]>
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 | |
b03a017b | 273 | #define CFLGS_OBJFREELIST_SLAB (0x40000000UL) |
1da177e4 | 274 | #define CFLGS_OFF_SLAB (0x80000000UL) |
b03a017b | 275 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
276 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
277 | ||
278 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
279 | /* |
280 | * Optimization question: fewer reaps means less probability for unnessary | |
281 | * cpucache drain/refill cycles. | |
1da177e4 | 282 | * |
dc6f3f27 | 283 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
284 | * which could lock up otherwise freeable slabs. |
285 | */ | |
5f0985bb JZ |
286 | #define REAPTIMEOUT_AC (2*HZ) |
287 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
288 | |
289 | #if STATS | |
290 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
291 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
292 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
293 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 294 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
295 | #define STATS_SET_HIGH(x) \ |
296 | do { \ | |
297 | if ((x)->num_active > (x)->high_mark) \ | |
298 | (x)->high_mark = (x)->num_active; \ | |
299 | } while (0) | |
1da177e4 LT |
300 | #define STATS_INC_ERR(x) ((x)->errors++) |
301 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 302 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 303 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
304 | #define STATS_SET_FREEABLE(x, i) \ |
305 | do { \ | |
306 | if ((x)->max_freeable < i) \ | |
307 | (x)->max_freeable = i; \ | |
308 | } while (0) | |
1da177e4 LT |
309 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
310 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
311 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
312 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
313 | #else | |
314 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
315 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
316 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
317 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 318 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
319 | #define STATS_SET_HIGH(x) do { } while (0) |
320 | #define STATS_INC_ERR(x) do { } while (0) | |
321 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 322 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 323 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 324 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
325 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
326 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
327 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
328 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
329 | #endif | |
330 | ||
331 | #if DEBUG | |
1da177e4 | 332 | |
a737b3e2 AM |
333 | /* |
334 | * memory layout of objects: | |
1da177e4 | 335 | * 0 : objp |
3dafccf2 | 336 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
337 | * the end of an object is aligned with the end of the real |
338 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 339 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 340 | * redzone word. |
3dafccf2 | 341 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
342 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
343 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 344 | * [BYTES_PER_WORD long] |
1da177e4 | 345 | */ |
343e0d7a | 346 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 347 | { |
3dafccf2 | 348 | return cachep->obj_offset; |
1da177e4 LT |
349 | } |
350 | ||
b46b8f19 | 351 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
352 | { |
353 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
354 | return (unsigned long long*) (objp + obj_offset(cachep) - |
355 | sizeof(unsigned long long)); | |
1da177e4 LT |
356 | } |
357 | ||
b46b8f19 | 358 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
359 | { |
360 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
361 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 362 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 363 | sizeof(unsigned long long) - |
87a927c7 | 364 | REDZONE_ALIGN); |
3b0efdfa | 365 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 366 | sizeof(unsigned long long)); |
1da177e4 LT |
367 | } |
368 | ||
343e0d7a | 369 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
370 | { |
371 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 372 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
373 | } |
374 | ||
375 | #else | |
376 | ||
3dafccf2 | 377 | #define obj_offset(x) 0 |
b46b8f19 DW |
378 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
379 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
380 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
381 | ||
382 | #endif | |
383 | ||
03787301 JK |
384 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
385 | ||
d31676df | 386 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 387 | { |
d31676df JK |
388 | return atomic_read(&cachep->store_user_clean) == 1; |
389 | } | |
03787301 | 390 | |
d31676df JK |
391 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
392 | { | |
393 | atomic_set(&cachep->store_user_clean, 1); | |
394 | } | |
03787301 | 395 | |
d31676df JK |
396 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
397 | { | |
398 | if (is_store_user_clean(cachep)) | |
399 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
400 | } |
401 | ||
402 | #else | |
d31676df | 403 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
404 | |
405 | #endif | |
406 | ||
1da177e4 | 407 | /* |
3df1cccd DR |
408 | * Do not go above this order unless 0 objects fit into the slab or |
409 | * overridden on the command line. | |
1da177e4 | 410 | */ |
543585cc DR |
411 | #define SLAB_MAX_ORDER_HI 1 |
412 | #define SLAB_MAX_ORDER_LO 0 | |
413 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 414 | static bool slab_max_order_set __initdata; |
1da177e4 | 415 | |
6ed5eb22 PE |
416 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
417 | { | |
b49af68f | 418 | struct page *page = virt_to_head_page(obj); |
35026088 | 419 | return page->slab_cache; |
6ed5eb22 PE |
420 | } |
421 | ||
8456a648 | 422 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
423 | unsigned int idx) |
424 | { | |
8456a648 | 425 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
426 | } |
427 | ||
6a2d7a95 | 428 | /* |
3b0efdfa CL |
429 | * We want to avoid an expensive divide : (offset / cache->size) |
430 | * Using the fact that size is a constant for a particular cache, | |
431 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
432 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
433 | */ | |
434 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 435 | const struct page *page, void *obj) |
8fea4e96 | 436 | { |
8456a648 | 437 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 438 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
439 | } |
440 | ||
6fb92430 | 441 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 442 | /* internal cache of cache description objs */ |
9b030cb8 | 443 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
444 | .batchcount = 1, |
445 | .limit = BOOT_CPUCACHE_ENTRIES, | |
446 | .shared = 1, | |
3b0efdfa | 447 | .size = sizeof(struct kmem_cache), |
b28a02de | 448 | .name = "kmem_cache", |
1da177e4 LT |
449 | }; |
450 | ||
edcad250 JK |
451 | #define BAD_ALIEN_MAGIC 0x01020304ul |
452 | ||
1871e52c | 453 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 454 | |
343e0d7a | 455 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 456 | { |
bf0dea23 | 457 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
458 | } |
459 | ||
a737b3e2 AM |
460 | /* |
461 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
462 | */ | |
70f75067 JK |
463 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
464 | unsigned long flags, size_t *left_over) | |
fbaccacf | 465 | { |
70f75067 | 466 | unsigned int num; |
fbaccacf | 467 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 468 | |
fbaccacf SR |
469 | /* |
470 | * The slab management structure can be either off the slab or | |
471 | * on it. For the latter case, the memory allocated for a | |
472 | * slab is used for: | |
473 | * | |
fbaccacf | 474 | * - @buffer_size bytes for each object |
2e6b3602 JK |
475 | * - One freelist_idx_t for each object |
476 | * | |
477 | * We don't need to consider alignment of freelist because | |
478 | * freelist will be at the end of slab page. The objects will be | |
479 | * at the correct alignment. | |
fbaccacf SR |
480 | * |
481 | * If the slab management structure is off the slab, then the | |
482 | * alignment will already be calculated into the size. Because | |
483 | * the slabs are all pages aligned, the objects will be at the | |
484 | * correct alignment when allocated. | |
485 | */ | |
b03a017b | 486 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 487 | num = slab_size / buffer_size; |
2e6b3602 | 488 | *left_over = slab_size % buffer_size; |
fbaccacf | 489 | } else { |
70f75067 | 490 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
491 | *left_over = slab_size % |
492 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 493 | } |
70f75067 JK |
494 | |
495 | return num; | |
1da177e4 LT |
496 | } |
497 | ||
f28510d3 | 498 | #if DEBUG |
d40cee24 | 499 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 500 | |
a737b3e2 AM |
501 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
502 | char *msg) | |
1da177e4 LT |
503 | { |
504 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 505 | function, cachep->name, msg); |
1da177e4 | 506 | dump_stack(); |
373d4d09 | 507 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 508 | } |
f28510d3 | 509 | #endif |
1da177e4 | 510 | |
3395ee05 PM |
511 | /* |
512 | * By default on NUMA we use alien caches to stage the freeing of | |
513 | * objects allocated from other nodes. This causes massive memory | |
514 | * inefficiencies when using fake NUMA setup to split memory into a | |
515 | * large number of small nodes, so it can be disabled on the command | |
516 | * line | |
517 | */ | |
518 | ||
519 | static int use_alien_caches __read_mostly = 1; | |
520 | static int __init noaliencache_setup(char *s) | |
521 | { | |
522 | use_alien_caches = 0; | |
523 | return 1; | |
524 | } | |
525 | __setup("noaliencache", noaliencache_setup); | |
526 | ||
3df1cccd DR |
527 | static int __init slab_max_order_setup(char *str) |
528 | { | |
529 | get_option(&str, &slab_max_order); | |
530 | slab_max_order = slab_max_order < 0 ? 0 : | |
531 | min(slab_max_order, MAX_ORDER - 1); | |
532 | slab_max_order_set = true; | |
533 | ||
534 | return 1; | |
535 | } | |
536 | __setup("slab_max_order=", slab_max_order_setup); | |
537 | ||
8fce4d8e CL |
538 | #ifdef CONFIG_NUMA |
539 | /* | |
540 | * Special reaping functions for NUMA systems called from cache_reap(). | |
541 | * These take care of doing round robin flushing of alien caches (containing | |
542 | * objects freed on different nodes from which they were allocated) and the | |
543 | * flushing of remote pcps by calling drain_node_pages. | |
544 | */ | |
1871e52c | 545 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
546 | |
547 | static void init_reap_node(int cpu) | |
548 | { | |
549 | int node; | |
550 | ||
7d6e6d09 | 551 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 552 | if (node == MAX_NUMNODES) |
442295c9 | 553 | node = first_node(node_online_map); |
8fce4d8e | 554 | |
1871e52c | 555 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
556 | } |
557 | ||
558 | static void next_reap_node(void) | |
559 | { | |
909ea964 | 560 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 561 | |
8fce4d8e CL |
562 | node = next_node(node, node_online_map); |
563 | if (unlikely(node >= MAX_NUMNODES)) | |
564 | node = first_node(node_online_map); | |
909ea964 | 565 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
566 | } |
567 | ||
568 | #else | |
569 | #define init_reap_node(cpu) do { } while (0) | |
570 | #define next_reap_node(void) do { } while (0) | |
571 | #endif | |
572 | ||
1da177e4 LT |
573 | /* |
574 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
575 | * via the workqueue/eventd. | |
576 | * Add the CPU number into the expiration time to minimize the possibility of | |
577 | * the CPUs getting into lockstep and contending for the global cache chain | |
578 | * lock. | |
579 | */ | |
0db0628d | 580 | static void start_cpu_timer(int cpu) |
1da177e4 | 581 | { |
1871e52c | 582 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
583 | |
584 | /* | |
585 | * When this gets called from do_initcalls via cpucache_init(), | |
586 | * init_workqueues() has already run, so keventd will be setup | |
587 | * at that time. | |
588 | */ | |
52bad64d | 589 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 590 | init_reap_node(cpu); |
203b42f7 | 591 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
592 | schedule_delayed_work_on(cpu, reap_work, |
593 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
594 | } |
595 | } | |
596 | ||
1fe00d50 | 597 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 598 | { |
d5cff635 CM |
599 | /* |
600 | * The array_cache structures contain pointers to free object. | |
25985edc | 601 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
602 | * cache the pointers are not cleared and they could be counted as |
603 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
604 | * not scan such objects. | |
605 | */ | |
1fe00d50 JK |
606 | kmemleak_no_scan(ac); |
607 | if (ac) { | |
608 | ac->avail = 0; | |
609 | ac->limit = limit; | |
610 | ac->batchcount = batch; | |
611 | ac->touched = 0; | |
1da177e4 | 612 | } |
1fe00d50 JK |
613 | } |
614 | ||
615 | static struct array_cache *alloc_arraycache(int node, int entries, | |
616 | int batchcount, gfp_t gfp) | |
617 | { | |
5e804789 | 618 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
619 | struct array_cache *ac = NULL; |
620 | ||
621 | ac = kmalloc_node(memsize, gfp, node); | |
622 | init_arraycache(ac, entries, batchcount); | |
623 | return ac; | |
1da177e4 LT |
624 | } |
625 | ||
8456a648 | 626 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 627 | { |
072bb0aa MG |
628 | return PageSlabPfmemalloc(page); |
629 | } | |
630 | ||
631 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
632 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
633 | struct array_cache *ac) | |
634 | { | |
18bf8541 | 635 | struct kmem_cache_node *n = get_node(cachep, numa_mem_id()); |
8456a648 | 636 | struct page *page; |
072bb0aa MG |
637 | unsigned long flags; |
638 | ||
639 | if (!pfmemalloc_active) | |
640 | return; | |
641 | ||
ce8eb6c4 | 642 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
643 | list_for_each_entry(page, &n->slabs_full, lru) |
644 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
645 | goto out; |
646 | ||
8456a648 JK |
647 | list_for_each_entry(page, &n->slabs_partial, lru) |
648 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
649 | goto out; |
650 | ||
8456a648 JK |
651 | list_for_each_entry(page, &n->slabs_free, lru) |
652 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
653 | goto out; |
654 | ||
655 | pfmemalloc_active = false; | |
656 | out: | |
ce8eb6c4 | 657 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
658 | } |
659 | ||
381760ea | 660 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
661 | gfp_t flags, bool force_refill) |
662 | { | |
663 | int i; | |
664 | void *objp = ac->entry[--ac->avail]; | |
665 | ||
666 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
667 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 668 | struct kmem_cache_node *n; |
072bb0aa MG |
669 | |
670 | if (gfp_pfmemalloc_allowed(flags)) { | |
671 | clear_obj_pfmemalloc(&objp); | |
672 | return objp; | |
673 | } | |
674 | ||
675 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 676 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
677 | /* If a !PFMEMALLOC object is found, swap them */ |
678 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
679 | objp = ac->entry[i]; | |
680 | ac->entry[i] = ac->entry[ac->avail]; | |
681 | ac->entry[ac->avail] = objp; | |
682 | return objp; | |
683 | } | |
684 | } | |
685 | ||
686 | /* | |
687 | * If there are empty slabs on the slabs_free list and we are | |
688 | * being forced to refill the cache, mark this one !pfmemalloc. | |
689 | */ | |
18bf8541 | 690 | n = get_node(cachep, numa_mem_id()); |
ce8eb6c4 | 691 | if (!list_empty(&n->slabs_free) && force_refill) { |
8456a648 | 692 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 693 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
694 | clear_obj_pfmemalloc(&objp); |
695 | recheck_pfmemalloc_active(cachep, ac); | |
696 | return objp; | |
697 | } | |
698 | ||
699 | /* No !PFMEMALLOC objects available */ | |
700 | ac->avail++; | |
701 | objp = NULL; | |
702 | } | |
703 | ||
704 | return objp; | |
705 | } | |
706 | ||
381760ea MG |
707 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
708 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
709 | { | |
710 | void *objp; | |
711 | ||
712 | if (unlikely(sk_memalloc_socks())) | |
713 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
714 | else | |
715 | objp = ac->entry[--ac->avail]; | |
716 | ||
717 | return objp; | |
718 | } | |
719 | ||
d3aec344 JK |
720 | static noinline void *__ac_put_obj(struct kmem_cache *cachep, |
721 | struct array_cache *ac, void *objp) | |
072bb0aa MG |
722 | { |
723 | if (unlikely(pfmemalloc_active)) { | |
724 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 725 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
726 | if (PageSlabPfmemalloc(page)) |
727 | set_obj_pfmemalloc(&objp); | |
728 | } | |
729 | ||
381760ea MG |
730 | return objp; |
731 | } | |
732 | ||
733 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
734 | void *objp) | |
735 | { | |
736 | if (unlikely(sk_memalloc_socks())) | |
737 | objp = __ac_put_obj(cachep, ac, objp); | |
738 | ||
072bb0aa MG |
739 | ac->entry[ac->avail++] = objp; |
740 | } | |
741 | ||
3ded175a CL |
742 | /* |
743 | * Transfer objects in one arraycache to another. | |
744 | * Locking must be handled by the caller. | |
745 | * | |
746 | * Return the number of entries transferred. | |
747 | */ | |
748 | static int transfer_objects(struct array_cache *to, | |
749 | struct array_cache *from, unsigned int max) | |
750 | { | |
751 | /* Figure out how many entries to transfer */ | |
732eacc0 | 752 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
753 | |
754 | if (!nr) | |
755 | return 0; | |
756 | ||
757 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
758 | sizeof(void *) *nr); | |
759 | ||
760 | from->avail -= nr; | |
761 | to->avail += nr; | |
3ded175a CL |
762 | return nr; |
763 | } | |
764 | ||
765c4507 CL |
765 | #ifndef CONFIG_NUMA |
