]>
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 | ||
1871e52c | 424 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 425 | |
343e0d7a | 426 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 427 | { |
bf0dea23 | 428 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
429 | } |
430 | ||
a737b3e2 AM |
431 | /* |
432 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
433 | */ | |
70f75067 JK |
434 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
435 | unsigned long flags, size_t *left_over) | |
fbaccacf | 436 | { |
70f75067 | 437 | unsigned int num; |
fbaccacf | 438 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 439 | |
fbaccacf SR |
440 | /* |
441 | * The slab management structure can be either off the slab or | |
442 | * on it. For the latter case, the memory allocated for a | |
443 | * slab is used for: | |
444 | * | |
fbaccacf | 445 | * - @buffer_size bytes for each object |
2e6b3602 JK |
446 | * - One freelist_idx_t for each object |
447 | * | |
448 | * We don't need to consider alignment of freelist because | |
449 | * freelist will be at the end of slab page. The objects will be | |
450 | * at the correct alignment. | |
fbaccacf SR |
451 | * |
452 | * If the slab management structure is off the slab, then the | |
453 | * alignment will already be calculated into the size. Because | |
454 | * the slabs are all pages aligned, the objects will be at the | |
455 | * correct alignment when allocated. | |
456 | */ | |
b03a017b | 457 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 458 | num = slab_size / buffer_size; |
2e6b3602 | 459 | *left_over = slab_size % buffer_size; |
fbaccacf | 460 | } else { |
70f75067 | 461 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
462 | *left_over = slab_size % |
463 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 464 | } |
70f75067 JK |
465 | |
466 | return num; | |
1da177e4 LT |
467 | } |
468 | ||
f28510d3 | 469 | #if DEBUG |
d40cee24 | 470 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 471 | |
a737b3e2 AM |
472 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
473 | char *msg) | |
1da177e4 | 474 | { |
1170532b | 475 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 476 | function, cachep->name, msg); |
1da177e4 | 477 | dump_stack(); |
373d4d09 | 478 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 479 | } |
f28510d3 | 480 | #endif |
1da177e4 | 481 | |
3395ee05 PM |
482 | /* |
483 | * By default on NUMA we use alien caches to stage the freeing of | |
484 | * objects allocated from other nodes. This causes massive memory | |
485 | * inefficiencies when using fake NUMA setup to split memory into a | |
486 | * large number of small nodes, so it can be disabled on the command | |
487 | * line | |
488 | */ | |
489 | ||
490 | static int use_alien_caches __read_mostly = 1; | |
491 | static int __init noaliencache_setup(char *s) | |
492 | { | |
493 | use_alien_caches = 0; | |
494 | return 1; | |
495 | } | |
496 | __setup("noaliencache", noaliencache_setup); | |
497 | ||
3df1cccd DR |
498 | static int __init slab_max_order_setup(char *str) |
499 | { | |
500 | get_option(&str, &slab_max_order); | |
501 | slab_max_order = slab_max_order < 0 ? 0 : | |
502 | min(slab_max_order, MAX_ORDER - 1); | |
503 | slab_max_order_set = true; | |
504 | ||
505 | return 1; | |
506 | } | |
507 | __setup("slab_max_order=", slab_max_order_setup); | |
508 | ||
8fce4d8e CL |
509 | #ifdef CONFIG_NUMA |
510 | /* | |
511 | * Special reaping functions for NUMA systems called from cache_reap(). | |
512 | * These take care of doing round robin flushing of alien caches (containing | |
513 | * objects freed on different nodes from which they were allocated) and the | |
514 | * flushing of remote pcps by calling drain_node_pages. | |
515 | */ | |
1871e52c | 516 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
517 | |
518 | static void init_reap_node(int cpu) | |
519 | { | |
520 | int node; | |
521 | ||
7d6e6d09 | 522 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 523 | if (node == MAX_NUMNODES) |
442295c9 | 524 | node = first_node(node_online_map); |
8fce4d8e | 525 | |
1871e52c | 526 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
527 | } |
528 | ||
529 | static void next_reap_node(void) | |
530 | { | |
909ea964 | 531 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 532 | |
8fce4d8e CL |
533 | node = next_node(node, node_online_map); |
534 | if (unlikely(node >= MAX_NUMNODES)) | |
535 | node = first_node(node_online_map); | |
909ea964 | 536 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
537 | } |
538 | ||
539 | #else | |
540 | #define init_reap_node(cpu) do { } while (0) | |
541 | #define next_reap_node(void) do { } while (0) | |
542 | #endif | |
543 | ||
1da177e4 LT |
544 | /* |
545 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
546 | * via the workqueue/eventd. | |
547 | * Add the CPU number into the expiration time to minimize the possibility of | |
548 | * the CPUs getting into lockstep and contending for the global cache chain | |
549 | * lock. | |
550 | */ | |
0db0628d | 551 | static void start_cpu_timer(int cpu) |
1da177e4 | 552 | { |
1871e52c | 553 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
554 | |
555 | /* | |
556 | * When this gets called from do_initcalls via cpucache_init(), | |
557 | * init_workqueues() has already run, so keventd will be setup | |
558 | * at that time. | |
559 | */ | |
52bad64d | 560 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 561 | init_reap_node(cpu); |
203b42f7 | 562 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
563 | schedule_delayed_work_on(cpu, reap_work, |
564 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
565 | } |
566 | } | |
567 | ||
1fe00d50 | 568 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 569 | { |
d5cff635 CM |
570 | /* |
571 | * The array_cache structures contain pointers to free object. | |
25985edc | 572 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
573 | * cache the pointers are not cleared and they could be counted as |
574 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
575 | * not scan such objects. | |
576 | */ | |
1fe00d50 JK |
577 | kmemleak_no_scan(ac); |
578 | if (ac) { | |
579 | ac->avail = 0; | |
580 | ac->limit = limit; | |
581 | ac->batchcount = batch; | |
582 | ac->touched = 0; | |
1da177e4 | 583 | } |
1fe00d50 JK |
584 | } |
585 | ||
586 | static struct array_cache *alloc_arraycache(int node, int entries, | |
587 | int batchcount, gfp_t gfp) | |
588 | { | |
5e804789 | 589 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
590 | struct array_cache *ac = NULL; |
591 | ||
592 | ac = kmalloc_node(memsize, gfp, node); | |
593 | init_arraycache(ac, entries, batchcount); | |
594 | return ac; | |
1da177e4 LT |
595 | } |
596 | ||
f68f8ddd JK |
597 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
598 | struct page *page, void *objp) | |
072bb0aa | 599 | { |
f68f8ddd JK |
600 | struct kmem_cache_node *n; |
601 | int page_node; | |
602 | LIST_HEAD(list); | |
072bb0aa | 603 | |
f68f8ddd JK |
604 | page_node = page_to_nid(page); |
605 | n = get_node(cachep, page_node); | |
381760ea | 606 | |
f68f8ddd JK |
607 | spin_lock(&n->list_lock); |
608 | free_block(cachep, &objp, 1, page_node, &list); | |
609 | spin_unlock(&n->list_lock); | |
381760ea | 610 | |
f68f8ddd | 611 | slabs_destroy(cachep, &list); |
072bb0aa MG |
612 | } |
613 | ||
3ded175a CL |
614 | /* |
615 | * Transfer objects in one arraycache to another. | |
616 | * Locking must be handled by the caller. | |
617 | * | |
618 | * Return the number of entries transferred. | |
619 | */ | |
620 | static int transfer_objects(struct array_cache *to, | |
621 | struct array_cache *from, unsigned int max) | |
622 | { | |
623 | /* Figure out how many entries to transfer */ | |
732eacc0 | 624 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
625 | |
626 | if (!nr) | |
627 | return 0; | |
628 | ||
629 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
630 | sizeof(void *) *nr); | |
631 | ||
632 | from->avail -= nr; | |
633 | to->avail += nr; | |
3ded175a CL |
634 | return nr; |
635 | } | |
636 | ||
765c4507 CL |
637 | #ifndef CONFIG_NUMA |
638 | ||
639 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 640 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 641 | |
c8522a3a JK |
642 | static inline struct alien_cache **alloc_alien_cache(int node, |
643 | int limit, gfp_t gfp) | |
765c4507 | 644 | { |
8888177e | 645 | return NULL; |
765c4507 CL |
646 | } |
647 | ||
c8522a3a | 648 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
649 | { |
650 | } | |
651 | ||
652 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
653 | { | |
654 | return 0; | |
655 | } | |
656 | ||
657 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
658 | gfp_t flags) | |
659 | { | |
660 | return NULL; | |
661 | } | |
662 | ||
8b98c169 | 663 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
664 | gfp_t flags, int nodeid) |
665 | { | |
666 | return NULL; | |
667 | } | |
668 | ||
4167e9b2 DR |
669 | static inline gfp_t gfp_exact_node(gfp_t flags) |
670 | { | |
444eb2a4 | 671 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
672 | } |
673 | ||
765c4507 CL |
674 | #else /* CONFIG_NUMA */ |
675 | ||
8b98c169 | 676 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 677 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 678 | |
c8522a3a JK |
679 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
680 | int batch, gfp_t gfp) | |
681 | { | |
5e804789 | 682 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
683 | struct alien_cache *alc = NULL; |
684 | ||
685 | alc = kmalloc_node(memsize, gfp, node); | |
686 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 687 | spin_lock_init(&alc->lock); |
c8522a3a JK |
688 | return alc; |
689 | } | |
690 | ||
691 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 692 | { |
c8522a3a | 693 | struct alien_cache **alc_ptr; |
5e804789 | 694 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
695 | int i; |
696 | ||
697 | if (limit > 1) | |
698 | limit = 12; | |
c8522a3a JK |
699 | alc_ptr = kzalloc_node(memsize, gfp, node); |
700 | if (!alc_ptr) | |
701 | return NULL; | |
702 | ||
703 | for_each_node(i) { | |
704 | if (i == node || !node_online(i)) | |
705 | continue; | |
706 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
707 | if (!alc_ptr[i]) { | |
708 | for (i--; i >= 0; i--) | |
709 | kfree(alc_ptr[i]); | |
710 | kfree(alc_ptr); | |
711 | return NULL; | |
e498be7d CL |
712 | } |
713 | } | |
c8522a3a | 714 | return alc_ptr; |
e498be7d CL |
715 | } |
716 | ||
c8522a3a | 717 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
718 | { |
719 | int i; | |
720 | ||
c8522a3a | 721 | if (!alc_ptr) |
e498be7d | 722 | return; |
e498be7d | 723 | for_each_node(i) |
c8522a3a JK |
724 | kfree(alc_ptr[i]); |
725 | kfree(alc_ptr); | |
e498be7d CL |
726 | } |
727 | ||
343e0d7a | 728 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
729 | struct array_cache *ac, int node, |
730 | struct list_head *list) | |
e498be7d | 731 | { |
18bf8541 | 732 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
733 | |
734 | if (ac->avail) { | |
ce8eb6c4 | 735 | spin_lock(&n->list_lock); |
e00946fe CL |
736 | /* |
737 | * Stuff objects into the remote nodes shared array first. | |
738 | * That way we could avoid the overhead of putting the objects | |
739 | * into the free lists and getting them back later. | |
740 | */ | |
ce8eb6c4 CL |
741 | if (n->shared) |
742 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 743 | |
833b706c | 744 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 745 | ac->avail = 0; |
ce8eb6c4 | 746 | spin_unlock(&n->list_lock); |
e498be7d CL |
747 | } |
748 | } | |
749 | ||
8fce4d8e CL |
750 | /* |
751 | * Called from cache_reap() to regularly drain alien caches round robin. | |
752 | */ | |
ce8eb6c4 | 753 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 754 | { |
909ea964 | 755 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 756 | |
ce8eb6c4 | 757 | if (n->alien) { |
c8522a3a JK |
758 | struct alien_cache *alc = n->alien[node]; |
759 | struct array_cache *ac; | |
760 | ||
761 | if (alc) { | |
762 | ac = &alc->ac; | |
49dfc304 | 763 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
764 | LIST_HEAD(list); |
765 | ||
766 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 767 | spin_unlock_irq(&alc->lock); |
833b706c | 768 | slabs_destroy(cachep, &list); |
c8522a3a | 769 | } |
8fce4d8e CL |
770 | } |
771 | } | |
772 | } | |
773 | ||
a737b3e2 | 774 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 775 | struct alien_cache **alien) |
e498be7d | 776 | { |
b28a02de | 777 | int i = 0; |
c8522a3a | 778 | struct alien_cache *alc; |
e498be7d CL |
779 | struct array_cache *ac; |
780 | unsigned long flags; | |
781 | ||
782 | for_each_online_node(i) { | |
c8522a3a JK |
783 | alc = alien[i]; |
784 | if (alc) { | |
833b706c JK |
785 | LIST_HEAD(list); |
786 | ||
c8522a3a | 787 | ac = &alc->ac; |
49dfc304 | 788 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 789 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 790 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 791 | slabs_destroy(cachep, &list); |
e498be7d CL |
792 | } |
793 | } | |
794 | } | |
729bd0b7 | 795 | |
25c4f304 JK |
796 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
797 | int node, int page_node) | |
729bd0b7 | 798 | { |
ce8eb6c4 | 799 | struct kmem_cache_node *n; |
c8522a3a JK |
800 | struct alien_cache *alien = NULL; |
801 | struct array_cache *ac; | |
97654dfa | 802 | LIST_HEAD(list); |
1ca4cb24 | 803 | |
18bf8541 | 804 | n = get_node(cachep, node); |
729bd0b7 | 805 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
806 | if (n->alien && n->alien[page_node]) { |
807 | alien = n->alien[page_node]; | |
c8522a3a | 808 | ac = &alien->ac; |
49dfc304 | 809 | spin_lock(&alien->lock); |
c8522a3a | 810 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 811 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 812 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 813 | } |
f68f8ddd | 814 | ac->entry[ac->avail++] = objp; |
49dfc304 | 815 | spin_unlock(&alien->lock); |
833b706c | 816 | slabs_destroy(cachep, &list); |
729bd0b7 | 817 | } else { |
25c4f304 | 818 | n = get_node(cachep, page_node); |
18bf8541 | 819 | spin_lock(&n->list_lock); |
25c4f304 | 820 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 821 | spin_unlock(&n->list_lock); |
97654dfa | 822 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
823 | } |
824 | return 1; | |
825 | } | |
25c4f304 JK |
826 | |
827 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
828 | { | |
829 | int page_node = page_to_nid(virt_to_page(objp)); | |
830 | int node = numa_mem_id(); | |
831 | /* | |
832 | * Make sure we are not freeing a object from another node to the array | |
833 | * cache on this cpu. | |
834 | */ | |
835 | if (likely(node == page_node)) | |
836 | return 0; | |
837 | ||
838 | return __cache_free_alien(cachep, objp, node, page_node); | |
839 | } | |
4167e9b2 DR |
840 | |
841 | /* | |
444eb2a4 MG |
842 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
843 | * warn about failures. | |
4167e9b2 DR |
844 | */ |
845 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
846 | { | |
444eb2a4 | 847 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 848 | } |
e498be7d CL |
849 | #endif |
850 | ||
8f9f8d9e | 851 | /* |
6a67368c | 852 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 853 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 854 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 855 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
856 | * already in use. |
857 | * | |
18004c5d | 858 | * Must hold slab_mutex. |
8f9f8d9e | 859 | */ |
6a67368c | 860 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
861 | { |
862 | struct kmem_cache *cachep; | |
ce8eb6c4 | 863 | struct kmem_cache_node *n; |
5e804789 | 864 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 865 | |
18004c5d | 866 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 867 | /* |
5f0985bb | 868 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
869 | * begin anything. Make sure some other cpu on this |
870 | * node has not already allocated this | |
871 | */ | |
18bf8541 CL |
872 | n = get_node(cachep, node); |
873 | if (!n) { | |
ce8eb6c4 CL |
874 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
875 | if (!n) | |
8f9f8d9e | 876 | return -ENOMEM; |
ce8eb6c4 | 877 | kmem_cache_node_init(n); |
5f0985bb JZ |
878 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
879 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
880 | |
881 | /* | |
5f0985bb JZ |
882 | * The kmem_cache_nodes don't come and go as CPUs |
883 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
884 | * protection here. |
885 | */ | |
ce8eb6c4 | 886 | cachep->node[node] = n; |
8f9f8d9e DR |
887 | } |
888 | ||
18bf8541 CL |
889 | spin_lock_irq(&n->list_lock); |
890 | n->free_limit = | |
8f9f8d9e DR |
891 | (1 + nr_cpus_node(node)) * |
892 | cachep->batchcount + cachep->num; | |
18bf8541 | 893 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
894 | } |
895 | return 0; | |
896 | } | |
897 | ||
0db0628d | 898 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
899 | { |
900 | struct kmem_cache *cachep; | |
ce8eb6c4 | 901 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 902 | int node = cpu_to_mem(cpu); |
a70f7302 | 903 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 904 | |
18004c5d | 905 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
906 | struct array_cache *nc; |
907 | struct array_cache *shared; | |
c8522a3a | 908 | struct alien_cache **alien; |
97654dfa | 909 | LIST_HEAD(list); |
fbf1e473 | 910 | |
18bf8541 | 911 | n = get_node(cachep, node); |
ce8eb6c4 | 912 | if (!n) |
bf0dea23 | 913 | continue; |
fbf1e473 | 914 | |
ce8eb6c4 | 915 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 916 | |
ce8eb6c4 CL |
917 | /* Free limit for this kmem_cache_node */ |
