |
dfb4f096 | 4395 | |
bc6697d8 | 4396 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4397 | if (!page) |
4398 | continue; | |
205ab99d | 4399 | |
ec3ab083 CL |
4400 | node = page_to_nid(page); |
4401 | if (flags & SO_TOTAL) | |
4402 | x = page->objects; | |
4403 | else if (flags & SO_OBJECTS) | |
4404 | x = page->inuse; | |
4405 | else | |
4406 | x = 1; | |
49e22585 | 4407 | |
ec3ab083 CL |
4408 | total += x; |
4409 | nodes[node] += x; | |
4410 | ||
4411 | page = ACCESS_ONCE(c->partial); | |
49e22585 CL |
4412 | if (page) { |
4413 | x = page->pobjects; | |
bc6697d8 ED |
4414 | total += x; |
4415 | nodes[node] += x; | |
49e22585 | 4416 | } |
ec3ab083 | 4417 | |
bc6697d8 | 4418 | per_cpu[node]++; |
81819f0f CL |
4419 | } |
4420 | } | |
4421 | ||
04d94879 | 4422 | lock_memory_hotplug(); |
ab4d5ed5 | 4423 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4424 | if (flags & SO_ALL) { |
4425 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4426 | struct kmem_cache_node *n = get_node(s, node); | |
4427 | ||
4428 | if (flags & SO_TOTAL) | |
4429 | x = atomic_long_read(&n->total_objects); | |
4430 | else if (flags & SO_OBJECTS) | |
4431 | x = atomic_long_read(&n->total_objects) - | |
4432 | count_partial(n, count_free); | |
81819f0f | 4433 | |
81819f0f | 4434 | else |
205ab99d | 4435 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4436 | total += x; |
4437 | nodes[node] += x; | |
4438 | } | |
4439 | ||
ab4d5ed5 CL |
4440 | } else |
4441 | #endif | |
4442 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4443 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4444 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4445 | |
205ab99d CL |
4446 | if (flags & SO_TOTAL) |
4447 | x = count_partial(n, count_total); | |
4448 | else if (flags & SO_OBJECTS) | |
4449 | x = count_partial(n, count_inuse); | |
81819f0f | 4450 | else |
205ab99d | 4451 | x = n->nr_partial; |
81819f0f CL |
4452 | total += x; |
4453 | nodes[node] += x; | |
4454 | } | |
4455 | } | |
81819f0f CL |
4456 | x = sprintf(buf, "%lu", total); |
4457 | #ifdef CONFIG_NUMA | |
f64dc58c | 4458 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4459 | if (nodes[node]) |
4460 | x += sprintf(buf + x, " N%d=%lu", | |
4461 | node, nodes[node]); | |
4462 | #endif | |
04d94879 | 4463 | unlock_memory_hotplug(); |
81819f0f CL |
4464 | kfree(nodes); |
4465 | return x + sprintf(buf + x, "\n"); | |
4466 | } | |
4467 | ||
ab4d5ed5 | 4468 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4469 | static int any_slab_objects(struct kmem_cache *s) |
4470 | { | |
4471 | int node; | |
81819f0f | 4472 | |
dfb4f096 | 4473 | for_each_online_node(node) { |
81819f0f CL |
4474 | struct kmem_cache_node *n = get_node(s, node); |
4475 | ||
dfb4f096 CL |
4476 | if (!n) |
4477 | continue; | |
4478 | ||
4ea33e2d | 4479 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4480 | return 1; |
4481 | } | |
4482 | return 0; | |
4483 | } | |
ab4d5ed5 | 4484 | #endif |
81819f0f CL |
4485 | |
4486 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4487 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4488 | |
4489 | struct slab_attribute { | |
4490 | struct attribute attr; | |
4491 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4492 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4493 | }; | |
4494 | ||
4495 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4496 | static struct slab_attribute _name##_attr = \ |
4497 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4498 | |
4499 | #define SLAB_ATTR(_name) \ | |
4500 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4501 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4502 | |
81819f0f CL |
4503 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4504 | { | |
4505 | return sprintf(buf, "%d\n", s->size); | |
4506 | } | |
4507 | SLAB_ATTR_RO(slab_size); | |
4508 | ||
4509 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4510 | { | |
4511 | return sprintf(buf, "%d\n", s->align); | |
