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> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
5a896d9e | 23 | #include <linux/kmemcheck.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b9049e23 | 30 | #include <linux/memory.h> |
f8bd2258 | 31 | #include <linux/math64.h> |
773ff60e | 32 | #include <linux/fault-inject.h> |
bfa71457 | 33 | #include <linux/stacktrace.h> |
4de900b4 | 34 | #include <linux/prefetch.h> |
2633d7a0 | 35 | #include <linux/memcontrol.h> |
6fa3eb70 | 36 | #include <linux/aee.h> |
81819f0f | 37 | |
4a92379b | 38 | #include <trace/events/kmem.h> |
6fa3eb70 | 39 | #include <mach/mtk_memcfg.h> |
4a92379b | 40 | |
072bb0aa | 41 | #include "internal.h" |
6fa3eb70 S |
42 | #ifdef CONFIG_ARM64 |
43 | #ifdef BUG | |
44 | #undef BUG | |
45 | #define BUG() \ | |
46 | do { \ | |
47 | pr_alert("BUG: failure at %s:%d/%s()!\n", __FILE__, __LINE__, __func__); \ | |
48 | *(char *)0 = 0;\ | |
49 | } while(0) | |
50 | #endif | |
51 | #endif | |
81819f0f CL |
52 | /* |
53 | * Lock order: | |
18004c5d | 54 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
55 | * 2. node->list_lock |
56 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 57 | * |
18004c5d | 58 | * slab_mutex |
881db7fb | 59 | * |
18004c5d | 60 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
61 | * and to synchronize major metadata changes to slab cache structures. |
62 | * | |
63 | * The slab_lock is only used for debugging and on arches that do not | |
64 | * have the ability to do a cmpxchg_double. It only protects the second | |
65 | * double word in the page struct. Meaning | |
66 | * A. page->freelist -> List of object free in a page | |
67 | * B. page->counters -> Counters of objects | |
68 | * C. page->frozen -> frozen state | |
69 | * | |
70 | * If a slab is frozen then it is exempt from list management. It is not | |
71 | * on any list. The processor that froze the slab is the one who can | |
72 | * perform list operations on the page. Other processors may put objects | |
73 | * onto the freelist but the processor that froze the slab is the only | |
74 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
75 | * |
76 | * The list_lock protects the partial and full list on each node and | |
77 | * the partial slab counter. If taken then no new slabs may be added or | |
78 | * removed from the lists nor make the number of partial slabs be modified. | |
79 | * (Note that the total number of slabs is an atomic value that may be | |
80 | * modified without taking the list lock). | |
81 | * | |
82 | * The list_lock is a centralized lock and thus we avoid taking it as | |
83 | * much as possible. As long as SLUB does not have to handle partial | |
84 | * slabs, operations can continue without any centralized lock. F.e. | |
85 | * allocating a long series of objects that fill up slabs does not require | |
86 | * the list lock. | |
81819f0f CL |
87 | * Interrupts are disabled during allocation and deallocation in order to |
88 | * make the slab allocator safe to use in the context of an irq. In addition | |
89 | * interrupts are disabled to ensure that the processor does not change | |
90 | * while handling per_cpu slabs, due to kernel preemption. | |
91 | * | |
92 | * SLUB assigns one slab for allocation to each processor. | |
93 | * Allocations only occur from these slabs called cpu slabs. | |
94 | * | |
672bba3a CL |
95 | * Slabs with free elements are kept on a partial list and during regular |
96 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 97 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
98 | * We track full slabs for debugging purposes though because otherwise we |
99 | * cannot scan all objects. | |
81819f0f CL |
100 | * |
101 | * Slabs are freed when they become empty. Teardown and setup is | |
102 | * minimal so we rely on the page allocators per cpu caches for | |
103 | * fast frees and allocs. | |
104 | * | |
105 | * Overloading of page flags that are otherwise used for LRU management. | |
106 | * | |
4b6f0750 CL |
107 | * PageActive The slab is frozen and exempt from list processing. |
108 | * This means that the slab is dedicated to a purpose | |
109 | * such as satisfying allocations for a specific | |
110 | * processor. Objects may be freed in the slab while | |
111 | * it is frozen but slab_free will then skip the usual | |
112 | * list operations. It is up to the processor holding | |
113 | * the slab to integrate the slab into the slab lists | |
114 | * when the slab is no longer needed. | |
115 | * | |
116 | * One use of this flag is to mark slabs that are | |
117 | * used for allocations. Then such a slab becomes a cpu | |
118 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 119 | * freelist that allows lockless access to |
894b8788 CL |
120 | * free objects in addition to the regular freelist |
121 | * that requires the slab lock. | |
81819f0f CL |
122 | * |
123 | * PageError Slab requires special handling due to debug | |
124 | * options set. This moves slab handling out of | |
894b8788 | 125 | * the fast path and disables lockless freelists. |
81819f0f CL |
126 | */ |
127 | ||
af537b0a CL |
128 | static inline int kmem_cache_debug(struct kmem_cache *s) |
129 | { | |
5577bd8a | 130 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 131 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 132 | #else |
af537b0a | 133 | return 0; |
5577bd8a | 134 | #endif |
af537b0a | 135 | } |
5577bd8a | 136 | |
81819f0f CL |
137 | /* |
138 | * Issues still to be resolved: | |
139 | * | |
81819f0f CL |
140 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
141 | * | |
81819f0f CL |
142 | * - Variable sizing of the per node arrays |
143 | */ | |
144 | ||
145 | /* Enable to test recovery from slab corruption on boot */ | |
146 | #undef SLUB_RESILIENCY_TEST | |
147 | ||
b789ef51 CL |
148 | /* Enable to log cmpxchg failures */ |
149 | #undef SLUB_DEBUG_CMPXCHG | |
150 | ||
2086d26a CL |
151 | /* |
152 | * Mininum number of partial slabs. These will be left on the partial | |
153 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
154 | */ | |
76be8950 | 155 | #define MIN_PARTIAL 5 |
e95eed57 | 156 | |
2086d26a CL |
157 | /* |
158 | * Maximum number of desirable partial slabs. | |
159 | * The existence of more partial slabs makes kmem_cache_shrink | |
160 | * sort the partial list by the number of objects in the. | |
161 | */ | |
162 | #define MAX_PARTIAL 10 | |
163 | ||
81819f0f CL |
164 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
165 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 166 | |
fa5ec8a1 | 167 | /* |
3de47213 DR |
168 | * Debugging flags that require metadata to be stored in the slab. These get |
169 | * disabled when slub_debug=O is used and a cache's min order increases with | |
170 | * metadata. | |
fa5ec8a1 | 171 | */ |
3de47213 | 172 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 173 | |
81819f0f CL |
174 | /* |
175 | * Set of flags that will prevent slab merging | |
176 | */ | |
177 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
178 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
179 | SLAB_FAILSLAB) | |
81819f0f CL |
180 | |
181 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 182 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 183 | |
210b5c06 CG |
184 | #define OO_SHIFT 16 |
185 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 186 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 187 | |
81819f0f | 188 | /* Internal SLUB flags */ |
f90ec390 | 189 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 190 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 191 | |
81819f0f CL |
192 | #ifdef CONFIG_SMP |
193 | static struct notifier_block slab_notifier; | |
194 | #endif | |
195 | ||
02cbc874 CL |
196 | /* |
197 | * Tracking user of a slab. | |
198 | */ | |
d6543e39 | 199 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 200 | struct track { |
ce71e27c | 201 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
202 | #ifdef CONFIG_STACKTRACE |
203 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
204 | #endif | |
02cbc874 CL |
205 | int cpu; /* Was running on cpu */ |
206 | int pid; /* Pid context */ | |
207 | unsigned long when; /* When did the operation occur */ | |
208 | }; | |
209 | ||
6fa3eb70 | 210 | enum track_item { TRACK_FREE, TRACK_ALLOC }; |
02cbc874 | 211 | |
ab4d5ed5 | 212 | #ifdef CONFIG_SYSFS |
81819f0f CL |
213 | static int sysfs_slab_add(struct kmem_cache *); |
214 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
215 | static void sysfs_slab_remove(struct kmem_cache *); | |
107dab5c | 216 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 217 | #else |
0c710013 CL |
218 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
219 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
220 | { return 0; } | |
db265eca | 221 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
8ff12cfc | 222 | |
107dab5c | 223 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
224 | #endif |
225 | ||
4fdccdfb | 226 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
227 | { |
228 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 229 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
230 | #endif |
231 | } | |
232 | ||
81819f0f CL |
233 | /******************************************************************** |
234 | * Core slab cache functions | |
235 | *******************************************************************/ | |
236 | ||
81819f0f CL |
237 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
238 | { | |
81819f0f | 239 | return s->node[node]; |
81819f0f CL |
240 | } |
241 | ||
6446faa2 | 242 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
243 | static inline int check_valid_pointer(struct kmem_cache *s, |
244 | struct page *page, const void *object) | |
245 | { | |
246 | void *base; | |
247 | ||
a973e9dd | 248 | if (!object) |
02cbc874 CL |
249 | return 1; |
250 | ||
a973e9dd | 251 | base = page_address(page); |
39b26464 | 252 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
253 | (object - base) % s->size) { |
254 | return 0; | |
255 | } | |
256 | ||
257 | return 1; | |
258 | } | |
259 | ||
7656c72b CL |
260 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
261 | { | |
262 | return *(void **)(object + s->offset); | |
263 | } | |
264 | ||
0ad9500e ED |
265 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
266 | { | |
267 | prefetch(object + s->offset); | |
268 | } | |
269 | ||
1393d9a1 CL |
270 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
271 | { | |
272 | void *p; | |
273 | ||
274 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
275 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
276 | #else | |
277 | p = get_freepointer(s, object); | |
278 | #endif | |
279 | return p; | |
280 | } | |
281 | ||
7656c72b CL |
282 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
283 | { | |
284 | *(void **)(object + s->offset) = fp; | |
285 | } | |
286 | ||
287 | /* Loop over all objects in a slab */ | |
224a88be CL |
288 | #define for_each_object(__p, __s, __addr, __objects) \ |
289 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
290 | __p += (__s)->size) |
291 | ||
7656c72b CL |
292 | /* Determine object index from a given position */ |
293 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
294 | { | |
295 | return (p - addr) / s->size; | |
296 | } | |
297 | ||
d71f606f MK |
298 | static inline size_t slab_ksize(const struct kmem_cache *s) |
299 | { | |
300 | #ifdef CONFIG_SLUB_DEBUG | |
301 | /* | |
302 | * Debugging requires use of the padding between object | |
303 | * and whatever may come after it. | |
304 | */ | |
305 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 306 | return s->object_size; |
d71f606f MK |
307 | |
308 | #endif | |
309 | /* | |
310 | * If we have the need to store the freelist pointer | |
311 | * back there or track user information then we can | |
312 | * only use the space before that information. | |
313 | */ | |
314 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
315 | return s->inuse; | |
316 | /* | |
317 | * Else we can use all the padding etc for the allocation | |
318 | */ | |
319 | return s->size; | |
320 | } | |
321 | ||
ab9a0f19 LJ |
322 | static inline int order_objects(int order, unsigned long size, int reserved) |
323 | { | |
324 | return ((PAGE_SIZE << order) - reserved) / size; | |
325 | } | |
326 | ||
834f3d11 | 327 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 328 | unsigned long size, int reserved) |
834f3d11 CL |
329 | { |
330 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 331 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
332 | }; |
333 | ||
334 | return x; | |
335 | } | |
336 | ||
337 | static inline int oo_order(struct kmem_cache_order_objects x) | |
338 | { | |
210b5c06 | 339 | return x.x >> OO_SHIFT; |
834f3d11 CL |
340 | } |
341 | ||
342 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
343 | { | |
210b5c06 | 344 | return x.x & OO_MASK; |
834f3d11 CL |
345 | } |
346 | ||
881db7fb CL |
347 | /* |
348 | * Per slab locking using the pagelock | |
349 | */ | |
350 | static __always_inline void slab_lock(struct page *page) | |
351 | { | |
352 | bit_spin_lock(PG_locked, &page->flags); | |
353 | } | |
354 | ||
355 | static __always_inline void slab_unlock(struct page *page) | |
356 | { | |
357 | __bit_spin_unlock(PG_locked, &page->flags); | |
358 | } | |
359 | ||
6fa3eb70 S |
360 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
361 | { | |
362 | struct page tmp; | |
363 | tmp.counters = counters_new; | |
364 | /* | |
365 | * page->counters can cover frozen/inuse/objects as well | |
366 | * as page->_count. If we assign to ->counters directly | |
367 | * we run the risk of losing updates to page->_count, so | |
368 | * be careful and only assign to the fields we need. | |
369 | */ | |
370 | page->frozen = tmp.frozen; | |
371 | page->inuse = tmp.inuse; | |
372 | page->objects = tmp.objects; | |
373 | } | |
374 | ||
1d07171c CL |
375 | /* Interrupts must be disabled (for the fallback code to work right) */ |
376 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
377 | void *freelist_old, unsigned long counters_old, | |
378 | void *freelist_new, unsigned long counters_new, | |
379 | const char *n) | |
380 | { | |
381 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
382 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
383 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 384 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 385 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
386 | freelist_old, counters_old, |
387 | freelist_new, counters_new)) | |
388 | return 1; | |
389 | } else | |
390 | #endif | |
391 | { | |
392 | slab_lock(page); | |
393 | if (page->freelist == freelist_old && page->counters == counters_old) { | |
394 | page->freelist = freelist_new; | |
6fa3eb70 | 395 | set_page_slub_counters(page, counters_new); |
1d07171c CL |
396 | slab_unlock(page); |
397 | return 1; | |
398 | } | |
399 | slab_unlock(page); | |
400 | } | |
401 | ||
402 | cpu_relax(); | |
403 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
404 | ||
405 | #ifdef SLUB_DEBUG_CMPXCHG | |
406 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
407 | #endif | |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
b789ef51 CL |
412 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
413 | void *freelist_old, unsigned long counters_old, | |
414 | void *freelist_new, unsigned long counters_new, | |
415 | const char *n) | |
416 | { | |
2565409f HC |
417 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
418 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 419 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 420 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
421 | freelist_old, counters_old, |
422 | freelist_new, counters_new)) | |
423 | return 1; | |
424 | } else | |
425 | #endif | |
426 | { | |
1d07171c CL |
427 | unsigned long flags; |
428 | ||
429 | local_irq_save(flags); | |
881db7fb | 430 | slab_lock(page); |
b789ef51 CL |
431 | if (page->freelist == freelist_old && page->counters == counters_old) { |
432 | page->freelist = freelist_new; | |
6fa3eb70 | 433 | set_page_slub_counters(page, counters_new); |
881db7fb | 434 | slab_unlock(page); |
1d07171c | 435 | local_irq_restore(flags); |
b789ef51 CL |
436 | return 1; |
437 | } | |
881db7fb | 438 | slab_unlock(page); |
1d07171c | 439 | local_irq_restore(flags); |
b789ef51 CL |
440 | } |
441 | ||
442 | cpu_relax(); | |
443 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
444 | ||
445 | #ifdef SLUB_DEBUG_CMPXCHG | |
446 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
447 | #endif | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
41ecc55b | 452 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
453 | /* |
454 | * Determine a map of object in use on a page. | |
455 | * | |
881db7fb | 456 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
457 | * not vanish from under us. |
458 | */ | |
459 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
460 | { | |
461 | void *p; | |
462 | void *addr = page_address(page); | |
463 | ||
464 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
465 | set_bit(slab_index(p, s, addr), map); | |
466 | } | |
467 | ||
41ecc55b CL |
468 | /* |
469 | * Debug settings: | |
470 | */ | |
f0630fff CL |
471 | #ifdef CONFIG_SLUB_DEBUG_ON |
472 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
473 | #else | |
41ecc55b | 474 | static int slub_debug; |
f0630fff | 475 | #endif |
41ecc55b CL |
476 | |
477 | static char *slub_debug_slabs; | |
fa5ec8a1 | 478 | static int disable_higher_order_debug; |
41ecc55b | 479 | |
81819f0f CL |
480 | /* |
481 | * Object debugging | |
482 | */ | |
483 | static void print_section(char *text, u8 *addr, unsigned int length) | |
484 | { | |
ffc79d28 SAS |
485 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
486 | length, 1); | |
81819f0f CL |
487 | } |
488 | ||
81819f0f CL |
489 | static struct track *get_track(struct kmem_cache *s, void *object, |
490 | enum track_item alloc) | |
491 | { | |
492 | struct track *p; | |
493 | ||
494 | if (s->offset) | |
495 | p = object + s->offset + sizeof(void *); | |
496 | else | |
497 | p = object + s->inuse; | |
498 | ||
499 | return p + alloc; | |
500 | } | |
501 | ||
502 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 503 | enum track_item alloc, unsigned long addr) |
81819f0f | 504 | { |
1a00df4a | 505 | struct track *p = get_track(s, object, alloc); |
81819f0f | 506 | |
81819f0f | 507 | if (addr) { |
d6543e39 BG |
508 | #ifdef CONFIG_STACKTRACE |
509 | struct stack_trace trace; | |
510 | int i; | |
511 | ||
512 | trace.nr_entries = 0; | |
513 | trace.max_entries = TRACK_ADDRS_COUNT; | |
514 | trace.entries = p->addrs; | |
515 | trace.skip = 3; | |
516 | save_stack_trace(&trace); | |
517 | ||
518 | /* See rant in lockdep.c */ | |
519 | if (trace.nr_entries != 0 && | |
520 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
521 | trace.nr_entries--; | |
522 | ||
523 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
524 | p->addrs[i] = 0; | |
525 | #endif | |
81819f0f CL |
526 | p->addr = addr; |
527 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 528 | p->pid = current->pid; |
81819f0f CL |
529 | p->when = jiffies; |
530 | } else | |
531 | memset(p, 0, sizeof(struct track)); | |
532 | } | |
533 | ||
81819f0f CL |
534 | static void init_tracking(struct kmem_cache *s, void *object) |
535 | { | |
24922684 CL |
536 | if (!(s->flags & SLAB_STORE_USER)) |
537 | return; | |
538 | ||
ce71e27c EGM |
539 | set_track(s, object, TRACK_FREE, 0UL); |
540 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
541 | } |
542 | ||
543 | static void print_track(const char *s, struct track *t) | |
544 | { | |
545 | if (!t->addr) | |
546 | return; | |
547 | ||
7daf705f | 548 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 549 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); |
d6543e39 BG |
550 | #ifdef CONFIG_STACKTRACE |
551 | { | |
552 | int i; | |
553 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
554 | if (t->addrs[i]) | |
555 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
556 | else | |
557 | break; | |
558 | } | |
559 | #endif | |
24922684 CL |
560 | } |
561 | ||
562 | static void print_tracking(struct kmem_cache *s, void *object) | |
563 | { | |
564 | if (!(s->flags & SLAB_STORE_USER)) | |
565 | return; | |
566 | ||
567 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
568 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
569 | } | |
570 | ||
571 | static void print_page_info(struct page *page) | |
572 | { | |
39b26464 CL |
573 | printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
574 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
575 | |
576 | } | |
577 | ||
578 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
579 | { | |
580 | va_list args; | |
581 | char buf[100]; | |
582 | ||
583 | va_start(args, fmt); | |
584 | vsnprintf(buf, sizeof(buf), fmt, args); | |
585 | va_end(args); | |
586 | printk(KERN_ERR "========================================" | |
587 | "=====================================\n"); | |
265d47e7 | 588 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
589 | printk(KERN_ERR "----------------------------------------" |
590 | "-------------------------------------\n\n"); | |
645df230 | 591 | |
373d4d09 | 592 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
593 | } |
594 | ||
24922684 CL |
595 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
596 | { | |
597 | va_list args; | |
598 | char buf[100]; | |
599 | ||
600 | va_start(args, fmt); | |
601 | vsnprintf(buf, sizeof(buf), fmt, args); | |
602 | va_end(args); | |
603 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
604 | } | |
605 | ||
606 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
607 | { |
608 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 609 | u8 *addr = page_address(page); |
24922684 CL |
610 | |
611 | print_tracking(s, p); | |
612 | ||
613 | print_page_info(page); | |
614 | ||
615 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
616 | p, p - addr, get_freepointer(s, p)); | |
617 | ||
618 | if (p > addr + 16) | |
ffc79d28 | 619 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 620 | |
3b0efdfa | 621 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 622 | PAGE_SIZE)); |
81819f0f | 623 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
624 | print_section("Redzone ", p + s->object_size, |
625 | s->inuse - s->object_size); | |
81819f0f | 626 | |
81819f0f CL |
627 | if (s->offset) |
628 | off = s->offset + sizeof(void *); | |
629 | else | |
630 | off = s->inuse; | |
631 | ||
24922684 | 632 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 633 | off += 2 * sizeof(struct track); |
81819f0f CL |
634 | |
635 | if (off != s->size) | |
636 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 637 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
638 | |
639 | dump_stack(); | |
81819f0f CL |
640 | } |
641 | ||
642 | static void object_err(struct kmem_cache *s, struct page *page, | |
643 | u8 *object, char *reason) | |
644 | { | |
3dc50637 | 645 | slab_bug(s, "%s", reason); |
24922684 | 646 | print_trailer(s, page, object); |
6fa3eb70 | 647 | BUG(); |
81819f0f CL |
648 | } |
649 | ||
945cf2b6 | 650 | static void slab_err(struct kmem_cache *s, struct page *page, const char *fmt, ...) |
81819f0f CL |
651 | { |
652 | va_list args; | |
653 | char buf[100]; | |
654 | ||
24922684 CL |
655 | va_start(args, fmt); |
656 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 657 | va_end(args); |
3dc50637 | 658 | slab_bug(s, "%s", buf); |
24922684 | 659 | print_page_info(page); |
81819f0f | 660 | dump_stack(); |
6fa3eb70 | 661 | BUG(); |
81819f0f CL |
662 | } |
663 | ||
f7cb1933 | 664 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
665 | { |
666 | u8 *p = object; | |
667 | ||
668 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
669 | memset(p, POISON_FREE, s->object_size - 1); |
670 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
671 | } |
672 | ||
673 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 674 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
675 | } |
676 | ||
24922684 CL |
677 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
678 | void *from, void *to) | |
679 | { | |
680 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
681 | memset(from, data, to - from); | |
682 | } | |
683 | ||
684 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
685 | u8 *object, char *what, | |
06428780 | 686 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
687 | { |
688 | u8 *fault; | |
689 | u8 *end; | |
690 | ||
79824820 | 691 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
692 | if (!fault) |
693 | return 1; | |
694 | ||
695 | end = start + bytes; | |
696 | while (end > fault && end[-1] == value) | |
