]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
f315e3fa JK |
160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
162 | ||
163 | #if FREELIST_BYTE_INDEX | |
164 | typedef unsigned char freelist_idx_t; | |
165 | #else | |
166 | typedef unsigned short freelist_idx_t; | |
167 | #endif | |
168 | ||
30321c7b | 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 170 | |
072bb0aa MG |
171 | /* |
172 | * true if a page was allocated from pfmemalloc reserves for network-based | |
173 | * swap | |
174 | */ | |
175 | static bool pfmemalloc_active __read_mostly; | |
176 | ||
1da177e4 LT |
177 | /* |
178 | * struct array_cache | |
179 | * | |
1da177e4 LT |
180 | * Purpose: |
181 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
182 | * - reduce the number of linked list operations | |
183 | * - reduce spinlock operations | |
184 | * | |
185 | * The limit is stored in the per-cpu structure to reduce the data cache | |
186 | * footprint. | |
187 | * | |
188 | */ | |
189 | struct array_cache { | |
190 | unsigned int avail; | |
191 | unsigned int limit; | |
192 | unsigned int batchcount; | |
193 | unsigned int touched; | |
bda5b655 | 194 | void *entry[]; /* |
a737b3e2 AM |
195 | * Must have this definition in here for the proper |
196 | * alignment of array_cache. Also simplifies accessing | |
197 | * the entries. | |
072bb0aa MG |
198 | * |
199 | * Entries should not be directly dereferenced as | |
200 | * entries belonging to slabs marked pfmemalloc will | |
201 | * have the lower bits set SLAB_OBJ_PFMEMALLOC | |
a737b3e2 | 202 | */ |
1da177e4 LT |
203 | }; |
204 | ||
c8522a3a JK |
205 | struct alien_cache { |
206 | spinlock_t lock; | |
207 | struct array_cache ac; | |
208 | }; | |
209 | ||
072bb0aa MG |
210 | #define SLAB_OBJ_PFMEMALLOC 1 |
211 | static inline bool is_obj_pfmemalloc(void *objp) | |
212 | { | |
213 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; | |
214 | } | |
215 | ||
216 | static inline void set_obj_pfmemalloc(void **objp) | |
217 | { | |
218 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); | |
219 | return; | |
220 | } | |
221 | ||
222 | static inline void clear_obj_pfmemalloc(void **objp) | |
223 | { | |
224 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); | |
225 | } | |
226 | ||
e498be7d CL |
227 | /* |
228 | * Need this for bootstrapping a per node allocator. | |
229 | */ | |
bf0dea23 | 230 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 231 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 232 | #define CACHE_CACHE 0 |
bf0dea23 | 233 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 234 | |
ed11d9eb | 235 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 236 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 237 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
238 | int node, struct list_head *list); |
239 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 240 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 241 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 242 | |
e0a42726 IM |
243 | static int slab_early_init = 1; |
244 | ||
ce8eb6c4 | 245 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 246 | |
ce8eb6c4 | 247 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
248 | { |
249 | INIT_LIST_HEAD(&parent->slabs_full); | |
250 | INIT_LIST_HEAD(&parent->slabs_partial); | |
251 | INIT_LIST_HEAD(&parent->slabs_free); | |
252 | parent->shared = NULL; | |
253 | parent->alien = NULL; | |
2e1217cf | 254 | parent->colour_next = 0; |
e498be7d CL |
255 | spin_lock_init(&parent->list_lock); |
256 | parent->free_objects = 0; | |
257 | parent->free_touched = 0; | |
258 | } | |
259 | ||
a737b3e2 AM |
260 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
261 | do { \ | |
262 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 263 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
264 | } while (0) |
265 | ||
a737b3e2 AM |
266 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
267 | do { \ | |
e498be7d CL |
268 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
269 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
270 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
271 | } while (0) | |
1da177e4 | 272 | |
1da177e4 LT |
273 | #define CFLGS_OFF_SLAB (0x80000000UL) |
274 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
275 | ||
276 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
277 | /* |
278 | * Optimization question: fewer reaps means less probability for unnessary | |
279 | * cpucache drain/refill cycles. | |
1da177e4 | 280 | * |
dc6f3f27 | 281 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
282 | * which could lock up otherwise freeable slabs. |
283 | */ | |
5f0985bb JZ |
284 | #define REAPTIMEOUT_AC (2*HZ) |
285 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
286 | |
287 | #if STATS | |
288 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
289 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
290 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
291 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 292 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
293 | #define STATS_SET_HIGH(x) \ |
294 | do { \ | |
295 | if ((x)->num_active > (x)->high_mark) \ | |
296 | (x)->high_mark = (x)->num_active; \ | |
297 | } while (0) | |
1da177e4 LT |
298 | #define STATS_INC_ERR(x) ((x)->errors++) |
299 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 300 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 301 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
302 | #define STATS_SET_FREEABLE(x, i) \ |
303 | do { \ | |
304 | if ((x)->max_freeable < i) \ | |
305 | (x)->max_freeable = i; \ | |
306 | } while (0) | |
1da177e4 LT |
307 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
308 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
309 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
310 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
311 | #else | |
312 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
313 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
314 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
315 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 316 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
317 | #define STATS_SET_HIGH(x) do { } while (0) |
318 | #define STATS_INC_ERR(x) do { } while (0) | |
319 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 320 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 321 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 322 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
323 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
324 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
325 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
326 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
327 | #endif | |
328 | ||
329 | #if DEBUG | |
1da177e4 | 330 | |
a737b3e2 AM |
331 | /* |
332 | * memory layout of objects: | |
1da177e4 | 333 | * 0 : objp |
3dafccf2 | 334 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
335 | * the end of an object is aligned with the end of the real |
336 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 337 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 338 | * redzone word. |
3dafccf2 | 339 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
340 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
341 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 342 | * [BYTES_PER_WORD long] |
1da177e4 | 343 | */ |
343e0d7a | 344 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 345 | { |
3dafccf2 | 346 | return cachep->obj_offset; |
1da177e4 LT |
347 | } |
348 | ||
b46b8f19 | 349 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
350 | { |
351 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
352 | return (unsigned long long*) (objp + obj_offset(cachep) - |
353 | sizeof(unsigned long long)); | |
1da177e4 LT |
354 | } |
355 | ||
b46b8f19 | 356 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
357 | { |
358 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
359 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 360 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 361 | sizeof(unsigned long long) - |
87a927c7 | 362 | REDZONE_ALIGN); |
3b0efdfa | 363 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 364 | sizeof(unsigned long long)); |
1da177e4 LT |
365 | } |
366 | ||
343e0d7a | 367 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
368 | { |
369 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 370 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
371 | } |
372 | ||
373 | #else | |
374 | ||
3dafccf2 | 375 | #define obj_offset(x) 0 |
b46b8f19 DW |
376 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
377 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
378 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
379 | ||
380 | #endif | |
381 | ||
03787301 JK |
382 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
383 | ||
d31676df | 384 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 385 | { |
d31676df JK |
386 | return atomic_read(&cachep->store_user_clean) == 1; |
387 | } | |
03787301 | 388 | |
d31676df JK |
389 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
390 | { | |
391 | atomic_set(&cachep->store_user_clean, 1); | |
392 | } | |
03787301 | 393 | |
d31676df JK |
394 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
395 | { | |
396 | if (is_store_user_clean(cachep)) | |
397 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
398 | } |
399 | ||
400 | #else | |
d31676df | 401 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
402 | |
403 | #endif | |
404 | ||
1da177e4 | 405 | /* |
3df1cccd DR |
406 | * Do not go above this order unless 0 objects fit into the slab or |
407 | * overridden on the command line. | |
1da177e4 | 408 | */ |
543585cc DR |
409 | #define SLAB_MAX_ORDER_HI 1 |
410 | #define SLAB_MAX_ORDER_LO 0 | |
411 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 412 | static bool slab_max_order_set __initdata; |
1da177e4 | 413 | |
6ed5eb22 PE |
414 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
415 | { | |
b49af68f | 416 | struct page *page = virt_to_head_page(obj); |
35026088 | 417 | return page->slab_cache; |
6ed5eb22 PE |
418 | } |
419 | ||
8456a648 | 420 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
421 | unsigned int idx) |
422 | { | |
8456a648 | 423 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
424 | } |
425 | ||
6a2d7a95 | 426 | /* |
3b0efdfa CL |
427 | * We want to avoid an expensive divide : (offset / cache->size) |
428 | * Using the fact that size is a constant for a particular cache, | |
429 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
430 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
431 | */ | |
432 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 433 | const struct page *page, void *obj) |
8fea4e96 | 434 | { |
8456a648 | 435 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 436 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
437 | } |
438 | ||
6fb92430 | 439 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 440 | /* internal cache of cache description objs */ |
9b030cb8 | 441 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
442 | .batchcount = 1, |
443 | .limit = BOOT_CPUCACHE_ENTRIES, | |
444 | .shared = 1, | |
3b0efdfa | 445 | .size = sizeof(struct kmem_cache), |
b28a02de | 446 | .name = "kmem_cache", |
1da177e4 LT |
447 | }; |
448 | ||
edcad250 JK |
449 | #define BAD_ALIEN_MAGIC 0x01020304ul |
450 | ||
1871e52c | 451 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 452 | |
343e0d7a | 453 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 454 | { |
bf0dea23 | 455 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
456 | } |
457 | ||
a737b3e2 AM |
458 | /* |
459 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
460 | */ | |
fbaccacf | 461 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
2e6b3602 | 462 | unsigned long flags, size_t *left_over, unsigned int *num) |
fbaccacf | 463 | { |
fbaccacf | 464 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 465 | |
fbaccacf SR |
466 | /* |
467 | * The slab management structure can be either off the slab or | |
468 | * on it. For the latter case, the memory allocated for a | |
469 | * slab is used for: | |
470 | * | |
fbaccacf | 471 | * - @buffer_size bytes for each object |
2e6b3602 JK |
472 | * - One freelist_idx_t for each object |
473 | * | |
474 | * We don't need to consider alignment of freelist because | |
475 | * freelist will be at the end of slab page. The objects will be | |
476 | * at the correct alignment. | |
fbaccacf SR |
477 | * |
478 | * If the slab management structure is off the slab, then the | |
479 | * alignment will already be calculated into the size. Because | |
480 | * the slabs are all pages aligned, the objects will be at the | |
481 | * correct alignment when allocated. | |
482 | */ | |
483 | if (flags & CFLGS_OFF_SLAB) { | |
2e6b3602 JK |
484 | *num = slab_size / buffer_size; |
485 | *left_over = slab_size % buffer_size; | |
fbaccacf | 486 | } else { |
2e6b3602 JK |
487 | *num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
488 | *left_over = slab_size % | |
489 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 490 | } |
1da177e4 LT |
491 | } |
492 | ||
f28510d3 | 493 | #if DEBUG |
d40cee24 | 494 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 495 | |
a737b3e2 AM |
496 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
497 | char *msg) | |
1da177e4 LT |
498 | { |
499 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 500 | function, cachep->name, msg); |
1da177e4 | 501 | dump_stack(); |
373d4d09 | 502 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 503 | } |
f28510d3 | 504 | #endif |
1da177e4 | 505 | |
3395ee05 PM |
506 | /* |
507 | * By default on NUMA we use alien caches to stage the freeing of | |
508 | * objects allocated from other nodes. This causes massive memory | |
509 | * inefficiencies when using fake NUMA setup to split memory into a | |
510 | * large number of small nodes, so it can be disabled on the command | |
511 | * line | |
512 | */ | |
513 | ||
514 | static int use_alien_caches __read_mostly = 1; | |
515 | static int __init noaliencache_setup(char *s) | |
516 | { | |
517 | use_alien_caches = 0; | |
518 | return 1; | |
519 | } | |
520 | __setup("noaliencache", noaliencache_setup); | |
521 | ||
3df1cccd DR |
522 | static int __init slab_max_order_setup(char *str) |
523 | { | |
524 | get_option(&str, &slab_max_order); | |
525 | slab_max_order = slab_max_order < 0 ? 0 : | |
526 | min(slab_max_order, MAX_ORDER - 1); | |
527 | slab_max_order_set = true; | |
528 | ||
529 | return 1; | |
530 | } | |
531 | __setup("slab_max_order=", slab_max_order_setup); | |
532 | ||
8fce4d8e CL |
533 | #ifdef CONFIG_NUMA |
534 | /* | |
535 | * Special reaping functions for NUMA systems called from cache_reap(). | |
536 | * These take care of doing round robin flushing of alien caches (containing | |
537 | * objects freed on different nodes from which they were allocated) and the | |
538 | * flushing of remote pcps by calling drain_node_pages. | |
539 | */ | |
1871e52c | 540 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
541 | |
542 | static void init_reap_node(int cpu) | |
543 | { | |
544 | int node; | |
545 | ||
7d6e6d09 | 546 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 547 | if (node == MAX_NUMNODES) |
442295c9 | 548 | node = first_node(node_online_map); |
8fce4d8e | 549 | |
1871e52c | 550 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
551 | } |
552 | ||
553 | static void next_reap_node(void) | |
554 | { | |
909ea964 | 555 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 556 | |
8fce4d8e CL |
557 | node = next_node(node, node_online_map); |
558 | if (unlikely(node >= MAX_NUMNODES)) | |
559 | node = first_node(node_online_map); | |
909ea964 | 560 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
561 | } |
562 | ||
563 | #else | |
564 | #define init_reap_node(cpu) do { } while (0) | |
565 | #define next_reap_node(void) do { } while (0) | |
566 | #endif | |
567 | ||
1da177e4 LT |
568 | /* |
569 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
570 | * via the workqueue/eventd. | |
571 | * Add the CPU number into the expiration time to minimize the possibility of | |
572 | * the CPUs getting into lockstep and contending for the global cache chain | |
573 | * lock. | |
574 | */ | |
0db0628d | 575 | static void start_cpu_timer(int cpu) |
1da177e4 | 576 | { |
1871e52c | 577 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
578 | |
579 | /* | |
580 | * When this gets called from do_initcalls via cpucache_init(), | |
581 | * init_workqueues() has already run, so keventd will be setup | |
582 | * at that time. | |
583 | */ | |
52bad64d | 584 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 585 | init_reap_node(cpu); |
203b42f7 | 586 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
587 | schedule_delayed_work_on(cpu, reap_work, |
588 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
589 | } |
590 | } | |
591 | ||
1fe00d50 | 592 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 593 | { |
d5cff635 CM |
594 | /* |
595 | * The array_cache structures contain pointers to free object. | |
25985edc | 596 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
597 | * cache the pointers are not cleared and they could be counted as |
598 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
599 | * not scan such objects. | |
600 | */ | |
1fe00d50 JK |
601 | kmemleak_no_scan(ac); |
602 | if (ac) { | |
603 | ac->avail = 0; | |
604 | ac->limit = limit; | |
605 | ac->batchcount = batch; | |
606 | ac->touched = 0; | |
1da177e4 | 607 | } |
1fe00d50 JK |
608 | } |
609 | ||
610 | static struct array_cache *alloc_arraycache(int node, int entries, | |
611 | int batchcount, gfp_t gfp) | |
612 | { | |
5e804789 | 613 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
614 | struct array_cache *ac = NULL; |
615 | ||
616 | ac = kmalloc_node(memsize, gfp, node); | |
617 | init_arraycache(ac, entries, batchcount); | |
618 | return ac; | |
1da177e4 LT |
619 | } |
620 | ||
8456a648 | 621 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 622 | { |
072bb0aa MG |
623 | return PageSlabPfmemalloc(page); |
624 | } | |
625 | ||
626 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
627 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
628 | struct array_cache *ac) | |
629 | { | |
18bf8541 | 630 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 631 | struct page *page; |
072bb0aa MG |
632 | unsigned long flags; |
633 | ||
634 | if (!pfmemalloc_active) | |
635 | return; | |
636 | ||
ce8eb6c4 | 637 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
638 | list_for_each_entry(page, &n->slabs_full, lru) |
639 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
640 | goto out; |
641 | ||
8456a648 JK |
642 | list_for_each_entry(page, &n->slabs_partial, lru) |
643 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
644 | goto out; |
645 | ||
8456a648 JK |