4512 | } | |
4513 | SLAB_ATTR_RO(align); | |
4514 | ||
4515 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4516 | { | |
3b0efdfa | 4517 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4518 | } |
4519 | SLAB_ATTR_RO(object_size); | |
4520 | ||
4521 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4522 | { | |
834f3d11 | 4523 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4524 | } |
4525 | SLAB_ATTR_RO(objs_per_slab); | |
4526 | ||
06b285dc CL |
4527 | static ssize_t order_store(struct kmem_cache *s, |
4528 | const char *buf, size_t length) | |
4529 | { | |
0121c619 CL |
4530 | unsigned long order; |
4531 | int err; | |
4532 | ||
4533 | err = strict_strtoul(buf, 10, &order); | |
4534 | if (err) | |
4535 | return err; | |
06b285dc CL |
4536 | |
4537 | if (order > slub_max_order || order < slub_min_order) | |
4538 | return -EINVAL; | |
4539 | ||
4540 | calculate_sizes(s, order); | |
4541 | return length; | |
4542 | } | |
4543 | ||
81819f0f CL |
4544 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4545 | { | |
834f3d11 | 4546 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4547 | } |
06b285dc | 4548 | SLAB_ATTR(order); |
81819f0f | 4549 | |
73d342b1 DR |
4550 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4551 | { | |
4552 | return sprintf(buf, "%lu\n", s->min_partial); | |
4553 | } | |
4554 | ||
4555 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4556 | size_t length) | |
4557 | { | |
4558 | unsigned long min; | |
4559 | int err; | |
4560 | ||
4561 | err = strict_strtoul(buf, 10, &min); | |
4562 | if (err) | |
4563 | return err; | |
4564 | ||
c0bdb232 | 4565 | set_min_partial(s, min); |
73d342b1 DR |
4566 | return length; |
4567 | } | |
4568 | SLAB_ATTR(min_partial); | |
4569 | ||
49e22585 CL |
4570 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4571 | { | |
4572 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4573 | } | |
4574 | ||
4575 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4576 | size_t length) | |
4577 | { | |
4578 | unsigned long objects; | |
4579 | int err; | |
4580 | ||
4581 | err = strict_strtoul(buf, 10, &objects); | |
4582 | if (err) | |
4583 | return err; | |
74ee4ef1 DR |
4584 | if (objects && kmem_cache_debug(s)) |
4585 | return -EINVAL; | |
49e22585 CL |
4586 | |
4587 | s->cpu_partial = objects; | |
4588 | flush_all(s); | |
4589 | return length; | |
4590 | } | |
4591 | SLAB_ATTR(cpu_partial); | |
4592 | ||
81819f0f CL |
4593 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4594 | { | |
62c70bce JP |
4595 | if (!s->ctor) |
4596 | return 0; | |
4597 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4598 | } |
4599 | SLAB_ATTR_RO(ctor); | |
4600 | ||
81819f0f CL |
4601 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4602 | { | |
4603 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4604 | } | |
4605 | SLAB_ATTR_RO(aliases); | |
4606 | ||
81819f0f CL |
4607 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4608 | { | |
d9acf4b7 | 4609 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4610 | } |
4611 | SLAB_ATTR_RO(partial); | |
4612 | ||
4613 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4614 | { | |
d9acf4b7 | 4615 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4616 | } |
4617 | SLAB_ATTR_RO(cpu_slabs); | |
4618 | ||
4619 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4620 | { | |
205ab99d | 4621 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4622 | } |
4623 | SLAB_ATTR_RO(objects); | |
4624 | ||
205ab99d CL |
4625 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4626 | { | |
4627 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4628 | } | |
4629 | SLAB_ATTR_RO(objects_partial); | |
4630 | ||
49e22585 CL |
4631 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4632 | { | |
4633 | int objects = 0; | |
4634 | int pages = 0; | |
4635 | int cpu; | |
4636 | int len; | |
4637 | ||
4638 | for_each_online_cpu(cpu) { | |
4639 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4640 | ||
