]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
f315e3fa JK |
160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
162 | ||
163 | #if FREELIST_BYTE_INDEX | |
164 | typedef unsigned char freelist_idx_t; | |
165 | #else | |
166 | typedef unsigned short freelist_idx_t; | |
167 | #endif | |
168 | ||
30321c7b | 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 170 | |
1da177e4 LT |
171 | /* |
172 | * struct array_cache | |
173 | * | |
1da177e4 LT |
174 | * Purpose: |
175 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
176 | * - reduce the number of linked list operations | |
177 | * - reduce spinlock operations | |
178 | * | |
179 | * The limit is stored in the per-cpu structure to reduce the data cache | |
180 | * footprint. | |
181 | * | |
182 | */ | |
183 | struct array_cache { | |
184 | unsigned int avail; | |
185 | unsigned int limit; | |
186 | unsigned int batchcount; | |
187 | unsigned int touched; | |
bda5b655 | 188 | void *entry[]; /* |
a737b3e2 AM |
189 | * Must have this definition in here for the proper |
190 | * alignment of array_cache. Also simplifies accessing | |
191 | * the entries. | |
a737b3e2 | 192 | */ |
1da177e4 LT |
193 | }; |
194 | ||
c8522a3a JK |
195 | struct alien_cache { |
196 | spinlock_t lock; | |
197 | struct array_cache ac; | |
198 | }; | |
199 | ||
e498be7d CL |
200 | /* |
201 | * Need this for bootstrapping a per node allocator. | |
202 | */ | |
bf0dea23 | 203 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 204 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 205 | #define CACHE_CACHE 0 |
bf0dea23 | 206 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 207 | |
ed11d9eb | 208 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 209 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 210 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
211 | int node, struct list_head *list); |
212 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 213 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 214 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 215 | |
76b342bd JK |
216 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
217 | void **list); | |
218 | static inline void fixup_slab_list(struct kmem_cache *cachep, | |
219 | struct kmem_cache_node *n, struct page *page, | |
220 | void **list); | |
e0a42726 IM |
221 | static int slab_early_init = 1; |
222 | ||
ce8eb6c4 | 223 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 224 | |
ce8eb6c4 | 225 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
226 | { |
227 | INIT_LIST_HEAD(&parent->slabs_full); | |
228 | INIT_LIST_HEAD(&parent->slabs_partial); | |
229 | INIT_LIST_HEAD(&parent->slabs_free); | |
bf00bd34 | 230 | parent->total_slabs = 0; |
f728b0a5 | 231 | parent->free_slabs = 0; |
e498be7d CL |
232 | parent->shared = NULL; |
233 | parent->alien = NULL; | |
2e1217cf | 234 | parent->colour_next = 0; |
e498be7d CL |
235 | spin_lock_init(&parent->list_lock); |
236 | parent->free_objects = 0; | |
237 | parent->free_touched = 0; | |
238 | } | |
239 | ||
a737b3e2 AM |
240 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
241 | do { \ | |
242 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 243 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
244 | } while (0) |
245 | ||
a737b3e2 AM |
246 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
247 | do { \ | |
e498be7d CL |
248 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
249 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
250 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
251 | } while (0) | |
1da177e4 | 252 | |
b03a017b | 253 | #define CFLGS_OBJFREELIST_SLAB (0x40000000UL) |
1da177e4 | 254 | #define CFLGS_OFF_SLAB (0x80000000UL) |
b03a017b | 255 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
256 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
257 | ||
258 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
259 | /* |
260 | * Optimization question: fewer reaps means less probability for unnessary | |
261 | * cpucache drain/refill cycles. | |
1da177e4 | 262 | * |
dc6f3f27 | 263 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
264 | * which could lock up otherwise freeable slabs. |
265 | */ | |
5f0985bb JZ |
266 | #define REAPTIMEOUT_AC (2*HZ) |
267 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
268 | |
269 | #if STATS | |
270 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
271 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
272 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
273 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 274 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
