]>
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 | |
1da177e4 LT |
171 | /* |
172 | * struct array_cache | |
173 | * | |
1da177e4 LT |
174 | * Purpose: |
175 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
176 | * - reduce the number of linked list operations | |
177 | * - reduce spinlock operations | |
178 | * | |
179 | * The limit is stored in the per-cpu structure to reduce the data cache | |
180 | * footprint. | |
181 | * | |
182 | */ | |
183 | struct array_cache { | |
184 | unsigned int avail; | |
185 | unsigned int limit; | |
186 | unsigned int batchcount; | |
187 | unsigned int touched; | |
bda5b655 | 188 | void *entry[]; /* |
a737b3e2 AM |
189 | * Must have this definition in here for the proper |
190 | * alignment of array_cache. Also simplifies accessing | |
191 | * the entries. | |
a737b3e2 | 192 | */ |
1da177e4 LT |
193 | }; |
194 | ||
c8522a3a JK |
195 | struct alien_cache { |
196 | spinlock_t lock; | |
197 | struct array_cache ac; | |
198 | }; | |
199 | ||
e498be7d CL |
200 | /* |
201 | * Need this for bootstrapping a per node allocator. | |
202 | */ | |
bf0dea23 | 203 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 204 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 205 | #define CACHE_CACHE 0 |
bf0dea23 | 206 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 207 | |
ed11d9eb | 208 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 209 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 210 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
211 | int node, struct list_head *list); |
212 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 213 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 214 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 215 | |
76b342bd JK |
216 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
217 | void **list); | |
218 | static inline void fixup_slab_list(struct kmem_cache *cachep, | |
219 | struct kmem_cache_node *n, struct page *page, | |
220 | void **list); | |
e0a42726 IM |
221 | static int slab_early_init = 1; |
222 | ||
ce8eb6c4 | 223 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 224 | |
ce8eb6c4 | 225 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
226 | { |
227 | INIT_LIST_HEAD(&parent->slabs_full); | |
228 | INIT_LIST_HEAD(&parent->slabs_partial); | |
229 | INIT_LIST_HEAD(&parent->slabs_free); | |
230 | parent->shared = NULL; | |
231 | parent->alien = NULL; | |
2e1217cf | 232 | parent->colour_next = 0; |
e498be7d CL |
233 | spin_lock_init(&parent->list_lock); |
234 | parent->free_objects = 0; | |
235 | parent->free_touched = 0; | |
236 | } | |
237 | ||
a737b3e2 AM |
238 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
239 | do { \ | |
240 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 241 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
242 | } while (0) |
243 | ||
a737b3e2 AM |
244 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
245 | do { \ | |
e498be7d CL |
246 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
247 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
248 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
249 | } while (0) | |
1da177e4 | 250 | |
b03a017b | 251 | #define CFLGS_OBJFREELIST_SLAB (0x40000000UL) |
1da177e4 | 252 | #define CFLGS_OFF_SLAB (0x80000000UL) |
b03a017b | 253 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
254 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
255 | ||
256 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
257 | /* |
258 | * Optimization question: fewer reaps means less probability for unnessary | |
259 | * cpucache drain/refill cycles. | |
1da177e4 | 260 | * |
dc6f3f27 | 261 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
262 | * which could lock up otherwise freeable slabs. |
263 | */ | |
5f0985bb JZ |
264 | #define REAPTIMEOUT_AC (2*HZ) |
265 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
266 | |
267 | #if STATS | |
268 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
269 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
270 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
271 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 272 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
273 | #define STATS_SET_HIGH(x) \ |
274 | do { \ | |
275 | if ((x)->num_active > (x)->high_mark) \ | |
276 | (x)->high_mark = (x)->num_active; \ | |
277 | } while (0) | |
1da177e4 LT |
278 | #define STATS_INC_ERR(x) ((x)->errors++) |
279 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 280 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 281 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
282 | #define STATS_SET_FREEABLE(x, i) \ |
283 | do { \ | |
284 | if ((x)->max_freeable < i) \ | |
285 | (x)->max_freeable = i; \ | |
286 | } while (0) | |
1da177e4 LT |
287 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
288 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
289 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
290 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
291 | #else | |
292 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
293 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
294 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
295 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 296 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
297 | #define STATS_SET_HIGH(x) do { } while (0) |
298 | #define STATS_INC_ERR(x) do { } while (0) | |
299 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 300 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 301 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 302 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
303 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
304 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
305 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
306 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
307 | #endif | |
308 | ||
309 | #if DEBUG | |
1da177e4 | 310 | |
a737b3e2 AM |
311 | /* |
312 | * memory layout of objects: | |
1da177e4 | 313 | * 0 : objp |
3dafccf2 | 314 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
315 | * the end of an object is aligned with the end of the real |
316 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 317 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 318 | * redzone word. |
3dafccf2 | 319 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
320 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
321 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 322 | * [BYTES_PER_WORD long] |
1da177e4 | 323 | */ |
343e0d7a | 324 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 325 | { |
3dafccf2 | 326 | return cachep->obj_offset; |
1da177e4 LT |
327 | } |
328 | ||
b46b8f19 | 329 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
330 | { |
331 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
332 | return (unsigned long long*) (objp + obj_offset(cachep) - |
333 | sizeof(unsigned long long)); | |
1da177e4 LT |
334 | } |
335 | ||
b46b8f19 | 336 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
337 | { |
338 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
339 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 340 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 341 | sizeof(unsigned long long) - |
87a927c7 | 342 | REDZONE_ALIGN); |
3b0efdfa | 343 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 344 | sizeof(unsigned long long)); |
1da177e4 LT |
345 | } |
346 | ||
343e0d7a | 347 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
348 | { |
349 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 350 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
351 | } |
352 | ||
353 | #else | |
354 | ||
3dafccf2 | 355 | #define obj_offset(x) 0 |
b46b8f19 DW |
356 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
357 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
358 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
359 | ||
360 | #endif | |
361 | ||
03787301 JK |
362 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
363 | ||
d31676df | 364 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 365 | { |
d31676df JK |
366 | return atomic_read(&cachep->store_user_clean) == 1; |
367 | } | |
03787301 | 368 | |
d31676df JK |
369 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
370 | { | |
371 | atomic_set(&cachep->store_user_clean, 1); | |
372 | } | |
03787301 | 373 | |
d31676df JK |
374 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
375 | { | |
376 | if (is_store_user_clean(cachep)) | |
377 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
378 | } |
379 | ||
380 | #else | |
d31676df | 381 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
382 | |
383 | #endif | |
384 | ||
1da177e4 | 385 | /* |
3df1cccd DR |
386 | * Do not go above this order unless 0 objects fit into the slab or |
387 | * overridden on the command line. | |
1da177e4 | 388 | */ |
543585cc DR |
389 | #define SLAB_MAX_ORDER_HI 1 |
390 | #define SLAB_MAX_ORDER_LO 0 | |
391 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 392 | static bool slab_max_order_set __initdata; |
1da177e4 | 393 | |
6ed5eb22 PE |
394 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
395 | { | |
b49af68f | 396 | struct page *page = virt_to_head_page(obj); |
35026088 | 397 | return page->slab_cache; |
6ed5eb22 PE |
398 | } |
399 | ||
8456a648 | 400 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
401 | unsigned int idx) |
402 | { | |
8456a648 | 403 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
404 | } |
405 | ||
6a2d7a95 | 406 | /* |
3b0efdfa CL |
407 | * We want to avoid an expensive divide : (offset / cache->size) |
408 | * Using the fact that size is a constant for a particular cache, | |
409 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
410 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
411 | */ | |
412 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 413 | const struct page *page, void *obj) |
8fea4e96 | 414 | { |
8456a648 | 415 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 416 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
417 | } |
418 | ||
6fb92430 | 419 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 420 | /* internal cache of cache description objs */ |
9b030cb8 | 421 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
422 | .batchcount = 1, |
423 | .limit = BOOT_CPUCACHE_ENTRIES, | |
424 | .shared = 1, | |
3b0efdfa | 425 | .size = sizeof(struct kmem_cache), |
b28a02de | 426 | .name = "kmem_cache", |
1da177e4 LT |
427 | }; |
428 | ||
1871e52c | 429 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 430 | |
343e0d7a | 431 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 432 | { |
bf0dea23 | 433 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
434 | } |
435 | ||
a737b3e2 AM |
436 | /* |
437 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
438 | */ | |
70f75067 JK |
439 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
440 | unsigned long flags, size_t *left_over) | |
fbaccacf | 441 | { |
70f75067 | 442 | unsigned int num; |
fbaccacf | 443 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 444 | |
fbaccacf SR |
445 | /* |
446 | * The slab management structure can be either off the slab or | |
447 | * on it. For the latter case, the memory allocated for a | |
448 | * slab is used for: | |
449 | * | |
fbaccacf | 450 | * - @buffer_size bytes for each object |
2e6b3602 JK |
451 | * - One freelist_idx_t for each object |
452 | * | |
453 | * We don't need to consider alignment of freelist because | |
454 | * freelist will be at the end of slab page. The objects will be | |
455 | * at the correct alignment. | |
fbaccacf SR |
456 | * |
457 | * If the slab management structure is off the slab, then the | |
458 | * alignment will already be calculated into the size. Because | |
459 | * the slabs are all pages aligned, the objects will be at the | |
460 | * correct alignment when allocated. | |
461 | */ | |
b03a017b | 462 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 463 | num = slab_size / buffer_size; |
2e6b3602 | 464 | *left_over = slab_size % buffer_size; |
fbaccacf | 465 | } else { |
70f75067 | 466 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
467 | *left_over = slab_size % |
468 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 469 | } |
70f75067 JK |
470 | |
471 | return num; | |
1da177e4 LT |
472 | } |
473 | ||
f28510d3 | 474 | #if DEBUG |
d40cee24 | 475 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 476 | |
a737b3e2 AM |
477 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
478 | char *msg) | |
1da177e4 | 479 | { |
1170532b | 480 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 481 | function, cachep->name, msg); |
1da177e4 | 482 | dump_stack(); |
373d4d09 | 483 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 484 | } |
f28510d3 | 485 | #endif |
1da177e4 | 486 | |
3395ee05 PM |
487 | /* |
488 | * By default on NUMA we use alien caches to stage the freeing of | |
489 | * objects allocated from other nodes. This causes massive memory | |
490 | * inefficiencies when using fake NUMA setup to split memory into a | |
491 | * large number of small nodes, so it can be disabled on the command | |
492 | * line | |
493 | */ | |
494 | ||
495 | static int use_alien_caches __read_mostly = 1; | |
496 | static int __init noaliencache_setup(char *s) | |
497 | { | |
498 | use_alien_caches = 0; | |
499 | return 1; | |
500 | } | |
501 | __setup("noaliencache", noaliencache_setup); | |
502 | ||
3df1cccd DR |
503 | static int __init slab_max_order_setup(char *str) |
504 | { | |
505 | get_option(&str, &slab_max_order); | |
506 | slab_max_order = slab_max_order < 0 ? 0 : | |
507 | min(slab_max_order, MAX_ORDER - 1); | |
508 | slab_max_order_set = true; | |
509 | ||
510 | return 1; | |
511 | } | |
512 | __setup("slab_max_order=", slab_max_order_setup); | |
513 | ||
8fce4d8e CL |
514 | #ifdef CONFIG_NUMA |
515 | /* | |
516 | * Special reaping functions for NUMA systems called from cache_reap(). | |
517 | * These take care of doing round robin flushing of alien caches (containing | |
518 | * objects freed on different nodes from which they were allocated) and the | |
519 | * flushing of remote pcps by calling drain_node_pages. | |
520 | */ | |
1871e52c | 521 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
522 | |
523 | static void init_reap_node(int cpu) | |
524 | { | |
525 | int node; | |
526 | ||
7d6e6d09 | 527 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 528 | if (node == MAX_NUMNODES) |
442295c9 | 529 | node = first_node(node_online_map); |
8fce4d8e | 530 | |
1871e52c | 531 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
532 | } |
533 | ||
534 | static void next_reap_node(void) | |
535 | { | |
909ea964 | 536 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 537 | |
8fce4d8e CL |
538 | node = next_node(node, node_online_map); |
539 | if (unlikely(node >= MAX_NUMNODES)) | |
540 | node = first_node(node_online_map); | |
909ea964 | 541 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
542 | } |
543 | ||
544 | #else | |
545 | #define init_reap_node(cpu) do { } while (0) | |
546 | #define next_reap_node(void) do { } while (0) | |
547 | #endif | |
548 | ||
1da177e4 LT |
549 | /* |
550 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
551 | * via the workqueue/eventd. | |
552 | * Add the CPU number into the expiration time to minimize the possibility of | |
553 | * the CPUs getting into lockstep and contending for the global cache chain | |
554 | * lock. | |
555 | */ | |
0db0628d | 556 | static void start_cpu_timer(int cpu) |
1da177e4 | 557 | { |
1871e52c | 558 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
559 | |
560 | /* | |
561 | * When this gets called from do_initcalls via cpucache_init(), | |
562 | * init_workqueues() has already run, so keventd will be setup | |
563 | * at that time. | |
564 | */ | |
52bad64d | 565 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 566 | init_reap_node(cpu); |
203b42f7 | 567 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
568 | schedule_delayed_work_on(cpu, reap_work, |
569 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
570 | } |
571 | } | |
572 | ||
1fe00d50 | 573 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 574 | { |
d5cff635 CM |
575 | /* |
576 | * The array_cache structures contain pointers to free object. | |
25985edc | 577 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
578 | * cache the pointers are not cleared and they could be counted as |
579 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
580 | * not scan such objects. | |
581 | */ | |
1fe00d50 JK |
582 | kmemleak_no_scan(ac); |
583 | if (ac) { | |
584 | ac->avail = 0; | |
585 | ac->limit = limit; | |
586 | ac->batchcount = batch; | |
587 | ac->touched = 0; | |
1da177e4 | 588 | } |
1fe00d50 JK |
589 | } |
590 | ||
591 | static struct array_cache *alloc_arraycache(int node, int entries, | |
592 | int batchcount, gfp_t gfp) | |
593 | { | |
5e804789 | 594 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
595 | struct array_cache *ac = NULL; |
596 | ||
597 | ac = kmalloc_node(memsize, gfp, node); | |
598 | init_arraycache(ac, entries, batchcount); | |
599 | return ac; | |
1da177e4 LT |
600 | } |
601 | ||
f68f8ddd JK |
602 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
603 | struct page *page, void *objp) | |
072bb0aa | 604 | { |
f68f8ddd JK |
605 | struct kmem_cache_node *n; |
606 | int page_node; | |
607 | LIST_HEAD(list); | |
072bb0aa | 608 | |
f68f8ddd JK |
609 | page_node = page_to_nid(page); |
610 | n = get_node(cachep, page_node); | |
381760ea | 611 | |
f68f8ddd JK |
612 | spin_lock(&n->list_lock); |
613 | free_block(cachep, &objp, 1, page_node, &list); | |
614 | spin_unlock(&n->list_lock); | |
381760ea | 615 | |
f68f8ddd | 616 | slabs_destroy(cachep, &list); |
072bb0aa MG |
617 | } |
618 | ||
3ded175a CL |
619 | /* |
620 | * Transfer objects in one arraycache to another. | |
621 | * Locking must be handled by the caller. | |
622 | * | |
623 | * Return the number of entries transferred. | |
624 | */ | |
625 | static int transfer_objects(struct array_cache *to, | |
626 | struct array_cache *from, unsigned int max) | |
627 | { | |
628 | /* Figure out how many entries to transfer */ | |
732eacc0 | 629 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
630 | |
631 | if (!nr) | |
632 | return 0; | |
633 | ||
634 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
635 | sizeof(void *) *nr); | |
636 | ||
637 | from->avail -= nr; | |
638 | to->avail += nr; | |
3ded175a CL |
639 | return nr; |
640 | } | |
641 | ||
765c4507 CL |
642 | #ifndef CONFIG_NUMA |
643 | ||
644 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 645 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 646 | |
c8522a3a JK |
647 | static inline struct alien_cache **alloc_alien_cache(int node, |
648 | int limit, gfp_t gfp) | |
765c4507 | 649 | { |
8888177e | 650 | return NULL; |
765c4507 CL |
651 | } |
652 | ||
c8522a3a | 653 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
654 | { |
655 | } | |
656 | ||
657 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
658 | { | |
659 | return 0; | |
660 | } | |
661 | ||
662 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