918 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
919 | |
920 | /* cpu is dead; no one can alloc from it. */ | |
921 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
922 | if (nc) { | |
97654dfa | 923 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
924 | nc->avail = 0; |
925 | } | |
fbf1e473 | 926 | |
58463c1f | 927 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 928 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 929 | goto free_slab; |
fbf1e473 AM |
930 | } |
931 | ||
ce8eb6c4 | 932 | shared = n->shared; |
fbf1e473 AM |
933 | if (shared) { |
934 | free_block(cachep, shared->entry, | |
97654dfa | 935 | shared->avail, node, &list); |
ce8eb6c4 | 936 | n->shared = NULL; |
fbf1e473 AM |
937 | } |
938 | ||
ce8eb6c4 CL |
939 | alien = n->alien; |
940 | n->alien = NULL; | |
fbf1e473 | 941 | |
ce8eb6c4 | 942 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
943 | |
944 | kfree(shared); | |
945 | if (alien) { | |
946 | drain_alien_cache(cachep, alien); | |
947 | free_alien_cache(alien); | |
948 | } | |
bf0dea23 JK |
949 | |
950 | free_slab: | |
97654dfa | 951 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
952 | } |
953 | /* | |
954 | * In the previous loop, all the objects were freed to | |
955 | * the respective cache's slabs, now we can go ahead and | |
956 | * shrink each nodelist to its limit. | |
957 | */ | |
18004c5d | 958 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 959 | n = get_node(cachep, node); |
ce8eb6c4 | 960 | if (!n) |
fbf1e473 | 961 | continue; |
a5aa63a5 | 962 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
963 | } |
964 | } | |
965 | ||
0db0628d | 966 | static int cpuup_prepare(long cpu) |
1da177e4 | 967 | { |
343e0d7a | 968 | struct kmem_cache *cachep; |
ce8eb6c4 | 969 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 970 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 971 | int err; |
1da177e4 | 972 | |
fbf1e473 AM |
973 | /* |
974 | * We need to do this right in the beginning since | |
975 | * alloc_arraycache's are going to use this list. | |
976 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 977 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 978 | */ |
6a67368c | 979 | err = init_cache_node_node(node); |
8f9f8d9e DR |
980 | if (err < 0) |
981 | goto bad; | |
fbf1e473 AM |
982 | |
983 | /* | |
984 | * Now we can go ahead with allocating the shared arrays and | |
985 | * array caches | |
986 | */ | |
18004c5d | 987 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 988 | struct array_cache *shared = NULL; |
c8522a3a | 989 | struct alien_cache **alien = NULL; |
fbf1e473 | 990 | |
fbf1e473 AM |
991 | if (cachep->shared) { |
992 | shared = alloc_arraycache(node, | |
993 | cachep->shared * cachep->batchcount, | |
83b519e8 | 994 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 995 | if (!shared) |
1da177e4 | 996 | goto bad; |
fbf1e473 AM |
997 | } |
998 | if (use_alien_caches) { | |
83b519e8 | 999 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1000 | if (!alien) { |
1001 | kfree(shared); | |
fbf1e473 | 1002 | goto bad; |
12d00f6a | 1003 | } |
fbf1e473 | 1004 | } |
18bf8541 | 1005 | n = get_node(cachep, node); |
ce8eb6c4 | 1006 | BUG_ON(!n); |
fbf1e473 | 1007 | |
ce8eb6c4 CL |
1008 | spin_lock_irq(&n->list_lock); |
1009 | if (!n->shared) { | |
fbf1e473 AM |
1010 | /* |
1011 | * We are serialised from CPU_DEAD or | |
1012 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1013 | */ | |
ce8eb6c4 | 1014 | n->shared = shared; |
fbf1e473 AM |
1015 | shared = NULL; |
1016 | } | |
4484ebf1 | 1017 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1018 | if (!n->alien) { |
1019 | n->alien = alien; | |
fbf1e473 | 1020 | alien = NULL; |
1da177e4 | 1021 | } |
fbf1e473 | 1022 | #endif |
ce8eb6c4 | 1023 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1024 | kfree(shared); |
1025 | free_alien_cache(alien); | |
1026 | } | |
ce79ddc8 | 1027 | |
fbf1e473 AM |
1028 | return 0; |
1029 | bad: | |
12d00f6a | 1030 | cpuup_canceled(cpu); |
fbf1e473 AM |
1031 | return -ENOMEM; |
1032 | } | |
1033 | ||
0db0628d | 1034 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1035 | unsigned long action, void *hcpu) |
1036 | { | |
1037 | long cpu = (long)hcpu; | |
1038 | int err = 0; | |
1039 | ||
1040 | switch (action) { | |
fbf1e473 AM |
1041 | case CPU_UP_PREPARE: |
1042 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1043 | mutex_lock(&slab_mutex); |
fbf1e473 | 1044 | err = cpuup_prepare(cpu); |
18004c5d | 1045 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1046 | break; |
1047 | case CPU_ONLINE: | |
8bb78442 | 1048 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1049 | start_cpu_timer(cpu); |
1050 | break; | |
1051 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1052 | case CPU_DOWN_PREPARE: |
8bb78442 | 1053 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1054 | /* |
18004c5d | 1055 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1056 | * held so that if cache_reap() is invoked it cannot do |
1057 | * anything expensive but will only modify reap_work | |
1058 | * and reschedule the timer. | |
1059 | */ | |
afe2c511 | 1060 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1061 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1062 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1063 | break; |
1064 | case CPU_DOWN_FAILED: | |
8bb78442 | 1065 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1066 | start_cpu_timer(cpu); |
1067 | break; | |
1da177e4 | 1068 | case CPU_DEAD: |
8bb78442 | 1069 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1070 | /* |
1071 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1072 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1073 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1074 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1075 | * structure is usually allocated from kmem_cache_create() and |
1076 | * gets destroyed at kmem_cache_destroy(). | |
1077 | */ | |
183ff22b | 1078 | /* fall through */ |
8f5be20b | 1079 | #endif |
1da177e4 | 1080 | case CPU_UP_CANCELED: |
8bb78442 | 1081 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1082 | mutex_lock(&slab_mutex); |
fbf1e473 | 1083 | cpuup_canceled(cpu); |
18004c5d | 1084 | mutex_unlock(&slab_mutex); |
1da177e4 | 1085 | break; |
1da177e4 | 1086 | } |
eac40680 | 1087 | return notifier_from_errno(err); |
1da177e4 LT |
1088 | } |
1089 | ||
0db0628d | 1090 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1091 | &cpuup_callback, NULL, 0 |
1092 | }; | |
1da177e4 | 1093 | |
8f9f8d9e DR |
1094 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1095 | /* | |
1096 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1097 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1098 | * removed. | |
1099 | * | |
18004c5d | 1100 | * Must hold slab_mutex. |
8f9f8d9e | 1101 | */ |
6a67368c | 1102 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1103 | { |
1104 | struct kmem_cache *cachep; | |
1105 | int ret = 0; | |
1106 | ||
18004c5d | 1107 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1108 | struct kmem_cache_node *n; |
8f9f8d9e | 1109 | |
18bf8541 | 1110 | n = get_node(cachep, node); |
ce8eb6c4 | 1111 | if (!n) |
8f9f8d9e DR |
1112 | continue; |
1113 | ||
a5aa63a5 | 1114 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1115 | |
ce8eb6c4 CL |
1116 | if (!list_empty(&n->slabs_full) || |
1117 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1118 | ret = -EBUSY; |
1119 | break; | |
1120 | } | |
1121 | } | |
1122 | return ret; | |
1123 | } | |
1124 | ||
1125 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1126 | unsigned long action, void *arg) | |
1127 | { | |
1128 | struct memory_notify *mnb = arg; | |
1129 | int ret = 0; | |
1130 | int nid; | |
1131 | ||
1132 | nid = mnb->status_change_nid; | |
1133 | if (nid < 0) | |
1134 | goto out; | |
1135 | ||
1136 | switch (action) { | |
1137 | case MEM_GOING_ONLINE: | |
18004c5d | 1138 | mutex_lock(&slab_mutex); |
6a67368c | 1139 | ret = init_cache_node_node(nid); |
18004c5d | 1140 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1141 | break; |
1142 | case MEM_GOING_OFFLINE: | |
18004c5d | 1143 | mutex_lock(&slab_mutex); |
6a67368c | 1144 | ret = drain_cache_node_node(nid); |
18004c5d | 1145 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1146 | break; |
1147 | case MEM_ONLINE: | |
1148 | case MEM_OFFLINE: | |
1149 | case MEM_CANCEL_ONLINE: | |
1150 | case MEM_CANCEL_OFFLINE: | |
1151 | break; | |
1152 | } | |
1153 | out: | |
5fda1bd5 | 1154 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1155 | } |
1156 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1157 | ||
e498be7d | 1158 | /* |
ce8eb6c4 | 1159 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1160 | */ |
6744f087 | 1161 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1162 | int nodeid) |
e498be7d | 1163 | { |
6744f087 | 1164 | struct kmem_cache_node *ptr; |
e498be7d | 1165 | |
6744f087 | 1166 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1167 | BUG_ON(!ptr); |
1168 | ||
6744f087 | 1169 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1170 | /* |
1171 | * Do not assume that spinlocks can be initialized via memcpy: | |
1172 | */ | |
1173 | spin_lock_init(&ptr->list_lock); | |
1174 | ||
e498be7d | 1175 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1176 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1177 | } |
1178 | ||
556a169d | 1179 | /* |
ce8eb6c4 CL |
1180 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1181 | * size of kmem_cache_node. | |
556a169d | 1182 | */ |
ce8eb6c4 | 1183 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1184 | { |
1185 | int node; | |
1186 | ||
1187 | for_each_online_node(node) { | |
ce8eb6c4 | 1188 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1189 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1190 | REAPTIMEOUT_NODE + |
1191 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1192 | } |
1193 | } | |
1194 | ||
a737b3e2 AM |
1195 | /* |
1196 | * Initialisation. Called after the page allocator have been initialised and | |
1197 | * before smp_init(). | |
1da177e4 LT |
1198 | */ |
1199 | void __init kmem_cache_init(void) | |
1200 | { | |
e498be7d CL |
1201 | int i; |
1202 | ||
68126702 JK |
1203 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1204 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1205 | kmem_cache = &kmem_cache_boot; |
1206 | ||
8888177e | 1207 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1208 | use_alien_caches = 0; |
1209 | ||
3c583465 | 1210 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1211 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1212 | |
1da177e4 LT |
1213 | /* |
1214 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1215 | * page orders on machines with more than 32MB of memory if |
1216 | * not overridden on the command line. | |
1da177e4 | 1217 | */ |
3df1cccd | 1218 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1219 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1220 | |
1da177e4 LT |
1221 | /* Bootstrap is tricky, because several objects are allocated |
1222 | * from caches that do not exist yet: | |
9b030cb8 CL |
1223 | * 1) initialize the kmem_cache cache: it contains the struct |
1224 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1225 | * kmem_cache is statically allocated. | |
e498be7d | 1226 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1227 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1228 | * array at the end of the bootstrap. |
1da177e4 | 1229 | * 2) Create the first kmalloc cache. |
343e0d7a | 1230 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1231 | * An __init data area is used for the head array. |
1232 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1233 | * head arrays. | |
9b030cb8 | 1234 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1235 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1236 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1237 | * the other cache's with kmalloc allocated memory. |
1238 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1239 | */ |
1240 | ||
9b030cb8 | 1241 | /* 1) create the kmem_cache */ |
1da177e4 | 1242 | |
8da3430d | 1243 | /* |
b56efcf0 | 1244 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1245 | */ |
2f9baa9f | 1246 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1247 | offsetof(struct kmem_cache, node) + |
6744f087 | 1248 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1249 | SLAB_HWCACHE_ALIGN); |
1250 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1251 | slab_state = PARTIAL; |
1da177e4 | 1252 | |
a737b3e2 | 1253 | /* |
bf0dea23 JK |
1254 | * Initialize the caches that provide memory for the kmem_cache_node |
1255 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1256 | */ |
bf0dea23 | 1257 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1258 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1259 | slab_state = PARTIAL_NODE; |
34cc6990 | 1260 | setup_kmalloc_cache_index_table(); |
e498be7d | 1261 | |
e0a42726 IM |
1262 | slab_early_init = 0; |
1263 | ||
ce8eb6c4 | 1264 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1265 | { |
1ca4cb24 PE |
1266 | int nid; |
1267 | ||
9c09a95c | 1268 | for_each_online_node(nid) { |
ce8eb6c4 | 1269 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1270 | |
bf0dea23 | 1271 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1272 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1273 | } |
1274 | } | |
1da177e4 | 1275 | |
f97d5f63 | 1276 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1277 | } |
1278 | ||
1279 | void __init kmem_cache_init_late(void) | |
1280 | { | |
1281 | struct kmem_cache *cachep; | |
1282 | ||
97d06609 | 1283 | slab_state = UP; |
52cef189 | 1284 | |
8429db5c | 1285 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1286 | mutex_lock(&slab_mutex); |
1287 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1288 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1289 | BUG(); | |
18004c5d | 1290 | mutex_unlock(&slab_mutex); |
056c6241 | 1291 | |
97d06609 CL |
1292 | /* Done! */ |
1293 | slab_state = FULL; | |
1294 | ||
a737b3e2 AM |
1295 | /* |
1296 | * Register a cpu startup notifier callback that initializes | |
1297 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1298 | */ |
1299 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1300 | |
8f9f8d9e DR |
1301 | #ifdef CONFIG_NUMA |
1302 | /* | |
1303 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1304 | * node. |
8f9f8d9e DR |
1305 | */ |
1306 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1307 | #endif | |
1308 | ||
a737b3e2 AM |
1309 | /* |
1310 | * The reap timers are started later, with a module init call: That part | |
1311 | * of the kernel is not yet operational. | |
1da177e4 LT |
1312 | */ |
1313 | } | |
1314 | ||
1315 | static int __init cpucache_init(void) | |
1316 | { | |
1317 | int cpu; | |
1318 | ||
a737b3e2 AM |
1319 | /* |
1320 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1321 | */ |
e498be7d | 1322 | for_each_online_cpu(cpu) |
a737b3e2 | 1323 | start_cpu_timer(cpu); |
a164f896 GC |
1324 | |
1325 | /* Done! */ | |
97d06609 | 1326 | slab_state = FULL; |
1da177e4 LT |
1327 | return 0; |
1328 | } | |
1da177e4 LT |
1329 | __initcall(cpucache_init); |
1330 | ||
8bdec192 RA |
1331 | static noinline void |
1332 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1333 | { | |
9a02d699 | 1334 | #if DEBUG |
ce8eb6c4 | 1335 | struct kmem_cache_node *n; |
8456a648 | 1336 | struct page *page; |
8bdec192 RA |
1337 | unsigned long flags; |
1338 | int node; | |
9a02d699 DR |
1339 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1340 | DEFAULT_RATELIMIT_BURST); | |
1341 | ||
1342 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1343 | return; | |
8bdec192 | 1344 | |
5b3810e5 VB |
1345 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1346 | nodeid, gfpflags, &gfpflags); | |
1347 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1348 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1349 | |
18bf8541 | 1350 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1351 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1352 | unsigned long active_slabs = 0, num_slabs = 0; | |
1353 | ||
ce8eb6c4 | 1354 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1355 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1356 | active_objs += cachep->num; |
1357 | active_slabs++; | |
1358 | } | |
8456a648 JK |
1359 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1360 | active_objs += page->active; | |
8bdec192 RA |
1361 | active_slabs++; |
1362 | } | |
8456a648 | 1363 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1364 | num_slabs++; |
1365 | ||
ce8eb6c4 CL |
1366 | free_objects += n->free_objects; |
1367 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1368 | |
1369 | num_slabs += active_slabs; | |
1370 | num_objs = num_slabs * cachep->num; | |
5b3810e5 | 1371 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", |
8bdec192 RA |
1372 | node, active_slabs, num_slabs, active_objs, num_objs, |
1373 | free_objects); | |
1374 | } | |
9a02d699 | 1375 | #endif |
8bdec192 RA |
1376 | } |
1377 | ||
1da177e4 | 1378 | /* |
8a7d9b43 WSH |
1379 | * Interface to system's page allocator. No need to hold the |
1380 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1381 | * |
1382 | * If we requested dmaable memory, we will get it. Even if we | |
1383 | * did not request dmaable memory, we might get it, but that | |
1384 | * would be relatively rare and ignorable. | |
1385 | */ | |
0c3aa83e JK |
1386 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1387 | int nodeid) | |
1da177e4 LT |
1388 | { |
1389 | struct page *page; | |
e1b6aa6f | 1390 | int nr_pages; |
765c4507 | 1391 | |
a618e89f | 1392 | flags |= cachep->allocflags; |
e12ba74d MG |
1393 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1394 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1395 | |
96db800f | 1396 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1397 | if (!page) { |
