]>
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 | |
e0a42726 IM |
216 | static int slab_early_init = 1; |
217 | ||
ce8eb6c4 | 218 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 219 | |
ce8eb6c4 | 220 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
221 | { |
222 | INIT_LIST_HEAD(&parent->slabs_full); | |
223 | INIT_LIST_HEAD(&parent->slabs_partial); | |
224 | INIT_LIST_HEAD(&parent->slabs_free); | |
225 | parent->shared = NULL; | |
226 | parent->alien = NULL; | |
2e1217cf | 227 | parent->colour_next = 0; |
e498be7d CL |
228 | spin_lock_init(&parent->list_lock); |
229 | parent->free_objects = 0; | |
230 | parent->free_touched = 0; | |
231 | } | |
232 | ||
a737b3e2 AM |
233 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
234 | do { \ | |
235 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 236 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
237 | } while (0) |
238 | ||
a737b3e2 AM |
239 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
240 | do { \ | |
e498be7d CL |
241 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
242 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
243 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
244 | } while (0) | |
1da177e4 | 245 | |
b03a017b | 246 | #define CFLGS_OBJFREELIST_SLAB (0x40000000UL) |
1da177e4 | 247 | #define CFLGS_OFF_SLAB (0x80000000UL) |
b03a017b | 248 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
249 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
250 | ||
251 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
252 | /* |
253 | * Optimization question: fewer reaps means less probability for unnessary | |
254 | * cpucache drain/refill cycles. | |
1da177e4 | 255 | * |
dc6f3f27 | 256 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
257 | * which could lock up otherwise freeable slabs. |
258 | */ | |
5f0985bb JZ |
259 | #define REAPTIMEOUT_AC (2*HZ) |
260 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
261 | |
262 | #if STATS | |
263 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
264 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
265 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
266 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 267 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
268 | #define STATS_SET_HIGH(x) \ |
269 | do { \ | |
270 | if ((x)->num_active > (x)->high_mark) \ | |
271 | (x)->high_mark = (x)->num_active; \ | |
272 | } while (0) | |
1da177e4 LT |
273 | #define STATS_INC_ERR(x) ((x)->errors++) |
274 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 275 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 276 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
277 | #define STATS_SET_FREEABLE(x, i) \ |
278 | do { \ | |
279 | if ((x)->max_freeable < i) \ | |
280 | (x)->max_freeable = i; \ | |
281 | } while (0) | |
1da177e4 LT |
282 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
283 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
284 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
285 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
286 | #else | |
287 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
288 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
289 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
290 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 291 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
292 | #define STATS_SET_HIGH(x) do { } while (0) |
293 | #define STATS_INC_ERR(x) do { } while (0) | |
294 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 295 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 296 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 297 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
298 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
299 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
300 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
301 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
302 | #endif | |
303 | ||
304 | #if DEBUG | |
1da177e4 | 305 | |
a737b3e2 AM |
306 | /* |
307 | * memory layout of objects: | |
1da177e4 | 308 | * 0 : objp |
3dafccf2 | 309 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
310 | * the end of an object is aligned with the end of the real |
311 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 312 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 313 | * redzone word. |
3dafccf2 | 314 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
315 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
316 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 317 | * [BYTES_PER_WORD long] |
1da177e4 | 318 | */ |
343e0d7a | 319 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 320 | { |
3dafccf2 | 321 | return cachep->obj_offset; |
1da177e4 LT |
322 | } |
323 | ||
b46b8f19 | 324 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
325 | { |
326 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
327 | return (unsigned long long*) (objp + obj_offset(cachep) - |
328 | sizeof(unsigned long long)); | |
1da177e4 LT |
329 | } |
330 | ||
b46b8f19 | 331 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
332 | { |
333 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
334 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 335 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 336 | sizeof(unsigned long long) - |
87a927c7 | 337 | REDZONE_ALIGN); |
3b0efdfa | 338 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 339 | sizeof(unsigned long long)); |
1da177e4 LT |
340 | } |
341 | ||
343e0d7a | 342 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
343 | { |
344 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 345 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
346 | } |
347 | ||
348 | #else | |
349 | ||
3dafccf2 | 350 | #define obj_offset(x) 0 |
b46b8f19 DW |
351 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
352 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
353 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
354 | ||
355 | #endif | |
356 | ||
03787301 JK |
357 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
358 | ||
d31676df | 359 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 360 | { |
d31676df JK |
361 | return atomic_read(&cachep->store_user_clean) == 1; |
362 | } | |
03787301 | 363 | |
d31676df JK |
364 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
365 | { | |
366 | atomic_set(&cachep->store_user_clean, 1); | |
367 | } | |
03787301 | 368 | |
d31676df JK |
369 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
370 | { | |
371 | if (is_store_user_clean(cachep)) | |
372 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
373 | } |
374 | ||
375 | #else | |
d31676df | 376 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
377 | |
378 | #endif | |
379 | ||
1da177e4 | 380 | /* |
3df1cccd DR |
381 | * Do not go above this order unless 0 objects fit into the slab or |
382 | * overridden on the command line. | |
1da177e4 | 383 | */ |
543585cc DR |
384 | #define SLAB_MAX_ORDER_HI 1 |
385 | #define SLAB_MAX_ORDER_LO 0 | |
386 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 387 | static bool slab_max_order_set __initdata; |
1da177e4 | 388 | |
6ed5eb22 PE |
389 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
390 | { | |
b49af68f | 391 | struct page *page = virt_to_head_page(obj); |
35026088 | 392 | return page->slab_cache; |
6ed5eb22 PE |
393 | } |
394 | ||
8456a648 | 395 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
396 | unsigned int idx) |
397 | { | |
8456a648 | 398 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
399 | } |
400 | ||
6a2d7a95 | 401 | /* |
3b0efdfa CL |
402 | * We want to avoid an expensive divide : (offset / cache->size) |
403 | * Using the fact that size is a constant for a particular cache, | |
404 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
405 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
406 | */ | |
407 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 408 | const struct page *page, void *obj) |
8fea4e96 | 409 | { |
8456a648 | 410 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 411 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
412 | } |
413 | ||
6fb92430 | 414 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 415 | /* internal cache of cache description objs */ |
9b030cb8 | 416 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
417 | .batchcount = 1, |
418 | .limit = BOOT_CPUCACHE_ENTRIES, | |
419 | .shared = 1, | |
3b0efdfa | 420 | .size = sizeof(struct kmem_cache), |
b28a02de | 421 | .name = "kmem_cache", |
1da177e4 LT |
422 | }; |
423 | ||
edcad250 JK |
424 | #define BAD_ALIEN_MAGIC 0x01020304ul |
425 | ||
1871e52c | 426 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 427 | |
343e0d7a | 428 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 429 | { |
bf0dea23 | 430 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
431 | } |
432 | ||
a737b3e2 AM |
433 | /* |
434 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
435 | */ | |
70f75067 JK |
436 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
437 | unsigned long flags, size_t *left_over) | |
fbaccacf | 438 | { |
70f75067 | 439 | unsigned int num; |
fbaccacf | 440 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 441 | |
fbaccacf SR |
442 | /* |
443 | * The slab management structure can be either off the slab or | |
444 | * on it. For the latter case, the memory allocated for a | |
445 | * slab is used for: | |
446 | * | |
fbaccacf | 447 | * - @buffer_size bytes for each object |
2e6b3602 JK |
448 | * - One freelist_idx_t for each object |
449 | * | |
450 | * We don't need to consider alignment of freelist because | |
451 | * freelist will be at the end of slab page. The objects will be | |
452 | * at the correct alignment. | |
fbaccacf SR |
453 | * |
454 | * If the slab management structure is off the slab, then the | |
455 | * alignment will already be calculated into the size. Because | |
456 | * the slabs are all pages aligned, the objects will be at the | |
457 | * correct alignment when allocated. | |
458 | */ | |
b03a017b | 459 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 460 | num = slab_size / buffer_size; |
2e6b3602 | 461 | *left_over = slab_size % buffer_size; |
fbaccacf | 462 | } else { |
70f75067 | 463 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
464 | *left_over = slab_size % |
465 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 466 | } |
70f75067 JK |
467 | |
468 | return num; | |
1da177e4 LT |
469 | } |
470 | ||
f28510d3 | 471 | #if DEBUG |
d40cee24 | 472 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 473 | |
a737b3e2 AM |
474 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
475 | char *msg) | |
1da177e4 | 476 | { |
1170532b | 477 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 478 | function, cachep->name, msg); |
1da177e4 | 479 | dump_stack(); |
373d4d09 | 480 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 481 | } |
f28510d3 | 482 | #endif |
1da177e4 | 483 | |
3395ee05 PM |
484 | /* |
485 | * By default on NUMA we use alien caches to stage the freeing of | |
486 | * objects allocated from other nodes. This causes massive memory | |
487 | * inefficiencies when using fake NUMA setup to split memory into a | |
488 | * large number of small nodes, so it can be disabled on the command | |
489 | * line | |
490 | */ | |
491 | ||
492 | static int use_alien_caches __read_mostly = 1; | |
493 | static int __init noaliencache_setup(char *s) | |
494 | { | |
495 | use_alien_caches = 0; | |
496 | return 1; | |
497 | } | |
498 | __setup("noaliencache", noaliencache_setup); | |
499 | ||
3df1cccd DR |
500 | static int __init slab_max_order_setup(char *str) |
501 | { | |
502 | get_option(&str, &slab_max_order); | |
503 | slab_max_order = slab_max_order < 0 ? 0 : | |
504 | min(slab_max_order, MAX_ORDER - 1); | |
505 | slab_max_order_set = true; | |
506 | ||
507 | return 1; | |
508 | } | |
509 | __setup("slab_max_order=", slab_max_order_setup); | |
510 | ||
8fce4d8e CL |
511 | #ifdef CONFIG_NUMA |
512 | /* | |
513 | * Special reaping functions for NUMA systems called from cache_reap(). | |
514 | * These take care of doing round robin flushing of alien caches (containing | |
515 | * objects freed on different nodes from which they were allocated) and the | |
516 | * flushing of remote pcps by calling drain_node_pages. | |
517 | */ | |
1871e52c | 518 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
519 | |
520 | static void init_reap_node(int cpu) | |
521 | { | |
522 | int node; | |
523 | ||
7d6e6d09 | 524 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 525 | if (node == MAX_NUMNODES) |
442295c9 | 526 | node = first_node(node_online_map); |
8fce4d8e | 527 | |
1871e52c | 528 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
529 | } |
530 | ||
531 | static void next_reap_node(void) | |
532 | { | |
909ea964 | 533 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 534 | |
8fce4d8e CL |
535 | node = next_node(node, node_online_map); |
536 | if (unlikely(node >= MAX_NUMNODES)) | |
537 | node = first_node(node_online_map); | |
909ea964 | 538 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
539 | } |
540 | ||
541 | #else | |
542 | #define init_reap_node(cpu) do { } while (0) | |
543 | #define next_reap_node(void) do { } while (0) | |
544 | #endif | |
545 | ||
1da177e4 LT |
546 | /* |
547 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
548 | * via the workqueue/eventd. | |
549 | * Add the CPU number into the expiration time to minimize the possibility of | |
550 | * the CPUs getting into lockstep and contending for the global cache chain | |
551 | * lock. | |
552 | */ | |
0db0628d | 553 | static void start_cpu_timer(int cpu) |
1da177e4 | 554 | { |
1871e52c | 555 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
556 | |
557 | /* | |
558 | * When this gets called from do_initcalls via cpucache_init(), | |
559 | * init_workqueues() has already run, so keventd will be setup | |
560 | * at that time. | |
561 | */ | |
52bad64d | 562 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 563 | init_reap_node(cpu); |
203b42f7 | 564 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
565 | schedule_delayed_work_on(cpu, reap_work, |
566 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
567 | } |
568 | } | |
569 | ||
1fe00d50 | 570 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 571 | { |
d5cff635 CM |
572 | /* |
573 | * The array_cache structures contain pointers to free object. | |
25985edc | 574 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
575 | * cache the pointers are not cleared and they could be counted as |
576 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
577 | * not scan such objects. | |
578 | */ | |
1fe00d50 JK |
579 | kmemleak_no_scan(ac); |
580 | if (ac) { | |
581 | ac->avail = 0; | |
582 | ac->limit = limit; | |
583 | ac->batchcount = batch; | |
584 | ac->touched = 0; | |
1da177e4 | 585 | } |
1fe00d50 JK |
586 | } |
587 | ||
588 | static struct array_cache *alloc_arraycache(int node, int entries, | |
589 | int batchcount, gfp_t gfp) | |
590 | { | |
5e804789 | 591 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
592 | struct array_cache *ac = NULL; |
593 | ||
594 | ac = kmalloc_node(memsize, gfp, node); | |
595 | init_arraycache(ac, entries, batchcount); | |
596 | return ac; | |
1da177e4 LT |
597 | } |
598 | ||
f68f8ddd JK |
599 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
600 | struct page *page, void *objp) | |
072bb0aa | 601 | { |
f68f8ddd JK |
602 | struct kmem_cache_node *n; |
603 | int page_node; | |
604 | LIST_HEAD(list); | |
072bb0aa | 605 | |
f68f8ddd JK |
606 | page_node = page_to_nid(page); |
607 | n = get_node(cachep, page_node); | |
381760ea | 608 | |
f68f8ddd JK |
609 | spin_lock(&n->list_lock); |
610 | free_block(cachep, &objp, 1, page_node, &list); | |
611 | spin_unlock(&n->list_lock); | |
381760ea | 612 | |
f68f8ddd | 613 | slabs_destroy(cachep, &list); |
072bb0aa MG |
614 | } |
615 | ||
3ded175a CL |
616 | /* |
617 | * Transfer objects in one arraycache to another. | |
618 | * Locking must be handled by the caller. | |
619 | * | |
620 | * Return the number of entries transferred. | |
621 | */ | |
622 | static int transfer_objects(struct array_cache *to, | |
623 | struct array_cache *from, unsigned int max) | |
624 | { | |
625 | /* Figure out how many entries to transfer */ | |
732eacc0 | 626 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
627 | |
628 | if (!nr) | |
629 | return 0; | |
630 | ||
631 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
632 | sizeof(void *) *nr); | |
633 | ||
634 | from->avail -= nr; | |
635 | to->avail += nr; | |
3ded175a CL |
636 | return nr; |
637 | } | |
638 | ||
765c4507 CL |
639 | #ifndef CONFIG_NUMA |
640 | ||
641 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 642 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 643 | |
c8522a3a JK |
644 | static inline struct alien_cache **alloc_alien_cache(int node, |
645 | int limit, gfp_t gfp) | |
765c4507 | 646 | { |
edcad250 | 647 | return (struct alien_cache **)BAD_ALIEN_MAGIC; |
765c4507 CL |
648 | } |
649 | ||
c8522a3a | 650 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
651 | { |
652 | } | |
653 | ||
654 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
655 | { | |
656 | return 0; | |
657 | } | |
658 | ||
659 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
660 | gfp_t flags) | |
661 | { | |
662 | return NULL; | |
663 | } | |
664 | ||
8b98c169 | 665 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
666 | gfp_t flags, int nodeid) |
667 | { | |
668 | return NULL; | |
669 | } | |
670 | ||
4167e9b2 DR |
671 | static inline gfp_t gfp_exact_node(gfp_t flags) |
672 | { | |
444eb2a4 | 673 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
674 | } |
675 | ||
765c4507 CL |
676 | #else /* CONFIG_NUMA */ |
677 | ||
8b98c169 | 678 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 679 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 680 | |
c8522a3a JK |
681 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
682 | int batch, gfp_t gfp) | |
683 | { | |
5e804789 | 684 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
685 | struct alien_cache *alc = NULL; |
686 | ||
687 | alc = kmalloc_node(memsize, gfp, node); | |
688 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 689 | spin_lock_init(&alc->lock); |
c8522a3a JK |
690 | return alc; |
691 | } | |
692 | ||
693 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 694 | { |
c8522a3a | 695 | struct alien_cache **alc_ptr; |
5e804789 | 696 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
697 | int i; |
698 | ||
699 | if (limit > 1) | |
700 | limit = 12; | |
c8522a3a JK |
701 | alc_ptr = kzalloc_node(memsize, gfp, node); |
702 | if (!alc_ptr) | |
703 | return NULL; | |
704 | ||
705 | for_each_node(i) { | |
706 | if (i == node || !node_online(i)) | |
707 | continue; | |
708 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
709 | if (!alc_ptr[i]) { | |
710 | for (i--; i >= 0; i--) | |
711 | kfree(alc_ptr[i]); | |
712 | kfree(alc_ptr); | |