4641 | if (page) { | |
4642 | pages += page->pages; | |
4643 | objects += page->pobjects; | |
4644 | } | |
4645 | } | |
4646 | ||
4647 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4648 | ||
4649 | #ifdef CONFIG_SMP | |
4650 | for_each_online_cpu(cpu) { | |
4651 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4652 | ||
4653 | if (page && len < PAGE_SIZE - 20) | |
4654 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4655 | page->pobjects, page->pages); | |
4656 | } | |
4657 | #endif | |
4658 | return len + sprintf(buf + len, "\n"); | |
4659 | } | |
4660 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4661 | ||
a5a84755 CL |
4662 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4663 | { | |
4664 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4665 | } | |
4666 | ||
4667 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4668 | const char *buf, size_t length) | |
4669 | { | |
4670 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4671 | if (buf[0] == '1') | |
4672 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4673 | return length; | |
4674 | } | |
4675 | SLAB_ATTR(reclaim_account); | |
4676 | ||
4677 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4678 | { | |
4679 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4680 | } | |
4681 | SLAB_ATTR_RO(hwcache_align); | |
4682 | ||
4683 | #ifdef CONFIG_ZONE_DMA | |
4684 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4685 | { | |
4686 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4687 | } | |
4688 | SLAB_ATTR_RO(cache_dma); | |
4689 | #endif | |
4690 | ||
4691 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4692 | { | |
4693 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4694 | } | |
4695 | SLAB_ATTR_RO(destroy_by_rcu); | |
4696 | ||
ab9a0f19 LJ |
4697 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4698 | { | |
4699 | return sprintf(buf, "%d\n", s->reserved); | |
4700 | } | |
4701 | SLAB_ATTR_RO(reserved); | |
4702 | ||
ab4d5ed5 | 4703 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4704 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4705 | { | |
4706 | return show_slab_objects(s, buf, SO_ALL); | |
4707 | } | |
4708 | SLAB_ATTR_RO(slabs); | |
4709 | ||
205ab99d CL |
4710 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4711 | { | |
4712 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4713 | } | |
4714 | SLAB_ATTR_RO(total_objects); | |
4715 | ||
81819f0f CL |
4716 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4717 | { | |
4718 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4719 | } | |
4720 | ||
4721 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4722 | const char *buf, size_t length) | |
4723 | { | |
4724 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4725 | if (buf[0] == '1') { |
4726 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4727 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4728 | } |
81819f0f CL |
4729 | return length; |
4730 | } | |
4731 | SLAB_ATTR(sanity_checks); | |
4732 | ||
4733 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4734 | { | |
4735 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4736 | } | |
4737 | ||
4738 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4739 | size_t length) | |
4740 | { | |
4741 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4742 | if (buf[0] == '1') { |
4743 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4744 | s->flags |= SLAB_TRACE; |
b789ef51 | 4745 | } |
81819f0f CL |
4746 | return length; |
4747 | } | |
4748 | SLAB_ATTR(trace); | |
4749 | ||
81819f0f CL |
4750 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4751 | { | |
4752 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4753 | } | |
4754 | ||
4755 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4756 | const char *buf, size_t length) | |
4757 | { | |
4758 | if (any_slab_objects(s)) | |
4759 | return -EBUSY; | |
4760 | ||
4761 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4762 | if (buf[0] == '1') { |