766 | ||
767 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 768 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 769 | |
c8522a3a JK |
770 | static inline struct alien_cache **alloc_alien_cache(int node, |
771 | int limit, gfp_t gfp) | |
765c4507 | 772 | { |
edcad250 | 773 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
774 | } |
775 | ||
c8522a3a | 776 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
777 | { |
778 | } | |
779 | ||
780 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
781 | { | |
782 | return 0; | |
783 | } | |
784 | ||
785 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
786 | gfp_t flags) | |
787 | { | |
788 | return NULL; | |
789 | } | |
790 | ||
8b98c169 | 791 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
792 | gfp_t flags, int nodeid) |
793 | { | |
794 | return NULL; | |
795 | } | |
796 | ||
4167e9b2 DR |
797 | static inline gfp_t gfp_exact_node(gfp_t flags) |
798 | { | |
799 | return flags; | |
800 | } | |
801 | ||
765c4507 CL |
802 | #else /* CONFIG_NUMA */ |
803 | ||
8b98c169 | 804 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 805 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 806 | |
c8522a3a JK |
807 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
808 | int batch, gfp_t gfp) | |
809 | { | |
5e804789 | 810 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
811 | struct alien_cache *alc = NULL; |
812 | ||
813 | alc = kmalloc_node(memsize, gfp, node); | |
814 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 815 | spin_lock_init(&alc->lock); |
c8522a3a JK |
816 | return alc; |
817 | } | |
818 | ||
819 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 820 | { |
c8522a3a | 821 | struct alien_cache **alc_ptr; |
5e804789 | 822 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
823 | int i; |
824 | ||
825 | if (limit > 1) | |
826 | limit = 12; | |
c8522a3a JK |
827 | alc_ptr = kzalloc_node(memsize, gfp, node); |
828 | if (!alc_ptr) | |
829 | return NULL; | |
830 | ||
831 | for_each_node(i) { | |
832 | if (i == node || !node_online(i)) | |
833 | continue; | |
834 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
835 | if (!alc_ptr[i]) { | |
836 | for (i--; i >= 0; i--) | |
837 | kfree(alc_ptr[i]); | |
838 | kfree(alc_ptr); | |
839 | return NULL; | |
e498be7d CL |
840 | } |
841 | } | |
c8522a3a | 842 | return alc_ptr; |
e498be7d CL |
843 | } |
844 | ||
c8522a3a | 845 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
846 | { |
847 | int i; | |
848 | ||
c8522a3a | 849 | if (!alc_ptr) |
e498be7d | 850 | return; |
e498be7d | 851 | for_each_node(i) |
c8522a3a JK |
852 | kfree(alc_ptr[i]); |
853 | kfree(alc_ptr); | |
e498be7d CL |
854 | } |
855 | ||
343e0d7a | 856 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
857 | struct array_cache *ac, int node, |
858 | struct list_head *list) | |
e498be7d | 859 | { |
18bf8541 | 860 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
861 | |
862 | if (ac->avail) { | |
ce8eb6c4 | 863 | spin_lock(&n->list_lock); |
e00946fe CL |
864 | /* |
865 | * Stuff objects into the remote nodes shared array first. | |
866 | * That way we could avoid the overhead of putting the objects | |
867 | * into the free lists and getting them back later. | |
868 | */ | |
ce8eb6c4 CL |
869 | if (n->shared) |
870 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 871 | |
833b706c | 872 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 873 | ac->avail = 0; |
ce8eb6c4 | 874 | spin_unlock(&n->list_lock); |
e498be7d CL |
875 | } |
876 | } | |
877 | ||
8fce4d8e CL |
878 | /* |
879 | * Called from cache_reap() to regularly drain alien caches round robin. | |
880 | */ | |
ce8eb6c4 | 881 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 882 | { |
909ea964 | 883 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 884 | |
ce8eb6c4 | 885 | if (n->alien) { |
c8522a3a JK |
886 | struct alien_cache *alc = n->alien[node]; |
887 | struct array_cache *ac; | |
888 | ||
889 | if (alc) { | |
890 | ac = &alc->ac; | |
49dfc304 | 891 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
892 | LIST_HEAD(list); |
893 | ||
894 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 895 | spin_unlock_irq(&alc->lock); |
833b706c | 896 | slabs_destroy(cachep, &list); |
c8522a3a | 897 | } |
8fce4d8e CL |
898 | } |
899 | } | |
900 | } | |
901 | ||
a737b3e2 | 902 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 903 | struct alien_cache **alien) |
e498be7d | 904 | { |
b28a02de | 905 | int i = 0; |
c8522a3a | 906 | struct alien_cache *alc; |
e498be7d CL |
907 | struct array_cache *ac; |
908 | unsigned long flags; | |
909 | ||
910 | for_each_online_node(i) { | |
c8522a3a JK |
911 | alc = alien[i]; |
912 | if (alc) { | |
833b706c JK |
913 | LIST_HEAD(list); |
914 | ||
c8522a3a | 915 | ac = &alc->ac; |
49dfc304 | 916 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 917 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 918 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 919 | slabs_destroy(cachep, &list); |
e498be7d CL |
920 | } |
921 | } | |
922 | } | |
729bd0b7 | 923 | |
25c4f304 JK |
924 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
925 | int node, int page_node) | |
729bd0b7 | 926 | { |
ce8eb6c4 | 927 | struct kmem_cache_node *n; |
c8522a3a JK |
928 | struct alien_cache *alien = NULL; |
929 | struct array_cache *ac; | |
97654dfa | 930 | LIST_HEAD(list); |
1ca4cb24 | 931 | |
18bf8541 | 932 | n = get_node(cachep, node); |
729bd0b7 | 933 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
934 | if (n->alien && n->alien[page_node]) { |
935 | alien = n->alien[page_node]; | |
c8522a3a | 936 | ac = &alien->ac; |
49dfc304 | 937 | spin_lock(&alien->lock); |
c8522a3a | 938 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 939 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 940 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 941 | } |
c8522a3a | 942 | ac_put_obj(cachep, ac, objp); |
49dfc304 | 943 | spin_unlock(&alien->lock); |
833b706c | 944 | slabs_destroy(cachep, &list); |
729bd0b7 | 945 | } else { |
25c4f304 | 946 | n = get_node(cachep, page_node); |
18bf8541 | 947 | spin_lock(&n->list_lock); |
25c4f304 | 948 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 949 | spin_unlock(&n->list_lock); |
97654dfa | 950 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
951 | } |
952 | return 1; | |
953 | } | |
25c4f304 JK |
954 | |
955 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
956 | { | |
957 | int page_node = page_to_nid(virt_to_page(objp)); | |
958 | int node = numa_mem_id(); | |
959 | /* | |
960 | * Make sure we are not freeing a object from another node to the array | |
961 | * cache on this cpu. | |
962 | */ | |
963 | if (likely(node == page_node)) | |
964 | return 0; | |
965 | ||
966 | return __cache_free_alien(cachep, objp, node, page_node); | |
967 | } | |
4167e9b2 DR |
968 | |
969 | /* | |
d0164adc MG |
970 | * Construct gfp mask to allocate from a specific node but do not direct reclaim |
971 | * or warn about failures. kswapd may still wake to reclaim in the background. | |
4167e9b2 DR |
972 | */ |
973 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
974 | { | |
d0164adc | 975 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM; |
4167e9b2 | 976 | } |
e498be7d CL |
977 | #endif |
978 | ||
8f9f8d9e | 979 | /* |
6a67368c | 980 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 981 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 982 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 983 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
984 | * already in use. |
985 | * | |
18004c5d | 986 | * Must hold slab_mutex. |
8f9f8d9e | 987 | */ |
6a67368c | 988 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
989 | { |
990 | struct kmem_cache *cachep; | |
ce8eb6c4 | 991 | struct kmem_cache_node *n; |
5e804789 | 992 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 993 | |
18004c5d | 994 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 995 | /* |
5f0985bb | 996 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
997 | * begin anything. Make sure some other cpu on this |
998 | * node has not already allocated this | |
999 | */ | |
18bf8541 CL |
1000 | n = get_node(cachep, node); |
1001 | if (!n) { | |
ce8eb6c4 CL |
1002 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1003 | if (!n) | |
8f9f8d9e | 1004 | return -ENOMEM; |
ce8eb6c4 | 1005 | kmem_cache_node_init(n); |
5f0985bb JZ |
1006 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1007 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1008 | |
1009 | /* | |
5f0985bb JZ |
1010 | * The kmem_cache_nodes don't come and go as CPUs |
1011 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1012 | * protection here. |
1013 | */ | |
ce8eb6c4 | 1014 | cachep->node[node] = n; |
8f9f8d9e DR |
1015 | } |
1016 | ||
18bf8541 CL |
1017 | spin_lock_irq(&n->list_lock); |
1018 | n->free_limit = | |
8f9f8d9e DR |
1019 | (1 + nr_cpus_node(node)) * |
1020 | cachep->batchcount + cachep->num; | |
18bf8541 | 1021 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
1022 | } |
1023 | return 0; | |
1024 | } | |
1025 | ||
0fa8103b WL |
1026 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1027 | struct kmem_cache_node *n) | |
1028 | { | |
1029 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1030 | } | |
1031 | ||
0db0628d | 1032 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1033 | { |
1034 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1035 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1036 | int node = cpu_to_mem(cpu); |
a70f7302 | 1037 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1038 | |
18004c5d | 1039 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1040 | struct array_cache *nc; |
1041 | struct array_cache *shared; | |
c8522a3a | 1042 | struct alien_cache **alien; |
97654dfa | 1043 | LIST_HEAD(list); |
fbf1e473 | 1044 | |
18bf8541 | 1045 | n = get_node(cachep, node); |
ce8eb6c4 | 1046 | if (!n) |
bf0dea23 | 1047 | continue; |
fbf1e473 | 1048 | |
ce8eb6c4 | 1049 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1050 | |
ce8eb6c4 CL |
1051 | /* Free limit for this kmem_cache_node */ |
1052 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1053 | |
1054 | /* cpu is dead; no one can alloc from it. */ | |
1055 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1056 | if (nc) { | |
97654dfa | 1057 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1058 | nc->avail = 0; |
1059 | } | |
fbf1e473 | 1060 | |
58463c1f | 1061 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1062 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1063 | goto free_slab; |
fbf1e473 AM |
1064 | } |
1065 | ||
ce8eb6c4 | 1066 | shared = n->shared; |
fbf1e473 AM |
1067 | if (shared) { |
1068 | free_block(cachep, shared->entry, | |
97654dfa | 1069 | shared->avail, node, &list); |
ce8eb6c4 | 1070 | n->shared = NULL; |
fbf1e473 AM |
1071 | } |
1072 | ||
ce8eb6c4 CL |
1073 | alien = n->alien; |
1074 | n->alien = NULL; | |
fbf1e473 | 1075 | |
ce8eb6c4 | 1076 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1077 | |
1078 | kfree(shared); | |
1079 | if (alien) { | |
1080 | drain_alien_cache(cachep, alien); | |
1081 | free_alien_cache(alien); | |
1082 | } | |
bf0dea23 JK |
1083 | |
1084 | free_slab: | |
97654dfa | 1085 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1086 | } |
1087 | /* | |
1088 | * In the previous loop, all the objects were freed to | |
1089 | * the respective cache's slabs, now we can go ahead and | |
1090 | * shrink each nodelist to its limit. | |
1091 | */ | |
18004c5d | 1092 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1093 | n = get_node(cachep, node); |
ce8eb6c4 | 1094 | if (!n) |
fbf1e473 | 1095 | continue; |
0fa8103b | 1096 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1097 | } |
1098 | } | |
1099 | ||
0db0628d | 1100 | static int cpuup_prepare(long cpu) |
1da177e4 | 1101 | { |
343e0d7a | 1102 | struct kmem_cache *cachep; |
ce8eb6c4 | 1103 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1104 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1105 | int err; |
1da177e4 | 1106 | |
fbf1e473 AM |
1107 | /* |
1108 | * We need to do this right in the beginning since | |
1109 | * alloc_arraycache's are going to use this list. | |
1110 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1111 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1112 | */ |
6a67368c | 1113 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1114 | if (err < 0) |
1115 | goto bad; | |
fbf1e473 AM |
1116 | |
1117 | /* | |
1118 | * Now we can go ahead with allocating the shared arrays and | |
1119 | * array caches | |
1120 | */ | |
18004c5d | 1121 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 1122 | struct array_cache *shared = NULL; |
c8522a3a | 1123 | struct alien_cache **alien = NULL; |
fbf1e473 | 1124 | |
fbf1e473 AM |
1125 | if (cachep->shared) { |
1126 | shared = alloc_arraycache(node, | |
1127 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1128 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1129 | if (!shared) |
1da177e4 | 1130 | goto bad; |
fbf1e473 AM |
1131 | } |
1132 | if (use_alien_caches) { | |
83b519e8 | 1133 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1134 | if (!alien) { |
1135 | kfree(shared); | |
fbf1e473 | 1136 | goto bad; |
12d00f6a | 1137 | } |
fbf1e473 | 1138 | } |
18bf8541 | 1139 | n = get_node(cachep, node); |
ce8eb6c4 | 1140 | BUG_ON(!n); |
fbf1e473 | 1141 | |
ce8eb6c4 CL |
1142 | spin_lock_irq(&n->list_lock); |
1143 | if (!n->shared) { | |
fbf1e473 AM |
1144 | /* |
1145 | * We are serialised from CPU_DEAD or | |
1146 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1147 | */ | |
ce8eb6c4 | 1148 | n->shared = shared; |
fbf1e473 AM |
1149 | shared = NULL; |
1150 | } | |
4484ebf1 | 1151 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1152 | if (!n->alien) { |
1153 | n->alien = alien; | |
fbf1e473 | 1154 | alien = NULL; |
1da177e4 | 1155 | } |
fbf1e473 | 1156 | #endif |
ce8eb6c4 | 1157 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1158 | kfree(shared); |
1159 | free_alien_cache(alien); | |
1160 | } | |
ce79ddc8 | 1161 | |
fbf1e473 AM |
1162 | return 0; |
1163 | bad: | |
12d00f6a | 1164 | cpuup_canceled(cpu); |
fbf1e473 AM |
1165 | return -ENOMEM; |
1166 | } | |
1167 | ||
0db0628d | 1168 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1169 | unsigned long action, void *hcpu) |
1170 | { | |
1171 | long cpu = (long)hcpu; | |
1172 | int err = 0; | |
1173 | ||
1174 | switch (action) { | |
fbf1e473 AM |
1175 | case CPU_UP_PREPARE: |
1176 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1177 | mutex_lock(&slab_mutex); |
fbf1e473 | 1178 | err = cpuup_prepare(cpu); |
18004c5d | 1179 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1180 | break; |
1181 | case CPU_ONLINE: | |
8bb78442 | 1182 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1183 | start_cpu_timer(cpu); |
1184 | break; | |
1185 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1186 | case CPU_DOWN_PREPARE: |
8bb78442 | 1187 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1188 | /* |
18004c5d | 1189 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1190 | * held so that if cache_reap() is invoked it cannot do |
1191 | * anything expensive but will only modify reap_work | |
1192 | * and reschedule the timer. | |
1193 | */ | |
afe2c511 | 1194 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1195 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1196 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1197 | break; |
1198 | case CPU_DOWN_FAILED: | |
8bb78442 | 1199 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1200 | start_cpu_timer(cpu); |
1201 | break; | |
1da177e4 | 1202 | case CPU_DEAD: |
8bb78442 | 1203 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1204 | /* |
1205 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1206 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1207 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1208 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1209 | * structure is usually allocated from kmem_cache_create() and |
1210 | * gets destroyed at kmem_cache_destroy(). | |
1211 | */ | |
183ff22b | 1212 | /* fall through */ |
8f5be20b | 1213 | #endif |
1da177e4 | 1214 | case CPU_UP_CANCELED: |
8bb78442 | 1215 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1216 | mutex_lock(&slab_mutex); |
fbf1e473 | 1217 | cpuup_canceled(cpu); |
18004c5d | 1218 | mutex_unlock(&slab_mutex); |
1da177e4 | 1219 | break; |
1da177e4 | 1220 | } |
eac40680 | 1221 | return notifier_from_errno(err); |
1da177e4 LT |
1222 | } |
1223 | ||
0db0628d | 1224 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1225 | &cpuup_callback, NULL, 0 |
1226 | }; | |
1da177e4 | 1227 | |
8f9f8d9e DR |
1228 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1229 | /* | |
1230 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1231 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1232 | * removed. | |
1233 | * | |
18004c5d | 1234 | * Must hold slab_mutex. |
8f9f8d9e | 1235 | */ |
6a67368c | 1236 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1237 | { |
1238 | struct kmem_cache *cachep; | |
1239 | int ret = 0; | |
1240 | ||
18004c5d | 1241 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1242 | struct kmem_cache_node *n; |
8f9f8d9e | 1243 | |
18bf8541 | 1244 | n = get_node(cachep, node); |
ce8eb6c4 | 1245 | if (!n) |
8f9f8d9e DR |
1246 | continue; |
1247 | ||
0fa8103b | 1248 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1249 | |
ce8eb6c4 CL |
1250 | if (!list_empty(&n->slabs_full) || |
1251 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1252 | ret = -EBUSY; |
1253 | break; | |
1254 | } | |
1255 | } | |
1256 | return ret; | |
1257 | } | |
1258 | ||
1259 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1260 | unsigned long action, void *arg) | |
1261 | { | |
1262 | struct memory_notify *mnb = arg; | |
1263 | int ret = 0; | |
1264 | int nid; | |
1265 | ||
1266 | nid = mnb->status_change_nid; | |
1267 | if (nid < 0) | |
1268 | goto out; | |
1269 | ||
1270 | switch (action) { | |
1271 | case MEM_GOING_ONLINE: | |
18004c5d | 1272 | mutex_lock(&slab_mutex); |
6a67368c | 1273 | ret = init_cache_node_node(nid); |
18004c5d | 1274 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1275 | break; |
1276 | case MEM_GOING_OFFLINE: | |
18004c5d | 1277 | mutex_lock(&slab_mutex); |
6a67368c | 1278 | ret = drain_cache_node_node(nid); |
18004c5d | 1279 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1280 | break; |
1281 | case MEM_ONLINE: | |
1282 | case MEM_OFFLINE: | |
1283 | case MEM_CANCEL_ONLINE: | |
1284 | case MEM_CANCEL_OFFLINE: | |
1285 | break; | |
1286 | } | |
1287 | out: | |
5fda1bd5 | 1288 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1289 | } |
1290 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1291 | ||
e498be7d | 1292 | /* |
ce8eb6c4 | 1293 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1294 | */ |