663 | gfp_t flags) | |
664 | { | |
665 | return NULL; | |
666 | } | |
667 | ||
8b98c169 | 668 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
669 | gfp_t flags, int nodeid) |
670 | { | |
671 | return NULL; | |
672 | } | |
673 | ||
4167e9b2 DR |
674 | static inline gfp_t gfp_exact_node(gfp_t flags) |
675 | { | |
444eb2a4 | 676 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
677 | } |
678 | ||
765c4507 CL |
679 | #else /* CONFIG_NUMA */ |
680 | ||
8b98c169 | 681 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 682 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 683 | |
c8522a3a JK |
684 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
685 | int batch, gfp_t gfp) | |
686 | { | |
5e804789 | 687 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
688 | struct alien_cache *alc = NULL; |
689 | ||
690 | alc = kmalloc_node(memsize, gfp, node); | |
691 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 692 | spin_lock_init(&alc->lock); |
c8522a3a JK |
693 | return alc; |
694 | } | |
695 | ||
696 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 697 | { |
c8522a3a | 698 | struct alien_cache **alc_ptr; |
5e804789 | 699 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
700 | int i; |
701 | ||
702 | if (limit > 1) | |
703 | limit = 12; | |
c8522a3a JK |
704 | alc_ptr = kzalloc_node(memsize, gfp, node); |
705 | if (!alc_ptr) | |
706 | return NULL; | |
707 | ||
708 | for_each_node(i) { | |
709 | if (i == node || !node_online(i)) | |
710 | continue; | |
711 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
712 | if (!alc_ptr[i]) { | |
713 | for (i--; i >= 0; i--) | |
714 | kfree(alc_ptr[i]); | |
715 | kfree(alc_ptr); | |
716 | return NULL; | |
e498be7d CL |
717 | } |
718 | } | |
c8522a3a | 719 | return alc_ptr; |
e498be7d CL |
720 | } |
721 | ||
c8522a3a | 722 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
723 | { |
724 | int i; | |
725 | ||
c8522a3a | 726 | if (!alc_ptr) |
e498be7d | 727 | return; |
e498be7d | 728 | for_each_node(i) |
c8522a3a JK |
729 | kfree(alc_ptr[i]); |
730 | kfree(alc_ptr); | |
e498be7d CL |
731 | } |
732 | ||
343e0d7a | 733 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
734 | struct array_cache *ac, int node, |
735 | struct list_head *list) | |
e498be7d | 736 | { |
18bf8541 | 737 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
738 | |
739 | if (ac->avail) { | |
ce8eb6c4 | 740 | spin_lock(&n->list_lock); |
e00946fe CL |
741 | /* |
742 | * Stuff objects into the remote nodes shared array first. | |
743 | * That way we could avoid the overhead of putting the objects | |
744 | * into the free lists and getting them back later. | |
745 | */ | |
ce8eb6c4 CL |
746 | if (n->shared) |
747 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 748 | |
833b706c | 749 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 750 | ac->avail = 0; |
ce8eb6c4 | 751 | spin_unlock(&n->list_lock); |
e498be7d CL |
752 | } |
753 | } | |
754 | ||
8fce4d8e CL |
755 | /* |
756 | * Called from cache_reap() to regularly drain alien caches round robin. | |
757 | */ | |
ce8eb6c4 | 758 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 759 | { |
909ea964 | 760 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 761 | |
ce8eb6c4 | 762 | if (n->alien) { |
c8522a3a JK |
763 | struct alien_cache *alc = n->alien[node]; |
764 | struct array_cache *ac; | |
765 | ||
766 | if (alc) { | |
767 | ac = &alc->ac; | |
49dfc304 | 768 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
769 | LIST_HEAD(list); |
770 | ||
771 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 772 | spin_unlock_irq(&alc->lock); |
833b706c | 773 | slabs_destroy(cachep, &list); |
c8522a3a | 774 | } |
8fce4d8e CL |
775 | } |
776 | } | |
777 | } | |
778 | ||
a737b3e2 | 779 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 780 | struct alien_cache **alien) |
e498be7d | 781 | { |
b28a02de | 782 | int i = 0; |
c8522a3a | 783 | struct alien_cache *alc; |
e498be7d CL |
784 | struct array_cache *ac; |
785 | unsigned long flags; | |
786 | ||
787 | for_each_online_node(i) { | |
c8522a3a JK |
788 | alc = alien[i]; |
789 | if (alc) { | |
833b706c JK |
790 | LIST_HEAD(list); |
791 | ||
c8522a3a | 792 | ac = &alc->ac; |
49dfc304 | 793 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 794 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 795 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 796 | slabs_destroy(cachep, &list); |
e498be7d CL |
797 | } |
798 | } | |
799 | } | |
729bd0b7 | 800 | |
25c4f304 JK |
801 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
802 | int node, int page_node) | |
729bd0b7 | 803 | { |
ce8eb6c4 | 804 | struct kmem_cache_node *n; |
c8522a3a JK |
805 | struct alien_cache *alien = NULL; |
806 | struct array_cache *ac; | |
97654dfa | 807 | LIST_HEAD(list); |
1ca4cb24 | 808 | |
18bf8541 | 809 | n = get_node(cachep, node); |
729bd0b7 | 810 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
811 | if (n->alien && n->alien[page_node]) { |
812 | alien = n->alien[page_node]; | |
c8522a3a | 813 | ac = &alien->ac; |
49dfc304 | 814 | spin_lock(&alien->lock); |
c8522a3a | 815 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 816 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 817 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 818 | } |
f68f8ddd | 819 | ac->entry[ac->avail++] = objp; |
49dfc304 | 820 | spin_unlock(&alien->lock); |
833b706c | 821 | slabs_destroy(cachep, &list); |
729bd0b7 | 822 | } else { |
25c4f304 | 823 | n = get_node(cachep, page_node); |
18bf8541 | 824 | spin_lock(&n->list_lock); |
25c4f304 | 825 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 826 | spin_unlock(&n->list_lock); |
97654dfa | 827 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
828 | } |
829 | return 1; | |
830 | } | |
25c4f304 JK |
831 | |
832 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
833 | { | |
834 | int page_node = page_to_nid(virt_to_page(objp)); | |
835 | int node = numa_mem_id(); | |
836 | /* | |
837 | * Make sure we are not freeing a object from another node to the array | |
838 | * cache on this cpu. | |
839 | */ | |
840 | if (likely(node == page_node)) | |
841 | return 0; | |
842 | ||
843 | return __cache_free_alien(cachep, objp, node, page_node); | |
844 | } | |
4167e9b2 DR |
845 | |
846 | /* | |
444eb2a4 MG |
847 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
848 | * warn about failures. | |
4167e9b2 DR |
849 | */ |
850 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
851 | { | |
444eb2a4 | 852 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 853 | } |
e498be7d CL |
854 | #endif |
855 | ||
ded0ecf6 JK |
856 | static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp) |
857 | { | |
858 | struct kmem_cache_node *n; | |
859 | ||
860 | /* | |
861 | * Set up the kmem_cache_node for cpu before we can | |
862 | * begin anything. Make sure some other cpu on this | |
863 | * node has not already allocated this | |
864 | */ | |
865 | n = get_node(cachep, node); | |
866 | if (n) { | |
867 | spin_lock_irq(&n->list_lock); | |
868 | n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount + | |
869 | cachep->num; | |
870 | spin_unlock_irq(&n->list_lock); | |
871 | ||
872 | return 0; | |
873 | } | |
874 | ||
875 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); | |
876 | if (!n) | |
877 | return -ENOMEM; | |
878 | ||
879 | kmem_cache_node_init(n); | |
880 | n->next_reap = jiffies + REAPTIMEOUT_NODE + | |
881 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
882 | ||
883 | n->free_limit = | |
884 | (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num; | |
885 | ||
886 | /* | |
887 | * The kmem_cache_nodes don't come and go as CPUs | |
888 | * come and go. slab_mutex is sufficient | |
889 | * protection here. | |
890 | */ | |
891 | cachep->node[node] = n; | |
892 | ||
893 | return 0; | |
894 | } | |
895 | ||
8f9f8d9e | 896 | /* |
6a67368c | 897 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 898 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 899 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 900 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
901 | * already in use. |
902 | * | |
18004c5d | 903 | * Must hold slab_mutex. |
8f9f8d9e | 904 | */ |
6a67368c | 905 | static int init_cache_node_node(int node) |
8f9f8d9e | 906 | { |
ded0ecf6 | 907 | int ret; |
8f9f8d9e | 908 | struct kmem_cache *cachep; |
8f9f8d9e | 909 | |
18004c5d | 910 | list_for_each_entry(cachep, &slab_caches, list) { |
ded0ecf6 JK |
911 | ret = init_cache_node(cachep, node, GFP_KERNEL); |
912 | if (ret) | |
913 | return ret; | |
8f9f8d9e | 914 | } |
ded0ecf6 | 915 | |
8f9f8d9e DR |
916 | return 0; |
917 | } | |
918 | ||
c3d332b6 JK |
919 | static int setup_kmem_cache_node(struct kmem_cache *cachep, |
920 | int node, gfp_t gfp, bool force_change) | |
921 | { | |
922 | int ret = -ENOMEM; | |
923 | struct kmem_cache_node *n; | |
924 | struct array_cache *old_shared = NULL; | |
925 | struct array_cache *new_shared = NULL; | |
926 | struct alien_cache **new_alien = NULL; | |
927 | LIST_HEAD(list); | |
928 | ||
929 | if (use_alien_caches) { | |
930 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
931 | if (!new_alien) | |
932 | goto fail; | |
933 | } | |
934 | ||
935 | if (cachep->shared) { | |
936 | new_shared = alloc_arraycache(node, | |
937 | cachep->shared * cachep->batchcount, 0xbaadf00d, gfp); | |
938 | if (!new_shared) | |
939 | goto fail; | |
940 | } | |
941 | ||
942 | ret = init_cache_node(cachep, node, gfp); | |
943 | if (ret) | |
944 | goto fail; | |
945 | ||
946 | n = get_node(cachep, node); | |
947 | spin_lock_irq(&n->list_lock); | |
948 | if (n->shared && force_change) { | |
949 | free_block(cachep, n->shared->entry, | |
950 | n->shared->avail, node, &list); | |
951 | n->shared->avail = 0; | |
952 | } | |
953 | ||
954 | if (!n->shared || force_change) { | |
955 | old_shared = n->shared; | |
956 | n->shared = new_shared; | |
957 | new_shared = NULL; | |
958 | } | |
959 | ||
960 | if (!n->alien) { | |
961 | n->alien = new_alien; | |
962 | new_alien = NULL; | |
963 | } | |
964 | ||
965 | spin_unlock_irq(&n->list_lock); | |
966 | slabs_destroy(cachep, &list); | |
967 | ||
968 | fail: | |
969 | kfree(old_shared); | |
970 | kfree(new_shared); | |
971 | free_alien_cache(new_alien); | |
972 | ||
973 | return ret; | |
974 | } | |
975 | ||
0db0628d | 976 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
977 | { |
978 | struct kmem_cache *cachep; | |
ce8eb6c4 | 979 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 980 | int node = cpu_to_mem(cpu); |
a70f7302 | 981 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 982 | |
18004c5d | 983 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
984 | struct array_cache *nc; |
985 | struct array_cache *shared; | |
c8522a3a | 986 | struct alien_cache **alien; |
97654dfa | 987 | LIST_HEAD(list); |
fbf1e473 | 988 | |
18bf8541 | 989 | n = get_node(cachep, node); |
ce8eb6c4 | 990 | if (!n) |
bf0dea23 | 991 | continue; |
fbf1e473 | 992 | |
ce8eb6c4 | 993 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 994 | |
ce8eb6c4 CL |
995 | /* Free limit for this kmem_cache_node */ |
996 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
997 | |
998 | /* cpu is dead; no one can alloc from it. */ | |
999 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1000 | if (nc) { | |
97654dfa | 1001 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1002 | nc->avail = 0; |
1003 | } | |
fbf1e473 | 1004 | |
58463c1f | 1005 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1006 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1007 | goto free_slab; |
fbf1e473 AM |
1008 | } |
1009 | ||
ce8eb6c4 | 1010 | shared = n->shared; |
fbf1e473 AM |
1011 | if (shared) { |
1012 | free_block(cachep, shared->entry, | |
97654dfa | 1013 | shared->avail, node, &list); |
ce8eb6c4 | 1014 | n->shared = NULL; |
fbf1e473 AM |
1015 | } |
1016 | ||
ce8eb6c4 CL |
1017 | alien = n->alien; |
1018 | n->alien = NULL; | |
fbf1e473 | 1019 | |
ce8eb6c4 | 1020 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1021 | |
1022 | kfree(shared); | |
1023 | if (alien) { | |
1024 | drain_alien_cache(cachep, alien); | |
1025 | free_alien_cache(alien); | |
1026 | } | |
bf0dea23 JK |
1027 | |
1028 | free_slab: | |
97654dfa | 1029 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1030 | } |
1031 | /* | |
1032 | * In the previous loop, all the objects were freed to | |
1033 | * the respective cache's slabs, now we can go ahead and | |
1034 | * shrink each nodelist to its limit. | |
1035 | */ | |
18004c5d | 1036 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1037 | n = get_node(cachep, node); |
ce8eb6c4 | 1038 | if (!n) |
fbf1e473 | 1039 | continue; |
a5aa63a5 | 1040 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
1041 | } |
1042 | } | |
1043 | ||
0db0628d | 1044 | static int cpuup_prepare(long cpu) |
1da177e4 | 1045 | { |
343e0d7a | 1046 | struct kmem_cache *cachep; |
7d6e6d09 | 1047 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1048 | int err; |
1da177e4 | 1049 | |
fbf1e473 AM |
1050 | /* |
1051 | * We need to do this right in the beginning since | |
1052 | * alloc_arraycache's are going to use this list. | |
1053 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1054 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1055 | */ |
6a67368c | 1056 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1057 | if (err < 0) |
1058 | goto bad; | |
fbf1e473 AM |
1059 | |
1060 | /* | |
1061 | * Now we can go ahead with allocating the shared arrays and | |
1062 | * array caches | |
1063 | */ | |
18004c5d | 1064 | list_for_each_entry(cachep, &slab_caches, list) { |
c3d332b6 JK |
1065 | err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false); |
1066 | if (err) | |
1067 | goto bad; | |
fbf1e473 | 1068 | } |
ce79ddc8 | 1069 | |
fbf1e473 AM |
1070 | return 0; |
1071 | bad: | |
12d00f6a | 1072 | cpuup_canceled(cpu); |
fbf1e473 AM |
1073 | return -ENOMEM; |
1074 | } | |
1075 | ||
0db0628d | 1076 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1077 | unsigned long action, void *hcpu) |
1078 | { | |
1079 | long cpu = (long)hcpu; | |
1080 | int err = 0; | |
1081 | ||
1082 | switch (action) { | |
fbf1e473 AM |
1083 | case CPU_UP_PREPARE: |
1084 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1085 | mutex_lock(&slab_mutex); |
fbf1e473 | 1086 | err = cpuup_prepare(cpu); |
18004c5d | 1087 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1088 | break; |
1089 | case CPU_ONLINE: | |
8bb78442 | 1090 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1091 | start_cpu_timer(cpu); |
1092 | break; | |
1093 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1094 | case CPU_DOWN_PREPARE: |
8bb78442 | 1095 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1096 | /* |
18004c5d | 1097 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1098 | * held so that if cache_reap() is invoked it cannot do |
1099 | * anything expensive but will only modify reap_work | |
1100 | * and reschedule the timer. | |
1101 | */ | |
afe2c511 | 1102 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1103 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1104 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1105 | break; |
1106 | case CPU_DOWN_FAILED: | |
8bb78442 | 1107 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1108 | start_cpu_timer(cpu); |
1109 | break; | |
1da177e4 | 1110 | case CPU_DEAD: |
8bb78442 | 1111 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1112 | /* |
1113 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1114 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1115 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1116 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1117 | * structure is usually allocated from kmem_cache_create() and |
1118 | * gets destroyed at kmem_cache_destroy(). | |
1119 | */ | |
183ff22b | 1120 | /* fall through */ |
8f5be20b | 1121 | #endif |
1da177e4 | 1122 | case CPU_UP_CANCELED: |
8bb78442 | 1123 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1124 | mutex_lock(&slab_mutex); |
fbf1e473 | 1125 | cpuup_canceled(cpu); |
18004c5d | 1126 | mutex_unlock(&slab_mutex); |
1da177e4 | 1127 | break; |
1da177e4 | 1128 | } |
eac40680 | 1129 | return notifier_from_errno(err); |
1da177e4 LT |
1130 | } |
1131 | ||
0db0628d | 1132 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1133 | &cpuup_callback, NULL, 0 |
1134 | }; | |
1da177e4 | 1135 | |
8f9f8d9e DR |
1136 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1137 | /* | |
1138 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1139 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1140 | * removed. | |
1141 | * | |
18004c5d | 1142 | * Must hold slab_mutex. |
8f9f8d9e | 1143 | */ |
6a67368c | 1144 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1145 | { |
1146 | struct kmem_cache *cachep; | |
1147 | int ret = 0; | |
1148 | ||
18004c5d | 1149 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1150 | struct kmem_cache_node *n; |
8f9f8d9e | 1151 | |
18bf8541 | 1152 | n = get_node(cachep, node); |
ce8eb6c4 | 1153 | if (!n) |
8f9f8d9e DR |
1154 | continue; |
1155 | ||
a5aa63a5 | 1156 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1157 | |
ce8eb6c4 CL |
1158 | if (!list_empty(&n->slabs_full) || |
1159 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1160 | ret = -EBUSY; |
1161 | break; | |
1162 | } | |
1163 | } | |
1164 | return ret; | |
1165 | } | |
1166 | ||
1167 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1168 | unsigned long action, void *arg) | |
1169 | { | |
1170 | struct memory_notify *mnb = arg; | |
1171 | int ret = 0; | |
1172 | int nid; | |
1173 | ||
1174 | nid = mnb->status_change_nid; | |
1175 | if (nid < 0) | |
1176 | goto out; | |
1177 | ||
1178 | switch (action) { | |
1179 | case MEM_GOING_ONLINE: | |
18004c5d | 1180 | mutex_lock(&slab_mutex); |
6a67368c | 1181 | ret = init_cache_node_node(nid); |
18004c5d | 1182 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1183 | break; |
1184 | case MEM_GOING_OFFLINE: | |
18004c5d | 1185 | mutex_lock(&slab_mutex); |
6a67368c | 1186 | ret = drain_cache_node_node(nid); |
18004c5d | 1187 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1188 | break; |
1189 | case MEM_ONLINE: | |
1190 | case MEM_OFFLINE: | |
1191 | case MEM_CANCEL_ONLINE: | |
1192 | case MEM_CANCEL_OFFLINE: | |
1193 | break; | |
1194 | } | |
1195 | out: | |
5fda1bd5 | 1196 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1197 | } |
1198 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1199 | ||
e498be7d | 1200 | /* |
ce8eb6c4 | 1201 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1202 | */ |
6744f087 | 1203 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1204 | int nodeid) |
e498be7d | 1205 | { |
6744f087 | 1206 | struct kmem_cache_node *ptr; |
e498be7d | 1207 | |
6744f087 | 1208 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1209 | BUG_ON(!ptr); |
1210 | ||
6744f087 | 1211 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1212 | /* |
1213 | * Do not assume that spinlocks can be initialized via memcpy: | |
1214 | */ | |
1215 | spin_lock_init(&ptr->list_lock); | |
1216 | ||