646 | list_for_each_entry(page, &n->slabs_free, lru) |
647 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
648 | goto out; |
649 | ||
650 | pfmemalloc_active = false; | |
651 | out: | |
ce8eb6c4 | 652 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
653 | } |
654 | ||
381760ea | 655 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
656 | gfp_t flags, bool force_refill) |
657 | { | |
658 | int i; | |
659 | void *objp = ac->entry[--ac->avail]; | |
660 | ||
661 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
662 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 663 | struct kmem_cache_node *n; |
072bb0aa MG |
664 | |
665 | if (gfp_pfmemalloc_allowed(flags)) { | |
666 | clear_obj_pfmemalloc(&objp); | |
667 | return objp; | |
668 | } | |
669 | ||
670 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 671 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
672 | /* If a !PFMEMALLOC object is found, swap them */ |
673 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
674 | objp = ac->entry[i]; | |
675 | ac->entry[i] = ac->entry[ac->avail]; | |
676 | ac->entry[ac->avail] = objp; | |
677 | return objp; | |
678 | } | |
679 | } | |
680 | ||
681 | /* | |
682 | * If there are empty slabs on the slabs_free list and we are | |
683 | * being forced to refill the cache, mark this one !pfmemalloc. | |
684 | */ | |
18bf8541 | 685 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 686 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 687 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 688 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
689 | clear_obj_pfmemalloc(&objp); |
690 | recheck_pfmemalloc_active(cachep, ac); | |
691 | return objp; | |
692 | } | |
693 | ||
694 | /* No !PFMEMALLOC objects available */ | |
695 | ac->avail++; | |
696 | objp = NULL; | |
697 | } | |
698 | ||
699 | return objp; | |
700 | } | |
701 | ||
381760ea MG |
702 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
703 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
704 | { | |
705 | void *objp; | |
706 | ||
707 | if (unlikely(sk_memalloc_socks())) | |
708 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
709 | else | |
710 | objp = ac->entry[--ac->avail]; | |
711 | ||
712 | return objp; | |
713 | } | |
714 | ||
d3aec344 JK |
715 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
716 | struct array_cache *ac, void *objp) | |
072bb0aa MG |
717 | { |
718 | if (unlikely(pfmemalloc_active)) { | |
719 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 720 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
721 | if (PageSlabPfmemalloc(page)) |
722 | set_obj_pfmemalloc(&objp); | |
723 | } | |
724 | ||
381760ea MG |
725 | return objp; |
726 | } | |
727 | ||
728 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
729 | void *objp) | |
730 | { | |
731 | if (unlikely(sk_memalloc_socks())) | |
732 | objp = __ac_put_obj(cachep, ac, objp); | |
733 | ||
072bb0aa MG |
734 | ac->entry[ac->avail++] = objp; |
735 | } | |
736 | ||
3ded175a CL |
737 | /* |
738 | * Transfer objects in one arraycache to another. | |
739 | * Locking must be handled by the caller. | |
740 | * | |
741 | * Return the number of entries transferred. | |
742 | */ | |
743 | static int transfer_objects(struct array_cache *to, | |
744 | struct array_cache *from, unsigned int max) | |
745 | { | |
746 | /* Figure out how many entries to transfer */ | |
732eacc0 | 747 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
748 | |
749 | if (!nr) | |
750 | return 0; | |
751 | ||
752 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
753 | sizeof(void *) *nr); | |
754 | ||
755 | from->avail -= nr; | |
756 | to->avail += nr; | |
3ded175a CL |
757 | return nr; |
758 | } | |
759 | ||
765c4507 CL |
760 | #ifndef CONFIG_NUMA |
761 | ||
762 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 763 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 764 | |
c8522a3a JK |
765 | static inline struct alien_cache **alloc_alien_cache(int node, |
766 | int limit, gfp_t gfp) | |
765c4507 | 767 | { |
edcad250 | 768 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
769 | } |
770 | ||
c8522a3a | 771 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
772 | { |
773 | } | |
774 | ||
775 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
776 | { | |
777 | return 0; | |
778 | } | |
779 | ||
780 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
781 | gfp_t flags) | |
782 | { | |
783 | return NULL; | |
784 | } | |
785 | ||
8b98c169 | 786 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
787 | gfp_t flags, int nodeid) |
788 | { | |
789 | return NULL; | |
790 | } | |
791 | ||
4167e9b2 DR |
792 | static inline gfp_t gfp_exact_node(gfp_t flags) |
793 | { | |
794 | return flags; | |
795 | } | |
796 | ||
765c4507 CL |
797 | #else /* CONFIG_NUMA */ |
798 | ||
8b98c169 | 799 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 800 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 801 | |
c8522a3a JK |
802 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
803 | int batch, gfp_t gfp) | |
804 | { | |
5e804789 | 805 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
806 | struct alien_cache *alc = NULL; |
807 | ||
808 | alc = kmalloc_node(memsize, gfp, node); | |
809 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 810 | spin_lock_init(&alc->lock); |
c8522a3a JK |
811 | return alc; |
812 | } | |
813 | ||
814 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 815 | { |
c8522a3a | 816 | struct alien_cache **alc_ptr; |
5e804789 | 817 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
818 | int i; |
819 | ||
820 | if (limit > 1) | |
821 | limit = 12; | |
c8522a3a JK |
822 | alc_ptr = kzalloc_node(memsize, gfp, node); |
823 | if (!alc_ptr) | |
824 | return NULL; | |
825 | ||
826 | for_each_node(i) { | |
827 | if (i == node || !node_online(i)) | |
828 | continue; | |
829 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
830 | if (!alc_ptr[i]) { | |
831 | for (i--; i >= 0; i--) | |
832 | kfree(alc_ptr[i]); | |
833 | kfree(alc_ptr); | |
834 | return NULL; | |
e498be7d CL |
835 | } |
836 | } | |
c8522a3a | 837 | return alc_ptr; |
e498be7d CL |
838 | } |
839 | ||
c8522a3a | 840 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
841 | { |
842 | int i; | |
843 | ||
c8522a3a | 844 | if (!alc_ptr) |
e498be7d | 845 | return; |
e498be7d | 846 | for_each_node(i) |
c8522a3a JK |
847 | kfree(alc_ptr[i]); |
848 | kfree(alc_ptr); | |
e498be7d CL |
849 | } |
850 | ||
343e0d7a | 851 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
852 | struct array_cache *ac, int node, |
853 | struct list_head *list) | |
e498be7d | 854 | { |
18bf8541 | 855 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
856 | |
857 | if (ac->avail) { | |
ce8eb6c4 | 858 | spin_lock(&n->list_lock); |
e00946fe CL |
859 | /* |
860 | * Stuff objects into the remote nodes shared array first. | |
861 | * That way we could avoid the overhead of putting the objects | |
862 | * into the free lists and getting them back later. | |
863 | */ | |
ce8eb6c4 CL |
864 | if (n->shared) |
865 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 866 | |
833b706c | 867 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 868 | ac->avail = 0; |
ce8eb6c4 | 869 | spin_unlock(&n->list_lock); |
e498be7d CL |
870 | } |
871 | } | |
872 | ||
8fce4d8e CL |
873 | /* |
874 | * Called from cache_reap() to regularly drain alien caches round robin. | |
875 | */ | |
ce8eb6c4 | 876 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 877 | { |
909ea964 | 878 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 879 | |
ce8eb6c4 | 880 | if (n->alien) { |
c8522a3a JK |
881 | struct alien_cache *alc = n->alien[node]; |
882 | struct array_cache *ac; | |
883 | ||
884 | if (alc) { | |
885 | ac = &alc->ac; | |
49dfc304 | 886 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
887 | LIST_HEAD(list); |
888 | ||
889 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 890 | spin_unlock_irq(&alc->lock); |
833b706c | 891 | slabs_destroy(cachep, &list); |
c8522a3a | 892 | } |
8fce4d8e CL |
893 | } |
894 | } | |
895 | } | |
896 | ||
a737b3e2 | 897 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 898 | struct alien_cache **alien) |
e498be7d | 899 | { |
b28a02de | 900 | int i = 0; |
c8522a3a | 901 | struct alien_cache *alc; |
e498be7d CL |
902 | struct array_cache *ac; |
903 | unsigned long flags; | |
904 | ||
905 | for_each_online_node(i) { | |
c8522a3a JK |
906 | alc = alien[i]; |
907 | if (alc) { | |
833b706c JK |
908 | LIST_HEAD(list); |
909 | ||
c8522a3a | 910 | ac = &alc->ac; |
49dfc304 | 911 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 912 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 913 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 914 | slabs_destroy(cachep, &list); |
e498be7d CL |
915 | } |
916 | } | |
917 | } | |
729bd0b7 | 918 | |
25c4f304 JK |
919 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
920 | int node, int page_node) | |
729bd0b7 | 921 | { |
ce8eb6c4 | 922 | struct kmem_cache_node *n; |
c8522a3a JK |
923 | struct alien_cache *alien = NULL; |
924 | struct array_cache *ac; | |
97654dfa | 925 | LIST_HEAD(list); |
1ca4cb24 | 926 | |
18bf8541 | 927 | n = get_node(cachep, node); |
729bd0b7 | 928 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
929 | if (n->alien && n->alien[page_node]) { |
930 | alien = n->alien[page_node]; | |
c8522a3a | 931 | ac = &alien->ac; |
49dfc304 | 932 | spin_lock(&alien->lock); |
c8522a3a | 933 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 934 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 935 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 936 | } |
c8522a3a | 937 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 938 | spin_unlock(&alien->lock); |
833b706c | 939 | slabs_destroy(cachep, &list); |
729bd0b7 | 940 | } else { |
25c4f304 | 941 | n = get_node(cachep, page_node); |
18bf8541 | 942 | spin_lock(&n->list_lock); |
25c4f304 | 943 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 944 | spin_unlock(&n->list_lock); |
97654dfa | 945 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
946 | } |
947 | return 1; | |
948 | } | |
25c4f304 JK |
949 | |
950 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
951 | { | |
952 | int page_node = page_to_nid(virt_to_page(objp)); | |
953 | int node = numa_mem_id(); | |
954 | /* | |
955 | * Make sure we are not freeing a object from another node to the array | |
956 | * cache on this cpu. | |
957 | */ | |
958 | if (likely(node == page_node)) | |
959 | return 0; | |
960 | ||
961 | return __cache_free_alien(cachep, objp, node, page_node); | |
962 | } | |
4167e9b2 DR |
963 | |
964 | /* | |
d0164adc MG |
965 | * Construct gfp mask to allocate from a specific node but do not direct reclaim |
966 | * or warn about failures. kswapd may still wake to reclaim in the background. | |
4167e9b2 DR |
967 | */ |
968 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
969 | { | |
d0164adc | 970 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM; |
4167e9b2 | 971 | } |
e498be7d CL |
972 | #endif |
973 | ||
8f9f8d9e | 974 | /* |
6a67368c | 975 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 976 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 977 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 978 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
979 | * already in use. |
980 | * | |
18004c5d | 981 | * Must hold slab_mutex. |
8f9f8d9e | 982 | */ |
6a67368c | 983 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
984 | { |
985 | struct kmem_cache *cachep; | |
ce8eb6c4 | 986 | struct kmem_cache_node *n; |
5e804789 | 987 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 988 | |
18004c5d | 989 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 990 | /* |
5f0985bb | 991 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
992 | * begin anything. Make sure some other cpu on this |
993 | * node has not already allocated this | |
994 | */ | |
18bf8541 CL |
995 | n = get_node(cachep, node); |
996 | if (!n) { | |
ce8eb6c4 CL |
997 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
998 | if (!n) | |
8f9f8d9e | 999 | return -ENOMEM; |
ce8eb6c4 | 1000 | kmem_cache_node_init(n); |
5f0985bb JZ |
1001 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1002 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1003 | |
1004 | /* | |
5f0985bb JZ |
1005 | * The kmem_cache_nodes don't come and go as CPUs |
1006 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1007 | * protection here. |
1008 | */ | |
ce8eb6c4 | 1009 | cachep->node[node] = n; |
8f9f8d9e DR |
1010 | } |
1011 | ||
18bf8541 CL |
1012 | spin_lock_irq(&n->list_lock); |
1013 | n->free_limit = | |
8f9f8d9e DR |
1014 | (1 + nr_cpus_node(node)) * |
1015 | cachep->batchcount + cachep->num; | |
18bf8541 | 1016 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1017 | } |
1018 | return 0; | |
1019 | } | |
1020 | ||
0fa8103b WL |
1021 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1022 | struct kmem_cache_node *n) | |
1023 | { | |
1024 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1025 | } | |
1026 | ||
0db0628d | 1027 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1028 | { |
1029 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1030 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1031 | int node = cpu_to_mem(cpu); |
a70f7302 | 1032 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1033 | |
18004c5d | 1034 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1035 | struct array_cache *nc; |
1036 | struct array_cache *shared; | |
c8522a3a | 1037 | struct alien_cache **alien; |
97654dfa | 1038 | LIST_HEAD(list); |
fbf1e473 | 1039 | |
18bf8541 | 1040 | n = get_node(cachep, node); |
ce8eb6c4 | 1041 | if (!n) |
bf0dea23 | 1042 | continue; |
fbf1e473 | 1043 | |
ce8eb6c4 | 1044 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1045 | |
ce8eb6c4 CL |
1046 | /* Free limit for this kmem_cache_node */ |
1047 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1048 | |
1049 | /* cpu is dead; no one can alloc from it. */ | |
1050 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1051 | if (nc) { | |
97654dfa | 1052 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1053 | nc->avail = 0; |
1054 | } | |
fbf1e473 | 1055 | |
58463c1f | 1056 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1057 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1058 | goto free_slab; |
fbf1e473 AM |
1059 | } |
1060 | ||
ce8eb6c4 | 1061 | shared = n->shared; |
fbf1e473 AM |
1062 | if (shared) { |
1063 | free_block(cachep, shared->entry, | |
97654dfa | 1064 | shared->avail, node, &list); |
ce8eb6c4 | 1065 | n->shared = NULL; |
fbf1e473 AM |
1066 | } |
1067 | ||
ce8eb6c4 CL |
1068 | alien = n->alien; |
1069 | n->alien = NULL; | |
fbf1e473 | 1070 | |
ce8eb6c4 | 1071 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1072 | |
1073 | kfree(shared); | |
1074 | if (alien) { | |
1075 | drain_alien_cache(cachep, alien); | |
1076 | free_alien_cache(alien); | |
1077 | } | |
bf0dea23 JK |
1078 | |
1079 | free_slab: | |
97654dfa | 1080 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1081 | } |
1082 | /* | |
1083 | * In the previous loop, all the objects were freed to | |
1084 | * the respective cache's slabs, now we can go ahead and | |
1085 | * shrink each nodelist to its limit. | |
1086 | */ | |
18004c5d | 1087 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1088 | n = get_node(cachep, node); |
ce8eb6c4 | 1089 | if (!n) |
fbf1e473 | 1090 | continue; |
0fa8103b | 1091 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1092 | } |
1093 | } | |
1094 | ||
0db0628d | 1095 | static int cpuup_prepare(long cpu) |
1da177e4 | 1096 | { |
343e0d7a | 1097 | struct kmem_cache *cachep; |
ce8eb6c4 | 1098 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1099 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1100 | int err; |
1da177e4 | 1101 | |
fbf1e473 AM |
1102 | /* |
1103 | * We need to do this right in the beginning since | |
1104 | * alloc_arraycache's are going to use this list. | |
1105 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1106 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1107 | */ |
6a67368c | 1108 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1109 | if (err < 0) |
1110 | goto bad; | |
fbf1e473 AM |
1111 | |
1112 | /* | |
1113 | * Now we can go ahead with allocating the shared arrays and | |
1114 | * array caches | |
1115 | */ | |
18004c5d | 1116 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 1117 | struct array_cache *shared = NULL; |
c8522a3a | 1118 | struct alien_cache **alien = NULL; |
fbf1e473 | 1119 | |
fbf1e473 AM |
1120 | if (cachep->shared) { |
1121 | shared = alloc_arraycache(node, | |
1122 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1123 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1124 | if (!shared) |
1da177e4 | 1125 | goto bad; |
fbf1e473 AM |
1126 | } |
1127 | if (use_alien_caches) { | |
83b519e8 | 1128 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1129 | if (!alien) { |
1130 | kfree(shared); | |
fbf1e473 | 1131 | goto bad; |
12d00f6a | 1132 | } |
fbf1e473 | 1133 | } |
18bf8541 | 1134 | n = get_node(cachep, node); |
ce8eb6c4 | 1135 | BUG_ON(!n); |
fbf1e473 | 1136 | |
ce8eb6c4 CL |
1137 | spin_lock_irq(&n->list_lock); |
1138 | if (!n->shared) { | |
fbf1e473 AM |
1139 | /* |
1140 | * We are serialised from CPU_DEAD or | |
1141 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1142 | */ | |
ce8eb6c4 | 1143 | n->shared = shared; |
fbf1e473 AM |
1144 | shared = NULL; |
1145 | } | |
4484ebf1 | 1146 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1147 | if (!n->alien) { |
1148 | n->alien = alien; | |
fbf1e473 | 1149 | alien = NULL; |
1da177e4 | 1150 | } |
fbf1e473 | 1151 | #endif |
ce8eb6c4 | 1152 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1153 | kfree(shared); |
1154 | free_alien_cache(alien); | |
1155 | } | |
ce79ddc8 | 1156 | |
fbf1e473 AM |
1157 | return 0; |
1158 | bad: | |
12d00f6a | 1159 | cpuup_canceled(cpu); |
fbf1e473 AM |
1160 | return -ENOMEM; |
1161 | } | |
1162 | ||
0db0628d | 1163 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1164 | unsigned long action, void *hcpu) |
1165 | { | |
1166 | long cpu = (long)hcpu; | |
1167 | int err = 0; | |
1168 | ||
1169 | switch (action) { | |
fbf1e473 AM |
1170 | case CPU_UP_PREPARE: |
1171 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1172 | mutex_lock(&slab_mutex); |
fbf1e473 | 1173 | err = cpuup_prepare(cpu); |
18004c5d | 1174 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1175 | break; |
1176 | case CPU_ONLINE: | |
8bb78442 | 1177 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1178 | start_cpu_timer(cpu); |
1179 | break; | |
1180 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1181 | case CPU_DOWN_PREPARE: |
8bb78442 | 1182 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1183 | /* |
18004c5d | 1184 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1185 | * held so that if cache_reap() is invoked it cannot do |
1186 | * anything expensive but will only modify reap_work | |
1187 | * and reschedule the timer. | |
