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 | ||
03787301 JK |
459 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
460 | { | |
461 | size_t freelist_size; | |
462 | ||
463 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
03787301 JK |
464 | if (align) |
465 | freelist_size = ALIGN(freelist_size, align); | |
466 | ||
467 | return freelist_size; | |
468 | } | |
469 | ||
9cef2e2b JK |
470 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
471 | size_t idx_size, size_t align) | |
1da177e4 | 472 | { |
9cef2e2b | 473 | int nr_objs; |
03787301 | 474 | size_t remained_size; |
9cef2e2b JK |
475 | size_t freelist_size; |
476 | ||
477 | /* | |
478 | * Ignore padding for the initial guess. The padding | |
479 | * is at most @align-1 bytes, and @buffer_size is at | |
480 | * least @align. In the worst case, this result will | |
481 | * be one greater than the number of objects that fit | |
482 | * into the memory allocation when taking the padding | |
483 | * into account. | |
484 | */ | |
249247b6 | 485 | nr_objs = slab_size / (buffer_size + idx_size); |
9cef2e2b JK |
486 | |
487 | /* | |
488 | * This calculated number will be either the right | |
489 | * amount, or one greater than what we want. | |
490 | */ | |
03787301 JK |
491 | remained_size = slab_size - nr_objs * buffer_size; |
492 | freelist_size = calculate_freelist_size(nr_objs, align); | |
493 | if (remained_size < freelist_size) | |
9cef2e2b JK |
494 | nr_objs--; |
495 | ||
496 | return nr_objs; | |
fbaccacf | 497 | } |
1da177e4 | 498 | |
a737b3e2 AM |
499 | /* |
500 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
501 | */ | |
fbaccacf SR |
502 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
503 | size_t align, int flags, size_t *left_over, | |
504 | unsigned int *num) | |
505 | { | |
506 | int nr_objs; | |
507 | size_t mgmt_size; | |
508 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 509 | |
fbaccacf SR |
510 | /* |
511 | * The slab management structure can be either off the slab or | |
512 | * on it. For the latter case, the memory allocated for a | |
513 | * slab is used for: | |
514 | * | |
12c61fe9 | 515 | * - One freelist_idx_t for each object |
fbaccacf SR |
516 | * - Padding to respect alignment of @align |
517 | * - @buffer_size bytes for each object | |
518 | * | |
519 | * If the slab management structure is off the slab, then the | |
520 | * alignment will already be calculated into the size. Because | |
521 | * the slabs are all pages aligned, the objects will be at the | |
522 | * correct alignment when allocated. | |
523 | */ | |
524 | if (flags & CFLGS_OFF_SLAB) { | |
525 | mgmt_size = 0; | |
526 | nr_objs = slab_size / buffer_size; | |
527 | ||
fbaccacf | 528 | } else { |
9cef2e2b | 529 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 530 | sizeof(freelist_idx_t), align); |
03787301 | 531 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
532 | } |
533 | *num = nr_objs; | |
534 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
535 | } |
536 | ||
f28510d3 | 537 | #if DEBUG |
d40cee24 | 538 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 539 | |
a737b3e2 AM |
540 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
541 | char *msg) | |
1da177e4 LT |
542 | { |
543 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 544 | function, cachep->name, msg); |
1da177e4 | 545 | dump_stack(); |
373d4d09 | 546 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 547 | } |
f28510d3 | 548 | #endif |
1da177e4 | 549 | |
3395ee05 PM |
550 | /* |
551 | * By default on NUMA we use alien caches to stage the freeing of | |
552 | * objects allocated from other nodes. This causes massive memory | |
553 | * inefficiencies when using fake NUMA setup to split memory into a | |
554 | * large number of small nodes, so it can be disabled on the command | |
555 | * line | |
556 | */ | |
557 | ||
558 | static int use_alien_caches __read_mostly = 1; | |
559 | static int __init noaliencache_setup(char *s) | |
560 | { | |
561 | use_alien_caches = 0; | |
562 | return 1; | |
563 | } | |
564 | __setup("noaliencache", noaliencache_setup); | |
565 | ||
3df1cccd DR |
566 | static int __init slab_max_order_setup(char *str) |
567 | { | |
568 | get_option(&str, &slab_max_order); | |
569 | slab_max_order = slab_max_order < 0 ? 0 : | |
570 | min(slab_max_order, MAX_ORDER - 1); | |
571 | slab_max_order_set = true; | |
572 | ||
573 | return 1; | |
574 | } | |
575 | __setup("slab_max_order=", slab_max_order_setup); | |
576 | ||
8fce4d8e CL |
577 | #ifdef CONFIG_NUMA |
578 | /* | |
579 | * Special reaping functions for NUMA systems called from cache_reap(). | |
580 | * These take care of doing round robin flushing of alien caches (containing | |
581 | * objects freed on different nodes from which they were allocated) and the | |
582 | * flushing of remote pcps by calling drain_node_pages. | |
583 | */ | |
1871e52c | 584 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
585 | |
586 | static void init_reap_node(int cpu) | |
587 | { | |
588 | int node; | |
589 | ||
7d6e6d09 | 590 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 591 | if (node == MAX_NUMNODES) |
442295c9 | 592 | node = first_node(node_online_map); |
8fce4d8e | 593 | |
1871e52c | 594 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
595 | } |
596 | ||
597 | static void next_reap_node(void) | |
598 | { | |
909ea964 | 599 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 600 | |
8fce4d8e CL |
601 | node = next_node(node, node_online_map); |
602 | if (unlikely(node >= MAX_NUMNODES)) | |
603 | node = first_node(node_online_map); | |
909ea964 | 604 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
605 | } |
606 | ||
607 | #else | |
608 | #define init_reap_node(cpu) do { } while (0) | |
609 | #define next_reap_node(void) do { } while (0) | |
610 | #endif | |
611 | ||
1da177e4 LT |
612 | /* |
613 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
614 | * via the workqueue/eventd. | |
615 | * Add the CPU number into the expiration time to minimize the possibility of | |
616 | * the CPUs getting into lockstep and contending for the global cache chain | |
617 | * lock. | |
618 | */ | |
0db0628d | 619 | static void start_cpu_timer(int cpu) |
1da177e4 | 620 | { |
1871e52c | 621 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
622 | |
623 | /* | |
624 | * When this gets called from do_initcalls via cpucache_init(), | |
625 | * init_workqueues() has already run, so keventd will be setup | |
626 | * at that time. | |
627 | */ | |
52bad64d | 628 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 629 | init_reap_node(cpu); |
203b42f7 | 630 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
631 | schedule_delayed_work_on(cpu, reap_work, |
632 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
633 | } |
634 | } | |
635 | ||
1fe00d50 | 636 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 637 | { |
d5cff635 CM |
638 | /* |
639 | * The array_cache structures contain pointers to free object. | |
25985edc | 640 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
641 | * cache the pointers are not cleared and they could be counted as |
642 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
643 | * not scan such objects. | |
644 | */ | |
1fe00d50 JK |
645 | kmemleak_no_scan(ac); |
646 | if (ac) { | |
647 | ac->avail = 0; | |
648 | ac->limit = limit; | |
649 | ac->batchcount = batch; | |
650 | ac->touched = 0; | |
1da177e4 | 651 | } |
1fe00d50 JK |
652 | } |
653 | ||
654 | static struct array_cache *alloc_arraycache(int node, int entries, | |
655 | int batchcount, gfp_t gfp) | |
656 | { | |
5e804789 | 657 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
658 | struct array_cache *ac = NULL; |
659 | ||
660 | ac = kmalloc_node(memsize, gfp, node); | |
661 | init_arraycache(ac, entries, batchcount); | |
662 | return ac; | |
1da177e4 LT |
663 | } |
664 | ||
8456a648 | 665 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 666 | { |
072bb0aa MG |
667 | return PageSlabPfmemalloc(page); |
668 | } | |
669 | ||
670 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
671 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
672 | struct array_cache *ac) | |
673 | { | |
18bf8541 | 674 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 675 | struct page *page; |
072bb0aa MG |
676 | unsigned long flags; |
677 | ||
678 | if (!pfmemalloc_active) | |
679 | return; | |
680 | ||
ce8eb6c4 | 681 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
682 | list_for_each_entry(page, &n->slabs_full, lru) |
683 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
684 | goto out; |
685 | ||
8456a648 JK |
686 | list_for_each_entry(page, &n->slabs_partial, lru) |
687 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
688 | goto out; |
689 | ||
8456a648 JK |
690 | list_for_each_entry(page, &n->slabs_free, lru) |
691 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
692 | goto out; |
693 | ||
694 | pfmemalloc_active = false; | |
695 | out: | |
ce8eb6c4 | 696 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
697 | } |
698 | ||
381760ea | 699 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
700 | gfp_t flags, bool force_refill) |
701 | { | |
702 | int i; | |
703 | void *objp = ac->entry[--ac->avail]; | |
704 | ||
705 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
706 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 707 | struct kmem_cache_node *n; |
072bb0aa MG |
708 | |
709 | if (gfp_pfmemalloc_allowed(flags)) { | |
710 | clear_obj_pfmemalloc(&objp); | |
711 | return objp; | |
712 | } | |
713 | ||
714 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 715 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
716 | /* If a !PFMEMALLOC object is found, swap them */ |
717 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
718 | objp = ac->entry[i]; | |
719 | ac->entry[i] = ac->entry[ac->avail]; | |
720 | ac->entry[ac->avail] = objp; | |
721 | return objp; | |
722 | } | |
723 | } | |
724 | ||
725 | /* | |
726 | * If there are empty slabs on the slabs_free list and we are | |
727 | * being forced to refill the cache, mark this one !pfmemalloc. | |
728 | */ | |
18bf8541 | 729 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 730 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 731 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 732 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
733 | clear_obj_pfmemalloc(&objp); |
734 | recheck_pfmemalloc_active(cachep, ac); | |
735 | return objp; | |
736 | } | |
737 | ||
738 | /* No !PFMEMALLOC objects available */ | |
739 | ac->avail++; | |
740 | objp = NULL; | |
741 | } | |
742 | ||
743 | return objp; | |
744 | } | |
745 | ||
381760ea MG |
746 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
747 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
748 | { | |
749 | void *objp; | |
750 | ||
751 | if (unlikely(sk_memalloc_socks())) | |
752 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
753 | else | |
754 | objp = ac->entry[--ac->avail]; | |
755 | ||
756 | return objp; | |
757 | } | |
758 | ||
d3aec344 JK |
759 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
760 | struct array_cache *ac, void *objp) | |
072bb0aa MG |
761 | { |
762 | if (unlikely(pfmemalloc_active)) { | |
763 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 764 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
765 | if (PageSlabPfmemalloc(page)) |
766 | set_obj_pfmemalloc(&objp); | |
767 | } | |
768 | ||
381760ea MG |
769 | return objp; |
770 | } | |
771 | ||
772 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
773 | void *objp) | |
774 | { | |
775 | if (unlikely(sk_memalloc_socks())) | |
776 | objp = __ac_put_obj(cachep, ac, objp); | |
777 | ||
072bb0aa MG |
778 | ac->entry[ac->avail++] = objp; |
779 | } | |
780 | ||
3ded175a CL |
781 | /* |
782 | * Transfer objects in one arraycache to another. | |
783 | * Locking must be handled by the caller. | |
784 | * | |
785 | * Return the number of entries transferred. | |
786 | */ | |
787 | static int transfer_objects(struct array_cache *to, | |
788 | struct array_cache *from, unsigned int max) | |
789 | { | |
790 | /* Figure out how many entries to transfer */ | |
732eacc0 | 791 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
792 | |
793 | if (!nr) | |
794 | return 0; | |
795 | ||
796 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
797 | sizeof(void *) *nr); | |
798 | ||
799 | from->avail -= nr; | |
800 | to->avail += nr; | |
3ded175a CL |
801 | return nr; |
802 | } | |
803 | ||
765c4507 CL |
804 | #ifndef CONFIG_NUMA |
805 | ||
806 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 807 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 808 | |
c8522a3a JK |
809 | static inline struct alien_cache **alloc_alien_cache(int node, |
810 | int limit, gfp_t gfp) | |
765c4507 | 811 | { |
edcad250 | 812 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
813 | } |
814 | ||
c8522a3a | 815 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
816 | { |
817 | } | |
818 | ||
819 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
820 | { | |
821 | return 0; | |
822 | } | |
823 | ||
824 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
825 | gfp_t flags) | |
826 | { | |
827 | return NULL; | |
828 | } | |
829 | ||
8b98c169 | 830 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
831 | gfp_t flags, int nodeid) |
832 | { | |
833 | return NULL; | |
834 | } | |
835 | ||
4167e9b2 DR |
836 | static inline gfp_t gfp_exact_node(gfp_t flags) |
837 | { | |
838 | return flags; | |
839 | } | |
840 | ||
765c4507 CL |
841 | #else /* CONFIG_NUMA */ |
842 | ||
8b98c169 | 843 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 844 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 845 | |
c8522a3a JK |
846 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
847 | int batch, gfp_t gfp) | |
848 | { | |
5e804789 | 849 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
850 | struct alien_cache *alc = NULL; |
851 | ||
852 | alc = kmalloc_node(memsize, gfp, node); | |
853 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 854 | spin_lock_init(&alc->lock); |
c8522a3a JK |
855 | return alc; |
856 | } | |
857 | ||
858 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 859 | { |
c8522a3a | 860 | struct alien_cache **alc_ptr; |
5e804789 | 861 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
862 | int i; |
863 | ||
864 | if (limit > 1) | |
865 | limit = 12; | |
c8522a3a JK |
866 | alc_ptr = kzalloc_node(memsize, gfp, node); |
867 | if (!alc_ptr) | |
868 | return NULL; | |
869 | ||
870 | for_each_node(i) { | |
871 | if (i == node || !node_online(i)) | |
872 | continue; | |
873 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
874 | if (!alc_ptr[i]) { | |
875 | for (i--; i >= 0; i--) | |
876 | kfree(alc_ptr[i]); | |
877 | kfree(alc_ptr); | |
878 | return NULL; | |
e498be7d CL |
879 | } |
880 | } | |
c8522a3a | 881 | return alc_ptr; |
e498be7d CL |
882 | } |
883 | ||
c8522a3a | 884 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
885 | { |
886 | int i; | |
887 | ||
c8522a3a | 888 | if (!alc_ptr) |
e498be7d | 889 | return; |
e498be7d | 890 | for_each_node(i) |
c8522a3a JK |
891 | kfree(alc_ptr[i]); |
892 | kfree(alc_ptr); | |
e498be7d CL |
893 | } |
894 | ||
343e0d7a | 895 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
896 | struct array_cache *ac, int node, |
897 | struct list_head *list) | |
e498be7d | 898 | { |
18bf8541 | 899 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
900 | |
901 | if (ac->avail) { | |
ce8eb6c4 | 902 | spin_lock(&n->list_lock); |
e00946fe CL |
903 | /* |
904 | * Stuff objects into the remote nodes shared array first. | |
905 | * That way we could avoid the overhead of putting the objects | |
906 | * into the free lists and getting them back later. | |
907 | */ | |
ce8eb6c4 CL |
908 | if (n->shared) |
909 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 910 | |
833b706c | 911 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 912 | ac->avail = 0; |
ce8eb6c4 | 913 | spin_unlock(&n->list_lock); |
e498be7d CL |
914 | } |
915 | } | |
916 | ||
8fce4d8e CL |
917 | /* |
918 | * Called from cache_reap() to regularly drain alien caches round robin. | |
919 | */ | |
ce8eb6c4 | 920 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 921 | { |
909ea964 | 922 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 923 | |
ce8eb6c4 | 924 | if (n->alien) { |
c8522a3a JK |
925 | struct alien_cache *alc = n->alien[node]; |
926 | struct array_cache *ac; | |
927 | ||
928 | if (alc) { | |
929 | ac = &alc->ac; | |
49dfc304 | 930 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
931 | LIST_HEAD(list); |
932 | ||
933 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 934 | spin_unlock_irq(&alc->lock); |
833b706c | 935 | slabs_destroy(cachep, &list); |
c8522a3a | 936 | } |
8fce4d8e CL |
937 | } |
938 | } | |
939 | } | |
940 | ||
a737b3e2 | 941 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 942 | struct alien_cache **alien) |
e498be7d | 943 | { |
b28a02de | 944 | int i = 0; |
c8522a3a | 945 | struct alien_cache *alc; |
e498be7d CL |
946 | struct array_cache *ac; |
947 | unsigned long flags; | |
948 | ||
949 | for_each_online_node(i) { | |
c8522a3a JK |
950 | alc = alien[i]; |
951 | if (alc) { | |
833b706c JK |
952 | LIST_HEAD(list); |
953 | ||
c8522a3a | 954 | ac = &alc->ac; |
49dfc304 | 955 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 956 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 957 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 958 | slabs_destroy(cachep, &list); |
e498be7d CL |
959 | } |
960 | } | |
961 | } | |
729bd0b7 | 962 | |
25c4f304 JK |
963 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
964 | int node, int page_node) | |