4763 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4764 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4765 | } |
06b285dc | 4766 | calculate_sizes(s, -1); |
81819f0f CL |
4767 | return length; |
4768 | } | |
4769 | SLAB_ATTR(red_zone); | |
4770 | ||
4771 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4772 | { | |
4773 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4774 | } | |
4775 | ||
4776 | static ssize_t poison_store(struct kmem_cache *s, | |
4777 | const char *buf, size_t length) | |
4778 | { | |
4779 | if (any_slab_objects(s)) | |
4780 | return -EBUSY; | |
4781 | ||
4782 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4783 | if (buf[0] == '1') { |
4784 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4785 | s->flags |= SLAB_POISON; |
b789ef51 | 4786 | } |
06b285dc | 4787 | calculate_sizes(s, -1); |
81819f0f CL |
4788 | return length; |
4789 | } | |
4790 | SLAB_ATTR(poison); | |
4791 | ||
4792 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4793 | { | |
4794 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4795 | } | |
4796 | ||
4797 | static ssize_t store_user_store(struct kmem_cache *s, | |
4798 | const char *buf, size_t length) | |
4799 | { | |
4800 | if (any_slab_objects(s)) | |
4801 | return -EBUSY; | |
4802 | ||
4803 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4804 | if (buf[0] == '1') { |
4805 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4806 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4807 | } |
06b285dc | 4808 | calculate_sizes(s, -1); |
81819f0f CL |
4809 | return length; |
4810 | } | |
4811 | SLAB_ATTR(store_user); | |
4812 | ||
53e15af0 CL |
4813 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4814 | { | |
4815 | return 0; | |
4816 | } | |
4817 | ||
4818 | static ssize_t validate_store(struct kmem_cache *s, | |
4819 | const char *buf, size_t length) | |
4820 | { | |
434e245d CL |
4821 | int ret = -EINVAL; |
4822 | ||
4823 | if (buf[0] == '1') { | |
4824 | ret = validate_slab_cache(s); | |
4825 | if (ret >= 0) | |
4826 | ret = length; | |
4827 | } | |
4828 | return ret; | |
53e15af0 CL |
4829 | } |
4830 | SLAB_ATTR(validate); | |
a5a84755 CL |
4831 | |
4832 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4833 | { | |
4834 | if (!(s->flags & SLAB_STORE_USER)) | |
4835 | return -ENOSYS; | |
4836 | return list_locations(s, buf, TRACK_ALLOC); | |
4837 | } | |
4838 | SLAB_ATTR_RO(alloc_calls); | |
4839 | ||
4840 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4841 | { | |
4842 | if (!(s->flags & SLAB_STORE_USER)) | |
4843 | return -ENOSYS; | |
4844 | return list_locations(s, buf, TRACK_FREE); | |
4845 | } | |
4846 | SLAB_ATTR_RO(free_calls); | |
4847 | #endif /* CONFIG_SLUB_DEBUG */ | |
4848 | ||
4849 | #ifdef CONFIG_FAILSLAB | |
4850 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4851 | { | |
4852 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4853 | } | |
4854 | ||
4855 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4856 | size_t length) | |
4857 | { | |
4858 | s->flags &= ~SLAB_FAILSLAB; | |
4859 | if (buf[0] == '1') | |
4860 | s->flags |= SLAB_FAILSLAB; | |
4861 | return length; | |
4862 | } | |
4863 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4864 | #endif |
53e15af0 | 4865 | |
2086d26a CL |
4866 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4867 | { | |
4868 | return 0; | |
4869 | } | |
4870 | ||
4871 | static ssize_t shrink_store(struct kmem_cache *s, | |
4872 | const char *buf, size_t length) | |
4873 | { | |
4874 | if (buf[0] == '1') { | |
4875 | int rc = kmem_cache_shrink(s); | |
4876 | ||
4877 | if (rc) | |
4878 | return rc; | |
4879 | } else | |
4880 | return -EINVAL; | |
4881 | return length; | |
4882 | } | |
4883 | SLAB_ATTR(shrink); | |
4884 | ||
81819f0f | 4885 | #ifdef CONFIG_NUMA |
9824601e | 4886 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4887 | { |
9824601e | 4888 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4889 | } |
