]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/slab.c | |
4 | * Written by Mark Hemment, 1996/97. | |
5 | * (markhe@nextd.demon.co.uk) | |
6 | * | |
7 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
8 | * | |
9 | * Major cleanup, different bufctl logic, per-cpu arrays | |
10 | * (c) 2000 Manfred Spraul | |
11 | * | |
12 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
13 | * (c) 2002 Manfred Spraul | |
14 | * | |
15 | * An implementation of the Slab Allocator as described in outline in; | |
16 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
17 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
18 | * or with a little more detail in; | |
19 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
20 | * Jeff Bonwick (Sun Microsystems). | |
21 | * Presented at: USENIX Summer 1994 Technical Conference | |
22 | * | |
23 | * The memory is organized in caches, one cache for each object type. | |
24 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
25 | * Each cache consists out of many slabs (they are small (usually one | |
26 | * page long) and always contiguous), and each slab contains multiple | |
27 | * initialized objects. | |
28 | * | |
29 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 30 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
31 | * kmem_cache_free. |
32 | * | |
33 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
34 | * normal). If you need a special memory type, then must create a new | |
35 | * cache for that memory type. | |
36 | * | |
37 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
38 | * full slabs with 0 free objects | |
39 | * partial slabs | |
40 | * empty slabs with no allocated objects | |
41 | * | |
42 | * If partial slabs exist, then new allocations come from these slabs, | |
43 | * otherwise from empty slabs or new slabs are allocated. | |
44 | * | |
45 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
46 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
47 | * | |
48 | * Each cache has a short per-cpu head array, most allocs | |
49 | * and frees go into that array, and if that array overflows, then 1/2 | |
50 | * of the entries in the array are given back into the global cache. | |
51 | * The head array is strictly LIFO and should improve the cache hit rates. | |
52 | * On SMP, it additionally reduces the spinlock operations. | |
53 | * | |
a737b3e2 | 54 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
55 | * it's changed with a smp_call_function(). |
56 | * | |
57 | * SMP synchronization: | |
58 | * constructors and destructors are called without any locking. | |
343e0d7a | 59 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
60 | * are accessed without any locking. |
61 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
62 | * and local interrupts are disabled so slab code is preempt-safe. | |
63 | * The non-constant members are protected with a per-cache irq spinlock. | |
64 | * | |
65 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
66 | * in 2000 - many ideas in the current implementation are derived from | |
67 | * his patch. | |
68 | * | |
69 | * Further notes from the original documentation: | |
70 | * | |
71 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 72 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
73 | * The sem is only needed when accessing/extending the cache-chain, which |
74 | * can never happen inside an interrupt (kmem_cache_create(), | |
75 | * kmem_cache_shrink() and kmem_cache_reap()). | |
76 | * | |
77 | * At present, each engine can be growing a cache. This should be blocked. | |
78 | * | |
e498be7d CL |
79 | * 15 March 2005. NUMA slab allocator. |
80 | * Shai Fultheim <shai@scalex86.org>. | |
81 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
82 | * Alok N Kataria <alokk@calsoftinc.com> | |
83 | * Christoph Lameter <christoph@lameter.com> | |
84 | * | |
85 | * Modified the slab allocator to be node aware on NUMA systems. | |
86 | * Each node has its own list of partial, free and full slabs. | |
87 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 LT |
90 | #include <linux/slab.h> |
91 | #include <linux/mm.h> | |
c9cf5528 | 92 | #include <linux/poison.h> |
1da177e4 LT |
93 | #include <linux/swap.h> |
94 | #include <linux/cache.h> | |
95 | #include <linux/interrupt.h> | |
96 | #include <linux/init.h> | |
97 | #include <linux/compiler.h> | |
101a5001 | 98 | #include <linux/cpuset.h> |
a0ec95a8 | 99 | #include <linux/proc_fs.h> |
1da177e4 LT |
100 | #include <linux/seq_file.h> |
101 | #include <linux/notifier.h> | |
102 | #include <linux/kallsyms.h> | |
103 | #include <linux/cpu.h> | |
104 | #include <linux/sysctl.h> | |
105 | #include <linux/module.h> | |
106 | #include <linux/rcupdate.h> | |
543537bd | 107 | #include <linux/string.h> |
138ae663 | 108 | #include <linux/uaccess.h> |
e498be7d | 109 | #include <linux/nodemask.h> |
d5cff635 | 110 | #include <linux/kmemleak.h> |
dc85da15 | 111 | #include <linux/mempolicy.h> |
fc0abb14 | 112 | #include <linux/mutex.h> |
8a8b6502 | 113 | #include <linux/fault-inject.h> |
e7eebaf6 | 114 | #include <linux/rtmutex.h> |
6a2d7a95 | 115 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 116 | #include <linux/debugobjects.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
3f8c2452 | 119 | #include <linux/sched/task_stack.h> |
1da177e4 | 120 | |
381760ea MG |
121 | #include <net/sock.h> |
122 | ||
1da177e4 LT |
123 | #include <asm/cacheflush.h> |
124 | #include <asm/tlbflush.h> | |
125 | #include <asm/page.h> | |
126 | ||
4dee6b64 SR |
127 | #include <trace/events/kmem.h> |
128 | ||
072bb0aa MG |
129 | #include "internal.h" |
130 | ||
b9ce5ef4 GC |
131 | #include "slab.h" |
132 | ||
1da177e4 | 133 | /* |
50953fe9 | 134 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
135 | * 0 for faster, smaller code (especially in the critical paths). |
136 | * | |
137 | * STATS - 1 to collect stats for /proc/slabinfo. | |
138 | * 0 for faster, smaller code (especially in the critical paths). | |
139 | * | |
140 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
141 | */ | |
142 | ||
143 | #ifdef CONFIG_DEBUG_SLAB | |
144 | #define DEBUG 1 | |
145 | #define STATS 1 | |
146 | #define FORCED_DEBUG 1 | |
147 | #else | |
148 | #define DEBUG 0 | |
149 | #define STATS 0 | |
150 | #define FORCED_DEBUG 0 | |
151 | #endif | |
152 | ||
1da177e4 LT |
153 | /* Shouldn't this be in a header file somewhere? */ |
154 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 155 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 156 | |
1da177e4 LT |
157 | #ifndef ARCH_KMALLOC_FLAGS |
158 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
159 | #endif | |
160 | ||
f315e3fa JK |
161 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
162 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
163 | ||
164 | #if FREELIST_BYTE_INDEX | |
165 | typedef unsigned char freelist_idx_t; | |
166 | #else | |
167 | typedef unsigned short freelist_idx_t; | |
168 | #endif | |
169 | ||
30321c7b | 170 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 171 | |
1da177e4 LT |
172 | /* |
173 | * struct array_cache | |
174 | * | |
1da177e4 LT |
175 | * Purpose: |
176 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
177 | * - reduce the number of linked list operations | |
178 | * - reduce spinlock operations | |
179 | * | |
180 | * The limit is stored in the per-cpu structure to reduce the data cache | |
181 | * footprint. | |
182 | * | |
183 | */ | |
184 | struct array_cache { | |
185 | unsigned int avail; | |
186 | unsigned int limit; | |
187 | unsigned int batchcount; | |
188 | unsigned int touched; | |
bda5b655 | 189 | void *entry[]; /* |
a737b3e2 AM |
190 | * Must have this definition in here for the proper |
191 | * alignment of array_cache. Also simplifies accessing | |
192 | * the entries. | |
a737b3e2 | 193 | */ |
1da177e4 LT |
194 | }; |
195 | ||
c8522a3a JK |
196 | struct alien_cache { |
197 | spinlock_t lock; | |
198 | struct array_cache ac; | |
199 | }; | |
200 | ||
e498be7d CL |
201 | /* |
202 | * Need this for bootstrapping a per node allocator. | |
203 | */ | |
bf0dea23 | 204 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 205 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 206 | #define CACHE_CACHE 0 |
bf0dea23 | 207 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 208 | |
ed11d9eb | 209 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 210 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 211 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
212 | int node, struct list_head *list); |
213 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 214 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 215 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 216 | |
76b342bd JK |
217 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
218 | void **list); | |
219 | static inline void fixup_slab_list(struct kmem_cache *cachep, | |
220 | struct kmem_cache_node *n, struct page *page, | |
221 | void **list); | |
e0a42726 IM |
222 | static int slab_early_init = 1; |
223 | ||
ce8eb6c4 | 224 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 225 | |
ce8eb6c4 | 226 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
227 | { |
228 | INIT_LIST_HEAD(&parent->slabs_full); | |
229 | INIT_LIST_HEAD(&parent->slabs_partial); | |
230 | INIT_LIST_HEAD(&parent->slabs_free); | |
bf00bd34 | 231 | parent->total_slabs = 0; |
f728b0a5 | 232 | parent->free_slabs = 0; |
e498be7d CL |
233 | parent->shared = NULL; |
234 | parent->alien = NULL; | |
2e1217cf | 235 | parent->colour_next = 0; |
e498be7d CL |
236 | spin_lock_init(&parent->list_lock); |
237 | parent->free_objects = 0; | |
238 | parent->free_touched = 0; | |
239 | } | |
240 | ||
a737b3e2 AM |
241 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
242 | do { \ | |
243 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 244 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
245 | } while (0) |
246 | ||
a737b3e2 AM |
247 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
248 | do { \ | |
e498be7d CL |
249 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
250 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
251 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
252 | } while (0) | |
1da177e4 | 253 | |
4fd0b46e AD |
254 | #define CFLGS_OBJFREELIST_SLAB ((slab_flags_t __force)0x40000000U) |
255 | #define CFLGS_OFF_SLAB ((slab_flags_t __force)0x80000000U) | |
b03a017b | 256 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
257 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
258 | ||
259 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
260 | /* |
261 | * Optimization question: fewer reaps means less probability for unnessary | |
262 | * cpucache drain/refill cycles. | |
1da177e4 | 263 | * |
dc6f3f27 | 264 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
265 | * which could lock up otherwise freeable slabs. |
266 | */ | |
5f0985bb JZ |
267 | #define REAPTIMEOUT_AC (2*HZ) |
268 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
269 | |
270 | #if STATS | |
271 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
272 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
273 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
274 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 275 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
276 | #define STATS_SET_HIGH(x) \ |
277 | do { \ | |
278 | if ((x)->num_active > (x)->high_mark) \ | |
279 | (x)->high_mark = (x)->num_active; \ | |
280 | } while (0) | |
1da177e4 LT |
281 | #define STATS_INC_ERR(x) ((x)->errors++) |
282 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 283 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 284 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
285 | #define STATS_SET_FREEABLE(x, i) \ |
286 | do { \ | |
287 | if ((x)->max_freeable < i) \ | |
288 | (x)->max_freeable = i; \ | |
289 | } while (0) | |
1da177e4 LT |
290 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
291 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
292 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
293 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
294 | #else | |
295 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
296 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
297 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
298 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 299 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
300 | #define STATS_SET_HIGH(x) do { } while (0) |
301 | #define STATS_INC_ERR(x) do { } while (0) | |
302 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 303 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 304 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 305 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
306 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
307 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
308 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
309 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
310 | #endif | |
311 | ||
312 | #if DEBUG | |
1da177e4 | 313 | |
a737b3e2 AM |
314 | /* |
315 | * memory layout of objects: | |
1da177e4 | 316 | * 0 : objp |
3dafccf2 | 317 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
318 | * the end of an object is aligned with the end of the real |
319 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 320 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 321 | * redzone word. |
3dafccf2 | 322 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
323 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
324 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 325 | * [BYTES_PER_WORD long] |
1da177e4 | 326 | */ |
343e0d7a | 327 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 328 | { |
3dafccf2 | 329 | return cachep->obj_offset; |
1da177e4 LT |
330 | } |
331 | ||
b46b8f19 | 332 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
333 | { |
334 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
335 | return (unsigned long long*) (objp + obj_offset(cachep) - |
336 | sizeof(unsigned long long)); | |
1da177e4 LT |
337 | } |
338 | ||
b46b8f19 | 339 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
340 | { |
341 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
342 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 343 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 344 | sizeof(unsigned long long) - |
87a927c7 | 345 | REDZONE_ALIGN); |
3b0efdfa | 346 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 347 | sizeof(unsigned long long)); |
1da177e4 LT |
348 | } |
349 | ||
343e0d7a | 350 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
351 | { |
352 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 353 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
354 | } |
355 | ||
356 | #else | |
357 | ||
3dafccf2 | 358 | #define obj_offset(x) 0 |
b46b8f19 DW |
359 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
360 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
361 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
362 | ||
363 | #endif | |
364 | ||
03787301 JK |
365 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
366 | ||
d31676df | 367 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 368 | { |
d31676df JK |
369 | return atomic_read(&cachep->store_user_clean) == 1; |
370 | } | |
03787301 | 371 | |
d31676df JK |
372 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
373 | { | |
374 | atomic_set(&cachep->store_user_clean, 1); | |
375 | } | |
03787301 | 376 | |
d31676df JK |
377 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
378 | { | |
379 | if (is_store_user_clean(cachep)) | |
380 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
381 | } |
382 | ||
383 | #else | |
d31676df | 384 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
385 | |
386 | #endif | |
387 | ||
1da177e4 | 388 | /* |
3df1cccd DR |
389 | * Do not go above this order unless 0 objects fit into the slab or |
390 | * overridden on the command line. | |
1da177e4 | 391 | */ |
543585cc DR |
392 | #define SLAB_MAX_ORDER_HI 1 |
393 | #define SLAB_MAX_ORDER_LO 0 | |
394 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 395 | static bool slab_max_order_set __initdata; |
1da177e4 | 396 | |
6ed5eb22 PE |
397 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
398 | { | |
b49af68f | 399 | struct page *page = virt_to_head_page(obj); |
35026088 | 400 | return page->slab_cache; |
6ed5eb22 PE |
401 | } |
402 | ||
8456a648 | 403 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
404 | unsigned int idx) |
405 | { | |
8456a648 | 406 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
407 | } |
408 | ||
6a2d7a95 | 409 | /* |
3b0efdfa CL |
410 | * We want to avoid an expensive divide : (offset / cache->size) |
411 | * Using the fact that size is a constant for a particular cache, | |
412 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
413 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
414 | */ | |
415 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 416 | const struct page *page, void *obj) |
8fea4e96 | 417 | { |
8456a648 | 418 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 419 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
420 | } |
421 | ||
6fb92430 | 422 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 423 | /* internal cache of cache description objs */ |
9b030cb8 | 424 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
425 | .batchcount = 1, |
426 | .limit = BOOT_CPUCACHE_ENTRIES, | |
427 | .shared = 1, | |
3b0efdfa | 428 | .size = sizeof(struct kmem_cache), |
b28a02de | 429 | .name = "kmem_cache", |
1da177e4 LT |
430 | }; |
431 | ||
1871e52c | 432 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 433 | |
343e0d7a | 434 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 435 | { |
bf0dea23 | 436 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
437 | } |
438 | ||
a737b3e2 AM |
439 | /* |
440 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
441 | */ | |
70f75067 | 442 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
d50112ed | 443 | slab_flags_t flags, size_t *left_over) |
fbaccacf | 444 | { |
70f75067 | 445 | unsigned int num; |
fbaccacf | 446 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 447 | |
fbaccacf SR |
448 | /* |
449 | * The slab management structure can be either off the slab or | |
450 | * on it. For the latter case, the memory allocated for a | |
451 | * slab is used for: | |
452 | * | |
fbaccacf | 453 | * - @buffer_size bytes for each object |
2e6b3602 JK |
454 | * - One freelist_idx_t for each object |
455 | * | |
456 | * We don't need to consider alignment of freelist because | |
457 | * freelist will be at the end of slab page. The objects will be | |
458 | * at the correct alignment. | |
fbaccacf SR |
459 | * |
460 | * If the slab management structure is off the slab, then the | |
461 | * alignment will already be calculated into the size. Because | |
462 | * the slabs are all pages aligned, the objects will be at the | |
463 | * correct alignment when allocated. | |
464 | */ | |
b03a017b | 465 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 466 | num = slab_size / buffer_size; |
2e6b3602 | 467 | *left_over = slab_size % buffer_size; |
fbaccacf | 468 | } else { |
70f75067 | 469 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
470 | *left_over = slab_size % |
471 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 472 | } |
70f75067 JK |
473 | |
474 | return num; | |
1da177e4 LT |
475 | } |
476 | ||
f28510d3 | 477 | #if DEBUG |
d40cee24 | 478 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 479 | |
a737b3e2 AM |
480 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
481 | char *msg) | |
1da177e4 | 482 | { |
1170532b | 483 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 484 | function, cachep->name, msg); |
1da177e4 | 485 | dump_stack(); |
373d4d09 | 486 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 487 | } |
f28510d3 | 488 | #endif |
1da177e4 | 489 | |
3395ee05 PM |
490 | /* |
491 | * By default on NUMA we use alien caches to stage the freeing of | |
492 | * objects allocated from other nodes. This causes massive memory | |
493 | * inefficiencies when using fake NUMA setup to split memory into a | |
494 | * large number of small nodes, so it can be disabled on the command | |
495 | * line | |
496 | */ | |
497 | ||
498 | static int use_alien_caches __read_mostly = 1; | |
499 | static int __init noaliencache_setup(char *s) | |
500 | { | |
501 | use_alien_caches = 0; | |
502 | return 1; | |
503 | } | |
504 | __setup("noaliencache", noaliencache_setup); | |
505 | ||
3df1cccd DR |
506 | static int __init slab_max_order_setup(char *str) |