697 | end--; | |
698 | ||
699 | slab_bug(s, "%s overwritten", what); | |
700 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
701 | fault, end - 1, fault[0], value); | |
702 | print_trailer(s, page, object); | |
703 | ||
6fa3eb70 S |
704 | /* trigger BUG before restore_bytes */ |
705 | BUG(); | |
24922684 | 706 | restore_bytes(s, what, value, fault, end); |
6fa3eb70 | 707 | |
24922684 | 708 | return 0; |
81819f0f CL |
709 | } |
710 | ||
81819f0f CL |
711 | /* |
712 | * Object layout: | |
713 | * | |
714 | * object address | |
715 | * Bytes of the object to be managed. | |
716 | * If the freepointer may overlay the object then the free | |
717 | * pointer is the first word of the object. | |
672bba3a | 718 | * |
81819f0f CL |
719 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
720 | * 0xa5 (POISON_END) | |
721 | * | |
3b0efdfa | 722 | * object + s->object_size |
81819f0f | 723 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 724 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 725 | * object_size == inuse. |
672bba3a | 726 | * |
81819f0f CL |
727 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
728 | * 0xcc (RED_ACTIVE) for objects in use. | |
729 | * | |
730 | * object + s->inuse | |
672bba3a CL |
731 | * Meta data starts here. |
732 | * | |
81819f0f CL |
733 | * A. Free pointer (if we cannot overwrite object on free) |
734 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 735 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 736 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
737 | * before the word boundary. |
738 | * | |
739 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
740 | * |
741 | * object + s->size | |
672bba3a | 742 | * Nothing is used beyond s->size. |
81819f0f | 743 | * |
3b0efdfa | 744 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 745 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
746 | * may be used with merged slabcaches. |
747 | */ | |
748 | ||
81819f0f CL |
749 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
750 | { | |
751 | unsigned long off = s->inuse; /* The end of info */ | |
752 | ||
753 | if (s->offset) | |
754 | /* Freepointer is placed after the object. */ | |
755 | off += sizeof(void *); | |
756 | ||
757 | if (s->flags & SLAB_STORE_USER) | |
758 | /* We also have user information there */ | |
759 | off += 2 * sizeof(struct track); | |
760 | ||
761 | if (s->size == off) | |
762 | return 1; | |
763 | ||
24922684 CL |
764 | return check_bytes_and_report(s, page, p, "Object padding", |
765 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
766 | } |
767 | ||
39b26464 | 768 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
769 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
770 | { | |
24922684 CL |
771 | u8 *start; |
772 | u8 *fault; | |
773 | u8 *end; | |
774 | int length; | |
775 | int remainder; | |
81819f0f CL |
776 | |
777 | if (!(s->flags & SLAB_POISON)) | |
778 | return 1; | |
779 | ||
a973e9dd | 780 | start = page_address(page); |
ab9a0f19 | 781 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
782 | end = start + length; |
783 | remainder = length % s->size; | |
81819f0f CL |
784 | if (!remainder) |
785 | return 1; | |
786 | ||
79824820 | 787 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
788 | if (!fault) |
789 | return 1; | |
790 | while (end > fault && end[-1] == POISON_INUSE) | |
791 | end--; | |
792 | ||
793 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 794 | print_section("Padding ", end - remainder, remainder); |
24922684 | 795 | |
8a3d271d | 796 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 797 | return 0; |
81819f0f CL |
798 | } |
799 | ||
800 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 801 | void *object, u8 val) |
81819f0f CL |
802 | { |
803 | u8 *p = object; | |
3b0efdfa | 804 | u8 *endobject = object + s->object_size; |
81819f0f CL |
805 | |
806 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 807 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 808 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 809 | return 0; |
81819f0f | 810 | } else { |
3b0efdfa | 811 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 812 | check_bytes_and_report(s, page, p, "Alignment padding", |
3b0efdfa | 813 | endobject, POISON_INUSE, s->inuse - s->object_size); |
3adbefee | 814 | } |
81819f0f CL |
815 | } |
816 | ||
817 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 818 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 819 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 820 | POISON_FREE, s->object_size - 1) || |
24922684 | 821 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 822 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 823 | return 0; |
81819f0f CL |
824 | /* |
825 | * check_pad_bytes cleans up on its own. | |
826 | */ | |
827 | check_pad_bytes(s, page, p); | |
828 | } | |
829 | ||
f7cb1933 | 830 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
831 | /* |
832 | * Object and freepointer overlap. Cannot check | |
833 | * freepointer while object is allocated. | |
834 | */ | |
835 | return 1; | |
836 | ||
837 | /* Check free pointer validity */ | |
838 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
839 | object_err(s, page, p, "Freepointer corrupt"); | |
840 | /* | |
9f6c708e | 841 | * No choice but to zap it and thus lose the remainder |
81819f0f | 842 | * of the free objects in this slab. May cause |
672bba3a | 843 | * another error because the object count is now wrong. |
81819f0f | 844 | */ |
a973e9dd | 845 | set_freepointer(s, p, NULL); |
81819f0f CL |
846 | return 0; |
847 | } | |
848 | return 1; | |
849 | } | |
850 | ||
851 | static int check_slab(struct kmem_cache *s, struct page *page) | |
852 | { | |
39b26464 CL |
853 | int maxobj; |
854 | ||
81819f0f CL |
855 | VM_BUG_ON(!irqs_disabled()); |
856 | ||
857 | if (!PageSlab(page)) { | |
24922684 | 858 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
859 | return 0; |
860 | } | |
39b26464 | 861 | |
ab9a0f19 | 862 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 | 863 | if (page->objects > maxobj) { |
6fa3eb70 S |
864 | pr_alert("page->objects: %d, maxobj: %d, comporder: %d", page->objects, |
865 | maxobj, compound_order(page)); | |
866 | pr_alert("s->size %d, s->reserved: %d", s->size, s->reserved); | |
867 | print_section("page: ", (void *)page, sizeof(struct page)); | |
868 | print_section("kmem_cache: ", (void *)s, sizeof(struct kmem_cache)); | |
39b26464 | 869 | slab_err(s, page, "objects %u > max %u", |
6fa3eb70 | 870 | page->objects, maxobj); |
39b26464 CL |
871 | return 0; |
872 | } | |
873 | if (page->inuse > page->objects) { | |
24922684 | 874 | slab_err(s, page, "inuse %u > max %u", |
6fa3eb70 | 875 | page->inuse, page->objects); |
81819f0f CL |
876 | return 0; |
877 | } | |
878 | /* Slab_pad_check fixes things up after itself */ | |
879 | slab_pad_check(s, page); | |
880 | return 1; | |
881 | } | |
882 | ||
883 | /* | |
672bba3a CL |
884 | * Determine if a certain object on a page is on the freelist. Must hold the |
885 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
886 | */ |
887 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
888 | { | |
889 | int nr = 0; | |
881db7fb | 890 | void *fp; |
81819f0f | 891 | void *object = NULL; |
224a88be | 892 | unsigned long max_objects; |
81819f0f | 893 | |
881db7fb | 894 | fp = page->freelist; |
39b26464 | 895 | while (fp && nr <= page->objects) { |
81819f0f CL |
896 | if (fp == search) |
897 | return 1; | |
898 | if (!check_valid_pointer(s, page, fp)) { | |
899 | if (object) { | |
900 | object_err(s, page, object, | |
901 | "Freechain corrupt"); | |
a973e9dd | 902 | set_freepointer(s, object, NULL); |
81819f0f CL |
903 | break; |
904 | } else { | |
24922684 | 905 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 906 | page->freelist = NULL; |
39b26464 | 907 | page->inuse = page->objects; |
24922684 | 908 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
909 | return 0; |
910 | } | |
911 | break; | |
912 | } | |
913 | object = fp; | |
914 | fp = get_freepointer(s, object); | |
915 | nr++; | |
916 | } | |
917 | ||
ab9a0f19 | 918 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
919 | if (max_objects > MAX_OBJS_PER_PAGE) |
920 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
921 | |
922 | if (page->objects != max_objects) { | |
923 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
924 | "should be %d", page->objects, max_objects); | |
925 | page->objects = max_objects; | |
926 | slab_fix(s, "Number of objects adjusted."); | |
927 | } | |
39b26464 | 928 | if (page->inuse != page->objects - nr) { |
70d71228 | 929 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
930 | "counted were %d", page->inuse, page->objects - nr); |
931 | page->inuse = page->objects - nr; | |
24922684 | 932 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
933 | } |
934 | return search == NULL; | |
935 | } | |
936 | ||
0121c619 CL |
937 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
938 | int alloc) | |
3ec09742 CL |
939 | { |
940 | if (s->flags & SLAB_TRACE) { | |
941 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
942 | s->name, | |
943 | alloc ? "alloc" : "free", | |
944 | object, page->inuse, | |
945 | page->freelist); | |
946 | ||
947 | if (!alloc) | |
3b0efdfa | 948 | print_section("Object ", (void *)object, s->object_size); |
3ec09742 CL |
949 | |
950 | dump_stack(); | |
951 | } | |
952 | } | |
953 | ||
c016b0bd CL |
954 | /* |
955 | * Hooks for other subsystems that check memory allocations. In a typical | |
956 | * production configuration these hooks all should produce no code at all. | |
957 | */ | |
958 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
959 | { | |
c1d50836 | 960 | flags &= gfp_allowed_mask; |
c016b0bd CL |
961 | lockdep_trace_alloc(flags); |
962 | might_sleep_if(flags & __GFP_WAIT); | |
963 | ||
3b0efdfa | 964 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
965 | } |
966 | ||
967 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | |
968 | { | |
c1d50836 | 969 | flags &= gfp_allowed_mask; |
b3d41885 | 970 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 971 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
972 | } |
973 | ||
974 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
975 | { | |
976 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 977 | |
d3f661d6 CL |
978 | /* |
979 | * Trouble is that we may no longer disable interupts in the fast path | |
980 | * So in order to make the debug calls that expect irqs to be | |
981 | * disabled we need to disable interrupts temporarily. | |
982 | */ | |
983 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
984 | { | |
985 | unsigned long flags; | |
986 | ||
987 | local_irq_save(flags); | |
3b0efdfa CL |
988 | kmemcheck_slab_free(s, x, s->object_size); |
989 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
990 | local_irq_restore(flags); |
991 | } | |
992 | #endif | |
f9b615de | 993 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 994 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
995 | } |
996 | ||
643b1138 | 997 | /* |
672bba3a | 998 | * Tracking of fully allocated slabs for debugging purposes. |
5cc6eee8 CL |
999 | * |
1000 | * list_lock must be held. | |
643b1138 | 1001 | */ |
5cc6eee8 CL |
1002 | static void add_full(struct kmem_cache *s, |
1003 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1004 | { |
5cc6eee8 CL |
1005 | if (!(s->flags & SLAB_STORE_USER)) |
1006 | return; | |
1007 | ||
643b1138 | 1008 | list_add(&page->lru, &n->full); |
643b1138 CL |
1009 | } |
1010 | ||
5cc6eee8 CL |
1011 | /* |
1012 | * list_lock must be held. | |
1013 | */ | |
643b1138 CL |
1014 | static void remove_full(struct kmem_cache *s, struct page *page) |
1015 | { | |
643b1138 CL |
1016 | if (!(s->flags & SLAB_STORE_USER)) |
1017 | return; | |
1018 | ||
643b1138 | 1019 | list_del(&page->lru); |
643b1138 CL |
1020 | } |
1021 | ||
0f389ec6 CL |
1022 | /* Tracking of the number of slabs for debugging purposes */ |
1023 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1024 | { | |
1025 | struct kmem_cache_node *n = get_node(s, node); | |
1026 | ||
1027 | return atomic_long_read(&n->nr_slabs); | |
1028 | } | |
1029 | ||
26c02cf0 AB |
1030 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1031 | { | |
1032 | return atomic_long_read(&n->nr_slabs); | |
1033 | } | |
1034 | ||
205ab99d | 1035 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1036 | { |
1037 | struct kmem_cache_node *n = get_node(s, node); | |
1038 | ||
1039 | /* | |
1040 | * May be called early in order to allocate a slab for the | |
1041 | * kmem_cache_node structure. Solve the chicken-egg | |
1042 | * dilemma by deferring the increment of the count during | |
1043 | * bootstrap (see early_kmem_cache_node_alloc). | |
1044 | */ | |
338b2642 | 1045 | if (likely(n)) { |
0f389ec6 | 1046 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1047 | atomic_long_add(objects, &n->total_objects); |
1048 | } | |
0f389ec6 | 1049 | } |
205ab99d | 1050 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1051 | { |
1052 | struct kmem_cache_node *n = get_node(s, node); | |
1053 | ||
1054 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1055 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1056 | } |
1057 | ||
1058 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1059 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1060 | void *object) | |
1061 | { | |
1062 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1063 | return; | |
1064 | ||
f7cb1933 | 1065 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1066 | init_tracking(s, object); |
1067 | } | |
1068 | ||
1537066c | 1069 | static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page, |
ce71e27c | 1070 | void *object, unsigned long addr) |
81819f0f CL |
1071 | { |
1072 | if (!check_slab(s, page)) | |
1073 | goto bad; | |
1074 | ||
81819f0f CL |
1075 | if (!check_valid_pointer(s, page, object)) { |
1076 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1077 | goto bad; |
81819f0f CL |
1078 | } |
1079 | ||
f7cb1933 | 1080 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1081 | goto bad; |
81819f0f | 1082 | |
3ec09742 CL |
1083 | /* Success perform special debug activities for allocs */ |
1084 | if (s->flags & SLAB_STORE_USER) | |
1085 | set_track(s, object, TRACK_ALLOC, addr); | |
1086 | trace(s, page, object, 1); | |
f7cb1933 | 1087 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1088 | return 1; |
3ec09742 | 1089 | |
81819f0f CL |
1090 | bad: |
1091 | if (PageSlab(page)) { | |
1092 | /* | |
1093 | * If this is a slab page then lets do the best we can | |
1094 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1095 | * as used avoids touching the remaining objects. |
81819f0f | 1096 | */ |
24922684 | 1097 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1098 | page->inuse = page->objects; |
a973e9dd | 1099 | page->freelist = NULL; |
81819f0f CL |
1100 | } |
1101 | return 0; | |
1102 | } | |
1103 | ||
19c7ff9e CL |
1104 | static noinline struct kmem_cache_node *free_debug_processing( |
1105 | struct kmem_cache *s, struct page *page, void *object, | |
1106 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1107 | { |
19c7ff9e | 1108 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1109 | |
19c7ff9e | 1110 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1111 | slab_lock(page); |
1112 | ||
81819f0f CL |
1113 | if (!check_slab(s, page)) |
1114 | goto fail; | |
1115 | ||
1116 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1117 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1118 | goto fail; |
1119 | } | |
1120 | ||
1121 | if (on_freelist(s, page, object)) { | |
24922684 | 1122 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1123 | goto fail; |
1124 | } | |
1125 | ||
f7cb1933 | 1126 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1127 | goto out; |
81819f0f | 1128 | |
1b4f59e3 | 1129 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1130 | if (!PageSlab(page)) { |
70d71228 CL |
1131 | slab_err(s, page, "Attempt to free object(0x%p) " |
1132 | "outside of slab", object); | |
1b4f59e3 | 1133 | } else if (!page->slab_cache) { |
81819f0f | 1134 | printk(KERN_ERR |
70d71228 | 1135 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1136 | object); |
70d71228 | 1137 | dump_stack(); |
06428780 | 1138 | } else |
24922684 CL |
1139 | object_err(s, page, object, |
1140 | "page slab pointer corrupt."); | |
81819f0f CL |
1141 | goto fail; |
1142 | } | |
3ec09742 | 1143 | |
3ec09742 CL |
1144 | if (s->flags & SLAB_STORE_USER) |
1145 | set_track(s, object, TRACK_FREE, addr); | |
1146 | trace(s, page, object, 0); | |
f7cb1933 | 1147 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1148 | out: |
881db7fb | 1149 | slab_unlock(page); |
19c7ff9e CL |
1150 | /* |
1151 | * Keep node_lock to preserve integrity | |
1152 | * until the object is actually freed | |
1153 | */ | |
1154 | return n; | |
3ec09742 | 1155 | |
81819f0f | 1156 | fail: |
19c7ff9e CL |
1157 | slab_unlock(page); |
1158 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1159 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1160 | return NULL; |
81819f0f CL |
1161 | } |
1162 | ||
41ecc55b CL |
1163 | static int __init setup_slub_debug(char *str) |
1164 | { | |
f0630fff CL |
1165 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1166 | if (*str++ != '=' || !*str) | |
1167 | /* | |
1168 | * No options specified. Switch on full debugging. | |
1169 | */ | |
1170 | goto out; | |
1171 | ||
1172 | if (*str == ',') | |
1173 | /* | |
1174 | * No options but restriction on slabs. This means full | |
1175 | * debugging for slabs matching a pattern. | |
1176 | */ | |
1177 | goto check_slabs; | |
1178 | ||
fa5ec8a1 DR |
1179 | if (tolower(*str) == 'o') { |
1180 | /* | |
1181 | * Avoid enabling debugging on caches if its minimum order | |
1182 | * would increase as a result. | |
1183 | */ | |
1184 | disable_higher_order_debug = 1; | |
1185 | goto out; | |
1186 | } | |
1187 | ||
f0630fff CL |
1188 | slub_debug = 0; |
1189 | if (*str == '-') | |
1190 | /* | |
1191 | * Switch off all debugging measures. | |
1192 | */ | |
1193 | goto out; | |
1194 | ||
1195 | /* | |
1196 | * Determine which debug features should be switched on | |
1197 | */ | |
06428780 | 1198 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1199 | switch (tolower(*str)) { |
1200 | case 'f': | |
1201 | slub_debug |= SLAB_DEBUG_FREE; | |
1202 | break; | |
1203 | case 'z': | |
1204 | slub_debug |= SLAB_RED_ZONE; | |
1205 | break; | |
1206 | case 'p': | |
1207 | slub_debug |= SLAB_POISON; | |
1208 | break; | |
1209 | case 'u': | |
1210 | slub_debug |= SLAB_STORE_USER; | |
1211 | break; | |
1212 | case 't': | |
1213 | slub_debug |= SLAB_TRACE; | |
1214 | break; | |
4c13dd3b DM |
1215 | case 'a': |
1216 | slub_debug |= SLAB_FAILSLAB; | |
1217 | break; | |
f0630fff CL |
1218 | default: |
1219 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1220 | "unknown. skipped\n", *str); |
f0630fff | 1221 | } |
41ecc55b CL |
1222 | } |
1223 | ||
f0630fff | 1224 | check_slabs: |
41ecc55b CL |
1225 | if (*str == ',') |
1226 | slub_debug_slabs = str + 1; | |
f0630fff | 1227 | out: |
41ecc55b CL |
1228 | return 1; |
1229 | } | |
1230 | ||
1231 | __setup("slub_debug", setup_slub_debug); | |
1232 | ||
3b0efdfa | 1233 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1234 | unsigned long flags, const char *name, |
51cc5068 | 1235 | void (*ctor)(void *)) |
41ecc55b CL |
1236 | { |
1237 | /* | |
e153362a | 1238 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1239 | */ |
6fa3eb70 S |
1240 | if(flags & SLAB_NO_DEBUG) { |
1241 | return flags; | |
1242 | } | |
1243 | ||
fe76407d CL |
1244 | if (slub_debug && (!slub_debug_slabs || (name && |
1245 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1246 | flags |= slub_debug; |
ba0268a8 CL |
1247 | |
1248 | return flags; | |
41ecc55b CL |
1249 | } |
1250 | #else | |
3ec09742 CL |
1251 | static inline void setup_object_debug(struct kmem_cache *s, |
1252 | struct page *page, void *object) {} | |
41ecc55b | 1253 | |
3ec09742 | 1254 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1255 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1256 | |
19c7ff9e CL |
1257 | static inline struct kmem_cache_node *free_debug_processing( |
1258 | struct kmem_cache *s, struct page *page, void *object, | |
1259 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1260 | |
41ecc55b CL |
1261 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1262 | { return 1; } | |
1263 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1264 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1265 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1266 | struct page *page) {} | |
2cfb7455 | 1267 | static inline void remove_full(struct kmem_cache *s, struct page *page) {} |
3b0efdfa | 1268 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1269 | unsigned long flags, const char *name, |
51cc5068 | 1270 | void (*ctor)(void *)) |
ba0268a8 CL |
1271 | { |
1272 | return flags; | |
1273 | } | |
41ecc55b | 1274 | #define slub_debug 0 |
0f389ec6 | 1275 | |
fdaa45e9 IM |
1276 | #define disable_higher_order_debug 0 |
1277 | ||
0f389ec6 CL |
1278 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1279 | { return 0; } | |
26c02cf0 AB |
1280 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1281 | { return 0; } | |
205ab99d CL |
1282 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1283 | int objects) {} | |
1284 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1285 | int objects) {} | |
7d550c56 CL |
1286 | |
1287 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |
1288 | { return 0; } | |
1289 | ||
1290 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
1291 | void *object) {} | |
1292 | ||
1293 | static inline void slab_free_hook(struct kmem_cache *s, void *x) {} | |
1294 | ||
ab4d5ed5 | 1295 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1296 | |
81819f0f CL |
1297 | /* |
1298 | * Slab allocation and freeing | |
1299 | */ | |
65c3376a CL |
1300 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1301 | struct kmem_cache_order_objects oo) | |
1302 | { | |
1303 | int order = oo_order(oo); | |
1304 | ||
b1eeab67 VN |
1305 | flags |= __GFP_NOTRACK; |
1306 | ||
2154a336 | 1307 | if (node == NUMA_NO_NODE) |
6fa3eb70 | 1308 | #ifndef CONFIG_MTK_PAGERECORDER |
65c3376a | 1309 | return alloc_pages(flags, order); |
6fa3eb70 S |
1310 | #else |
1311 | return alloc_pages_nopagedebug(flags, order); | |
1312 | #endif | |
65c3376a | 1313 | else |
6b65aaf3 | 1314 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1315 | } |
1316 | ||
81819f0f CL |
1317 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1318 | { | |
06428780 | 1319 | struct page *page; |
834f3d11 | 1320 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1321 | gfp_t alloc_gfp; |
81819f0f | 1322 | |
7e0528da CL |
1323 | flags &= gfp_allowed_mask; |
1324 | ||
1325 | if (flags & __GFP_WAIT) | |
1326 | local_irq_enable(); | |
1327 | ||
b7a49f0d | 1328 | flags |= s->allocflags; |
e12ba74d | 1329 | |
ba52270d PE |
1330 | /* |
1331 | * Let the initial higher-order allocation fail under memory pressure | |
1332 | * so we fall-back to the minimum order allocation. | |
1333 | */ | |
1334 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1335 | ||
1336 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1337 | if (unlikely(!page)) { |
1338 | oo = s->min; | |
1339 | /* | |
1340 | * Allocation may have failed due to fragmentation. | |
1341 | * Try a lower order alloc if possible | |
1342 | */ | |
1343 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1344 | |
7e0528da CL |
1345 | if (page) |
1346 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1347 | } |
5a896d9e | 1348 | |
737b719e | 1349 | if (kmemcheck_enabled && page |
5086c389 | 1350 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1351 | int pages = 1 << oo_order(oo); |
1352 | ||
1353 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1354 | ||
1355 | /* | |
1356 | * Objects from caches that have a constructor don't get | |
1357 | * cleared when they're allocated, so we need to do it here. | |
1358 | */ | |
1359 | if (s->ctor) | |
1360 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1361 | else | |
1362 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1363 | } |
1364 | ||
737b719e DR |
1365 | if (flags & __GFP_WAIT) |
1366 | local_irq_disable(); | |
1367 | if (!page) | |
1368 | return NULL; | |
1369 | ||
834f3d11 | 1370 | page->objects = oo_objects(oo); |
81819f0f CL |
1371 | mod_zone_page_state(page_zone(page), |
1372 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1373 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1374 | 1 << oo_order(oo)); |
81819f0f CL |
1375 | |
1376 | return page; | |
1377 | } | |
1378 | ||