6744f087 | 1295 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1296 | int nodeid) |
e498be7d | 1297 | { |
6744f087 | 1298 | struct kmem_cache_node *ptr; |
e498be7d | 1299 | |
6744f087 | 1300 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1301 | BUG_ON(!ptr); |
1302 | ||
6744f087 | 1303 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1304 | /* |
1305 | * Do not assume that spinlocks can be initialized via memcpy: | |
1306 | */ | |
1307 | spin_lock_init(&ptr->list_lock); | |
1308 | ||
e498be7d | 1309 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1310 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1311 | } |
1312 | ||
556a169d | 1313 | /* |
ce8eb6c4 CL |
1314 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1315 | * size of kmem_cache_node. | |
556a169d | 1316 | */ |
ce8eb6c4 | 1317 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1318 | { |
1319 | int node; | |
1320 | ||
1321 | for_each_online_node(node) { | |
ce8eb6c4 | 1322 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1323 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1324 | REAPTIMEOUT_NODE + |
1325 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1326 | } |
1327 | } | |
1328 | ||
a737b3e2 AM |
1329 | /* |
1330 | * Initialisation. Called after the page allocator have been initialised and | |
1331 | * before smp_init(). | |
1da177e4 LT |
1332 | */ |
1333 | void __init kmem_cache_init(void) | |
1334 | { | |
e498be7d CL |
1335 | int i; |
1336 | ||
68126702 JK |
1337 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1338 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1339 | kmem_cache = &kmem_cache_boot; |
1340 | ||
b6e68bc1 | 1341 | if (num_possible_nodes() == 1) |
62918a03 SS |
1342 | use_alien_caches = 0; |
1343 | ||
3c583465 | 1344 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1345 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1346 | |
1da177e4 LT |
1347 | /* |
1348 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1349 | * page orders on machines with more than 32MB of memory if |
1350 | * not overridden on the command line. | |
1da177e4 | 1351 | */ |
3df1cccd | 1352 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1353 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1354 | |
1da177e4 LT |
1355 | /* Bootstrap is tricky, because several objects are allocated |
1356 | * from caches that do not exist yet: | |
9b030cb8 CL |
1357 | * 1) initialize the kmem_cache cache: it contains the struct |
1358 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1359 | * kmem_cache is statically allocated. | |
e498be7d | 1360 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1361 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1362 | * array at the end of the bootstrap. |
1da177e4 | 1363 | * 2) Create the first kmalloc cache. |
343e0d7a | 1364 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1365 | * An __init data area is used for the head array. |
1366 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1367 | * head arrays. | |
9b030cb8 | 1368 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1369 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1370 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1371 | * the other cache's with kmalloc allocated memory. |
1372 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1373 | */ |
1374 | ||
9b030cb8 | 1375 | /* 1) create the kmem_cache */ |
1da177e4 | 1376 | |
8da3430d | 1377 | /* |
b56efcf0 | 1378 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1379 | */ |
2f9baa9f | 1380 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1381 | offsetof(struct kmem_cache, node) + |
6744f087 | 1382 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1383 | SLAB_HWCACHE_ALIGN); |
1384 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1385 | slab_state = PARTIAL; |
1da177e4 | 1386 | |
a737b3e2 | 1387 | /* |
bf0dea23 JK |
1388 | * Initialize the caches that provide memory for the kmem_cache_node |
1389 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1390 | */ |
bf0dea23 | 1391 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1392 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1393 | slab_state = PARTIAL_NODE; |
34cc6990 | 1394 | setup_kmalloc_cache_index_table(); |
e498be7d | 1395 | |
e0a42726 IM |
1396 | slab_early_init = 0; |
1397 | ||
ce8eb6c4 | 1398 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1399 | { |
1ca4cb24 PE |
1400 | int nid; |
1401 | ||
9c09a95c | 1402 | for_each_online_node(nid) { |
ce8eb6c4 | 1403 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1404 | |
bf0dea23 | 1405 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1406 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1407 | } |
1408 | } | |
1da177e4 | 1409 | |
f97d5f63 | 1410 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1411 | } |
1412 | ||
1413 | void __init kmem_cache_init_late(void) | |
1414 | { | |
1415 | struct kmem_cache *cachep; | |
1416 | ||
97d06609 | 1417 | slab_state = UP; |
52cef189 | 1418 | |
8429db5c | 1419 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1420 | mutex_lock(&slab_mutex); |
1421 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1422 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1423 | BUG(); | |
18004c5d | 1424 | mutex_unlock(&slab_mutex); |
056c6241 | 1425 | |
97d06609 CL |
1426 | /* Done! */ |
1427 | slab_state = FULL; | |
1428 | ||
a737b3e2 AM |
1429 | /* |
1430 | * Register a cpu startup notifier callback that initializes | |
1431 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1432 | */ |
1433 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1434 | |
8f9f8d9e DR |
1435 | #ifdef CONFIG_NUMA |
1436 | /* | |
1437 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1438 | * node. |
8f9f8d9e DR |
1439 | */ |
1440 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1441 | #endif | |
1442 | ||
a737b3e2 AM |
1443 | /* |
1444 | * The reap timers are started later, with a module init call: That part | |
1445 | * of the kernel is not yet operational. | |
1da177e4 LT |
1446 | */ |
1447 | } | |
1448 | ||
1449 | static int __init cpucache_init(void) | |
1450 | { | |
1451 | int cpu; | |
1452 | ||
a737b3e2 AM |
1453 | /* |
1454 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1455 | */ |
e498be7d | 1456 | for_each_online_cpu(cpu) |
a737b3e2 | 1457 | start_cpu_timer(cpu); |
a164f896 GC |
1458 | |
1459 | /* Done! */ | |
97d06609 | 1460 | slab_state = FULL; |
1da177e4 LT |
1461 | return 0; |
1462 | } | |
1da177e4 LT |
1463 | __initcall(cpucache_init); |
1464 | ||
8bdec192 RA |
1465 | static noinline void |
1466 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1467 | { | |
9a02d699 | 1468 | #if DEBUG |
ce8eb6c4 | 1469 | struct kmem_cache_node *n; |
8456a648 | 1470 | struct page *page; |
8bdec192 RA |
1471 | unsigned long flags; |
1472 | int node; | |
9a02d699 DR |
1473 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1474 | DEFAULT_RATELIMIT_BURST); | |
1475 | ||
1476 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1477 | return; | |
8bdec192 RA |
1478 | |
1479 | printk(KERN_WARNING | |
1480 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1481 | nodeid, gfpflags); | |
1482 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1483 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1484 | |
18bf8541 | 1485 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1486 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1487 | unsigned long active_slabs = 0, num_slabs = 0; | |
1488 | ||
ce8eb6c4 | 1489 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1490 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1491 | active_objs += cachep->num; |
1492 | active_slabs++; | |
1493 | } | |
8456a648 JK |
1494 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1495 | active_objs += page->active; | |
8bdec192 RA |
1496 | active_slabs++; |
1497 | } | |
8456a648 | 1498 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1499 | num_slabs++; |
1500 | ||
ce8eb6c4 CL |
1501 | free_objects += n->free_objects; |
1502 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1503 | |
1504 | num_slabs += active_slabs; | |
1505 | num_objs = num_slabs * cachep->num; | |
1506 | printk(KERN_WARNING | |
1507 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1508 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1509 | free_objects); | |
1510 | } | |
9a02d699 | 1511 | #endif |
8bdec192 RA |
1512 | } |
1513 | ||
1da177e4 | 1514 | /* |
8a7d9b43 WSH |
1515 | * Interface to system's page allocator. No need to hold the |
1516 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1517 | * |
1518 | * If we requested dmaable memory, we will get it. Even if we | |
1519 | * did not request dmaable memory, we might get it, but that | |
1520 | * would be relatively rare and ignorable. | |
1521 | */ | |
0c3aa83e JK |
1522 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1523 | int nodeid) | |
1da177e4 LT |
1524 | { |
1525 | struct page *page; | |
e1b6aa6f | 1526 | int nr_pages; |
765c4507 | 1527 | |
a618e89f | 1528 | flags |= cachep->allocflags; |
e12ba74d MG |
1529 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1530 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1531 | |
96db800f | 1532 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1533 | if (!page) { |
9a02d699 | 1534 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1535 | return NULL; |
8bdec192 | 1536 | } |
1da177e4 | 1537 | |
f3ccb2c4 VD |
1538 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1539 | __free_pages(page, cachep->gfporder); | |
1540 | return NULL; | |
1541 | } | |
1542 | ||
b37f1dd0 | 1543 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
2f064f34 | 1544 | if (page_is_pfmemalloc(page)) |
072bb0aa MG |
1545 | pfmemalloc_active = true; |
1546 | ||
e1b6aa6f | 1547 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1548 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1549 | add_zone_page_state(page_zone(page), |
1550 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1551 | else | |
1552 | add_zone_page_state(page_zone(page), | |
1553 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 | 1554 | __SetPageSlab(page); |
2f064f34 | 1555 | if (page_is_pfmemalloc(page)) |
a57a4988 | 1556 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1557 | |
b1eeab67 VN |
1558 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1559 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1560 | ||
1561 | if (cachep->ctor) | |
1562 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1563 | else | |
1564 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1565 | } | |
c175eea4 | 1566 | |
0c3aa83e | 1567 | return page; |
1da177e4 LT |
1568 | } |
1569 | ||
1570 | /* | |
1571 | * Interface to system's page release. | |
1572 | */ | |
0c3aa83e | 1573 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1574 | { |
a57a4988 | 1575 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1576 | |
b1eeab67 | 1577 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1578 | |
972d1a7b CL |
1579 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1580 | sub_zone_page_state(page_zone(page), | |
1581 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1582 | else | |
1583 | sub_zone_page_state(page_zone(page), | |
1584 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1585 | |
a57a4988 | 1586 | BUG_ON(!PageSlab(page)); |
73293c2f | 1587 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1588 | __ClearPageSlab(page); |
8456a648 JK |
1589 | page_mapcount_reset(page); |
1590 | page->mapping = NULL; | |
1f458cbf | 1591 | |
1da177e4 LT |
1592 | if (current->reclaim_state) |
1593 | current->reclaim_state->reclaimed_slab += nr_freed; | |
f3ccb2c4 | 1594 | __free_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1595 | } |
1596 | ||
1597 | static void kmem_rcu_free(struct rcu_head *head) | |
1598 | { | |
68126702 JK |
1599 | struct kmem_cache *cachep; |
1600 | struct page *page; | |
1da177e4 | 1601 | |
68126702 JK |
1602 | page = container_of(head, struct page, rcu_head); |
1603 | cachep = page->slab_cache; | |
1604 | ||
1605 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1606 | } |
1607 | ||
1608 | #if DEBUG | |
40b44137 JK |
1609 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1610 | { | |
1611 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1612 | (cachep->size % PAGE_SIZE) == 0) | |
1613 | return true; | |
1614 | ||
1615 | return false; | |
1616 | } | |
1da177e4 LT |
1617 | |
1618 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1619 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1620 | unsigned long caller) |
1da177e4 | 1621 | { |
8c138bc0 | 1622 | int size = cachep->object_size; |
1da177e4 | 1623 | |
3dafccf2 | 1624 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1625 | |
b28a02de | 1626 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1627 | return; |
1628 | ||
b28a02de PE |
1629 | *addr++ = 0x12345678; |
1630 | *addr++ = caller; | |
1631 | *addr++ = smp_processor_id(); | |
1632 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1633 | { |
1634 | unsigned long *sptr = &caller; | |
1635 | unsigned long svalue; | |
1636 | ||
1637 | while (!kstack_end(sptr)) { | |
1638 | svalue = *sptr++; | |
1639 | if (kernel_text_address(svalue)) { | |
b28a02de | 1640 | *addr++ = svalue; |
1da177e4 LT |
1641 | size -= sizeof(unsigned long); |
1642 | if (size <= sizeof(unsigned long)) | |
1643 | break; | |
1644 | } | |
1645 | } | |
1646 | ||
1647 | } | |
b28a02de | 1648 | *addr++ = 0x87654321; |
1da177e4 | 1649 | } |
40b44137 JK |
1650 | |
1651 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1652 | int map, unsigned long caller) | |
1653 | { | |
1654 | if (!is_debug_pagealloc_cache(cachep)) | |
1655 | return; | |
1656 | ||
1657 | if (caller) | |
1658 | store_stackinfo(cachep, objp, caller); | |
1659 | ||
1660 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1661 | } | |
1662 | ||
1663 | #else | |
1664 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1665 | int map, unsigned long caller) {} | |
1666 | ||
1da177e4 LT |
1667 | #endif |
1668 | ||
343e0d7a | 1669 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1670 | { |
8c138bc0 | 1671 | int size = cachep->object_size; |
3dafccf2 | 1672 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1673 | |
1674 | memset(addr, val, size); | |
b28a02de | 1675 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | static void dump_line(char *data, int offset, int limit) | |
1679 | { | |
1680 | int i; | |
aa83aa40 DJ |
1681 | unsigned char error = 0; |
1682 | int bad_count = 0; | |
1683 | ||
fdde6abb | 1684 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1685 | for (i = 0; i < limit; i++) { |
1686 | if (data[offset + i] != POISON_FREE) { | |
1687 | error = data[offset + i]; | |
1688 | bad_count++; | |
1689 | } | |
aa83aa40 | 1690 | } |
fdde6abb SAS |
1691 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1692 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1693 | |
1694 | if (bad_count == 1) { | |
1695 | error ^= POISON_FREE; | |
1696 | if (!(error & (error - 1))) { | |
1697 | printk(KERN_ERR "Single bit error detected. Probably " | |
1698 | "bad RAM.\n"); | |
1699 | #ifdef CONFIG_X86 | |
1700 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1701 | "test tool.\n"); | |
1702 | #else | |
1703 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1704 | #endif | |
1705 | } | |
1706 | } | |
1da177e4 LT |
1707 | } |
1708 | #endif | |
1709 | ||
1710 | #if DEBUG | |
1711 | ||
343e0d7a | 1712 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1713 | { |
1714 | int i, size; | |
1715 | char *realobj; | |
1716 | ||
1717 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1718 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1719 | *dbg_redzone1(cachep, objp), |
1720 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1721 | } |
1722 | ||
1723 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1724 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1725 | *dbg_userword(cachep, objp), | |
1726 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1727 | } |
3dafccf2 | 1728 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1729 | size = cachep->object_size; |
b28a02de | 1730 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1731 | int limit; |
1732 | limit = 16; | |
b28a02de PE |
1733 | if (i + limit > size) |
1734 | limit = size - i; | |
1da177e4 LT |
1735 | dump_line(realobj, i, limit); |
1736 | } | |
1737 | } | |
1738 | ||
343e0d7a | 1739 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1740 | { |
1741 | char *realobj; | |
1742 | int size, i; | |
1743 | int lines = 0; | |
1744 | ||
40b44137 JK |
1745 | if (is_debug_pagealloc_cache(cachep)) |
1746 | return; | |
1747 | ||
3dafccf2 | 1748 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1749 | size = cachep->object_size; |
1da177e4 | 1750 | |
b28a02de | 1751 | for (i = 0; i < size; i++) { |
1da177e4 | 1752 | char exp = POISON_FREE; |
b28a02de | 1753 | if (i == size - 1) |
1da177e4 LT |
1754 | exp = POISON_END; |
1755 | if (realobj[i] != exp) { | |
1756 | int limit; | |
1757 | /* Mismatch ! */ | |
1758 | /* Print header */ | |
1759 | if (lines == 0) { | |
b28a02de | 1760 | printk(KERN_ERR |
face37f5 DJ |
1761 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1762 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1763 | print_objinfo(cachep, objp, 0); |
1764 | } | |
1765 | /* Hexdump the affected line */ | |
b28a02de | 1766 | i = (i / 16) * 16; |
1da177e4 | 1767 | limit = 16; |
b28a02de PE |
1768 | if (i + limit > size) |
1769 | limit = size - i; | |
1da177e4 LT |
1770 | dump_line(realobj, i, limit); |
1771 | i += 16; | |
1772 | lines++; | |
1773 | /* Limit to 5 lines */ | |
1774 | if (lines > 5) | |
1775 | break; | |
1776 | } | |
1777 | } | |
1778 | if (lines != 0) { | |
1779 | /* Print some data about the neighboring objects, if they | |
1780 | * exist: | |
1781 | */ | |
8456a648 | 1782 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1783 | unsigned int objnr; |
1da177e4 | 1784 | |
8456a648 | 1785 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1786 | if (objnr) { |
8456a648 | 1787 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1788 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1789 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1790 | realobj, size); |
1da177e4 LT |
1791 | print_objinfo(cachep, objp, 2); |
1792 | } | |
b28a02de | 1793 | if (objnr + 1 < cachep->num) { |
8456a648 | 1794 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1795 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1796 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1797 | realobj, size); |
1da177e4 LT |
1798 | print_objinfo(cachep, objp, 2); |
1799 | } | |
1800 | } | |
1801 | } | |
1802 | #endif | |
1803 | ||
12dd36fa | 1804 | #if DEBUG |
8456a648 JK |
1805 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1806 | struct page *page) | |