e498be7d | 1217 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1218 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1219 | } |
1220 | ||
556a169d | 1221 | /* |
ce8eb6c4 CL |
1222 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1223 | * size of kmem_cache_node. | |
556a169d | 1224 | */ |
ce8eb6c4 | 1225 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1226 | { |
1227 | int node; | |
1228 | ||
1229 | for_each_online_node(node) { | |
ce8eb6c4 | 1230 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1231 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1232 | REAPTIMEOUT_NODE + |
1233 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1234 | } |
1235 | } | |
1236 | ||
a737b3e2 AM |
1237 | /* |
1238 | * Initialisation. Called after the page allocator have been initialised and | |
1239 | * before smp_init(). | |
1da177e4 LT |
1240 | */ |
1241 | void __init kmem_cache_init(void) | |
1242 | { | |
e498be7d CL |
1243 | int i; |
1244 | ||
68126702 JK |
1245 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1246 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1247 | kmem_cache = &kmem_cache_boot; |
1248 | ||
8888177e | 1249 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1250 | use_alien_caches = 0; |
1251 | ||
3c583465 | 1252 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1253 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1254 | |
1da177e4 LT |
1255 | /* |
1256 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1257 | * page orders on machines with more than 32MB of memory if |
1258 | * not overridden on the command line. | |
1da177e4 | 1259 | */ |
3df1cccd | 1260 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1261 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1262 | |
1da177e4 LT |
1263 | /* Bootstrap is tricky, because several objects are allocated |
1264 | * from caches that do not exist yet: | |
9b030cb8 CL |
1265 | * 1) initialize the kmem_cache cache: it contains the struct |
1266 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1267 | * kmem_cache is statically allocated. | |
e498be7d | 1268 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1269 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1270 | * array at the end of the bootstrap. |
1da177e4 | 1271 | * 2) Create the first kmalloc cache. |
343e0d7a | 1272 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1273 | * An __init data area is used for the head array. |
1274 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1275 | * head arrays. | |
9b030cb8 | 1276 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1277 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1278 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1279 | * the other cache's with kmalloc allocated memory. |
1280 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1281 | */ |
1282 | ||
9b030cb8 | 1283 | /* 1) create the kmem_cache */ |
1da177e4 | 1284 | |
8da3430d | 1285 | /* |
b56efcf0 | 1286 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1287 | */ |
2f9baa9f | 1288 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1289 | offsetof(struct kmem_cache, node) + |
6744f087 | 1290 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1291 | SLAB_HWCACHE_ALIGN); |
1292 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1293 | slab_state = PARTIAL; |
1da177e4 | 1294 | |
a737b3e2 | 1295 | /* |
bf0dea23 JK |
1296 | * Initialize the caches that provide memory for the kmem_cache_node |
1297 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1298 | */ |
bf0dea23 | 1299 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1300 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1301 | slab_state = PARTIAL_NODE; |
34cc6990 | 1302 | setup_kmalloc_cache_index_table(); |
e498be7d | 1303 | |
e0a42726 IM |
1304 | slab_early_init = 0; |
1305 | ||
ce8eb6c4 | 1306 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1307 | { |
1ca4cb24 PE |
1308 | int nid; |
1309 | ||
9c09a95c | 1310 | for_each_online_node(nid) { |
ce8eb6c4 | 1311 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1312 | |
bf0dea23 | 1313 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1314 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1315 | } |
1316 | } | |
1da177e4 | 1317 | |
f97d5f63 | 1318 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1319 | } |
1320 | ||
1321 | void __init kmem_cache_init_late(void) | |
1322 | { | |
1323 | struct kmem_cache *cachep; | |
1324 | ||
97d06609 | 1325 | slab_state = UP; |
52cef189 | 1326 | |
8429db5c | 1327 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1328 | mutex_lock(&slab_mutex); |
1329 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1330 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1331 | BUG(); | |
18004c5d | 1332 | mutex_unlock(&slab_mutex); |
056c6241 | 1333 | |
97d06609 CL |
1334 | /* Done! */ |
1335 | slab_state = FULL; | |
1336 | ||
a737b3e2 AM |
1337 | /* |
1338 | * Register a cpu startup notifier callback that initializes | |
1339 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1340 | */ |
1341 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1342 | |
8f9f8d9e DR |
1343 | #ifdef CONFIG_NUMA |
1344 | /* | |
1345 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1346 | * node. |
8f9f8d9e DR |
1347 | */ |
1348 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1349 | #endif | |
1350 | ||
a737b3e2 AM |
1351 | /* |
1352 | * The reap timers are started later, with a module init call: That part | |
1353 | * of the kernel is not yet operational. | |
1da177e4 LT |
1354 | */ |
1355 | } | |
1356 | ||
1357 | static int __init cpucache_init(void) | |
1358 | { | |
1359 | int cpu; | |
1360 | ||
a737b3e2 AM |
1361 | /* |
1362 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1363 | */ |
e498be7d | 1364 | for_each_online_cpu(cpu) |
a737b3e2 | 1365 | start_cpu_timer(cpu); |
a164f896 GC |
1366 | |
1367 | /* Done! */ | |
97d06609 | 1368 | slab_state = FULL; |
1da177e4 LT |
1369 | return 0; |
1370 | } | |
1da177e4 LT |
1371 | __initcall(cpucache_init); |
1372 | ||
8bdec192 RA |
1373 | static noinline void |
1374 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1375 | { | |
9a02d699 | 1376 | #if DEBUG |
ce8eb6c4 | 1377 | struct kmem_cache_node *n; |
8456a648 | 1378 | struct page *page; |
8bdec192 RA |
1379 | unsigned long flags; |
1380 | int node; | |
9a02d699 DR |
1381 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1382 | DEFAULT_RATELIMIT_BURST); | |
1383 | ||
1384 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1385 | return; | |
8bdec192 | 1386 | |
5b3810e5 VB |
1387 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1388 | nodeid, gfpflags, &gfpflags); | |
1389 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1390 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1391 | |
18bf8541 | 1392 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1393 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1394 | unsigned long active_slabs = 0, num_slabs = 0; | |
1395 | ||
ce8eb6c4 | 1396 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1397 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1398 | active_objs += cachep->num; |
1399 | active_slabs++; | |
1400 | } | |
8456a648 JK |
1401 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1402 | active_objs += page->active; | |
8bdec192 RA |
1403 | active_slabs++; |
1404 | } | |
8456a648 | 1405 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1406 | num_slabs++; |
1407 | ||
ce8eb6c4 CL |
1408 | free_objects += n->free_objects; |
1409 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1410 | |
1411 | num_slabs += active_slabs; | |
1412 | num_objs = num_slabs * cachep->num; | |
5b3810e5 | 1413 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", |
8bdec192 RA |
1414 | node, active_slabs, num_slabs, active_objs, num_objs, |
1415 | free_objects); | |
1416 | } | |
9a02d699 | 1417 | #endif |
8bdec192 RA |
1418 | } |
1419 | ||
1da177e4 | 1420 | /* |
8a7d9b43 WSH |
1421 | * Interface to system's page allocator. No need to hold the |
1422 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1423 | * |
1424 | * If we requested dmaable memory, we will get it. Even if we | |
1425 | * did not request dmaable memory, we might get it, but that | |
1426 | * would be relatively rare and ignorable. | |
1427 | */ | |
0c3aa83e JK |
1428 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1429 | int nodeid) | |
1da177e4 LT |
1430 | { |
1431 | struct page *page; | |
e1b6aa6f | 1432 | int nr_pages; |
765c4507 | 1433 | |
a618e89f | 1434 | flags |= cachep->allocflags; |
e12ba74d MG |
1435 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1436 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1437 | |
96db800f | 1438 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1439 | if (!page) { |
9a02d699 | 1440 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1441 | return NULL; |
8bdec192 | 1442 | } |
1da177e4 | 1443 | |
f3ccb2c4 VD |
1444 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1445 | __free_pages(page, cachep->gfporder); | |
1446 | return NULL; | |
1447 | } | |
1448 | ||
e1b6aa6f | 1449 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1450 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1451 | add_zone_page_state(page_zone(page), |
1452 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1453 | else | |
1454 | add_zone_page_state(page_zone(page), | |
1455 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
f68f8ddd | 1456 | |
a57a4988 | 1457 | __SetPageSlab(page); |
f68f8ddd JK |
1458 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1459 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1460 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1461 | |
b1eeab67 VN |
1462 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1463 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1464 | ||
1465 | if (cachep->ctor) | |
1466 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1467 | else | |
1468 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1469 | } | |
c175eea4 | 1470 | |
0c3aa83e | 1471 | return page; |
1da177e4 LT |
1472 | } |
1473 | ||
1474 | /* | |
1475 | * Interface to system's page release. | |
1476 | */ | |
0c3aa83e | 1477 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1478 | { |
27ee57c9 VD |
1479 | int order = cachep->gfporder; |
1480 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1481 | |
27ee57c9 | 1482 | kmemcheck_free_shadow(page, order); |
c175eea4 | 1483 | |
972d1a7b CL |
1484 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1485 | sub_zone_page_state(page_zone(page), | |
1486 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1487 | else | |
1488 | sub_zone_page_state(page_zone(page), | |
1489 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1490 | |
a57a4988 | 1491 | BUG_ON(!PageSlab(page)); |
73293c2f | 1492 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1493 | __ClearPageSlab(page); |
8456a648 JK |
1494 | page_mapcount_reset(page); |
1495 | page->mapping = NULL; | |
1f458cbf | 1496 | |
1da177e4 LT |
1497 | if (current->reclaim_state) |
1498 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1499 | memcg_uncharge_slab(page, order, cachep); |
1500 | __free_pages(page, order); | |
1da177e4 LT |
1501 | } |
1502 | ||
1503 | static void kmem_rcu_free(struct rcu_head *head) | |
1504 | { | |
68126702 JK |
1505 | struct kmem_cache *cachep; |
1506 | struct page *page; | |
1da177e4 | 1507 | |
68126702 JK |
1508 | page = container_of(head, struct page, rcu_head); |
1509 | cachep = page->slab_cache; | |
1510 | ||
1511 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | #if DEBUG | |
40b44137 JK |
1515 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1516 | { | |
1517 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1518 | (cachep->size % PAGE_SIZE) == 0) | |
1519 | return true; | |
1520 | ||
1521 | return false; | |
1522 | } | |
1da177e4 LT |
1523 | |
1524 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1525 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1526 | unsigned long caller) |
1da177e4 | 1527 | { |
8c138bc0 | 1528 | int size = cachep->object_size; |
1da177e4 | 1529 | |
3dafccf2 | 1530 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1531 | |
b28a02de | 1532 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1533 | return; |
1534 | ||
b28a02de PE |
1535 | *addr++ = 0x12345678; |
1536 | *addr++ = caller; | |
1537 | *addr++ = smp_processor_id(); | |
1538 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1539 | { |
1540 | unsigned long *sptr = &caller; | |
1541 | unsigned long svalue; | |
1542 | ||
1543 | while (!kstack_end(sptr)) { | |
1544 | svalue = *sptr++; | |
1545 | if (kernel_text_address(svalue)) { | |
b28a02de | 1546 | *addr++ = svalue; |
1da177e4 LT |
1547 | size -= sizeof(unsigned long); |
1548 | if (size <= sizeof(unsigned long)) | |
1549 | break; | |
1550 | } | |
1551 | } | |
1552 | ||
1553 | } | |
b28a02de | 1554 | *addr++ = 0x87654321; |
1da177e4 | 1555 | } |
40b44137 JK |
1556 | |
1557 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1558 | int map, unsigned long caller) | |
1559 | { | |
1560 | if (!is_debug_pagealloc_cache(cachep)) | |
1561 | return; | |
1562 | ||
1563 | if (caller) | |
1564 | store_stackinfo(cachep, objp, caller); | |
1565 | ||
1566 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1567 | } | |
1568 | ||
1569 | #else | |
1570 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1571 | int map, unsigned long caller) {} | |
1572 | ||
1da177e4 LT |
1573 | #endif |
1574 | ||
343e0d7a | 1575 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1576 | { |
8c138bc0 | 1577 | int size = cachep->object_size; |
3dafccf2 | 1578 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1579 | |
1580 | memset(addr, val, size); | |
b28a02de | 1581 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1582 | } |
1583 | ||
1584 | static void dump_line(char *data, int offset, int limit) | |
1585 | { | |
1586 | int i; | |
aa83aa40 DJ |
1587 | unsigned char error = 0; |
1588 | int bad_count = 0; | |
1589 | ||
1170532b | 1590 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1591 | for (i = 0; i < limit; i++) { |
1592 | if (data[offset + i] != POISON_FREE) { | |
1593 | error = data[offset + i]; | |
1594 | bad_count++; | |
1595 | } | |
aa83aa40 | 1596 | } |
fdde6abb SAS |
1597 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1598 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1599 | |
1600 | if (bad_count == 1) { | |
1601 | error ^= POISON_FREE; | |
1602 | if (!(error & (error - 1))) { | |
1170532b | 1603 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1604 | #ifdef CONFIG_X86 |
1170532b | 1605 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1606 | #else |
1170532b | 1607 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1608 | #endif |
1609 | } | |
1610 | } | |
1da177e4 LT |
1611 | } |
1612 | #endif | |
1613 | ||
1614 | #if DEBUG | |
1615 | ||
343e0d7a | 1616 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1617 | { |
1618 | int i, size; | |
1619 | char *realobj; | |
1620 | ||
1621 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1622 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1623 | *dbg_redzone1(cachep, objp), | |
1624 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1625 | } |
1626 | ||
1627 | if (cachep->flags & SLAB_STORE_USER) { | |
1170532b | 1628 | pr_err("Last user: [<%p>](%pSR)\n", |
071361d3 JP |
1629 | *dbg_userword(cachep, objp), |
1630 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1631 | } |
3dafccf2 | 1632 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1633 | size = cachep->object_size; |
b28a02de | 1634 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1635 | int limit; |
1636 | limit = 16; | |
b28a02de PE |
1637 | if (i + limit > size) |
1638 | limit = size - i; | |
1da177e4 LT |
1639 | dump_line(realobj, i, limit); |
1640 | } | |
1641 | } | |
1642 | ||
343e0d7a | 1643 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1644 | { |
1645 | char *realobj; | |
1646 | int size, i; | |
1647 | int lines = 0; | |
1648 | ||
40b44137 JK |
1649 | if (is_debug_pagealloc_cache(cachep)) |
1650 | return; | |
1651 | ||
3dafccf2 | 1652 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1653 | size = cachep->object_size; |
1da177e4 | 1654 | |
b28a02de | 1655 | for (i = 0; i < size; i++) { |
1da177e4 | 1656 | char exp = POISON_FREE; |
b28a02de | 1657 | if (i == size - 1) |
1da177e4 LT |
1658 | exp = POISON_END; |
1659 | if (realobj[i] != exp) { | |
1660 | int limit; | |
1661 | /* Mismatch ! */ | |
1662 | /* Print header */ | |
1663 | if (lines == 0) { | |
1170532b JP |
1664 | pr_err("Slab corruption (%s): %s start=%p, len=%d\n", |
1665 | print_tainted(), cachep->name, | |
1666 | realobj, size); | |
1da177e4 LT |
1667 | print_objinfo(cachep, objp, 0); |
1668 | } | |
1669 | /* Hexdump the affected line */ | |
b28a02de | 1670 | i = (i / 16) * 16; |
1da177e4 | 1671 | limit = 16; |
b28a02de PE |
1672 | if (i + limit > size) |
1673 | limit = size - i; | |
1da177e4 LT |
1674 | dump_line(realobj, i, limit); |
1675 | i += 16; | |
1676 | lines++; | |
1677 | /* Limit to 5 lines */ | |
1678 | if (lines > 5) | |
1679 | break; | |
1680 | } | |
1681 | } | |
1682 | if (lines != 0) { | |
1683 | /* Print some data about the neighboring objects, if they | |
1684 | * exist: | |
1685 | */ | |
8456a648 | 1686 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1687 | unsigned int objnr; |
1da177e4 | 1688 | |
8456a648 | 1689 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1690 | if (objnr) { |
8456a648 | 1691 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1692 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1693 | pr_err("Prev obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1694 | print_objinfo(cachep, objp, 2); |
1695 | } | |
b28a02de | 1696 | if (objnr + 1 < cachep->num) { |
8456a648 | 1697 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1698 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1699 | pr_err("Next obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1700 | print_objinfo(cachep, objp, 2); |
1701 | } | |
1702 | } | |
1703 | } | |
1704 | #endif | |
1705 | ||
12dd36fa | 1706 | #if DEBUG |
8456a648 JK |
1707 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1708 | struct page *page) | |
1da177e4 | 1709 | { |
1da177e4 | 1710 | int i; |
b03a017b JK |
1711 | |
1712 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1713 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1714 | POISON_FREE); | |
1715 | } | |
1716 | ||
1da177e4 | 1717 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1718 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1719 | |
1720 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1721 | check_poison_obj(cachep, objp); |
40b44137 | 1722 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1723 | } |
1724 | if (cachep->flags & SLAB_RED_ZONE) { | |
1725 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1726 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1727 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1728 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1729 | } |
1da177e4 | 1730 | } |
12dd36fa | 1731 | } |
1da177e4 | 1732 | #else |
8456a648 JK |
1733 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1734 | struct page *page) | |