1379 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1380 | void *object) | |
1381 | { | |
3ec09742 | 1382 | setup_object_debug(s, page, object); |
4f104934 | 1383 | if (unlikely(s->ctor)) |
51cc5068 | 1384 | s->ctor(object); |
81819f0f CL |
1385 | } |
1386 | ||
1387 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1388 | { | |
1389 | struct page *page; | |
81819f0f | 1390 | void *start; |
81819f0f CL |
1391 | void *last; |
1392 | void *p; | |
1f458cbf | 1393 | int order; |
81819f0f | 1394 | |
6cb06229 | 1395 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1396 | |
6cb06229 CL |
1397 | page = allocate_slab(s, |
1398 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1399 | if (!page) |
1400 | goto out; | |
1401 | ||
1f458cbf | 1402 | order = compound_order(page); |
205ab99d | 1403 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1f458cbf | 1404 | memcg_bind_pages(s, order); |
1b4f59e3 | 1405 | page->slab_cache = s; |
c03f94cc | 1406 | __SetPageSlab(page); |
072bb0aa MG |
1407 | if (page->pfmemalloc) |
1408 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1409 | |
1410 | start = page_address(page); | |
81819f0f CL |
1411 | |
1412 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1413 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f CL |
1414 | |
1415 | last = start; | |
224a88be | 1416 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1417 | setup_object(s, page, last); |
1418 | set_freepointer(s, last, p); | |
1419 | last = p; | |
1420 | } | |
1421 | setup_object(s, page, last); | |
a973e9dd | 1422 | set_freepointer(s, last, NULL); |
81819f0f CL |
1423 | |
1424 | page->freelist = start; | |
e6e82ea1 | 1425 | page->inuse = page->objects; |
8cb0a506 | 1426 | page->frozen = 1; |
81819f0f | 1427 | out: |
81819f0f CL |
1428 | return page; |
1429 | } | |
1430 | ||
1431 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1432 | { | |
834f3d11 CL |
1433 | int order = compound_order(page); |
1434 | int pages = 1 << order; | |
81819f0f | 1435 | |
af537b0a | 1436 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1437 | void *p; |
1438 | ||
1439 | slab_pad_check(s, page); | |
224a88be CL |
1440 | for_each_object(p, s, page_address(page), |
1441 | page->objects) | |
f7cb1933 | 1442 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1443 | } |
1444 | ||
b1eeab67 | 1445 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1446 | |
81819f0f CL |
1447 | mod_zone_page_state(page_zone(page), |
1448 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1449 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1450 | -pages); |
81819f0f | 1451 | |
072bb0aa | 1452 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1453 | __ClearPageSlab(page); |
1f458cbf GC |
1454 | |
1455 | memcg_release_pages(s, order); | |
22b751c3 | 1456 | page_mapcount_reset(page); |
1eb5ac64 NP |
1457 | if (current->reclaim_state) |
1458 | current->reclaim_state->reclaimed_slab += pages; | |
d79923fa | 1459 | __free_memcg_kmem_pages(page, order); |
81819f0f CL |
1460 | } |
1461 | ||
da9a638c LJ |
1462 | #define need_reserve_slab_rcu \ |
1463 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1464 | ||
81819f0f CL |
1465 | static void rcu_free_slab(struct rcu_head *h) |
1466 | { | |
1467 | struct page *page; | |
1468 | ||
da9a638c LJ |
1469 | if (need_reserve_slab_rcu) |
1470 | page = virt_to_head_page(h); | |
1471 | else | |
1472 | page = container_of((struct list_head *)h, struct page, lru); | |
1473 | ||
1b4f59e3 | 1474 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1475 | } |
1476 | ||
1477 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1478 | { | |
1479 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1480 | struct rcu_head *head; |
1481 | ||
1482 | if (need_reserve_slab_rcu) { | |
1483 | int order = compound_order(page); | |
1484 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1485 | ||
1486 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1487 | head = page_address(page) + offset; | |
1488 | } else { | |
1489 | /* | |
1490 | * RCU free overloads the RCU head over the LRU | |
1491 | */ | |
1492 | head = (void *)&page->lru; | |
1493 | } | |
81819f0f CL |
1494 | |
1495 | call_rcu(head, rcu_free_slab); | |
1496 | } else | |
1497 | __free_slab(s, page); | |
1498 | } | |
1499 | ||
1500 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1501 | { | |
205ab99d | 1502 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1503 | free_slab(s, page); |
1504 | } | |
1505 | ||
1506 | /* | |
5cc6eee8 CL |
1507 | * Management of partially allocated slabs. |
1508 | * | |
1509 | * list_lock must be held. | |
81819f0f | 1510 | */ |
5cc6eee8 | 1511 | static inline void add_partial(struct kmem_cache_node *n, |
7c2e132c | 1512 | struct page *page, int tail) |
81819f0f | 1513 | { |
e95eed57 | 1514 | n->nr_partial++; |
136333d1 | 1515 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1516 | list_add_tail(&page->lru, &n->partial); |
1517 | else | |
1518 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1519 | } |
1520 | ||
5cc6eee8 CL |
1521 | /* |
1522 | * list_lock must be held. | |
1523 | */ | |
1524 | static inline void remove_partial(struct kmem_cache_node *n, | |
62e346a8 CL |
1525 | struct page *page) |
1526 | { | |
1527 | list_del(&page->lru); | |
1528 | n->nr_partial--; | |
1529 | } | |
1530 | ||
81819f0f | 1531 | /* |
7ced3719 CL |
1532 | * Remove slab from the partial list, freeze it and |
1533 | * return the pointer to the freelist. | |
81819f0f | 1534 | * |
497b66f2 CL |
1535 | * Returns a list of objects or NULL if it fails. |
1536 | * | |
7ced3719 | 1537 | * Must hold list_lock since we modify the partial list. |
81819f0f | 1538 | */ |
497b66f2 | 1539 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1540 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1541 | int mode, int *objects) |
81819f0f | 1542 | { |
2cfb7455 CL |
1543 | void *freelist; |
1544 | unsigned long counters; | |
1545 | struct page new; | |
1546 | ||
2cfb7455 CL |
1547 | /* |
1548 | * Zap the freelist and set the frozen bit. | |
1549 | * The old freelist is the list of objects for the | |
1550 | * per cpu allocation list. | |
1551 | */ | |
7ced3719 CL |
1552 | freelist = page->freelist; |
1553 | counters = page->counters; | |
1554 | new.counters = counters; | |
633b0764 | 1555 | *objects = new.objects - new.inuse; |
23910c50 | 1556 | if (mode) { |
7ced3719 | 1557 | new.inuse = page->objects; |
23910c50 PE |
1558 | new.freelist = NULL; |
1559 | } else { | |
1560 | new.freelist = freelist; | |
1561 | } | |
2cfb7455 | 1562 | |
7ced3719 CL |
1563 | VM_BUG_ON(new.frozen); |
1564 | new.frozen = 1; | |
2cfb7455 | 1565 | |
7ced3719 | 1566 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1567 | freelist, counters, |
02d7633f | 1568 | new.freelist, new.counters, |
7ced3719 | 1569 | "acquire_slab")) |
7ced3719 | 1570 | return NULL; |
2cfb7455 CL |
1571 | |
1572 | remove_partial(n, page); | |
7ced3719 | 1573 | WARN_ON(!freelist); |
49e22585 | 1574 | return freelist; |
81819f0f CL |
1575 | } |
1576 | ||
633b0764 | 1577 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1578 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1579 | |
81819f0f | 1580 | /* |
672bba3a | 1581 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1582 | */ |
8ba00bb6 JK |
1583 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1584 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1585 | { |
49e22585 CL |
1586 | struct page *page, *page2; |
1587 | void *object = NULL; | |
633b0764 JK |
1588 | int available = 0; |
1589 | int objects; | |
81819f0f CL |
1590 | |
1591 | /* | |
1592 | * Racy check. If we mistakenly see no partial slabs then we | |
1593 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1594 | * partial slab and there is none available then get_partials() |
1595 | * will return NULL. | |
81819f0f CL |
1596 | */ |
1597 | if (!n || !n->nr_partial) | |
1598 | return NULL; | |
1599 | ||
1600 | spin_lock(&n->list_lock); | |
49e22585 | 1601 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1602 | void *t; |
49e22585 | 1603 | |
8ba00bb6 JK |
1604 | if (!pfmemalloc_match(page, flags)) |
1605 | continue; | |
1606 | ||
633b0764 | 1607 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1608 | if (!t) |
1609 | break; | |
1610 | ||
633b0764 | 1611 | available += objects; |
12d79634 | 1612 | if (!object) { |
49e22585 | 1613 | c->page = page; |
49e22585 | 1614 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1615 | object = t; |
49e22585 | 1616 | } else { |
633b0764 | 1617 | put_cpu_partial(s, page, 0); |
8028dcea | 1618 | stat(s, CPU_PARTIAL_NODE); |
49e22585 CL |
1619 | } |
1620 | if (kmem_cache_debug(s) || available > s->cpu_partial / 2) | |
1621 | break; | |
1622 | ||
497b66f2 | 1623 | } |
81819f0f | 1624 | spin_unlock(&n->list_lock); |
497b66f2 | 1625 | return object; |
81819f0f CL |
1626 | } |
1627 | ||
1628 | /* | |
672bba3a | 1629 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1630 | */ |
de3ec035 | 1631 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1632 | struct kmem_cache_cpu *c) |
81819f0f CL |
1633 | { |
1634 | #ifdef CONFIG_NUMA | |
1635 | struct zonelist *zonelist; | |
dd1a239f | 1636 | struct zoneref *z; |
54a6eb5c MG |
1637 | struct zone *zone; |
1638 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1639 | void *object; |
cc9a6c87 | 1640 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1641 | |
1642 | /* | |
672bba3a CL |
1643 | * The defrag ratio allows a configuration of the tradeoffs between |
1644 | * inter node defragmentation and node local allocations. A lower | |
1645 | * defrag_ratio increases the tendency to do local allocations | |
1646 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1647 | * |
672bba3a CL |
1648 | * If the defrag_ratio is set to 0 then kmalloc() always |
1649 | * returns node local objects. If the ratio is higher then kmalloc() | |
1650 | * may return off node objects because partial slabs are obtained | |
1651 | * from other nodes and filled up. | |
81819f0f | 1652 | * |
6446faa2 | 1653 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1654 | * defrag_ratio = 1000) then every (well almost) allocation will |
1655 | * first attempt to defrag slab caches on other nodes. This means | |
1656 | * scanning over all nodes to look for partial slabs which may be | |
1657 | * expensive if we do it every time we are trying to find a slab | |
1658 | * with available objects. | |
81819f0f | 1659 | */ |
9824601e CL |
1660 | if (!s->remote_node_defrag_ratio || |
1661 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1662 | return NULL; |
1663 | ||
cc9a6c87 MG |
1664 | do { |
1665 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 1666 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 MG |
1667 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1668 | struct kmem_cache_node *n; | |
1669 | ||
1670 | n = get_node(s, zone_to_nid(zone)); | |
1671 | ||
1672 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1673 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1674 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1675 | if (object) { |
1676 | /* | |
1677 | * Return the object even if | |
1678 | * put_mems_allowed indicated that | |
1679 | * the cpuset mems_allowed was | |
1680 | * updated in parallel. It's a | |
1681 | * harmless race between the alloc | |
1682 | * and the cpuset update. | |
1683 | */ | |
1684 | put_mems_allowed(cpuset_mems_cookie); | |
1685 | return object; | |
1686 | } | |
c0ff7453 | 1687 | } |
81819f0f | 1688 | } |
cc9a6c87 | 1689 | } while (!put_mems_allowed(cpuset_mems_cookie)); |
81819f0f CL |
1690 | #endif |
1691 | return NULL; | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * Get a partial page, lock it and return it. | |
1696 | */ | |
497b66f2 | 1697 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1698 | struct kmem_cache_cpu *c) |
81819f0f | 1699 | { |
497b66f2 | 1700 | void *object; |
2154a336 | 1701 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1702 | |
8ba00bb6 | 1703 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1704 | if (object || node != NUMA_NO_NODE) |
1705 | return object; | |
81819f0f | 1706 | |
acd19fd1 | 1707 | return get_any_partial(s, flags, c); |
81819f0f CL |
1708 | } |
1709 | ||
8a5ec0ba CL |
1710 | #ifdef CONFIG_PREEMPT |
1711 | /* | |
1712 | * Calculate the next globally unique transaction for disambiguiation | |
1713 | * during cmpxchg. The transactions start with the cpu number and are then | |
1714 | * incremented by CONFIG_NR_CPUS. | |
1715 | */ | |
1716 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1717 | #else | |
1718 | /* | |
1719 | * No preemption supported therefore also no need to check for | |
1720 | * different cpus. | |
1721 | */ | |
1722 | #define TID_STEP 1 | |
1723 | #endif | |
1724 | ||
1725 | static inline unsigned long next_tid(unsigned long tid) | |
1726 | { | |
1727 | return tid + TID_STEP; | |
1728 | } | |
1729 | ||
1730 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1731 | { | |
1732 | return tid % TID_STEP; | |
1733 | } | |
1734 | ||
1735 | static inline unsigned long tid_to_event(unsigned long tid) | |
1736 | { | |
1737 | return tid / TID_STEP; | |
1738 | } | |
1739 | ||
1740 | static inline unsigned int init_tid(int cpu) | |
1741 | { | |
1742 | return cpu; | |
1743 | } | |
1744 | ||
1745 | static inline void note_cmpxchg_failure(const char *n, | |
1746 | const struct kmem_cache *s, unsigned long tid) | |
1747 | { | |
1748 | #ifdef SLUB_DEBUG_CMPXCHG | |
1749 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1750 | ||
1751 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1752 | ||
1753 | #ifdef CONFIG_PREEMPT | |
1754 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1755 | printk("due to cpu change %d -> %d\n", | |
1756 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1757 | else | |
1758 | #endif | |
1759 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1760 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1761 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1762 | else | |
1763 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1764 | actual_tid, tid, next_tid(tid)); | |
1765 | #endif | |
4fdccdfb | 1766 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1767 | } |
1768 | ||
788e1aad | 1769 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1770 | { |
8a5ec0ba CL |
1771 | int cpu; |
1772 | ||
1773 | for_each_possible_cpu(cpu) | |
1774 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1775 | } |
2cfb7455 | 1776 | |
81819f0f CL |
1777 | /* |
1778 | * Remove the cpu slab | |
1779 | */ | |
c17dda40 | 1780 | static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist) |
81819f0f | 1781 | { |
2cfb7455 | 1782 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1783 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1784 | int lock = 0; | |
1785 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1786 | void *nextfree; |
136333d1 | 1787 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1788 | struct page new; |
1789 | struct page old; | |
1790 | ||
1791 | if (page->freelist) { | |
84e554e6 | 1792 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1793 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1794 | } |
1795 | ||
894b8788 | 1796 | /* |
2cfb7455 CL |
1797 | * Stage one: Free all available per cpu objects back |
1798 | * to the page freelist while it is still frozen. Leave the | |
1799 | * last one. | |
1800 | * | |
1801 | * There is no need to take the list->lock because the page | |
1802 | * is still frozen. | |
1803 | */ | |
1804 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1805 | void *prior; | |
1806 | unsigned long counters; | |
1807 | ||
1808 | do { | |
1809 | prior = page->freelist; | |
1810 | counters = page->counters; | |
1811 | set_freepointer(s, freelist, prior); | |
1812 | new.counters = counters; | |
1813 | new.inuse--; | |
1814 | VM_BUG_ON(!new.frozen); | |
1815 | ||
1d07171c | 1816 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1817 | prior, counters, |
1818 | freelist, new.counters, | |
1819 | "drain percpu freelist")); | |
1820 | ||
1821 | freelist = nextfree; | |
1822 | } | |
1823 | ||
894b8788 | 1824 | /* |
2cfb7455 CL |
1825 | * Stage two: Ensure that the page is unfrozen while the |
1826 | * list presence reflects the actual number of objects | |
1827 | * during unfreeze. | |
1828 | * | |
1829 | * We setup the list membership and then perform a cmpxchg | |
1830 | * with the count. If there is a mismatch then the page | |
1831 | * is not unfrozen but the page is on the wrong list. | |
1832 | * | |
1833 | * Then we restart the process which may have to remove | |
1834 | * the page from the list that we just put it on again | |
1835 | * because the number of objects in the slab may have | |
1836 | * changed. | |
894b8788 | 1837 | */ |
2cfb7455 | 1838 | redo: |
894b8788 | 1839 | |
2cfb7455 CL |
1840 | old.freelist = page->freelist; |
1841 | old.counters = page->counters; | |
1842 | VM_BUG_ON(!old.frozen); | |
7c2e132c | 1843 | |
2cfb7455 CL |
1844 | /* Determine target state of the slab */ |
1845 | new.counters = old.counters; | |
1846 | if (freelist) { | |
1847 | new.inuse--; | |
1848 | set_freepointer(s, freelist, old.freelist); | |
1849 | new.freelist = freelist; | |
1850 | } else | |
1851 | new.freelist = old.freelist; | |
1852 | ||
1853 | new.frozen = 0; | |
1854 | ||
81107188 | 1855 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1856 | m = M_FREE; |
1857 | else if (new.freelist) { | |
1858 | m = M_PARTIAL; | |
1859 | if (!lock) { | |
1860 | lock = 1; | |
1861 | /* | |
1862 | * Taking the spinlock removes the possiblity | |
1863 | * that acquire_slab() will see a slab page that | |
1864 | * is frozen | |
1865 | */ | |
1866 | spin_lock(&n->list_lock); | |
1867 | } | |
1868 | } else { | |
1869 | m = M_FULL; | |
1870 | if (kmem_cache_debug(s) && !lock) { | |
1871 | lock = 1; | |
1872 | /* | |
1873 | * This also ensures that the scanning of full | |
1874 | * slabs from diagnostic functions will not see | |
1875 | * any frozen slabs. | |
1876 | */ | |
1877 | spin_lock(&n->list_lock); | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | if (l != m) { | |
1882 | ||
1883 | if (l == M_PARTIAL) | |
1884 | ||
1885 | remove_partial(n, page); | |
1886 | ||
1887 | else if (l == M_FULL) | |
894b8788 | 1888 | |
2cfb7455 CL |
1889 | remove_full(s, page); |
1890 | ||
1891 | if (m == M_PARTIAL) { | |
1892 | ||
1893 | add_partial(n, page, tail); | |
136333d1 | 1894 | stat(s, tail); |
2cfb7455 CL |
1895 | |
1896 | } else if (m == M_FULL) { | |
894b8788 | 1897 | |
2cfb7455 CL |
1898 | stat(s, DEACTIVATE_FULL); |
1899 | add_full(s, n, page); | |
1900 | ||
1901 | } | |
1902 | } | |
1903 | ||
1904 | l = m; | |
1d07171c | 1905 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1906 | old.freelist, old.counters, |
1907 | new.freelist, new.counters, | |
1908 | "unfreezing slab")) | |
1909 | goto redo; | |
1910 | ||
2cfb7455 CL |
1911 | if (lock) |
1912 | spin_unlock(&n->list_lock); | |
1913 | ||
1914 | if (m == M_FREE) { | |
1915 | stat(s, DEACTIVATE_EMPTY); | |
1916 | discard_slab(s, page); | |
1917 | stat(s, FREE_SLAB); | |
894b8788 | 1918 | } |
81819f0f CL |
1919 | } |
1920 | ||
d24ac77f JK |
1921 | /* |
1922 | * Unfreeze all the cpu partial slabs. | |
1923 | * | |
59a09917 CL |
1924 | * This function must be called with interrupts disabled |
1925 | * for the cpu using c (or some other guarantee must be there | |
1926 | * to guarantee no concurrent accesses). | |
d24ac77f | 1927 | */ |
59a09917 CL |
1928 | static void unfreeze_partials(struct kmem_cache *s, |
1929 | struct kmem_cache_cpu *c) | |
49e22585 | 1930 | { |
43d77867 | 1931 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1932 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1933 | |
1934 | while ((page = c->partial)) { | |
49e22585 CL |
1935 | struct page new; |
1936 | struct page old; | |
1937 | ||
1938 | c->partial = page->next; | |
43d77867 JK |
1939 | |
1940 | n2 = get_node(s, page_to_nid(page)); | |
1941 | if (n != n2) { | |
1942 | if (n) | |
1943 | spin_unlock(&n->list_lock); | |
1944 | ||
1945 | n = n2; | |
1946 | spin_lock(&n->list_lock); | |
1947 | } | |
49e22585 CL |
1948 | |
1949 | do { | |
1950 | ||
1951 | old.freelist = page->freelist; | |
1952 | old.counters = page->counters; | |
1953 | VM_BUG_ON(!old.frozen); | |
1954 | ||
1955 | new.counters = old.counters; | |
1956 | new.freelist = old.freelist; | |
1957 | ||
1958 | new.frozen = 0; | |
1959 | ||
d24ac77f | 1960 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1961 | old.freelist, old.counters, |
1962 | new.freelist, new.counters, | |
1963 | "unfreezing slab")); | |
1964 | ||
43d77867 | 1965 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1966 | page->next = discard_page; |
1967 | discard_page = page; | |
43d77867 JK |
1968 | } else { |
1969 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1970 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1971 | } |
1972 | } | |
1973 | ||
1974 | if (n) | |
1975 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1976 | |
1977 | while (discard_page) { | |
1978 | page = discard_page; | |
1979 | discard_page = discard_page->next; | |
1980 | ||
1981 | stat(s, DEACTIVATE_EMPTY); | |
1982 | discard_slab(s, page); | |
1983 | stat(s, FREE_SLAB); | |
1984 | } | |
49e22585 CL |
1985 | } |
1986 | ||
1987 | /* | |
1988 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1989 | * slot if available. This is done without interrupts disabled and without | |
1990 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
1991 | * onto a random cpus partial slot. | |
1992 | * | |
1993 | * If we did not find a slot then simply move all the partials to the | |
1994 | * per node partial list. | |
1995 | */ | |
633b0764 | 1996 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 CL |
1997 | { |
1998 | struct page *oldpage; | |
1999 | int pages; | |
2000 | int pobjects; | |
2001 | ||
2002 | do { | |
2003 | pages = 0; | |
2004 | pobjects = 0; | |
2005 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2006 | ||
2007 | if (oldpage) { | |
2008 | pobjects = oldpage->pobjects; | |
2009 | pages = oldpage->pages; | |
2010 | if (drain && pobjects > s->cpu_partial) { | |
2011 | unsigned long flags; | |
2012 | /* | |
2013 | * partial array is full. Move the existing | |
2014 | * set to the per node partial list. | |
2015 | */ | |
2016 | local_irq_save(flags); | |
59a09917 | 2017 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2018 | local_irq_restore(flags); |
e24fc410 | 2019 | oldpage = NULL; |
49e22585 CL |
2020 | pobjects = 0; |
2021 | pages = 0; | |
8028dcea | 2022 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2023 | } |
2024 | } | |
2025 | ||
2026 | pages++; | |
2027 | pobjects += page->objects - page->inuse; | |
2028 | ||
2029 | page->pages = pages; | |
2030 | page->pobjects = pobjects; | |
2031 | page->next = oldpage; | |
2032 | ||
933393f5 | 2033 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage); |
49e22585 CL |
2034 | } |
2035 | ||
dfb4f096 | 2036 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2037 | { |
84e554e6 | 2038 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2039 | deactivate_slab(s, c->page, c->freelist); |
2040 | ||
2041 | c->tid = next_tid(c->tid); | |
2042 | c->page = NULL; | |
2043 | c->freelist = NULL; | |
81819f0f CL |
2044 | } |
2045 | ||
2046 | /* | |
2047 | * Flush cpu slab. | |
6446faa2 | 2048 | * |
81819f0f CL |
2049 | * Called from IPI handler with interrupts disabled. |
2050 | */ | |
0c710013 | 2051 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2052 | { |
9dfc6e68 | 2053 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2054 | |
49e22585 CL |
2055 | if (likely(c)) { |
2056 | if (c->page) | |
2057 | flush_slab(s, c); | |
2058 | ||
59a09917 | 2059 | unfreeze_partials(s, c); |
49e22585 | 2060 | } |
81819f0f CL |
2061 | } |
2062 | ||
2063 | static void flush_cpu_slab(void *d) | |
2064 | { | |
2065 | struct kmem_cache *s = d; | |
81819f0f | 2066 | |
dfb4f096 | 2067 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2068 | } |
2069 | ||
a8364d55 GBY |
2070 | static bool has_cpu_slab(int cpu, void *info) |
2071 | { | |
2072 | struct kmem_cache *s = info; | |
2073 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2074 | ||
02e1a9cd | 2075 | return c->page || c->partial; |
a8364d55 GBY |
2076 | } |
2077 | ||
81819f0f CL |
2078 | static void flush_all(struct kmem_cache *s) |
2079 | { | |
a8364d55 | 2080 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2081 | } |
2082 | ||
dfb4f096 CL |
2083 | /* |
2084 | * Check if the objects in a per cpu structure fit numa | |
2085 | * locality expectations. | |
2086 | */ | |
57d437d2 | 2087 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2088 | { |
2089 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2090 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2091 | return 0; |