1188 | */ | |
afe2c511 | 1189 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1190 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1191 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1192 | break; |
1193 | case CPU_DOWN_FAILED: | |
8bb78442 | 1194 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1195 | start_cpu_timer(cpu); |
1196 | break; | |
1da177e4 | 1197 | case CPU_DEAD: |
8bb78442 | 1198 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1199 | /* |
1200 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1201 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1202 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1203 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1204 | * structure is usually allocated from kmem_cache_create() and |
1205 | * gets destroyed at kmem_cache_destroy(). | |
1206 | */ | |
183ff22b | 1207 | /* fall through */ |
8f5be20b | 1208 | #endif |
1da177e4 | 1209 | case CPU_UP_CANCELED: |
8bb78442 | 1210 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1211 | mutex_lock(&slab_mutex); |
fbf1e473 | 1212 | cpuup_canceled(cpu); |
18004c5d | 1213 | mutex_unlock(&slab_mutex); |
1da177e4 | 1214 | break; |
1da177e4 | 1215 | } |
eac40680 | 1216 | return notifier_from_errno(err); |
1da177e4 LT |
1217 | } |
1218 | ||
0db0628d | 1219 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1220 | &cpuup_callback, NULL, 0 |
1221 | }; | |
1da177e4 | 1222 | |
8f9f8d9e DR |
1223 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1224 | /* | |
1225 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1226 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1227 | * removed. | |
1228 | * | |
18004c5d | 1229 | * Must hold slab_mutex. |
8f9f8d9e | 1230 | */ |
6a67368c | 1231 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1232 | { |
1233 | struct kmem_cache *cachep; | |
1234 | int ret = 0; | |
1235 | ||
18004c5d | 1236 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1237 | struct kmem_cache_node *n; |
8f9f8d9e | 1238 | |
18bf8541 | 1239 | n = get_node(cachep, node); |
ce8eb6c4 | 1240 | if (!n) |
8f9f8d9e DR |
1241 | continue; |
1242 | ||
0fa8103b | 1243 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1244 | |
ce8eb6c4 CL |
1245 | if (!list_empty(&n->slabs_full) || |
1246 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1247 | ret = -EBUSY; |
1248 | break; | |
1249 | } | |
1250 | } | |
1251 | return ret; | |
1252 | } | |
1253 | ||
1254 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1255 | unsigned long action, void *arg) | |
1256 | { | |
1257 | struct memory_notify *mnb = arg; | |
1258 | int ret = 0; | |
1259 | int nid; | |
1260 | ||
1261 | nid = mnb->status_change_nid; | |
1262 | if (nid < 0) | |
1263 | goto out; | |
1264 | ||
1265 | switch (action) { | |
1266 | case MEM_GOING_ONLINE: | |
18004c5d | 1267 | mutex_lock(&slab_mutex); |
6a67368c | 1268 | ret = init_cache_node_node(nid); |
18004c5d | 1269 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1270 | break; |
1271 | case MEM_GOING_OFFLINE: | |
18004c5d | 1272 | mutex_lock(&slab_mutex); |
6a67368c | 1273 | ret = drain_cache_node_node(nid); |
18004c5d | 1274 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1275 | break; |
1276 | case MEM_ONLINE: | |
1277 | case MEM_OFFLINE: | |
1278 | case MEM_CANCEL_ONLINE: | |
1279 | case MEM_CANCEL_OFFLINE: | |
1280 | break; | |
1281 | } | |
1282 | out: | |
5fda1bd5 | 1283 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1284 | } |
1285 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1286 | ||
e498be7d | 1287 | /* |
ce8eb6c4 | 1288 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1289 | */ |
6744f087 | 1290 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1291 | int nodeid) |
e498be7d | 1292 | { |
6744f087 | 1293 | struct kmem_cache_node *ptr; |
e498be7d | 1294 | |
6744f087 | 1295 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1296 | BUG_ON(!ptr); |
1297 | ||
6744f087 | 1298 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1299 | /* |
1300 | * Do not assume that spinlocks can be initialized via memcpy: | |
1301 | */ | |
1302 | spin_lock_init(&ptr->list_lock); | |
1303 | ||
e498be7d | 1304 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1305 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1306 | } |
1307 | ||
556a169d | 1308 | /* |
ce8eb6c4 CL |
1309 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1310 | * size of kmem_cache_node. | |
556a169d | 1311 | */ |
ce8eb6c4 | 1312 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1313 | { |
1314 | int node; | |
1315 | ||
1316 | for_each_online_node(node) { | |
ce8eb6c4 | 1317 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1318 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1319 | REAPTIMEOUT_NODE + |
1320 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1321 | } |
1322 | } | |
1323 | ||
a737b3e2 AM |
1324 | /* |
1325 | * Initialisation. Called after the page allocator have been initialised and | |
1326 | * before smp_init(). | |
1da177e4 LT |
1327 | */ |
1328 | void __init kmem_cache_init(void) | |
1329 | { | |
e498be7d CL |
1330 | int i; |
1331 | ||
68126702 JK |
1332 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1333 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1334 | kmem_cache = &kmem_cache_boot; |
1335 | ||
b6e68bc1 | 1336 | if (num_possible_nodes() == 1) |
62918a03 SS |
1337 | use_alien_caches = 0; |
1338 | ||
3c583465 | 1339 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1340 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1341 | |
1da177e4 LT |
1342 | /* |
1343 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1344 | * page orders on machines with more than 32MB of memory if |
1345 | * not overridden on the command line. | |
1da177e4 | 1346 | */ |
3df1cccd | 1347 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1348 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1349 | |
1da177e4 LT |
1350 | /* Bootstrap is tricky, because several objects are allocated |
1351 | * from caches that do not exist yet: | |
9b030cb8 CL |
1352 | * 1) initialize the kmem_cache cache: it contains the struct |
1353 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1354 | * kmem_cache is statically allocated. | |
e498be7d | 1355 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1356 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1357 | * array at the end of the bootstrap. |
1da177e4 | 1358 | * 2) Create the first kmalloc cache. |
343e0d7a | 1359 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1360 | * An __init data area is used for the head array. |
1361 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1362 | * head arrays. | |
9b030cb8 | 1363 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1364 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1365 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1366 | * the other cache's with kmalloc allocated memory. |
1367 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1368 | */ |
1369 | ||
9b030cb8 | 1370 | /* 1) create the kmem_cache */ |
1da177e4 | 1371 | |
8da3430d | 1372 | /* |
b56efcf0 | 1373 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1374 | */ |
2f9baa9f | 1375 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1376 | offsetof(struct kmem_cache, node) + |
6744f087 | 1377 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1378 | SLAB_HWCACHE_ALIGN); |
1379 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1380 | slab_state = PARTIAL; |
1da177e4 | 1381 | |
a737b3e2 | 1382 | /* |
bf0dea23 JK |
1383 | * Initialize the caches that provide memory for the kmem_cache_node |
1384 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1385 | */ |
bf0dea23 | 1386 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1387 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1388 | slab_state = PARTIAL_NODE; |
34cc6990 | 1389 | setup_kmalloc_cache_index_table(); |
e498be7d | 1390 | |
e0a42726 IM |
1391 | slab_early_init = 0; |
1392 | ||
ce8eb6c4 | 1393 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1394 | { |
1ca4cb24 PE |
1395 | int nid; |
1396 | ||
9c09a95c | 1397 | for_each_online_node(nid) { |
ce8eb6c4 | 1398 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1399 | |
bf0dea23 | 1400 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1401 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1402 | } |
1403 | } | |
1da177e4 | 1404 | |
f97d5f63 | 1405 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1406 | } |
1407 | ||
1408 | void __init kmem_cache_init_late(void) | |
1409 | { | |
1410 | struct kmem_cache *cachep; | |
1411 | ||
97d06609 | 1412 | slab_state = UP; |
52cef189 | 1413 | |
8429db5c | 1414 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1415 | mutex_lock(&slab_mutex); |
1416 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1417 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1418 | BUG(); | |
18004c5d | 1419 | mutex_unlock(&slab_mutex); |
056c6241 | 1420 | |
97d06609 CL |
1421 | /* Done! */ |
1422 | slab_state = FULL; | |
1423 | ||
a737b3e2 AM |
1424 | /* |
1425 | * Register a cpu startup notifier callback that initializes | |
1426 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1427 | */ |
1428 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1429 | |
8f9f8d9e DR |
1430 | #ifdef CONFIG_NUMA |
1431 | /* | |
1432 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1433 | * node. |
8f9f8d9e DR |
1434 | */ |
1435 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1436 | #endif | |
1437 | ||
a737b3e2 AM |
1438 | /* |
1439 | * The reap timers are started later, with a module init call: That part | |
1440 | * of the kernel is not yet operational. | |
1da177e4 LT |
1441 | */ |
1442 | } | |
1443 | ||
1444 | static int __init cpucache_init(void) | |
1445 | { | |
1446 | int cpu; | |
1447 | ||
a737b3e2 AM |
1448 | /* |
1449 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1450 | */ |
e498be7d | 1451 | for_each_online_cpu(cpu) |
a737b3e2 | 1452 | start_cpu_timer(cpu); |
a164f896 GC |
1453 | |
1454 | /* Done! */ | |
97d06609 | 1455 | slab_state = FULL; |
1da177e4 LT |
1456 | return 0; |
1457 | } | |
1da177e4 LT |
1458 | __initcall(cpucache_init); |
1459 | ||
8bdec192 RA |
1460 | static noinline void |
1461 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1462 | { | |
9a02d699 | 1463 | #if DEBUG |
ce8eb6c4 | 1464 | struct kmem_cache_node *n; |
8456a648 | 1465 | struct page *page; |
8bdec192 RA |
1466 | unsigned long flags; |
1467 | int node; | |
9a02d699 DR |
1468 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1469 | DEFAULT_RATELIMIT_BURST); | |
1470 | ||
1471 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1472 | return; | |
8bdec192 RA |
1473 | |
1474 | printk(KERN_WARNING | |
1475 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1476 | nodeid, gfpflags); | |
1477 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1478 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1479 | |
18bf8541 | 1480 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1481 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1482 | unsigned long active_slabs = 0, num_slabs = 0; | |
1483 | ||
ce8eb6c4 | 1484 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1485 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1486 | active_objs += cachep->num; |
1487 | active_slabs++; | |
1488 | } | |
8456a648 JK |
1489 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1490 | active_objs += page->active; | |
8bdec192 RA |
1491 | active_slabs++; |
1492 | } | |
8456a648 | 1493 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1494 | num_slabs++; |
1495 | ||
ce8eb6c4 CL |
1496 | free_objects += n->free_objects; |
1497 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1498 | |
1499 | num_slabs += active_slabs; | |
1500 | num_objs = num_slabs * cachep->num; | |
1501 | printk(KERN_WARNING | |
1502 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1503 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1504 | free_objects); | |
1505 | } | |
9a02d699 | 1506 | #endif |
8bdec192 RA |
1507 | } |
1508 | ||
1da177e4 | 1509 | /* |
8a7d9b43 WSH |
1510 | * Interface to system's page allocator. No need to hold the |
1511 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1512 | * |
1513 | * If we requested dmaable memory, we will get it. Even if we | |
1514 | * did not request dmaable memory, we might get it, but that | |
1515 | * would be relatively rare and ignorable. | |
1516 | */ | |
0c3aa83e JK |
1517 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1518 | int nodeid) | |
1da177e4 LT |
1519 | { |
1520 | struct page *page; | |
e1b6aa6f | 1521 | int nr_pages; |
765c4507 | 1522 | |
a618e89f | 1523 | flags |= cachep->allocflags; |
e12ba74d MG |
1524 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1525 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1526 | |
96db800f | 1527 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1528 | if (!page) { |
9a02d699 | 1529 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1530 | return NULL; |
8bdec192 | 1531 | } |
1da177e4 | 1532 | |
f3ccb2c4 VD |
1533 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1534 | __free_pages(page, cachep->gfporder); | |
1535 | return NULL; | |
1536 | } | |
1537 | ||
b37f1dd0 | 1538 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
2f064f34 | 1539 | if (page_is_pfmemalloc(page)) |
072bb0aa MG |
1540 | pfmemalloc_active = true; |
1541 | ||
e1b6aa6f | 1542 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1543 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1544 | add_zone_page_state(page_zone(page), |
1545 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1546 | else | |
1547 | add_zone_page_state(page_zone(page), | |
1548 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 | 1549 | __SetPageSlab(page); |
2f064f34 | 1550 | if (page_is_pfmemalloc(page)) |
a57a4988 | 1551 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1552 | |
b1eeab67 VN |
1553 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1554 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1555 | ||
1556 | if (cachep->ctor) | |
1557 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1558 | else | |
1559 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1560 | } | |
c175eea4 | 1561 | |
0c3aa83e | 1562 | return page; |
1da177e4 LT |
1563 | } |
1564 | ||
1565 | /* | |
1566 | * Interface to system's page release. | |
1567 | */ | |
0c3aa83e | 1568 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1569 | { |
a57a4988 | 1570 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1571 | |
b1eeab67 | 1572 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1573 | |
972d1a7b CL |
1574 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1575 | sub_zone_page_state(page_zone(page), | |
1576 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1577 | else | |
1578 | sub_zone_page_state(page_zone(page), | |
1579 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1580 | |
a57a4988 | 1581 | BUG_ON(!PageSlab(page)); |
73293c2f | 1582 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1583 | __ClearPageSlab(page); |
8456a648 JK |
1584 | page_mapcount_reset(page); |
1585 | page->mapping = NULL; | |
1f458cbf | 1586 | |
1da177e4 LT |
1587 | if (current->reclaim_state) |
1588 | current->reclaim_state->reclaimed_slab += nr_freed; | |
f3ccb2c4 | 1589 | __free_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1590 | } |
1591 | ||
1592 | static void kmem_rcu_free(struct rcu_head *head) | |
1593 | { | |
68126702 JK |
1594 | struct kmem_cache *cachep; |
1595 | struct page *page; | |
1da177e4 | 1596 | |
68126702 JK |
1597 | page = container_of(head, struct page, rcu_head); |
1598 | cachep = page->slab_cache; | |
1599 | ||
1600 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1601 | } |
1602 | ||
1603 | #if DEBUG | |
40b44137 JK |
1604 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1605 | { | |
1606 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1607 | (cachep->size % PAGE_SIZE) == 0) | |
1608 | return true; | |
1609 | ||
1610 | return false; | |
1611 | } | |
1da177e4 LT |
1612 | |
1613 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1614 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1615 | unsigned long caller) |
1da177e4 | 1616 | { |
8c138bc0 | 1617 | int size = cachep->object_size; |
1da177e4 | 1618 | |
3dafccf2 | 1619 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1620 | |
b28a02de | 1621 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1622 | return; |
1623 | ||
b28a02de PE |
1624 | *addr++ = 0x12345678; |
1625 | *addr++ = caller; | |
1626 | *addr++ = smp_processor_id(); | |
1627 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1628 | { |
1629 | unsigned long *sptr = &caller; | |
1630 | unsigned long svalue; | |
1631 | ||
1632 | while (!kstack_end(sptr)) { | |
1633 | svalue = *sptr++; | |
1634 | if (kernel_text_address(svalue)) { | |
b28a02de | 1635 | *addr++ = svalue; |
1da177e4 LT |
1636 | size -= sizeof(unsigned long); |
1637 | if (size <= sizeof(unsigned long)) | |
1638 | break; | |
1639 | } | |
1640 | } | |
1641 | ||
1642 | } | |
b28a02de | 1643 | *addr++ = 0x87654321; |
1da177e4 | 1644 | } |
40b44137 JK |
1645 | |
1646 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1647 | int map, unsigned long caller) | |
1648 | { | |
1649 | if (!is_debug_pagealloc_cache(cachep)) | |
1650 | return; | |
1651 | ||
1652 | if (caller) | |
1653 | store_stackinfo(cachep, objp, caller); | |
1654 | ||
1655 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1656 | } | |
1657 | ||
1658 | #else | |
1659 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1660 | int map, unsigned long caller) {} | |
1661 | ||
1da177e4 LT |
1662 | #endif |
1663 | ||
343e0d7a | 1664 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1665 | { |
8c138bc0 | 1666 | int size = cachep->object_size; |
3dafccf2 | 1667 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1668 | |
1669 | memset(addr, val, size); | |
b28a02de | 1670 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1671 | } |
1672 | ||
1673 | static void dump_line(char *data, int offset, int limit) | |
1674 | { | |
1675 | int i; | |
aa83aa40 DJ |
1676 | unsigned char error = 0; |
1677 | int bad_count = 0; | |
1678 | ||
fdde6abb | 1679 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1680 | for (i = 0; i < limit; i++) { |
1681 | if (data[offset + i] != POISON_FREE) { | |
1682 | error = data[offset + i]; | |
1683 | bad_count++; | |
1684 | } | |
aa83aa40 | 1685 | } |
fdde6abb SAS |
1686 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1687 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1688 | |
1689 | if (bad_count == 1) { | |
1690 | error ^= POISON_FREE; | |
1691 | if (!(error & (error - 1))) { | |
1692 | printk(KERN_ERR "Single bit error detected. Probably " | |
1693 | "bad RAM.\n"); | |
1694 | #ifdef CONFIG_X86 | |
1695 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1696 | "test tool.\n"); | |
1697 | #else | |
1698 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1699 | #endif | |
1700 | } | |
1701 | } | |