4890 | ||
9824601e | 4891 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4892 | const char *buf, size_t length) |
4893 | { | |
0121c619 CL |
4894 | unsigned long ratio; |
4895 | int err; | |
4896 | ||
4897 | err = strict_strtoul(buf, 10, &ratio); | |
4898 | if (err) | |
4899 | return err; | |
4900 | ||
e2cb96b7 | 4901 | if (ratio <= 100) |
0121c619 | 4902 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4903 | |
81819f0f CL |
4904 | return length; |
4905 | } | |
9824601e | 4906 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4907 | #endif |
4908 | ||
8ff12cfc | 4909 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4910 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4911 | { | |
4912 | unsigned long sum = 0; | |
4913 | int cpu; | |
4914 | int len; | |
4915 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4916 | ||
4917 | if (!data) | |
4918 | return -ENOMEM; | |
4919 | ||
4920 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4921 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4922 | |
4923 | data[cpu] = x; | |
4924 | sum += x; | |
4925 | } | |
4926 | ||
4927 | len = sprintf(buf, "%lu", sum); | |
4928 | ||
50ef37b9 | 4929 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4930 | for_each_online_cpu(cpu) { |
4931 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4932 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4933 | } |
50ef37b9 | 4934 | #endif |
8ff12cfc CL |
4935 | kfree(data); |
4936 | return len + sprintf(buf + len, "\n"); | |
4937 | } | |
4938 | ||
78eb00cc DR |
4939 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4940 | { | |
4941 | int cpu; | |
4942 | ||
4943 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4944 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4945 | } |
4946 | ||
8ff12cfc CL |
4947 | #define STAT_ATTR(si, text) \ |
4948 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4949 | { \ | |
4950 | return show_stat(s, buf, si); \ | |
4951 | } \ | |
78eb00cc DR |
4952 | static ssize_t text##_store(struct kmem_cache *s, \ |
4953 | const char *buf, size_t length) \ | |
4954 | { \ | |
4955 | if (buf[0] != '0') \ | |
4956 | return -EINVAL; \ | |
4957 | clear_stat(s, si); \ | |
4958 | return length; \ | |
4959 | } \ | |
4960 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4961 | |
4962 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4963 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4964 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4965 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4966 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4967 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4968 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4969 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4970 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4971 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4972 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4973 | STAT_ATTR(FREE_SLAB, free_slab); |
4974 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4975 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4976 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4977 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4978 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4979 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4980 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4981 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4982 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4983 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4984 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4985 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4986 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4987 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4988 | #endif |
4989 | ||
06428780 | 4990 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4991 | &slab_size_attr.attr, |
4992 | &object_size_attr.attr, | |
4993 | &objs_per_slab_attr.attr, | |
