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) | |
d4322d88 | 275 | #define OFF_SLAB_MIN_SIZE (max_t(size_t, PAGE_SIZE >> 5, KMALLOC_MIN_SIZE + 1)) |
1da177e4 LT |
276 | |
277 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
278 | /* |
279 | * Optimization question: fewer reaps means less probability for unnessary | |
280 | * cpucache drain/refill cycles. | |
1da177e4 | 281 | * |
dc6f3f27 | 282 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
283 | * which could lock up otherwise freeable slabs. |
284 | */ | |
5f0985bb JZ |
285 | #define REAPTIMEOUT_AC (2*HZ) |
286 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
287 | |
288 | #if STATS | |
289 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
290 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
291 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
292 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 293 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
294 | #define STATS_SET_HIGH(x) \ |
295 | do { \ | |
296 | if ((x)->num_active > (x)->high_mark) \ | |
297 | (x)->high_mark = (x)->num_active; \ | |
298 | } while (0) | |
1da177e4 LT |
299 | #define STATS_INC_ERR(x) ((x)->errors++) |
300 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 301 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 302 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
303 | #define STATS_SET_FREEABLE(x, i) \ |
304 | do { \ | |
305 | if ((x)->max_freeable < i) \ | |
306 | (x)->max_freeable = i; \ | |
307 | } while (0) | |
1da177e4 LT |
308 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
309 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
310 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
311 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
312 | #else | |
313 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
314 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
315 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
316 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 317 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
318 | #define STATS_SET_HIGH(x) do { } while (0) |
319 | #define STATS_INC_ERR(x) do { } while (0) | |
320 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 321 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 322 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 323 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
324 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
325 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
326 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
327 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
328 | #endif | |
329 | ||
330 | #if DEBUG | |
1da177e4 | 331 | |
a737b3e2 AM |
332 | /* |
333 | * memory layout of objects: | |
1da177e4 | 334 | * 0 : objp |
3dafccf2 | 335 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
336 | * the end of an object is aligned with the end of the real |
337 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 338 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 339 | * redzone word. |
3dafccf2 | 340 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
341 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
342 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 343 | * [BYTES_PER_WORD long] |
1da177e4 | 344 | */ |
343e0d7a | 345 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 346 | { |
3dafccf2 | 347 | return cachep->obj_offset; |
1da177e4 LT |
348 | } |
349 | ||
b46b8f19 | 350 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
351 | { |
352 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
353 | return (unsigned long long*) (objp + obj_offset(cachep) - |
354 | sizeof(unsigned long long)); | |
1da177e4 LT |
355 | } |
356 | ||
b46b8f19 | 357 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
358 | { |
359 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
360 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 361 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 362 | sizeof(unsigned long long) - |
87a927c7 | 363 | REDZONE_ALIGN); |
3b0efdfa | 364 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 365 | sizeof(unsigned long long)); |
1da177e4 LT |
366 | } |
367 | ||
343e0d7a | 368 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
369 | { |
370 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 371 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
372 | } |
373 | ||
374 | #else | |
375 | ||
3dafccf2 | 376 | #define obj_offset(x) 0 |
b46b8f19 DW |
377 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
378 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
379 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
380 | ||
381 | #endif | |
382 | ||
03787301 JK |
383 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
384 | ||
d31676df | 385 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 386 | { |
d31676df JK |
387 | return atomic_read(&cachep->store_user_clean) == 1; |
388 | } | |
03787301 | 389 | |
d31676df JK |
390 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
391 | { | |
392 | atomic_set(&cachep->store_user_clean, 1); | |
393 | } | |
03787301 | 394 | |
d31676df JK |
395 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
396 | { | |
397 | if (is_store_user_clean(cachep)) | |
398 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
399 | } |
400 | ||
401 | #else | |
d31676df | 402 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
403 | |
404 | #endif | |
405 | ||
1da177e4 | 406 | /* |
3df1cccd DR |
407 | * Do not go above this order unless 0 objects fit into the slab or |
408 | * overridden on the command line. | |
1da177e4 | 409 | */ |
543585cc DR |
410 | #define SLAB_MAX_ORDER_HI 1 |
411 | #define SLAB_MAX_ORDER_LO 0 | |
412 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 413 | static bool slab_max_order_set __initdata; |
1da177e4 | 414 | |
6ed5eb22 PE |
415 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
416 | { | |
b49af68f | 417 | struct page *page = virt_to_head_page(obj); |
35026088 | 418 | return page->slab_cache; |
6ed5eb22 PE |
419 | } |
420 | ||
8456a648 | 421 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
422 | unsigned int idx) |
423 | { | |
8456a648 | 424 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
425 | } |
426 | ||
6a2d7a95 | 427 | /* |
3b0efdfa CL |
428 | * We want to avoid an expensive divide : (offset / cache->size) |
429 | * Using the fact that size is a constant for a particular cache, | |
430 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
431 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
432 | */ | |
433 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 434 | const struct page *page, void *obj) |
8fea4e96 | 435 | { |
8456a648 | 436 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 437 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
438 | } |
439 | ||
6fb92430 | 440 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 441 | /* internal cache of cache description objs */ |
9b030cb8 | 442 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
443 | .batchcount = 1, |
444 | .limit = BOOT_CPUCACHE_ENTRIES, | |
445 | .shared = 1, | |
3b0efdfa | 446 | .size = sizeof(struct kmem_cache), |
b28a02de | 447 | .name = "kmem_cache", |
1da177e4 LT |
448 | }; |
449 | ||
edcad250 JK |
450 | #define BAD_ALIEN_MAGIC 0x01020304ul |
451 | ||
1871e52c | 452 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 453 | |
343e0d7a | 454 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 455 | { |
bf0dea23 | 456 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
457 | } |
458 | ||
a737b3e2 AM |
459 | /* |
460 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
461 | */ | |
fbaccacf | 462 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
2e6b3602 | 463 | unsigned long flags, size_t *left_over, unsigned int *num) |
fbaccacf | 464 | { |
fbaccacf | 465 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 466 | |
fbaccacf SR |
467 | /* |
468 | * The slab management structure can be either off the slab or | |
469 | * on it. For the latter case, the memory allocated for a | |
470 | * slab is used for: | |
471 | * | |
fbaccacf | 472 | * - @buffer_size bytes for each object |
2e6b3602 JK |
473 | * - One freelist_idx_t for each object |
474 | * | |
475 | * We don't need to consider alignment of freelist because | |
476 | * freelist will be at the end of slab page. The objects will be | |
477 | * at the correct alignment. | |
fbaccacf SR |
478 | * |
479 | * If the slab management structure is off the slab, then the | |
480 | * alignment will already be calculated into the size. Because | |
481 | * the slabs are all pages aligned, the objects will be at the | |
482 | * correct alignment when allocated. | |
483 | */ | |
484 | if (flags & CFLGS_OFF_SLAB) { | |
2e6b3602 JK |
485 | *num = slab_size / buffer_size; |
486 | *left_over = slab_size % buffer_size; | |
fbaccacf | 487 | } else { |
2e6b3602 JK |
488 | *num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
489 | *left_over = slab_size % | |
490 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 491 | } |
1da177e4 LT |
492 | } |
493 | ||
f28510d3 | 494 | #if DEBUG |
d40cee24 | 495 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 496 | |
a737b3e2 AM |
497 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
498 | char *msg) | |
1da177e4 LT |
499 | { |
500 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 501 | function, cachep->name, msg); |
1da177e4 | 502 | dump_stack(); |
373d4d09 | 503 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 504 | } |
f28510d3 | 505 | #endif |
1da177e4 | 506 | |
3395ee05 PM |
507 | /* |
508 | * By default on NUMA we use alien caches to stage the freeing of | |
509 | * objects allocated from other nodes. This causes massive memory | |
510 | * inefficiencies when using fake NUMA setup to split memory into a | |
511 | * large number of small nodes, so it can be disabled on the command | |
512 | * line | |
513 | */ | |
514 | ||
515 | static int use_alien_caches __read_mostly = 1; | |
516 | static int __init noaliencache_setup(char *s) | |
517 | { | |
518 | use_alien_caches = 0; | |
519 | return 1; | |
520 | } | |
521 | __setup("noaliencache", noaliencache_setup); | |
522 | ||
3df1cccd DR |
523 | static int __init slab_max_order_setup(char *str) |
524 | { | |
525 | get_option(&str, &slab_max_order); | |
526 | slab_max_order = slab_max_order < 0 ? 0 : | |
527 | min(slab_max_order, MAX_ORDER - 1); | |
528 | slab_max_order_set = true; | |
529 | ||
530 | return 1; | |
531 | } | |
532 | __setup("slab_max_order=", slab_max_order_setup); | |
533 | ||
8fce4d8e CL |
534 | #ifdef CONFIG_NUMA |
535 | /* | |
536 | * Special reaping functions for NUMA systems called from cache_reap(). | |
537 | * These take care of doing round robin flushing of alien caches (containing | |
538 | * objects freed on different nodes from which they were allocated) and the | |
539 | * flushing of remote pcps by calling drain_node_pages. | |
540 | */ | |
1871e52c | 541 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
542 | |
543 | static void init_reap_node(int cpu) | |
544 | { | |
545 | int node; | |
546 | ||
7d6e6d09 | 547 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 548 | if (node == MAX_NUMNODES) |
442295c9 | 549 | node = first_node(node_online_map); |
8fce4d8e | 550 | |
1871e52c | 551 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
552 | } |
553 | ||
554 | static void next_reap_node(void) | |
555 | { | |
909ea964 | 556 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 557 | |
8fce4d8e CL |
558 | node = next_node(node, node_online_map); |
559 | if (unlikely(node >= MAX_NUMNODES)) | |
560 | node = first_node(node_online_map); | |
909ea964 | 561 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
562 | } |
563 | ||
564 | #else | |
565 | #define init_reap_node(cpu) do { } while (0) | |
566 | #define next_reap_node(void) do { } while (0) | |
567 | #endif | |
568 | ||
1da177e4 LT |
569 | /* |
570 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
571 | * via the workqueue/eventd. | |
572 | * Add the CPU number into the expiration time to minimize the possibility of | |
573 | * the CPUs getting into lockstep and contending for the global cache chain | |
574 | * lock. | |
575 | */ | |
0db0628d | 576 | static void start_cpu_timer(int cpu) |
1da177e4 | 577 | { |
1871e52c | 578 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
579 | |
580 | /* | |
581 | * When this gets called from do_initcalls via cpucache_init(), | |
582 | * init_workqueues() has already run, so keventd will be setup | |
583 | * at that time. | |
584 | */ | |
52bad64d | 585 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 586 | init_reap_node(cpu); |
203b42f7 | 587 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
588 | schedule_delayed_work_on(cpu, reap_work, |
589 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
590 | } |
591 | } | |
592 | ||
1fe00d50 | 593 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 594 | { |
d5cff635 CM |
595 | /* |
596 | * The array_cache structures contain pointers to free object. | |
25985edc | 597 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
598 | * cache the pointers are not cleared and they could be counted as |
599 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
600 | * not scan such objects. | |
601 | */ | |
1fe00d50 JK |
602 | kmemleak_no_scan(ac); |
603 | if (ac) { | |
604 | ac->avail = 0; | |
605 | ac->limit = limit; | |
606 | ac->batchcount = batch; | |
607 | ac->touched = 0; | |
1da177e4 | 608 | } |
1fe00d50 JK |
609 | } |
610 | ||
611 | static struct array_cache *alloc_arraycache(int node, int entries, | |
612 | int batchcount, gfp_t gfp) | |
613 | { | |
5e804789 | 614 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
615 | struct array_cache *ac = NULL; |
616 | ||
617 | ac = kmalloc_node(memsize, gfp, node); | |
618 | init_arraycache(ac, entries, batchcount); | |
619 | return ac; | |
1da177e4 LT |
620 | } |
621 | ||
8456a648 | 622 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 623 | { |
072bb0aa MG |
624 | return PageSlabPfmemalloc(page); |
625 | } | |
626 | ||
627 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
628 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
629 | struct array_cache *ac) | |
630 | { | |
18bf8541 | 631 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 632 | struct page *page; |
072bb0aa MG |
633 | unsigned long flags; |
634 | ||
635 | if (!pfmemalloc_active) | |
636 | return; | |
637 | ||
ce8eb6c4 | 638 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
639 | list_for_each_entry(page, &n->slabs_full, lru) |
640 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
641 | goto out; |
642 | ||
8456a648 JK |
643 | list_for_each_entry(page, &n->slabs_partial, lru) |
644 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
645 | goto out; |
646 | ||
8456a648 JK |
647 | list_for_each_entry(page, &n->slabs_free, lru) |
648 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
649 | goto out; |
650 | ||
651 | pfmemalloc_active = false; | |
652 | out: | |
ce8eb6c4 | 653 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
654 | } |
655 | ||
381760ea | 656 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
657 | gfp_t flags, bool force_refill) |
658 | { | |
659 | int i; | |
660 | void *objp = ac->entry[--ac->avail]; | |
661 | ||
662 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
663 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 664 | struct kmem_cache_node *n; |
072bb0aa MG |
665 | |
666 | if (gfp_pfmemalloc_allowed(flags)) { | |
667 | clear_obj_pfmemalloc(&objp); | |
668 | return objp; | |
669 | } | |
670 | ||
671 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 672 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
673 | /* If a !PFMEMALLOC object is found, swap them */ |
674 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
675 | objp = ac->entry[i]; | |
676 | ac->entry[i] = ac->entry[ac->avail]; | |
677 | ac->entry[ac->avail] = objp; | |
678 | return objp; | |
679 | } | |
680 | } | |
681 | ||
682 | /* | |
683 | * If there are empty slabs on the slabs_free list and we are | |
684 | * being forced to refill the cache, mark this one !pfmemalloc. | |
685 | */ | |
18bf8541 | 686 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 687 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 688 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 689 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
690 | clear_obj_pfmemalloc(&objp); |
691 | recheck_pfmemalloc_active(cachep, ac); | |
692 | return objp; | |
693 | } | |
694 | ||
695 | /* No !PFMEMALLOC objects available */ | |
696 | ac->avail++; | |
697 | objp = NULL; | |
698 | } | |
699 | ||
700 | return objp; | |
701 | } | |
702 | ||
381760ea MG |
703 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
704 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
705 | { | |
706 | void *objp; | |
707 | ||
708 | if (unlikely(sk_memalloc_socks())) | |
709 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
710 | else | |
711 | objp = ac->entry[--ac->avail]; | |
712 | ||
713 | return objp; | |
714 | } | |
715 | ||
d3aec344 JK |
716 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
717 | struct array_cache *ac, void *objp) | |
072bb0aa MG |
718 | { |
719 | if (unlikely(pfmemalloc_active)) { | |
720 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 721 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
722 | if (PageSlabPfmemalloc(page)) |
723 | set_obj_pfmemalloc(&objp); | |
724 | } | |
725 | ||
381760ea MG |
726 | return objp; |
727 | } | |
728 | ||
729 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
730 | void *objp) | |
731 | { | |
732 | if (unlikely(sk_memalloc_socks())) | |
733 | objp = __ac_put_obj(cachep, ac, objp); | |
734 | ||
072bb0aa MG |
735 | ac->entry[ac->avail++] = objp; |
736 | } | |
737 | ||
3ded175a CL |
738 | /* |
739 | * Transfer objects in one arraycache to another. | |
740 | * Locking must be handled by the caller. | |
741 | * | |
742 | * Return the number of entries transferred. | |
743 | */ | |
744 | static int transfer_objects(struct array_cache *to, | |
745 | struct array_cache *from, unsigned int max) | |
746 | { | |
747 | /* Figure out how many entries to transfer */ | |
732eacc0 | 748 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
749 | |
750 | if (!nr) | |
751 | return 0; | |
752 | ||
753 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