12dd36fa | 1735 | { |
12dd36fa | 1736 | } |
1da177e4 LT |
1737 | #endif |
1738 | ||
911851e6 RD |
1739 | /** |
1740 | * slab_destroy - destroy and release all objects in a slab | |
1741 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1742 | * @page: page pointer being destroyed |
911851e6 | 1743 | * |
8a7d9b43 WSH |
1744 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1745 | * Before calling the slab page must have been unlinked from the cache. The | |
1746 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1747 | */ |
8456a648 | 1748 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1749 | { |
7e007355 | 1750 | void *freelist; |
12dd36fa | 1751 | |
8456a648 JK |
1752 | freelist = page->freelist; |
1753 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1754 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1755 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1756 | else | |
0c3aa83e | 1757 | kmem_freepages(cachep, page); |
68126702 JK |
1758 | |
1759 | /* | |
8456a648 | 1760 | * From now on, we don't use freelist |
68126702 JK |
1761 | * although actual page can be freed in rcu context |
1762 | */ | |
1763 | if (OFF_SLAB(cachep)) | |
8456a648 | 1764 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1765 | } |
1766 | ||
97654dfa JK |
1767 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1768 | { | |
1769 | struct page *page, *n; | |
1770 | ||
1771 | list_for_each_entry_safe(page, n, list, lru) { | |
1772 | list_del(&page->lru); | |
1773 | slab_destroy(cachep, page); | |
1774 | } | |
1775 | } | |
1776 | ||
4d268eba | 1777 | /** |
a70773dd RD |
1778 | * calculate_slab_order - calculate size (page order) of slabs |
1779 | * @cachep: pointer to the cache that is being created | |
1780 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1781 | * @flags: slab allocation flags |
1782 | * | |
1783 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1784 | * |
1785 | * This could be made much more intelligent. For now, try to avoid using | |
1786 | * high order pages for slabs. When the gfp() functions are more friendly | |
1787 | * towards high-order requests, this should be changed. | |
1788 | */ | |
a737b3e2 | 1789 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1790 | size_t size, unsigned long flags) |
4d268eba PE |
1791 | { |
1792 | size_t left_over = 0; | |
9888e6fa | 1793 | int gfporder; |
4d268eba | 1794 | |
0aa817f0 | 1795 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1796 | unsigned int num; |
1797 | size_t remainder; | |
1798 | ||
70f75067 | 1799 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1800 | if (!num) |
1801 | continue; | |
9888e6fa | 1802 | |
f315e3fa JK |
1803 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1804 | if (num > SLAB_OBJ_MAX_NUM) | |
1805 | break; | |
1806 | ||
b1ab41c4 | 1807 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1808 | struct kmem_cache *freelist_cache; |
1809 | size_t freelist_size; | |
1810 | ||
1811 | freelist_size = num * sizeof(freelist_idx_t); | |
1812 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1813 | if (!freelist_cache) | |
1814 | continue; | |
1815 | ||
b1ab41c4 | 1816 | /* |
3217fd9b | 1817 | * Needed to avoid possible looping condition |
76b342bd | 1818 | * in cache_grow_begin() |
b1ab41c4 | 1819 | */ |
3217fd9b JK |
1820 | if (OFF_SLAB(freelist_cache)) |
1821 | continue; | |
b1ab41c4 | 1822 | |
3217fd9b JK |
1823 | /* check if off slab has enough benefit */ |
1824 | if (freelist_cache->size > cachep->size / 2) | |
1825 | continue; | |
b1ab41c4 | 1826 | } |
4d268eba | 1827 | |
9888e6fa | 1828 | /* Found something acceptable - save it away */ |
4d268eba | 1829 | cachep->num = num; |
9888e6fa | 1830 | cachep->gfporder = gfporder; |
4d268eba PE |
1831 | left_over = remainder; |
1832 | ||
f78bb8ad LT |
1833 | /* |
1834 | * A VFS-reclaimable slab tends to have most allocations | |
1835 | * as GFP_NOFS and we really don't want to have to be allocating | |
1836 | * higher-order pages when we are unable to shrink dcache. | |
1837 | */ | |
1838 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1839 | break; | |
1840 | ||
4d268eba PE |
1841 | /* |
1842 | * Large number of objects is good, but very large slabs are | |
1843 | * currently bad for the gfp()s. | |
1844 | */ | |
543585cc | 1845 | if (gfporder >= slab_max_order) |
4d268eba PE |
1846 | break; |
1847 | ||
9888e6fa LT |
1848 | /* |
1849 | * Acceptable internal fragmentation? | |
1850 | */ | |
a737b3e2 | 1851 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1852 | break; |
1853 | } | |
1854 | return left_over; | |
1855 | } | |
1856 | ||
bf0dea23 JK |
1857 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1858 | struct kmem_cache *cachep, int entries, int batchcount) | |
1859 | { | |
1860 | int cpu; | |
1861 | size_t size; | |
1862 | struct array_cache __percpu *cpu_cache; | |
1863 | ||
1864 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1865 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1866 | |
1867 | if (!cpu_cache) | |
1868 | return NULL; | |
1869 | ||
1870 | for_each_possible_cpu(cpu) { | |
1871 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1872 | entries, batchcount); | |
1873 | } | |
1874 | ||
1875 | return cpu_cache; | |
1876 | } | |
1877 | ||
83b519e8 | 1878 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1879 | { |
97d06609 | 1880 | if (slab_state >= FULL) |
83b519e8 | 1881 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1882 | |
bf0dea23 JK |
1883 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1884 | if (!cachep->cpu_cache) | |
1885 | return 1; | |
1886 | ||
97d06609 | 1887 | if (slab_state == DOWN) { |
bf0dea23 JK |
1888 | /* Creation of first cache (kmem_cache). */ |
1889 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1890 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1891 | /* For kmem_cache_node */ |
1892 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1893 | } else { |
bf0dea23 | 1894 | int node; |
f30cf7d1 | 1895 | |
bf0dea23 JK |
1896 | for_each_online_node(node) { |
1897 | cachep->node[node] = kmalloc_node( | |
1898 | sizeof(struct kmem_cache_node), gfp, node); | |
1899 | BUG_ON(!cachep->node[node]); | |
1900 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1901 | } |
1902 | } | |
bf0dea23 | 1903 | |
6a67368c | 1904 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1905 | jiffies + REAPTIMEOUT_NODE + |
1906 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1907 | |
1908 | cpu_cache_get(cachep)->avail = 0; | |
1909 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1910 | cpu_cache_get(cachep)->batchcount = 1; | |
1911 | cpu_cache_get(cachep)->touched = 0; | |
1912 | cachep->batchcount = 1; | |
1913 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1914 | return 0; |
f30cf7d1 PE |
1915 | } |
1916 | ||
12220dea JK |
1917 | unsigned long kmem_cache_flags(unsigned long object_size, |
1918 | unsigned long flags, const char *name, | |
1919 | void (*ctor)(void *)) | |
1920 | { | |
1921 | return flags; | |
1922 | } | |
1923 | ||
1924 | struct kmem_cache * | |
1925 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
1926 | unsigned long flags, void (*ctor)(void *)) | |
1927 | { | |
1928 | struct kmem_cache *cachep; | |
1929 | ||
1930 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1931 | if (cachep) { | |
1932 | cachep->refcount++; | |
1933 | ||
1934 | /* | |
1935 | * Adjust the object sizes so that we clear | |
1936 | * the complete object on kzalloc. | |
1937 | */ | |
1938 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1939 | } | |
1940 | return cachep; | |
1941 | } | |
1942 | ||
b03a017b JK |
1943 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
1944 | size_t size, unsigned long flags) | |
1945 | { | |
1946 | size_t left; | |
1947 | ||
1948 | cachep->num = 0; | |
1949 | ||
1950 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
1951 | return false; | |
1952 | ||
1953 | left = calculate_slab_order(cachep, size, | |
1954 | flags | CFLGS_OBJFREELIST_SLAB); | |
1955 | if (!cachep->num) | |
1956 | return false; | |
1957 | ||
1958 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1959 | return false; | |
1960 | ||
1961 | cachep->colour = left / cachep->colour_off; | |
1962 | ||
1963 | return true; | |
1964 | } | |
1965 | ||
158e319b JK |
1966 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
1967 | size_t size, unsigned long flags) | |
1968 | { | |
1969 | size_t left; | |
1970 | ||
1971 | cachep->num = 0; | |
1972 | ||
1973 | /* | |
3217fd9b JK |
1974 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1975 | * to avoid recursive calls into kmemleak. | |
158e319b | 1976 | */ |
158e319b JK |
1977 | if (flags & SLAB_NOLEAKTRACE) |
1978 | return false; | |
1979 | ||
1980 | /* | |
1981 | * Size is large, assume best to place the slab management obj | |
1982 | * off-slab (should allow better packing of objs). | |
1983 | */ | |
1984 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1985 | if (!cachep->num) | |
1986 | return false; | |
1987 | ||
1988 | /* | |
1989 | * If the slab has been placed off-slab, and we have enough space then | |
1990 | * move it on-slab. This is at the expense of any extra colouring. | |
1991 | */ | |
1992 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1993 | return false; | |
1994 | ||
1995 | cachep->colour = left / cachep->colour_off; | |
1996 | ||
1997 | return true; | |
1998 | } | |
1999 | ||
2000 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
2001 | size_t size, unsigned long flags) | |
2002 | { | |
2003 | size_t left; | |
2004 | ||
2005 | cachep->num = 0; | |
2006 | ||
2007 | left = calculate_slab_order(cachep, size, flags); | |
2008 | if (!cachep->num) | |
2009 | return false; | |
2010 | ||
2011 | cachep->colour = left / cachep->colour_off; | |
2012 | ||
2013 | return true; | |
2014 | } | |
2015 | ||
1da177e4 | 2016 | /** |
039363f3 | 2017 | * __kmem_cache_create - Create a cache. |
a755b76a | 2018 | * @cachep: cache management descriptor |
1da177e4 | 2019 | * @flags: SLAB flags |
1da177e4 LT |
2020 | * |
2021 | * Returns a ptr to the cache on success, NULL on failure. | |
2022 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2023 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2024 | * |
1da177e4 LT |
2025 | * The flags are |
2026 | * | |
2027 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2028 | * to catch references to uninitialised memory. | |
2029 | * | |
2030 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2031 | * for buffer overruns. | |
2032 | * | |
1da177e4 LT |
2033 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2034 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2035 | * as davem. | |
2036 | */ | |
278b1bb1 | 2037 | int |
8a13a4cc | 2038 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2039 | { |
d4a5fca5 | 2040 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2041 | gfp_t gfp; |
278b1bb1 | 2042 | int err; |
8a13a4cc | 2043 | size_t size = cachep->size; |
1da177e4 | 2044 | |
1da177e4 | 2045 | #if DEBUG |
1da177e4 LT |
2046 | #if FORCED_DEBUG |
2047 | /* | |
2048 | * Enable redzoning and last user accounting, except for caches with | |
2049 | * large objects, if the increased size would increase the object size | |
2050 | * above the next power of two: caches with object sizes just above a | |
2051 | * power of two have a significant amount of internal fragmentation. | |
2052 | */ | |
87a927c7 DW |
2053 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2054 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2055 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2056 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2057 | flags |= SLAB_POISON; | |
2058 | #endif | |
1da177e4 | 2059 | #endif |
1da177e4 | 2060 | |
a737b3e2 AM |
2061 | /* |
2062 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2063 | * unaligned accesses for some archs when redzoning is used, and makes |
2064 | * sure any on-slab bufctl's are also correctly aligned. | |
2065 | */ | |
b28a02de PE |
2066 | if (size & (BYTES_PER_WORD - 1)) { |
2067 | size += (BYTES_PER_WORD - 1); | |
2068 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2069 | } |
2070 | ||
87a927c7 DW |
2071 | if (flags & SLAB_RED_ZONE) { |
2072 | ralign = REDZONE_ALIGN; | |
2073 | /* If redzoning, ensure that the second redzone is suitably | |
2074 | * aligned, by adjusting the object size accordingly. */ | |
2075 | size += REDZONE_ALIGN - 1; | |
2076 | size &= ~(REDZONE_ALIGN - 1); | |
2077 | } | |
ca5f9703 | 2078 | |
a44b56d3 | 2079 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2080 | if (ralign < cachep->align) { |
2081 | ralign = cachep->align; | |
1da177e4 | 2082 | } |
3ff84a7f PE |
2083 | /* disable debug if necessary */ |
2084 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2085 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2086 | /* |
ca5f9703 | 2087 | * 4) Store it. |
1da177e4 | 2088 | */ |
8a13a4cc | 2089 | cachep->align = ralign; |
158e319b JK |
2090 | cachep->colour_off = cache_line_size(); |
2091 | /* Offset must be a multiple of the alignment. */ | |
2092 | if (cachep->colour_off < cachep->align) | |
2093 | cachep->colour_off = cachep->align; | |
1da177e4 | 2094 | |
83b519e8 PE |
2095 | if (slab_is_available()) |
2096 | gfp = GFP_KERNEL; | |
2097 | else | |
2098 | gfp = GFP_NOWAIT; | |
2099 | ||
1da177e4 | 2100 | #if DEBUG |
1da177e4 | 2101 | |
ca5f9703 PE |
2102 | /* |
2103 | * Both debugging options require word-alignment which is calculated | |
2104 | * into align above. | |
2105 | */ | |
1da177e4 | 2106 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2107 | /* add space for red zone words */ |
3ff84a7f PE |
2108 | cachep->obj_offset += sizeof(unsigned long long); |
2109 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2110 | } |
2111 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2112 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2113 | * the real object. But if the second red zone needs to be |
2114 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2115 | */ |
87a927c7 DW |
2116 | if (flags & SLAB_RED_ZONE) |
2117 | size += REDZONE_ALIGN; | |
2118 | else | |
2119 | size += BYTES_PER_WORD; | |
1da177e4 | 2120 | } |
832a15d2 JK |
2121 | #endif |
2122 | ||
7ed2f9e6 AP |
2123 | kasan_cache_create(cachep, &size, &flags); |
2124 | ||
832a15d2 JK |
2125 | size = ALIGN(size, cachep->align); |
2126 | /* | |
2127 | * We should restrict the number of objects in a slab to implement | |
2128 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2129 | */ | |
2130 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2131 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2132 | ||
2133 | #if DEBUG | |
03a2d2a3 JK |
2134 | /* |
2135 | * To activate debug pagealloc, off-slab management is necessary | |
2136 | * requirement. In early phase of initialization, small sized slab | |
2137 | * doesn't get initialized so it would not be possible. So, we need | |
2138 | * to check size >= 256. It guarantees that all necessary small | |
2139 | * sized slab is initialized in current slab initialization sequence. | |
2140 | */ | |
40323278 | 2141 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2142 | size >= 256 && cachep->object_size > cache_line_size()) { |
2143 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2144 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2145 | ||
2146 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2147 | flags |= CFLGS_OFF_SLAB; | |
2148 | cachep->obj_offset += tmp_size - size; | |
2149 | size = tmp_size; | |
2150 | goto done; | |
2151 | } | |
2152 | } | |
1da177e4 | 2153 | } |
1da177e4 LT |
2154 | #endif |
2155 | ||
b03a017b JK |
2156 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2157 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2158 | goto done; | |
2159 | } | |
2160 | ||
158e319b | 2161 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2162 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2163 | goto done; |
832a15d2 | 2164 | } |
1da177e4 | 2165 | |
158e319b JK |
2166 | if (set_on_slab_cache(cachep, size, flags)) |
2167 | goto done; | |
1da177e4 | 2168 | |
158e319b | 2169 | return -E2BIG; |
1da177e4 | 2170 | |
158e319b JK |
2171 | done: |
2172 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2173 | cachep->flags = flags; |
a57a4988 | 2174 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2175 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2176 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2177 | cachep->size = size; |
6a2d7a95 | 2178 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2179 | |
40b44137 JK |
2180 | #if DEBUG |
2181 | /* | |
2182 | * If we're going to use the generic kernel_map_pages() | |
2183 | * poisoning, then it's going to smash the contents of | |
2184 | * the redzone and userword anyhow, so switch them off. | |
2185 | */ | |
2186 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2187 | (cachep->flags & SLAB_POISON) && | |
2188 | is_debug_pagealloc_cache(cachep)) | |
2189 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2190 | #endif | |
2191 | ||
2192 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2193 | cachep->freelist_cache = |
2194 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2195 | } |
1da177e4 | 2196 | |
278b1bb1 CL |
2197 | err = setup_cpu_cache(cachep, gfp); |
2198 | if (err) { | |
52b4b950 | 2199 | __kmem_cache_release(cachep); |
278b1bb1 | 2200 | return err; |
2ed3a4ef | 2201 | } |
1da177e4 | 2202 | |
278b1bb1 | 2203 | return 0; |
1da177e4 | 2204 | } |
1da177e4 LT |
2205 | |
2206 | #if DEBUG | |
2207 | static void check_irq_off(void) | |
2208 | { | |
2209 | BUG_ON(!irqs_disabled()); | |
2210 | } | |
2211 | ||
2212 | static void check_irq_on(void) | |
2213 | { | |
2214 | BUG_ON(irqs_disabled()); | |
2215 | } | |
2216 | ||
18726ca8 JK |
2217 | static void check_mutex_acquired(void) |
2218 | { | |
2219 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2220 | } | |
2221 | ||
343e0d7a | 2222 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2223 | { |
2224 | #ifdef CONFIG_SMP | |
2225 | check_irq_off(); | |
18bf8541 | 2226 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2227 | #endif |
2228 | } | |
e498be7d | 2229 | |
343e0d7a | 2230 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2231 | { |
2232 | #ifdef CONFIG_SMP | |
2233 | check_irq_off(); | |
18bf8541 | 2234 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2235 | #endif |
2236 | } | |
2237 | ||
1da177e4 LT |
2238 | #else |
2239 | #define check_irq_off() do { } while(0) | |
2240 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2241 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2242 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2243 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2244 | #endif |
2245 | ||
18726ca8 JK |
2246 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2247 | int node, bool free_all, struct list_head *list) | |
2248 | { | |
2249 | int tofree; | |
2250 | ||
2251 | if (!ac || !ac->avail) | |
2252 | return; | |
2253 | ||
2254 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2255 | if (tofree > ac->avail) | |
2256 | tofree = (ac->avail + 1) / 2; | |
2257 | ||
2258 | free_block(cachep, ac->entry, tofree, node, list); | |