507 | { | |
508 | get_option(&str, &slab_max_order); | |
509 | slab_max_order = slab_max_order < 0 ? 0 : | |
510 | min(slab_max_order, MAX_ORDER - 1); | |
511 | slab_max_order_set = true; | |
512 | ||
513 | return 1; | |
514 | } | |
515 | __setup("slab_max_order=", slab_max_order_setup); | |
516 | ||
8fce4d8e CL |
517 | #ifdef CONFIG_NUMA |
518 | /* | |
519 | * Special reaping functions for NUMA systems called from cache_reap(). | |
520 | * These take care of doing round robin flushing of alien caches (containing | |
521 | * objects freed on different nodes from which they were allocated) and the | |
522 | * flushing of remote pcps by calling drain_node_pages. | |
523 | */ | |
1871e52c | 524 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
525 | |
526 | static void init_reap_node(int cpu) | |
527 | { | |
0edaf86c AM |
528 | per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu), |
529 | node_online_map); | |
8fce4d8e CL |
530 | } |
531 | ||
532 | static void next_reap_node(void) | |
533 | { | |
909ea964 | 534 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 535 | |
0edaf86c | 536 | node = next_node_in(node, node_online_map); |
909ea964 | 537 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
538 | } |
539 | ||
540 | #else | |
541 | #define init_reap_node(cpu) do { } while (0) | |
542 | #define next_reap_node(void) do { } while (0) | |
543 | #endif | |
544 | ||
1da177e4 LT |
545 | /* |
546 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
547 | * via the workqueue/eventd. | |
548 | * Add the CPU number into the expiration time to minimize the possibility of | |
549 | * the CPUs getting into lockstep and contending for the global cache chain | |
550 | * lock. | |
551 | */ | |
0db0628d | 552 | static void start_cpu_timer(int cpu) |
1da177e4 | 553 | { |
1871e52c | 554 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 | 555 | |
eac0337a | 556 | if (reap_work->work.func == NULL) { |
8fce4d8e | 557 | init_reap_node(cpu); |
203b42f7 | 558 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
559 | schedule_delayed_work_on(cpu, reap_work, |
560 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
561 | } |
562 | } | |
563 | ||
1fe00d50 | 564 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 565 | { |
d5cff635 CM |
566 | /* |
567 | * The array_cache structures contain pointers to free object. | |
25985edc | 568 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
569 | * cache the pointers are not cleared and they could be counted as |
570 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
571 | * not scan such objects. | |
572 | */ | |
1fe00d50 JK |
573 | kmemleak_no_scan(ac); |
574 | if (ac) { | |
575 | ac->avail = 0; | |
576 | ac->limit = limit; | |
577 | ac->batchcount = batch; | |
578 | ac->touched = 0; | |
1da177e4 | 579 | } |
1fe00d50 JK |
580 | } |
581 | ||
582 | static struct array_cache *alloc_arraycache(int node, int entries, | |
583 | int batchcount, gfp_t gfp) | |
584 | { | |
5e804789 | 585 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
586 | struct array_cache *ac = NULL; |
587 | ||
588 | ac = kmalloc_node(memsize, gfp, node); | |
589 | init_arraycache(ac, entries, batchcount); | |
590 | return ac; | |
1da177e4 LT |
591 | } |
592 | ||
f68f8ddd JK |
593 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
594 | struct page *page, void *objp) | |
072bb0aa | 595 | { |
f68f8ddd JK |
596 | struct kmem_cache_node *n; |
597 | int page_node; | |
598 | LIST_HEAD(list); | |
072bb0aa | 599 | |
f68f8ddd JK |
600 | page_node = page_to_nid(page); |
601 | n = get_node(cachep, page_node); | |
381760ea | 602 | |
f68f8ddd JK |
603 | spin_lock(&n->list_lock); |
604 | free_block(cachep, &objp, 1, page_node, &list); | |
605 | spin_unlock(&n->list_lock); | |
381760ea | 606 | |
f68f8ddd | 607 | slabs_destroy(cachep, &list); |
072bb0aa MG |
608 | } |
609 | ||
3ded175a CL |
610 | /* |
611 | * Transfer objects in one arraycache to another. | |
612 | * Locking must be handled by the caller. | |
613 | * | |
614 | * Return the number of entries transferred. | |
615 | */ | |
616 | static int transfer_objects(struct array_cache *to, | |
617 | struct array_cache *from, unsigned int max) | |
618 | { | |
619 | /* Figure out how many entries to transfer */ | |
732eacc0 | 620 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
621 | |
622 | if (!nr) | |
623 | return 0; | |
624 | ||
625 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
626 | sizeof(void *) *nr); | |
627 | ||
628 | from->avail -= nr; | |
629 | to->avail += nr; | |
3ded175a CL |
630 | return nr; |
631 | } | |
632 | ||
765c4507 CL |
633 | #ifndef CONFIG_NUMA |
634 | ||
635 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 636 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 637 | |
c8522a3a JK |
638 | static inline struct alien_cache **alloc_alien_cache(int node, |
639 | int limit, gfp_t gfp) | |
765c4507 | 640 | { |
8888177e | 641 | return NULL; |
765c4507 CL |
642 | } |
643 | ||
c8522a3a | 644 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
645 | { |
646 | } | |
647 | ||
648 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
649 | { | |
650 | return 0; | |
651 | } | |
652 | ||
653 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
654 | gfp_t flags) | |
655 | { | |
656 | return NULL; | |
657 | } | |
658 | ||
8b98c169 | 659 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
660 | gfp_t flags, int nodeid) |
661 | { | |
662 | return NULL; | |
663 | } | |
664 | ||
4167e9b2 DR |
665 | static inline gfp_t gfp_exact_node(gfp_t flags) |
666 | { | |
444eb2a4 | 667 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
668 | } |
669 | ||
765c4507 CL |
670 | #else /* CONFIG_NUMA */ |
671 | ||
8b98c169 | 672 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 673 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 674 | |
c8522a3a JK |
675 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
676 | int batch, gfp_t gfp) | |
677 | { | |
5e804789 | 678 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
679 | struct alien_cache *alc = NULL; |
680 | ||
681 | alc = kmalloc_node(memsize, gfp, node); | |
a527aabb CL |
682 | if (alc) { |
683 | init_arraycache(&alc->ac, entries, batch); | |
684 | spin_lock_init(&alc->lock); | |
685 | } | |
c8522a3a JK |
686 | return alc; |
687 | } | |
688 | ||
689 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 690 | { |
c8522a3a | 691 | struct alien_cache **alc_ptr; |
5e804789 | 692 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
693 | int i; |
694 | ||
695 | if (limit > 1) | |
696 | limit = 12; | |
c8522a3a JK |
697 | alc_ptr = kzalloc_node(memsize, gfp, node); |
698 | if (!alc_ptr) | |
699 | return NULL; | |
700 | ||
701 | for_each_node(i) { | |
702 | if (i == node || !node_online(i)) | |
703 | continue; | |
704 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
705 | if (!alc_ptr[i]) { | |
706 | for (i--; i >= 0; i--) | |
707 | kfree(alc_ptr[i]); | |
708 | kfree(alc_ptr); | |
709 | return NULL; | |
e498be7d CL |
710 | } |
711 | } | |
c8522a3a | 712 | return alc_ptr; |
e498be7d CL |
713 | } |
714 | ||
c8522a3a | 715 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
716 | { |
717 | int i; | |
718 | ||
c8522a3a | 719 | if (!alc_ptr) |
e498be7d | 720 | return; |
e498be7d | 721 | for_each_node(i) |
c8522a3a JK |
722 | kfree(alc_ptr[i]); |
723 | kfree(alc_ptr); | |
e498be7d CL |
724 | } |
725 | ||
343e0d7a | 726 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
727 | struct array_cache *ac, int node, |
728 | struct list_head *list) | |
e498be7d | 729 | { |
18bf8541 | 730 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
731 | |
732 | if (ac->avail) { | |
ce8eb6c4 | 733 | spin_lock(&n->list_lock); |
e00946fe CL |
734 | /* |
735 | * Stuff objects into the remote nodes shared array first. | |
736 | * That way we could avoid the overhead of putting the objects | |
737 | * into the free lists and getting them back later. | |
738 | */ | |
ce8eb6c4 CL |
739 | if (n->shared) |
740 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 741 | |
833b706c | 742 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 743 | ac->avail = 0; |
ce8eb6c4 | 744 | spin_unlock(&n->list_lock); |
e498be7d CL |
745 | } |
746 | } | |
747 | ||
8fce4d8e CL |
748 | /* |
749 | * Called from cache_reap() to regularly drain alien caches round robin. | |
750 | */ | |
ce8eb6c4 | 751 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 752 | { |
909ea964 | 753 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 754 | |
ce8eb6c4 | 755 | if (n->alien) { |
c8522a3a JK |
756 | struct alien_cache *alc = n->alien[node]; |
757 | struct array_cache *ac; | |
758 | ||
759 | if (alc) { | |
760 | ac = &alc->ac; | |
49dfc304 | 761 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
762 | LIST_HEAD(list); |
763 | ||
764 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 765 | spin_unlock_irq(&alc->lock); |
833b706c | 766 | slabs_destroy(cachep, &list); |
c8522a3a | 767 | } |
8fce4d8e CL |
768 | } |
769 | } | |
770 | } | |
771 | ||
a737b3e2 | 772 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 773 | struct alien_cache **alien) |
e498be7d | 774 | { |
b28a02de | 775 | int i = 0; |
c8522a3a | 776 | struct alien_cache *alc; |
e498be7d CL |
777 | struct array_cache *ac; |
778 | unsigned long flags; | |
779 | ||
780 | for_each_online_node(i) { | |
c8522a3a JK |
781 | alc = alien[i]; |
782 | if (alc) { | |
833b706c JK |
783 | LIST_HEAD(list); |
784 | ||
c8522a3a | 785 | ac = &alc->ac; |
49dfc304 | 786 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 787 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 788 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 789 | slabs_destroy(cachep, &list); |
e498be7d CL |
790 | } |
791 | } | |
792 | } | |
729bd0b7 | 793 | |
25c4f304 JK |
794 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
795 | int node, int page_node) | |
729bd0b7 | 796 | { |
ce8eb6c4 | 797 | struct kmem_cache_node *n; |
c8522a3a JK |
798 | struct alien_cache *alien = NULL; |
799 | struct array_cache *ac; | |
97654dfa | 800 | LIST_HEAD(list); |
1ca4cb24 | 801 | |
18bf8541 | 802 | n = get_node(cachep, node); |
729bd0b7 | 803 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
804 | if (n->alien && n->alien[page_node]) { |
805 | alien = n->alien[page_node]; | |
c8522a3a | 806 | ac = &alien->ac; |
49dfc304 | 807 | spin_lock(&alien->lock); |
c8522a3a | 808 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 809 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 810 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 811 | } |
f68f8ddd | 812 | ac->entry[ac->avail++] = objp; |
49dfc304 | 813 | spin_unlock(&alien->lock); |
833b706c | 814 | slabs_destroy(cachep, &list); |
729bd0b7 | 815 | } else { |
25c4f304 | 816 | n = get_node(cachep, page_node); |
18bf8541 | 817 | spin_lock(&n->list_lock); |
25c4f304 | 818 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 819 | spin_unlock(&n->list_lock); |
97654dfa | 820 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
821 | } |
822 | return 1; | |
823 | } | |
25c4f304 JK |
824 | |
825 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
826 | { | |
827 | int page_node = page_to_nid(virt_to_page(objp)); | |
828 | int node = numa_mem_id(); | |
829 | /* | |
830 | * Make sure we are not freeing a object from another node to the array | |
831 | * cache on this cpu. | |
832 | */ | |
833 | if (likely(node == page_node)) | |
834 | return 0; | |
835 | ||
836 | return __cache_free_alien(cachep, objp, node, page_node); | |
837 | } | |
4167e9b2 DR |
838 | |
839 | /* | |
444eb2a4 MG |
840 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
841 | * warn about failures. | |
4167e9b2 DR |
842 | */ |
843 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
844 | { | |
444eb2a4 | 845 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 846 | } |
e498be7d CL |
847 | #endif |
848 | ||
ded0ecf6 JK |
849 | static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp) |
850 | { | |
851 | struct kmem_cache_node *n; | |
852 | ||
853 | /* | |
854 | * Set up the kmem_cache_node for cpu before we can | |
855 | * begin anything. Make sure some other cpu on this | |
856 | * node has not already allocated this | |
857 | */ | |
858 | n = get_node(cachep, node); | |
859 | if (n) { | |
860 | spin_lock_irq(&n->list_lock); | |
861 | n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount + | |
862 | cachep->num; | |
863 | spin_unlock_irq(&n->list_lock); | |
864 | ||
865 | return 0; | |
866 | } | |
867 | ||
868 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); | |
869 | if (!n) | |
870 | return -ENOMEM; | |
871 | ||
872 | kmem_cache_node_init(n); | |
873 | n->next_reap = jiffies + REAPTIMEOUT_NODE + | |
874 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
875 | ||
876 | n->free_limit = | |
877 | (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num; | |
878 | ||
879 | /* | |
880 | * The kmem_cache_nodes don't come and go as CPUs | |
881 | * come and go. slab_mutex is sufficient | |
882 | * protection here. | |
883 | */ | |
884 | cachep->node[node] = n; | |
885 | ||
886 | return 0; | |
887 | } | |
888 | ||
6731d4f1 | 889 | #if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP) |
8f9f8d9e | 890 | /* |
6a67368c | 891 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 892 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 893 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 894 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
895 | * already in use. |
896 | * | |
18004c5d | 897 | * Must hold slab_mutex. |
8f9f8d9e | 898 | */ |
6a67368c | 899 | static int init_cache_node_node(int node) |
8f9f8d9e | 900 | { |
ded0ecf6 | 901 | int ret; |
8f9f8d9e | 902 | struct kmem_cache *cachep; |
8f9f8d9e | 903 | |
18004c5d | 904 | list_for_each_entry(cachep, &slab_caches, list) { |
ded0ecf6 JK |
905 | ret = init_cache_node(cachep, node, GFP_KERNEL); |
906 | if (ret) | |
907 | return ret; | |
8f9f8d9e | 908 | } |
ded0ecf6 | 909 | |
8f9f8d9e DR |
910 | return 0; |
911 | } | |
6731d4f1 | 912 | #endif |
8f9f8d9e | 913 | |
c3d332b6 JK |
914 | static int setup_kmem_cache_node(struct kmem_cache *cachep, |
915 | int node, gfp_t gfp, bool force_change) | |
916 | { | |
917 | int ret = -ENOMEM; | |
918 | struct kmem_cache_node *n; | |
919 | struct array_cache *old_shared = NULL; | |
920 | struct array_cache *new_shared = NULL; | |
921 | struct alien_cache **new_alien = NULL; | |
922 | LIST_HEAD(list); | |
923 | ||
924 | if (use_alien_caches) { | |
925 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
926 | if (!new_alien) | |
927 | goto fail; | |
928 | } | |
929 | ||
930 | if (cachep->shared) { | |
931 | new_shared = alloc_arraycache(node, | |
932 | cachep->shared * cachep->batchcount, 0xbaadf00d, gfp); | |
933 | if (!new_shared) | |
934 | goto fail; | |
935 | } | |
936 | ||
937 | ret = init_cache_node(cachep, node, gfp); | |
938 | if (ret) | |
939 | goto fail; | |
940 | ||
941 | n = get_node(cachep, node); | |
942 | spin_lock_irq(&n->list_lock); | |
943 | if (n->shared && force_change) { | |
944 | free_block(cachep, n->shared->entry, | |
945 | n->shared->avail, node, &list); | |
946 | n->shared->avail = 0; | |
947 | } | |
948 | ||
949 | if (!n->shared || force_change) { | |
950 | old_shared = n->shared; | |
951 | n->shared = new_shared; | |
952 | new_shared = NULL; | |
953 | } | |
954 | ||
955 | if (!n->alien) { | |
956 | n->alien = new_alien; | |
957 | new_alien = NULL; | |
958 | } | |
959 | ||
960 | spin_unlock_irq(&n->list_lock); | |
961 | slabs_destroy(cachep, &list); | |
962 | ||
801faf0d JK |
963 | /* |
964 | * To protect lockless access to n->shared during irq disabled context. | |
965 | * If n->shared isn't NULL in irq disabled context, accessing to it is | |
966 | * guaranteed to be valid until irq is re-enabled, because it will be | |
967 | * freed after synchronize_sched(). | |
968 | */ | |
86d9f485 | 969 | if (old_shared && force_change) |
801faf0d JK |
970 | synchronize_sched(); |
971 | ||
c3d332b6 JK |
972 | fail: |
973 | kfree(old_shared); | |
974 | kfree(new_shared); | |
975 | free_alien_cache(new_alien); | |
976 | ||
977 | return ret; | |
978 | } | |
979 | ||
6731d4f1 SAS |
980 | #ifdef CONFIG_SMP |
981 | ||
0db0628d | 982 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
983 | { |
984 | struct kmem_cache *cachep; | |
ce8eb6c4 | 985 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 986 | int node = cpu_to_mem(cpu); |
a70f7302 | 987 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 988 | |
18004c5d | 989 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
990 | struct array_cache *nc; |
991 | struct array_cache *shared; | |
c8522a3a | 992 | struct alien_cache **alien; |
97654dfa | 993 | LIST_HEAD(list); |
fbf1e473 | 994 | |
18bf8541 | 995 | n = get_node(cachep, node); |
ce8eb6c4 | 996 | if (!n) |
bf0dea23 | 997 | continue; |
fbf1e473 | 998 | |
ce8eb6c4 | 999 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1000 | |
ce8eb6c4 CL |
1001 | /* Free limit for this kmem_cache_node */ |
1002 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1003 | |
1004 | /* cpu is dead; no one can alloc from it. */ | |
1005 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1006 | if (nc) { | |
97654dfa | 1007 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1008 | nc->avail = 0; |
1009 | } | |
fbf1e473 | 1010 | |
58463c1f | 1011 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1012 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1013 | goto free_slab; |
fbf1e473 AM |
1014 | } |
1015 | ||
ce8eb6c4 | 1016 | shared = n->shared; |
fbf1e473 AM |
1017 | if (shared) { |
1018 | free_block(cachep, shared->entry, | |
97654dfa | 1019 | shared->avail, node, &list); |
ce8eb6c4 | 1020 | n->shared = NULL; |
fbf1e473 AM |
1021 | } |
1022 | ||
ce8eb6c4 CL |
1023 | alien = n->alien; |
1024 | n->alien = NULL; | |
fbf1e473 | 1025 | |
ce8eb6c4 | 1026 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1027 | |
1028 | kfree(shared); | |
1029 | if (alien) { | |
1030 | drain_alien_cache(cachep, alien); | |
1031 | free_alien_cache(alien); | |
1032 | } | |
bf0dea23 JK |
1033 | |
1034 | free_slab: | |
97654dfa | 1035 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1036 | } |
1037 | /* | |
1038 | * In the previous loop, all the objects were freed to | |
1039 | * the respective cache's slabs, now we can go ahead and | |
1040 | * shrink each nodelist to its limit. | |
1041 | */ | |
18004c5d | 1042 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1043 | n = get_node(cachep, node); |
ce8eb6c4 | 1044 | if (!n) |
fbf1e473 | 1045 | continue; |
a5aa63a5 | 1046 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
1047 | } |
1048 | } | |
1049 | ||
0db0628d | 1050 | static int cpuup_prepare(long cpu) |
1da177e4 | 1051 | { |
343e0d7a | 1052 | struct kmem_cache *cachep; |
7d6e6d09 | 1053 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1054 | int err; |
1da177e4 | 1055 | |
fbf1e473 AM |
1056 | /* |
1057 | * We need to do this right in the beginning since | |
1058 | * alloc_arraycache's are going to use this list. | |
1059 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1060 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1061 | */ |
6a67368c | 1062 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1063 | if (err < 0) |
1064 | goto bad; | |
fbf1e473 AM |
1065 | |
1066 | /* | |
1067 | * Now we can go ahead with allocating the shared arrays and | |
1068 | * array caches | |
1069 | */ | |
18004c5d | 1070 | list_for_each_entry(cachep, &slab_caches, list) { |
c3d332b6 JK |
1071 | err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false); |
1072 | if (err) | |
1073 | goto bad; | |
fbf1e473 | 1074 | } |
ce79ddc8 | 1075 | |
fbf1e473 AM |
1076 | return 0; |
1077 | bad: | |
12d00f6a | 1078 | cpuup_canceled(cpu); |
fbf1e473 AM |
1079 | return -ENOMEM; |
1080 | } | |
1081 | ||
6731d4f1 | 1082 | int slab_prepare_cpu(unsigned int cpu) |
fbf1e473 | 1083 | { |
6731d4f1 | 1084 | int err; |
fbf1e473 | 1085 | |
6731d4f1 SAS |
1086 | mutex_lock(&slab_mutex); |
1087 | err = cpuup_prepare(cpu); | |
1088 | mutex_unlock(&slab_mutex); | |
1089 | return err; | |
1090 | } | |
1091 | ||
1092 | /* | |
1093 | * This is called for a failed online attempt and for a successful | |
1094 | * offline. | |
1095 | * | |
1096 | * Even if all the cpus of a node are down, we don't free the | |
1097 | * kmem_list3 of any cache. This to avoid a race between cpu_down, and | |
1098 | * a kmalloc allocation from another cpu for memory from the node of | |
1099 | * the cpu going down. The list3 structure is usually allocated from | |
1100 | * kmem_cache_create() and gets destroyed at kmem_cache_destroy(). | |