2092 | #endif | |
2093 | return 1; | |
2094 | } | |
2095 | ||
781b2ba6 PE |
2096 | static int count_free(struct page *page) |
2097 | { | |
2098 | return page->objects - page->inuse; | |
2099 | } | |
2100 | ||
2101 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2102 | int (*get_count)(struct page *)) | |
2103 | { | |
2104 | unsigned long flags; | |
2105 | unsigned long x = 0; | |
2106 | struct page *page; | |
2107 | ||
2108 | spin_lock_irqsave(&n->list_lock, flags); | |
2109 | list_for_each_entry(page, &n->partial, lru) | |
2110 | x += get_count(page); | |
2111 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2112 | return x; | |
2113 | } | |
2114 | ||
26c02cf0 AB |
2115 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2116 | { | |
2117 | #ifdef CONFIG_SLUB_DEBUG | |
2118 | return atomic_long_read(&n->total_objects); | |
2119 | #else | |
2120 | return 0; | |
2121 | #endif | |
2122 | } | |
2123 | ||
781b2ba6 PE |
2124 | static noinline void |
2125 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2126 | { | |
2127 | int node; | |
2128 | ||
2129 | printk(KERN_WARNING | |
2130 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2131 | nid, gfpflags); | |
2132 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2133 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2134 | s->size, oo_order(s->oo), oo_order(s->min)); |
2135 | ||
3b0efdfa | 2136 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2137 | printk(KERN_WARNING " %s debugging increased min order, use " |
2138 | "slub_debug=O to disable.\n", s->name); | |
2139 | ||
781b2ba6 PE |
2140 | for_each_online_node(node) { |
2141 | struct kmem_cache_node *n = get_node(s, node); | |
2142 | unsigned long nr_slabs; | |
2143 | unsigned long nr_objs; | |
2144 | unsigned long nr_free; | |
2145 | ||
2146 | if (!n) | |
2147 | continue; | |
2148 | ||
26c02cf0 AB |
2149 | nr_free = count_partial(n, count_free); |
2150 | nr_slabs = node_nr_slabs(n); | |
2151 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2152 | |
2153 | printk(KERN_WARNING | |
2154 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2155 | node, nr_slabs, nr_objs, nr_free); | |
2156 | } | |
2157 | } | |
2158 | ||
497b66f2 CL |
2159 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2160 | int node, struct kmem_cache_cpu **pc) | |
2161 | { | |
6faa6833 | 2162 | void *freelist; |
188fd063 CL |
2163 | struct kmem_cache_cpu *c = *pc; |
2164 | struct page *page; | |
497b66f2 | 2165 | |
188fd063 | 2166 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2167 | |
188fd063 CL |
2168 | if (freelist) |
2169 | return freelist; | |
2170 | ||
2171 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2172 | if (page) { |
2173 | c = __this_cpu_ptr(s->cpu_slab); | |
2174 | if (c->page) | |
2175 | flush_slab(s, c); | |
2176 | ||
2177 | /* | |
2178 | * No other reference to the page yet so we can | |
2179 | * muck around with it freely without cmpxchg | |
2180 | */ | |
6faa6833 | 2181 | freelist = page->freelist; |
497b66f2 CL |
2182 | page->freelist = NULL; |
2183 | ||
2184 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2185 | c->page = page; |
2186 | *pc = c; | |
2187 | } else | |
6faa6833 | 2188 | freelist = NULL; |
497b66f2 | 2189 | |
6faa6833 | 2190 | return freelist; |
497b66f2 CL |
2191 | } |
2192 | ||
072bb0aa MG |
2193 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2194 | { | |
2195 | if (unlikely(PageSlabPfmemalloc(page))) | |
2196 | return gfp_pfmemalloc_allowed(gfpflags); | |
2197 | ||
2198 | return true; | |
2199 | } | |
2200 | ||
213eeb9f CL |
2201 | /* |
2202 | * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist | |
2203 | * or deactivate the page. | |
2204 | * | |
2205 | * The page is still frozen if the return value is not NULL. | |
2206 | * | |
2207 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2208 | * |
2209 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2210 | */ |
2211 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2212 | { | |
2213 | struct page new; | |
2214 | unsigned long counters; | |
2215 | void *freelist; | |
2216 | ||
2217 | do { | |
2218 | freelist = page->freelist; | |
2219 | counters = page->counters; | |
6faa6833 | 2220 | |
213eeb9f CL |
2221 | new.counters = counters; |
2222 | VM_BUG_ON(!new.frozen); | |
2223 | ||
2224 | new.inuse = page->objects; | |
2225 | new.frozen = freelist != NULL; | |
2226 | ||
d24ac77f | 2227 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2228 | freelist, counters, |
2229 | NULL, new.counters, | |
2230 | "get_freelist")); | |
2231 | ||
2232 | return freelist; | |
2233 | } | |
2234 | ||
81819f0f | 2235 | /* |
894b8788 CL |
2236 | * Slow path. The lockless freelist is empty or we need to perform |
2237 | * debugging duties. | |
2238 | * | |
894b8788 CL |
2239 | * Processing is still very fast if new objects have been freed to the |
2240 | * regular freelist. In that case we simply take over the regular freelist | |
2241 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2242 | * |
894b8788 CL |
2243 | * If that is not working then we fall back to the partial lists. We take the |
2244 | * first element of the freelist as the object to allocate now and move the | |
2245 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2246 | * |
894b8788 | 2247 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2248 | * we need to allocate a new slab. This is the slowest path since it involves |
2249 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2250 | */ |
ce71e27c EGM |
2251 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2252 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2253 | { |
6faa6833 | 2254 | void *freelist; |
f6e7def7 | 2255 | struct page *page; |
8a5ec0ba CL |
2256 | unsigned long flags; |
2257 | ||
2258 | local_irq_save(flags); | |
2259 | #ifdef CONFIG_PREEMPT | |
2260 | /* | |
2261 | * We may have been preempted and rescheduled on a different | |
2262 | * cpu before disabling interrupts. Need to reload cpu area | |
2263 | * pointer. | |
2264 | */ | |
2265 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2266 | #endif |
81819f0f | 2267 | |
f6e7def7 CL |
2268 | page = c->page; |
2269 | if (!page) | |
81819f0f | 2270 | goto new_slab; |
49e22585 | 2271 | redo: |
6faa6833 | 2272 | |
57d437d2 | 2273 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2274 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2275 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2276 | c->page = NULL; |
2277 | c->freelist = NULL; | |
fc59c053 CL |
2278 | goto new_slab; |
2279 | } | |
6446faa2 | 2280 | |
072bb0aa MG |
2281 | /* |
2282 | * By rights, we should be searching for a slab page that was | |
2283 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2284 | * information when the page leaves the per-cpu allocator | |
2285 | */ | |
2286 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2287 | deactivate_slab(s, page, c->freelist); | |
2288 | c->page = NULL; | |
2289 | c->freelist = NULL; | |
2290 | goto new_slab; | |
2291 | } | |
2292 | ||
73736e03 | 2293 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2294 | freelist = c->freelist; |
2295 | if (freelist) | |
73736e03 | 2296 | goto load_freelist; |
03e404af | 2297 | |
2cfb7455 | 2298 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2299 | |
f6e7def7 | 2300 | freelist = get_freelist(s, page); |
6446faa2 | 2301 | |
6faa6833 | 2302 | if (!freelist) { |
03e404af CL |
2303 | c->page = NULL; |
2304 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2305 | goto new_slab; |
03e404af | 2306 | } |
6446faa2 | 2307 | |
84e554e6 | 2308 | stat(s, ALLOC_REFILL); |
6446faa2 | 2309 | |
894b8788 | 2310 | load_freelist: |
507effea CL |
2311 | /* |
2312 | * freelist is pointing to the list of objects to be used. | |
2313 | * page is pointing to the page from which the objects are obtained. | |
2314 | * That page must be frozen for per cpu allocations to work. | |
2315 | */ | |
2316 | VM_BUG_ON(!c->page->frozen); | |
6faa6833 | 2317 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2318 | c->tid = next_tid(c->tid); |
2319 | local_irq_restore(flags); | |
6faa6833 | 2320 | return freelist; |
81819f0f | 2321 | |
81819f0f | 2322 | new_slab: |
2cfb7455 | 2323 | |
49e22585 | 2324 | if (c->partial) { |
f6e7def7 CL |
2325 | page = c->page = c->partial; |
2326 | c->partial = page->next; | |
49e22585 CL |
2327 | stat(s, CPU_PARTIAL_ALLOC); |
2328 | c->freelist = NULL; | |
2329 | goto redo; | |
81819f0f CL |
2330 | } |
2331 | ||
188fd063 | 2332 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2333 | |
f4697436 CL |
2334 | if (unlikely(!freelist)) { |
2335 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2336 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2337 | |
f4697436 CL |
2338 | local_irq_restore(flags); |
2339 | return NULL; | |
81819f0f | 2340 | } |
2cfb7455 | 2341 | |
f6e7def7 | 2342 | page = c->page; |
5091b74a | 2343 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2344 | goto load_freelist; |
2cfb7455 | 2345 | |
497b66f2 | 2346 | /* Only entered in the debug case */ |
5091b74a | 2347 | if (kmem_cache_debug(s) && !alloc_debug_processing(s, page, freelist, addr)) |
497b66f2 | 2348 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2349 | |
f6e7def7 | 2350 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2351 | c->page = NULL; |
2352 | c->freelist = NULL; | |
a71ae47a | 2353 | local_irq_restore(flags); |
6faa6833 | 2354 | return freelist; |
894b8788 CL |
2355 | } |
2356 | ||
2357 | /* | |
2358 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2359 | * have the fastpath folded into their functions. So no function call | |
2360 | * overhead for requests that can be satisfied on the fastpath. | |
2361 | * | |
2362 | * The fastpath works by first checking if the lockless freelist can be used. | |
2363 | * If not then __slab_alloc is called for slow processing. | |
2364 | * | |
2365 | * Otherwise we can simply pick the next object from the lockless free list. | |
2366 | */ | |
2b847c3c | 2367 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2368 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2369 | { |
894b8788 | 2370 | void **object; |
dfb4f096 | 2371 | struct kmem_cache_cpu *c; |
57d437d2 | 2372 | struct page *page; |
8a5ec0ba | 2373 | unsigned long tid; |
1f84260c | 2374 | |
c016b0bd | 2375 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2376 | return NULL; |
1f84260c | 2377 | |
d79923fa | 2378 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2379 | redo: |
8a5ec0ba CL |
2380 | /* |
2381 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2382 | * enabled. We may switch back and forth between cpus while | |
2383 | * reading from one cpu area. That does not matter as long | |
2384 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2385 | * |
2386 | * Preemption is disabled for the retrieval of the tid because that | |
2387 | * must occur from the current processor. We cannot allow rescheduling | |
2388 | * on a different processor between the determination of the pointer | |
2389 | * and the retrieval of the tid. | |
8a5ec0ba | 2390 | */ |
7cccd80b | 2391 | preempt_disable(); |
9dfc6e68 | 2392 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2393 | |
8a5ec0ba CL |
2394 | /* |
2395 | * The transaction ids are globally unique per cpu and per operation on | |
2396 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2397 | * occurs on the right processor and that there was no operation on the | |
2398 | * linked list in between. | |
2399 | */ | |
2400 | tid = c->tid; | |
7cccd80b | 2401 | preempt_enable(); |
8a5ec0ba | 2402 | |
9dfc6e68 | 2403 | object = c->freelist; |
57d437d2 | 2404 | page = c->page; |
5091b74a | 2405 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2406 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2407 | |
2408 | else { | |
0ad9500e ED |
2409 | void *next_object = get_freepointer_safe(s, object); |
2410 | ||
8a5ec0ba | 2411 | /* |
25985edc | 2412 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2413 | * operation and if we are on the right processor. |
2414 | * | |
2415 | * The cmpxchg does the following atomically (without lock semantics!) | |
2416 | * 1. Relocate first pointer to the current per cpu area. | |
2417 | * 2. Verify that tid and freelist have not been changed | |
2418 | * 3. If they were not changed replace tid and freelist | |
2419 | * | |
2420 | * Since this is without lock semantics the protection is only against | |
2421 | * code executing on this cpu *not* from access by other cpus. | |
2422 | */ | |
933393f5 | 2423 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2424 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2425 | object, tid, | |
0ad9500e | 2426 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2427 | |
2428 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2429 | goto redo; | |
2430 | } | |
0ad9500e | 2431 | prefetch_freepointer(s, next_object); |
84e554e6 | 2432 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2433 | } |
8a5ec0ba | 2434 | |
74e2134f | 2435 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2436 | memset(object, 0, s->object_size); |
d07dbea4 | 2437 | |
c016b0bd | 2438 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2439 | |
894b8788 | 2440 | return object; |
81819f0f CL |
2441 | } |
2442 | ||
2b847c3c EG |
2443 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2444 | gfp_t gfpflags, unsigned long addr) | |
2445 | { | |
2446 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2447 | } | |
2448 | ||
81819f0f CL |
2449 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2450 | { | |
2b847c3c | 2451 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2452 | |
3b0efdfa | 2453 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags); |
5b882be4 EGM |
2454 | |
2455 | return ret; | |
81819f0f CL |
2456 | } |
2457 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2458 | ||
0f24f128 | 2459 | #ifdef CONFIG_TRACING |
4a92379b RK |
2460 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2461 | { | |
2b847c3c | 2462 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2463 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2464 | return ret; | |
2465 | } | |
2466 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
2467 | ||
2468 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
5b882be4 | 2469 | { |
4a92379b RK |
2470 | void *ret = kmalloc_order(size, flags, order); |
2471 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
2472 | return ret; | |
5b882be4 | 2473 | } |
4a92379b | 2474 | EXPORT_SYMBOL(kmalloc_order_trace); |
5b882be4 EGM |
2475 | #endif |
2476 | ||
81819f0f CL |
2477 | #ifdef CONFIG_NUMA |
2478 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2479 | { | |
2b847c3c | 2480 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2481 | |
ca2b84cb | 2482 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2483 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2484 | |
2485 | return ret; | |
81819f0f CL |
2486 | } |
2487 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2488 | |
0f24f128 | 2489 | #ifdef CONFIG_TRACING |
4a92379b | 2490 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2491 | gfp_t gfpflags, |
4a92379b | 2492 | int node, size_t size) |
5b882be4 | 2493 | { |
2b847c3c | 2494 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2495 | |
2496 | trace_kmalloc_node(_RET_IP_, ret, | |
2497 | size, s->size, gfpflags, node); | |
2498 | return ret; | |
5b882be4 | 2499 | } |
4a92379b | 2500 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2501 | #endif |
5d1f57e4 | 2502 | #endif |
5b882be4 | 2503 | |
81819f0f | 2504 | /* |
894b8788 CL |
2505 | * Slow patch handling. This may still be called frequently since objects |
2506 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2507 | * |
894b8788 CL |
2508 | * So we still attempt to reduce cache line usage. Just take the slab |
2509 | * lock and free the item. If there is no additional partial page | |
2510 | * handling required then we can return immediately. | |
81819f0f | 2511 | */ |
894b8788 | 2512 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2513 | void *x, unsigned long addr) |
81819f0f CL |
2514 | { |
2515 | void *prior; | |
2516 | void **object = (void *)x; | |
2cfb7455 | 2517 | int was_frozen; |
2cfb7455 CL |
2518 | struct page new; |
2519 | unsigned long counters; | |
2520 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2521 | unsigned long uninitialized_var(flags); |
81819f0f | 2522 | |
8a5ec0ba | 2523 | stat(s, FREE_SLOWPATH); |
81819f0f | 2524 | |
19c7ff9e CL |
2525 | if (kmem_cache_debug(s) && |
2526 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2527 | return; |
6446faa2 | 2528 | |
2cfb7455 | 2529 | do { |
837d678d JK |
2530 | if (unlikely(n)) { |
2531 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2532 | n = NULL; | |
2533 | } | |
2cfb7455 CL |
2534 | prior = page->freelist; |
2535 | counters = page->counters; | |
2536 | set_freepointer(s, object, prior); | |
2537 | new.counters = counters; | |
2538 | was_frozen = new.frozen; | |
2539 | new.inuse--; | |
837d678d | 2540 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 CL |
2541 | |
2542 | if (!kmem_cache_debug(s) && !prior) | |
2543 | ||
2544 | /* | |
2545 | * Slab was on no list before and will be partially empty | |
2546 | * We can defer the list move and instead freeze it. | |
2547 | */ | |
2548 | new.frozen = 1; | |
2549 | ||
2550 | else { /* Needs to be taken off a list */ | |
2551 | ||
2552 | n = get_node(s, page_to_nid(page)); | |
2553 | /* | |
2554 | * Speculatively acquire the list_lock. | |
2555 | * If the cmpxchg does not succeed then we may | |
2556 | * drop the list_lock without any processing. | |
2557 | * | |
2558 | * Otherwise the list_lock will synchronize with | |
2559 | * other processors updating the list of slabs. | |
2560 | */ | |
2561 | spin_lock_irqsave(&n->list_lock, flags); | |
2562 | ||
2563 | } | |
2cfb7455 | 2564 | } |
81819f0f | 2565 | |
2cfb7455 CL |
2566 | } while (!cmpxchg_double_slab(s, page, |
2567 | prior, counters, | |
2568 | object, new.counters, | |
2569 | "__slab_free")); | |
81819f0f | 2570 | |
2cfb7455 | 2571 | if (likely(!n)) { |
49e22585 CL |
2572 | |
2573 | /* | |
2574 | * If we just froze the page then put it onto the | |
2575 | * per cpu partial list. | |
2576 | */ | |
8028dcea | 2577 | if (new.frozen && !was_frozen) { |
49e22585 | 2578 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2579 | stat(s, CPU_PARTIAL_FREE); |
2580 | } | |
49e22585 | 2581 | /* |
2cfb7455 CL |
2582 | * The list lock was not taken therefore no list |
2583 | * activity can be necessary. | |
2584 | */ | |
2585 | if (was_frozen) | |
2586 | stat(s, FREE_FROZEN); | |
80f08c19 | 2587 | return; |
2cfb7455 | 2588 | } |
81819f0f | 2589 | |
837d678d JK |
2590 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) |
2591 | goto slab_empty; | |
2592 | ||
81819f0f | 2593 | /* |
837d678d JK |
2594 | * Objects left in the slab. If it was not on the partial list before |
2595 | * then add it. | |
81819f0f | 2596 | */ |
837d678d JK |
2597 | if (kmem_cache_debug(s) && unlikely(!prior)) { |
2598 | remove_full(s, page); | |
2599 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2600 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2601 | } |
80f08c19 | 2602 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2603 | return; |
2604 | ||
2605 | slab_empty: | |
a973e9dd | 2606 | if (prior) { |
81819f0f | 2607 | /* |
6fbabb20 | 2608 | * Slab on the partial list. |
81819f0f | 2609 | */ |
5cc6eee8 | 2610 | remove_partial(n, page); |
84e554e6 | 2611 | stat(s, FREE_REMOVE_PARTIAL); |
6fbabb20 CL |
2612 | } else |
2613 | /* Slab must be on the full list */ | |
2614 | remove_full(s, page); | |
2cfb7455 | 2615 | |
80f08c19 | 2616 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2617 | stat(s, FREE_SLAB); |
81819f0f | 2618 | discard_slab(s, page); |
81819f0f CL |
2619 | } |
2620 | ||
894b8788 CL |
2621 | /* |
2622 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2623 | * can perform fastpath freeing without additional function calls. | |
2624 | * | |
2625 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2626 | * of this processor. This typically the case if we have just allocated | |
2627 | * the item before. | |
2628 | * | |
2629 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2630 | * with all sorts of special processing. | |
2631 | */ | |
06428780 | 2632 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2633 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2634 | { |
2635 | void **object = (void *)x; | |
dfb4f096 | 2636 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2637 | unsigned long tid; |
1f84260c | 2638 | |
c016b0bd CL |
2639 | slab_free_hook(s, x); |
2640 | ||
8a5ec0ba CL |
2641 | redo: |
2642 | /* | |
2643 | * Determine the currently cpus per cpu slab. | |
2644 | * The cpu may change afterward. However that does not matter since | |
2645 | * data is retrieved via this pointer. If we are on the same cpu | |
2646 | * during the cmpxchg then the free will succedd. | |
2647 | */ | |
7cccd80b | 2648 | preempt_disable(); |
9dfc6e68 | 2649 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2650 | |
8a5ec0ba | 2651 | tid = c->tid; |
7cccd80b | 2652 | preempt_enable(); |
c016b0bd | 2653 | |
442b06bc | 2654 | if (likely(page == c->page)) { |
ff12059e | 2655 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2656 | |
933393f5 | 2657 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2658 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2659 | c->freelist, tid, | |
2660 | object, next_tid(tid)))) { | |
2661 | ||
2662 | note_cmpxchg_failure("slab_free", s, tid); | |
2663 | goto redo; | |
2664 | } | |
84e554e6 | 2665 | stat(s, FREE_FASTPATH); |
894b8788 | 2666 | } else |
ff12059e | 2667 | __slab_free(s, page, x, addr); |
894b8788 | 2668 | |
894b8788 CL |
2669 | } |
2670 | ||
81819f0f CL |
2671 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2672 | { | |
b9ce5ef4 GC |
2673 | s = cache_from_obj(s, x); |
2674 | if (!s) | |
79576102 | 2675 | return; |
b9ce5ef4 | 2676 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2677 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2678 | } |
2679 | EXPORT_SYMBOL(kmem_cache_free); | |
2680 | ||
81819f0f | 2681 | /* |
672bba3a CL |
2682 | * Object placement in a slab is made very easy because we always start at |
2683 | * offset 0. If we tune the size of the object to the alignment then we can | |
2684 | * get the required alignment by putting one properly sized object after | |
2685 | * another. | |
81819f0f CL |
2686 | * |
2687 | * Notice that the allocation order determines the sizes of the per cpu | |
2688 | * caches. Each processor has always one slab available for allocations. | |
2689 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2690 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2691 | * locking overhead. |
81819f0f CL |
2692 | */ |
2693 | ||
2694 | /* | |
2695 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2696 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2697 | * and increases the number of allocations possible without having to | |
2698 | * take the list_lock. | |
2699 | */ | |
2700 | static int slub_min_order; | |
114e9e89 | 2701 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2702 | static int slub_min_objects; |
81819f0f CL |
2703 | |
2704 | /* | |
2705 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2706 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2707 | */ |
2708 | static int slub_nomerge; | |
2709 | ||
81819f0f CL |
2710 | /* |
2711 | * Calculate the order of allocation given an slab object size. | |
2712 | * | |
672bba3a CL |
2713 | * The order of allocation has significant impact on performance and other |
2714 | * system components. Generally order 0 allocations should be preferred since | |
2715 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2716 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2717 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2718 | * would be wasted. |
2719 | * | |
2720 | * In order to reach satisfactory performance we must ensure that a minimum | |
2721 | * number of objects is in one slab. Otherwise we may generate too much | |
2722 | * activity on the partial lists which requires taking the list_lock. This is | |
2723 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2724 | * |
672bba3a CL |
2725 | * slub_max_order specifies the order where we begin to stop considering the |
2726 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2727 | * we try to keep the page order as low as possible. So we accept more waste | |
2728 | * of space in favor of a small page order. | |
81819f0f | 2729 | * |
672bba3a CL |
2730 | * Higher order allocations also allow the placement of more objects in a |