9a02d699 | 1398 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1399 | return NULL; |
8bdec192 | 1400 | } |
1da177e4 | 1401 | |
f3ccb2c4 VD |
1402 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1403 | __free_pages(page, cachep->gfporder); | |
1404 | return NULL; | |
1405 | } | |
1406 | ||
e1b6aa6f | 1407 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1408 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1409 | add_zone_page_state(page_zone(page), |
1410 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1411 | else | |
1412 | add_zone_page_state(page_zone(page), | |
1413 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
f68f8ddd | 1414 | |
a57a4988 | 1415 | __SetPageSlab(page); |
f68f8ddd JK |
1416 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1417 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1418 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1419 | |
b1eeab67 VN |
1420 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1421 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1422 | ||
1423 | if (cachep->ctor) | |
1424 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1425 | else | |
1426 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1427 | } | |
c175eea4 | 1428 | |
0c3aa83e | 1429 | return page; |
1da177e4 LT |
1430 | } |
1431 | ||
1432 | /* | |
1433 | * Interface to system's page release. | |
1434 | */ | |
0c3aa83e | 1435 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1436 | { |
27ee57c9 VD |
1437 | int order = cachep->gfporder; |
1438 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1439 | |
27ee57c9 | 1440 | kmemcheck_free_shadow(page, order); |
c175eea4 | 1441 | |
972d1a7b CL |
1442 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1443 | sub_zone_page_state(page_zone(page), | |
1444 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1445 | else | |
1446 | sub_zone_page_state(page_zone(page), | |
1447 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1448 | |
a57a4988 | 1449 | BUG_ON(!PageSlab(page)); |
73293c2f | 1450 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1451 | __ClearPageSlab(page); |
8456a648 JK |
1452 | page_mapcount_reset(page); |
1453 | page->mapping = NULL; | |
1f458cbf | 1454 | |
1da177e4 LT |
1455 | if (current->reclaim_state) |
1456 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1457 | memcg_uncharge_slab(page, order, cachep); |
1458 | __free_pages(page, order); | |
1da177e4 LT |
1459 | } |
1460 | ||
1461 | static void kmem_rcu_free(struct rcu_head *head) | |
1462 | { | |
68126702 JK |
1463 | struct kmem_cache *cachep; |
1464 | struct page *page; | |
1da177e4 | 1465 | |
68126702 JK |
1466 | page = container_of(head, struct page, rcu_head); |
1467 | cachep = page->slab_cache; | |
1468 | ||
1469 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1470 | } |
1471 | ||
1472 | #if DEBUG | |
40b44137 JK |
1473 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1474 | { | |
1475 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1476 | (cachep->size % PAGE_SIZE) == 0) | |
1477 | return true; | |
1478 | ||
1479 | return false; | |
1480 | } | |
1da177e4 LT |
1481 | |
1482 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1483 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1484 | unsigned long caller) |
1da177e4 | 1485 | { |
8c138bc0 | 1486 | int size = cachep->object_size; |
1da177e4 | 1487 | |
3dafccf2 | 1488 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1489 | |
b28a02de | 1490 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1491 | return; |
1492 | ||
b28a02de PE |
1493 | *addr++ = 0x12345678; |
1494 | *addr++ = caller; | |
1495 | *addr++ = smp_processor_id(); | |
1496 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1497 | { |
1498 | unsigned long *sptr = &caller; | |
1499 | unsigned long svalue; | |
1500 | ||
1501 | while (!kstack_end(sptr)) { | |
1502 | svalue = *sptr++; | |
1503 | if (kernel_text_address(svalue)) { | |
b28a02de | 1504 | *addr++ = svalue; |
1da177e4 LT |
1505 | size -= sizeof(unsigned long); |
1506 | if (size <= sizeof(unsigned long)) | |
1507 | break; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | } | |
b28a02de | 1512 | *addr++ = 0x87654321; |
1da177e4 | 1513 | } |
40b44137 JK |
1514 | |
1515 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1516 | int map, unsigned long caller) | |
1517 | { | |
1518 | if (!is_debug_pagealloc_cache(cachep)) | |
1519 | return; | |
1520 | ||
1521 | if (caller) | |
1522 | store_stackinfo(cachep, objp, caller); | |
1523 | ||
1524 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1525 | } | |
1526 | ||
1527 | #else | |
1528 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1529 | int map, unsigned long caller) {} | |
1530 | ||
1da177e4 LT |
1531 | #endif |
1532 | ||
343e0d7a | 1533 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1534 | { |
8c138bc0 | 1535 | int size = cachep->object_size; |
3dafccf2 | 1536 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1537 | |
1538 | memset(addr, val, size); | |
b28a02de | 1539 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1540 | } |
1541 | ||
1542 | static void dump_line(char *data, int offset, int limit) | |
1543 | { | |
1544 | int i; | |
aa83aa40 DJ |
1545 | unsigned char error = 0; |
1546 | int bad_count = 0; | |
1547 | ||
1170532b | 1548 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1549 | for (i = 0; i < limit; i++) { |
1550 | if (data[offset + i] != POISON_FREE) { | |
1551 | error = data[offset + i]; | |
1552 | bad_count++; | |
1553 | } | |
aa83aa40 | 1554 | } |
fdde6abb SAS |
1555 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1556 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1557 | |
1558 | if (bad_count == 1) { | |
1559 | error ^= POISON_FREE; | |
1560 | if (!(error & (error - 1))) { | |
1170532b | 1561 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1562 | #ifdef CONFIG_X86 |
1170532b | 1563 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1564 | #else |
1170532b | 1565 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1566 | #endif |
1567 | } | |
1568 | } | |
1da177e4 LT |
1569 | } |
1570 | #endif | |
1571 | ||
1572 | #if DEBUG | |
1573 | ||
343e0d7a | 1574 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1575 | { |
1576 | int i, size; | |
1577 | char *realobj; | |
1578 | ||
1579 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1580 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1581 | *dbg_redzone1(cachep, objp), | |
1582 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1583 | } |
1584 | ||
1585 | if (cachep->flags & SLAB_STORE_USER) { | |
1170532b | 1586 | pr_err("Last user: [<%p>](%pSR)\n", |
071361d3 JP |
1587 | *dbg_userword(cachep, objp), |
1588 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1589 | } |
3dafccf2 | 1590 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1591 | size = cachep->object_size; |
b28a02de | 1592 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1593 | int limit; |
1594 | limit = 16; | |
b28a02de PE |
1595 | if (i + limit > size) |
1596 | limit = size - i; | |
1da177e4 LT |
1597 | dump_line(realobj, i, limit); |
1598 | } | |
1599 | } | |
1600 | ||
343e0d7a | 1601 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1602 | { |
1603 | char *realobj; | |
1604 | int size, i; | |
1605 | int lines = 0; | |
1606 | ||
40b44137 JK |
1607 | if (is_debug_pagealloc_cache(cachep)) |
1608 | return; | |
1609 | ||
3dafccf2 | 1610 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1611 | size = cachep->object_size; |
1da177e4 | 1612 | |
b28a02de | 1613 | for (i = 0; i < size; i++) { |
1da177e4 | 1614 | char exp = POISON_FREE; |
b28a02de | 1615 | if (i == size - 1) |
1da177e4 LT |
1616 | exp = POISON_END; |
1617 | if (realobj[i] != exp) { | |
1618 | int limit; | |
1619 | /* Mismatch ! */ | |
1620 | /* Print header */ | |
1621 | if (lines == 0) { | |
1170532b JP |
1622 | pr_err("Slab corruption (%s): %s start=%p, len=%d\n", |
1623 | print_tainted(), cachep->name, | |
1624 | realobj, size); | |
1da177e4 LT |
1625 | print_objinfo(cachep, objp, 0); |
1626 | } | |
1627 | /* Hexdump the affected line */ | |
b28a02de | 1628 | i = (i / 16) * 16; |
1da177e4 | 1629 | limit = 16; |
b28a02de PE |
1630 | if (i + limit > size) |
1631 | limit = size - i; | |
1da177e4 LT |
1632 | dump_line(realobj, i, limit); |
1633 | i += 16; | |
1634 | lines++; | |
1635 | /* Limit to 5 lines */ | |
1636 | if (lines > 5) | |
1637 | break; | |
1638 | } | |
1639 | } | |
1640 | if (lines != 0) { | |
1641 | /* Print some data about the neighboring objects, if they | |
1642 | * exist: | |
1643 | */ | |
8456a648 | 1644 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1645 | unsigned int objnr; |
1da177e4 | 1646 | |
8456a648 | 1647 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1648 | if (objnr) { |
8456a648 | 1649 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1650 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1651 | pr_err("Prev obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1652 | print_objinfo(cachep, objp, 2); |
1653 | } | |
b28a02de | 1654 | if (objnr + 1 < cachep->num) { |
8456a648 | 1655 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1656 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1657 | pr_err("Next obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1658 | print_objinfo(cachep, objp, 2); |
1659 | } | |
1660 | } | |
1661 | } | |
1662 | #endif | |
1663 | ||
12dd36fa | 1664 | #if DEBUG |
8456a648 JK |
1665 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1666 | struct page *page) | |
1da177e4 | 1667 | { |
1da177e4 | 1668 | int i; |
b03a017b JK |
1669 | |
1670 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1671 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1672 | POISON_FREE); | |
1673 | } | |
1674 | ||
1da177e4 | 1675 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1676 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1677 | |
1678 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1679 | check_poison_obj(cachep, objp); |
40b44137 | 1680 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1681 | } |
1682 | if (cachep->flags & SLAB_RED_ZONE) { | |
1683 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1684 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1685 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1686 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1687 | } |
1da177e4 | 1688 | } |
12dd36fa | 1689 | } |
1da177e4 | 1690 | #else |
8456a648 JK |
1691 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1692 | struct page *page) | |
12dd36fa | 1693 | { |
12dd36fa | 1694 | } |
1da177e4 LT |
1695 | #endif |
1696 | ||
911851e6 RD |
1697 | /** |
1698 | * slab_destroy - destroy and release all objects in a slab | |
1699 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1700 | * @page: page pointer being destroyed |
911851e6 | 1701 | * |
8a7d9b43 WSH |
1702 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1703 | * Before calling the slab page must have been unlinked from the cache. The | |
1704 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1705 | */ |
8456a648 | 1706 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1707 | { |
7e007355 | 1708 | void *freelist; |
12dd36fa | 1709 | |
8456a648 JK |
1710 | freelist = page->freelist; |
1711 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1712 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1713 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1714 | else | |
0c3aa83e | 1715 | kmem_freepages(cachep, page); |
68126702 JK |
1716 | |
1717 | /* | |
8456a648 | 1718 | * From now on, we don't use freelist |
68126702 JK |
1719 | * although actual page can be freed in rcu context |
1720 | */ | |
1721 | if (OFF_SLAB(cachep)) | |
8456a648 | 1722 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1723 | } |
1724 | ||
97654dfa JK |
1725 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1726 | { | |
1727 | struct page *page, *n; | |
1728 | ||
1729 | list_for_each_entry_safe(page, n, list, lru) { | |
1730 | list_del(&page->lru); | |
1731 | slab_destroy(cachep, page); | |
1732 | } | |
1733 | } | |
1734 | ||
4d268eba | 1735 | /** |
a70773dd RD |
1736 | * calculate_slab_order - calculate size (page order) of slabs |
1737 | * @cachep: pointer to the cache that is being created | |
1738 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1739 | * @flags: slab allocation flags |
1740 | * | |
1741 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1742 | * |
1743 | * This could be made much more intelligent. For now, try to avoid using | |
1744 | * high order pages for slabs. When the gfp() functions are more friendly | |
1745 | * towards high-order requests, this should be changed. | |
1746 | */ | |
a737b3e2 | 1747 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1748 | size_t size, unsigned long flags) |
4d268eba PE |
1749 | { |
1750 | size_t left_over = 0; | |
9888e6fa | 1751 | int gfporder; |
4d268eba | 1752 | |
0aa817f0 | 1753 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1754 | unsigned int num; |
1755 | size_t remainder; | |
1756 | ||
70f75067 | 1757 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1758 | if (!num) |
1759 | continue; | |
9888e6fa | 1760 | |
f315e3fa JK |
1761 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1762 | if (num > SLAB_OBJ_MAX_NUM) | |
1763 | break; | |
1764 | ||
b1ab41c4 | 1765 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1766 | struct kmem_cache *freelist_cache; |
1767 | size_t freelist_size; | |
1768 | ||
1769 | freelist_size = num * sizeof(freelist_idx_t); | |
1770 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1771 | if (!freelist_cache) | |
1772 | continue; | |
1773 | ||
b1ab41c4 | 1774 | /* |
3217fd9b JK |
1775 | * Needed to avoid possible looping condition |
1776 | * in cache_grow() | |
b1ab41c4 | 1777 | */ |
3217fd9b JK |
1778 | if (OFF_SLAB(freelist_cache)) |
1779 | continue; | |
b1ab41c4 | 1780 | |
3217fd9b JK |
1781 | /* check if off slab has enough benefit */ |
1782 | if (freelist_cache->size > cachep->size / 2) | |
1783 | continue; | |
b1ab41c4 | 1784 | } |
4d268eba | 1785 | |
9888e6fa | 1786 | /* Found something acceptable - save it away */ |
4d268eba | 1787 | cachep->num = num; |
9888e6fa | 1788 | cachep->gfporder = gfporder; |
4d268eba PE |
1789 | left_over = remainder; |
1790 | ||
f78bb8ad LT |
1791 | /* |
1792 | * A VFS-reclaimable slab tends to have most allocations | |
1793 | * as GFP_NOFS and we really don't want to have to be allocating | |
1794 | * higher-order pages when we are unable to shrink dcache. | |
1795 | */ | |
1796 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1797 | break; | |
1798 | ||
4d268eba PE |
1799 | /* |
1800 | * Large number of objects is good, but very large slabs are | |
1801 | * currently bad for the gfp()s. | |
1802 | */ | |
543585cc | 1803 | if (gfporder >= slab_max_order) |
4d268eba PE |
1804 | break; |
1805 | ||
9888e6fa LT |
1806 | /* |
1807 | * Acceptable internal fragmentation? | |
1808 | */ | |
a737b3e2 | 1809 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1810 | break; |
1811 | } | |
1812 | return left_over; | |
1813 | } | |
1814 | ||
bf0dea23 JK |
1815 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1816 | struct kmem_cache *cachep, int entries, int batchcount) | |
1817 | { | |
1818 | int cpu; | |
1819 | size_t size; | |
1820 | struct array_cache __percpu *cpu_cache; | |
1821 | ||
1822 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1823 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1824 | |
1825 | if (!cpu_cache) | |
1826 | return NULL; | |
1827 | ||
1828 | for_each_possible_cpu(cpu) { | |
1829 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1830 | entries, batchcount); | |
1831 | } | |
1832 | ||
1833 | return cpu_cache; | |
1834 | } | |
1835 | ||
83b519e8 | 1836 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1837 | { |
97d06609 | 1838 | if (slab_state >= FULL) |
83b519e8 | 1839 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1840 | |
bf0dea23 JK |
1841 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1842 | if (!cachep->cpu_cache) | |
1843 | return 1; | |
1844 | ||
97d06609 | 1845 | if (slab_state == DOWN) { |
bf0dea23 JK |
1846 | /* Creation of first cache (kmem_cache). */ |
1847 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1848 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1849 | /* For kmem_cache_node */ |
1850 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1851 | } else { |
bf0dea23 | 1852 | int node; |
f30cf7d1 | 1853 | |
bf0dea23 JK |
1854 | for_each_online_node(node) { |
1855 | cachep->node[node] = kmalloc_node( | |
1856 | sizeof(struct kmem_cache_node), gfp, node); | |
1857 | BUG_ON(!cachep->node[node]); | |
1858 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1859 | } |
1860 | } | |
bf0dea23 | 1861 | |
6a67368c | 1862 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1863 | jiffies + REAPTIMEOUT_NODE + |
1864 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1865 | |
1866 | cpu_cache_get(cachep)->avail = 0; | |
1867 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1868 | cpu_cache_get(cachep)->batchcount = 1; | |
1869 | cpu_cache_get(cachep)->touched = 0; | |
1870 | cachep->batchcount = 1; | |
1871 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1872 | return 0; |
f30cf7d1 PE |
1873 | } |
1874 | ||
12220dea JK |
1875 | unsigned long kmem_cache_flags(unsigned long object_size, |
1876 | unsigned long flags, const char *name, | |
1877 | void (*ctor)(void *)) | |
1878 | { | |
1879 | return flags; | |
1880 | } | |
1881 | ||
1882 | struct kmem_cache * | |
1883 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
1884 | unsigned long flags, void (*ctor)(void *)) | |
1885 | { | |
1886 | struct kmem_cache *cachep; | |
1887 | ||
1888 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1889 | if (cachep) { | |
1890 | cachep->refcount++; | |
1891 | ||
1892 | /* | |
1893 | * Adjust the object sizes so that we clear | |
1894 | * the complete object on kzalloc. | |