713 | return NULL; | |
e498be7d CL |
714 | } |
715 | } | |
c8522a3a | 716 | return alc_ptr; |
e498be7d CL |
717 | } |
718 | ||
c8522a3a | 719 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
720 | { |
721 | int i; | |
722 | ||
c8522a3a | 723 | if (!alc_ptr) |
e498be7d | 724 | return; |
e498be7d | 725 | for_each_node(i) |
c8522a3a JK |
726 | kfree(alc_ptr[i]); |
727 | kfree(alc_ptr); | |
e498be7d CL |
728 | } |
729 | ||
343e0d7a | 730 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
731 | struct array_cache *ac, int node, |
732 | struct list_head *list) | |
e498be7d | 733 | { |
18bf8541 | 734 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
735 | |
736 | if (ac->avail) { | |
ce8eb6c4 | 737 | spin_lock(&n->list_lock); |
e00946fe CL |
738 | /* |
739 | * Stuff objects into the remote nodes shared array first. | |
740 | * That way we could avoid the overhead of putting the objects | |
741 | * into the free lists and getting them back later. | |
742 | */ | |
ce8eb6c4 CL |
743 | if (n->shared) |
744 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 745 | |
833b706c | 746 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 747 | ac->avail = 0; |
ce8eb6c4 | 748 | spin_unlock(&n->list_lock); |
e498be7d CL |
749 | } |
750 | } | |
751 | ||
8fce4d8e CL |
752 | /* |
753 | * Called from cache_reap() to regularly drain alien caches round robin. | |
754 | */ | |
ce8eb6c4 | 755 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 756 | { |
909ea964 | 757 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 758 | |
ce8eb6c4 | 759 | if (n->alien) { |
c8522a3a JK |
760 | struct alien_cache *alc = n->alien[node]; |
761 | struct array_cache *ac; | |
762 | ||
763 | if (alc) { | |
764 | ac = &alc->ac; | |
49dfc304 | 765 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
766 | LIST_HEAD(list); |
767 | ||
768 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 769 | spin_unlock_irq(&alc->lock); |
833b706c | 770 | slabs_destroy(cachep, &list); |
c8522a3a | 771 | } |
8fce4d8e CL |
772 | } |
773 | } | |
774 | } | |
775 | ||
a737b3e2 | 776 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 777 | struct alien_cache **alien) |
e498be7d | 778 | { |
b28a02de | 779 | int i = 0; |
c8522a3a | 780 | struct alien_cache *alc; |
e498be7d CL |
781 | struct array_cache *ac; |
782 | unsigned long flags; | |
783 | ||
784 | for_each_online_node(i) { | |
c8522a3a JK |
785 | alc = alien[i]; |
786 | if (alc) { | |
833b706c JK |
787 | LIST_HEAD(list); |
788 | ||
c8522a3a | 789 | ac = &alc->ac; |
49dfc304 | 790 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 791 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 792 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 793 | slabs_destroy(cachep, &list); |
e498be7d CL |
794 | } |
795 | } | |
796 | } | |
729bd0b7 | 797 | |
25c4f304 JK |
798 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
799 | int node, int page_node) | |
729bd0b7 | 800 | { |
ce8eb6c4 | 801 | struct kmem_cache_node *n; |
c8522a3a JK |
802 | struct alien_cache *alien = NULL; |
803 | struct array_cache *ac; | |
97654dfa | 804 | LIST_HEAD(list); |
1ca4cb24 | 805 | |
18bf8541 | 806 | n = get_node(cachep, node); |
729bd0b7 | 807 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
808 | if (n->alien && n->alien[page_node]) { |
809 | alien = n->alien[page_node]; | |
c8522a3a | 810 | ac = &alien->ac; |
49dfc304 | 811 | spin_lock(&alien->lock); |
c8522a3a | 812 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 813 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 814 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 815 | } |
f68f8ddd | 816 | ac->entry[ac->avail++] = objp; |
49dfc304 | 817 | spin_unlock(&alien->lock); |
833b706c | 818 | slabs_destroy(cachep, &list); |
729bd0b7 | 819 | } else { |
25c4f304 | 820 | n = get_node(cachep, page_node); |
18bf8541 | 821 | spin_lock(&n->list_lock); |
25c4f304 | 822 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 823 | spin_unlock(&n->list_lock); |
97654dfa | 824 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
825 | } |
826 | return 1; | |
827 | } | |
25c4f304 JK |
828 | |
829 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
830 | { | |
831 | int page_node = page_to_nid(virt_to_page(objp)); | |
832 | int node = numa_mem_id(); | |
833 | /* | |
834 | * Make sure we are not freeing a object from another node to the array | |
835 | * cache on this cpu. | |
836 | */ | |
837 | if (likely(node == page_node)) | |
838 | return 0; | |
839 | ||
840 | return __cache_free_alien(cachep, objp, node, page_node); | |
841 | } | |
4167e9b2 DR |
842 | |
843 | /* | |
444eb2a4 MG |
844 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
845 | * warn about failures. | |
4167e9b2 DR |
846 | */ |
847 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
848 | { | |
444eb2a4 | 849 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 850 | } |
e498be7d CL |
851 | #endif |
852 | ||
8f9f8d9e | 853 | /* |
6a67368c | 854 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 855 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 856 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 857 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
858 | * already in use. |
859 | * | |
18004c5d | 860 | * Must hold slab_mutex. |
8f9f8d9e | 861 | */ |
6a67368c | 862 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
863 | { |
864 | struct kmem_cache *cachep; | |
ce8eb6c4 | 865 | struct kmem_cache_node *n; |
5e804789 | 866 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 867 | |
18004c5d | 868 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 869 | /* |
5f0985bb | 870 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
871 | * begin anything. Make sure some other cpu on this |
872 | * node has not already allocated this | |
873 | */ | |
18bf8541 CL |
874 | n = get_node(cachep, node); |
875 | if (!n) { | |
ce8eb6c4 CL |
876 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
877 | if (!n) | |
8f9f8d9e | 878 | return -ENOMEM; |
ce8eb6c4 | 879 | kmem_cache_node_init(n); |
5f0985bb JZ |
880 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
881 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
882 | |
883 | /* | |
5f0985bb JZ |
884 | * The kmem_cache_nodes don't come and go as CPUs |
885 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
886 | * protection here. |
887 | */ | |
ce8eb6c4 | 888 | cachep->node[node] = n; |
8f9f8d9e DR |
889 | } |
890 | ||
18bf8541 CL |
891 | spin_lock_irq(&n->list_lock); |
892 | n->free_limit = | |
8f9f8d9e DR |
893 | (1 + nr_cpus_node(node)) * |
894 | cachep->batchcount + cachep->num; | |
18bf8541 | 895 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
896 | } |
897 | return 0; | |
898 | } | |
899 | ||
0fa8103b WL |
900 | static inline int slabs_tofree(struct kmem_cache *cachep, |
901 | struct kmem_cache_node *n) | |
902 | { | |
903 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
904 | } | |
905 | ||
0db0628d | 906 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
907 | { |
908 | struct kmem_cache *cachep; | |
ce8eb6c4 | 909 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 910 | int node = cpu_to_mem(cpu); |
a70f7302 | 911 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 912 | |
18004c5d | 913 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
914 | struct array_cache *nc; |
915 | struct array_cache *shared; | |
c8522a3a | 916 | struct alien_cache **alien; |
97654dfa | 917 | LIST_HEAD(list); |
fbf1e473 | 918 | |
18bf8541 | 919 | n = get_node(cachep, node); |
ce8eb6c4 | 920 | if (!n) |
bf0dea23 | 921 | continue; |
fbf1e473 | 922 | |
ce8eb6c4 | 923 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 924 | |
ce8eb6c4 CL |
925 | /* Free limit for this kmem_cache_node */ |
926 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
927 | |
928 | /* cpu is dead; no one can alloc from it. */ | |
929 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
930 | if (nc) { | |
97654dfa | 931 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
932 | nc->avail = 0; |
933 | } | |
fbf1e473 | 934 | |
58463c1f | 935 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 936 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 937 | goto free_slab; |
fbf1e473 AM |
938 | } |
939 | ||
ce8eb6c4 | 940 | shared = n->shared; |
fbf1e473 AM |
941 | if (shared) { |
942 | free_block(cachep, shared->entry, | |
97654dfa | 943 | shared->avail, node, &list); |
ce8eb6c4 | 944 | n->shared = NULL; |
fbf1e473 AM |
945 | } |
946 | ||
ce8eb6c4 CL |
947 | alien = n->alien; |
948 | n->alien = NULL; | |
fbf1e473 | 949 | |
ce8eb6c4 | 950 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
951 | |
952 | kfree(shared); | |
953 | if (alien) { | |
954 | drain_alien_cache(cachep, alien); | |
955 | free_alien_cache(alien); | |
956 | } | |
bf0dea23 JK |
957 | |
958 | free_slab: | |
97654dfa | 959 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
960 | } |
961 | /* | |
962 | * In the previous loop, all the objects were freed to | |
963 | * the respective cache's slabs, now we can go ahead and | |
964 | * shrink each nodelist to its limit. | |
965 | */ | |
18004c5d | 966 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 967 | n = get_node(cachep, node); |
ce8eb6c4 | 968 | if (!n) |
fbf1e473 | 969 | continue; |
0fa8103b | 970 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
971 | } |
972 | } | |
973 | ||
0db0628d | 974 | static int cpuup_prepare(long cpu) |
1da177e4 | 975 | { |
343e0d7a | 976 | struct kmem_cache *cachep; |
ce8eb6c4 | 977 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 978 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 979 | int err; |
1da177e4 | 980 | |
fbf1e473 AM |
981 | /* |
982 | * We need to do this right in the beginning since | |
983 | * alloc_arraycache's are going to use this list. | |
984 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 985 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 986 | */ |
6a67368c | 987 | err = init_cache_node_node(node); |
8f9f8d9e DR |
988 | if (err < 0) |
989 | goto bad; | |
fbf1e473 AM |
990 | |
991 | /* | |
992 | * Now we can go ahead with allocating the shared arrays and | |
993 | * array caches | |
994 | */ | |
18004c5d | 995 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 996 | struct array_cache *shared = NULL; |
c8522a3a | 997 | struct alien_cache **alien = NULL; |
fbf1e473 | 998 | |
fbf1e473 AM |
999 | if (cachep->shared) { |
1000 | shared = alloc_arraycache(node, | |
1001 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1002 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1003 | if (!shared) |
1da177e4 | 1004 | goto bad; |
fbf1e473 AM |
1005 | } |
1006 | if (use_alien_caches) { | |
83b519e8 | 1007 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1008 | if (!alien) { |
1009 | kfree(shared); | |
fbf1e473 | 1010 | goto bad; |
12d00f6a | 1011 | } |
fbf1e473 | 1012 | } |
18bf8541 | 1013 | n = get_node(cachep, node); |
ce8eb6c4 | 1014 | BUG_ON(!n); |
fbf1e473 | 1015 | |
ce8eb6c4 CL |
1016 | spin_lock_irq(&n->list_lock); |
1017 | if (!n->shared) { | |
fbf1e473 AM |
1018 | /* |
1019 | * We are serialised from CPU_DEAD or | |
1020 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1021 | */ | |
ce8eb6c4 | 1022 | n->shared = shared; |
fbf1e473 AM |
1023 | shared = NULL; |
1024 | } | |
4484ebf1 | 1025 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1026 | if (!n->alien) { |
1027 | n->alien = alien; | |
fbf1e473 | 1028 | alien = NULL; |
1da177e4 | 1029 | } |
fbf1e473 | 1030 | #endif |
ce8eb6c4 | 1031 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1032 | kfree(shared); |
1033 | free_alien_cache(alien); | |
1034 | } | |
ce79ddc8 | 1035 | |
fbf1e473 AM |
1036 | return 0; |
1037 | bad: | |
12d00f6a | 1038 | cpuup_canceled(cpu); |
fbf1e473 AM |
1039 | return -ENOMEM; |
1040 | } | |
1041 | ||
0db0628d | 1042 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1043 | unsigned long action, void *hcpu) |
1044 | { | |
1045 | long cpu = (long)hcpu; | |
1046 | int err = 0; | |
1047 | ||
1048 | switch (action) { | |
fbf1e473 AM |
1049 | case CPU_UP_PREPARE: |
1050 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1051 | mutex_lock(&slab_mutex); |
fbf1e473 | 1052 | err = cpuup_prepare(cpu); |
18004c5d | 1053 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1054 | break; |
1055 | case CPU_ONLINE: | |
8bb78442 | 1056 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1057 | start_cpu_timer(cpu); |
1058 | break; | |
1059 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1060 | case CPU_DOWN_PREPARE: |
8bb78442 | 1061 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1062 | /* |
18004c5d | 1063 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1064 | * held so that if cache_reap() is invoked it cannot do |
1065 | * anything expensive but will only modify reap_work | |
1066 | * and reschedule the timer. | |
1067 | */ | |
afe2c511 | 1068 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1069 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1070 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1071 | break; |
1072 | case CPU_DOWN_FAILED: | |
8bb78442 | 1073 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1074 | start_cpu_timer(cpu); |
1075 | break; | |
1da177e4 | 1076 | case CPU_DEAD: |
8bb78442 | 1077 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1078 | /* |
1079 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1080 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1081 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1082 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1083 | * structure is usually allocated from kmem_cache_create() and |
1084 | * gets destroyed at kmem_cache_destroy(). | |
1085 | */ | |
183ff22b | 1086 | /* fall through */ |
8f5be20b | 1087 | #endif |
1da177e4 | 1088 | case CPU_UP_CANCELED: |
8bb78442 | 1089 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1090 | mutex_lock(&slab_mutex); |
fbf1e473 | 1091 | cpuup_canceled(cpu); |
18004c5d | 1092 | mutex_unlock(&slab_mutex); |
1da177e4 | 1093 | break; |
1da177e4 | 1094 | } |
eac40680 | 1095 | return notifier_from_errno(err); |
1da177e4 LT |
1096 | } |
1097 | ||
0db0628d | 1098 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1099 | &cpuup_callback, NULL, 0 |
1100 | }; | |
1da177e4 | 1101 | |
8f9f8d9e DR |
1102 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1103 | /* | |
1104 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1105 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1106 | * removed. | |
1107 | * | |
18004c5d | 1108 | * Must hold slab_mutex. |
8f9f8d9e | 1109 | */ |
6a67368c | 1110 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1111 | { |
1112 | struct kmem_cache *cachep; | |
1113 | int ret = 0; | |
1114 | ||
18004c5d | 1115 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1116 | struct kmem_cache_node *n; |
8f9f8d9e | 1117 | |
18bf8541 | 1118 | n = get_node(cachep, node); |
ce8eb6c4 | 1119 | if (!n) |
8f9f8d9e DR |
1120 | continue; |
1121 | ||
0fa8103b | 1122 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1123 | |
ce8eb6c4 CL |
1124 | if (!list_empty(&n->slabs_full) || |
1125 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1126 | ret = -EBUSY; |
1127 | break; | |
1128 | } | |
1129 | } | |
1130 | return ret; | |
1131 | } | |
1132 | ||
1133 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1134 | unsigned long action, void *arg) | |
1135 | { | |
1136 | struct memory_notify *mnb = arg; | |
1137 | int ret = 0; | |
1138 | int nid; | |
1139 | ||
1140 | nid = mnb->status_change_nid; | |
1141 | if (nid < 0) | |
1142 | goto out; | |
1143 | ||
1144 | switch (action) { | |
1145 | case MEM_GOING_ONLINE: | |
18004c5d | 1146 | mutex_lock(&slab_mutex); |
6a67368c | 1147 | ret = init_cache_node_node(nid); |
18004c5d | 1148 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1149 | break; |
1150 | case MEM_GOING_OFFLINE: | |
18004c5d | 1151 | mutex_lock(&slab_mutex); |
6a67368c | 1152 | ret = drain_cache_node_node(nid); |
18004c5d | 1153 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1154 | break; |
1155 | case MEM_ONLINE: | |
1156 | case MEM_OFFLINE: | |
1157 | case MEM_CANCEL_ONLINE: | |
1158 | case MEM_CANCEL_OFFLINE: | |
1159 | break; | |
1160 | } | |
1161 | out: | |
5fda1bd5 | 1162 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1163 | } |
1164 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1165 | ||
e498be7d | 1166 | /* |
ce8eb6c4 | 1167 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1168 | */ |
6744f087 | 1169 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1170 | int nodeid) |
e498be7d | 1171 | { |
6744f087 | 1172 | struct kmem_cache_node *ptr; |
e498be7d | 1173 | |
6744f087 | 1174 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1175 | BUG_ON(!ptr); |
1176 | ||
6744f087 | 1177 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1178 | /* |
1179 | * Do not assume that spinlocks can be initialized via memcpy: | |
1180 | */ | |
1181 | spin_lock_init(&ptr->list_lock); | |
1182 | ||
e498be7d | 1183 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1184 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1185 | } |
1186 | ||
556a169d | 1187 | /* |
ce8eb6c4 CL |
1188 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1189 | * size of kmem_cache_node. | |
556a169d | 1190 | */ |
ce8eb6c4 | 1191 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1192 | { |
1193 | int node; | |
1194 | ||
1195 | for_each_online_node(node) { | |
ce8eb6c4 | 1196 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1197 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1198 | REAPTIMEOUT_NODE + |
1199 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1200 | } |
1201 | } | |
1202 | ||
a737b3e2 AM |
1203 | /* |
1204 | * Initialisation. Called after the page allocator have been initialised and | |
1205 | * before smp_init(). | |
1da177e4 LT |
1206 | */ |
1207 | void __init kmem_cache_init(void) | |
1208 | { | |
e498be7d CL |
1209 | int i; |
1210 | ||
68126702 JK |
1211 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1212 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1213 | kmem_cache = &kmem_cache_boot; |
1214 | ||
b6e68bc1 | 1215 | if (num_possible_nodes() == 1) |
62918a03 SS |
1216 | use_alien_caches = 0; |
1217 | ||
3c583465 | 1218 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1219 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1220 | |
1da177e4 LT |
1221 | /* |
1222 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1223 | * page orders on machines with more than 32MB of memory if |
1224 | * not overridden on the command line. | |
1da177e4 | 1225 | */ |
3df1cccd | 1226 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1227 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1228 | |
1da177e4 LT |
1229 | /* Bootstrap is tricky, because several objects are allocated |
1230 | * from caches that do not exist yet: | |
9b030cb8 CL |
1231 | * 1) initialize the kmem_cache cache: it contains the struct |
1232 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1233 | * kmem_cache is statically allocated. | |