754 | sizeof(void *) *nr); | |
755 | ||
756 | from->avail -= nr; | |
757 | to->avail += nr; | |
3ded175a CL |
758 | return nr; |
759 | } | |
760 | ||
765c4507 CL |
761 | #ifndef CONFIG_NUMA |
762 | ||
763 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 764 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 765 | |
c8522a3a JK |
766 | static inline struct alien_cache **alloc_alien_cache(int node, |
767 | int limit, gfp_t gfp) | |
765c4507 | 768 | { |
edcad250 | 769 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
770 | } |
771 | ||
c8522a3a | 772 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
773 | { |
774 | } | |
775 | ||
776 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
777 | { | |
778 | return 0; | |
779 | } | |
780 | ||
781 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
782 | gfp_t flags) | |
783 | { | |
784 | return NULL; | |
785 | } | |
786 | ||
8b98c169 | 787 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
788 | gfp_t flags, int nodeid) |
789 | { | |
790 | return NULL; | |
791 | } | |
792 | ||
4167e9b2 DR |
793 | static inline gfp_t gfp_exact_node(gfp_t flags) |
794 | { | |
795 | return flags; | |
796 | } | |
797 | ||
765c4507 CL |
798 | #else /* CONFIG_NUMA */ |
799 | ||
8b98c169 | 800 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 801 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 802 | |
c8522a3a JK |
803 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
804 | int batch, gfp_t gfp) | |
805 | { | |
5e804789 | 806 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
807 | struct alien_cache *alc = NULL; |
808 | ||
809 | alc = kmalloc_node(memsize, gfp, node); | |
810 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 811 | spin_lock_init(&alc->lock); |
c8522a3a JK |
812 | return alc; |
813 | } | |
814 | ||
815 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 816 | { |
c8522a3a | 817 | struct alien_cache **alc_ptr; |
5e804789 | 818 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
819 | int i; |
820 | ||
821 | if (limit > 1) | |
822 | limit = 12; | |
c8522a3a JK |
823 | alc_ptr = kzalloc_node(memsize, gfp, node); |
824 | if (!alc_ptr) | |
825 | return NULL; | |
826 | ||
827 | for_each_node(i) { | |
828 | if (i == node || !node_online(i)) | |
829 | continue; | |
830 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
831 | if (!alc_ptr[i]) { | |
832 | for (i--; i >= 0; i--) | |
833 | kfree(alc_ptr[i]); | |
834 | kfree(alc_ptr); | |
835 | return NULL; | |
e498be7d CL |
836 | } |
837 | } | |
c8522a3a | 838 | return alc_ptr; |
e498be7d CL |
839 | } |
840 | ||
c8522a3a | 841 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
842 | { |
843 | int i; | |
844 | ||
c8522a3a | 845 | if (!alc_ptr) |
e498be7d | 846 | return; |
e498be7d | 847 | for_each_node(i) |
c8522a3a JK |
848 | kfree(alc_ptr[i]); |
849 | kfree(alc_ptr); | |
e498be7d CL |
850 | } |
851 | ||
343e0d7a | 852 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
853 | struct array_cache *ac, int node, |
854 | struct list_head *list) | |
e498be7d | 855 | { |
18bf8541 | 856 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
857 | |
858 | if (ac->avail) { | |
ce8eb6c4 | 859 | spin_lock(&n->list_lock); |
e00946fe CL |
860 | /* |
861 | * Stuff objects into the remote nodes shared array first. | |
862 | * That way we could avoid the overhead of putting the objects | |
863 | * into the free lists and getting them back later. | |
864 | */ | |
ce8eb6c4 CL |
865 | if (n->shared) |
866 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 867 | |
833b706c | 868 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 869 | ac->avail = 0; |
ce8eb6c4 | 870 | spin_unlock(&n->list_lock); |
e498be7d CL |
871 | } |
872 | } | |
873 | ||
8fce4d8e CL |
874 | /* |
875 | * Called from cache_reap() to regularly drain alien caches round robin. | |
876 | */ | |
ce8eb6c4 | 877 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 878 | { |
909ea964 | 879 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 880 | |
ce8eb6c4 | 881 | if (n->alien) { |
c8522a3a JK |
882 | struct alien_cache *alc = n->alien[node]; |
883 | struct array_cache *ac; | |
884 | ||
885 | if (alc) { | |
886 | ac = &alc->ac; | |
49dfc304 | 887 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
888 | LIST_HEAD(list); |
889 | ||
890 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 891 | spin_unlock_irq(&alc->lock); |
833b706c | 892 | slabs_destroy(cachep, &list); |
c8522a3a | 893 | } |
8fce4d8e CL |
894 | } |
895 | } | |
896 | } | |
897 | ||
a737b3e2 | 898 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 899 | struct alien_cache **alien) |
e498be7d | 900 | { |
b28a02de | 901 | int i = 0; |
c8522a3a | 902 | struct alien_cache *alc; |
e498be7d CL |
903 | struct array_cache *ac; |
904 | unsigned long flags; | |
905 | ||
906 | for_each_online_node(i) { | |
c8522a3a JK |
907 | alc = alien[i]; |
908 | if (alc) { | |
833b706c JK |
909 | LIST_HEAD(list); |
910 | ||
c8522a3a | 911 | ac = &alc->ac; |
49dfc304 | 912 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 913 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 914 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 915 | slabs_destroy(cachep, &list); |
e498be7d CL |
916 | } |
917 | } | |
918 | } | |
729bd0b7 | 919 | |
25c4f304 JK |
920 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
921 | int node, int page_node) | |
729bd0b7 | 922 | { |
ce8eb6c4 | 923 | struct kmem_cache_node *n; |
c8522a3a JK |
924 | struct alien_cache *alien = NULL; |
925 | struct array_cache *ac; | |
97654dfa | 926 | LIST_HEAD(list); |
1ca4cb24 | 927 | |
18bf8541 | 928 | n = get_node(cachep, node); |
729bd0b7 | 929 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
930 | if (n->alien && n->alien[page_node]) { |
931 | alien = n->alien[page_node]; | |
c8522a3a | 932 | ac = &alien->ac; |
49dfc304 | 933 | spin_lock(&alien->lock); |
c8522a3a | 934 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 935 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 936 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 937 | } |
c8522a3a | 938 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 939 | spin_unlock(&alien->lock); |
833b706c | 940 | slabs_destroy(cachep, &list); |
729bd0b7 | 941 | } else { |
25c4f304 | 942 | n = get_node(cachep, page_node); |
18bf8541 | 943 | spin_lock(&n->list_lock); |
25c4f304 | 944 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 945 | spin_unlock(&n->list_lock); |
97654dfa | 946 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
947 | } |
948 | return 1; | |
949 | } | |
25c4f304 JK |
950 | |
951 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
952 | { | |
953 | int page_node = page_to_nid(virt_to_page(objp)); | |
954 | int node = numa_mem_id(); | |
955 | /* | |
956 | * Make sure we are not freeing a object from another node to the array | |
957 | * cache on this cpu. | |
958 | */ | |
959 | if (likely(node == page_node)) | |
960 | return 0; | |
961 | ||
962 | return __cache_free_alien(cachep, objp, node, page_node); | |
963 | } | |
4167e9b2 DR |
964 | |
965 | /* | |
d0164adc MG |
966 | * Construct gfp mask to allocate from a specific node but do not direct reclaim |
967 | * or warn about failures. kswapd may still wake to reclaim in the background. | |
4167e9b2 DR |
968 | */ |
969 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
970 | { | |
d0164adc | 971 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM; |
4167e9b2 | 972 | } |
e498be7d CL |
973 | #endif |
974 | ||
8f9f8d9e | 975 | /* |
6a67368c | 976 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 977 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 978 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 979 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
980 | * already in use. |
981 | * | |
18004c5d | 982 | * Must hold slab_mutex. |
8f9f8d9e | 983 | */ |
6a67368c | 984 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
985 | { |
986 | struct kmem_cache *cachep; | |
ce8eb6c4 | 987 | struct kmem_cache_node *n; |
5e804789 | 988 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 989 | |
18004c5d | 990 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 991 | /* |
5f0985bb | 992 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
993 | * begin anything. Make sure some other cpu on this |
994 | * node has not already allocated this | |
995 | */ | |
18bf8541 CL |
996 | n = get_node(cachep, node); |
997 | if (!n) { | |
ce8eb6c4 CL |
998 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
999 | if (!n) | |
8f9f8d9e | 1000 | return -ENOMEM; |
ce8eb6c4 | 1001 | kmem_cache_node_init(n); |
5f0985bb JZ |
1002 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1003 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1004 | |
1005 | /* | |
5f0985bb JZ |
1006 | * The kmem_cache_nodes don't come and go as CPUs |
1007 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1008 | * protection here. |
1009 | */ | |
ce8eb6c4 | 1010 | cachep->node[node] = n; |
8f9f8d9e DR |
1011 | } |
1012 | ||
18bf8541 CL |
1013 | spin_lock_irq(&n->list_lock); |
1014 | n->free_limit = | |
8f9f8d9e DR |
1015 | (1 + nr_cpus_node(node)) * |
1016 | cachep->batchcount + cachep->num; | |
18bf8541 | 1017 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1018 | } |
1019 | return 0; | |
1020 | } | |
1021 | ||
0fa8103b WL |
1022 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1023 | struct kmem_cache_node *n) | |
1024 | { | |
1025 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1026 | } | |
1027 | ||
0db0628d | 1028 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1029 | { |
1030 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1031 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1032 | int node = cpu_to_mem(cpu); |
a70f7302 | 1033 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1034 | |
18004c5d | 1035 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1036 | struct array_cache *nc; |
1037 | struct array_cache *shared; | |
c8522a3a | 1038 | struct alien_cache **alien; |
97654dfa | 1039 | LIST_HEAD(list); |
fbf1e473 | 1040 | |
18bf8541 | 1041 | n = get_node(cachep, node); |
ce8eb6c4 | 1042 | if (!n) |
bf0dea23 | 1043 | continue; |
fbf1e473 | 1044 | |
ce8eb6c4 | 1045 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1046 | |
ce8eb6c4 CL |
1047 | /* Free limit for this kmem_cache_node */ |
1048 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1049 | |
1050 | /* cpu is dead; no one can alloc from it. */ | |
1051 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1052 | if (nc) { | |
97654dfa | 1053 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1054 | nc->avail = 0; |
1055 | } | |
fbf1e473 | 1056 | |
58463c1f | 1057 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1058 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1059 | goto free_slab; |
fbf1e473 AM |
1060 | } |
1061 | ||
ce8eb6c4 | 1062 | shared = n->shared; |
fbf1e473 AM |
1063 | if (shared) { |
1064 | free_block(cachep, shared->entry, | |
97654dfa | 1065 | shared->avail, node, &list); |
ce8eb6c4 | 1066 | n->shared = NULL; |
fbf1e473 AM |
1067 | } |
1068 | ||
ce8eb6c4 CL |
1069 | alien = n->alien; |
1070 | n->alien = NULL; | |
fbf1e473 | 1071 | |
ce8eb6c4 | 1072 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1073 | |
1074 | kfree(shared); | |
1075 | if (alien) { | |
1076 | drain_alien_cache(cachep, alien); | |
1077 | free_alien_cache(alien); | |
1078 | } | |
bf0dea23 JK |
1079 | |
1080 | free_slab: | |
97654dfa | 1081 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1082 | } |
1083 | /* | |
1084 | * In the previous loop, all the objects were freed to | |
1085 | * the respective cache's slabs, now we can go ahead and | |
1086 | * shrink each nodelist to its limit. | |
1087 | */ | |
18004c5d | 1088 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1089 | n = get_node(cachep, node); |
ce8eb6c4 | 1090 | if (!n) |
fbf1e473 | 1091 | continue; |
0fa8103b | 1092 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1093 | } |
1094 | } | |
1095 | ||
0db0628d | 1096 | static int cpuup_prepare(long cpu) |
1da177e4 | 1097 | { |
343e0d7a | 1098 | struct kmem_cache *cachep; |
ce8eb6c4 | 1099 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1100 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1101 | int err; |
1da177e4 | 1102 | |
fbf1e473 AM |
1103 | /* |
1104 | * We need to do this right in the beginning since | |
1105 | * alloc_arraycache's are going to use this list. | |
1106 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1107 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1108 | */ |
6a67368c | 1109 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1110 | if (err < 0) |
1111 | goto bad; | |
fbf1e473 AM |
1112 | |
1113 | /* | |
1114 | * Now we can go ahead with allocating the shared arrays and | |
1115 | * array caches | |
1116 | */ | |
18004c5d | 1117 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 1118 | struct array_cache *shared = NULL; |
c8522a3a | 1119 | struct alien_cache **alien = NULL; |
fbf1e473 | 1120 | |
fbf1e473 AM |
1121 | if (cachep->shared) { |
1122 | shared = alloc_arraycache(node, | |
1123 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1124 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1125 | if (!shared) |
1da177e4 | 1126 | goto bad; |
fbf1e473 AM |
1127 | } |
1128 | if (use_alien_caches) { | |
83b519e8 | 1129 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1130 | if (!alien) { |
1131 | kfree(shared); | |
fbf1e473 | 1132 | goto bad; |
12d00f6a | 1133 | } |
fbf1e473 | 1134 | } |
18bf8541 | 1135 | n = get_node(cachep, node); |
ce8eb6c4 | 1136 | BUG_ON(!n); |
fbf1e473 | 1137 | |
ce8eb6c4 CL |
1138 | spin_lock_irq(&n->list_lock); |
1139 | if (!n->shared) { | |
fbf1e473 AM |
1140 | /* |
1141 | * We are serialised from CPU_DEAD or | |
1142 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1143 | */ | |
ce8eb6c4 | 1144 | n->shared = shared; |
fbf1e473 AM |
1145 | shared = NULL; |
1146 | } | |
4484ebf1 | 1147 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1148 | if (!n->alien) { |
1149 | n->alien = alien; | |
fbf1e473 | 1150 | alien = NULL; |
1da177e4 | 1151 | } |
fbf1e473 | 1152 | #endif |
ce8eb6c4 | 1153 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1154 | kfree(shared); |
1155 | free_alien_cache(alien); | |
1156 | } | |
ce79ddc8 | 1157 | |
fbf1e473 AM |
1158 | return 0; |
1159 | bad: | |
12d00f6a | 1160 | cpuup_canceled(cpu); |
fbf1e473 AM |
1161 | return -ENOMEM; |
1162 | } | |
1163 | ||
0db0628d | 1164 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1165 | unsigned long action, void *hcpu) |
1166 | { | |
1167 | long cpu = (long)hcpu; | |
1168 | int err = 0; | |
1169 | ||
1170 | switch (action) { | |
fbf1e473 AM |
1171 | case CPU_UP_PREPARE: |
1172 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1173 | mutex_lock(&slab_mutex); |
fbf1e473 | 1174 | err = cpuup_prepare(cpu); |
18004c5d | 1175 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1176 | break; |
1177 | case CPU_ONLINE: | |
8bb78442 | 1178 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1179 | start_cpu_timer(cpu); |
1180 | break; | |
1181 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1182 | case CPU_DOWN_PREPARE: |
8bb78442 | 1183 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1184 | /* |
18004c5d | 1185 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1186 | * held so that if cache_reap() is invoked it cannot do |
1187 | * anything expensive but will only modify reap_work | |
1188 | * and reschedule the timer. | |
1189 | */ | |
afe2c511 | 1190 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1191 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1192 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1193 | break; |
1194 | case CPU_DOWN_FAILED: | |
8bb78442 | 1195 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1196 | start_cpu_timer(cpu); |
1197 | break; | |
1da177e4 | 1198 | case CPU_DEAD: |
8bb78442 | 1199 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1200 | /* |
1201 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1202 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1203 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1204 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1205 | * structure is usually allocated from kmem_cache_create() and |
1206 | * gets destroyed at kmem_cache_destroy(). | |
1207 | */ | |
183ff22b | 1208 | /* fall through */ |
8f5be20b | 1209 | #endif |
1da177e4 | 1210 | case CPU_UP_CANCELED: |
8bb78442 | 1211 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1212 | mutex_lock(&slab_mutex); |
fbf1e473 | 1213 | cpuup_canceled(cpu); |
18004c5d | 1214 | mutex_unlock(&slab_mutex); |
1da177e4 | 1215 | break; |
1da177e4 | 1216 | } |
eac40680 | 1217 | return notifier_from_errno(err); |
1da177e4 LT |
1218 | } |
1219 | ||
0db0628d | 1220 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1221 | &cpuup_callback, NULL, 0 |
1222 | }; | |
1da177e4 | 1223 | |
8f9f8d9e DR |
1224 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1225 | /* | |
1226 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1227 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1228 | * removed. | |
1229 | * | |
18004c5d | 1230 | * Must hold slab_mutex. |
8f9f8d9e | 1231 | */ |
6a67368c | 1232 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1233 | { |
1234 | struct kmem_cache *cachep; | |
1235 | int ret = 0; | |
1236 | ||
18004c5d | 1237 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1238 | struct kmem_cache_node *n; |
8f9f8d9e | 1239 | |
18bf8541 | 1240 | n = get_node(cachep, node); |
ce8eb6c4 | 1241 | if (!n) |
8f9f8d9e DR |
1242 | continue; |
1243 | ||
0fa8103b | 1244 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1245 | |
ce8eb6c4 CL |
1246 | if (!list_empty(&n->slabs_full) || |
1247 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1248 | ret = -EBUSY; |
1249 | break; | |
1250 | } | |
1251 | } | |
1252 | return ret; | |
1253 | } | |
1254 | ||
1255 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1256 | unsigned long action, void *arg) | |
1257 | { | |
1258 | struct memory_notify *mnb = arg; | |
1259 | int ret = 0; | |
1260 | int nid; | |
1261 | ||
1262 | nid = mnb->status_change_nid; | |
1263 | if (nid < 0) | |
1264 | goto out; | |
1265 | ||
1266 | switch (action) { | |
1267 | case MEM_GOING_ONLINE: | |
18004c5d | 1268 | mutex_lock(&slab_mutex); |
6a67368c | 1269 | ret = init_cache_node_node(nid); |
18004c5d | 1270 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1271 | break; |