729bd0b7 | 965 | { |
ce8eb6c4 | 966 | struct kmem_cache_node *n; |
c8522a3a JK |
967 | struct alien_cache *alien = NULL; |
968 | struct array_cache *ac; | |
97654dfa | 969 | LIST_HEAD(list); |
1ca4cb24 | 970 | |
18bf8541 | 971 | n = get_node(cachep, node); |
729bd0b7 | 972 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
973 | if (n->alien && n->alien[page_node]) { |
974 | alien = n->alien[page_node]; | |
c8522a3a | 975 | ac = &alien->ac; |
49dfc304 | 976 | spin_lock(&alien->lock); |
c8522a3a | 977 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 978 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 979 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 980 | } |
c8522a3a | 981 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 982 | spin_unlock(&alien->lock); |
833b706c | 983 | slabs_destroy(cachep, &list); |
729bd0b7 | 984 | } else { |
25c4f304 | 985 | n = get_node(cachep, page_node); |
18bf8541 | 986 | spin_lock(&n->list_lock); |
25c4f304 | 987 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 988 | spin_unlock(&n->list_lock); |
97654dfa | 989 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
990 | } |
991 | return 1; | |
992 | } | |
25c4f304 JK |
993 | |
994 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
995 | { | |
996 | int page_node = page_to_nid(virt_to_page(objp)); | |
997 | int node = numa_mem_id(); | |
998 | /* | |
999 | * Make sure we are not freeing a object from another node to the array | |
1000 | * cache on this cpu. | |
1001 | */ | |
1002 | if (likely(node == page_node)) | |
1003 | return 0; | |
1004 | ||
1005 | return __cache_free_alien(cachep, objp, node, page_node); | |
1006 | } | |
4167e9b2 DR |
1007 | |
1008 | /* | |
d0164adc MG |
1009 | * Construct gfp mask to allocate from a specific node but do not direct reclaim |
1010 | * or warn about failures. kswapd may still wake to reclaim in the background. | |
4167e9b2 DR |
1011 | */ |
1012 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
1013 | { | |
d0164adc | 1014 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM; |
4167e9b2 | 1015 | } |
e498be7d CL |
1016 | #endif |
1017 | ||
8f9f8d9e | 1018 | /* |
6a67368c | 1019 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1020 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1021 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1022 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1023 | * already in use. |
1024 | * | |
18004c5d | 1025 | * Must hold slab_mutex. |
8f9f8d9e | 1026 | */ |
6a67368c | 1027 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1028 | { |
1029 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1030 | struct kmem_cache_node *n; |
5e804789 | 1031 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1032 | |
18004c5d | 1033 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1034 | /* |
5f0985bb | 1035 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1036 | * begin anything. Make sure some other cpu on this |
1037 | * node has not already allocated this | |
1038 | */ | |
18bf8541 CL |
1039 | n = get_node(cachep, node); |
1040 | if (!n) { | |
ce8eb6c4 CL |
1041 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1042 | if (!n) | |
8f9f8d9e | 1043 | return -ENOMEM; |
ce8eb6c4 | 1044 | kmem_cache_node_init(n); |
5f0985bb JZ |
1045 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1046 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1047 | |
1048 | /* | |
5f0985bb JZ |
1049 | * The kmem_cache_nodes don't come and go as CPUs |
1050 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1051 | * protection here. |
1052 | */ | |
ce8eb6c4 | 1053 | cachep->node[node] = n; |
8f9f8d9e DR |
1054 | } |
1055 | ||
18bf8541 CL |
1056 | spin_lock_irq(&n->list_lock); |
1057 | n->free_limit = | |
8f9f8d9e DR |
1058 | (1 + nr_cpus_node(node)) * |
1059 | cachep->batchcount + cachep->num; | |
18bf8541 | 1060 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1061 | } |
1062 | return 0; | |
1063 | } | |
1064 | ||
0fa8103b WL |
1065 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1066 | struct kmem_cache_node *n) | |
1067 | { | |
1068 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1069 | } | |
1070 | ||
0db0628d | 1071 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1072 | { |
1073 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1074 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1075 | int node = cpu_to_mem(cpu); |
a70f7302 | 1076 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1077 | |
18004c5d | 1078 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1079 | struct array_cache *nc; |
1080 | struct array_cache *shared; | |
c8522a3a | 1081 | struct alien_cache **alien; |
97654dfa | 1082 | LIST_HEAD(list); |
fbf1e473 | 1083 | |
18bf8541 | 1084 | n = get_node(cachep, node); |
ce8eb6c4 | 1085 | if (!n) |
bf0dea23 | 1086 | continue; |
fbf1e473 | 1087 | |
ce8eb6c4 | 1088 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1089 | |
ce8eb6c4 CL |
1090 | /* Free limit for this kmem_cache_node */ |
1091 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1092 | |
1093 | /* cpu is dead; no one can alloc from it. */ | |
1094 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1095 | if (nc) { | |
97654dfa | 1096 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1097 | nc->avail = 0; |
1098 | } | |
fbf1e473 | 1099 | |
58463c1f | 1100 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1101 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1102 | goto free_slab; |
fbf1e473 AM |
1103 | } |
1104 | ||
ce8eb6c4 | 1105 | shared = n->shared; |
fbf1e473 AM |
1106 | if (shared) { |
1107 | free_block(cachep, shared->entry, | |
97654dfa | 1108 | shared->avail, node, &list); |
ce8eb6c4 | 1109 | n->shared = NULL; |
fbf1e473 AM |
1110 | } |
1111 | ||
ce8eb6c4 CL |
1112 | alien = n->alien; |
1113 | n->alien = NULL; | |
fbf1e473 | 1114 | |
ce8eb6c4 | 1115 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1116 | |
1117 | kfree(shared); | |
1118 | if (alien) { | |
1119 | drain_alien_cache(cachep, alien); | |
1120 | free_alien_cache(alien); | |
1121 | } | |
bf0dea23 JK |
1122 | |
1123 | free_slab: | |
97654dfa | 1124 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1125 | } |
1126 | /* | |
1127 | * In the previous loop, all the objects were freed to | |
1128 | * the respective cache's slabs, now we can go ahead and | |
1129 | * shrink each nodelist to its limit. | |
1130 | */ | |
18004c5d | 1131 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1132 | n = get_node(cachep, node); |
ce8eb6c4 | 1133 | if (!n) |
fbf1e473 | 1134 | continue; |
0fa8103b | 1135 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1136 | } |
1137 | } | |
1138 | ||
0db0628d | 1139 | static int cpuup_prepare(long cpu) |
1da177e4 | 1140 | { |
343e0d7a | 1141 | struct kmem_cache *cachep; |
ce8eb6c4 | 1142 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1143 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1144 | int err; |
1da177e4 | 1145 | |
fbf1e473 AM |
1146 | /* |
1147 | * We need to do this right in the beginning since | |
1148 | * alloc_arraycache's are going to use this list. | |
1149 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1150 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1151 | */ |
6a67368c | 1152 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1153 | if (err < 0) |
1154 | goto bad; | |
fbf1e473 AM |
1155 | |
1156 | /* | |
1157 | * Now we can go ahead with allocating the shared arrays and | |
1158 | * array caches | |
1159 | */ | |
18004c5d | 1160 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 1161 | struct array_cache *shared = NULL; |
c8522a3a | 1162 | struct alien_cache **alien = NULL; |
fbf1e473 | 1163 | |
fbf1e473 AM |
1164 | if (cachep->shared) { |
1165 | shared = alloc_arraycache(node, | |
1166 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1167 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1168 | if (!shared) |
1da177e4 | 1169 | goto bad; |
fbf1e473 AM |
1170 | } |
1171 | if (use_alien_caches) { | |
83b519e8 | 1172 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1173 | if (!alien) { |
1174 | kfree(shared); | |
fbf1e473 | 1175 | goto bad; |
12d00f6a | 1176 | } |
fbf1e473 | 1177 | } |
18bf8541 | 1178 | n = get_node(cachep, node); |
ce8eb6c4 | 1179 | BUG_ON(!n); |
fbf1e473 | 1180 | |
ce8eb6c4 CL |
1181 | spin_lock_irq(&n->list_lock); |
1182 | if (!n->shared) { | |
fbf1e473 AM |
1183 | /* |
1184 | * We are serialised from CPU_DEAD or | |
1185 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1186 | */ | |
ce8eb6c4 | 1187 | n->shared = shared; |
fbf1e473 AM |
1188 | shared = NULL; |
1189 | } | |
4484ebf1 | 1190 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1191 | if (!n->alien) { |
1192 | n->alien = alien; | |
fbf1e473 | 1193 | alien = NULL; |
1da177e4 | 1194 | } |
fbf1e473 | 1195 | #endif |
ce8eb6c4 | 1196 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1197 | kfree(shared); |
1198 | free_alien_cache(alien); | |
1199 | } | |
ce79ddc8 | 1200 | |
fbf1e473 AM |
1201 | return 0; |
1202 | bad: | |
12d00f6a | 1203 | cpuup_canceled(cpu); |
fbf1e473 AM |
1204 | return -ENOMEM; |
1205 | } | |
1206 | ||
0db0628d | 1207 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1208 | unsigned long action, void *hcpu) |
1209 | { | |
1210 | long cpu = (long)hcpu; | |
1211 | int err = 0; | |
1212 | ||
1213 | switch (action) { | |
fbf1e473 AM |
1214 | case CPU_UP_PREPARE: |
1215 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1216 | mutex_lock(&slab_mutex); |
fbf1e473 | 1217 | err = cpuup_prepare(cpu); |
18004c5d | 1218 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1219 | break; |
1220 | case CPU_ONLINE: | |
8bb78442 | 1221 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1222 | start_cpu_timer(cpu); |
1223 | break; | |
1224 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1225 | case CPU_DOWN_PREPARE: |
8bb78442 | 1226 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1227 | /* |
18004c5d | 1228 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1229 | * held so that if cache_reap() is invoked it cannot do |
1230 | * anything expensive but will only modify reap_work | |
1231 | * and reschedule the timer. | |
1232 | */ | |
afe2c511 | 1233 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1234 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1235 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1236 | break; |
1237 | case CPU_DOWN_FAILED: | |
8bb78442 | 1238 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1239 | start_cpu_timer(cpu); |
1240 | break; | |
1da177e4 | 1241 | case CPU_DEAD: |
8bb78442 | 1242 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1243 | /* |
1244 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1245 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1246 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1247 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1248 | * structure is usually allocated from kmem_cache_create() and |
1249 | * gets destroyed at kmem_cache_destroy(). | |
1250 | */ | |
183ff22b | 1251 | /* fall through */ |
8f5be20b | 1252 | #endif |
1da177e4 | 1253 | case CPU_UP_CANCELED: |
8bb78442 | 1254 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1255 | mutex_lock(&slab_mutex); |
fbf1e473 | 1256 | cpuup_canceled(cpu); |
18004c5d | 1257 | mutex_unlock(&slab_mutex); |
1da177e4 | 1258 | break; |
1da177e4 | 1259 | } |
eac40680 | 1260 | return notifier_from_errno(err); |
1da177e4 LT |
1261 | } |
1262 | ||
0db0628d | 1263 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1264 | &cpuup_callback, NULL, 0 |
1265 | }; | |
1da177e4 | 1266 | |
8f9f8d9e DR |
1267 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1268 | /* | |
1269 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1270 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1271 | * removed. | |
1272 | * | |
18004c5d | 1273 | * Must hold slab_mutex. |
8f9f8d9e | 1274 | */ |
6a67368c | 1275 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1276 | { |
1277 | struct kmem_cache *cachep; | |
1278 | int ret = 0; | |
1279 | ||
18004c5d | 1280 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1281 | struct kmem_cache_node *n; |
8f9f8d9e | 1282 | |
18bf8541 | 1283 | n = get_node(cachep, node); |
ce8eb6c4 | 1284 | if (!n) |
8f9f8d9e DR |
1285 | continue; |
1286 | ||
0fa8103b | 1287 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1288 | |
ce8eb6c4 CL |
1289 | if (!list_empty(&n->slabs_full) || |
1290 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1291 | ret = -EBUSY; |
1292 | break; | |
1293 | } | |
1294 | } | |
1295 | return ret; | |
1296 | } | |
1297 | ||
1298 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1299 | unsigned long action, void *arg) | |
1300 | { | |
1301 | struct memory_notify *mnb = arg; | |
1302 | int ret = 0; | |
1303 | int nid; | |
1304 | ||
1305 | nid = mnb->status_change_nid; | |
1306 | if (nid < 0) | |
1307 | goto out; | |
1308 | ||
1309 | switch (action) { | |
1310 | case MEM_GOING_ONLINE: | |
18004c5d | 1311 | mutex_lock(&slab_mutex); |
6a67368c | 1312 | ret = init_cache_node_node(nid); |
18004c5d | 1313 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1314 | break; |
1315 | case MEM_GOING_OFFLINE: | |
18004c5d | 1316 | mutex_lock(&slab_mutex); |
6a67368c | 1317 | ret = drain_cache_node_node(nid); |
18004c5d | 1318 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1319 | break; |
1320 | case MEM_ONLINE: | |
1321 | case MEM_OFFLINE: | |
1322 | case MEM_CANCEL_ONLINE: | |
1323 | case MEM_CANCEL_OFFLINE: | |
1324 | break; | |
1325 | } | |
1326 | out: | |
5fda1bd5 | 1327 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1328 | } |
1329 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1330 | ||
e498be7d | 1331 | /* |
ce8eb6c4 | 1332 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1333 | */ |
6744f087 | 1334 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1335 | int nodeid) |
e498be7d | 1336 | { |
6744f087 | 1337 | struct kmem_cache_node *ptr; |
e498be7d | 1338 | |
6744f087 | 1339 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1340 | BUG_ON(!ptr); |
1341 | ||
6744f087 | 1342 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1343 | /* |
1344 | * Do not assume that spinlocks can be initialized via memcpy: | |
1345 | */ | |
1346 | spin_lock_init(&ptr->list_lock); | |
1347 | ||
e498be7d | 1348 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1349 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1350 | } |
1351 | ||
556a169d | 1352 | /* |
ce8eb6c4 CL |
1353 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1354 | * size of kmem_cache_node. | |
556a169d | 1355 | */ |
ce8eb6c4 | 1356 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1357 | { |
1358 | int node; | |
1359 | ||
1360 | for_each_online_node(node) { | |
ce8eb6c4 | 1361 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1362 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1363 | REAPTIMEOUT_NODE + |
1364 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1365 | } |
1366 | } | |
1367 | ||
a737b3e2 AM |
1368 | /* |
1369 | * Initialisation. Called after the page allocator have been initialised and | |
1370 | * before smp_init(). | |
1da177e4 LT |
1371 | */ |
1372 | void __init kmem_cache_init(void) | |
1373 | { | |
e498be7d CL |
1374 | int i; |
1375 | ||
68126702 JK |
1376 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1377 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1378 | kmem_cache = &kmem_cache_boot; |
1379 | ||
b6e68bc1 | 1380 | if (num_possible_nodes() == 1) |
62918a03 SS |
1381 | use_alien_caches = 0; |
1382 | ||
3c583465 | 1383 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1384 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1385 | |
1da177e4 LT |
1386 | /* |
1387 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1388 | * page orders on machines with more than 32MB of memory if |
1389 | * not overridden on the command line. | |
1da177e4 | 1390 | */ |
3df1cccd | 1391 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1392 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1393 | |
1da177e4 LT |
1394 | /* Bootstrap is tricky, because several objects are allocated |
1395 | * from caches that do not exist yet: | |
9b030cb8 CL |
1396 | * 1) initialize the kmem_cache cache: it contains the struct |
1397 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1398 | * kmem_cache is statically allocated. | |
e498be7d | 1399 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1400 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1401 | * array at the end of the bootstrap. |
1da177e4 | 1402 | * 2) Create the first kmalloc cache. |
343e0d7a | 1403 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1404 | * An __init data area is used for the head array. |
1405 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1406 | * head arrays. | |
9b030cb8 | 1407 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1408 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1409 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1410 | * the other cache's with kmalloc allocated memory. |
1411 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1412 | */ |
1413 | ||
9b030cb8 | 1414 | /* 1) create the kmem_cache */ |
1da177e4 | 1415 | |
8da3430d | 1416 | /* |
b56efcf0 | 1417 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1418 | */ |
2f9baa9f | 1419 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1420 | offsetof(struct kmem_cache, node) + |
6744f087 | 1421 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1422 | SLAB_HWCACHE_ALIGN); |
1423 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1424 | slab_state = PARTIAL; |
1da177e4 | 1425 | |
a737b3e2 | 1426 | /* |
bf0dea23 JK |