1101 | */ | |
1102 | int slab_dead_cpu(unsigned int cpu) | |
1103 | { | |
1104 | mutex_lock(&slab_mutex); | |
1105 | cpuup_canceled(cpu); | |
1106 | mutex_unlock(&slab_mutex); | |
1107 | return 0; | |
1108 | } | |
8f5be20b | 1109 | #endif |
6731d4f1 SAS |
1110 | |
1111 | static int slab_online_cpu(unsigned int cpu) | |
1112 | { | |
1113 | start_cpu_timer(cpu); | |
1114 | return 0; | |
1da177e4 LT |
1115 | } |
1116 | ||
6731d4f1 SAS |
1117 | static int slab_offline_cpu(unsigned int cpu) |
1118 | { | |
1119 | /* | |
1120 | * Shutdown cache reaper. Note that the slab_mutex is held so | |
1121 | * that if cache_reap() is invoked it cannot do anything | |
1122 | * expensive but will only modify reap_work and reschedule the | |
1123 | * timer. | |
1124 | */ | |
1125 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); | |
1126 | /* Now the cache_reaper is guaranteed to be not running. */ | |
1127 | per_cpu(slab_reap_work, cpu).work.func = NULL; | |
1128 | return 0; | |
1129 | } | |
1da177e4 | 1130 | |
8f9f8d9e DR |
1131 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1132 | /* | |
1133 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1134 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1135 | * removed. | |
1136 | * | |
18004c5d | 1137 | * Must hold slab_mutex. |
8f9f8d9e | 1138 | */ |
6a67368c | 1139 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1140 | { |
1141 | struct kmem_cache *cachep; | |
1142 | int ret = 0; | |
1143 | ||
18004c5d | 1144 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1145 | struct kmem_cache_node *n; |
8f9f8d9e | 1146 | |
18bf8541 | 1147 | n = get_node(cachep, node); |
ce8eb6c4 | 1148 | if (!n) |
8f9f8d9e DR |
1149 | continue; |
1150 | ||
a5aa63a5 | 1151 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1152 | |
ce8eb6c4 CL |
1153 | if (!list_empty(&n->slabs_full) || |
1154 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1155 | ret = -EBUSY; |
1156 | break; | |
1157 | } | |
1158 | } | |
1159 | return ret; | |
1160 | } | |
1161 | ||
1162 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1163 | unsigned long action, void *arg) | |
1164 | { | |
1165 | struct memory_notify *mnb = arg; | |
1166 | int ret = 0; | |
1167 | int nid; | |
1168 | ||
1169 | nid = mnb->status_change_nid; | |
1170 | if (nid < 0) | |
1171 | goto out; | |
1172 | ||
1173 | switch (action) { | |
1174 | case MEM_GOING_ONLINE: | |
18004c5d | 1175 | mutex_lock(&slab_mutex); |
6a67368c | 1176 | ret = init_cache_node_node(nid); |
18004c5d | 1177 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1178 | break; |
1179 | case MEM_GOING_OFFLINE: | |
18004c5d | 1180 | mutex_lock(&slab_mutex); |
6a67368c | 1181 | ret = drain_cache_node_node(nid); |
18004c5d | 1182 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1183 | break; |
1184 | case MEM_ONLINE: | |
1185 | case MEM_OFFLINE: | |
1186 | case MEM_CANCEL_ONLINE: | |
1187 | case MEM_CANCEL_OFFLINE: | |
1188 | break; | |
1189 | } | |
1190 | out: | |
5fda1bd5 | 1191 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1192 | } |
1193 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1194 | ||
e498be7d | 1195 | /* |
ce8eb6c4 | 1196 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1197 | */ |
6744f087 | 1198 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1199 | int nodeid) |
e498be7d | 1200 | { |
6744f087 | 1201 | struct kmem_cache_node *ptr; |
e498be7d | 1202 | |
6744f087 | 1203 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1204 | BUG_ON(!ptr); |
1205 | ||
6744f087 | 1206 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1207 | /* |
1208 | * Do not assume that spinlocks can be initialized via memcpy: | |
1209 | */ | |
1210 | spin_lock_init(&ptr->list_lock); | |
1211 | ||
e498be7d | 1212 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1213 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1214 | } |
1215 | ||
556a169d | 1216 | /* |
ce8eb6c4 CL |
1217 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1218 | * size of kmem_cache_node. | |
556a169d | 1219 | */ |
ce8eb6c4 | 1220 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1221 | { |
1222 | int node; | |
1223 | ||
1224 | for_each_online_node(node) { | |
ce8eb6c4 | 1225 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1226 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1227 | REAPTIMEOUT_NODE + |
1228 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1229 | } |
1230 | } | |
1231 | ||
a737b3e2 AM |
1232 | /* |
1233 | * Initialisation. Called after the page allocator have been initialised and | |
1234 | * before smp_init(). | |
1da177e4 LT |
1235 | */ |
1236 | void __init kmem_cache_init(void) | |
1237 | { | |
e498be7d CL |
1238 | int i; |
1239 | ||
68126702 JK |
1240 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1241 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1242 | kmem_cache = &kmem_cache_boot; |
1243 | ||
8888177e | 1244 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1245 | use_alien_caches = 0; |
1246 | ||
3c583465 | 1247 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1248 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1249 | |
1da177e4 LT |
1250 | /* |
1251 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1252 | * page orders on machines with more than 32MB of memory if |
1253 | * not overridden on the command line. | |
1da177e4 | 1254 | */ |
3df1cccd | 1255 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1256 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1257 | |
1da177e4 LT |
1258 | /* Bootstrap is tricky, because several objects are allocated |
1259 | * from caches that do not exist yet: | |
9b030cb8 CL |
1260 | * 1) initialize the kmem_cache cache: it contains the struct |
1261 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1262 | * kmem_cache is statically allocated. | |
e498be7d | 1263 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1264 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1265 | * array at the end of the bootstrap. |
1da177e4 | 1266 | * 2) Create the first kmalloc cache. |
343e0d7a | 1267 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1268 | * An __init data area is used for the head array. |
1269 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1270 | * head arrays. | |
9b030cb8 | 1271 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1272 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1273 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1274 | * the other cache's with kmalloc allocated memory. |
1275 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1276 | */ |
1277 | ||
9b030cb8 | 1278 | /* 1) create the kmem_cache */ |
1da177e4 | 1279 | |
8da3430d | 1280 | /* |
b56efcf0 | 1281 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1282 | */ |
2f9baa9f | 1283 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1284 | offsetof(struct kmem_cache, node) + |
6744f087 | 1285 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1286 | SLAB_HWCACHE_ALIGN); |
1287 | list_add(&kmem_cache->list, &slab_caches); | |
eb220838 | 1288 | memcg_link_cache(kmem_cache); |
bf0dea23 | 1289 | slab_state = PARTIAL; |
1da177e4 | 1290 | |
a737b3e2 | 1291 | /* |
bf0dea23 JK |
1292 | * Initialize the caches that provide memory for the kmem_cache_node |
1293 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1294 | */ |
af3b5f87 VB |
1295 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache( |
1296 | kmalloc_info[INDEX_NODE].name, | |
ce8eb6c4 | 1297 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1298 | slab_state = PARTIAL_NODE; |
34cc6990 | 1299 | setup_kmalloc_cache_index_table(); |
e498be7d | 1300 | |
e0a42726 IM |
1301 | slab_early_init = 0; |
1302 | ||
ce8eb6c4 | 1303 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1304 | { |
1ca4cb24 PE |
1305 | int nid; |
1306 | ||
9c09a95c | 1307 | for_each_online_node(nid) { |
ce8eb6c4 | 1308 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1309 | |
bf0dea23 | 1310 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1311 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1312 | } |
1313 | } | |
1da177e4 | 1314 | |
f97d5f63 | 1315 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1316 | } |
1317 | ||
1318 | void __init kmem_cache_init_late(void) | |
1319 | { | |
1320 | struct kmem_cache *cachep; | |
1321 | ||
97d06609 | 1322 | slab_state = UP; |
52cef189 | 1323 | |
8429db5c | 1324 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1325 | mutex_lock(&slab_mutex); |
1326 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1327 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1328 | BUG(); | |
18004c5d | 1329 | mutex_unlock(&slab_mutex); |
056c6241 | 1330 | |
97d06609 CL |
1331 | /* Done! */ |
1332 | slab_state = FULL; | |
1333 | ||
8f9f8d9e DR |
1334 | #ifdef CONFIG_NUMA |
1335 | /* | |
1336 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1337 | * node. |
8f9f8d9e DR |
1338 | */ |
1339 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1340 | #endif | |
1341 | ||
a737b3e2 AM |
1342 | /* |
1343 | * The reap timers are started later, with a module init call: That part | |
1344 | * of the kernel is not yet operational. | |
1da177e4 LT |
1345 | */ |
1346 | } | |
1347 | ||
1348 | static int __init cpucache_init(void) | |
1349 | { | |
6731d4f1 | 1350 | int ret; |
1da177e4 | 1351 | |
a737b3e2 AM |
1352 | /* |
1353 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1354 | */ |
6731d4f1 SAS |
1355 | ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online", |
1356 | slab_online_cpu, slab_offline_cpu); | |
1357 | WARN_ON(ret < 0); | |
a164f896 GC |
1358 | |
1359 | /* Done! */ | |
97d06609 | 1360 | slab_state = FULL; |
1da177e4 LT |
1361 | return 0; |
1362 | } | |
1da177e4 LT |
1363 | __initcall(cpucache_init); |
1364 | ||
8bdec192 RA |
1365 | static noinline void |
1366 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1367 | { | |
9a02d699 | 1368 | #if DEBUG |
ce8eb6c4 | 1369 | struct kmem_cache_node *n; |
8bdec192 RA |
1370 | unsigned long flags; |
1371 | int node; | |
9a02d699 DR |
1372 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1373 | DEFAULT_RATELIMIT_BURST); | |
1374 | ||
1375 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1376 | return; | |
8bdec192 | 1377 | |
5b3810e5 VB |
1378 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1379 | nodeid, gfpflags, &gfpflags); | |
1380 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1381 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1382 | |
18bf8541 | 1383 | for_each_kmem_cache_node(cachep, node, n) { |
bf00bd34 | 1384 | unsigned long total_slabs, free_slabs, free_objs; |
8bdec192 | 1385 | |
ce8eb6c4 | 1386 | spin_lock_irqsave(&n->list_lock, flags); |
bf00bd34 DR |
1387 | total_slabs = n->total_slabs; |
1388 | free_slabs = n->free_slabs; | |
1389 | free_objs = n->free_objects; | |
ce8eb6c4 | 1390 | spin_unlock_irqrestore(&n->list_lock, flags); |
8bdec192 | 1391 | |
bf00bd34 DR |
1392 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n", |
1393 | node, total_slabs - free_slabs, total_slabs, | |
1394 | (total_slabs * cachep->num) - free_objs, | |
1395 | total_slabs * cachep->num); | |
8bdec192 | 1396 | } |
9a02d699 | 1397 | #endif |
8bdec192 RA |
1398 | } |
1399 | ||
1da177e4 | 1400 | /* |
8a7d9b43 WSH |
1401 | * Interface to system's page allocator. No need to hold the |
1402 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1403 | * |
1404 | * If we requested dmaable memory, we will get it. Even if we | |
1405 | * did not request dmaable memory, we might get it, but that | |
1406 | * would be relatively rare and ignorable. | |
1407 | */ | |
0c3aa83e JK |
1408 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1409 | int nodeid) | |
1da177e4 LT |
1410 | { |
1411 | struct page *page; | |
e1b6aa6f | 1412 | int nr_pages; |
765c4507 | 1413 | |
a618e89f | 1414 | flags |= cachep->allocflags; |
e1b6aa6f | 1415 | |
75f296d9 | 1416 | page = __alloc_pages_node(nodeid, flags, cachep->gfporder); |
8bdec192 | 1417 | if (!page) { |
9a02d699 | 1418 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1419 | return NULL; |
8bdec192 | 1420 | } |
1da177e4 | 1421 | |
f3ccb2c4 VD |
1422 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1423 | __free_pages(page, cachep->gfporder); | |
1424 | return NULL; | |
1425 | } | |
1426 | ||
e1b6aa6f | 1427 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1428 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
7779f212 | 1429 | mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, nr_pages); |
972d1a7b | 1430 | else |
7779f212 | 1431 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, nr_pages); |
f68f8ddd | 1432 | |
a57a4988 | 1433 | __SetPageSlab(page); |
f68f8ddd JK |
1434 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1435 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1436 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1437 | |
0c3aa83e | 1438 | return page; |
1da177e4 LT |
1439 | } |
1440 | ||
1441 | /* | |
1442 | * Interface to system's page release. | |
1443 | */ | |
0c3aa83e | 1444 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1445 | { |
27ee57c9 VD |
1446 | int order = cachep->gfporder; |
1447 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1448 | |
972d1a7b | 1449 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
7779f212 | 1450 | mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, -nr_freed); |
972d1a7b | 1451 | else |
7779f212 | 1452 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, -nr_freed); |
73293c2f | 1453 | |
a57a4988 | 1454 | BUG_ON(!PageSlab(page)); |
73293c2f | 1455 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1456 | __ClearPageSlab(page); |
8456a648 JK |
1457 | page_mapcount_reset(page); |
1458 | page->mapping = NULL; | |
1f458cbf | 1459 | |
1da177e4 LT |
1460 | if (current->reclaim_state) |
1461 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1462 | memcg_uncharge_slab(page, order, cachep); |
1463 | __free_pages(page, order); | |
1da177e4 LT |
1464 | } |
1465 | ||
1466 | static void kmem_rcu_free(struct rcu_head *head) | |
1467 | { | |
68126702 JK |
1468 | struct kmem_cache *cachep; |
1469 | struct page *page; | |
1da177e4 | 1470 | |
68126702 JK |
1471 | page = container_of(head, struct page, rcu_head); |
1472 | cachep = page->slab_cache; | |
1473 | ||
1474 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1475 | } |
1476 | ||
1477 | #if DEBUG | |
40b44137 JK |
1478 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1479 | { | |
1480 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1481 | (cachep->size % PAGE_SIZE) == 0) | |
1482 | return true; | |
1483 | ||
1484 | return false; | |
1485 | } | |
1da177e4 LT |
1486 | |
1487 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1488 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1489 | unsigned long caller) |
1da177e4 | 1490 | { |
8c138bc0 | 1491 | int size = cachep->object_size; |
1da177e4 | 1492 | |
3dafccf2 | 1493 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1494 | |
b28a02de | 1495 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1496 | return; |
1497 | ||
b28a02de PE |
1498 | *addr++ = 0x12345678; |
1499 | *addr++ = caller; | |
1500 | *addr++ = smp_processor_id(); | |
1501 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1502 | { |
1503 | unsigned long *sptr = &caller; | |
1504 | unsigned long svalue; | |
1505 | ||
1506 | while (!kstack_end(sptr)) { | |
1507 | svalue = *sptr++; | |
1508 | if (kernel_text_address(svalue)) { | |
b28a02de | 1509 | *addr++ = svalue; |
1da177e4 LT |
1510 | size -= sizeof(unsigned long); |
1511 | if (size <= sizeof(unsigned long)) | |
1512 | break; | |
1513 | } | |
1514 | } | |
1515 | ||
1516 | } | |
b28a02de | 1517 | *addr++ = 0x87654321; |
1da177e4 | 1518 | } |
40b44137 JK |
1519 | |
1520 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1521 | int map, unsigned long caller) | |
1522 | { | |
1523 | if (!is_debug_pagealloc_cache(cachep)) | |
1524 | return; | |
1525 | ||
1526 | if (caller) | |
1527 | store_stackinfo(cachep, objp, caller); | |
1528 | ||
1529 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1530 | } | |
1531 | ||
1532 | #else | |
1533 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1534 | int map, unsigned long caller) {} | |
1535 | ||
1da177e4 LT |
1536 | #endif |
1537 | ||
343e0d7a | 1538 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1539 | { |
8c138bc0 | 1540 | int size = cachep->object_size; |
3dafccf2 | 1541 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1542 | |
1543 | memset(addr, val, size); | |
b28a02de | 1544 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | static void dump_line(char *data, int offset, int limit) | |
1548 | { | |
1549 | int i; | |
aa83aa40 DJ |
1550 | unsigned char error = 0; |
1551 | int bad_count = 0; | |
1552 | ||
1170532b | 1553 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1554 | for (i = 0; i < limit; i++) { |
1555 | if (data[offset + i] != POISON_FREE) { | |
1556 | error = data[offset + i]; | |
1557 | bad_count++; | |
1558 | } | |
aa83aa40 | 1559 | } |
fdde6abb SAS |
1560 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1561 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1562 | |
1563 | if (bad_count == 1) { | |
1564 | error ^= POISON_FREE; | |
1565 | if (!(error & (error - 1))) { | |
1170532b | 1566 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1567 | #ifdef CONFIG_X86 |
1170532b | 1568 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1569 | #else |
1170532b | 1570 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1571 | #endif |
1572 | } | |
1573 | } | |
1da177e4 LT |
1574 | } |
1575 | #endif | |
1576 | ||
1577 | #if DEBUG | |
1578 | ||
343e0d7a | 1579 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1580 | { |
1581 | int i, size; | |
1582 | char *realobj; | |
1583 | ||
1584 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1585 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1586 | *dbg_redzone1(cachep, objp), | |
1587 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1588 | } |
1589 | ||
85c3e4a5 GU |
1590 | if (cachep->flags & SLAB_STORE_USER) |
1591 | pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp)); | |
3dafccf2 | 1592 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1593 | size = cachep->object_size; |
b28a02de | 1594 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1595 | int limit; |
1596 | limit = 16; | |
b28a02de PE |
1597 | if (i + limit > size) |
1598 | limit = size - i; | |
1da177e4 LT |
1599 | dump_line(realobj, i, limit); |
1600 | } | |
1601 | } | |
1602 | ||
343e0d7a | 1603 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1604 | { |
1605 | char *realobj; | |
1606 | int size, i; | |
1607 | int lines = 0; | |
1608 | ||
40b44137 JK |
1609 | if (is_debug_pagealloc_cache(cachep)) |
1610 | return; | |
1611 | ||
3dafccf2 | 1612 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1613 | size = cachep->object_size; |
1da177e4 | 1614 | |
b28a02de | 1615 | for (i = 0; i < size; i++) { |
1da177e4 | 1616 | char exp = POISON_FREE; |
b28a02de | 1617 | if (i == size - 1) |
1da177e4 LT |
1618 | exp = POISON_END; |
1619 | if (realobj[i] != exp) { | |
1620 | int limit; | |
1621 | /* Mismatch ! */ | |
1622 | /* Print header */ | |
1623 | if (lines == 0) { | |
85c3e4a5 | 1624 | pr_err("Slab corruption (%s): %s start=%px, len=%d\n", |
1170532b JP |
1625 | print_tainted(), cachep->name, |
1626 | realobj, size); | |
1da177e4 LT |
1627 | print_objinfo(cachep, objp, 0); |
1628 | } | |
1629 | /* Hexdump the affected line */ | |
b28a02de | 1630 | i = (i / 16) * 16; |
1da177e4 | 1631 | limit = 16; |
b28a02de PE |
1632 | if (i + limit > size) |
1633 | limit = size - i; | |
1da177e4 LT |
1634 | dump_line(realobj, i, limit); |
1635 | i += 16; | |
1636 | lines++; | |
1637 | /* Limit to 5 lines */ | |
1638 | if (lines > 5) | |
1639 | break; | |
1640 | } | |
1641 | } | |
1642 | if (lines != 0) { | |
1643 | /* Print some data about the neighboring objects, if they | |
1644 | * exist: | |
1645 | */ | |
8456a648 | 1646 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1647 | unsigned int objnr; |
1da177e4 | 1648 | |
8456a648 | 1649 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1650 | if (objnr) { |
8456a648 | 1651 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1652 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1653 | pr_err("Prev obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1654 | print_objinfo(cachep, objp, 2); |
1655 | } | |
b28a02de | 1656 | if (objnr + 1 < cachep->num) { |
8456a648 | 1657 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1658 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1659 | pr_err("Next obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1660 | print_objinfo(cachep, objp, 2); |