e498be7d | 1234 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1235 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1236 | * array at the end of the bootstrap. |
1da177e4 | 1237 | * 2) Create the first kmalloc cache. |
343e0d7a | 1238 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1239 | * An __init data area is used for the head array. |
1240 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1241 | * head arrays. | |
9b030cb8 | 1242 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1243 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1244 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1245 | * the other cache's with kmalloc allocated memory. |
1246 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1247 | */ |
1248 | ||
9b030cb8 | 1249 | /* 1) create the kmem_cache */ |
1da177e4 | 1250 | |
8da3430d | 1251 | /* |
b56efcf0 | 1252 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1253 | */ |
2f9baa9f | 1254 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1255 | offsetof(struct kmem_cache, node) + |
6744f087 | 1256 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1257 | SLAB_HWCACHE_ALIGN); |
1258 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1259 | slab_state = PARTIAL; |
1da177e4 | 1260 | |
a737b3e2 | 1261 | /* |
bf0dea23 JK |
1262 | * Initialize the caches that provide memory for the kmem_cache_node |
1263 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1264 | */ |
bf0dea23 | 1265 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1266 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1267 | slab_state = PARTIAL_NODE; |
34cc6990 | 1268 | setup_kmalloc_cache_index_table(); |
e498be7d | 1269 | |
e0a42726 IM |
1270 | slab_early_init = 0; |
1271 | ||
ce8eb6c4 | 1272 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1273 | { |
1ca4cb24 PE |
1274 | int nid; |
1275 | ||
9c09a95c | 1276 | for_each_online_node(nid) { |
ce8eb6c4 | 1277 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1278 | |
bf0dea23 | 1279 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1280 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1281 | } |
1282 | } | |
1da177e4 | 1283 | |
f97d5f63 | 1284 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1285 | } |
1286 | ||
1287 | void __init kmem_cache_init_late(void) | |
1288 | { | |
1289 | struct kmem_cache *cachep; | |
1290 | ||
97d06609 | 1291 | slab_state = UP; |
52cef189 | 1292 | |
8429db5c | 1293 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1294 | mutex_lock(&slab_mutex); |
1295 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1296 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1297 | BUG(); | |
18004c5d | 1298 | mutex_unlock(&slab_mutex); |
056c6241 | 1299 | |
97d06609 CL |
1300 | /* Done! */ |
1301 | slab_state = FULL; | |
1302 | ||
a737b3e2 AM |
1303 | /* |
1304 | * Register a cpu startup notifier callback that initializes | |
1305 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1306 | */ |
1307 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1308 | |
8f9f8d9e DR |
1309 | #ifdef CONFIG_NUMA |
1310 | /* | |
1311 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1312 | * node. |
8f9f8d9e DR |
1313 | */ |
1314 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1315 | #endif | |
1316 | ||
a737b3e2 AM |
1317 | /* |
1318 | * The reap timers are started later, with a module init call: That part | |
1319 | * of the kernel is not yet operational. | |
1da177e4 LT |
1320 | */ |
1321 | } | |
1322 | ||
1323 | static int __init cpucache_init(void) | |
1324 | { | |
1325 | int cpu; | |
1326 | ||
a737b3e2 AM |
1327 | /* |
1328 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1329 | */ |
e498be7d | 1330 | for_each_online_cpu(cpu) |
a737b3e2 | 1331 | start_cpu_timer(cpu); |
a164f896 GC |
1332 | |
1333 | /* Done! */ | |
97d06609 | 1334 | slab_state = FULL; |
1da177e4 LT |
1335 | return 0; |
1336 | } | |
1da177e4 LT |
1337 | __initcall(cpucache_init); |
1338 | ||
8bdec192 RA |
1339 | static noinline void |
1340 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1341 | { | |
9a02d699 | 1342 | #if DEBUG |
ce8eb6c4 | 1343 | struct kmem_cache_node *n; |
8456a648 | 1344 | struct page *page; |
8bdec192 RA |
1345 | unsigned long flags; |
1346 | int node; | |
9a02d699 DR |
1347 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1348 | DEFAULT_RATELIMIT_BURST); | |
1349 | ||
1350 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1351 | return; | |
8bdec192 | 1352 | |
5b3810e5 VB |
1353 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1354 | nodeid, gfpflags, &gfpflags); | |
1355 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1356 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1357 | |
18bf8541 | 1358 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1359 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1360 | unsigned long active_slabs = 0, num_slabs = 0; | |
1361 | ||
ce8eb6c4 | 1362 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1363 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1364 | active_objs += cachep->num; |
1365 | active_slabs++; | |
1366 | } | |
8456a648 JK |
1367 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1368 | active_objs += page->active; | |
8bdec192 RA |
1369 | active_slabs++; |
1370 | } | |
8456a648 | 1371 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1372 | num_slabs++; |
1373 | ||
ce8eb6c4 CL |
1374 | free_objects += n->free_objects; |
1375 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1376 | |
1377 | num_slabs += active_slabs; | |
1378 | num_objs = num_slabs * cachep->num; | |
5b3810e5 | 1379 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", |
8bdec192 RA |
1380 | node, active_slabs, num_slabs, active_objs, num_objs, |
1381 | free_objects); | |
1382 | } | |
9a02d699 | 1383 | #endif |
8bdec192 RA |
1384 | } |
1385 | ||
1da177e4 | 1386 | /* |
8a7d9b43 WSH |
1387 | * Interface to system's page allocator. No need to hold the |
1388 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1389 | * |
1390 | * If we requested dmaable memory, we will get it. Even if we | |
1391 | * did not request dmaable memory, we might get it, but that | |
1392 | * would be relatively rare and ignorable. | |
1393 | */ | |
0c3aa83e JK |
1394 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1395 | int nodeid) | |
1da177e4 LT |
1396 | { |
1397 | struct page *page; | |
e1b6aa6f | 1398 | int nr_pages; |
765c4507 | 1399 | |
a618e89f | 1400 | flags |= cachep->allocflags; |
e12ba74d MG |
1401 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1402 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1403 | |
96db800f | 1404 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1405 | if (!page) { |
9a02d699 | 1406 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1407 | return NULL; |
8bdec192 | 1408 | } |
1da177e4 | 1409 | |
f3ccb2c4 VD |
1410 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1411 | __free_pages(page, cachep->gfporder); | |
1412 | return NULL; | |
1413 | } | |
1414 | ||
e1b6aa6f | 1415 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1416 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1417 | add_zone_page_state(page_zone(page), |
1418 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1419 | else | |
1420 | add_zone_page_state(page_zone(page), | |
1421 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
f68f8ddd | 1422 | |
a57a4988 | 1423 | __SetPageSlab(page); |
f68f8ddd JK |
1424 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1425 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1426 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1427 | |
b1eeab67 VN |
1428 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1429 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1430 | ||
1431 | if (cachep->ctor) | |
1432 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1433 | else | |
1434 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1435 | } | |
c175eea4 | 1436 | |
0c3aa83e | 1437 | return page; |
1da177e4 LT |
1438 | } |
1439 | ||
1440 | /* | |
1441 | * Interface to system's page release. | |
1442 | */ | |
0c3aa83e | 1443 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1444 | { |
27ee57c9 VD |
1445 | int order = cachep->gfporder; |
1446 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1447 | |
27ee57c9 | 1448 | kmemcheck_free_shadow(page, order); |
c175eea4 | 1449 | |
972d1a7b CL |
1450 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1451 | sub_zone_page_state(page_zone(page), | |
1452 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1453 | else | |
1454 | sub_zone_page_state(page_zone(page), | |
1455 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1456 | |
a57a4988 | 1457 | BUG_ON(!PageSlab(page)); |
73293c2f | 1458 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1459 | __ClearPageSlab(page); |
8456a648 JK |
1460 | page_mapcount_reset(page); |
1461 | page->mapping = NULL; | |
1f458cbf | 1462 | |
1da177e4 LT |
1463 | if (current->reclaim_state) |
1464 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1465 | memcg_uncharge_slab(page, order, cachep); |
1466 | __free_pages(page, order); | |
1da177e4 LT |
1467 | } |
1468 | ||
1469 | static void kmem_rcu_free(struct rcu_head *head) | |
1470 | { | |
68126702 JK |
1471 | struct kmem_cache *cachep; |
1472 | struct page *page; | |
1da177e4 | 1473 | |
68126702 JK |
1474 | page = container_of(head, struct page, rcu_head); |
1475 | cachep = page->slab_cache; | |
1476 | ||
1477 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | #if DEBUG | |
40b44137 JK |
1481 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1482 | { | |
1483 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1484 | (cachep->size % PAGE_SIZE) == 0) | |
1485 | return true; | |
1486 | ||
1487 | return false; | |
1488 | } | |
1da177e4 LT |
1489 | |
1490 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1491 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1492 | unsigned long caller) |
1da177e4 | 1493 | { |
8c138bc0 | 1494 | int size = cachep->object_size; |
1da177e4 | 1495 | |
3dafccf2 | 1496 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1497 | |
b28a02de | 1498 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1499 | return; |
1500 | ||
b28a02de PE |
1501 | *addr++ = 0x12345678; |
1502 | *addr++ = caller; | |
1503 | *addr++ = smp_processor_id(); | |
1504 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1505 | { |
1506 | unsigned long *sptr = &caller; | |
1507 | unsigned long svalue; | |
1508 | ||
1509 | while (!kstack_end(sptr)) { | |
1510 | svalue = *sptr++; | |
1511 | if (kernel_text_address(svalue)) { | |
b28a02de | 1512 | *addr++ = svalue; |
1da177e4 LT |
1513 | size -= sizeof(unsigned long); |
1514 | if (size <= sizeof(unsigned long)) | |
1515 | break; | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | } | |
b28a02de | 1520 | *addr++ = 0x87654321; |
1da177e4 | 1521 | } |
40b44137 JK |
1522 | |
1523 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1524 | int map, unsigned long caller) | |
1525 | { | |
1526 | if (!is_debug_pagealloc_cache(cachep)) | |
1527 | return; | |
1528 | ||
1529 | if (caller) | |
1530 | store_stackinfo(cachep, objp, caller); | |
1531 | ||
1532 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1533 | } | |
1534 | ||
1535 | #else | |
1536 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1537 | int map, unsigned long caller) {} | |
1538 | ||
1da177e4 LT |
1539 | #endif |
1540 | ||
343e0d7a | 1541 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1542 | { |
8c138bc0 | 1543 | int size = cachep->object_size; |
3dafccf2 | 1544 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1545 | |
1546 | memset(addr, val, size); | |
b28a02de | 1547 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1548 | } |
1549 | ||
1550 | static void dump_line(char *data, int offset, int limit) | |
1551 | { | |
1552 | int i; | |
aa83aa40 DJ |
1553 | unsigned char error = 0; |
1554 | int bad_count = 0; | |
1555 | ||
1170532b | 1556 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1557 | for (i = 0; i < limit; i++) { |
1558 | if (data[offset + i] != POISON_FREE) { | |
1559 | error = data[offset + i]; | |
1560 | bad_count++; | |
1561 | } | |
aa83aa40 | 1562 | } |
fdde6abb SAS |
1563 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1564 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1565 | |
1566 | if (bad_count == 1) { | |
1567 | error ^= POISON_FREE; | |
1568 | if (!(error & (error - 1))) { | |
1170532b | 1569 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1570 | #ifdef CONFIG_X86 |
1170532b | 1571 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1572 | #else |
1170532b | 1573 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1574 | #endif |
1575 | } | |
1576 | } | |
1da177e4 LT |
1577 | } |
1578 | #endif | |
1579 | ||
1580 | #if DEBUG | |
1581 | ||
343e0d7a | 1582 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1583 | { |
1584 | int i, size; | |
1585 | char *realobj; | |
1586 | ||
1587 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1588 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1589 | *dbg_redzone1(cachep, objp), | |
1590 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1591 | } |
1592 | ||
1593 | if (cachep->flags & SLAB_STORE_USER) { | |
1170532b | 1594 | pr_err("Last user: [<%p>](%pSR)\n", |
071361d3 JP |
1595 | *dbg_userword(cachep, objp), |
1596 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1597 | } |
3dafccf2 | 1598 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1599 | size = cachep->object_size; |
b28a02de | 1600 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1601 | int limit; |
1602 | limit = 16; | |
b28a02de PE |
1603 | if (i + limit > size) |
1604 | limit = size - i; | |
1da177e4 LT |
1605 | dump_line(realobj, i, limit); |
1606 | } | |
1607 | } | |
1608 | ||
343e0d7a | 1609 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1610 | { |
1611 | char *realobj; | |
1612 | int size, i; | |
1613 | int lines = 0; | |
1614 | ||
40b44137 JK |
1615 | if (is_debug_pagealloc_cache(cachep)) |
1616 | return; | |
1617 | ||
3dafccf2 | 1618 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1619 | size = cachep->object_size; |
1da177e4 | 1620 | |
b28a02de | 1621 | for (i = 0; i < size; i++) { |
1da177e4 | 1622 | char exp = POISON_FREE; |
b28a02de | 1623 | if (i == size - 1) |
1da177e4 LT |
1624 | exp = POISON_END; |
1625 | if (realobj[i] != exp) { | |
1626 | int limit; | |
1627 | /* Mismatch ! */ | |
1628 | /* Print header */ | |
1629 | if (lines == 0) { | |
1170532b JP |
1630 | pr_err("Slab corruption (%s): %s start=%p, len=%d\n", |
1631 | print_tainted(), cachep->name, | |
1632 | realobj, size); | |
1da177e4 LT |
1633 | print_objinfo(cachep, objp, 0); |
1634 | } | |
1635 | /* Hexdump the affected line */ | |
b28a02de | 1636 | i = (i / 16) * 16; |
1da177e4 | 1637 | limit = 16; |
b28a02de PE |
1638 | if (i + limit > size) |
1639 | limit = size - i; | |
1da177e4 LT |
1640 | dump_line(realobj, i, limit); |
1641 | i += 16; | |
1642 | lines++; | |
1643 | /* Limit to 5 lines */ | |
1644 | if (lines > 5) | |
1645 | break; | |
1646 | } | |
1647 | } | |
1648 | if (lines != 0) { | |
1649 | /* Print some data about the neighboring objects, if they | |
1650 | * exist: | |
1651 | */ | |
8456a648 | 1652 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1653 | unsigned int objnr; |
1da177e4 | 1654 | |
8456a648 | 1655 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1656 | if (objnr) { |
8456a648 | 1657 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1658 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1659 | pr_err("Prev obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1660 | print_objinfo(cachep, objp, 2); |
1661 | } | |
b28a02de | 1662 | if (objnr + 1 < cachep->num) { |
8456a648 | 1663 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1664 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1665 | pr_err("Next obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1666 | print_objinfo(cachep, objp, 2); |
1667 | } | |
1668 | } | |
1669 | } | |
1670 | #endif | |
1671 | ||
12dd36fa | 1672 | #if DEBUG |
8456a648 JK |
1673 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1674 | struct page *page) | |
1da177e4 | 1675 | { |
1da177e4 | 1676 | int i; |
b03a017b JK |
1677 | |
1678 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1679 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1680 | POISON_FREE); | |
1681 | } | |
1682 | ||
1da177e4 | 1683 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1684 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1685 | |
1686 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1687 | check_poison_obj(cachep, objp); |
40b44137 | 1688 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1689 | } |
1690 | if (cachep->flags & SLAB_RED_ZONE) { | |
1691 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1692 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1693 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1694 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1695 | } |
1da177e4 | 1696 | } |
12dd36fa | 1697 | } |
1da177e4 | 1698 | #else |
8456a648 JK |
1699 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1700 | struct page *page) | |
12dd36fa | 1701 | { |
12dd36fa | 1702 | } |
1da177e4 LT |
1703 | #endif |
1704 | ||
911851e6 RD |
1705 | /** |
1706 | * slab_destroy - destroy and release all objects in a slab | |
1707 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1708 | * @page: page pointer being destroyed |
911851e6 | 1709 | * |
8a7d9b43 WSH |
1710 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1711 | * Before calling the slab page must have been unlinked from the cache. The | |
1712 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1713 | */ |
8456a648 | 1714 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1715 | { |
7e007355 | 1716 | void *freelist; |
12dd36fa | 1717 | |
8456a648 JK |
1718 | freelist = page->freelist; |
1719 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1720 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1721 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1722 | else | |
0c3aa83e | 1723 | kmem_freepages(cachep, page); |
68126702 JK |
1724 | |
1725 | /* | |
8456a648 | 1726 | * From now on, we don't use freelist |
68126702 JK |
1727 | * although actual page can be freed in rcu context |
1728 | */ | |
1729 | if (OFF_SLAB(cachep)) | |
8456a648 | 1730 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1731 | } |
1732 | ||
97654dfa JK |
1733 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1734 | { | |
1735 | struct page *page, *n; | |
1736 | ||
1737 | list_for_each_entry_safe(page, n, list, lru) { | |
1738 | list_del(&page->lru); | |
1739 | slab_destroy(cachep, page); | |
1740 | } | |
1741 | } | |
1742 | ||
4d268eba | 1743 | /** |
a70773dd RD |
1744 | * calculate_slab_order - calculate size (page order) of slabs |
1745 | * @cachep: pointer to the cache that is being created | |
1746 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1747 | * @flags: slab allocation flags |
1748 | * | |
1749 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1750 | * |