1427 | * Initialize the caches that provide memory for the kmem_cache_node |
1428 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1429 | */ |
bf0dea23 | 1430 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1431 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1432 | slab_state = PARTIAL_NODE; |
34cc6990 | 1433 | setup_kmalloc_cache_index_table(); |
e498be7d | 1434 | |
e0a42726 IM |
1435 | slab_early_init = 0; |
1436 | ||
ce8eb6c4 | 1437 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1438 | { |
1ca4cb24 PE |
1439 | int nid; |
1440 | ||
9c09a95c | 1441 | for_each_online_node(nid) { |
ce8eb6c4 | 1442 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1443 | |
bf0dea23 | 1444 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1445 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1446 | } |
1447 | } | |
1da177e4 | 1448 | |
f97d5f63 | 1449 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1450 | } |
1451 | ||
1452 | void __init kmem_cache_init_late(void) | |
1453 | { | |
1454 | struct kmem_cache *cachep; | |
1455 | ||
97d06609 | 1456 | slab_state = UP; |
52cef189 | 1457 | |
8429db5c | 1458 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1459 | mutex_lock(&slab_mutex); |
1460 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1461 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1462 | BUG(); | |
18004c5d | 1463 | mutex_unlock(&slab_mutex); |
056c6241 | 1464 | |
97d06609 CL |
1465 | /* Done! */ |
1466 | slab_state = FULL; | |
1467 | ||
a737b3e2 AM |
1468 | /* |
1469 | * Register a cpu startup notifier callback that initializes | |
1470 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1471 | */ |
1472 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1473 | |
8f9f8d9e DR |
1474 | #ifdef CONFIG_NUMA |
1475 | /* | |
1476 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1477 | * node. |
8f9f8d9e DR |
1478 | */ |
1479 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1480 | #endif | |
1481 | ||
a737b3e2 AM |
1482 | /* |
1483 | * The reap timers are started later, with a module init call: That part | |
1484 | * of the kernel is not yet operational. | |
1da177e4 LT |
1485 | */ |
1486 | } | |
1487 | ||
1488 | static int __init cpucache_init(void) | |
1489 | { | |
1490 | int cpu; | |
1491 | ||
a737b3e2 AM |
1492 | /* |
1493 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1494 | */ |
e498be7d | 1495 | for_each_online_cpu(cpu) |
a737b3e2 | 1496 | start_cpu_timer(cpu); |
a164f896 GC |
1497 | |
1498 | /* Done! */ | |
97d06609 | 1499 | slab_state = FULL; |
1da177e4 LT |
1500 | return 0; |
1501 | } | |
1da177e4 LT |
1502 | __initcall(cpucache_init); |
1503 | ||
8bdec192 RA |
1504 | static noinline void |
1505 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1506 | { | |
9a02d699 | 1507 | #if DEBUG |
ce8eb6c4 | 1508 | struct kmem_cache_node *n; |
8456a648 | 1509 | struct page *page; |
8bdec192 RA |
1510 | unsigned long flags; |
1511 | int node; | |
9a02d699 DR |
1512 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1513 | DEFAULT_RATELIMIT_BURST); | |
1514 | ||
1515 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1516 | return; | |
8bdec192 RA |
1517 | |
1518 | printk(KERN_WARNING | |
1519 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1520 | nodeid, gfpflags); | |
1521 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1522 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1523 | |
18bf8541 | 1524 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1525 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1526 | unsigned long active_slabs = 0, num_slabs = 0; | |
1527 | ||
ce8eb6c4 | 1528 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1529 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1530 | active_objs += cachep->num; |
1531 | active_slabs++; | |
1532 | } | |
8456a648 JK |
1533 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1534 | active_objs += page->active; | |
8bdec192 RA |
1535 | active_slabs++; |
1536 | } | |
8456a648 | 1537 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1538 | num_slabs++; |
1539 | ||
ce8eb6c4 CL |
1540 | free_objects += n->free_objects; |
1541 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1542 | |
1543 | num_slabs += active_slabs; | |
1544 | num_objs = num_slabs * cachep->num; | |
1545 | printk(KERN_WARNING | |
1546 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1547 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1548 | free_objects); | |
1549 | } | |
9a02d699 | 1550 | #endif |
8bdec192 RA |
1551 | } |
1552 | ||
1da177e4 | 1553 | /* |
8a7d9b43 WSH |
1554 | * Interface to system's page allocator. No need to hold the |
1555 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1556 | * |
1557 | * If we requested dmaable memory, we will get it. Even if we | |
1558 | * did not request dmaable memory, we might get it, but that | |
1559 | * would be relatively rare and ignorable. | |
1560 | */ | |
0c3aa83e JK |
1561 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1562 | int nodeid) | |
1da177e4 LT |
1563 | { |
1564 | struct page *page; | |
e1b6aa6f | 1565 | int nr_pages; |
765c4507 | 1566 | |
a618e89f | 1567 | flags |= cachep->allocflags; |
e12ba74d MG |
1568 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1569 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1570 | |
96db800f | 1571 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1572 | if (!page) { |
9a02d699 | 1573 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1574 | return NULL; |
8bdec192 | 1575 | } |
1da177e4 | 1576 | |
f3ccb2c4 VD |
1577 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1578 | __free_pages(page, cachep->gfporder); | |
1579 | return NULL; | |
1580 | } | |
1581 | ||
b37f1dd0 | 1582 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
2f064f34 | 1583 | if (page_is_pfmemalloc(page)) |
072bb0aa MG |
1584 | pfmemalloc_active = true; |
1585 | ||
e1b6aa6f | 1586 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1587 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1588 | add_zone_page_state(page_zone(page), |
1589 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1590 | else | |
1591 | add_zone_page_state(page_zone(page), | |
1592 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 | 1593 | __SetPageSlab(page); |
2f064f34 | 1594 | if (page_is_pfmemalloc(page)) |
a57a4988 | 1595 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1596 | |
b1eeab67 VN |
1597 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1598 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1599 | ||
1600 | if (cachep->ctor) | |
1601 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1602 | else | |
1603 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1604 | } | |
c175eea4 | 1605 | |
0c3aa83e | 1606 | return page; |
1da177e4 LT |
1607 | } |
1608 | ||
1609 | /* | |
1610 | * Interface to system's page release. | |
1611 | */ | |
0c3aa83e | 1612 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1613 | { |
a57a4988 | 1614 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1615 | |
b1eeab67 | 1616 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1617 | |
972d1a7b CL |
1618 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1619 | sub_zone_page_state(page_zone(page), | |
1620 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1621 | else | |
1622 | sub_zone_page_state(page_zone(page), | |
1623 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1624 | |
a57a4988 | 1625 | BUG_ON(!PageSlab(page)); |
73293c2f | 1626 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1627 | __ClearPageSlab(page); |
8456a648 JK |
1628 | page_mapcount_reset(page); |
1629 | page->mapping = NULL; | |
1f458cbf | 1630 | |
1da177e4 LT |
1631 | if (current->reclaim_state) |
1632 | current->reclaim_state->reclaimed_slab += nr_freed; | |
f3ccb2c4 | 1633 | __free_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1634 | } |
1635 | ||
1636 | static void kmem_rcu_free(struct rcu_head *head) | |
1637 | { | |
68126702 JK |
1638 | struct kmem_cache *cachep; |
1639 | struct page *page; | |
1da177e4 | 1640 | |
68126702 JK |
1641 | page = container_of(head, struct page, rcu_head); |
1642 | cachep = page->slab_cache; | |
1643 | ||
1644 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1645 | } |
1646 | ||
1647 | #if DEBUG | |
40b44137 JK |
1648 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1649 | { | |
1650 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1651 | (cachep->size % PAGE_SIZE) == 0) | |
1652 | return true; | |
1653 | ||
1654 | return false; | |
1655 | } | |
1da177e4 LT |
1656 | |
1657 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1658 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1659 | unsigned long caller) |
1da177e4 | 1660 | { |
8c138bc0 | 1661 | int size = cachep->object_size; |
1da177e4 | 1662 | |
3dafccf2 | 1663 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1664 | |
b28a02de | 1665 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1666 | return; |
1667 | ||
b28a02de PE |
1668 | *addr++ = 0x12345678; |
1669 | *addr++ = caller; | |
1670 | *addr++ = smp_processor_id(); | |
1671 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1672 | { |
1673 | unsigned long *sptr = &caller; | |
1674 | unsigned long svalue; | |
1675 | ||
1676 | while (!kstack_end(sptr)) { | |
1677 | svalue = *sptr++; | |
1678 | if (kernel_text_address(svalue)) { | |
b28a02de | 1679 | *addr++ = svalue; |
1da177e4 LT |
1680 | size -= sizeof(unsigned long); |
1681 | if (size <= sizeof(unsigned long)) | |
1682 | break; | |
1683 | } | |
1684 | } | |
1685 | ||
1686 | } | |
b28a02de | 1687 | *addr++ = 0x87654321; |
1da177e4 | 1688 | } |
40b44137 JK |
1689 | |
1690 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1691 | int map, unsigned long caller) | |
1692 | { | |
1693 | if (!is_debug_pagealloc_cache(cachep)) | |
1694 | return; | |
1695 | ||
1696 | if (caller) | |
1697 | store_stackinfo(cachep, objp, caller); | |
1698 | ||
1699 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1700 | } | |
1701 | ||
1702 | #else | |
1703 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1704 | int map, unsigned long caller) {} | |
1705 | ||
1da177e4 LT |
1706 | #endif |
1707 | ||
343e0d7a | 1708 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1709 | { |
8c138bc0 | 1710 | int size = cachep->object_size; |
3dafccf2 | 1711 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1712 | |
1713 | memset(addr, val, size); | |
b28a02de | 1714 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | static void dump_line(char *data, int offset, int limit) | |
1718 | { | |
1719 | int i; | |
aa83aa40 DJ |
1720 | unsigned char error = 0; |
1721 | int bad_count = 0; | |
1722 | ||
fdde6abb | 1723 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1724 | for (i = 0; i < limit; i++) { |
1725 | if (data[offset + i] != POISON_FREE) { | |
1726 | error = data[offset + i]; | |
1727 | bad_count++; | |
1728 | } | |
aa83aa40 | 1729 | } |
fdde6abb SAS |
1730 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1731 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1732 | |
1733 | if (bad_count == 1) { | |
1734 | error ^= POISON_FREE; | |
1735 | if (!(error & (error - 1))) { | |
1736 | printk(KERN_ERR "Single bit error detected. Probably " | |
1737 | "bad RAM.\n"); | |
1738 | #ifdef CONFIG_X86 | |
1739 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1740 | "test tool.\n"); | |
1741 | #else | |
1742 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1743 | #endif | |
1744 | } | |
1745 | } | |
1da177e4 LT |
1746 | } |
1747 | #endif | |
1748 | ||
1749 | #if DEBUG | |
1750 | ||
343e0d7a | 1751 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1752 | { |
1753 | int i, size; | |
1754 | char *realobj; | |
1755 | ||
1756 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1757 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1758 | *dbg_redzone1(cachep, objp), |
1759 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1760 | } |
1761 | ||
1762 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1763 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1764 | *dbg_userword(cachep, objp), | |
1765 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1766 | } |
3dafccf2 | 1767 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1768 | size = cachep->object_size; |
b28a02de | 1769 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1770 | int limit; |
1771 | limit = 16; | |
b28a02de PE |
1772 | if (i + limit > size) |
1773 | limit = size - i; | |
1da177e4 LT |
1774 | dump_line(realobj, i, limit); |
1775 | } | |
1776 | } | |
1777 | ||
343e0d7a | 1778 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1779 | { |
1780 | char *realobj; | |
1781 | int size, i; | |
1782 | int lines = 0; | |
1783 | ||
40b44137 JK |
1784 | if (is_debug_pagealloc_cache(cachep)) |
1785 | return; | |
1786 | ||
3dafccf2 | 1787 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1788 | size = cachep->object_size; |
1da177e4 | 1789 | |
b28a02de | 1790 | for (i = 0; i < size; i++) { |
1da177e4 | 1791 | char exp = POISON_FREE; |
b28a02de | 1792 | if (i == size - 1) |
1da177e4 LT |
1793 | exp = POISON_END; |
1794 | if (realobj[i] != exp) { | |
1795 | int limit; | |
1796 | /* Mismatch ! */ | |
1797 | /* Print header */ | |
1798 | if (lines == 0) { | |
b28a02de | 1799 | printk(KERN_ERR |
face37f5 DJ |
1800 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1801 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1802 | print_objinfo(cachep, objp, 0); |
1803 | } | |
1804 | /* Hexdump the affected line */ | |
b28a02de | 1805 | i = (i / 16) * 16; |
1da177e4 | 1806 | limit = 16; |
b28a02de PE |
1807 | if (i + limit > size) |
1808 | limit = size - i; | |
1da177e4 LT |
1809 | dump_line(realobj, i, limit); |
1810 | i += 16; | |
1811 | lines++; | |
1812 | /* Limit to 5 lines */ | |
1813 | if (lines > 5) | |
1814 | break; | |
1815 | } | |
1816 | } | |
1817 | if (lines != 0) { | |
1818 | /* Print some data about the neighboring objects, if they | |
1819 | * exist: | |
1820 | */ | |
8456a648 | 1821 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1822 | unsigned int objnr; |
1da177e4 | 1823 | |
8456a648 | 1824 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1825 | if (objnr) { |
8456a648 | 1826 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1827 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1828 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1829 | realobj, size); |
1da177e4 LT |
1830 | print_objinfo(cachep, objp, 2); |
1831 | } | |
b28a02de | 1832 | if (objnr + 1 < cachep->num) { |
8456a648 | 1833 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1834 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1835 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1836 | realobj, size); |
1da177e4 LT |
1837 | print_objinfo(cachep, objp, 2); |
1838 | } | |
1839 | } | |
1840 | } | |
1841 | #endif | |
1842 | ||
12dd36fa | 1843 | #if DEBUG |
8456a648 JK |
1844 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1845 | struct page *page) | |
1da177e4 | 1846 | { |
1da177e4 LT |
1847 | int i; |
1848 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1849 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1850 | |
1851 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1852 | check_poison_obj(cachep, objp); |
40b44137 | 1853 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1854 | } |
1855 | if (cachep->flags & SLAB_RED_ZONE) { | |
1856 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1857 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1858 | "was overwritten"); |
1da177e4 LT |
1859 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1860 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1861 | "was overwritten"); |
1da177e4 | 1862 | } |
1da177e4 | 1863 | } |
12dd36fa | 1864 | } |
1da177e4 | 1865 | #else |
8456a648 JK |
1866 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1867 | struct page *page) | |
12dd36fa | 1868 | { |
12dd36fa | 1869 | } |
1da177e4 LT |
1870 | #endif |
1871 | ||
911851e6 RD |
1872 | /** |
1873 | * slab_destroy - destroy and release all objects in a slab | |
1874 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1875 | * @page: page pointer being destroyed |
911851e6 | 1876 | * |
8a7d9b43 WSH |
1877 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1878 | * Before calling the slab page must have been unlinked from the cache. The | |
1879 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1880 | */ |
8456a648 | 1881 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1882 | { |
7e007355 | 1883 | void *freelist; |
12dd36fa | 1884 | |
8456a648 JK |
1885 | freelist = page->freelist; |
1886 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1887 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1888 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1889 | else | |
0c3aa83e | 1890 | kmem_freepages(cachep, page); |
68126702 JK |
1891 | |
1892 | /* | |
8456a648 | 1893 | * From now on, we don't use freelist |
68126702 JK |
1894 | * although actual page can be freed in rcu context |
1895 | */ | |
1896 | if (OFF_SLAB(cachep)) | |
8456a648 | 1897 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1898 | } |
1899 | ||
97654dfa JK |
1900 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1901 | { | |
1902 | struct page *page, *n; | |
1903 | ||
1904 | list_for_each_entry_safe(page, n, list, lru) { | |
1905 | list_del(&page->lru); | |
1906 | slab_destroy(cachep, page); | |
1907 | } | |
1908 | } | |
1909 | ||
4d268eba | 1910 | /** |
a70773dd RD |
1911 | * calculate_slab_order - calculate size (page order) of slabs |
1912 | * @cachep: pointer to the cache that is being created | |
1913 | * @size: size of objects to be created in this cache. | |
1914 | * @align: required alignment for the objects. | |