1272 | case MEM_GOING_OFFLINE: | |
18004c5d | 1273 | mutex_lock(&slab_mutex); |
6a67368c | 1274 | ret = drain_cache_node_node(nid); |
18004c5d | 1275 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1276 | break; |
1277 | case MEM_ONLINE: | |
1278 | case MEM_OFFLINE: | |
1279 | case MEM_CANCEL_ONLINE: | |
1280 | case MEM_CANCEL_OFFLINE: | |
1281 | break; | |
1282 | } | |
1283 | out: | |
5fda1bd5 | 1284 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1285 | } |
1286 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1287 | ||
e498be7d | 1288 | /* |
ce8eb6c4 | 1289 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1290 | */ |
6744f087 | 1291 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1292 | int nodeid) |
e498be7d | 1293 | { |
6744f087 | 1294 | struct kmem_cache_node *ptr; |
e498be7d | 1295 | |
6744f087 | 1296 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1297 | BUG_ON(!ptr); |
1298 | ||
6744f087 | 1299 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1300 | /* |
1301 | * Do not assume that spinlocks can be initialized via memcpy: | |
1302 | */ | |
1303 | spin_lock_init(&ptr->list_lock); | |
1304 | ||
e498be7d | 1305 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1306 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1307 | } |
1308 | ||
556a169d | 1309 | /* |
ce8eb6c4 CL |
1310 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1311 | * size of kmem_cache_node. | |
556a169d | 1312 | */ |
ce8eb6c4 | 1313 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1314 | { |
1315 | int node; | |
1316 | ||
1317 | for_each_online_node(node) { | |
ce8eb6c4 | 1318 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1319 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1320 | REAPTIMEOUT_NODE + |
1321 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1322 | } |
1323 | } | |
1324 | ||
a737b3e2 AM |
1325 | /* |
1326 | * Initialisation. Called after the page allocator have been initialised and | |
1327 | * before smp_init(). | |
1da177e4 LT |
1328 | */ |
1329 | void __init kmem_cache_init(void) | |
1330 | { | |
e498be7d CL |
1331 | int i; |
1332 | ||
68126702 JK |
1333 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1334 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1335 | kmem_cache = &kmem_cache_boot; |
1336 | ||
b6e68bc1 | 1337 | if (num_possible_nodes() == 1) |
62918a03 SS |
1338 | use_alien_caches = 0; |
1339 | ||
3c583465 | 1340 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1341 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1342 | |
1da177e4 LT |
1343 | /* |
1344 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1345 | * page orders on machines with more than 32MB of memory if |
1346 | * not overridden on the command line. | |
1da177e4 | 1347 | */ |
3df1cccd | 1348 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1349 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1350 | |
1da177e4 LT |
1351 | /* Bootstrap is tricky, because several objects are allocated |
1352 | * from caches that do not exist yet: | |
9b030cb8 CL |
1353 | * 1) initialize the kmem_cache cache: it contains the struct |
1354 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1355 | * kmem_cache is statically allocated. | |
e498be7d | 1356 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1357 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1358 | * array at the end of the bootstrap. |
1da177e4 | 1359 | * 2) Create the first kmalloc cache. |
343e0d7a | 1360 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1361 | * An __init data area is used for the head array. |
1362 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1363 | * head arrays. | |
9b030cb8 | 1364 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1365 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1366 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1367 | * the other cache's with kmalloc allocated memory. |
1368 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1369 | */ |
1370 | ||
9b030cb8 | 1371 | /* 1) create the kmem_cache */ |
1da177e4 | 1372 | |
8da3430d | 1373 | /* |
b56efcf0 | 1374 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1375 | */ |
2f9baa9f | 1376 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1377 | offsetof(struct kmem_cache, node) + |
6744f087 | 1378 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1379 | SLAB_HWCACHE_ALIGN); |
1380 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1381 | slab_state = PARTIAL; |
1da177e4 | 1382 | |
a737b3e2 | 1383 | /* |
bf0dea23 JK |
1384 | * Initialize the caches that provide memory for the kmem_cache_node |
1385 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1386 | */ |
bf0dea23 | 1387 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1388 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1389 | slab_state = PARTIAL_NODE; |
34cc6990 | 1390 | setup_kmalloc_cache_index_table(); |
e498be7d | 1391 | |
e0a42726 IM |
1392 | slab_early_init = 0; |
1393 | ||
ce8eb6c4 | 1394 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1395 | { |
1ca4cb24 PE |
1396 | int nid; |
1397 | ||
9c09a95c | 1398 | for_each_online_node(nid) { |
ce8eb6c4 | 1399 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1400 | |
bf0dea23 | 1401 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1402 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1403 | } |
1404 | } | |
1da177e4 | 1405 | |
f97d5f63 | 1406 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1407 | } |
1408 | ||
1409 | void __init kmem_cache_init_late(void) | |
1410 | { | |
1411 | struct kmem_cache *cachep; | |
1412 | ||
97d06609 | 1413 | slab_state = UP; |
52cef189 | 1414 | |
8429db5c | 1415 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1416 | mutex_lock(&slab_mutex); |
1417 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1418 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1419 | BUG(); | |
18004c5d | 1420 | mutex_unlock(&slab_mutex); |
056c6241 | 1421 | |
97d06609 CL |
1422 | /* Done! */ |
1423 | slab_state = FULL; | |
1424 | ||
a737b3e2 AM |
1425 | /* |
1426 | * Register a cpu startup notifier callback that initializes | |
1427 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1428 | */ |
1429 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1430 | |
8f9f8d9e DR |
1431 | #ifdef CONFIG_NUMA |
1432 | /* | |
1433 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1434 | * node. |
8f9f8d9e DR |
1435 | */ |
1436 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1437 | #endif | |
1438 | ||
a737b3e2 AM |
1439 | /* |
1440 | * The reap timers are started later, with a module init call: That part | |
1441 | * of the kernel is not yet operational. | |
1da177e4 LT |
1442 | */ |
1443 | } | |
1444 | ||
1445 | static int __init cpucache_init(void) | |
1446 | { | |
1447 | int cpu; | |
1448 | ||
a737b3e2 AM |
1449 | /* |
1450 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1451 | */ |
e498be7d | 1452 | for_each_online_cpu(cpu) |
a737b3e2 | 1453 | start_cpu_timer(cpu); |
a164f896 GC |
1454 | |
1455 | /* Done! */ | |
97d06609 | 1456 | slab_state = FULL; |
1da177e4 LT |
1457 | return 0; |
1458 | } | |
1da177e4 LT |
1459 | __initcall(cpucache_init); |
1460 | ||
8bdec192 RA |
1461 | static noinline void |
1462 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1463 | { | |
9a02d699 | 1464 | #if DEBUG |
ce8eb6c4 | 1465 | struct kmem_cache_node *n; |
8456a648 | 1466 | struct page *page; |
8bdec192 RA |
1467 | unsigned long flags; |
1468 | int node; | |
9a02d699 DR |
1469 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1470 | DEFAULT_RATELIMIT_BURST); | |
1471 | ||
1472 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1473 | return; | |
8bdec192 RA |
1474 | |
1475 | printk(KERN_WARNING | |
1476 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1477 | nodeid, gfpflags); | |
1478 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1479 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1480 | |
18bf8541 | 1481 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1482 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1483 | unsigned long active_slabs = 0, num_slabs = 0; | |
1484 | ||
ce8eb6c4 | 1485 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1486 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1487 | active_objs += cachep->num; |
1488 | active_slabs++; | |
1489 | } | |
8456a648 JK |
1490 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1491 | active_objs += page->active; | |
8bdec192 RA |
1492 | active_slabs++; |
1493 | } | |
8456a648 | 1494 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1495 | num_slabs++; |
1496 | ||
ce8eb6c4 CL |
1497 | free_objects += n->free_objects; |
1498 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1499 | |
1500 | num_slabs += active_slabs; | |
1501 | num_objs = num_slabs * cachep->num; | |
1502 | printk(KERN_WARNING | |
1503 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1504 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1505 | free_objects); | |
1506 | } | |
9a02d699 | 1507 | #endif |
8bdec192 RA |
1508 | } |
1509 | ||
1da177e4 | 1510 | /* |
8a7d9b43 WSH |
1511 | * Interface to system's page allocator. No need to hold the |
1512 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1513 | * |
1514 | * If we requested dmaable memory, we will get it. Even if we | |
1515 | * did not request dmaable memory, we might get it, but that | |
1516 | * would be relatively rare and ignorable. | |
1517 | */ | |
0c3aa83e JK |
1518 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1519 | int nodeid) | |
1da177e4 LT |
1520 | { |
1521 | struct page *page; | |
e1b6aa6f | 1522 | int nr_pages; |
765c4507 | 1523 | |
a618e89f | 1524 | flags |= cachep->allocflags; |
e12ba74d MG |
1525 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1526 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1527 | |
96db800f | 1528 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1529 | if (!page) { |
9a02d699 | 1530 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1531 | return NULL; |
8bdec192 | 1532 | } |
1da177e4 | 1533 | |
f3ccb2c4 VD |
1534 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1535 | __free_pages(page, cachep->gfporder); | |
1536 | return NULL; | |
1537 | } | |
1538 | ||
b37f1dd0 | 1539 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
2f064f34 | 1540 | if (page_is_pfmemalloc(page)) |
072bb0aa MG |
1541 | pfmemalloc_active = true; |
1542 | ||
e1b6aa6f | 1543 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1544 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1545 | add_zone_page_state(page_zone(page), |
1546 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1547 | else | |
1548 | add_zone_page_state(page_zone(page), | |
1549 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 | 1550 | __SetPageSlab(page); |
2f064f34 | 1551 | if (page_is_pfmemalloc(page)) |
a57a4988 | 1552 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1553 | |
b1eeab67 VN |
1554 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1555 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1556 | ||
1557 | if (cachep->ctor) | |
1558 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1559 | else | |
1560 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1561 | } | |
c175eea4 | 1562 | |
0c3aa83e | 1563 | return page; |
1da177e4 LT |
1564 | } |
1565 | ||
1566 | /* | |
1567 | * Interface to system's page release. | |
1568 | */ | |
0c3aa83e | 1569 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1570 | { |
a57a4988 | 1571 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1572 | |
b1eeab67 | 1573 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1574 | |
972d1a7b CL |
1575 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1576 | sub_zone_page_state(page_zone(page), | |
1577 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1578 | else | |
1579 | sub_zone_page_state(page_zone(page), | |
1580 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1581 | |
a57a4988 | 1582 | BUG_ON(!PageSlab(page)); |
73293c2f | 1583 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1584 | __ClearPageSlab(page); |
8456a648 JK |
1585 | page_mapcount_reset(page); |
1586 | page->mapping = NULL; | |
1f458cbf | 1587 | |
1da177e4 LT |
1588 | if (current->reclaim_state) |
1589 | current->reclaim_state->reclaimed_slab += nr_freed; | |
f3ccb2c4 | 1590 | __free_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1591 | } |
1592 | ||
1593 | static void kmem_rcu_free(struct rcu_head *head) | |
1594 | { | |
68126702 JK |
1595 | struct kmem_cache *cachep; |
1596 | struct page *page; | |
1da177e4 | 1597 | |
68126702 JK |
1598 | page = container_of(head, struct page, rcu_head); |
1599 | cachep = page->slab_cache; | |
1600 | ||
1601 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1602 | } |
1603 | ||
1604 | #if DEBUG | |
40b44137 JK |
1605 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1606 | { | |
1607 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1608 | (cachep->size % PAGE_SIZE) == 0) | |
1609 | return true; | |
1610 | ||
1611 | return false; | |
1612 | } | |
1da177e4 LT |
1613 | |
1614 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1615 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1616 | unsigned long caller) |
1da177e4 | 1617 | { |
8c138bc0 | 1618 | int size = cachep->object_size; |
1da177e4 | 1619 | |
3dafccf2 | 1620 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1621 | |
b28a02de | 1622 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1623 | return; |
1624 | ||
b28a02de PE |
1625 | *addr++ = 0x12345678; |
1626 | *addr++ = caller; | |
1627 | *addr++ = smp_processor_id(); | |
1628 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1629 | { |
1630 | unsigned long *sptr = &caller; | |
1631 | unsigned long svalue; | |
1632 | ||
1633 | while (!kstack_end(sptr)) { | |
1634 | svalue = *sptr++; | |
1635 | if (kernel_text_address(svalue)) { | |
b28a02de | 1636 | *addr++ = svalue; |
1da177e4 LT |
1637 | size -= sizeof(unsigned long); |
1638 | if (size <= sizeof(unsigned long)) | |
1639 | break; | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | } | |
b28a02de | 1644 | *addr++ = 0x87654321; |
1da177e4 | 1645 | } |
40b44137 JK |
1646 | |
1647 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1648 | int map, unsigned long caller) | |
1649 | { | |
1650 | if (!is_debug_pagealloc_cache(cachep)) | |
1651 | return; | |
1652 | ||
1653 | if (caller) | |
1654 | store_stackinfo(cachep, objp, caller); | |
1655 | ||
1656 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1657 | } | |
1658 | ||
1659 | #else | |
1660 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1661 | int map, unsigned long caller) {} | |
1662 | ||
1da177e4 LT |
1663 | #endif |
1664 | ||
343e0d7a | 1665 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1666 | { |
8c138bc0 | 1667 | int size = cachep->object_size; |
3dafccf2 | 1668 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1669 | |
1670 | memset(addr, val, size); | |
b28a02de | 1671 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1672 | } |
1673 | ||
1674 | static void dump_line(char *data, int offset, int limit) | |
1675 | { | |
1676 | int i; | |
aa83aa40 DJ |
1677 | unsigned char error = 0; |
1678 | int bad_count = 0; | |
1679 | ||
fdde6abb | 1680 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1681 | for (i = 0; i < limit; i++) { |
1682 | if (data[offset + i] != POISON_FREE) { | |
1683 | error = data[offset + i]; | |
1684 | bad_count++; | |
1685 | } | |
aa83aa40 | 1686 | } |
fdde6abb SAS |
1687 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1688 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1689 | |
1690 | if (bad_count == 1) { | |
1691 | error ^= POISON_FREE; | |
1692 | if (!(error & (error - 1))) { | |
1693 | printk(KERN_ERR "Single bit error detected. Probably " | |
1694 | "bad RAM.\n"); | |
1695 | #ifdef CONFIG_X86 | |
1696 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1697 | "test tool.\n"); | |
1698 | #else | |
1699 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1700 | #endif | |
1701 | } | |
1702 | } | |
1da177e4 LT |
1703 | } |
1704 | #endif | |
1705 | ||
1706 | #if DEBUG | |
1707 | ||
343e0d7a | 1708 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1709 | { |
1710 | int i, size; | |
1711 | char *realobj; | |
1712 | ||
1713 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1714 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1715 | *dbg_redzone1(cachep, objp), |
1716 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1717 | } |
1718 | ||
1719 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1720 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1721 | *dbg_userword(cachep, objp), | |
1722 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1723 | } |
3dafccf2 | 1724 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1725 | size = cachep->object_size; |
b28a02de | 1726 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1727 | int limit; |
1728 | limit = 16; | |
b28a02de PE |
1729 | if (i + limit > size) |
1730 | limit = size - i; | |
1da177e4 LT |
1731 | dump_line(realobj, i, limit); |
1732 | } | |
1733 | } | |
1734 | ||
343e0d7a | 1735 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1736 | { |
1737 | char *realobj; | |
1738 | int size, i; | |
1739 | int lines = 0; | |
1740 | ||
40b44137 JK |
1741 | if (is_debug_pagealloc_cache(cachep)) |
1742 | return; | |
1743 | ||
3dafccf2 | 1744 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1745 | size = cachep->object_size; |
1da177e4 | 1746 | |
b28a02de | 1747 | for (i = 0; i < size; i++) { |
1da177e4 | 1748 | char exp = POISON_FREE; |
b28a02de | 1749 | if (i == size - 1) |
1da177e4 LT |
1750 | exp = POISON_END; |
1751 | if (realobj[i] != exp) { | |
1752 | int limit; | |
1753 | /* Mismatch ! */ | |
1754 | /* Print header */ | |
1755 | if (lines == 0) { | |
b28a02de | 1756 | printk(KERN_ERR |
face37f5 DJ |
1757 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1758 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1759 | print_objinfo(cachep, objp, 0); |
1760 | } | |
1761 | /* Hexdump the affected line */ | |
b28a02de | 1762 | i = (i / 16) * 16; |
1da177e4 | 1763 | limit = 16; |
b28a02de PE |
1764 | if (i + limit > size) |
1765 | limit = size - i; | |
1da177e4 LT |
1766 | dump_line(realobj, i, limit); |
1767 | i += 16; | |
1768 | lines++; | |
1769 | /* Limit to 5 lines */ | |
1770 | if (lines > 5) | |
1771 | break; | |
1772 | } | |
1773 | } | |
1774 | if (lines != 0) { | |