275 | #define STATS_SET_HIGH(x) \ |
276 | do { \ | |
277 | if ((x)->num_active > (x)->high_mark) \ | |
278 | (x)->high_mark = (x)->num_active; \ | |
279 | } while (0) | |
1da177e4 LT |
280 | #define STATS_INC_ERR(x) ((x)->errors++) |
281 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 282 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 283 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
284 | #define STATS_SET_FREEABLE(x, i) \ |
285 | do { \ | |
286 | if ((x)->max_freeable < i) \ | |
287 | (x)->max_freeable = i; \ | |
288 | } while (0) | |
1da177e4 LT |
289 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
290 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
291 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
292 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
293 | #else | |
294 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
295 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
296 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
297 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 298 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
299 | #define STATS_SET_HIGH(x) do { } while (0) |
300 | #define STATS_INC_ERR(x) do { } while (0) | |
301 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 302 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 303 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 304 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
305 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
306 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
307 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
308 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
309 | #endif | |
310 | ||
311 | #if DEBUG | |
1da177e4 | 312 | |
a737b3e2 AM |
313 | /* |
314 | * memory layout of objects: | |
1da177e4 | 315 | * 0 : objp |
3dafccf2 | 316 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
317 | * the end of an object is aligned with the end of the real |
318 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 319 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 320 | * redzone word. |
3dafccf2 | 321 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
322 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
323 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 324 | * [BYTES_PER_WORD long] |
1da177e4 | 325 | */ |
343e0d7a | 326 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 327 | { |
3dafccf2 | 328 | return cachep->obj_offset; |
1da177e4 LT |
329 | } |
330 | ||
b46b8f19 | 331 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
332 | { |
333 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
334 | return (unsigned long long*) (objp + obj_offset(cachep) - |
335 | sizeof(unsigned long long)); | |
1da177e4 LT |
336 | } |
337 | ||
b46b8f19 | 338 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
339 | { |
340 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
341 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 342 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 343 | sizeof(unsigned long long) - |
87a927c7 | 344 | REDZONE_ALIGN); |
3b0efdfa | 345 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 346 | sizeof(unsigned long long)); |
1da177e4 LT |
347 | } |
348 | ||
343e0d7a | 349 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
350 | { |
351 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 352 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
353 | } |
354 | ||
355 | #else | |
356 | ||
3dafccf2 | 357 | #define obj_offset(x) 0 |
b46b8f19 DW |
358 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
359 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
360 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
361 | ||
362 | #endif | |
363 | ||
03787301 JK |
364 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
365 | ||
d31676df | 366 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 367 | { |
d31676df JK |
368 | return atomic_read(&cachep->store_user_clean) == 1; |
369 | } | |
03787301 | 370 | |
d31676df JK |
371 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
372 | { | |
373 | atomic_set(&cachep->store_user_clean, 1); | |
374 | } | |
03787301 | 375 | |
d31676df JK |
376 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
377 | { | |
378 | if (is_store_user_clean(cachep)) | |
379 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
380 | } |
381 | ||
382 | #else | |
d31676df | 383 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
384 | |
385 | #endif | |
386 | ||
1da177e4 | 387 | /* |
3df1cccd DR |
388 | * Do not go above this order unless 0 objects fit into the slab or |
389 | * overridden on the command line. | |