1915 | * @flags: slab allocation flags | |
1916 | * | |
1917 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1918 | * |
1919 | * This could be made much more intelligent. For now, try to avoid using | |
1920 | * high order pages for slabs. When the gfp() functions are more friendly | |
1921 | * towards high-order requests, this should be changed. | |
1922 | */ | |
a737b3e2 | 1923 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1924 | size_t size, size_t align, unsigned long flags) |
4d268eba | 1925 | { |
b1ab41c4 | 1926 | unsigned long offslab_limit; |
4d268eba | 1927 | size_t left_over = 0; |
9888e6fa | 1928 | int gfporder; |
4d268eba | 1929 | |
0aa817f0 | 1930 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1931 | unsigned int num; |
1932 | size_t remainder; | |
1933 | ||
9888e6fa | 1934 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1935 | if (!num) |
1936 | continue; | |
9888e6fa | 1937 | |
f315e3fa JK |
1938 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1939 | if (num > SLAB_OBJ_MAX_NUM) | |
1940 | break; | |
1941 | ||
b1ab41c4 IM |
1942 | if (flags & CFLGS_OFF_SLAB) { |
1943 | /* | |
1944 | * Max number of objs-per-slab for caches which | |
1945 | * use off-slab slabs. Needed to avoid a possible | |
1946 | * looping condition in cache_grow(). | |
1947 | */ | |
8456a648 | 1948 | offslab_limit = size; |
249247b6 | 1949 | offslab_limit /= sizeof(freelist_idx_t); |
b1ab41c4 IM |
1950 | |
1951 | if (num > offslab_limit) | |
1952 | break; | |
1953 | } | |
4d268eba | 1954 | |
9888e6fa | 1955 | /* Found something acceptable - save it away */ |
4d268eba | 1956 | cachep->num = num; |
9888e6fa | 1957 | cachep->gfporder = gfporder; |
4d268eba PE |
1958 | left_over = remainder; |
1959 | ||
f78bb8ad LT |
1960 | /* |
1961 | * A VFS-reclaimable slab tends to have most allocations | |
1962 | * as GFP_NOFS and we really don't want to have to be allocating | |
1963 | * higher-order pages when we are unable to shrink dcache. | |
1964 | */ | |
1965 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1966 | break; | |
1967 | ||
4d268eba PE |
1968 | /* |
1969 | * Large number of objects is good, but very large slabs are | |
1970 | * currently bad for the gfp()s. | |
1971 | */ | |
543585cc | 1972 | if (gfporder >= slab_max_order) |
4d268eba PE |
1973 | break; |
1974 | ||
9888e6fa LT |
1975 | /* |
1976 | * Acceptable internal fragmentation? | |
1977 | */ | |
a737b3e2 | 1978 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1979 | break; |
1980 | } | |
1981 | return left_over; | |
1982 | } | |
1983 | ||
bf0dea23 JK |
1984 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1985 | struct kmem_cache *cachep, int entries, int batchcount) | |
1986 | { | |
1987 | int cpu; | |
1988 | size_t size; | |
1989 | struct array_cache __percpu *cpu_cache; | |
1990 | ||
1991 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1992 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1993 | |
1994 | if (!cpu_cache) | |
1995 | return NULL; | |
1996 | ||
1997 | for_each_possible_cpu(cpu) { | |
1998 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1999 | entries, batchcount); | |
2000 | } | |
2001 | ||
2002 | return cpu_cache; | |
2003 | } | |
2004 | ||
83b519e8 | 2005 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2006 | { |
97d06609 | 2007 | if (slab_state >= FULL) |
83b519e8 | 2008 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2009 | |
bf0dea23 JK |
2010 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
2011 | if (!cachep->cpu_cache) | |
2012 | return 1; | |
2013 | ||
97d06609 | 2014 | if (slab_state == DOWN) { |
bf0dea23 JK |
2015 | /* Creation of first cache (kmem_cache). */ |
2016 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 2017 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
2018 | /* For kmem_cache_node */ |
2019 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 2020 | } else { |
bf0dea23 | 2021 | int node; |
f30cf7d1 | 2022 | |
bf0dea23 JK |
2023 | for_each_online_node(node) { |
2024 | cachep->node[node] = kmalloc_node( | |
2025 | sizeof(struct kmem_cache_node), gfp, node); | |
2026 | BUG_ON(!cachep->node[node]); | |
2027 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
2028 | } |
2029 | } | |
bf0dea23 | 2030 | |
6a67368c | 2031 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2032 | jiffies + REAPTIMEOUT_NODE + |
2033 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2034 | |
2035 | cpu_cache_get(cachep)->avail = 0; | |
2036 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2037 | cpu_cache_get(cachep)->batchcount = 1; | |
2038 | cpu_cache_get(cachep)->touched = 0; | |
2039 | cachep->batchcount = 1; | |
2040 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2041 | return 0; |
f30cf7d1 PE |
2042 | } |
2043 | ||
12220dea JK |
2044 | unsigned long kmem_cache_flags(unsigned long object_size, |
2045 | unsigned long flags, const char *name, | |
2046 | void (*ctor)(void *)) | |
2047 | { | |
2048 | return flags; | |
2049 | } | |
2050 | ||
2051 | struct kmem_cache * | |
2052 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
2053 | unsigned long flags, void (*ctor)(void *)) | |
2054 | { | |
2055 | struct kmem_cache *cachep; | |
2056 | ||
2057 | cachep = find_mergeable(size, align, flags, name, ctor); | |
2058 | if (cachep) { | |
2059 | cachep->refcount++; | |
2060 | ||
2061 | /* | |
2062 | * Adjust the object sizes so that we clear | |
2063 | * the complete object on kzalloc. | |
2064 | */ | |
2065 | cachep->object_size = max_t(int, cachep->object_size, size); | |
2066 | } | |
2067 | return cachep; | |
2068 | } | |
2069 | ||
1da177e4 | 2070 | /** |
039363f3 | 2071 | * __kmem_cache_create - Create a cache. |
a755b76a | 2072 | * @cachep: cache management descriptor |
1da177e4 | 2073 | * @flags: SLAB flags |
1da177e4 LT |
2074 | * |
2075 | * Returns a ptr to the cache on success, NULL on failure. | |
2076 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2077 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2078 | * |
1da177e4 LT |
2079 | * The flags are |
2080 | * | |
2081 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2082 | * to catch references to uninitialised memory. | |
2083 | * | |
2084 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2085 | * for buffer overruns. | |
2086 | * | |
1da177e4 LT |
2087 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2088 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2089 | * as davem. | |
2090 | */ | |
278b1bb1 | 2091 | int |
8a13a4cc | 2092 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2093 | { |
d4a5fca5 DR |
2094 | size_t left_over, freelist_size; |
2095 | size_t ralign = BYTES_PER_WORD; | |
83b519e8 | 2096 | gfp_t gfp; |
278b1bb1 | 2097 | int err; |
8a13a4cc | 2098 | size_t size = cachep->size; |
1da177e4 | 2099 | |
1da177e4 | 2100 | #if DEBUG |
1da177e4 LT |
2101 | #if FORCED_DEBUG |
2102 | /* | |
2103 | * Enable redzoning and last user accounting, except for caches with | |
2104 | * large objects, if the increased size would increase the object size | |
2105 | * above the next power of two: caches with object sizes just above a | |
2106 | * power of two have a significant amount of internal fragmentation. | |
2107 | */ | |
87a927c7 DW |
2108 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2109 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2110 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2111 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2112 | flags |= SLAB_POISON; | |
2113 | #endif | |
1da177e4 | 2114 | #endif |
1da177e4 | 2115 | |
a737b3e2 AM |
2116 | /* |
2117 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2118 | * unaligned accesses for some archs when redzoning is used, and makes |
2119 | * sure any on-slab bufctl's are also correctly aligned. | |
2120 | */ | |
b28a02de PE |
2121 | if (size & (BYTES_PER_WORD - 1)) { |
2122 | size += (BYTES_PER_WORD - 1); | |
2123 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2124 | } |
2125 | ||
87a927c7 DW |
2126 | if (flags & SLAB_RED_ZONE) { |
2127 | ralign = REDZONE_ALIGN; | |
2128 | /* If redzoning, ensure that the second redzone is suitably | |
2129 | * aligned, by adjusting the object size accordingly. */ | |
2130 | size += REDZONE_ALIGN - 1; | |
2131 | size &= ~(REDZONE_ALIGN - 1); | |
2132 | } | |
ca5f9703 | 2133 | |
a44b56d3 | 2134 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2135 | if (ralign < cachep->align) { |
2136 | ralign = cachep->align; | |
1da177e4 | 2137 | } |
3ff84a7f PE |
2138 | /* disable debug if necessary */ |
2139 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2140 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2141 | /* |
ca5f9703 | 2142 | * 4) Store it. |
1da177e4 | 2143 | */ |
8a13a4cc | 2144 | cachep->align = ralign; |
1da177e4 | 2145 | |
83b519e8 PE |
2146 | if (slab_is_available()) |
2147 | gfp = GFP_KERNEL; | |
2148 | else | |
2149 | gfp = GFP_NOWAIT; | |
2150 | ||
1da177e4 | 2151 | #if DEBUG |
1da177e4 | 2152 | |
ca5f9703 PE |
2153 | /* |
2154 | * Both debugging options require word-alignment which is calculated | |
2155 | * into align above. | |
2156 | */ | |
1da177e4 | 2157 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2158 | /* add space for red zone words */ |
3ff84a7f PE |
2159 | cachep->obj_offset += sizeof(unsigned long long); |
2160 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2161 | } |
2162 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2163 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2164 | * the real object. But if the second red zone needs to be |
2165 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2166 | */ |
87a927c7 DW |
2167 | if (flags & SLAB_RED_ZONE) |
2168 | size += REDZONE_ALIGN; | |
2169 | else | |
2170 | size += BYTES_PER_WORD; | |
1da177e4 | 2171 | } |
03a2d2a3 JK |
2172 | /* |
2173 | * To activate debug pagealloc, off-slab management is necessary | |
2174 | * requirement. In early phase of initialization, small sized slab | |
2175 | * doesn't get initialized so it would not be possible. So, we need | |
2176 | * to check size >= 256. It guarantees that all necessary small | |
2177 | * sized slab is initialized in current slab initialization sequence. | |
2178 | */ | |
40323278 | 2179 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
a307ebd4 | 2180 | !slab_early_init && size >= kmalloc_size(INDEX_NODE) && |
03a2d2a3 JK |
2181 | size >= 256 && cachep->object_size > cache_line_size() && |
2182 | ALIGN(size, cachep->align) < PAGE_SIZE) { | |
608da7e3 | 2183 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); |
1da177e4 LT |
2184 | size = PAGE_SIZE; |
2185 | } | |
1da177e4 LT |
2186 | #endif |
2187 | ||
e0a42726 IM |
2188 | /* |
2189 | * Determine if the slab management is 'on' or 'off' slab. | |
2190 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2191 | * it too early on. Always use on-slab management when |
2192 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2193 | */ |
d4322d88 | 2194 | if (size >= OFF_SLAB_MIN_SIZE && !slab_early_init && |
e7cb55b9 | 2195 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2196 | /* |
2197 | * Size is large, assume best to place the slab management obj | |
2198 | * off-slab (should allow better packing of objs). | |
2199 | */ | |
2200 | flags |= CFLGS_OFF_SLAB; | |
2201 | ||
8a13a4cc | 2202 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2203 | /* |
2204 | * We should restrict the number of objects in a slab to implement | |
2205 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2206 | */ | |
2207 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2208 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2209 | |
8a13a4cc | 2210 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2211 | |
8a13a4cc | 2212 | if (!cachep->num) |
278b1bb1 | 2213 | return -E2BIG; |
1da177e4 | 2214 | |
03787301 | 2215 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2216 | |
2217 | /* | |
2218 | * If the slab has been placed off-slab, and we have enough space then | |
2219 | * move it on-slab. This is at the expense of any extra colouring. | |
2220 | */ | |
8456a648 | 2221 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2222 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2223 | left_over -= freelist_size; |
1da177e4 LT |
2224 | } |
2225 | ||
2226 | if (flags & CFLGS_OFF_SLAB) { | |
2227 | /* really off slab. No need for manual alignment */ | |
03787301 | 2228 | freelist_size = calculate_freelist_size(cachep->num, 0); |
1da177e4 LT |
2229 | } |
2230 | ||
2231 | cachep->colour_off = cache_line_size(); | |
2232 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2233 | if (cachep->colour_off < cachep->align) |
2234 | cachep->colour_off = cachep->align; | |
b28a02de | 2235 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2236 | cachep->freelist_size = freelist_size; |
1da177e4 | 2237 | cachep->flags = flags; |
a57a4988 | 2238 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2239 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2240 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2241 | cachep->size = size; |
6a2d7a95 | 2242 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2243 | |
40b44137 JK |
2244 | #if DEBUG |
2245 | /* | |
2246 | * If we're going to use the generic kernel_map_pages() | |
2247 | * poisoning, then it's going to smash the contents of | |
2248 | * the redzone and userword anyhow, so switch them off. | |
2249 | */ | |
2250 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2251 | (cachep->flags & SLAB_POISON) && | |
2252 | is_debug_pagealloc_cache(cachep)) | |
2253 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2254 | #endif | |
2255 | ||
2256 | if (OFF_SLAB(cachep)) { | |
8456a648 | 2257 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2258 | /* |
5f0985bb | 2259 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2260 | * But since we go off slab only for object size greater than |
d4322d88 | 2261 | * OFF_SLAB_MIN_SIZE, and kmalloc_{dma,}_caches get created |
5f0985bb | 2262 | * in ascending order,this should not happen at all. |
e5ac9c5a RT |
2263 | * But leave a BUG_ON for some lucky dude. |
2264 | */ | |
8456a648 | 2265 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2266 | } |
1da177e4 | 2267 | |
278b1bb1 CL |
2268 | err = setup_cpu_cache(cachep, gfp); |
2269 | if (err) { | |
52b4b950 | 2270 | __kmem_cache_release(cachep); |
278b1bb1 | 2271 | return err; |
2ed3a4ef | 2272 | } |
1da177e4 | 2273 | |
278b1bb1 | 2274 | return 0; |
1da177e4 | 2275 | } |
1da177e4 LT |
2276 | |
2277 | #if DEBUG | |
2278 | static void check_irq_off(void) | |
2279 | { | |
2280 | BUG_ON(!irqs_disabled()); | |
2281 | } | |
2282 | ||
2283 | static void check_irq_on(void) | |
2284 | { | |
2285 | BUG_ON(irqs_disabled()); | |
2286 | } | |
2287 | ||
343e0d7a | 2288 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2289 | { |
2290 | #ifdef CONFIG_SMP | |
2291 | check_irq_off(); | |
18bf8541 | 2292 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2293 | #endif |
2294 | } | |
e498be7d | 2295 | |
343e0d7a | 2296 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2297 | { |
2298 | #ifdef CONFIG_SMP | |
2299 | check_irq_off(); | |
18bf8541 | 2300 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2301 | #endif |
2302 | } | |
2303 | ||
1da177e4 LT |
2304 | #else |
2305 | #define check_irq_off() do { } while(0) | |
2306 | #define check_irq_on() do { } while(0) | |
2307 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2308 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2309 | #endif |
2310 | ||
ce8eb6c4 | 2311 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2312 | struct array_cache *ac, |
2313 | int force, int node); | |
2314 | ||
1da177e4 LT |
2315 | static void do_drain(void *arg) |
2316 | { | |
a737b3e2 | 2317 | struct kmem_cache *cachep = arg; |
1da177e4 | 2318 | struct array_cache *ac; |
7d6e6d09 | 2319 | int node = numa_mem_id(); |
18bf8541 | 2320 | struct kmem_cache_node *n; |
97654dfa | 2321 | LIST_HEAD(list); |
1da177e4 LT |
2322 | |
2323 | check_irq_off(); | |
9a2dba4b | 2324 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2325 | n = get_node(cachep, node); |
2326 | spin_lock(&n->list_lock); | |
97654dfa | 2327 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2328 | spin_unlock(&n->list_lock); |
97654dfa | 2329 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2330 | ac->avail = 0; |
2331 | } | |
2332 | ||
343e0d7a | 2333 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2334 | { |
ce8eb6c4 | 2335 | struct kmem_cache_node *n; |
e498be7d CL |
2336 | int node; |
2337 | ||
15c8b6c1 | 2338 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2339 | check_irq_on(); |
18bf8541 CL |
2340 | for_each_kmem_cache_node(cachep, node, n) |
2341 | if (n->alien) | |
ce8eb6c4 | 2342 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2343 | |
18bf8541 CL |
2344 | for_each_kmem_cache_node(cachep, node, n) |
2345 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2346 | } |
2347 | ||
ed11d9eb CL |
2348 | /* |
2349 | * Remove slabs from the list of free slabs. | |
2350 | * Specify the number of slabs to drain in tofree. | |
2351 | * | |
2352 | * Returns the actual number of slabs released. | |
2353 | */ | |
2354 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2355 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2356 | { |
ed11d9eb CL |
2357 | struct list_head *p; |
2358 | int nr_freed; | |
8456a648 | 2359 | struct page *page; |
1da177e4 | 2360 | |
ed11d9eb | 2361 | nr_freed = 0; |
ce8eb6c4 | 2362 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2363 | |
ce8eb6c4 CL |
2364 | spin_lock_irq(&n->list_lock); |
2365 | p = n->slabs_free.prev; | |
2366 | if (p == &n->slabs_free) { | |
2367 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2368 | goto out; |