1775 | /* Print some data about the neighboring objects, if they | |
1776 | * exist: | |
1777 | */ | |
8456a648 | 1778 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1779 | unsigned int objnr; |
1da177e4 | 1780 | |
8456a648 | 1781 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1782 | if (objnr) { |
8456a648 | 1783 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1784 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1785 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1786 | realobj, size); |
1da177e4 LT |
1787 | print_objinfo(cachep, objp, 2); |
1788 | } | |
b28a02de | 1789 | if (objnr + 1 < cachep->num) { |
8456a648 | 1790 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1791 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1792 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1793 | realobj, size); |
1da177e4 LT |
1794 | print_objinfo(cachep, objp, 2); |
1795 | } | |
1796 | } | |
1797 | } | |
1798 | #endif | |
1799 | ||
12dd36fa | 1800 | #if DEBUG |
8456a648 JK |
1801 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1802 | struct page *page) | |
1da177e4 | 1803 | { |
1da177e4 LT |
1804 | int i; |
1805 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1806 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1807 | |
1808 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1809 | check_poison_obj(cachep, objp); |
40b44137 | 1810 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1811 | } |
1812 | if (cachep->flags & SLAB_RED_ZONE) { | |
1813 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1814 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1815 | "was overwritten"); |
1da177e4 LT |
1816 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1817 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1818 | "was overwritten"); |
1da177e4 | 1819 | } |
1da177e4 | 1820 | } |
12dd36fa | 1821 | } |
1da177e4 | 1822 | #else |
8456a648 JK |
1823 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1824 | struct page *page) | |
12dd36fa | 1825 | { |
12dd36fa | 1826 | } |
1da177e4 LT |
1827 | #endif |
1828 | ||
911851e6 RD |
1829 | /** |
1830 | * slab_destroy - destroy and release all objects in a slab | |
1831 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1832 | * @page: page pointer being destroyed |
911851e6 | 1833 | * |
8a7d9b43 WSH |
1834 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1835 | * Before calling the slab page must have been unlinked from the cache. The | |
1836 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1837 | */ |
8456a648 | 1838 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1839 | { |
7e007355 | 1840 | void *freelist; |
12dd36fa | 1841 | |
8456a648 JK |
1842 | freelist = page->freelist; |
1843 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1844 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1845 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1846 | else | |
0c3aa83e | 1847 | kmem_freepages(cachep, page); |
68126702 JK |
1848 | |
1849 | /* | |
8456a648 | 1850 | * From now on, we don't use freelist |
68126702 JK |
1851 | * although actual page can be freed in rcu context |
1852 | */ | |
1853 | if (OFF_SLAB(cachep)) | |
8456a648 | 1854 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1855 | } |
1856 | ||
97654dfa JK |
1857 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1858 | { | |
1859 | struct page *page, *n; | |
1860 | ||
1861 | list_for_each_entry_safe(page, n, list, lru) { | |
1862 | list_del(&page->lru); | |
1863 | slab_destroy(cachep, page); | |
1864 | } | |
1865 | } | |
1866 | ||
4d268eba | 1867 | /** |
a70773dd RD |
1868 | * calculate_slab_order - calculate size (page order) of slabs |
1869 | * @cachep: pointer to the cache that is being created | |
1870 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1871 | * @flags: slab allocation flags |
1872 | * | |
1873 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1874 | * |
1875 | * This could be made much more intelligent. For now, try to avoid using | |
1876 | * high order pages for slabs. When the gfp() functions are more friendly | |
1877 | * towards high-order requests, this should be changed. | |
1878 | */ | |
a737b3e2 | 1879 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1880 | size_t size, unsigned long flags) |
4d268eba | 1881 | { |
b1ab41c4 | 1882 | unsigned long offslab_limit; |
4d268eba | 1883 | size_t left_over = 0; |
9888e6fa | 1884 | int gfporder; |
4d268eba | 1885 | |
0aa817f0 | 1886 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1887 | unsigned int num; |
1888 | size_t remainder; | |
1889 | ||
2e6b3602 | 1890 | cache_estimate(gfporder, size, flags, &remainder, &num); |
4d268eba PE |
1891 | if (!num) |
1892 | continue; | |
9888e6fa | 1893 | |
f315e3fa JK |
1894 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1895 | if (num > SLAB_OBJ_MAX_NUM) | |
1896 | break; | |
1897 | ||
b1ab41c4 IM |
1898 | if (flags & CFLGS_OFF_SLAB) { |
1899 | /* | |
1900 | * Max number of objs-per-slab for caches which | |
1901 | * use off-slab slabs. Needed to avoid a possible | |
1902 | * looping condition in cache_grow(). | |
1903 | */ | |
8456a648 | 1904 | offslab_limit = size; |
249247b6 | 1905 | offslab_limit /= sizeof(freelist_idx_t); |
b1ab41c4 IM |
1906 | |
1907 | if (num > offslab_limit) | |
1908 | break; | |
1909 | } | |
4d268eba | 1910 | |
9888e6fa | 1911 | /* Found something acceptable - save it away */ |
4d268eba | 1912 | cachep->num = num; |
9888e6fa | 1913 | cachep->gfporder = gfporder; |
4d268eba PE |
1914 | left_over = remainder; |
1915 | ||
f78bb8ad LT |
1916 | /* |
1917 | * A VFS-reclaimable slab tends to have most allocations | |
1918 | * as GFP_NOFS and we really don't want to have to be allocating | |
1919 | * higher-order pages when we are unable to shrink dcache. | |
1920 | */ | |
1921 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1922 | break; | |
1923 | ||
4d268eba PE |
1924 | /* |
1925 | * Large number of objects is good, but very large slabs are | |
1926 | * currently bad for the gfp()s. | |
1927 | */ | |
543585cc | 1928 | if (gfporder >= slab_max_order) |
4d268eba PE |
1929 | break; |
1930 | ||
9888e6fa LT |
1931 | /* |
1932 | * Acceptable internal fragmentation? | |
1933 | */ | |
a737b3e2 | 1934 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1935 | break; |
1936 | } | |
1937 | return left_over; | |
1938 | } | |
1939 | ||
bf0dea23 JK |
1940 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1941 | struct kmem_cache *cachep, int entries, int batchcount) | |
1942 | { | |
1943 | int cpu; | |
1944 | size_t size; | |
1945 | struct array_cache __percpu *cpu_cache; | |
1946 | ||
1947 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1948 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1949 | |
1950 | if (!cpu_cache) | |
1951 | return NULL; | |
1952 | ||
1953 | for_each_possible_cpu(cpu) { | |
1954 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1955 | entries, batchcount); | |
1956 | } | |
1957 | ||
1958 | return cpu_cache; | |
1959 | } | |
1960 | ||
83b519e8 | 1961 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1962 | { |
97d06609 | 1963 | if (slab_state >= FULL) |
83b519e8 | 1964 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1965 | |
bf0dea23 JK |
1966 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1967 | if (!cachep->cpu_cache) | |
1968 | return 1; | |
1969 | ||
97d06609 | 1970 | if (slab_state == DOWN) { |
bf0dea23 JK |
1971 | /* Creation of first cache (kmem_cache). */ |
1972 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1973 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1974 | /* For kmem_cache_node */ |
1975 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1976 | } else { |
bf0dea23 | 1977 | int node; |
f30cf7d1 | 1978 | |
bf0dea23 JK |
1979 | for_each_online_node(node) { |
1980 | cachep->node[node] = kmalloc_node( | |
1981 | sizeof(struct kmem_cache_node), gfp, node); | |
1982 | BUG_ON(!cachep->node[node]); | |
1983 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1984 | } |
1985 | } | |
bf0dea23 | 1986 | |
6a67368c | 1987 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1988 | jiffies + REAPTIMEOUT_NODE + |
1989 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1990 | |
1991 | cpu_cache_get(cachep)->avail = 0; | |
1992 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1993 | cpu_cache_get(cachep)->batchcount = 1; | |
1994 | cpu_cache_get(cachep)->touched = 0; | |
1995 | cachep->batchcount = 1; | |
1996 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1997 | return 0; |
f30cf7d1 PE |
1998 | } |
1999 | ||
12220dea JK |
2000 | unsigned long kmem_cache_flags(unsigned long object_size, |
2001 | unsigned long flags, const char *name, | |
2002 | void (*ctor)(void *)) | |
2003 | { | |
2004 | return flags; | |
2005 | } | |
2006 | ||
2007 | struct kmem_cache * | |
2008 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
2009 | unsigned long flags, void (*ctor)(void *)) | |
2010 | { | |
2011 | struct kmem_cache *cachep; | |
2012 | ||
2013 | cachep = find_mergeable(size, align, flags, name, ctor); | |
2014 | if (cachep) { | |
2015 | cachep->refcount++; | |
2016 | ||
2017 | /* | |
2018 | * Adjust the object sizes so that we clear | |
2019 | * the complete object on kzalloc. | |
2020 | */ | |
2021 | cachep->object_size = max_t(int, cachep->object_size, size); | |
2022 | } | |
2023 | return cachep; | |
2024 | } | |
2025 | ||
1da177e4 | 2026 | /** |
039363f3 | 2027 | * __kmem_cache_create - Create a cache. |
a755b76a | 2028 | * @cachep: cache management descriptor |
1da177e4 | 2029 | * @flags: SLAB flags |
1da177e4 LT |
2030 | * |
2031 | * Returns a ptr to the cache on success, NULL on failure. | |
2032 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2033 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2034 | * |
1da177e4 LT |
2035 | * The flags are |
2036 | * | |
2037 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2038 | * to catch references to uninitialised memory. | |
2039 | * | |
2040 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2041 | * for buffer overruns. | |
2042 | * | |
1da177e4 LT |
2043 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2044 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2045 | * as davem. | |
2046 | */ | |
278b1bb1 | 2047 | int |
8a13a4cc | 2048 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2049 | { |
d4a5fca5 DR |
2050 | size_t left_over, freelist_size; |
2051 | size_t ralign = BYTES_PER_WORD; | |
83b519e8 | 2052 | gfp_t gfp; |
278b1bb1 | 2053 | int err; |
8a13a4cc | 2054 | size_t size = cachep->size; |
1da177e4 | 2055 | |
1da177e4 | 2056 | #if DEBUG |
1da177e4 LT |
2057 | #if FORCED_DEBUG |
2058 | /* | |
2059 | * Enable redzoning and last user accounting, except for caches with | |
2060 | * large objects, if the increased size would increase the object size | |
2061 | * above the next power of two: caches with object sizes just above a | |
2062 | * power of two have a significant amount of internal fragmentation. | |
2063 | */ | |
87a927c7 DW |
2064 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2065 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2066 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2067 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2068 | flags |= SLAB_POISON; | |
2069 | #endif | |
1da177e4 | 2070 | #endif |
1da177e4 | 2071 | |
a737b3e2 AM |
2072 | /* |
2073 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2074 | * unaligned accesses for some archs when redzoning is used, and makes |
2075 | * sure any on-slab bufctl's are also correctly aligned. | |
2076 | */ | |
b28a02de PE |
2077 | if (size & (BYTES_PER_WORD - 1)) { |
2078 | size += (BYTES_PER_WORD - 1); | |
2079 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2080 | } |
2081 | ||
87a927c7 DW |
2082 | if (flags & SLAB_RED_ZONE) { |
2083 | ralign = REDZONE_ALIGN; | |
2084 | /* If redzoning, ensure that the second redzone is suitably | |
2085 | * aligned, by adjusting the object size accordingly. */ | |
2086 | size += REDZONE_ALIGN - 1; | |
2087 | size &= ~(REDZONE_ALIGN - 1); | |
2088 | } | |
ca5f9703 | 2089 | |
a44b56d3 | 2090 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2091 | if (ralign < cachep->align) { |
2092 | ralign = cachep->align; | |
1da177e4 | 2093 | } |
3ff84a7f PE |
2094 | /* disable debug if necessary */ |
2095 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2096 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2097 | /* |
ca5f9703 | 2098 | * 4) Store it. |
1da177e4 | 2099 | */ |
8a13a4cc | 2100 | cachep->align = ralign; |
1da177e4 | 2101 | |
83b519e8 PE |
2102 | if (slab_is_available()) |
2103 | gfp = GFP_KERNEL; | |
2104 | else | |
2105 | gfp = GFP_NOWAIT; | |
2106 | ||
1da177e4 | 2107 | #if DEBUG |
1da177e4 | 2108 | |
ca5f9703 PE |
2109 | /* |
2110 | * Both debugging options require word-alignment which is calculated | |
2111 | * into align above. | |
2112 | */ | |
1da177e4 | 2113 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2114 | /* add space for red zone words */ |
3ff84a7f PE |
2115 | cachep->obj_offset += sizeof(unsigned long long); |
2116 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2117 | } |
2118 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2119 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2120 | * the real object. But if the second red zone needs to be |
2121 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2122 | */ |
87a927c7 DW |
2123 | if (flags & SLAB_RED_ZONE) |
2124 | size += REDZONE_ALIGN; | |
2125 | else | |
2126 | size += BYTES_PER_WORD; | |
1da177e4 | 2127 | } |
832a15d2 JK |
2128 | #endif |
2129 | ||
2130 | size = ALIGN(size, cachep->align); | |
2131 | /* | |
2132 | * We should restrict the number of objects in a slab to implement | |
2133 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2134 | */ | |
2135 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2136 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2137 | ||
2138 | #if DEBUG | |
03a2d2a3 JK |
2139 | /* |
2140 | * To activate debug pagealloc, off-slab management is necessary | |
2141 | * requirement. In early phase of initialization, small sized slab | |
2142 | * doesn't get initialized so it would not be possible. So, we need | |
2143 | * to check size >= 256. It guarantees that all necessary small | |
2144 | * sized slab is initialized in current slab initialization sequence. | |
2145 | */ | |
40323278 | 2146 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
a307ebd4 | 2147 | !slab_early_init && size >= kmalloc_size(INDEX_NODE) && |
03a2d2a3 | 2148 | size >= 256 && cachep->object_size > cache_line_size() && |
832a15d2 JK |
2149 | size < PAGE_SIZE) { |
2150 | cachep->obj_offset += PAGE_SIZE - size; | |
1da177e4 LT |
2151 | size = PAGE_SIZE; |
2152 | } | |
1da177e4 LT |
2153 | #endif |
2154 | ||
e0a42726 IM |
2155 | /* |
2156 | * Determine if the slab management is 'on' or 'off' slab. | |
2157 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2158 | * it too early on. Always use on-slab management when |
2159 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2160 | */ |
d4322d88 | 2161 | if (size >= OFF_SLAB_MIN_SIZE && !slab_early_init && |
832a15d2 | 2162 | !(flags & SLAB_NOLEAKTRACE)) { |
1da177e4 LT |
2163 | /* |
2164 | * Size is large, assume best to place the slab management obj | |
2165 | * off-slab (should allow better packing of objs). | |
2166 | */ | |
2167 | flags |= CFLGS_OFF_SLAB; | |
832a15d2 | 2168 | } |
1da177e4 | 2169 | |
2e6b3602 | 2170 | left_over = calculate_slab_order(cachep, size, flags); |
1da177e4 | 2171 | |
8a13a4cc | 2172 | if (!cachep->num) |
278b1bb1 | 2173 | return -E2BIG; |
1da177e4 | 2174 | |
2e6b3602 | 2175 | freelist_size = cachep->num * sizeof(freelist_idx_t); |
1da177e4 LT |
2176 | |
2177 | /* | |
2178 | * If the slab has been placed off-slab, and we have enough space then | |
2179 | * move it on-slab. This is at the expense of any extra colouring. | |
2180 | */ | |
8456a648 | 2181 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2182 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2183 | left_over -= freelist_size; |
1da177e4 LT |
2184 | } |
2185 | ||
1da177e4 LT |
2186 | cachep->colour_off = cache_line_size(); |
2187 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2188 | if (cachep->colour_off < cachep->align) |
2189 | cachep->colour_off = cachep->align; | |
b28a02de | 2190 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2191 | cachep->freelist_size = freelist_size; |
1da177e4 | 2192 | cachep->flags = flags; |
a57a4988 | 2193 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2194 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2195 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2196 | cachep->size = size; |
6a2d7a95 | 2197 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2198 | |
40b44137 JK |
2199 | #if DEBUG |
2200 | /* | |
2201 | * If we're going to use the generic kernel_map_pages() | |
2202 | * poisoning, then it's going to smash the contents of | |
2203 | * the redzone and userword anyhow, so switch them off. | |
2204 | */ | |
2205 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2206 | (cachep->flags & SLAB_POISON) && | |
2207 | is_debug_pagealloc_cache(cachep)) | |
2208 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2209 | #endif | |
2210 | ||
2211 | if (OFF_SLAB(cachep)) { | |
8456a648 | 2212 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2213 | /* |
5f0985bb | 2214 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2215 | * But since we go off slab only for object size greater than |
d4322d88 | 2216 | * OFF_SLAB_MIN_SIZE, and kmalloc_{dma,}_caches get created |
5f0985bb | 2217 | * in ascending order,this should not happen at all. |
e5ac9c5a RT |
2218 | * But leave a BUG_ON for some lucky dude. |
2219 | */ | |
8456a648 | 2220 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2221 | } |
1da177e4 | 2222 | |
278b1bb1 CL |
2223 | err = setup_cpu_cache(cachep, gfp); |
2224 | if (err) { | |
52b4b950 | 2225 | __kmem_cache_release(cachep); |
278b1bb1 | 2226 | return err; |
2ed3a4ef | 2227 | } |
1da177e4 | 2228 | |
278b1bb1 | 2229 | return 0; |
1da177e4 | 2230 | } |
1da177e4 LT |
2231 | |
2232 | #if DEBUG | |
2233 | static void check_irq_off(void) | |
2234 | { | |
2235 | BUG_ON(!irqs_disabled()); | |
2236 | } | |
2237 | ||
2238 | static void check_irq_on(void) | |
2239 | { | |
2240 | BUG_ON(irqs_disabled()); | |
2241 | } | |
2242 | ||
343e0d7a | 2243 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2244 | { |