1da177e4 | 1807 | { |
1da177e4 | 1808 | int i; |
b03a017b JK |
1809 | |
1810 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1811 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1812 | POISON_FREE); | |
1813 | } | |
1814 | ||
1da177e4 | 1815 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1816 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1817 | |
1818 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1819 | check_poison_obj(cachep, objp); |
40b44137 | 1820 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1821 | } |
1822 | if (cachep->flags & SLAB_RED_ZONE) { | |
1823 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1824 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1825 | "was overwritten"); |
1da177e4 LT |
1826 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1827 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1828 | "was overwritten"); |
1da177e4 | 1829 | } |
1da177e4 | 1830 | } |
12dd36fa | 1831 | } |
1da177e4 | 1832 | #else |
8456a648 JK |
1833 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1834 | struct page *page) | |
12dd36fa | 1835 | { |
12dd36fa | 1836 | } |
1da177e4 LT |
1837 | #endif |
1838 | ||
911851e6 RD |
1839 | /** |
1840 | * slab_destroy - destroy and release all objects in a slab | |
1841 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1842 | * @page: page pointer being destroyed |
911851e6 | 1843 | * |
8a7d9b43 WSH |
1844 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1845 | * Before calling the slab page must have been unlinked from the cache. The | |
1846 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1847 | */ |
8456a648 | 1848 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1849 | { |
7e007355 | 1850 | void *freelist; |
12dd36fa | 1851 | |
8456a648 JK |
1852 | freelist = page->freelist; |
1853 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1854 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1855 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1856 | else | |
0c3aa83e | 1857 | kmem_freepages(cachep, page); |
68126702 JK |
1858 | |
1859 | /* | |
8456a648 | 1860 | * From now on, we don't use freelist |
68126702 JK |
1861 | * although actual page can be freed in rcu context |
1862 | */ | |
1863 | if (OFF_SLAB(cachep)) | |
8456a648 | 1864 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1865 | } |
1866 | ||
97654dfa JK |
1867 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1868 | { | |
1869 | struct page *page, *n; | |
1870 | ||
1871 | list_for_each_entry_safe(page, n, list, lru) { | |
1872 | list_del(&page->lru); | |
1873 | slab_destroy(cachep, page); | |
1874 | } | |
1875 | } | |
1876 | ||
4d268eba | 1877 | /** |
a70773dd RD |
1878 | * calculate_slab_order - calculate size (page order) of slabs |
1879 | * @cachep: pointer to the cache that is being created | |
1880 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1881 | * @flags: slab allocation flags |
1882 | * | |
1883 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1884 | * |
1885 | * This could be made much more intelligent. For now, try to avoid using | |
1886 | * high order pages for slabs. When the gfp() functions are more friendly | |
1887 | * towards high-order requests, this should be changed. | |
1888 | */ | |
a737b3e2 | 1889 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1890 | size_t size, unsigned long flags) |
4d268eba PE |
1891 | { |
1892 | size_t left_over = 0; | |
9888e6fa | 1893 | int gfporder; |
4d268eba | 1894 | |
0aa817f0 | 1895 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1896 | unsigned int num; |
1897 | size_t remainder; | |
1898 | ||
70f75067 | 1899 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1900 | if (!num) |
1901 | continue; | |
9888e6fa | 1902 | |
f315e3fa JK |
1903 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1904 | if (num > SLAB_OBJ_MAX_NUM) | |
1905 | break; | |
1906 | ||
b1ab41c4 | 1907 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1908 | struct kmem_cache *freelist_cache; |
1909 | size_t freelist_size; | |
1910 | ||
1911 | freelist_size = num * sizeof(freelist_idx_t); | |
1912 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1913 | if (!freelist_cache) | |
1914 | continue; | |
1915 | ||
b1ab41c4 | 1916 | /* |
3217fd9b JK |
1917 | * Needed to avoid possible looping condition |
1918 | * in cache_grow() | |
b1ab41c4 | 1919 | */ |
3217fd9b JK |
1920 | if (OFF_SLAB(freelist_cache)) |
1921 | continue; | |
b1ab41c4 | 1922 | |
3217fd9b JK |
1923 | /* check if off slab has enough benefit */ |
1924 | if (freelist_cache->size > cachep->size / 2) | |
1925 | continue; | |
b1ab41c4 | 1926 | } |
4d268eba | 1927 | |
9888e6fa | 1928 | /* Found something acceptable - save it away */ |
4d268eba | 1929 | cachep->num = num; |
9888e6fa | 1930 | cachep->gfporder = gfporder; |
4d268eba PE |
1931 | left_over = remainder; |
1932 | ||
f78bb8ad LT |
1933 | /* |
1934 | * A VFS-reclaimable slab tends to have most allocations | |
1935 | * as GFP_NOFS and we really don't want to have to be allocating | |
1936 | * higher-order pages when we are unable to shrink dcache. | |
1937 | */ | |
1938 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1939 | break; | |
1940 | ||
4d268eba PE |
1941 | /* |
1942 | * Large number of objects is good, but very large slabs are | |
1943 | * currently bad for the gfp()s. | |
1944 | */ | |
543585cc | 1945 | if (gfporder >= slab_max_order) |
4d268eba PE |
1946 | break; |
1947 | ||
9888e6fa LT |
1948 | /* |
1949 | * Acceptable internal fragmentation? | |
1950 | */ | |
a737b3e2 | 1951 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1952 | break; |
1953 | } | |
1954 | return left_over; | |
1955 | } | |
1956 | ||
bf0dea23 JK |
1957 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1958 | struct kmem_cache *cachep, int entries, int batchcount) | |
1959 | { | |
1960 | int cpu; | |
1961 | size_t size; | |
1962 | struct array_cache __percpu *cpu_cache; | |
1963 | ||
1964 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1965 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1966 | |
1967 | if (!cpu_cache) | |
1968 | return NULL; | |
1969 | ||
1970 | for_each_possible_cpu(cpu) { | |
1971 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1972 | entries, batchcount); | |
1973 | } | |
1974 | ||
1975 | return cpu_cache; | |
1976 | } | |
1977 | ||
83b519e8 | 1978 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1979 | { |
97d06609 | 1980 | if (slab_state >= FULL) |
83b519e8 | 1981 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1982 | |
bf0dea23 JK |
1983 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1984 | if (!cachep->cpu_cache) | |
1985 | return 1; | |
1986 | ||
97d06609 | 1987 | if (slab_state == DOWN) { |
bf0dea23 JK |
1988 | /* Creation of first cache (kmem_cache). */ |
1989 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1990 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1991 | /* For kmem_cache_node */ |
1992 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1993 | } else { |
bf0dea23 | 1994 | int node; |
f30cf7d1 | 1995 | |
bf0dea23 JK |
1996 | for_each_online_node(node) { |
1997 | cachep->node[node] = kmalloc_node( | |
1998 | sizeof(struct kmem_cache_node), gfp, node); | |
1999 | BUG_ON(!cachep->node[node]); | |
2000 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
2001 | } |
2002 | } | |
bf0dea23 | 2003 | |
6a67368c | 2004 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2005 | jiffies + REAPTIMEOUT_NODE + |
2006 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2007 | |
2008 | cpu_cache_get(cachep)->avail = 0; | |
2009 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2010 | cpu_cache_get(cachep)->batchcount = 1; | |
2011 | cpu_cache_get(cachep)->touched = 0; | |
2012 | cachep->batchcount = 1; | |
2013 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2014 | return 0; |
f30cf7d1 PE |
2015 | } |
2016 | ||
12220dea JK |
2017 | unsigned long kmem_cache_flags(unsigned long object_size, |
2018 | unsigned long flags, const char *name, | |
2019 | void (*ctor)(void *)) | |
2020 | { | |
2021 | return flags; | |
2022 | } | |
2023 | ||
2024 | struct kmem_cache * | |
2025 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
2026 | unsigned long flags, void (*ctor)(void *)) | |
2027 | { | |
2028 | struct kmem_cache *cachep; | |
2029 | ||
2030 | cachep = find_mergeable(size, align, flags, name, ctor); | |
2031 | if (cachep) { | |
2032 | cachep->refcount++; | |
2033 | ||
2034 | /* | |
2035 | * Adjust the object sizes so that we clear | |
2036 | * the complete object on kzalloc. | |
2037 | */ | |
2038 | cachep->object_size = max_t(int, cachep->object_size, size); | |
2039 | } | |
2040 | return cachep; | |
2041 | } | |
2042 | ||
b03a017b JK |
2043 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
2044 | size_t size, unsigned long flags) | |
2045 | { | |
2046 | size_t left; | |
2047 | ||
2048 | cachep->num = 0; | |
2049 | ||
2050 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
2051 | return false; | |
2052 | ||
2053 | left = calculate_slab_order(cachep, size, | |
2054 | flags | CFLGS_OBJFREELIST_SLAB); | |
2055 | if (!cachep->num) | |
2056 | return false; | |
2057 | ||
2058 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
2059 | return false; | |
2060 | ||
2061 | cachep->colour = left / cachep->colour_off; | |
2062 | ||
2063 | return true; | |
2064 | } | |
2065 | ||
158e319b JK |
2066 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
2067 | size_t size, unsigned long flags) | |
2068 | { | |
2069 | size_t left; | |
2070 | ||
2071 | cachep->num = 0; | |
2072 | ||
2073 | /* | |
3217fd9b JK |
2074 | * Always use on-slab management when SLAB_NOLEAKTRACE |
2075 | * to avoid recursive calls into kmemleak. | |
158e319b | 2076 | */ |
158e319b JK |
2077 | if (flags & SLAB_NOLEAKTRACE) |
2078 | return false; | |
2079 | ||
2080 | /* | |
2081 | * Size is large, assume best to place the slab management obj | |
2082 | * off-slab (should allow better packing of objs). | |
2083 | */ | |
2084 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
2085 | if (!cachep->num) | |
2086 | return false; | |
2087 | ||
2088 | /* | |
2089 | * If the slab has been placed off-slab, and we have enough space then | |
2090 | * move it on-slab. This is at the expense of any extra colouring. | |
2091 | */ | |
2092 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
2093 | return false; | |
2094 | ||
2095 | cachep->colour = left / cachep->colour_off; | |
2096 | ||
2097 | return true; | |
2098 | } | |
2099 | ||
2100 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
2101 | size_t size, unsigned long flags) | |
2102 | { | |
2103 | size_t left; | |
2104 | ||
2105 | cachep->num = 0; | |
2106 | ||
2107 | left = calculate_slab_order(cachep, size, flags); | |
2108 | if (!cachep->num) | |
2109 | return false; | |
2110 | ||
2111 | cachep->colour = left / cachep->colour_off; | |
2112 | ||
2113 | return true; | |
2114 | } | |
2115 | ||
1da177e4 | 2116 | /** |
039363f3 | 2117 | * __kmem_cache_create - Create a cache. |
a755b76a | 2118 | * @cachep: cache management descriptor |
1da177e4 | 2119 | * @flags: SLAB flags |
1da177e4 LT |
2120 | * |
2121 | * Returns a ptr to the cache on success, NULL on failure. | |
2122 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2123 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2124 | * |
1da177e4 LT |
2125 | * The flags are |
2126 | * | |
2127 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2128 | * to catch references to uninitialised memory. | |
2129 | * | |
2130 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2131 | * for buffer overruns. | |
2132 | * | |
1da177e4 LT |
2133 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2134 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2135 | * as davem. | |
2136 | */ | |
278b1bb1 | 2137 | int |
8a13a4cc | 2138 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2139 | { |
d4a5fca5 | 2140 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2141 | gfp_t gfp; |
278b1bb1 | 2142 | int err; |
8a13a4cc | 2143 | size_t size = cachep->size; |
1da177e4 | 2144 | |
1da177e4 | 2145 | #if DEBUG |
1da177e4 LT |
2146 | #if FORCED_DEBUG |
2147 | /* | |
2148 | * Enable redzoning and last user accounting, except for caches with | |
2149 | * large objects, if the increased size would increase the object size | |
2150 | * above the next power of two: caches with object sizes just above a | |
2151 | * power of two have a significant amount of internal fragmentation. | |
2152 | */ | |
87a927c7 DW |
2153 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2154 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2155 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2156 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2157 | flags |= SLAB_POISON; | |
2158 | #endif | |
1da177e4 | 2159 | #endif |
1da177e4 | 2160 | |
a737b3e2 AM |
2161 | /* |
2162 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2163 | * unaligned accesses for some archs when redzoning is used, and makes |
2164 | * sure any on-slab bufctl's are also correctly aligned. | |
2165 | */ | |
b28a02de PE |
2166 | if (size & (BYTES_PER_WORD - 1)) { |
2167 | size += (BYTES_PER_WORD - 1); | |
2168 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2169 | } |
2170 | ||
87a927c7 DW |
2171 | if (flags & SLAB_RED_ZONE) { |
2172 | ralign = REDZONE_ALIGN; | |
2173 | /* If redzoning, ensure that the second redzone is suitably | |
2174 | * aligned, by adjusting the object size accordingly. */ | |
2175 | size += REDZONE_ALIGN - 1; | |
2176 | size &= ~(REDZONE_ALIGN - 1); | |
2177 | } | |
ca5f9703 | 2178 | |
a44b56d3 | 2179 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2180 | if (ralign < cachep->align) { |
2181 | ralign = cachep->align; | |
1da177e4 | 2182 | } |
3ff84a7f PE |
2183 | /* disable debug if necessary */ |
2184 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2185 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2186 | /* |
ca5f9703 | 2187 | * 4) Store it. |
1da177e4 | 2188 | */ |
8a13a4cc | 2189 | cachep->align = ralign; |
158e319b JK |
2190 | cachep->colour_off = cache_line_size(); |
2191 | /* Offset must be a multiple of the alignment. */ | |
2192 | if (cachep->colour_off < cachep->align) | |
2193 | cachep->colour_off = cachep->align; | |
1da177e4 | 2194 | |
83b519e8 PE |
2195 | if (slab_is_available()) |
2196 | gfp = GFP_KERNEL; | |
2197 | else | |
2198 | gfp = GFP_NOWAIT; | |
2199 | ||
1da177e4 | 2200 | #if DEBUG |
1da177e4 | 2201 | |
ca5f9703 PE |
2202 | /* |
2203 | * Both debugging options require word-alignment which is calculated | |
2204 | * into align above. | |
2205 | */ | |
1da177e4 | 2206 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2207 | /* add space for red zone words */ |
3ff84a7f PE |
2208 | cachep->obj_offset += sizeof(unsigned long long); |
2209 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2210 | } |
2211 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2212 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2213 | * the real object. But if the second red zone needs to be |
2214 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2215 | */ |
87a927c7 DW |
2216 | if (flags & SLAB_RED_ZONE) |
2217 | size += REDZONE_ALIGN; | |
2218 | else | |
2219 | size += BYTES_PER_WORD; | |
1da177e4 | 2220 | } |
832a15d2 JK |
2221 | #endif |
2222 | ||
2223 | size = ALIGN(size, cachep->align); | |
2224 | /* | |
2225 | * We should restrict the number of objects in a slab to implement | |
2226 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2227 | */ | |
2228 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2229 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2230 | ||
2231 | #if DEBUG | |
03a2d2a3 JK |
2232 | /* |
2233 | * To activate debug pagealloc, off-slab management is necessary | |
2234 | * requirement. In early phase of initialization, small sized slab | |
2235 | * doesn't get initialized so it would not be possible. So, we need | |
2236 | * to check size >= 256. It guarantees that all necessary small | |
2237 | * sized slab is initialized in current slab initialization sequence. | |
2238 | */ | |
40323278 | 2239 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2240 | size >= 256 && cachep->object_size > cache_line_size()) { |
2241 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2242 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2243 | ||
2244 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2245 | flags |= CFLGS_OFF_SLAB; | |
2246 | cachep->obj_offset += tmp_size - size; | |
2247 | size = tmp_size; | |
2248 | goto done; | |
2249 | } | |
2250 | } | |
1da177e4 | 2251 | } |
1da177e4 LT |
2252 | #endif |
2253 | ||
b03a017b JK |
2254 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2255 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2256 | goto done; | |
2257 | } | |
2258 | ||
158e319b | 2259 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2260 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2261 | goto done; |
832a15d2 | 2262 | } |
1da177e4 | 2263 | |
158e319b JK |
2264 | if (set_on_slab_cache(cachep, size, flags)) |
2265 | goto done; | |
1da177e4 | 2266 | |
158e319b | 2267 | return -E2BIG; |
1da177e4 | 2268 | |
158e319b JK |
2269 | done: |
2270 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2271 | cachep->flags = flags; |
a57a4988 | 2272 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2273 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2274 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2275 | cachep->size = size; |
6a2d7a95 | 2276 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2277 | |
40b44137 JK |
2278 | #if DEBUG |
2279 | /* | |
2280 | * If we're going to use the generic kernel_map_pages() | |
2281 | * poisoning, then it's going to smash the contents of | |
2282 | * the redzone and userword anyhow, so switch them off. | |
2283 | */ | |
2284 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2285 | (cachep->flags & SLAB_POISON) && | |
2286 | is_debug_pagealloc_cache(cachep)) | |
2287 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2288 | #endif | |
2289 | ||
2290 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2291 | cachep->freelist_cache = |
2292 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2293 | } |
1da177e4 | 2294 | |
278b1bb1 CL |
2295 | err = setup_cpu_cache(cachep, gfp); |
2296 | if (err) { | |
52b4b950 | 2297 | __kmem_cache_release(cachep); |
278b1bb1 | 2298 | return err; |
2ed3a4ef | 2299 | } |
1da177e4 | 2300 | |
278b1bb1 | 2301 | return 0; |
1da177e4 | 2302 | } |
1da177e4 LT |
2303 | |
2304 | #if DEBUG | |
2305 | static void check_irq_off(void) | |
2306 | { | |
2307 | BUG_ON(!irqs_disabled()); | |
2308 | } | |
2309 | ||
2310 | static void check_irq_on(void) | |
2311 | { | |
2312 | BUG_ON(irqs_disabled()); | |
2313 | } | |
2314 | ||
343e0d7a | 2315 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2316 | { |
2317 | #ifdef CONFIG_SMP | |