4994 | &order_attr.attr, | |
73d342b1 | 4995 | &min_partial_attr.attr, |
49e22585 | 4996 | &cpu_partial_attr.attr, |
81819f0f | 4997 | &objects_attr.attr, |
205ab99d | 4998 | &objects_partial_attr.attr, |
81819f0f CL |
4999 | &partial_attr.attr, |
5000 | &cpu_slabs_attr.attr, | |
5001 | &ctor_attr.attr, | |
81819f0f CL |
5002 | &aliases_attr.attr, |
5003 | &align_attr.attr, | |
81819f0f CL |
5004 | &hwcache_align_attr.attr, |
5005 | &reclaim_account_attr.attr, | |
5006 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5007 | &shrink_attr.attr, |
ab9a0f19 | 5008 | &reserved_attr.attr, |
49e22585 | 5009 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5010 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5011 | &total_objects_attr.attr, |
5012 | &slabs_attr.attr, | |
5013 | &sanity_checks_attr.attr, | |
5014 | &trace_attr.attr, | |
81819f0f CL |
5015 | &red_zone_attr.attr, |
5016 | &poison_attr.attr, | |
5017 | &store_user_attr.attr, | |
53e15af0 | 5018 | &validate_attr.attr, |
88a420e4 CL |
5019 | &alloc_calls_attr.attr, |
5020 | &free_calls_attr.attr, | |
ab4d5ed5 | 5021 | #endif |
81819f0f CL |
5022 | #ifdef CONFIG_ZONE_DMA |
5023 | &cache_dma_attr.attr, | |
5024 | #endif | |
5025 | #ifdef CONFIG_NUMA | |
9824601e | 5026 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5027 | #endif |
5028 | #ifdef CONFIG_SLUB_STATS | |
5029 | &alloc_fastpath_attr.attr, | |
5030 | &alloc_slowpath_attr.attr, | |
5031 | &free_fastpath_attr.attr, | |
5032 | &free_slowpath_attr.attr, | |
5033 | &free_frozen_attr.attr, | |
5034 | &free_add_partial_attr.attr, | |
5035 | &free_remove_partial_attr.attr, | |
5036 | &alloc_from_partial_attr.attr, | |
5037 | &alloc_slab_attr.attr, | |
5038 | &alloc_refill_attr.attr, | |
e36a2652 | 5039 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5040 | &free_slab_attr.attr, |
5041 | &cpuslab_flush_attr.attr, | |
5042 | &deactivate_full_attr.attr, | |
5043 | &deactivate_empty_attr.attr, | |
5044 | &deactivate_to_head_attr.attr, | |
5045 | &deactivate_to_tail_attr.attr, | |
5046 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5047 | &deactivate_bypass_attr.attr, |
65c3376a | 5048 | &order_fallback_attr.attr, |
b789ef51 CL |
5049 | &cmpxchg_double_fail_attr.attr, |
5050 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5051 | &cpu_partial_alloc_attr.attr, |
5052 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5053 | &cpu_partial_node_attr.attr, |
5054 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5055 | #endif |
4c13dd3b DM |
5056 | #ifdef CONFIG_FAILSLAB |
5057 | &failslab_attr.attr, | |
5058 | #endif | |
5059 | ||
81819f0f CL |
5060 | NULL |
5061 | }; | |
5062 | ||
5063 | static struct attribute_group slab_attr_group = { | |
5064 | .attrs = slab_attrs, | |
5065 | }; | |
5066 | ||
5067 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5068 | struct attribute *attr, | |
5069 | char *buf) | |
5070 | { | |
5071 | struct slab_attribute *attribute; | |
5072 | struct kmem_cache *s; | |
5073 | int err; | |
5074 | ||
5075 | attribute = to_slab_attr(attr); | |
5076 | s = to_slab(kobj); | |
5077 | ||
5078 | if (!attribute->show) | |
5079 | return -EIO; | |
5080 | ||
5081 | err = attribute->show(s, buf); | |
5082 | ||
5083 | return err; | |
5084 | } | |
5085 | ||
5086 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5087 | struct attribute *attr, | |
5088 | const char *buf, size_t len) | |
5089 | { | |
5090 | struct slab_attribute *attribute; | |
5091 | struct kmem_cache *s; | |
5092 | int err; | |
5093 | ||
5094 | attribute = to_slab_attr(attr); | |
5095 | s = to_slab(kobj); | |
5096 | ||
5097 | if (!attribute->store) | |
5098 | return -EIO; | |
5099 | ||
5100 | err = attribute->store(s, buf, len); | |
5101 | ||
5102 | return err; | |
5103 | } | |
5104 | ||
52cf25d0 | 5105 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5106 | .show = slab_attr_show, |
5107 | .store = slab_attr_store, | |
5108 | }; | |
5109 | ||