2245 | #ifdef CONFIG_SMP | |
2246 | check_irq_off(); | |
18bf8541 | 2247 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2248 | #endif |
2249 | } | |
e498be7d | 2250 | |
343e0d7a | 2251 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2252 | { |
2253 | #ifdef CONFIG_SMP | |
2254 | check_irq_off(); | |
18bf8541 | 2255 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2256 | #endif |
2257 | } | |
2258 | ||
1da177e4 LT |
2259 | #else |
2260 | #define check_irq_off() do { } while(0) | |
2261 | #define check_irq_on() do { } while(0) | |
2262 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2263 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2264 | #endif |
2265 | ||
ce8eb6c4 | 2266 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2267 | struct array_cache *ac, |
2268 | int force, int node); | |
2269 | ||
1da177e4 LT |
2270 | static void do_drain(void *arg) |
2271 | { | |
a737b3e2 | 2272 | struct kmem_cache *cachep = arg; |
1da177e4 | 2273 | struct array_cache *ac; |
7d6e6d09 | 2274 | int node = numa_mem_id(); |
18bf8541 | 2275 | struct kmem_cache_node *n; |
97654dfa | 2276 | LIST_HEAD(list); |
1da177e4 LT |
2277 | |
2278 | check_irq_off(); | |
9a2dba4b | 2279 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2280 | n = get_node(cachep, node); |
2281 | spin_lock(&n->list_lock); | |
97654dfa | 2282 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2283 | spin_unlock(&n->list_lock); |
97654dfa | 2284 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2285 | ac->avail = 0; |
2286 | } | |
2287 | ||
343e0d7a | 2288 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2289 | { |
ce8eb6c4 | 2290 | struct kmem_cache_node *n; |
e498be7d CL |
2291 | int node; |
2292 | ||
15c8b6c1 | 2293 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2294 | check_irq_on(); |
18bf8541 CL |
2295 | for_each_kmem_cache_node(cachep, node, n) |
2296 | if (n->alien) | |
ce8eb6c4 | 2297 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2298 | |
18bf8541 CL |
2299 | for_each_kmem_cache_node(cachep, node, n) |
2300 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2301 | } |
2302 | ||
ed11d9eb CL |
2303 | /* |
2304 | * Remove slabs from the list of free slabs. | |
2305 | * Specify the number of slabs to drain in tofree. | |
2306 | * | |
2307 | * Returns the actual number of slabs released. | |
2308 | */ | |
2309 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2310 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2311 | { |
ed11d9eb CL |
2312 | struct list_head *p; |
2313 | int nr_freed; | |
8456a648 | 2314 | struct page *page; |
1da177e4 | 2315 | |
ed11d9eb | 2316 | nr_freed = 0; |
ce8eb6c4 | 2317 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2318 | |
ce8eb6c4 CL |
2319 | spin_lock_irq(&n->list_lock); |
2320 | p = n->slabs_free.prev; | |
2321 | if (p == &n->slabs_free) { | |
2322 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2323 | goto out; |
2324 | } | |
1da177e4 | 2325 | |
8456a648 | 2326 | page = list_entry(p, struct page, lru); |
8456a648 | 2327 | list_del(&page->lru); |
ed11d9eb CL |
2328 | /* |
2329 | * Safe to drop the lock. The slab is no longer linked | |
2330 | * to the cache. | |
2331 | */ | |
ce8eb6c4 CL |
2332 | n->free_objects -= cache->num; |
2333 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2334 | slab_destroy(cache, page); |
ed11d9eb | 2335 | nr_freed++; |
1da177e4 | 2336 | } |
ed11d9eb CL |
2337 | out: |
2338 | return nr_freed; | |
1da177e4 LT |
2339 | } |
2340 | ||
d6e0b7fa | 2341 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2342 | { |
18bf8541 CL |
2343 | int ret = 0; |
2344 | int node; | |
ce8eb6c4 | 2345 | struct kmem_cache_node *n; |
e498be7d CL |
2346 | |
2347 | drain_cpu_caches(cachep); | |
2348 | ||
2349 | check_irq_on(); | |
18bf8541 | 2350 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2351 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2352 | |
ce8eb6c4 CL |
2353 | ret += !list_empty(&n->slabs_full) || |
2354 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2355 | } |
2356 | return (ret ? 1 : 0); | |
2357 | } | |
2358 | ||
945cf2b6 | 2359 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2360 | { |
2361 | return __kmem_cache_shrink(cachep, false); | |
2362 | } | |
2363 | ||
2364 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2365 | { |
12c3667f | 2366 | int i; |
ce8eb6c4 | 2367 | struct kmem_cache_node *n; |
1da177e4 | 2368 | |
bf0dea23 | 2369 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2370 | |
ce8eb6c4 | 2371 | /* NUMA: free the node structures */ |
18bf8541 CL |
2372 | for_each_kmem_cache_node(cachep, i, n) { |
2373 | kfree(n->shared); | |
2374 | free_alien_cache(n->alien); | |
2375 | kfree(n); | |
2376 | cachep->node[i] = NULL; | |
12c3667f | 2377 | } |
1da177e4 | 2378 | } |
1da177e4 | 2379 | |
e5ac9c5a RT |
2380 | /* |
2381 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2382 | * |
2383 | * For a slab cache when the slab descriptor is off-slab, the | |
2384 | * slab descriptor can't come from the same cache which is being created, | |
2385 | * Because if it is the case, that means we defer the creation of | |
2386 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2387 | * And we eventually call down to __kmem_cache_create(), which | |
2388 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2389 | * This is a "chicken-and-egg" problem. | |
2390 | * | |
2391 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2392 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2393 | */ |
7e007355 | 2394 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2395 | struct page *page, int colour_off, |
2396 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2397 | { |
7e007355 | 2398 | void *freelist; |
0c3aa83e | 2399 | void *addr = page_address(page); |
b28a02de | 2400 | |
2e6b3602 JK |
2401 | page->s_mem = addr + colour_off; |
2402 | page->active = 0; | |
2403 | ||
1da177e4 LT |
2404 | if (OFF_SLAB(cachep)) { |
2405 | /* Slab management obj is off-slab. */ | |
8456a648 | 2406 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2407 | local_flags, nodeid); |
8456a648 | 2408 | if (!freelist) |
1da177e4 LT |
2409 | return NULL; |
2410 | } else { | |
2e6b3602 JK |
2411 | /* We will use last bytes at the slab for freelist */ |
2412 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2413 | cachep->freelist_size; | |
1da177e4 | 2414 | } |
2e6b3602 | 2415 | |
8456a648 | 2416 | return freelist; |
1da177e4 LT |
2417 | } |
2418 | ||
7cc68973 | 2419 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2420 | { |
a41adfaa | 2421 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2422 | } |
2423 | ||
2424 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2425 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2426 | { |
a41adfaa | 2427 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2428 | } |
2429 | ||
343e0d7a | 2430 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2431 | struct page *page) |
1da177e4 LT |
2432 | { |
2433 | int i; | |
2434 | ||
2435 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2436 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 | 2437 | #if DEBUG |
1da177e4 LT |
2438 | if (cachep->flags & SLAB_STORE_USER) |
2439 | *dbg_userword(cachep, objp) = NULL; | |
2440 | ||
2441 | if (cachep->flags & SLAB_RED_ZONE) { | |
2442 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2443 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2444 | } | |
2445 | /* | |
a737b3e2 AM |
2446 | * Constructors are not allowed to allocate memory from the same |
2447 | * cache which they are a constructor for. Otherwise, deadlock. | |
2448 | * They must also be threaded. | |
1da177e4 LT |
2449 | */ |
2450 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2451 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2452 | |
2453 | if (cachep->flags & SLAB_RED_ZONE) { | |
2454 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2455 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2456 | " end of an object"); |
1da177e4 LT |
2457 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2458 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2459 | " start of an object"); |
1da177e4 | 2460 | } |
40b44137 JK |
2461 | /* need to poison the objs? */ |
2462 | if (cachep->flags & SLAB_POISON) { | |
2463 | poison_obj(cachep, objp, POISON_FREE); | |
2464 | slab_kernel_map(cachep, objp, 0, 0); | |
2465 | } | |
1da177e4 LT |
2466 | #else |
2467 | if (cachep->ctor) | |
51cc5068 | 2468 | cachep->ctor(objp); |
1da177e4 | 2469 | #endif |
e5c58dfd | 2470 | set_free_obj(page, i, i); |
1da177e4 | 2471 | } |
1da177e4 LT |
2472 | } |
2473 | ||
343e0d7a | 2474 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2475 | { |
4b51d669 CL |
2476 | if (CONFIG_ZONE_DMA_FLAG) { |
2477 | if (flags & GFP_DMA) | |
a618e89f | 2478 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2479 | else |
a618e89f | 2480 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2481 | } |
1da177e4 LT |
2482 | } |
2483 | ||
260b61dd | 2484 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2485 | { |
b1cb0982 | 2486 | void *objp; |
78d382d7 | 2487 | |
e5c58dfd | 2488 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2489 | page->active++; |
78d382d7 | 2490 | |
d31676df JK |
2491 | #if DEBUG |
2492 | if (cachep->flags & SLAB_STORE_USER) | |
2493 | set_store_user_dirty(cachep); | |
2494 | #endif | |
2495 | ||
78d382d7 MD |
2496 | return objp; |
2497 | } | |
2498 | ||
260b61dd JK |
2499 | static void slab_put_obj(struct kmem_cache *cachep, |
2500 | struct page *page, void *objp) | |
78d382d7 | 2501 | { |
8456a648 | 2502 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2503 | #if DEBUG |
16025177 | 2504 | unsigned int i; |
b1cb0982 | 2505 | |
b1cb0982 | 2506 | /* Verify double free bug */ |
8456a648 | 2507 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2508 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2509 | printk(KERN_ERR "slab: double free detected in cache " |
2510 | "'%s', objp %p\n", cachep->name, objp); | |
2511 | BUG(); | |
2512 | } | |
78d382d7 MD |
2513 | } |
2514 | #endif | |
8456a648 | 2515 | page->active--; |
e5c58dfd | 2516 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2517 | } |
2518 | ||
4776874f PE |
2519 | /* |
2520 | * Map pages beginning at addr to the given cache and slab. This is required | |
2521 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2522 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2523 | */ |
8456a648 | 2524 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2525 | void *freelist) |
1da177e4 | 2526 | { |
a57a4988 | 2527 | page->slab_cache = cache; |
8456a648 | 2528 | page->freelist = freelist; |
1da177e4 LT |
2529 | } |
2530 | ||
2531 | /* | |
2532 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2533 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2534 | */ | |
3c517a61 | 2535 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2536 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2537 | { |
7e007355 | 2538 | void *freelist; |
b28a02de PE |
2539 | size_t offset; |
2540 | gfp_t local_flags; | |
ce8eb6c4 | 2541 | struct kmem_cache_node *n; |
1da177e4 | 2542 | |
a737b3e2 AM |
2543 | /* |
2544 | * Be lazy and only check for valid flags here, keeping it out of the | |
2545 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2546 | */ |
c871ac4e AM |
2547 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2548 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2549 | BUG(); | |
2550 | } | |
6cb06229 | 2551 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2552 | |
ce8eb6c4 | 2553 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2554 | check_irq_off(); |
18bf8541 | 2555 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2556 | spin_lock(&n->list_lock); |
1da177e4 LT |
2557 | |
2558 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2559 | offset = n->colour_next; |
2560 | n->colour_next++; | |
2561 | if (n->colour_next >= cachep->colour) | |
2562 | n->colour_next = 0; | |
2563 | spin_unlock(&n->list_lock); | |
1da177e4 | 2564 | |
2e1217cf | 2565 | offset *= cachep->colour_off; |
1da177e4 | 2566 | |
d0164adc | 2567 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2568 | local_irq_enable(); |
2569 | ||
2570 | /* | |
2571 | * The test for missing atomic flag is performed here, rather than | |
2572 | * the more obvious place, simply to reduce the critical path length | |
2573 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2574 | * will eventually be caught here (where it matters). | |
2575 | */ | |
2576 | kmem_flagcheck(cachep, flags); | |
2577 | ||
a737b3e2 AM |
2578 | /* |
2579 | * Get mem for the objs. Attempt to allocate a physical page from | |
2580 | * 'nodeid'. | |
e498be7d | 2581 | */ |
0c3aa83e JK |
2582 | if (!page) |
2583 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2584 | if (!page) | |
1da177e4 LT |
2585 | goto failed; |
2586 | ||
2587 | /* Get slab management. */ | |
8456a648 | 2588 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2589 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2590 | if (!freelist) |
1da177e4 LT |
2591 | goto opps1; |
2592 | ||
8456a648 | 2593 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2594 | |
8456a648 | 2595 | cache_init_objs(cachep, page); |
1da177e4 | 2596 | |
d0164adc | 2597 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2598 | local_irq_disable(); |
2599 | check_irq_off(); | |
ce8eb6c4 | 2600 | spin_lock(&n->list_lock); |
1da177e4 LT |
2601 | |
2602 | /* Make slab active. */ | |
8456a648 | 2603 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2604 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2605 | n->free_objects += cachep->num; |
2606 | spin_unlock(&n->list_lock); | |
1da177e4 | 2607 | return 1; |
a737b3e2 | 2608 | opps1: |
0c3aa83e | 2609 | kmem_freepages(cachep, page); |
a737b3e2 | 2610 | failed: |
d0164adc | 2611 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2612 | local_irq_disable(); |
2613 | return 0; | |
2614 | } | |
2615 | ||
2616 | #if DEBUG | |
2617 | ||
2618 | /* | |
2619 | * Perform extra freeing checks: | |
2620 | * - detect bad pointers. | |
2621 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2622 | */ |
2623 | static void kfree_debugcheck(const void *objp) | |
2624 | { | |
1da177e4 LT |
2625 | if (!virt_addr_valid(objp)) { |
2626 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2627 | (unsigned long)objp); |
2628 | BUG(); | |
1da177e4 | 2629 | } |
1da177e4 LT |
2630 | } |
2631 | ||
58ce1fd5 PE |
2632 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2633 | { | |
b46b8f19 | 2634 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2635 | |
2636 | redzone1 = *dbg_redzone1(cache, obj); | |
2637 | redzone2 = *dbg_redzone2(cache, obj); | |
2638 | ||
2639 | /* | |
2640 | * Redzone is ok. | |
2641 | */ | |
2642 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2643 | return; | |
2644 | ||
2645 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2646 | slab_error(cache, "double free detected"); | |
2647 | else | |
2648 | slab_error(cache, "memory outside object was overwritten"); | |
2649 | ||
b46b8f19 | 2650 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2651 | obj, redzone1, redzone2); |
2652 | } | |
2653 | ||
343e0d7a | 2654 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2655 | unsigned long caller) |
1da177e4 | 2656 | { |
1da177e4 | 2657 | unsigned int objnr; |
8456a648 | 2658 | struct page *page; |
1da177e4 | 2659 | |
80cbd911 MW |
2660 | BUG_ON(virt_to_cache(objp) != cachep); |
2661 | ||
3dafccf2 | 2662 | objp -= obj_offset(cachep); |
1da177e4 | 2663 | kfree_debugcheck(objp); |
b49af68f | 2664 | page = virt_to_head_page(objp); |
1da177e4 | 2665 | |
1da177e4 | 2666 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2667 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2668 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2669 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2670 | } | |
d31676df JK |
2671 | if (cachep->flags & SLAB_STORE_USER) { |
2672 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2673 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2674 | } |
1da177e4 | 2675 | |
8456a648 | 2676 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2677 | |
2678 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2679 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2680 | |
1da177e4 | 2681 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2682 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2683 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2684 | } |
2685 | return objp; | |
2686 | } | |
2687 | ||
1da177e4 LT |
2688 | #else |
2689 | #define kfree_debugcheck(x) do { } while(0) | |
2690 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2691 | #endif |
2692 | ||
7aa0d227 GT |
2693 | static struct page *get_first_slab(struct kmem_cache_node *n) |
2694 | { | |
2695 | struct page *page; | |
2696 | ||
2697 | page = list_first_entry_or_null(&n->slabs_partial, | |
2698 | struct page, lru); | |
2699 | if (!page) { | |
2700 | n->free_touched = 1; | |
2701 | page = list_first_entry_or_null(&n->slabs_free, | |
2702 | struct page, lru); | |
2703 | } | |
2704 | ||
2705 | return page; | |
2706 | } | |
2707 | ||
072bb0aa MG |
2708 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2709 | bool force_refill) | |
1da177e4 LT |
2710 | { |
2711 | int batchcount; | |
ce8eb6c4 | 2712 | struct kmem_cache_node *n; |
1da177e4 | 2713 | struct array_cache *ac; |
1ca4cb24 PE |
2714 | int node; |
2715 | ||
1da177e4 | 2716 | check_irq_off(); |
7d6e6d09 | 2717 | node = numa_mem_id(); |
072bb0aa MG |
2718 | if (unlikely(force_refill)) |
2719 | goto force_grow; | |
2720 | retry: | |
9a2dba4b | 2721 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2722 | batchcount = ac->batchcount; |
2723 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2724 | /* |
2725 | * If there was little recent activity on this cache, then | |
2726 | * perform only a partial refill. Otherwise we could generate | |
2727 | * refill bouncing. | |
1da177e4 LT |
2728 | */ |
2729 | batchcount = BATCHREFILL_LIMIT; | |
2730 | } | |
18bf8541 | 2731 | n = get_node(cachep, node); |
e498be7d | 2732 | |
ce8eb6c4 CL |
2733 | BUG_ON(ac->avail > 0 || !n); |
2734 | spin_lock(&n->list_lock); | |
1da177e4 | 2735 | |
3ded175a | 2736 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2737 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2738 | n->shared->touched = 1; | |
3ded175a | 2739 | goto alloc_done; |
44b57f1c | 2740 | } |