2318 | check_irq_off(); | |
18bf8541 | 2319 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2320 | #endif |
2321 | } | |
e498be7d | 2322 | |
343e0d7a | 2323 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2324 | { |
2325 | #ifdef CONFIG_SMP | |
2326 | check_irq_off(); | |
18bf8541 | 2327 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2328 | #endif |
2329 | } | |
2330 | ||
1da177e4 LT |
2331 | #else |
2332 | #define check_irq_off() do { } while(0) | |
2333 | #define check_irq_on() do { } while(0) | |
2334 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2335 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2336 | #endif |
2337 | ||
ce8eb6c4 | 2338 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2339 | struct array_cache *ac, |
2340 | int force, int node); | |
2341 | ||
1da177e4 LT |
2342 | static void do_drain(void *arg) |
2343 | { | |
a737b3e2 | 2344 | struct kmem_cache *cachep = arg; |
1da177e4 | 2345 | struct array_cache *ac; |
7d6e6d09 | 2346 | int node = numa_mem_id(); |
18bf8541 | 2347 | struct kmem_cache_node *n; |
97654dfa | 2348 | LIST_HEAD(list); |
1da177e4 LT |
2349 | |
2350 | check_irq_off(); | |
9a2dba4b | 2351 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2352 | n = get_node(cachep, node); |
2353 | spin_lock(&n->list_lock); | |
97654dfa | 2354 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2355 | spin_unlock(&n->list_lock); |
97654dfa | 2356 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2357 | ac->avail = 0; |
2358 | } | |
2359 | ||
343e0d7a | 2360 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2361 | { |
ce8eb6c4 | 2362 | struct kmem_cache_node *n; |
e498be7d CL |
2363 | int node; |
2364 | ||
15c8b6c1 | 2365 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2366 | check_irq_on(); |
18bf8541 CL |
2367 | for_each_kmem_cache_node(cachep, node, n) |
2368 | if (n->alien) | |
ce8eb6c4 | 2369 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2370 | |
18bf8541 CL |
2371 | for_each_kmem_cache_node(cachep, node, n) |
2372 | drain_array(cachep, n, n->shared, 1, node); | |
1da177e4 LT |
2373 | } |
2374 | ||
ed11d9eb CL |
2375 | /* |
2376 | * Remove slabs from the list of free slabs. | |
2377 | * Specify the number of slabs to drain in tofree. | |
2378 | * | |
2379 | * Returns the actual number of slabs released. | |
2380 | */ | |
2381 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2382 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2383 | { |
ed11d9eb CL |
2384 | struct list_head *p; |
2385 | int nr_freed; | |
8456a648 | 2386 | struct page *page; |
1da177e4 | 2387 | |
ed11d9eb | 2388 | nr_freed = 0; |
ce8eb6c4 | 2389 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2390 | |
ce8eb6c4 CL |
2391 | spin_lock_irq(&n->list_lock); |
2392 | p = n->slabs_free.prev; | |
2393 | if (p == &n->slabs_free) { | |
2394 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2395 | goto out; |
2396 | } | |
1da177e4 | 2397 | |
8456a648 | 2398 | page = list_entry(p, struct page, lru); |
8456a648 | 2399 | list_del(&page->lru); |
ed11d9eb CL |
2400 | /* |
2401 | * Safe to drop the lock. The slab is no longer linked | |
2402 | * to the cache. | |
2403 | */ | |
ce8eb6c4 CL |
2404 | n->free_objects -= cache->num; |
2405 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2406 | slab_destroy(cache, page); |
ed11d9eb | 2407 | nr_freed++; |
1da177e4 | 2408 | } |
ed11d9eb CL |
2409 | out: |
2410 | return nr_freed; | |
1da177e4 LT |
2411 | } |
2412 | ||
d6e0b7fa | 2413 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2414 | { |
18bf8541 CL |
2415 | int ret = 0; |
2416 | int node; | |
ce8eb6c4 | 2417 | struct kmem_cache_node *n; |
e498be7d CL |
2418 | |
2419 | drain_cpu_caches(cachep); | |
2420 | ||
2421 | check_irq_on(); | |
18bf8541 | 2422 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2423 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2424 | |
ce8eb6c4 CL |
2425 | ret += !list_empty(&n->slabs_full) || |
2426 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2427 | } |
2428 | return (ret ? 1 : 0); | |
2429 | } | |
2430 | ||
945cf2b6 | 2431 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2432 | { |
2433 | return __kmem_cache_shrink(cachep, false); | |
2434 | } | |
2435 | ||
2436 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2437 | { |
12c3667f | 2438 | int i; |
ce8eb6c4 | 2439 | struct kmem_cache_node *n; |
1da177e4 | 2440 | |
bf0dea23 | 2441 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2442 | |
ce8eb6c4 | 2443 | /* NUMA: free the node structures */ |
18bf8541 CL |
2444 | for_each_kmem_cache_node(cachep, i, n) { |
2445 | kfree(n->shared); | |
2446 | free_alien_cache(n->alien); | |
2447 | kfree(n); | |
2448 | cachep->node[i] = NULL; | |
12c3667f | 2449 | } |
1da177e4 | 2450 | } |
1da177e4 | 2451 | |
e5ac9c5a RT |
2452 | /* |
2453 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2454 | * |
2455 | * For a slab cache when the slab descriptor is off-slab, the | |
2456 | * slab descriptor can't come from the same cache which is being created, | |
2457 | * Because if it is the case, that means we defer the creation of | |
2458 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2459 | * And we eventually call down to __kmem_cache_create(), which | |
2460 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2461 | * This is a "chicken-and-egg" problem. | |
2462 | * | |
2463 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2464 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2465 | */ |
7e007355 | 2466 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2467 | struct page *page, int colour_off, |
2468 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2469 | { |
7e007355 | 2470 | void *freelist; |
0c3aa83e | 2471 | void *addr = page_address(page); |
b28a02de | 2472 | |
2e6b3602 JK |
2473 | page->s_mem = addr + colour_off; |
2474 | page->active = 0; | |
2475 | ||
b03a017b JK |
2476 | if (OBJFREELIST_SLAB(cachep)) |
2477 | freelist = NULL; | |
2478 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2479 | /* Slab management obj is off-slab. */ |
8456a648 | 2480 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2481 | local_flags, nodeid); |
8456a648 | 2482 | if (!freelist) |
1da177e4 LT |
2483 | return NULL; |
2484 | } else { | |
2e6b3602 JK |
2485 | /* We will use last bytes at the slab for freelist */ |
2486 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2487 | cachep->freelist_size; | |
1da177e4 | 2488 | } |
2e6b3602 | 2489 | |
8456a648 | 2490 | return freelist; |
1da177e4 LT |
2491 | } |
2492 | ||
7cc68973 | 2493 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2494 | { |
a41adfaa | 2495 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2496 | } |
2497 | ||
2498 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2499 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2500 | { |
a41adfaa | 2501 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2502 | } |
2503 | ||
10b2e9e8 | 2504 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2505 | { |
10b2e9e8 | 2506 | #if DEBUG |
1da177e4 LT |
2507 | int i; |
2508 | ||
2509 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2510 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2511 | |
1da177e4 LT |
2512 | if (cachep->flags & SLAB_STORE_USER) |
2513 | *dbg_userword(cachep, objp) = NULL; | |
2514 | ||
2515 | if (cachep->flags & SLAB_RED_ZONE) { | |
2516 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2517 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2518 | } | |
2519 | /* | |
a737b3e2 AM |
2520 | * Constructors are not allowed to allocate memory from the same |
2521 | * cache which they are a constructor for. Otherwise, deadlock. | |
2522 | * They must also be threaded. | |
1da177e4 LT |
2523 | */ |
2524 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2525 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2526 | |
2527 | if (cachep->flags & SLAB_RED_ZONE) { | |
2528 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2529 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2530 | " end of an object"); |
1da177e4 LT |
2531 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2532 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2533 | " start of an object"); |
1da177e4 | 2534 | } |
40b44137 JK |
2535 | /* need to poison the objs? */ |
2536 | if (cachep->flags & SLAB_POISON) { | |
2537 | poison_obj(cachep, objp, POISON_FREE); | |
2538 | slab_kernel_map(cachep, objp, 0, 0); | |
2539 | } | |
10b2e9e8 | 2540 | } |
1da177e4 | 2541 | #endif |
10b2e9e8 JK |
2542 | } |
2543 | ||
2544 | static void cache_init_objs(struct kmem_cache *cachep, | |
2545 | struct page *page) | |
2546 | { | |
2547 | int i; | |
2548 | ||
2549 | cache_init_objs_debug(cachep, page); | |
2550 | ||
b03a017b JK |
2551 | if (OBJFREELIST_SLAB(cachep)) { |
2552 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + | |
2553 | obj_offset(cachep); | |
2554 | } | |
2555 | ||
10b2e9e8 JK |
2556 | for (i = 0; i < cachep->num; i++) { |
2557 | /* constructor could break poison info */ | |
2558 | if (DEBUG == 0 && cachep->ctor) | |
2559 | cachep->ctor(index_to_obj(cachep, page, i)); | |
2560 | ||
e5c58dfd | 2561 | set_free_obj(page, i, i); |
1da177e4 | 2562 | } |
1da177e4 LT |
2563 | } |
2564 | ||
343e0d7a | 2565 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2566 | { |
4b51d669 CL |
2567 | if (CONFIG_ZONE_DMA_FLAG) { |
2568 | if (flags & GFP_DMA) | |
a618e89f | 2569 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2570 | else |
a618e89f | 2571 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2572 | } |
1da177e4 LT |
2573 | } |
2574 | ||
260b61dd | 2575 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2576 | { |
b1cb0982 | 2577 | void *objp; |
78d382d7 | 2578 | |
e5c58dfd | 2579 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2580 | page->active++; |
78d382d7 | 2581 | |
d31676df JK |
2582 | #if DEBUG |
2583 | if (cachep->flags & SLAB_STORE_USER) | |
2584 | set_store_user_dirty(cachep); | |
2585 | #endif | |
2586 | ||
78d382d7 MD |
2587 | return objp; |
2588 | } | |
2589 | ||
260b61dd JK |
2590 | static void slab_put_obj(struct kmem_cache *cachep, |
2591 | struct page *page, void *objp) | |
78d382d7 | 2592 | { |
8456a648 | 2593 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2594 | #if DEBUG |
16025177 | 2595 | unsigned int i; |
b1cb0982 | 2596 | |
b1cb0982 | 2597 | /* Verify double free bug */ |
8456a648 | 2598 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2599 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2600 | printk(KERN_ERR "slab: double free detected in cache " |
2601 | "'%s', objp %p\n", cachep->name, objp); | |
2602 | BUG(); | |
2603 | } | |
78d382d7 MD |
2604 | } |
2605 | #endif | |
8456a648 | 2606 | page->active--; |
b03a017b JK |
2607 | if (!page->freelist) |
2608 | page->freelist = objp + obj_offset(cachep); | |
2609 | ||
e5c58dfd | 2610 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2611 | } |
2612 | ||
4776874f PE |
2613 | /* |
2614 | * Map pages beginning at addr to the given cache and slab. This is required | |
2615 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2616 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2617 | */ |
8456a648 | 2618 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2619 | void *freelist) |
1da177e4 | 2620 | { |
a57a4988 | 2621 | page->slab_cache = cache; |
8456a648 | 2622 | page->freelist = freelist; |
1da177e4 LT |
2623 | } |
2624 | ||
2625 | /* | |
2626 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2627 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2628 | */ | |
3c517a61 | 2629 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2630 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2631 | { |
7e007355 | 2632 | void *freelist; |
b28a02de PE |
2633 | size_t offset; |
2634 | gfp_t local_flags; | |
ce8eb6c4 | 2635 | struct kmem_cache_node *n; |
1da177e4 | 2636 | |
a737b3e2 AM |
2637 | /* |
2638 | * Be lazy and only check for valid flags here, keeping it out of the | |
2639 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2640 | */ |
c871ac4e AM |
2641 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2642 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2643 | BUG(); | |
2644 | } | |
6cb06229 | 2645 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2646 | |
ce8eb6c4 | 2647 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2648 | check_irq_off(); |
18bf8541 | 2649 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2650 | spin_lock(&n->list_lock); |
1da177e4 LT |
2651 | |
2652 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2653 | offset = n->colour_next; |
2654 | n->colour_next++; | |
2655 | if (n->colour_next >= cachep->colour) | |
2656 | n->colour_next = 0; | |
2657 | spin_unlock(&n->list_lock); | |
1da177e4 | 2658 | |
2e1217cf | 2659 | offset *= cachep->colour_off; |
1da177e4 | 2660 | |
d0164adc | 2661 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2662 | local_irq_enable(); |
2663 | ||
2664 | /* | |
2665 | * The test for missing atomic flag is performed here, rather than | |
2666 | * the more obvious place, simply to reduce the critical path length | |
2667 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2668 | * will eventually be caught here (where it matters). | |
2669 | */ | |
2670 | kmem_flagcheck(cachep, flags); | |
2671 | ||
a737b3e2 AM |
2672 | /* |
2673 | * Get mem for the objs. Attempt to allocate a physical page from | |
2674 | * 'nodeid'. | |
e498be7d | 2675 | */ |
0c3aa83e JK |
2676 | if (!page) |
2677 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2678 | if (!page) | |
1da177e4 LT |
2679 | goto failed; |
2680 | ||
2681 | /* Get slab management. */ | |
8456a648 | 2682 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2683 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
b03a017b | 2684 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2685 | goto opps1; |
2686 | ||
8456a648 | 2687 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2688 | |
8456a648 | 2689 | cache_init_objs(cachep, page); |
1da177e4 | 2690 | |
d0164adc | 2691 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2692 | local_irq_disable(); |
2693 | check_irq_off(); | |
ce8eb6c4 | 2694 | spin_lock(&n->list_lock); |
1da177e4 LT |
2695 | |
2696 | /* Make slab active. */ | |
8456a648 | 2697 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2698 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2699 | n->free_objects += cachep->num; |
2700 | spin_unlock(&n->list_lock); | |
1da177e4 | 2701 | return 1; |
a737b3e2 | 2702 | opps1: |
0c3aa83e | 2703 | kmem_freepages(cachep, page); |
a737b3e2 | 2704 | failed: |
d0164adc | 2705 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2706 | local_irq_disable(); |
2707 | return 0; | |
2708 | } | |
2709 | ||
2710 | #if DEBUG | |
2711 | ||
2712 | /* | |
2713 | * Perform extra freeing checks: | |
2714 | * - detect bad pointers. | |
2715 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2716 | */ |
2717 | static void kfree_debugcheck(const void *objp) | |
2718 | { | |
1da177e4 LT |
2719 | if (!virt_addr_valid(objp)) { |
2720 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2721 | (unsigned long)objp); |
2722 | BUG(); | |
1da177e4 | 2723 | } |
1da177e4 LT |
2724 | } |
2725 | ||
58ce1fd5 PE |
2726 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2727 | { | |
b46b8f19 | 2728 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2729 | |
2730 | redzone1 = *dbg_redzone1(cache, obj); | |
2731 | redzone2 = *dbg_redzone2(cache, obj); | |
2732 | ||
2733 | /* | |
2734 | * Redzone is ok. | |
2735 | */ | |
2736 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2737 | return; | |
2738 | ||
2739 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2740 | slab_error(cache, "double free detected"); | |
2741 | else | |
2742 | slab_error(cache, "memory outside object was overwritten"); | |
2743 | ||
b46b8f19 | 2744 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2745 | obj, redzone1, redzone2); |
2746 | } | |
2747 | ||
343e0d7a | 2748 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2749 | unsigned long caller) |
1da177e4 | 2750 | { |
1da177e4 | 2751 | unsigned int objnr; |
8456a648 | 2752 | struct page *page; |
1da177e4 | 2753 | |
80cbd911 MW |
2754 | BUG_ON(virt_to_cache(objp) != cachep); |
2755 | ||
3dafccf2 | 2756 | objp -= obj_offset(cachep); |
1da177e4 | 2757 | kfree_debugcheck(objp); |
b49af68f | 2758 | page = virt_to_head_page(objp); |
1da177e4 | 2759 | |
1da177e4 | 2760 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2761 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2762 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2763 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2764 | } | |
d31676df JK |
2765 | if (cachep->flags & SLAB_STORE_USER) { |
2766 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2767 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2768 | } |
1da177e4 | 2769 | |
8456a648 | 2770 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2771 | |
2772 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2773 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2774 | |
1da177e4 | 2775 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2776 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2777 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2778 | } |
2779 | return objp; | |
2780 | } | |
2781 | ||
1da177e4 LT |
2782 | #else |
2783 | #define kfree_debugcheck(x) do { } while(0) | |
2784 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2785 | #endif |
2786 | ||
b03a017b JK |
2787 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2788 | void **list) | |
2789 | { | |
2790 | #if DEBUG | |
2791 | void *next = *list; | |
2792 | void *objp; | |
2793 | ||
2794 | while (next) { | |
2795 | objp = next - obj_offset(cachep); | |
2796 | next = *(void **)next; | |
2797 | poison_obj(cachep, objp, POISON_FREE); | |
2798 | } | |
2799 | #endif | |
2800 | } | |
2801 | ||
d8410234 | 2802 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2803 | struct kmem_cache_node *n, struct page *page, |
2804 | void **list) | |
d8410234 JK |
2805 | { |
2806 | /* move slabp to correct slabp list: */ | |
2807 | list_del(&page->lru); | |
b03a017b | 2808 | if (page->active == cachep->num) { |
d8410234 | 2809 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2810 | if (OBJFREELIST_SLAB(cachep)) { |
2811 | #if DEBUG | |
2812 | /* Poisoning will be done without holding the lock */ | |
2813 | if (cachep->flags & SLAB_POISON) { | |
2814 | void **objp = page->freelist; | |
2815 | ||
2816 | *objp = *list; | |
2817 | *list = objp; | |
2818 | } | |
2819 | #endif | |
2820 | page->freelist = NULL; | |
2821 | } | |
2822 | } else | |
d8410234 JK |
2823 | list_add(&page->lru, &n->slabs_partial); |
2824 | } | |
2825 | ||
7aa0d227 GT |
2826 | static struct page *get_first_slab(struct kmem_cache_node *n) |
2827 | { | |
2828 | struct page *page; | |
2829 | ||