1661 | } | |
1662 | } | |
1663 | } | |
1664 | #endif | |
1665 | ||
12dd36fa | 1666 | #if DEBUG |
8456a648 JK |
1667 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1668 | struct page *page) | |
1da177e4 | 1669 | { |
1da177e4 | 1670 | int i; |
b03a017b JK |
1671 | |
1672 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1673 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1674 | POISON_FREE); | |
1675 | } | |
1676 | ||
1da177e4 | 1677 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1678 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1679 | |
1680 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1681 | check_poison_obj(cachep, objp); |
40b44137 | 1682 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1683 | } |
1684 | if (cachep->flags & SLAB_RED_ZONE) { | |
1685 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1686 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1687 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1688 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1689 | } |
1da177e4 | 1690 | } |
12dd36fa | 1691 | } |
1da177e4 | 1692 | #else |
8456a648 JK |
1693 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1694 | struct page *page) | |
12dd36fa | 1695 | { |
12dd36fa | 1696 | } |
1da177e4 LT |
1697 | #endif |
1698 | ||
911851e6 RD |
1699 | /** |
1700 | * slab_destroy - destroy and release all objects in a slab | |
1701 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1702 | * @page: page pointer being destroyed |
911851e6 | 1703 | * |
8a7d9b43 WSH |
1704 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1705 | * Before calling the slab page must have been unlinked from the cache. The | |
1706 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1707 | */ |
8456a648 | 1708 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1709 | { |
7e007355 | 1710 | void *freelist; |
12dd36fa | 1711 | |
8456a648 JK |
1712 | freelist = page->freelist; |
1713 | slab_destroy_debugcheck(cachep, page); | |
5f0d5a3a | 1714 | if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU)) |
bc4f610d KS |
1715 | call_rcu(&page->rcu_head, kmem_rcu_free); |
1716 | else | |
0c3aa83e | 1717 | kmem_freepages(cachep, page); |
68126702 JK |
1718 | |
1719 | /* | |
8456a648 | 1720 | * From now on, we don't use freelist |
68126702 JK |
1721 | * although actual page can be freed in rcu context |
1722 | */ | |
1723 | if (OFF_SLAB(cachep)) | |
8456a648 | 1724 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1725 | } |
1726 | ||
97654dfa JK |
1727 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1728 | { | |
1729 | struct page *page, *n; | |
1730 | ||
1731 | list_for_each_entry_safe(page, n, list, lru) { | |
1732 | list_del(&page->lru); | |
1733 | slab_destroy(cachep, page); | |
1734 | } | |
1735 | } | |
1736 | ||
4d268eba | 1737 | /** |
a70773dd RD |
1738 | * calculate_slab_order - calculate size (page order) of slabs |
1739 | * @cachep: pointer to the cache that is being created | |
1740 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1741 | * @flags: slab allocation flags |
1742 | * | |
1743 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1744 | * |
1745 | * This could be made much more intelligent. For now, try to avoid using | |
1746 | * high order pages for slabs. When the gfp() functions are more friendly | |
1747 | * towards high-order requests, this should be changed. | |
1748 | */ | |
a737b3e2 | 1749 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
d50112ed | 1750 | size_t size, slab_flags_t flags) |
4d268eba PE |
1751 | { |
1752 | size_t left_over = 0; | |
9888e6fa | 1753 | int gfporder; |
4d268eba | 1754 | |
0aa817f0 | 1755 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1756 | unsigned int num; |
1757 | size_t remainder; | |
1758 | ||
70f75067 | 1759 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1760 | if (!num) |
1761 | continue; | |
9888e6fa | 1762 | |
f315e3fa JK |
1763 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1764 | if (num > SLAB_OBJ_MAX_NUM) | |
1765 | break; | |
1766 | ||
b1ab41c4 | 1767 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1768 | struct kmem_cache *freelist_cache; |
1769 | size_t freelist_size; | |
1770 | ||
1771 | freelist_size = num * sizeof(freelist_idx_t); | |
1772 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1773 | if (!freelist_cache) | |
1774 | continue; | |
1775 | ||
b1ab41c4 | 1776 | /* |
3217fd9b | 1777 | * Needed to avoid possible looping condition |
76b342bd | 1778 | * in cache_grow_begin() |
b1ab41c4 | 1779 | */ |
3217fd9b JK |
1780 | if (OFF_SLAB(freelist_cache)) |
1781 | continue; | |
b1ab41c4 | 1782 | |
3217fd9b JK |
1783 | /* check if off slab has enough benefit */ |
1784 | if (freelist_cache->size > cachep->size / 2) | |
1785 | continue; | |
b1ab41c4 | 1786 | } |
4d268eba | 1787 | |
9888e6fa | 1788 | /* Found something acceptable - save it away */ |
4d268eba | 1789 | cachep->num = num; |
9888e6fa | 1790 | cachep->gfporder = gfporder; |
4d268eba PE |
1791 | left_over = remainder; |
1792 | ||
f78bb8ad LT |
1793 | /* |
1794 | * A VFS-reclaimable slab tends to have most allocations | |
1795 | * as GFP_NOFS and we really don't want to have to be allocating | |
1796 | * higher-order pages when we are unable to shrink dcache. | |
1797 | */ | |
1798 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1799 | break; | |
1800 | ||
4d268eba PE |
1801 | /* |
1802 | * Large number of objects is good, but very large slabs are | |
1803 | * currently bad for the gfp()s. | |
1804 | */ | |
543585cc | 1805 | if (gfporder >= slab_max_order) |
4d268eba PE |
1806 | break; |
1807 | ||
9888e6fa LT |
1808 | /* |
1809 | * Acceptable internal fragmentation? | |
1810 | */ | |
a737b3e2 | 1811 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1812 | break; |
1813 | } | |
1814 | return left_over; | |
1815 | } | |
1816 | ||
bf0dea23 JK |
1817 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1818 | struct kmem_cache *cachep, int entries, int batchcount) | |
1819 | { | |
1820 | int cpu; | |
1821 | size_t size; | |
1822 | struct array_cache __percpu *cpu_cache; | |
1823 | ||
1824 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1825 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1826 | |
1827 | if (!cpu_cache) | |
1828 | return NULL; | |
1829 | ||
1830 | for_each_possible_cpu(cpu) { | |
1831 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1832 | entries, batchcount); | |
1833 | } | |
1834 | ||
1835 | return cpu_cache; | |
1836 | } | |
1837 | ||
bd721ea7 | 1838 | static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1839 | { |
97d06609 | 1840 | if (slab_state >= FULL) |
83b519e8 | 1841 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1842 | |
bf0dea23 JK |
1843 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1844 | if (!cachep->cpu_cache) | |
1845 | return 1; | |
1846 | ||
97d06609 | 1847 | if (slab_state == DOWN) { |
bf0dea23 JK |
1848 | /* Creation of first cache (kmem_cache). */ |
1849 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1850 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1851 | /* For kmem_cache_node */ |
1852 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1853 | } else { |
bf0dea23 | 1854 | int node; |
f30cf7d1 | 1855 | |
bf0dea23 JK |
1856 | for_each_online_node(node) { |
1857 | cachep->node[node] = kmalloc_node( | |
1858 | sizeof(struct kmem_cache_node), gfp, node); | |
1859 | BUG_ON(!cachep->node[node]); | |
1860 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1861 | } |
1862 | } | |
bf0dea23 | 1863 | |
6a67368c | 1864 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1865 | jiffies + REAPTIMEOUT_NODE + |
1866 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1867 | |
1868 | cpu_cache_get(cachep)->avail = 0; | |
1869 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1870 | cpu_cache_get(cachep)->batchcount = 1; | |
1871 | cpu_cache_get(cachep)->touched = 0; | |
1872 | cachep->batchcount = 1; | |
1873 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1874 | return 0; |
f30cf7d1 PE |
1875 | } |
1876 | ||
d50112ed AD |
1877 | slab_flags_t kmem_cache_flags(unsigned long object_size, |
1878 | slab_flags_t flags, const char *name, | |
12220dea JK |
1879 | void (*ctor)(void *)) |
1880 | { | |
1881 | return flags; | |
1882 | } | |
1883 | ||
1884 | struct kmem_cache * | |
1885 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
d50112ed | 1886 | slab_flags_t flags, void (*ctor)(void *)) |
12220dea JK |
1887 | { |
1888 | struct kmem_cache *cachep; | |
1889 | ||
1890 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1891 | if (cachep) { | |
1892 | cachep->refcount++; | |
1893 | ||
1894 | /* | |
1895 | * Adjust the object sizes so that we clear | |
1896 | * the complete object on kzalloc. | |
1897 | */ | |
1898 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1899 | } | |
1900 | return cachep; | |
1901 | } | |
1902 | ||
b03a017b | 1903 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1904 | size_t size, slab_flags_t flags) |
b03a017b JK |
1905 | { |
1906 | size_t left; | |
1907 | ||
1908 | cachep->num = 0; | |
1909 | ||
5f0d5a3a | 1910 | if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU) |
b03a017b JK |
1911 | return false; |
1912 | ||
1913 | left = calculate_slab_order(cachep, size, | |
1914 | flags | CFLGS_OBJFREELIST_SLAB); | |
1915 | if (!cachep->num) | |
1916 | return false; | |
1917 | ||
1918 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1919 | return false; | |
1920 | ||
1921 | cachep->colour = left / cachep->colour_off; | |
1922 | ||
1923 | return true; | |
1924 | } | |
1925 | ||
158e319b | 1926 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1927 | size_t size, slab_flags_t flags) |
158e319b JK |
1928 | { |
1929 | size_t left; | |
1930 | ||
1931 | cachep->num = 0; | |
1932 | ||
1933 | /* | |
3217fd9b JK |
1934 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1935 | * to avoid recursive calls into kmemleak. | |
158e319b | 1936 | */ |
158e319b JK |
1937 | if (flags & SLAB_NOLEAKTRACE) |
1938 | return false; | |
1939 | ||
1940 | /* | |
1941 | * Size is large, assume best to place the slab management obj | |
1942 | * off-slab (should allow better packing of objs). | |
1943 | */ | |
1944 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1945 | if (!cachep->num) | |
1946 | return false; | |
1947 | ||
1948 | /* | |
1949 | * If the slab has been placed off-slab, and we have enough space then | |
1950 | * move it on-slab. This is at the expense of any extra colouring. | |
1951 | */ | |
1952 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1953 | return false; | |
1954 | ||
1955 | cachep->colour = left / cachep->colour_off; | |
1956 | ||
1957 | return true; | |
1958 | } | |
1959 | ||
1960 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
d50112ed | 1961 | size_t size, slab_flags_t flags) |
158e319b JK |
1962 | { |
1963 | size_t left; | |
1964 | ||
1965 | cachep->num = 0; | |
1966 | ||
1967 | left = calculate_slab_order(cachep, size, flags); | |
1968 | if (!cachep->num) | |
1969 | return false; | |
1970 | ||
1971 | cachep->colour = left / cachep->colour_off; | |
1972 | ||
1973 | return true; | |
1974 | } | |
1975 | ||
1da177e4 | 1976 | /** |
039363f3 | 1977 | * __kmem_cache_create - Create a cache. |
a755b76a | 1978 | * @cachep: cache management descriptor |
1da177e4 | 1979 | * @flags: SLAB flags |
1da177e4 LT |
1980 | * |
1981 | * Returns a ptr to the cache on success, NULL on failure. | |
1982 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1983 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1984 | * |
1da177e4 LT |
1985 | * The flags are |
1986 | * | |
1987 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1988 | * to catch references to uninitialised memory. | |
1989 | * | |
1990 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1991 | * for buffer overruns. | |
1992 | * | |
1da177e4 LT |
1993 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1994 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1995 | * as davem. | |
1996 | */ | |
d50112ed | 1997 | int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags) |
1da177e4 | 1998 | { |
d4a5fca5 | 1999 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2000 | gfp_t gfp; |
278b1bb1 | 2001 | int err; |
8a13a4cc | 2002 | size_t size = cachep->size; |
1da177e4 | 2003 | |
1da177e4 | 2004 | #if DEBUG |
1da177e4 LT |
2005 | #if FORCED_DEBUG |
2006 | /* | |
2007 | * Enable redzoning and last user accounting, except for caches with | |
2008 | * large objects, if the increased size would increase the object size | |
2009 | * above the next power of two: caches with object sizes just above a | |
2010 | * power of two have a significant amount of internal fragmentation. | |
2011 | */ | |
87a927c7 DW |
2012 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2013 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2014 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
5f0d5a3a | 2015 | if (!(flags & SLAB_TYPESAFE_BY_RCU)) |
1da177e4 LT |
2016 | flags |= SLAB_POISON; |
2017 | #endif | |
1da177e4 | 2018 | #endif |
1da177e4 | 2019 | |
a737b3e2 AM |
2020 | /* |
2021 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2022 | * unaligned accesses for some archs when redzoning is used, and makes |
2023 | * sure any on-slab bufctl's are also correctly aligned. | |
2024 | */ | |
e0771950 | 2025 | size = ALIGN(size, BYTES_PER_WORD); |
1da177e4 | 2026 | |
87a927c7 DW |
2027 | if (flags & SLAB_RED_ZONE) { |
2028 | ralign = REDZONE_ALIGN; | |
2029 | /* If redzoning, ensure that the second redzone is suitably | |
2030 | * aligned, by adjusting the object size accordingly. */ | |
e0771950 | 2031 | size = ALIGN(size, REDZONE_ALIGN); |
87a927c7 | 2032 | } |
ca5f9703 | 2033 | |
a44b56d3 | 2034 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2035 | if (ralign < cachep->align) { |
2036 | ralign = cachep->align; | |
1da177e4 | 2037 | } |
3ff84a7f PE |
2038 | /* disable debug if necessary */ |
2039 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2040 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2041 | /* |
ca5f9703 | 2042 | * 4) Store it. |
1da177e4 | 2043 | */ |
8a13a4cc | 2044 | cachep->align = ralign; |
158e319b JK |
2045 | cachep->colour_off = cache_line_size(); |
2046 | /* Offset must be a multiple of the alignment. */ | |
2047 | if (cachep->colour_off < cachep->align) | |
2048 | cachep->colour_off = cachep->align; | |
1da177e4 | 2049 | |
83b519e8 PE |
2050 | if (slab_is_available()) |
2051 | gfp = GFP_KERNEL; | |
2052 | else | |
2053 | gfp = GFP_NOWAIT; | |
2054 | ||
1da177e4 | 2055 | #if DEBUG |
1da177e4 | 2056 | |
ca5f9703 PE |
2057 | /* |
2058 | * Both debugging options require word-alignment which is calculated | |
2059 | * into align above. | |
2060 | */ | |
1da177e4 | 2061 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2062 | /* add space for red zone words */ |
3ff84a7f PE |
2063 | cachep->obj_offset += sizeof(unsigned long long); |
2064 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2065 | } |
2066 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2067 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2068 | * the real object. But if the second red zone needs to be |
2069 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2070 | */ |
87a927c7 DW |
2071 | if (flags & SLAB_RED_ZONE) |
2072 | size += REDZONE_ALIGN; | |
2073 | else | |
2074 | size += BYTES_PER_WORD; | |
1da177e4 | 2075 | } |
832a15d2 JK |
2076 | #endif |
2077 | ||
7ed2f9e6 AP |
2078 | kasan_cache_create(cachep, &size, &flags); |
2079 | ||
832a15d2 JK |
2080 | size = ALIGN(size, cachep->align); |
2081 | /* | |
2082 | * We should restrict the number of objects in a slab to implement | |
2083 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2084 | */ | |
2085 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2086 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2087 | ||
2088 | #if DEBUG | |
03a2d2a3 JK |
2089 | /* |
2090 | * To activate debug pagealloc, off-slab management is necessary | |
2091 | * requirement. In early phase of initialization, small sized slab | |
2092 | * doesn't get initialized so it would not be possible. So, we need | |
2093 | * to check size >= 256. It guarantees that all necessary small | |
2094 | * sized slab is initialized in current slab initialization sequence. | |
2095 | */ | |
40323278 | 2096 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2097 | size >= 256 && cachep->object_size > cache_line_size()) { |
2098 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2099 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2100 | ||
2101 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2102 | flags |= CFLGS_OFF_SLAB; | |
2103 | cachep->obj_offset += tmp_size - size; | |
2104 | size = tmp_size; | |
2105 | goto done; | |
2106 | } | |
2107 | } | |
1da177e4 | 2108 | } |
1da177e4 LT |
2109 | #endif |
2110 | ||
b03a017b JK |
2111 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2112 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2113 | goto done; | |
2114 | } | |
2115 | ||
158e319b | 2116 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2117 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2118 | goto done; |
832a15d2 | 2119 | } |
1da177e4 | 2120 | |
158e319b JK |
2121 | if (set_on_slab_cache(cachep, size, flags)) |
2122 | goto done; | |
1da177e4 | 2123 | |
158e319b | 2124 | return -E2BIG; |
1da177e4 | 2125 | |
158e319b JK |
2126 | done: |
2127 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2128 | cachep->flags = flags; |
a57a4988 | 2129 | cachep->allocflags = __GFP_COMP; |
a3187e43 | 2130 | if (flags & SLAB_CACHE_DMA) |
a618e89f | 2131 | cachep->allocflags |= GFP_DMA; |
26f42db9 NB |
2132 | if (flags & SLAB_CACHE_DMA32) |
2133 | cachep->allocflags |= GFP_DMA32; | |
a3ba0744 DR |
2134 | if (flags & SLAB_RECLAIM_ACCOUNT) |
2135 | cachep->allocflags |= __GFP_RECLAIMABLE; | |
3b0efdfa | 2136 | cachep->size = size; |
6a2d7a95 | 2137 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2138 | |
40b44137 JK |
2139 | #if DEBUG |
2140 | /* | |
2141 | * If we're going to use the generic kernel_map_pages() | |
2142 | * poisoning, then it's going to smash the contents of | |
2143 | * the redzone and userword anyhow, so switch them off. | |
2144 | */ | |
2145 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2146 | (cachep->flags & SLAB_POISON) && | |
2147 | is_debug_pagealloc_cache(cachep)) | |
2148 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2149 | #endif | |
2150 | ||
2151 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2152 | cachep->freelist_cache = |
2153 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2154 | } |
1da177e4 | 2155 | |
278b1bb1 CL |
2156 | err = setup_cpu_cache(cachep, gfp); |
2157 | if (err) { | |
52b4b950 | 2158 | __kmem_cache_release(cachep); |
278b1bb1 | 2159 | return err; |
2ed3a4ef | 2160 | } |
1da177e4 | 2161 | |
278b1bb1 | 2162 | return 0; |
1da177e4 | 2163 | } |
1da177e4 LT |
2164 | |
2165 | #if DEBUG | |
2166 | static void check_irq_off(void) | |
2167 | { | |
2168 | BUG_ON(!irqs_disabled()); | |
2169 | } | |
2170 | ||
2171 | static void check_irq_on(void) | |
2172 | { | |
2173 | BUG_ON(irqs_disabled()); | |
2174 | } | |
2175 | ||
18726ca8 JK |
2176 | static void check_mutex_acquired(void) |
2177 | { | |
2178 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2179 | } | |
2180 | ||
343e0d7a | 2181 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2182 | { |
2183 | #ifdef CONFIG_SMP | |
2184 | check_irq_off(); | |
18bf8541 | 2185 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2186 | #endif |
2187 | } | |
e498be7d | 2188 | |
343e0d7a | 2189 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2190 | { |
2191 | #ifdef CONFIG_SMP | |
2192 | check_irq_off(); | |
18bf8541 | 2193 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2194 | #endif |
2195 | } | |
2196 | ||
1da177e4 LT |
2197 | #else |
2198 | #define check_irq_off() do { } while(0) | |
2199 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2200 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2201 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2202 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2203 | #endif |
2204 | ||
18726ca8 JK |
2205 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2206 | int node, bool free_all, struct list_head *list) | |
2207 | { | |
2208 | int tofree; | |
2209 | ||
2210 | if (!ac || !ac->avail) | |
2211 | return; | |
2212 | ||
2213 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2214 | if (tofree > ac->avail) | |
2215 | tofree = (ac->avail + 1) / 2; | |
2216 | ||
2217 | free_block(cachep, ac->entry, tofree, node, list); | |