3ded175a | 2741 | |
1da177e4 | 2742 | while (batchcount > 0) { |
8456a648 | 2743 | struct page *page; |
1da177e4 | 2744 | /* Get slab alloc is to come from. */ |
7aa0d227 GT |
2745 | page = get_first_slab(n); |
2746 | if (!page) | |
2747 | goto must_grow; | |
1da177e4 | 2748 | |
1da177e4 | 2749 | check_spinlock_acquired(cachep); |
714b8171 PE |
2750 | |
2751 | /* | |
2752 | * The slab was either on partial or free list so | |
2753 | * there must be at least one object available for | |
2754 | * allocation. | |
2755 | */ | |
8456a648 | 2756 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2757 | |
8456a648 | 2758 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2759 | STATS_INC_ALLOCED(cachep); |
2760 | STATS_INC_ACTIVE(cachep); | |
2761 | STATS_SET_HIGH(cachep); | |
2762 | ||
260b61dd | 2763 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page)); |
1da177e4 | 2764 | } |
1da177e4 LT |
2765 | |
2766 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2767 | list_del(&page->lru); |
2768 | if (page->active == cachep->num) | |
34bf6ef9 | 2769 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2770 | else |
34bf6ef9 | 2771 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2772 | } |
2773 | ||
a737b3e2 | 2774 | must_grow: |
ce8eb6c4 | 2775 | n->free_objects -= ac->avail; |
a737b3e2 | 2776 | alloc_done: |
ce8eb6c4 | 2777 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2778 | |
2779 | if (unlikely(!ac->avail)) { | |
2780 | int x; | |
072bb0aa | 2781 | force_grow: |
4167e9b2 | 2782 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2783 | |
a737b3e2 | 2784 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2785 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2786 | node = numa_mem_id(); |
072bb0aa MG |
2787 | |
2788 | /* no objects in sight? abort */ | |
2789 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2790 | return NULL; |
2791 | ||
a737b3e2 | 2792 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2793 | goto retry; |
2794 | } | |
2795 | ac->touched = 1; | |
072bb0aa MG |
2796 | |
2797 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2798 | } |
2799 | ||
a737b3e2 AM |
2800 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2801 | gfp_t flags) | |
1da177e4 | 2802 | { |
d0164adc | 2803 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2804 | #if DEBUG |
2805 | kmem_flagcheck(cachep, flags); | |
2806 | #endif | |
2807 | } | |
2808 | ||
2809 | #if DEBUG | |
a737b3e2 | 2810 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2811 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2812 | { |
b28a02de | 2813 | if (!objp) |
1da177e4 | 2814 | return objp; |
b28a02de | 2815 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2816 | check_poison_obj(cachep, objp); |
40b44137 | 2817 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2818 | poison_obj(cachep, objp, POISON_INUSE); |
2819 | } | |
2820 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2821 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2822 | |
2823 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2824 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2825 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2826 | slab_error(cachep, "double free, or memory outside" | |
2827 | " object was overwritten"); | |
b28a02de | 2828 | printk(KERN_ERR |
b46b8f19 | 2829 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2830 | objp, *dbg_redzone1(cachep, objp), |
2831 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2832 | } |
2833 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2834 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2835 | } | |
03787301 | 2836 | |
3dafccf2 | 2837 | objp += obj_offset(cachep); |
4f104934 | 2838 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2839 | cachep->ctor(objp); |
7ea466f2 TH |
2840 | if (ARCH_SLAB_MINALIGN && |
2841 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2842 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2843 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2844 | } |
1da177e4 LT |
2845 | return objp; |
2846 | } | |
2847 | #else | |
2848 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2849 | #endif | |
2850 | ||
343e0d7a | 2851 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2852 | { |
b28a02de | 2853 | void *objp; |
1da177e4 | 2854 | struct array_cache *ac; |
072bb0aa | 2855 | bool force_refill = false; |
1da177e4 | 2856 | |
5c382300 | 2857 | check_irq_off(); |
8a8b6502 | 2858 | |
9a2dba4b | 2859 | ac = cpu_cache_get(cachep); |
1da177e4 | 2860 | if (likely(ac->avail)) { |
1da177e4 | 2861 | ac->touched = 1; |
072bb0aa MG |
2862 | objp = ac_get_obj(cachep, ac, flags, false); |
2863 | ||
ddbf2e83 | 2864 | /* |
072bb0aa MG |
2865 | * Allow for the possibility all avail objects are not allowed |
2866 | * by the current flags | |
ddbf2e83 | 2867 | */ |
072bb0aa MG |
2868 | if (objp) { |
2869 | STATS_INC_ALLOCHIT(cachep); | |
2870 | goto out; | |
2871 | } | |
2872 | force_refill = true; | |
1da177e4 | 2873 | } |
072bb0aa MG |
2874 | |
2875 | STATS_INC_ALLOCMISS(cachep); | |
2876 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2877 | /* | |
2878 | * the 'ac' may be updated by cache_alloc_refill(), | |
2879 | * and kmemleak_erase() requires its correct value. | |
2880 | */ | |
2881 | ac = cpu_cache_get(cachep); | |
2882 | ||
2883 | out: | |
d5cff635 CM |
2884 | /* |
2885 | * To avoid a false negative, if an object that is in one of the | |
2886 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2887 | * treat the array pointers as a reference to the object. | |
2888 | */ | |
f3d8b53a O |
2889 | if (objp) |
2890 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2891 | return objp; |
2892 | } | |
2893 | ||
e498be7d | 2894 | #ifdef CONFIG_NUMA |
c61afb18 | 2895 | /* |
2ad654bc | 2896 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2897 | * |
2898 | * If we are in_interrupt, then process context, including cpusets and | |
2899 | * mempolicy, may not apply and should not be used for allocation policy. | |
2900 | */ | |
2901 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2902 | { | |
2903 | int nid_alloc, nid_here; | |
2904 | ||
765c4507 | 2905 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 2906 | return NULL; |
7d6e6d09 | 2907 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 2908 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 2909 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 2910 | else if (current->mempolicy) |
2a389610 | 2911 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 2912 | if (nid_alloc != nid_here) |
8b98c169 | 2913 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
2914 | return NULL; |
2915 | } | |
2916 | ||
765c4507 CL |
2917 | /* |
2918 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 2919 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 2920 | * available node for available objects. If that fails then we |
3c517a61 CL |
2921 | * perform an allocation without specifying a node. This allows the page |
2922 | * allocator to do its reclaim / fallback magic. We then insert the | |
2923 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 2924 | */ |
8c8cc2c1 | 2925 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 2926 | { |
8c8cc2c1 PE |
2927 | struct zonelist *zonelist; |
2928 | gfp_t local_flags; | |
dd1a239f | 2929 | struct zoneref *z; |
54a6eb5c MG |
2930 | struct zone *zone; |
2931 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 2932 | void *obj = NULL; |
3c517a61 | 2933 | int nid; |
cc9a6c87 | 2934 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
2935 | |
2936 | if (flags & __GFP_THISNODE) | |
2937 | return NULL; | |
2938 | ||
6cb06229 | 2939 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 2940 | |
cc9a6c87 | 2941 | retry_cpuset: |
d26914d1 | 2942 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2943 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 2944 | |
3c517a61 CL |
2945 | retry: |
2946 | /* | |
2947 | * Look through allowed nodes for objects available | |
2948 | * from existing per node queues. | |
2949 | */ | |
54a6eb5c MG |
2950 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
2951 | nid = zone_to_nid(zone); | |
aedb0eb1 | 2952 | |
061d7074 | 2953 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
2954 | get_node(cache, nid) && |
2955 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 2956 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 2957 | gfp_exact_node(flags), nid); |
481c5346 CL |
2958 | if (obj) |
2959 | break; | |
2960 | } | |
3c517a61 CL |
2961 | } |
2962 | ||
cfce6604 | 2963 | if (!obj) { |
3c517a61 CL |
2964 | /* |
2965 | * This allocation will be performed within the constraints | |
2966 | * of the current cpuset / memory policy requirements. | |
2967 | * We may trigger various forms of reclaim on the allowed | |
2968 | * set and go into memory reserves if necessary. | |
2969 | */ | |
0c3aa83e JK |
2970 | struct page *page; |
2971 | ||
d0164adc | 2972 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
2973 | local_irq_enable(); |
2974 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 2975 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 2976 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 2977 | local_irq_disable(); |
0c3aa83e | 2978 | if (page) { |
3c517a61 CL |
2979 | /* |
2980 | * Insert into the appropriate per node queues | |
2981 | */ | |
0c3aa83e JK |
2982 | nid = page_to_nid(page); |
2983 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 2984 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 2985 | gfp_exact_node(flags), nid); |
3c517a61 CL |
2986 | if (!obj) |
2987 | /* | |
2988 | * Another processor may allocate the | |
2989 | * objects in the slab since we are | |
2990 | * not holding any locks. | |
2991 | */ | |
2992 | goto retry; | |
2993 | } else { | |
b6a60451 | 2994 | /* cache_grow already freed obj */ |
3c517a61 CL |
2995 | obj = NULL; |
2996 | } | |
2997 | } | |
aedb0eb1 | 2998 | } |
cc9a6c87 | 2999 | |
d26914d1 | 3000 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3001 | goto retry_cpuset; |
765c4507 CL |
3002 | return obj; |
3003 | } | |
3004 | ||
e498be7d CL |
3005 | /* |
3006 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3007 | */ |
8b98c169 | 3008 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3009 | int nodeid) |
e498be7d | 3010 | { |
8456a648 | 3011 | struct page *page; |
ce8eb6c4 | 3012 | struct kmem_cache_node *n; |
b28a02de | 3013 | void *obj; |
b28a02de PE |
3014 | int x; |
3015 | ||
7c3fbbdd | 3016 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3017 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3018 | BUG_ON(!n); |
b28a02de | 3019 | |
a737b3e2 | 3020 | retry: |
ca3b9b91 | 3021 | check_irq_off(); |
ce8eb6c4 | 3022 | spin_lock(&n->list_lock); |
7aa0d227 GT |
3023 | page = get_first_slab(n); |
3024 | if (!page) | |
3025 | goto must_grow; | |
b28a02de | 3026 | |
b28a02de | 3027 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3028 | |
3029 | STATS_INC_NODEALLOCS(cachep); | |
3030 | STATS_INC_ACTIVE(cachep); | |
3031 | STATS_SET_HIGH(cachep); | |
3032 | ||
8456a648 | 3033 | BUG_ON(page->active == cachep->num); |
b28a02de | 3034 | |
260b61dd | 3035 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3036 | n->free_objects--; |
b28a02de | 3037 | /* move slabp to correct slabp list: */ |
8456a648 | 3038 | list_del(&page->lru); |
b28a02de | 3039 | |
8456a648 JK |
3040 | if (page->active == cachep->num) |
3041 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3042 | else |
8456a648 | 3043 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3044 | |
ce8eb6c4 | 3045 | spin_unlock(&n->list_lock); |
b28a02de | 3046 | goto done; |
e498be7d | 3047 | |
a737b3e2 | 3048 | must_grow: |
ce8eb6c4 | 3049 | spin_unlock(&n->list_lock); |
4167e9b2 | 3050 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3051 | if (x) |
3052 | goto retry; | |
1da177e4 | 3053 | |
8c8cc2c1 | 3054 | return fallback_alloc(cachep, flags); |
e498be7d | 3055 | |
a737b3e2 | 3056 | done: |
b28a02de | 3057 | return obj; |
e498be7d | 3058 | } |
8c8cc2c1 | 3059 | |
8c8cc2c1 | 3060 | static __always_inline void * |
48356303 | 3061 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3062 | unsigned long caller) |
8c8cc2c1 PE |
3063 | { |
3064 | unsigned long save_flags; | |
3065 | void *ptr; | |
7d6e6d09 | 3066 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3067 | |
dcce284a | 3068 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3069 | cachep = slab_pre_alloc_hook(cachep, flags); |
3070 | if (unlikely(!cachep)) | |
824ebef1 AM |
3071 | return NULL; |
3072 | ||
8c8cc2c1 PE |
3073 | cache_alloc_debugcheck_before(cachep, flags); |
3074 | local_irq_save(save_flags); | |
3075 | ||
eacbbae3 | 3076 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3077 | nodeid = slab_node; |
8c8cc2c1 | 3078 | |
18bf8541 | 3079 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3080 | /* Node not bootstrapped yet */ |
3081 | ptr = fallback_alloc(cachep, flags); | |
3082 | goto out; | |
3083 | } | |
3084 | ||
7d6e6d09 | 3085 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3086 | /* |
3087 | * Use the locally cached objects if possible. | |
3088 | * However ____cache_alloc does not allow fallback | |
3089 | * to other nodes. It may fail while we still have | |
3090 | * objects on other nodes available. | |
3091 | */ | |
3092 | ptr = ____cache_alloc(cachep, flags); | |
3093 | if (ptr) | |
3094 | goto out; | |
3095 | } | |
3096 | /* ___cache_alloc_node can fall back to other nodes */ | |
3097 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3098 | out: | |
3099 | local_irq_restore(save_flags); | |
3100 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3101 | ||
d5e3ed66 JDB |
3102 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3103 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3104 | |
d5e3ed66 | 3105 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3106 | return ptr; |
3107 | } | |
3108 | ||
3109 | static __always_inline void * | |
3110 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3111 | { | |
3112 | void *objp; | |
3113 | ||
2ad654bc | 3114 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3115 | objp = alternate_node_alloc(cache, flags); |
3116 | if (objp) | |
3117 | goto out; | |
3118 | } | |
3119 | objp = ____cache_alloc(cache, flags); | |
3120 | ||
3121 | /* | |
3122 | * We may just have run out of memory on the local node. | |
3123 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3124 | */ | |
7d6e6d09 LS |
3125 | if (!objp) |
3126 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3127 | |
3128 | out: | |
3129 | return objp; | |
3130 | } | |
3131 | #else | |
3132 | ||
3133 | static __always_inline void * | |
3134 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3135 | { | |
3136 | return ____cache_alloc(cachep, flags); | |
3137 | } | |
3138 | ||
3139 | #endif /* CONFIG_NUMA */ | |
3140 | ||
3141 | static __always_inline void * | |
48356303 | 3142 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3143 | { |
3144 | unsigned long save_flags; | |
3145 | void *objp; | |
3146 | ||
dcce284a | 3147 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3148 | cachep = slab_pre_alloc_hook(cachep, flags); |
3149 | if (unlikely(!cachep)) | |
824ebef1 AM |
3150 | return NULL; |
3151 | ||
8c8cc2c1 PE |
3152 | cache_alloc_debugcheck_before(cachep, flags); |
3153 | local_irq_save(save_flags); | |
3154 | objp = __do_cache_alloc(cachep, flags); | |
3155 | local_irq_restore(save_flags); | |
3156 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3157 | prefetchw(objp); | |
3158 | ||
d5e3ed66 JDB |
3159 | if (unlikely(flags & __GFP_ZERO) && objp) |
3160 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3161 | |
d5e3ed66 | 3162 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3163 | return objp; |
3164 | } | |
e498be7d CL |
3165 | |
3166 | /* | |
5f0985bb | 3167 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3168 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3169 | */ |
97654dfa JK |
3170 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3171 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3172 | { |
3173 | int i; | |
25c063fb | 3174 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3175 | |
3176 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3177 | void *objp; |
8456a648 | 3178 | struct page *page; |
1da177e4 | 3179 | |
072bb0aa MG |
3180 | clear_obj_pfmemalloc(&objpp[i]); |
3181 | objp = objpp[i]; | |
3182 | ||
8456a648 | 3183 | page = virt_to_head_page(objp); |
8456a648 | 3184 | list_del(&page->lru); |
ff69416e | 3185 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3186 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3187 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3188 | n->free_objects++; |
1da177e4 LT |
3189 | |
3190 | /* fixup slab chains */ | |
8456a648 | 3191 | if (page->active == 0) { |
ce8eb6c4 CL |
3192 | if (n->free_objects > n->free_limit) { |
3193 | n->free_objects -= cachep->num; | |
97654dfa | 3194 | list_add_tail(&page->lru, list); |
1da177e4 | 3195 | } else { |
8456a648 | 3196 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3197 | } |
3198 | } else { | |
3199 | /* Unconditionally move a slab to the end of the | |
3200 | * partial list on free - maximum time for the | |
3201 | * other objects to be freed, too. | |
3202 | */ | |
8456a648 | 3203 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3204 | } |
3205 | } | |
3206 | } | |
3207 | ||
343e0d7a | 3208 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3209 | { |
3210 | int batchcount; | |
ce8eb6c4 | 3211 | struct kmem_cache_node *n; |
7d6e6d09 | 3212 | int node = numa_mem_id(); |
97654dfa | 3213 | LIST_HEAD(list); |
1da177e4 LT |
3214 | |
3215 | batchcount = ac->batchcount; | |
260b61dd | 3216 | |
1da177e4 | 3217 | check_irq_off(); |
18bf8541 | 3218 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3219 | spin_lock(&n->list_lock); |
3220 | if (n->shared) { | |
3221 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3222 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3223 | if (max) { |
3224 | if (batchcount > max) | |
3225 | batchcount = max; | |
e498be7d | 3226 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3227 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3228 | shared_array->avail += batchcount; |
3229 | goto free_done; | |
3230 | } | |
3231 | } | |
3232 | ||
97654dfa | 3233 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3234 | free_done: |