1da177e4 LT |
1702 | } |
1703 | #endif | |
1704 | ||
1705 | #if DEBUG | |
1706 | ||
343e0d7a | 1707 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1708 | { |
1709 | int i, size; | |
1710 | char *realobj; | |
1711 | ||
1712 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1713 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1714 | *dbg_redzone1(cachep, objp), |
1715 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1716 | } |
1717 | ||
1718 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1719 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1720 | *dbg_userword(cachep, objp), | |
1721 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1722 | } |
3dafccf2 | 1723 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1724 | size = cachep->object_size; |
b28a02de | 1725 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1726 | int limit; |
1727 | limit = 16; | |
b28a02de PE |
1728 | if (i + limit > size) |
1729 | limit = size - i; | |
1da177e4 LT |
1730 | dump_line(realobj, i, limit); |
1731 | } | |
1732 | } | |
1733 | ||
343e0d7a | 1734 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1735 | { |
1736 | char *realobj; | |
1737 | int size, i; | |
1738 | int lines = 0; | |
1739 | ||
40b44137 JK |
1740 | if (is_debug_pagealloc_cache(cachep)) |
1741 | return; | |
1742 | ||
3dafccf2 | 1743 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1744 | size = cachep->object_size; |
1da177e4 | 1745 | |
b28a02de | 1746 | for (i = 0; i < size; i++) { |
1da177e4 | 1747 | char exp = POISON_FREE; |
b28a02de | 1748 | if (i == size - 1) |
1da177e4 LT |
1749 | exp = POISON_END; |
1750 | if (realobj[i] != exp) { | |
1751 | int limit; | |
1752 | /* Mismatch ! */ | |
1753 | /* Print header */ | |
1754 | if (lines == 0) { | |
b28a02de | 1755 | printk(KERN_ERR |
face37f5 DJ |
1756 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1757 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1758 | print_objinfo(cachep, objp, 0); |
1759 | } | |
1760 | /* Hexdump the affected line */ | |
b28a02de | 1761 | i = (i / 16) * 16; |
1da177e4 | 1762 | limit = 16; |
b28a02de PE |
1763 | if (i + limit > size) |
1764 | limit = size - i; | |
1da177e4 LT |
1765 | dump_line(realobj, i, limit); |
1766 | i += 16; | |
1767 | lines++; | |
1768 | /* Limit to 5 lines */ | |
1769 | if (lines > 5) | |
1770 | break; | |
1771 | } | |
1772 | } | |
1773 | if (lines != 0) { | |
1774 | /* Print some data about the neighboring objects, if they | |
1775 | * exist: | |
1776 | */ | |
8456a648 | 1777 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1778 | unsigned int objnr; |
1da177e4 | 1779 | |
8456a648 | 1780 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1781 | if (objnr) { |
8456a648 | 1782 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1783 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1784 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1785 | realobj, size); |
1da177e4 LT |
1786 | print_objinfo(cachep, objp, 2); |
1787 | } | |
b28a02de | 1788 | if (objnr + 1 < cachep->num) { |
8456a648 | 1789 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1790 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1791 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1792 | realobj, size); |
1da177e4 LT |
1793 | print_objinfo(cachep, objp, 2); |
1794 | } | |
1795 | } | |
1796 | } | |
1797 | #endif | |
1798 | ||
12dd36fa | 1799 | #if DEBUG |
8456a648 JK |
1800 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1801 | struct page *page) | |
1da177e4 | 1802 | { |
1da177e4 LT |
1803 | int i; |
1804 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1805 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1806 | |
1807 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1808 | check_poison_obj(cachep, objp); |
40b44137 | 1809 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1810 | } |
1811 | if (cachep->flags & SLAB_RED_ZONE) { | |
1812 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1813 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1814 | "was overwritten"); |
1da177e4 LT |
1815 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1816 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1817 | "was overwritten"); |
1da177e4 | 1818 | } |
1da177e4 | 1819 | } |
12dd36fa | 1820 | } |
1da177e4 | 1821 | #else |
8456a648 JK |
1822 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1823 | struct page *page) | |
12dd36fa | 1824 | { |
12dd36fa | 1825 | } |
1da177e4 LT |
1826 | #endif |
1827 | ||
911851e6 RD |
1828 | /** |
1829 | * slab_destroy - destroy and release all objects in a slab | |
1830 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1831 | * @page: page pointer being destroyed |
911851e6 | 1832 | * |
8a7d9b43 WSH |
1833 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1834 | * Before calling the slab page must have been unlinked from the cache. The | |
1835 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1836 | */ |
8456a648 | 1837 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1838 | { |
7e007355 | 1839 | void *freelist; |
12dd36fa | 1840 | |
8456a648 JK |
1841 | freelist = page->freelist; |
1842 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1843 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1844 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1845 | else | |
0c3aa83e | 1846 | kmem_freepages(cachep, page); |
68126702 JK |
1847 | |
1848 | /* | |
8456a648 | 1849 | * From now on, we don't use freelist |
68126702 JK |
1850 | * although actual page can be freed in rcu context |
1851 | */ | |
1852 | if (OFF_SLAB(cachep)) | |
8456a648 | 1853 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1854 | } |
1855 | ||
97654dfa JK |
1856 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1857 | { | |
1858 | struct page *page, *n; | |
1859 | ||
1860 | list_for_each_entry_safe(page, n, list, lru) { | |
1861 | list_del(&page->lru); | |
1862 | slab_destroy(cachep, page); | |
1863 | } | |
1864 | } | |
1865 | ||
4d268eba | 1866 | /** |
a70773dd RD |
1867 | * calculate_slab_order - calculate size (page order) of slabs |
1868 | * @cachep: pointer to the cache that is being created | |
1869 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1870 | * @flags: slab allocation flags |
1871 | * | |
1872 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1873 | * |
1874 | * This could be made much more intelligent. For now, try to avoid using | |
1875 | * high order pages for slabs. When the gfp() functions are more friendly | |
1876 | * towards high-order requests, this should be changed. | |
1877 | */ | |
a737b3e2 | 1878 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1879 | size_t size, unsigned long flags) |
4d268eba PE |
1880 | { |
1881 | size_t left_over = 0; | |
9888e6fa | 1882 | int gfporder; |
4d268eba | 1883 | |
0aa817f0 | 1884 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1885 | unsigned int num; |
1886 | size_t remainder; | |
1887 | ||
2e6b3602 | 1888 | cache_estimate(gfporder, size, flags, &remainder, &num); |
4d268eba PE |
1889 | if (!num) |
1890 | continue; | |
9888e6fa | 1891 | |
f315e3fa JK |
1892 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1893 | if (num > SLAB_OBJ_MAX_NUM) | |
1894 | break; | |
1895 | ||
b1ab41c4 | 1896 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1897 | struct kmem_cache *freelist_cache; |
1898 | size_t freelist_size; | |
1899 | ||
1900 | freelist_size = num * sizeof(freelist_idx_t); | |
1901 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1902 | if (!freelist_cache) | |
1903 | continue; | |
1904 | ||
b1ab41c4 | 1905 | /* |
3217fd9b JK |
1906 | * Needed to avoid possible looping condition |
1907 | * in cache_grow() | |
b1ab41c4 | 1908 | */ |
3217fd9b JK |
1909 | if (OFF_SLAB(freelist_cache)) |
1910 | continue; | |
b1ab41c4 | 1911 | |
3217fd9b JK |
1912 | /* check if off slab has enough benefit */ |
1913 | if (freelist_cache->size > cachep->size / 2) | |
1914 | continue; | |
b1ab41c4 | 1915 | } |
4d268eba | 1916 | |
9888e6fa | 1917 | /* Found something acceptable - save it away */ |
4d268eba | 1918 | cachep->num = num; |
9888e6fa | 1919 | cachep->gfporder = gfporder; |
4d268eba PE |
1920 | left_over = remainder; |
1921 | ||
f78bb8ad LT |
1922 | /* |
1923 | * A VFS-reclaimable slab tends to have most allocations | |
1924 | * as GFP_NOFS and we really don't want to have to be allocating | |
1925 | * higher-order pages when we are unable to shrink dcache. | |
1926 | */ | |
1927 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1928 | break; | |
1929 | ||
4d268eba PE |
1930 | /* |
1931 | * Large number of objects is good, but very large slabs are | |
1932 | * currently bad for the gfp()s. | |
1933 | */ | |
543585cc | 1934 | if (gfporder >= slab_max_order) |
4d268eba PE |
1935 | break; |
1936 | ||
9888e6fa LT |
1937 | /* |
1938 | * Acceptable internal fragmentation? | |
1939 | */ | |
a737b3e2 | 1940 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1941 | break; |
1942 | } | |
1943 | return left_over; | |
1944 | } | |
1945 | ||
bf0dea23 JK |
1946 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1947 | struct kmem_cache *cachep, int entries, int batchcount) | |
1948 | { | |
1949 | int cpu; | |
1950 | size_t size; | |
1951 | struct array_cache __percpu *cpu_cache; | |
1952 | ||
1953 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1954 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1955 | |
1956 | if (!cpu_cache) | |
1957 | return NULL; | |
1958 | ||
1959 | for_each_possible_cpu(cpu) { | |
1960 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1961 | entries, batchcount); | |
1962 | } | |
1963 | ||
1964 | return cpu_cache; | |
1965 | } | |
1966 | ||
83b519e8 | 1967 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1968 | { |
97d06609 | 1969 | if (slab_state >= FULL) |
83b519e8 | 1970 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1971 | |
bf0dea23 JK |
1972 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1973 | if (!cachep->cpu_cache) | |
1974 | return 1; | |
1975 | ||
97d06609 | 1976 | if (slab_state == DOWN) { |
bf0dea23 JK |
1977 | /* Creation of first cache (kmem_cache). */ |
1978 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1979 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1980 | /* For kmem_cache_node */ |
1981 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1982 | } else { |
bf0dea23 | 1983 | int node; |
f30cf7d1 | 1984 | |
bf0dea23 JK |
1985 | for_each_online_node(node) { |
1986 | cachep->node[node] = kmalloc_node( | |
1987 | sizeof(struct kmem_cache_node), gfp, node); | |
1988 | BUG_ON(!cachep->node[node]); | |
1989 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1990 | } |
1991 | } | |
bf0dea23 | 1992 | |
6a67368c | 1993 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1994 | jiffies + REAPTIMEOUT_NODE + |
1995 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1996 | |
1997 | cpu_cache_get(cachep)->avail = 0; | |
1998 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1999 | cpu_cache_get(cachep)->batchcount = 1; | |
2000 | cpu_cache_get(cachep)->touched = 0; | |
2001 | cachep->batchcount = 1; | |
2002 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2003 | return 0; |
f30cf7d1 PE |
2004 | } |
2005 | ||
12220dea JK |
2006 | unsigned long kmem_cache_flags(unsigned long object_size, |
2007 | unsigned long flags, const char *name, | |
2008 | void (*ctor)(void *)) | |
2009 | { | |
2010 | return flags; | |
2011 | } | |
2012 | ||
2013 | struct kmem_cache * | |
2014 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
2015 | unsigned long flags, void (*ctor)(void *)) | |
2016 | { | |
2017 | struct kmem_cache *cachep; | |
2018 | ||
2019 | cachep = find_mergeable(size, align, flags, name, ctor); | |
2020 | if (cachep) { | |
2021 | cachep->refcount++; | |
2022 | ||
2023 | /* | |
2024 | * Adjust the object sizes so that we clear | |
2025 | * the complete object on kzalloc. | |
2026 | */ | |
2027 | cachep->object_size = max_t(int, cachep->object_size, size); | |
2028 | } | |
2029 | return cachep; | |
2030 | } | |
2031 | ||
158e319b JK |
2032 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
2033 | size_t size, unsigned long flags) | |
2034 | { | |
2035 | size_t left; | |
2036 | ||
2037 | cachep->num = 0; | |
2038 | ||
2039 | /* | |
3217fd9b JK |
2040 | * Always use on-slab management when SLAB_NOLEAKTRACE |
2041 | * to avoid recursive calls into kmemleak. | |
158e319b | 2042 | */ |
158e319b JK |
2043 | if (flags & SLAB_NOLEAKTRACE) |
2044 | return false; | |
2045 | ||
2046 | /* | |
2047 | * Size is large, assume best to place the slab management obj | |
2048 | * off-slab (should allow better packing of objs). | |
2049 | */ | |
2050 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
2051 | if (!cachep->num) | |
2052 | return false; | |
2053 | ||
2054 | /* | |
2055 | * If the slab has been placed off-slab, and we have enough space then | |
2056 | * move it on-slab. This is at the expense of any extra colouring. | |
2057 | */ | |
2058 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
2059 | return false; | |
2060 | ||
2061 | cachep->colour = left / cachep->colour_off; | |
2062 | ||
2063 | return true; | |
2064 | } | |
2065 | ||
2066 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
2067 | size_t size, unsigned long flags) | |
2068 | { | |
2069 | size_t left; | |
2070 | ||
2071 | cachep->num = 0; | |
2072 | ||
2073 | left = calculate_slab_order(cachep, size, flags); | |
2074 | if (!cachep->num) | |
2075 | return false; | |
2076 | ||
2077 | cachep->colour = left / cachep->colour_off; | |
2078 | ||
2079 | return true; | |
2080 | } | |
2081 | ||
1da177e4 | 2082 | /** |
039363f3 | 2083 | * __kmem_cache_create - Create a cache. |
a755b76a | 2084 | * @cachep: cache management descriptor |
1da177e4 | 2085 | * @flags: SLAB flags |
1da177e4 LT |
2086 | * |
2087 | * Returns a ptr to the cache on success, NULL on failure. | |
2088 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2089 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2090 | * |
1da177e4 LT |
2091 | * The flags are |
2092 | * | |
2093 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2094 | * to catch references to uninitialised memory. | |
2095 | * | |
2096 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2097 | * for buffer overruns. | |
2098 | * | |
1da177e4 LT |
2099 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2100 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2101 | * as davem. | |
2102 | */ | |
278b1bb1 | 2103 | int |
8a13a4cc | 2104 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2105 | { |
d4a5fca5 | 2106 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2107 | gfp_t gfp; |
278b1bb1 | 2108 | int err; |
8a13a4cc | 2109 | size_t size = cachep->size; |
1da177e4 | 2110 | |
1da177e4 | 2111 | #if DEBUG |
1da177e4 LT |
2112 | #if FORCED_DEBUG |
2113 | /* | |
2114 | * Enable redzoning and last user accounting, except for caches with | |
2115 | * large objects, if the increased size would increase the object size | |
2116 | * above the next power of two: caches with object sizes just above a | |
2117 | * power of two have a significant amount of internal fragmentation. | |
2118 | */ | |
87a927c7 DW |
2119 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2120 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2121 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2122 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2123 | flags |= SLAB_POISON; | |
2124 | #endif | |
1da177e4 | 2125 | #endif |
1da177e4 | 2126 | |
a737b3e2 AM |
2127 | /* |
2128 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2129 | * unaligned accesses for some archs when redzoning is used, and makes |
2130 | * sure any on-slab bufctl's are also correctly aligned. | |
2131 | */ | |
b28a02de PE |
2132 | if (size & (BYTES_PER_WORD - 1)) { |
2133 | size += (BYTES_PER_WORD - 1); | |
2134 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2135 | } |
2136 | ||
87a927c7 DW |
2137 | if (flags & SLAB_RED_ZONE) { |
2138 | ralign = REDZONE_ALIGN; | |
2139 | /* If redzoning, ensure that the second redzone is suitably | |
2140 | * aligned, by adjusting the object size accordingly. */ | |
2141 | size += REDZONE_ALIGN - 1; | |
2142 | size &= ~(REDZONE_ALIGN - 1); | |
2143 | } | |
ca5f9703 | 2144 | |
a44b56d3 | 2145 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2146 | if (ralign < cachep->align) { |
2147 | ralign = cachep->align; | |
1da177e4 | 2148 | } |
3ff84a7f PE |
2149 | /* disable debug if necessary */ |
2150 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2151 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2152 | /* |
ca5f9703 | 2153 | * 4) Store it. |
1da177e4 | 2154 | */ |
8a13a4cc | 2155 | cachep->align = ralign; |
158e319b JK |
2156 | cachep->colour_off = cache_line_size(); |
2157 | /* Offset must be a multiple of the alignment. */ | |
2158 | if (cachep->colour_off < cachep->align) | |
2159 | cachep->colour_off = cachep->align; | |
1da177e4 | 2160 | |
83b519e8 PE |
2161 | if (slab_is_available()) |
2162 | gfp = GFP_KERNEL; | |
2163 | else | |
2164 | gfp = GFP_NOWAIT; | |
2165 | ||
1da177e4 | 2166 | #if DEBUG |
1da177e4 | 2167 | |
ca5f9703 PE |
2168 | /* |
2169 | * Both debugging options require word-alignment which is calculated | |
2170 | * into align above. | |
2171 | */ | |
1da177e4 | 2172 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2173 | /* add space for red zone words */ |
3ff84a7f PE |
2174 | cachep->obj_offset += sizeof(unsigned long long); |
2175 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2176 | } |
2177 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2178 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2179 | * the real object. But if the second red zone needs to be |
2180 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2181 | */ |
87a927c7 DW |
2182 | if (flags & SLAB_RED_ZONE) |
2183 | size += REDZONE_ALIGN; | |
2184 | else | |
2185 | size += BYTES_PER_WORD; | |
1da177e4 | 2186 | } |
832a15d2 JK |
2187 | #endif |
2188 | ||
2189 | size = ALIGN(size, cachep->align); | |
2190 | /* | |
2191 | * We should restrict the number of objects in a slab to implement | |
2192 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2193 | */ | |
2194 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2195 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2196 | ||
2197 | #if DEBUG | |
03a2d2a3 JK |
2198 | /* |
2199 | * To activate debug pagealloc, off-slab management is necessary | |
2200 | * requirement. In early phase of initialization, small sized slab | |
2201 | * doesn't get initialized so it would not be possible. So, we need | |
2202 | * to check size >= 256. It guarantees that all necessary small | |
2203 | * sized slab is initialized in current slab initialization sequence. | |
2204 | */ | |
40323278 | 2205 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2206 | size >= 256 && cachep->object_size > cache_line_size()) { |
2207 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2208 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2209 | ||