2259 | ac->avail -= tofree; | |
2260 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2261 | } | |
aab2207c | 2262 | |
1da177e4 LT |
2263 | static void do_drain(void *arg) |
2264 | { | |
a737b3e2 | 2265 | struct kmem_cache *cachep = arg; |
1da177e4 | 2266 | struct array_cache *ac; |
7d6e6d09 | 2267 | int node = numa_mem_id(); |
18bf8541 | 2268 | struct kmem_cache_node *n; |
97654dfa | 2269 | LIST_HEAD(list); |
1da177e4 LT |
2270 | |
2271 | check_irq_off(); | |
9a2dba4b | 2272 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2273 | n = get_node(cachep, node); |
2274 | spin_lock(&n->list_lock); | |
97654dfa | 2275 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2276 | spin_unlock(&n->list_lock); |
97654dfa | 2277 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2278 | ac->avail = 0; |
2279 | } | |
2280 | ||
343e0d7a | 2281 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2282 | { |
ce8eb6c4 | 2283 | struct kmem_cache_node *n; |
e498be7d | 2284 | int node; |
18726ca8 | 2285 | LIST_HEAD(list); |
e498be7d | 2286 | |
15c8b6c1 | 2287 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2288 | check_irq_on(); |
18bf8541 CL |
2289 | for_each_kmem_cache_node(cachep, node, n) |
2290 | if (n->alien) | |
ce8eb6c4 | 2291 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2292 | |
18726ca8 JK |
2293 | for_each_kmem_cache_node(cachep, node, n) { |
2294 | spin_lock_irq(&n->list_lock); | |
2295 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2296 | spin_unlock_irq(&n->list_lock); | |
2297 | ||
2298 | slabs_destroy(cachep, &list); | |
2299 | } | |
1da177e4 LT |
2300 | } |
2301 | ||
ed11d9eb CL |
2302 | /* |
2303 | * Remove slabs from the list of free slabs. | |
2304 | * Specify the number of slabs to drain in tofree. | |
2305 | * | |
2306 | * Returns the actual number of slabs released. | |
2307 | */ | |
2308 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2309 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2310 | { |
ed11d9eb CL |
2311 | struct list_head *p; |
2312 | int nr_freed; | |
8456a648 | 2313 | struct page *page; |
1da177e4 | 2314 | |
ed11d9eb | 2315 | nr_freed = 0; |
ce8eb6c4 | 2316 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2317 | |
ce8eb6c4 CL |
2318 | spin_lock_irq(&n->list_lock); |
2319 | p = n->slabs_free.prev; | |
2320 | if (p == &n->slabs_free) { | |
2321 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2322 | goto out; |
2323 | } | |
1da177e4 | 2324 | |
8456a648 | 2325 | page = list_entry(p, struct page, lru); |
8456a648 | 2326 | list_del(&page->lru); |
ed11d9eb CL |
2327 | /* |
2328 | * Safe to drop the lock. The slab is no longer linked | |
2329 | * to the cache. | |
2330 | */ | |
ce8eb6c4 CL |
2331 | n->free_objects -= cache->num; |
2332 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2333 | slab_destroy(cache, page); |
ed11d9eb | 2334 | nr_freed++; |
1da177e4 | 2335 | } |
ed11d9eb CL |
2336 | out: |
2337 | return nr_freed; | |
1da177e4 LT |
2338 | } |
2339 | ||
d6e0b7fa | 2340 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2341 | { |
18bf8541 CL |
2342 | int ret = 0; |
2343 | int node; | |
ce8eb6c4 | 2344 | struct kmem_cache_node *n; |
e498be7d CL |
2345 | |
2346 | drain_cpu_caches(cachep); | |
2347 | ||
2348 | check_irq_on(); | |
18bf8541 | 2349 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2350 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2351 | |
ce8eb6c4 CL |
2352 | ret += !list_empty(&n->slabs_full) || |
2353 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2354 | } |
2355 | return (ret ? 1 : 0); | |
2356 | } | |
2357 | ||
945cf2b6 | 2358 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2359 | { |
2360 | return __kmem_cache_shrink(cachep, false); | |
2361 | } | |
2362 | ||
2363 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2364 | { |
12c3667f | 2365 | int i; |
ce8eb6c4 | 2366 | struct kmem_cache_node *n; |
1da177e4 | 2367 | |
bf0dea23 | 2368 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2369 | |
ce8eb6c4 | 2370 | /* NUMA: free the node structures */ |
18bf8541 CL |
2371 | for_each_kmem_cache_node(cachep, i, n) { |
2372 | kfree(n->shared); | |
2373 | free_alien_cache(n->alien); | |
2374 | kfree(n); | |
2375 | cachep->node[i] = NULL; | |
12c3667f | 2376 | } |
1da177e4 | 2377 | } |
1da177e4 | 2378 | |
e5ac9c5a RT |
2379 | /* |
2380 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2381 | * |
2382 | * For a slab cache when the slab descriptor is off-slab, the | |
2383 | * slab descriptor can't come from the same cache which is being created, | |
2384 | * Because if it is the case, that means we defer the creation of | |
2385 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2386 | * And we eventually call down to __kmem_cache_create(), which | |
2387 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2388 | * This is a "chicken-and-egg" problem. | |
2389 | * | |
2390 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2391 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2392 | */ |
7e007355 | 2393 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2394 | struct page *page, int colour_off, |
2395 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2396 | { |
7e007355 | 2397 | void *freelist; |
0c3aa83e | 2398 | void *addr = page_address(page); |
b28a02de | 2399 | |
2e6b3602 JK |
2400 | page->s_mem = addr + colour_off; |
2401 | page->active = 0; | |
2402 | ||
b03a017b JK |
2403 | if (OBJFREELIST_SLAB(cachep)) |
2404 | freelist = NULL; | |
2405 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2406 | /* Slab management obj is off-slab. */ |
8456a648 | 2407 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2408 | local_flags, nodeid); |
8456a648 | 2409 | if (!freelist) |
1da177e4 LT |
2410 | return NULL; |
2411 | } else { | |
2e6b3602 JK |
2412 | /* We will use last bytes at the slab for freelist */ |
2413 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2414 | cachep->freelist_size; | |
1da177e4 | 2415 | } |
2e6b3602 | 2416 | |
8456a648 | 2417 | return freelist; |
1da177e4 LT |
2418 | } |
2419 | ||
7cc68973 | 2420 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2421 | { |
a41adfaa | 2422 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2423 | } |
2424 | ||
2425 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2426 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2427 | { |
a41adfaa | 2428 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2429 | } |
2430 | ||
10b2e9e8 | 2431 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2432 | { |
10b2e9e8 | 2433 | #if DEBUG |
1da177e4 LT |
2434 | int i; |
2435 | ||
2436 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2437 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2438 | |
1da177e4 LT |
2439 | if (cachep->flags & SLAB_STORE_USER) |
2440 | *dbg_userword(cachep, objp) = NULL; | |
2441 | ||
2442 | if (cachep->flags & SLAB_RED_ZONE) { | |
2443 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2444 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2445 | } | |
2446 | /* | |
a737b3e2 AM |
2447 | * Constructors are not allowed to allocate memory from the same |
2448 | * cache which they are a constructor for. Otherwise, deadlock. | |
2449 | * They must also be threaded. | |
1da177e4 | 2450 | */ |
7ed2f9e6 AP |
2451 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2452 | kasan_unpoison_object_data(cachep, | |
2453 | objp + obj_offset(cachep)); | |
51cc5068 | 2454 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2455 | kasan_poison_object_data( |
2456 | cachep, objp + obj_offset(cachep)); | |
2457 | } | |
1da177e4 LT |
2458 | |
2459 | if (cachep->flags & SLAB_RED_ZONE) { | |
2460 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2461 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2462 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2463 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2464 | } |
40b44137 JK |
2465 | /* need to poison the objs? */ |
2466 | if (cachep->flags & SLAB_POISON) { | |
2467 | poison_obj(cachep, objp, POISON_FREE); | |
2468 | slab_kernel_map(cachep, objp, 0, 0); | |
2469 | } | |
10b2e9e8 | 2470 | } |
1da177e4 | 2471 | #endif |
10b2e9e8 JK |
2472 | } |
2473 | ||
2474 | static void cache_init_objs(struct kmem_cache *cachep, | |
2475 | struct page *page) | |
2476 | { | |
2477 | int i; | |
7ed2f9e6 | 2478 | void *objp; |
10b2e9e8 JK |
2479 | |
2480 | cache_init_objs_debug(cachep, page); | |
2481 | ||
b03a017b JK |
2482 | if (OBJFREELIST_SLAB(cachep)) { |
2483 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + | |
2484 | obj_offset(cachep); | |
2485 | } | |
2486 | ||
10b2e9e8 JK |
2487 | for (i = 0; i < cachep->num; i++) { |
2488 | /* constructor could break poison info */ | |
7ed2f9e6 AP |
2489 | if (DEBUG == 0 && cachep->ctor) { |
2490 | objp = index_to_obj(cachep, page, i); | |
2491 | kasan_unpoison_object_data(cachep, objp); | |
2492 | cachep->ctor(objp); | |
2493 | kasan_poison_object_data(cachep, objp); | |
2494 | } | |
10b2e9e8 | 2495 | |
e5c58dfd | 2496 | set_free_obj(page, i, i); |
1da177e4 | 2497 | } |
1da177e4 LT |
2498 | } |
2499 | ||
343e0d7a | 2500 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2501 | { |
4b51d669 CL |
2502 | if (CONFIG_ZONE_DMA_FLAG) { |
2503 | if (flags & GFP_DMA) | |
a618e89f | 2504 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2505 | else |
a618e89f | 2506 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2507 | } |
1da177e4 LT |
2508 | } |
2509 | ||
260b61dd | 2510 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2511 | { |
b1cb0982 | 2512 | void *objp; |
78d382d7 | 2513 | |
e5c58dfd | 2514 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2515 | page->active++; |
78d382d7 | 2516 | |
d31676df JK |
2517 | #if DEBUG |
2518 | if (cachep->flags & SLAB_STORE_USER) | |
2519 | set_store_user_dirty(cachep); | |
2520 | #endif | |
2521 | ||
78d382d7 MD |
2522 | return objp; |
2523 | } | |
2524 | ||
260b61dd JK |
2525 | static void slab_put_obj(struct kmem_cache *cachep, |
2526 | struct page *page, void *objp) | |
78d382d7 | 2527 | { |
8456a648 | 2528 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2529 | #if DEBUG |
16025177 | 2530 | unsigned int i; |
b1cb0982 | 2531 | |
b1cb0982 | 2532 | /* Verify double free bug */ |
8456a648 | 2533 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2534 | if (get_free_obj(page, i) == objnr) { |
1170532b | 2535 | pr_err("slab: double free detected in cache '%s', objp %p\n", |
756a025f | 2536 | cachep->name, objp); |
b1cb0982 JK |
2537 | BUG(); |
2538 | } | |
78d382d7 MD |
2539 | } |
2540 | #endif | |
8456a648 | 2541 | page->active--; |
b03a017b JK |
2542 | if (!page->freelist) |
2543 | page->freelist = objp + obj_offset(cachep); | |
2544 | ||
e5c58dfd | 2545 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2546 | } |
2547 | ||
4776874f PE |
2548 | /* |
2549 | * Map pages beginning at addr to the given cache and slab. This is required | |
2550 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2551 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2552 | */ |
8456a648 | 2553 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2554 | void *freelist) |
1da177e4 | 2555 | { |
a57a4988 | 2556 | page->slab_cache = cache; |
8456a648 | 2557 | page->freelist = freelist; |
1da177e4 LT |
2558 | } |
2559 | ||
2560 | /* | |
2561 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2562 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2563 | */ | |
76b342bd JK |
2564 | static struct page *cache_grow_begin(struct kmem_cache *cachep, |
2565 | gfp_t flags, int nodeid) | |
1da177e4 | 2566 | { |
7e007355 | 2567 | void *freelist; |
b28a02de PE |
2568 | size_t offset; |
2569 | gfp_t local_flags; | |
511e3a05 | 2570 | int page_node; |
ce8eb6c4 | 2571 | struct kmem_cache_node *n; |
511e3a05 | 2572 | struct page *page; |
1da177e4 | 2573 | |
a737b3e2 AM |
2574 | /* |
2575 | * Be lazy and only check for valid flags here, keeping it out of the | |
2576 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2577 | */ |
c871ac4e AM |
2578 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2579 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2580 | BUG(); | |
2581 | } | |
6cb06229 | 2582 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2583 | |
1da177e4 | 2584 | check_irq_off(); |
d0164adc | 2585 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2586 | local_irq_enable(); |
2587 | ||
2588 | /* | |
2589 | * The test for missing atomic flag is performed here, rather than | |
2590 | * the more obvious place, simply to reduce the critical path length | |
2591 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2592 | * will eventually be caught here (where it matters). | |
2593 | */ | |
2594 | kmem_flagcheck(cachep, flags); | |
2595 | ||
a737b3e2 AM |
2596 | /* |
2597 | * Get mem for the objs. Attempt to allocate a physical page from | |
2598 | * 'nodeid'. | |
e498be7d | 2599 | */ |
511e3a05 | 2600 | page = kmem_getpages(cachep, local_flags, nodeid); |
0c3aa83e | 2601 | if (!page) |
1da177e4 LT |
2602 | goto failed; |
2603 | ||
511e3a05 JK |
2604 | page_node = page_to_nid(page); |
2605 | n = get_node(cachep, page_node); | |
03d1d43a JK |
2606 | |
2607 | /* Get colour for the slab, and cal the next value. */ | |
2608 | n->colour_next++; | |
2609 | if (n->colour_next >= cachep->colour) | |
2610 | n->colour_next = 0; | |
2611 | ||
2612 | offset = n->colour_next; | |
2613 | if (offset >= cachep->colour) | |
2614 | offset = 0; | |
2615 | ||
2616 | offset *= cachep->colour_off; | |
2617 | ||
1da177e4 | 2618 | /* Get slab management. */ |
8456a648 | 2619 | freelist = alloc_slabmgmt(cachep, page, offset, |
511e3a05 | 2620 | local_flags & ~GFP_CONSTRAINT_MASK, page_node); |
b03a017b | 2621 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2622 | goto opps1; |
2623 | ||
8456a648 | 2624 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2625 | |
7ed2f9e6 | 2626 | kasan_poison_slab(page); |
8456a648 | 2627 | cache_init_objs(cachep, page); |
1da177e4 | 2628 | |
d0164adc | 2629 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2630 | local_irq_disable(); |
1da177e4 | 2631 | |
76b342bd JK |
2632 | return page; |
2633 | ||
a737b3e2 | 2634 | opps1: |
0c3aa83e | 2635 | kmem_freepages(cachep, page); |
a737b3e2 | 2636 | failed: |
d0164adc | 2637 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2638 | local_irq_disable(); |
76b342bd JK |
2639 | return NULL; |
2640 | } | |
2641 | ||
2642 | static void cache_grow_end(struct kmem_cache *cachep, struct page *page) | |
2643 | { | |
2644 | struct kmem_cache_node *n; | |
2645 | void *list = NULL; | |
2646 | ||
2647 | check_irq_off(); | |
2648 | ||
2649 | if (!page) | |
2650 | return; | |
2651 | ||
2652 | INIT_LIST_HEAD(&page->lru); | |
2653 | n = get_node(cachep, page_to_nid(page)); | |
2654 | ||
2655 | spin_lock(&n->list_lock); | |
2656 | if (!page->active) | |
2657 | list_add_tail(&page->lru, &(n->slabs_free)); | |
2658 | else | |
2659 | fixup_slab_list(cachep, n, page, &list); | |
2660 | STATS_INC_GROWN(cachep); | |
2661 | n->free_objects += cachep->num - page->active; | |
2662 | spin_unlock(&n->list_lock); | |
2663 | ||
2664 | fixup_objfreelist_debug(cachep, &list); | |
1da177e4 LT |
2665 | } |
2666 | ||
2667 | #if DEBUG | |
2668 | ||
2669 | /* | |
2670 | * Perform extra freeing checks: | |
2671 | * - detect bad pointers. | |
2672 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2673 | */ |
2674 | static void kfree_debugcheck(const void *objp) | |
2675 | { | |
1da177e4 | 2676 | if (!virt_addr_valid(objp)) { |
1170532b | 2677 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2678 | (unsigned long)objp); |
2679 | BUG(); | |
1da177e4 | 2680 | } |
1da177e4 LT |
2681 | } |
2682 | ||
58ce1fd5 PE |
2683 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2684 | { | |
b46b8f19 | 2685 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2686 | |
2687 | redzone1 = *dbg_redzone1(cache, obj); | |
2688 | redzone2 = *dbg_redzone2(cache, obj); | |
2689 | ||
2690 | /* | |
2691 | * Redzone is ok. | |
2692 | */ | |
2693 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2694 | return; | |
2695 | ||
2696 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2697 | slab_error(cache, "double free detected"); | |
2698 | else | |
2699 | slab_error(cache, "memory outside object was overwritten"); | |
2700 | ||
1170532b JP |
2701 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2702 | obj, redzone1, redzone2); | |
58ce1fd5 PE |
2703 | } |
2704 | ||
343e0d7a | 2705 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2706 | unsigned long caller) |
1da177e4 | 2707 | { |
1da177e4 | 2708 | unsigned int objnr; |
8456a648 | 2709 | struct page *page; |
1da177e4 | 2710 | |
80cbd911 MW |
2711 | BUG_ON(virt_to_cache(objp) != cachep); |
2712 | ||
3dafccf2 | 2713 | objp -= obj_offset(cachep); |
1da177e4 | 2714 | kfree_debugcheck(objp); |
b49af68f | 2715 | page = virt_to_head_page(objp); |
1da177e4 | 2716 | |
1da177e4 | 2717 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2718 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2719 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2720 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2721 | } | |
d31676df JK |
2722 | if (cachep->flags & SLAB_STORE_USER) { |
2723 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2724 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2725 | } |
1da177e4 | 2726 | |
8456a648 | 2727 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2728 | |
2729 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2730 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2731 | |
1da177e4 | 2732 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2733 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2734 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2735 | } |
2736 | return objp; | |
2737 | } | |
2738 | ||
1da177e4 LT |
2739 | #else |
2740 | #define kfree_debugcheck(x) do { } while(0) | |
2741 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2742 | #endif |
2743 | ||
b03a017b JK |
2744 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2745 | void **list) | |
2746 | { | |
2747 | #if DEBUG | |
2748 | void *next = *list; | |
2749 | void *objp; | |
2750 | ||
2751 | while (next) { | |
2752 | objp = next - obj_offset(cachep); | |
2753 | next = *(void **)next; | |
2754 | poison_obj(cachep, objp, POISON_FREE); | |
2755 | } | |
2756 | #endif | |
2757 | } | |
2758 | ||
d8410234 | 2759 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2760 | struct kmem_cache_node *n, struct page *page, |
2761 | void **list) | |
d8410234 JK |
2762 | { |
2763 | /* move slabp to correct slabp list: */ | |
2764 | list_del(&page->lru); | |
b03a017b | 2765 | if (page->active == cachep->num) { |
d8410234 | 2766 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2767 | if (OBJFREELIST_SLAB(cachep)) { |
2768 | #if DEBUG | |
2769 | /* Poisoning will be done without holding the lock */ | |
2770 | if (cachep->flags & SLAB_POISON) { | |
2771 | void **objp = page->freelist; | |
2772 | ||
2773 | *objp = *list; | |
2774 | *list = objp; | |
2775 | } | |
2776 | #endif | |
2777 | page->freelist = NULL; | |
2778 | } | |
2779 | } else | |
d8410234 JK |
2780 | list_add(&page->lru, &n->slabs_partial); |