2218 | ac->avail -= tofree; | |
2219 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2220 | } | |
aab2207c | 2221 | |
1da177e4 LT |
2222 | static void do_drain(void *arg) |
2223 | { | |
a737b3e2 | 2224 | struct kmem_cache *cachep = arg; |
1da177e4 | 2225 | struct array_cache *ac; |
7d6e6d09 | 2226 | int node = numa_mem_id(); |
18bf8541 | 2227 | struct kmem_cache_node *n; |
97654dfa | 2228 | LIST_HEAD(list); |
1da177e4 LT |
2229 | |
2230 | check_irq_off(); | |
9a2dba4b | 2231 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2232 | n = get_node(cachep, node); |
2233 | spin_lock(&n->list_lock); | |
97654dfa | 2234 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2235 | spin_unlock(&n->list_lock); |
97654dfa | 2236 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2237 | ac->avail = 0; |
2238 | } | |
2239 | ||
343e0d7a | 2240 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2241 | { |
ce8eb6c4 | 2242 | struct kmem_cache_node *n; |
e498be7d | 2243 | int node; |
18726ca8 | 2244 | LIST_HEAD(list); |
e498be7d | 2245 | |
15c8b6c1 | 2246 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2247 | check_irq_on(); |
18bf8541 CL |
2248 | for_each_kmem_cache_node(cachep, node, n) |
2249 | if (n->alien) | |
ce8eb6c4 | 2250 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2251 | |
18726ca8 JK |
2252 | for_each_kmem_cache_node(cachep, node, n) { |
2253 | spin_lock_irq(&n->list_lock); | |
2254 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2255 | spin_unlock_irq(&n->list_lock); | |
2256 | ||
2257 | slabs_destroy(cachep, &list); | |
2258 | } | |
1da177e4 LT |
2259 | } |
2260 | ||
ed11d9eb CL |
2261 | /* |
2262 | * Remove slabs from the list of free slabs. | |
2263 | * Specify the number of slabs to drain in tofree. | |
2264 | * | |
2265 | * Returns the actual number of slabs released. | |
2266 | */ | |
2267 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2268 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2269 | { |
ed11d9eb CL |
2270 | struct list_head *p; |
2271 | int nr_freed; | |
8456a648 | 2272 | struct page *page; |
1da177e4 | 2273 | |
ed11d9eb | 2274 | nr_freed = 0; |
ce8eb6c4 | 2275 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2276 | |
ce8eb6c4 CL |
2277 | spin_lock_irq(&n->list_lock); |
2278 | p = n->slabs_free.prev; | |
2279 | if (p == &n->slabs_free) { | |
2280 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2281 | goto out; |
2282 | } | |
1da177e4 | 2283 | |
8456a648 | 2284 | page = list_entry(p, struct page, lru); |
8456a648 | 2285 | list_del(&page->lru); |
f728b0a5 | 2286 | n->free_slabs--; |
bf00bd34 | 2287 | n->total_slabs--; |
ed11d9eb CL |
2288 | /* |
2289 | * Safe to drop the lock. The slab is no longer linked | |
2290 | * to the cache. | |
2291 | */ | |
ce8eb6c4 CL |
2292 | n->free_objects -= cache->num; |
2293 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2294 | slab_destroy(cache, page); |
ed11d9eb | 2295 | nr_freed++; |
1da177e4 | 2296 | } |
ed11d9eb CL |
2297 | out: |
2298 | return nr_freed; | |
1da177e4 LT |
2299 | } |
2300 | ||
c9fc5864 | 2301 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2302 | { |
18bf8541 CL |
2303 | int ret = 0; |
2304 | int node; | |
ce8eb6c4 | 2305 | struct kmem_cache_node *n; |
e498be7d CL |
2306 | |
2307 | drain_cpu_caches(cachep); | |
2308 | ||
2309 | check_irq_on(); | |
18bf8541 | 2310 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2311 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2312 | |
ce8eb6c4 CL |
2313 | ret += !list_empty(&n->slabs_full) || |
2314 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2315 | } |
2316 | return (ret ? 1 : 0); | |
2317 | } | |
2318 | ||
c9fc5864 TH |
2319 | #ifdef CONFIG_MEMCG |
2320 | void __kmemcg_cache_deactivate(struct kmem_cache *cachep) | |
2321 | { | |
2322 | __kmem_cache_shrink(cachep); | |
2323 | } | |
2324 | #endif | |
2325 | ||
945cf2b6 | 2326 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 | 2327 | { |
c9fc5864 | 2328 | return __kmem_cache_shrink(cachep); |
52b4b950 DS |
2329 | } |
2330 | ||
2331 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2332 | { |
12c3667f | 2333 | int i; |
ce8eb6c4 | 2334 | struct kmem_cache_node *n; |
1da177e4 | 2335 | |
c7ce4f60 TG |
2336 | cache_random_seq_destroy(cachep); |
2337 | ||
bf0dea23 | 2338 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2339 | |
ce8eb6c4 | 2340 | /* NUMA: free the node structures */ |
18bf8541 CL |
2341 | for_each_kmem_cache_node(cachep, i, n) { |
2342 | kfree(n->shared); | |
2343 | free_alien_cache(n->alien); | |
2344 | kfree(n); | |
2345 | cachep->node[i] = NULL; | |
12c3667f | 2346 | } |
1da177e4 | 2347 | } |
1da177e4 | 2348 | |
e5ac9c5a RT |
2349 | /* |
2350 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2351 | * |
2352 | * For a slab cache when the slab descriptor is off-slab, the | |
2353 | * slab descriptor can't come from the same cache which is being created, | |
2354 | * Because if it is the case, that means we defer the creation of | |
2355 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2356 | * And we eventually call down to __kmem_cache_create(), which | |
2357 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2358 | * This is a "chicken-and-egg" problem. | |
2359 | * | |
2360 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2361 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2362 | */ |
7e007355 | 2363 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2364 | struct page *page, int colour_off, |
2365 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2366 | { |
7e007355 | 2367 | void *freelist; |
0c3aa83e | 2368 | void *addr = page_address(page); |
b28a02de | 2369 | |
2e6b3602 JK |
2370 | page->s_mem = addr + colour_off; |
2371 | page->active = 0; | |
2372 | ||
b03a017b JK |
2373 | if (OBJFREELIST_SLAB(cachep)) |
2374 | freelist = NULL; | |
2375 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2376 | /* Slab management obj is off-slab. */ |
8456a648 | 2377 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2378 | local_flags, nodeid); |
8456a648 | 2379 | if (!freelist) |
1da177e4 LT |
2380 | return NULL; |
2381 | } else { | |
2e6b3602 JK |
2382 | /* We will use last bytes at the slab for freelist */ |
2383 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2384 | cachep->freelist_size; | |
1da177e4 | 2385 | } |
2e6b3602 | 2386 | |
8456a648 | 2387 | return freelist; |
1da177e4 LT |
2388 | } |
2389 | ||
7cc68973 | 2390 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2391 | { |
a41adfaa | 2392 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2393 | } |
2394 | ||
2395 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2396 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2397 | { |
a41adfaa | 2398 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2399 | } |
2400 | ||
10b2e9e8 | 2401 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2402 | { |
10b2e9e8 | 2403 | #if DEBUG |
1da177e4 LT |
2404 | int i; |
2405 | ||
2406 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2407 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2408 | |
1da177e4 LT |
2409 | if (cachep->flags & SLAB_STORE_USER) |
2410 | *dbg_userword(cachep, objp) = NULL; | |
2411 | ||
2412 | if (cachep->flags & SLAB_RED_ZONE) { | |
2413 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2414 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2415 | } | |
2416 | /* | |
a737b3e2 AM |
2417 | * Constructors are not allowed to allocate memory from the same |
2418 | * cache which they are a constructor for. Otherwise, deadlock. | |
2419 | * They must also be threaded. | |
1da177e4 | 2420 | */ |
7ed2f9e6 AP |
2421 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2422 | kasan_unpoison_object_data(cachep, | |
2423 | objp + obj_offset(cachep)); | |
51cc5068 | 2424 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2425 | kasan_poison_object_data( |
2426 | cachep, objp + obj_offset(cachep)); | |
2427 | } | |
1da177e4 LT |
2428 | |
2429 | if (cachep->flags & SLAB_RED_ZONE) { | |
2430 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2431 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2432 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2433 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2434 | } |
40b44137 JK |
2435 | /* need to poison the objs? */ |
2436 | if (cachep->flags & SLAB_POISON) { | |
2437 | poison_obj(cachep, objp, POISON_FREE); | |
2438 | slab_kernel_map(cachep, objp, 0, 0); | |
2439 | } | |
10b2e9e8 | 2440 | } |
1da177e4 | 2441 | #endif |
10b2e9e8 JK |
2442 | } |
2443 | ||
c7ce4f60 TG |
2444 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
2445 | /* Hold information during a freelist initialization */ | |
2446 | union freelist_init_state { | |
2447 | struct { | |
2448 | unsigned int pos; | |
7c00fce9 | 2449 | unsigned int *list; |
c7ce4f60 | 2450 | unsigned int count; |
c7ce4f60 TG |
2451 | }; |
2452 | struct rnd_state rnd_state; | |
2453 | }; | |
2454 | ||
2455 | /* | |
2456 | * Initialize the state based on the randomization methode available. | |
2457 | * return true if the pre-computed list is available, false otherwize. | |
2458 | */ | |
2459 | static bool freelist_state_initialize(union freelist_init_state *state, | |
2460 | struct kmem_cache *cachep, | |
2461 | unsigned int count) | |
2462 | { | |
2463 | bool ret; | |
2464 | unsigned int rand; | |
2465 | ||
2466 | /* Use best entropy available to define a random shift */ | |
7c00fce9 | 2467 | rand = get_random_int(); |
c7ce4f60 TG |
2468 | |
2469 | /* Use a random state if the pre-computed list is not available */ | |
2470 | if (!cachep->random_seq) { | |
2471 | prandom_seed_state(&state->rnd_state, rand); | |
2472 | ret = false; | |
2473 | } else { | |
2474 | state->list = cachep->random_seq; | |
2475 | state->count = count; | |
c4e490cf | 2476 | state->pos = rand % count; |
c7ce4f60 TG |
2477 | ret = true; |
2478 | } | |
2479 | return ret; | |
2480 | } | |
2481 | ||
2482 | /* Get the next entry on the list and randomize it using a random shift */ | |
2483 | static freelist_idx_t next_random_slot(union freelist_init_state *state) | |
2484 | { | |
c4e490cf JS |
2485 | if (state->pos >= state->count) |
2486 | state->pos = 0; | |
2487 | return state->list[state->pos++]; | |
c7ce4f60 TG |
2488 | } |
2489 | ||
7c00fce9 TG |
2490 | /* Swap two freelist entries */ |
2491 | static void swap_free_obj(struct page *page, unsigned int a, unsigned int b) | |
2492 | { | |
2493 | swap(((freelist_idx_t *)page->freelist)[a], | |
2494 | ((freelist_idx_t *)page->freelist)[b]); | |
2495 | } | |
2496 | ||
c7ce4f60 TG |
2497 | /* |
2498 | * Shuffle the freelist initialization state based on pre-computed lists. | |
2499 | * return true if the list was successfully shuffled, false otherwise. | |
2500 | */ | |
2501 | static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page) | |
2502 | { | |
7c00fce9 | 2503 | unsigned int objfreelist = 0, i, rand, count = cachep->num; |
c7ce4f60 TG |
2504 | union freelist_init_state state; |
2505 | bool precomputed; | |
2506 | ||
2507 | if (count < 2) | |
2508 | return false; | |
2509 | ||
2510 | precomputed = freelist_state_initialize(&state, cachep, count); | |
2511 | ||
2512 | /* Take a random entry as the objfreelist */ | |
2513 | if (OBJFREELIST_SLAB(cachep)) { | |
2514 | if (!precomputed) | |
2515 | objfreelist = count - 1; | |
2516 | else | |
2517 | objfreelist = next_random_slot(&state); | |
2518 | page->freelist = index_to_obj(cachep, page, objfreelist) + | |
2519 | obj_offset(cachep); | |
2520 | count--; | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * On early boot, generate the list dynamically. | |
2525 | * Later use a pre-computed list for speed. | |
2526 | */ | |
2527 | if (!precomputed) { | |
7c00fce9 TG |
2528 | for (i = 0; i < count; i++) |
2529 | set_free_obj(page, i, i); | |
2530 | ||
2531 | /* Fisher-Yates shuffle */ | |
2532 | for (i = count - 1; i > 0; i--) { | |
2533 | rand = prandom_u32_state(&state.rnd_state); | |
2534 | rand %= (i + 1); | |
2535 | swap_free_obj(page, i, rand); | |
2536 | } | |
c7ce4f60 TG |
2537 | } else { |
2538 | for (i = 0; i < count; i++) | |
2539 | set_free_obj(page, i, next_random_slot(&state)); | |
2540 | } | |
2541 | ||
2542 | if (OBJFREELIST_SLAB(cachep)) | |
2543 | set_free_obj(page, cachep->num - 1, objfreelist); | |
2544 | ||
2545 | return true; | |
2546 | } | |
2547 | #else | |
2548 | static inline bool shuffle_freelist(struct kmem_cache *cachep, | |
2549 | struct page *page) | |
2550 | { | |
2551 | return false; | |
2552 | } | |
2553 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
2554 | ||
10b2e9e8 JK |
2555 | static void cache_init_objs(struct kmem_cache *cachep, |
2556 | struct page *page) | |
2557 | { | |
2558 | int i; | |
7ed2f9e6 | 2559 | void *objp; |
c7ce4f60 | 2560 | bool shuffled; |
10b2e9e8 JK |
2561 | |
2562 | cache_init_objs_debug(cachep, page); | |
2563 | ||
c7ce4f60 TG |
2564 | /* Try to randomize the freelist if enabled */ |
2565 | shuffled = shuffle_freelist(cachep, page); | |
2566 | ||
2567 | if (!shuffled && OBJFREELIST_SLAB(cachep)) { | |
b03a017b JK |
2568 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + |
2569 | obj_offset(cachep); | |
2570 | } | |
2571 | ||
10b2e9e8 | 2572 | for (i = 0; i < cachep->num; i++) { |
b3cbd9bf AR |
2573 | objp = index_to_obj(cachep, page, i); |
2574 | kasan_init_slab_obj(cachep, objp); | |
2575 | ||
10b2e9e8 | 2576 | /* constructor could break poison info */ |
7ed2f9e6 | 2577 | if (DEBUG == 0 && cachep->ctor) { |
7ed2f9e6 AP |
2578 | kasan_unpoison_object_data(cachep, objp); |
2579 | cachep->ctor(objp); | |
2580 | kasan_poison_object_data(cachep, objp); | |
2581 | } | |
10b2e9e8 | 2582 | |
c7ce4f60 TG |
2583 | if (!shuffled) |
2584 | set_free_obj(page, i, i); | |
1da177e4 | 2585 | } |
1da177e4 LT |
2586 | } |
2587 | ||
260b61dd | 2588 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2589 | { |
b1cb0982 | 2590 | void *objp; |
78d382d7 | 2591 | |
e5c58dfd | 2592 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2593 | page->active++; |
78d382d7 | 2594 | |
d31676df JK |
2595 | #if DEBUG |
2596 | if (cachep->flags & SLAB_STORE_USER) | |
2597 | set_store_user_dirty(cachep); | |
2598 | #endif | |
2599 | ||
78d382d7 MD |
2600 | return objp; |
2601 | } | |
2602 | ||
260b61dd JK |
2603 | static void slab_put_obj(struct kmem_cache *cachep, |
2604 | struct page *page, void *objp) | |
78d382d7 | 2605 | { |
8456a648 | 2606 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2607 | #if DEBUG |
16025177 | 2608 | unsigned int i; |
b1cb0982 | 2609 | |
b1cb0982 | 2610 | /* Verify double free bug */ |
8456a648 | 2611 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2612 | if (get_free_obj(page, i) == objnr) { |
85c3e4a5 | 2613 | pr_err("slab: double free detected in cache '%s', objp %px\n", |
756a025f | 2614 | cachep->name, objp); |
b1cb0982 JK |
2615 | BUG(); |
2616 | } | |
78d382d7 MD |
2617 | } |
2618 | #endif | |
8456a648 | 2619 | page->active--; |
b03a017b JK |
2620 | if (!page->freelist) |
2621 | page->freelist = objp + obj_offset(cachep); | |
2622 | ||
e5c58dfd | 2623 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2624 | } |
2625 | ||
4776874f PE |
2626 | /* |
2627 | * Map pages beginning at addr to the given cache and slab. This is required | |
2628 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2629 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2630 | */ |
8456a648 | 2631 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2632 | void *freelist) |
1da177e4 | 2633 | { |
a57a4988 | 2634 | page->slab_cache = cache; |
8456a648 | 2635 | page->freelist = freelist; |
1da177e4 LT |
2636 | } |
2637 | ||
2638 | /* | |
2639 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2640 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2641 | */ | |
76b342bd JK |
2642 | static struct page *cache_grow_begin(struct kmem_cache *cachep, |
2643 | gfp_t flags, int nodeid) | |
1da177e4 | 2644 | { |
7e007355 | 2645 | void *freelist; |
b28a02de PE |
2646 | size_t offset; |
2647 | gfp_t local_flags; | |
511e3a05 | 2648 | int page_node; |
ce8eb6c4 | 2649 | struct kmem_cache_node *n; |
511e3a05 | 2650 | struct page *page; |
1da177e4 | 2651 | |
a737b3e2 AM |
2652 | /* |
2653 | * Be lazy and only check for valid flags here, keeping it out of the | |
2654 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2655 | */ |
c871ac4e | 2656 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
bacdcb34 | 2657 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
2658 | flags &= ~GFP_SLAB_BUG_MASK; |
2659 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
2660 | invalid_mask, &invalid_mask, flags, &flags); | |
2661 | dump_stack(); | |
c871ac4e | 2662 | } |
6cb06229 | 2663 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2664 | |
1da177e4 | 2665 | check_irq_off(); |
d0164adc | 2666 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2667 | local_irq_enable(); |
2668 | ||
a737b3e2 AM |
2669 | /* |
2670 | * Get mem for the objs. Attempt to allocate a physical page from | |
2671 | * 'nodeid'. | |
e498be7d | 2672 | */ |
511e3a05 | 2673 | page = kmem_getpages(cachep, local_flags, nodeid); |
0c3aa83e | 2674 | if (!page) |
1da177e4 LT |
2675 | goto failed; |
2676 | ||
511e3a05 JK |
2677 | page_node = page_to_nid(page); |
2678 | n = get_node(cachep, page_node); | |
03d1d43a JK |
2679 | |
2680 | /* Get colour for the slab, and cal the next value. */ | |
2681 | n->colour_next++; | |
2682 | if (n->colour_next >= cachep->colour) | |
2683 | n->colour_next = 0; | |
2684 | ||
2685 | offset = n->colour_next; | |
2686 | if (offset >= cachep->colour) | |
2687 | offset = 0; | |
2688 | ||
2689 | offset *= cachep->colour_off; | |
2690 | ||
1da177e4 | 2691 | /* Get slab management. */ |
8456a648 | 2692 | freelist = alloc_slabmgmt(cachep, page, offset, |
511e3a05 | 2693 | local_flags & ~GFP_CONSTRAINT_MASK, page_node); |
b03a017b | 2694 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2695 | goto opps1; |
2696 | ||
8456a648 | 2697 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2698 | |
7ed2f9e6 | 2699 | kasan_poison_slab(page); |
8456a648 | 2700 | cache_init_objs(cachep, page); |
1da177e4 | 2701 | |
d0164adc | 2702 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2703 | local_irq_disable(); |
1da177e4 | 2704 | |
76b342bd JK |
2705 | return page; |
2706 | ||
a737b3e2 | 2707 | opps1: |
0c3aa83e | 2708 | kmem_freepages(cachep, page); |
a737b3e2 | 2709 | failed: |
d0164adc | 2710 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2711 | local_irq_disable(); |
76b342bd JK |
2712 | return NULL; |
2713 | } | |
2714 | ||
2715 | static void cache_grow_end(struct kmem_cache *cachep, struct page *page) | |
2716 | { | |
2717 | struct kmem_cache_node *n; | |
2718 | void *list = NULL; | |
2719 | ||
2720 | check_irq_off(); | |
2721 | ||
2722 | if (!page) | |
2723 | return; | |
2724 | ||
2725 | INIT_LIST_HEAD(&page->lru); | |
2726 | n = get_node(cachep, page_to_nid(page)); | |
2727 | ||
2728 | spin_lock(&n->list_lock); | |
bf00bd34 | 2729 | n->total_slabs++; |
f728b0a5 | 2730 | if (!page->active) { |
76b342bd | 2731 | list_add_tail(&page->lru, &(n->slabs_free)); |
f728b0a5 | 2732 | n->free_slabs++; |
bf00bd34 | 2733 | } else |
76b342bd | 2734 | fixup_slab_list(cachep, n, page, &list); |
07a63c41 | 2735 | |
76b342bd JK |
2736 | STATS_INC_GROWN(cachep); |
2737 | n->free_objects += cachep->num - page->active; | |
2738 | spin_unlock(&n->list_lock); | |
2739 | ||
2740 | fixup_objfreelist_debug(cachep, &list); | |
1da177e4 LT |
2741 | } |
2742 | ||
2743 | #if DEBUG | |
2744 | ||
2745 | /* | |
2746 | * Perform extra freeing checks: | |
2747 | * - detect bad pointers. | |
2748 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2749 | */ |
2750 | static void kfree_debugcheck(const void *objp) | |
2751 | { | |
1da177e4 | 2752 | if (!virt_addr_valid(objp)) { |
1170532b | 2753 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2754 | (unsigned long)objp); |
2755 | BUG(); | |
1da177e4 | 2756 | } |
1da177e4 LT |
2757 | } |
2758 | ||
58ce1fd5 PE |
2759 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2760 | { | |
b46b8f19 | 2761 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2762 | |
2763 | redzone1 = *dbg_redzone1(cache, obj); | |
2764 | redzone2 = *dbg_redzone2(cache, obj); | |
2765 | ||
2766 | /* | |
2767 | * Redzone is ok. | |
2768 | */ | |
2769 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2770 | return; | |
2771 | ||
2772 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2773 | slab_error(cache, "double free detected"); | |
2774 | else | |
2775 | slab_error(cache, "memory outside object was overwritten"); | |
2776 | ||