1da177e4 | 390 | */ |
543585cc DR |
391 | #define SLAB_MAX_ORDER_HI 1 |
392 | #define SLAB_MAX_ORDER_LO 0 | |
393 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 394 | static bool slab_max_order_set __initdata; |
1da177e4 | 395 | |
6ed5eb22 PE |
396 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
397 | { | |
b49af68f | 398 | struct page *page = virt_to_head_page(obj); |
35026088 | 399 | return page->slab_cache; |
6ed5eb22 PE |
400 | } |
401 | ||
8456a648 | 402 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
403 | unsigned int idx) |
404 | { | |
8456a648 | 405 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
406 | } |
407 | ||
6a2d7a95 | 408 | /* |
3b0efdfa CL |
409 | * We want to avoid an expensive divide : (offset / cache->size) |
410 | * Using the fact that size is a constant for a particular cache, | |
411 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
412 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
413 | */ | |
414 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 415 | const struct page *page, void *obj) |
8fea4e96 | 416 | { |
8456a648 | 417 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 418 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
419 | } |
420 | ||
6fb92430 | 421 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 422 | /* internal cache of cache description objs */ |
9b030cb8 | 423 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
424 | .batchcount = 1, |
425 | .limit = BOOT_CPUCACHE_ENTRIES, | |
426 | .shared = 1, | |
3b0efdfa | 427 | .size = sizeof(struct kmem_cache), |
b28a02de | 428 | .name = "kmem_cache", |
1da177e4 LT |
429 | }; |
430 | ||
1871e52c | 431 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 432 | |
343e0d7a | 433 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 434 | { |
bf0dea23 | 435 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
436 | } |
437 | ||
a737b3e2 AM |
438 | /* |
439 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
440 | */ | |
70f75067 JK |
441 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
442 | unsigned long flags, size_t *left_over) | |
fbaccacf | 443 | { |
70f75067 | 444 | unsigned int num; |
fbaccacf | 445 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 446 | |
fbaccacf SR |
447 | /* |
448 | * The slab management structure can be either off the slab or | |
449 | * on it. For the latter case, the memory allocated for a | |
450 | * slab is used for: | |
451 | * | |
fbaccacf | 452 | * - @buffer_size bytes for each object |
2e6b3602 JK |
453 | * - One freelist_idx_t for each object |
454 | * | |
455 | * We don't need to consider alignment of freelist because | |
456 | * freelist will be at the end of slab page. The objects will be | |
457 | * at the correct alignment. | |
fbaccacf SR |
458 | * |
459 | * If the slab management structure is off the slab, then the | |
460 | * alignment will already be calculated into the size. Because | |
461 | * the slabs are all pages aligned, the objects will be at the | |
462 | * correct alignment when allocated. | |
463 | */ | |
b03a017b | 464 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 465 | num = slab_size / buffer_size; |
2e6b3602 | 466 | *left_over = slab_size % buffer_size; |
fbaccacf | 467 | } else { |
70f75067 | 468 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
469 | *left_over = slab_size % |
470 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 471 | } |
70f75067 JK |
472 | |
473 | return num; | |
1da177e4 LT |
474 | } |
475 | ||
f28510d3 | 476 | #if DEBUG |
d40cee24 | 477 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 478 | |
a737b3e2 AM |
479 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
480 | char *msg) | |
1da177e4 | 481 | { |
1170532b | 482 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 483 | function, cachep->name, msg); |
1da177e4 | 484 | dump_stack(); |
373d4d09 | 485 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 486 | } |
f28510d3 | 487 | #endif |
1da177e4 | 488 | |
3395ee05 PM |
489 | /* |
490 | * By default on NUMA we use alien caches to stage the freeing of | |
491 | * objects allocated from other nodes. This causes massive memory | |
492 | * inefficiencies when using fake NUMA setup to split memory into a | |
493 | * large number of small nodes, so it can be disabled on the command | |
494 | * line | |
495 | */ | |
496 | ||
497 | static int use_alien_caches __read_mostly = 1; | |
498 | static int __init noaliencache_setup(char *s) | |
499 | { | |
500 | use_alien_caches = 0; | |
501 | return 1; | |
502 | } | |
503 | __setup("noaliencache", noaliencache_setup); | |
504 | ||
3df1cccd DR |