1895 | */ | |
1896 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1897 | } | |
1898 | return cachep; | |
1899 | } | |
1900 | ||
b03a017b JK |
1901 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
1902 | size_t size, unsigned long flags) | |
1903 | { | |
1904 | size_t left; | |
1905 | ||
1906 | cachep->num = 0; | |
1907 | ||
1908 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
1909 | return false; | |
1910 | ||
1911 | left = calculate_slab_order(cachep, size, | |
1912 | flags | CFLGS_OBJFREELIST_SLAB); | |
1913 | if (!cachep->num) | |
1914 | return false; | |
1915 | ||
1916 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1917 | return false; | |
1918 | ||
1919 | cachep->colour = left / cachep->colour_off; | |
1920 | ||
1921 | return true; | |
1922 | } | |
1923 | ||
158e319b JK |
1924 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
1925 | size_t size, unsigned long flags) | |
1926 | { | |
1927 | size_t left; | |
1928 | ||
1929 | cachep->num = 0; | |
1930 | ||
1931 | /* | |
3217fd9b JK |
1932 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1933 | * to avoid recursive calls into kmemleak. | |
158e319b | 1934 | */ |
158e319b JK |
1935 | if (flags & SLAB_NOLEAKTRACE) |
1936 | return false; | |
1937 | ||
1938 | /* | |
1939 | * Size is large, assume best to place the slab management obj | |
1940 | * off-slab (should allow better packing of objs). | |
1941 | */ | |
1942 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1943 | if (!cachep->num) | |
1944 | return false; | |
1945 | ||
1946 | /* | |
1947 | * If the slab has been placed off-slab, and we have enough space then | |
1948 | * move it on-slab. This is at the expense of any extra colouring. | |
1949 | */ | |
1950 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1951 | return false; | |
1952 | ||
1953 | cachep->colour = left / cachep->colour_off; | |
1954 | ||
1955 | return true; | |
1956 | } | |
1957 | ||
1958 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
1959 | size_t size, unsigned long flags) | |
1960 | { | |
1961 | size_t left; | |
1962 | ||
1963 | cachep->num = 0; | |
1964 | ||
1965 | left = calculate_slab_order(cachep, size, flags); | |
1966 | if (!cachep->num) | |
1967 | return false; | |
1968 | ||
1969 | cachep->colour = left / cachep->colour_off; | |
1970 | ||
1971 | return true; | |
1972 | } | |
1973 | ||
1da177e4 | 1974 | /** |
039363f3 | 1975 | * __kmem_cache_create - Create a cache. |
a755b76a | 1976 | * @cachep: cache management descriptor |
1da177e4 | 1977 | * @flags: SLAB flags |
1da177e4 LT |
1978 | * |
1979 | * Returns a ptr to the cache on success, NULL on failure. | |
1980 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1981 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1982 | * |
1da177e4 LT |
1983 | * The flags are |
1984 | * | |
1985 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1986 | * to catch references to uninitialised memory. | |
1987 | * | |
1988 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1989 | * for buffer overruns. | |
1990 | * | |
1da177e4 LT |
1991 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1992 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1993 | * as davem. | |
1994 | */ | |
278b1bb1 | 1995 | int |
8a13a4cc | 1996 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 1997 | { |
d4a5fca5 | 1998 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 1999 | gfp_t gfp; |
278b1bb1 | 2000 | int err; |
8a13a4cc | 2001 | size_t size = cachep->size; |
1da177e4 | 2002 | |
1da177e4 | 2003 | #if DEBUG |
1da177e4 LT |
2004 | #if FORCED_DEBUG |
2005 | /* | |
2006 | * Enable redzoning and last user accounting, except for caches with | |
2007 | * large objects, if the increased size would increase the object size | |
2008 | * above the next power of two: caches with object sizes just above a | |
2009 | * power of two have a significant amount of internal fragmentation. | |
2010 | */ | |
87a927c7 DW |
2011 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2012 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2013 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2014 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2015 | flags |= SLAB_POISON; | |
2016 | #endif | |
1da177e4 | 2017 | #endif |
1da177e4 | 2018 | |
a737b3e2 AM |
2019 | /* |
2020 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2021 | * unaligned accesses for some archs when redzoning is used, and makes |
2022 | * sure any on-slab bufctl's are also correctly aligned. | |
2023 | */ | |
b28a02de PE |
2024 | if (size & (BYTES_PER_WORD - 1)) { |
2025 | size += (BYTES_PER_WORD - 1); | |
2026 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2027 | } |
2028 | ||
87a927c7 DW |
2029 | if (flags & SLAB_RED_ZONE) { |
2030 | ralign = REDZONE_ALIGN; | |
2031 | /* If redzoning, ensure that the second redzone is suitably | |
2032 | * aligned, by adjusting the object size accordingly. */ | |
2033 | size += REDZONE_ALIGN - 1; | |
2034 | size &= ~(REDZONE_ALIGN - 1); | |
2035 | } | |
ca5f9703 | 2036 | |
a44b56d3 | 2037 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2038 | if (ralign < cachep->align) { |
2039 | ralign = cachep->align; | |
1da177e4 | 2040 | } |
3ff84a7f PE |
2041 | /* disable debug if necessary */ |
2042 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2043 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2044 | /* |
ca5f9703 | 2045 | * 4) Store it. |
1da177e4 | 2046 | */ |
8a13a4cc | 2047 | cachep->align = ralign; |
158e319b JK |
2048 | cachep->colour_off = cache_line_size(); |
2049 | /* Offset must be a multiple of the alignment. */ | |
2050 | if (cachep->colour_off < cachep->align) | |
2051 | cachep->colour_off = cachep->align; | |
1da177e4 | 2052 | |
83b519e8 PE |
2053 | if (slab_is_available()) |
2054 | gfp = GFP_KERNEL; | |
2055 | else | |
2056 | gfp = GFP_NOWAIT; | |
2057 | ||
1da177e4 | 2058 | #if DEBUG |
1da177e4 | 2059 | |
ca5f9703 PE |
2060 | /* |
2061 | * Both debugging options require word-alignment which is calculated | |
2062 | * into align above. | |
2063 | */ | |
1da177e4 | 2064 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2065 | /* add space for red zone words */ |
3ff84a7f PE |
2066 | cachep->obj_offset += sizeof(unsigned long long); |
2067 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2068 | } |
2069 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2070 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2071 | * the real object. But if the second red zone needs to be |
2072 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2073 | */ |
87a927c7 DW |
2074 | if (flags & SLAB_RED_ZONE) |
2075 | size += REDZONE_ALIGN; | |
2076 | else | |
2077 | size += BYTES_PER_WORD; | |
1da177e4 | 2078 | } |
832a15d2 JK |
2079 | #endif |
2080 | ||
7ed2f9e6 AP |
2081 | kasan_cache_create(cachep, &size, &flags); |
2082 | ||
832a15d2 JK |
2083 | size = ALIGN(size, cachep->align); |
2084 | /* | |
2085 | * We should restrict the number of objects in a slab to implement | |
2086 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2087 | */ | |
2088 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2089 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2090 | ||
2091 | #if DEBUG | |
03a2d2a3 JK |
2092 | /* |
2093 | * To activate debug pagealloc, off-slab management is necessary | |
2094 | * requirement. In early phase of initialization, small sized slab | |
2095 | * doesn't get initialized so it would not be possible. So, we need | |
2096 | * to check size >= 256. It guarantees that all necessary small | |
2097 | * sized slab is initialized in current slab initialization sequence. | |
2098 | */ | |
40323278 | 2099 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2100 | size >= 256 && cachep->object_size > cache_line_size()) { |
2101 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2102 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2103 | ||
2104 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2105 | flags |= CFLGS_OFF_SLAB; | |
2106 | cachep->obj_offset += tmp_size - size; | |
2107 | size = tmp_size; | |
2108 | goto done; | |
2109 | } | |
2110 | } | |
1da177e4 | 2111 | } |
1da177e4 LT |
2112 | #endif |
2113 | ||
b03a017b JK |
2114 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2115 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2116 | goto done; | |
2117 | } | |
2118 | ||
158e319b | 2119 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2120 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2121 | goto done; |
832a15d2 | 2122 | } |
1da177e4 | 2123 | |
158e319b JK |
2124 | if (set_on_slab_cache(cachep, size, flags)) |
2125 | goto done; | |
1da177e4 | 2126 | |
158e319b | 2127 | return -E2BIG; |
1da177e4 | 2128 | |
158e319b JK |
2129 | done: |
2130 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2131 | cachep->flags = flags; |
a57a4988 | 2132 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2133 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2134 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2135 | cachep->size = size; |
6a2d7a95 | 2136 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2137 | |
40b44137 JK |
2138 | #if DEBUG |
2139 | /* | |
2140 | * If we're going to use the generic kernel_map_pages() | |
2141 | * poisoning, then it's going to smash the contents of | |
2142 | * the redzone and userword anyhow, so switch them off. | |
2143 | */ | |
2144 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2145 | (cachep->flags & SLAB_POISON) && | |
2146 | is_debug_pagealloc_cache(cachep)) | |
2147 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2148 | #endif | |
2149 | ||
2150 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2151 | cachep->freelist_cache = |
2152 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2153 | } |
1da177e4 | 2154 | |
278b1bb1 CL |
2155 | err = setup_cpu_cache(cachep, gfp); |
2156 | if (err) { | |
52b4b950 | 2157 | __kmem_cache_release(cachep); |
278b1bb1 | 2158 | return err; |
2ed3a4ef | 2159 | } |
1da177e4 | 2160 | |
278b1bb1 | 2161 | return 0; |
1da177e4 | 2162 | } |
1da177e4 LT |
2163 | |
2164 | #if DEBUG | |
2165 | static void check_irq_off(void) | |
2166 | { | |
2167 | BUG_ON(!irqs_disabled()); | |
2168 | } | |
2169 | ||
2170 | static void check_irq_on(void) | |
2171 | { | |
2172 | BUG_ON(irqs_disabled()); | |
2173 | } | |
2174 | ||
18726ca8 JK |
2175 | static void check_mutex_acquired(void) |
2176 | { | |
2177 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2178 | } | |
2179 | ||
343e0d7a | 2180 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2181 | { |
2182 | #ifdef CONFIG_SMP | |
2183 | check_irq_off(); | |
18bf8541 | 2184 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2185 | #endif |
2186 | } | |
e498be7d | 2187 | |
343e0d7a | 2188 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2189 | { |
2190 | #ifdef CONFIG_SMP | |
2191 | check_irq_off(); | |
18bf8541 | 2192 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2193 | #endif |
2194 | } | |
2195 | ||
1da177e4 LT |
2196 | #else |
2197 | #define check_irq_off() do { } while(0) | |
2198 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2199 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2200 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2201 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2202 | #endif |
2203 | ||
18726ca8 JK |
2204 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2205 | int node, bool free_all, struct list_head *list) | |
2206 | { | |
2207 | int tofree; | |
2208 | ||
2209 | if (!ac || !ac->avail) | |
2210 | return; | |
2211 | ||
2212 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2213 | if (tofree > ac->avail) | |
2214 | tofree = (ac->avail + 1) / 2; | |
2215 | ||
2216 | free_block(cachep, ac->entry, tofree, node, list); | |
2217 | ac->avail -= tofree; | |
2218 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2219 | } | |
aab2207c | 2220 | |
1da177e4 LT |
2221 | static void do_drain(void *arg) |
2222 | { | |
a737b3e2 | 2223 | struct kmem_cache *cachep = arg; |
1da177e4 | 2224 | struct array_cache *ac; |
7d6e6d09 | 2225 | int node = numa_mem_id(); |
18bf8541 | 2226 | struct kmem_cache_node *n; |
97654dfa | 2227 | LIST_HEAD(list); |
1da177e4 LT |
2228 | |
2229 | check_irq_off(); | |
9a2dba4b | 2230 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2231 | n = get_node(cachep, node); |
2232 | spin_lock(&n->list_lock); | |
97654dfa | 2233 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2234 | spin_unlock(&n->list_lock); |
97654dfa | 2235 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2236 | ac->avail = 0; |
2237 | } | |
2238 | ||
343e0d7a | 2239 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2240 | { |
ce8eb6c4 | 2241 | struct kmem_cache_node *n; |
e498be7d | 2242 | int node; |
18726ca8 | 2243 | LIST_HEAD(list); |
e498be7d | 2244 | |
15c8b6c1 | 2245 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2246 | check_irq_on(); |
18bf8541 CL |
2247 | for_each_kmem_cache_node(cachep, node, n) |
2248 | if (n->alien) | |
ce8eb6c4 | 2249 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2250 | |
18726ca8 JK |
2251 | for_each_kmem_cache_node(cachep, node, n) { |
2252 | spin_lock_irq(&n->list_lock); | |
2253 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2254 | spin_unlock_irq(&n->list_lock); | |
2255 | ||
2256 | slabs_destroy(cachep, &list); | |
2257 | } | |
1da177e4 LT |
2258 | } |
2259 | ||
ed11d9eb CL |
2260 | /* |
2261 | * Remove slabs from the list of free slabs. | |
2262 | * Specify the number of slabs to drain in tofree. | |
2263 | * | |
2264 | * Returns the actual number of slabs released. | |
2265 | */ | |
2266 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2267 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2268 | { |
ed11d9eb CL |
2269 | struct list_head *p; |
2270 | int nr_freed; | |
8456a648 | 2271 | struct page *page; |
1da177e4 | 2272 | |
ed11d9eb | 2273 | nr_freed = 0; |
ce8eb6c4 | 2274 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2275 | |
ce8eb6c4 CL |
2276 | spin_lock_irq(&n->list_lock); |
2277 | p = n->slabs_free.prev; | |
2278 | if (p == &n->slabs_free) { | |
2279 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2280 | goto out; |
2281 | } | |
1da177e4 | 2282 | |
8456a648 | 2283 | page = list_entry(p, struct page, lru); |
8456a648 | 2284 | list_del(&page->lru); |
ed11d9eb CL |
2285 | /* |
2286 | * Safe to drop the lock. The slab is no longer linked | |
2287 | * to the cache. | |
2288 | */ | |
ce8eb6c4 CL |
2289 | n->free_objects -= cache->num; |
2290 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2291 | slab_destroy(cache, page); |
ed11d9eb | 2292 | nr_freed++; |
1da177e4 | 2293 | } |
ed11d9eb CL |
2294 | out: |
2295 | return nr_freed; | |
1da177e4 LT |
2296 | } |
2297 | ||
d6e0b7fa | 2298 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2299 | { |
18bf8541 CL |
2300 | int ret = 0; |
2301 | int node; | |
ce8eb6c4 | 2302 | struct kmem_cache_node *n; |
e498be7d CL |
2303 | |
2304 | drain_cpu_caches(cachep); | |
2305 | ||
2306 | check_irq_on(); | |
18bf8541 | 2307 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2308 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2309 | |
ce8eb6c4 CL |
2310 | ret += !list_empty(&n->slabs_full) || |
2311 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2312 | } |
2313 | return (ret ? 1 : 0); | |
2314 | } | |
2315 | ||
945cf2b6 | 2316 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2317 | { |
2318 | return __kmem_cache_shrink(cachep, false); | |
2319 | } | |
2320 | ||
2321 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2322 | { |
12c3667f | 2323 | int i; |
ce8eb6c4 | 2324 | struct kmem_cache_node *n; |
1da177e4 | 2325 | |
bf0dea23 | 2326 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2327 | |
ce8eb6c4 | 2328 | /* NUMA: free the node structures */ |
18bf8541 CL |
2329 | for_each_kmem_cache_node(cachep, i, n) { |
2330 | kfree(n->shared); | |
2331 | free_alien_cache(n->alien); | |
2332 | kfree(n); | |
2333 | cachep->node[i] = NULL; | |
12c3667f | 2334 | } |
1da177e4 | 2335 | } |
1da177e4 | 2336 | |
e5ac9c5a RT |
2337 | /* |
2338 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2339 | * |
2340 | * For a slab cache when the slab descriptor is off-slab, the | |
2341 | * slab descriptor can't come from the same cache which is being created, | |
2342 | * Because if it is the case, that means we defer the creation of | |
2343 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2344 | * And we eventually call down to __kmem_cache_create(), which | |
2345 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2346 | * This is a "chicken-and-egg" problem. | |
2347 | * | |
2348 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2349 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2350 | */ |
7e007355 | 2351 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2352 | struct page *page, int colour_off, |
2353 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2354 | { |
7e007355 | 2355 | void *freelist; |
0c3aa83e | 2356 | void *addr = page_address(page); |
b28a02de | 2357 | |
2e6b3602 JK |
2358 | page->s_mem = addr + colour_off; |
2359 | page->active = 0; | |
2360 | ||
b03a017b JK |
2361 | if (OBJFREELIST_SLAB(cachep)) |
2362 | freelist = NULL; | |
2363 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2364 | /* Slab management obj is off-slab. */ |
8456a648 | 2365 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2366 | local_flags, nodeid); |
8456a648 | 2367 | if (!freelist) |
1da177e4 LT |
2368 | return NULL; |
2369 | } else { | |
2e6b3602 JK |
2370 | /* We will use last bytes at the slab for freelist */ |
2371 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2372 | cachep->freelist_size; | |
1da177e4 | 2373 | } |
2e6b3602 | 2374 | |
8456a648 | 2375 | return freelist; |
1da177e4 LT |
2376 | } |
2377 | ||
7cc68973 | 2378 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2379 | { |
a41adfaa | 2380 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2381 | } |
2382 | ||