85c3e4a5 | 2777 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b | 2778 | obj, redzone1, redzone2); |
58ce1fd5 PE |
2779 | } |
2780 | ||
343e0d7a | 2781 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2782 | unsigned long caller) |
1da177e4 | 2783 | { |
1da177e4 | 2784 | unsigned int objnr; |
8456a648 | 2785 | struct page *page; |
1da177e4 | 2786 | |
80cbd911 MW |
2787 | BUG_ON(virt_to_cache(objp) != cachep); |
2788 | ||
3dafccf2 | 2789 | objp -= obj_offset(cachep); |
1da177e4 | 2790 | kfree_debugcheck(objp); |
b49af68f | 2791 | page = virt_to_head_page(objp); |
1da177e4 | 2792 | |
1da177e4 | 2793 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2794 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2795 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2796 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2797 | } | |
d31676df JK |
2798 | if (cachep->flags & SLAB_STORE_USER) { |
2799 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2800 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2801 | } |
1da177e4 | 2802 | |
8456a648 | 2803 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2804 | |
2805 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2806 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2807 | |
1da177e4 | 2808 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2809 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2810 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2811 | } |
2812 | return objp; | |
2813 | } | |
2814 | ||
1da177e4 LT |
2815 | #else |
2816 | #define kfree_debugcheck(x) do { } while(0) | |
2817 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2818 | #endif |
2819 | ||
b03a017b JK |
2820 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2821 | void **list) | |
2822 | { | |
2823 | #if DEBUG | |
2824 | void *next = *list; | |
2825 | void *objp; | |
2826 | ||
2827 | while (next) { | |
2828 | objp = next - obj_offset(cachep); | |
2829 | next = *(void **)next; | |
2830 | poison_obj(cachep, objp, POISON_FREE); | |
2831 | } | |
2832 | #endif | |
2833 | } | |
2834 | ||
d8410234 | 2835 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2836 | struct kmem_cache_node *n, struct page *page, |
2837 | void **list) | |
d8410234 JK |
2838 | { |
2839 | /* move slabp to correct slabp list: */ | |
2840 | list_del(&page->lru); | |
b03a017b | 2841 | if (page->active == cachep->num) { |
d8410234 | 2842 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2843 | if (OBJFREELIST_SLAB(cachep)) { |
2844 | #if DEBUG | |
2845 | /* Poisoning will be done without holding the lock */ | |
2846 | if (cachep->flags & SLAB_POISON) { | |
2847 | void **objp = page->freelist; | |
2848 | ||
2849 | *objp = *list; | |
2850 | *list = objp; | |
2851 | } | |
2852 | #endif | |
2853 | page->freelist = NULL; | |
2854 | } | |
2855 | } else | |
d8410234 JK |
2856 | list_add(&page->lru, &n->slabs_partial); |
2857 | } | |
2858 | ||
f68f8ddd JK |
2859 | /* Try to find non-pfmemalloc slab if needed */ |
2860 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
bf00bd34 | 2861 | struct page *page, bool pfmemalloc) |
f68f8ddd JK |
2862 | { |
2863 | if (!page) | |
2864 | return NULL; | |
2865 | ||
2866 | if (pfmemalloc) | |
2867 | return page; | |
2868 | ||
2869 | if (!PageSlabPfmemalloc(page)) | |
2870 | return page; | |
2871 | ||
2872 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2873 | if (n->free_objects > n->free_limit) { | |
2874 | ClearPageSlabPfmemalloc(page); | |
2875 | return page; | |
2876 | } | |
2877 | ||
2878 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2879 | list_del(&page->lru); | |
bf00bd34 | 2880 | if (!page->active) { |
f68f8ddd | 2881 | list_add_tail(&page->lru, &n->slabs_free); |
bf00bd34 | 2882 | n->free_slabs++; |
f728b0a5 | 2883 | } else |
f68f8ddd JK |
2884 | list_add_tail(&page->lru, &n->slabs_partial); |
2885 | ||
2886 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2887 | if (!PageSlabPfmemalloc(page)) | |
2888 | return page; | |
2889 | } | |
2890 | ||
f728b0a5 | 2891 | n->free_touched = 1; |
f68f8ddd | 2892 | list_for_each_entry(page, &n->slabs_free, lru) { |
f728b0a5 | 2893 | if (!PageSlabPfmemalloc(page)) { |
bf00bd34 | 2894 | n->free_slabs--; |
f68f8ddd | 2895 | return page; |
f728b0a5 | 2896 | } |
f68f8ddd JK |
2897 | } |
2898 | ||
2899 | return NULL; | |
2900 | } | |
2901 | ||
2902 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2903 | { |
2904 | struct page *page; | |
2905 | ||
f728b0a5 | 2906 | assert_spin_locked(&n->list_lock); |
bf00bd34 | 2907 | page = list_first_entry_or_null(&n->slabs_partial, struct page, lru); |
7aa0d227 GT |
2908 | if (!page) { |
2909 | n->free_touched = 1; | |
bf00bd34 DR |
2910 | page = list_first_entry_or_null(&n->slabs_free, struct page, |
2911 | lru); | |
f728b0a5 | 2912 | if (page) |
bf00bd34 | 2913 | n->free_slabs--; |
7aa0d227 GT |
2914 | } |
2915 | ||
f68f8ddd | 2916 | if (sk_memalloc_socks()) |
bf00bd34 | 2917 | page = get_valid_first_slab(n, page, pfmemalloc); |
f68f8ddd | 2918 | |
7aa0d227 GT |
2919 | return page; |
2920 | } | |
2921 | ||
f68f8ddd JK |
2922 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2923 | struct kmem_cache_node *n, gfp_t flags) | |
2924 | { | |
2925 | struct page *page; | |
2926 | void *obj; | |
2927 | void *list = NULL; | |
2928 | ||
2929 | if (!gfp_pfmemalloc_allowed(flags)) | |
2930 | return NULL; | |
2931 | ||
2932 | spin_lock(&n->list_lock); | |
2933 | page = get_first_slab(n, true); | |
2934 | if (!page) { | |
2935 | spin_unlock(&n->list_lock); | |
2936 | return NULL; | |
2937 | } | |
2938 | ||
2939 | obj = slab_get_obj(cachep, page); | |
2940 | n->free_objects--; | |
2941 | ||
2942 | fixup_slab_list(cachep, n, page, &list); | |
2943 | ||
2944 | spin_unlock(&n->list_lock); | |
2945 | fixup_objfreelist_debug(cachep, &list); | |
2946 | ||
2947 | return obj; | |
2948 | } | |
2949 | ||
213b4695 JK |
2950 | /* |
2951 | * Slab list should be fixed up by fixup_slab_list() for existing slab | |
2952 | * or cache_grow_end() for new slab | |
2953 | */ | |
2954 | static __always_inline int alloc_block(struct kmem_cache *cachep, | |
2955 | struct array_cache *ac, struct page *page, int batchcount) | |
2956 | { | |
2957 | /* | |
2958 | * There must be at least one object available for | |
2959 | * allocation. | |
2960 | */ | |
2961 | BUG_ON(page->active >= cachep->num); | |
2962 | ||
2963 | while (page->active < cachep->num && batchcount--) { | |
2964 | STATS_INC_ALLOCED(cachep); | |
2965 | STATS_INC_ACTIVE(cachep); | |
2966 | STATS_SET_HIGH(cachep); | |
2967 | ||
2968 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); | |
2969 | } | |
2970 | ||
2971 | return batchcount; | |
2972 | } | |
2973 | ||
f68f8ddd | 2974 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2975 | { |
2976 | int batchcount; | |
ce8eb6c4 | 2977 | struct kmem_cache_node *n; |
801faf0d | 2978 | struct array_cache *ac, *shared; |
1ca4cb24 | 2979 | int node; |
b03a017b | 2980 | void *list = NULL; |
76b342bd | 2981 | struct page *page; |
1ca4cb24 | 2982 | |
1da177e4 | 2983 | check_irq_off(); |
7d6e6d09 | 2984 | node = numa_mem_id(); |
f68f8ddd | 2985 | |
9a2dba4b | 2986 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2987 | batchcount = ac->batchcount; |
2988 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2989 | /* |
2990 | * If there was little recent activity on this cache, then | |
2991 | * perform only a partial refill. Otherwise we could generate | |
2992 | * refill bouncing. | |
1da177e4 LT |
2993 | */ |
2994 | batchcount = BATCHREFILL_LIMIT; | |
2995 | } | |
18bf8541 | 2996 | n = get_node(cachep, node); |
e498be7d | 2997 | |
ce8eb6c4 | 2998 | BUG_ON(ac->avail > 0 || !n); |
801faf0d JK |
2999 | shared = READ_ONCE(n->shared); |
3000 | if (!n->free_objects && (!shared || !shared->avail)) | |
3001 | goto direct_grow; | |
3002 | ||
ce8eb6c4 | 3003 | spin_lock(&n->list_lock); |
801faf0d | 3004 | shared = READ_ONCE(n->shared); |
1da177e4 | 3005 | |
3ded175a | 3006 | /* See if we can refill from the shared array */ |
801faf0d JK |
3007 | if (shared && transfer_objects(ac, shared, batchcount)) { |
3008 | shared->touched = 1; | |
3ded175a | 3009 | goto alloc_done; |
44b57f1c | 3010 | } |
3ded175a | 3011 | |
1da177e4 | 3012 | while (batchcount > 0) { |
1da177e4 | 3013 | /* Get slab alloc is to come from. */ |
f68f8ddd | 3014 | page = get_first_slab(n, false); |
7aa0d227 GT |
3015 | if (!page) |
3016 | goto must_grow; | |
1da177e4 | 3017 | |
1da177e4 | 3018 | check_spinlock_acquired(cachep); |
714b8171 | 3019 | |
213b4695 | 3020 | batchcount = alloc_block(cachep, ac, page, batchcount); |
b03a017b | 3021 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
3022 | } |
3023 | ||
a737b3e2 | 3024 | must_grow: |
ce8eb6c4 | 3025 | n->free_objects -= ac->avail; |
a737b3e2 | 3026 | alloc_done: |
ce8eb6c4 | 3027 | spin_unlock(&n->list_lock); |
b03a017b | 3028 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 | 3029 | |
801faf0d | 3030 | direct_grow: |
1da177e4 | 3031 | if (unlikely(!ac->avail)) { |
f68f8ddd JK |
3032 | /* Check if we can use obj in pfmemalloc slab */ |
3033 | if (sk_memalloc_socks()) { | |
3034 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
3035 | ||
3036 | if (obj) | |
3037 | return obj; | |
3038 | } | |
3039 | ||
76b342bd | 3040 | page = cache_grow_begin(cachep, gfp_exact_node(flags), node); |
e498be7d | 3041 | |
76b342bd JK |
3042 | /* |
3043 | * cache_grow_begin() can reenable interrupts, | |
3044 | * then ac could change. | |
3045 | */ | |
9a2dba4b | 3046 | ac = cpu_cache_get(cachep); |
213b4695 JK |
3047 | if (!ac->avail && page) |
3048 | alloc_block(cachep, ac, page, batchcount); | |
3049 | cache_grow_end(cachep, page); | |
072bb0aa | 3050 | |
213b4695 | 3051 | if (!ac->avail) |
1da177e4 | 3052 | return NULL; |
1da177e4 LT |
3053 | } |
3054 | ac->touched = 1; | |
072bb0aa | 3055 | |
f68f8ddd | 3056 | return ac->entry[--ac->avail]; |
1da177e4 LT |
3057 | } |
3058 | ||
a737b3e2 AM |
3059 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3060 | gfp_t flags) | |
1da177e4 | 3061 | { |
d0164adc | 3062 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
3063 | } |
3064 | ||
3065 | #if DEBUG | |
a737b3e2 | 3066 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3067 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3068 | { |
b28a02de | 3069 | if (!objp) |
1da177e4 | 3070 | return objp; |
b28a02de | 3071 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3072 | check_poison_obj(cachep, objp); |
40b44137 | 3073 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
3074 | poison_obj(cachep, objp, POISON_INUSE); |
3075 | } | |
3076 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3077 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3078 | |
3079 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3080 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3081 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 3082 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
85c3e4a5 | 3083 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b JP |
3084 | objp, *dbg_redzone1(cachep, objp), |
3085 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3086 | } |
3087 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3088 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3089 | } | |
03787301 | 3090 | |
3dafccf2 | 3091 | objp += obj_offset(cachep); |
4f104934 | 3092 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3093 | cachep->ctor(objp); |
7ea466f2 TH |
3094 | if (ARCH_SLAB_MINALIGN && |
3095 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
85c3e4a5 | 3096 | pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3097 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3098 | } |
1da177e4 LT |
3099 | return objp; |
3100 | } | |
3101 | #else | |
3102 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3103 | #endif | |
3104 | ||
343e0d7a | 3105 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3106 | { |
b28a02de | 3107 | void *objp; |
1da177e4 LT |
3108 | struct array_cache *ac; |
3109 | ||
5c382300 | 3110 | check_irq_off(); |
8a8b6502 | 3111 | |
9a2dba4b | 3112 | ac = cpu_cache_get(cachep); |
1da177e4 | 3113 | if (likely(ac->avail)) { |
1da177e4 | 3114 | ac->touched = 1; |
f68f8ddd | 3115 | objp = ac->entry[--ac->avail]; |
072bb0aa | 3116 | |
f68f8ddd JK |
3117 | STATS_INC_ALLOCHIT(cachep); |
3118 | goto out; | |
1da177e4 | 3119 | } |
072bb0aa MG |
3120 | |
3121 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 3122 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
3123 | /* |
3124 | * the 'ac' may be updated by cache_alloc_refill(), | |
3125 | * and kmemleak_erase() requires its correct value. | |
3126 | */ | |
3127 | ac = cpu_cache_get(cachep); | |
3128 | ||
3129 | out: | |
d5cff635 CM |
3130 | /* |
3131 | * To avoid a false negative, if an object that is in one of the | |
3132 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3133 | * treat the array pointers as a reference to the object. | |
3134 | */ | |
f3d8b53a O |
3135 | if (objp) |
3136 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3137 | return objp; |
3138 | } | |
3139 | ||
e498be7d | 3140 | #ifdef CONFIG_NUMA |
c61afb18 | 3141 | /* |
2ad654bc | 3142 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3143 | * |
3144 | * If we are in_interrupt, then process context, including cpusets and | |
3145 | * mempolicy, may not apply and should not be used for allocation policy. | |
3146 | */ | |
3147 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3148 | { | |
3149 | int nid_alloc, nid_here; | |
3150 | ||
765c4507 | 3151 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3152 | return NULL; |
7d6e6d09 | 3153 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3154 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3155 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3156 | else if (current->mempolicy) |
2a389610 | 3157 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3158 | if (nid_alloc != nid_here) |
8b98c169 | 3159 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3160 | return NULL; |
3161 | } | |
3162 | ||
765c4507 CL |
3163 | /* |
3164 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3165 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3166 | * available node for available objects. If that fails then we |
3c517a61 CL |
3167 | * perform an allocation without specifying a node. This allows the page |
3168 | * allocator to do its reclaim / fallback magic. We then insert the | |
3169 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3170 | */ |
8c8cc2c1 | 3171 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3172 | { |
8c8cc2c1 | 3173 | struct zonelist *zonelist; |
dd1a239f | 3174 | struct zoneref *z; |
54a6eb5c MG |
3175 | struct zone *zone; |
3176 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3177 | void *obj = NULL; |
76b342bd | 3178 | struct page *page; |
3c517a61 | 3179 | int nid; |
cc9a6c87 | 3180 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3181 | |
3182 | if (flags & __GFP_THISNODE) | |
3183 | return NULL; | |
3184 | ||
cc9a6c87 | 3185 | retry_cpuset: |
d26914d1 | 3186 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3187 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3188 | |
3c517a61 CL |
3189 | retry: |
3190 | /* | |
3191 | * Look through allowed nodes for objects available | |
3192 | * from existing per node queues. | |
3193 | */ | |
54a6eb5c MG |
3194 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3195 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3196 | |
061d7074 | 3197 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3198 | get_node(cache, nid) && |
3199 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3200 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3201 | gfp_exact_node(flags), nid); |
481c5346 CL |
3202 | if (obj) |
3203 | break; | |
3204 | } | |
3c517a61 CL |
3205 | } |
3206 | ||
cfce6604 | 3207 | if (!obj) { |
3c517a61 CL |
3208 | /* |
3209 | * This allocation will be performed within the constraints | |
3210 | * of the current cpuset / memory policy requirements. | |
3211 | * We may trigger various forms of reclaim on the allowed | |
3212 | * set and go into memory reserves if necessary. | |
3213 | */ | |
76b342bd JK |
3214 | page = cache_grow_begin(cache, flags, numa_mem_id()); |
3215 | cache_grow_end(cache, page); | |
3216 | if (page) { | |
3217 | nid = page_to_nid(page); | |
511e3a05 JK |
3218 | obj = ____cache_alloc_node(cache, |
3219 | gfp_exact_node(flags), nid); | |
0c3aa83e | 3220 | |
3c517a61 | 3221 | /* |
511e3a05 JK |
3222 | * Another processor may allocate the objects in |
3223 | * the slab since we are not holding any locks. | |
3c517a61 | 3224 | */ |
511e3a05 JK |
3225 | if (!obj) |
3226 | goto retry; | |
3c517a61 | 3227 | } |
aedb0eb1 | 3228 | } |
cc9a6c87 | 3229 | |
d26914d1 | 3230 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3231 | goto retry_cpuset; |
765c4507 CL |
3232 | return obj; |
3233 | } | |
3234 | ||
e498be7d CL |
3235 | /* |
3236 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3237 | */ |
8b98c169 | 3238 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3239 | int nodeid) |
e498be7d | 3240 | { |
8456a648 | 3241 | struct page *page; |
ce8eb6c4 | 3242 | struct kmem_cache_node *n; |
213b4695 | 3243 | void *obj = NULL; |
b03a017b | 3244 | void *list = NULL; |
b28a02de | 3245 | |
7c3fbbdd | 3246 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3247 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3248 | BUG_ON(!n); |
b28a02de | 3249 | |
ca3b9b91 | 3250 | check_irq_off(); |
ce8eb6c4 | 3251 | spin_lock(&n->list_lock); |
f68f8ddd | 3252 | page = get_first_slab(n, false); |
7aa0d227 GT |
3253 | if (!page) |
3254 | goto must_grow; | |
b28a02de | 3255 | |
b28a02de | 3256 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3257 | |
3258 | STATS_INC_NODEALLOCS(cachep); | |
3259 | STATS_INC_ACTIVE(cachep); | |
3260 | STATS_SET_HIGH(cachep); | |
3261 | ||
8456a648 | 3262 | BUG_ON(page->active == cachep->num); |
b28a02de | 3263 | |
260b61dd | 3264 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3265 | n->free_objects--; |
b28a02de | 3266 | |
b03a017b | 3267 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3268 | |
ce8eb6c4 | 3269 | spin_unlock(&n->list_lock); |
b03a017b | 3270 | fixup_objfreelist_debug(cachep, &list); |
213b4695 | 3271 | return obj; |
e498be7d | 3272 | |
a737b3e2 | 3273 | must_grow: |
ce8eb6c4 | 3274 | spin_unlock(&n->list_lock); |
76b342bd | 3275 | page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid); |
213b4695 JK |
3276 | if (page) { |
3277 | /* This slab isn't counted yet so don't update free_objects */ | |
3278 | obj = slab_get_obj(cachep, page); | |
3279 | } | |
76b342bd | 3280 | cache_grow_end(cachep, page); |
1da177e4 | 3281 | |
213b4695 | 3282 | return obj ? obj : fallback_alloc(cachep, flags); |
e498be7d | 3283 | } |
8c8cc2c1 | 3284 | |
8c8cc2c1 | 3285 | static __always_inline void * |
48356303 | 3286 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3287 | unsigned long caller) |
8c8cc2c1 PE |
3288 | { |
3289 | unsigned long save_flags; | |
3290 | void *ptr; | |
7d6e6d09 | 3291 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3292 | |
dcce284a | 3293 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3294 | cachep = slab_pre_alloc_hook(cachep, flags); |
3295 | if (unlikely(!cachep)) | |
824ebef1 AM |
3296 | return NULL; |
3297 | ||
8c8cc2c1 PE |
3298 | cache_alloc_debugcheck_before(cachep, flags); |
3299 | local_irq_save(save_flags); | |
3300 | ||
eacbbae3 | 3301 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3302 | nodeid = slab_node; |
8c8cc2c1 | 3303 | |
18bf8541 | 3304 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3305 | /* Node not bootstrapped yet */ |
3306 | ptr = fallback_alloc(cachep, flags); | |
3307 | goto out; | |
3308 | } | |
3309 | ||
7d6e6d09 | 3310 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3311 | /* |
3312 | * Use the locally cached objects if possible. | |
3313 | * However ____cache_alloc does not allow fallback | |
3314 | * to other nodes. It may fail while we still have | |
3315 | * objects on other nodes available. | |
3316 | */ | |
3317 | ptr = ____cache_alloc(cachep, flags); | |
3318 | if (ptr) | |
3319 | goto out; | |
3320 | } | |
3321 | /* ___cache_alloc_node can fall back to other nodes */ | |
3322 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3323 | out: | |
3324 | local_irq_restore(save_flags); | |
3325 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3326 | ||
d5e3ed66 JDB |
3327 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3328 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3329 | |
d5e3ed66 | 3330 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3331 | return ptr; |