2830 | page = list_first_entry_or_null(&n->slabs_partial, | |
2831 | struct page, lru); | |
2832 | if (!page) { | |
2833 | n->free_touched = 1; | |
2834 | page = list_first_entry_or_null(&n->slabs_free, | |
2835 | struct page, lru); | |
2836 | } | |
2837 | ||
2838 | return page; | |
2839 | } | |
2840 | ||
072bb0aa MG |
2841 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2842 | bool force_refill) | |
1da177e4 LT |
2843 | { |
2844 | int batchcount; | |
ce8eb6c4 | 2845 | struct kmem_cache_node *n; |
1da177e4 | 2846 | struct array_cache *ac; |
1ca4cb24 | 2847 | int node; |
b03a017b | 2848 | void *list = NULL; |
1ca4cb24 | 2849 | |
1da177e4 | 2850 | check_irq_off(); |
7d6e6d09 | 2851 | node = numa_mem_id(); |
072bb0aa MG |
2852 | if (unlikely(force_refill)) |
2853 | goto force_grow; | |
2854 | retry: | |
9a2dba4b | 2855 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2856 | batchcount = ac->batchcount; |
2857 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2858 | /* |
2859 | * If there was little recent activity on this cache, then | |
2860 | * perform only a partial refill. Otherwise we could generate | |
2861 | * refill bouncing. | |
1da177e4 LT |
2862 | */ |
2863 | batchcount = BATCHREFILL_LIMIT; | |
2864 | } | |
18bf8541 | 2865 | n = get_node(cachep, node); |
e498be7d | 2866 | |
ce8eb6c4 CL |
2867 | BUG_ON(ac->avail > 0 || !n); |
2868 | spin_lock(&n->list_lock); | |
1da177e4 | 2869 | |
3ded175a | 2870 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2871 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2872 | n->shared->touched = 1; | |
3ded175a | 2873 | goto alloc_done; |
44b57f1c | 2874 | } |
3ded175a | 2875 | |
1da177e4 | 2876 | while (batchcount > 0) { |
8456a648 | 2877 | struct page *page; |
1da177e4 | 2878 | /* Get slab alloc is to come from. */ |
7aa0d227 GT |
2879 | page = get_first_slab(n); |
2880 | if (!page) | |
2881 | goto must_grow; | |
1da177e4 | 2882 | |
1da177e4 | 2883 | check_spinlock_acquired(cachep); |
714b8171 PE |
2884 | |
2885 | /* | |
2886 | * The slab was either on partial or free list so | |
2887 | * there must be at least one object available for | |
2888 | * allocation. | |
2889 | */ | |
8456a648 | 2890 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2891 | |
8456a648 | 2892 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2893 | STATS_INC_ALLOCED(cachep); |
2894 | STATS_INC_ACTIVE(cachep); | |
2895 | STATS_SET_HIGH(cachep); | |
2896 | ||
260b61dd | 2897 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page)); |
1da177e4 | 2898 | } |
1da177e4 | 2899 | |
b03a017b | 2900 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2901 | } |
2902 | ||
a737b3e2 | 2903 | must_grow: |
ce8eb6c4 | 2904 | n->free_objects -= ac->avail; |
a737b3e2 | 2905 | alloc_done: |
ce8eb6c4 | 2906 | spin_unlock(&n->list_lock); |
b03a017b | 2907 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 LT |
2908 | |
2909 | if (unlikely(!ac->avail)) { | |
2910 | int x; | |
072bb0aa | 2911 | force_grow: |
4167e9b2 | 2912 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2913 | |
a737b3e2 | 2914 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2915 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2916 | node = numa_mem_id(); |
072bb0aa MG |
2917 | |
2918 | /* no objects in sight? abort */ | |
2919 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2920 | return NULL; |
2921 | ||
a737b3e2 | 2922 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2923 | goto retry; |
2924 | } | |
2925 | ac->touched = 1; | |
072bb0aa MG |
2926 | |
2927 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2928 | } |
2929 | ||
a737b3e2 AM |
2930 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2931 | gfp_t flags) | |
1da177e4 | 2932 | { |
d0164adc | 2933 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2934 | #if DEBUG |
2935 | kmem_flagcheck(cachep, flags); | |
2936 | #endif | |
2937 | } | |
2938 | ||
2939 | #if DEBUG | |
a737b3e2 | 2940 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2941 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2942 | { |
b28a02de | 2943 | if (!objp) |
1da177e4 | 2944 | return objp; |
b28a02de | 2945 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2946 | check_poison_obj(cachep, objp); |
40b44137 | 2947 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2948 | poison_obj(cachep, objp, POISON_INUSE); |
2949 | } | |
2950 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2951 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2952 | |
2953 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2954 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2955 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2956 | slab_error(cachep, "double free, or memory outside" | |
2957 | " object was overwritten"); | |
b28a02de | 2958 | printk(KERN_ERR |
b46b8f19 | 2959 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2960 | objp, *dbg_redzone1(cachep, objp), |
2961 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2962 | } |
2963 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2964 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2965 | } | |
03787301 | 2966 | |
3dafccf2 | 2967 | objp += obj_offset(cachep); |
4f104934 | 2968 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2969 | cachep->ctor(objp); |
7ea466f2 TH |
2970 | if (ARCH_SLAB_MINALIGN && |
2971 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 2972 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2973 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2974 | } |
1da177e4 LT |
2975 | return objp; |
2976 | } | |
2977 | #else | |
2978 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2979 | #endif | |
2980 | ||
343e0d7a | 2981 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2982 | { |
b28a02de | 2983 | void *objp; |
1da177e4 | 2984 | struct array_cache *ac; |
072bb0aa | 2985 | bool force_refill = false; |
1da177e4 | 2986 | |
5c382300 | 2987 | check_irq_off(); |
8a8b6502 | 2988 | |
9a2dba4b | 2989 | ac = cpu_cache_get(cachep); |
1da177e4 | 2990 | if (likely(ac->avail)) { |
1da177e4 | 2991 | ac->touched = 1; |
072bb0aa MG |
2992 | objp = ac_get_obj(cachep, ac, flags, false); |
2993 | ||
ddbf2e83 | 2994 | /* |
072bb0aa MG |
2995 | * Allow for the possibility all avail objects are not allowed |
2996 | * by the current flags | |
ddbf2e83 | 2997 | */ |
072bb0aa MG |
2998 | if (objp) { |
2999 | STATS_INC_ALLOCHIT(cachep); | |
3000 | goto out; | |
3001 | } | |
3002 | force_refill = true; | |
1da177e4 | 3003 | } |
072bb0aa MG |
3004 | |
3005 | STATS_INC_ALLOCMISS(cachep); | |
3006 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
3007 | /* | |
3008 | * the 'ac' may be updated by cache_alloc_refill(), | |
3009 | * and kmemleak_erase() requires its correct value. | |
3010 | */ | |
3011 | ac = cpu_cache_get(cachep); | |
3012 | ||
3013 | out: | |
d5cff635 CM |
3014 | /* |
3015 | * To avoid a false negative, if an object that is in one of the | |
3016 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3017 | * treat the array pointers as a reference to the object. | |
3018 | */ | |
f3d8b53a O |
3019 | if (objp) |
3020 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3021 | return objp; |
3022 | } | |
3023 | ||
e498be7d | 3024 | #ifdef CONFIG_NUMA |
c61afb18 | 3025 | /* |
2ad654bc | 3026 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3027 | * |
3028 | * If we are in_interrupt, then process context, including cpusets and | |
3029 | * mempolicy, may not apply and should not be used for allocation policy. | |
3030 | */ | |
3031 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3032 | { | |
3033 | int nid_alloc, nid_here; | |
3034 | ||
765c4507 | 3035 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3036 | return NULL; |
7d6e6d09 | 3037 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3038 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3039 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3040 | else if (current->mempolicy) |
2a389610 | 3041 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3042 | if (nid_alloc != nid_here) |
8b98c169 | 3043 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3044 | return NULL; |
3045 | } | |
3046 | ||
765c4507 CL |
3047 | /* |
3048 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3049 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3050 | * available node for available objects. If that fails then we |
3c517a61 CL |
3051 | * perform an allocation without specifying a node. This allows the page |
3052 | * allocator to do its reclaim / fallback magic. We then insert the | |
3053 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3054 | */ |
8c8cc2c1 | 3055 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3056 | { |
8c8cc2c1 PE |
3057 | struct zonelist *zonelist; |
3058 | gfp_t local_flags; | |
dd1a239f | 3059 | struct zoneref *z; |
54a6eb5c MG |
3060 | struct zone *zone; |
3061 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3062 | void *obj = NULL; |
3c517a61 | 3063 | int nid; |
cc9a6c87 | 3064 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3065 | |
3066 | if (flags & __GFP_THISNODE) | |
3067 | return NULL; | |
3068 | ||
6cb06229 | 3069 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3070 | |
cc9a6c87 | 3071 | retry_cpuset: |
d26914d1 | 3072 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3073 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3074 | |
3c517a61 CL |
3075 | retry: |
3076 | /* | |
3077 | * Look through allowed nodes for objects available | |
3078 | * from existing per node queues. | |
3079 | */ | |
54a6eb5c MG |
3080 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3081 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3082 | |
061d7074 | 3083 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3084 | get_node(cache, nid) && |
3085 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3086 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3087 | gfp_exact_node(flags), nid); |
481c5346 CL |
3088 | if (obj) |
3089 | break; | |
3090 | } | |
3c517a61 CL |
3091 | } |
3092 | ||
cfce6604 | 3093 | if (!obj) { |
3c517a61 CL |
3094 | /* |
3095 | * This allocation will be performed within the constraints | |
3096 | * of the current cpuset / memory policy requirements. | |
3097 | * We may trigger various forms of reclaim on the allowed | |
3098 | * set and go into memory reserves if necessary. | |
3099 | */ | |
0c3aa83e JK |
3100 | struct page *page; |
3101 | ||
d0164adc | 3102 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3103 | local_irq_enable(); |
3104 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3105 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3106 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3107 | local_irq_disable(); |
0c3aa83e | 3108 | if (page) { |
3c517a61 CL |
3109 | /* |
3110 | * Insert into the appropriate per node queues | |
3111 | */ | |
0c3aa83e JK |
3112 | nid = page_to_nid(page); |
3113 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3114 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3115 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3116 | if (!obj) |
3117 | /* | |
3118 | * Another processor may allocate the | |
3119 | * objects in the slab since we are | |
3120 | * not holding any locks. | |
3121 | */ | |
3122 | goto retry; | |
3123 | } else { | |
b6a60451 | 3124 | /* cache_grow already freed obj */ |
3c517a61 CL |
3125 | obj = NULL; |
3126 | } | |
3127 | } | |
aedb0eb1 | 3128 | } |
cc9a6c87 | 3129 | |
d26914d1 | 3130 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3131 | goto retry_cpuset; |
765c4507 CL |
3132 | return obj; |
3133 | } | |
3134 | ||
e498be7d CL |
3135 | /* |
3136 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3137 | */ |
8b98c169 | 3138 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3139 | int nodeid) |
e498be7d | 3140 | { |
8456a648 | 3141 | struct page *page; |
ce8eb6c4 | 3142 | struct kmem_cache_node *n; |
b28a02de | 3143 | void *obj; |
b03a017b | 3144 | void *list = NULL; |
b28a02de PE |
3145 | int x; |
3146 | ||
7c3fbbdd | 3147 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3148 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3149 | BUG_ON(!n); |
b28a02de | 3150 | |
a737b3e2 | 3151 | retry: |
ca3b9b91 | 3152 | check_irq_off(); |
ce8eb6c4 | 3153 | spin_lock(&n->list_lock); |
7aa0d227 GT |
3154 | page = get_first_slab(n); |
3155 | if (!page) | |
3156 | goto must_grow; | |
b28a02de | 3157 | |
b28a02de | 3158 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3159 | |
3160 | STATS_INC_NODEALLOCS(cachep); | |
3161 | STATS_INC_ACTIVE(cachep); | |
3162 | STATS_SET_HIGH(cachep); | |
3163 | ||
8456a648 | 3164 | BUG_ON(page->active == cachep->num); |
b28a02de | 3165 | |
260b61dd | 3166 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3167 | n->free_objects--; |
b28a02de | 3168 | |
b03a017b | 3169 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3170 | |
ce8eb6c4 | 3171 | spin_unlock(&n->list_lock); |
b03a017b | 3172 | fixup_objfreelist_debug(cachep, &list); |
b28a02de | 3173 | goto done; |
e498be7d | 3174 | |
a737b3e2 | 3175 | must_grow: |
ce8eb6c4 | 3176 | spin_unlock(&n->list_lock); |
4167e9b2 | 3177 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3178 | if (x) |
3179 | goto retry; | |
1da177e4 | 3180 | |
8c8cc2c1 | 3181 | return fallback_alloc(cachep, flags); |
e498be7d | 3182 | |
a737b3e2 | 3183 | done: |
b28a02de | 3184 | return obj; |
e498be7d | 3185 | } |
8c8cc2c1 | 3186 | |
8c8cc2c1 | 3187 | static __always_inline void * |
48356303 | 3188 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3189 | unsigned long caller) |
8c8cc2c1 PE |
3190 | { |
3191 | unsigned long save_flags; | |
3192 | void *ptr; | |
7d6e6d09 | 3193 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3194 | |
dcce284a | 3195 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3196 | cachep = slab_pre_alloc_hook(cachep, flags); |
3197 | if (unlikely(!cachep)) | |
824ebef1 AM |
3198 | return NULL; |
3199 | ||
8c8cc2c1 PE |
3200 | cache_alloc_debugcheck_before(cachep, flags); |
3201 | local_irq_save(save_flags); | |
3202 | ||
eacbbae3 | 3203 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3204 | nodeid = slab_node; |
8c8cc2c1 | 3205 | |
18bf8541 | 3206 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3207 | /* Node not bootstrapped yet */ |
3208 | ptr = fallback_alloc(cachep, flags); | |
3209 | goto out; | |
3210 | } | |
3211 | ||
7d6e6d09 | 3212 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3213 | /* |
3214 | * Use the locally cached objects if possible. | |
3215 | * However ____cache_alloc does not allow fallback | |
3216 | * to other nodes. It may fail while we still have | |
3217 | * objects on other nodes available. | |
3218 | */ | |
3219 | ptr = ____cache_alloc(cachep, flags); | |
3220 | if (ptr) | |
3221 | goto out; | |
3222 | } | |
3223 | /* ___cache_alloc_node can fall back to other nodes */ | |
3224 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3225 | out: | |
3226 | local_irq_restore(save_flags); | |
3227 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3228 | ||
d5e3ed66 JDB |
3229 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3230 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3231 | |
d5e3ed66 | 3232 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3233 | return ptr; |
3234 | } | |
3235 | ||
3236 | static __always_inline void * | |
3237 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3238 | { | |
3239 | void *objp; | |
3240 | ||
2ad654bc | 3241 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3242 | objp = alternate_node_alloc(cache, flags); |
3243 | if (objp) | |
3244 | goto out; | |
3245 | } | |
3246 | objp = ____cache_alloc(cache, flags); | |
3247 | ||
3248 | /* | |
3249 | * We may just have run out of memory on the local node. | |
3250 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3251 | */ | |
7d6e6d09 LS |
3252 | if (!objp) |
3253 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3254 | |
3255 | out: | |
3256 | return objp; | |
3257 | } | |
3258 | #else | |
3259 | ||
3260 | static __always_inline void * | |
3261 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3262 | { | |
3263 | return ____cache_alloc(cachep, flags); | |
3264 | } | |
3265 | ||
3266 | #endif /* CONFIG_NUMA */ | |
3267 | ||
3268 | static __always_inline void * | |
48356303 | 3269 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3270 | { |
3271 | unsigned long save_flags; | |
3272 | void *objp; | |
3273 | ||
dcce284a | 3274 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3275 | cachep = slab_pre_alloc_hook(cachep, flags); |
3276 | if (unlikely(!cachep)) | |
824ebef1 AM |
3277 | return NULL; |
3278 | ||
8c8cc2c1 PE |
3279 | cache_alloc_debugcheck_before(cachep, flags); |
3280 | local_irq_save(save_flags); | |
3281 | objp = __do_cache_alloc(cachep, flags); | |
3282 | local_irq_restore(save_flags); | |
3283 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3284 | prefetchw(objp); | |
3285 | ||
d5e3ed66 JDB |
3286 | if (unlikely(flags & __GFP_ZERO) && objp) |
3287 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3288 | |
d5e3ed66 | 3289 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3290 | return objp; |
3291 | } | |
e498be7d CL |
3292 | |
3293 | /* | |
5f0985bb | 3294 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3295 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3296 | */ |
97654dfa JK |
3297 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3298 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3299 | { |
3300 | int i; | |
25c063fb | 3301 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3302 | |
3303 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3304 | void *objp; |
8456a648 | 3305 | struct page *page; |
1da177e4 | 3306 | |
072bb0aa MG |
3307 | clear_obj_pfmemalloc(&objpp[i]); |
3308 | objp = objpp[i]; | |
3309 | ||
8456a648 | 3310 | page = virt_to_head_page(objp); |
8456a648 | 3311 | list_del(&page->lru); |
ff69416e | 3312 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3313 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3314 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3315 | n->free_objects++; |
1da177e4 LT |
3316 | |
3317 | /* fixup slab chains */ | |
8456a648 | 3318 | if (page->active == 0) { |
ce8eb6c4 CL |
3319 | if (n->free_objects > n->free_limit) { |
3320 | n->free_objects -= cachep->num; | |
97654dfa | 3321 | list_add_tail(&page->lru, list); |
1da177e4 | 3322 | } else { |
8456a648 | 3323 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3324 | } |
3325 | } else { | |
3326 | /* Unconditionally move a slab to the end of the | |
3327 | * partial list on free - maximum time for the | |
3328 | * other objects to be freed, too. | |
3329 | */ | |
8456a648 | 3330 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3331 | } |
3332 | } | |
3333 | } | |
3334 | ||
343e0d7a | 3335 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3336 | { |
3337 | int batchcount; | |