2383 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2384 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2385 | { |
a41adfaa | 2386 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2387 | } |
2388 | ||
10b2e9e8 | 2389 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2390 | { |
10b2e9e8 | 2391 | #if DEBUG |
1da177e4 LT |
2392 | int i; |
2393 | ||
2394 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2395 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2396 | |
1da177e4 LT |
2397 | if (cachep->flags & SLAB_STORE_USER) |
2398 | *dbg_userword(cachep, objp) = NULL; | |
2399 | ||
2400 | if (cachep->flags & SLAB_RED_ZONE) { | |
2401 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2402 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2403 | } | |
2404 | /* | |
a737b3e2 AM |
2405 | * Constructors are not allowed to allocate memory from the same |
2406 | * cache which they are a constructor for. Otherwise, deadlock. | |
2407 | * They must also be threaded. | |
1da177e4 | 2408 | */ |
7ed2f9e6 AP |
2409 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2410 | kasan_unpoison_object_data(cachep, | |
2411 | objp + obj_offset(cachep)); | |
51cc5068 | 2412 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2413 | kasan_poison_object_data( |
2414 | cachep, objp + obj_offset(cachep)); | |
2415 | } | |
1da177e4 LT |
2416 | |
2417 | if (cachep->flags & SLAB_RED_ZONE) { | |
2418 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2419 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2420 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2421 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2422 | } |
40b44137 JK |
2423 | /* need to poison the objs? */ |
2424 | if (cachep->flags & SLAB_POISON) { | |
2425 | poison_obj(cachep, objp, POISON_FREE); | |
2426 | slab_kernel_map(cachep, objp, 0, 0); | |
2427 | } | |
10b2e9e8 | 2428 | } |
1da177e4 | 2429 | #endif |
10b2e9e8 JK |
2430 | } |
2431 | ||
2432 | static void cache_init_objs(struct kmem_cache *cachep, | |
2433 | struct page *page) | |
2434 | { | |
2435 | int i; | |
7ed2f9e6 | 2436 | void *objp; |
10b2e9e8 JK |
2437 | |
2438 | cache_init_objs_debug(cachep, page); | |
2439 | ||
b03a017b JK |
2440 | if (OBJFREELIST_SLAB(cachep)) { |
2441 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + | |
2442 | obj_offset(cachep); | |
2443 | } | |
2444 | ||
10b2e9e8 JK |
2445 | for (i = 0; i < cachep->num; i++) { |
2446 | /* constructor could break poison info */ | |
7ed2f9e6 AP |
2447 | if (DEBUG == 0 && cachep->ctor) { |
2448 | objp = index_to_obj(cachep, page, i); | |
2449 | kasan_unpoison_object_data(cachep, objp); | |
2450 | cachep->ctor(objp); | |
2451 | kasan_poison_object_data(cachep, objp); | |
2452 | } | |
10b2e9e8 | 2453 | |
e5c58dfd | 2454 | set_free_obj(page, i, i); |
1da177e4 | 2455 | } |
1da177e4 LT |
2456 | } |
2457 | ||
343e0d7a | 2458 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2459 | { |
4b51d669 CL |
2460 | if (CONFIG_ZONE_DMA_FLAG) { |
2461 | if (flags & GFP_DMA) | |
a618e89f | 2462 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2463 | else |
a618e89f | 2464 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2465 | } |
1da177e4 LT |
2466 | } |
2467 | ||
260b61dd | 2468 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2469 | { |
b1cb0982 | 2470 | void *objp; |
78d382d7 | 2471 | |
e5c58dfd | 2472 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2473 | page->active++; |
78d382d7 | 2474 | |
d31676df JK |
2475 | #if DEBUG |
2476 | if (cachep->flags & SLAB_STORE_USER) | |
2477 | set_store_user_dirty(cachep); | |
2478 | #endif | |
2479 | ||
78d382d7 MD |
2480 | return objp; |
2481 | } | |
2482 | ||
260b61dd JK |
2483 | static void slab_put_obj(struct kmem_cache *cachep, |
2484 | struct page *page, void *objp) | |
78d382d7 | 2485 | { |
8456a648 | 2486 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2487 | #if DEBUG |
16025177 | 2488 | unsigned int i; |
b1cb0982 | 2489 | |
b1cb0982 | 2490 | /* Verify double free bug */ |
8456a648 | 2491 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2492 | if (get_free_obj(page, i) == objnr) { |
1170532b | 2493 | pr_err("slab: double free detected in cache '%s', objp %p\n", |
756a025f | 2494 | cachep->name, objp); |
b1cb0982 JK |
2495 | BUG(); |
2496 | } | |
78d382d7 MD |
2497 | } |
2498 | #endif | |
8456a648 | 2499 | page->active--; |
b03a017b JK |
2500 | if (!page->freelist) |
2501 | page->freelist = objp + obj_offset(cachep); | |
2502 | ||
e5c58dfd | 2503 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2504 | } |
2505 | ||
4776874f PE |
2506 | /* |
2507 | * Map pages beginning at addr to the given cache and slab. This is required | |
2508 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2509 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2510 | */ |
8456a648 | 2511 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2512 | void *freelist) |
1da177e4 | 2513 | { |
a57a4988 | 2514 | page->slab_cache = cache; |
8456a648 | 2515 | page->freelist = freelist; |
1da177e4 LT |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2520 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2521 | */ | |
3c517a61 | 2522 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2523 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2524 | { |
7e007355 | 2525 | void *freelist; |
b28a02de PE |
2526 | size_t offset; |
2527 | gfp_t local_flags; | |
ce8eb6c4 | 2528 | struct kmem_cache_node *n; |
1da177e4 | 2529 | |
a737b3e2 AM |
2530 | /* |
2531 | * Be lazy and only check for valid flags here, keeping it out of the | |
2532 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2533 | */ |
c871ac4e AM |
2534 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2535 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2536 | BUG(); | |
2537 | } | |
6cb06229 | 2538 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2539 | |
ce8eb6c4 | 2540 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2541 | check_irq_off(); |
18bf8541 | 2542 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2543 | spin_lock(&n->list_lock); |
1da177e4 LT |
2544 | |
2545 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2546 | offset = n->colour_next; |
2547 | n->colour_next++; | |
2548 | if (n->colour_next >= cachep->colour) | |
2549 | n->colour_next = 0; | |
2550 | spin_unlock(&n->list_lock); | |
1da177e4 | 2551 | |
2e1217cf | 2552 | offset *= cachep->colour_off; |
1da177e4 | 2553 | |
d0164adc | 2554 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2555 | local_irq_enable(); |
2556 | ||
2557 | /* | |
2558 | * The test for missing atomic flag is performed here, rather than | |
2559 | * the more obvious place, simply to reduce the critical path length | |
2560 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2561 | * will eventually be caught here (where it matters). | |
2562 | */ | |
2563 | kmem_flagcheck(cachep, flags); | |
2564 | ||
a737b3e2 AM |
2565 | /* |
2566 | * Get mem for the objs. Attempt to allocate a physical page from | |
2567 | * 'nodeid'. | |
e498be7d | 2568 | */ |
0c3aa83e JK |
2569 | if (!page) |
2570 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2571 | if (!page) | |
1da177e4 LT |
2572 | goto failed; |
2573 | ||
2574 | /* Get slab management. */ | |
8456a648 | 2575 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2576 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
b03a017b | 2577 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2578 | goto opps1; |
2579 | ||
8456a648 | 2580 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2581 | |
7ed2f9e6 | 2582 | kasan_poison_slab(page); |
8456a648 | 2583 | cache_init_objs(cachep, page); |
1da177e4 | 2584 | |
d0164adc | 2585 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2586 | local_irq_disable(); |
2587 | check_irq_off(); | |
ce8eb6c4 | 2588 | spin_lock(&n->list_lock); |
1da177e4 LT |
2589 | |
2590 | /* Make slab active. */ | |
8456a648 | 2591 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2592 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2593 | n->free_objects += cachep->num; |
2594 | spin_unlock(&n->list_lock); | |
1da177e4 | 2595 | return 1; |
a737b3e2 | 2596 | opps1: |
0c3aa83e | 2597 | kmem_freepages(cachep, page); |
a737b3e2 | 2598 | failed: |
d0164adc | 2599 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2600 | local_irq_disable(); |
2601 | return 0; | |
2602 | } | |
2603 | ||
2604 | #if DEBUG | |
2605 | ||
2606 | /* | |
2607 | * Perform extra freeing checks: | |
2608 | * - detect bad pointers. | |
2609 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2610 | */ |
2611 | static void kfree_debugcheck(const void *objp) | |
2612 | { | |
1da177e4 | 2613 | if (!virt_addr_valid(objp)) { |
1170532b | 2614 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2615 | (unsigned long)objp); |
2616 | BUG(); | |
1da177e4 | 2617 | } |
1da177e4 LT |
2618 | } |
2619 | ||
58ce1fd5 PE |
2620 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2621 | { | |
b46b8f19 | 2622 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2623 | |
2624 | redzone1 = *dbg_redzone1(cache, obj); | |
2625 | redzone2 = *dbg_redzone2(cache, obj); | |
2626 | ||
2627 | /* | |
2628 | * Redzone is ok. | |
2629 | */ | |
2630 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2631 | return; | |
2632 | ||
2633 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2634 | slab_error(cache, "double free detected"); | |
2635 | else | |
2636 | slab_error(cache, "memory outside object was overwritten"); | |
2637 | ||
1170532b JP |
2638 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2639 | obj, redzone1, redzone2); | |
58ce1fd5 PE |
2640 | } |
2641 | ||
343e0d7a | 2642 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2643 | unsigned long caller) |
1da177e4 | 2644 | { |
1da177e4 | 2645 | unsigned int objnr; |
8456a648 | 2646 | struct page *page; |
1da177e4 | 2647 | |
80cbd911 MW |
2648 | BUG_ON(virt_to_cache(objp) != cachep); |
2649 | ||
3dafccf2 | 2650 | objp -= obj_offset(cachep); |
1da177e4 | 2651 | kfree_debugcheck(objp); |
b49af68f | 2652 | page = virt_to_head_page(objp); |
1da177e4 | 2653 | |
1da177e4 | 2654 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2655 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2656 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2657 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2658 | } | |
d31676df JK |
2659 | if (cachep->flags & SLAB_STORE_USER) { |
2660 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2661 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2662 | } |
1da177e4 | 2663 | |
8456a648 | 2664 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2665 | |
2666 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2667 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2668 | |
1da177e4 | 2669 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2670 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2671 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2672 | } |
2673 | return objp; | |
2674 | } | |
2675 | ||
1da177e4 LT |
2676 | #else |
2677 | #define kfree_debugcheck(x) do { } while(0) | |
2678 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2679 | #endif |
2680 | ||
b03a017b JK |
2681 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2682 | void **list) | |
2683 | { | |
2684 | #if DEBUG | |
2685 | void *next = *list; | |
2686 | void *objp; | |
2687 | ||
2688 | while (next) { | |
2689 | objp = next - obj_offset(cachep); | |
2690 | next = *(void **)next; | |
2691 | poison_obj(cachep, objp, POISON_FREE); | |
2692 | } | |
2693 | #endif | |
2694 | } | |
2695 | ||
d8410234 | 2696 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2697 | struct kmem_cache_node *n, struct page *page, |
2698 | void **list) | |
d8410234 JK |
2699 | { |
2700 | /* move slabp to correct slabp list: */ | |
2701 | list_del(&page->lru); | |
b03a017b | 2702 | if (page->active == cachep->num) { |
d8410234 | 2703 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2704 | if (OBJFREELIST_SLAB(cachep)) { |
2705 | #if DEBUG | |
2706 | /* Poisoning will be done without holding the lock */ | |
2707 | if (cachep->flags & SLAB_POISON) { | |
2708 | void **objp = page->freelist; | |
2709 | ||
2710 | *objp = *list; | |
2711 | *list = objp; | |
2712 | } | |
2713 | #endif | |
2714 | page->freelist = NULL; | |
2715 | } | |
2716 | } else | |
d8410234 JK |
2717 | list_add(&page->lru, &n->slabs_partial); |
2718 | } | |
2719 | ||
f68f8ddd JK |
2720 | /* Try to find non-pfmemalloc slab if needed */ |
2721 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
2722 | struct page *page, bool pfmemalloc) | |
2723 | { | |
2724 | if (!page) | |
2725 | return NULL; | |
2726 | ||
2727 | if (pfmemalloc) | |
2728 | return page; | |
2729 | ||
2730 | if (!PageSlabPfmemalloc(page)) | |
2731 | return page; | |
2732 | ||
2733 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2734 | if (n->free_objects > n->free_limit) { | |
2735 | ClearPageSlabPfmemalloc(page); | |
2736 | return page; | |
2737 | } | |
2738 | ||
2739 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2740 | list_del(&page->lru); | |
2741 | if (!page->active) | |
2742 | list_add_tail(&page->lru, &n->slabs_free); | |
2743 | else | |
2744 | list_add_tail(&page->lru, &n->slabs_partial); | |
2745 | ||
2746 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2747 | if (!PageSlabPfmemalloc(page)) | |
2748 | return page; | |
2749 | } | |
2750 | ||
2751 | list_for_each_entry(page, &n->slabs_free, lru) { | |
2752 | if (!PageSlabPfmemalloc(page)) | |
2753 | return page; | |
2754 | } | |
2755 | ||
2756 | return NULL; | |
2757 | } | |
2758 | ||
2759 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2760 | { |
2761 | struct page *page; | |
2762 | ||
2763 | page = list_first_entry_or_null(&n->slabs_partial, | |
2764 | struct page, lru); | |
2765 | if (!page) { | |
2766 | n->free_touched = 1; | |
2767 | page = list_first_entry_or_null(&n->slabs_free, | |
2768 | struct page, lru); | |
2769 | } | |
2770 | ||
f68f8ddd JK |
2771 | if (sk_memalloc_socks()) |
2772 | return get_valid_first_slab(n, page, pfmemalloc); | |
2773 | ||
7aa0d227 GT |
2774 | return page; |
2775 | } | |
2776 | ||
f68f8ddd JK |
2777 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2778 | struct kmem_cache_node *n, gfp_t flags) | |
2779 | { | |
2780 | struct page *page; | |
2781 | void *obj; | |
2782 | void *list = NULL; | |
2783 | ||
2784 | if (!gfp_pfmemalloc_allowed(flags)) | |
2785 | return NULL; | |
2786 | ||
2787 | spin_lock(&n->list_lock); | |
2788 | page = get_first_slab(n, true); | |
2789 | if (!page) { | |
2790 | spin_unlock(&n->list_lock); | |
2791 | return NULL; | |
2792 | } | |
2793 | ||
2794 | obj = slab_get_obj(cachep, page); | |
2795 | n->free_objects--; | |
2796 | ||
2797 | fixup_slab_list(cachep, n, page, &list); | |
2798 | ||
2799 | spin_unlock(&n->list_lock); | |
2800 | fixup_objfreelist_debug(cachep, &list); | |
2801 | ||
2802 | return obj; | |
2803 | } | |
2804 | ||
2805 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) | |
1da177e4 LT |
2806 | { |
2807 | int batchcount; | |
ce8eb6c4 | 2808 | struct kmem_cache_node *n; |
1da177e4 | 2809 | struct array_cache *ac; |
1ca4cb24 | 2810 | int node; |
b03a017b | 2811 | void *list = NULL; |
1ca4cb24 | 2812 | |
1da177e4 | 2813 | check_irq_off(); |
7d6e6d09 | 2814 | node = numa_mem_id(); |
f68f8ddd | 2815 | |
072bb0aa | 2816 | retry: |
9a2dba4b | 2817 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2818 | batchcount = ac->batchcount; |
2819 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2820 | /* |
2821 | * If there was little recent activity on this cache, then | |
2822 | * perform only a partial refill. Otherwise we could generate | |
2823 | * refill bouncing. | |
1da177e4 LT |
2824 | */ |
2825 | batchcount = BATCHREFILL_LIMIT; | |
2826 | } | |
18bf8541 | 2827 | n = get_node(cachep, node); |
e498be7d | 2828 | |
ce8eb6c4 CL |
2829 | BUG_ON(ac->avail > 0 || !n); |
2830 | spin_lock(&n->list_lock); | |
1da177e4 | 2831 | |
3ded175a | 2832 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2833 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2834 | n->shared->touched = 1; | |
3ded175a | 2835 | goto alloc_done; |
44b57f1c | 2836 | } |
3ded175a | 2837 | |
1da177e4 | 2838 | while (batchcount > 0) { |
8456a648 | 2839 | struct page *page; |
1da177e4 | 2840 | /* Get slab alloc is to come from. */ |
f68f8ddd | 2841 | page = get_first_slab(n, false); |
7aa0d227 GT |
2842 | if (!page) |
2843 | goto must_grow; | |
1da177e4 | 2844 | |
1da177e4 | 2845 | check_spinlock_acquired(cachep); |
714b8171 PE |
2846 | |
2847 | /* | |
2848 | * The slab was either on partial or free list so | |
2849 | * there must be at least one object available for | |
2850 | * allocation. | |
2851 | */ | |
8456a648 | 2852 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2853 | |
8456a648 | 2854 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2855 | STATS_INC_ALLOCED(cachep); |
2856 | STATS_INC_ACTIVE(cachep); | |
2857 | STATS_SET_HIGH(cachep); | |
2858 | ||
f68f8ddd | 2859 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); |
1da177e4 | 2860 | } |
1da177e4 | 2861 | |
b03a017b | 2862 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2863 | } |
2864 | ||
a737b3e2 | 2865 | must_grow: |
ce8eb6c4 | 2866 | n->free_objects -= ac->avail; |
a737b3e2 | 2867 | alloc_done: |
ce8eb6c4 | 2868 | spin_unlock(&n->list_lock); |
b03a017b | 2869 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 LT |
2870 | |
2871 | if (unlikely(!ac->avail)) { | |
2872 | int x; | |
f68f8ddd JK |
2873 | |
2874 | /* Check if we can use obj in pfmemalloc slab */ | |
2875 | if (sk_memalloc_socks()) { | |
2876 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
2877 | ||
2878 | if (obj) | |
2879 | return obj; | |
2880 | } | |
2881 | ||
4167e9b2 | 2882 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2883 | |
a737b3e2 | 2884 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2885 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2886 | node = numa_mem_id(); |
072bb0aa MG |
2887 | |