2210 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2211 | flags |= CFLGS_OFF_SLAB; | |
2212 | cachep->obj_offset += tmp_size - size; | |
2213 | size = tmp_size; | |
2214 | goto done; | |
2215 | } | |
2216 | } | |
1da177e4 | 2217 | } |
1da177e4 LT |
2218 | #endif |
2219 | ||
158e319b | 2220 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2221 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2222 | goto done; |
832a15d2 | 2223 | } |
1da177e4 | 2224 | |
158e319b JK |
2225 | if (set_on_slab_cache(cachep, size, flags)) |
2226 | goto done; | |
1da177e4 | 2227 | |
158e319b | 2228 | return -E2BIG; |
1da177e4 | 2229 | |
158e319b JK |
2230 | done: |
2231 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2232 | cachep->flags = flags; |
a57a4988 | 2233 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2234 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2235 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2236 | cachep->size = size; |
6a2d7a95 | 2237 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2238 | |
40b44137 JK |
2239 | #if DEBUG |
2240 | /* | |
2241 | * If we're going to use the generic kernel_map_pages() | |
2242 | * poisoning, then it's going to smash the contents of | |
2243 | * the redzone and userword anyhow, so switch them off. | |
2244 | */ | |
2245 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2246 | (cachep->flags & SLAB_POISON) && | |
2247 | is_debug_pagealloc_cache(cachep)) | |
2248 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2249 | #endif | |
2250 | ||
2251 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2252 | cachep->freelist_cache = |
2253 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2254 | } |
1da177e4 | 2255 | |
278b1bb1 CL |
2256 | err = setup_cpu_cache(cachep, gfp); |
2257 | if (err) { | |
52b4b950 | 2258 | __kmem_cache_release(cachep); |
278b1bb1 | 2259 | return err; |
2ed3a4ef | 2260 | } |
1da177e4 | 2261 | |
278b1bb1 | 2262 | return 0; |
1da177e4 | 2263 | } |
1da177e4 LT |
2264 | |
2265 | #if DEBUG | |
2266 | static void check_irq_off(void) | |
2267 | { | |
2268 | BUG_ON(!irqs_disabled()); | |
2269 | } | |
2270 | ||
2271 | static void check_irq_on(void) | |
2272 | { | |
2273 | BUG_ON(irqs_disabled()); | |
2274 | } | |
2275 | ||
343e0d7a | 2276 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2277 | { |
2278 | #ifdef CONFIG_SMP | |
2279 | check_irq_off(); | |
18bf8541 | 2280 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2281 | #endif |
2282 | } | |
e498be7d | 2283 | |
343e0d7a | 2284 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2285 | { |
2286 | #ifdef CONFIG_SMP | |
2287 | check_irq_off(); | |
18bf8541 | 2288 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2289 | #endif |
2290 | } | |
2291 | ||
1da177e4 LT |
2292 | #else |
2293 | #define check_irq_off() do { } while(0) | |
2294 | #define check_irq_on() do { } while(0) | |
2295 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2296 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2297 | #endif |
2298 | ||
ce8eb6c4 | 2299 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2300 | struct array_cache *ac, |
2301 | int force, int node); | |
2302 | ||
1da177e4 LT |
2303 | static void do_drain(void *arg) |
2304 | { | |
a737b3e2 | 2305 | struct kmem_cache *cachep = arg; |
1da177e4 | 2306 | struct array_cache *ac; |
7d6e6d09 | 2307 | int node = numa_mem_id(); |
18bf8541 | 2308 | struct kmem_cache_node *n; |
97654dfa | 2309 | LIST_HEAD(list); |
1da177e4 LT |
2310 | |
2311 | check_irq_off(); | |
9a2dba4b | 2312 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2313 | n = get_node(cachep, node); |
2314 | spin_lock(&n->list_lock); | |
97654dfa | 2315 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2316 | spin_unlock(&n->list_lock); |
97654dfa | 2317 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2318 | ac->avail = 0; |
2319 | } | |
2320 | ||
343e0d7a | 2321 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2322 | { |
ce8eb6c4 | 2323 | struct kmem_cache_node *n; |
e498be7d CL |
2324 | int node; |
2325 | ||
15c8b6c1 | 2326 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2327 | check_irq_on(); |
18bf8541 CL |
2328 | for_each_kmem_cache_node(cachep, node, n) |
2329 | if (n->alien) | |
ce8eb6c4 | 2330 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2331 | |
18bf8541 CL |
2332 | for_each_kmem_cache_node(cachep, node, n) |
2333 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2334 | } |
2335 | ||
ed11d9eb CL |
2336 | /* |
2337 | * Remove slabs from the list of free slabs. | |
2338 | * Specify the number of slabs to drain in tofree. | |
2339 | * | |
2340 | * Returns the actual number of slabs released. | |
2341 | */ | |
2342 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2343 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2344 | { |
ed11d9eb CL |
2345 | struct list_head *p; |
2346 | int nr_freed; | |
8456a648 | 2347 | struct page *page; |
1da177e4 | 2348 | |
ed11d9eb | 2349 | nr_freed = 0; |
ce8eb6c4 | 2350 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2351 | |
ce8eb6c4 CL |
2352 | spin_lock_irq(&n->list_lock); |
2353 | p = n->slabs_free.prev; | |
2354 | if (p == &n->slabs_free) { | |
2355 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2356 | goto out; |
2357 | } | |
1da177e4 | 2358 | |
8456a648 | 2359 | page = list_entry(p, struct page, lru); |
8456a648 | 2360 | list_del(&page->lru); |
ed11d9eb CL |
2361 | /* |
2362 | * Safe to drop the lock. The slab is no longer linked | |
2363 | * to the cache. | |
2364 | */ | |
ce8eb6c4 CL |
2365 | n->free_objects -= cache->num; |
2366 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2367 | slab_destroy(cache, page); |
ed11d9eb | 2368 | nr_freed++; |
1da177e4 | 2369 | } |
ed11d9eb CL |
2370 | out: |
2371 | return nr_freed; | |
1da177e4 LT |
2372 | } |
2373 | ||
d6e0b7fa | 2374 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2375 | { |
18bf8541 CL |
2376 | int ret = 0; |
2377 | int node; | |
ce8eb6c4 | 2378 | struct kmem_cache_node *n; |
e498be7d CL |
2379 | |
2380 | drain_cpu_caches(cachep); | |
2381 | ||
2382 | check_irq_on(); | |
18bf8541 | 2383 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2384 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2385 | |
ce8eb6c4 CL |
2386 | ret += !list_empty(&n->slabs_full) || |
2387 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2388 | } |
2389 | return (ret ? 1 : 0); | |
2390 | } | |
2391 | ||
945cf2b6 | 2392 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2393 | { |
2394 | return __kmem_cache_shrink(cachep, false); | |
2395 | } | |
2396 | ||
2397 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2398 | { |
12c3667f | 2399 | int i; |
ce8eb6c4 | 2400 | struct kmem_cache_node *n; |
1da177e4 | 2401 | |
bf0dea23 | 2402 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2403 | |
ce8eb6c4 | 2404 | /* NUMA: free the node structures */ |
18bf8541 CL |
2405 | for_each_kmem_cache_node(cachep, i, n) { |
2406 | kfree(n->shared); | |
2407 | free_alien_cache(n->alien); | |
2408 | kfree(n); | |
2409 | cachep->node[i] = NULL; | |
12c3667f | 2410 | } |
1da177e4 | 2411 | } |
1da177e4 | 2412 | |
e5ac9c5a RT |
2413 | /* |
2414 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2415 | * |
2416 | * For a slab cache when the slab descriptor is off-slab, the | |
2417 | * slab descriptor can't come from the same cache which is being created, | |
2418 | * Because if it is the case, that means we defer the creation of | |
2419 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2420 | * And we eventually call down to __kmem_cache_create(), which | |
2421 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2422 | * This is a "chicken-and-egg" problem. | |
2423 | * | |
2424 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2425 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2426 | */ |
7e007355 | 2427 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2428 | struct page *page, int colour_off, |
2429 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2430 | { |
7e007355 | 2431 | void *freelist; |
0c3aa83e | 2432 | void *addr = page_address(page); |
b28a02de | 2433 | |
2e6b3602 JK |
2434 | page->s_mem = addr + colour_off; |
2435 | page->active = 0; | |
2436 | ||
1da177e4 LT |
2437 | if (OFF_SLAB(cachep)) { |
2438 | /* Slab management obj is off-slab. */ | |
8456a648 | 2439 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2440 | local_flags, nodeid); |
8456a648 | 2441 | if (!freelist) |
1da177e4 LT |
2442 | return NULL; |
2443 | } else { | |
2e6b3602 JK |
2444 | /* We will use last bytes at the slab for freelist */ |
2445 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2446 | cachep->freelist_size; | |
1da177e4 | 2447 | } |
2e6b3602 | 2448 | |
8456a648 | 2449 | return freelist; |
1da177e4 LT |
2450 | } |
2451 | ||
7cc68973 | 2452 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2453 | { |
a41adfaa | 2454 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2455 | } |
2456 | ||
2457 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2458 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2459 | { |
a41adfaa | 2460 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2461 | } |
2462 | ||
343e0d7a | 2463 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2464 | struct page *page) |
1da177e4 LT |
2465 | { |
2466 | int i; | |
2467 | ||
2468 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2469 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 | 2470 | #if DEBUG |
1da177e4 LT |
2471 | if (cachep->flags & SLAB_STORE_USER) |
2472 | *dbg_userword(cachep, objp) = NULL; | |
2473 | ||
2474 | if (cachep->flags & SLAB_RED_ZONE) { | |
2475 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2476 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2477 | } | |
2478 | /* | |
a737b3e2 AM |
2479 | * Constructors are not allowed to allocate memory from the same |
2480 | * cache which they are a constructor for. Otherwise, deadlock. | |
2481 | * They must also be threaded. | |
1da177e4 LT |
2482 | */ |
2483 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2484 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2485 | |
2486 | if (cachep->flags & SLAB_RED_ZONE) { | |
2487 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2488 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2489 | " end of an object"); |
1da177e4 LT |
2490 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2491 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2492 | " start of an object"); |
1da177e4 | 2493 | } |
40b44137 JK |
2494 | /* need to poison the objs? */ |
2495 | if (cachep->flags & SLAB_POISON) { | |
2496 | poison_obj(cachep, objp, POISON_FREE); | |
2497 | slab_kernel_map(cachep, objp, 0, 0); | |
2498 | } | |
1da177e4 LT |
2499 | #else |
2500 | if (cachep->ctor) | |
51cc5068 | 2501 | cachep->ctor(objp); |
1da177e4 | 2502 | #endif |
e5c58dfd | 2503 | set_free_obj(page, i, i); |
1da177e4 | 2504 | } |
1da177e4 LT |
2505 | } |
2506 | ||
343e0d7a | 2507 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2508 | { |
4b51d669 CL |
2509 | if (CONFIG_ZONE_DMA_FLAG) { |
2510 | if (flags & GFP_DMA) | |
a618e89f | 2511 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2512 | else |
a618e89f | 2513 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2514 | } |
1da177e4 LT |
2515 | } |
2516 | ||
260b61dd | 2517 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2518 | { |
b1cb0982 | 2519 | void *objp; |
78d382d7 | 2520 | |
e5c58dfd | 2521 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2522 | page->active++; |
78d382d7 | 2523 | |
d31676df JK |
2524 | #if DEBUG |
2525 | if (cachep->flags & SLAB_STORE_USER) | |
2526 | set_store_user_dirty(cachep); | |
2527 | #endif | |
2528 | ||
78d382d7 MD |
2529 | return objp; |
2530 | } | |
2531 | ||
260b61dd JK |
2532 | static void slab_put_obj(struct kmem_cache *cachep, |
2533 | struct page *page, void *objp) | |
78d382d7 | 2534 | { |
8456a648 | 2535 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2536 | #if DEBUG |
16025177 | 2537 | unsigned int i; |
b1cb0982 | 2538 | |
b1cb0982 | 2539 | /* Verify double free bug */ |
8456a648 | 2540 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2541 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2542 | printk(KERN_ERR "slab: double free detected in cache " |
2543 | "'%s', objp %p\n", cachep->name, objp); | |
2544 | BUG(); | |
2545 | } | |
78d382d7 MD |
2546 | } |
2547 | #endif | |
8456a648 | 2548 | page->active--; |
e5c58dfd | 2549 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2550 | } |
2551 | ||
4776874f PE |
2552 | /* |
2553 | * Map pages beginning at addr to the given cache and slab. This is required | |
2554 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2555 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2556 | */ |
8456a648 | 2557 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2558 | void *freelist) |
1da177e4 | 2559 | { |
a57a4988 | 2560 | page->slab_cache = cache; |
8456a648 | 2561 | page->freelist = freelist; |
1da177e4 LT |
2562 | } |
2563 | ||
2564 | /* | |
2565 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2566 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2567 | */ | |
3c517a61 | 2568 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2569 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2570 | { |
7e007355 | 2571 | void *freelist; |
b28a02de PE |
2572 | size_t offset; |
2573 | gfp_t local_flags; | |
ce8eb6c4 | 2574 | struct kmem_cache_node *n; |
1da177e4 | 2575 | |
a737b3e2 AM |
2576 | /* |
2577 | * Be lazy and only check for valid flags here, keeping it out of the | |
2578 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2579 | */ |
c871ac4e AM |
2580 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2581 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2582 | BUG(); | |
2583 | } | |
6cb06229 | 2584 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2585 | |
ce8eb6c4 | 2586 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2587 | check_irq_off(); |
18bf8541 | 2588 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2589 | spin_lock(&n->list_lock); |
1da177e4 LT |
2590 | |
2591 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2592 | offset = n->colour_next; |
2593 | n->colour_next++; | |
2594 | if (n->colour_next >= cachep->colour) | |
2595 | n->colour_next = 0; | |
2596 | spin_unlock(&n->list_lock); | |
1da177e4 | 2597 | |
2e1217cf | 2598 | offset *= cachep->colour_off; |
1da177e4 | 2599 | |
d0164adc | 2600 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2601 | local_irq_enable(); |
2602 | ||
2603 | /* | |
2604 | * The test for missing atomic flag is performed here, rather than | |
2605 | * the more obvious place, simply to reduce the critical path length | |
2606 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2607 | * will eventually be caught here (where it matters). | |
2608 | */ | |
2609 | kmem_flagcheck(cachep, flags); | |
2610 | ||
a737b3e2 AM |
2611 | /* |
2612 | * Get mem for the objs. Attempt to allocate a physical page from | |
2613 | * 'nodeid'. | |
e498be7d | 2614 | */ |
0c3aa83e JK |
2615 | if (!page) |
2616 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2617 | if (!page) | |
1da177e4 LT |
2618 | goto failed; |
2619 | ||
2620 | /* Get slab management. */ | |
8456a648 | 2621 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2622 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2623 | if (!freelist) |
1da177e4 LT |
2624 | goto opps1; |
2625 | ||
8456a648 | 2626 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2627 | |
8456a648 | 2628 | cache_init_objs(cachep, page); |
1da177e4 | 2629 | |
d0164adc | 2630 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2631 | local_irq_disable(); |
2632 | check_irq_off(); | |
ce8eb6c4 | 2633 | spin_lock(&n->list_lock); |
1da177e4 LT |
2634 | |
2635 | /* Make slab active. */ | |
8456a648 | 2636 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2637 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2638 | n->free_objects += cachep->num; |
2639 | spin_unlock(&n->list_lock); | |
1da177e4 | 2640 | return 1; |
a737b3e2 | 2641 | opps1: |
0c3aa83e | 2642 | kmem_freepages(cachep, page); |
a737b3e2 | 2643 | failed: |
d0164adc | 2644 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2645 | local_irq_disable(); |
2646 | return 0; | |
2647 | } | |
2648 | ||
2649 | #if DEBUG | |
2650 | ||
2651 | /* | |
2652 | * Perform extra freeing checks: | |
2653 | * - detect bad pointers. | |
2654 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2655 | */ |
2656 | static void kfree_debugcheck(const void *objp) | |
2657 | { | |
1da177e4 LT |
2658 | if (!virt_addr_valid(objp)) { |
2659 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2660 | (unsigned long)objp); |
2661 | BUG(); | |
1da177e4 | 2662 | } |
1da177e4 LT |
2663 | } |
2664 | ||
58ce1fd5 PE |
2665 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2666 | { | |
b46b8f19 | 2667 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2668 | |
2669 | redzone1 = *dbg_redzone1(cache, obj); | |
2670 | redzone2 = *dbg_redzone2(cache, obj); | |
2671 | ||
2672 | /* | |
2673 | * Redzone is ok. | |
2674 | */ | |
2675 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2676 | return; | |
2677 | ||
2678 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2679 | slab_error(cache, "double free detected"); | |
2680 | else | |
2681 | slab_error(cache, "memory outside object was overwritten"); | |
2682 | ||
b46b8f19 | 2683 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2684 | obj, redzone1, redzone2); |
2685 | } | |
2686 | ||
343e0d7a | 2687 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2688 | unsigned long caller) |
1da177e4 | 2689 | { |
1da177e4 | 2690 | unsigned int objnr; |
8456a648 | 2691 | struct page *page; |
1da177e4 | 2692 | |
80cbd911 MW |
2693 | BUG_ON(virt_to_cache(objp) != cachep); |
2694 | ||
3dafccf2 | 2695 | objp -= obj_offset(cachep); |
1da177e4 | 2696 | kfree_debugcheck(objp); |
b49af68f | 2697 | page = virt_to_head_page(objp); |
1da177e4 | 2698 | |
1da177e4 | 2699 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2700 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2701 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2702 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2703 | } | |
d31676df JK |
2704 | if (cachep->flags & SLAB_STORE_USER) { |
2705 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2706 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2707 | } |
1da177e4 | 2708 | |
8456a648 | 2709 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2710 | |
2711 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2712 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2713 | |
1da177e4 | 2714 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2715 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2716 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2717 | } |
2718 | return objp; | |
2719 | } | |
2720 | ||
1da177e4 LT |
2721 | #else |
2722 | #define kfree_debugcheck(x) do { } while(0) | |
2723 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2724 | #endif |