ce8eb6c4 | 3338 | struct kmem_cache_node *n; |
7d6e6d09 | 3339 | int node = numa_mem_id(); |
97654dfa | 3340 | LIST_HEAD(list); |
1da177e4 LT |
3341 | |
3342 | batchcount = ac->batchcount; | |
260b61dd | 3343 | |
1da177e4 | 3344 | check_irq_off(); |
18bf8541 | 3345 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3346 | spin_lock(&n->list_lock); |
3347 | if (n->shared) { | |
3348 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3349 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3350 | if (max) { |
3351 | if (batchcount > max) | |
3352 | batchcount = max; | |
e498be7d | 3353 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3354 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3355 | shared_array->avail += batchcount; |
3356 | goto free_done; | |
3357 | } | |
3358 | } | |
3359 | ||
97654dfa | 3360 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3361 | free_done: |
1da177e4 LT |
3362 | #if STATS |
3363 | { | |
3364 | int i = 0; | |
73c0219d | 3365 | struct page *page; |
1da177e4 | 3366 | |
73c0219d | 3367 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3368 | BUG_ON(page->active); |
1da177e4 LT |
3369 | |
3370 | i++; | |
1da177e4 LT |
3371 | } |
3372 | STATS_SET_FREEABLE(cachep, i); | |
3373 | } | |
3374 | #endif | |
ce8eb6c4 | 3375 | spin_unlock(&n->list_lock); |
97654dfa | 3376 | slabs_destroy(cachep, &list); |
1da177e4 | 3377 | ac->avail -= batchcount; |
a737b3e2 | 3378 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3379 | } |
3380 | ||
3381 | /* | |
a737b3e2 AM |
3382 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3383 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3384 | */ |
a947eb95 | 3385 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3386 | unsigned long caller) |
1da177e4 | 3387 | { |
9a2dba4b | 3388 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3389 | |
3390 | check_irq_off(); | |
d5cff635 | 3391 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3392 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3393 | |
8c138bc0 | 3394 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3395 | |
1807a1aa SS |
3396 | /* |
3397 | * Skip calling cache_free_alien() when the platform is not numa. | |
3398 | * This will avoid cache misses that happen while accessing slabp (which | |
3399 | * is per page memory reference) to get nodeid. Instead use a global | |
3400 | * variable to skip the call, which is mostly likely to be present in | |
3401 | * the cache. | |
3402 | */ | |
b6e68bc1 | 3403 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3404 | return; |
3405 | ||
3d880194 | 3406 | if (ac->avail < ac->limit) { |
1da177e4 | 3407 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3408 | } else { |
3409 | STATS_INC_FREEMISS(cachep); | |
3410 | cache_flusharray(cachep, ac); | |
1da177e4 | 3411 | } |
42c8c99c | 3412 | |
072bb0aa | 3413 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3414 | } |
3415 | ||
3416 | /** | |
3417 | * kmem_cache_alloc - Allocate an object | |
3418 | * @cachep: The cache to allocate from. | |
3419 | * @flags: See kmalloc(). | |
3420 | * | |
3421 | * Allocate an object from this cache. The flags are only relevant | |
3422 | * if the cache has no available objects. | |
3423 | */ | |
343e0d7a | 3424 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3425 | { |
48356303 | 3426 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3427 | |
ca2b84cb | 3428 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3429 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3430 | |
3431 | return ret; | |
1da177e4 LT |
3432 | } |
3433 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3434 | ||
7b0501dd JDB |
3435 | static __always_inline void |
3436 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3437 | size_t size, void **p, unsigned long caller) | |
3438 | { | |
3439 | size_t i; | |
3440 | ||
3441 | for (i = 0; i < size; i++) | |
3442 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3443 | } | |
3444 | ||
865762a8 | 3445 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3446 | void **p) |
484748f0 | 3447 | { |
2a777eac JDB |
3448 | size_t i; |
3449 | ||
3450 | s = slab_pre_alloc_hook(s, flags); | |
3451 | if (!s) | |
3452 | return 0; | |
3453 | ||
3454 | cache_alloc_debugcheck_before(s, flags); | |
3455 | ||
3456 | local_irq_disable(); | |
3457 | for (i = 0; i < size; i++) { | |
3458 | void *objp = __do_cache_alloc(s, flags); | |
3459 | ||
2a777eac JDB |
3460 | if (unlikely(!objp)) |
3461 | goto error; | |
3462 | p[i] = objp; | |
3463 | } | |
3464 | local_irq_enable(); | |
3465 | ||
7b0501dd JDB |
3466 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3467 | ||
2a777eac JDB |
3468 | /* Clear memory outside IRQ disabled section */ |
3469 | if (unlikely(flags & __GFP_ZERO)) | |
3470 | for (i = 0; i < size; i++) | |
3471 | memset(p[i], 0, s->object_size); | |
3472 | ||
3473 | slab_post_alloc_hook(s, flags, size, p); | |
3474 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3475 | return size; | |
3476 | error: | |
3477 | local_irq_enable(); | |
7b0501dd | 3478 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3479 | slab_post_alloc_hook(s, flags, i, p); |
3480 | __kmem_cache_free_bulk(s, i, p); | |
3481 | return 0; | |
484748f0 CL |
3482 | } |
3483 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3484 | ||
0f24f128 | 3485 | #ifdef CONFIG_TRACING |
85beb586 | 3486 | void * |
4052147c | 3487 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3488 | { |
85beb586 SR |
3489 | void *ret; |
3490 | ||
48356303 | 3491 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3492 | |
3493 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3494 | size, cachep->size, flags); |
85beb586 | 3495 | return ret; |
36555751 | 3496 | } |
85beb586 | 3497 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3498 | #endif |
3499 | ||
1da177e4 | 3500 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3501 | /** |
3502 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3503 | * @cachep: The cache to allocate from. | |
3504 | * @flags: See kmalloc(). | |
3505 | * @nodeid: node number of the target node. | |
3506 | * | |
3507 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3508 | * node, which can improve the performance for cpu bound structures. | |
3509 | * | |
3510 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3511 | */ | |
8b98c169 CH |
3512 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3513 | { | |
48356303 | 3514 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3515 | |
ca2b84cb | 3516 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3517 | cachep->object_size, cachep->size, |
ca2b84cb | 3518 | flags, nodeid); |
36555751 EGM |
3519 | |
3520 | return ret; | |
8b98c169 | 3521 | } |
1da177e4 LT |
3522 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3523 | ||
0f24f128 | 3524 | #ifdef CONFIG_TRACING |
4052147c | 3525 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3526 | gfp_t flags, |
4052147c EG |
3527 | int nodeid, |
3528 | size_t size) | |
36555751 | 3529 | { |
85beb586 SR |
3530 | void *ret; |
3531 | ||
592f4145 | 3532 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3533 | |
85beb586 | 3534 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3535 | size, cachep->size, |
85beb586 SR |
3536 | flags, nodeid); |
3537 | return ret; | |
36555751 | 3538 | } |
85beb586 | 3539 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3540 | #endif |
3541 | ||
8b98c169 | 3542 | static __always_inline void * |
7c0cb9c6 | 3543 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3544 | { |
343e0d7a | 3545 | struct kmem_cache *cachep; |
97e2bde4 | 3546 | |
2c59dd65 | 3547 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3548 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3549 | return cachep; | |
4052147c | 3550 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3551 | } |
8b98c169 | 3552 | |
8b98c169 CH |
3553 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3554 | { | |
7c0cb9c6 | 3555 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3556 | } |
dbe5e69d | 3557 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3558 | |
3559 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3560 | int node, unsigned long caller) |
8b98c169 | 3561 | { |
7c0cb9c6 | 3562 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3563 | } |
3564 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3565 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3566 | |
3567 | /** | |
800590f5 | 3568 | * __do_kmalloc - allocate memory |
1da177e4 | 3569 | * @size: how many bytes of memory are required. |
800590f5 | 3570 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3571 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3572 | */ |
7fd6b141 | 3573 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3574 | unsigned long caller) |
1da177e4 | 3575 | { |
343e0d7a | 3576 | struct kmem_cache *cachep; |
36555751 | 3577 | void *ret; |
1da177e4 | 3578 | |
2c59dd65 | 3579 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3580 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3581 | return cachep; | |
48356303 | 3582 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3583 | |
7c0cb9c6 | 3584 | trace_kmalloc(caller, ret, |
3b0efdfa | 3585 | size, cachep->size, flags); |
36555751 EGM |
3586 | |
3587 | return ret; | |
7fd6b141 PE |
3588 | } |
3589 | ||
7fd6b141 PE |
3590 | void *__kmalloc(size_t size, gfp_t flags) |
3591 | { | |
7c0cb9c6 | 3592 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3593 | } |
3594 | EXPORT_SYMBOL(__kmalloc); | |
3595 | ||
ce71e27c | 3596 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3597 | { |
7c0cb9c6 | 3598 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3599 | } |
3600 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3601 | |
1da177e4 LT |
3602 | /** |
3603 | * kmem_cache_free - Deallocate an object | |
3604 | * @cachep: The cache the allocation was from. | |
3605 | * @objp: The previously allocated object. | |
3606 | * | |
3607 | * Free an object which was previously allocated from this | |
3608 | * cache. | |
3609 | */ | |
343e0d7a | 3610 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3611 | { |
3612 | unsigned long flags; | |
b9ce5ef4 GC |
3613 | cachep = cache_from_obj(cachep, objp); |
3614 | if (!cachep) | |
3615 | return; | |
1da177e4 LT |
3616 | |
3617 | local_irq_save(flags); | |
d97d476b | 3618 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3619 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3620 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3621 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3622 | local_irq_restore(flags); |
36555751 | 3623 | |
ca2b84cb | 3624 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3625 | } |
3626 | EXPORT_SYMBOL(kmem_cache_free); | |
3627 | ||
e6cdb58d JDB |
3628 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3629 | { | |
3630 | struct kmem_cache *s; | |
3631 | size_t i; | |
3632 | ||
3633 | local_irq_disable(); | |
3634 | for (i = 0; i < size; i++) { | |
3635 | void *objp = p[i]; | |
3636 | ||
ca257195 JDB |
3637 | if (!orig_s) /* called via kfree_bulk */ |
3638 | s = virt_to_cache(objp); | |
3639 | else | |
3640 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3641 | |
3642 | debug_check_no_locks_freed(objp, s->object_size); | |
3643 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3644 | debug_check_no_obj_freed(objp, s->object_size); | |
3645 | ||
3646 | __cache_free(s, objp, _RET_IP_); | |
3647 | } | |
3648 | local_irq_enable(); | |
3649 | ||
3650 | /* FIXME: add tracing */ | |
3651 | } | |
3652 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3653 | ||
1da177e4 LT |
3654 | /** |
3655 | * kfree - free previously allocated memory | |
3656 | * @objp: pointer returned by kmalloc. | |
3657 | * | |
80e93eff PE |
3658 | * If @objp is NULL, no operation is performed. |
3659 | * | |
1da177e4 LT |
3660 | * Don't free memory not originally allocated by kmalloc() |
3661 | * or you will run into trouble. | |
3662 | */ | |
3663 | void kfree(const void *objp) | |
3664 | { | |
343e0d7a | 3665 | struct kmem_cache *c; |
1da177e4 LT |
3666 | unsigned long flags; |
3667 | ||
2121db74 PE |
3668 | trace_kfree(_RET_IP_, objp); |
3669 | ||
6cb8f913 | 3670 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3671 | return; |
3672 | local_irq_save(flags); | |
3673 | kfree_debugcheck(objp); | |
6ed5eb22 | 3674 | c = virt_to_cache(objp); |
8c138bc0 CL |
3675 | debug_check_no_locks_freed(objp, c->object_size); |
3676 | ||
3677 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3678 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3679 | local_irq_restore(flags); |
3680 | } | |
3681 | EXPORT_SYMBOL(kfree); | |
3682 | ||
e498be7d | 3683 | /* |
ce8eb6c4 | 3684 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3685 | */ |
5f0985bb | 3686 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3687 | { |
3688 | int node; | |
ce8eb6c4 | 3689 | struct kmem_cache_node *n; |
cafeb02e | 3690 | struct array_cache *new_shared; |
c8522a3a | 3691 | struct alien_cache **new_alien = NULL; |
e498be7d | 3692 | |
9c09a95c | 3693 | for_each_online_node(node) { |
cafeb02e | 3694 | |
b455def2 L |
3695 | if (use_alien_caches) { |
3696 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3697 | if (!new_alien) | |
3698 | goto fail; | |
3699 | } | |
cafeb02e | 3700 | |
63109846 ED |
3701 | new_shared = NULL; |
3702 | if (cachep->shared) { | |
3703 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3704 | cachep->shared*cachep->batchcount, |
83b519e8 | 3705 | 0xbaadf00d, gfp); |
63109846 ED |
3706 | if (!new_shared) { |
3707 | free_alien_cache(new_alien); | |
3708 | goto fail; | |
3709 | } | |
0718dc2a | 3710 | } |
cafeb02e | 3711 | |
18bf8541 | 3712 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3713 | if (n) { |
3714 | struct array_cache *shared = n->shared; | |
97654dfa | 3715 | LIST_HEAD(list); |
cafeb02e | 3716 | |
ce8eb6c4 | 3717 | spin_lock_irq(&n->list_lock); |
e498be7d | 3718 | |
cafeb02e | 3719 | if (shared) |
0718dc2a | 3720 | free_block(cachep, shared->entry, |
97654dfa | 3721 | shared->avail, node, &list); |
e498be7d | 3722 | |
ce8eb6c4 CL |
3723 | n->shared = new_shared; |
3724 | if (!n->alien) { | |
3725 | n->alien = new_alien; | |
e498be7d CL |
3726 | new_alien = NULL; |
3727 | } | |
ce8eb6c4 | 3728 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3729 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3730 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3731 | slabs_destroy(cachep, &list); |
cafeb02e | 3732 | kfree(shared); |
e498be7d CL |
3733 | free_alien_cache(new_alien); |
3734 | continue; | |
3735 | } | |
ce8eb6c4 CL |
3736 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3737 | if (!n) { | |
0718dc2a CL |
3738 | free_alien_cache(new_alien); |
3739 | kfree(new_shared); | |
e498be7d | 3740 | goto fail; |
0718dc2a | 3741 | } |
e498be7d | 3742 | |
ce8eb6c4 | 3743 | kmem_cache_node_init(n); |
5f0985bb JZ |
3744 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3745 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3746 | n->shared = new_shared; |
3747 | n->alien = new_alien; | |
3748 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3749 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3750 | cachep->node[node] = n; |
e498be7d | 3751 | } |
cafeb02e | 3752 | return 0; |
0718dc2a | 3753 | |
a737b3e2 | 3754 | fail: |
3b0efdfa | 3755 | if (!cachep->list.next) { |
0718dc2a CL |
3756 | /* Cache is not active yet. Roll back what we did */ |
3757 | node--; | |
3758 | while (node >= 0) { | |
18bf8541 CL |
3759 | n = get_node(cachep, node); |
3760 | if (n) { | |
ce8eb6c4 CL |
3761 | kfree(n->shared); |
3762 | free_alien_cache(n->alien); | |
3763 | kfree(n); | |
6a67368c | 3764 | cachep->node[node] = NULL; |
0718dc2a CL |
3765 | } |
3766 | node--; | |
3767 | } | |
3768 | } | |
cafeb02e | 3769 | return -ENOMEM; |
e498be7d CL |
3770 | } |
3771 | ||
18004c5d | 3772 | /* Always called with the slab_mutex held */ |
943a451a | 3773 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3774 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3775 | { |
bf0dea23 JK |
3776 | struct array_cache __percpu *cpu_cache, *prev; |
3777 | int cpu; | |
1da177e4 | 3778 | |
bf0dea23 JK |
3779 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3780 | if (!cpu_cache) | |
d2e7b7d0 SS |
3781 | return -ENOMEM; |
3782 | ||
bf0dea23 JK |
3783 | prev = cachep->cpu_cache; |
3784 | cachep->cpu_cache = cpu_cache; | |
3785 | kick_all_cpus_sync(); | |
e498be7d | 3786 | |
1da177e4 | 3787 | check_irq_on(); |
1da177e4 LT |
3788 | cachep->batchcount = batchcount; |
3789 | cachep->limit = limit; | |
e498be7d | 3790 | cachep->shared = shared; |
1da177e4 | 3791 | |
bf0dea23 JK |
3792 | if (!prev) |
3793 | goto alloc_node; | |
3794 | ||
3795 | for_each_online_cpu(cpu) { | |
97654dfa | 3796 | LIST_HEAD(list); |
18bf8541 CL |
3797 | int node; |
3798 | struct kmem_cache_node *n; | |
bf0dea23 | 3799 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3800 | |
bf0dea23 | 3801 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3802 | n = get_node(cachep, node); |
3803 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3804 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3805 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3806 | slabs_destroy(cachep, &list); |
1da177e4 | 3807 | } |
bf0dea23 JK |
3808 | free_percpu(prev); |
3809 | ||
3810 | alloc_node: | |
5f0985bb | 3811 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3812 | } |
3813 | ||
943a451a GC |
3814 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3815 | int batchcount, int shared, gfp_t gfp) | |
3816 | { | |
3817 | int ret; | |
426589f5 | 3818 | struct kmem_cache *c; |
943a451a GC |
3819 | |
3820 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3821 | ||
3822 | if (slab_state < FULL) | |
3823 | return ret; | |
3824 | ||
3825 | if ((ret < 0) || !is_root_cache(cachep)) | |
3826 | return ret; | |
3827 | ||
426589f5 VD |
3828 | lockdep_assert_held(&slab_mutex); |
3829 | for_each_memcg_cache(c, cachep) { | |
3830 | /* return value determined by the root cache only */ | |
3831 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3832 | } |
3833 | ||
3834 | return ret; | |
3835 | } | |
3836 | ||
18004c5d | 3837 | /* Called with slab_mutex held always */ |
83b519e8 | 3838 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3839 | { |
3840 | int err; | |
943a451a GC |
3841 | int limit = 0; |
3842 | int shared = 0; | |
3843 | int batchcount = 0; | |
3844 | ||
3845 | if (!is_root_cache(cachep)) { | |
3846 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3847 | limit = root->limit; | |
3848 | shared = root->shared; | |
3849 | batchcount = root->batchcount; | |
3850 | } | |
1da177e4 | 3851 | |
943a451a GC |
3852 | if (limit && shared && batchcount) |
3853 | goto skip_setup; | |
a737b3e2 AM |
3854 | /* |
3855 | * The head array serves three purposes: | |
1da177e4 LT |