5110 | static struct kobj_type slab_ktype = { | |
5111 | .sysfs_ops = &slab_sysfs_ops, | |
5112 | }; | |
5113 | ||
5114 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5115 | { | |
5116 | struct kobj_type *ktype = get_ktype(kobj); | |
5117 | ||
5118 | if (ktype == &slab_ktype) | |
5119 | return 1; | |
5120 | return 0; | |
5121 | } | |
5122 | ||
9cd43611 | 5123 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5124 | .filter = uevent_filter, |
5125 | }; | |
5126 | ||
27c3a314 | 5127 | static struct kset *slab_kset; |
81819f0f CL |
5128 | |
5129 | #define ID_STR_LENGTH 64 | |
5130 | ||
5131 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5132 | * |
5133 | * Format :[flags-]size | |
81819f0f CL |
5134 | */ |
5135 | static char *create_unique_id(struct kmem_cache *s) | |
5136 | { | |
5137 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5138 | char *p = name; | |
5139 | ||
5140 | BUG_ON(!name); | |
5141 | ||
5142 | *p++ = ':'; | |
5143 | /* | |
5144 | * First flags affecting slabcache operations. We will only | |
5145 | * get here for aliasable slabs so we do not need to support | |
5146 | * too many flags. The flags here must cover all flags that | |
5147 | * are matched during merging to guarantee that the id is | |
5148 | * unique. | |
5149 | */ | |
5150 | if (s->flags & SLAB_CACHE_DMA) | |
5151 | *p++ = 'd'; | |
5152 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5153 | *p++ = 'a'; | |
5154 | if (s->flags & SLAB_DEBUG_FREE) | |
5155 | *p++ = 'F'; | |
5a896d9e VN |
5156 | if (!(s->flags & SLAB_NOTRACK)) |
5157 | *p++ = 't'; | |
81819f0f CL |
5158 | if (p != name + 1) |
5159 | *p++ = '-'; | |
5160 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 GC |
5161 | |
5162 | #ifdef CONFIG_MEMCG_KMEM | |
5163 | if (!is_root_cache(s)) | |
5164 | p += sprintf(p, "-%08d", memcg_cache_id(s->memcg_params->memcg)); | |
5165 | #endif | |
5166 | ||
81819f0f CL |
5167 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5168 | return name; | |
5169 | } | |
5170 | ||
5171 | static int sysfs_slab_add(struct kmem_cache *s) | |
5172 | { | |
5173 | int err; | |
5174 | const char *name; | |
45530c44 | 5175 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5176 | |
81819f0f CL |
5177 | if (unmergeable) { |
5178 | /* | |
5179 | * Slabcache can never be merged so we can use the name proper. | |
5180 | * This is typically the case for debug situations. In that | |
5181 | * case we can catch duplicate names easily. | |
5182 | */ | |
27c3a314 | 5183 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5184 | name = s->name; |
5185 | } else { | |
5186 | /* | |
5187 | * Create a unique name for the slab as a target | |
5188 | * for the symlinks. | |
5189 | */ | |
5190 | name = create_unique_id(s); | |
5191 | } | |
5192 | ||
27c3a314 | 5193 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5194 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5195 | if (err) { | |
5196 | kobject_put(&s->kobj); | |
81819f0f | 5197 | return err; |
1eada11c | 5198 | } |
81819f0f CL |
5199 | |
5200 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5201 | if (err) { |
5202 | kobject_del(&s->kobj); | |
5203 | kobject_put(&s->kobj); | |
81819f0f | 5204 | return err; |
5788d8ad | 5205 | } |
81819f0f CL |
5206 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5207 | if (!unmergeable) { | |
5208 | /* Setup first alias */ | |
5209 | sysfs_slab_alias(s, s->name); | |
5210 | kfree(name); | |
5211 | } | |
5212 | return 0; | |
5213 | } | |
5214 | ||
5215 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5216 | { | |
97d06609 | 5217 | if (slab_state < FULL) |
2bce6485 CL |
5218 | /* |
5219 | * Sysfs has not been setup yet so no need to remove the | |
5220 | * cache from sysfs. | |
5221 | */ | |
5222 | return; | |
5223 | ||
81819f0f CL |
5224 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5225 | kobject_del(&s->kobj); | |
151c602f | 5226 | kobject_put(&s->kobj); |
81819f0f CL |
5227 | } |
5228 | ||
5229 | /* | |