2781 | } | |
2782 | ||
f68f8ddd JK |
2783 | /* Try to find non-pfmemalloc slab if needed */ |
2784 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
2785 | struct page *page, bool pfmemalloc) | |
2786 | { | |
2787 | if (!page) | |
2788 | return NULL; | |
2789 | ||
2790 | if (pfmemalloc) | |
2791 | return page; | |
2792 | ||
2793 | if (!PageSlabPfmemalloc(page)) | |
2794 | return page; | |
2795 | ||
2796 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2797 | if (n->free_objects > n->free_limit) { | |
2798 | ClearPageSlabPfmemalloc(page); | |
2799 | return page; | |
2800 | } | |
2801 | ||
2802 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2803 | list_del(&page->lru); | |
2804 | if (!page->active) | |
2805 | list_add_tail(&page->lru, &n->slabs_free); | |
2806 | else | |
2807 | list_add_tail(&page->lru, &n->slabs_partial); | |
2808 | ||
2809 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2810 | if (!PageSlabPfmemalloc(page)) | |
2811 | return page; | |
2812 | } | |
2813 | ||
2814 | list_for_each_entry(page, &n->slabs_free, lru) { | |
2815 | if (!PageSlabPfmemalloc(page)) | |
2816 | return page; | |
2817 | } | |
2818 | ||
2819 | return NULL; | |
2820 | } | |
2821 | ||
2822 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2823 | { |
2824 | struct page *page; | |
2825 | ||
2826 | page = list_first_entry_or_null(&n->slabs_partial, | |
2827 | struct page, lru); | |
2828 | if (!page) { | |
2829 | n->free_touched = 1; | |
2830 | page = list_first_entry_or_null(&n->slabs_free, | |
2831 | struct page, lru); | |
2832 | } | |
2833 | ||
f68f8ddd JK |
2834 | if (sk_memalloc_socks()) |
2835 | return get_valid_first_slab(n, page, pfmemalloc); | |
2836 | ||
7aa0d227 GT |
2837 | return page; |
2838 | } | |
2839 | ||
f68f8ddd JK |
2840 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2841 | struct kmem_cache_node *n, gfp_t flags) | |
2842 | { | |
2843 | struct page *page; | |
2844 | void *obj; | |
2845 | void *list = NULL; | |
2846 | ||
2847 | if (!gfp_pfmemalloc_allowed(flags)) | |
2848 | return NULL; | |
2849 | ||
2850 | spin_lock(&n->list_lock); | |
2851 | page = get_first_slab(n, true); | |
2852 | if (!page) { | |
2853 | spin_unlock(&n->list_lock); | |
2854 | return NULL; | |
2855 | } | |
2856 | ||
2857 | obj = slab_get_obj(cachep, page); | |
2858 | n->free_objects--; | |
2859 | ||
2860 | fixup_slab_list(cachep, n, page, &list); | |
2861 | ||
2862 | spin_unlock(&n->list_lock); | |
2863 | fixup_objfreelist_debug(cachep, &list); | |
2864 | ||
2865 | return obj; | |
2866 | } | |
2867 | ||
213b4695 JK |
2868 | /* |
2869 | * Slab list should be fixed up by fixup_slab_list() for existing slab | |
2870 | * or cache_grow_end() for new slab | |
2871 | */ | |
2872 | static __always_inline int alloc_block(struct kmem_cache *cachep, | |
2873 | struct array_cache *ac, struct page *page, int batchcount) | |
2874 | { | |
2875 | /* | |
2876 | * There must be at least one object available for | |
2877 | * allocation. | |
2878 | */ | |
2879 | BUG_ON(page->active >= cachep->num); | |
2880 | ||
2881 | while (page->active < cachep->num && batchcount--) { | |
2882 | STATS_INC_ALLOCED(cachep); | |
2883 | STATS_INC_ACTIVE(cachep); | |
2884 | STATS_SET_HIGH(cachep); | |
2885 | ||
2886 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); | |
2887 | } | |
2888 | ||
2889 | return batchcount; | |
2890 | } | |
2891 | ||
f68f8ddd | 2892 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2893 | { |
2894 | int batchcount; | |
ce8eb6c4 | 2895 | struct kmem_cache_node *n; |
1da177e4 | 2896 | struct array_cache *ac; |
1ca4cb24 | 2897 | int node; |
b03a017b | 2898 | void *list = NULL; |
76b342bd | 2899 | struct page *page; |
1ca4cb24 | 2900 | |
1da177e4 | 2901 | check_irq_off(); |
7d6e6d09 | 2902 | node = numa_mem_id(); |
f68f8ddd | 2903 | |
9a2dba4b | 2904 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2905 | batchcount = ac->batchcount; |
2906 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2907 | /* |
2908 | * If there was little recent activity on this cache, then | |
2909 | * perform only a partial refill. Otherwise we could generate | |
2910 | * refill bouncing. | |
1da177e4 LT |
2911 | */ |
2912 | batchcount = BATCHREFILL_LIMIT; | |
2913 | } | |
18bf8541 | 2914 | n = get_node(cachep, node); |
e498be7d | 2915 | |
ce8eb6c4 CL |
2916 | BUG_ON(ac->avail > 0 || !n); |
2917 | spin_lock(&n->list_lock); | |
1da177e4 | 2918 | |
3ded175a | 2919 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2920 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2921 | n->shared->touched = 1; | |
3ded175a | 2922 | goto alloc_done; |
44b57f1c | 2923 | } |
3ded175a | 2924 | |
1da177e4 | 2925 | while (batchcount > 0) { |
1da177e4 | 2926 | /* Get slab alloc is to come from. */ |
f68f8ddd | 2927 | page = get_first_slab(n, false); |
7aa0d227 GT |
2928 | if (!page) |
2929 | goto must_grow; | |
1da177e4 | 2930 | |
1da177e4 | 2931 | check_spinlock_acquired(cachep); |
714b8171 | 2932 | |
213b4695 | 2933 | batchcount = alloc_block(cachep, ac, page, batchcount); |
b03a017b | 2934 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2935 | } |
2936 | ||
a737b3e2 | 2937 | must_grow: |
ce8eb6c4 | 2938 | n->free_objects -= ac->avail; |
a737b3e2 | 2939 | alloc_done: |
ce8eb6c4 | 2940 | spin_unlock(&n->list_lock); |
b03a017b | 2941 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 LT |
2942 | |
2943 | if (unlikely(!ac->avail)) { | |
f68f8ddd JK |
2944 | /* Check if we can use obj in pfmemalloc slab */ |
2945 | if (sk_memalloc_socks()) { | |
2946 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
2947 | ||
2948 | if (obj) | |
2949 | return obj; | |
2950 | } | |
2951 | ||
76b342bd | 2952 | page = cache_grow_begin(cachep, gfp_exact_node(flags), node); |
e498be7d | 2953 | |
76b342bd JK |
2954 | /* |
2955 | * cache_grow_begin() can reenable interrupts, | |
2956 | * then ac could change. | |
2957 | */ | |
9a2dba4b | 2958 | ac = cpu_cache_get(cachep); |
213b4695 JK |
2959 | if (!ac->avail && page) |
2960 | alloc_block(cachep, ac, page, batchcount); | |
2961 | cache_grow_end(cachep, page); | |
072bb0aa | 2962 | |
213b4695 | 2963 | if (!ac->avail) |
1da177e4 | 2964 | return NULL; |
1da177e4 LT |
2965 | } |
2966 | ac->touched = 1; | |
072bb0aa | 2967 | |
f68f8ddd | 2968 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2969 | } |
2970 | ||
a737b3e2 AM |
2971 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2972 | gfp_t flags) | |
1da177e4 | 2973 | { |
d0164adc | 2974 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2975 | #if DEBUG |
2976 | kmem_flagcheck(cachep, flags); | |
2977 | #endif | |
2978 | } | |
2979 | ||
2980 | #if DEBUG | |
a737b3e2 | 2981 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2982 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2983 | { |
b28a02de | 2984 | if (!objp) |
1da177e4 | 2985 | return objp; |
b28a02de | 2986 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2987 | check_poison_obj(cachep, objp); |
40b44137 | 2988 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2989 | poison_obj(cachep, objp, POISON_INUSE); |
2990 | } | |
2991 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2992 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2993 | |
2994 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2995 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2996 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 2997 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
1170532b JP |
2998 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2999 | objp, *dbg_redzone1(cachep, objp), | |
3000 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3001 | } |
3002 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3003 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3004 | } | |
03787301 | 3005 | |
3dafccf2 | 3006 | objp += obj_offset(cachep); |
4f104934 | 3007 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3008 | cachep->ctor(objp); |
7ea466f2 TH |
3009 | if (ARCH_SLAB_MINALIGN && |
3010 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
1170532b | 3011 | pr_err("0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3012 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3013 | } |
1da177e4 LT |
3014 | return objp; |
3015 | } | |
3016 | #else | |
3017 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3018 | #endif | |
3019 | ||
343e0d7a | 3020 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3021 | { |
b28a02de | 3022 | void *objp; |
1da177e4 LT |
3023 | struct array_cache *ac; |
3024 | ||
5c382300 | 3025 | check_irq_off(); |
8a8b6502 | 3026 | |
9a2dba4b | 3027 | ac = cpu_cache_get(cachep); |
1da177e4 | 3028 | if (likely(ac->avail)) { |
1da177e4 | 3029 | ac->touched = 1; |
f68f8ddd | 3030 | objp = ac->entry[--ac->avail]; |
072bb0aa | 3031 | |
f68f8ddd JK |
3032 | STATS_INC_ALLOCHIT(cachep); |
3033 | goto out; | |
1da177e4 | 3034 | } |
072bb0aa MG |
3035 | |
3036 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 3037 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
3038 | /* |
3039 | * the 'ac' may be updated by cache_alloc_refill(), | |
3040 | * and kmemleak_erase() requires its correct value. | |
3041 | */ | |
3042 | ac = cpu_cache_get(cachep); | |
3043 | ||
3044 | out: | |
d5cff635 CM |
3045 | /* |
3046 | * To avoid a false negative, if an object that is in one of the | |
3047 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3048 | * treat the array pointers as a reference to the object. | |
3049 | */ | |
f3d8b53a O |
3050 | if (objp) |
3051 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3052 | return objp; |
3053 | } | |
3054 | ||
e498be7d | 3055 | #ifdef CONFIG_NUMA |
c61afb18 | 3056 | /* |
2ad654bc | 3057 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3058 | * |
3059 | * If we are in_interrupt, then process context, including cpusets and | |
3060 | * mempolicy, may not apply and should not be used for allocation policy. | |
3061 | */ | |
3062 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3063 | { | |
3064 | int nid_alloc, nid_here; | |
3065 | ||
765c4507 | 3066 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3067 | return NULL; |
7d6e6d09 | 3068 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3069 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3070 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3071 | else if (current->mempolicy) |
2a389610 | 3072 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3073 | if (nid_alloc != nid_here) |
8b98c169 | 3074 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3075 | return NULL; |
3076 | } | |
3077 | ||
765c4507 CL |
3078 | /* |
3079 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3080 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3081 | * available node for available objects. If that fails then we |
3c517a61 CL |
3082 | * perform an allocation without specifying a node. This allows the page |
3083 | * allocator to do its reclaim / fallback magic. We then insert the | |
3084 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3085 | */ |
8c8cc2c1 | 3086 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3087 | { |
8c8cc2c1 | 3088 | struct zonelist *zonelist; |
dd1a239f | 3089 | struct zoneref *z; |
54a6eb5c MG |
3090 | struct zone *zone; |
3091 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3092 | void *obj = NULL; |
76b342bd | 3093 | struct page *page; |
3c517a61 | 3094 | int nid; |
cc9a6c87 | 3095 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3096 | |
3097 | if (flags & __GFP_THISNODE) | |
3098 | return NULL; | |
3099 | ||
cc9a6c87 | 3100 | retry_cpuset: |
d26914d1 | 3101 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3102 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3103 | |
3c517a61 CL |
3104 | retry: |
3105 | /* | |
3106 | * Look through allowed nodes for objects available | |
3107 | * from existing per node queues. | |
3108 | */ | |
54a6eb5c MG |
3109 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3110 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3111 | |
061d7074 | 3112 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3113 | get_node(cache, nid) && |
3114 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3115 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3116 | gfp_exact_node(flags), nid); |
481c5346 CL |
3117 | if (obj) |
3118 | break; | |
3119 | } | |
3c517a61 CL |
3120 | } |
3121 | ||
cfce6604 | 3122 | if (!obj) { |
3c517a61 CL |
3123 | /* |
3124 | * This allocation will be performed within the constraints | |
3125 | * of the current cpuset / memory policy requirements. | |
3126 | * We may trigger various forms of reclaim on the allowed | |
3127 | * set and go into memory reserves if necessary. | |
3128 | */ | |
76b342bd JK |
3129 | page = cache_grow_begin(cache, flags, numa_mem_id()); |
3130 | cache_grow_end(cache, page); | |
3131 | if (page) { | |
3132 | nid = page_to_nid(page); | |
511e3a05 JK |
3133 | obj = ____cache_alloc_node(cache, |
3134 | gfp_exact_node(flags), nid); | |
0c3aa83e | 3135 | |
3c517a61 | 3136 | /* |
511e3a05 JK |
3137 | * Another processor may allocate the objects in |
3138 | * the slab since we are not holding any locks. | |
3c517a61 | 3139 | */ |
511e3a05 JK |
3140 | if (!obj) |
3141 | goto retry; | |
3c517a61 | 3142 | } |
aedb0eb1 | 3143 | } |
cc9a6c87 | 3144 | |
d26914d1 | 3145 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3146 | goto retry_cpuset; |
765c4507 CL |
3147 | return obj; |
3148 | } | |
3149 | ||
e498be7d CL |
3150 | /* |
3151 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3152 | */ |
8b98c169 | 3153 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3154 | int nodeid) |
e498be7d | 3155 | { |
8456a648 | 3156 | struct page *page; |
ce8eb6c4 | 3157 | struct kmem_cache_node *n; |
213b4695 | 3158 | void *obj = NULL; |
b03a017b | 3159 | void *list = NULL; |
b28a02de | 3160 | |
7c3fbbdd | 3161 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3162 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3163 | BUG_ON(!n); |
b28a02de | 3164 | |
ca3b9b91 | 3165 | check_irq_off(); |
ce8eb6c4 | 3166 | spin_lock(&n->list_lock); |
f68f8ddd | 3167 | page = get_first_slab(n, false); |
7aa0d227 GT |
3168 | if (!page) |
3169 | goto must_grow; | |
b28a02de | 3170 | |
b28a02de | 3171 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3172 | |
3173 | STATS_INC_NODEALLOCS(cachep); | |
3174 | STATS_INC_ACTIVE(cachep); | |
3175 | STATS_SET_HIGH(cachep); | |
3176 | ||
8456a648 | 3177 | BUG_ON(page->active == cachep->num); |
b28a02de | 3178 | |
260b61dd | 3179 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3180 | n->free_objects--; |
b28a02de | 3181 | |
b03a017b | 3182 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3183 | |
ce8eb6c4 | 3184 | spin_unlock(&n->list_lock); |
b03a017b | 3185 | fixup_objfreelist_debug(cachep, &list); |
213b4695 | 3186 | return obj; |
e498be7d | 3187 | |
a737b3e2 | 3188 | must_grow: |
ce8eb6c4 | 3189 | spin_unlock(&n->list_lock); |
76b342bd | 3190 | page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid); |
213b4695 JK |
3191 | if (page) { |
3192 | /* This slab isn't counted yet so don't update free_objects */ | |
3193 | obj = slab_get_obj(cachep, page); | |
3194 | } | |
76b342bd | 3195 | cache_grow_end(cachep, page); |
1da177e4 | 3196 | |
213b4695 | 3197 | return obj ? obj : fallback_alloc(cachep, flags); |
e498be7d | 3198 | } |
8c8cc2c1 | 3199 | |
8c8cc2c1 | 3200 | static __always_inline void * |
48356303 | 3201 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3202 | unsigned long caller) |
8c8cc2c1 PE |
3203 | { |
3204 | unsigned long save_flags; | |
3205 | void *ptr; | |
7d6e6d09 | 3206 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3207 | |
dcce284a | 3208 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3209 | cachep = slab_pre_alloc_hook(cachep, flags); |
3210 | if (unlikely(!cachep)) | |
824ebef1 AM |
3211 | return NULL; |
3212 | ||
8c8cc2c1 PE |
3213 | cache_alloc_debugcheck_before(cachep, flags); |
3214 | local_irq_save(save_flags); | |
3215 | ||
eacbbae3 | 3216 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3217 | nodeid = slab_node; |
8c8cc2c1 | 3218 | |
18bf8541 | 3219 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3220 | /* Node not bootstrapped yet */ |
3221 | ptr = fallback_alloc(cachep, flags); | |
3222 | goto out; | |
3223 | } | |
3224 | ||
7d6e6d09 | 3225 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3226 | /* |
3227 | * Use the locally cached objects if possible. | |
3228 | * However ____cache_alloc does not allow fallback | |
3229 | * to other nodes. It may fail while we still have | |
3230 | * objects on other nodes available. | |
3231 | */ | |
3232 | ptr = ____cache_alloc(cachep, flags); | |
3233 | if (ptr) | |
3234 | goto out; | |
3235 | } | |
3236 | /* ___cache_alloc_node can fall back to other nodes */ | |
3237 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3238 | out: | |
3239 | local_irq_restore(save_flags); | |
3240 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3241 | ||
d5e3ed66 JDB |
3242 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3243 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3244 | |
d5e3ed66 | 3245 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3246 | return ptr; |
3247 | } | |
3248 | ||
3249 | static __always_inline void * | |
3250 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3251 | { | |
3252 | void *objp; | |
3253 | ||
2ad654bc | 3254 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3255 | objp = alternate_node_alloc(cache, flags); |
3256 | if (objp) | |
3257 | goto out; | |
3258 | } | |
3259 | objp = ____cache_alloc(cache, flags); | |
3260 | ||
3261 | /* | |
3262 | * We may just have run out of memory on the local node. | |
3263 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3264 | */ | |
7d6e6d09 LS |
3265 | if (!objp) |
3266 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3267 | |
3268 | out: | |
3269 | return objp; | |
3270 | } | |
3271 | #else | |
3272 | ||
3273 | static __always_inline void * | |
3274 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3275 | { | |
3276 | return ____cache_alloc(cachep, flags); | |
3277 | } | |
3278 | ||
3279 | #endif /* CONFIG_NUMA */ | |
3280 | ||
3281 | static __always_inline void * | |
48356303 | 3282 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3283 | { |
3284 | unsigned long save_flags; | |
3285 | void *objp; | |
3286 | ||
dcce284a | 3287 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3288 | cachep = slab_pre_alloc_hook(cachep, flags); |
3289 | if (unlikely(!cachep)) | |
824ebef1 AM |
3290 | return NULL; |
3291 | ||
8c8cc2c1 PE |
3292 | cache_alloc_debugcheck_before(cachep, flags); |
3293 | local_irq_save(save_flags); | |
3294 | objp = __do_cache_alloc(cachep, flags); | |
3295 | local_irq_restore(save_flags); | |
3296 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3297 | prefetchw(objp); | |
3298 | ||
d5e3ed66 JDB |
3299 | if (unlikely(flags & __GFP_ZERO) && objp) |
3300 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3301 | |
d5e3ed66 | 3302 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3303 | return objp; |
3304 | } | |
e498be7d CL |
3305 | |
3306 | /* | |