3856 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3857 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3858 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3859 | * bufctl chains: array operations are cheaper. |
3860 | * The numbers are guessed, we should auto-tune as described by | |
3861 | * Bonwick. | |
3862 | */ | |
3b0efdfa | 3863 | if (cachep->size > 131072) |
1da177e4 | 3864 | limit = 1; |
3b0efdfa | 3865 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3866 | limit = 8; |
3b0efdfa | 3867 | else if (cachep->size > 1024) |
1da177e4 | 3868 | limit = 24; |
3b0efdfa | 3869 | else if (cachep->size > 256) |
1da177e4 LT |
3870 | limit = 54; |
3871 | else | |
3872 | limit = 120; | |
3873 | ||
a737b3e2 AM |
3874 | /* |
3875 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3876 | * allocation behaviour: Most allocs on one cpu, most free operations |
3877 | * on another cpu. For these cases, an efficient object passing between | |
3878 | * cpus is necessary. This is provided by a shared array. The array | |
3879 | * replaces Bonwick's magazine layer. | |
3880 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3881 | * to a larger limit. Thus disabled by default. | |
3882 | */ | |
3883 | shared = 0; | |
3b0efdfa | 3884 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3885 | shared = 8; |
1da177e4 LT |
3886 | |
3887 | #if DEBUG | |
a737b3e2 AM |
3888 | /* |
3889 | * With debugging enabled, large batchcount lead to excessively long | |
3890 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3891 | */ |
3892 | if (limit > 32) | |
3893 | limit = 32; | |
3894 | #endif | |
943a451a GC |
3895 | batchcount = (limit + 1) / 2; |
3896 | skip_setup: | |
3897 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3898 | if (err) |
3899 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3900 | cachep->name, -err); |
2ed3a4ef | 3901 | return err; |
1da177e4 LT |
3902 | } |
3903 | ||
1b55253a | 3904 | /* |
ce8eb6c4 CL |
3905 | * Drain an array if it contains any elements taking the node lock only if |
3906 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3907 | * if drain_array() is used on the shared array. |
1b55253a | 3908 | */ |
ce8eb6c4 | 3909 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3910 | struct array_cache *ac, int force, int node) |
1da177e4 | 3911 | { |
97654dfa | 3912 | LIST_HEAD(list); |
1da177e4 LT |
3913 | int tofree; |
3914 | ||
1b55253a CL |
3915 | if (!ac || !ac->avail) |
3916 | return; | |
1da177e4 LT |
3917 | if (ac->touched && !force) { |
3918 | ac->touched = 0; | |
b18e7e65 | 3919 | } else { |
ce8eb6c4 | 3920 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3921 | if (ac->avail) { |
3922 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3923 | if (tofree > ac->avail) | |
3924 | tofree = (ac->avail + 1) / 2; | |
97654dfa | 3925 | free_block(cachep, ac->entry, tofree, node, &list); |
b18e7e65 CL |
3926 | ac->avail -= tofree; |
3927 | memmove(ac->entry, &(ac->entry[tofree]), | |
3928 | sizeof(void *) * ac->avail); | |
3929 | } | |
ce8eb6c4 | 3930 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3931 | slabs_destroy(cachep, &list); |
1da177e4 LT |
3932 | } |
3933 | } | |
3934 | ||
3935 | /** | |
3936 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3937 | * @w: work descriptor |
1da177e4 LT |
3938 | * |
3939 | * Called from workqueue/eventd every few seconds. | |
3940 | * Purpose: | |
3941 | * - clear the per-cpu caches for this CPU. | |
3942 | * - return freeable pages to the main free memory pool. | |
3943 | * | |
a737b3e2 AM |
3944 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3945 | * again on the next iteration. | |
1da177e4 | 3946 | */ |
7c5cae36 | 3947 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3948 | { |
7a7c381d | 3949 | struct kmem_cache *searchp; |
ce8eb6c4 | 3950 | struct kmem_cache_node *n; |
7d6e6d09 | 3951 | int node = numa_mem_id(); |
bf6aede7 | 3952 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3953 | |
18004c5d | 3954 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3955 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3956 | goto out; |
1da177e4 | 3957 | |
18004c5d | 3958 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3959 | check_irq_on(); |
3960 | ||
35386e3b | 3961 | /* |
ce8eb6c4 | 3962 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3963 | * have established with reasonable certainty that |
3964 | * we can do some work if the lock was obtained. | |
3965 | */ | |
18bf8541 | 3966 | n = get_node(searchp, node); |
35386e3b | 3967 | |
ce8eb6c4 | 3968 | reap_alien(searchp, n); |
1da177e4 | 3969 | |
ce8eb6c4 | 3970 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3971 | |
35386e3b CL |
3972 | /* |
3973 | * These are racy checks but it does not matter | |
3974 | * if we skip one check or scan twice. | |
3975 | */ | |
ce8eb6c4 | 3976 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3977 | goto next; |
1da177e4 | 3978 | |
5f0985bb | 3979 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3980 | |
ce8eb6c4 | 3981 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 3982 | |
ce8eb6c4 CL |
3983 | if (n->free_touched) |
3984 | n->free_touched = 0; | |
ed11d9eb CL |
3985 | else { |
3986 | int freed; | |
1da177e4 | 3987 | |
ce8eb6c4 | 3988 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3989 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3990 | STATS_ADD_REAPED(searchp, freed); | |
3991 | } | |
35386e3b | 3992 | next: |
1da177e4 LT |
3993 | cond_resched(); |
3994 | } | |
3995 | check_irq_on(); | |
18004c5d | 3996 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3997 | next_reap_node(); |
7c5cae36 | 3998 | out: |
a737b3e2 | 3999 | /* Set up the next iteration */ |
5f0985bb | 4000 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
4001 | } |
4002 | ||
158a9624 | 4003 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4004 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4005 | { |
8456a648 | 4006 | struct page *page; |
b28a02de PE |
4007 | unsigned long active_objs; |
4008 | unsigned long num_objs; | |
4009 | unsigned long active_slabs = 0; | |
4010 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4011 | const char *name; |
1da177e4 | 4012 | char *error = NULL; |
e498be7d | 4013 | int node; |
ce8eb6c4 | 4014 | struct kmem_cache_node *n; |
1da177e4 | 4015 | |
1da177e4 LT |
4016 | active_objs = 0; |
4017 | num_slabs = 0; | |
18bf8541 | 4018 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 4019 | |
ca3b9b91 | 4020 | check_irq_on(); |
ce8eb6c4 | 4021 | spin_lock_irq(&n->list_lock); |
e498be7d | 4022 | |
8456a648 JK |
4023 | list_for_each_entry(page, &n->slabs_full, lru) { |
4024 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4025 | error = "slabs_full accounting error"; |
4026 | active_objs += cachep->num; | |
4027 | active_slabs++; | |
4028 | } | |
8456a648 JK |
4029 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4030 | if (page->active == cachep->num && !error) | |
106a74e1 | 4031 | error = "slabs_partial accounting error"; |
8456a648 | 4032 | if (!page->active && !error) |
106a74e1 | 4033 | error = "slabs_partial accounting error"; |
8456a648 | 4034 | active_objs += page->active; |
e498be7d CL |
4035 | active_slabs++; |
4036 | } | |
8456a648 JK |
4037 | list_for_each_entry(page, &n->slabs_free, lru) { |
4038 | if (page->active && !error) | |
106a74e1 | 4039 | error = "slabs_free accounting error"; |
e498be7d CL |
4040 | num_slabs++; |
4041 | } | |
ce8eb6c4 CL |
4042 | free_objects += n->free_objects; |
4043 | if (n->shared) | |
4044 | shared_avail += n->shared->avail; | |
e498be7d | 4045 | |
ce8eb6c4 | 4046 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4047 | } |
b28a02de PE |
4048 | num_slabs += active_slabs; |
4049 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4050 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4051 | error = "free_objects accounting error"; |
4052 | ||
b28a02de | 4053 | name = cachep->name; |
1da177e4 LT |
4054 | if (error) |
4055 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4056 | ||
0d7561c6 GC |
4057 | sinfo->active_objs = active_objs; |
4058 | sinfo->num_objs = num_objs; | |
4059 | sinfo->active_slabs = active_slabs; | |
4060 | sinfo->num_slabs = num_slabs; | |
4061 | sinfo->shared_avail = shared_avail; | |
4062 | sinfo->limit = cachep->limit; | |
4063 | sinfo->batchcount = cachep->batchcount; | |
4064 | sinfo->shared = cachep->shared; | |
4065 | sinfo->objects_per_slab = cachep->num; | |
4066 | sinfo->cache_order = cachep->gfporder; | |
4067 | } | |
4068 | ||
4069 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4070 | { | |
1da177e4 | 4071 | #if STATS |
ce8eb6c4 | 4072 | { /* node stats */ |
1da177e4 LT |
4073 | unsigned long high = cachep->high_mark; |
4074 | unsigned long allocs = cachep->num_allocations; | |
4075 | unsigned long grown = cachep->grown; | |
4076 | unsigned long reaped = cachep->reaped; | |
4077 | unsigned long errors = cachep->errors; | |
4078 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4079 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4080 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4081 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4082 | |
e92dd4fd JP |
4083 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4084 | "%4lu %4lu %4lu %4lu %4lu", | |
4085 | allocs, high, grown, | |
4086 | reaped, errors, max_freeable, node_allocs, | |
4087 | node_frees, overflows); | |
1da177e4 LT |
4088 | } |
4089 | /* cpu stats */ | |
4090 | { | |
4091 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4092 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4093 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4094 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4095 | ||
4096 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4097 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4098 | } |
4099 | #endif | |
1da177e4 LT |
4100 | } |
4101 | ||
1da177e4 LT |
4102 | #define MAX_SLABINFO_WRITE 128 |
4103 | /** | |
4104 | * slabinfo_write - Tuning for the slab allocator | |
4105 | * @file: unused | |
4106 | * @buffer: user buffer | |
4107 | * @count: data length | |
4108 | * @ppos: unused | |
4109 | */ | |
b7454ad3 | 4110 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4111 | size_t count, loff_t *ppos) |
1da177e4 | 4112 | { |
b28a02de | 4113 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4114 | int limit, batchcount, shared, res; |
7a7c381d | 4115 | struct kmem_cache *cachep; |
b28a02de | 4116 | |
1da177e4 LT |
4117 | if (count > MAX_SLABINFO_WRITE) |
4118 | return -EINVAL; | |
4119 | if (copy_from_user(&kbuf, buffer, count)) | |
4120 | return -EFAULT; | |
b28a02de | 4121 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4122 | |
4123 | tmp = strchr(kbuf, ' '); | |
4124 | if (!tmp) | |
4125 | return -EINVAL; | |
4126 | *tmp = '\0'; | |
4127 | tmp++; | |
4128 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4129 | return -EINVAL; | |
4130 | ||
4131 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4132 | mutex_lock(&slab_mutex); |
1da177e4 | 4133 | res = -EINVAL; |
18004c5d | 4134 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4135 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4136 | if (limit < 1 || batchcount < 1 || |
4137 | batchcount > limit || shared < 0) { | |
e498be7d | 4138 | res = 0; |
1da177e4 | 4139 | } else { |
e498be7d | 4140 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4141 | batchcount, shared, |
4142 | GFP_KERNEL); | |
1da177e4 LT |
4143 | } |
4144 | break; | |
4145 | } | |
4146 | } | |
18004c5d | 4147 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4148 | if (res >= 0) |
4149 | res = count; | |
4150 | return res; | |
4151 | } | |
871751e2 AV |
4152 | |
4153 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4154 | ||
871751e2 AV |
4155 | static inline int add_caller(unsigned long *n, unsigned long v) |
4156 | { | |
4157 | unsigned long *p; | |
4158 | int l; | |
4159 | if (!v) | |
4160 | return 1; | |
4161 | l = n[1]; | |
4162 | p = n + 2; | |
4163 | while (l) { | |
4164 | int i = l/2; | |
4165 | unsigned long *q = p + 2 * i; | |
4166 | if (*q == v) { | |
4167 | q[1]++; | |
4168 | return 1; | |
4169 | } | |
4170 | if (*q > v) { | |
4171 | l = i; | |
4172 | } else { | |
4173 | p = q + 2; | |
4174 | l -= i + 1; | |
4175 | } | |
4176 | } | |
4177 | if (++n[1] == n[0]) | |
4178 | return 0; | |
4179 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4180 | p[0] = v; | |
4181 | p[1] = 1; | |
4182 | return 1; | |
4183 | } | |
4184 | ||
8456a648 JK |
4185 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4186 | struct page *page) | |
871751e2 AV |
4187 | { |
4188 | void *p; | |
d31676df JK |
4189 | int i, j; |
4190 | unsigned long v; | |
b1cb0982 | 4191 | |
871751e2 AV |
4192 | if (n[0] == n[1]) |
4193 | return; | |
8456a648 | 4194 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4195 | bool active = true; |
4196 | ||
4197 | for (j = page->active; j < c->num; j++) { | |
4198 | if (get_free_obj(page, j) == i) { | |
4199 | active = false; | |
4200 | break; | |
4201 | } | |
4202 | } | |
4203 | ||
4204 | if (!active) | |
871751e2 | 4205 | continue; |
b1cb0982 | 4206 | |
d31676df JK |
4207 | /* |
4208 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4209 | * mapping is established when actual object allocation and | |
4210 | * we could mistakenly access the unmapped object in the cpu | |
4211 | * cache. | |
4212 | */ | |
4213 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4214 | continue; | |
4215 | ||
4216 | if (!add_caller(n, v)) | |
871751e2 AV |
4217 | return; |
4218 | } | |
4219 | } | |
4220 | ||
4221 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4222 | { | |
4223 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4224 | unsigned long offset, size; |
9281acea | 4225 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4226 | |
a5c43dae | 4227 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4228 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4229 | if (modname[0]) |
871751e2 AV |
4230 | seq_printf(m, " [%s]", modname); |
4231 | return; | |
4232 | } | |
4233 | #endif | |
4234 | seq_printf(m, "%p", (void *)address); | |
4235 | } | |
4236 | ||
4237 | static int leaks_show(struct seq_file *m, void *p) | |
4238 | { | |
0672aa7c | 4239 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4240 | struct page *page; |
ce8eb6c4 | 4241 | struct kmem_cache_node *n; |
871751e2 | 4242 | const char *name; |
db845067 | 4243 | unsigned long *x = m->private; |
871751e2 AV |
4244 | int node; |
4245 | int i; | |
4246 | ||
4247 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4248 | return 0; | |
4249 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4250 | return 0; | |
4251 | ||
d31676df JK |
4252 | /* |
4253 | * Set store_user_clean and start to grab stored user information | |
4254 | * for all objects on this cache. If some alloc/free requests comes | |
4255 | * during the processing, information would be wrong so restart | |
4256 | * whole processing. | |
4257 | */ | |
4258 | do { | |
4259 | set_store_user_clean(cachep); | |
4260 | drain_cpu_caches(cachep); | |
4261 | ||
4262 | x[1] = 0; | |
871751e2 | 4263 | |
d31676df | 4264 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4265 | |
d31676df JK |
4266 | check_irq_on(); |
4267 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4268 | |
d31676df JK |
4269 | list_for_each_entry(page, &n->slabs_full, lru) |
4270 | handle_slab(x, cachep, page); | |
4271 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4272 | handle_slab(x, cachep, page); | |
4273 | spin_unlock_irq(&n->list_lock); | |
4274 | } | |
4275 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4276 | |
871751e2 | 4277 | name = cachep->name; |
db845067 | 4278 | if (x[0] == x[1]) { |
871751e2 | 4279 | /* Increase the buffer size */ |
18004c5d | 4280 | mutex_unlock(&slab_mutex); |
db845067 | 4281 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4282 | if (!m->private) { |
4283 | /* Too bad, we are really out */ | |
db845067 | 4284 | m->private = x; |
18004c5d | 4285 | mutex_lock(&slab_mutex); |
871751e2 AV |
4286 | return -ENOMEM; |
4287 | } | |
db845067 CL |
4288 | *(unsigned long *)m->private = x[0] * 2; |
4289 | kfree(x); | |
18004c5d | 4290 | mutex_lock(&slab_mutex); |
871751e2 AV |
4291 | /* Now make sure this entry will be retried */ |
4292 | m->count = m->size; | |
4293 | return 0; | |
4294 | } | |
db845067 CL |
4295 | for (i = 0; i < x[1]; i++) { |
4296 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4297 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4298 | seq_putc(m, '\n'); |
4299 | } | |
d2e7b7d0 | 4300 | |
871751e2 AV |
4301 | return 0; |
4302 | } | |
4303 | ||
a0ec95a8 | 4304 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4305 | .start = slab_start, |
276a2439 WL |
4306 | .next = slab_next, |
4307 | .stop = slab_stop, | |
871751e2 AV |
4308 | .show = leaks_show, |
4309 | }; | |
a0ec95a8 AD |
4310 | |
4311 | static int slabstats_open(struct inode *inode, struct file *file) | |
4312 | { | |
b208ce32 RJ |
4313 | unsigned long *n; |
4314 | ||
4315 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4316 | if (!n) | |
4317 | return -ENOMEM; | |
4318 | ||
4319 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4320 | ||
4321 | return 0; | |
a0ec95a8 AD |
4322 | } |
4323 | ||
4324 | static const struct file_operations proc_slabstats_operations = { | |
4325 | .open = slabstats_open, | |
4326 | .read = seq_read, | |
4327 | .llseek = seq_lseek, | |
4328 | .release = seq_release_private, | |
4329 | }; | |
4330 | #endif | |
4331 | ||
4332 | static int __init slab_proc_init(void) | |
4333 | { | |
4334 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4335 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4336 | #endif |
a0ec95a8 AD |
4337 | return 0; |
4338 | } | |
4339 | module_init(slab_proc_init); | |
1da177e4 LT |
4340 | #endif |
4341 | ||
00e145b6 MS |
4342 | /** |
4343 | * ksize - get the actual amount of memory allocated for a given object | |
4344 | * @objp: Pointer to the object | |
4345 | * | |
4346 | * kmalloc may internally round up allocations and return more memory | |
4347 | * than requested. ksize() can be used to determine the actual amount of | |
4348 | * memory allocated. The caller may use this additional memory, even though | |
4349 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4350 | * The caller must guarantee that objp points to a valid object previously | |
4351 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4352 | * must not be freed during the duration of the call. | |
4353 | */ | |
fd76bab2 | 4354 | size_t ksize(const void *objp) |
1da177e4 | 4355 | { |
ef8b4520 CL |
4356 | BUG_ON(!objp); |
4357 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4358 | return 0; |
1da177e4 | 4359 | |
8c138bc0 | 4360 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4361 | } |
b1aabecd | 4362 | EXPORT_SYMBOL(ksize); |