2369 | } | |
1da177e4 | 2370 | |
8456a648 | 2371 | page = list_entry(p, struct page, lru); |
8456a648 | 2372 | list_del(&page->lru); |
ed11d9eb CL |
2373 | /* |
2374 | * Safe to drop the lock. The slab is no longer linked | |
2375 | * to the cache. | |
2376 | */ | |
ce8eb6c4 CL |
2377 | n->free_objects -= cache->num; |
2378 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2379 | slab_destroy(cache, page); |
ed11d9eb | 2380 | nr_freed++; |
1da177e4 | 2381 | } |
ed11d9eb CL |
2382 | out: |
2383 | return nr_freed; | |
1da177e4 LT |
2384 | } |
2385 | ||
d6e0b7fa | 2386 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2387 | { |
18bf8541 CL |
2388 | int ret = 0; |
2389 | int node; | |
ce8eb6c4 | 2390 | struct kmem_cache_node *n; |
e498be7d CL |
2391 | |
2392 | drain_cpu_caches(cachep); | |
2393 | ||
2394 | check_irq_on(); | |
18bf8541 | 2395 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2396 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2397 | |
ce8eb6c4 CL |
2398 | ret += !list_empty(&n->slabs_full) || |
2399 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2400 | } |
2401 | return (ret ? 1 : 0); | |
2402 | } | |
2403 | ||
945cf2b6 | 2404 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2405 | { |
2406 | return __kmem_cache_shrink(cachep, false); | |
2407 | } | |
2408 | ||
2409 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2410 | { |
12c3667f | 2411 | int i; |
ce8eb6c4 | 2412 | struct kmem_cache_node *n; |
1da177e4 | 2413 | |
bf0dea23 | 2414 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2415 | |
ce8eb6c4 | 2416 | /* NUMA: free the node structures */ |
18bf8541 CL |
2417 | for_each_kmem_cache_node(cachep, i, n) { |
2418 | kfree(n->shared); | |
2419 | free_alien_cache(n->alien); | |
2420 | kfree(n); | |
2421 | cachep->node[i] = NULL; | |
12c3667f | 2422 | } |
1da177e4 | 2423 | } |
1da177e4 | 2424 | |
e5ac9c5a RT |
2425 | /* |
2426 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2427 | * |
2428 | * For a slab cache when the slab descriptor is off-slab, the | |
2429 | * slab descriptor can't come from the same cache which is being created, | |
2430 | * Because if it is the case, that means we defer the creation of | |
2431 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2432 | * And we eventually call down to __kmem_cache_create(), which | |
2433 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2434 | * This is a "chicken-and-egg" problem. | |
2435 | * | |
2436 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2437 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2438 | */ |
7e007355 | 2439 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2440 | struct page *page, int colour_off, |
2441 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2442 | { |
7e007355 | 2443 | void *freelist; |
0c3aa83e | 2444 | void *addr = page_address(page); |
b28a02de | 2445 | |
1da177e4 LT |
2446 | if (OFF_SLAB(cachep)) { |
2447 | /* Slab management obj is off-slab. */ | |
8456a648 | 2448 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2449 | local_flags, nodeid); |
8456a648 | 2450 | if (!freelist) |
1da177e4 LT |
2451 | return NULL; |
2452 | } else { | |
8456a648 JK |
2453 | freelist = addr + colour_off; |
2454 | colour_off += cachep->freelist_size; | |
1da177e4 | 2455 | } |
8456a648 JK |
2456 | page->active = 0; |
2457 | page->s_mem = addr + colour_off; | |
2458 | return freelist; | |
1da177e4 LT |
2459 | } |
2460 | ||
7cc68973 | 2461 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2462 | { |
a41adfaa | 2463 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2464 | } |
2465 | ||
2466 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2467 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2468 | { |
a41adfaa | 2469 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2470 | } |
2471 | ||
343e0d7a | 2472 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2473 | struct page *page) |
1da177e4 LT |
2474 | { |
2475 | int i; | |
2476 | ||
2477 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2478 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 | 2479 | #if DEBUG |
1da177e4 LT |
2480 | if (cachep->flags & SLAB_STORE_USER) |
2481 | *dbg_userword(cachep, objp) = NULL; | |
2482 | ||
2483 | if (cachep->flags & SLAB_RED_ZONE) { | |
2484 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2485 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2486 | } | |
2487 | /* | |
a737b3e2 AM |
2488 | * Constructors are not allowed to allocate memory from the same |
2489 | * cache which they are a constructor for. Otherwise, deadlock. | |
2490 | * They must also be threaded. | |
1da177e4 LT |
2491 | */ |
2492 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2493 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2494 | |
2495 | if (cachep->flags & SLAB_RED_ZONE) { | |
2496 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2497 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2498 | " end of an object"); |
1da177e4 LT |
2499 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2500 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2501 | " start of an object"); |
1da177e4 | 2502 | } |
40b44137 JK |
2503 | /* need to poison the objs? */ |
2504 | if (cachep->flags & SLAB_POISON) { | |
2505 | poison_obj(cachep, objp, POISON_FREE); | |
2506 | slab_kernel_map(cachep, objp, 0, 0); | |
2507 | } | |
1da177e4 LT |
2508 | #else |
2509 | if (cachep->ctor) | |
51cc5068 | 2510 | cachep->ctor(objp); |
1da177e4 | 2511 | #endif |
e5c58dfd | 2512 | set_free_obj(page, i, i); |
1da177e4 | 2513 | } |
1da177e4 LT |
2514 | } |
2515 | ||
343e0d7a | 2516 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2517 | { |
4b51d669 CL |
2518 | if (CONFIG_ZONE_DMA_FLAG) { |
2519 | if (flags & GFP_DMA) | |
a618e89f | 2520 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2521 | else |
a618e89f | 2522 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2523 | } |
1da177e4 LT |
2524 | } |
2525 | ||
260b61dd | 2526 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2527 | { |
b1cb0982 | 2528 | void *objp; |
78d382d7 | 2529 | |
e5c58dfd | 2530 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2531 | page->active++; |
78d382d7 | 2532 | |
d31676df JK |
2533 | #if DEBUG |
2534 | if (cachep->flags & SLAB_STORE_USER) | |
2535 | set_store_user_dirty(cachep); | |
2536 | #endif | |
2537 | ||
78d382d7 MD |
2538 | return objp; |
2539 | } | |
2540 | ||
260b61dd JK |
2541 | static void slab_put_obj(struct kmem_cache *cachep, |
2542 | struct page *page, void *objp) | |
78d382d7 | 2543 | { |
8456a648 | 2544 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2545 | #if DEBUG |
16025177 | 2546 | unsigned int i; |
b1cb0982 | 2547 | |
b1cb0982 | 2548 | /* Verify double free bug */ |
8456a648 | 2549 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2550 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2551 | printk(KERN_ERR "slab: double free detected in cache " |
2552 | "'%s', objp %p\n", cachep->name, objp); | |
2553 | BUG(); | |
2554 | } | |
78d382d7 MD |
2555 | } |
2556 | #endif | |
8456a648 | 2557 | page->active--; |
e5c58dfd | 2558 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2559 | } |
2560 | ||
4776874f PE |
2561 | /* |
2562 | * Map pages beginning at addr to the given cache and slab. This is required | |
2563 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2564 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2565 | */ |
8456a648 | 2566 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2567 | void *freelist) |
1da177e4 | 2568 | { |
a57a4988 | 2569 | page->slab_cache = cache; |
8456a648 | 2570 | page->freelist = freelist; |
1da177e4 LT |
2571 | } |
2572 | ||
2573 | /* | |
2574 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2575 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2576 | */ | |
3c517a61 | 2577 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2578 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2579 | { |
7e007355 | 2580 | void *freelist; |
b28a02de PE |
2581 | size_t offset; |
2582 | gfp_t local_flags; | |
ce8eb6c4 | 2583 | struct kmem_cache_node *n; |
1da177e4 | 2584 | |
a737b3e2 AM |
2585 | /* |
2586 | * Be lazy and only check for valid flags here, keeping it out of the | |
2587 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2588 | */ |
c871ac4e AM |
2589 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2590 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2591 | BUG(); | |
2592 | } | |
6cb06229 | 2593 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2594 | |
ce8eb6c4 | 2595 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2596 | check_irq_off(); |
18bf8541 | 2597 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2598 | spin_lock(&n->list_lock); |
1da177e4 LT |
2599 | |
2600 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2601 | offset = n->colour_next; |
2602 | n->colour_next++; | |
2603 | if (n->colour_next >= cachep->colour) | |
2604 | n->colour_next = 0; | |
2605 | spin_unlock(&n->list_lock); | |
1da177e4 | 2606 | |
2e1217cf | 2607 | offset *= cachep->colour_off; |
1da177e4 | 2608 | |
d0164adc | 2609 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2610 | local_irq_enable(); |
2611 | ||
2612 | /* | |
2613 | * The test for missing atomic flag is performed here, rather than | |
2614 | * the more obvious place, simply to reduce the critical path length | |
2615 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2616 | * will eventually be caught here (where it matters). | |
2617 | */ | |
2618 | kmem_flagcheck(cachep, flags); | |
2619 | ||
a737b3e2 AM |
2620 | /* |
2621 | * Get mem for the objs. Attempt to allocate a physical page from | |
2622 | * 'nodeid'. | |
e498be7d | 2623 | */ |
0c3aa83e JK |
2624 | if (!page) |
2625 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2626 | if (!page) | |
1da177e4 LT |
2627 | goto failed; |
2628 | ||
2629 | /* Get slab management. */ | |
8456a648 | 2630 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2631 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2632 | if (!freelist) |
1da177e4 LT |
2633 | goto opps1; |
2634 | ||
8456a648 | 2635 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2636 | |
8456a648 | 2637 | cache_init_objs(cachep, page); |
1da177e4 | 2638 | |
d0164adc | 2639 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2640 | local_irq_disable(); |
2641 | check_irq_off(); | |
ce8eb6c4 | 2642 | spin_lock(&n->list_lock); |
1da177e4 LT |
2643 | |
2644 | /* Make slab active. */ | |
8456a648 | 2645 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2646 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2647 | n->free_objects += cachep->num; |
2648 | spin_unlock(&n->list_lock); | |
1da177e4 | 2649 | return 1; |
a737b3e2 | 2650 | opps1: |
0c3aa83e | 2651 | kmem_freepages(cachep, page); |
a737b3e2 | 2652 | failed: |
d0164adc | 2653 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2654 | local_irq_disable(); |
2655 | return 0; | |
2656 | } | |
2657 | ||
2658 | #if DEBUG | |
2659 | ||
2660 | /* | |
2661 | * Perform extra freeing checks: | |
2662 | * - detect bad pointers. | |
2663 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2664 | */ |
2665 | static void kfree_debugcheck(const void *objp) | |
2666 | { | |
1da177e4 LT |
2667 | if (!virt_addr_valid(objp)) { |
2668 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2669 | (unsigned long)objp); |
2670 | BUG(); | |
1da177e4 | 2671 | } |
1da177e4 LT |
2672 | } |
2673 | ||
58ce1fd5 PE |
2674 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2675 | { | |
b46b8f19 | 2676 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2677 | |
2678 | redzone1 = *dbg_redzone1(cache, obj); | |
2679 | redzone2 = *dbg_redzone2(cache, obj); | |
2680 | ||
2681 | /* | |
2682 | * Redzone is ok. | |
2683 | */ | |
2684 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2685 | return; | |
2686 | ||
2687 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2688 | slab_error(cache, "double free detected"); | |
2689 | else | |
2690 | slab_error(cache, "memory outside object was overwritten"); | |
2691 | ||
b46b8f19 | 2692 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2693 | obj, redzone1, redzone2); |
2694 | } | |
2695 | ||
343e0d7a | 2696 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2697 | unsigned long caller) |
1da177e4 | 2698 | { |
1da177e4 | 2699 | unsigned int objnr; |
8456a648 | 2700 | struct page *page; |
1da177e4 | 2701 | |
80cbd911 MW |
2702 | BUG_ON(virt_to_cache(objp) != cachep); |
2703 | ||
3dafccf2 | 2704 | objp -= obj_offset(cachep); |
1da177e4 | 2705 | kfree_debugcheck(objp); |
b49af68f | 2706 | page = virt_to_head_page(objp); |
1da177e4 | 2707 | |
1da177e4 | 2708 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2709 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2710 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2711 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2712 | } | |
d31676df JK |
2713 | if (cachep->flags & SLAB_STORE_USER) { |
2714 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2715 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2716 | } |
1da177e4 | 2717 | |
8456a648 | 2718 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2719 | |
2720 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2721 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2722 | |
1da177e4 | 2723 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2724 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2725 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2726 | } |
2727 | return objp; | |
2728 | } | |
2729 | ||
1da177e4 LT |
2730 | #else |
2731 | #define kfree_debugcheck(x) do { } while(0) | |
2732 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2733 | #endif |
2734 | ||
7aa0d227 GT |
2735 | static struct page *get_first_slab(struct kmem_cache_node *n) |
2736 | { | |
2737 | struct page *page; | |
2738 | ||
2739 | page = list_first_entry_or_null(&n->slabs_partial, | |
2740 | struct page, lru); | |
2741 | if (!page) { | |
2742 | n->free_touched = 1; | |
2743 | page = list_first_entry_or_null(&n->slabs_free, | |
2744 | struct page, lru); | |
2745 | } | |
2746 | ||
2747 | return page; | |
2748 | } | |
2749 | ||
072bb0aa MG |
2750 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2751 | bool force_refill) | |
1da177e4 LT |
2752 | { |
2753 | int batchcount; | |
ce8eb6c4 | 2754 | struct kmem_cache_node *n; |
1da177e4 | 2755 | struct array_cache *ac; |
1ca4cb24 PE |
2756 | int node; |
2757 | ||
1da177e4 | 2758 | check_irq_off(); |
7d6e6d09 | 2759 | node = numa_mem_id(); |
072bb0aa MG |
2760 | if (unlikely(force_refill)) |
2761 | goto force_grow; | |
2762 | retry: | |
9a2dba4b | 2763 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2764 | batchcount = ac->batchcount; |
2765 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2766 | /* |
2767 | * If there was little recent activity on this cache, then | |
2768 | * perform only a partial refill. Otherwise we could generate | |
2769 | * refill bouncing. | |
1da177e4 LT |
2770 | */ |
2771 | batchcount = BATCHREFILL_LIMIT; | |
2772 | } | |
18bf8541 | 2773 | n = get_node(cachep, node); |
e498be7d | 2774 | |
ce8eb6c4 CL |
2775 | BUG_ON(ac->avail > 0 || !n); |
2776 | spin_lock(&n->list_lock); | |
1da177e4 | 2777 | |
3ded175a | 2778 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2779 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2780 | n->shared->touched = 1; | |
3ded175a | 2781 | goto alloc_done; |
44b57f1c | 2782 | } |
3ded175a | 2783 | |
1da177e4 | 2784 | while (batchcount > 0) { |
8456a648 | 2785 | struct page *page; |
1da177e4 | 2786 | /* Get slab alloc is to come from. */ |
7aa0d227 GT |
2787 | page = get_first_slab(n); |
2788 | if (!page) | |
2789 | goto must_grow; | |
1da177e4 | 2790 | |
1da177e4 | 2791 | check_spinlock_acquired(cachep); |
714b8171 PE |
2792 | |
2793 | /* | |
2794 | * The slab was either on partial or free list so | |
2795 | * there must be at least one object available for | |
2796 | * allocation. | |
2797 | */ | |
8456a648 | 2798 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2799 | |
8456a648 | 2800 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2801 | STATS_INC_ALLOCED(cachep); |
2802 | STATS_INC_ACTIVE(cachep); | |
2803 | STATS_SET_HIGH(cachep); | |
2804 | ||
260b61dd | 2805 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page)); |
1da177e4 | 2806 | } |
1da177e4 LT |
2807 | |
2808 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2809 | list_del(&page->lru); |
2810 | if (page->active == cachep->num) | |
34bf6ef9 | 2811 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2812 | else |
34bf6ef9 | 2813 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2814 | } |
2815 | ||
a737b3e2 | 2816 | must_grow: |
ce8eb6c4 | 2817 | n->free_objects -= ac->avail; |
a737b3e2 | 2818 | alloc_done: |
ce8eb6c4 | 2819 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2820 | |
2821 | if (unlikely(!ac->avail)) { | |
2822 | int x; | |
072bb0aa | 2823 | force_grow: |
4167e9b2 | 2824 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2825 | |
a737b3e2 | 2826 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2827 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2828 | node = numa_mem_id(); |
072bb0aa MG |
2829 | |
2830 | /* no objects in sight? abort */ | |
2831 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2832 | return NULL; |
2833 | ||
a737b3e2 | 2834 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2835 | goto retry; |
2836 | } | |
2837 | ac->touched = 1; | |
072bb0aa MG |
2838 | |
2839 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2840 | } |
2841 | ||
a737b3e2 AM |
2842 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2843 | gfp_t flags) | |
1da177e4 | 2844 | { |