1751 | * This could be made much more intelligent. For now, try to avoid using | |
1752 | * high order pages for slabs. When the gfp() functions are more friendly | |
1753 | * towards high-order requests, this should be changed. | |
1754 | */ | |
a737b3e2 | 1755 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1756 | size_t size, unsigned long flags) |
4d268eba PE |
1757 | { |
1758 | size_t left_over = 0; | |
9888e6fa | 1759 | int gfporder; |
4d268eba | 1760 | |
0aa817f0 | 1761 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1762 | unsigned int num; |
1763 | size_t remainder; | |
1764 | ||
70f75067 | 1765 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1766 | if (!num) |
1767 | continue; | |
9888e6fa | 1768 | |
f315e3fa JK |
1769 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1770 | if (num > SLAB_OBJ_MAX_NUM) | |
1771 | break; | |
1772 | ||
b1ab41c4 | 1773 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1774 | struct kmem_cache *freelist_cache; |
1775 | size_t freelist_size; | |
1776 | ||
1777 | freelist_size = num * sizeof(freelist_idx_t); | |
1778 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1779 | if (!freelist_cache) | |
1780 | continue; | |
1781 | ||
b1ab41c4 | 1782 | /* |
3217fd9b JK |
1783 | * Needed to avoid possible looping condition |
1784 | * in cache_grow() | |
b1ab41c4 | 1785 | */ |
3217fd9b JK |
1786 | if (OFF_SLAB(freelist_cache)) |
1787 | continue; | |
b1ab41c4 | 1788 | |
3217fd9b JK |
1789 | /* check if off slab has enough benefit */ |
1790 | if (freelist_cache->size > cachep->size / 2) | |
1791 | continue; | |
b1ab41c4 | 1792 | } |
4d268eba | 1793 | |
9888e6fa | 1794 | /* Found something acceptable - save it away */ |
4d268eba | 1795 | cachep->num = num; |
9888e6fa | 1796 | cachep->gfporder = gfporder; |
4d268eba PE |
1797 | left_over = remainder; |
1798 | ||
f78bb8ad LT |
1799 | /* |
1800 | * A VFS-reclaimable slab tends to have most allocations | |
1801 | * as GFP_NOFS and we really don't want to have to be allocating | |
1802 | * higher-order pages when we are unable to shrink dcache. | |
1803 | */ | |
1804 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1805 | break; | |
1806 | ||
4d268eba PE |
1807 | /* |
1808 | * Large number of objects is good, but very large slabs are | |
1809 | * currently bad for the gfp()s. | |
1810 | */ | |
543585cc | 1811 | if (gfporder >= slab_max_order) |
4d268eba PE |
1812 | break; |
1813 | ||
9888e6fa LT |
1814 | /* |
1815 | * Acceptable internal fragmentation? | |
1816 | */ | |
a737b3e2 | 1817 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1818 | break; |
1819 | } | |
1820 | return left_over; | |
1821 | } | |
1822 | ||
bf0dea23 JK |
1823 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1824 | struct kmem_cache *cachep, int entries, int batchcount) | |
1825 | { | |
1826 | int cpu; | |
1827 | size_t size; | |
1828 | struct array_cache __percpu *cpu_cache; | |
1829 | ||
1830 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1831 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1832 | |
1833 | if (!cpu_cache) | |
1834 | return NULL; | |
1835 | ||
1836 | for_each_possible_cpu(cpu) { | |
1837 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1838 | entries, batchcount); | |
1839 | } | |
1840 | ||
1841 | return cpu_cache; | |
1842 | } | |
1843 | ||
83b519e8 | 1844 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1845 | { |
97d06609 | 1846 | if (slab_state >= FULL) |
83b519e8 | 1847 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1848 | |
bf0dea23 JK |
1849 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1850 | if (!cachep->cpu_cache) | |
1851 | return 1; | |
1852 | ||
97d06609 | 1853 | if (slab_state == DOWN) { |
bf0dea23 JK |
1854 | /* Creation of first cache (kmem_cache). */ |
1855 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1856 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1857 | /* For kmem_cache_node */ |
1858 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1859 | } else { |
bf0dea23 | 1860 | int node; |
f30cf7d1 | 1861 | |
bf0dea23 JK |
1862 | for_each_online_node(node) { |
1863 | cachep->node[node] = kmalloc_node( | |
1864 | sizeof(struct kmem_cache_node), gfp, node); | |
1865 | BUG_ON(!cachep->node[node]); | |
1866 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1867 | } |
1868 | } | |
bf0dea23 | 1869 | |
6a67368c | 1870 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1871 | jiffies + REAPTIMEOUT_NODE + |
1872 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1873 | |
1874 | cpu_cache_get(cachep)->avail = 0; | |
1875 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1876 | cpu_cache_get(cachep)->batchcount = 1; | |
1877 | cpu_cache_get(cachep)->touched = 0; | |
1878 | cachep->batchcount = 1; | |
1879 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1880 | return 0; |
f30cf7d1 PE |
1881 | } |
1882 | ||
12220dea JK |
1883 | unsigned long kmem_cache_flags(unsigned long object_size, |
1884 | unsigned long flags, const char *name, | |
1885 | void (*ctor)(void *)) | |
1886 | { | |
1887 | return flags; | |
1888 | } | |
1889 | ||
1890 | struct kmem_cache * | |
1891 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
1892 | unsigned long flags, void (*ctor)(void *)) | |
1893 | { | |
1894 | struct kmem_cache *cachep; | |
1895 | ||
1896 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1897 | if (cachep) { | |
1898 | cachep->refcount++; | |
1899 | ||
1900 | /* | |
1901 | * Adjust the object sizes so that we clear | |
1902 | * the complete object on kzalloc. | |
1903 | */ | |
1904 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1905 | } | |
1906 | return cachep; | |
1907 | } | |
1908 | ||
b03a017b JK |
1909 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
1910 | size_t size, unsigned long flags) | |
1911 | { | |
1912 | size_t left; | |
1913 | ||
1914 | cachep->num = 0; | |
1915 | ||
1916 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
1917 | return false; | |
1918 | ||
1919 | left = calculate_slab_order(cachep, size, | |
1920 | flags | CFLGS_OBJFREELIST_SLAB); | |
1921 | if (!cachep->num) | |
1922 | return false; | |
1923 | ||
1924 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1925 | return false; | |
1926 | ||
1927 | cachep->colour = left / cachep->colour_off; | |
1928 | ||
1929 | return true; | |
1930 | } | |
1931 | ||
158e319b JK |
1932 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
1933 | size_t size, unsigned long flags) | |
1934 | { | |
1935 | size_t left; | |
1936 | ||
1937 | cachep->num = 0; | |
1938 | ||
1939 | /* | |
3217fd9b JK |
1940 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1941 | * to avoid recursive calls into kmemleak. | |
158e319b | 1942 | */ |
158e319b JK |
1943 | if (flags & SLAB_NOLEAKTRACE) |
1944 | return false; | |
1945 | ||
1946 | /* | |
1947 | * Size is large, assume best to place the slab management obj | |
1948 | * off-slab (should allow better packing of objs). | |
1949 | */ | |
1950 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1951 | if (!cachep->num) | |
1952 | return false; | |
1953 | ||
1954 | /* | |
1955 | * If the slab has been placed off-slab, and we have enough space then | |
1956 | * move it on-slab. This is at the expense of any extra colouring. | |
1957 | */ | |
1958 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1959 | return false; | |
1960 | ||
1961 | cachep->colour = left / cachep->colour_off; | |
1962 | ||
1963 | return true; | |
1964 | } | |
1965 | ||
1966 | static bool set_on_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 | left = calculate_slab_order(cachep, size, flags); | |
1974 | if (!cachep->num) | |
1975 | return false; | |
1976 | ||
1977 | cachep->colour = left / cachep->colour_off; | |
1978 | ||
1979 | return true; | |
1980 | } | |
1981 | ||
1da177e4 | 1982 | /** |
039363f3 | 1983 | * __kmem_cache_create - Create a cache. |
a755b76a | 1984 | * @cachep: cache management descriptor |
1da177e4 | 1985 | * @flags: SLAB flags |
1da177e4 LT |
1986 | * |
1987 | * Returns a ptr to the cache on success, NULL on failure. | |
1988 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1989 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1990 | * |
1da177e4 LT |
1991 | * The flags are |
1992 | * | |
1993 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1994 | * to catch references to uninitialised memory. | |
1995 | * | |
1996 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1997 | * for buffer overruns. | |
1998 | * | |
1da177e4 LT |
1999 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2000 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2001 | * as davem. | |
2002 | */ | |
278b1bb1 | 2003 | int |
8a13a4cc | 2004 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2005 | { |
d4a5fca5 | 2006 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2007 | gfp_t gfp; |
278b1bb1 | 2008 | int err; |
8a13a4cc | 2009 | size_t size = cachep->size; |
1da177e4 | 2010 | |
1da177e4 | 2011 | #if DEBUG |
1da177e4 LT |
2012 | #if FORCED_DEBUG |
2013 | /* | |
2014 | * Enable redzoning and last user accounting, except for caches with | |
2015 | * large objects, if the increased size would increase the object size | |
2016 | * above the next power of two: caches with object sizes just above a | |
2017 | * power of two have a significant amount of internal fragmentation. | |
2018 | */ | |
87a927c7 DW |
2019 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2020 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2021 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2022 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2023 | flags |= SLAB_POISON; | |
2024 | #endif | |
1da177e4 | 2025 | #endif |
1da177e4 | 2026 | |
a737b3e2 AM |
2027 | /* |
2028 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2029 | * unaligned accesses for some archs when redzoning is used, and makes |
2030 | * sure any on-slab bufctl's are also correctly aligned. | |
2031 | */ | |
b28a02de PE |
2032 | if (size & (BYTES_PER_WORD - 1)) { |
2033 | size += (BYTES_PER_WORD - 1); | |
2034 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2035 | } |
2036 | ||
87a927c7 DW |
2037 | if (flags & SLAB_RED_ZONE) { |
2038 | ralign = REDZONE_ALIGN; | |
2039 | /* If redzoning, ensure that the second redzone is suitably | |
2040 | * aligned, by adjusting the object size accordingly. */ | |
2041 | size += REDZONE_ALIGN - 1; | |
2042 | size &= ~(REDZONE_ALIGN - 1); | |
2043 | } | |
ca5f9703 | 2044 | |
a44b56d3 | 2045 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2046 | if (ralign < cachep->align) { |
2047 | ralign = cachep->align; | |
1da177e4 | 2048 | } |
3ff84a7f PE |
2049 | /* disable debug if necessary */ |
2050 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2051 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2052 | /* |
ca5f9703 | 2053 | * 4) Store it. |
1da177e4 | 2054 | */ |
8a13a4cc | 2055 | cachep->align = ralign; |
158e319b JK |
2056 | cachep->colour_off = cache_line_size(); |
2057 | /* Offset must be a multiple of the alignment. */ | |
2058 | if (cachep->colour_off < cachep->align) | |
2059 | cachep->colour_off = cachep->align; | |
1da177e4 | 2060 | |
83b519e8 PE |
2061 | if (slab_is_available()) |
2062 | gfp = GFP_KERNEL; | |
2063 | else | |
2064 | gfp = GFP_NOWAIT; | |
2065 | ||
1da177e4 | 2066 | #if DEBUG |
1da177e4 | 2067 | |
ca5f9703 PE |
2068 | /* |
2069 | * Both debugging options require word-alignment which is calculated | |
2070 | * into align above. | |
2071 | */ | |
1da177e4 | 2072 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2073 | /* add space for red zone words */ |
3ff84a7f PE |
2074 | cachep->obj_offset += sizeof(unsigned long long); |
2075 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2076 | } |
2077 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2078 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2079 | * the real object. But if the second red zone needs to be |
2080 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2081 | */ |
87a927c7 DW |
2082 | if (flags & SLAB_RED_ZONE) |
2083 | size += REDZONE_ALIGN; | |
2084 | else | |
2085 | size += BYTES_PER_WORD; | |
1da177e4 | 2086 | } |
832a15d2 JK |
2087 | #endif |
2088 | ||
7ed2f9e6 AP |
2089 | kasan_cache_create(cachep, &size, &flags); |
2090 | ||
832a15d2 JK |
2091 | size = ALIGN(size, cachep->align); |
2092 | /* | |
2093 | * We should restrict the number of objects in a slab to implement | |
2094 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2095 | */ | |
2096 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2097 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2098 | ||
2099 | #if DEBUG | |
03a2d2a3 JK |
2100 | /* |
2101 | * To activate debug pagealloc, off-slab management is necessary | |
2102 | * requirement. In early phase of initialization, small sized slab | |
2103 | * doesn't get initialized so it would not be possible. So, we need | |
2104 | * to check size >= 256. It guarantees that all necessary small | |
2105 | * sized slab is initialized in current slab initialization sequence. | |
2106 | */ | |
40323278 | 2107 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2108 | size >= 256 && cachep->object_size > cache_line_size()) { |
2109 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2110 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2111 | ||
2112 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2113 | flags |= CFLGS_OFF_SLAB; | |
2114 | cachep->obj_offset += tmp_size - size; | |
2115 | size = tmp_size; | |
2116 | goto done; | |
2117 | } | |
2118 | } | |
1da177e4 | 2119 | } |
1da177e4 LT |
2120 | #endif |
2121 | ||
b03a017b JK |
2122 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2123 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2124 | goto done; | |
2125 | } | |
2126 | ||
158e319b | 2127 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2128 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2129 | goto done; |
832a15d2 | 2130 | } |
1da177e4 | 2131 | |
158e319b JK |
2132 | if (set_on_slab_cache(cachep, size, flags)) |
2133 | goto done; | |
1da177e4 | 2134 | |
158e319b | 2135 | return -E2BIG; |
1da177e4 | 2136 | |
158e319b JK |
2137 | done: |
2138 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2139 | cachep->flags = flags; |
a57a4988 | 2140 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2141 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2142 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2143 | cachep->size = size; |
6a2d7a95 | 2144 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2145 | |
40b44137 JK |
2146 | #if DEBUG |
2147 | /* | |
2148 | * If we're going to use the generic kernel_map_pages() | |
2149 | * poisoning, then it's going to smash the contents of | |
2150 | * the redzone and userword anyhow, so switch them off. | |
2151 | */ | |
2152 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2153 | (cachep->flags & SLAB_POISON) && | |
2154 | is_debug_pagealloc_cache(cachep)) | |
2155 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2156 | #endif | |
2157 | ||
2158 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2159 | cachep->freelist_cache = |
2160 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2161 | } |
1da177e4 | 2162 | |
278b1bb1 CL |
2163 | err = setup_cpu_cache(cachep, gfp); |
2164 | if (err) { | |
52b4b950 | 2165 | __kmem_cache_release(cachep); |
278b1bb1 | 2166 | return err; |
2ed3a4ef | 2167 | } |
1da177e4 | 2168 | |
278b1bb1 | 2169 | return 0; |
1da177e4 | 2170 | } |
1da177e4 LT |
2171 | |
2172 | #if DEBUG | |
2173 | static void check_irq_off(void) | |
2174 | { | |
2175 | BUG_ON(!irqs_disabled()); | |
2176 | } | |
2177 | ||
2178 | static void check_irq_on(void) | |
2179 | { | |
2180 | BUG_ON(irqs_disabled()); | |
2181 | } | |
2182 | ||
18726ca8 JK |
2183 | static void check_mutex_acquired(void) |
2184 | { | |
2185 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2186 | } | |
2187 | ||
343e0d7a | 2188 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2189 | { |
2190 | #ifdef CONFIG_SMP | |
2191 | check_irq_off(); | |
18bf8541 | 2192 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2193 | #endif |
2194 | } | |
e498be7d | 2195 | |
343e0d7a | 2196 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2197 | { |
2198 | #ifdef CONFIG_SMP | |
2199 | check_irq_off(); | |
18bf8541 | 2200 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2201 | #endif |
2202 | } | |
2203 | ||
1da177e4 LT |
2204 | #else |
2205 | #define check_irq_off() do { } while(0) | |
2206 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2207 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2208 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2209 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2210 | #endif |
2211 | ||
18726ca8 JK |
2212 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2213 | int node, bool free_all, struct list_head *list) | |
2214 | { | |
2215 | int tofree; | |
2216 | ||
2217 | if (!ac || !ac->avail) | |
2218 | return; | |
2219 | ||
2220 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2221 | if (tofree > ac->avail) | |
2222 | tofree = (ac->avail + 1) / 2; | |
2223 | ||
2224 | free_block(cachep, ac->entry, tofree, node, list); | |
2225 | ac->avail -= tofree; | |
2226 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2227 | } | |
aab2207c | 2228 | |
1da177e4 LT |
2229 | static void do_drain(void *arg) |
2230 | { | |
a737b3e2 | 2231 | struct kmem_cache *cachep = arg; |
1da177e4 | 2232 | struct array_cache *ac; |
7d6e6d09 | 2233 | int node = numa_mem_id(); |
18bf8541 | 2234 | struct kmem_cache_node *n; |
97654dfa | 2235 | LIST_HEAD(list); |
1da177e4 LT |
2236 | |
2237 | check_irq_off(); | |
9a2dba4b | 2238 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2239 | n = get_node(cachep, node); |
2240 | spin_lock(&n->list_lock); | |
97654dfa | 2241 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2242 | spin_unlock(&n->list_lock); |
97654dfa | 2243 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2244 | ac->avail = 0; |
2245 | } | |
2246 | ||
343e0d7a | 2247 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2248 | { |
ce8eb6c4 | 2249 | struct kmem_cache_node *n; |
e498be7d | 2250 | int node; |
18726ca8 | 2251 | LIST_HEAD(list); |
e498be7d | 2252 | |
15c8b6c1 | 2253 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2254 | check_irq_on(); |
18bf8541 CL |
2255 | for_each_kmem_cache_node(cachep, node, n) |
2256 | if (n->alien) | |
ce8eb6c4 | 2257 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2258 | |
18726ca8 JK |
2259 | for_each_kmem_cache_node(cachep, node, n) { |
2260 | spin_lock_irq(&n->list_lock); | |
2261 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2262 | spin_unlock_irq(&n->list_lock); | |
2263 | ||
2264 | slabs_destroy(cachep, &list); | |
2265 | } | |
1da177e4 LT |
2266 | } |
2267 | ||
ed11d9eb CL |
2268 | /* |
2269 | * Remove slabs from the list of free slabs. | |
2270 | * Specify the number of slabs to drain in tofree. | |
2271 | * | |
2272 | * Returns the actual number of slabs released. | |
2273 | */ | |
2274 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2275 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2276 | { |