2731 | * slab and thereby reduce object handling overhead. If the user has | |
2732 | * requested a higher mininum order then we start with that one instead of | |
2733 | * the smallest order which will fit the object. | |
81819f0f | 2734 | */ |
5e6d444e | 2735 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2736 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2737 | { |
2738 | int order; | |
2739 | int rem; | |
6300ea75 | 2740 | int min_order = slub_min_order; |
81819f0f | 2741 | |
ab9a0f19 | 2742 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2743 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2744 | |
6300ea75 | 2745 | for (order = max(min_order, |
5e6d444e CL |
2746 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2747 | order <= max_order; order++) { | |
81819f0f | 2748 | |
5e6d444e | 2749 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2750 | |
ab9a0f19 | 2751 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2752 | continue; |
2753 | ||
ab9a0f19 | 2754 | rem = (slab_size - reserved) % size; |
81819f0f | 2755 | |
5e6d444e | 2756 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2757 | break; |
2758 | ||
2759 | } | |
672bba3a | 2760 | |
81819f0f CL |
2761 | return order; |
2762 | } | |
2763 | ||
ab9a0f19 | 2764 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2765 | { |
2766 | int order; | |
2767 | int min_objects; | |
2768 | int fraction; | |
e8120ff1 | 2769 | int max_objects; |
5e6d444e CL |
2770 | |
2771 | /* | |
2772 | * Attempt to find best configuration for a slab. This | |
2773 | * works by first attempting to generate a layout with | |
2774 | * the best configuration and backing off gradually. | |
2775 | * | |
2776 | * First we reduce the acceptable waste in a slab. Then | |
2777 | * we reduce the minimum objects required in a slab. | |
2778 | */ | |
2779 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2780 | if (!min_objects) |
2781 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2782 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2783 | min_objects = min(min_objects, max_objects); |
2784 | ||
5e6d444e | 2785 | while (min_objects > 1) { |
c124f5b5 | 2786 | fraction = 16; |
5e6d444e CL |
2787 | while (fraction >= 4) { |
2788 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2789 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2790 | if (order <= slub_max_order) |
2791 | return order; | |
2792 | fraction /= 2; | |
2793 | } | |
5086c389 | 2794 | min_objects--; |
5e6d444e CL |
2795 | } |
2796 | ||
2797 | /* | |
2798 | * We were unable to place multiple objects in a slab. Now | |
2799 | * lets see if we can place a single object there. | |
2800 | */ | |
ab9a0f19 | 2801 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2802 | if (order <= slub_max_order) |
2803 | return order; | |
2804 | ||
2805 | /* | |
2806 | * Doh this slab cannot be placed using slub_max_order. | |
2807 | */ | |
ab9a0f19 | 2808 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2809 | if (order < MAX_ORDER) |
5e6d444e CL |
2810 | return order; |
2811 | return -ENOSYS; | |
2812 | } | |
2813 | ||
5595cffc | 2814 | static void |
4053497d | 2815 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2816 | { |
2817 | n->nr_partial = 0; | |
81819f0f CL |
2818 | spin_lock_init(&n->list_lock); |
2819 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2820 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2821 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2822 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2823 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2824 | #endif |
81819f0f CL |
2825 | } |
2826 | ||
55136592 | 2827 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2828 | { |
6c182dc0 | 2829 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2830 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2831 | |
8a5ec0ba | 2832 | /* |
d4d84fef CM |
2833 | * Must align to double word boundary for the double cmpxchg |
2834 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2835 | */ |
d4d84fef CM |
2836 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2837 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2838 | |
2839 | if (!s->cpu_slab) | |
2840 | return 0; | |
2841 | ||
2842 | init_kmem_cache_cpus(s); | |
4c93c355 | 2843 | |
8a5ec0ba | 2844 | return 1; |
4c93c355 | 2845 | } |
4c93c355 | 2846 | |
51df1142 CL |
2847 | static struct kmem_cache *kmem_cache_node; |
2848 | ||
81819f0f CL |
2849 | /* |
2850 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2851 | * slab on the node for this slabcache. There are no concurrent accesses | |
2852 | * possible. | |
2853 | * | |
2854 | * Note that this function only works on the kmalloc_node_cache | |
4c93c355 CL |
2855 | * when allocating for the kmalloc_node_cache. This is used for bootstrapping |
2856 | * memory on a fresh node that has no slab structures yet. | |
81819f0f | 2857 | */ |
55136592 | 2858 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2859 | { |
2860 | struct page *page; | |
2861 | struct kmem_cache_node *n; | |
2862 | ||
51df1142 | 2863 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2864 | |
51df1142 | 2865 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2866 | |
2867 | BUG_ON(!page); | |
a2f92ee7 CL |
2868 | if (page_to_nid(page) != node) { |
2869 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2870 | "node %d\n", node); | |
2871 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2872 | "in order to be able to continue\n"); | |
2873 | } | |
2874 | ||
81819f0f CL |
2875 | n = page->freelist; |
2876 | BUG_ON(!n); | |
51df1142 | 2877 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2878 | page->inuse = 1; |
8cb0a506 | 2879 | page->frozen = 0; |
51df1142 | 2880 | kmem_cache_node->node[node] = n; |
8ab1372f | 2881 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2882 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2883 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2884 | #endif |
4053497d | 2885 | init_kmem_cache_node(n); |
51df1142 | 2886 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2887 | |
136333d1 | 2888 | add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2889 | } |
2890 | ||
2891 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2892 | { | |
2893 | int node; | |
2894 | ||
f64dc58c | 2895 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2896 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2897 | |
73367bd8 | 2898 | if (n) |
51df1142 CL |
2899 | kmem_cache_free(kmem_cache_node, n); |
2900 | ||
81819f0f CL |
2901 | s->node[node] = NULL; |
2902 | } | |
2903 | } | |
2904 | ||
55136592 | 2905 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2906 | { |
2907 | int node; | |
81819f0f | 2908 | |
f64dc58c | 2909 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2910 | struct kmem_cache_node *n; |
2911 | ||
73367bd8 | 2912 | if (slab_state == DOWN) { |
55136592 | 2913 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2914 | continue; |
2915 | } | |
51df1142 | 2916 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2917 | GFP_KERNEL, node); |
81819f0f | 2918 | |
73367bd8 AD |
2919 | if (!n) { |
2920 | free_kmem_cache_nodes(s); | |
2921 | return 0; | |
81819f0f | 2922 | } |
73367bd8 | 2923 | |
81819f0f | 2924 | s->node[node] = n; |
4053497d | 2925 | init_kmem_cache_node(n); |
81819f0f CL |
2926 | } |
2927 | return 1; | |
2928 | } | |
81819f0f | 2929 | |
c0bdb232 | 2930 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2931 | { |
2932 | if (min < MIN_PARTIAL) | |
2933 | min = MIN_PARTIAL; | |
2934 | else if (min > MAX_PARTIAL) | |
2935 | min = MAX_PARTIAL; | |
2936 | s->min_partial = min; | |
2937 | } | |
2938 | ||
81819f0f CL |
2939 | /* |
2940 | * calculate_sizes() determines the order and the distribution of data within | |
2941 | * a slab object. | |
2942 | */ | |
06b285dc | 2943 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2944 | { |
2945 | unsigned long flags = s->flags; | |
3b0efdfa | 2946 | unsigned long size = s->object_size; |
834f3d11 | 2947 | int order; |
81819f0f | 2948 | |
d8b42bf5 CL |
2949 | /* |
2950 | * Round up object size to the next word boundary. We can only | |
2951 | * place the free pointer at word boundaries and this determines | |
2952 | * the possible location of the free pointer. | |
2953 | */ | |
2954 | size = ALIGN(size, sizeof(void *)); | |
2955 | ||
2956 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2957 | /* |
2958 | * Determine if we can poison the object itself. If the user of | |
2959 | * the slab may touch the object after free or before allocation | |
2960 | * then we should never poison the object itself. | |
2961 | */ | |
2962 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2963 | !s->ctor) |
81819f0f CL |
2964 | s->flags |= __OBJECT_POISON; |
2965 | else | |
2966 | s->flags &= ~__OBJECT_POISON; | |
2967 | ||
81819f0f CL |
2968 | |
2969 | /* | |
672bba3a | 2970 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2971 | * end of the object and the free pointer. If not then add an |
672bba3a | 2972 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2973 | */ |
3b0efdfa | 2974 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2975 | size += sizeof(void *); |
41ecc55b | 2976 | #endif |
81819f0f CL |
2977 | |
2978 | /* | |
672bba3a CL |
2979 | * With that we have determined the number of bytes in actual use |
2980 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2981 | */ |
2982 | s->inuse = size; | |
2983 | ||
2984 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 2985 | s->ctor)) { |
81819f0f CL |
2986 | /* |
2987 | * Relocate free pointer after the object if it is not | |
2988 | * permitted to overwrite the first word of the object on | |
2989 | * kmem_cache_free. | |
2990 | * | |
2991 | * This is the case if we do RCU, have a constructor or | |
2992 | * destructor or are poisoning the objects. | |
2993 | */ | |
2994 | s->offset = size; | |
2995 | size += sizeof(void *); | |
2996 | } | |
2997 | ||
c12b3c62 | 2998 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
2999 | if (flags & SLAB_STORE_USER) |
3000 | /* | |
3001 | * Need to store information about allocs and frees after | |
3002 | * the object. | |
3003 | */ | |
3004 | size += 2 * sizeof(struct track); | |
3005 | ||
be7b3fbc | 3006 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3007 | /* |
3008 | * Add some empty padding so that we can catch | |
3009 | * overwrites from earlier objects rather than let | |
3010 | * tracking information or the free pointer be | |
0211a9c8 | 3011 | * corrupted if a user writes before the start |
81819f0f CL |
3012 | * of the object. |
3013 | */ | |
3014 | size += sizeof(void *); | |
41ecc55b | 3015 | #endif |
672bba3a | 3016 | |
81819f0f CL |
3017 | /* |
3018 | * SLUB stores one object immediately after another beginning from | |
3019 | * offset 0. In order to align the objects we have to simply size | |
3020 | * each object to conform to the alignment. | |
3021 | */ | |
45906855 | 3022 | size = ALIGN(size, s->align); |
81819f0f | 3023 | s->size = size; |
06b285dc CL |
3024 | if (forced_order >= 0) |
3025 | order = forced_order; | |
3026 | else | |
ab9a0f19 | 3027 | order = calculate_order(size, s->reserved); |
81819f0f | 3028 | |
834f3d11 | 3029 | if (order < 0) |
81819f0f CL |
3030 | return 0; |
3031 | ||
b7a49f0d | 3032 | s->allocflags = 0; |
834f3d11 | 3033 | if (order) |
b7a49f0d CL |
3034 | s->allocflags |= __GFP_COMP; |
3035 | ||
3036 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3037 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3038 | |
3039 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3040 | s->allocflags |= __GFP_RECLAIMABLE; | |
3041 | ||
81819f0f CL |
3042 | /* |
3043 | * Determine the number of objects per slab | |
3044 | */ | |
ab9a0f19 LJ |
3045 | s->oo = oo_make(order, size, s->reserved); |
3046 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3047 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3048 | s->max = s->oo; | |
81819f0f | 3049 | |
834f3d11 | 3050 | return !!oo_objects(s->oo); |
81819f0f CL |
3051 | } |
3052 | ||
8a13a4cc | 3053 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3054 | { |
8a13a4cc | 3055 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3056 | s->reserved = 0; |
81819f0f | 3057 | |
da9a638c LJ |
3058 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3059 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3060 | |
06b285dc | 3061 | if (!calculate_sizes(s, -1)) |
81819f0f | 3062 | goto error; |
3de47213 DR |
3063 | if (disable_higher_order_debug) { |
3064 | /* | |
3065 | * Disable debugging flags that store metadata if the min slab | |
3066 | * order increased. | |
3067 | */ | |
3b0efdfa | 3068 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3069 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3070 | s->offset = 0; | |
3071 | if (!calculate_sizes(s, -1)) | |
3072 | goto error; | |
3073 | } | |
3074 | } | |
81819f0f | 3075 | |
2565409f HC |
3076 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3077 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3078 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3079 | /* Enable fast mode */ | |
3080 | s->flags |= __CMPXCHG_DOUBLE; | |
3081 | #endif | |
3082 | ||
3b89d7d8 DR |
3083 | /* |
3084 | * The larger the object size is, the more pages we want on the partial | |
3085 | * list to avoid pounding the page allocator excessively. | |
3086 | */ | |
49e22585 CL |
3087 | set_min_partial(s, ilog2(s->size) / 2); |
3088 | ||
3089 | /* | |
3090 | * cpu_partial determined the maximum number of objects kept in the | |
3091 | * per cpu partial lists of a processor. | |
3092 | * | |
3093 | * Per cpu partial lists mainly contain slabs that just have one | |
3094 | * object freed. If they are used for allocation then they can be | |
3095 | * filled up again with minimal effort. The slab will never hit the | |
3096 | * per node partial lists and therefore no locking will be required. | |
3097 | * | |
3098 | * This setting also determines | |
3099 | * | |
3100 | * A) The number of objects from per cpu partial slabs dumped to the | |
3101 | * per node list when we reach the limit. | |
9f264904 | 3102 | * B) The number of objects in cpu partial slabs to extract from the |
49e22585 CL |
3103 | * per node list when we run out of per cpu objects. We only fetch 50% |
3104 | * to keep some capacity around for frees. | |
3105 | */ | |
8f1e33da CL |
3106 | if (kmem_cache_debug(s)) |
3107 | s->cpu_partial = 0; | |
3108 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3109 | s->cpu_partial = 2; |
3110 | else if (s->size >= 1024) | |
3111 | s->cpu_partial = 6; | |
3112 | else if (s->size >= 256) | |
3113 | s->cpu_partial = 13; | |
3114 | else | |
3115 | s->cpu_partial = 30; | |
3116 | ||
81819f0f | 3117 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3118 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3119 | #endif |
55136592 | 3120 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3121 | goto error; |
81819f0f | 3122 | |
55136592 | 3123 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3124 | return 0; |
ff12059e | 3125 | |
4c93c355 | 3126 | free_kmem_cache_nodes(s); |
81819f0f CL |
3127 | error: |
3128 | if (flags & SLAB_PANIC) | |
3129 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3130 | "order=%u offset=%u flags=%lx\n", | |
8a13a4cc | 3131 | s->name, (unsigned long)s->size, s->size, oo_order(s->oo), |
81819f0f | 3132 | s->offset, flags); |
278b1bb1 | 3133 | return -EINVAL; |
81819f0f | 3134 | } |
81819f0f | 3135 | |
33b12c38 CL |
3136 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3137 | const char *text) | |
3138 | { | |
3139 | #ifdef CONFIG_SLUB_DEBUG | |
3140 | void *addr = page_address(page); | |
3141 | void *p; | |
a5dd5c11 NK |
3142 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3143 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3144 | if (!map) |
3145 | return; | |
945cf2b6 | 3146 | slab_err(s, page, text, s->name); |
33b12c38 | 3147 | slab_lock(page); |
33b12c38 | 3148 | |
5f80b13a | 3149 | get_map(s, page, map); |
33b12c38 CL |
3150 | for_each_object(p, s, addr, page->objects) { |
3151 | ||
3152 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3153 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3154 | p, p - addr); | |
3155 | print_tracking(s, p); | |
3156 | } | |
3157 | } | |
3158 | slab_unlock(page); | |
bbd7d57b | 3159 | kfree(map); |
33b12c38 CL |
3160 | #endif |
3161 | } | |
3162 | ||
81819f0f | 3163 | /* |
599870b1 | 3164 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3165 | * This is called from kmem_cache_close(). We must be the last thread |
3166 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3167 | */ |
599870b1 | 3168 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3169 | { |
81819f0f CL |
3170 | struct page *page, *h; |
3171 | ||
33b12c38 | 3172 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3173 | if (!page->inuse) { |
5cc6eee8 | 3174 | remove_partial(n, page); |
81819f0f | 3175 | discard_slab(s, page); |
33b12c38 CL |
3176 | } else { |
3177 | list_slab_objects(s, page, | |
945cf2b6 | 3178 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3179 | } |
33b12c38 | 3180 | } |
81819f0f CL |
3181 | } |
3182 | ||
3183 | /* | |
672bba3a | 3184 | * Release all resources used by a slab cache. |
81819f0f | 3185 | */ |
0c710013 | 3186 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3187 | { |
3188 | int node; | |
3189 | ||
3190 | flush_all(s); | |
81819f0f | 3191 | /* Attempt to free all objects */ |
f64dc58c | 3192 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3193 | struct kmem_cache_node *n = get_node(s, node); |
3194 | ||
599870b1 CL |
3195 | free_partial(s, n); |
3196 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3197 | return 1; |
3198 | } | |
945cf2b6 | 3199 | free_percpu(s->cpu_slab); |
81819f0f CL |
3200 | free_kmem_cache_nodes(s); |
3201 | return 0; | |
3202 | } | |
3203 | ||
945cf2b6 | 3204 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3205 | { |
12c3667f | 3206 | int rc = kmem_cache_close(s); |
945cf2b6 | 3207 | |
5413dfba GC |
3208 | if (!rc) { |
3209 | /* | |
3210 | * We do the same lock strategy around sysfs_slab_add, see | |
3211 | * __kmem_cache_create. Because this is pretty much the last | |
3212 | * operation we do and the lock will be released shortly after | |
3213 | * that in slab_common.c, we could just move sysfs_slab_remove | |
3214 | * to a later point in common code. We should do that when we | |
3215 | * have a common sysfs framework for all allocators. | |
3216 | */ | |
3217 | mutex_unlock(&slab_mutex); | |
81819f0f | 3218 | sysfs_slab_remove(s); |
5413dfba GC |
3219 | mutex_lock(&slab_mutex); |
3220 | } | |
12c3667f CL |
3221 | |
3222 | return rc; | |
81819f0f | 3223 | } |
81819f0f CL |
3224 | |
3225 | /******************************************************************** | |
3226 | * Kmalloc subsystem | |
3227 | *******************************************************************/ | |
3228 | ||
81819f0f CL |
3229 | static int __init setup_slub_min_order(char *str) |
3230 | { | |
06428780 | 3231 | get_option(&str, &slub_min_order); |
81819f0f CL |
3232 | |
3233 | return 1; | |
3234 | } | |
3235 | ||
3236 | __setup("slub_min_order=", setup_slub_min_order); | |
3237 | ||
3238 | static int __init setup_slub_max_order(char *str) | |
3239 | { | |
06428780 | 3240 | get_option(&str, &slub_max_order); |
818cf590 | 3241 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3242 | |
3243 | return 1; | |
3244 | } | |
3245 | ||
3246 | __setup("slub_max_order=", setup_slub_max_order); | |
3247 | ||
3248 | static int __init setup_slub_min_objects(char *str) | |
3249 | { | |
06428780 | 3250 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3251 | |
3252 | return 1; | |
3253 | } | |
3254 | ||
3255 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3256 | ||
3257 | static int __init setup_slub_nomerge(char *str) | |
3258 | { | |
3259 | slub_nomerge = 1; | |
3260 | return 1; | |
3261 | } | |
3262 | ||
3263 | __setup("slub_nomerge", setup_slub_nomerge); | |
3264 | ||
81819f0f CL |
3265 | void *__kmalloc(size_t size, gfp_t flags) |
3266 | { | |
aadb4bc4 | 3267 | struct kmem_cache *s; |
5b882be4 | 3268 | void *ret; |
81819f0f | 3269 | |
95a05b42 | 3270 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3271 | return kmalloc_large(size, flags); |
aadb4bc4 | 3272 | |
2c59dd65 | 3273 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3274 | |
3275 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3276 | return s; |
3277 | ||
2b847c3c | 3278 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3279 | |
ca2b84cb | 3280 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3281 | |
3282 | return ret; | |
81819f0f CL |
3283 | } |
3284 | EXPORT_SYMBOL(__kmalloc); | |
3285 | ||
5d1f57e4 | 3286 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3287 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3288 | { | |
b1eeab67 | 3289 | struct page *page; |
e4f7c0b4 | 3290 | void *ptr = NULL; |
f619cfe1 | 3291 | |
d79923fa | 3292 | flags |= __GFP_COMP | __GFP_NOTRACK | __GFP_KMEMCG; |
b1eeab67 | 3293 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3294 | if (page) |
e4f7c0b4 CM |
3295 | ptr = page_address(page); |
3296 | ||
3297 | kmemleak_alloc(ptr, size, 1, flags); | |
3298 | return ptr; | |
f619cfe1 CL |
3299 | } |
3300 | ||
81819f0f CL |
3301 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3302 | { | |
aadb4bc4 | 3303 | struct kmem_cache *s; |
5b882be4 | 3304 | void *ret; |
81819f0f | 3305 | |
95a05b42 | 3306 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3307 | ret = kmalloc_large_node(size, flags, node); |
3308 | ||
ca2b84cb EGM |
3309 | trace_kmalloc_node(_RET_IP_, ret, |
3310 | size, PAGE_SIZE << get_order(size), | |
3311 | flags, node); | |
5b882be4 EGM |
3312 | |
3313 | return ret; | |
3314 | } | |
aadb4bc4 | 3315 | |
2c59dd65 | 3316 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3317 | |
3318 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3319 | return s; |
3320 | ||
2b847c3c | 3321 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3322 | |
ca2b84cb | 3323 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3324 | |
3325 | return ret; | |
81819f0f CL |
3326 | } |
3327 | EXPORT_SYMBOL(__kmalloc_node); | |
3328 | #endif | |
3329 | ||
3330 | size_t ksize(const void *object) | |
3331 | { | |
272c1d21 | 3332 | struct page *page; |
81819f0f | 3333 | |
ef8b4520 | 3334 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3335 | return 0; |
3336 | ||
294a80a8 | 3337 | page = virt_to_head_page(object); |
294a80a8 | 3338 | |
76994412 PE |
3339 | if (unlikely(!PageSlab(page))) { |
3340 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3341 | return PAGE_SIZE << compound_order(page); |
76994412 | 3342 | } |
81819f0f | 3343 | |
1b4f59e3 | 3344 | return slab_ksize(page->slab_cache); |
81819f0f | 3345 | } |
b1aabecd | 3346 | EXPORT_SYMBOL(ksize); |
81819f0f | 3347 | |
d18a90dd BG |
3348 | #ifdef CONFIG_SLUB_DEBUG |
3349 | bool verify_mem_not_deleted(const void *x) | |
3350 | { | |
3351 | struct page *page; | |
3352 | void *object = (void *)x; | |
3353 | unsigned long flags; | |
3354 | bool rv; | |
3355 | ||
3356 | if (unlikely(ZERO_OR_NULL_PTR(x))) | |
3357 | return false; | |
3358 | ||
3359 | local_irq_save(flags); | |
3360 | ||
3361 | page = virt_to_head_page(x); | |
3362 | if (unlikely(!PageSlab(page))) { | |
3363 | /* maybe it was from stack? */ | |
3364 | rv = true; | |
3365 | goto out_unlock; | |
3366 | } | |
3367 | ||
3368 | slab_lock(page); | |
1b4f59e3 GC |
3369 | if (on_freelist(page->slab_cache, page, object)) { |
3370 | object_err(page->slab_cache, page, object, "Object is on free-list"); | |
d18a90dd BG |
3371 | rv = false; |
3372 | } else { | |
3373 | rv = true; | |
3374 | } | |
3375 | slab_unlock(page); | |
3376 | ||
3377 | out_unlock: | |
3378 | local_irq_restore(flags); | |
3379 | return rv; | |
3380 | } | |
3381 | EXPORT_SYMBOL(verify_mem_not_deleted); | |
3382 | #endif | |
3383 | ||
81819f0f CL |
3384 | void kfree(const void *x) |
3385 | { | |
81819f0f | 3386 | struct page *page; |
5bb983b0 | 3387 | void *object = (void *)x; |
81819f0f | 3388 | |
2121db74 PE |
3389 | trace_kfree(_RET_IP_, x); |
3390 | ||
2408c550 | 3391 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3392 | return; |
3393 | ||
b49af68f | 3394 | page = virt_to_head_page(x); |
aadb4bc4 | 3395 | if (unlikely(!PageSlab(page))) { |
0937502a | 3396 | BUG_ON(!PageCompound(page)); |
e4f7c0b4 | 3397 | kmemleak_free(x); |
d79923fa | 3398 | __free_memcg_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3399 | return; |
3400 | } | |
1b4f59e3 | 3401 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3402 | } |
3403 | EXPORT_SYMBOL(kfree); | |
3404 | ||
2086d26a | 3405 | /* |
672bba3a CL |