5230 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5231 | * available lest we lose that information. |
81819f0f CL |
5232 | */ |
5233 | struct saved_alias { | |
5234 | struct kmem_cache *s; | |
5235 | const char *name; | |
5236 | struct saved_alias *next; | |
5237 | }; | |
5238 | ||
5af328a5 | 5239 | static struct saved_alias *alias_list; |
81819f0f CL |
5240 | |
5241 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5242 | { | |
5243 | struct saved_alias *al; | |
5244 | ||
97d06609 | 5245 | if (slab_state == FULL) { |
81819f0f CL |
5246 | /* |
5247 | * If we have a leftover link then remove it. | |
5248 | */ | |
27c3a314 GKH |
5249 | sysfs_remove_link(&slab_kset->kobj, name); |
5250 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5251 | } |
5252 | ||
5253 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5254 | if (!al) | |
5255 | return -ENOMEM; | |
5256 | ||
5257 | al->s = s; | |
5258 | al->name = name; | |
5259 | al->next = alias_list; | |
5260 | alias_list = al; | |
5261 | return 0; | |
5262 | } | |
5263 | ||
5264 | static int __init slab_sysfs_init(void) | |
5265 | { | |
5b95a4ac | 5266 | struct kmem_cache *s; |
81819f0f CL |
5267 | int err; |
5268 | ||
18004c5d | 5269 | mutex_lock(&slab_mutex); |
2bce6485 | 5270 | |
0ff21e46 | 5271 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5272 | if (!slab_kset) { |
18004c5d | 5273 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5274 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5275 | return -ENOSYS; | |
5276 | } | |
5277 | ||
97d06609 | 5278 | slab_state = FULL; |
26a7bd03 | 5279 | |
5b95a4ac | 5280 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5281 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5282 | if (err) |
5283 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5284 | " to sysfs\n", s->name); | |
26a7bd03 | 5285 | } |
81819f0f CL |
5286 | |
5287 | while (alias_list) { | |
5288 | struct saved_alias *al = alias_list; | |
5289 | ||
5290 | alias_list = alias_list->next; | |
5291 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5292 | if (err) |
5293 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5294 | " %s to sysfs\n", al->name); |
81819f0f CL |
5295 | kfree(al); |
5296 | } | |
5297 | ||
18004c5d | 5298 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5299 | resiliency_test(); |
5300 | return 0; | |
5301 | } | |
5302 | ||
5303 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5304 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5305 | |
5306 | /* | |
5307 | * The /proc/slabinfo ABI | |
5308 | */ | |
158a9624 | 5309 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5310 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda PE |
5311 | { |
5312 | unsigned long nr_partials = 0; | |
5313 | unsigned long nr_slabs = 0; | |
205ab99d CL |
5314 | unsigned long nr_objs = 0; |
5315 | unsigned long nr_free = 0; | |
57ed3eda PE |
5316 | int node; |
5317 | ||
57ed3eda PE |
5318 | for_each_online_node(node) { |
5319 | struct kmem_cache_node *n = get_node(s, node); | |
5320 | ||
5321 | if (!n) | |
5322 | continue; | |
5323 | ||
5324 | nr_partials += n->nr_partial; | |
5325 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5326 | nr_objs += atomic_long_read(&n->total_objects); |
5327 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5328 | } |
5329 | ||
0d7561c6 GC |
5330 | sinfo->active_objs = nr_objs - nr_free; |
5331 | sinfo->num_objs = nr_objs; | |
5332 | sinfo->active_slabs = nr_slabs; | |
5333 | sinfo->num_slabs = nr_slabs; | |
5334 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5335 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5336 | } |
5337 | ||
0d7561c6 | 5338 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5339 | { |
7b3c3a50 AD |
5340 | } |
5341 | ||
b7454ad3 GC |
5342 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5343 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5344 | { |
b7454ad3 | 5345 | return -EIO; |
7b3c3a50 | 5346 | } |
158a9624 | 5347 | #endif /* CONFIG_SLABINFO */ |