2888 | /* no objects in sight? abort */ | |
f68f8ddd | 2889 | if (!x && ac->avail == 0) |
1da177e4 LT |
2890 | return NULL; |
2891 | ||
a737b3e2 | 2892 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2893 | goto retry; |
2894 | } | |
2895 | ac->touched = 1; | |
072bb0aa | 2896 | |
f68f8ddd | 2897 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2898 | } |
2899 | ||
a737b3e2 AM |
2900 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2901 | gfp_t flags) | |
1da177e4 | 2902 | { |
d0164adc | 2903 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2904 | #if DEBUG |
2905 | kmem_flagcheck(cachep, flags); | |
2906 | #endif | |
2907 | } | |
2908 | ||
2909 | #if DEBUG | |
a737b3e2 | 2910 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2911 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2912 | { |
b28a02de | 2913 | if (!objp) |
1da177e4 | 2914 | return objp; |
b28a02de | 2915 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2916 | check_poison_obj(cachep, objp); |
40b44137 | 2917 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2918 | poison_obj(cachep, objp, POISON_INUSE); |
2919 | } | |
2920 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2921 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2922 | |
2923 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2924 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2925 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 2926 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
1170532b JP |
2927 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2928 | objp, *dbg_redzone1(cachep, objp), | |
2929 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2930 | } |
2931 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2932 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2933 | } | |
03787301 | 2934 | |
3dafccf2 | 2935 | objp += obj_offset(cachep); |
4f104934 | 2936 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2937 | cachep->ctor(objp); |
7ea466f2 TH |
2938 | if (ARCH_SLAB_MINALIGN && |
2939 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
1170532b | 2940 | pr_err("0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2941 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2942 | } |
1da177e4 LT |
2943 | return objp; |
2944 | } | |
2945 | #else | |
2946 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2947 | #endif | |
2948 | ||
343e0d7a | 2949 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2950 | { |
b28a02de | 2951 | void *objp; |
1da177e4 LT |
2952 | struct array_cache *ac; |
2953 | ||
5c382300 | 2954 | check_irq_off(); |
8a8b6502 | 2955 | |
9a2dba4b | 2956 | ac = cpu_cache_get(cachep); |
1da177e4 | 2957 | if (likely(ac->avail)) { |
1da177e4 | 2958 | ac->touched = 1; |
f68f8ddd | 2959 | objp = ac->entry[--ac->avail]; |
072bb0aa | 2960 | |
f68f8ddd JK |
2961 | STATS_INC_ALLOCHIT(cachep); |
2962 | goto out; | |
1da177e4 | 2963 | } |
072bb0aa MG |
2964 | |
2965 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 2966 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
2967 | /* |
2968 | * the 'ac' may be updated by cache_alloc_refill(), | |
2969 | * and kmemleak_erase() requires its correct value. | |
2970 | */ | |
2971 | ac = cpu_cache_get(cachep); | |
2972 | ||
2973 | out: | |
d5cff635 CM |
2974 | /* |
2975 | * To avoid a false negative, if an object that is in one of the | |
2976 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2977 | * treat the array pointers as a reference to the object. | |
2978 | */ | |
f3d8b53a O |
2979 | if (objp) |
2980 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2981 | return objp; |
2982 | } | |
2983 | ||
e498be7d | 2984 | #ifdef CONFIG_NUMA |
c61afb18 | 2985 | /* |
2ad654bc | 2986 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2987 | * |
2988 | * If we are in_interrupt, then process context, including cpusets and | |
2989 | * mempolicy, may not apply and should not be used for allocation policy. | |
2990 | */ | |
2991 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2992 | { | |
2993 | int nid_alloc, nid_here; | |
2994 | ||
765c4507 | 2995 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 2996 | return NULL; |
7d6e6d09 | 2997 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 2998 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 2999 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3000 | else if (current->mempolicy) |
2a389610 | 3001 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3002 | if (nid_alloc != nid_here) |
8b98c169 | 3003 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3004 | return NULL; |
3005 | } | |
3006 | ||
765c4507 CL |
3007 | /* |
3008 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3009 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3010 | * available node for available objects. If that fails then we |
3c517a61 CL |
3011 | * perform an allocation without specifying a node. This allows the page |
3012 | * allocator to do its reclaim / fallback magic. We then insert the | |
3013 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3014 | */ |
8c8cc2c1 | 3015 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3016 | { |
8c8cc2c1 PE |
3017 | struct zonelist *zonelist; |
3018 | gfp_t local_flags; | |
dd1a239f | 3019 | struct zoneref *z; |
54a6eb5c MG |
3020 | struct zone *zone; |
3021 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3022 | void *obj = NULL; |
3c517a61 | 3023 | int nid; |
cc9a6c87 | 3024 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3025 | |
3026 | if (flags & __GFP_THISNODE) | |
3027 | return NULL; | |
3028 | ||
6cb06229 | 3029 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3030 | |
cc9a6c87 | 3031 | retry_cpuset: |
d26914d1 | 3032 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3033 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3034 | |
3c517a61 CL |
3035 | retry: |
3036 | /* | |
3037 | * Look through allowed nodes for objects available | |
3038 | * from existing per node queues. | |
3039 | */ | |
54a6eb5c MG |
3040 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3041 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3042 | |
061d7074 | 3043 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3044 | get_node(cache, nid) && |
3045 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3046 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3047 | gfp_exact_node(flags), nid); |
481c5346 CL |
3048 | if (obj) |
3049 | break; | |
3050 | } | |
3c517a61 CL |
3051 | } |
3052 | ||
cfce6604 | 3053 | if (!obj) { |
3c517a61 CL |
3054 | /* |
3055 | * This allocation will be performed within the constraints | |
3056 | * of the current cpuset / memory policy requirements. | |
3057 | * We may trigger various forms of reclaim on the allowed | |
3058 | * set and go into memory reserves if necessary. | |
3059 | */ | |
0c3aa83e JK |
3060 | struct page *page; |
3061 | ||
d0164adc | 3062 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3063 | local_irq_enable(); |
3064 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3065 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3066 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3067 | local_irq_disable(); |
0c3aa83e | 3068 | if (page) { |
3c517a61 CL |
3069 | /* |
3070 | * Insert into the appropriate per node queues | |
3071 | */ | |
0c3aa83e JK |
3072 | nid = page_to_nid(page); |
3073 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3074 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3075 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3076 | if (!obj) |
3077 | /* | |
3078 | * Another processor may allocate the | |
3079 | * objects in the slab since we are | |
3080 | * not holding any locks. | |
3081 | */ | |
3082 | goto retry; | |
3083 | } else { | |
b6a60451 | 3084 | /* cache_grow already freed obj */ |
3c517a61 CL |
3085 | obj = NULL; |
3086 | } | |
3087 | } | |
aedb0eb1 | 3088 | } |
cc9a6c87 | 3089 | |
d26914d1 | 3090 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3091 | goto retry_cpuset; |
765c4507 CL |
3092 | return obj; |
3093 | } | |
3094 | ||
e498be7d CL |
3095 | /* |
3096 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3097 | */ |
8b98c169 | 3098 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3099 | int nodeid) |
e498be7d | 3100 | { |
8456a648 | 3101 | struct page *page; |
ce8eb6c4 | 3102 | struct kmem_cache_node *n; |
b28a02de | 3103 | void *obj; |
b03a017b | 3104 | void *list = NULL; |
b28a02de PE |
3105 | int x; |
3106 | ||
7c3fbbdd | 3107 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3108 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3109 | BUG_ON(!n); |
b28a02de | 3110 | |
a737b3e2 | 3111 | retry: |
ca3b9b91 | 3112 | check_irq_off(); |
ce8eb6c4 | 3113 | spin_lock(&n->list_lock); |
f68f8ddd | 3114 | page = get_first_slab(n, false); |
7aa0d227 GT |
3115 | if (!page) |
3116 | goto must_grow; | |
b28a02de | 3117 | |
b28a02de | 3118 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3119 | |
3120 | STATS_INC_NODEALLOCS(cachep); | |
3121 | STATS_INC_ACTIVE(cachep); | |
3122 | STATS_SET_HIGH(cachep); | |
3123 | ||
8456a648 | 3124 | BUG_ON(page->active == cachep->num); |
b28a02de | 3125 | |
260b61dd | 3126 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3127 | n->free_objects--; |
b28a02de | 3128 | |
b03a017b | 3129 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3130 | |
ce8eb6c4 | 3131 | spin_unlock(&n->list_lock); |
b03a017b | 3132 | fixup_objfreelist_debug(cachep, &list); |
b28a02de | 3133 | goto done; |
e498be7d | 3134 | |
a737b3e2 | 3135 | must_grow: |
ce8eb6c4 | 3136 | spin_unlock(&n->list_lock); |
4167e9b2 | 3137 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3138 | if (x) |
3139 | goto retry; | |
1da177e4 | 3140 | |
8c8cc2c1 | 3141 | return fallback_alloc(cachep, flags); |
e498be7d | 3142 | |
a737b3e2 | 3143 | done: |
b28a02de | 3144 | return obj; |
e498be7d | 3145 | } |
8c8cc2c1 | 3146 | |
8c8cc2c1 | 3147 | static __always_inline void * |
48356303 | 3148 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3149 | unsigned long caller) |
8c8cc2c1 PE |
3150 | { |
3151 | unsigned long save_flags; | |
3152 | void *ptr; | |
7d6e6d09 | 3153 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3154 | |
dcce284a | 3155 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3156 | cachep = slab_pre_alloc_hook(cachep, flags); |
3157 | if (unlikely(!cachep)) | |
824ebef1 AM |
3158 | return NULL; |
3159 | ||
8c8cc2c1 PE |
3160 | cache_alloc_debugcheck_before(cachep, flags); |
3161 | local_irq_save(save_flags); | |
3162 | ||
eacbbae3 | 3163 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3164 | nodeid = slab_node; |
8c8cc2c1 | 3165 | |
18bf8541 | 3166 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3167 | /* Node not bootstrapped yet */ |
3168 | ptr = fallback_alloc(cachep, flags); | |
3169 | goto out; | |
3170 | } | |
3171 | ||
7d6e6d09 | 3172 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3173 | /* |
3174 | * Use the locally cached objects if possible. | |
3175 | * However ____cache_alloc does not allow fallback | |
3176 | * to other nodes. It may fail while we still have | |
3177 | * objects on other nodes available. | |
3178 | */ | |
3179 | ptr = ____cache_alloc(cachep, flags); | |
3180 | if (ptr) | |
3181 | goto out; | |
3182 | } | |
3183 | /* ___cache_alloc_node can fall back to other nodes */ | |
3184 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3185 | out: | |
3186 | local_irq_restore(save_flags); | |
3187 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3188 | ||
d5e3ed66 JDB |
3189 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3190 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3191 | |
d5e3ed66 | 3192 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3193 | return ptr; |
3194 | } | |
3195 | ||
3196 | static __always_inline void * | |
3197 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3198 | { | |
3199 | void *objp; | |
3200 | ||
2ad654bc | 3201 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3202 | objp = alternate_node_alloc(cache, flags); |
3203 | if (objp) | |
3204 | goto out; | |
3205 | } | |
3206 | objp = ____cache_alloc(cache, flags); | |
3207 | ||
3208 | /* | |
3209 | * We may just have run out of memory on the local node. | |
3210 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3211 | */ | |
7d6e6d09 LS |
3212 | if (!objp) |
3213 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3214 | |
3215 | out: | |
3216 | return objp; | |
3217 | } | |
3218 | #else | |
3219 | ||
3220 | static __always_inline void * | |
3221 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3222 | { | |
3223 | return ____cache_alloc(cachep, flags); | |
3224 | } | |
3225 | ||
3226 | #endif /* CONFIG_NUMA */ | |
3227 | ||
3228 | static __always_inline void * | |
48356303 | 3229 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3230 | { |
3231 | unsigned long save_flags; | |
3232 | void *objp; | |
3233 | ||
dcce284a | 3234 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3235 | cachep = slab_pre_alloc_hook(cachep, flags); |
3236 | if (unlikely(!cachep)) | |
824ebef1 AM |
3237 | return NULL; |
3238 | ||
8c8cc2c1 PE |
3239 | cache_alloc_debugcheck_before(cachep, flags); |
3240 | local_irq_save(save_flags); | |
3241 | objp = __do_cache_alloc(cachep, flags); | |
3242 | local_irq_restore(save_flags); | |
3243 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3244 | prefetchw(objp); | |
3245 | ||
d5e3ed66 JDB |
3246 | if (unlikely(flags & __GFP_ZERO) && objp) |
3247 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3248 | |
d5e3ed66 | 3249 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3250 | return objp; |
3251 | } | |
e498be7d CL |
3252 | |
3253 | /* | |
5f0985bb | 3254 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3255 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3256 | */ |
97654dfa JK |
3257 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3258 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3259 | { |
3260 | int i; | |
25c063fb | 3261 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3262 | |
3263 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3264 | void *objp; |
8456a648 | 3265 | struct page *page; |
1da177e4 | 3266 | |
072bb0aa MG |
3267 | objp = objpp[i]; |
3268 | ||
8456a648 | 3269 | page = virt_to_head_page(objp); |
8456a648 | 3270 | list_del(&page->lru); |
ff69416e | 3271 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3272 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3273 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3274 | n->free_objects++; |
1da177e4 LT |
3275 | |
3276 | /* fixup slab chains */ | |
8456a648 | 3277 | if (page->active == 0) { |
ce8eb6c4 CL |
3278 | if (n->free_objects > n->free_limit) { |
3279 | n->free_objects -= cachep->num; | |
97654dfa | 3280 | list_add_tail(&page->lru, list); |
1da177e4 | 3281 | } else { |
8456a648 | 3282 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3283 | } |
3284 | } else { | |
3285 | /* Unconditionally move a slab to the end of the | |
3286 | * partial list on free - maximum time for the | |
3287 | * other objects to be freed, too. | |
3288 | */ | |
8456a648 | 3289 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3290 | } |
3291 | } | |
3292 | } | |
3293 | ||
343e0d7a | 3294 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3295 | { |
3296 | int batchcount; | |
ce8eb6c4 | 3297 | struct kmem_cache_node *n; |
7d6e6d09 | 3298 | int node = numa_mem_id(); |
97654dfa | 3299 | LIST_HEAD(list); |
1da177e4 LT |
3300 | |
3301 | batchcount = ac->batchcount; | |
260b61dd | 3302 | |
1da177e4 | 3303 | check_irq_off(); |
18bf8541 | 3304 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3305 | spin_lock(&n->list_lock); |
3306 | if (n->shared) { | |
3307 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3308 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3309 | if (max) { |
3310 | if (batchcount > max) | |
3311 | batchcount = max; | |
e498be7d | 3312 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3313 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3314 | shared_array->avail += batchcount; |
3315 | goto free_done; | |
3316 | } | |
3317 | } | |
3318 | ||
97654dfa | 3319 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3320 | free_done: |
1da177e4 LT |
3321 | #if STATS |
3322 | { | |
3323 | int i = 0; | |
73c0219d | 3324 | struct page *page; |
1da177e4 | 3325 | |
73c0219d | 3326 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3327 | BUG_ON(page->active); |
1da177e4 LT |
3328 | |
3329 | i++; | |
1da177e4 LT |
3330 | } |
3331 | STATS_SET_FREEABLE(cachep, i); | |
3332 | } | |
3333 | #endif | |
ce8eb6c4 | 3334 | spin_unlock(&n->list_lock); |
97654dfa | 3335 | slabs_destroy(cachep, &list); |
1da177e4 | 3336 | ac->avail -= batchcount; |
a737b3e2 | 3337 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3338 | } |
3339 | ||
3340 | /* | |
a737b3e2 AM |
3341 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3342 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3343 | */ |
a947eb95 | 3344 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3345 | unsigned long caller) |
1da177e4 | 3346 | { |
9a2dba4b | 3347 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 | 3348 | |
7ed2f9e6 AP |
3349 | kasan_slab_free(cachep, objp); |
3350 | ||
1da177e4 | 3351 | check_irq_off(); |
d5cff635 | 3352 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3353 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3354 | |
8c138bc0 | 3355 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3356 | |
1807a1aa SS |
3357 | /* |
3358 | * Skip calling cache_free_alien() when the platform is not numa. | |
3359 | * This will avoid cache misses that happen while accessing slabp (which | |
3360 | * is per page memory reference) to get nodeid. Instead use a global | |
3361 | * variable to skip the call, which is mostly likely to be present in | |
3362 | * the cache. | |
3363 | */ | |
b6e68bc1 | 3364 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3365 | return; |
3366 | ||