3406 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3407 | * the remaining slabs by the number of items in use. The slabs with the | |
3408 | * most items in use come first. New allocations will then fill those up | |
3409 | * and thus they can be removed from the partial lists. | |
3410 | * | |
3411 | * The slabs with the least items are placed last. This results in them | |
3412 | * being allocated from last increasing the chance that the last objects | |
3413 | * are freed in them. | |
2086d26a CL |
3414 | */ |
3415 | int kmem_cache_shrink(struct kmem_cache *s) | |
3416 | { | |
3417 | int node; | |
3418 | int i; | |
3419 | struct kmem_cache_node *n; | |
3420 | struct page *page; | |
3421 | struct page *t; | |
205ab99d | 3422 | int objects = oo_objects(s->max); |
2086d26a | 3423 | struct list_head *slabs_by_inuse = |
834f3d11 | 3424 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3425 | unsigned long flags; |
3426 | ||
3427 | if (!slabs_by_inuse) | |
3428 | return -ENOMEM; | |
3429 | ||
3430 | flush_all(s); | |
f64dc58c | 3431 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3432 | n = get_node(s, node); |
3433 | ||
3434 | if (!n->nr_partial) | |
3435 | continue; | |
3436 | ||
834f3d11 | 3437 | for (i = 0; i < objects; i++) |
2086d26a CL |
3438 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3439 | ||
3440 | spin_lock_irqsave(&n->list_lock, flags); | |
3441 | ||
3442 | /* | |
672bba3a | 3443 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3444 | * |
672bba3a CL |
3445 | * Note that concurrent frees may occur while we hold the |
3446 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3447 | */ |
3448 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3449 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3450 | if (!page->inuse) | |
3451 | n->nr_partial--; | |
2086d26a CL |
3452 | } |
3453 | ||
2086d26a | 3454 | /* |
672bba3a CL |
3455 | * Rebuild the partial list with the slabs filled up most |
3456 | * first and the least used slabs at the end. | |
2086d26a | 3457 | */ |
69cb8e6b | 3458 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3459 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3460 | ||
2086d26a | 3461 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3462 | |
3463 | /* Release empty slabs */ | |
3464 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3465 | discard_slab(s, page); | |
2086d26a CL |
3466 | } |
3467 | ||
3468 | kfree(slabs_by_inuse); | |
3469 | return 0; | |
3470 | } | |
3471 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3472 | ||
b9049e23 YG |
3473 | static int slab_mem_going_offline_callback(void *arg) |
3474 | { | |
3475 | struct kmem_cache *s; | |
3476 | ||
18004c5d | 3477 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3478 | list_for_each_entry(s, &slab_caches, list) |
3479 | kmem_cache_shrink(s); | |
18004c5d | 3480 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3481 | |
3482 | return 0; | |
3483 | } | |
3484 | ||
3485 | static void slab_mem_offline_callback(void *arg) | |
3486 | { | |
3487 | struct kmem_cache_node *n; | |
3488 | struct kmem_cache *s; | |
3489 | struct memory_notify *marg = arg; | |
3490 | int offline_node; | |
3491 | ||
b9d5ab25 | 3492 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3493 | |
3494 | /* | |
3495 | * If the node still has available memory. we need kmem_cache_node | |
3496 | * for it yet. | |
3497 | */ | |
3498 | if (offline_node < 0) | |
3499 | return; | |
3500 | ||
18004c5d | 3501 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3502 | list_for_each_entry(s, &slab_caches, list) { |
3503 | n = get_node(s, offline_node); | |
3504 | if (n) { | |
3505 | /* | |
3506 | * if n->nr_slabs > 0, slabs still exist on the node | |
3507 | * that is going down. We were unable to free them, | |
c9404c9c | 3508 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3509 | * callback. So, we must fail. |
3510 | */ | |
0f389ec6 | 3511 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3512 | |
3513 | s->node[offline_node] = NULL; | |
8de66a0c | 3514 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3515 | } |
3516 | } | |
18004c5d | 3517 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3518 | } |
3519 | ||
3520 | static int slab_mem_going_online_callback(void *arg) | |
3521 | { | |
3522 | struct kmem_cache_node *n; | |
3523 | struct kmem_cache *s; | |
3524 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3525 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3526 | int ret = 0; |
3527 | ||
3528 | /* | |
3529 | * If the node's memory is already available, then kmem_cache_node is | |
3530 | * already created. Nothing to do. | |
3531 | */ | |
3532 | if (nid < 0) | |
3533 | return 0; | |
3534 | ||
3535 | /* | |
0121c619 | 3536 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3537 | * allocate a kmem_cache_node structure in order to bring the node |
3538 | * online. | |
3539 | */ | |
18004c5d | 3540 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3541 | list_for_each_entry(s, &slab_caches, list) { |
3542 | /* | |
3543 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3544 | * since memory is not yet available from the node that | |
3545 | * is brought up. | |
3546 | */ | |
8de66a0c | 3547 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3548 | if (!n) { |
3549 | ret = -ENOMEM; | |
3550 | goto out; | |
3551 | } | |
4053497d | 3552 | init_kmem_cache_node(n); |
b9049e23 YG |
3553 | s->node[nid] = n; |
3554 | } | |
3555 | out: | |
18004c5d | 3556 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3557 | return ret; |
3558 | } | |
3559 | ||
3560 | static int slab_memory_callback(struct notifier_block *self, | |
3561 | unsigned long action, void *arg) | |
3562 | { | |
3563 | int ret = 0; | |
3564 | ||
3565 | switch (action) { | |
3566 | case MEM_GOING_ONLINE: | |
3567 | ret = slab_mem_going_online_callback(arg); | |
3568 | break; | |
3569 | case MEM_GOING_OFFLINE: | |
3570 | ret = slab_mem_going_offline_callback(arg); | |
3571 | break; | |
3572 | case MEM_OFFLINE: | |
3573 | case MEM_CANCEL_ONLINE: | |
3574 | slab_mem_offline_callback(arg); | |
3575 | break; | |
3576 | case MEM_ONLINE: | |
3577 | case MEM_CANCEL_OFFLINE: | |
3578 | break; | |
3579 | } | |
dc19f9db KH |
3580 | if (ret) |
3581 | ret = notifier_from_errno(ret); | |
3582 | else | |
3583 | ret = NOTIFY_OK; | |
b9049e23 YG |
3584 | return ret; |
3585 | } | |
3586 | ||
3ac38faa AM |
3587 | static struct notifier_block slab_memory_callback_nb = { |
3588 | .notifier_call = slab_memory_callback, | |
3589 | .priority = SLAB_CALLBACK_PRI, | |
3590 | }; | |
b9049e23 | 3591 | |
81819f0f CL |
3592 | /******************************************************************** |
3593 | * Basic setup of slabs | |
3594 | *******************************************************************/ | |
3595 | ||
51df1142 CL |
3596 | /* |
3597 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3598 | * the page allocator. Allocate them properly then fix up the pointers |
3599 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3600 | */ |
3601 | ||
dffb4d60 | 3602 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3603 | { |
3604 | int node; | |
dffb4d60 | 3605 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
51df1142 | 3606 | |
dffb4d60 | 3607 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3608 | |
7d557b3c GC |
3609 | /* |
3610 | * This runs very early, and only the boot processor is supposed to be | |
3611 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3612 | * IPIs around. | |
3613 | */ | |
3614 | __flush_cpu_slab(s, smp_processor_id()); | |
51df1142 CL |
3615 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3616 | struct kmem_cache_node *n = get_node(s, node); | |
3617 | struct page *p; | |
3618 | ||
3619 | if (n) { | |
3620 | list_for_each_entry(p, &n->partial, lru) | |
1b4f59e3 | 3621 | p->slab_cache = s; |
51df1142 | 3622 | |
607bf324 | 3623 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 | 3624 | list_for_each_entry(p, &n->full, lru) |
1b4f59e3 | 3625 | p->slab_cache = s; |
51df1142 CL |
3626 | #endif |
3627 | } | |
3628 | } | |
dffb4d60 CL |
3629 | list_add(&s->list, &slab_caches); |
3630 | return s; | |
51df1142 CL |
3631 | } |
3632 | ||
81819f0f CL |
3633 | void __init kmem_cache_init(void) |
3634 | { | |
dffb4d60 CL |
3635 | static __initdata struct kmem_cache boot_kmem_cache, |
3636 | boot_kmem_cache_node; | |
51df1142 | 3637 | |
fc8d8620 SG |
3638 | if (debug_guardpage_minorder()) |
3639 | slub_max_order = 0; | |
3640 | ||
dffb4d60 CL |
3641 | kmem_cache_node = &boot_kmem_cache_node; |
3642 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3643 | |
dffb4d60 CL |
3644 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3645 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3646 | |
3ac38faa | 3647 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3648 | |
3649 | /* Able to allocate the per node structures */ | |
3650 | slab_state = PARTIAL; | |
3651 | ||
dffb4d60 CL |
3652 | create_boot_cache(kmem_cache, "kmem_cache", |
3653 | offsetof(struct kmem_cache, node) + | |
3654 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3655 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3656 | |
dffb4d60 | 3657 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3658 | |
51df1142 CL |
3659 | /* |
3660 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3661 | * kmem_cache_node is separately allocated so no need to | |
3662 | * update any list pointers. | |
3663 | */ | |
dffb4d60 | 3664 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3665 | |
3666 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3667 | create_kmalloc_caches(0); |
81819f0f CL |
3668 | |
3669 | #ifdef CONFIG_SMP | |
3670 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3671 | #endif |
81819f0f | 3672 | |
3adbefee | 3673 | printk(KERN_INFO |
f97d5f63 | 3674 | "SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d," |
4b356be0 | 3675 | " CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3676 | cache_line_size(), |
81819f0f CL |
3677 | slub_min_order, slub_max_order, slub_min_objects, |
3678 | nr_cpu_ids, nr_node_ids); | |
3679 | } | |
3680 | ||
7e85ee0c PE |
3681 | void __init kmem_cache_init_late(void) |
3682 | { | |
7e85ee0c PE |
3683 | } |
3684 | ||
81819f0f CL |
3685 | /* |
3686 | * Find a mergeable slab cache | |
3687 | */ | |
3688 | static int slab_unmergeable(struct kmem_cache *s) | |
3689 | { | |
3690 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3691 | return 1; | |
3692 | ||
c59def9f | 3693 | if (s->ctor) |
81819f0f CL |
3694 | return 1; |
3695 | ||
8ffa6875 CL |
3696 | /* |
3697 | * We may have set a slab to be unmergeable during bootstrap. | |
3698 | */ | |
3699 | if (s->refcount < 0) | |
3700 | return 1; | |
3701 | ||
81819f0f CL |
3702 | return 0; |
3703 | } | |
3704 | ||
2633d7a0 | 3705 | static struct kmem_cache *find_mergeable(struct mem_cgroup *memcg, size_t size, |
ba0268a8 | 3706 | size_t align, unsigned long flags, const char *name, |
51cc5068 | 3707 | void (*ctor)(void *)) |
81819f0f | 3708 | { |
5b95a4ac | 3709 | struct kmem_cache *s; |
81819f0f CL |
3710 | |
3711 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3712 | return NULL; | |
3713 | ||
c59def9f | 3714 | if (ctor) |
81819f0f CL |
3715 | return NULL; |
3716 | ||
3717 | size = ALIGN(size, sizeof(void *)); | |
3718 | align = calculate_alignment(flags, align, size); | |
3719 | size = ALIGN(size, align); | |
ba0268a8 | 3720 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3721 | |
5b95a4ac | 3722 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3723 | if (slab_unmergeable(s)) |
3724 | continue; | |
3725 | ||
3726 | if (size > s->size) | |
3727 | continue; | |
3728 | ||
ba0268a8 | 3729 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
81819f0f CL |
3730 | continue; |
3731 | /* | |
3732 | * Check if alignment is compatible. | |
3733 | * Courtesy of Adrian Drzewiecki | |
3734 | */ | |
06428780 | 3735 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3736 | continue; |
3737 | ||
3738 | if (s->size - size >= sizeof(void *)) | |
3739 | continue; | |
3740 | ||
2633d7a0 GC |
3741 | if (!cache_match_memcg(s, memcg)) |
3742 | continue; | |
3743 | ||
81819f0f CL |
3744 | return s; |
3745 | } | |
3746 | return NULL; | |
3747 | } | |
3748 | ||
2633d7a0 GC |
3749 | struct kmem_cache * |
3750 | __kmem_cache_alias(struct mem_cgroup *memcg, const char *name, size_t size, | |
3751 | size_t align, unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3752 | { |
3753 | struct kmem_cache *s; | |
3754 | ||
2633d7a0 | 3755 | s = find_mergeable(memcg, size, align, flags, name, ctor); |
81819f0f CL |
3756 | if (s) { |
3757 | s->refcount++; | |
3758 | /* | |
3759 | * Adjust the object sizes so that we clear | |
3760 | * the complete object on kzalloc. | |
3761 | */ | |
3b0efdfa | 3762 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3763 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3764 | |
7b8f3b66 | 3765 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3766 | s->refcount--; |
cbb79694 | 3767 | s = NULL; |
7b8f3b66 | 3768 | } |
a0e1d1be | 3769 | } |
6446faa2 | 3770 | |
cbb79694 CL |
3771 | return s; |
3772 | } | |
84c1cf62 | 3773 | |
8a13a4cc | 3774 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3775 | { |
aac3a166 PE |
3776 | int err; |
3777 | ||
3778 | err = kmem_cache_open(s, flags); | |
3779 | if (err) | |
3780 | return err; | |
20cea968 | 3781 | |
45530c44 CL |
3782 | /* Mutex is not taken during early boot */ |
3783 | if (slab_state <= UP) | |
3784 | return 0; | |
3785 | ||
107dab5c | 3786 | memcg_propagate_slab_attrs(s); |
aac3a166 PE |
3787 | mutex_unlock(&slab_mutex); |
3788 | err = sysfs_slab_add(s); | |
3789 | mutex_lock(&slab_mutex); | |
20cea968 | 3790 | |
aac3a166 PE |
3791 | if (err) |
3792 | kmem_cache_close(s); | |
20cea968 | 3793 | |
aac3a166 | 3794 | return err; |
81819f0f | 3795 | } |
81819f0f | 3796 | |
81819f0f | 3797 | #ifdef CONFIG_SMP |
81819f0f | 3798 | /* |
672bba3a CL |
3799 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3800 | * necessary. | |
81819f0f CL |
3801 | */ |
3802 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
3803 | unsigned long action, void *hcpu) | |
3804 | { | |
3805 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3806 | struct kmem_cache *s; |
3807 | unsigned long flags; | |
81819f0f CL |
3808 | |
3809 | switch (action) { | |
3810 | case CPU_UP_CANCELED: | |
8bb78442 | 3811 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3812 | case CPU_DEAD: |
8bb78442 | 3813 | case CPU_DEAD_FROZEN: |
18004c5d | 3814 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3815 | list_for_each_entry(s, &slab_caches, list) { |
3816 | local_irq_save(flags); | |
3817 | __flush_cpu_slab(s, cpu); | |
3818 | local_irq_restore(flags); | |
3819 | } | |
18004c5d | 3820 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3821 | break; |
3822 | default: | |
3823 | break; | |
3824 | } | |
3825 | return NOTIFY_OK; | |
3826 | } | |
3827 | ||
06428780 | 3828 | static struct notifier_block __cpuinitdata slab_notifier = { |
3adbefee | 3829 | .notifier_call = slab_cpuup_callback |
06428780 | 3830 | }; |
81819f0f CL |
3831 | |
3832 | #endif | |
3833 | ||
ce71e27c | 3834 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3835 | { |
aadb4bc4 | 3836 | struct kmem_cache *s; |
94b528d0 | 3837 | void *ret; |
aadb4bc4 | 3838 | |
95a05b42 | 3839 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3840 | return kmalloc_large(size, gfpflags); |
3841 | ||
2c59dd65 | 3842 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3843 | |
2408c550 | 3844 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3845 | return s; |
81819f0f | 3846 | |
2b847c3c | 3847 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3848 | |
25985edc | 3849 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3850 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3851 | |
3852 | return ret; | |
81819f0f CL |
3853 | } |
3854 | ||
5d1f57e4 | 3855 | #ifdef CONFIG_NUMA |
81819f0f | 3856 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3857 | int node, unsigned long caller) |
81819f0f | 3858 | { |
aadb4bc4 | 3859 | struct kmem_cache *s; |
94b528d0 | 3860 | void *ret; |
aadb4bc4 | 3861 | |
95a05b42 | 3862 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3863 | ret = kmalloc_large_node(size, gfpflags, node); |
3864 | ||
3865 | trace_kmalloc_node(caller, ret, | |
3866 | size, PAGE_SIZE << get_order(size), | |
3867 | gfpflags, node); | |
3868 | ||
3869 | return ret; | |
3870 | } | |
eada35ef | 3871 | |
2c59dd65 | 3872 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3873 | |
2408c550 | 3874 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3875 | return s; |
81819f0f | 3876 | |
2b847c3c | 3877 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3878 | |
25985edc | 3879 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3880 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3881 | |
3882 | return ret; | |
81819f0f | 3883 | } |
5d1f57e4 | 3884 | #endif |
81819f0f | 3885 | |
ab4d5ed5 | 3886 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3887 | static int count_inuse(struct page *page) |
3888 | { | |
3889 | return page->inuse; | |
3890 | } | |
3891 | ||
3892 | static int count_total(struct page *page) | |
3893 | { | |
3894 | return page->objects; | |
3895 | } | |
ab4d5ed5 | 3896 | #endif |
205ab99d | 3897 | |
ab4d5ed5 | 3898 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3899 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3900 | unsigned long *map) | |
53e15af0 CL |
3901 | { |
3902 | void *p; | |
a973e9dd | 3903 | void *addr = page_address(page); |
53e15af0 CL |
3904 | |
3905 | if (!check_slab(s, page) || | |
3906 | !on_freelist(s, page, NULL)) | |
3907 | return 0; | |
3908 | ||
3909 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3910 | bitmap_zero(map, page->objects); |
53e15af0 | 3911 | |
5f80b13a CL |
3912 | get_map(s, page, map); |
3913 | for_each_object(p, s, addr, page->objects) { | |
3914 | if (test_bit(slab_index(p, s, addr), map)) | |
3915 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3916 | return 0; | |
53e15af0 CL |
3917 | } |
3918 | ||
224a88be | 3919 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3920 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3921 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3922 | return 0; |
3923 | return 1; | |
3924 | } | |
3925 | ||
434e245d CL |
3926 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3927 | unsigned long *map) | |
53e15af0 | 3928 | { |
881db7fb CL |
3929 | slab_lock(page); |
3930 | validate_slab(s, page, map); | |
3931 | slab_unlock(page); | |
53e15af0 CL |
3932 | } |
3933 | ||
434e245d CL |
3934 | static int validate_slab_node(struct kmem_cache *s, |
3935 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3936 | { |
3937 | unsigned long count = 0; | |
3938 | struct page *page; | |
3939 | unsigned long flags; | |
3940 | ||
3941 | spin_lock_irqsave(&n->list_lock, flags); | |
3942 | ||
3943 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3944 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3945 | count++; |
3946 | } | |
3947 | if (count != n->nr_partial) | |
3948 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3949 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3950 | ||
3951 | if (!(s->flags & SLAB_STORE_USER)) | |
3952 | goto out; | |
3953 | ||
3954 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3955 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3956 | count++; |
3957 | } | |
3958 | if (count != atomic_long_read(&n->nr_slabs)) | |
3959 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3960 | "counter=%ld\n", s->name, count, | |
3961 | atomic_long_read(&n->nr_slabs)); | |
3962 | ||
3963 | out: | |
3964 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3965 | return count; | |
3966 | } | |
3967 | ||
434e245d | 3968 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3969 | { |
3970 | int node; | |
3971 | unsigned long count = 0; | |
205ab99d | 3972 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3973 | sizeof(unsigned long), GFP_KERNEL); |
3974 | ||
3975 | if (!map) | |
3976 | return -ENOMEM; | |
53e15af0 CL |
3977 | |
3978 | flush_all(s); | |
f64dc58c | 3979 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3980 | struct kmem_cache_node *n = get_node(s, node); |
3981 | ||
434e245d | 3982 | count += validate_slab_node(s, n, map); |
53e15af0 | 3983 | } |
434e245d | 3984 | kfree(map); |
53e15af0 CL |
3985 | return count; |
3986 | } | |
88a420e4 | 3987 | /* |
672bba3a | 3988 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3989 | * and freed. |
3990 | */ | |
3991 | ||
6fa3eb70 S |
3992 | #ifdef CONFIG_MTK_MEMCFG |
3993 | #define MTK_MEMCFG_SLABTRACE_CNT 4 | |
3994 | /* MTK_MEMCFG_SLABTRACE_CNT should be always <= TRACK_ADDRS_COUNT */ | |
3995 | #if (MTK_MEMCFG_SLABTRACE_CNT > TRACK_ADDRS_COUNT) | |
3996 | #error (MTK_MEMCFG_SLABTRACE_CNT > TRACK_ADDRS_COUNT) | |
3997 | #endif | |
3998 | #endif | |
3999 | ||
88a420e4 CL |
4000 | struct location { |
4001 | unsigned long count; | |
ce71e27c | 4002 | unsigned long addr; |
6fa3eb70 S |
4003 | #ifdef CONFIG_MTK_MEMCFG |
4004 | #ifdef CONFIG_STACKTRACE | |
4005 | unsigned long addrs[MTK_MEMCFG_SLABTRACE_CNT]; /* Called from address */ | |
4006 | #endif | |
4007 | #endif | |
45edfa58 CL |
4008 | long long sum_time; |
4009 | long min_time; | |
4010 | long max_time; | |
4011 | long min_pid; | |
4012 | long max_pid; | |
174596a0 | 4013 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4014 | nodemask_t nodes; |
88a420e4 CL |
4015 | }; |
4016 | ||
4017 | struct loc_track { | |
4018 | unsigned long max; | |
4019 | unsigned long count; | |
4020 | struct location *loc; | |
4021 | }; | |
4022 | ||
4023 | static void free_loc_track(struct loc_track *t) | |
4024 | { | |
4025 | if (t->max) | |
6fa3eb70 | 4026 | #ifndef CONFIG_MTK_PAGERECORDER |
88a420e4 CL |
4027 | free_pages((unsigned long)t->loc, |
4028 | get_order(sizeof(struct location) * t->max)); | |
6fa3eb70 S |
4029 | #else |
4030 | __free_pages_nopagedebug((struct page *)t->loc, | |
4031 | get_order(sizeof(struct location) * t->max)); | |
4032 | #endif | |
88a420e4 CL |
4033 | } |
4034 | ||
68dff6a9 | 4035 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4036 | { |
4037 | struct location *l; | |
4038 | int order; | |
4039 | ||
88a420e4 CL |
4040 | order = get_order(sizeof(struct location) * max); |
4041 | ||
6fa3eb70 | 4042 | #ifndef CONFIG_MTK_PAGERECORDER |
68dff6a9 | 4043 | l = (void *)__get_free_pages(flags, order); |
6fa3eb70 S |
4044 | #else |
4045 | l = (void *)__get_free_pages_nopagedebug(flags, order); | |
4046 | #endif | |
88a420e4 CL |
4047 | if (!l) |
4048 | return 0; | |
4049 | ||
4050 | if (t->count) { | |
4051 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4052 | free_loc_track(t); | |
4053 | } | |
4054 | t->max = max; | |
4055 | t->loc = l; | |
4056 | return 1; | |
4057 | } | |
4058 | ||
4059 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4060 | const struct track *track) |
88a420e4 CL |
4061 | { |
4062 | long start, end, pos; | |
4063 | struct location *l; | |
ce71e27c | 4064 | unsigned long caddr; |
45edfa58 | 4065 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4066 | |
4067 | start = -1; | |
4068 | end = t->count; | |
4069 | ||
4070 | for ( ; ; ) { | |
4071 | pos = start + (end - start + 1) / 2; | |
4072 | ||
4073 | /* | |
4074 | * There is nothing at "end". If we end up there | |
4075 | * we need to add something to before end. | |
4076 | */ | |
4077 | if (pos == end) | |
4078 | break; | |
4079 | ||
4080 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4081 | if (track->addr == caddr) { |
4082 | ||
4083 | l = &t->loc[pos]; | |
4084 | l->count++; | |
4085 | if (track->when) { | |
4086 | l->sum_time += age; | |
4087 | if (age < l->min_time) | |
4088 | l->min_time = age; | |
4089 | if (age > l->max_time) | |
4090 | l->max_time = age; | |
4091 | ||
4092 | if (track->pid < l->min_pid) | |
4093 | l->min_pid = track->pid; | |
4094 | if (track->pid > l->max_pid) | |
4095 | l->max_pid = track->pid; | |
4096 | ||
174596a0 RR |
4097 | cpumask_set_cpu(track->cpu, |
4098 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4099 | } |
4100 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4101 | return 1; |
4102 | } | |
4103 | ||
45edfa58 | 4104 | if (track->addr < caddr) |
88a420e4 CL |
4105 | end = pos; |
4106 | else | |
4107 | start = pos; | |
4108 | } | |
4109 | ||
4110 | /* | |
672bba3a | 4111 | * Not found. Insert new tracking element. |
88a420e4 | 4112 | */ |
68dff6a9 | 4113 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4114 | return 0; |
4115 | ||
4116 | l = t->loc + pos; | |
4117 | if (pos < t->count) | |
4118 | memmove(l + 1, l, | |
4119 | (t->count - pos) * sizeof(struct location)); | |
4120 | t->count++; | |
4121 | l->count = 1; | |