d0164adc | 2845 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2846 | #if DEBUG |
2847 | kmem_flagcheck(cachep, flags); | |
2848 | #endif | |
2849 | } | |
2850 | ||
2851 | #if DEBUG | |
a737b3e2 | 2852 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2853 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2854 | { |
b28a02de | 2855 | if (!objp) |
1da177e4 | 2856 | return objp; |
b28a02de | 2857 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2858 | check_poison_obj(cachep, objp); |
40b44137 | 2859 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2860 | poison_obj(cachep, objp, POISON_INUSE); |
2861 | } | |
2862 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2863 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2864 | |
2865 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2866 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2867 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2868 | slab_error(cachep, "double free, or memory outside" | |
2869 | " object was overwritten"); | |
b28a02de | 2870 | printk(KERN_ERR |
b46b8f19 | 2871 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2872 | objp, *dbg_redzone1(cachep, objp), |
2873 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2874 | } |
2875 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2876 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2877 | } | |
03787301 | 2878 | |
3dafccf2 | 2879 | objp += obj_offset(cachep); |
4f104934 | 2880 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2881 | cachep->ctor(objp); |
7ea466f2 TH |
2882 | if (ARCH_SLAB_MINALIGN && |
2883 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2884 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2885 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2886 | } |
1da177e4 LT |
2887 | return objp; |
2888 | } | |
2889 | #else | |
2890 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2891 | #endif | |
2892 | ||
343e0d7a | 2893 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2894 | { |
b28a02de | 2895 | void *objp; |
1da177e4 | 2896 | struct array_cache *ac; |
072bb0aa | 2897 | bool force_refill = false; |
1da177e4 | 2898 | |
5c382300 | 2899 | check_irq_off(); |
8a8b6502 | 2900 | |
9a2dba4b | 2901 | ac = cpu_cache_get(cachep); |
1da177e4 | 2902 | if (likely(ac->avail)) { |
1da177e4 | 2903 | ac->touched = 1; |
072bb0aa MG |
2904 | objp = ac_get_obj(cachep, ac, flags, false); |
2905 | ||
ddbf2e83 | 2906 | /* |
072bb0aa MG |
2907 | * Allow for the possibility all avail objects are not allowed |
2908 | * by the current flags | |
ddbf2e83 | 2909 | */ |
072bb0aa MG |
2910 | if (objp) { |
2911 | STATS_INC_ALLOCHIT(cachep); | |
2912 | goto out; | |
2913 | } | |
2914 | force_refill = true; | |
1da177e4 | 2915 | } |
072bb0aa MG |
2916 | |
2917 | STATS_INC_ALLOCMISS(cachep); | |
2918 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
2919 | /* | |
2920 | * the 'ac' may be updated by cache_alloc_refill(), | |
2921 | * and kmemleak_erase() requires its correct value. | |
2922 | */ | |
2923 | ac = cpu_cache_get(cachep); | |
2924 | ||
2925 | out: | |
d5cff635 CM |
2926 | /* |
2927 | * To avoid a false negative, if an object that is in one of the | |
2928 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2929 | * treat the array pointers as a reference to the object. | |
2930 | */ | |
f3d8b53a O |
2931 | if (objp) |
2932 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2933 | return objp; |
2934 | } | |
2935 | ||
e498be7d | 2936 | #ifdef CONFIG_NUMA |
c61afb18 | 2937 | /* |
2ad654bc | 2938 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2939 | * |
2940 | * If we are in_interrupt, then process context, including cpusets and | |
2941 | * mempolicy, may not apply and should not be used for allocation policy. | |
2942 | */ | |
2943 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2944 | { | |
2945 | int nid_alloc, nid_here; | |
2946 | ||
765c4507 | 2947 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 2948 | return NULL; |
7d6e6d09 | 2949 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 2950 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 2951 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 2952 | else if (current->mempolicy) |
2a389610 | 2953 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 2954 | if (nid_alloc != nid_here) |
8b98c169 | 2955 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
2956 | return NULL; |
2957 | } | |
2958 | ||
765c4507 CL |
2959 | /* |
2960 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 2961 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 2962 | * available node for available objects. If that fails then we |
3c517a61 CL |
2963 | * perform an allocation without specifying a node. This allows the page |
2964 | * allocator to do its reclaim / fallback magic. We then insert the | |
2965 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 2966 | */ |
8c8cc2c1 | 2967 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 2968 | { |
8c8cc2c1 PE |
2969 | struct zonelist *zonelist; |
2970 | gfp_t local_flags; | |
dd1a239f | 2971 | struct zoneref *z; |
54a6eb5c MG |
2972 | struct zone *zone; |
2973 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 2974 | void *obj = NULL; |
3c517a61 | 2975 | int nid; |
cc9a6c87 | 2976 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
2977 | |
2978 | if (flags & __GFP_THISNODE) | |
2979 | return NULL; | |
2980 | ||
6cb06229 | 2981 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 2982 | |
cc9a6c87 | 2983 | retry_cpuset: |
d26914d1 | 2984 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2985 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 2986 | |
3c517a61 CL |
2987 | retry: |
2988 | /* | |
2989 | * Look through allowed nodes for objects available | |
2990 | * from existing per node queues. | |
2991 | */ | |
54a6eb5c MG |
2992 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
2993 | nid = zone_to_nid(zone); | |
aedb0eb1 | 2994 | |
061d7074 | 2995 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
2996 | get_node(cache, nid) && |
2997 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 2998 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 2999 | gfp_exact_node(flags), nid); |
481c5346 CL |
3000 | if (obj) |
3001 | break; | |
3002 | } | |
3c517a61 CL |
3003 | } |
3004 | ||
cfce6604 | 3005 | if (!obj) { |
3c517a61 CL |
3006 | /* |
3007 | * This allocation will be performed within the constraints | |
3008 | * of the current cpuset / memory policy requirements. | |
3009 | * We may trigger various forms of reclaim on the allowed | |
3010 | * set and go into memory reserves if necessary. | |
3011 | */ | |
0c3aa83e JK |
3012 | struct page *page; |
3013 | ||
d0164adc | 3014 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3015 | local_irq_enable(); |
3016 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3017 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3018 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3019 | local_irq_disable(); |
0c3aa83e | 3020 | if (page) { |
3c517a61 CL |
3021 | /* |
3022 | * Insert into the appropriate per node queues | |
3023 | */ | |
0c3aa83e JK |
3024 | nid = page_to_nid(page); |
3025 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3026 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3027 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3028 | if (!obj) |
3029 | /* | |
3030 | * Another processor may allocate the | |
3031 | * objects in the slab since we are | |
3032 | * not holding any locks. | |
3033 | */ | |
3034 | goto retry; | |
3035 | } else { | |
b6a60451 | 3036 | /* cache_grow already freed obj */ |
3c517a61 CL |
3037 | obj = NULL; |
3038 | } | |
3039 | } | |
aedb0eb1 | 3040 | } |
cc9a6c87 | 3041 | |
d26914d1 | 3042 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3043 | goto retry_cpuset; |
765c4507 CL |
3044 | return obj; |
3045 | } | |
3046 | ||
e498be7d CL |
3047 | /* |
3048 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3049 | */ |
8b98c169 | 3050 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3051 | int nodeid) |
e498be7d | 3052 | { |
8456a648 | 3053 | struct page *page; |
ce8eb6c4 | 3054 | struct kmem_cache_node *n; |
b28a02de | 3055 | void *obj; |
b28a02de PE |
3056 | int x; |
3057 | ||
7c3fbbdd | 3058 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3059 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3060 | BUG_ON(!n); |
b28a02de | 3061 | |
a737b3e2 | 3062 | retry: |
ca3b9b91 | 3063 | check_irq_off(); |
ce8eb6c4 | 3064 | spin_lock(&n->list_lock); |
7aa0d227 GT |
3065 | page = get_first_slab(n); |
3066 | if (!page) | |
3067 | goto must_grow; | |
b28a02de | 3068 | |
b28a02de | 3069 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3070 | |
3071 | STATS_INC_NODEALLOCS(cachep); | |
3072 | STATS_INC_ACTIVE(cachep); | |
3073 | STATS_SET_HIGH(cachep); | |
3074 | ||
8456a648 | 3075 | BUG_ON(page->active == cachep->num); |
b28a02de | 3076 | |
260b61dd | 3077 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3078 | n->free_objects--; |
b28a02de | 3079 | /* move slabp to correct slabp list: */ |
8456a648 | 3080 | list_del(&page->lru); |
b28a02de | 3081 | |
8456a648 JK |
3082 | if (page->active == cachep->num) |
3083 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3084 | else |
8456a648 | 3085 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3086 | |
ce8eb6c4 | 3087 | spin_unlock(&n->list_lock); |
b28a02de | 3088 | goto done; |
e498be7d | 3089 | |
a737b3e2 | 3090 | must_grow: |
ce8eb6c4 | 3091 | spin_unlock(&n->list_lock); |
4167e9b2 | 3092 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3093 | if (x) |
3094 | goto retry; | |
1da177e4 | 3095 | |
8c8cc2c1 | 3096 | return fallback_alloc(cachep, flags); |
e498be7d | 3097 | |
a737b3e2 | 3098 | done: |
b28a02de | 3099 | return obj; |
e498be7d | 3100 | } |
8c8cc2c1 | 3101 | |
8c8cc2c1 | 3102 | static __always_inline void * |
48356303 | 3103 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3104 | unsigned long caller) |
8c8cc2c1 PE |
3105 | { |
3106 | unsigned long save_flags; | |
3107 | void *ptr; | |
7d6e6d09 | 3108 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3109 | |
dcce284a | 3110 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3111 | cachep = slab_pre_alloc_hook(cachep, flags); |
3112 | if (unlikely(!cachep)) | |
824ebef1 AM |
3113 | return NULL; |
3114 | ||
8c8cc2c1 PE |
3115 | cache_alloc_debugcheck_before(cachep, flags); |
3116 | local_irq_save(save_flags); | |
3117 | ||
eacbbae3 | 3118 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3119 | nodeid = slab_node; |
8c8cc2c1 | 3120 | |
18bf8541 | 3121 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3122 | /* Node not bootstrapped yet */ |
3123 | ptr = fallback_alloc(cachep, flags); | |
3124 | goto out; | |
3125 | } | |
3126 | ||
7d6e6d09 | 3127 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3128 | /* |
3129 | * Use the locally cached objects if possible. | |
3130 | * However ____cache_alloc does not allow fallback | |
3131 | * to other nodes. It may fail while we still have | |
3132 | * objects on other nodes available. | |
3133 | */ | |
3134 | ptr = ____cache_alloc(cachep, flags); | |
3135 | if (ptr) | |
3136 | goto out; | |
3137 | } | |
3138 | /* ___cache_alloc_node can fall back to other nodes */ | |
3139 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3140 | out: | |
3141 | local_irq_restore(save_flags); | |
3142 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3143 | ||
d5e3ed66 JDB |
3144 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3145 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3146 | |
d5e3ed66 | 3147 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3148 | return ptr; |
3149 | } | |
3150 | ||
3151 | static __always_inline void * | |
3152 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3153 | { | |
3154 | void *objp; | |
3155 | ||
2ad654bc | 3156 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3157 | objp = alternate_node_alloc(cache, flags); |
3158 | if (objp) | |
3159 | goto out; | |
3160 | } | |
3161 | objp = ____cache_alloc(cache, flags); | |
3162 | ||
3163 | /* | |
3164 | * We may just have run out of memory on the local node. | |
3165 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3166 | */ | |
7d6e6d09 LS |
3167 | if (!objp) |
3168 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3169 | |
3170 | out: | |
3171 | return objp; | |
3172 | } | |
3173 | #else | |
3174 | ||
3175 | static __always_inline void * | |
3176 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3177 | { | |
3178 | return ____cache_alloc(cachep, flags); | |
3179 | } | |
3180 | ||
3181 | #endif /* CONFIG_NUMA */ | |
3182 | ||
3183 | static __always_inline void * | |
48356303 | 3184 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3185 | { |
3186 | unsigned long save_flags; | |
3187 | void *objp; | |
3188 | ||
dcce284a | 3189 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3190 | cachep = slab_pre_alloc_hook(cachep, flags); |
3191 | if (unlikely(!cachep)) | |
824ebef1 AM |
3192 | return NULL; |
3193 | ||
8c8cc2c1 PE |
3194 | cache_alloc_debugcheck_before(cachep, flags); |
3195 | local_irq_save(save_flags); | |
3196 | objp = __do_cache_alloc(cachep, flags); | |
3197 | local_irq_restore(save_flags); | |
3198 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3199 | prefetchw(objp); | |
3200 | ||
d5e3ed66 JDB |
3201 | if (unlikely(flags & __GFP_ZERO) && objp) |
3202 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3203 | |
d5e3ed66 | 3204 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3205 | return objp; |
3206 | } | |
e498be7d CL |
3207 | |
3208 | /* | |
5f0985bb | 3209 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3210 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3211 | */ |
97654dfa JK |
3212 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3213 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3214 | { |
3215 | int i; | |
25c063fb | 3216 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3217 | |
3218 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3219 | void *objp; |
8456a648 | 3220 | struct page *page; |
1da177e4 | 3221 | |
072bb0aa MG |
3222 | clear_obj_pfmemalloc(&objpp[i]); |
3223 | objp = objpp[i]; | |
3224 | ||
8456a648 | 3225 | page = virt_to_head_page(objp); |
8456a648 | 3226 | list_del(&page->lru); |
ff69416e | 3227 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3228 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3229 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3230 | n->free_objects++; |
1da177e4 LT |
3231 | |
3232 | /* fixup slab chains */ | |
8456a648 | 3233 | if (page->active == 0) { |
ce8eb6c4 CL |
3234 | if (n->free_objects > n->free_limit) { |
3235 | n->free_objects -= cachep->num; | |
97654dfa | 3236 | list_add_tail(&page->lru, list); |
1da177e4 | 3237 | } else { |
8456a648 | 3238 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3239 | } |
3240 | } else { | |
3241 | /* Unconditionally move a slab to the end of the | |
3242 | * partial list on free - maximum time for the | |
3243 | * other objects to be freed, too. | |
3244 | */ | |
8456a648 | 3245 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3246 | } |
3247 | } | |
3248 | } | |
3249 | ||
343e0d7a | 3250 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3251 | { |
3252 | int batchcount; | |
ce8eb6c4 | 3253 | struct kmem_cache_node *n; |
7d6e6d09 | 3254 | int node = numa_mem_id(); |
97654dfa | 3255 | LIST_HEAD(list); |
1da177e4 LT |
3256 | |
3257 | batchcount = ac->batchcount; | |
260b61dd | 3258 | |
1da177e4 | 3259 | check_irq_off(); |
18bf8541 | 3260 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3261 | spin_lock(&n->list_lock); |
3262 | if (n->shared) { | |
3263 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3264 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3265 | if (max) { |
3266 | if (batchcount > max) | |
3267 | batchcount = max; | |
e498be7d | 3268 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3269 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3270 | shared_array->avail += batchcount; |
3271 | goto free_done; | |
3272 | } | |
3273 | } | |
3274 | ||
97654dfa | 3275 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3276 | free_done: |
1da177e4 LT |
3277 | #if STATS |
3278 | { | |
3279 | int i = 0; | |
73c0219d | 3280 | struct page *page; |
1da177e4 | 3281 | |
73c0219d | 3282 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3283 | BUG_ON(page->active); |
1da177e4 LT |
3284 | |
3285 | i++; | |
1da177e4 LT |
3286 | } |
3287 | STATS_SET_FREEABLE(cachep, i); | |
3288 | } | |
3289 | #endif | |
ce8eb6c4 | 3290 | spin_unlock(&n->list_lock); |
97654dfa | 3291 | slabs_destroy(cachep, &list); |
1da177e4 | 3292 | ac->avail -= batchcount; |
a737b3e2 | 3293 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3294 | } |
3295 | ||
3296 | /* | |
a737b3e2 AM |
3297 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3298 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3299 | */ |
a947eb95 | 3300 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3301 | unsigned long caller) |
1da177e4 | 3302 | { |
9a2dba4b | 3303 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3304 | |
3305 | check_irq_off(); | |
d5cff635 | 3306 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3307 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3308 | |
8c138bc0 | 3309 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3310 | |
1807a1aa SS |
3311 | /* |
3312 | * Skip calling cache_free_alien() when the platform is not numa. | |
3313 | * This will avoid cache misses that happen while accessing slabp (which | |
3314 | * is per page memory reference) to get nodeid. Instead use a global | |