3d880194 | 3367 | if (ac->avail < ac->limit) { |
1da177e4 | 3368 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3369 | } else { |
3370 | STATS_INC_FREEMISS(cachep); | |
3371 | cache_flusharray(cachep, ac); | |
1da177e4 | 3372 | } |
42c8c99c | 3373 | |
f68f8ddd JK |
3374 | if (sk_memalloc_socks()) { |
3375 | struct page *page = virt_to_head_page(objp); | |
3376 | ||
3377 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3378 | cache_free_pfmemalloc(cachep, page, objp); | |
3379 | return; | |
3380 | } | |
3381 | } | |
3382 | ||
3383 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3384 | } |
3385 | ||
3386 | /** | |
3387 | * kmem_cache_alloc - Allocate an object | |
3388 | * @cachep: The cache to allocate from. | |
3389 | * @flags: See kmalloc(). | |
3390 | * | |
3391 | * Allocate an object from this cache. The flags are only relevant | |
3392 | * if the cache has no available objects. | |
3393 | */ | |
343e0d7a | 3394 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3395 | { |
48356303 | 3396 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3397 | |
505f5dcb | 3398 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3399 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3400 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3401 | |
3402 | return ret; | |
1da177e4 LT |
3403 | } |
3404 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3405 | ||
7b0501dd JDB |
3406 | static __always_inline void |
3407 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3408 | size_t size, void **p, unsigned long caller) | |
3409 | { | |
3410 | size_t i; | |
3411 | ||
3412 | for (i = 0; i < size; i++) | |
3413 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3414 | } | |
3415 | ||
865762a8 | 3416 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3417 | void **p) |
484748f0 | 3418 | { |
2a777eac JDB |
3419 | size_t i; |
3420 | ||
3421 | s = slab_pre_alloc_hook(s, flags); | |
3422 | if (!s) | |
3423 | return 0; | |
3424 | ||
3425 | cache_alloc_debugcheck_before(s, flags); | |
3426 | ||
3427 | local_irq_disable(); | |
3428 | for (i = 0; i < size; i++) { | |
3429 | void *objp = __do_cache_alloc(s, flags); | |
3430 | ||
2a777eac JDB |
3431 | if (unlikely(!objp)) |
3432 | goto error; | |
3433 | p[i] = objp; | |
3434 | } | |
3435 | local_irq_enable(); | |
3436 | ||
7b0501dd JDB |
3437 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3438 | ||
2a777eac JDB |
3439 | /* Clear memory outside IRQ disabled section */ |
3440 | if (unlikely(flags & __GFP_ZERO)) | |
3441 | for (i = 0; i < size; i++) | |
3442 | memset(p[i], 0, s->object_size); | |
3443 | ||
3444 | slab_post_alloc_hook(s, flags, size, p); | |
3445 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3446 | return size; | |
3447 | error: | |
3448 | local_irq_enable(); | |
7b0501dd | 3449 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3450 | slab_post_alloc_hook(s, flags, i, p); |
3451 | __kmem_cache_free_bulk(s, i, p); | |
3452 | return 0; | |
484748f0 CL |
3453 | } |
3454 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3455 | ||
0f24f128 | 3456 | #ifdef CONFIG_TRACING |
85beb586 | 3457 | void * |
4052147c | 3458 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3459 | { |
85beb586 SR |
3460 | void *ret; |
3461 | ||
48356303 | 3462 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3463 | |
505f5dcb | 3464 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3465 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3466 | size, cachep->size, flags); |
85beb586 | 3467 | return ret; |
36555751 | 3468 | } |
85beb586 | 3469 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3470 | #endif |
3471 | ||
1da177e4 | 3472 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3473 | /** |
3474 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3475 | * @cachep: The cache to allocate from. | |
3476 | * @flags: See kmalloc(). | |
3477 | * @nodeid: node number of the target node. | |
3478 | * | |
3479 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3480 | * node, which can improve the performance for cpu bound structures. | |
3481 | * | |
3482 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3483 | */ | |
8b98c169 CH |
3484 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3485 | { | |
48356303 | 3486 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3487 | |
505f5dcb | 3488 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3489 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3490 | cachep->object_size, cachep->size, |
ca2b84cb | 3491 | flags, nodeid); |
36555751 EGM |
3492 | |
3493 | return ret; | |
8b98c169 | 3494 | } |
1da177e4 LT |
3495 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3496 | ||
0f24f128 | 3497 | #ifdef CONFIG_TRACING |
4052147c | 3498 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3499 | gfp_t flags, |
4052147c EG |
3500 | int nodeid, |
3501 | size_t size) | |
36555751 | 3502 | { |
85beb586 SR |
3503 | void *ret; |
3504 | ||
592f4145 | 3505 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3506 | |
3507 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3508 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3509 | size, cachep->size, |
85beb586 SR |
3510 | flags, nodeid); |
3511 | return ret; | |
36555751 | 3512 | } |
85beb586 | 3513 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3514 | #endif |
3515 | ||
8b98c169 | 3516 | static __always_inline void * |
7c0cb9c6 | 3517 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3518 | { |
343e0d7a | 3519 | struct kmem_cache *cachep; |
7ed2f9e6 | 3520 | void *ret; |
97e2bde4 | 3521 | |
2c59dd65 | 3522 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3523 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3524 | return cachep; | |
7ed2f9e6 | 3525 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3526 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3527 | |
3528 | return ret; | |
97e2bde4 | 3529 | } |
8b98c169 | 3530 | |
8b98c169 CH |
3531 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3532 | { | |
7c0cb9c6 | 3533 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3534 | } |
dbe5e69d | 3535 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3536 | |
3537 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3538 | int node, unsigned long caller) |
8b98c169 | 3539 | { |
7c0cb9c6 | 3540 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3541 | } |
3542 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3543 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3544 | |
3545 | /** | |
800590f5 | 3546 | * __do_kmalloc - allocate memory |
1da177e4 | 3547 | * @size: how many bytes of memory are required. |
800590f5 | 3548 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3549 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3550 | */ |
7fd6b141 | 3551 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3552 | unsigned long caller) |
1da177e4 | 3553 | { |
343e0d7a | 3554 | struct kmem_cache *cachep; |
36555751 | 3555 | void *ret; |
1da177e4 | 3556 | |
2c59dd65 | 3557 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3558 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3559 | return cachep; | |
48356303 | 3560 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3561 | |
505f5dcb | 3562 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3563 | trace_kmalloc(caller, ret, |
3b0efdfa | 3564 | size, cachep->size, flags); |
36555751 EGM |
3565 | |
3566 | return ret; | |
7fd6b141 PE |
3567 | } |
3568 | ||
7fd6b141 PE |
3569 | void *__kmalloc(size_t size, gfp_t flags) |
3570 | { | |
7c0cb9c6 | 3571 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3572 | } |
3573 | EXPORT_SYMBOL(__kmalloc); | |
3574 | ||
ce71e27c | 3575 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3576 | { |
7c0cb9c6 | 3577 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3578 | } |
3579 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3580 | |
1da177e4 LT |
3581 | /** |
3582 | * kmem_cache_free - Deallocate an object | |
3583 | * @cachep: The cache the allocation was from. | |
3584 | * @objp: The previously allocated object. | |
3585 | * | |
3586 | * Free an object which was previously allocated from this | |
3587 | * cache. | |
3588 | */ | |
343e0d7a | 3589 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3590 | { |
3591 | unsigned long flags; | |
b9ce5ef4 GC |
3592 | cachep = cache_from_obj(cachep, objp); |
3593 | if (!cachep) | |
3594 | return; | |
1da177e4 LT |
3595 | |
3596 | local_irq_save(flags); | |
d97d476b | 3597 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3598 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3599 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3600 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3601 | local_irq_restore(flags); |
36555751 | 3602 | |
ca2b84cb | 3603 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3604 | } |
3605 | EXPORT_SYMBOL(kmem_cache_free); | |
3606 | ||
e6cdb58d JDB |
3607 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3608 | { | |
3609 | struct kmem_cache *s; | |
3610 | size_t i; | |
3611 | ||
3612 | local_irq_disable(); | |
3613 | for (i = 0; i < size; i++) { | |
3614 | void *objp = p[i]; | |
3615 | ||
ca257195 JDB |
3616 | if (!orig_s) /* called via kfree_bulk */ |
3617 | s = virt_to_cache(objp); | |
3618 | else | |
3619 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3620 | |
3621 | debug_check_no_locks_freed(objp, s->object_size); | |
3622 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3623 | debug_check_no_obj_freed(objp, s->object_size); | |
3624 | ||
3625 | __cache_free(s, objp, _RET_IP_); | |
3626 | } | |
3627 | local_irq_enable(); | |
3628 | ||
3629 | /* FIXME: add tracing */ | |
3630 | } | |
3631 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3632 | ||
1da177e4 LT |
3633 | /** |
3634 | * kfree - free previously allocated memory | |
3635 | * @objp: pointer returned by kmalloc. | |
3636 | * | |
80e93eff PE |
3637 | * If @objp is NULL, no operation is performed. |
3638 | * | |
1da177e4 LT |
3639 | * Don't free memory not originally allocated by kmalloc() |
3640 | * or you will run into trouble. | |
3641 | */ | |
3642 | void kfree(const void *objp) | |
3643 | { | |
343e0d7a | 3644 | struct kmem_cache *c; |
1da177e4 LT |
3645 | unsigned long flags; |
3646 | ||
2121db74 PE |
3647 | trace_kfree(_RET_IP_, objp); |
3648 | ||
6cb8f913 | 3649 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3650 | return; |
3651 | local_irq_save(flags); | |
3652 | kfree_debugcheck(objp); | |
6ed5eb22 | 3653 | c = virt_to_cache(objp); |
8c138bc0 CL |
3654 | debug_check_no_locks_freed(objp, c->object_size); |
3655 | ||
3656 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3657 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3658 | local_irq_restore(flags); |
3659 | } | |
3660 | EXPORT_SYMBOL(kfree); | |
3661 | ||
e498be7d | 3662 | /* |
ce8eb6c4 | 3663 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3664 | */ |
5f0985bb | 3665 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3666 | { |
3667 | int node; | |
ce8eb6c4 | 3668 | struct kmem_cache_node *n; |
cafeb02e | 3669 | struct array_cache *new_shared; |
c8522a3a | 3670 | struct alien_cache **new_alien = NULL; |
e498be7d | 3671 | |
9c09a95c | 3672 | for_each_online_node(node) { |
cafeb02e | 3673 | |
b455def2 L |
3674 | if (use_alien_caches) { |
3675 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3676 | if (!new_alien) | |
3677 | goto fail; | |
3678 | } | |
cafeb02e | 3679 | |
63109846 ED |
3680 | new_shared = NULL; |
3681 | if (cachep->shared) { | |
3682 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3683 | cachep->shared*cachep->batchcount, |
83b519e8 | 3684 | 0xbaadf00d, gfp); |
63109846 ED |
3685 | if (!new_shared) { |
3686 | free_alien_cache(new_alien); | |
3687 | goto fail; | |
3688 | } | |
0718dc2a | 3689 | } |
cafeb02e | 3690 | |
18bf8541 | 3691 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3692 | if (n) { |
3693 | struct array_cache *shared = n->shared; | |
97654dfa | 3694 | LIST_HEAD(list); |
cafeb02e | 3695 | |
ce8eb6c4 | 3696 | spin_lock_irq(&n->list_lock); |
e498be7d | 3697 | |
cafeb02e | 3698 | if (shared) |
0718dc2a | 3699 | free_block(cachep, shared->entry, |
97654dfa | 3700 | shared->avail, node, &list); |
e498be7d | 3701 | |
ce8eb6c4 CL |
3702 | n->shared = new_shared; |
3703 | if (!n->alien) { | |
3704 | n->alien = new_alien; | |
e498be7d CL |
3705 | new_alien = NULL; |
3706 | } | |
ce8eb6c4 | 3707 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3708 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3709 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3710 | slabs_destroy(cachep, &list); |
cafeb02e | 3711 | kfree(shared); |
e498be7d CL |
3712 | free_alien_cache(new_alien); |
3713 | continue; | |
3714 | } | |
ce8eb6c4 CL |
3715 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3716 | if (!n) { | |
0718dc2a CL |
3717 | free_alien_cache(new_alien); |
3718 | kfree(new_shared); | |
e498be7d | 3719 | goto fail; |
0718dc2a | 3720 | } |
e498be7d | 3721 | |
ce8eb6c4 | 3722 | kmem_cache_node_init(n); |
5f0985bb JZ |
3723 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3724 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3725 | n->shared = new_shared; |
3726 | n->alien = new_alien; | |
3727 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3728 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3729 | cachep->node[node] = n; |
e498be7d | 3730 | } |
cafeb02e | 3731 | return 0; |
0718dc2a | 3732 | |
a737b3e2 | 3733 | fail: |
3b0efdfa | 3734 | if (!cachep->list.next) { |
0718dc2a CL |
3735 | /* Cache is not active yet. Roll back what we did */ |
3736 | node--; | |
3737 | while (node >= 0) { | |
18bf8541 CL |
3738 | n = get_node(cachep, node); |
3739 | if (n) { | |
ce8eb6c4 CL |
3740 | kfree(n->shared); |
3741 | free_alien_cache(n->alien); | |
3742 | kfree(n); | |
6a67368c | 3743 | cachep->node[node] = NULL; |
0718dc2a CL |
3744 | } |
3745 | node--; | |
3746 | } | |
3747 | } | |
cafeb02e | 3748 | return -ENOMEM; |
e498be7d CL |
3749 | } |
3750 | ||
18004c5d | 3751 | /* Always called with the slab_mutex held */ |
943a451a | 3752 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3753 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3754 | { |
bf0dea23 JK |
3755 | struct array_cache __percpu *cpu_cache, *prev; |
3756 | int cpu; | |
1da177e4 | 3757 | |
bf0dea23 JK |
3758 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3759 | if (!cpu_cache) | |
d2e7b7d0 SS |
3760 | return -ENOMEM; |
3761 | ||
bf0dea23 JK |
3762 | prev = cachep->cpu_cache; |
3763 | cachep->cpu_cache = cpu_cache; | |
3764 | kick_all_cpus_sync(); | |
e498be7d | 3765 | |
1da177e4 | 3766 | check_irq_on(); |
1da177e4 LT |
3767 | cachep->batchcount = batchcount; |
3768 | cachep->limit = limit; | |
e498be7d | 3769 | cachep->shared = shared; |
1da177e4 | 3770 | |
bf0dea23 JK |
3771 | if (!prev) |
3772 | goto alloc_node; | |
3773 | ||
3774 | for_each_online_cpu(cpu) { | |
97654dfa | 3775 | LIST_HEAD(list); |
18bf8541 CL |
3776 | int node; |
3777 | struct kmem_cache_node *n; | |
bf0dea23 | 3778 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3779 | |
bf0dea23 | 3780 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3781 | n = get_node(cachep, node); |
3782 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3783 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3784 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3785 | slabs_destroy(cachep, &list); |
1da177e4 | 3786 | } |
bf0dea23 JK |
3787 | free_percpu(prev); |
3788 | ||
3789 | alloc_node: | |
5f0985bb | 3790 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3791 | } |
3792 | ||
943a451a GC |
3793 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3794 | int batchcount, int shared, gfp_t gfp) | |
3795 | { | |
3796 | int ret; | |
426589f5 | 3797 | struct kmem_cache *c; |
943a451a GC |
3798 | |
3799 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3800 | ||
3801 | if (slab_state < FULL) | |
3802 | return ret; | |
3803 | ||
3804 | if ((ret < 0) || !is_root_cache(cachep)) | |
3805 | return ret; | |
3806 | ||
426589f5 VD |
3807 | lockdep_assert_held(&slab_mutex); |
3808 | for_each_memcg_cache(c, cachep) { | |
3809 | /* return value determined by the root cache only */ | |
3810 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3811 | } |
3812 | ||
3813 | return ret; | |
3814 | } | |
3815 | ||
18004c5d | 3816 | /* Called with slab_mutex held always */ |
83b519e8 | 3817 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3818 | { |
3819 | int err; | |
943a451a GC |
3820 | int limit = 0; |
3821 | int shared = 0; | |
3822 | int batchcount = 0; | |
3823 | ||
3824 | if (!is_root_cache(cachep)) { | |
3825 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3826 | limit = root->limit; | |
3827 | shared = root->shared; | |
3828 | batchcount = root->batchcount; | |
3829 | } | |
1da177e4 | 3830 | |
943a451a GC |
3831 | if (limit && shared && batchcount) |
3832 | goto skip_setup; | |
a737b3e2 AM |
3833 | /* |
3834 | * The head array serves three purposes: | |
1da177e4 LT |
3835 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3836 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3837 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3838 | * bufctl chains: array operations are cheaper. |
3839 | * The numbers are guessed, we should auto-tune as described by | |
3840 | * Bonwick. | |
3841 | */ | |
3b0efdfa | 3842 | if (cachep->size > 131072) |
1da177e4 | 3843 | limit = 1; |
3b0efdfa | 3844 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3845 | limit = 8; |
3b0efdfa | 3846 | else if (cachep->size > 1024) |
1da177e4 | 3847 | limit = 24; |
3b0efdfa | 3848 | else if (cachep->size > 256) |
1da177e4 LT |
3849 | limit = 54; |
3850 | else | |
3851 | limit = 120; | |
3852 | ||
a737b3e2 AM |
3853 | /* |
3854 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3855 | * allocation behaviour: Most allocs on one cpu, most free operations |
3856 | * on another cpu. For these cases, an efficient object passing between | |