3332 | } | |
3333 | ||
3334 | static __always_inline void * | |
3335 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3336 | { | |
3337 | void *objp; | |
3338 | ||
2ad654bc | 3339 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3340 | objp = alternate_node_alloc(cache, flags); |
3341 | if (objp) | |
3342 | goto out; | |
3343 | } | |
3344 | objp = ____cache_alloc(cache, flags); | |
3345 | ||
3346 | /* | |
3347 | * We may just have run out of memory on the local node. | |
3348 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3349 | */ | |
7d6e6d09 LS |
3350 | if (!objp) |
3351 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3352 | |
3353 | out: | |
3354 | return objp; | |
3355 | } | |
3356 | #else | |
3357 | ||
3358 | static __always_inline void * | |
3359 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3360 | { | |
3361 | return ____cache_alloc(cachep, flags); | |
3362 | } | |
3363 | ||
3364 | #endif /* CONFIG_NUMA */ | |
3365 | ||
3366 | static __always_inline void * | |
48356303 | 3367 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3368 | { |
3369 | unsigned long save_flags; | |
3370 | void *objp; | |
3371 | ||
dcce284a | 3372 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3373 | cachep = slab_pre_alloc_hook(cachep, flags); |
3374 | if (unlikely(!cachep)) | |
824ebef1 AM |
3375 | return NULL; |
3376 | ||
8c8cc2c1 PE |
3377 | cache_alloc_debugcheck_before(cachep, flags); |
3378 | local_irq_save(save_flags); | |
3379 | objp = __do_cache_alloc(cachep, flags); | |
3380 | local_irq_restore(save_flags); | |
3381 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3382 | prefetchw(objp); | |
3383 | ||
d5e3ed66 JDB |
3384 | if (unlikely(flags & __GFP_ZERO) && objp) |
3385 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3386 | |
d5e3ed66 | 3387 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3388 | return objp; |
3389 | } | |
e498be7d CL |
3390 | |
3391 | /* | |
5f0985bb | 3392 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3393 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3394 | */ |
97654dfa JK |
3395 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3396 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3397 | { |
3398 | int i; | |
25c063fb | 3399 | struct kmem_cache_node *n = get_node(cachep, node); |
6052b788 JK |
3400 | struct page *page; |
3401 | ||
3402 | n->free_objects += nr_objects; | |
1da177e4 LT |
3403 | |
3404 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3405 | void *objp; |
8456a648 | 3406 | struct page *page; |
1da177e4 | 3407 | |
072bb0aa MG |
3408 | objp = objpp[i]; |
3409 | ||
8456a648 | 3410 | page = virt_to_head_page(objp); |
8456a648 | 3411 | list_del(&page->lru); |
ff69416e | 3412 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3413 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3414 | STATS_DEC_ACTIVE(cachep); |
1da177e4 LT |
3415 | |
3416 | /* fixup slab chains */ | |
f728b0a5 | 3417 | if (page->active == 0) { |
6052b788 | 3418 | list_add(&page->lru, &n->slabs_free); |
f728b0a5 | 3419 | n->free_slabs++; |
f728b0a5 | 3420 | } else { |
1da177e4 LT |
3421 | /* Unconditionally move a slab to the end of the |
3422 | * partial list on free - maximum time for the | |
3423 | * other objects to be freed, too. | |
3424 | */ | |
8456a648 | 3425 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3426 | } |
3427 | } | |
6052b788 JK |
3428 | |
3429 | while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) { | |
3430 | n->free_objects -= cachep->num; | |
3431 | ||
3432 | page = list_last_entry(&n->slabs_free, struct page, lru); | |
de24baec | 3433 | list_move(&page->lru, list); |
f728b0a5 | 3434 | n->free_slabs--; |
bf00bd34 | 3435 | n->total_slabs--; |
6052b788 | 3436 | } |
1da177e4 LT |
3437 | } |
3438 | ||
343e0d7a | 3439 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3440 | { |
3441 | int batchcount; | |
ce8eb6c4 | 3442 | struct kmem_cache_node *n; |
7d6e6d09 | 3443 | int node = numa_mem_id(); |
97654dfa | 3444 | LIST_HEAD(list); |
1da177e4 LT |
3445 | |
3446 | batchcount = ac->batchcount; | |
260b61dd | 3447 | |
1da177e4 | 3448 | check_irq_off(); |
18bf8541 | 3449 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3450 | spin_lock(&n->list_lock); |
3451 | if (n->shared) { | |
3452 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3453 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3454 | if (max) { |
3455 | if (batchcount > max) | |
3456 | batchcount = max; | |
e498be7d | 3457 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3458 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3459 | shared_array->avail += batchcount; |
3460 | goto free_done; | |
3461 | } | |
3462 | } | |
3463 | ||
97654dfa | 3464 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3465 | free_done: |
1da177e4 LT |
3466 | #if STATS |
3467 | { | |
3468 | int i = 0; | |
73c0219d | 3469 | struct page *page; |
1da177e4 | 3470 | |
73c0219d | 3471 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3472 | BUG_ON(page->active); |
1da177e4 LT |
3473 | |
3474 | i++; | |
1da177e4 LT |
3475 | } |
3476 | STATS_SET_FREEABLE(cachep, i); | |
3477 | } | |
3478 | #endif | |
ce8eb6c4 | 3479 | spin_unlock(&n->list_lock); |
97654dfa | 3480 | slabs_destroy(cachep, &list); |
1da177e4 | 3481 | ac->avail -= batchcount; |
a737b3e2 | 3482 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3483 | } |
3484 | ||
3485 | /* | |
a737b3e2 AM |
3486 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3487 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3488 | */ |
a947eb95 | 3489 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3490 | unsigned long caller) |
1da177e4 | 3491 | { |
55834c59 AP |
3492 | /* Put the object into the quarantine, don't touch it for now. */ |
3493 | if (kasan_slab_free(cachep, objp)) | |
3494 | return; | |
3495 | ||
3496 | ___cache_free(cachep, objp, caller); | |
3497 | } | |
1da177e4 | 3498 | |
55834c59 AP |
3499 | void ___cache_free(struct kmem_cache *cachep, void *objp, |
3500 | unsigned long caller) | |
3501 | { | |
3502 | struct array_cache *ac = cpu_cache_get(cachep); | |
7ed2f9e6 | 3503 | |
1da177e4 | 3504 | check_irq_off(); |
d5cff635 | 3505 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3506 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3507 | |
1807a1aa SS |
3508 | /* |
3509 | * Skip calling cache_free_alien() when the platform is not numa. | |
3510 | * This will avoid cache misses that happen while accessing slabp (which | |
3511 | * is per page memory reference) to get nodeid. Instead use a global | |
3512 | * variable to skip the call, which is mostly likely to be present in | |
3513 | * the cache. | |
3514 | */ | |
b6e68bc1 | 3515 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3516 | return; |
3517 | ||
3d880194 | 3518 | if (ac->avail < ac->limit) { |
1da177e4 | 3519 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3520 | } else { |
3521 | STATS_INC_FREEMISS(cachep); | |
3522 | cache_flusharray(cachep, ac); | |
1da177e4 | 3523 | } |
42c8c99c | 3524 | |
f68f8ddd JK |
3525 | if (sk_memalloc_socks()) { |
3526 | struct page *page = virt_to_head_page(objp); | |
3527 | ||
3528 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3529 | cache_free_pfmemalloc(cachep, page, objp); | |
3530 | return; | |
3531 | } | |
3532 | } | |
3533 | ||
3534 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3535 | } |
3536 | ||
3537 | /** | |
3538 | * kmem_cache_alloc - Allocate an object | |
3539 | * @cachep: The cache to allocate from. | |
3540 | * @flags: See kmalloc(). | |
3541 | * | |
3542 | * Allocate an object from this cache. The flags are only relevant | |
3543 | * if the cache has no available objects. | |
3544 | */ | |
343e0d7a | 3545 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3546 | { |
48356303 | 3547 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3548 | |
505f5dcb | 3549 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3550 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3551 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3552 | |
3553 | return ret; | |
1da177e4 LT |
3554 | } |
3555 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3556 | ||
7b0501dd JDB |
3557 | static __always_inline void |
3558 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3559 | size_t size, void **p, unsigned long caller) | |
3560 | { | |
3561 | size_t i; | |
3562 | ||
3563 | for (i = 0; i < size; i++) | |
3564 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3565 | } | |
3566 | ||
865762a8 | 3567 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3568 | void **p) |
484748f0 | 3569 | { |
2a777eac JDB |
3570 | size_t i; |
3571 | ||
3572 | s = slab_pre_alloc_hook(s, flags); | |
3573 | if (!s) | |
3574 | return 0; | |
3575 | ||
3576 | cache_alloc_debugcheck_before(s, flags); | |
3577 | ||
3578 | local_irq_disable(); | |
3579 | for (i = 0; i < size; i++) { | |
3580 | void *objp = __do_cache_alloc(s, flags); | |
3581 | ||
2a777eac JDB |
3582 | if (unlikely(!objp)) |
3583 | goto error; | |
3584 | p[i] = objp; | |
3585 | } | |
3586 | local_irq_enable(); | |
3587 | ||
7b0501dd JDB |
3588 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3589 | ||
2a777eac JDB |
3590 | /* Clear memory outside IRQ disabled section */ |
3591 | if (unlikely(flags & __GFP_ZERO)) | |
3592 | for (i = 0; i < size; i++) | |
3593 | memset(p[i], 0, s->object_size); | |
3594 | ||
3595 | slab_post_alloc_hook(s, flags, size, p); | |
3596 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3597 | return size; | |
3598 | error: | |
3599 | local_irq_enable(); | |
7b0501dd | 3600 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3601 | slab_post_alloc_hook(s, flags, i, p); |
3602 | __kmem_cache_free_bulk(s, i, p); | |
3603 | return 0; | |
484748f0 CL |
3604 | } |
3605 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3606 | ||
0f24f128 | 3607 | #ifdef CONFIG_TRACING |
85beb586 | 3608 | void * |
4052147c | 3609 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3610 | { |
85beb586 SR |
3611 | void *ret; |
3612 | ||
48356303 | 3613 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3614 | |
505f5dcb | 3615 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3616 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3617 | size, cachep->size, flags); |
85beb586 | 3618 | return ret; |
36555751 | 3619 | } |
85beb586 | 3620 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3621 | #endif |
3622 | ||
1da177e4 | 3623 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3624 | /** |
3625 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3626 | * @cachep: The cache to allocate from. | |
3627 | * @flags: See kmalloc(). | |
3628 | * @nodeid: node number of the target node. | |
3629 | * | |
3630 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3631 | * node, which can improve the performance for cpu bound structures. | |
3632 | * | |
3633 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3634 | */ | |
8b98c169 CH |
3635 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3636 | { | |
48356303 | 3637 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3638 | |
505f5dcb | 3639 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3640 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3641 | cachep->object_size, cachep->size, |
ca2b84cb | 3642 | flags, nodeid); |
36555751 EGM |
3643 | |
3644 | return ret; | |
8b98c169 | 3645 | } |
1da177e4 LT |
3646 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3647 | ||
0f24f128 | 3648 | #ifdef CONFIG_TRACING |
4052147c | 3649 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3650 | gfp_t flags, |
4052147c EG |
3651 | int nodeid, |
3652 | size_t size) | |
36555751 | 3653 | { |
85beb586 SR |
3654 | void *ret; |
3655 | ||
592f4145 | 3656 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3657 | |
3658 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3659 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3660 | size, cachep->size, |
85beb586 SR |
3661 | flags, nodeid); |
3662 | return ret; | |
36555751 | 3663 | } |
85beb586 | 3664 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3665 | #endif |
3666 | ||
8b98c169 | 3667 | static __always_inline void * |
7c0cb9c6 | 3668 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3669 | { |
343e0d7a | 3670 | struct kmem_cache *cachep; |
7ed2f9e6 | 3671 | void *ret; |
97e2bde4 | 3672 | |
333cfbd3 DV |
3673 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3674 | return NULL; | |
2c59dd65 | 3675 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3676 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3677 | return cachep; | |
7ed2f9e6 | 3678 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3679 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3680 | |
3681 | return ret; | |
97e2bde4 | 3682 | } |
8b98c169 | 3683 | |
8b98c169 CH |
3684 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3685 | { | |
7c0cb9c6 | 3686 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3687 | } |
dbe5e69d | 3688 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3689 | |
3690 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3691 | int node, unsigned long caller) |
8b98c169 | 3692 | { |
7c0cb9c6 | 3693 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3694 | } |
3695 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3696 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3697 | |
3698 | /** | |
800590f5 | 3699 | * __do_kmalloc - allocate memory |
1da177e4 | 3700 | * @size: how many bytes of memory are required. |
800590f5 | 3701 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3702 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3703 | */ |
7fd6b141 | 3704 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3705 | unsigned long caller) |
1da177e4 | 3706 | { |
343e0d7a | 3707 | struct kmem_cache *cachep; |
36555751 | 3708 | void *ret; |
1da177e4 | 3709 | |
333cfbd3 DV |
3710 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3711 | return NULL; | |
2c59dd65 | 3712 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3713 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3714 | return cachep; | |
48356303 | 3715 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3716 | |
505f5dcb | 3717 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3718 | trace_kmalloc(caller, ret, |
3b0efdfa | 3719 | size, cachep->size, flags); |
36555751 EGM |
3720 | |
3721 | return ret; | |
7fd6b141 PE |
3722 | } |
3723 | ||
7fd6b141 PE |
3724 | void *__kmalloc(size_t size, gfp_t flags) |
3725 | { | |
7c0cb9c6 | 3726 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3727 | } |
3728 | EXPORT_SYMBOL(__kmalloc); | |
3729 | ||
ce71e27c | 3730 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3731 | { |
7c0cb9c6 | 3732 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3733 | } |
3734 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3735 | |
1da177e4 LT |
3736 | /** |
3737 | * kmem_cache_free - Deallocate an object | |
3738 | * @cachep: The cache the allocation was from. | |
3739 | * @objp: The previously allocated object. | |
3740 | * | |
3741 | * Free an object which was previously allocated from this | |
3742 | * cache. | |
3743 | */ | |
343e0d7a | 3744 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3745 | { |
3746 | unsigned long flags; | |
b9ce5ef4 GC |
3747 | cachep = cache_from_obj(cachep, objp); |
3748 | if (!cachep) | |
3749 | return; | |
1da177e4 LT |
3750 | |
3751 | local_irq_save(flags); | |
d97d476b | 3752 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3753 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3754 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3755 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3756 | local_irq_restore(flags); |
36555751 | 3757 | |
ca2b84cb | 3758 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3759 | } |
3760 | EXPORT_SYMBOL(kmem_cache_free); | |
3761 | ||
e6cdb58d JDB |
3762 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3763 | { | |
3764 | struct kmem_cache *s; | |
3765 | size_t i; | |
3766 | ||
3767 | local_irq_disable(); | |
3768 | for (i = 0; i < size; i++) { | |
3769 | void *objp = p[i]; | |
3770 | ||
ca257195 JDB |
3771 | if (!orig_s) /* called via kfree_bulk */ |
3772 | s = virt_to_cache(objp); | |
3773 | else | |
3774 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3775 | |
3776 | debug_check_no_locks_freed(objp, s->object_size); | |
3777 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3778 | debug_check_no_obj_freed(objp, s->object_size); | |
3779 | ||
3780 | __cache_free(s, objp, _RET_IP_); | |
3781 | } | |
3782 | local_irq_enable(); | |
3783 | ||
3784 | /* FIXME: add tracing */ | |
3785 | } | |
3786 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3787 | ||
1da177e4 LT |
3788 | /** |
3789 | * kfree - free previously allocated memory | |
3790 | * @objp: pointer returned by kmalloc. | |
3791 | * | |
80e93eff PE |
3792 | * If @objp is NULL, no operation is performed. |
3793 | * | |
1da177e4 LT |
3794 | * Don't free memory not originally allocated by kmalloc() |
3795 | * or you will run into trouble. | |
3796 | */ | |
3797 | void kfree(const void *objp) | |
3798 | { | |
343e0d7a | 3799 | struct kmem_cache *c; |
1da177e4 LT |
3800 | unsigned long flags; |
3801 | ||
2121db74 PE |
3802 | trace_kfree(_RET_IP_, objp); |
3803 | ||
6cb8f913 | 3804 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3805 | return; |
3806 | local_irq_save(flags); | |
3807 | kfree_debugcheck(objp); | |
6ed5eb22 | 3808 | c = virt_to_cache(objp); |
8c138bc0 CL |
3809 | debug_check_no_locks_freed(objp, c->object_size); |
3810 | ||
3811 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3812 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3813 | local_irq_restore(flags); |
3814 | } | |
3815 | EXPORT_SYMBOL(kfree); | |
3816 | ||
e498be7d | 3817 | /* |
ce8eb6c4 | 3818 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3819 | */ |
c3d332b6 | 3820 | static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d | 3821 | { |
c3d332b6 | 3822 | int ret; |
e498be7d | 3823 | int node; |
ce8eb6c4 | 3824 | struct kmem_cache_node *n; |
e498be7d | 3825 | |
9c09a95c | 3826 | for_each_online_node(node) { |
c3d332b6 JK |
3827 | ret = setup_kmem_cache_node(cachep, node, gfp, true); |
3828 | if (ret) | |
e498be7d CL |
3829 | goto fail; |
3830 | ||
e498be7d | 3831 | } |
c3d332b6 | 3832 | |
cafeb02e | 3833 | return 0; |
0718dc2a | 3834 | |
a737b3e2 | 3835 | fail: |
3b0efdfa | 3836 | if (!cachep->list.next) { |
0718dc2a CL |
3837 | /* Cache is not active yet. Roll back what we did */ |
3838 | node--; | |
3839 | while (node >= 0) { | |
18bf8541 CL |
3840 | n = get_node(cachep, node); |
3841 | if (n) { | |
ce8eb6c4 CL |
3842 | kfree(n->shared); |
3843 | free_alien_cache(n->alien); | |
3844 | kfree(n); | |
6a67368c | 3845 | cachep->node[node] = NULL; |
0718dc2a CL |
3846 | } |
3847 | node--; | |
3848 | } | |
3849 | } | |
cafeb02e | 3850 | return -ENOMEM; |
e498be7d CL |
3851 | } |
3852 | ||
18004c5d | 3853 | /* Always called with the slab_mutex held */ |
943a451a | 3854 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3855 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3856 | { |
bf0dea23 JK |
3857 | struct array_cache __percpu *cpu_cache, *prev; |
3858 | int cpu; | |
1da177e4 | 3859 | |
bf0dea23 JK |
3860 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3861 | if (!cpu_cache) | |
d2e7b7d0 SS |
3862 | return -ENOMEM; |
3863 | ||
bf0dea23 JK |
3864 | prev = cachep->cpu_cache; |
3865 | cachep->cpu_cache = cpu_cache; | |
a87c75fb GT |
3866 | /* |
3867 | * Without a previous cpu_cache there's no need to synchronize remote | |
3868 | * cpus, so skip the IPIs. | |
3869 | */ | |
3870 | if (prev) | |
3871 | kick_all_cpus_sync(); | |
e498be7d | 3872 | |
1da177e4 | 3873 | check_irq_on(); |
1da177e4 LT |
3874 | cachep->batchcount = batchcount; |
3875 | cachep->limit = limit; | |
e498be7d | 3876 | cachep->shared = shared; |
1da177e4 | 3877 | |
bf0dea23 | 3878 | if (!prev) |
c3d332b6 | 3879 | goto setup_node; |
bf0dea23 JK |
3880 | |
3881 | for_each_online_cpu(cpu) { | |
97654dfa | 3882 | LIST_HEAD(list); |
18bf8541 CL |
3883 | int node; |
3884 | struct kmem_cache_node *n; | |
bf0dea23 | 3885 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3886 | |
bf0dea23 | 3887 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3888 | n = get_node(cachep, node); |
3889 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3890 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3891 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3892 | slabs_destroy(cachep, &list); |
1da177e4 | 3893 | } |
bf0dea23 JK |
3894 | free_percpu(prev); |
3895 | ||
c3d332b6 JK |
3896 | setup_node: |