5f0985bb | 3307 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3308 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3309 | */ |
97654dfa JK |
3310 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3311 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3312 | { |
3313 | int i; | |
25c063fb | 3314 | struct kmem_cache_node *n = get_node(cachep, node); |
6052b788 JK |
3315 | struct page *page; |
3316 | ||
3317 | n->free_objects += nr_objects; | |
1da177e4 LT |
3318 | |
3319 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3320 | void *objp; |
8456a648 | 3321 | struct page *page; |
1da177e4 | 3322 | |
072bb0aa MG |
3323 | objp = objpp[i]; |
3324 | ||
8456a648 | 3325 | page = virt_to_head_page(objp); |
8456a648 | 3326 | list_del(&page->lru); |
ff69416e | 3327 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3328 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3329 | STATS_DEC_ACTIVE(cachep); |
1da177e4 LT |
3330 | |
3331 | /* fixup slab chains */ | |
6052b788 JK |
3332 | if (page->active == 0) |
3333 | list_add(&page->lru, &n->slabs_free); | |
3334 | else { | |
1da177e4 LT |
3335 | /* Unconditionally move a slab to the end of the |
3336 | * partial list on free - maximum time for the | |
3337 | * other objects to be freed, too. | |
3338 | */ | |
8456a648 | 3339 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3340 | } |
3341 | } | |
6052b788 JK |
3342 | |
3343 | while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) { | |
3344 | n->free_objects -= cachep->num; | |
3345 | ||
3346 | page = list_last_entry(&n->slabs_free, struct page, lru); | |
3347 | list_del(&page->lru); | |
3348 | list_add(&page->lru, list); | |
3349 | } | |
1da177e4 LT |
3350 | } |
3351 | ||
343e0d7a | 3352 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3353 | { |
3354 | int batchcount; | |
ce8eb6c4 | 3355 | struct kmem_cache_node *n; |
7d6e6d09 | 3356 | int node = numa_mem_id(); |
97654dfa | 3357 | LIST_HEAD(list); |
1da177e4 LT |
3358 | |
3359 | batchcount = ac->batchcount; | |
260b61dd | 3360 | |
1da177e4 | 3361 | check_irq_off(); |
18bf8541 | 3362 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3363 | spin_lock(&n->list_lock); |
3364 | if (n->shared) { | |
3365 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3366 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3367 | if (max) { |
3368 | if (batchcount > max) | |
3369 | batchcount = max; | |
e498be7d | 3370 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3371 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3372 | shared_array->avail += batchcount; |
3373 | goto free_done; | |
3374 | } | |
3375 | } | |
3376 | ||
97654dfa | 3377 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3378 | free_done: |
1da177e4 LT |
3379 | #if STATS |
3380 | { | |
3381 | int i = 0; | |
73c0219d | 3382 | struct page *page; |
1da177e4 | 3383 | |
73c0219d | 3384 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3385 | BUG_ON(page->active); |
1da177e4 LT |
3386 | |
3387 | i++; | |
1da177e4 LT |
3388 | } |
3389 | STATS_SET_FREEABLE(cachep, i); | |
3390 | } | |
3391 | #endif | |
ce8eb6c4 | 3392 | spin_unlock(&n->list_lock); |
97654dfa | 3393 | slabs_destroy(cachep, &list); |
1da177e4 | 3394 | ac->avail -= batchcount; |
a737b3e2 | 3395 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3396 | } |
3397 | ||
3398 | /* | |
a737b3e2 AM |
3399 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3400 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3401 | */ |
a947eb95 | 3402 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3403 | unsigned long caller) |
1da177e4 | 3404 | { |
9a2dba4b | 3405 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 | 3406 | |
7ed2f9e6 AP |
3407 | kasan_slab_free(cachep, objp); |
3408 | ||
1da177e4 | 3409 | check_irq_off(); |
d5cff635 | 3410 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3411 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3412 | |
8c138bc0 | 3413 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3414 | |
1807a1aa SS |
3415 | /* |
3416 | * Skip calling cache_free_alien() when the platform is not numa. | |
3417 | * This will avoid cache misses that happen while accessing slabp (which | |
3418 | * is per page memory reference) to get nodeid. Instead use a global | |
3419 | * variable to skip the call, which is mostly likely to be present in | |
3420 | * the cache. | |
3421 | */ | |
b6e68bc1 | 3422 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3423 | return; |
3424 | ||
3d880194 | 3425 | if (ac->avail < ac->limit) { |
1da177e4 | 3426 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3427 | } else { |
3428 | STATS_INC_FREEMISS(cachep); | |
3429 | cache_flusharray(cachep, ac); | |
1da177e4 | 3430 | } |
42c8c99c | 3431 | |
f68f8ddd JK |
3432 | if (sk_memalloc_socks()) { |
3433 | struct page *page = virt_to_head_page(objp); | |
3434 | ||
3435 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3436 | cache_free_pfmemalloc(cachep, page, objp); | |
3437 | return; | |
3438 | } | |
3439 | } | |
3440 | ||
3441 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3442 | } |
3443 | ||
3444 | /** | |
3445 | * kmem_cache_alloc - Allocate an object | |
3446 | * @cachep: The cache to allocate from. | |
3447 | * @flags: See kmalloc(). | |
3448 | * | |
3449 | * Allocate an object from this cache. The flags are only relevant | |
3450 | * if the cache has no available objects. | |
3451 | */ | |
343e0d7a | 3452 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3453 | { |
48356303 | 3454 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3455 | |
505f5dcb | 3456 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3457 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3458 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3459 | |
3460 | return ret; | |
1da177e4 LT |
3461 | } |
3462 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3463 | ||
7b0501dd JDB |
3464 | static __always_inline void |
3465 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3466 | size_t size, void **p, unsigned long caller) | |
3467 | { | |
3468 | size_t i; | |
3469 | ||
3470 | for (i = 0; i < size; i++) | |
3471 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3472 | } | |
3473 | ||
865762a8 | 3474 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3475 | void **p) |
484748f0 | 3476 | { |
2a777eac JDB |
3477 | size_t i; |
3478 | ||
3479 | s = slab_pre_alloc_hook(s, flags); | |
3480 | if (!s) | |
3481 | return 0; | |
3482 | ||
3483 | cache_alloc_debugcheck_before(s, flags); | |
3484 | ||
3485 | local_irq_disable(); | |
3486 | for (i = 0; i < size; i++) { | |
3487 | void *objp = __do_cache_alloc(s, flags); | |
3488 | ||
2a777eac JDB |
3489 | if (unlikely(!objp)) |
3490 | goto error; | |
3491 | p[i] = objp; | |
3492 | } | |
3493 | local_irq_enable(); | |
3494 | ||
7b0501dd JDB |
3495 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3496 | ||
2a777eac JDB |
3497 | /* Clear memory outside IRQ disabled section */ |
3498 | if (unlikely(flags & __GFP_ZERO)) | |
3499 | for (i = 0; i < size; i++) | |
3500 | memset(p[i], 0, s->object_size); | |
3501 | ||
3502 | slab_post_alloc_hook(s, flags, size, p); | |
3503 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3504 | return size; | |
3505 | error: | |
3506 | local_irq_enable(); | |
7b0501dd | 3507 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3508 | slab_post_alloc_hook(s, flags, i, p); |
3509 | __kmem_cache_free_bulk(s, i, p); | |
3510 | return 0; | |
484748f0 CL |
3511 | } |
3512 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3513 | ||
0f24f128 | 3514 | #ifdef CONFIG_TRACING |
85beb586 | 3515 | void * |
4052147c | 3516 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3517 | { |
85beb586 SR |
3518 | void *ret; |
3519 | ||
48356303 | 3520 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3521 | |
505f5dcb | 3522 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3523 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3524 | size, cachep->size, flags); |
85beb586 | 3525 | return ret; |
36555751 | 3526 | } |
85beb586 | 3527 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3528 | #endif |
3529 | ||
1da177e4 | 3530 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3531 | /** |
3532 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3533 | * @cachep: The cache to allocate from. | |
3534 | * @flags: See kmalloc(). | |
3535 | * @nodeid: node number of the target node. | |
3536 | * | |
3537 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3538 | * node, which can improve the performance for cpu bound structures. | |
3539 | * | |
3540 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3541 | */ | |
8b98c169 CH |
3542 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3543 | { | |
48356303 | 3544 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3545 | |
505f5dcb | 3546 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3547 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3548 | cachep->object_size, cachep->size, |
ca2b84cb | 3549 | flags, nodeid); |
36555751 EGM |
3550 | |
3551 | return ret; | |
8b98c169 | 3552 | } |
1da177e4 LT |
3553 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3554 | ||
0f24f128 | 3555 | #ifdef CONFIG_TRACING |
4052147c | 3556 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3557 | gfp_t flags, |
4052147c EG |
3558 | int nodeid, |
3559 | size_t size) | |
36555751 | 3560 | { |
85beb586 SR |
3561 | void *ret; |
3562 | ||
592f4145 | 3563 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3564 | |
3565 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3566 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3567 | size, cachep->size, |
85beb586 SR |
3568 | flags, nodeid); |
3569 | return ret; | |
36555751 | 3570 | } |
85beb586 | 3571 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3572 | #endif |
3573 | ||
8b98c169 | 3574 | static __always_inline void * |
7c0cb9c6 | 3575 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3576 | { |
343e0d7a | 3577 | struct kmem_cache *cachep; |
7ed2f9e6 | 3578 | void *ret; |
97e2bde4 | 3579 | |
2c59dd65 | 3580 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3581 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3582 | return cachep; | |
7ed2f9e6 | 3583 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3584 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3585 | |
3586 | return ret; | |
97e2bde4 | 3587 | } |
8b98c169 | 3588 | |
8b98c169 CH |
3589 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3590 | { | |
7c0cb9c6 | 3591 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3592 | } |
dbe5e69d | 3593 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3594 | |
3595 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3596 | int node, unsigned long caller) |
8b98c169 | 3597 | { |
7c0cb9c6 | 3598 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3599 | } |
3600 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3601 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3602 | |
3603 | /** | |
800590f5 | 3604 | * __do_kmalloc - allocate memory |
1da177e4 | 3605 | * @size: how many bytes of memory are required. |
800590f5 | 3606 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3607 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3608 | */ |
7fd6b141 | 3609 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3610 | unsigned long caller) |
1da177e4 | 3611 | { |
343e0d7a | 3612 | struct kmem_cache *cachep; |
36555751 | 3613 | void *ret; |
1da177e4 | 3614 | |
2c59dd65 | 3615 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3616 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3617 | return cachep; | |
48356303 | 3618 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3619 | |
505f5dcb | 3620 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3621 | trace_kmalloc(caller, ret, |
3b0efdfa | 3622 | size, cachep->size, flags); |
36555751 EGM |
3623 | |
3624 | return ret; | |
7fd6b141 PE |
3625 | } |
3626 | ||
7fd6b141 PE |
3627 | void *__kmalloc(size_t size, gfp_t flags) |
3628 | { | |
7c0cb9c6 | 3629 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3630 | } |
3631 | EXPORT_SYMBOL(__kmalloc); | |
3632 | ||
ce71e27c | 3633 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3634 | { |
7c0cb9c6 | 3635 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3636 | } |
3637 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3638 | |
1da177e4 LT |
3639 | /** |
3640 | * kmem_cache_free - Deallocate an object | |
3641 | * @cachep: The cache the allocation was from. | |
3642 | * @objp: The previously allocated object. | |
3643 | * | |
3644 | * Free an object which was previously allocated from this | |
3645 | * cache. | |
3646 | */ | |
343e0d7a | 3647 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3648 | { |
3649 | unsigned long flags; | |
b9ce5ef4 GC |
3650 | cachep = cache_from_obj(cachep, objp); |
3651 | if (!cachep) | |
3652 | return; | |
1da177e4 LT |
3653 | |
3654 | local_irq_save(flags); | |
d97d476b | 3655 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3656 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3657 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3658 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3659 | local_irq_restore(flags); |
36555751 | 3660 | |
ca2b84cb | 3661 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3662 | } |
3663 | EXPORT_SYMBOL(kmem_cache_free); | |
3664 | ||
e6cdb58d JDB |
3665 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3666 | { | |
3667 | struct kmem_cache *s; | |
3668 | size_t i; | |
3669 | ||
3670 | local_irq_disable(); | |
3671 | for (i = 0; i < size; i++) { | |
3672 | void *objp = p[i]; | |
3673 | ||
ca257195 JDB |
3674 | if (!orig_s) /* called via kfree_bulk */ |
3675 | s = virt_to_cache(objp); | |
3676 | else | |
3677 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3678 | |
3679 | debug_check_no_locks_freed(objp, s->object_size); | |
3680 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3681 | debug_check_no_obj_freed(objp, s->object_size); | |
3682 | ||
3683 | __cache_free(s, objp, _RET_IP_); | |
3684 | } | |
3685 | local_irq_enable(); | |
3686 | ||
3687 | /* FIXME: add tracing */ | |
3688 | } | |
3689 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3690 | ||
1da177e4 LT |
3691 | /** |
3692 | * kfree - free previously allocated memory | |
3693 | * @objp: pointer returned by kmalloc. | |
3694 | * | |
80e93eff PE |
3695 | * If @objp is NULL, no operation is performed. |
3696 | * | |
1da177e4 LT |
3697 | * Don't free memory not originally allocated by kmalloc() |
3698 | * or you will run into trouble. | |
3699 | */ | |
3700 | void kfree(const void *objp) | |
3701 | { | |
343e0d7a | 3702 | struct kmem_cache *c; |
1da177e4 LT |
3703 | unsigned long flags; |
3704 | ||
2121db74 PE |
3705 | trace_kfree(_RET_IP_, objp); |
3706 | ||
6cb8f913 | 3707 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3708 | return; |
3709 | local_irq_save(flags); | |
3710 | kfree_debugcheck(objp); | |
6ed5eb22 | 3711 | c = virt_to_cache(objp); |
8c138bc0 CL |
3712 | debug_check_no_locks_freed(objp, c->object_size); |
3713 | ||
3714 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3715 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3716 | local_irq_restore(flags); |
3717 | } | |
3718 | EXPORT_SYMBOL(kfree); | |
3719 | ||
e498be7d | 3720 | /* |
ce8eb6c4 | 3721 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3722 | */ |
c3d332b6 | 3723 | static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d | 3724 | { |
c3d332b6 | 3725 | int ret; |
e498be7d | 3726 | int node; |
ce8eb6c4 | 3727 | struct kmem_cache_node *n; |
e498be7d | 3728 | |
9c09a95c | 3729 | for_each_online_node(node) { |
c3d332b6 JK |
3730 | ret = setup_kmem_cache_node(cachep, node, gfp, true); |
3731 | if (ret) | |
e498be7d CL |
3732 | goto fail; |
3733 | ||
e498be7d | 3734 | } |
c3d332b6 | 3735 | |
cafeb02e | 3736 | return 0; |
0718dc2a | 3737 | |
a737b3e2 | 3738 | fail: |
3b0efdfa | 3739 | if (!cachep->list.next) { |
0718dc2a CL |
3740 | /* Cache is not active yet. Roll back what we did */ |
3741 | node--; | |
3742 | while (node >= 0) { | |
18bf8541 CL |
3743 | n = get_node(cachep, node); |
3744 | if (n) { | |
ce8eb6c4 CL |
3745 | kfree(n->shared); |
3746 | free_alien_cache(n->alien); | |
3747 | kfree(n); | |
6a67368c | 3748 | cachep->node[node] = NULL; |
0718dc2a CL |
3749 | } |
3750 | node--; | |
3751 | } | |
3752 | } | |
cafeb02e | 3753 | return -ENOMEM; |
e498be7d CL |
3754 | } |
3755 | ||
18004c5d | 3756 | /* Always called with the slab_mutex held */ |
943a451a | 3757 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3758 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3759 | { |
bf0dea23 JK |
3760 | struct array_cache __percpu *cpu_cache, *prev; |
3761 | int cpu; | |
1da177e4 | 3762 | |
bf0dea23 JK |
3763 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3764 | if (!cpu_cache) | |
d2e7b7d0 SS |
3765 | return -ENOMEM; |
3766 | ||
bf0dea23 JK |
3767 | prev = cachep->cpu_cache; |
3768 | cachep->cpu_cache = cpu_cache; | |
3769 | kick_all_cpus_sync(); | |
e498be7d | 3770 | |
1da177e4 | 3771 | check_irq_on(); |
1da177e4 LT |
3772 | cachep->batchcount = batchcount; |
3773 | cachep->limit = limit; | |
e498be7d | 3774 | cachep->shared = shared; |
1da177e4 | 3775 | |
bf0dea23 | 3776 | if (!prev) |
c3d332b6 | 3777 | goto setup_node; |
bf0dea23 JK |
3778 | |
3779 | for_each_online_cpu(cpu) { | |
97654dfa | 3780 | LIST_HEAD(list); |
18bf8541 CL |
3781 | int node; |
3782 | struct kmem_cache_node *n; | |
bf0dea23 | 3783 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3784 | |
bf0dea23 | 3785 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3786 | n = get_node(cachep, node); |
3787 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3788 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3789 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3790 | slabs_destroy(cachep, &list); |
1da177e4 | 3791 | } |
bf0dea23 JK |
3792 | free_percpu(prev); |
3793 | ||
c3d332b6 JK |
3794 | setup_node: |
3795 | return setup_kmem_cache_nodes(cachep, gfp); | |
1da177e4 LT |
3796 | } |
3797 | ||
943a451a GC |
3798 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3799 | int batchcount, int shared, gfp_t gfp) | |
3800 | { | |
3801 | int ret; | |
426589f5 | 3802 | struct kmem_cache *c; |
943a451a GC |
3803 | |
3804 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3805 | ||
3806 | if (slab_state < FULL) | |
3807 | return ret; | |
3808 | ||
3809 | if ((ret < 0) || !is_root_cache(cachep)) | |
3810 | return ret; | |
3811 | ||
426589f5 VD |
3812 | lockdep_assert_held(&slab_mutex); |
3813 | for_each_memcg_cache(c, cachep) { | |
3814 | /* return value determined by the root cache only */ | |
3815 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3816 | } |
3817 | ||
3818 | return ret; | |
3819 | } | |
3820 | ||
18004c5d | 3821 | /* Called with slab_mutex held always */ |
83b519e8 | 3822 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3823 | { |
3824 | int err; | |
943a451a GC |
3825 | int limit = 0; |
3826 | int shared = 0; | |
3827 | int batchcount = 0; | |
3828 | ||
3829 | if (!is_root_cache(cachep)) { | |
3830 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3831 | limit = root->limit; | |
3832 | shared = root->shared; | |