ed11d9eb CL |
2277 | struct list_head *p; |
2278 | int nr_freed; | |
8456a648 | 2279 | struct page *page; |
1da177e4 | 2280 | |
ed11d9eb | 2281 | nr_freed = 0; |
ce8eb6c4 | 2282 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2283 | |
ce8eb6c4 CL |
2284 | spin_lock_irq(&n->list_lock); |
2285 | p = n->slabs_free.prev; | |
2286 | if (p == &n->slabs_free) { | |
2287 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2288 | goto out; |
2289 | } | |
1da177e4 | 2290 | |
8456a648 | 2291 | page = list_entry(p, struct page, lru); |
8456a648 | 2292 | list_del(&page->lru); |
ed11d9eb CL |
2293 | /* |
2294 | * Safe to drop the lock. The slab is no longer linked | |
2295 | * to the cache. | |
2296 | */ | |
ce8eb6c4 CL |
2297 | n->free_objects -= cache->num; |
2298 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2299 | slab_destroy(cache, page); |
ed11d9eb | 2300 | nr_freed++; |
1da177e4 | 2301 | } |
ed11d9eb CL |
2302 | out: |
2303 | return nr_freed; | |
1da177e4 LT |
2304 | } |
2305 | ||
d6e0b7fa | 2306 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2307 | { |
18bf8541 CL |
2308 | int ret = 0; |
2309 | int node; | |
ce8eb6c4 | 2310 | struct kmem_cache_node *n; |
e498be7d CL |
2311 | |
2312 | drain_cpu_caches(cachep); | |
2313 | ||
2314 | check_irq_on(); | |
18bf8541 | 2315 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2316 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2317 | |
ce8eb6c4 CL |
2318 | ret += !list_empty(&n->slabs_full) || |
2319 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2320 | } |
2321 | return (ret ? 1 : 0); | |
2322 | } | |
2323 | ||
945cf2b6 | 2324 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2325 | { |
2326 | return __kmem_cache_shrink(cachep, false); | |
2327 | } | |
2328 | ||
2329 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2330 | { |
12c3667f | 2331 | int i; |
ce8eb6c4 | 2332 | struct kmem_cache_node *n; |
1da177e4 | 2333 | |
bf0dea23 | 2334 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2335 | |
ce8eb6c4 | 2336 | /* NUMA: free the node structures */ |
18bf8541 CL |
2337 | for_each_kmem_cache_node(cachep, i, n) { |
2338 | kfree(n->shared); | |
2339 | free_alien_cache(n->alien); | |
2340 | kfree(n); | |
2341 | cachep->node[i] = NULL; | |
12c3667f | 2342 | } |
1da177e4 | 2343 | } |
1da177e4 | 2344 | |
e5ac9c5a RT |
2345 | /* |
2346 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2347 | * |
2348 | * For a slab cache when the slab descriptor is off-slab, the | |
2349 | * slab descriptor can't come from the same cache which is being created, | |
2350 | * Because if it is the case, that means we defer the creation of | |
2351 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2352 | * And we eventually call down to __kmem_cache_create(), which | |
2353 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2354 | * This is a "chicken-and-egg" problem. | |
2355 | * | |
2356 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2357 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2358 | */ |
7e007355 | 2359 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2360 | struct page *page, int colour_off, |
2361 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2362 | { |
7e007355 | 2363 | void *freelist; |
0c3aa83e | 2364 | void *addr = page_address(page); |
b28a02de | 2365 | |
2e6b3602 JK |
2366 | page->s_mem = addr + colour_off; |
2367 | page->active = 0; | |
2368 | ||
b03a017b JK |
2369 | if (OBJFREELIST_SLAB(cachep)) |
2370 | freelist = NULL; | |
2371 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2372 | /* Slab management obj is off-slab. */ |
8456a648 | 2373 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2374 | local_flags, nodeid); |
8456a648 | 2375 | if (!freelist) |
1da177e4 LT |
2376 | return NULL; |
2377 | } else { | |
2e6b3602 JK |
2378 | /* We will use last bytes at the slab for freelist */ |
2379 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2380 | cachep->freelist_size; | |
1da177e4 | 2381 | } |
2e6b3602 | 2382 | |
8456a648 | 2383 | return freelist; |
1da177e4 LT |
2384 | } |
2385 | ||
7cc68973 | 2386 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2387 | { |
a41adfaa | 2388 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2389 | } |
2390 | ||
2391 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2392 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2393 | { |
a41adfaa | 2394 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2395 | } |
2396 | ||
10b2e9e8 | 2397 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2398 | { |
10b2e9e8 | 2399 | #if DEBUG |
1da177e4 LT |
2400 | int i; |
2401 | ||
2402 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2403 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2404 | |
1da177e4 LT |
2405 | if (cachep->flags & SLAB_STORE_USER) |
2406 | *dbg_userword(cachep, objp) = NULL; | |
2407 | ||
2408 | if (cachep->flags & SLAB_RED_ZONE) { | |
2409 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2410 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2411 | } | |
2412 | /* | |
a737b3e2 AM |
2413 | * Constructors are not allowed to allocate memory from the same |
2414 | * cache which they are a constructor for. Otherwise, deadlock. | |
2415 | * They must also be threaded. | |
1da177e4 | 2416 | */ |
7ed2f9e6 AP |
2417 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2418 | kasan_unpoison_object_data(cachep, | |
2419 | objp + obj_offset(cachep)); | |
51cc5068 | 2420 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2421 | kasan_poison_object_data( |
2422 | cachep, objp + obj_offset(cachep)); | |
2423 | } | |
1da177e4 LT |
2424 | |
2425 | if (cachep->flags & SLAB_RED_ZONE) { | |
2426 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2427 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2428 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2429 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2430 | } |
40b44137 JK |
2431 | /* need to poison the objs? */ |
2432 | if (cachep->flags & SLAB_POISON) { | |
2433 | poison_obj(cachep, objp, POISON_FREE); | |
2434 | slab_kernel_map(cachep, objp, 0, 0); | |
2435 | } | |
10b2e9e8 | 2436 | } |
1da177e4 | 2437 | #endif |
10b2e9e8 JK |
2438 | } |
2439 | ||
2440 | static void cache_init_objs(struct kmem_cache *cachep, | |
2441 | struct page *page) | |
2442 | { | |
2443 | int i; | |
7ed2f9e6 | 2444 | void *objp; |
10b2e9e8 JK |
2445 | |
2446 | cache_init_objs_debug(cachep, page); | |
2447 | ||
b03a017b JK |
2448 | if (OBJFREELIST_SLAB(cachep)) { |
2449 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + | |
2450 | obj_offset(cachep); | |
2451 | } | |
2452 | ||
10b2e9e8 JK |
2453 | for (i = 0; i < cachep->num; i++) { |
2454 | /* constructor could break poison info */ | |
7ed2f9e6 AP |
2455 | if (DEBUG == 0 && cachep->ctor) { |
2456 | objp = index_to_obj(cachep, page, i); | |
2457 | kasan_unpoison_object_data(cachep, objp); | |
2458 | cachep->ctor(objp); | |
2459 | kasan_poison_object_data(cachep, objp); | |
2460 | } | |
10b2e9e8 | 2461 | |
e5c58dfd | 2462 | set_free_obj(page, i, i); |
1da177e4 | 2463 | } |
1da177e4 LT |
2464 | } |
2465 | ||
343e0d7a | 2466 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2467 | { |
4b51d669 CL |
2468 | if (CONFIG_ZONE_DMA_FLAG) { |
2469 | if (flags & GFP_DMA) | |
a618e89f | 2470 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2471 | else |
a618e89f | 2472 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2473 | } |
1da177e4 LT |
2474 | } |
2475 | ||
260b61dd | 2476 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2477 | { |
b1cb0982 | 2478 | void *objp; |
78d382d7 | 2479 | |
e5c58dfd | 2480 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2481 | page->active++; |
78d382d7 | 2482 | |
d31676df JK |
2483 | #if DEBUG |
2484 | if (cachep->flags & SLAB_STORE_USER) | |
2485 | set_store_user_dirty(cachep); | |
2486 | #endif | |
2487 | ||
78d382d7 MD |
2488 | return objp; |
2489 | } | |
2490 | ||
260b61dd JK |
2491 | static void slab_put_obj(struct kmem_cache *cachep, |
2492 | struct page *page, void *objp) | |
78d382d7 | 2493 | { |
8456a648 | 2494 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2495 | #if DEBUG |
16025177 | 2496 | unsigned int i; |
b1cb0982 | 2497 | |
b1cb0982 | 2498 | /* Verify double free bug */ |
8456a648 | 2499 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2500 | if (get_free_obj(page, i) == objnr) { |
1170532b | 2501 | pr_err("slab: double free detected in cache '%s', objp %p\n", |
756a025f | 2502 | cachep->name, objp); |
b1cb0982 JK |
2503 | BUG(); |
2504 | } | |
78d382d7 MD |
2505 | } |
2506 | #endif | |
8456a648 | 2507 | page->active--; |
b03a017b JK |
2508 | if (!page->freelist) |
2509 | page->freelist = objp + obj_offset(cachep); | |
2510 | ||
e5c58dfd | 2511 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2512 | } |
2513 | ||
4776874f PE |
2514 | /* |
2515 | * Map pages beginning at addr to the given cache and slab. This is required | |
2516 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2517 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2518 | */ |
8456a648 | 2519 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2520 | void *freelist) |
1da177e4 | 2521 | { |
a57a4988 | 2522 | page->slab_cache = cache; |
8456a648 | 2523 | page->freelist = freelist; |
1da177e4 LT |
2524 | } |
2525 | ||
2526 | /* | |
2527 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2528 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2529 | */ | |
3c517a61 | 2530 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2531 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2532 | { |
7e007355 | 2533 | void *freelist; |
b28a02de PE |
2534 | size_t offset; |
2535 | gfp_t local_flags; | |
ce8eb6c4 | 2536 | struct kmem_cache_node *n; |
1da177e4 | 2537 | |
a737b3e2 AM |
2538 | /* |
2539 | * Be lazy and only check for valid flags here, keeping it out of the | |
2540 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2541 | */ |
c871ac4e AM |
2542 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2543 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2544 | BUG(); | |
2545 | } | |
6cb06229 | 2546 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2547 | |
ce8eb6c4 | 2548 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2549 | check_irq_off(); |
18bf8541 | 2550 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2551 | spin_lock(&n->list_lock); |
1da177e4 LT |
2552 | |
2553 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2554 | offset = n->colour_next; |
2555 | n->colour_next++; | |
2556 | if (n->colour_next >= cachep->colour) | |
2557 | n->colour_next = 0; | |
2558 | spin_unlock(&n->list_lock); | |
1da177e4 | 2559 | |
2e1217cf | 2560 | offset *= cachep->colour_off; |
1da177e4 | 2561 | |
d0164adc | 2562 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2563 | local_irq_enable(); |
2564 | ||
2565 | /* | |
2566 | * The test for missing atomic flag is performed here, rather than | |
2567 | * the more obvious place, simply to reduce the critical path length | |
2568 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2569 | * will eventually be caught here (where it matters). | |
2570 | */ | |
2571 | kmem_flagcheck(cachep, flags); | |
2572 | ||
a737b3e2 AM |
2573 | /* |
2574 | * Get mem for the objs. Attempt to allocate a physical page from | |
2575 | * 'nodeid'. | |
e498be7d | 2576 | */ |
0c3aa83e JK |
2577 | if (!page) |
2578 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2579 | if (!page) | |
1da177e4 LT |
2580 | goto failed; |
2581 | ||
2582 | /* Get slab management. */ | |
8456a648 | 2583 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2584 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
b03a017b | 2585 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2586 | goto opps1; |
2587 | ||
8456a648 | 2588 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2589 | |
7ed2f9e6 | 2590 | kasan_poison_slab(page); |
8456a648 | 2591 | cache_init_objs(cachep, page); |
1da177e4 | 2592 | |
d0164adc | 2593 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2594 | local_irq_disable(); |
2595 | check_irq_off(); | |
ce8eb6c4 | 2596 | spin_lock(&n->list_lock); |
1da177e4 LT |
2597 | |
2598 | /* Make slab active. */ | |
8456a648 | 2599 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2600 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2601 | n->free_objects += cachep->num; |
2602 | spin_unlock(&n->list_lock); | |
1da177e4 | 2603 | return 1; |
a737b3e2 | 2604 | opps1: |
0c3aa83e | 2605 | kmem_freepages(cachep, page); |
a737b3e2 | 2606 | failed: |
d0164adc | 2607 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2608 | local_irq_disable(); |
2609 | return 0; | |
2610 | } | |
2611 | ||
2612 | #if DEBUG | |
2613 | ||
2614 | /* | |
2615 | * Perform extra freeing checks: | |
2616 | * - detect bad pointers. | |
2617 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2618 | */ |
2619 | static void kfree_debugcheck(const void *objp) | |
2620 | { | |
1da177e4 | 2621 | if (!virt_addr_valid(objp)) { |
1170532b | 2622 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2623 | (unsigned long)objp); |
2624 | BUG(); | |
1da177e4 | 2625 | } |
1da177e4 LT |
2626 | } |
2627 | ||
58ce1fd5 PE |
2628 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2629 | { | |
b46b8f19 | 2630 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2631 | |
2632 | redzone1 = *dbg_redzone1(cache, obj); | |
2633 | redzone2 = *dbg_redzone2(cache, obj); | |
2634 | ||
2635 | /* | |
2636 | * Redzone is ok. | |
2637 | */ | |
2638 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2639 | return; | |
2640 | ||
2641 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2642 | slab_error(cache, "double free detected"); | |
2643 | else | |
2644 | slab_error(cache, "memory outside object was overwritten"); | |
2645 | ||
1170532b JP |
2646 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2647 | obj, redzone1, redzone2); | |
58ce1fd5 PE |
2648 | } |
2649 | ||
343e0d7a | 2650 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2651 | unsigned long caller) |
1da177e4 | 2652 | { |
1da177e4 | 2653 | unsigned int objnr; |
8456a648 | 2654 | struct page *page; |
1da177e4 | 2655 | |
80cbd911 MW |
2656 | BUG_ON(virt_to_cache(objp) != cachep); |
2657 | ||
3dafccf2 | 2658 | objp -= obj_offset(cachep); |
1da177e4 | 2659 | kfree_debugcheck(objp); |
b49af68f | 2660 | page = virt_to_head_page(objp); |
1da177e4 | 2661 | |
1da177e4 | 2662 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2663 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2664 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2665 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2666 | } | |
d31676df JK |
2667 | if (cachep->flags & SLAB_STORE_USER) { |
2668 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2669 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2670 | } |
1da177e4 | 2671 | |
8456a648 | 2672 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2673 | |
2674 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2675 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2676 | |
1da177e4 | 2677 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2678 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2679 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2680 | } |
2681 | return objp; | |
2682 | } | |
2683 | ||
1da177e4 LT |
2684 | #else |
2685 | #define kfree_debugcheck(x) do { } while(0) | |
2686 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2687 | #endif |
2688 | ||
b03a017b JK |
2689 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2690 | void **list) | |
2691 | { | |
2692 | #if DEBUG | |
2693 | void *next = *list; | |
2694 | void *objp; | |
2695 | ||
2696 | while (next) { | |
2697 | objp = next - obj_offset(cachep); | |
2698 | next = *(void **)next; | |
2699 | poison_obj(cachep, objp, POISON_FREE); | |
2700 | } | |
2701 | #endif | |
2702 | } | |
2703 | ||
d8410234 | 2704 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2705 | struct kmem_cache_node *n, struct page *page, |
2706 | void **list) | |
d8410234 JK |
2707 | { |
2708 | /* move slabp to correct slabp list: */ | |
2709 | list_del(&page->lru); | |
b03a017b | 2710 | if (page->active == cachep->num) { |
d8410234 | 2711 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2712 | if (OBJFREELIST_SLAB(cachep)) { |
2713 | #if DEBUG | |
2714 | /* Poisoning will be done without holding the lock */ | |
2715 | if (cachep->flags & SLAB_POISON) { | |
2716 | void **objp = page->freelist; | |
2717 | ||
2718 | *objp = *list; | |
2719 | *list = objp; | |
2720 | } | |
2721 | #endif | |
2722 | page->freelist = NULL; | |
2723 | } | |
2724 | } else | |
d8410234 JK |
2725 | list_add(&page->lru, &n->slabs_partial); |
2726 | } | |
2727 | ||
f68f8ddd JK |
2728 | /* Try to find non-pfmemalloc slab if needed */ |
2729 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
2730 | struct page *page, bool pfmemalloc) | |
2731 | { | |
2732 | if (!page) | |
2733 | return NULL; | |
2734 | ||
2735 | if (pfmemalloc) | |
2736 | return page; | |
2737 | ||
2738 | if (!PageSlabPfmemalloc(page)) | |
2739 | return page; | |
2740 | ||
2741 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2742 | if (n->free_objects > n->free_limit) { | |
2743 | ClearPageSlabPfmemalloc(page); | |
2744 | return page; | |
2745 | } | |
2746 | ||
2747 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2748 | list_del(&page->lru); | |
2749 | if (!page->active) | |
2750 | list_add_tail(&page->lru, &n->slabs_free); | |
2751 | else | |
2752 | list_add_tail(&page->lru, &n->slabs_partial); | |
2753 | ||
2754 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2755 | if (!PageSlabPfmemalloc(page)) | |
2756 | return page; | |
2757 | } | |
2758 | ||
2759 | list_for_each_entry(page, &n->slabs_free, lru) { | |
2760 | if (!PageSlabPfmemalloc(page)) | |
2761 | return page; | |
2762 | } | |
2763 | ||
2764 | return NULL; | |
2765 | } | |
2766 | ||
2767 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2768 | { |
2769 | struct page *page; | |
2770 | ||
2771 | page = list_first_entry_or_null(&n->slabs_partial, | |
2772 | struct page, lru); | |
2773 | if (!page) { | |
2774 | n->free_touched = 1; | |
2775 | page = list_first_entry_or_null(&n->slabs_free, | |
2776 | struct page, lru); | |
2777 | } | |
2778 | ||
f68f8ddd JK |
2779 | if (sk_memalloc_socks()) |
2780 | return get_valid_first_slab(n, page, pfmemalloc); | |
2781 | ||
7aa0d227 GT |
2782 | return page; |
2783 | } | |
2784 | ||
f68f8ddd JK |
2785 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2786 | struct kmem_cache_node *n, gfp_t flags) | |
2787 | { | |
2788 | struct page *page; | |
2789 | void *obj; | |
2790 | void *list = NULL; | |
2791 | ||
2792 | if (!gfp_pfmemalloc_allowed(flags)) | |
2793 | return NULL; | |
2794 | ||
2795 | spin_lock(&n->list_lock); | |
2796 | page = get_first_slab(n, true); | |
2797 | if (!page) { | |
2798 | spin_unlock(&n->list_lock); | |
2799 | return NULL; | |
2800 | } | |
2801 | ||
2802 | obj = slab_get_obj(cachep, page); | |
2803 | n->free_objects--; | |
2804 | ||