3897 | return setup_kmem_cache_nodes(cachep, gfp); | |
1da177e4 LT |
3898 | } |
3899 | ||
943a451a GC |
3900 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3901 | int batchcount, int shared, gfp_t gfp) | |
3902 | { | |
3903 | int ret; | |
426589f5 | 3904 | struct kmem_cache *c; |
943a451a GC |
3905 | |
3906 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3907 | ||
3908 | if (slab_state < FULL) | |
3909 | return ret; | |
3910 | ||
3911 | if ((ret < 0) || !is_root_cache(cachep)) | |
3912 | return ret; | |
3913 | ||
426589f5 VD |
3914 | lockdep_assert_held(&slab_mutex); |
3915 | for_each_memcg_cache(c, cachep) { | |
3916 | /* return value determined by the root cache only */ | |
3917 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3918 | } |
3919 | ||
3920 | return ret; | |
3921 | } | |
3922 | ||
18004c5d | 3923 | /* Called with slab_mutex held always */ |
83b519e8 | 3924 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3925 | { |
3926 | int err; | |
943a451a GC |
3927 | int limit = 0; |
3928 | int shared = 0; | |
3929 | int batchcount = 0; | |
3930 | ||
7c00fce9 | 3931 | err = cache_random_seq_create(cachep, cachep->num, gfp); |
c7ce4f60 TG |
3932 | if (err) |
3933 | goto end; | |
3934 | ||
943a451a GC |
3935 | if (!is_root_cache(cachep)) { |
3936 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3937 | limit = root->limit; | |
3938 | shared = root->shared; | |
3939 | batchcount = root->batchcount; | |
3940 | } | |
1da177e4 | 3941 | |
943a451a GC |
3942 | if (limit && shared && batchcount) |
3943 | goto skip_setup; | |
a737b3e2 AM |
3944 | /* |
3945 | * The head array serves three purposes: | |
1da177e4 LT |
3946 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3947 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3948 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3949 | * bufctl chains: array operations are cheaper. |
3950 | * The numbers are guessed, we should auto-tune as described by | |
3951 | * Bonwick. | |
3952 | */ | |
3b0efdfa | 3953 | if (cachep->size > 131072) |
1da177e4 | 3954 | limit = 1; |
3b0efdfa | 3955 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3956 | limit = 8; |
3b0efdfa | 3957 | else if (cachep->size > 1024) |
1da177e4 | 3958 | limit = 24; |
3b0efdfa | 3959 | else if (cachep->size > 256) |
1da177e4 LT |
3960 | limit = 54; |
3961 | else | |
3962 | limit = 120; | |
3963 | ||
a737b3e2 AM |
3964 | /* |
3965 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3966 | * allocation behaviour: Most allocs on one cpu, most free operations |
3967 | * on another cpu. For these cases, an efficient object passing between | |
3968 | * cpus is necessary. This is provided by a shared array. The array | |
3969 | * replaces Bonwick's magazine layer. | |
3970 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3971 | * to a larger limit. Thus disabled by default. | |
3972 | */ | |
3973 | shared = 0; | |
3b0efdfa | 3974 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3975 | shared = 8; |
1da177e4 LT |
3976 | |
3977 | #if DEBUG | |
a737b3e2 AM |
3978 | /* |
3979 | * With debugging enabled, large batchcount lead to excessively long | |
3980 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3981 | */ |
3982 | if (limit > 32) | |
3983 | limit = 32; | |
3984 | #endif | |
943a451a GC |
3985 | batchcount = (limit + 1) / 2; |
3986 | skip_setup: | |
3987 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
c7ce4f60 | 3988 | end: |
1da177e4 | 3989 | if (err) |
1170532b | 3990 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3991 | cachep->name, -err); |
2ed3a4ef | 3992 | return err; |
1da177e4 LT |
3993 | } |
3994 | ||
1b55253a | 3995 | /* |
ce8eb6c4 CL |
3996 | * Drain an array if it contains any elements taking the node lock only if |
3997 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3998 | * if drain_array() is used on the shared array. |
1b55253a | 3999 | */ |
ce8eb6c4 | 4000 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 4001 | struct array_cache *ac, int node) |
1da177e4 | 4002 | { |
97654dfa | 4003 | LIST_HEAD(list); |
18726ca8 JK |
4004 | |
4005 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
4006 | check_mutex_acquired(); | |
1da177e4 | 4007 | |
1b55253a CL |
4008 | if (!ac || !ac->avail) |
4009 | return; | |
18726ca8 JK |
4010 | |
4011 | if (ac->touched) { | |
1da177e4 | 4012 | ac->touched = 0; |
18726ca8 | 4013 | return; |
1da177e4 | 4014 | } |
18726ca8 JK |
4015 | |
4016 | spin_lock_irq(&n->list_lock); | |
4017 | drain_array_locked(cachep, ac, node, false, &list); | |
4018 | spin_unlock_irq(&n->list_lock); | |
4019 | ||
4020 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
4021 | } |
4022 | ||
4023 | /** | |
4024 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4025 | * @w: work descriptor |
1da177e4 LT |
4026 | * |
4027 | * Called from workqueue/eventd every few seconds. | |
4028 | * Purpose: | |
4029 | * - clear the per-cpu caches for this CPU. | |
4030 | * - return freeable pages to the main free memory pool. | |
4031 | * | |
a737b3e2 AM |
4032 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4033 | * again on the next iteration. | |
1da177e4 | 4034 | */ |
7c5cae36 | 4035 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4036 | { |
7a7c381d | 4037 | struct kmem_cache *searchp; |
ce8eb6c4 | 4038 | struct kmem_cache_node *n; |
7d6e6d09 | 4039 | int node = numa_mem_id(); |
bf6aede7 | 4040 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4041 | |
18004c5d | 4042 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4043 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4044 | goto out; |
1da177e4 | 4045 | |
18004c5d | 4046 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4047 | check_irq_on(); |
4048 | ||
35386e3b | 4049 | /* |
ce8eb6c4 | 4050 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4051 | * have established with reasonable certainty that |
4052 | * we can do some work if the lock was obtained. | |
4053 | */ | |
18bf8541 | 4054 | n = get_node(searchp, node); |
35386e3b | 4055 | |
ce8eb6c4 | 4056 | reap_alien(searchp, n); |
1da177e4 | 4057 | |
18726ca8 | 4058 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 4059 | |
35386e3b CL |
4060 | /* |
4061 | * These are racy checks but it does not matter | |
4062 | * if we skip one check or scan twice. | |
4063 | */ | |
ce8eb6c4 | 4064 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4065 | goto next; |
1da177e4 | 4066 | |
5f0985bb | 4067 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4068 | |
18726ca8 | 4069 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 4070 | |
ce8eb6c4 CL |
4071 | if (n->free_touched) |
4072 | n->free_touched = 0; | |
ed11d9eb CL |
4073 | else { |
4074 | int freed; | |
1da177e4 | 4075 | |
ce8eb6c4 | 4076 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4077 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4078 | STATS_ADD_REAPED(searchp, freed); | |
4079 | } | |
35386e3b | 4080 | next: |
1da177e4 LT |
4081 | cond_resched(); |
4082 | } | |
4083 | check_irq_on(); | |
18004c5d | 4084 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4085 | next_reap_node(); |
7c5cae36 | 4086 | out: |
a737b3e2 | 4087 | /* Set up the next iteration */ |
7f251b20 VB |
4088 | schedule_delayed_work_on(smp_processor_id(), work, |
4089 | round_jiffies_relative(REAPTIMEOUT_AC)); | |
1da177e4 LT |
4090 | } |
4091 | ||
0d7561c6 | 4092 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4093 | { |
f728b0a5 | 4094 | unsigned long active_objs, num_objs, active_slabs; |
bf00bd34 DR |
4095 | unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0; |
4096 | unsigned long free_slabs = 0; | |
e498be7d | 4097 | int node; |
ce8eb6c4 | 4098 | struct kmem_cache_node *n; |
1da177e4 | 4099 | |
18bf8541 | 4100 | for_each_kmem_cache_node(cachep, node, n) { |
ca3b9b91 | 4101 | check_irq_on(); |
ce8eb6c4 | 4102 | spin_lock_irq(&n->list_lock); |
e498be7d | 4103 | |
bf00bd34 DR |
4104 | total_slabs += n->total_slabs; |
4105 | free_slabs += n->free_slabs; | |
f728b0a5 | 4106 | free_objs += n->free_objects; |
07a63c41 | 4107 | |
ce8eb6c4 CL |
4108 | if (n->shared) |
4109 | shared_avail += n->shared->avail; | |
e498be7d | 4110 | |
ce8eb6c4 | 4111 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4112 | } |
bf00bd34 DR |
4113 | num_objs = total_slabs * cachep->num; |
4114 | active_slabs = total_slabs - free_slabs; | |
f728b0a5 | 4115 | active_objs = num_objs - free_objs; |
1da177e4 | 4116 | |
0d7561c6 GC |
4117 | sinfo->active_objs = active_objs; |
4118 | sinfo->num_objs = num_objs; | |
4119 | sinfo->active_slabs = active_slabs; | |
bf00bd34 | 4120 | sinfo->num_slabs = total_slabs; |
0d7561c6 GC |
4121 | sinfo->shared_avail = shared_avail; |
4122 | sinfo->limit = cachep->limit; | |
4123 | sinfo->batchcount = cachep->batchcount; | |
4124 | sinfo->shared = cachep->shared; | |
4125 | sinfo->objects_per_slab = cachep->num; | |
4126 | sinfo->cache_order = cachep->gfporder; | |
4127 | } | |
4128 | ||
4129 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4130 | { | |
1da177e4 | 4131 | #if STATS |
ce8eb6c4 | 4132 | { /* node stats */ |
1da177e4 LT |
4133 | unsigned long high = cachep->high_mark; |
4134 | unsigned long allocs = cachep->num_allocations; | |
4135 | unsigned long grown = cachep->grown; | |
4136 | unsigned long reaped = cachep->reaped; | |
4137 | unsigned long errors = cachep->errors; | |
4138 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4139 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4140 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4141 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4142 | |
756a025f | 4143 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4144 | allocs, high, grown, |
4145 | reaped, errors, max_freeable, node_allocs, | |
4146 | node_frees, overflows); | |
1da177e4 LT |
4147 | } |
4148 | /* cpu stats */ | |
4149 | { | |
4150 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4151 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4152 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4153 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4154 | ||
4155 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4156 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4157 | } |
4158 | #endif | |
1da177e4 LT |
4159 | } |
4160 | ||
1da177e4 LT |
4161 | #define MAX_SLABINFO_WRITE 128 |
4162 | /** | |
4163 | * slabinfo_write - Tuning for the slab allocator | |
4164 | * @file: unused | |
4165 | * @buffer: user buffer | |
4166 | * @count: data length | |
4167 | * @ppos: unused | |
4168 | */ | |
b7454ad3 | 4169 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4170 | size_t count, loff_t *ppos) |
1da177e4 | 4171 | { |
b28a02de | 4172 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4173 | int limit, batchcount, shared, res; |
7a7c381d | 4174 | struct kmem_cache *cachep; |
b28a02de | 4175 | |
1da177e4 LT |
4176 | if (count > MAX_SLABINFO_WRITE) |
4177 | return -EINVAL; | |
4178 | if (copy_from_user(&kbuf, buffer, count)) | |
4179 | return -EFAULT; | |
b28a02de | 4180 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4181 | |
4182 | tmp = strchr(kbuf, ' '); | |
4183 | if (!tmp) | |
4184 | return -EINVAL; | |
4185 | *tmp = '\0'; | |
4186 | tmp++; | |
4187 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4188 | return -EINVAL; | |
4189 | ||
4190 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4191 | mutex_lock(&slab_mutex); |
1da177e4 | 4192 | res = -EINVAL; |
18004c5d | 4193 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4194 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4195 | if (limit < 1 || batchcount < 1 || |
4196 | batchcount > limit || shared < 0) { | |
e498be7d | 4197 | res = 0; |
1da177e4 | 4198 | } else { |
e498be7d | 4199 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4200 | batchcount, shared, |
4201 | GFP_KERNEL); | |
1da177e4 LT |
4202 | } |
4203 | break; | |
4204 | } | |
4205 | } | |
18004c5d | 4206 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4207 | if (res >= 0) |
4208 | res = count; | |
4209 | return res; | |
4210 | } | |
871751e2 AV |
4211 | |
4212 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4213 | ||
871751e2 AV |
4214 | static inline int add_caller(unsigned long *n, unsigned long v) |
4215 | { | |
4216 | unsigned long *p; | |
4217 | int l; | |
4218 | if (!v) | |
4219 | return 1; | |
4220 | l = n[1]; | |
4221 | p = n + 2; | |
4222 | while (l) { | |
4223 | int i = l/2; | |
4224 | unsigned long *q = p + 2 * i; | |
4225 | if (*q == v) { | |
4226 | q[1]++; | |
4227 | return 1; | |
4228 | } | |
4229 | if (*q > v) { | |
4230 | l = i; | |
4231 | } else { | |
4232 | p = q + 2; | |
4233 | l -= i + 1; | |
4234 | } | |
4235 | } | |
4236 | if (++n[1] == n[0]) | |
4237 | return 0; | |
4238 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4239 | p[0] = v; | |
4240 | p[1] = 1; | |
4241 | return 1; | |
4242 | } | |
4243 | ||
8456a648 JK |
4244 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4245 | struct page *page) | |
871751e2 AV |
4246 | { |
4247 | void *p; | |
d31676df JK |
4248 | int i, j; |
4249 | unsigned long v; | |
b1cb0982 | 4250 | |
871751e2 AV |
4251 | if (n[0] == n[1]) |
4252 | return; | |
8456a648 | 4253 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4254 | bool active = true; |
4255 | ||
4256 | for (j = page->active; j < c->num; j++) { | |
4257 | if (get_free_obj(page, j) == i) { | |
4258 | active = false; | |
4259 | break; | |
4260 | } | |
4261 | } | |
4262 | ||
4263 | if (!active) | |
871751e2 | 4264 | continue; |
b1cb0982 | 4265 | |
d31676df JK |
4266 | /* |
4267 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4268 | * mapping is established when actual object allocation and | |
4269 | * we could mistakenly access the unmapped object in the cpu | |
4270 | * cache. | |
4271 | */ | |
4272 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4273 | continue; | |
4274 | ||
4275 | if (!add_caller(n, v)) | |
871751e2 AV |
4276 | return; |
4277 | } | |
4278 | } | |
4279 | ||
4280 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4281 | { | |
4282 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4283 | unsigned long offset, size; |
9281acea | 4284 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4285 | |
a5c43dae | 4286 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4287 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4288 | if (modname[0]) |
871751e2 AV |
4289 | seq_printf(m, " [%s]", modname); |
4290 | return; | |
4291 | } | |
4292 | #endif | |
85c3e4a5 | 4293 | seq_printf(m, "%px", (void *)address); |
871751e2 AV |
4294 | } |
4295 | ||
4296 | static int leaks_show(struct seq_file *m, void *p) | |
4297 | { | |
0672aa7c | 4298 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4299 | struct page *page; |
ce8eb6c4 | 4300 | struct kmem_cache_node *n; |
871751e2 | 4301 | const char *name; |
db845067 | 4302 | unsigned long *x = m->private; |
871751e2 AV |
4303 | int node; |
4304 | int i; | |
4305 | ||
4306 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4307 | return 0; | |
4308 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4309 | return 0; | |
4310 | ||
d31676df JK |
4311 | /* |
4312 | * Set store_user_clean and start to grab stored user information | |
4313 | * for all objects on this cache. If some alloc/free requests comes | |
4314 | * during the processing, information would be wrong so restart | |
4315 | * whole processing. | |
4316 | */ | |
4317 | do { | |
4318 | set_store_user_clean(cachep); | |
4319 | drain_cpu_caches(cachep); | |
4320 | ||
4321 | x[1] = 0; | |
871751e2 | 4322 | |
d31676df | 4323 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4324 | |
d31676df JK |
4325 | check_irq_on(); |
4326 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4327 | |
d31676df JK |
4328 | list_for_each_entry(page, &n->slabs_full, lru) |
4329 | handle_slab(x, cachep, page); | |
4330 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4331 | handle_slab(x, cachep, page); | |
4332 | spin_unlock_irq(&n->list_lock); | |
4333 | } | |
4334 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4335 | |
871751e2 | 4336 | name = cachep->name; |
db845067 | 4337 | if (x[0] == x[1]) { |
871751e2 | 4338 | /* Increase the buffer size */ |
18004c5d | 4339 | mutex_unlock(&slab_mutex); |
db845067 | 4340 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4341 | if (!m->private) { |
4342 | /* Too bad, we are really out */ | |
db845067 | 4343 | m->private = x; |
18004c5d | 4344 | mutex_lock(&slab_mutex); |
871751e2 AV |
4345 | return -ENOMEM; |
4346 | } | |
db845067 CL |
4347 | *(unsigned long *)m->private = x[0] * 2; |
4348 | kfree(x); | |
18004c5d | 4349 | mutex_lock(&slab_mutex); |
871751e2 AV |
4350 | /* Now make sure this entry will be retried */ |
4351 | m->count = m->size; | |
4352 | return 0; | |
4353 | } | |
db845067 CL |
4354 | for (i = 0; i < x[1]; i++) { |
4355 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4356 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4357 | seq_putc(m, '\n'); |
4358 | } | |
d2e7b7d0 | 4359 | |
871751e2 AV |
4360 | return 0; |
4361 | } | |
4362 | ||
a0ec95a8 | 4363 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4364 | .start = slab_start, |
276a2439 WL |
4365 | .next = slab_next, |
4366 | .stop = slab_stop, | |
871751e2 AV |
4367 | .show = leaks_show, |
4368 | }; | |
a0ec95a8 AD |
4369 | |
4370 | static int slabstats_open(struct inode *inode, struct file *file) | |
4371 | { | |
b208ce32 RJ |
4372 | unsigned long *n; |
4373 | ||
4374 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4375 | if (!n) | |
4376 | return -ENOMEM; | |
4377 | ||
4378 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4379 | ||
4380 | return 0; | |
a0ec95a8 AD |
4381 | } |
4382 | ||
4383 | static const struct file_operations proc_slabstats_operations = { | |
4384 | .open = slabstats_open, | |
4385 | .read = seq_read, | |
4386 | .llseek = seq_lseek, | |
4387 | .release = seq_release_private, | |
4388 | }; | |
4389 | #endif | |
4390 | ||
4391 | static int __init slab_proc_init(void) | |
4392 | { | |
4393 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4394 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4395 | #endif |
a0ec95a8 AD |
4396 | return 0; |
4397 | } | |
4398 | module_init(slab_proc_init); | |
1da177e4 | 4399 | |
04385fc5 KC |
4400 | #ifdef CONFIG_HARDENED_USERCOPY |
4401 | /* | |
4402 | * Rejects objects that are incorrectly sized. | |
4403 | * | |
4404 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4405 | * to indicate an error. | |
4406 | */ | |
4407 | const char *__check_heap_object(const void *ptr, unsigned long n, | |
4408 | struct page *page) | |
4409 | { | |
4410 | struct kmem_cache *cachep; | |
4411 | unsigned int objnr; | |
4412 | unsigned long offset; | |
4413 | ||
4414 | /* Find and validate object. */ | |
4415 | cachep = page->slab_cache; | |
4416 | objnr = obj_to_index(cachep, page, (void *)ptr); | |
4417 | BUG_ON(objnr >= cachep->num); | |
4418 | ||
4419 | /* Find offset within object. */ | |
4420 | offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep); | |
4421 | ||
4422 | /* Allow address range falling entirely within object size. */ | |
4423 | if (offset <= cachep->object_size && n <= cachep->object_size - offset) | |
4424 | return NULL; | |
4425 | ||
4426 | return cachep->name; | |
4427 | } | |
4428 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4429 | ||
00e145b6 MS |
4430 | /** |
4431 | * ksize - get the actual amount of memory allocated for a given object | |
4432 | * @objp: Pointer to the object | |
4433 | * | |
4434 | * kmalloc may internally round up allocations and return more memory | |
4435 | * than requested. ksize() can be used to determine the actual amount of | |
4436 | * memory allocated. The caller may use this additional memory, even though | |
4437 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4438 | * The caller must guarantee that objp points to a valid object previously | |
4439 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4440 | * must not be freed during the duration of the call. | |
4441 | */ | |
fd76bab2 | 4442 | size_t ksize(const void *objp) |
1da177e4 | 4443 | { |
7ed2f9e6 AP |
4444 | size_t size; |
4445 | ||
ef8b4520 CL |
4446 | BUG_ON(!objp); |
4447 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4448 | return 0; |
1da177e4 | 4449 | |
7ed2f9e6 AP |
4450 | size = virt_to_cache(objp)->object_size; |
4451 | /* We assume that ksize callers could use the whole allocated area, | |
4452 | * so we need to unpoison this area. | |
4453 | */ | |
4ebb31a4 | 4454 | kasan_unpoison_shadow(objp, size); |
7ed2f9e6 AP |
4455 | |
4456 | return size; | |
1da177e4 | 4457 | } |
b1aabecd | 4458 | EXPORT_SYMBOL(ksize); |