3315 | * variable to skip the call, which is mostly likely to be present in | |
3316 | * the cache. | |
3317 | */ | |
b6e68bc1 | 3318 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3319 | return; |
3320 | ||
3d880194 | 3321 | if (ac->avail < ac->limit) { |
1da177e4 | 3322 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3323 | } else { |
3324 | STATS_INC_FREEMISS(cachep); | |
3325 | cache_flusharray(cachep, ac); | |
1da177e4 | 3326 | } |
42c8c99c | 3327 | |
072bb0aa | 3328 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3329 | } |
3330 | ||
3331 | /** | |
3332 | * kmem_cache_alloc - Allocate an object | |
3333 | * @cachep: The cache to allocate from. | |
3334 | * @flags: See kmalloc(). | |
3335 | * | |
3336 | * Allocate an object from this cache. The flags are only relevant | |
3337 | * if the cache has no available objects. | |
3338 | */ | |
343e0d7a | 3339 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3340 | { |
48356303 | 3341 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3342 | |
ca2b84cb | 3343 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3344 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3345 | |
3346 | return ret; | |
1da177e4 LT |
3347 | } |
3348 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3349 | ||
7b0501dd JDB |
3350 | static __always_inline void |
3351 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3352 | size_t size, void **p, unsigned long caller) | |
3353 | { | |
3354 | size_t i; | |
3355 | ||
3356 | for (i = 0; i < size; i++) | |
3357 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3358 | } | |
3359 | ||
865762a8 | 3360 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3361 | void **p) |
484748f0 | 3362 | { |
2a777eac JDB |
3363 | size_t i; |
3364 | ||
3365 | s = slab_pre_alloc_hook(s, flags); | |
3366 | if (!s) | |
3367 | return 0; | |
3368 | ||
3369 | cache_alloc_debugcheck_before(s, flags); | |
3370 | ||
3371 | local_irq_disable(); | |
3372 | for (i = 0; i < size; i++) { | |
3373 | void *objp = __do_cache_alloc(s, flags); | |
3374 | ||
2a777eac JDB |
3375 | if (unlikely(!objp)) |
3376 | goto error; | |
3377 | p[i] = objp; | |
3378 | } | |
3379 | local_irq_enable(); | |
3380 | ||
7b0501dd JDB |
3381 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3382 | ||
2a777eac JDB |
3383 | /* Clear memory outside IRQ disabled section */ |
3384 | if (unlikely(flags & __GFP_ZERO)) | |
3385 | for (i = 0; i < size; i++) | |
3386 | memset(p[i], 0, s->object_size); | |
3387 | ||
3388 | slab_post_alloc_hook(s, flags, size, p); | |
3389 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3390 | return size; | |
3391 | error: | |
3392 | local_irq_enable(); | |
7b0501dd | 3393 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3394 | slab_post_alloc_hook(s, flags, i, p); |
3395 | __kmem_cache_free_bulk(s, i, p); | |
3396 | return 0; | |
484748f0 CL |
3397 | } |
3398 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3399 | ||
0f24f128 | 3400 | #ifdef CONFIG_TRACING |
85beb586 | 3401 | void * |
4052147c | 3402 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3403 | { |
85beb586 SR |
3404 | void *ret; |
3405 | ||
48356303 | 3406 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3407 | |
3408 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3409 | size, cachep->size, flags); |
85beb586 | 3410 | return ret; |
36555751 | 3411 | } |
85beb586 | 3412 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3413 | #endif |
3414 | ||
1da177e4 | 3415 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3416 | /** |
3417 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3418 | * @cachep: The cache to allocate from. | |
3419 | * @flags: See kmalloc(). | |
3420 | * @nodeid: node number of the target node. | |
3421 | * | |
3422 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3423 | * node, which can improve the performance for cpu bound structures. | |
3424 | * | |
3425 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3426 | */ | |
8b98c169 CH |
3427 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3428 | { | |
48356303 | 3429 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3430 | |
ca2b84cb | 3431 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3432 | cachep->object_size, cachep->size, |
ca2b84cb | 3433 | flags, nodeid); |
36555751 EGM |
3434 | |
3435 | return ret; | |
8b98c169 | 3436 | } |
1da177e4 LT |
3437 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3438 | ||
0f24f128 | 3439 | #ifdef CONFIG_TRACING |
4052147c | 3440 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3441 | gfp_t flags, |
4052147c EG |
3442 | int nodeid, |
3443 | size_t size) | |
36555751 | 3444 | { |
85beb586 SR |
3445 | void *ret; |
3446 | ||
592f4145 | 3447 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3448 | |
85beb586 | 3449 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3450 | size, cachep->size, |
85beb586 SR |
3451 | flags, nodeid); |
3452 | return ret; | |
36555751 | 3453 | } |
85beb586 | 3454 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3455 | #endif |
3456 | ||
8b98c169 | 3457 | static __always_inline void * |
7c0cb9c6 | 3458 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3459 | { |
343e0d7a | 3460 | struct kmem_cache *cachep; |
97e2bde4 | 3461 | |
2c59dd65 | 3462 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3463 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3464 | return cachep; | |
4052147c | 3465 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3466 | } |
8b98c169 | 3467 | |
8b98c169 CH |
3468 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3469 | { | |
7c0cb9c6 | 3470 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3471 | } |
dbe5e69d | 3472 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3473 | |
3474 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3475 | int node, unsigned long caller) |
8b98c169 | 3476 | { |
7c0cb9c6 | 3477 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3478 | } |
3479 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3480 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3481 | |
3482 | /** | |
800590f5 | 3483 | * __do_kmalloc - allocate memory |
1da177e4 | 3484 | * @size: how many bytes of memory are required. |
800590f5 | 3485 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3486 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3487 | */ |
7fd6b141 | 3488 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3489 | unsigned long caller) |
1da177e4 | 3490 | { |
343e0d7a | 3491 | struct kmem_cache *cachep; |
36555751 | 3492 | void *ret; |
1da177e4 | 3493 | |
2c59dd65 | 3494 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3495 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3496 | return cachep; | |
48356303 | 3497 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3498 | |
7c0cb9c6 | 3499 | trace_kmalloc(caller, ret, |
3b0efdfa | 3500 | size, cachep->size, flags); |
36555751 EGM |
3501 | |
3502 | return ret; | |
7fd6b141 PE |
3503 | } |
3504 | ||
7fd6b141 PE |
3505 | void *__kmalloc(size_t size, gfp_t flags) |
3506 | { | |
7c0cb9c6 | 3507 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3508 | } |
3509 | EXPORT_SYMBOL(__kmalloc); | |
3510 | ||
ce71e27c | 3511 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3512 | { |
7c0cb9c6 | 3513 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3514 | } |
3515 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3516 | |
1da177e4 LT |
3517 | /** |
3518 | * kmem_cache_free - Deallocate an object | |
3519 | * @cachep: The cache the allocation was from. | |
3520 | * @objp: The previously allocated object. | |
3521 | * | |
3522 | * Free an object which was previously allocated from this | |
3523 | * cache. | |
3524 | */ | |
343e0d7a | 3525 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3526 | { |
3527 | unsigned long flags; | |
b9ce5ef4 GC |
3528 | cachep = cache_from_obj(cachep, objp); |
3529 | if (!cachep) | |
3530 | return; | |
1da177e4 LT |
3531 | |
3532 | local_irq_save(flags); | |
d97d476b | 3533 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3534 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3535 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3536 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3537 | local_irq_restore(flags); |
36555751 | 3538 | |
ca2b84cb | 3539 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3540 | } |
3541 | EXPORT_SYMBOL(kmem_cache_free); | |
3542 | ||
e6cdb58d JDB |
3543 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3544 | { | |
3545 | struct kmem_cache *s; | |
3546 | size_t i; | |
3547 | ||
3548 | local_irq_disable(); | |
3549 | for (i = 0; i < size; i++) { | |
3550 | void *objp = p[i]; | |
3551 | ||
ca257195 JDB |
3552 | if (!orig_s) /* called via kfree_bulk */ |
3553 | s = virt_to_cache(objp); | |
3554 | else | |
3555 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3556 | |
3557 | debug_check_no_locks_freed(objp, s->object_size); | |
3558 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3559 | debug_check_no_obj_freed(objp, s->object_size); | |
3560 | ||
3561 | __cache_free(s, objp, _RET_IP_); | |
3562 | } | |
3563 | local_irq_enable(); | |
3564 | ||
3565 | /* FIXME: add tracing */ | |
3566 | } | |
3567 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3568 | ||
1da177e4 LT |
3569 | /** |
3570 | * kfree - free previously allocated memory | |
3571 | * @objp: pointer returned by kmalloc. | |
3572 | * | |
80e93eff PE |
3573 | * If @objp is NULL, no operation is performed. |
3574 | * | |
1da177e4 LT |
3575 | * Don't free memory not originally allocated by kmalloc() |
3576 | * or you will run into trouble. | |
3577 | */ | |
3578 | void kfree(const void *objp) | |
3579 | { | |
343e0d7a | 3580 | struct kmem_cache *c; |
1da177e4 LT |
3581 | unsigned long flags; |
3582 | ||
2121db74 PE |
3583 | trace_kfree(_RET_IP_, objp); |
3584 | ||
6cb8f913 | 3585 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3586 | return; |
3587 | local_irq_save(flags); | |
3588 | kfree_debugcheck(objp); | |
6ed5eb22 | 3589 | c = virt_to_cache(objp); |
8c138bc0 CL |
3590 | debug_check_no_locks_freed(objp, c->object_size); |
3591 | ||
3592 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3593 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3594 | local_irq_restore(flags); |
3595 | } | |
3596 | EXPORT_SYMBOL(kfree); | |
3597 | ||
e498be7d | 3598 | /* |
ce8eb6c4 | 3599 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3600 | */ |
5f0985bb | 3601 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3602 | { |
3603 | int node; | |
ce8eb6c4 | 3604 | struct kmem_cache_node *n; |
cafeb02e | 3605 | struct array_cache *new_shared; |
c8522a3a | 3606 | struct alien_cache **new_alien = NULL; |
e498be7d | 3607 | |
9c09a95c | 3608 | for_each_online_node(node) { |
cafeb02e | 3609 | |
b455def2 L |
3610 | if (use_alien_caches) { |
3611 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3612 | if (!new_alien) | |
3613 | goto fail; | |
3614 | } | |
cafeb02e | 3615 | |
63109846 ED |
3616 | new_shared = NULL; |
3617 | if (cachep->shared) { | |
3618 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3619 | cachep->shared*cachep->batchcount, |
83b519e8 | 3620 | 0xbaadf00d, gfp); |
63109846 ED |
3621 | if (!new_shared) { |
3622 | free_alien_cache(new_alien); | |
3623 | goto fail; | |
3624 | } | |
0718dc2a | 3625 | } |
cafeb02e | 3626 | |
18bf8541 | 3627 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3628 | if (n) { |
3629 | struct array_cache *shared = n->shared; | |
97654dfa | 3630 | LIST_HEAD(list); |
cafeb02e | 3631 | |
ce8eb6c4 | 3632 | spin_lock_irq(&n->list_lock); |
e498be7d | 3633 | |
cafeb02e | 3634 | if (shared) |
0718dc2a | 3635 | free_block(cachep, shared->entry, |
97654dfa | 3636 | shared->avail, node, &list); |
e498be7d | 3637 | |
ce8eb6c4 CL |
3638 | n->shared = new_shared; |
3639 | if (!n->alien) { | |
3640 | n->alien = new_alien; | |
e498be7d CL |
3641 | new_alien = NULL; |
3642 | } | |
ce8eb6c4 | 3643 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3644 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3645 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3646 | slabs_destroy(cachep, &list); |
cafeb02e | 3647 | kfree(shared); |
e498be7d CL |
3648 | free_alien_cache(new_alien); |
3649 | continue; | |
3650 | } | |
ce8eb6c4 CL |
3651 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3652 | if (!n) { | |
0718dc2a CL |
3653 | free_alien_cache(new_alien); |
3654 | kfree(new_shared); | |
e498be7d | 3655 | goto fail; |
0718dc2a | 3656 | } |
e498be7d | 3657 | |
ce8eb6c4 | 3658 | kmem_cache_node_init(n); |
5f0985bb JZ |
3659 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3660 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3661 | n->shared = new_shared; |
3662 | n->alien = new_alien; | |
3663 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3664 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3665 | cachep->node[node] = n; |
e498be7d | 3666 | } |
cafeb02e | 3667 | return 0; |
0718dc2a | 3668 | |
a737b3e2 | 3669 | fail: |
3b0efdfa | 3670 | if (!cachep->list.next) { |
0718dc2a CL |
3671 | /* Cache is not active yet. Roll back what we did */ |
3672 | node--; | |
3673 | while (node >= 0) { | |
18bf8541 CL |
3674 | n = get_node(cachep, node); |
3675 | if (n) { | |
ce8eb6c4 CL |
3676 | kfree(n->shared); |
3677 | free_alien_cache(n->alien); | |
3678 | kfree(n); | |
6a67368c | 3679 | cachep->node[node] = NULL; |
0718dc2a CL |
3680 | } |
3681 | node--; | |
3682 | } | |
3683 | } | |
cafeb02e | 3684 | return -ENOMEM; |
e498be7d CL |
3685 | } |
3686 | ||
18004c5d | 3687 | /* Always called with the slab_mutex held */ |
943a451a | 3688 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3689 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3690 | { |
bf0dea23 JK |
3691 | struct array_cache __percpu *cpu_cache, *prev; |
3692 | int cpu; | |
1da177e4 | 3693 | |
bf0dea23 JK |
3694 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3695 | if (!cpu_cache) | |
d2e7b7d0 SS |
3696 | return -ENOMEM; |
3697 | ||
bf0dea23 JK |
3698 | prev = cachep->cpu_cache; |
3699 | cachep->cpu_cache = cpu_cache; | |
3700 | kick_all_cpus_sync(); | |
e498be7d | 3701 | |
1da177e4 | 3702 | check_irq_on(); |
1da177e4 LT |
3703 | cachep->batchcount = batchcount; |
3704 | cachep->limit = limit; | |
e498be7d | 3705 | cachep->shared = shared; |
1da177e4 | 3706 | |
bf0dea23 JK |
3707 | if (!prev) |
3708 | goto alloc_node; | |
3709 | ||
3710 | for_each_online_cpu(cpu) { | |
97654dfa | 3711 | LIST_HEAD(list); |
18bf8541 CL |
3712 | int node; |
3713 | struct kmem_cache_node *n; | |
bf0dea23 | 3714 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3715 | |
bf0dea23 | 3716 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3717 | n = get_node(cachep, node); |
3718 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3719 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3720 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3721 | slabs_destroy(cachep, &list); |
1da177e4 | 3722 | } |
bf0dea23 JK |
3723 | free_percpu(prev); |
3724 | ||
3725 | alloc_node: | |
5f0985bb | 3726 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3727 | } |
3728 | ||
943a451a GC |
3729 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3730 | int batchcount, int shared, gfp_t gfp) | |
3731 | { | |
3732 | int ret; | |
426589f5 | 3733 | struct kmem_cache *c; |
943a451a GC |
3734 | |
3735 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3736 | ||
3737 | if (slab_state < FULL) | |
3738 | return ret; | |
3739 | ||
3740 | if ((ret < 0) || !is_root_cache(cachep)) | |
3741 | return ret; | |
3742 | ||
426589f5 VD |
3743 | lockdep_assert_held(&slab_mutex); |
3744 | for_each_memcg_cache(c, cachep) { | |
3745 | /* return value determined by the root cache only */ | |
3746 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3747 | } |
3748 | ||
3749 | return ret; | |
3750 | } | |
3751 | ||
18004c5d | 3752 | /* Called with slab_mutex held always */ |
83b519e8 | 3753 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3754 | { |
3755 | int err; | |
943a451a GC |
3756 | int limit = 0; |
3757 | int shared = 0; | |
3758 | int batchcount = 0; | |
3759 | ||
3760 | if (!is_root_cache(cachep)) { | |
3761 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3762 | limit = root->limit; | |
3763 | shared = root->shared; | |
3764 | batchcount = root->batchcount; | |
3765 | } | |
1da177e4 | 3766 | |
943a451a GC |
3767 | if (limit && shared && batchcount) |
3768 | goto skip_setup; | |
a737b3e2 AM |
3769 | /* |
3770 | * The head array serves three purposes: | |
1da177e4 LT |
3771 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3772 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3773 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3774 | * bufctl chains: array operations are cheaper. |
3775 | * The numbers are guessed, we should auto-tune as described by | |
3776 | * Bonwick. | |
3777 | */ | |
3b0efdfa | 3778 | if (cachep->size > 131072) |
1da177e4 | 3779 | limit = 1; |
3b0efdfa | 3780 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3781 | limit = 8; |
3b0efdfa | 3782 | else if (cachep->size > 1024) |
1da177e4 | 3783 | limit = 24; |
3b0efdfa | 3784 | else if (cachep->size > 256) |
1da177e4 LT |
3785 | limit = 54; |
3786 | else | |
3787 | limit = 120; | |
3788 | ||
a737b3e2 AM |
3789 | /* |
3790 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3791 | * allocation behaviour: Most allocs on one cpu, most free operations |
3792 | * on another cpu. For these cases, an efficient object passing between | |
3793 | * cpus is necessary. This is provided by a shared array. The array | |
3794 | * replaces Bonwick's magazine layer. | |
3795 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3796 | * to a larger limit. Thus disabled by default. | |