2805 | fixup_slab_list(cachep, n, page, &list); | |
2806 | ||
2807 | spin_unlock(&n->list_lock); | |
2808 | fixup_objfreelist_debug(cachep, &list); | |
2809 | ||
2810 | return obj; | |
2811 | } | |
2812 | ||
2813 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) | |
1da177e4 LT |
2814 | { |
2815 | int batchcount; | |
ce8eb6c4 | 2816 | struct kmem_cache_node *n; |
1da177e4 | 2817 | struct array_cache *ac; |
1ca4cb24 | 2818 | int node; |
b03a017b | 2819 | void *list = NULL; |
1ca4cb24 | 2820 | |
1da177e4 | 2821 | check_irq_off(); |
7d6e6d09 | 2822 | node = numa_mem_id(); |
f68f8ddd | 2823 | |
072bb0aa | 2824 | retry: |
9a2dba4b | 2825 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2826 | batchcount = ac->batchcount; |
2827 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2828 | /* |
2829 | * If there was little recent activity on this cache, then | |
2830 | * perform only a partial refill. Otherwise we could generate | |
2831 | * refill bouncing. | |
1da177e4 LT |
2832 | */ |
2833 | batchcount = BATCHREFILL_LIMIT; | |
2834 | } | |
18bf8541 | 2835 | n = get_node(cachep, node); |
e498be7d | 2836 | |
ce8eb6c4 CL |
2837 | BUG_ON(ac->avail > 0 || !n); |
2838 | spin_lock(&n->list_lock); | |
1da177e4 | 2839 | |
3ded175a | 2840 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2841 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2842 | n->shared->touched = 1; | |
3ded175a | 2843 | goto alloc_done; |
44b57f1c | 2844 | } |
3ded175a | 2845 | |
1da177e4 | 2846 | while (batchcount > 0) { |
8456a648 | 2847 | struct page *page; |
1da177e4 | 2848 | /* Get slab alloc is to come from. */ |
f68f8ddd | 2849 | page = get_first_slab(n, false); |
7aa0d227 GT |
2850 | if (!page) |
2851 | goto must_grow; | |
1da177e4 | 2852 | |
1da177e4 | 2853 | check_spinlock_acquired(cachep); |
714b8171 PE |
2854 | |
2855 | /* | |
2856 | * The slab was either on partial or free list so | |
2857 | * there must be at least one object available for | |
2858 | * allocation. | |
2859 | */ | |
8456a648 | 2860 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2861 | |
8456a648 | 2862 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2863 | STATS_INC_ALLOCED(cachep); |
2864 | STATS_INC_ACTIVE(cachep); | |
2865 | STATS_SET_HIGH(cachep); | |
2866 | ||
f68f8ddd | 2867 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); |
1da177e4 | 2868 | } |
1da177e4 | 2869 | |
b03a017b | 2870 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2871 | } |
2872 | ||
a737b3e2 | 2873 | must_grow: |
ce8eb6c4 | 2874 | n->free_objects -= ac->avail; |
a737b3e2 | 2875 | alloc_done: |
ce8eb6c4 | 2876 | spin_unlock(&n->list_lock); |
b03a017b | 2877 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 LT |
2878 | |
2879 | if (unlikely(!ac->avail)) { | |
2880 | int x; | |
f68f8ddd JK |
2881 | |
2882 | /* Check if we can use obj in pfmemalloc slab */ | |
2883 | if (sk_memalloc_socks()) { | |
2884 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
2885 | ||
2886 | if (obj) | |
2887 | return obj; | |
2888 | } | |
2889 | ||
4167e9b2 | 2890 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2891 | |
a737b3e2 | 2892 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2893 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2894 | node = numa_mem_id(); |
072bb0aa MG |
2895 | |
2896 | /* no objects in sight? abort */ | |
f68f8ddd | 2897 | if (!x && ac->avail == 0) |
1da177e4 LT |
2898 | return NULL; |
2899 | ||
a737b3e2 | 2900 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2901 | goto retry; |
2902 | } | |
2903 | ac->touched = 1; | |
072bb0aa | 2904 | |
f68f8ddd | 2905 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2906 | } |
2907 | ||
a737b3e2 AM |
2908 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2909 | gfp_t flags) | |
1da177e4 | 2910 | { |
d0164adc | 2911 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2912 | #if DEBUG |
2913 | kmem_flagcheck(cachep, flags); | |
2914 | #endif | |
2915 | } | |
2916 | ||
2917 | #if DEBUG | |
a737b3e2 | 2918 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2919 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2920 | { |
b28a02de | 2921 | if (!objp) |
1da177e4 | 2922 | return objp; |
b28a02de | 2923 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2924 | check_poison_obj(cachep, objp); |
40b44137 | 2925 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2926 | poison_obj(cachep, objp, POISON_INUSE); |
2927 | } | |
2928 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2929 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2930 | |
2931 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2932 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2933 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 2934 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
1170532b JP |
2935 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2936 | objp, *dbg_redzone1(cachep, objp), | |
2937 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2938 | } |
2939 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2940 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2941 | } | |
03787301 | 2942 | |
3dafccf2 | 2943 | objp += obj_offset(cachep); |
4f104934 | 2944 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2945 | cachep->ctor(objp); |
7ea466f2 TH |
2946 | if (ARCH_SLAB_MINALIGN && |
2947 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
1170532b | 2948 | pr_err("0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2949 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2950 | } |
1da177e4 LT |
2951 | return objp; |
2952 | } | |
2953 | #else | |
2954 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2955 | #endif | |
2956 | ||
343e0d7a | 2957 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2958 | { |
b28a02de | 2959 | void *objp; |
1da177e4 LT |
2960 | struct array_cache *ac; |
2961 | ||
5c382300 | 2962 | check_irq_off(); |
8a8b6502 | 2963 | |
9a2dba4b | 2964 | ac = cpu_cache_get(cachep); |
1da177e4 | 2965 | if (likely(ac->avail)) { |
1da177e4 | 2966 | ac->touched = 1; |
f68f8ddd | 2967 | objp = ac->entry[--ac->avail]; |
072bb0aa | 2968 | |
f68f8ddd JK |
2969 | STATS_INC_ALLOCHIT(cachep); |
2970 | goto out; | |
1da177e4 | 2971 | } |
072bb0aa MG |
2972 | |
2973 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 2974 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
2975 | /* |
2976 | * the 'ac' may be updated by cache_alloc_refill(), | |
2977 | * and kmemleak_erase() requires its correct value. | |
2978 | */ | |
2979 | ac = cpu_cache_get(cachep); | |
2980 | ||
2981 | out: | |
d5cff635 CM |
2982 | /* |
2983 | * To avoid a false negative, if an object that is in one of the | |
2984 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2985 | * treat the array pointers as a reference to the object. | |
2986 | */ | |
f3d8b53a O |
2987 | if (objp) |
2988 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2989 | return objp; |
2990 | } | |
2991 | ||
e498be7d | 2992 | #ifdef CONFIG_NUMA |
c61afb18 | 2993 | /* |
2ad654bc | 2994 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2995 | * |
2996 | * If we are in_interrupt, then process context, including cpusets and | |
2997 | * mempolicy, may not apply and should not be used for allocation policy. | |
2998 | */ | |
2999 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3000 | { | |
3001 | int nid_alloc, nid_here; | |
3002 | ||
765c4507 | 3003 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3004 | return NULL; |
7d6e6d09 | 3005 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3006 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3007 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3008 | else if (current->mempolicy) |
2a389610 | 3009 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3010 | if (nid_alloc != nid_here) |
8b98c169 | 3011 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3012 | return NULL; |
3013 | } | |
3014 | ||
765c4507 CL |
3015 | /* |
3016 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3017 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3018 | * available node for available objects. If that fails then we |
3c517a61 CL |
3019 | * perform an allocation without specifying a node. This allows the page |
3020 | * allocator to do its reclaim / fallback magic. We then insert the | |
3021 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3022 | */ |
8c8cc2c1 | 3023 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3024 | { |
8c8cc2c1 PE |
3025 | struct zonelist *zonelist; |
3026 | gfp_t local_flags; | |
dd1a239f | 3027 | struct zoneref *z; |
54a6eb5c MG |
3028 | struct zone *zone; |
3029 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3030 | void *obj = NULL; |
3c517a61 | 3031 | int nid; |
cc9a6c87 | 3032 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3033 | |
3034 | if (flags & __GFP_THISNODE) | |
3035 | return NULL; | |
3036 | ||
6cb06229 | 3037 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3038 | |
cc9a6c87 | 3039 | retry_cpuset: |
d26914d1 | 3040 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3041 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3042 | |
3c517a61 CL |
3043 | retry: |
3044 | /* | |
3045 | * Look through allowed nodes for objects available | |
3046 | * from existing per node queues. | |
3047 | */ | |
54a6eb5c MG |
3048 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3049 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3050 | |
061d7074 | 3051 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3052 | get_node(cache, nid) && |
3053 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3054 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3055 | gfp_exact_node(flags), nid); |
481c5346 CL |
3056 | if (obj) |
3057 | break; | |
3058 | } | |
3c517a61 CL |
3059 | } |
3060 | ||
cfce6604 | 3061 | if (!obj) { |
3c517a61 CL |
3062 | /* |
3063 | * This allocation will be performed within the constraints | |
3064 | * of the current cpuset / memory policy requirements. | |
3065 | * We may trigger various forms of reclaim on the allowed | |
3066 | * set and go into memory reserves if necessary. | |
3067 | */ | |
0c3aa83e JK |
3068 | struct page *page; |
3069 | ||
d0164adc | 3070 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3071 | local_irq_enable(); |
3072 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3073 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3074 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3075 | local_irq_disable(); |
0c3aa83e | 3076 | if (page) { |
3c517a61 CL |
3077 | /* |
3078 | * Insert into the appropriate per node queues | |
3079 | */ | |
0c3aa83e JK |
3080 | nid = page_to_nid(page); |
3081 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3082 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3083 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3084 | if (!obj) |
3085 | /* | |
3086 | * Another processor may allocate the | |
3087 | * objects in the slab since we are | |
3088 | * not holding any locks. | |
3089 | */ | |
3090 | goto retry; | |
3091 | } else { | |
b6a60451 | 3092 | /* cache_grow already freed obj */ |
3c517a61 CL |
3093 | obj = NULL; |
3094 | } | |
3095 | } | |
aedb0eb1 | 3096 | } |
cc9a6c87 | 3097 | |
d26914d1 | 3098 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3099 | goto retry_cpuset; |
765c4507 CL |
3100 | return obj; |
3101 | } | |
3102 | ||
e498be7d CL |
3103 | /* |
3104 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3105 | */ |
8b98c169 | 3106 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3107 | int nodeid) |
e498be7d | 3108 | { |
8456a648 | 3109 | struct page *page; |
ce8eb6c4 | 3110 | struct kmem_cache_node *n; |
b28a02de | 3111 | void *obj; |
b03a017b | 3112 | void *list = NULL; |
b28a02de PE |
3113 | int x; |
3114 | ||
7c3fbbdd | 3115 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3116 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3117 | BUG_ON(!n); |
b28a02de | 3118 | |
a737b3e2 | 3119 | retry: |
ca3b9b91 | 3120 | check_irq_off(); |
ce8eb6c4 | 3121 | spin_lock(&n->list_lock); |
f68f8ddd | 3122 | page = get_first_slab(n, false); |
7aa0d227 GT |
3123 | if (!page) |
3124 | goto must_grow; | |
b28a02de | 3125 | |
b28a02de | 3126 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3127 | |
3128 | STATS_INC_NODEALLOCS(cachep); | |
3129 | STATS_INC_ACTIVE(cachep); | |
3130 | STATS_SET_HIGH(cachep); | |
3131 | ||
8456a648 | 3132 | BUG_ON(page->active == cachep->num); |
b28a02de | 3133 | |
260b61dd | 3134 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3135 | n->free_objects--; |
b28a02de | 3136 | |
b03a017b | 3137 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3138 | |
ce8eb6c4 | 3139 | spin_unlock(&n->list_lock); |
b03a017b | 3140 | fixup_objfreelist_debug(cachep, &list); |
b28a02de | 3141 | goto done; |
e498be7d | 3142 | |
a737b3e2 | 3143 | must_grow: |
ce8eb6c4 | 3144 | spin_unlock(&n->list_lock); |
4167e9b2 | 3145 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3146 | if (x) |
3147 | goto retry; | |
1da177e4 | 3148 | |
8c8cc2c1 | 3149 | return fallback_alloc(cachep, flags); |
e498be7d | 3150 | |
a737b3e2 | 3151 | done: |
b28a02de | 3152 | return obj; |
e498be7d | 3153 | } |
8c8cc2c1 | 3154 | |
8c8cc2c1 | 3155 | static __always_inline void * |
48356303 | 3156 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3157 | unsigned long caller) |
8c8cc2c1 PE |
3158 | { |
3159 | unsigned long save_flags; | |
3160 | void *ptr; | |
7d6e6d09 | 3161 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3162 | |
dcce284a | 3163 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3164 | cachep = slab_pre_alloc_hook(cachep, flags); |
3165 | if (unlikely(!cachep)) | |
824ebef1 AM |
3166 | return NULL; |
3167 | ||
8c8cc2c1 PE |
3168 | cache_alloc_debugcheck_before(cachep, flags); |
3169 | local_irq_save(save_flags); | |
3170 | ||
eacbbae3 | 3171 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3172 | nodeid = slab_node; |
8c8cc2c1 | 3173 | |
18bf8541 | 3174 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3175 | /* Node not bootstrapped yet */ |
3176 | ptr = fallback_alloc(cachep, flags); | |
3177 | goto out; | |
3178 | } | |
3179 | ||
7d6e6d09 | 3180 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3181 | /* |
3182 | * Use the locally cached objects if possible. | |
3183 | * However ____cache_alloc does not allow fallback | |
3184 | * to other nodes. It may fail while we still have | |
3185 | * objects on other nodes available. | |
3186 | */ | |
3187 | ptr = ____cache_alloc(cachep, flags); | |
3188 | if (ptr) | |
3189 | goto out; | |
3190 | } | |
3191 | /* ___cache_alloc_node can fall back to other nodes */ | |
3192 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3193 | out: | |
3194 | local_irq_restore(save_flags); | |
3195 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3196 | ||
d5e3ed66 JDB |
3197 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3198 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3199 | |
d5e3ed66 | 3200 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3201 | return ptr; |
3202 | } | |
3203 | ||
3204 | static __always_inline void * | |
3205 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3206 | { | |
3207 | void *objp; | |
3208 | ||
2ad654bc | 3209 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3210 | objp = alternate_node_alloc(cache, flags); |
3211 | if (objp) | |
3212 | goto out; | |
3213 | } | |
3214 | objp = ____cache_alloc(cache, flags); | |
3215 | ||
3216 | /* | |
3217 | * We may just have run out of memory on the local node. | |
3218 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3219 | */ | |
7d6e6d09 LS |
3220 | if (!objp) |
3221 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3222 | |
3223 | out: | |
3224 | return objp; | |
3225 | } | |
3226 | #else | |
3227 | ||
3228 | static __always_inline void * | |
3229 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3230 | { | |
3231 | return ____cache_alloc(cachep, flags); | |
3232 | } | |
3233 | ||
3234 | #endif /* CONFIG_NUMA */ | |
3235 | ||
3236 | static __always_inline void * | |
48356303 | 3237 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3238 | { |
3239 | unsigned long save_flags; | |
3240 | void *objp; | |
3241 | ||
dcce284a | 3242 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3243 | cachep = slab_pre_alloc_hook(cachep, flags); |
3244 | if (unlikely(!cachep)) | |
824ebef1 AM |
3245 | return NULL; |
3246 | ||
8c8cc2c1 PE |
3247 | cache_alloc_debugcheck_before(cachep, flags); |
3248 | local_irq_save(save_flags); | |
3249 | objp = __do_cache_alloc(cachep, flags); | |
3250 | local_irq_restore(save_flags); | |
3251 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3252 | prefetchw(objp); | |
3253 | ||
d5e3ed66 JDB |
3254 | if (unlikely(flags & __GFP_ZERO) && objp) |
3255 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3256 | |
d5e3ed66 | 3257 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3258 | return objp; |
3259 | } | |
e498be7d CL |
3260 | |
3261 | /* | |
5f0985bb | 3262 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3263 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3264 | */ |
97654dfa JK |
3265 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3266 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3267 | { |
3268 | int i; | |
25c063fb | 3269 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3270 | |
3271 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3272 | void *objp; |
8456a648 | 3273 | struct page *page; |
1da177e4 | 3274 | |
072bb0aa MG |
3275 | objp = objpp[i]; |
3276 | ||
8456a648 | 3277 | page = virt_to_head_page(objp); |
8456a648 | 3278 | list_del(&page->lru); |
ff69416e | 3279 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3280 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3281 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3282 | n->free_objects++; |
1da177e4 LT |
3283 | |
3284 | /* fixup slab chains */ | |
8456a648 | 3285 | if (page->active == 0) { |
ce8eb6c4 CL |
3286 | if (n->free_objects > n->free_limit) { |
3287 | n->free_objects -= cachep->num; | |
97654dfa | 3288 | list_add_tail(&page->lru, list); |
1da177e4 | 3289 | } else { |
8456a648 | 3290 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3291 | } |
3292 | } else { | |
3293 | /* Unconditionally move a slab to the end of the | |
3294 | * partial list on free - maximum time for the | |
3295 | * other objects to be freed, too. | |
3296 | */ | |
8456a648 | 3297 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3298 | } |
3299 | } | |
3300 | } | |
3301 | ||
343e0d7a | 3302 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3303 | { |
3304 | int batchcount; | |
ce8eb6c4 | 3305 | struct kmem_cache_node *n; |
7d6e6d09 | 3306 | int node = numa_mem_id(); |
97654dfa | 3307 | LIST_HEAD(list); |
1da177e4 LT |
3308 | |
3309 | batchcount = ac->batchcount; | |
260b61dd | 3310 | |
1da177e4 | 3311 | check_irq_off(); |
18bf8541 | 3312 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3313 | spin_lock(&n->list_lock); |
3314 | if (n->shared) { | |
3315 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3316 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3317 | if (max) { |