3833 | batchcount = root->batchcount; | |
3834 | } | |
1da177e4 | 3835 | |
943a451a GC |
3836 | if (limit && shared && batchcount) |
3837 | goto skip_setup; | |
a737b3e2 AM |
3838 | /* |
3839 | * The head array serves three purposes: | |
1da177e4 LT |
3840 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3841 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3842 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3843 | * bufctl chains: array operations are cheaper. |
3844 | * The numbers are guessed, we should auto-tune as described by | |
3845 | * Bonwick. | |
3846 | */ | |
3b0efdfa | 3847 | if (cachep->size > 131072) |
1da177e4 | 3848 | limit = 1; |
3b0efdfa | 3849 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3850 | limit = 8; |
3b0efdfa | 3851 | else if (cachep->size > 1024) |
1da177e4 | 3852 | limit = 24; |
3b0efdfa | 3853 | else if (cachep->size > 256) |
1da177e4 LT |
3854 | limit = 54; |
3855 | else | |
3856 | limit = 120; | |
3857 | ||
a737b3e2 AM |
3858 | /* |
3859 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3860 | * allocation behaviour: Most allocs on one cpu, most free operations |
3861 | * on another cpu. For these cases, an efficient object passing between | |
3862 | * cpus is necessary. This is provided by a shared array. The array | |
3863 | * replaces Bonwick's magazine layer. | |
3864 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3865 | * to a larger limit. Thus disabled by default. | |
3866 | */ | |
3867 | shared = 0; | |
3b0efdfa | 3868 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3869 | shared = 8; |
1da177e4 LT |
3870 | |
3871 | #if DEBUG | |
a737b3e2 AM |
3872 | /* |
3873 | * With debugging enabled, large batchcount lead to excessively long | |
3874 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3875 | */ |
3876 | if (limit > 32) | |
3877 | limit = 32; | |
3878 | #endif | |
943a451a GC |
3879 | batchcount = (limit + 1) / 2; |
3880 | skip_setup: | |
3881 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 | 3882 | if (err) |
1170532b | 3883 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3884 | cachep->name, -err); |
2ed3a4ef | 3885 | return err; |
1da177e4 LT |
3886 | } |
3887 | ||
1b55253a | 3888 | /* |
ce8eb6c4 CL |
3889 | * Drain an array if it contains any elements taking the node lock only if |
3890 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3891 | * if drain_array() is used on the shared array. |
1b55253a | 3892 | */ |
ce8eb6c4 | 3893 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 3894 | struct array_cache *ac, int node) |
1da177e4 | 3895 | { |
97654dfa | 3896 | LIST_HEAD(list); |
18726ca8 JK |
3897 | |
3898 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
3899 | check_mutex_acquired(); | |
1da177e4 | 3900 | |
1b55253a CL |
3901 | if (!ac || !ac->avail) |
3902 | return; | |
18726ca8 JK |
3903 | |
3904 | if (ac->touched) { | |
1da177e4 | 3905 | ac->touched = 0; |
18726ca8 | 3906 | return; |
1da177e4 | 3907 | } |
18726ca8 JK |
3908 | |
3909 | spin_lock_irq(&n->list_lock); | |
3910 | drain_array_locked(cachep, ac, node, false, &list); | |
3911 | spin_unlock_irq(&n->list_lock); | |
3912 | ||
3913 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
3914 | } |
3915 | ||
3916 | /** | |
3917 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3918 | * @w: work descriptor |
1da177e4 LT |
3919 | * |
3920 | * Called from workqueue/eventd every few seconds. | |
3921 | * Purpose: | |
3922 | * - clear the per-cpu caches for this CPU. | |
3923 | * - return freeable pages to the main free memory pool. | |
3924 | * | |
a737b3e2 AM |
3925 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3926 | * again on the next iteration. | |
1da177e4 | 3927 | */ |
7c5cae36 | 3928 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3929 | { |
7a7c381d | 3930 | struct kmem_cache *searchp; |
ce8eb6c4 | 3931 | struct kmem_cache_node *n; |
7d6e6d09 | 3932 | int node = numa_mem_id(); |
bf6aede7 | 3933 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3934 | |
18004c5d | 3935 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3936 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3937 | goto out; |
1da177e4 | 3938 | |
18004c5d | 3939 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3940 | check_irq_on(); |
3941 | ||
35386e3b | 3942 | /* |
ce8eb6c4 | 3943 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3944 | * have established with reasonable certainty that |
3945 | * we can do some work if the lock was obtained. | |
3946 | */ | |
18bf8541 | 3947 | n = get_node(searchp, node); |
35386e3b | 3948 | |
ce8eb6c4 | 3949 | reap_alien(searchp, n); |
1da177e4 | 3950 | |
18726ca8 | 3951 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 3952 | |
35386e3b CL |
3953 | /* |
3954 | * These are racy checks but it does not matter | |
3955 | * if we skip one check or scan twice. | |
3956 | */ | |
ce8eb6c4 | 3957 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3958 | goto next; |
1da177e4 | 3959 | |
5f0985bb | 3960 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3961 | |
18726ca8 | 3962 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 3963 | |
ce8eb6c4 CL |
3964 | if (n->free_touched) |
3965 | n->free_touched = 0; | |
ed11d9eb CL |
3966 | else { |
3967 | int freed; | |
1da177e4 | 3968 | |
ce8eb6c4 | 3969 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3970 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3971 | STATS_ADD_REAPED(searchp, freed); | |
3972 | } | |
35386e3b | 3973 | next: |
1da177e4 LT |
3974 | cond_resched(); |
3975 | } | |
3976 | check_irq_on(); | |
18004c5d | 3977 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3978 | next_reap_node(); |
7c5cae36 | 3979 | out: |
a737b3e2 | 3980 | /* Set up the next iteration */ |
5f0985bb | 3981 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3982 | } |
3983 | ||
158a9624 | 3984 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3985 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3986 | { |
8456a648 | 3987 | struct page *page; |
b28a02de PE |
3988 | unsigned long active_objs; |
3989 | unsigned long num_objs; | |
3990 | unsigned long active_slabs = 0; | |
3991 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3992 | const char *name; |
1da177e4 | 3993 | char *error = NULL; |
e498be7d | 3994 | int node; |
ce8eb6c4 | 3995 | struct kmem_cache_node *n; |
1da177e4 | 3996 | |
1da177e4 LT |
3997 | active_objs = 0; |
3998 | num_slabs = 0; | |
18bf8541 | 3999 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 4000 | |
ca3b9b91 | 4001 | check_irq_on(); |
ce8eb6c4 | 4002 | spin_lock_irq(&n->list_lock); |
e498be7d | 4003 | |
8456a648 JK |
4004 | list_for_each_entry(page, &n->slabs_full, lru) { |
4005 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4006 | error = "slabs_full accounting error"; |
4007 | active_objs += cachep->num; | |
4008 | active_slabs++; | |
4009 | } | |
8456a648 JK |
4010 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4011 | if (page->active == cachep->num && !error) | |
106a74e1 | 4012 | error = "slabs_partial accounting error"; |
8456a648 | 4013 | if (!page->active && !error) |
106a74e1 | 4014 | error = "slabs_partial accounting error"; |
8456a648 | 4015 | active_objs += page->active; |
e498be7d CL |
4016 | active_slabs++; |
4017 | } | |
8456a648 JK |
4018 | list_for_each_entry(page, &n->slabs_free, lru) { |
4019 | if (page->active && !error) | |
106a74e1 | 4020 | error = "slabs_free accounting error"; |
e498be7d CL |
4021 | num_slabs++; |
4022 | } | |
ce8eb6c4 CL |
4023 | free_objects += n->free_objects; |
4024 | if (n->shared) | |
4025 | shared_avail += n->shared->avail; | |
e498be7d | 4026 | |
ce8eb6c4 | 4027 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4028 | } |
b28a02de PE |
4029 | num_slabs += active_slabs; |
4030 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4031 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4032 | error = "free_objects accounting error"; |
4033 | ||
b28a02de | 4034 | name = cachep->name; |
1da177e4 | 4035 | if (error) |
1170532b | 4036 | pr_err("slab: cache %s error: %s\n", name, error); |
1da177e4 | 4037 | |
0d7561c6 GC |
4038 | sinfo->active_objs = active_objs; |
4039 | sinfo->num_objs = num_objs; | |
4040 | sinfo->active_slabs = active_slabs; | |
4041 | sinfo->num_slabs = num_slabs; | |
4042 | sinfo->shared_avail = shared_avail; | |
4043 | sinfo->limit = cachep->limit; | |
4044 | sinfo->batchcount = cachep->batchcount; | |
4045 | sinfo->shared = cachep->shared; | |
4046 | sinfo->objects_per_slab = cachep->num; | |
4047 | sinfo->cache_order = cachep->gfporder; | |
4048 | } | |
4049 | ||
4050 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4051 | { | |
1da177e4 | 4052 | #if STATS |
ce8eb6c4 | 4053 | { /* node stats */ |
1da177e4 LT |
4054 | unsigned long high = cachep->high_mark; |
4055 | unsigned long allocs = cachep->num_allocations; | |
4056 | unsigned long grown = cachep->grown; | |
4057 | unsigned long reaped = cachep->reaped; | |
4058 | unsigned long errors = cachep->errors; | |
4059 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4060 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4061 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4062 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4063 | |
756a025f | 4064 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4065 | allocs, high, grown, |
4066 | reaped, errors, max_freeable, node_allocs, | |
4067 | node_frees, overflows); | |
1da177e4 LT |
4068 | } |
4069 | /* cpu stats */ | |
4070 | { | |
4071 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4072 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4073 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4074 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4075 | ||
4076 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4077 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4078 | } |
4079 | #endif | |
1da177e4 LT |
4080 | } |
4081 | ||
1da177e4 LT |
4082 | #define MAX_SLABINFO_WRITE 128 |
4083 | /** | |
4084 | * slabinfo_write - Tuning for the slab allocator | |
4085 | * @file: unused | |
4086 | * @buffer: user buffer | |
4087 | * @count: data length | |
4088 | * @ppos: unused | |
4089 | */ | |
b7454ad3 | 4090 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4091 | size_t count, loff_t *ppos) |
1da177e4 | 4092 | { |
b28a02de | 4093 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4094 | int limit, batchcount, shared, res; |
7a7c381d | 4095 | struct kmem_cache *cachep; |
b28a02de | 4096 | |
1da177e4 LT |
4097 | if (count > MAX_SLABINFO_WRITE) |
4098 | return -EINVAL; | |
4099 | if (copy_from_user(&kbuf, buffer, count)) | |
4100 | return -EFAULT; | |
b28a02de | 4101 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4102 | |
4103 | tmp = strchr(kbuf, ' '); | |
4104 | if (!tmp) | |
4105 | return -EINVAL; | |
4106 | *tmp = '\0'; | |
4107 | tmp++; | |
4108 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4109 | return -EINVAL; | |
4110 | ||
4111 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4112 | mutex_lock(&slab_mutex); |
1da177e4 | 4113 | res = -EINVAL; |
18004c5d | 4114 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4115 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4116 | if (limit < 1 || batchcount < 1 || |
4117 | batchcount > limit || shared < 0) { | |
e498be7d | 4118 | res = 0; |
1da177e4 | 4119 | } else { |
e498be7d | 4120 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4121 | batchcount, shared, |
4122 | GFP_KERNEL); | |
1da177e4 LT |
4123 | } |
4124 | break; | |
4125 | } | |
4126 | } | |
18004c5d | 4127 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4128 | if (res >= 0) |
4129 | res = count; | |
4130 | return res; | |
4131 | } | |
871751e2 AV |
4132 | |
4133 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4134 | ||
871751e2 AV |
4135 | static inline int add_caller(unsigned long *n, unsigned long v) |
4136 | { | |
4137 | unsigned long *p; | |
4138 | int l; | |
4139 | if (!v) | |
4140 | return 1; | |
4141 | l = n[1]; | |
4142 | p = n + 2; | |
4143 | while (l) { | |
4144 | int i = l/2; | |
4145 | unsigned long *q = p + 2 * i; | |
4146 | if (*q == v) { | |
4147 | q[1]++; | |
4148 | return 1; | |
4149 | } | |
4150 | if (*q > v) { | |
4151 | l = i; | |
4152 | } else { | |
4153 | p = q + 2; | |
4154 | l -= i + 1; | |
4155 | } | |
4156 | } | |
4157 | if (++n[1] == n[0]) | |
4158 | return 0; | |
4159 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4160 | p[0] = v; | |
4161 | p[1] = 1; | |
4162 | return 1; | |
4163 | } | |
4164 | ||
8456a648 JK |
4165 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4166 | struct page *page) | |
871751e2 AV |
4167 | { |
4168 | void *p; | |
d31676df JK |
4169 | int i, j; |
4170 | unsigned long v; | |
b1cb0982 | 4171 | |
871751e2 AV |
4172 | if (n[0] == n[1]) |
4173 | return; | |
8456a648 | 4174 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4175 | bool active = true; |
4176 | ||
4177 | for (j = page->active; j < c->num; j++) { | |
4178 | if (get_free_obj(page, j) == i) { | |
4179 | active = false; | |
4180 | break; | |
4181 | } | |
4182 | } | |
4183 | ||
4184 | if (!active) | |
871751e2 | 4185 | continue; |
b1cb0982 | 4186 | |
d31676df JK |
4187 | /* |
4188 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4189 | * mapping is established when actual object allocation and | |
4190 | * we could mistakenly access the unmapped object in the cpu | |
4191 | * cache. | |
4192 | */ | |
4193 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4194 | continue; | |
4195 | ||
4196 | if (!add_caller(n, v)) | |
871751e2 AV |
4197 | return; |
4198 | } | |
4199 | } | |
4200 | ||
4201 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4202 | { | |
4203 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4204 | unsigned long offset, size; |
9281acea | 4205 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4206 | |
a5c43dae | 4207 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4208 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4209 | if (modname[0]) |
871751e2 AV |
4210 | seq_printf(m, " [%s]", modname); |
4211 | return; | |
4212 | } | |
4213 | #endif | |
4214 | seq_printf(m, "%p", (void *)address); | |
4215 | } | |
4216 | ||
4217 | static int leaks_show(struct seq_file *m, void *p) | |
4218 | { | |
0672aa7c | 4219 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4220 | struct page *page; |
ce8eb6c4 | 4221 | struct kmem_cache_node *n; |
871751e2 | 4222 | const char *name; |
db845067 | 4223 | unsigned long *x = m->private; |
871751e2 AV |
4224 | int node; |
4225 | int i; | |
4226 | ||
4227 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4228 | return 0; | |
4229 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4230 | return 0; | |
4231 | ||
d31676df JK |
4232 | /* |
4233 | * Set store_user_clean and start to grab stored user information | |
4234 | * for all objects on this cache. If some alloc/free requests comes | |
4235 | * during the processing, information would be wrong so restart | |
4236 | * whole processing. | |
4237 | */ | |
4238 | do { | |
4239 | set_store_user_clean(cachep); | |
4240 | drain_cpu_caches(cachep); | |
4241 | ||
4242 | x[1] = 0; | |
871751e2 | 4243 | |
d31676df | 4244 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4245 | |
d31676df JK |
4246 | check_irq_on(); |
4247 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4248 | |
d31676df JK |
4249 | list_for_each_entry(page, &n->slabs_full, lru) |
4250 | handle_slab(x, cachep, page); | |
4251 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4252 | handle_slab(x, cachep, page); | |
4253 | spin_unlock_irq(&n->list_lock); | |
4254 | } | |
4255 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4256 | |
871751e2 | 4257 | name = cachep->name; |
db845067 | 4258 | if (x[0] == x[1]) { |
871751e2 | 4259 | /* Increase the buffer size */ |
18004c5d | 4260 | mutex_unlock(&slab_mutex); |
db845067 | 4261 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4262 | if (!m->private) { |
4263 | /* Too bad, we are really out */ | |
db845067 | 4264 | m->private = x; |
18004c5d | 4265 | mutex_lock(&slab_mutex); |
871751e2 AV |
4266 | return -ENOMEM; |
4267 | } | |
db845067 CL |
4268 | *(unsigned long *)m->private = x[0] * 2; |
4269 | kfree(x); | |
18004c5d | 4270 | mutex_lock(&slab_mutex); |
871751e2 AV |
4271 | /* Now make sure this entry will be retried */ |
4272 | m->count = m->size; | |
4273 | return 0; | |
4274 | } | |
db845067 CL |
4275 | for (i = 0; i < x[1]; i++) { |
4276 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4277 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4278 | seq_putc(m, '\n'); |
4279 | } | |
d2e7b7d0 | 4280 | |
871751e2 AV |
4281 | return 0; |
4282 | } | |
4283 | ||
a0ec95a8 | 4284 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4285 | .start = slab_start, |
276a2439 WL |
4286 | .next = slab_next, |
4287 | .stop = slab_stop, | |
871751e2 AV |
4288 | .show = leaks_show, |
4289 | }; | |
a0ec95a8 AD |
4290 | |
4291 | static int slabstats_open(struct inode *inode, struct file *file) | |
4292 | { | |
b208ce32 RJ |
4293 | unsigned long *n; |
4294 | ||
4295 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4296 | if (!n) | |
4297 | return -ENOMEM; | |
4298 | ||
4299 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4300 | ||
4301 | return 0; | |
a0ec95a8 AD |
4302 | } |
4303 | ||
4304 | static const struct file_operations proc_slabstats_operations = { | |
4305 | .open = slabstats_open, | |
4306 | .read = seq_read, | |
4307 | .llseek = seq_lseek, | |
4308 | .release = seq_release_private, | |
4309 | }; | |
4310 | #endif | |
4311 | ||
4312 | static int __init slab_proc_init(void) | |
4313 | { | |
4314 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4315 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4316 | #endif |
a0ec95a8 AD |
4317 | return 0; |
4318 | } | |
4319 | module_init(slab_proc_init); | |
1da177e4 LT |
4320 | #endif |
4321 | ||
00e145b6 MS |
4322 | /** |
4323 | * ksize - get the actual amount of memory allocated for a given object | |
4324 | * @objp: Pointer to the object | |
4325 | * | |
4326 | * kmalloc may internally round up allocations and return more memory | |
4327 | * than requested. ksize() can be used to determine the actual amount of | |
4328 | * memory allocated. The caller may use this additional memory, even though | |
4329 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4330 | * The caller must guarantee that objp points to a valid object previously | |
4331 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4332 | * must not be freed during the duration of the call. | |
4333 | */ | |
fd76bab2 | 4334 | size_t ksize(const void *objp) |
1da177e4 | 4335 | { |
7ed2f9e6 AP |
4336 | size_t size; |
4337 | ||
ef8b4520 CL |
4338 | BUG_ON(!objp); |
4339 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4340 | return 0; |
1da177e4 | 4341 | |
7ed2f9e6 AP |
4342 | size = virt_to_cache(objp)->object_size; |
4343 | /* We assume that ksize callers could use the whole allocated area, | |
4344 | * so we need to unpoison this area. | |
4345 | */ | |
505f5dcb | 4346 | kasan_krealloc(objp, size, GFP_NOWAIT); |
7ed2f9e6 AP |
4347 | |
4348 | return size; | |
1da177e4 | 4349 | } |
b1aabecd | 4350 | EXPORT_SYMBOL(ksize); |