1da177e4 LT |
3235 | #if STATS |
3236 | { | |
3237 | int i = 0; | |
73c0219d | 3238 | struct page *page; |
1da177e4 | 3239 | |
73c0219d | 3240 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3241 | BUG_ON(page->active); |
1da177e4 LT |
3242 | |
3243 | i++; | |
1da177e4 LT |
3244 | } |
3245 | STATS_SET_FREEABLE(cachep, i); | |
3246 | } | |
3247 | #endif | |
ce8eb6c4 | 3248 | spin_unlock(&n->list_lock); |
97654dfa | 3249 | slabs_destroy(cachep, &list); |
1da177e4 | 3250 | ac->avail -= batchcount; |
a737b3e2 | 3251 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3252 | } |
3253 | ||
3254 | /* | |
a737b3e2 AM |
3255 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3256 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3257 | */ |
a947eb95 | 3258 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3259 | unsigned long caller) |
1da177e4 | 3260 | { |
9a2dba4b | 3261 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3262 | |
3263 | check_irq_off(); | |
d5cff635 | 3264 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3265 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3266 | |
8c138bc0 | 3267 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3268 | |
1807a1aa SS |
3269 | /* |
3270 | * Skip calling cache_free_alien() when the platform is not numa. | |
3271 | * This will avoid cache misses that happen while accessing slabp (which | |
3272 | * is per page memory reference) to get nodeid. Instead use a global | |
3273 | * variable to skip the call, which is mostly likely to be present in | |
3274 | * the cache. | |
3275 | */ | |
b6e68bc1 | 3276 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3277 | return; |
3278 | ||
3d880194 | 3279 | if (ac->avail < ac->limit) { |
1da177e4 | 3280 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3281 | } else { |
3282 | STATS_INC_FREEMISS(cachep); | |
3283 | cache_flusharray(cachep, ac); | |
1da177e4 | 3284 | } |
42c8c99c | 3285 | |
072bb0aa | 3286 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3287 | } |
3288 | ||
3289 | /** | |
3290 | * kmem_cache_alloc - Allocate an object | |
3291 | * @cachep: The cache to allocate from. | |
3292 | * @flags: See kmalloc(). | |
3293 | * | |
3294 | * Allocate an object from this cache. The flags are only relevant | |
3295 | * if the cache has no available objects. | |
3296 | */ | |
343e0d7a | 3297 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3298 | { |
48356303 | 3299 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3300 | |
ca2b84cb | 3301 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3302 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3303 | |
3304 | return ret; | |
1da177e4 LT |
3305 | } |
3306 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3307 | ||
7b0501dd JDB |
3308 | static __always_inline void |
3309 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3310 | size_t size, void **p, unsigned long caller) | |
3311 | { | |
3312 | size_t i; | |
3313 | ||
3314 | for (i = 0; i < size; i++) | |
3315 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3316 | } | |
3317 | ||
865762a8 | 3318 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3319 | void **p) |
484748f0 | 3320 | { |
2a777eac JDB |
3321 | size_t i; |
3322 | ||
3323 | s = slab_pre_alloc_hook(s, flags); | |
3324 | if (!s) | |
3325 | return 0; | |
3326 | ||
3327 | cache_alloc_debugcheck_before(s, flags); | |
3328 | ||
3329 | local_irq_disable(); | |
3330 | for (i = 0; i < size; i++) { | |
3331 | void *objp = __do_cache_alloc(s, flags); | |
3332 | ||
2a777eac JDB |
3333 | if (unlikely(!objp)) |
3334 | goto error; | |
3335 | p[i] = objp; | |
3336 | } | |
3337 | local_irq_enable(); | |
3338 | ||
7b0501dd JDB |
3339 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3340 | ||
2a777eac JDB |
3341 | /* Clear memory outside IRQ disabled section */ |
3342 | if (unlikely(flags & __GFP_ZERO)) | |
3343 | for (i = 0; i < size; i++) | |
3344 | memset(p[i], 0, s->object_size); | |
3345 | ||
3346 | slab_post_alloc_hook(s, flags, size, p); | |
3347 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3348 | return size; | |
3349 | error: | |
3350 | local_irq_enable(); | |
7b0501dd | 3351 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3352 | slab_post_alloc_hook(s, flags, i, p); |
3353 | __kmem_cache_free_bulk(s, i, p); | |
3354 | return 0; | |
484748f0 CL |
3355 | } |
3356 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3357 | ||
0f24f128 | 3358 | #ifdef CONFIG_TRACING |
85beb586 | 3359 | void * |
4052147c | 3360 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3361 | { |
85beb586 SR |
3362 | void *ret; |
3363 | ||
48356303 | 3364 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3365 | |
3366 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3367 | size, cachep->size, flags); |
85beb586 | 3368 | return ret; |
36555751 | 3369 | } |
85beb586 | 3370 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3371 | #endif |
3372 | ||
1da177e4 | 3373 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3374 | /** |
3375 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3376 | * @cachep: The cache to allocate from. | |
3377 | * @flags: See kmalloc(). | |
3378 | * @nodeid: node number of the target node. | |
3379 | * | |
3380 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3381 | * node, which can improve the performance for cpu bound structures. | |
3382 | * | |
3383 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3384 | */ | |
8b98c169 CH |
3385 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3386 | { | |
48356303 | 3387 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3388 | |
ca2b84cb | 3389 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3390 | cachep->object_size, cachep->size, |
ca2b84cb | 3391 | flags, nodeid); |
36555751 EGM |
3392 | |
3393 | return ret; | |
8b98c169 | 3394 | } |
1da177e4 LT |
3395 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3396 | ||
0f24f128 | 3397 | #ifdef CONFIG_TRACING |
4052147c | 3398 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3399 | gfp_t flags, |
4052147c EG |
3400 | int nodeid, |
3401 | size_t size) | |
36555751 | 3402 | { |
85beb586 SR |
3403 | void *ret; |
3404 | ||
592f4145 | 3405 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3406 | |
85beb586 | 3407 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3408 | size, cachep->size, |
85beb586 SR |
3409 | flags, nodeid); |
3410 | return ret; | |
36555751 | 3411 | } |
85beb586 | 3412 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3413 | #endif |
3414 | ||
8b98c169 | 3415 | static __always_inline void * |
7c0cb9c6 | 3416 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3417 | { |
343e0d7a | 3418 | struct kmem_cache *cachep; |
97e2bde4 | 3419 | |
2c59dd65 | 3420 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3421 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3422 | return cachep; | |
4052147c | 3423 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3424 | } |
8b98c169 | 3425 | |
8b98c169 CH |
3426 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3427 | { | |
7c0cb9c6 | 3428 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3429 | } |
dbe5e69d | 3430 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3431 | |
3432 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3433 | int node, unsigned long caller) |
8b98c169 | 3434 | { |
7c0cb9c6 | 3435 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3436 | } |
3437 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3438 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3439 | |
3440 | /** | |
800590f5 | 3441 | * __do_kmalloc - allocate memory |
1da177e4 | 3442 | * @size: how many bytes of memory are required. |
800590f5 | 3443 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3444 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3445 | */ |
7fd6b141 | 3446 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3447 | unsigned long caller) |
1da177e4 | 3448 | { |
343e0d7a | 3449 | struct kmem_cache *cachep; |
36555751 | 3450 | void *ret; |
1da177e4 | 3451 | |
2c59dd65 | 3452 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3453 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3454 | return cachep; | |
48356303 | 3455 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3456 | |
7c0cb9c6 | 3457 | trace_kmalloc(caller, ret, |
3b0efdfa | 3458 | size, cachep->size, flags); |
36555751 EGM |
3459 | |
3460 | return ret; | |
7fd6b141 PE |
3461 | } |
3462 | ||
7fd6b141 PE |
3463 | void *__kmalloc(size_t size, gfp_t flags) |
3464 | { | |
7c0cb9c6 | 3465 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3466 | } |
3467 | EXPORT_SYMBOL(__kmalloc); | |
3468 | ||
ce71e27c | 3469 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3470 | { |
7c0cb9c6 | 3471 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3472 | } |
3473 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3474 | |
1da177e4 LT |
3475 | /** |
3476 | * kmem_cache_free - Deallocate an object | |
3477 | * @cachep: The cache the allocation was from. | |
3478 | * @objp: The previously allocated object. | |
3479 | * | |
3480 | * Free an object which was previously allocated from this | |
3481 | * cache. | |
3482 | */ | |
343e0d7a | 3483 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3484 | { |
3485 | unsigned long flags; | |
b9ce5ef4 GC |
3486 | cachep = cache_from_obj(cachep, objp); |
3487 | if (!cachep) | |
3488 | return; | |
1da177e4 LT |
3489 | |
3490 | local_irq_save(flags); | |
d97d476b | 3491 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3492 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3493 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3494 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3495 | local_irq_restore(flags); |
36555751 | 3496 | |
ca2b84cb | 3497 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3498 | } |
3499 | EXPORT_SYMBOL(kmem_cache_free); | |
3500 | ||
e6cdb58d JDB |
3501 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3502 | { | |
3503 | struct kmem_cache *s; | |
3504 | size_t i; | |
3505 | ||
3506 | local_irq_disable(); | |
3507 | for (i = 0; i < size; i++) { | |
3508 | void *objp = p[i]; | |
3509 | ||
ca257195 JDB |
3510 | if (!orig_s) /* called via kfree_bulk */ |
3511 | s = virt_to_cache(objp); | |
3512 | else | |
3513 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3514 | |
3515 | debug_check_no_locks_freed(objp, s->object_size); | |
3516 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3517 | debug_check_no_obj_freed(objp, s->object_size); | |
3518 | ||
3519 | __cache_free(s, objp, _RET_IP_); | |
3520 | } | |
3521 | local_irq_enable(); | |
3522 | ||
3523 | /* FIXME: add tracing */ | |
3524 | } | |
3525 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3526 | ||
1da177e4 LT |
3527 | /** |
3528 | * kfree - free previously allocated memory | |
3529 | * @objp: pointer returned by kmalloc. | |
3530 | * | |
80e93eff PE |
3531 | * If @objp is NULL, no operation is performed. |
3532 | * | |
1da177e4 LT |
3533 | * Don't free memory not originally allocated by kmalloc() |
3534 | * or you will run into trouble. | |
3535 | */ | |
3536 | void kfree(const void *objp) | |
3537 | { | |
343e0d7a | 3538 | struct kmem_cache *c; |
1da177e4 LT |
3539 | unsigned long flags; |
3540 | ||
2121db74 PE |
3541 | trace_kfree(_RET_IP_, objp); |
3542 | ||
6cb8f913 | 3543 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3544 | return; |
3545 | local_irq_save(flags); | |
3546 | kfree_debugcheck(objp); | |
6ed5eb22 | 3547 | c = virt_to_cache(objp); |
8c138bc0 CL |
3548 | debug_check_no_locks_freed(objp, c->object_size); |
3549 | ||
3550 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3551 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3552 | local_irq_restore(flags); |
3553 | } | |
3554 | EXPORT_SYMBOL(kfree); | |
3555 | ||
e498be7d | 3556 | /* |
ce8eb6c4 | 3557 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3558 | */ |
5f0985bb | 3559 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3560 | { |
3561 | int node; | |
ce8eb6c4 | 3562 | struct kmem_cache_node *n; |
cafeb02e | 3563 | struct array_cache *new_shared; |
c8522a3a | 3564 | struct alien_cache **new_alien = NULL; |
e498be7d | 3565 | |
9c09a95c | 3566 | for_each_online_node(node) { |
cafeb02e | 3567 | |
b455def2 L |
3568 | if (use_alien_caches) { |
3569 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3570 | if (!new_alien) | |
3571 | goto fail; | |
3572 | } | |
cafeb02e | 3573 | |
63109846 ED |
3574 | new_shared = NULL; |
3575 | if (cachep->shared) { | |
3576 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3577 | cachep->shared*cachep->batchcount, |
83b519e8 | 3578 | 0xbaadf00d, gfp); |
63109846 ED |
3579 | if (!new_shared) { |
3580 | free_alien_cache(new_alien); | |
3581 | goto fail; | |
3582 | } | |
0718dc2a | 3583 | } |
cafeb02e | 3584 | |
18bf8541 | 3585 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3586 | if (n) { |
3587 | struct array_cache *shared = n->shared; | |
97654dfa | 3588 | LIST_HEAD(list); |
cafeb02e | 3589 | |
ce8eb6c4 | 3590 | spin_lock_irq(&n->list_lock); |
e498be7d | 3591 | |
cafeb02e | 3592 | if (shared) |
0718dc2a | 3593 | free_block(cachep, shared->entry, |
97654dfa | 3594 | shared->avail, node, &list); |
e498be7d | 3595 | |
ce8eb6c4 CL |
3596 | n->shared = new_shared; |
3597 | if (!n->alien) { | |
3598 | n->alien = new_alien; | |
e498be7d CL |
3599 | new_alien = NULL; |
3600 | } | |
ce8eb6c4 | 3601 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3602 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3603 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3604 | slabs_destroy(cachep, &list); |
cafeb02e | 3605 | kfree(shared); |
e498be7d CL |
3606 | free_alien_cache(new_alien); |
3607 | continue; | |
3608 | } | |
ce8eb6c4 CL |
3609 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3610 | if (!n) { | |
0718dc2a CL |
3611 | free_alien_cache(new_alien); |
3612 | kfree(new_shared); | |
e498be7d | 3613 | goto fail; |
0718dc2a | 3614 | } |
e498be7d | 3615 | |
ce8eb6c4 | 3616 | kmem_cache_node_init(n); |
5f0985bb JZ |
3617 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3618 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3619 | n->shared = new_shared; |
3620 | n->alien = new_alien; | |
3621 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3622 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3623 | cachep->node[node] = n; |
e498be7d | 3624 | } |
cafeb02e | 3625 | return 0; |
0718dc2a | 3626 | |
a737b3e2 | 3627 | fail: |
3b0efdfa | 3628 | if (!cachep->list.next) { |
0718dc2a CL |
3629 | /* Cache is not active yet. Roll back what we did */ |
3630 | node--; | |
3631 | while (node >= 0) { | |
18bf8541 CL |
3632 | n = get_node(cachep, node); |
3633 | if (n) { | |
ce8eb6c4 CL |
3634 | kfree(n->shared); |
3635 | free_alien_cache(n->alien); | |
3636 | kfree(n); | |
6a67368c | 3637 | cachep->node[node] = NULL; |
0718dc2a CL |
3638 | } |
3639 | node--; | |
3640 | } | |
3641 | } | |
cafeb02e | 3642 | return -ENOMEM; |
e498be7d CL |
3643 | } |
3644 | ||
18004c5d | 3645 | /* Always called with the slab_mutex held */ |
943a451a | 3646 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3647 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3648 | { |
bf0dea23 JK |
3649 | struct array_cache __percpu *cpu_cache, *prev; |
3650 | int cpu; | |
1da177e4 | 3651 | |
bf0dea23 JK |
3652 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3653 | if (!cpu_cache) | |
d2e7b7d0 SS |
3654 | return -ENOMEM; |
3655 | ||
bf0dea23 JK |
3656 | prev = cachep->cpu_cache; |
3657 | cachep->cpu_cache = cpu_cache; | |
3658 | kick_all_cpus_sync(); | |
e498be7d | 3659 | |
1da177e4 | 3660 | check_irq_on(); |
1da177e4 LT |
3661 | cachep->batchcount = batchcount; |
3662 | cachep->limit = limit; | |
e498be7d | 3663 | cachep->shared = shared; |
1da177e4 | 3664 | |
bf0dea23 JK |
3665 | if (!prev) |
3666 | goto alloc_node; | |
3667 | ||
3668 | for_each_online_cpu(cpu) { | |
97654dfa | 3669 | LIST_HEAD(list); |
18bf8541 CL |
3670 | int node; |
3671 | struct kmem_cache_node *n; | |
bf0dea23 | 3672 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3673 | |
bf0dea23 | 3674 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3675 | n = get_node(cachep, node); |
3676 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3677 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3678 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3679 | slabs_destroy(cachep, &list); |
1da177e4 | 3680 | } |
bf0dea23 JK |
3681 | free_percpu(prev); |
3682 | ||
3683 | alloc_node: | |
5f0985bb | 3684 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3685 | } |
3686 | ||
943a451a GC |
3687 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3688 | int batchcount, int shared, gfp_t gfp) | |
3689 | { | |
3690 | int ret; | |
426589f5 | 3691 | struct kmem_cache *c; |
943a451a GC |
3692 | |
3693 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3694 | ||
3695 | if (slab_state < FULL) | |
3696 | return ret; | |
3697 | ||
3698 | if ((ret < 0) || !is_root_cache(cachep)) | |
3699 | return ret; | |
3700 | ||
426589f5 VD |
3701 | lockdep_assert_held(&slab_mutex); |
3702 | for_each_memcg_cache(c, cachep) { | |
3703 | /* return value determined by the root cache only */ | |
3704 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3705 | } |
3706 | ||
3707 | return ret; | |
3708 | } | |
3709 | ||
18004c5d | 3710 | /* Called with slab_mutex held always */ |
83b519e8 | 3711 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3712 | { |
3713 | int err; | |
943a451a GC |
3714 | int limit = 0; |
3715 | int shared = 0; | |
3716 | int batchcount = 0; | |
3717 | ||
3718 | if (!is_root_cache(cachep)) { | |
3719 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3720 | limit = root->limit; | |
3721 | shared = root->shared; | |
3722 | batchcount = root->batchcount; | |
3723 | } | |
1da177e4 | 3724 | |
943a451a GC |
3725 | if (limit && shared && batchcount) |
3726 | goto skip_setup; | |
a737b3e2 AM |
3727 | /* |
3728 | * The head array serves three purposes: | |
1da177e4 LT |
3729 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3730 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3731 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3732 | * bufctl chains: array operations are cheaper. |
3733 | * The numbers are guessed, we should auto-tune as described by | |
3734 | * Bonwick. | |
3735 | */ | |
3b0efdfa | 3736 | if (cachep->size > 131072) |
1da177e4 | 3737 | limit = 1; |