3318 | if (batchcount > max) | |
3319 | batchcount = max; | |
e498be7d | 3320 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3321 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3322 | shared_array->avail += batchcount; |
3323 | goto free_done; | |
3324 | } | |
3325 | } | |
3326 | ||
97654dfa | 3327 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3328 | free_done: |
1da177e4 LT |
3329 | #if STATS |
3330 | { | |
3331 | int i = 0; | |
73c0219d | 3332 | struct page *page; |
1da177e4 | 3333 | |
73c0219d | 3334 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3335 | BUG_ON(page->active); |
1da177e4 LT |
3336 | |
3337 | i++; | |
1da177e4 LT |
3338 | } |
3339 | STATS_SET_FREEABLE(cachep, i); | |
3340 | } | |
3341 | #endif | |
ce8eb6c4 | 3342 | spin_unlock(&n->list_lock); |
97654dfa | 3343 | slabs_destroy(cachep, &list); |
1da177e4 | 3344 | ac->avail -= batchcount; |
a737b3e2 | 3345 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3346 | } |
3347 | ||
3348 | /* | |
a737b3e2 AM |
3349 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3350 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3351 | */ |
a947eb95 | 3352 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3353 | unsigned long caller) |
1da177e4 | 3354 | { |
9a2dba4b | 3355 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 | 3356 | |
7ed2f9e6 AP |
3357 | kasan_slab_free(cachep, objp); |
3358 | ||
1da177e4 | 3359 | check_irq_off(); |
d5cff635 | 3360 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3361 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3362 | |
8c138bc0 | 3363 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3364 | |
1807a1aa SS |
3365 | /* |
3366 | * Skip calling cache_free_alien() when the platform is not numa. | |
3367 | * This will avoid cache misses that happen while accessing slabp (which | |
3368 | * is per page memory reference) to get nodeid. Instead use a global | |
3369 | * variable to skip the call, which is mostly likely to be present in | |
3370 | * the cache. | |
3371 | */ | |
b6e68bc1 | 3372 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3373 | return; |
3374 | ||
3d880194 | 3375 | if (ac->avail < ac->limit) { |
1da177e4 | 3376 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3377 | } else { |
3378 | STATS_INC_FREEMISS(cachep); | |
3379 | cache_flusharray(cachep, ac); | |
1da177e4 | 3380 | } |
42c8c99c | 3381 | |
f68f8ddd JK |
3382 | if (sk_memalloc_socks()) { |
3383 | struct page *page = virt_to_head_page(objp); | |
3384 | ||
3385 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3386 | cache_free_pfmemalloc(cachep, page, objp); | |
3387 | return; | |
3388 | } | |
3389 | } | |
3390 | ||
3391 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3392 | } |
3393 | ||
3394 | /** | |
3395 | * kmem_cache_alloc - Allocate an object | |
3396 | * @cachep: The cache to allocate from. | |
3397 | * @flags: See kmalloc(). | |
3398 | * | |
3399 | * Allocate an object from this cache. The flags are only relevant | |
3400 | * if the cache has no available objects. | |
3401 | */ | |
343e0d7a | 3402 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3403 | { |
48356303 | 3404 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3405 | |
505f5dcb | 3406 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3407 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3408 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3409 | |
3410 | return ret; | |
1da177e4 LT |
3411 | } |
3412 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3413 | ||
7b0501dd JDB |
3414 | static __always_inline void |
3415 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3416 | size_t size, void **p, unsigned long caller) | |
3417 | { | |
3418 | size_t i; | |
3419 | ||
3420 | for (i = 0; i < size; i++) | |
3421 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3422 | } | |
3423 | ||
865762a8 | 3424 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3425 | void **p) |
484748f0 | 3426 | { |
2a777eac JDB |
3427 | size_t i; |
3428 | ||
3429 | s = slab_pre_alloc_hook(s, flags); | |
3430 | if (!s) | |
3431 | return 0; | |
3432 | ||
3433 | cache_alloc_debugcheck_before(s, flags); | |
3434 | ||
3435 | local_irq_disable(); | |
3436 | for (i = 0; i < size; i++) { | |
3437 | void *objp = __do_cache_alloc(s, flags); | |
3438 | ||
2a777eac JDB |
3439 | if (unlikely(!objp)) |
3440 | goto error; | |
3441 | p[i] = objp; | |
3442 | } | |
3443 | local_irq_enable(); | |
3444 | ||
7b0501dd JDB |
3445 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3446 | ||
2a777eac JDB |
3447 | /* Clear memory outside IRQ disabled section */ |
3448 | if (unlikely(flags & __GFP_ZERO)) | |
3449 | for (i = 0; i < size; i++) | |
3450 | memset(p[i], 0, s->object_size); | |
3451 | ||
3452 | slab_post_alloc_hook(s, flags, size, p); | |
3453 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3454 | return size; | |
3455 | error: | |
3456 | local_irq_enable(); | |
7b0501dd | 3457 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3458 | slab_post_alloc_hook(s, flags, i, p); |
3459 | __kmem_cache_free_bulk(s, i, p); | |
3460 | return 0; | |
484748f0 CL |
3461 | } |
3462 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3463 | ||
0f24f128 | 3464 | #ifdef CONFIG_TRACING |
85beb586 | 3465 | void * |
4052147c | 3466 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3467 | { |
85beb586 SR |
3468 | void *ret; |
3469 | ||
48356303 | 3470 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3471 | |
505f5dcb | 3472 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3473 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3474 | size, cachep->size, flags); |
85beb586 | 3475 | return ret; |
36555751 | 3476 | } |
85beb586 | 3477 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3478 | #endif |
3479 | ||
1da177e4 | 3480 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3481 | /** |
3482 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3483 | * @cachep: The cache to allocate from. | |
3484 | * @flags: See kmalloc(). | |
3485 | * @nodeid: node number of the target node. | |
3486 | * | |
3487 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3488 | * node, which can improve the performance for cpu bound structures. | |
3489 | * | |
3490 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3491 | */ | |
8b98c169 CH |
3492 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3493 | { | |
48356303 | 3494 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3495 | |
505f5dcb | 3496 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3497 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3498 | cachep->object_size, cachep->size, |
ca2b84cb | 3499 | flags, nodeid); |
36555751 EGM |
3500 | |
3501 | return ret; | |
8b98c169 | 3502 | } |
1da177e4 LT |
3503 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3504 | ||
0f24f128 | 3505 | #ifdef CONFIG_TRACING |
4052147c | 3506 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3507 | gfp_t flags, |
4052147c EG |
3508 | int nodeid, |
3509 | size_t size) | |
36555751 | 3510 | { |
85beb586 SR |
3511 | void *ret; |
3512 | ||
592f4145 | 3513 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3514 | |
3515 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3516 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3517 | size, cachep->size, |
85beb586 SR |
3518 | flags, nodeid); |
3519 | return ret; | |
36555751 | 3520 | } |
85beb586 | 3521 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3522 | #endif |
3523 | ||
8b98c169 | 3524 | static __always_inline void * |
7c0cb9c6 | 3525 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3526 | { |
343e0d7a | 3527 | struct kmem_cache *cachep; |
7ed2f9e6 | 3528 | void *ret; |
97e2bde4 | 3529 | |
2c59dd65 | 3530 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3531 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3532 | return cachep; | |
7ed2f9e6 | 3533 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3534 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3535 | |
3536 | return ret; | |
97e2bde4 | 3537 | } |
8b98c169 | 3538 | |
8b98c169 CH |
3539 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3540 | { | |
7c0cb9c6 | 3541 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3542 | } |
dbe5e69d | 3543 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3544 | |
3545 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3546 | int node, unsigned long caller) |
8b98c169 | 3547 | { |
7c0cb9c6 | 3548 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3549 | } |
3550 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3551 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3552 | |
3553 | /** | |
800590f5 | 3554 | * __do_kmalloc - allocate memory |
1da177e4 | 3555 | * @size: how many bytes of memory are required. |
800590f5 | 3556 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3557 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3558 | */ |
7fd6b141 | 3559 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3560 | unsigned long caller) |
1da177e4 | 3561 | { |
343e0d7a | 3562 | struct kmem_cache *cachep; |
36555751 | 3563 | void *ret; |
1da177e4 | 3564 | |
2c59dd65 | 3565 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3566 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3567 | return cachep; | |
48356303 | 3568 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3569 | |
505f5dcb | 3570 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3571 | trace_kmalloc(caller, ret, |
3b0efdfa | 3572 | size, cachep->size, flags); |
36555751 EGM |
3573 | |
3574 | return ret; | |
7fd6b141 PE |
3575 | } |
3576 | ||
7fd6b141 PE |
3577 | void *__kmalloc(size_t size, gfp_t flags) |
3578 | { | |
7c0cb9c6 | 3579 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3580 | } |
3581 | EXPORT_SYMBOL(__kmalloc); | |
3582 | ||
ce71e27c | 3583 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3584 | { |
7c0cb9c6 | 3585 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3586 | } |
3587 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3588 | |
1da177e4 LT |
3589 | /** |
3590 | * kmem_cache_free - Deallocate an object | |
3591 | * @cachep: The cache the allocation was from. | |
3592 | * @objp: The previously allocated object. | |
3593 | * | |
3594 | * Free an object which was previously allocated from this | |
3595 | * cache. | |
3596 | */ | |
343e0d7a | 3597 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3598 | { |
3599 | unsigned long flags; | |
b9ce5ef4 GC |
3600 | cachep = cache_from_obj(cachep, objp); |
3601 | if (!cachep) | |
3602 | return; | |
1da177e4 LT |
3603 | |
3604 | local_irq_save(flags); | |
d97d476b | 3605 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3606 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3607 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3608 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3609 | local_irq_restore(flags); |
36555751 | 3610 | |
ca2b84cb | 3611 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3612 | } |
3613 | EXPORT_SYMBOL(kmem_cache_free); | |
3614 | ||
e6cdb58d JDB |
3615 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3616 | { | |
3617 | struct kmem_cache *s; | |
3618 | size_t i; | |
3619 | ||
3620 | local_irq_disable(); | |
3621 | for (i = 0; i < size; i++) { | |
3622 | void *objp = p[i]; | |
3623 | ||
ca257195 JDB |
3624 | if (!orig_s) /* called via kfree_bulk */ |
3625 | s = virt_to_cache(objp); | |
3626 | else | |
3627 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3628 | |
3629 | debug_check_no_locks_freed(objp, s->object_size); | |
3630 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3631 | debug_check_no_obj_freed(objp, s->object_size); | |
3632 | ||
3633 | __cache_free(s, objp, _RET_IP_); | |
3634 | } | |
3635 | local_irq_enable(); | |
3636 | ||
3637 | /* FIXME: add tracing */ | |
3638 | } | |
3639 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3640 | ||
1da177e4 LT |
3641 | /** |
3642 | * kfree - free previously allocated memory | |
3643 | * @objp: pointer returned by kmalloc. | |
3644 | * | |
80e93eff PE |
3645 | * If @objp is NULL, no operation is performed. |
3646 | * | |
1da177e4 LT |
3647 | * Don't free memory not originally allocated by kmalloc() |
3648 | * or you will run into trouble. | |
3649 | */ | |
3650 | void kfree(const void *objp) | |
3651 | { | |
343e0d7a | 3652 | struct kmem_cache *c; |
1da177e4 LT |
3653 | unsigned long flags; |
3654 | ||
2121db74 PE |
3655 | trace_kfree(_RET_IP_, objp); |
3656 | ||
6cb8f913 | 3657 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3658 | return; |
3659 | local_irq_save(flags); | |
3660 | kfree_debugcheck(objp); | |
6ed5eb22 | 3661 | c = virt_to_cache(objp); |
8c138bc0 CL |
3662 | debug_check_no_locks_freed(objp, c->object_size); |
3663 | ||
3664 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3665 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3666 | local_irq_restore(flags); |
3667 | } | |
3668 | EXPORT_SYMBOL(kfree); | |
3669 | ||
e498be7d | 3670 | /* |
ce8eb6c4 | 3671 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3672 | */ |
5f0985bb | 3673 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3674 | { |
3675 | int node; | |
ce8eb6c4 | 3676 | struct kmem_cache_node *n; |
cafeb02e | 3677 | struct array_cache *new_shared; |
c8522a3a | 3678 | struct alien_cache **new_alien = NULL; |
e498be7d | 3679 | |
9c09a95c | 3680 | for_each_online_node(node) { |
cafeb02e | 3681 | |
b455def2 L |
3682 | if (use_alien_caches) { |
3683 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3684 | if (!new_alien) | |
3685 | goto fail; | |
3686 | } | |
cafeb02e | 3687 | |
63109846 ED |
3688 | new_shared = NULL; |
3689 | if (cachep->shared) { | |
3690 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3691 | cachep->shared*cachep->batchcount, |
83b519e8 | 3692 | 0xbaadf00d, gfp); |
63109846 ED |
3693 | if (!new_shared) { |
3694 | free_alien_cache(new_alien); | |
3695 | goto fail; | |
3696 | } | |
0718dc2a | 3697 | } |
cafeb02e | 3698 | |
18bf8541 | 3699 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3700 | if (n) { |
3701 | struct array_cache *shared = n->shared; | |
97654dfa | 3702 | LIST_HEAD(list); |
cafeb02e | 3703 | |
ce8eb6c4 | 3704 | spin_lock_irq(&n->list_lock); |
e498be7d | 3705 | |
cafeb02e | 3706 | if (shared) |
0718dc2a | 3707 | free_block(cachep, shared->entry, |
97654dfa | 3708 | shared->avail, node, &list); |
e498be7d | 3709 | |
ce8eb6c4 CL |
3710 | n->shared = new_shared; |
3711 | if (!n->alien) { | |
3712 | n->alien = new_alien; | |
e498be7d CL |
3713 | new_alien = NULL; |
3714 | } | |
ce8eb6c4 | 3715 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3716 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3717 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3718 | slabs_destroy(cachep, &list); |
cafeb02e | 3719 | kfree(shared); |
e498be7d CL |
3720 | free_alien_cache(new_alien); |
3721 | continue; | |
3722 | } | |
ce8eb6c4 CL |
3723 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3724 | if (!n) { | |
0718dc2a CL |
3725 | free_alien_cache(new_alien); |
3726 | kfree(new_shared); | |
e498be7d | 3727 | goto fail; |
0718dc2a | 3728 | } |
e498be7d | 3729 | |
ce8eb6c4 | 3730 | kmem_cache_node_init(n); |
5f0985bb JZ |
3731 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3732 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3733 | n->shared = new_shared; |
3734 | n->alien = new_alien; | |
3735 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3736 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3737 | cachep->node[node] = n; |
e498be7d | 3738 | } |
cafeb02e | 3739 | return 0; |
0718dc2a | 3740 | |
a737b3e2 | 3741 | fail: |
3b0efdfa | 3742 | if (!cachep->list.next) { |
0718dc2a CL |
3743 | /* Cache is not active yet. Roll back what we did */ |
3744 | node--; | |
3745 | while (node >= 0) { | |
18bf8541 CL |
3746 | n = get_node(cachep, node); |
3747 | if (n) { | |
ce8eb6c4 CL |
3748 | kfree(n->shared); |
3749 | free_alien_cache(n->alien); | |
3750 | kfree(n); | |
6a67368c | 3751 | cachep->node[node] = NULL; |
0718dc2a CL |
3752 | } |
3753 | node--; | |
3754 | } | |
3755 | } | |
cafeb02e | 3756 | return -ENOMEM; |
e498be7d CL |
3757 | } |
3758 | ||
18004c5d | 3759 | /* Always called with the slab_mutex held */ |
943a451a | 3760 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3761 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3762 | { |
bf0dea23 JK |
3763 | struct array_cache __percpu *cpu_cache, *prev; |
3764 | int cpu; | |
1da177e4 | 3765 | |
bf0dea23 JK |
3766 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3767 | if (!cpu_cache) | |
d2e7b7d0 SS |
3768 | return -ENOMEM; |
3769 | ||
bf0dea23 JK |
3770 | prev = cachep->cpu_cache; |
3771 | cachep->cpu_cache = cpu_cache; | |
3772 | kick_all_cpus_sync(); | |
e498be7d | 3773 | |
1da177e4 | 3774 | check_irq_on(); |
1da177e4 LT |
3775 | cachep->batchcount = batchcount; |
3776 | cachep->limit = limit; | |
e498be7d | 3777 | cachep->shared = shared; |
1da177e4 | 3778 | |
bf0dea23 JK |
3779 | if (!prev) |
3780 | goto alloc_node; | |
3781 | ||
3782 | for_each_online_cpu(cpu) { | |
97654dfa | 3783 | LIST_HEAD(list); |
18bf8541 CL |
3784 | int node; |
3785 | struct kmem_cache_node *n; | |
bf0dea23 | 3786 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3787 | |
bf0dea23 | 3788 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3789 | n = get_node(cachep, node); |
3790 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3791 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3792 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3793 | slabs_destroy(cachep, &list); |
1da177e4 | 3794 | } |
bf0dea23 JK |
3795 | free_percpu(prev); |
3796 | ||
3797 | alloc_node: | |
5f0985bb | 3798 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3799 | } |
3800 | ||
943a451a GC |
3801 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3802 | int batchcount, int shared, gfp_t gfp) | |
3803 | { | |
3804 | int ret; | |
426589f5 | 3805 | struct kmem_cache *c; |
943a451a GC |
3806 | |
3807 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3808 | ||
3809 | if (slab_state < FULL) | |
3810 | return ret; | |
3811 | ||
3812 | if ((ret < 0) || !is_root_cache(cachep)) | |
3813 | return ret; | |
3814 | ||
426589f5 VD |
3815 | lockdep_assert_held(&slab_mutex); |
3816 | for_each_memcg_cache(c, cachep) { | |
3817 | /* return value determined by the root cache only */ | |
3818 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3819 | } |
3820 | ||
3821 | return ret; | |
3822 | } | |
3823 | ||
18004c5d | 3824 | /* Called with slab_mutex held always */ |
83b519e8 | 3825 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3826 | { |
3827 | int err; | |
943a451a GC |
3828 | int limit = 0; |
3829 | int shared = 0; | |
3830 | int batchcount = 0; | |
3831 | ||
3832 | if (!is_root_cache(cachep)) { | |
3833 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3834 | limit = root->limit; | |
3835 | shared = root->shared; | |
3836 | batchcount = root->batchcount; | |
3837 | } | |
1da177e4 | 3838 | |
943a451a GC |
3839 | if (limit && shared && batchcount) |