45edfa58 CL |
4122 | l->addr = track->addr; |
4123 | l->sum_time = age; | |
4124 | l->min_time = age; | |
4125 | l->max_time = age; | |
4126 | l->min_pid = track->pid; | |
4127 | l->max_pid = track->pid; | |
174596a0 RR |
4128 | cpumask_clear(to_cpumask(l->cpus)); |
4129 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4130 | nodes_clear(l->nodes); |
4131 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4132 | return 1; |
4133 | } | |
4134 | ||
4135 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4136 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4137 | unsigned long *map) |
88a420e4 | 4138 | { |
a973e9dd | 4139 | void *addr = page_address(page); |
88a420e4 CL |
4140 | void *p; |
4141 | ||
39b26464 | 4142 | bitmap_zero(map, page->objects); |
5f80b13a | 4143 | get_map(s, page, map); |
88a420e4 | 4144 | |
224a88be | 4145 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4146 | if (!test_bit(slab_index(p, s, addr), map)) |
4147 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4148 | } |
4149 | ||
4150 | static int list_locations(struct kmem_cache *s, char *buf, | |
4151 | enum track_item alloc) | |
4152 | { | |
e374d483 | 4153 | int len = 0; |
88a420e4 | 4154 | unsigned long i; |
68dff6a9 | 4155 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4156 | int node; |
bbd7d57b ED |
4157 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4158 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4159 | |
bbd7d57b ED |
4160 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4161 | GFP_TEMPORARY)) { | |
4162 | kfree(map); | |
68dff6a9 | 4163 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4164 | } |
88a420e4 CL |
4165 | /* Push back cpu slabs */ |
4166 | flush_all(s); | |
4167 | ||
f64dc58c | 4168 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4169 | struct kmem_cache_node *n = get_node(s, node); |
4170 | unsigned long flags; | |
4171 | struct page *page; | |
4172 | ||
9e86943b | 4173 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4174 | continue; |
4175 | ||
4176 | spin_lock_irqsave(&n->list_lock, flags); | |
4177 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4178 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4179 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4180 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4181 | spin_unlock_irqrestore(&n->list_lock, flags); |
4182 | } | |
4183 | ||
4184 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4185 | struct location *l = &t.loc[i]; |
88a420e4 | 4186 | |
9c246247 | 4187 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4188 | break; |
e374d483 | 4189 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4190 | |
4191 | if (l->addr) | |
62c70bce | 4192 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4193 | else |
e374d483 | 4194 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4195 | |
4196 | if (l->sum_time != l->min_time) { | |
e374d483 | 4197 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4198 | l->min_time, |
4199 | (long)div_u64(l->sum_time, l->count), | |
4200 | l->max_time); | |
45edfa58 | 4201 | } else |
e374d483 | 4202 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4203 | l->min_time); |
4204 | ||
4205 | if (l->min_pid != l->max_pid) | |
e374d483 | 4206 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4207 | l->min_pid, l->max_pid); |
4208 | else | |
e374d483 | 4209 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4210 | l->min_pid); |
4211 | ||
174596a0 RR |
4212 | if (num_online_cpus() > 1 && |
4213 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4214 | len < PAGE_SIZE - 60) { |
4215 | len += sprintf(buf + len, " cpus="); | |
4216 | len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
174596a0 | 4217 | to_cpumask(l->cpus)); |
45edfa58 CL |
4218 | } |
4219 | ||
62bc62a8 | 4220 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4221 | len < PAGE_SIZE - 60) { |
4222 | len += sprintf(buf + len, " nodes="); | |
4223 | len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50, | |
45edfa58 CL |
4224 | l->nodes); |
4225 | } | |
4226 | ||
e374d483 | 4227 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4228 | } |
4229 | ||
4230 | free_loc_track(&t); | |
bbd7d57b | 4231 | kfree(map); |
88a420e4 | 4232 | if (!t.count) |
e374d483 HH |
4233 | len += sprintf(buf, "No data\n"); |
4234 | return len; | |
88a420e4 | 4235 | } |
ab4d5ed5 | 4236 | #endif |
88a420e4 | 4237 | |
a5a84755 CL |
4238 | #ifdef SLUB_RESILIENCY_TEST |
4239 | static void resiliency_test(void) | |
4240 | { | |
4241 | u8 *p; | |
4242 | ||
95a05b42 | 4243 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 CL |
4244 | |
4245 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4246 | printk(KERN_ERR "-----------------------\n"); | |
4247 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4248 | ||
4249 | p = kzalloc(16, GFP_KERNEL); | |
4250 | p[16] = 0x12; | |
4251 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4252 | " 0x12->0x%p\n\n", p + 16); | |
4253 | ||
4254 | validate_slab_cache(kmalloc_caches[4]); | |
4255 | ||
4256 | /* Hmmm... The next two are dangerous */ | |
4257 | p = kzalloc(32, GFP_KERNEL); | |
4258 | p[32 + sizeof(void *)] = 0x34; | |
4259 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4260 | " 0x34 -> -0x%p\n", p); | |
4261 | printk(KERN_ERR | |
4262 | "If allocated object is overwritten then not detectable\n\n"); | |
4263 | ||
4264 | validate_slab_cache(kmalloc_caches[5]); | |
4265 | p = kzalloc(64, GFP_KERNEL); | |
4266 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4267 | *p = 0x56; | |
4268 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4269 | p); | |
4270 | printk(KERN_ERR | |
4271 | "If allocated object is overwritten then not detectable\n\n"); | |
4272 | validate_slab_cache(kmalloc_caches[6]); | |
4273 | ||
4274 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4275 | p = kzalloc(128, GFP_KERNEL); | |
4276 | kfree(p); | |
4277 | *p = 0x78; | |
4278 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4279 | validate_slab_cache(kmalloc_caches[7]); | |
4280 | ||
4281 | p = kzalloc(256, GFP_KERNEL); | |
4282 | kfree(p); | |
4283 | p[50] = 0x9a; | |
4284 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4285 | p); | |
4286 | validate_slab_cache(kmalloc_caches[8]); | |
4287 | ||
4288 | p = kzalloc(512, GFP_KERNEL); | |
4289 | kfree(p); | |
4290 | p[512] = 0xab; | |
4291 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4292 | validate_slab_cache(kmalloc_caches[9]); | |
4293 | } | |
4294 | #else | |
4295 | #ifdef CONFIG_SYSFS | |
4296 | static void resiliency_test(void) {}; | |
4297 | #endif | |
4298 | #endif | |
4299 | ||
ab4d5ed5 | 4300 | #ifdef CONFIG_SYSFS |
81819f0f | 4301 | enum slab_stat_type { |
205ab99d CL |
4302 | SL_ALL, /* All slabs */ |
4303 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4304 | SL_CPU, /* Only slabs used for cpu caches */ | |
4305 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4306 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4307 | }; |
4308 | ||
205ab99d | 4309 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4310 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4311 | #define SO_CPU (1 << SL_CPU) | |
4312 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4313 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4314 | |
62e5c4b4 CG |
4315 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4316 | char *buf, unsigned long flags) | |
81819f0f CL |
4317 | { |
4318 | unsigned long total = 0; | |
81819f0f CL |
4319 | int node; |
4320 | int x; | |
4321 | unsigned long *nodes; | |
4322 | unsigned long *per_cpu; | |
4323 | ||
4324 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
62e5c4b4 CG |
4325 | if (!nodes) |
4326 | return -ENOMEM; | |
81819f0f CL |
4327 | per_cpu = nodes + nr_node_ids; |
4328 | ||
205ab99d CL |
4329 | if (flags & SO_CPU) { |
4330 | int cpu; | |
81819f0f | 4331 | |
205ab99d | 4332 | for_each_possible_cpu(cpu) { |
9dfc6e68 | 4333 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
ec3ab083 | 4334 | int node; |
49e22585 | 4335 | struct page *page; |
dfb4f096 | 4336 | |
bc6697d8 | 4337 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4338 | if (!page) |
4339 | continue; | |
205ab99d | 4340 | |
ec3ab083 CL |
4341 | node = page_to_nid(page); |
4342 | if (flags & SO_TOTAL) | |
4343 | x = page->objects; | |
4344 | else if (flags & SO_OBJECTS) | |
4345 | x = page->inuse; | |
4346 | else | |
4347 | x = 1; | |
49e22585 | 4348 | |
ec3ab083 CL |
4349 | total += x; |
4350 | nodes[node] += x; | |
4351 | ||
4352 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4353 | if (page) { |
d6d76c66 LZ |
4354 | node = page_to_nid(page); |
4355 | if (flags & SO_TOTAL) | |
4356 | WARN_ON_ONCE(1); | |
4357 | else if (flags & SO_OBJECTS) | |
4358 | WARN_ON_ONCE(1); | |
4359 | else | |
4360 | x = page->pages; | |
bc6697d8 ED |
4361 | total += x; |
4362 | nodes[node] += x; | |
49e22585 | 4363 | } |
ec3ab083 | 4364 | |
bc6697d8 | 4365 | per_cpu[node]++; |
81819f0f CL |
4366 | } |
4367 | } | |
4368 | ||
04d94879 | 4369 | lock_memory_hotplug(); |
ab4d5ed5 | 4370 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4371 | if (flags & SO_ALL) { |
4372 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4373 | struct kmem_cache_node *n = get_node(s, node); | |
4374 | ||
4375 | if (flags & SO_TOTAL) | |
4376 | x = atomic_long_read(&n->total_objects); | |
4377 | else if (flags & SO_OBJECTS) | |
4378 | x = atomic_long_read(&n->total_objects) - | |
4379 | count_partial(n, count_free); | |
81819f0f | 4380 | |
81819f0f | 4381 | else |
205ab99d | 4382 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4383 | total += x; |
4384 | nodes[node] += x; | |
4385 | } | |
4386 | ||
ab4d5ed5 CL |
4387 | } else |
4388 | #endif | |
4389 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4390 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4391 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4392 | |
205ab99d CL |
4393 | if (flags & SO_TOTAL) |
4394 | x = count_partial(n, count_total); | |
4395 | else if (flags & SO_OBJECTS) | |
4396 | x = count_partial(n, count_inuse); | |
81819f0f | 4397 | else |
205ab99d | 4398 | x = n->nr_partial; |
81819f0f CL |
4399 | total += x; |
4400 | nodes[node] += x; | |
4401 | } | |
4402 | } | |
81819f0f CL |
4403 | x = sprintf(buf, "%lu", total); |
4404 | #ifdef CONFIG_NUMA | |
f64dc58c | 4405 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4406 | if (nodes[node]) |
4407 | x += sprintf(buf + x, " N%d=%lu", | |
4408 | node, nodes[node]); | |
4409 | #endif | |
04d94879 | 4410 | unlock_memory_hotplug(); |
81819f0f CL |
4411 | kfree(nodes); |
4412 | return x + sprintf(buf + x, "\n"); | |
4413 | } | |
4414 | ||
ab4d5ed5 | 4415 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4416 | static int any_slab_objects(struct kmem_cache *s) |
4417 | { | |
4418 | int node; | |
81819f0f | 4419 | |
dfb4f096 | 4420 | for_each_online_node(node) { |
81819f0f CL |
4421 | struct kmem_cache_node *n = get_node(s, node); |
4422 | ||
dfb4f096 CL |
4423 | if (!n) |
4424 | continue; | |
4425 | ||
4ea33e2d | 4426 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4427 | return 1; |
4428 | } | |
4429 | return 0; | |
4430 | } | |
ab4d5ed5 | 4431 | #endif |
81819f0f CL |
4432 | |
4433 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4434 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4435 | |
4436 | struct slab_attribute { | |
4437 | struct attribute attr; | |
4438 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4439 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4440 | }; | |
4441 | ||
4442 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4443 | static struct slab_attribute _name##_attr = \ |
4444 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4445 | |
4446 | #define SLAB_ATTR(_name) \ | |
4447 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4448 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4449 | |
81819f0f CL |
4450 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4451 | { | |
4452 | return sprintf(buf, "%d\n", s->size); | |
4453 | } | |
4454 | SLAB_ATTR_RO(slab_size); | |
4455 | ||
4456 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4457 | { | |
4458 | return sprintf(buf, "%d\n", s->align); | |
4459 | } | |
4460 | SLAB_ATTR_RO(align); | |
4461 | ||
4462 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4463 | { | |
3b0efdfa | 4464 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4465 | } |
4466 | SLAB_ATTR_RO(object_size); | |
4467 | ||
4468 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4469 | { | |
834f3d11 | 4470 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4471 | } |
4472 | SLAB_ATTR_RO(objs_per_slab); | |
4473 | ||
06b285dc CL |
4474 | static ssize_t order_store(struct kmem_cache *s, |
4475 | const char *buf, size_t length) | |
4476 | { | |
0121c619 CL |
4477 | unsigned long order; |
4478 | int err; | |
4479 | ||
4480 | err = strict_strtoul(buf, 10, &order); | |
4481 | if (err) | |
4482 | return err; | |
06b285dc CL |
4483 | |
4484 | if (order > slub_max_order || order < slub_min_order) | |
4485 | return -EINVAL; | |
4486 | ||
4487 | calculate_sizes(s, order); | |
4488 | return length; | |
4489 | } | |
4490 | ||
81819f0f CL |
4491 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4492 | { | |
834f3d11 | 4493 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4494 | } |
06b285dc | 4495 | SLAB_ATTR(order); |
81819f0f | 4496 | |
73d342b1 DR |
4497 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4498 | { | |
4499 | return sprintf(buf, "%lu\n", s->min_partial); | |
4500 | } | |
4501 | ||
4502 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4503 | size_t length) | |
4504 | { | |
4505 | unsigned long min; | |
4506 | int err; | |
4507 | ||
4508 | err = strict_strtoul(buf, 10, &min); | |
4509 | if (err) | |
4510 | return err; | |
4511 | ||
c0bdb232 | 4512 | set_min_partial(s, min); |
73d342b1 DR |
4513 | return length; |
4514 | } | |
4515 | SLAB_ATTR(min_partial); | |
4516 | ||
49e22585 CL |
4517 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4518 | { | |
4519 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4520 | } | |
4521 | ||
4522 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4523 | size_t length) | |
4524 | { | |
4525 | unsigned long objects; | |
4526 | int err; | |
4527 | ||
4528 | err = strict_strtoul(buf, 10, &objects); | |
4529 | if (err) | |
4530 | return err; | |
74ee4ef1 DR |
4531 | if (objects && kmem_cache_debug(s)) |
4532 | return -EINVAL; | |
49e22585 CL |
4533 | |
4534 | s->cpu_partial = objects; | |
4535 | flush_all(s); | |
4536 | return length; | |
4537 | } | |
4538 | SLAB_ATTR(cpu_partial); | |
4539 | ||
81819f0f CL |
4540 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4541 | { | |
62c70bce JP |
4542 | if (!s->ctor) |
4543 | return 0; | |
4544 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4545 | } |
4546 | SLAB_ATTR_RO(ctor); | |
4547 | ||
81819f0f CL |
4548 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4549 | { | |
4550 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4551 | } | |
4552 | SLAB_ATTR_RO(aliases); | |
4553 | ||
81819f0f CL |
4554 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4555 | { | |
d9acf4b7 | 4556 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4557 | } |
4558 | SLAB_ATTR_RO(partial); | |
4559 | ||
4560 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4561 | { | |
d9acf4b7 | 4562 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4563 | } |
4564 | SLAB_ATTR_RO(cpu_slabs); | |
4565 | ||
4566 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4567 | { | |
205ab99d | 4568 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4569 | } |
4570 | SLAB_ATTR_RO(objects); | |
4571 | ||
205ab99d CL |
4572 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4573 | { | |
4574 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4575 | } | |
4576 | SLAB_ATTR_RO(objects_partial); | |
4577 | ||
49e22585 CL |
4578 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4579 | { | |
4580 | int objects = 0; | |
4581 | int pages = 0; | |
4582 | int cpu; | |
4583 | int len; | |
4584 | ||
4585 | for_each_online_cpu(cpu) { | |
4586 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4587 | ||
4588 | if (page) { | |
4589 | pages += page->pages; | |
4590 | objects += page->pobjects; | |
4591 | } | |
4592 | } | |
4593 | ||
4594 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4595 | ||
4596 | #ifdef CONFIG_SMP | |
4597 | for_each_online_cpu(cpu) { | |
4598 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4599 | ||
4600 | if (page && len < PAGE_SIZE - 20) | |
4601 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4602 | page->pobjects, page->pages); | |
4603 | } | |
4604 | #endif | |
4605 | return len + sprintf(buf + len, "\n"); | |
4606 | } | |
4607 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4608 | ||
a5a84755 CL |
4609 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4610 | { | |
4611 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4612 | } | |
4613 | ||
4614 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4615 | const char *buf, size_t length) | |
4616 | { | |
4617 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4618 | if (buf[0] == '1') | |
4619 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4620 | return length; | |
4621 | } | |
4622 | SLAB_ATTR(reclaim_account); | |
4623 | ||
4624 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4625 | { | |
4626 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4627 | } | |
4628 | SLAB_ATTR_RO(hwcache_align); | |
4629 | ||
4630 | #ifdef CONFIG_ZONE_DMA | |
4631 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4632 | { | |
4633 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4634 | } | |
4635 | SLAB_ATTR_RO(cache_dma); | |
4636 | #endif | |
4637 | ||
4638 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4639 | { | |
4640 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4641 | } | |
4642 | SLAB_ATTR_RO(destroy_by_rcu); | |
4643 | ||
ab9a0f19 LJ |
4644 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4645 | { | |
4646 | return sprintf(buf, "%d\n", s->reserved); | |
4647 | } | |
4648 | SLAB_ATTR_RO(reserved); | |
4649 | ||
ab4d5ed5 | 4650 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4651 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4652 | { | |
4653 | return show_slab_objects(s, buf, SO_ALL); | |
4654 | } | |
4655 | SLAB_ATTR_RO(slabs); | |
4656 | ||
205ab99d CL |
4657 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4658 | { | |
4659 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4660 | } | |
4661 | SLAB_ATTR_RO(total_objects); | |
4662 | ||
81819f0f CL |
4663 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4664 | { | |
4665 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4666 | } | |
4667 | ||
4668 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4669 | const char *buf, size_t length) | |
4670 | { | |
4671 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4672 | if (buf[0] == '1') { |
4673 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4674 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4675 | } |
81819f0f CL |
4676 | return length; |
4677 | } | |
4678 | SLAB_ATTR(sanity_checks); | |
4679 | ||
4680 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4681 | { | |
4682 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4683 | } | |
4684 | ||
4685 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4686 | size_t length) | |
4687 | { | |
4688 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4689 | if (buf[0] == '1') { |
4690 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4691 | s->flags |= SLAB_TRACE; |
b789ef51 | 4692 | } |
81819f0f CL |
4693 | return length; |
4694 | } | |
4695 | SLAB_ATTR(trace); | |
4696 | ||
81819f0f CL |
4697 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4698 | { | |
4699 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4700 | } | |
4701 | ||
4702 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4703 | const char *buf, size_t length) | |
4704 | { | |
4705 | if (any_slab_objects(s)) | |
4706 | return -EBUSY; | |
4707 | ||
4708 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4709 | if (buf[0] == '1') { |
4710 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4711 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4712 | } |
06b285dc | 4713 | calculate_sizes(s, -1); |
81819f0f CL |
4714 | return length; |
4715 | } | |
4716 | SLAB_ATTR(red_zone); | |
4717 | ||
4718 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4719 | { | |
4720 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4721 | } | |
4722 | ||
4723 | static ssize_t poison_store(struct kmem_cache *s, | |
4724 | const char *buf, size_t length) | |
4725 | { | |
4726 | if (any_slab_objects(s)) | |
4727 | return -EBUSY; | |
4728 | ||
4729 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4730 | if (buf[0] == '1') { |
4731 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4732 | s->flags |= SLAB_POISON; |
b789ef51 | 4733 | } |
06b285dc | 4734 | calculate_sizes(s, -1); |
81819f0f CL |
4735 | return length; |
4736 | } | |
4737 | SLAB_ATTR(poison); | |
4738 | ||
4739 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4740 | { | |
4741 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4742 | } | |
4743 | ||
4744 | static ssize_t store_user_store(struct kmem_cache *s, | |
4745 | const char *buf, size_t length) | |
4746 | { | |
4747 | if (any_slab_objects(s)) | |
4748 | return -EBUSY; | |
4749 | ||
4750 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4751 | if (buf[0] == '1') { |
4752 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4753 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4754 | } |
06b285dc | 4755 | calculate_sizes(s, -1); |
81819f0f CL |
4756 | return length; |
4757 | } | |
4758 | SLAB_ATTR(store_user); | |
4759 | ||
53e15af0 CL |
4760 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4761 | { | |
4762 | return 0; | |
4763 | } | |
4764 | ||
4765 | static ssize_t validate_store(struct kmem_cache *s, | |
4766 | const char *buf, size_t length) | |
4767 | { | |
434e245d CL |
4768 | int ret = -EINVAL; |
4769 | ||
4770 | if (buf[0] == '1') { | |
4771 | ret = validate_slab_cache(s); | |
4772 | if (ret >= 0) | |
4773 | ret = length; | |
4774 | } | |
4775 | return ret; | |
53e15af0 CL |
4776 | } |
4777 | SLAB_ATTR(validate); | |
a5a84755 CL |
4778 | |
4779 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4780 | { | |
4781 | if (!(s->flags & SLAB_STORE_USER)) | |
4782 | return -ENOSYS; | |
4783 | return list_locations(s, buf, TRACK_ALLOC); | |
4784 | } | |
4785 | SLAB_ATTR_RO(alloc_calls); | |
4786 | ||
4787 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4788 | { | |
4789 | if (!(s->flags & SLAB_STORE_USER)) | |
4790 | return -ENOSYS; | |
4791 | return list_locations(s, buf, TRACK_FREE); | |
4792 | } | |
4793 | SLAB_ATTR_RO(free_calls); | |
4794 | #endif /* CONFIG_SLUB_DEBUG */ | |
4795 | ||
4796 | #ifdef CONFIG_FAILSLAB | |
4797 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4798 | { | |
4799 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4800 | } | |
4801 | ||
4802 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4803 | size_t length) | |
4804 | { | |
4805 | s->flags &= ~SLAB_FAILSLAB; | |
4806 | if (buf[0] == '1') | |
4807 | s->flags |= SLAB_FAILSLAB; | |
4808 | return length; | |
4809 | } | |
4810 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4811 | #endif |
53e15af0 | 4812 | |
2086d26a CL |
4813 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4814 | { | |
4815 | return 0; | |
4816 | } | |
4817 | ||
4818 | static ssize_t shrink_store(struct kmem_cache *s, | |
4819 | const char *buf, size_t length) | |
4820 | { | |
4821 | if (buf[0] == '1') { | |
4822 | int rc = kmem_cache_shrink(s); | |
4823 | ||
4824 | if (rc) | |
4825 | return rc; | |
4826 | } else | |
4827 | return -EINVAL; | |
4828 | return length; | |
4829 | } | |
4830 | SLAB_ATTR(shrink); | |
4831 | ||
81819f0f | 4832 | #ifdef CONFIG_NUMA |
9824601e | 4833 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4834 | { |
9824601e | 4835 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4836 | } |
4837 | ||
9824601e | 4838 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4839 | const char *buf, size_t length) |
4840 | { | |
0121c619 CL |
4841 | unsigned long ratio; |
4842 | int err; | |
4843 | ||
4844 | err = strict_strtoul(buf, 10, &ratio); | |
4845 | if (err) | |
4846 | return err; | |
4847 | ||
e2cb96b7 | 4848 | if (ratio <= 100) |
0121c619 | 4849 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4850 | |
81819f0f CL |
4851 | return length; |
4852 | } | |
9824601e | 4853 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4854 | #endif |
4855 | ||
8ff12cfc | 4856 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4857 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4858 | { | |