3b0efdfa | 3738 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3739 | limit = 8; |
3b0efdfa | 3740 | else if (cachep->size > 1024) |
1da177e4 | 3741 | limit = 24; |
3b0efdfa | 3742 | else if (cachep->size > 256) |
1da177e4 LT |
3743 | limit = 54; |
3744 | else | |
3745 | limit = 120; | |
3746 | ||
a737b3e2 AM |
3747 | /* |
3748 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3749 | * allocation behaviour: Most allocs on one cpu, most free operations |
3750 | * on another cpu. For these cases, an efficient object passing between | |
3751 | * cpus is necessary. This is provided by a shared array. The array | |
3752 | * replaces Bonwick's magazine layer. | |
3753 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3754 | * to a larger limit. Thus disabled by default. | |
3755 | */ | |
3756 | shared = 0; | |
3b0efdfa | 3757 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3758 | shared = 8; |
1da177e4 LT |
3759 | |
3760 | #if DEBUG | |
a737b3e2 AM |
3761 | /* |
3762 | * With debugging enabled, large batchcount lead to excessively long | |
3763 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3764 | */ |
3765 | if (limit > 32) | |
3766 | limit = 32; | |
3767 | #endif | |
943a451a GC |
3768 | batchcount = (limit + 1) / 2; |
3769 | skip_setup: | |
3770 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3771 | if (err) |
3772 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3773 | cachep->name, -err); |
2ed3a4ef | 3774 | return err; |
1da177e4 LT |
3775 | } |
3776 | ||
1b55253a | 3777 | /* |
ce8eb6c4 CL |
3778 | * Drain an array if it contains any elements taking the node lock only if |
3779 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3780 | * if drain_array() is used on the shared array. |
1b55253a | 3781 | */ |
ce8eb6c4 | 3782 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3783 | struct array_cache *ac, int force, int node) |
1da177e4 | 3784 | { |
97654dfa | 3785 | LIST_HEAD(list); |
1da177e4 LT |
3786 | int tofree; |
3787 | ||
1b55253a CL |
3788 | if (!ac || !ac->avail) |
3789 | return; | |
1da177e4 LT |
3790 | if (ac->touched && !force) { |
3791 | ac->touched = 0; | |
b18e7e65 | 3792 | } else { |
ce8eb6c4 | 3793 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3794 | if (ac->avail) { |
3795 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3796 | if (tofree > ac->avail) | |
3797 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3798 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3799 | ac->avail -= tofree; |
3800 | memmove(ac->entry, &(ac->entry[tofree]), | |
3801 | sizeof(void *) * ac->avail); | |
3802 | } | |
ce8eb6c4 | 3803 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3804 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3805 | } |
3806 | } | |
3807 | ||
3808 | /** | |
3809 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3810 | * @w: work descriptor |
1da177e4 LT |
3811 | * |
3812 | * Called from workqueue/eventd every few seconds. | |
3813 | * Purpose: | |
3814 | * - clear the per-cpu caches for this CPU. | |
3815 | * - return freeable pages to the main free memory pool. | |
3816 | * | |
a737b3e2 AM |
3817 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3818 | * again on the next iteration. | |
1da177e4 | 3819 | */ |
7c5cae36 | 3820 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3821 | { |
7a7c381d | 3822 | struct kmem_cache *searchp; |
ce8eb6c4 | 3823 | struct kmem_cache_node *n; |
7d6e6d09 | 3824 | int node = numa_mem_id(); |
bf6aede7 | 3825 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3826 | |
18004c5d | 3827 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3828 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3829 | goto out; |
1da177e4 | 3830 | |
18004c5d | 3831 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3832 | check_irq_on(); |
3833 | ||
35386e3b | 3834 | /* |
ce8eb6c4 | 3835 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3836 | * have established with reasonable certainty that |
3837 | * we can do some work if the lock was obtained. | |
3838 | */ | |
18bf8541 | 3839 | n = get_node(searchp, node); |
35386e3b | 3840 | |
ce8eb6c4 | 3841 | reap_alien(searchp, n); |
1da177e4 | 3842 | |
ce8eb6c4 | 3843 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3844 | |
35386e3b CL |
3845 | /* |
3846 | * These are racy checks but it does not matter | |
3847 | * if we skip one check or scan twice. | |
3848 | */ | |
ce8eb6c4 | 3849 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3850 | goto next; |
1da177e4 | 3851 | |
5f0985bb | 3852 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3853 | |
ce8eb6c4 | 3854 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3855 | |
ce8eb6c4 CL |
3856 | if (n->free_touched) |
3857 | n->free_touched = 0; | |
ed11d9eb CL |
3858 | else { |
3859 | int freed; | |
1da177e4 | 3860 | |
ce8eb6c4 | 3861 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3862 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3863 | STATS_ADD_REAPED(searchp, freed); | |
3864 | } | |
35386e3b | 3865 | next: |
1da177e4 LT |
3866 | cond_resched(); |
3867 | } | |
3868 | check_irq_on(); | |
18004c5d | 3869 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3870 | next_reap_node(); |
7c5cae36 | 3871 | out: |
a737b3e2 | 3872 | /* Set up the next iteration */ |
5f0985bb | 3873 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3874 | } |
3875 | ||
158a9624 | 3876 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3877 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3878 | { |
8456a648 | 3879 | struct page *page; |
b28a02de PE |
3880 | unsigned long active_objs; |
3881 | unsigned long num_objs; | |
3882 | unsigned long active_slabs = 0; | |
3883 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3884 | const char *name; |
1da177e4 | 3885 | char *error = NULL; |
e498be7d | 3886 | int node; |
ce8eb6c4 | 3887 | struct kmem_cache_node *n; |
1da177e4 | 3888 | |
1da177e4 LT |
3889 | active_objs = 0; |
3890 | num_slabs = 0; | |
18bf8541 | 3891 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3892 | |
ca3b9b91 | 3893 | check_irq_on(); |
ce8eb6c4 | 3894 | spin_lock_irq(&n->list_lock); |
e498be7d | 3895 | |
8456a648 JK |
3896 | list_for_each_entry(page, &n->slabs_full, lru) { |
3897 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3898 | error = "slabs_full accounting error"; |
3899 | active_objs += cachep->num; | |
3900 | active_slabs++; | |
3901 | } | |
8456a648 JK |
3902 | list_for_each_entry(page, &n->slabs_partial, lru) { |
3903 | if (page->active == cachep->num && !error) | |
106a74e1 | 3904 | error = "slabs_partial accounting error"; |
8456a648 | 3905 | if (!page->active && !error) |
106a74e1 | 3906 | error = "slabs_partial accounting error"; |
8456a648 | 3907 | active_objs += page->active; |
e498be7d CL |
3908 | active_slabs++; |
3909 | } | |
8456a648 JK |
3910 | list_for_each_entry(page, &n->slabs_free, lru) { |
3911 | if (page->active && !error) | |
106a74e1 | 3912 | error = "slabs_free accounting error"; |
e498be7d CL |
3913 | num_slabs++; |
3914 | } | |
ce8eb6c4 CL |
3915 | free_objects += n->free_objects; |
3916 | if (n->shared) | |
3917 | shared_avail += n->shared->avail; | |
e498be7d | 3918 | |
ce8eb6c4 | 3919 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 3920 | } |
b28a02de PE |
3921 | num_slabs += active_slabs; |
3922 | num_objs = num_slabs * cachep->num; | |
e498be7d | 3923 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
3924 | error = "free_objects accounting error"; |
3925 | ||
b28a02de | 3926 | name = cachep->name; |
1da177e4 LT |
3927 | if (error) |
3928 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
3929 | ||
0d7561c6 GC |
3930 | sinfo->active_objs = active_objs; |
3931 | sinfo->num_objs = num_objs; | |
3932 | sinfo->active_slabs = active_slabs; | |
3933 | sinfo->num_slabs = num_slabs; | |
3934 | sinfo->shared_avail = shared_avail; | |
3935 | sinfo->limit = cachep->limit; | |
3936 | sinfo->batchcount = cachep->batchcount; | |
3937 | sinfo->shared = cachep->shared; | |
3938 | sinfo->objects_per_slab = cachep->num; | |
3939 | sinfo->cache_order = cachep->gfporder; | |
3940 | } | |
3941 | ||
3942 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
3943 | { | |
1da177e4 | 3944 | #if STATS |
ce8eb6c4 | 3945 | { /* node stats */ |
1da177e4 LT |
3946 | unsigned long high = cachep->high_mark; |
3947 | unsigned long allocs = cachep->num_allocations; | |
3948 | unsigned long grown = cachep->grown; | |
3949 | unsigned long reaped = cachep->reaped; | |
3950 | unsigned long errors = cachep->errors; | |
3951 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 3952 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 3953 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 3954 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 3955 | |
e92dd4fd JP |
3956 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
3957 | "%4lu %4lu %4lu %4lu %4lu", | |
3958 | allocs, high, grown, | |
3959 | reaped, errors, max_freeable, node_allocs, | |
3960 | node_frees, overflows); | |
1da177e4 LT |
3961 | } |
3962 | /* cpu stats */ | |
3963 | { | |
3964 | unsigned long allochit = atomic_read(&cachep->allochit); | |
3965 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
3966 | unsigned long freehit = atomic_read(&cachep->freehit); | |
3967 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
3968 | ||
3969 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 3970 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
3971 | } |
3972 | #endif | |
1da177e4 LT |
3973 | } |
3974 | ||
1da177e4 LT |
3975 | #define MAX_SLABINFO_WRITE 128 |
3976 | /** | |
3977 | * slabinfo_write - Tuning for the slab allocator | |
3978 | * @file: unused | |
3979 | * @buffer: user buffer | |
3980 | * @count: data length | |
3981 | * @ppos: unused | |
3982 | */ | |
b7454ad3 | 3983 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 3984 | size_t count, loff_t *ppos) |
1da177e4 | 3985 | { |
b28a02de | 3986 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 3987 | int limit, batchcount, shared, res; |
7a7c381d | 3988 | struct kmem_cache *cachep; |
b28a02de | 3989 | |
1da177e4 LT |
3990 | if (count > MAX_SLABINFO_WRITE) |
3991 | return -EINVAL; | |
3992 | if (copy_from_user(&kbuf, buffer, count)) | |
3993 | return -EFAULT; | |
b28a02de | 3994 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
3995 | |
3996 | tmp = strchr(kbuf, ' '); | |
3997 | if (!tmp) | |
3998 | return -EINVAL; | |
3999 | *tmp = '\0'; | |
4000 | tmp++; | |
4001 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4002 | return -EINVAL; | |
4003 | ||
4004 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4005 | mutex_lock(&slab_mutex); |
1da177e4 | 4006 | res = -EINVAL; |
18004c5d | 4007 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4008 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4009 | if (limit < 1 || batchcount < 1 || |
4010 | batchcount > limit || shared < 0) { | |
e498be7d | 4011 | res = 0; |
1da177e4 | 4012 | } else { |
e498be7d | 4013 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4014 | batchcount, shared, |
4015 | GFP_KERNEL); | |
1da177e4 LT |
4016 | } |
4017 | break; | |
4018 | } | |
4019 | } | |
18004c5d | 4020 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4021 | if (res >= 0) |
4022 | res = count; | |
4023 | return res; | |
4024 | } | |
871751e2 AV |
4025 | |
4026 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4027 | ||
871751e2 AV |
4028 | static inline int add_caller(unsigned long *n, unsigned long v) |
4029 | { | |
4030 | unsigned long *p; | |
4031 | int l; | |
4032 | if (!v) | |
4033 | return 1; | |
4034 | l = n[1]; | |
4035 | p = n + 2; | |
4036 | while (l) { | |
4037 | int i = l/2; | |
4038 | unsigned long *q = p + 2 * i; | |
4039 | if (*q == v) { | |
4040 | q[1]++; | |
4041 | return 1; | |
4042 | } | |
4043 | if (*q > v) { | |
4044 | l = i; | |
4045 | } else { | |
4046 | p = q + 2; | |
4047 | l -= i + 1; | |
4048 | } | |
4049 | } | |
4050 | if (++n[1] == n[0]) | |
4051 | return 0; | |
4052 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4053 | p[0] = v; | |
4054 | p[1] = 1; | |
4055 | return 1; | |
4056 | } | |
4057 | ||
8456a648 JK |
4058 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4059 | struct page *page) | |
871751e2 AV |
4060 | { |
4061 | void *p; | |
d31676df JK |
4062 | int i, j; |
4063 | unsigned long v; | |
b1cb0982 | 4064 | |
871751e2 AV |
4065 | if (n[0] == n[1]) |
4066 | return; | |
8456a648 | 4067 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4068 | bool active = true; |
4069 | ||
4070 | for (j = page->active; j < c->num; j++) { | |
4071 | if (get_free_obj(page, j) == i) { | |
4072 | active = false; | |
4073 | break; | |
4074 | } | |
4075 | } | |
4076 | ||
4077 | if (!active) | |
871751e2 | 4078 | continue; |
b1cb0982 | 4079 | |
d31676df JK |
4080 | /* |
4081 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4082 | * mapping is established when actual object allocation and | |
4083 | * we could mistakenly access the unmapped object in the cpu | |
4084 | * cache. | |
4085 | */ | |
4086 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4087 | continue; | |
4088 | ||
4089 | if (!add_caller(n, v)) | |
871751e2 AV |
4090 | return; |
4091 | } | |
4092 | } | |
4093 | ||
4094 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4095 | { | |
4096 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4097 | unsigned long offset, size; |
9281acea | 4098 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4099 | |
a5c43dae | 4100 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4101 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4102 | if (modname[0]) |
871751e2 AV |
4103 | seq_printf(m, " [%s]", modname); |
4104 | return; | |
4105 | } | |
4106 | #endif | |
4107 | seq_printf(m, "%p", (void *)address); | |
4108 | } | |
4109 | ||
4110 | static int leaks_show(struct seq_file *m, void *p) | |
4111 | { | |
0672aa7c | 4112 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4113 | struct page *page; |
ce8eb6c4 | 4114 | struct kmem_cache_node *n; |
871751e2 | 4115 | const char *name; |
db845067 | 4116 | unsigned long *x = m->private; |
871751e2 AV |
4117 | int node; |
4118 | int i; | |
4119 | ||
4120 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4121 | return 0; | |
4122 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4123 | return 0; | |
4124 | ||
d31676df JK |
4125 | /* |
4126 | * Set store_user_clean and start to grab stored user information | |
4127 | * for all objects on this cache. If some alloc/free requests comes | |
4128 | * during the processing, information would be wrong so restart | |
4129 | * whole processing. | |
4130 | */ | |
4131 | do { | |
4132 | set_store_user_clean(cachep); | |
4133 | drain_cpu_caches(cachep); | |
4134 | ||
4135 | x[1] = 0; | |
871751e2 | 4136 | |
d31676df | 4137 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4138 | |
d31676df JK |
4139 | check_irq_on(); |
4140 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4141 | |
d31676df JK |
4142 | list_for_each_entry(page, &n->slabs_full, lru) |
4143 | handle_slab(x, cachep, page); | |
4144 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4145 | handle_slab(x, cachep, page); | |
4146 | spin_unlock_irq(&n->list_lock); | |
4147 | } | |
4148 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4149 | |
871751e2 | 4150 | name = cachep->name; |
db845067 | 4151 | if (x[0] == x[1]) { |
871751e2 | 4152 | /* Increase the buffer size */ |
18004c5d | 4153 | mutex_unlock(&slab_mutex); |
db845067 | 4154 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4155 | if (!m->private) { |
4156 | /* Too bad, we are really out */ | |
db845067 | 4157 | m->private = x; |
18004c5d | 4158 | mutex_lock(&slab_mutex); |
871751e2 AV |
4159 | return -ENOMEM; |
4160 | } | |
db845067 CL |
4161 | *(unsigned long *)m->private = x[0] * 2; |
4162 | kfree(x); | |
18004c5d | 4163 | mutex_lock(&slab_mutex); |
871751e2 AV |
4164 | /* Now make sure this entry will be retried */ |
4165 | m->count = m->size; | |
4166 | return 0; | |
4167 | } | |
db845067 CL |
4168 | for (i = 0; i < x[1]; i++) { |
4169 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4170 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4171 | seq_putc(m, '\n'); |
4172 | } | |
d2e7b7d0 | 4173 | |
871751e2 AV |
4174 | return 0; |
4175 | } | |
4176 | ||
a0ec95a8 | 4177 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4178 | .start = slab_start, |
276a2439 WL |
4179 | .next = slab_next, |
4180 | .stop = slab_stop, | |
871751e2 AV |
4181 | .show = leaks_show, |
4182 | }; | |
a0ec95a8 AD |
4183 | |
4184 | static int slabstats_open(struct inode *inode, struct file *file) | |
4185 | { | |
b208ce32 RJ |
4186 | unsigned long *n; |
4187 | ||
4188 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4189 | if (!n) | |
4190 | return -ENOMEM; | |
4191 | ||
4192 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4193 | ||
4194 | return 0; | |
a0ec95a8 AD |
4195 | } |
4196 | ||
4197 | static const struct file_operations proc_slabstats_operations = { | |
4198 | .open = slabstats_open, | |
4199 | .read = seq_read, | |
4200 | .llseek = seq_lseek, | |
4201 | .release = seq_release_private, | |
4202 | }; | |
4203 | #endif | |
4204 | ||
4205 | static int __init slab_proc_init(void) | |
4206 | { | |
4207 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4208 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4209 | #endif |
a0ec95a8 AD |
4210 | return 0; |
4211 | } | |
4212 | module_init(slab_proc_init); | |
1da177e4 LT |
4213 | #endif |
4214 | ||
00e145b6 MS |
4215 | /** |
4216 | * ksize - get the actual amount of memory allocated for a given object | |
4217 | * @objp: Pointer to the object | |
4218 | * | |
4219 | * kmalloc may internally round up allocations and return more memory | |
4220 | * than requested. ksize() can be used to determine the actual amount of | |
4221 | * memory allocated. The caller may use this additional memory, even though | |
4222 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4223 | * The caller must guarantee that objp points to a valid object previously | |
4224 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4225 | * must not be freed during the duration of the call. | |
4226 | */ | |
fd76bab2 | 4227 | size_t ksize(const void *objp) |
1da177e4 | 4228 | { |
ef8b4520 CL |
4229 | BUG_ON(!objp); |
4230 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4231 | return 0; |
1da177e4 | 4232 | |
8c138bc0 | 4233 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4234 | } |
b1aabecd | 4235 | EXPORT_SYMBOL(ksize); |