3857 | * cpus is necessary. This is provided by a shared array. The array | |
3858 | * replaces Bonwick's magazine layer. | |
3859 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3860 | * to a larger limit. Thus disabled by default. | |
3861 | */ | |
3862 | shared = 0; | |
3b0efdfa | 3863 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3864 | shared = 8; |
1da177e4 LT |
3865 | |
3866 | #if DEBUG | |
a737b3e2 AM |
3867 | /* |
3868 | * With debugging enabled, large batchcount lead to excessively long | |
3869 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3870 | */ |
3871 | if (limit > 32) | |
3872 | limit = 32; | |
3873 | #endif | |
943a451a GC |
3874 | batchcount = (limit + 1) / 2; |
3875 | skip_setup: | |
3876 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 | 3877 | if (err) |
1170532b | 3878 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3879 | cachep->name, -err); |
2ed3a4ef | 3880 | return err; |
1da177e4 LT |
3881 | } |
3882 | ||
1b55253a | 3883 | /* |
ce8eb6c4 CL |
3884 | * Drain an array if it contains any elements taking the node lock only if |
3885 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3886 | * if drain_array() is used on the shared array. |
1b55253a | 3887 | */ |
ce8eb6c4 | 3888 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 3889 | struct array_cache *ac, int node) |
1da177e4 | 3890 | { |
97654dfa | 3891 | LIST_HEAD(list); |
18726ca8 JK |
3892 | |
3893 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
3894 | check_mutex_acquired(); | |
1da177e4 | 3895 | |
1b55253a CL |
3896 | if (!ac || !ac->avail) |
3897 | return; | |
18726ca8 JK |
3898 | |
3899 | if (ac->touched) { | |
1da177e4 | 3900 | ac->touched = 0; |
18726ca8 | 3901 | return; |
1da177e4 | 3902 | } |
18726ca8 JK |
3903 | |
3904 | spin_lock_irq(&n->list_lock); | |
3905 | drain_array_locked(cachep, ac, node, false, &list); | |
3906 | spin_unlock_irq(&n->list_lock); | |
3907 | ||
3908 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
3909 | } |
3910 | ||
3911 | /** | |
3912 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3913 | * @w: work descriptor |
1da177e4 LT |
3914 | * |
3915 | * Called from workqueue/eventd every few seconds. | |
3916 | * Purpose: | |
3917 | * - clear the per-cpu caches for this CPU. | |
3918 | * - return freeable pages to the main free memory pool. | |
3919 | * | |
a737b3e2 AM |
3920 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3921 | * again on the next iteration. | |
1da177e4 | 3922 | */ |
7c5cae36 | 3923 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3924 | { |
7a7c381d | 3925 | struct kmem_cache *searchp; |
ce8eb6c4 | 3926 | struct kmem_cache_node *n; |
7d6e6d09 | 3927 | int node = numa_mem_id(); |
bf6aede7 | 3928 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3929 | |
18004c5d | 3930 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3931 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3932 | goto out; |
1da177e4 | 3933 | |
18004c5d | 3934 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3935 | check_irq_on(); |
3936 | ||
35386e3b | 3937 | /* |
ce8eb6c4 | 3938 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3939 | * have established with reasonable certainty that |
3940 | * we can do some work if the lock was obtained. | |
3941 | */ | |
18bf8541 | 3942 | n = get_node(searchp, node); |
35386e3b | 3943 | |
ce8eb6c4 | 3944 | reap_alien(searchp, n); |
1da177e4 | 3945 | |
18726ca8 | 3946 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 3947 | |
35386e3b CL |
3948 | /* |
3949 | * These are racy checks but it does not matter | |
3950 | * if we skip one check or scan twice. | |
3951 | */ | |
ce8eb6c4 | 3952 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3953 | goto next; |
1da177e4 | 3954 | |
5f0985bb | 3955 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3956 | |
18726ca8 | 3957 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 3958 | |
ce8eb6c4 CL |
3959 | if (n->free_touched) |
3960 | n->free_touched = 0; | |
ed11d9eb CL |
3961 | else { |
3962 | int freed; | |
1da177e4 | 3963 | |
ce8eb6c4 | 3964 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3965 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3966 | STATS_ADD_REAPED(searchp, freed); | |
3967 | } | |
35386e3b | 3968 | next: |
1da177e4 LT |
3969 | cond_resched(); |
3970 | } | |
3971 | check_irq_on(); | |
18004c5d | 3972 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3973 | next_reap_node(); |
7c5cae36 | 3974 | out: |
a737b3e2 | 3975 | /* Set up the next iteration */ |
5f0985bb | 3976 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3977 | } |
3978 | ||
158a9624 | 3979 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3980 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3981 | { |
8456a648 | 3982 | struct page *page; |
b28a02de PE |
3983 | unsigned long active_objs; |
3984 | unsigned long num_objs; | |
3985 | unsigned long active_slabs = 0; | |
3986 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3987 | const char *name; |
1da177e4 | 3988 | char *error = NULL; |
e498be7d | 3989 | int node; |
ce8eb6c4 | 3990 | struct kmem_cache_node *n; |
1da177e4 | 3991 | |
1da177e4 LT |
3992 | active_objs = 0; |
3993 | num_slabs = 0; | |
18bf8541 | 3994 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 3995 | |
ca3b9b91 | 3996 | check_irq_on(); |
ce8eb6c4 | 3997 | spin_lock_irq(&n->list_lock); |
e498be7d | 3998 | |
8456a648 JK |
3999 | list_for_each_entry(page, &n->slabs_full, lru) { |
4000 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4001 | error = "slabs_full accounting error"; |
4002 | active_objs += cachep->num; | |
4003 | active_slabs++; | |
4004 | } | |
8456a648 JK |
4005 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4006 | if (page->active == cachep->num && !error) | |
106a74e1 | 4007 | error = "slabs_partial accounting error"; |
8456a648 | 4008 | if (!page->active && !error) |
106a74e1 | 4009 | error = "slabs_partial accounting error"; |
8456a648 | 4010 | active_objs += page->active; |
e498be7d CL |
4011 | active_slabs++; |
4012 | } | |
8456a648 JK |
4013 | list_for_each_entry(page, &n->slabs_free, lru) { |
4014 | if (page->active && !error) | |
106a74e1 | 4015 | error = "slabs_free accounting error"; |
e498be7d CL |
4016 | num_slabs++; |
4017 | } | |
ce8eb6c4 CL |
4018 | free_objects += n->free_objects; |
4019 | if (n->shared) | |
4020 | shared_avail += n->shared->avail; | |
e498be7d | 4021 | |
ce8eb6c4 | 4022 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4023 | } |
b28a02de PE |
4024 | num_slabs += active_slabs; |
4025 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4026 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4027 | error = "free_objects accounting error"; |
4028 | ||
b28a02de | 4029 | name = cachep->name; |
1da177e4 | 4030 | if (error) |
1170532b | 4031 | pr_err("slab: cache %s error: %s\n", name, error); |
1da177e4 | 4032 | |
0d7561c6 GC |
4033 | sinfo->active_objs = active_objs; |
4034 | sinfo->num_objs = num_objs; | |
4035 | sinfo->active_slabs = active_slabs; | |
4036 | sinfo->num_slabs = num_slabs; | |
4037 | sinfo->shared_avail = shared_avail; | |
4038 | sinfo->limit = cachep->limit; | |
4039 | sinfo->batchcount = cachep->batchcount; | |
4040 | sinfo->shared = cachep->shared; | |
4041 | sinfo->objects_per_slab = cachep->num; | |
4042 | sinfo->cache_order = cachep->gfporder; | |
4043 | } | |
4044 | ||
4045 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4046 | { | |
1da177e4 | 4047 | #if STATS |
ce8eb6c4 | 4048 | { /* node stats */ |
1da177e4 LT |
4049 | unsigned long high = cachep->high_mark; |
4050 | unsigned long allocs = cachep->num_allocations; | |
4051 | unsigned long grown = cachep->grown; | |
4052 | unsigned long reaped = cachep->reaped; | |
4053 | unsigned long errors = cachep->errors; | |
4054 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4055 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4056 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4057 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4058 | |
756a025f | 4059 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4060 | allocs, high, grown, |
4061 | reaped, errors, max_freeable, node_allocs, | |
4062 | node_frees, overflows); | |
1da177e4 LT |
4063 | } |
4064 | /* cpu stats */ | |
4065 | { | |
4066 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4067 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4068 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4069 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4070 | ||
4071 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4072 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4073 | } |
4074 | #endif | |
1da177e4 LT |
4075 | } |
4076 | ||
1da177e4 LT |
4077 | #define MAX_SLABINFO_WRITE 128 |
4078 | /** | |
4079 | * slabinfo_write - Tuning for the slab allocator | |
4080 | * @file: unused | |
4081 | * @buffer: user buffer | |
4082 | * @count: data length | |
4083 | * @ppos: unused | |
4084 | */ | |
b7454ad3 | 4085 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4086 | size_t count, loff_t *ppos) |
1da177e4 | 4087 | { |
b28a02de | 4088 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4089 | int limit, batchcount, shared, res; |
7a7c381d | 4090 | struct kmem_cache *cachep; |
b28a02de | 4091 | |
1da177e4 LT |
4092 | if (count > MAX_SLABINFO_WRITE) |
4093 | return -EINVAL; | |
4094 | if (copy_from_user(&kbuf, buffer, count)) | |
4095 | return -EFAULT; | |
b28a02de | 4096 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4097 | |
4098 | tmp = strchr(kbuf, ' '); | |
4099 | if (!tmp) | |
4100 | return -EINVAL; | |
4101 | *tmp = '\0'; | |
4102 | tmp++; | |
4103 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4104 | return -EINVAL; | |
4105 | ||
4106 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4107 | mutex_lock(&slab_mutex); |
1da177e4 | 4108 | res = -EINVAL; |
18004c5d | 4109 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4110 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4111 | if (limit < 1 || batchcount < 1 || |
4112 | batchcount > limit || shared < 0) { | |
e498be7d | 4113 | res = 0; |
1da177e4 | 4114 | } else { |
e498be7d | 4115 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4116 | batchcount, shared, |
4117 | GFP_KERNEL); | |
1da177e4 LT |
4118 | } |
4119 | break; | |
4120 | } | |
4121 | } | |
18004c5d | 4122 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4123 | if (res >= 0) |
4124 | res = count; | |
4125 | return res; | |
4126 | } | |
871751e2 AV |
4127 | |
4128 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4129 | ||
871751e2 AV |
4130 | static inline int add_caller(unsigned long *n, unsigned long v) |
4131 | { | |
4132 | unsigned long *p; | |
4133 | int l; | |
4134 | if (!v) | |
4135 | return 1; | |
4136 | l = n[1]; | |
4137 | p = n + 2; | |
4138 | while (l) { | |
4139 | int i = l/2; | |
4140 | unsigned long *q = p + 2 * i; | |
4141 | if (*q == v) { | |
4142 | q[1]++; | |
4143 | return 1; | |
4144 | } | |
4145 | if (*q > v) { | |
4146 | l = i; | |
4147 | } else { | |
4148 | p = q + 2; | |
4149 | l -= i + 1; | |
4150 | } | |
4151 | } | |
4152 | if (++n[1] == n[0]) | |
4153 | return 0; | |
4154 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4155 | p[0] = v; | |
4156 | p[1] = 1; | |
4157 | return 1; | |
4158 | } | |
4159 | ||
8456a648 JK |
4160 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4161 | struct page *page) | |
871751e2 AV |
4162 | { |
4163 | void *p; | |
d31676df JK |
4164 | int i, j; |
4165 | unsigned long v; | |
b1cb0982 | 4166 | |
871751e2 AV |
4167 | if (n[0] == n[1]) |
4168 | return; | |
8456a648 | 4169 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4170 | bool active = true; |
4171 | ||
4172 | for (j = page->active; j < c->num; j++) { | |
4173 | if (get_free_obj(page, j) == i) { | |
4174 | active = false; | |
4175 | break; | |
4176 | } | |
4177 | } | |
4178 | ||
4179 | if (!active) | |
871751e2 | 4180 | continue; |
b1cb0982 | 4181 | |
d31676df JK |
4182 | /* |
4183 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4184 | * mapping is established when actual object allocation and | |
4185 | * we could mistakenly access the unmapped object in the cpu | |
4186 | * cache. | |
4187 | */ | |
4188 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4189 | continue; | |
4190 | ||
4191 | if (!add_caller(n, v)) | |
871751e2 AV |
4192 | return; |
4193 | } | |
4194 | } | |
4195 | ||
4196 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4197 | { | |
4198 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4199 | unsigned long offset, size; |
9281acea | 4200 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4201 | |
a5c43dae | 4202 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4203 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4204 | if (modname[0]) |
871751e2 AV |
4205 | seq_printf(m, " [%s]", modname); |
4206 | return; | |
4207 | } | |
4208 | #endif | |
4209 | seq_printf(m, "%p", (void *)address); | |
4210 | } | |
4211 | ||
4212 | static int leaks_show(struct seq_file *m, void *p) | |
4213 | { | |
0672aa7c | 4214 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4215 | struct page *page; |
ce8eb6c4 | 4216 | struct kmem_cache_node *n; |
871751e2 | 4217 | const char *name; |
db845067 | 4218 | unsigned long *x = m->private; |
871751e2 AV |
4219 | int node; |
4220 | int i; | |
4221 | ||
4222 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4223 | return 0; | |
4224 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4225 | return 0; | |
4226 | ||
d31676df JK |
4227 | /* |
4228 | * Set store_user_clean and start to grab stored user information | |
4229 | * for all objects on this cache. If some alloc/free requests comes | |
4230 | * during the processing, information would be wrong so restart | |
4231 | * whole processing. | |
4232 | */ | |
4233 | do { | |
4234 | set_store_user_clean(cachep); | |
4235 | drain_cpu_caches(cachep); | |
4236 | ||
4237 | x[1] = 0; | |
871751e2 | 4238 | |
d31676df | 4239 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4240 | |
d31676df JK |
4241 | check_irq_on(); |
4242 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4243 | |
d31676df JK |
4244 | list_for_each_entry(page, &n->slabs_full, lru) |
4245 | handle_slab(x, cachep, page); | |
4246 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4247 | handle_slab(x, cachep, page); | |
4248 | spin_unlock_irq(&n->list_lock); | |
4249 | } | |
4250 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4251 | |
871751e2 | 4252 | name = cachep->name; |
db845067 | 4253 | if (x[0] == x[1]) { |
871751e2 | 4254 | /* Increase the buffer size */ |
18004c5d | 4255 | mutex_unlock(&slab_mutex); |
db845067 | 4256 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4257 | if (!m->private) { |
4258 | /* Too bad, we are really out */ | |
db845067 | 4259 | m->private = x; |
18004c5d | 4260 | mutex_lock(&slab_mutex); |
871751e2 AV |
4261 | return -ENOMEM; |
4262 | } | |
db845067 CL |
4263 | *(unsigned long *)m->private = x[0] * 2; |
4264 | kfree(x); | |
18004c5d | 4265 | mutex_lock(&slab_mutex); |
871751e2 AV |
4266 | /* Now make sure this entry will be retried */ |
4267 | m->count = m->size; | |
4268 | return 0; | |
4269 | } | |
db845067 CL |
4270 | for (i = 0; i < x[1]; i++) { |
4271 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4272 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4273 | seq_putc(m, '\n'); |
4274 | } | |
d2e7b7d0 | 4275 | |
871751e2 AV |
4276 | return 0; |
4277 | } | |
4278 | ||
a0ec95a8 | 4279 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4280 | .start = slab_start, |
276a2439 WL |
4281 | .next = slab_next, |
4282 | .stop = slab_stop, | |
871751e2 AV |
4283 | .show = leaks_show, |
4284 | }; | |
a0ec95a8 AD |
4285 | |
4286 | static int slabstats_open(struct inode *inode, struct file *file) | |
4287 | { | |
b208ce32 RJ |
4288 | unsigned long *n; |
4289 | ||
4290 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4291 | if (!n) | |
4292 | return -ENOMEM; | |
4293 | ||
4294 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4295 | ||
4296 | return 0; | |
a0ec95a8 AD |
4297 | } |
4298 | ||
4299 | static const struct file_operations proc_slabstats_operations = { | |
4300 | .open = slabstats_open, | |
4301 | .read = seq_read, | |
4302 | .llseek = seq_lseek, | |
4303 | .release = seq_release_private, | |
4304 | }; | |
4305 | #endif | |
4306 | ||
4307 | static int __init slab_proc_init(void) | |
4308 | { | |
4309 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4310 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4311 | #endif |
a0ec95a8 AD |
4312 | return 0; |
4313 | } | |
4314 | module_init(slab_proc_init); | |
1da177e4 LT |
4315 | #endif |
4316 | ||
00e145b6 MS |
4317 | /** |
4318 | * ksize - get the actual amount of memory allocated for a given object | |
4319 | * @objp: Pointer to the object | |
4320 | * | |
4321 | * kmalloc may internally round up allocations and return more memory | |
4322 | * than requested. ksize() can be used to determine the actual amount of | |
4323 | * memory allocated. The caller may use this additional memory, even though | |
4324 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4325 | * The caller must guarantee that objp points to a valid object previously | |
4326 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4327 | * must not be freed during the duration of the call. | |
4328 | */ | |
fd76bab2 | 4329 | size_t ksize(const void *objp) |
1da177e4 | 4330 | { |
7ed2f9e6 AP |
4331 | size_t size; |
4332 | ||
ef8b4520 CL |
4333 | BUG_ON(!objp); |
4334 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4335 | return 0; |
1da177e4 | 4336 | |
7ed2f9e6 AP |
4337 | size = virt_to_cache(objp)->object_size; |
4338 | /* We assume that ksize callers could use the whole allocated area, | |
4339 | * so we need to unpoison this area. | |
4340 | */ | |
505f5dcb | 4341 | kasan_krealloc(objp, size, GFP_NOWAIT); |
7ed2f9e6 AP |
4342 | |
4343 | return size; | |
1da177e4 | 4344 | } |
b1aabecd | 4345 | EXPORT_SYMBOL(ksize); |