4859 | unsigned long sum = 0; | |
4860 | int cpu; | |
4861 | int len; | |
4862 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4863 | ||
4864 | if (!data) | |
4865 | return -ENOMEM; | |
4866 | ||
4867 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4868 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4869 | |
4870 | data[cpu] = x; | |
4871 | sum += x; | |
4872 | } | |
4873 | ||
4874 | len = sprintf(buf, "%lu", sum); | |
4875 | ||
50ef37b9 | 4876 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4877 | for_each_online_cpu(cpu) { |
4878 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4879 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4880 | } |
50ef37b9 | 4881 | #endif |
8ff12cfc CL |
4882 | kfree(data); |
4883 | return len + sprintf(buf + len, "\n"); | |
4884 | } | |
4885 | ||
78eb00cc DR |
4886 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4887 | { | |
4888 | int cpu; | |
4889 | ||
4890 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4891 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4892 | } |
4893 | ||
8ff12cfc CL |
4894 | #define STAT_ATTR(si, text) \ |
4895 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4896 | { \ | |
4897 | return show_stat(s, buf, si); \ | |
4898 | } \ | |
78eb00cc DR |
4899 | static ssize_t text##_store(struct kmem_cache *s, \ |
4900 | const char *buf, size_t length) \ | |
4901 | { \ | |
4902 | if (buf[0] != '0') \ | |
4903 | return -EINVAL; \ | |
4904 | clear_stat(s, si); \ | |
4905 | return length; \ | |
4906 | } \ | |
4907 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4908 | |
4909 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4910 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4911 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4912 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4913 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4914 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4915 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4916 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4917 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4918 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4919 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4920 | STAT_ATTR(FREE_SLAB, free_slab); |
4921 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4922 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4923 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4924 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4925 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4926 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4927 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4928 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4929 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4930 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4931 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4932 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4933 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4934 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4935 | #endif |
4936 | ||
06428780 | 4937 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4938 | &slab_size_attr.attr, |
4939 | &object_size_attr.attr, | |
4940 | &objs_per_slab_attr.attr, | |
4941 | &order_attr.attr, | |
73d342b1 | 4942 | &min_partial_attr.attr, |
49e22585 | 4943 | &cpu_partial_attr.attr, |
81819f0f | 4944 | &objects_attr.attr, |
205ab99d | 4945 | &objects_partial_attr.attr, |
81819f0f CL |
4946 | &partial_attr.attr, |
4947 | &cpu_slabs_attr.attr, | |
4948 | &ctor_attr.attr, | |
81819f0f CL |
4949 | &aliases_attr.attr, |
4950 | &align_attr.attr, | |
81819f0f CL |
4951 | &hwcache_align_attr.attr, |
4952 | &reclaim_account_attr.attr, | |
4953 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4954 | &shrink_attr.attr, |
ab9a0f19 | 4955 | &reserved_attr.attr, |
49e22585 | 4956 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4957 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4958 | &total_objects_attr.attr, |
4959 | &slabs_attr.attr, | |
4960 | &sanity_checks_attr.attr, | |
4961 | &trace_attr.attr, | |
81819f0f CL |
4962 | &red_zone_attr.attr, |
4963 | &poison_attr.attr, | |
4964 | &store_user_attr.attr, | |
53e15af0 | 4965 | &validate_attr.attr, |
88a420e4 CL |
4966 | &alloc_calls_attr.attr, |
4967 | &free_calls_attr.attr, | |
ab4d5ed5 | 4968 | #endif |
81819f0f CL |
4969 | #ifdef CONFIG_ZONE_DMA |
4970 | &cache_dma_attr.attr, | |
4971 | #endif | |
4972 | #ifdef CONFIG_NUMA | |
9824601e | 4973 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4974 | #endif |
4975 | #ifdef CONFIG_SLUB_STATS | |
4976 | &alloc_fastpath_attr.attr, | |
4977 | &alloc_slowpath_attr.attr, | |
4978 | &free_fastpath_attr.attr, | |
4979 | &free_slowpath_attr.attr, | |
4980 | &free_frozen_attr.attr, | |
4981 | &free_add_partial_attr.attr, | |
4982 | &free_remove_partial_attr.attr, | |
4983 | &alloc_from_partial_attr.attr, | |
4984 | &alloc_slab_attr.attr, | |
4985 | &alloc_refill_attr.attr, | |
e36a2652 | 4986 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4987 | &free_slab_attr.attr, |
4988 | &cpuslab_flush_attr.attr, | |
4989 | &deactivate_full_attr.attr, | |
4990 | &deactivate_empty_attr.attr, | |
4991 | &deactivate_to_head_attr.attr, | |
4992 | &deactivate_to_tail_attr.attr, | |
4993 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4994 | &deactivate_bypass_attr.attr, |
65c3376a | 4995 | &order_fallback_attr.attr, |
b789ef51 CL |
4996 | &cmpxchg_double_fail_attr.attr, |
4997 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4998 | &cpu_partial_alloc_attr.attr, |
4999 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5000 | &cpu_partial_node_attr.attr, |
5001 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5002 | #endif |
4c13dd3b DM |
5003 | #ifdef CONFIG_FAILSLAB |
5004 | &failslab_attr.attr, | |
5005 | #endif | |
5006 | ||
81819f0f CL |
5007 | NULL |
5008 | }; | |
5009 | ||
5010 | static struct attribute_group slab_attr_group = { | |
5011 | .attrs = slab_attrs, | |
5012 | }; | |
5013 | ||
5014 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5015 | struct attribute *attr, | |
5016 | char *buf) | |
5017 | { | |
5018 | struct slab_attribute *attribute; | |
5019 | struct kmem_cache *s; | |
5020 | int err; | |
5021 | ||
5022 | attribute = to_slab_attr(attr); | |
5023 | s = to_slab(kobj); | |
5024 | ||
5025 | if (!attribute->show) | |
5026 | return -EIO; | |
5027 | ||
5028 | err = attribute->show(s, buf); | |
5029 | ||
5030 | return err; | |
5031 | } | |
5032 | ||
5033 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5034 | struct attribute *attr, | |
5035 | const char *buf, size_t len) | |
5036 | { | |
5037 | struct slab_attribute *attribute; | |
5038 | struct kmem_cache *s; | |
5039 | int err; | |
5040 | ||
5041 | attribute = to_slab_attr(attr); | |
5042 | s = to_slab(kobj); | |
5043 | ||
5044 | if (!attribute->store) | |
5045 | return -EIO; | |
5046 | ||
5047 | err = attribute->store(s, buf, len); | |
107dab5c GC |
5048 | #ifdef CONFIG_MEMCG_KMEM |
5049 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
5050 | int i; | |
81819f0f | 5051 | |
107dab5c GC |
5052 | mutex_lock(&slab_mutex); |
5053 | if (s->max_attr_size < len) | |
5054 | s->max_attr_size = len; | |
5055 | ||
ebe945c2 GC |
5056 | /* |
5057 | * This is a best effort propagation, so this function's return | |
5058 | * value will be determined by the parent cache only. This is | |
5059 | * basically because not all attributes will have a well | |
5060 | * defined semantics for rollbacks - most of the actions will | |
5061 | * have permanent effects. | |
5062 | * | |
5063 | * Returning the error value of any of the children that fail | |
5064 | * is not 100 % defined, in the sense that users seeing the | |
5065 | * error code won't be able to know anything about the state of | |
5066 | * the cache. | |
5067 | * | |
5068 | * Only returning the error code for the parent cache at least | |
5069 | * has well defined semantics. The cache being written to | |
5070 | * directly either failed or succeeded, in which case we loop | |
5071 | * through the descendants with best-effort propagation. | |
5072 | */ | |
107dab5c GC |
5073 | for_each_memcg_cache_index(i) { |
5074 | struct kmem_cache *c = cache_from_memcg(s, i); | |
107dab5c GC |
5075 | if (c) |
5076 | attribute->store(c, buf, len); | |
5077 | } | |
5078 | mutex_unlock(&slab_mutex); | |
5079 | } | |
5080 | #endif | |
81819f0f CL |
5081 | return err; |
5082 | } | |
5083 | ||
107dab5c GC |
5084 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5085 | { | |
5086 | #ifdef CONFIG_MEMCG_KMEM | |
5087 | int i; | |
5088 | char *buffer = NULL; | |
5089 | ||
5090 | if (!is_root_cache(s)) | |
5091 | return; | |
5092 | ||
5093 | /* | |
5094 | * This mean this cache had no attribute written. Therefore, no point | |
5095 | * in copying default values around | |
5096 | */ | |
5097 | if (!s->max_attr_size) | |
5098 | return; | |
5099 | ||
5100 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5101 | char mbuf[64]; | |
5102 | char *buf; | |
5103 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5104 | ||
5105 | if (!attr || !attr->store || !attr->show) | |
5106 | continue; | |
5107 | ||
5108 | /* | |
5109 | * It is really bad that we have to allocate here, so we will | |
5110 | * do it only as a fallback. If we actually allocate, though, | |
5111 | * we can just use the allocated buffer until the end. | |
5112 | * | |
5113 | * Most of the slub attributes will tend to be very small in | |
5114 | * size, but sysfs allows buffers up to a page, so they can | |
5115 | * theoretically happen. | |
5116 | */ | |
5117 | if (buffer) | |
5118 | buf = buffer; | |
5119 | else if (s->max_attr_size < ARRAY_SIZE(mbuf)) | |
5120 | buf = mbuf; | |
5121 | else { | |
5122 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5123 | if (WARN_ON(!buffer)) | |
5124 | continue; | |
5125 | buf = buffer; | |
5126 | } | |
5127 | ||
5128 | attr->show(s->memcg_params->root_cache, buf); | |
5129 | attr->store(s, buf, strlen(buf)); | |
5130 | } | |
5131 | ||
5132 | if (buffer) | |
5133 | free_page((unsigned long)buffer); | |
5134 | #endif | |
5135 | } | |
5136 | ||
52cf25d0 | 5137 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5138 | .show = slab_attr_show, |
5139 | .store = slab_attr_store, | |
5140 | }; | |
5141 | ||
5142 | static struct kobj_type slab_ktype = { | |
5143 | .sysfs_ops = &slab_sysfs_ops, | |
5144 | }; | |
5145 | ||
5146 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5147 | { | |
5148 | struct kobj_type *ktype = get_ktype(kobj); | |
5149 | ||
5150 | if (ktype == &slab_ktype) | |
5151 | return 1; | |
5152 | return 0; | |
5153 | } | |
5154 | ||
9cd43611 | 5155 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5156 | .filter = uevent_filter, |
5157 | }; | |
5158 | ||
27c3a314 | 5159 | static struct kset *slab_kset; |
81819f0f CL |
5160 | |
5161 | #define ID_STR_LENGTH 64 | |
5162 | ||
5163 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5164 | * |
5165 | * Format :[flags-]size | |
81819f0f CL |
5166 | */ |
5167 | static char *create_unique_id(struct kmem_cache *s) | |
5168 | { | |
5169 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5170 | char *p = name; | |
5171 | ||
5172 | BUG_ON(!name); | |
5173 | ||
5174 | *p++ = ':'; | |
5175 | /* | |
5176 | * First flags affecting slabcache operations. We will only | |
5177 | * get here for aliasable slabs so we do not need to support | |
5178 | * too many flags. The flags here must cover all flags that | |
5179 | * are matched during merging to guarantee that the id is | |
5180 | * unique. | |
5181 | */ | |
5182 | if (s->flags & SLAB_CACHE_DMA) | |
5183 | *p++ = 'd'; | |
5184 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5185 | *p++ = 'a'; | |
5186 | if (s->flags & SLAB_DEBUG_FREE) | |
5187 | *p++ = 'F'; | |
5a896d9e VN |
5188 | if (!(s->flags & SLAB_NOTRACK)) |
5189 | *p++ = 't'; | |
81819f0f CL |
5190 | if (p != name + 1) |
5191 | *p++ = '-'; | |
5192 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 GC |
5193 | |
5194 | #ifdef CONFIG_MEMCG_KMEM | |
5195 | if (!is_root_cache(s)) | |
5196 | p += sprintf(p, "-%08d", memcg_cache_id(s->memcg_params->memcg)); | |
5197 | #endif | |
5198 | ||
81819f0f CL |
5199 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5200 | return name; | |
5201 | } | |
5202 | ||
5203 | static int sysfs_slab_add(struct kmem_cache *s) | |
5204 | { | |
5205 | int err; | |
5206 | const char *name; | |
45530c44 | 5207 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5208 | |
81819f0f CL |
5209 | if (unmergeable) { |
5210 | /* | |
5211 | * Slabcache can never be merged so we can use the name proper. | |
5212 | * This is typically the case for debug situations. In that | |
5213 | * case we can catch duplicate names easily. | |
5214 | */ | |
27c3a314 | 5215 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5216 | name = s->name; |
5217 | } else { | |
5218 | /* | |
5219 | * Create a unique name for the slab as a target | |
5220 | * for the symlinks. | |
5221 | */ | |
5222 | name = create_unique_id(s); | |
5223 | } | |
5224 | ||
27c3a314 | 5225 | s->kobj.kset = slab_kset; |
1eada11c GKH |
5226 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name); |
5227 | if (err) { | |
5228 | kobject_put(&s->kobj); | |
81819f0f | 5229 | return err; |
1eada11c | 5230 | } |
81819f0f CL |
5231 | |
5232 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5233 | if (err) { |
5234 | kobject_del(&s->kobj); | |
5235 | kobject_put(&s->kobj); | |
81819f0f | 5236 | return err; |
5788d8ad | 5237 | } |
81819f0f CL |
5238 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5239 | if (!unmergeable) { | |
5240 | /* Setup first alias */ | |
5241 | sysfs_slab_alias(s, s->name); | |
5242 | kfree(name); | |
5243 | } | |
5244 | return 0; | |
5245 | } | |
5246 | ||
5247 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5248 | { | |
97d06609 | 5249 | if (slab_state < FULL) |
2bce6485 CL |
5250 | /* |
5251 | * Sysfs has not been setup yet so no need to remove the | |
5252 | * cache from sysfs. | |
5253 | */ | |
5254 | return; | |
5255 | ||
81819f0f CL |
5256 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5257 | kobject_del(&s->kobj); | |
151c602f | 5258 | kobject_put(&s->kobj); |
81819f0f CL |
5259 | } |
5260 | ||
5261 | /* | |
5262 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5263 | * available lest we lose that information. |
81819f0f CL |
5264 | */ |
5265 | struct saved_alias { | |
5266 | struct kmem_cache *s; | |
5267 | const char *name; | |
5268 | struct saved_alias *next; | |
5269 | }; | |
5270 | ||
5af328a5 | 5271 | static struct saved_alias *alias_list; |
81819f0f CL |
5272 | |
5273 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5274 | { | |
5275 | struct saved_alias *al; | |
5276 | ||
97d06609 | 5277 | if (slab_state == FULL) { |
81819f0f CL |
5278 | /* |
5279 | * If we have a leftover link then remove it. | |
5280 | */ | |
27c3a314 GKH |
5281 | sysfs_remove_link(&slab_kset->kobj, name); |
5282 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5283 | } |
5284 | ||
5285 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5286 | if (!al) | |
5287 | return -ENOMEM; | |
5288 | ||
5289 | al->s = s; | |
5290 | al->name = name; | |
5291 | al->next = alias_list; | |
5292 | alias_list = al; | |
5293 | return 0; | |
5294 | } | |
5295 | ||
5296 | static int __init slab_sysfs_init(void) | |
5297 | { | |
5b95a4ac | 5298 | struct kmem_cache *s; |
81819f0f CL |
5299 | int err; |
5300 | ||
18004c5d | 5301 | mutex_lock(&slab_mutex); |
2bce6485 | 5302 | |
0ff21e46 | 5303 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5304 | if (!slab_kset) { |
18004c5d | 5305 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5306 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5307 | return -ENOSYS; | |
5308 | } | |
5309 | ||
97d06609 | 5310 | slab_state = FULL; |
26a7bd03 | 5311 | |
5b95a4ac | 5312 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5313 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5314 | if (err) |
5315 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5316 | " to sysfs\n", s->name); | |
26a7bd03 | 5317 | } |
81819f0f CL |
5318 | |
5319 | while (alias_list) { | |
5320 | struct saved_alias *al = alias_list; | |
5321 | ||
5322 | alias_list = alias_list->next; | |
5323 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5324 | if (err) |
5325 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5326 | " %s to sysfs\n", al->name); |
81819f0f CL |
5327 | kfree(al); |
5328 | } | |
5329 | ||
18004c5d | 5330 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5331 | resiliency_test(); |
5332 | return 0; | |
5333 | } | |
5334 | ||
5335 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5336 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5337 | |
5338 | /* | |
5339 | * The /proc/slabinfo ABI | |
5340 | */ | |
158a9624 | 5341 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5342 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda PE |
5343 | { |
5344 | unsigned long nr_partials = 0; | |
5345 | unsigned long nr_slabs = 0; | |
205ab99d CL |
5346 | unsigned long nr_objs = 0; |
5347 | unsigned long nr_free = 0; | |
57ed3eda PE |
5348 | int node; |
5349 | ||
57ed3eda PE |
5350 | for_each_online_node(node) { |
5351 | struct kmem_cache_node *n = get_node(s, node); | |
5352 | ||
5353 | if (!n) | |
5354 | continue; | |
5355 | ||
5356 | nr_partials += n->nr_partial; | |
5357 | nr_slabs += atomic_long_read(&n->nr_slabs); | |
205ab99d CL |
5358 | nr_objs += atomic_long_read(&n->total_objects); |
5359 | nr_free += count_partial(n, count_free); | |
57ed3eda PE |
5360 | } |
5361 | ||
0d7561c6 GC |
5362 | sinfo->active_objs = nr_objs - nr_free; |
5363 | sinfo->num_objs = nr_objs; | |
5364 | sinfo->active_slabs = nr_slabs; | |
5365 | sinfo->num_slabs = nr_slabs; | |
5366 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5367 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5368 | } |
5369 | ||
0d7561c6 | 5370 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5371 | { |
7b3c3a50 AD |
5372 | } |
5373 | ||
b7454ad3 GC |
5374 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5375 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5376 | { |
b7454ad3 | 5377 | return -EIO; |
7b3c3a50 | 5378 | } |
6fa3eb70 S |
5379 | |
5380 | #ifdef CONFIG_MTK_MEMCFG | |
5381 | ||
5382 | static int mtk_memcfg_add_location(struct loc_track *t, struct kmem_cache *s, | |
5383 | const struct track *track) | |
5384 | { | |
5385 | long start, end, pos; | |
5386 | struct location *l; | |
5387 | unsigned long (*caddrs)[MTK_MEMCFG_SLABTRACE_CNT]; /* Called from addresses */ | |
5388 | unsigned long taddrs[MTK_MEMCFG_SLABTRACE_CNT] | |
5389 | = { [0 ... MTK_MEMCFG_SLABTRACE_CNT - 1] = 0,}; /* Called from addresses of track */ | |
5390 | unsigned long age = jiffies - track->when; | |
5391 | int i, cnt; | |
5392 | ||
5393 | start = -1; | |
5394 | end = t->count; | |
5395 | ||
5396 | /* find the index of track->addr */ | |
5397 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) { | |
5398 | if ((track->addr == track->addrs[i]) || | |
5399 | (track->addr - 4 == track->addrs[i])) | |
5400 | break; | |
5401 | } | |
5402 | cnt = min(MTK_MEMCFG_SLABTRACE_CNT, TRACK_ADDRS_COUNT - i); | |
5403 | memcpy(taddrs, track->addrs + i, (cnt * sizeof (unsigned long))); | |
5404 | ||
5405 | for ( ; ; ) { | |
5406 | pos = start + (end - start + 1) / 2; | |
5407 | ||
5408 | /* | |
5409 | * There is nothing at "end". If we end up there | |
5410 | * we need to add something to before end. | |
5411 | */ | |
5412 | if (pos == end) | |
5413 | break; | |
5414 | ||
5415 | caddrs = &(t->loc[pos].addrs); | |
5416 | if (!memcmp(caddrs, taddrs, MTK_MEMCFG_SLABTRACE_CNT * sizeof (unsigned long))) { | |
5417 | ||
5418 | l = &t->loc[pos]; | |
5419 | l->count++; | |
5420 | if (track->when) { | |
5421 | l->sum_time += age; | |
5422 | if (age < l->min_time) | |
5423 | l->min_time = age; | |
5424 | if (age > l->max_time) | |
5425 | l->max_time = age; | |
5426 | ||
5427 | if (track->pid < l->min_pid) | |
5428 | l->min_pid = track->pid; | |
5429 | if (track->pid > l->max_pid) | |
5430 | l->max_pid = track->pid; | |
5431 | ||
5432 | cpumask_set_cpu(track->cpu, | |
5433 | to_cpumask(l->cpus)); | |
5434 | } | |
5435 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
5436 | return 1; | |
5437 | } | |
5438 | ||
5439 | if (memcmp(caddrs, taddrs, MTK_MEMCFG_SLABTRACE_CNT * sizeof (unsigned long)) < 0) | |
5440 | end = pos; | |
5441 | else | |
5442 | start = pos; | |
5443 | } | |
5444 | ||
5445 | /* | |
5446 | * Not found. Insert new tracking element. | |
5447 | */ | |
5448 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) | |
5449 | return 0; | |
5450 | ||
5451 | l = t->loc + pos; | |
5452 | if (pos < t->count) | |
5453 | memmove(l + 1, l, | |
5454 | (t->count - pos) * sizeof(struct location)); | |
5455 | t->count++; | |
5456 | l->count = 1; | |
5457 | l->addr = track->addr; | |
5458 | memcpy(l->addrs, taddrs, MTK_MEMCFG_SLABTRACE_CNT * sizeof (unsigned long)); | |
5459 | l->sum_time = age; | |
5460 | l->min_time = age; | |
5461 | l->max_time = age; | |
5462 | l->min_pid = track->pid; | |
5463 | l->max_pid = track->pid; | |
5464 | cpumask_clear(to_cpumask(l->cpus)); | |
5465 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
5466 | nodes_clear(l->nodes); | |
5467 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
5468 | return 1; | |
5469 | } | |
5470 | ||
5471 | static void mtk_memcfg_process_slab(struct loc_track *t, struct kmem_cache *s, | |
5472 | struct page *page, enum track_item alloc, | |
5473 | unsigned long *map) | |
5474 | { | |
5475 | void *addr = page_address(page); | |
5476 | void *p; | |
5477 | ||
5478 | bitmap_zero(map, page->objects); | |
5479 | get_map(s, page, map); | |
5480 | ||
5481 | for_each_object(p, s, addr, page->objects) | |
5482 | if (!test_bit(slab_index(p, s, addr), map)) | |
5483 | mtk_memcfg_add_location(t, s, get_track(s, p, alloc)); | |
5484 | } | |
5485 | ||
5486 | static int mtk_memcfg_list_locations(struct kmem_cache *s, struct seq_file *m, | |
5487 | enum track_item alloc) | |
5488 | { | |
5489 | unsigned long i, j; | |
5490 | struct loc_track t = { 0, 0, NULL }; | |
5491 | int node; | |
5492 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * | |
5493 | sizeof(unsigned long), GFP_KERNEL); | |
5494 | ||
5495 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), | |
5496 | GFP_TEMPORARY)) { | |
5497 | kfree(map); | |
5498 | return seq_printf(m, "Out of memory\n"); | |
5499 | } | |
5500 | /* Push back cpu slabs */ | |
5501 | flush_all(s); | |
5502 | ||
5503 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
5504 | struct kmem_cache_node *n = get_node(s, node); | |
5505 | unsigned long flags; | |
5506 | struct page *page; | |
5507 | ||
5508 | if (!atomic_long_read(&n->nr_slabs)) | |
5509 | continue; | |
5510 | ||
5511 | spin_lock_irqsave(&n->list_lock, flags); | |
5512 | list_for_each_entry(page, &n->partial, lru) | |
5513 | mtk_memcfg_process_slab(&t, s, page, alloc, map); | |
5514 | list_for_each_entry(page, &n->full, lru) | |
5515 | mtk_memcfg_process_slab(&t, s, page, alloc, map); | |
5516 | spin_unlock_irqrestore(&n->list_lock, flags); | |
5517 | } | |
5518 | ||
5519 | for (i = 0; i < t.count; i++) { | |
5520 | struct location *l = &t.loc[i]; | |
5521 | ||
5522 | seq_printf(m, "%7ld ", l->count); | |
5523 | ||
5524 | if (l->addr) | |
5525 | seq_printf(m, "%pS", (void *)l->addr); | |
5526 | else | |
5527 | seq_printf(m, "<not-available>"); | |
5528 | ||
5529 | for (j = 0; j < MTK_MEMCFG_SLABTRACE_CNT; j++) | |
5530 | if (l->addrs[j]) | |
5531 | seq_printf(m, " %p", (void *)l->addrs[j]); | |
5532 | ||
5533 | seq_printf(m, "\n"); | |
5534 | } | |
5535 | ||
5536 | free_loc_track(&t); | |
5537 | kfree(map); | |
5538 | ||
5539 | if (!t.count) | |
5540 | seq_printf(m, "No data\n"); | |
5541 | return 0; | |
5542 | } | |
5543 | ||
5544 | static int mtk_memcfg_slabtrace_show(struct seq_file *m, void *p) | |
5545 | { | |
5546 | struct kmem_cache *s; | |
5547 | mutex_lock(&slab_mutex); | |
5548 | list_for_each_entry(s, &slab_caches, list) { | |
5549 | seq_printf(m, "========== kmem_cache: %s alloc_calls ==========\n", s->name); | |
5550 | if (!(s->flags & SLAB_STORE_USER)) { | |
5551 | continue; | |
5552 | } else { | |
5553 | mtk_memcfg_list_locations(s, m, TRACK_ALLOC); | |
5554 | } | |
5555 | } | |
5556 | mutex_unlock(&slab_mutex); | |
5557 | return 0; | |
5558 | } | |
5559 | ||
5560 | int slabtrace_open(struct inode *inode, struct file *file) | |
5561 | { | |
5562 | return single_open(file, mtk_memcfg_slabtrace_show, NULL); | |
5563 | } | |
5564 | ||
5565 | #endif | |
5566 | ||
158a9624 | 5567 | #endif /* CONFIG_SLABINFO */ |