2725 | ||
d8410234 JK |
2726 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
2727 | struct kmem_cache_node *n, struct page *page) | |
2728 | { | |
2729 | /* move slabp to correct slabp list: */ | |
2730 | list_del(&page->lru); | |
2731 | if (page->active == cachep->num) | |
2732 | list_add(&page->lru, &n->slabs_full); | |
2733 | else | |
2734 | list_add(&page->lru, &n->slabs_partial); | |
2735 | } | |
2736 | ||
7aa0d227 GT |
2737 | static struct page *get_first_slab(struct kmem_cache_node *n) |
2738 | { | |
2739 | struct page *page; | |
2740 | ||
2741 | page = list_first_entry_or_null(&n->slabs_partial, | |
2742 | struct page, lru); | |
2743 | if (!page) { | |
2744 | n->free_touched = 1; | |
2745 | page = list_first_entry_or_null(&n->slabs_free, | |
2746 | struct page, lru); | |
2747 | } | |
2748 | ||
2749 | return page; | |
2750 | } | |
2751 | ||
072bb0aa MG |
2752 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2753 | bool force_refill) | |
1da177e4 LT |
2754 | { |
2755 | int batchcount; | |
ce8eb6c4 | 2756 | struct kmem_cache_node *n; |
1da177e4 | 2757 | struct array_cache *ac; |
1ca4cb24 PE |
2758 | int node; |
2759 | ||
1da177e4 | 2760 | check_irq_off(); |
7d6e6d09 | 2761 | node = numa_mem_id(); |
072bb0aa MG |
2762 | if (unlikely(force_refill)) |
2763 | goto force_grow; | |
2764 | retry: | |
9a2dba4b | 2765 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2766 | batchcount = ac->batchcount; |
2767 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2768 | /* |
2769 | * If there was little recent activity on this cache, then | |
2770 | * perform only a partial refill. Otherwise we could generate | |
2771 | * refill bouncing. | |
1da177e4 LT |
2772 | */ |
2773 | batchcount = BATCHREFILL_LIMIT; | |
2774 | } | |
18bf8541 | 2775 | n = get_node(cachep, node); |
e498be7d | 2776 | |
ce8eb6c4 CL |
2777 | BUG_ON(ac->avail > 0 || !n); |
2778 | spin_lock(&n->list_lock); | |
1da177e4 | 2779 | |
3ded175a | 2780 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2781 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2782 | n->shared->touched = 1; | |
3ded175a | 2783 | goto alloc_done; |
44b57f1c | 2784 | } |
3ded175a | 2785 | |
1da177e4 | 2786 | while (batchcount > 0) { |
8456a648 | 2787 | struct page *page; |
1da177e4 | 2788 | /* Get slab alloc is to come from. */ |
7aa0d227 GT |
2789 | page = get_first_slab(n); |
2790 | if (!page) | |
2791 | goto must_grow; | |
1da177e4 | 2792 | |
1da177e4 | 2793 | check_spinlock_acquired(cachep); |
714b8171 PE |
2794 | |
2795 | /* | |
2796 | * The slab was either on partial or free list so | |
2797 | * there must be at least one object available for | |
2798 | * allocation. | |
2799 | */ | |
8456a648 | 2800 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2801 | |
8456a648 | 2802 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2803 | STATS_INC_ALLOCED(cachep); |
2804 | STATS_INC_ACTIVE(cachep); | |
2805 | STATS_SET_HIGH(cachep); | |
2806 | ||
260b61dd | 2807 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page)); |
1da177e4 | 2808 | } |
1da177e4 | 2809 | |
d8410234 | 2810 | fixup_slab_list(cachep, n, page); |
1da177e4 LT |
2811 | } |
2812 | ||
a737b3e2 | 2813 | must_grow: |
ce8eb6c4 | 2814 | n->free_objects -= ac->avail; |
a737b3e2 | 2815 | alloc_done: |
ce8eb6c4 | 2816 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2817 | |
2818 | if (unlikely(!ac->avail)) { | |
2819 | int x; | |
072bb0aa | 2820 | force_grow: |
4167e9b2 | 2821 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2822 | |
a737b3e2 | 2823 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2824 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2825 | node = numa_mem_id(); |
072bb0aa MG |
2826 | |
2827 | /* no objects in sight? abort */ | |
2828 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2829 | return NULL; |
2830 | ||
a737b3e2 | 2831 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2832 | goto retry; |
2833 | } | |
2834 | ac->touched = 1; | |
072bb0aa MG |
2835 | |
2836 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2837 | } |
2838 | ||
a737b3e2 AM |
2839 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2840 | gfp_t flags) | |
1da177e4 | 2841 | { |
d0164adc | 2842 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2843 | #if DEBUG |
2844 | kmem_flagcheck(cachep, flags); | |
2845 | #endif | |
2846 | } | |
2847 | ||
2848 | #if DEBUG | |
a737b3e2 | 2849 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2850 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2851 | { |
b28a02de | 2852 | if (!objp) |
1da177e4 | 2853 | return objp; |
b28a02de | 2854 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2855 | check_poison_obj(cachep, objp); |
40b44137 | 2856 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2857 | poison_obj(cachep, objp, POISON_INUSE); |
2858 | } | |
2859 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2860 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2861 | |
2862 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2863 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2864 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2865 | slab_error(cachep, "double free, or memory outside" | |
2866 | " object was overwritten"); | |
b28a02de | 2867 | printk(KERN_ERR |
b46b8f19 | 2868 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2869 | objp, *dbg_redzone1(cachep, objp), |
2870 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2871 | } |
2872 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2873 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2874 | } | |
03787301 | 2875 | |
3dafccf2 | 2876 | objp += obj_offset(cachep); |
4f104934 | 2877 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2878 | cachep->ctor(objp); |
7ea466f2 TH |
2879 | if (ARCH_SLAB_MINALIGN && |
2880 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2881 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2882 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2883 | } |
1da177e4 LT |
2884 | return objp; |
2885 | } | |
2886 | #else | |
2887 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2888 | #endif | |
2889 | ||
343e0d7a | 2890 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2891 | { |
b28a02de | 2892 | void *objp; |
1da177e4 | 2893 | struct array_cache *ac; |
072bb0aa | 2894 | bool force_refill = false; |
1da177e4 | 2895 | |
5c382300 | 2896 | check_irq_off(); |
8a8b6502 | 2897 | |
9a2dba4b | 2898 | ac = cpu_cache_get(cachep); |
1da177e4 | 2899 | if (likely(ac->avail)) { |
1da177e4 | 2900 | ac->touched = 1; |
072bb0aa MG |
2901 | objp = ac_get_obj(cachep, ac, flags, false); |
2902 | ||
ddbf2e83 | 2903 | /* |
072bb0aa MG |
2904 | * Allow for the possibility all avail objects are not allowed |
2905 | * by the current flags | |
ddbf2e83 | 2906 | */ |
072bb0aa MG |
2907 | if (objp) { |
2908 | STATS_INC_ALLOCHIT(cachep); | |
2909 | goto out; | |
2910 | } | |
2911 | force_refill = true; | |
1da177e4 | 2912 | } |
072bb0aa MG |
2913 | |
2914 | STATS_INC_ALLOCMISS(cachep); | |
2915 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2916 | /* | |
2917 | * the 'ac' may be updated by cache_alloc_refill(), | |
2918 | * and kmemleak_erase() requires its correct value. | |
2919 | */ | |
2920 | ac = cpu_cache_get(cachep); | |
2921 | ||
2922 | out: | |
d5cff635 CM |
2923 | /* |
2924 | * To avoid a false negative, if an object that is in one of the | |
2925 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2926 | * treat the array pointers as a reference to the object. | |
2927 | */ | |
f3d8b53a O |
2928 | if (objp) |
2929 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2930 | return objp; |
2931 | } | |
2932 | ||
e498be7d | 2933 | #ifdef CONFIG_NUMA |
c61afb18 | 2934 | /* |
2ad654bc | 2935 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2936 | * |
2937 | * If we are in_interrupt, then process context, including cpusets and | |
2938 | * mempolicy, may not apply and should not be used for allocation policy. | |
2939 | */ | |
2940 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2941 | { | |
2942 | int nid_alloc, nid_here; | |
2943 | ||
765c4507 | 2944 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 2945 | return NULL; |
7d6e6d09 | 2946 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 2947 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 2948 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 2949 | else if (current->mempolicy) |
2a389610 | 2950 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 2951 | if (nid_alloc != nid_here) |
8b98c169 | 2952 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
2953 | return NULL; |
2954 | } | |
2955 | ||
765c4507 CL |
2956 | /* |
2957 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 2958 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 2959 | * available node for available objects. If that fails then we |
3c517a61 CL |
2960 | * perform an allocation without specifying a node. This allows the page |
2961 | * allocator to do its reclaim / fallback magic. We then insert the | |
2962 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 2963 | */ |
8c8cc2c1 | 2964 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 2965 | { |
8c8cc2c1 PE |
2966 | struct zonelist *zonelist; |
2967 | gfp_t local_flags; | |
dd1a239f | 2968 | struct zoneref *z; |
54a6eb5c MG |
2969 | struct zone *zone; |
2970 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 2971 | void *obj = NULL; |
3c517a61 | 2972 | int nid; |
cc9a6c87 | 2973 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
2974 | |
2975 | if (flags & __GFP_THISNODE) | |
2976 | return NULL; | |
2977 | ||
6cb06229 | 2978 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 2979 | |
cc9a6c87 | 2980 | retry_cpuset: |
d26914d1 | 2981 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2982 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 2983 | |
3c517a61 CL |
2984 | retry: |
2985 | /* | |
2986 | * Look through allowed nodes for objects available | |
2987 | * from existing per node queues. | |
2988 | */ | |
54a6eb5c MG |
2989 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
2990 | nid = zone_to_nid(zone); | |
aedb0eb1 | 2991 | |
061d7074 | 2992 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
2993 | get_node(cache, nid) && |
2994 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 2995 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 2996 | gfp_exact_node(flags), nid); |
481c5346 CL |
2997 | if (obj) |
2998 | break; | |
2999 | } | |
3c517a61 CL |
3000 | } |
3001 | ||
cfce6604 | 3002 | if (!obj) { |
3c517a61 CL |
3003 | /* |
3004 | * This allocation will be performed within the constraints | |
3005 | * of the current cpuset / memory policy requirements. | |
3006 | * We may trigger various forms of reclaim on the allowed | |
3007 | * set and go into memory reserves if necessary. | |
3008 | */ | |
0c3aa83e JK |
3009 | struct page *page; |
3010 | ||
d0164adc | 3011 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3012 | local_irq_enable(); |
3013 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3014 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3015 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3016 | local_irq_disable(); |
0c3aa83e | 3017 | if (page) { |
3c517a61 CL |
3018 | /* |
3019 | * Insert into the appropriate per node queues | |
3020 | */ | |
0c3aa83e JK |
3021 | nid = page_to_nid(page); |
3022 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3023 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3024 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3025 | if (!obj) |
3026 | /* | |
3027 | * Another processor may allocate the | |
3028 | * objects in the slab since we are | |
3029 | * not holding any locks. | |
3030 | */ | |
3031 | goto retry; | |
3032 | } else { | |
b6a60451 | 3033 | /* cache_grow already freed obj */ |
3c517a61 CL |
3034 | obj = NULL; |
3035 | } | |
3036 | } | |
aedb0eb1 | 3037 | } |
cc9a6c87 | 3038 | |
d26914d1 | 3039 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3040 | goto retry_cpuset; |
765c4507 CL |
3041 | return obj; |
3042 | } | |
3043 | ||
e498be7d CL |
3044 | /* |
3045 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3046 | */ |
8b98c169 | 3047 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3048 | int nodeid) |
e498be7d | 3049 | { |
8456a648 | 3050 | struct page *page; |
ce8eb6c4 | 3051 | struct kmem_cache_node *n; |
b28a02de | 3052 | void *obj; |
b28a02de PE |
3053 | int x; |
3054 | ||
7c3fbbdd | 3055 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3056 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3057 | BUG_ON(!n); |
b28a02de | 3058 | |
a737b3e2 | 3059 | retry: |
ca3b9b91 | 3060 | check_irq_off(); |
ce8eb6c4 | 3061 | spin_lock(&n->list_lock); |
7aa0d227 GT |
3062 | page = get_first_slab(n); |
3063 | if (!page) | |
3064 | goto must_grow; | |
b28a02de | 3065 | |
b28a02de | 3066 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3067 | |
3068 | STATS_INC_NODEALLOCS(cachep); | |
3069 | STATS_INC_ACTIVE(cachep); | |
3070 | STATS_SET_HIGH(cachep); | |
3071 | ||
8456a648 | 3072 | BUG_ON(page->active == cachep->num); |
b28a02de | 3073 | |
260b61dd | 3074 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3075 | n->free_objects--; |
b28a02de | 3076 | |
d8410234 | 3077 | fixup_slab_list(cachep, n, page); |
e498be7d | 3078 | |
ce8eb6c4 | 3079 | spin_unlock(&n->list_lock); |
b28a02de | 3080 | goto done; |
e498be7d | 3081 | |
a737b3e2 | 3082 | must_grow: |
ce8eb6c4 | 3083 | spin_unlock(&n->list_lock); |
4167e9b2 | 3084 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3085 | if (x) |
3086 | goto retry; | |
1da177e4 | 3087 | |
8c8cc2c1 | 3088 | return fallback_alloc(cachep, flags); |
e498be7d | 3089 | |
a737b3e2 | 3090 | done: |
b28a02de | 3091 | return obj; |
e498be7d | 3092 | } |
8c8cc2c1 | 3093 | |
8c8cc2c1 | 3094 | static __always_inline void * |
48356303 | 3095 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3096 | unsigned long caller) |
8c8cc2c1 PE |
3097 | { |
3098 | unsigned long save_flags; | |
3099 | void *ptr; | |
7d6e6d09 | 3100 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3101 | |
dcce284a | 3102 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3103 | cachep = slab_pre_alloc_hook(cachep, flags); |
3104 | if (unlikely(!cachep)) | |
824ebef1 AM |
3105 | return NULL; |
3106 | ||
8c8cc2c1 PE |
3107 | cache_alloc_debugcheck_before(cachep, flags); |
3108 | local_irq_save(save_flags); | |
3109 | ||
eacbbae3 | 3110 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3111 | nodeid = slab_node; |
8c8cc2c1 | 3112 | |
18bf8541 | 3113 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3114 | /* Node not bootstrapped yet */ |
3115 | ptr = fallback_alloc(cachep, flags); | |
3116 | goto out; | |
3117 | } | |
3118 | ||
7d6e6d09 | 3119 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3120 | /* |
3121 | * Use the locally cached objects if possible. | |
3122 | * However ____cache_alloc does not allow fallback | |
3123 | * to other nodes. It may fail while we still have | |
3124 | * objects on other nodes available. | |
3125 | */ | |
3126 | ptr = ____cache_alloc(cachep, flags); | |
3127 | if (ptr) | |
3128 | goto out; | |
3129 | } | |
3130 | /* ___cache_alloc_node can fall back to other nodes */ | |
3131 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3132 | out: | |
3133 | local_irq_restore(save_flags); | |
3134 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3135 | ||
d5e3ed66 JDB |
3136 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3137 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3138 | |
d5e3ed66 | 3139 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3140 | return ptr; |
3141 | } | |
3142 | ||
3143 | static __always_inline void * | |
3144 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3145 | { | |
3146 | void *objp; | |
3147 | ||
2ad654bc | 3148 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3149 | objp = alternate_node_alloc(cache, flags); |
3150 | if (objp) | |
3151 | goto out; | |
3152 | } | |
3153 | objp = ____cache_alloc(cache, flags); | |
3154 | ||
3155 | /* | |
3156 | * We may just have run out of memory on the local node. | |
3157 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3158 | */ | |
7d6e6d09 LS |
3159 | if (!objp) |
3160 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3161 | |
3162 | out: | |
3163 | return objp; | |
3164 | } | |
3165 | #else | |
3166 | ||
3167 | static __always_inline void * | |
3168 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3169 | { | |
3170 | return ____cache_alloc(cachep, flags); | |
3171 | } | |
3172 | ||
3173 | #endif /* CONFIG_NUMA */ | |
3174 | ||
3175 | static __always_inline void * | |
48356303 | 3176 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3177 | { |
3178 | unsigned long save_flags; | |
3179 | void *objp; | |
3180 | ||
dcce284a | 3181 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3182 | cachep = slab_pre_alloc_hook(cachep, flags); |
3183 | if (unlikely(!cachep)) | |
824ebef1 AM |
3184 | return NULL; |
3185 | ||
8c8cc2c1 PE |
3186 | cache_alloc_debugcheck_before(cachep, flags); |
3187 | local_irq_save(save_flags); | |
3188 | objp = __do_cache_alloc(cachep, flags); | |
3189 | local_irq_restore(save_flags); | |
3190 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3191 | prefetchw(objp); | |
3192 | ||
d5e3ed66 JDB |
3193 | if (unlikely(flags & __GFP_ZERO) && objp) |
3194 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3195 | |
d5e3ed66 | 3196 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3197 | return objp; |
3198 | } | |
e498be7d CL |
3199 | |
3200 | /* | |
5f0985bb | 3201 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3202 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3203 | */ |
97654dfa JK |
3204 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3205 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3206 | { |
3207 | int i; | |
25c063fb | 3208 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3209 | |
3210 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3211 | void *objp; |
8456a648 | 3212 | struct page *page; |
1da177e4 | 3213 | |
072bb0aa MG |
3214 | clear_obj_pfmemalloc(&objpp[i]); |
3215 | objp = objpp[i]; | |
3216 | ||
8456a648 | 3217 | page = virt_to_head_page(objp); |
8456a648 | 3218 | list_del(&page->lru); |
ff69416e | 3219 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3220 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3221 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3222 | n->free_objects++; |
1da177e4 LT |
3223 | |
3224 | /* fixup slab chains */ | |
8456a648 | 3225 | if (page->active == 0) { |
ce8eb6c4 CL |
3226 | if (n->free_objects > n->free_limit) { |
3227 | n->free_objects -= cachep->num; | |
97654dfa | 3228 | list_add_tail(&page->lru, list); |
1da177e4 | 3229 | } else { |
8456a648 | 3230 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3231 | } |
3232 | } else { | |
3233 | /* Unconditionally move a slab to the end of the | |
3234 | * partial list on free - maximum time for the | |
3235 | * other objects to be freed, too. | |