3797 | */ | |
3798 | shared = 0; | |
3b0efdfa | 3799 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3800 | shared = 8; |
1da177e4 LT |
3801 | |
3802 | #if DEBUG | |
a737b3e2 AM |
3803 | /* |
3804 | * With debugging enabled, large batchcount lead to excessively long | |
3805 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3806 | */ |
3807 | if (limit > 32) | |
3808 | limit = 32; | |
3809 | #endif | |
943a451a GC |
3810 | batchcount = (limit + 1) / 2; |
3811 | skip_setup: | |
3812 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3813 | if (err) |
3814 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3815 | cachep->name, -err); |
2ed3a4ef | 3816 | return err; |
1da177e4 LT |
3817 | } |
3818 | ||
1b55253a | 3819 | /* |
ce8eb6c4 CL |
3820 | * Drain an array if it contains any elements taking the node lock only if |
3821 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3822 | * if drain_array() is used on the shared array. |
1b55253a | 3823 | */ |
ce8eb6c4 | 3824 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3825 | struct array_cache *ac, int force, int node) |
1da177e4 | 3826 | { |
97654dfa | 3827 | LIST_HEAD(list); |
1da177e4 LT |
3828 | int tofree; |
3829 | ||
1b55253a CL |
3830 | if (!ac || !ac->avail) |
3831 | return; | |
1da177e4 LT |
3832 | if (ac->touched && !force) { |
3833 | ac->touched = 0; | |
b18e7e65 | 3834 | } else { |
ce8eb6c4 | 3835 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3836 | if (ac->avail) { |
3837 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3838 | if (tofree > ac->avail) | |
3839 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3840 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3841 | ac->avail -= tofree; |
3842 | memmove(ac->entry, &(ac->entry[tofree]), | |
3843 | sizeof(void *) * ac->avail); | |
3844 | } | |
ce8eb6c4 | 3845 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3846 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3847 | } |
3848 | } | |
3849 | ||
3850 | /** | |
3851 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3852 | * @w: work descriptor |
1da177e4 LT |
3853 | * |
3854 | * Called from workqueue/eventd every few seconds. | |
3855 | * Purpose: | |
3856 | * - clear the per-cpu caches for this CPU. | |
3857 | * - return freeable pages to the main free memory pool. | |
3858 | * | |
a737b3e2 AM |
3859 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3860 | * again on the next iteration. | |
1da177e4 | 3861 | */ |
7c5cae36 | 3862 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3863 | { |
7a7c381d | 3864 | struct kmem_cache *searchp; |
ce8eb6c4 | 3865 | struct kmem_cache_node *n; |
7d6e6d09 | 3866 | int node = numa_mem_id(); |
bf6aede7 | 3867 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3868 | |
18004c5d | 3869 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3870 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3871 | goto out; |
1da177e4 | 3872 | |
18004c5d | 3873 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3874 | check_irq_on(); |
3875 | ||
35386e3b | 3876 | /* |
ce8eb6c4 | 3877 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3878 | * have established with reasonable certainty that |
3879 | * we can do some work if the lock was obtained. | |
3880 | */ | |
18bf8541 | 3881 | n = get_node(searchp, node); |
35386e3b | 3882 | |
ce8eb6c4 | 3883 | reap_alien(searchp, n); |
1da177e4 | 3884 | |
ce8eb6c4 | 3885 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3886 | |
35386e3b CL |
3887 | /* |
3888 | * These are racy checks but it does not matter | |
3889 | * if we skip one check or scan twice. | |
3890 | */ | |
ce8eb6c4 | 3891 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3892 | goto next; |
1da177e4 | 3893 | |
5f0985bb | 3894 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3895 | |
ce8eb6c4 | 3896 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3897 | |
ce8eb6c4 CL |
3898 | if (n->free_touched) |
3899 | n->free_touched = 0; | |
ed11d9eb CL |
3900 | else { |
3901 | int freed; | |
1da177e4 | 3902 | |
ce8eb6c4 | 3903 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3904 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3905 | STATS_ADD_REAPED(searchp, freed); | |
3906 | } | |
35386e3b | 3907 | next: |
1da177e4 LT |
3908 | cond_resched(); |
3909 | } | |
3910 | check_irq_on(); | |
18004c5d | 3911 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3912 | next_reap_node(); |
7c5cae36 | 3913 | out: |
a737b3e2 | 3914 | /* Set up the next iteration */ |
5f0985bb | 3915 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3916 | } |
3917 | ||
158a9624 | 3918 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3919 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3920 | { |
8456a648 | 3921 | struct page *page; |
b28a02de PE |
3922 | unsigned long active_objs; |
3923 | unsigned long num_objs; | |
3924 | unsigned long active_slabs = 0; | |
3925 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3926 | const char *name; |
1da177e4 | 3927 | char *error = NULL; |
e498be7d | 3928 | int node; |
ce8eb6c4 | 3929 | struct kmem_cache_node *n; |
1da177e4 | 3930 | |
1da177e4 LT |
3931 | active_objs = 0; |
3932 | num_slabs = 0; | |
18bf8541 | 3933 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3934 | |
ca3b9b91 | 3935 | check_irq_on(); |
ce8eb6c4 | 3936 | spin_lock_irq(&n->list_lock); |
e498be7d | 3937 | |
8456a648 JK |
3938 | list_for_each_entry(page, &n->slabs_full, lru) { |
3939 | if (page->active != cachep->num && !error) | |
e498be7d CL |
3940 | error = "slabs_full accounting error"; |
3941 | active_objs += cachep->num; | |
3942 | active_slabs++; | |
3943 | } | |
8456a648 JK |
3944 | list_for_each_entry(page, &n->slabs_partial, lru) { |
3945 | if (page->active == cachep->num && !error) | |
106a74e1 | 3946 | error = "slabs_partial accounting error"; |
8456a648 | 3947 | if (!page->active && !error) |
106a74e1 | 3948 | error = "slabs_partial accounting error"; |
8456a648 | 3949 | active_objs += page->active; |
e498be7d CL |
3950 | active_slabs++; |
3951 | } | |
8456a648 JK |
3952 | list_for_each_entry(page, &n->slabs_free, lru) { |
3953 | if (page->active && !error) | |
106a74e1 | 3954 | error = "slabs_free accounting error"; |
e498be7d CL |
3955 | num_slabs++; |
3956 | } | |
ce8eb6c4 CL |
3957 | free_objects += n->free_objects; |
3958 | if (n->shared) | |
3959 | shared_avail += n->shared->avail; | |
e498be7d | 3960 | |
ce8eb6c4 | 3961 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 3962 | } |
b28a02de PE |
3963 | num_slabs += active_slabs; |
3964 | num_objs = num_slabs * cachep->num; | |
e498be7d | 3965 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
3966 | error = "free_objects accounting error"; |
3967 | ||
b28a02de | 3968 | name = cachep->name; |
1da177e4 LT |
3969 | if (error) |
3970 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
3971 | ||
0d7561c6 GC |
3972 | sinfo->active_objs = active_objs; |
3973 | sinfo->num_objs = num_objs; | |
3974 | sinfo->active_slabs = active_slabs; | |
3975 | sinfo->num_slabs = num_slabs; | |
3976 | sinfo->shared_avail = shared_avail; | |
3977 | sinfo->limit = cachep->limit; | |
3978 | sinfo->batchcount = cachep->batchcount; | |
3979 | sinfo->shared = cachep->shared; | |
3980 | sinfo->objects_per_slab = cachep->num; | |
3981 | sinfo->cache_order = cachep->gfporder; | |
3982 | } | |
3983 | ||
3984 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
3985 | { | |
1da177e4 | 3986 | #if STATS |
ce8eb6c4 | 3987 | { /* node stats */ |
1da177e4 LT |
3988 | unsigned long high = cachep->high_mark; |
3989 | unsigned long allocs = cachep->num_allocations; | |
3990 | unsigned long grown = cachep->grown; | |
3991 | unsigned long reaped = cachep->reaped; | |
3992 | unsigned long errors = cachep->errors; | |
3993 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 3994 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 3995 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 3996 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 3997 | |
e92dd4fd JP |
3998 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
3999 | "%4lu %4lu %4lu %4lu %4lu", | |
4000 | allocs, high, grown, | |
4001 | reaped, errors, max_freeable, node_allocs, | |
4002 | node_frees, overflows); | |
1da177e4 LT |
4003 | } |
4004 | /* cpu stats */ | |
4005 | { | |
4006 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4007 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4008 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4009 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4010 | ||
4011 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4012 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4013 | } |
4014 | #endif | |
1da177e4 LT |
4015 | } |
4016 | ||
1da177e4 LT |
4017 | #define MAX_SLABINFO_WRITE 128 |
4018 | /** | |
4019 | * slabinfo_write - Tuning for the slab allocator | |
4020 | * @file: unused | |
4021 | * @buffer: user buffer | |
4022 | * @count: data length | |
4023 | * @ppos: unused | |
4024 | */ | |
b7454ad3 | 4025 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4026 | size_t count, loff_t *ppos) |
1da177e4 | 4027 | { |
b28a02de | 4028 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4029 | int limit, batchcount, shared, res; |
7a7c381d | 4030 | struct kmem_cache *cachep; |
b28a02de | 4031 | |
1da177e4 LT |
4032 | if (count > MAX_SLABINFO_WRITE) |
4033 | return -EINVAL; | |
4034 | if (copy_from_user(&kbuf, buffer, count)) | |
4035 | return -EFAULT; | |
b28a02de | 4036 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4037 | |
4038 | tmp = strchr(kbuf, ' '); | |
4039 | if (!tmp) | |
4040 | return -EINVAL; | |
4041 | *tmp = '\0'; | |
4042 | tmp++; | |
4043 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4044 | return -EINVAL; | |
4045 | ||
4046 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4047 | mutex_lock(&slab_mutex); |
1da177e4 | 4048 | res = -EINVAL; |
18004c5d | 4049 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4050 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4051 | if (limit < 1 || batchcount < 1 || |
4052 | batchcount > limit || shared < 0) { | |
e498be7d | 4053 | res = 0; |
1da177e4 | 4054 | } else { |
e498be7d | 4055 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4056 | batchcount, shared, |
4057 | GFP_KERNEL); | |
1da177e4 LT |
4058 | } |
4059 | break; | |
4060 | } | |
4061 | } | |
18004c5d | 4062 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4063 | if (res >= 0) |
4064 | res = count; | |
4065 | return res; | |
4066 | } | |
871751e2 AV |
4067 | |
4068 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4069 | ||
871751e2 AV |
4070 | static inline int add_caller(unsigned long *n, unsigned long v) |
4071 | { | |
4072 | unsigned long *p; | |
4073 | int l; | |
4074 | if (!v) | |
4075 | return 1; | |
4076 | l = n[1]; | |
4077 | p = n + 2; | |
4078 | while (l) { | |
4079 | int i = l/2; | |
4080 | unsigned long *q = p + 2 * i; | |
4081 | if (*q == v) { | |
4082 | q[1]++; | |
4083 | return 1; | |
4084 | } | |
4085 | if (*q > v) { | |
4086 | l = i; | |
4087 | } else { | |
4088 | p = q + 2; | |
4089 | l -= i + 1; | |
4090 | } | |
4091 | } | |
4092 | if (++n[1] == n[0]) | |
4093 | return 0; | |
4094 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4095 | p[0] = v; | |
4096 | p[1] = 1; | |
4097 | return 1; | |
4098 | } | |
4099 | ||
8456a648 JK |
4100 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4101 | struct page *page) | |
871751e2 AV |
4102 | { |
4103 | void *p; | |
d31676df JK |
4104 | int i, j; |
4105 | unsigned long v; | |
b1cb0982 | 4106 | |
871751e2 AV |
4107 | if (n[0] == n[1]) |
4108 | return; | |
8456a648 | 4109 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4110 | bool active = true; |
4111 | ||
4112 | for (j = page->active; j < c->num; j++) { | |
4113 | if (get_free_obj(page, j) == i) { | |
4114 | active = false; | |
4115 | break; | |
4116 | } | |
4117 | } | |
4118 | ||
4119 | if (!active) | |
871751e2 | 4120 | continue; |
b1cb0982 | 4121 | |
d31676df JK |
4122 | /* |
4123 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4124 | * mapping is established when actual object allocation and | |
4125 | * we could mistakenly access the unmapped object in the cpu | |
4126 | * cache. | |
4127 | */ | |
4128 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4129 | continue; | |
4130 | ||
4131 | if (!add_caller(n, v)) | |
871751e2 AV |
4132 | return; |
4133 | } | |
4134 | } | |
4135 | ||
4136 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4137 | { | |
4138 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4139 | unsigned long offset, size; |
9281acea | 4140 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4141 | |
a5c43dae | 4142 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4143 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4144 | if (modname[0]) |
871751e2 AV |
4145 | seq_printf(m, " [%s]", modname); |
4146 | return; | |
4147 | } | |
4148 | #endif | |
4149 | seq_printf(m, "%p", (void *)address); | |
4150 | } | |
4151 | ||
4152 | static int leaks_show(struct seq_file *m, void *p) | |
4153 | { | |
0672aa7c | 4154 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4155 | struct page *page; |
ce8eb6c4 | 4156 | struct kmem_cache_node *n; |
871751e2 | 4157 | const char *name; |
db845067 | 4158 | unsigned long *x = m->private; |
871751e2 AV |
4159 | int node; |
4160 | int i; | |
4161 | ||
4162 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4163 | return 0; | |
4164 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4165 | return 0; | |
4166 | ||
d31676df JK |
4167 | /* |
4168 | * Set store_user_clean and start to grab stored user information | |
4169 | * for all objects on this cache. If some alloc/free requests comes | |
4170 | * during the processing, information would be wrong so restart | |
4171 | * whole processing. | |
4172 | */ | |
4173 | do { | |
4174 | set_store_user_clean(cachep); | |
4175 | drain_cpu_caches(cachep); | |
4176 | ||
4177 | x[1] = 0; | |
871751e2 | 4178 | |
d31676df | 4179 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4180 | |
d31676df JK |
4181 | check_irq_on(); |
4182 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4183 | |
d31676df JK |
4184 | list_for_each_entry(page, &n->slabs_full, lru) |
4185 | handle_slab(x, cachep, page); | |
4186 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4187 | handle_slab(x, cachep, page); | |
4188 | spin_unlock_irq(&n->list_lock); | |
4189 | } | |
4190 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4191 | |
871751e2 | 4192 | name = cachep->name; |
db845067 | 4193 | if (x[0] == x[1]) { |
871751e2 | 4194 | /* Increase the buffer size */ |
18004c5d | 4195 | mutex_unlock(&slab_mutex); |
db845067 | 4196 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4197 | if (!m->private) { |
4198 | /* Too bad, we are really out */ | |
db845067 | 4199 | m->private = x; |
18004c5d | 4200 | mutex_lock(&slab_mutex); |
871751e2 AV |
4201 | return -ENOMEM; |
4202 | } | |
db845067 CL |
4203 | *(unsigned long *)m->private = x[0] * 2; |
4204 | kfree(x); | |
18004c5d | 4205 | mutex_lock(&slab_mutex); |
871751e2 AV |
4206 | /* Now make sure this entry will be retried */ |
4207 | m->count = m->size; | |
4208 | return 0; | |
4209 | } | |
db845067 CL |
4210 | for (i = 0; i < x[1]; i++) { |
4211 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4212 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4213 | seq_putc(m, '\n'); |
4214 | } | |
d2e7b7d0 | 4215 | |
871751e2 AV |
4216 | return 0; |
4217 | } | |
4218 | ||
a0ec95a8 | 4219 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4220 | .start = slab_start, |
276a2439 WL |
4221 | .next = slab_next, |
4222 | .stop = slab_stop, | |
871751e2 AV |
4223 | .show = leaks_show, |
4224 | }; | |
a0ec95a8 AD |
4225 | |
4226 | static int slabstats_open(struct inode *inode, struct file *file) | |
4227 | { | |
b208ce32 RJ |
4228 | unsigned long *n; |
4229 | ||
4230 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4231 | if (!n) | |
4232 | return -ENOMEM; | |
4233 | ||
4234 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4235 | ||
4236 | return 0; | |
a0ec95a8 AD |
4237 | } |
4238 | ||
4239 | static const struct file_operations proc_slabstats_operations = { | |
4240 | .open = slabstats_open, | |
4241 | .read = seq_read, | |
4242 | .llseek = seq_lseek, | |
4243 | .release = seq_release_private, | |
4244 | }; | |
4245 | #endif | |
4246 | ||
4247 | static int __init slab_proc_init(void) | |
4248 | { | |
4249 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4250 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4251 | #endif |
a0ec95a8 AD |
4252 | return 0; |
4253 | } | |
4254 | module_init(slab_proc_init); | |
1da177e4 LT |
4255 | #endif |
4256 | ||
00e145b6 MS |
4257 | /** |
4258 | * ksize - get the actual amount of memory allocated for a given object | |
4259 | * @objp: Pointer to the object | |
4260 | * | |
4261 | * kmalloc may internally round up allocations and return more memory | |
4262 | * than requested. ksize() can be used to determine the actual amount of | |
4263 | * memory allocated. The caller may use this additional memory, even though | |
4264 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4265 | * The caller must guarantee that objp points to a valid object previously | |
4266 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4267 | * must not be freed during the duration of the call. | |
4268 | */ | |
fd76bab2 | 4269 | size_t ksize(const void *objp) |
1da177e4 | 4270 | { |
ef8b4520 CL |
4271 | BUG_ON(!objp); |
4272 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4273 | return 0; |
1da177e4 | 4274 | |
8c138bc0 | 4275 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4276 | } |
b1aabecd | 4277 | EXPORT_SYMBOL(ksize); |