3840 | goto skip_setup; | |
a737b3e2 AM |
3841 | /* |
3842 | * The head array serves three purposes: | |
1da177e4 LT |
3843 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3844 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3845 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3846 | * bufctl chains: array operations are cheaper. |
3847 | * The numbers are guessed, we should auto-tune as described by | |
3848 | * Bonwick. | |
3849 | */ | |
3b0efdfa | 3850 | if (cachep->size > 131072) |
1da177e4 | 3851 | limit = 1; |
3b0efdfa | 3852 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3853 | limit = 8; |
3b0efdfa | 3854 | else if (cachep->size > 1024) |
1da177e4 | 3855 | limit = 24; |
3b0efdfa | 3856 | else if (cachep->size > 256) |
1da177e4 LT |
3857 | limit = 54; |
3858 | else | |
3859 | limit = 120; | |
3860 | ||
a737b3e2 AM |
3861 | /* |
3862 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3863 | * allocation behaviour: Most allocs on one cpu, most free operations |
3864 | * on another cpu. For these cases, an efficient object passing between | |
3865 | * cpus is necessary. This is provided by a shared array. The array | |
3866 | * replaces Bonwick's magazine layer. | |
3867 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3868 | * to a larger limit. Thus disabled by default. | |
3869 | */ | |
3870 | shared = 0; | |
3b0efdfa | 3871 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3872 | shared = 8; |
1da177e4 LT |
3873 | |
3874 | #if DEBUG | |
a737b3e2 AM |
3875 | /* |
3876 | * With debugging enabled, large batchcount lead to excessively long | |
3877 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3878 | */ |
3879 | if (limit > 32) | |
3880 | limit = 32; | |
3881 | #endif | |
943a451a GC |
3882 | batchcount = (limit + 1) / 2; |
3883 | skip_setup: | |
3884 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 | 3885 | if (err) |
1170532b | 3886 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3887 | cachep->name, -err); |
2ed3a4ef | 3888 | return err; |
1da177e4 LT |
3889 | } |
3890 | ||
1b55253a | 3891 | /* |
ce8eb6c4 CL |
3892 | * Drain an array if it contains any elements taking the node lock only if |
3893 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3894 | * if drain_array() is used on the shared array. |
1b55253a | 3895 | */ |
ce8eb6c4 | 3896 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 3897 | struct array_cache *ac, int node) |
1da177e4 | 3898 | { |
97654dfa | 3899 | LIST_HEAD(list); |
18726ca8 JK |
3900 | |
3901 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
3902 | check_mutex_acquired(); | |
1da177e4 | 3903 | |
1b55253a CL |
3904 | if (!ac || !ac->avail) |
3905 | return; | |
18726ca8 JK |
3906 | |
3907 | if (ac->touched) { | |
1da177e4 | 3908 | ac->touched = 0; |
18726ca8 | 3909 | return; |
1da177e4 | 3910 | } |
18726ca8 JK |
3911 | |
3912 | spin_lock_irq(&n->list_lock); | |
3913 | drain_array_locked(cachep, ac, node, false, &list); | |
3914 | spin_unlock_irq(&n->list_lock); | |
3915 | ||
3916 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
3917 | } |
3918 | ||
3919 | /** | |
3920 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3921 | * @w: work descriptor |
1da177e4 LT |
3922 | * |
3923 | * Called from workqueue/eventd every few seconds. | |
3924 | * Purpose: | |
3925 | * - clear the per-cpu caches for this CPU. | |
3926 | * - return freeable pages to the main free memory pool. | |
3927 | * | |
a737b3e2 AM |
3928 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3929 | * again on the next iteration. | |
1da177e4 | 3930 | */ |
7c5cae36 | 3931 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3932 | { |
7a7c381d | 3933 | struct kmem_cache *searchp; |
ce8eb6c4 | 3934 | struct kmem_cache_node *n; |
7d6e6d09 | 3935 | int node = numa_mem_id(); |
bf6aede7 | 3936 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3937 | |
18004c5d | 3938 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3939 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3940 | goto out; |
1da177e4 | 3941 | |
18004c5d | 3942 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3943 | check_irq_on(); |
3944 | ||
35386e3b | 3945 | /* |
ce8eb6c4 | 3946 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3947 | * have established with reasonable certainty that |
3948 | * we can do some work if the lock was obtained. | |
3949 | */ | |
18bf8541 | 3950 | n = get_node(searchp, node); |
35386e3b | 3951 | |
ce8eb6c4 | 3952 | reap_alien(searchp, n); |
1da177e4 | 3953 | |
18726ca8 | 3954 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 3955 | |
35386e3b CL |
3956 | /* |
3957 | * These are racy checks but it does not matter | |
3958 | * if we skip one check or scan twice. | |
3959 | */ | |
ce8eb6c4 | 3960 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3961 | goto next; |
1da177e4 | 3962 | |
5f0985bb | 3963 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3964 | |
18726ca8 | 3965 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 3966 | |
ce8eb6c4 CL |
3967 | if (n->free_touched) |
3968 | n->free_touched = 0; | |
ed11d9eb CL |
3969 | else { |
3970 | int freed; | |
1da177e4 | 3971 | |
ce8eb6c4 | 3972 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3973 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3974 | STATS_ADD_REAPED(searchp, freed); | |
3975 | } | |
35386e3b | 3976 | next: |
1da177e4 LT |
3977 | cond_resched(); |
3978 | } | |
3979 | check_irq_on(); | |
18004c5d | 3980 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3981 | next_reap_node(); |
7c5cae36 | 3982 | out: |
a737b3e2 | 3983 | /* Set up the next iteration */ |
5f0985bb | 3984 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3985 | } |
3986 | ||
158a9624 | 3987 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3988 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3989 | { |
8456a648 | 3990 | struct page *page; |
b28a02de PE |
3991 | unsigned long active_objs; |
3992 | unsigned long num_objs; | |
3993 | unsigned long active_slabs = 0; | |
3994 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3995 | const char *name; |
1da177e4 | 3996 | char *error = NULL; |
e498be7d | 3997 | int node; |
ce8eb6c4 | 3998 | struct kmem_cache_node *n; |
1da177e4 | 3999 | |
1da177e4 LT |
4000 | active_objs = 0; |
4001 | num_slabs = 0; | |
18bf8541 | 4002 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 4003 | |
ca3b9b91 | 4004 | check_irq_on(); |
ce8eb6c4 | 4005 | spin_lock_irq(&n->list_lock); |
e498be7d | 4006 | |
8456a648 JK |
4007 | list_for_each_entry(page, &n->slabs_full, lru) { |
4008 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4009 | error = "slabs_full accounting error"; |
4010 | active_objs += cachep->num; | |
4011 | active_slabs++; | |
4012 | } | |
8456a648 JK |
4013 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4014 | if (page->active == cachep->num && !error) | |
106a74e1 | 4015 | error = "slabs_partial accounting error"; |
8456a648 | 4016 | if (!page->active && !error) |
106a74e1 | 4017 | error = "slabs_partial accounting error"; |
8456a648 | 4018 | active_objs += page->active; |
e498be7d CL |
4019 | active_slabs++; |
4020 | } | |
8456a648 JK |
4021 | list_for_each_entry(page, &n->slabs_free, lru) { |
4022 | if (page->active && !error) | |
106a74e1 | 4023 | error = "slabs_free accounting error"; |
e498be7d CL |
4024 | num_slabs++; |
4025 | } | |
ce8eb6c4 CL |
4026 | free_objects += n->free_objects; |
4027 | if (n->shared) | |
4028 | shared_avail += n->shared->avail; | |
e498be7d | 4029 | |
ce8eb6c4 | 4030 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4031 | } |
b28a02de PE |
4032 | num_slabs += active_slabs; |
4033 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4034 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4035 | error = "free_objects accounting error"; |
4036 | ||
b28a02de | 4037 | name = cachep->name; |
1da177e4 | 4038 | if (error) |
1170532b | 4039 | pr_err("slab: cache %s error: %s\n", name, error); |
1da177e4 | 4040 | |
0d7561c6 GC |
4041 | sinfo->active_objs = active_objs; |
4042 | sinfo->num_objs = num_objs; | |
4043 | sinfo->active_slabs = active_slabs; | |
4044 | sinfo->num_slabs = num_slabs; | |
4045 | sinfo->shared_avail = shared_avail; | |
4046 | sinfo->limit = cachep->limit; | |
4047 | sinfo->batchcount = cachep->batchcount; | |
4048 | sinfo->shared = cachep->shared; | |
4049 | sinfo->objects_per_slab = cachep->num; | |
4050 | sinfo->cache_order = cachep->gfporder; | |
4051 | } | |
4052 | ||
4053 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4054 | { | |
1da177e4 | 4055 | #if STATS |
ce8eb6c4 | 4056 | { /* node stats */ |
1da177e4 LT |
4057 | unsigned long high = cachep->high_mark; |
4058 | unsigned long allocs = cachep->num_allocations; | |
4059 | unsigned long grown = cachep->grown; | |
4060 | unsigned long reaped = cachep->reaped; | |
4061 | unsigned long errors = cachep->errors; | |
4062 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4063 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4064 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4065 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4066 | |
756a025f | 4067 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4068 | allocs, high, grown, |
4069 | reaped, errors, max_freeable, node_allocs, | |
4070 | node_frees, overflows); | |
1da177e4 LT |
4071 | } |
4072 | /* cpu stats */ | |
4073 | { | |
4074 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4075 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4076 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4077 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4078 | ||
4079 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4080 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4081 | } |
4082 | #endif | |
1da177e4 LT |
4083 | } |
4084 | ||
1da177e4 LT |
4085 | #define MAX_SLABINFO_WRITE 128 |
4086 | /** | |
4087 | * slabinfo_write - Tuning for the slab allocator | |
4088 | * @file: unused | |
4089 | * @buffer: user buffer | |
4090 | * @count: data length | |
4091 | * @ppos: unused | |
4092 | */ | |
b7454ad3 | 4093 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4094 | size_t count, loff_t *ppos) |
1da177e4 | 4095 | { |
b28a02de | 4096 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4097 | int limit, batchcount, shared, res; |
7a7c381d | 4098 | struct kmem_cache *cachep; |
b28a02de | 4099 | |
1da177e4 LT |
4100 | if (count > MAX_SLABINFO_WRITE) |
4101 | return -EINVAL; | |
4102 | if (copy_from_user(&kbuf, buffer, count)) | |
4103 | return -EFAULT; | |
b28a02de | 4104 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4105 | |
4106 | tmp = strchr(kbuf, ' '); | |
4107 | if (!tmp) | |
4108 | return -EINVAL; | |
4109 | *tmp = '\0'; | |
4110 | tmp++; | |
4111 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4112 | return -EINVAL; | |
4113 | ||
4114 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4115 | mutex_lock(&slab_mutex); |
1da177e4 | 4116 | res = -EINVAL; |
18004c5d | 4117 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4118 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4119 | if (limit < 1 || batchcount < 1 || |
4120 | batchcount > limit || shared < 0) { | |
e498be7d | 4121 | res = 0; |
1da177e4 | 4122 | } else { |
e498be7d | 4123 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4124 | batchcount, shared, |
4125 | GFP_KERNEL); | |
1da177e4 LT |
4126 | } |
4127 | break; | |
4128 | } | |
4129 | } | |
18004c5d | 4130 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4131 | if (res >= 0) |
4132 | res = count; | |
4133 | return res; | |
4134 | } | |
871751e2 AV |
4135 | |
4136 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4137 | ||
871751e2 AV |
4138 | static inline int add_caller(unsigned long *n, unsigned long v) |
4139 | { | |
4140 | unsigned long *p; | |
4141 | int l; | |
4142 | if (!v) | |
4143 | return 1; | |
4144 | l = n[1]; | |
4145 | p = n + 2; | |
4146 | while (l) { | |
4147 | int i = l/2; | |
4148 | unsigned long *q = p + 2 * i; | |
4149 | if (*q == v) { | |
4150 | q[1]++; | |
4151 | return 1; | |
4152 | } | |
4153 | if (*q > v) { | |
4154 | l = i; | |
4155 | } else { | |
4156 | p = q + 2; | |
4157 | l -= i + 1; | |
4158 | } | |
4159 | } | |
4160 | if (++n[1] == n[0]) | |
4161 | return 0; | |
4162 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4163 | p[0] = v; | |
4164 | p[1] = 1; | |
4165 | return 1; | |
4166 | } | |
4167 | ||
8456a648 JK |
4168 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4169 | struct page *page) | |
871751e2 AV |
4170 | { |
4171 | void *p; | |
d31676df JK |
4172 | int i, j; |
4173 | unsigned long v; | |
b1cb0982 | 4174 | |
871751e2 AV |
4175 | if (n[0] == n[1]) |
4176 | return; | |
8456a648 | 4177 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4178 | bool active = true; |
4179 | ||
4180 | for (j = page->active; j < c->num; j++) { | |
4181 | if (get_free_obj(page, j) == i) { | |
4182 | active = false; | |
4183 | break; | |
4184 | } | |
4185 | } | |
4186 | ||
4187 | if (!active) | |
871751e2 | 4188 | continue; |
b1cb0982 | 4189 | |
d31676df JK |
4190 | /* |
4191 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4192 | * mapping is established when actual object allocation and | |
4193 | * we could mistakenly access the unmapped object in the cpu | |
4194 | * cache. | |
4195 | */ | |
4196 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4197 | continue; | |
4198 | ||
4199 | if (!add_caller(n, v)) | |
871751e2 AV |
4200 | return; |
4201 | } | |
4202 | } | |
4203 | ||
4204 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4205 | { | |
4206 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4207 | unsigned long offset, size; |
9281acea | 4208 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4209 | |
a5c43dae | 4210 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4211 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4212 | if (modname[0]) |
871751e2 AV |
4213 | seq_printf(m, " [%s]", modname); |
4214 | return; | |
4215 | } | |
4216 | #endif | |
4217 | seq_printf(m, "%p", (void *)address); | |
4218 | } | |
4219 | ||
4220 | static int leaks_show(struct seq_file *m, void *p) | |
4221 | { | |
0672aa7c | 4222 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4223 | struct page *page; |
ce8eb6c4 | 4224 | struct kmem_cache_node *n; |
871751e2 | 4225 | const char *name; |
db845067 | 4226 | unsigned long *x = m->private; |
871751e2 AV |
4227 | int node; |
4228 | int i; | |
4229 | ||
4230 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4231 | return 0; | |
4232 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4233 | return 0; | |
4234 | ||
d31676df JK |
4235 | /* |
4236 | * Set store_user_clean and start to grab stored user information | |
4237 | * for all objects on this cache. If some alloc/free requests comes | |
4238 | * during the processing, information would be wrong so restart | |
4239 | * whole processing. | |
4240 | */ | |
4241 | do { | |
4242 | set_store_user_clean(cachep); | |
4243 | drain_cpu_caches(cachep); | |
4244 | ||
4245 | x[1] = 0; | |
871751e2 | 4246 | |
d31676df | 4247 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4248 | |
d31676df JK |
4249 | check_irq_on(); |
4250 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4251 | |
d31676df JK |
4252 | list_for_each_entry(page, &n->slabs_full, lru) |
4253 | handle_slab(x, cachep, page); | |
4254 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4255 | handle_slab(x, cachep, page); | |
4256 | spin_unlock_irq(&n->list_lock); | |
4257 | } | |
4258 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4259 | |
871751e2 | 4260 | name = cachep->name; |
db845067 | 4261 | if (x[0] == x[1]) { |
871751e2 | 4262 | /* Increase the buffer size */ |
18004c5d | 4263 | mutex_unlock(&slab_mutex); |
db845067 | 4264 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4265 | if (!m->private) { |
4266 | /* Too bad, we are really out */ | |
db845067 | 4267 | m->private = x; |
18004c5d | 4268 | mutex_lock(&slab_mutex); |
871751e2 AV |
4269 | return -ENOMEM; |
4270 | } | |
db845067 CL |
4271 | *(unsigned long *)m->private = x[0] * 2; |
4272 | kfree(x); | |
18004c5d | 4273 | mutex_lock(&slab_mutex); |
871751e2 AV |
4274 | /* Now make sure this entry will be retried */ |
4275 | m->count = m->size; | |
4276 | return 0; | |
4277 | } | |
db845067 CL |
4278 | for (i = 0; i < x[1]; i++) { |
4279 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4280 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4281 | seq_putc(m, '\n'); |
4282 | } | |
d2e7b7d0 | 4283 | |
871751e2 AV |
4284 | return 0; |
4285 | } | |
4286 | ||
a0ec95a8 | 4287 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4288 | .start = slab_start, |
276a2439 WL |
4289 | .next = slab_next, |
4290 | .stop = slab_stop, | |
871751e2 AV |
4291 | .show = leaks_show, |
4292 | }; | |
a0ec95a8 AD |
4293 | |
4294 | static int slabstats_open(struct inode *inode, struct file *file) | |
4295 | { | |
b208ce32 RJ |
4296 | unsigned long *n; |
4297 | ||
4298 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4299 | if (!n) | |
4300 | return -ENOMEM; | |
4301 | ||
4302 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4303 | ||
4304 | return 0; | |
a0ec95a8 AD |
4305 | } |
4306 | ||
4307 | static const struct file_operations proc_slabstats_operations = { | |
4308 | .open = slabstats_open, | |
4309 | .read = seq_read, | |
4310 | .llseek = seq_lseek, | |
4311 | .release = seq_release_private, | |
4312 | }; | |
4313 | #endif | |
4314 | ||
4315 | static int __init slab_proc_init(void) | |
4316 | { | |
4317 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4318 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4319 | #endif |
a0ec95a8 AD |
4320 | return 0; |
4321 | } | |
4322 | module_init(slab_proc_init); | |
1da177e4 LT |
4323 | #endif |
4324 | ||
00e145b6 MS |
4325 | /** |
4326 | * ksize - get the actual amount of memory allocated for a given object | |
4327 | * @objp: Pointer to the object | |
4328 | * | |
4329 | * kmalloc may internally round up allocations and return more memory | |
4330 | * than requested. ksize() can be used to determine the actual amount of | |
4331 | * memory allocated. The caller may use this additional memory, even though | |
4332 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4333 | * The caller must guarantee that objp points to a valid object previously | |
4334 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4335 | * must not be freed during the duration of the call. | |
4336 | */ | |
fd76bab2 | 4337 | size_t ksize(const void *objp) |
1da177e4 | 4338 | { |
7ed2f9e6 AP |
4339 | size_t size; |
4340 | ||
ef8b4520 CL |
4341 | BUG_ON(!objp); |
4342 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4343 | return 0; |
1da177e4 | 4344 | |
7ed2f9e6 AP |
4345 | size = virt_to_cache(objp)->object_size; |
4346 | /* We assume that ksize callers could use the whole allocated area, | |
4347 | * so we need to unpoison this area. | |
4348 | */ | |
505f5dcb | 4349 | kasan_krealloc(objp, size, GFP_NOWAIT); |
7ed2f9e6 AP |
4350 | |
4351 | return size; | |
1da177e4 | 4352 | } |
b1aabecd | 4353 | EXPORT_SYMBOL(ksize); |