3236 | */ | |
8456a648 | 3237 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3238 | } |
3239 | } | |
3240 | } | |
3241 | ||
343e0d7a | 3242 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3243 | { |
3244 | int batchcount; | |
ce8eb6c4 | 3245 | struct kmem_cache_node *n; |
7d6e6d09 | 3246 | int node = numa_mem_id(); |
97654dfa | 3247 | LIST_HEAD(list); |
1da177e4 LT |
3248 | |
3249 | batchcount = ac->batchcount; | |
260b61dd | 3250 | |
1da177e4 | 3251 | check_irq_off(); |
18bf8541 | 3252 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3253 | spin_lock(&n->list_lock); |
3254 | if (n->shared) { | |
3255 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3256 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3257 | if (max) { |
3258 | if (batchcount > max) | |
3259 | batchcount = max; | |
e498be7d | 3260 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3261 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3262 | shared_array->avail += batchcount; |
3263 | goto free_done; | |
3264 | } | |
3265 | } | |
3266 | ||
97654dfa | 3267 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3268 | free_done: |
1da177e4 LT |
3269 | #if STATS |
3270 | { | |
3271 | int i = 0; | |
73c0219d | 3272 | struct page *page; |
1da177e4 | 3273 | |
73c0219d | 3274 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3275 | BUG_ON(page->active); |
1da177e4 LT |
3276 | |
3277 | i++; | |
1da177e4 LT |
3278 | } |
3279 | STATS_SET_FREEABLE(cachep, i); | |
3280 | } | |
3281 | #endif | |
ce8eb6c4 | 3282 | spin_unlock(&n->list_lock); |
97654dfa | 3283 | slabs_destroy(cachep, &list); |
1da177e4 | 3284 | ac->avail -= batchcount; |
a737b3e2 | 3285 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3286 | } |
3287 | ||
3288 | /* | |
a737b3e2 AM |
3289 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3290 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3291 | */ |
a947eb95 | 3292 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3293 | unsigned long caller) |
1da177e4 | 3294 | { |
9a2dba4b | 3295 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3296 | |
3297 | check_irq_off(); | |
d5cff635 | 3298 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3299 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3300 | |
8c138bc0 | 3301 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3302 | |
1807a1aa SS |
3303 | /* |
3304 | * Skip calling cache_free_alien() when the platform is not numa. | |
3305 | * This will avoid cache misses that happen while accessing slabp (which | |
3306 | * is per page memory reference) to get nodeid. Instead use a global | |
3307 | * variable to skip the call, which is mostly likely to be present in | |
3308 | * the cache. | |
3309 | */ | |
b6e68bc1 | 3310 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3311 | return; |
3312 | ||
3d880194 | 3313 | if (ac->avail < ac->limit) { |
1da177e4 | 3314 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3315 | } else { |
3316 | STATS_INC_FREEMISS(cachep); | |
3317 | cache_flusharray(cachep, ac); | |
1da177e4 | 3318 | } |
42c8c99c | 3319 | |
072bb0aa | 3320 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3321 | } |
3322 | ||
3323 | /** | |
3324 | * kmem_cache_alloc - Allocate an object | |
3325 | * @cachep: The cache to allocate from. | |
3326 | * @flags: See kmalloc(). | |
3327 | * | |
3328 | * Allocate an object from this cache. The flags are only relevant | |
3329 | * if the cache has no available objects. | |
3330 | */ | |
343e0d7a | 3331 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3332 | { |
48356303 | 3333 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3334 | |
ca2b84cb | 3335 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3336 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3337 | |
3338 | return ret; | |
1da177e4 LT |
3339 | } |
3340 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3341 | ||
7b0501dd JDB |
3342 | static __always_inline void |
3343 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3344 | size_t size, void **p, unsigned long caller) | |
3345 | { | |
3346 | size_t i; | |
3347 | ||
3348 | for (i = 0; i < size; i++) | |
3349 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3350 | } | |
3351 | ||
865762a8 | 3352 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3353 | void **p) |
484748f0 | 3354 | { |
2a777eac JDB |
3355 | size_t i; |
3356 | ||
3357 | s = slab_pre_alloc_hook(s, flags); | |
3358 | if (!s) | |
3359 | return 0; | |
3360 | ||
3361 | cache_alloc_debugcheck_before(s, flags); | |
3362 | ||
3363 | local_irq_disable(); | |
3364 | for (i = 0; i < size; i++) { | |
3365 | void *objp = __do_cache_alloc(s, flags); | |
3366 | ||
2a777eac JDB |
3367 | if (unlikely(!objp)) |
3368 | goto error; | |
3369 | p[i] = objp; | |
3370 | } | |
3371 | local_irq_enable(); | |
3372 | ||
7b0501dd JDB |
3373 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3374 | ||
2a777eac JDB |
3375 | /* Clear memory outside IRQ disabled section */ |
3376 | if (unlikely(flags & __GFP_ZERO)) | |
3377 | for (i = 0; i < size; i++) | |
3378 | memset(p[i], 0, s->object_size); | |
3379 | ||
3380 | slab_post_alloc_hook(s, flags, size, p); | |
3381 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3382 | return size; | |
3383 | error: | |
3384 | local_irq_enable(); | |
7b0501dd | 3385 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3386 | slab_post_alloc_hook(s, flags, i, p); |
3387 | __kmem_cache_free_bulk(s, i, p); | |
3388 | return 0; | |
484748f0 CL |
3389 | } |
3390 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3391 | ||
0f24f128 | 3392 | #ifdef CONFIG_TRACING |
85beb586 | 3393 | void * |
4052147c | 3394 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3395 | { |
85beb586 SR |
3396 | void *ret; |
3397 | ||
48356303 | 3398 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3399 | |
3400 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3401 | size, cachep->size, flags); |
85beb586 | 3402 | return ret; |
36555751 | 3403 | } |
85beb586 | 3404 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3405 | #endif |
3406 | ||
1da177e4 | 3407 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3408 | /** |
3409 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3410 | * @cachep: The cache to allocate from. | |
3411 | * @flags: See kmalloc(). | |
3412 | * @nodeid: node number of the target node. | |
3413 | * | |
3414 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3415 | * node, which can improve the performance for cpu bound structures. | |
3416 | * | |
3417 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3418 | */ | |
8b98c169 CH |
3419 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3420 | { | |
48356303 | 3421 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3422 | |
ca2b84cb | 3423 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3424 | cachep->object_size, cachep->size, |
ca2b84cb | 3425 | flags, nodeid); |
36555751 EGM |
3426 | |
3427 | return ret; | |
8b98c169 | 3428 | } |
1da177e4 LT |
3429 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3430 | ||
0f24f128 | 3431 | #ifdef CONFIG_TRACING |
4052147c | 3432 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3433 | gfp_t flags, |
4052147c EG |
3434 | int nodeid, |
3435 | size_t size) | |
36555751 | 3436 | { |
85beb586 SR |
3437 | void *ret; |
3438 | ||
592f4145 | 3439 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3440 | |
85beb586 | 3441 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3442 | size, cachep->size, |
85beb586 SR |
3443 | flags, nodeid); |
3444 | return ret; | |
36555751 | 3445 | } |
85beb586 | 3446 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3447 | #endif |
3448 | ||
8b98c169 | 3449 | static __always_inline void * |
7c0cb9c6 | 3450 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3451 | { |
343e0d7a | 3452 | struct kmem_cache *cachep; |
97e2bde4 | 3453 | |
2c59dd65 | 3454 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3455 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3456 | return cachep; | |
4052147c | 3457 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3458 | } |
8b98c169 | 3459 | |
8b98c169 CH |
3460 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3461 | { | |
7c0cb9c6 | 3462 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3463 | } |
dbe5e69d | 3464 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3465 | |
3466 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3467 | int node, unsigned long caller) |
8b98c169 | 3468 | { |
7c0cb9c6 | 3469 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3470 | } |
3471 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3472 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3473 | |
3474 | /** | |
800590f5 | 3475 | * __do_kmalloc - allocate memory |
1da177e4 | 3476 | * @size: how many bytes of memory are required. |
800590f5 | 3477 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3478 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3479 | */ |
7fd6b141 | 3480 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3481 | unsigned long caller) |
1da177e4 | 3482 | { |
343e0d7a | 3483 | struct kmem_cache *cachep; |
36555751 | 3484 | void *ret; |
1da177e4 | 3485 | |
2c59dd65 | 3486 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3487 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3488 | return cachep; | |
48356303 | 3489 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3490 | |
7c0cb9c6 | 3491 | trace_kmalloc(caller, ret, |
3b0efdfa | 3492 | size, cachep->size, flags); |
36555751 EGM |
3493 | |
3494 | return ret; | |
7fd6b141 PE |
3495 | } |
3496 | ||
7fd6b141 PE |
3497 | void *__kmalloc(size_t size, gfp_t flags) |
3498 | { | |
7c0cb9c6 | 3499 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3500 | } |
3501 | EXPORT_SYMBOL(__kmalloc); | |
3502 | ||
ce71e27c | 3503 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3504 | { |
7c0cb9c6 | 3505 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3506 | } |
3507 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3508 | |
1da177e4 LT |
3509 | /** |
3510 | * kmem_cache_free - Deallocate an object | |
3511 | * @cachep: The cache the allocation was from. | |
3512 | * @objp: The previously allocated object. | |
3513 | * | |
3514 | * Free an object which was previously allocated from this | |
3515 | * cache. | |
3516 | */ | |
343e0d7a | 3517 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3518 | { |
3519 | unsigned long flags; | |
b9ce5ef4 GC |
3520 | cachep = cache_from_obj(cachep, objp); |
3521 | if (!cachep) | |
3522 | return; | |
1da177e4 LT |
3523 | |
3524 | local_irq_save(flags); | |
d97d476b | 3525 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3526 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3527 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3528 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3529 | local_irq_restore(flags); |
36555751 | 3530 | |
ca2b84cb | 3531 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3532 | } |
3533 | EXPORT_SYMBOL(kmem_cache_free); | |
3534 | ||
e6cdb58d JDB |
3535 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3536 | { | |
3537 | struct kmem_cache *s; | |
3538 | size_t i; | |
3539 | ||
3540 | local_irq_disable(); | |
3541 | for (i = 0; i < size; i++) { | |
3542 | void *objp = p[i]; | |
3543 | ||
ca257195 JDB |
3544 | if (!orig_s) /* called via kfree_bulk */ |
3545 | s = virt_to_cache(objp); | |
3546 | else | |
3547 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3548 | |
3549 | debug_check_no_locks_freed(objp, s->object_size); | |
3550 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3551 | debug_check_no_obj_freed(objp, s->object_size); | |
3552 | ||
3553 | __cache_free(s, objp, _RET_IP_); | |
3554 | } | |
3555 | local_irq_enable(); | |
3556 | ||
3557 | /* FIXME: add tracing */ | |
3558 | } | |
3559 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3560 | ||
1da177e4 LT |
3561 | /** |
3562 | * kfree - free previously allocated memory | |
3563 | * @objp: pointer returned by kmalloc. | |
3564 | * | |
80e93eff PE |
3565 | * If @objp is NULL, no operation is performed. |
3566 | * | |
1da177e4 LT |
3567 | * Don't free memory not originally allocated by kmalloc() |
3568 | * or you will run into trouble. | |
3569 | */ | |
3570 | void kfree(const void *objp) | |
3571 | { | |
343e0d7a | 3572 | struct kmem_cache *c; |
1da177e4 LT |
3573 | unsigned long flags; |
3574 | ||
2121db74 PE |
3575 | trace_kfree(_RET_IP_, objp); |
3576 | ||
6cb8f913 | 3577 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3578 | return; |
3579 | local_irq_save(flags); | |
3580 | kfree_debugcheck(objp); | |
6ed5eb22 | 3581 | c = virt_to_cache(objp); |
8c138bc0 CL |
3582 | debug_check_no_locks_freed(objp, c->object_size); |
3583 | ||
3584 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3585 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3586 | local_irq_restore(flags); |
3587 | } | |
3588 | EXPORT_SYMBOL(kfree); | |
3589 | ||
e498be7d | 3590 | /* |
ce8eb6c4 | 3591 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3592 | */ |
5f0985bb | 3593 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3594 | { |
3595 | int node; | |
ce8eb6c4 | 3596 | struct kmem_cache_node *n; |
cafeb02e | 3597 | struct array_cache *new_shared; |
c8522a3a | 3598 | struct alien_cache **new_alien = NULL; |
e498be7d | 3599 | |
9c09a95c | 3600 | for_each_online_node(node) { |
cafeb02e | 3601 | |
b455def2 L |
3602 | if (use_alien_caches) { |
3603 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3604 | if (!new_alien) | |
3605 | goto fail; | |
3606 | } | |
cafeb02e | 3607 | |
63109846 ED |
3608 | new_shared = NULL; |
3609 | if (cachep->shared) { | |
3610 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3611 | cachep->shared*cachep->batchcount, |
83b519e8 | 3612 | 0xbaadf00d, gfp); |
63109846 ED |
3613 | if (!new_shared) { |
3614 | free_alien_cache(new_alien); | |
3615 | goto fail; | |
3616 | } | |
0718dc2a | 3617 | } |
cafeb02e | 3618 | |
18bf8541 | 3619 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3620 | if (n) { |
3621 | struct array_cache *shared = n->shared; | |
97654dfa | 3622 | LIST_HEAD(list); |
cafeb02e | 3623 | |
ce8eb6c4 | 3624 | spin_lock_irq(&n->list_lock); |
e498be7d | 3625 | |
cafeb02e | 3626 | if (shared) |
0718dc2a | 3627 | free_block(cachep, shared->entry, |
97654dfa | 3628 | shared->avail, node, &list); |
e498be7d | 3629 | |
ce8eb6c4 CL |
3630 | n->shared = new_shared; |
3631 | if (!n->alien) { | |
3632 | n->alien = new_alien; | |
e498be7d CL |
3633 | new_alien = NULL; |
3634 | } | |
ce8eb6c4 | 3635 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3636 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3637 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3638 | slabs_destroy(cachep, &list); |
cafeb02e | 3639 | kfree(shared); |
e498be7d CL |
3640 | free_alien_cache(new_alien); |
3641 | continue; | |
3642 | } | |
ce8eb6c4 CL |
3643 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3644 | if (!n) { | |
0718dc2a CL |
3645 | free_alien_cache(new_alien); |
3646 | kfree(new_shared); | |
e498be7d | 3647 | goto fail; |
0718dc2a | 3648 | } |
e498be7d | 3649 | |
ce8eb6c4 | 3650 | kmem_cache_node_init(n); |
5f0985bb JZ |
3651 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3652 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3653 | n->shared = new_shared; |
3654 | n->alien = new_alien; | |
3655 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3656 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3657 | cachep->node[node] = n; |
e498be7d | 3658 | } |
cafeb02e | 3659 | return 0; |
0718dc2a | 3660 | |
a737b3e2 | 3661 | fail: |
3b0efdfa | 3662 | if (!cachep->list.next) { |
0718dc2a CL |
3663 | /* Cache is not active yet. Roll back what we did */ |
3664 | node--; | |
3665 | while (node >= 0) { | |
18bf8541 CL |
3666 | n = get_node(cachep, node); |
3667 | if (n) { | |
ce8eb6c4 CL |
3668 | kfree(n->shared); |
3669 | free_alien_cache(n->alien); | |
3670 | kfree(n); | |
6a67368c | 3671 | cachep->node[node] = NULL; |
0718dc2a CL |
3672 | } |
3673 | node--; | |
3674 | } | |
3675 | } | |
cafeb02e | 3676 | return -ENOMEM; |
e498be7d CL |
3677 | } |
3678 | ||
18004c5d | 3679 | /* Always called with the slab_mutex held */ |
943a451a | 3680 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3681 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3682 | { |
bf0dea23 JK |
3683 | struct array_cache __percpu *cpu_cache, *prev; |
3684 | int cpu; | |
1da177e4 | 3685 | |
bf0dea23 JK |
3686 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3687 | if (!cpu_cache) | |
d2e7b7d0 SS |
3688 | return -ENOMEM; |
3689 | ||
bf0dea23 JK |
3690 | prev = cachep->cpu_cache; |
3691 | cachep->cpu_cache = cpu_cache; | |
3692 | kick_all_cpus_sync(); | |
e498be7d | 3693 | |
1da177e4 | 3694 | check_irq_on(); |
1da177e4 LT |
3695 | cachep->batchcount = batchcount; |
3696 | cachep->limit = limit; | |
e498be7d | 3697 | cachep->shared = shared; |
1da177e4 | 3698 | |
bf0dea23 JK |
3699 | if (!prev) |
3700 | goto alloc_node; | |
3701 | ||
3702 | for_each_online_cpu(cpu) { | |
97654dfa | 3703 | LIST_HEAD(list); |
18bf8541 CL |
3704 | int node; |
3705 | struct kmem_cache_node *n; | |
bf0dea23 | 3706 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3707 | |
bf0dea23 | 3708 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3709 | n = get_node(cachep, node); |
3710 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3711 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3712 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3713 | slabs_destroy(cachep, &list); |
1da177e4 | 3714 | } |
bf0dea23 JK |
3715 | free_percpu(prev); |
3716 | ||
3717 | alloc_node: | |
5f0985bb | 3718 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3719 | } |
3720 | ||
943a451a GC |
3721 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3722 | int batchcount, int shared, gfp_t gfp) | |
3723 | { | |
3724 | int ret; | |
426589f5 | 3725 | struct kmem_cache *c; |
943a451a GC |
3726 | |
3727 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3728 | ||
3729 | if (slab_state < FULL) | |
3730 | return ret; | |
3731 | ||
3732 | if ((ret < 0) || !is_root_cache(cachep)) | |
3733 | return ret; | |
3734 | ||
426589f5 VD |
3735 | lockdep_assert_held(&slab_mutex); |
3736 | for_each_memcg_cache(c, cachep) { | |
3737 | /* return value determined by the root cache only */ | |
3738 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3739 | } |
3740 | ||
3741 | return ret; | |
3742 | } | |
3743 | ||
18004c5d | 3744 | /* Called with slab_mutex held always */ |
83b519e8 | 3745 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3746 | { |
3747 | int err; | |
943a451a GC |
3748 | int limit = 0; |
3749 | int shared = 0; | |
3750 | int batchcount = 0; | |
3751 | ||
3752 | if (!is_root_cache(cachep)) { | |
3753 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3754 | limit = root->limit; | |
3755 | shared = root->shared; | |
3756 | batchcount = root->batchcount; | |
3757 | } | |
1da177e4 | 3758 | |
943a451a GC |
3759 | if (limit && shared && batchcount) |
3760 | goto skip_setup; | |
a737b3e2 AM |
3761 | /* |
3762 | * The head array serves three purposes: | |
1da177e4 LT |
3763 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3764 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3765 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3766 | * bufctl chains: array operations are cheaper. |
3767 | * The numbers are guessed, we should auto-tune as described by | |
3768 | * Bonwick. | |
3769 | */ | |
3b0efdfa | 3770 | if (cachep->size > 131072) |
1da177e4 | 3771 | limit = 1; |
3b0efdfa | 3772 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3773 | limit = 8; |
3b0efdfa | 3774 | else if (cachep->size > 1024) |
1da177e4 | 3775 | limit = 24; |
3b0efdfa | 3776 | else if (cachep->size > 256) |
1da177e4 LT |
3777 | limit = 54; |
3778 | else | |
3779 | limit = 120; | |
3780 | ||
a737b3e2 AM |
3781 | /* |
3782 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3783 | * allocation behaviour: Most allocs on one cpu, most free operations |
3784 | * on another cpu. For these cases, an efficient object passing between | |
3785 | * cpus is necessary. This is provided by a shared array. The array | |
3786 | * replaces Bonwick's magazine layer. | |
3787 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3788 | * to a larger limit. Thus disabled by default. | |
3789 | */ | |
3790 | shared = 0; | |
3b0efdfa | 3791 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3792 | shared = 8; |
1da177e4 LT |
3793 | |
3794 | #if DEBUG | |
a737b3e2 AM |
3795 | /* |
3796 | * With debugging enabled, large batchcount lead to excessively long | |
3797 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3798 | */ |
3799 | if (limit > 32) | |
3800 | limit = 32; | |
3801 | #endif | |
943a451a GC |
3802 | batchcount = (limit + 1) / 2; |
3803 | skip_setup: | |
3804 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3805 | if (err) |
3806 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3807 | cachep->name, -err); |
2ed3a4ef | 3808 | return err; |
1da177e4 LT |
3809 | } |
3810 | ||
1b55253a | 3811 | /* |
ce8eb6c4 CL |
3812 | * Drain an array if it contains any elements taking the node lock only if |
3813 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3814 | * if drain_array() is used on the shared array. |
1b55253a | 3815 | */ |
ce8eb6c4 | 3816 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3817 | struct array_cache *ac, int force, int node) |
1da177e4 | 3818 | { |
97654dfa | 3819 | LIST_HEAD(list); |
1da177e4 LT |
3820 | int tofree; |
3821 | ||
1b55253a CL |
3822 | if (!ac || !ac->avail) |
3823 | return; | |
1da177e4 LT |
3824 | if (ac->touched && !force) { |
3825 | ac->touched = 0; | |
b18e7e65 | 3826 | } else { |
ce8eb6c4 | 3827 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3828 | if (ac->avail) { |
3829 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3830 | if (tofree > ac->avail) | |
3831 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3832 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3833 | ac->avail -= tofree; |
3834 | memmove(ac->entry, &(ac->entry[tofree]), | |
3835 | sizeof(void *) * ac->avail); | |
3836 | } | |
ce8eb6c4 | 3837 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3838 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3839 | } |
3840 | } | |
3841 | ||
3842 | /** | |
3843 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3844 | * @w: work descriptor |
1da177e4 LT |
3845 | * |
3846 | * Called from workqueue/eventd every few seconds. | |
3847 | * Purpose: | |
3848 | * - clear the per-cpu caches for this CPU. | |
3849 | * - return freeable pages to the main free memory pool. | |
3850 | * | |
a737b3e2 AM |
3851 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3852 | * again on the next iteration. | |
1da177e4 | 3853 | */ |
7c5cae36 | 3854 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3855 | { |
7a7c381d | 3856 | struct kmem_cache *searchp; |
ce8eb6c4 | 3857 | struct kmem_cache_node *n; |
7d6e6d09 | 3858 | int node = numa_mem_id(); |
bf6aede7 | 3859 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3860 | |
18004c5d | 3861 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3862 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3863 | goto out; |
1da177e4 | 3864 | |
18004c5d | 3865 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3866 | check_irq_on(); |
3867 | ||
35386e3b | 3868 | /* |
ce8eb6c4 | 3869 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3870 | * have established with reasonable certainty that |
3871 | * we can do some work if the lock was obtained. | |
3872 | */ | |
18bf8541 | 3873 | n = get_node(searchp, node); |
35386e3b | 3874 | |
ce8eb6c4 | 3875 | reap_alien(searchp, n); |
1da177e4 | 3876 | |
ce8eb6c4 | 3877 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3878 | |
35386e3b CL |
3879 | /* |
3880 | * These are racy checks but it does not matter | |
3881 | * if we skip one check or scan twice. | |
3882 | */ | |
ce8eb6c4 | 3883 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3884 | goto next; |
1da177e4 | 3885 | |
5f0985bb | 3886 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3887 | |
ce8eb6c4 | 3888 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3889 | |
ce8eb6c4 CL |
3890 | if (n->free_touched) |
3891 | n->free_touched = 0; | |
ed11d9eb CL |
3892 | else { |
3893 | int freed; | |
1da177e4 | 3894 | |
ce8eb6c4 | 3895 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3896 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3897 | STATS_ADD_REAPED(searchp, freed); | |
3898 | } | |
35386e3b | 3899 | next: |
1da177e4 LT |
3900 | cond_resched(); |
3901 | } | |
3902 | check_irq_on(); | |
18004c5d | 3903 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3904 | next_reap_node(); |
7c5cae36 | 3905 | out: |
a737b3e2 | 3906 | /* Set up the next iteration */ |
5f0985bb | 3907 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3908 | } |
3909 | ||
158a9624 | 3910 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3911 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3912 | { |
8456a648 | 3913 | struct page *page; |
b28a02de PE |
3914 | unsigned long active_objs; |
3915 | unsigned long num_objs; | |
3916 | unsigned long active_slabs = 0; | |
3917 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3918 | const char *name; |
1da177e4 | 3919 | char *error = NULL; |
e498be7d | 3920 | int node; |
ce8eb6c4 | 3921 | struct kmem_cache_node *n; |
1da177e4 | 3922 | |
1da177e4 LT |
3923 | active_objs = 0; |
3924 | num_slabs = 0; | |
18bf8541 | 3925 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3926 | |
ca3b9b91 | 3927 | check_irq_on(); |
ce8eb6c4 | 3928 | spin_lock_irq(&n->list_lock); |
e498be7d | 3929 | |
8456a648 JK |
3930 | list_for_each_entry(page, &n->slabs_full, lru) { |
3931 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3932 | error = "slabs_full accounting error"; |
3933 | active_objs += cachep->num; | |
3934 | active_slabs++; | |
3935 | } | |
8456a648 JK |
3936 | list_for_each_entry(page, &n->slabs_partial, lru) { |
3937 | if (page->active == cachep->num && !error) | |
106a74e1 | 3938 | error = "slabs_partial accounting error"; |
8456a648 | 3939 | if (!page->active && !error) |
106a74e1 | 3940 | error = "slabs_partial accounting error"; |
8456a648 | 3941 | active_objs += page->active; |
e498be7d CL |
3942 | active_slabs++; |
3943 | } | |
8456a648 JK |
3944 | list_for_each_entry(page, &n->slabs_free, lru) { |
3945 | if (page->active && !error) | |
106a74e1 | 3946 | error = "slabs_free accounting error"; |
e498be7d CL |
3947 | num_slabs++; |
3948 | } | |
ce8eb6c4 CL |
3949 | free_objects += n->free_objects; |
3950 | if (n->shared) | |
3951 | shared_avail += n->shared->avail; | |
e498be7d | 3952 | |
ce8eb6c4 | 3953 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 3954 | } |
b28a02de PE |
3955 | num_slabs += active_slabs; |
3956 | num_objs = num_slabs * cachep->num; | |
e498be7d | 3957 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
3958 | error = "free_objects accounting error"; |
3959 | ||
b28a02de | 3960 | name = cachep->name; |
1da177e4 LT |
3961 | if (error) |
3962 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
3963 | ||
0d7561c6 GC |
3964 | sinfo->active_objs = active_objs; |
3965 | sinfo->num_objs = num_objs; | |
3966 | sinfo->active_slabs = active_slabs; | |
3967 | sinfo->num_slabs = num_slabs; | |
3968 | sinfo->shared_avail = shared_avail; | |
3969 | sinfo->limit = cachep->limit; | |
3970 | sinfo->batchcount = cachep->batchcount; | |
3971 | sinfo->shared = cachep->shared; | |
3972 | sinfo->objects_per_slab = cachep->num; | |
3973 | sinfo->cache_order = cachep->gfporder; | |
3974 | } | |
3975 | ||
3976 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
3977 | { | |
1da177e4 | 3978 | #if STATS |
ce8eb6c4 | 3979 | { /* node stats */ |
1da177e4 LT |
3980 | unsigned long high = cachep->high_mark; |
3981 | unsigned long allocs = cachep->num_allocations; | |
3982 | unsigned long grown = cachep->grown; | |
3983 | unsigned long reaped = cachep->reaped; | |
3984 | unsigned long errors = cachep->errors; | |
3985 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 3986 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 3987 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 3988 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 3989 | |
e92dd4fd JP |
3990 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
3991 | "%4lu %4lu %4lu %4lu %4lu", | |
3992 | allocs, high, grown, | |
3993 | reaped, errors, max_freeable, node_allocs, | |
3994 | node_frees, overflows); | |
1da177e4 LT |
3995 | } |
3996 | /* cpu stats */ | |
3997 | { | |
3998 | unsigned long allochit = atomic_read(&cachep->allochit); | |
3999 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4000 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4001 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4002 | ||
4003 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4004 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4005 | } |
4006 | #endif | |
1da177e4 LT |
4007 | } |
4008 | ||
1da177e4 LT |
4009 | #define MAX_SLABINFO_WRITE 128 |
4010 | /** | |
4011 | * slabinfo_write - Tuning for the slab allocator | |
4012 | * @file: unused | |
4013 | * @buffer: user buffer | |
4014 | * @count: data length | |
4015 | * @ppos: unused | |
4016 | */ | |
b7454ad3 | 4017 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4018 | size_t count, loff_t *ppos) |
1da177e4 | 4019 | { |
b28a02de | 4020 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4021 | int limit, batchcount, shared, res; |
7a7c381d | 4022 | struct kmem_cache *cachep; |
b28a02de | 4023 | |
1da177e4 LT |
4024 | if (count > MAX_SLABINFO_WRITE) |
4025 | return -EINVAL; | |
4026 | if (copy_from_user(&kbuf, buffer, count)) | |
4027 | return -EFAULT; | |
b28a02de | 4028 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4029 | |
4030 | tmp = strchr(kbuf, ' '); | |
4031 | if (!tmp) | |
4032 | return -EINVAL; | |
4033 | *tmp = '\0'; | |
4034 | tmp++; | |
4035 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4036 | return -EINVAL; | |
4037 | ||
4038 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4039 | mutex_lock(&slab_mutex); |
1da177e4 | 4040 | res = -EINVAL; |
18004c5d | 4041 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4042 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4043 | if (limit < 1 || batchcount < 1 || |
4044 | batchcount > limit || shared < 0) { | |
e498be7d | 4045 | res = 0; |
1da177e4 | 4046 | } else { |
e498be7d | 4047 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4048 | batchcount, shared, |
4049 | GFP_KERNEL); | |
1da177e4 LT |
4050 | } |
4051 | break; | |
4052 | } | |
4053 | } | |
18004c5d | 4054 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4055 | if (res >= 0) |
4056 | res = count; | |
4057 | return res; | |
4058 | } | |
871751e2 AV |
4059 | |
4060 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4061 | ||
871751e2 AV |
4062 | static inline int add_caller(unsigned long *n, unsigned long v) |
4063 | { | |
4064 | unsigned long *p; | |
4065 | int l; | |
4066 | if (!v) | |
4067 | return 1; | |
4068 | l = n[1]; | |
4069 | p = n + 2; | |
4070 | while (l) { | |
4071 | int i = l/2; | |
4072 | unsigned long *q = p + 2 * i; | |
4073 | if (*q == v) { | |
4074 | q[1]++; | |
4075 | return 1; | |
4076 | } | |
4077 | if (*q > v) { | |
4078 | l = i; | |
4079 | } else { | |
4080 | p = q + 2; | |
4081 | l -= i + 1; | |
4082 | } | |
4083 | } | |
4084 | if (++n[1] == n[0]) | |
4085 | return 0; | |
4086 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4087 | p[0] = v; | |
4088 | p[1] = 1; | |
4089 | return 1; | |
4090 | } | |
4091 | ||
8456a648 JK |
4092 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4093 | struct page *page) | |
871751e2 AV |
4094 | { |
4095 | void *p; | |
d31676df JK |
4096 | int i, j; |
4097 | unsigned long v; | |
b1cb0982 | 4098 | |
871751e2 AV |
4099 | if (n[0] == n[1]) |
4100 | return; | |
8456a648 | 4101 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4102 | bool active = true; |
4103 | ||
4104 | for (j = page->active; j < c->num; j++) { | |
4105 | if (get_free_obj(page, j) == i) { | |
4106 | active = false; | |
4107 | break; | |
4108 | } | |
4109 | } | |
4110 | ||
4111 | if (!active) | |
871751e2 | 4112 | continue; |
b1cb0982 | 4113 | |
d31676df JK |
4114 | /* |
4115 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4116 | * mapping is established when actual object allocation and | |
4117 | * we could mistakenly access the unmapped object in the cpu | |
4118 | * cache. | |
4119 | */ | |
4120 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4121 | continue; | |
4122 | ||
4123 | if (!add_caller(n, v)) | |
871751e2 AV |
4124 | return; |
4125 | } | |
4126 | } | |
4127 | ||
4128 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4129 | { | |
4130 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4131 | unsigned long offset, size; |
9281acea | 4132 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4133 | |
a5c43dae | 4134 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4135 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4136 | if (modname[0]) |
871751e2 AV |
4137 | seq_printf(m, " [%s]", modname); |
4138 | return; | |
4139 | } | |
4140 | #endif | |
4141 | seq_printf(m, "%p", (void *)address); | |
4142 | } | |
4143 | ||
4144 | static int leaks_show(struct seq_file *m, void *p) | |
4145 | { | |
0672aa7c | 4146 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4147 | struct page *page; |
ce8eb6c4 | 4148 | struct kmem_cache_node *n; |
871751e2 | 4149 | const char *name; |
db845067 | 4150 | unsigned long *x = m->private; |
871751e2 AV |
4151 | int node; |
4152 | int i; | |
4153 | ||
4154 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4155 | return 0; | |
4156 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4157 | return 0; | |
4158 | ||
d31676df JK |
4159 | /* |
4160 | * Set store_user_clean and start to grab stored user information | |
4161 | * for all objects on this cache. If some alloc/free requests comes | |
4162 | * during the processing, information would be wrong so restart | |
4163 | * whole processing. | |
4164 | */ | |
4165 | do { | |
4166 | set_store_user_clean(cachep); | |
4167 | drain_cpu_caches(cachep); | |
4168 | ||
4169 | x[1] = 0; | |
871751e2 | 4170 | |
d31676df | 4171 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4172 | |
d31676df JK |
4173 | check_irq_on(); |
4174 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4175 | |
d31676df JK |
4176 | list_for_each_entry(page, &n->slabs_full, lru) |
4177 | handle_slab(x, cachep, page); | |
4178 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4179 | handle_slab(x, cachep, page); | |
4180 | spin_unlock_irq(&n->list_lock); | |
4181 | } | |
4182 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4183 | |
871751e2 | 4184 | name = cachep->name; |
db845067 | 4185 | if (x[0] == x[1]) { |
871751e2 | 4186 | /* Increase the buffer size */ |
18004c5d | 4187 | mutex_unlock(&slab_mutex); |
db845067 | 4188 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4189 | if (!m->private) { |
4190 | /* Too bad, we are really out */ | |
db845067 | 4191 | m->private = x; |
18004c5d | 4192 | mutex_lock(&slab_mutex); |
871751e2 AV |
4193 | return -ENOMEM; |
4194 | } | |
db845067 CL |
4195 | *(unsigned long *)m->private = x[0] * 2; |
4196 | kfree(x); | |
18004c5d | 4197 | mutex_lock(&slab_mutex); |
871751e2 AV |
4198 | /* Now make sure this entry will be retried */ |
4199 | m->count = m->size; | |
4200 | return 0; | |
4201 | } | |
db845067 CL |
4202 | for (i = 0; i < x[1]; i++) { |
4203 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4204 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4205 | seq_putc(m, '\n'); |
4206 | } | |
d2e7b7d0 | 4207 | |
871751e2 AV |
4208 | return 0; |
4209 | } | |
4210 | ||
a0ec95a8 | 4211 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4212 | .start = slab_start, |
276a2439 WL |
4213 | .next = slab_next, |
4214 | .stop = slab_stop, | |
871751e2 AV |
4215 | .show = leaks_show, |
4216 | }; | |
a0ec95a8 AD |
4217 | |
4218 | static int slabstats_open(struct inode *inode, struct file *file) | |
4219 | { | |
b208ce32 RJ |
4220 | unsigned long *n; |
4221 | ||
4222 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4223 | if (!n) | |
4224 | return -ENOMEM; | |
4225 | ||
4226 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4227 | ||
4228 | return 0; | |
a0ec95a8 AD |
4229 | } |
4230 | ||
4231 | static const struct file_operations proc_slabstats_operations = { | |
4232 | .open = slabstats_open, | |
4233 | .read = seq_read, | |
4234 | .llseek = seq_lseek, | |
4235 | .release = seq_release_private, | |
4236 | }; | |
4237 | #endif | |
4238 | ||
4239 | static int __init slab_proc_init(void) | |
4240 | { | |
4241 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4242 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4243 | #endif |
a0ec95a8 AD |
4244 | return 0; |
4245 | } | |
4246 | module_init(slab_proc_init); | |
1da177e4 LT |
4247 | #endif |
4248 | ||
00e145b6 MS |
4249 | /** |
4250 | * ksize - get the actual amount of memory allocated for a given object | |
4251 | * @objp: Pointer to the object | |
4252 | * | |
4253 | * kmalloc may internally round up allocations and return more memory | |
4254 | * than requested. ksize() can be used to determine the actual amount of | |
4255 | * memory allocated. The caller may use this additional memory, even though | |
4256 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4257 | * The caller must guarantee that objp points to a valid object previously | |
4258 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4259 | * must not be freed during the duration of the call. | |
4260 | */ | |
fd76bab2 | 4261 | size_t ksize(const void *objp) |
1da177e4 | 4262 | { |
ef8b4520 CL |
4263 | BUG_ON(!objp); |
4264 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4265 | return 0; |
1da177e4 | 4266 | |
8c138bc0 | 4267 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4268 | } |
b1aabecd | 4269 | EXPORT_SYMBOL(ksize); |