505 | static int __init slab_max_order_setup(char *str) |
506 | { | |
507 | get_option(&str, &slab_max_order); | |
508 | slab_max_order = slab_max_order < 0 ? 0 : | |
509 | min(slab_max_order, MAX_ORDER - 1); | |
510 | slab_max_order_set = true; | |
511 | ||
512 | return 1; | |
513 | } | |
514 | __setup("slab_max_order=", slab_max_order_setup); | |
515 | ||
8fce4d8e CL |
516 | #ifdef CONFIG_NUMA |
517 | /* | |
518 | * Special reaping functions for NUMA systems called from cache_reap(). | |
519 | * These take care of doing round robin flushing of alien caches (containing | |
520 | * objects freed on different nodes from which they were allocated) and the | |
521 | * flushing of remote pcps by calling drain_node_pages. | |
522 | */ | |
1871e52c | 523 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
524 | |
525 | static void init_reap_node(int cpu) | |
526 | { | |
0edaf86c AM |
527 | per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu), |
528 | node_online_map); | |
8fce4d8e CL |
529 | } |
530 | ||
531 | static void next_reap_node(void) | |
532 | { | |
909ea964 | 533 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 534 | |
0edaf86c | 535 | node = next_node_in(node, node_online_map); |
909ea964 | 536 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
537 | } |
538 | ||
539 | #else | |
540 | #define init_reap_node(cpu) do { } while (0) | |
541 | #define next_reap_node(void) do { } while (0) | |
542 | #endif | |
543 | ||
1da177e4 LT |
544 | /* |
545 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
546 | * via the workqueue/eventd. | |
547 | * Add the CPU number into the expiration time to minimize the possibility of | |
548 | * the CPUs getting into lockstep and contending for the global cache chain | |
549 | * lock. | |
550 | */ | |
0db0628d | 551 | static void start_cpu_timer(int cpu) |
1da177e4 | 552 | { |
1871e52c | 553 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
554 | |
555 | /* | |
556 | * When this gets called from do_initcalls via cpucache_init(), | |
557 | * init_workqueues() has already run, so keventd will be setup | |
558 | * at that time. | |
559 | */ | |
52bad64d | 560 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 561 | init_reap_node(cpu); |
203b42f7 | 562 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
563 | schedule_delayed_work_on(cpu, reap_work, |
564 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
565 | } |
566 | } | |
567 | ||
1fe00d50 | 568 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 569 | { |
d5cff635 CM |
570 | /* |
571 | * The array_cache structures contain pointers to free object. | |
25985edc | 572 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
573 | * cache the pointers are not cleared and they could be counted as |
574 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
575 | * not scan such objects. | |
576 | */ | |
1fe00d50 JK |
577 | kmemleak_no_scan(ac); |
578 | if (ac) { | |
579 | ac->avail = 0; | |
580 | ac->limit = limit; | |
581 | ac->batchcount = batch; | |
582 | ac->touched = 0; | |
1da177e4 | 583 | } |
1fe00d50 JK |
584 | } |
585 | ||
586 | static struct array_cache *alloc_arraycache(int node, int entries, | |
587 | int batchcount, gfp_t gfp) | |
588 | { | |
5e804789 | 589 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
590 | struct array_cache *ac = NULL; |
591 | ||
592 | ac = kmalloc_node(memsize, gfp, node); | |
593 | init_arraycache(ac, entries, batchcount); | |
594 | return ac; | |
1da177e4 LT |
595 | } |
596 | ||
f68f8ddd JK |
597 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
598 | struct page *page, void *objp) | |
072bb0aa | 599 | { |
f68f8ddd JK |
600 | struct kmem_cache_node *n; |
601 | int page_node; | |
602 | LIST_HEAD(list); | |
072bb0aa | 603 | |
f68f8ddd JK |
604 | page_node = page_to_nid(page); |
605 | n = get_node(cachep, page_node); | |
381760ea | 606 | |
f68f8ddd JK |
607 | spin_lock(&n->list_lock); |
608 | free_block(cachep, &objp, 1, page_node, &list); | |
609 | spin_unlock(&n->list_lock); | |
381760ea | 610 | |
f68f8ddd | 611 | slabs_destroy(cachep, &list); |
072bb0aa MG |
612 | } |
613 | ||
3ded175a CL |
614 | /* |
615 | * Transfer objects in one arraycache to another. | |
616 | * Locking must be handled by the caller. | |
617 | * | |
618 | * Return the number of entries transferred. | |
619 | */ | |
620 | static int transfer_objects(struct array_cache *to, | |
621 | struct array_cache *from, unsigned int max) | |
622 | { | |
623 | /* Figure out how many entries to transfer */ | |
732eacc0 | 624 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
625 | |
626 | if (!nr) | |
627 | return 0; | |
628 | ||
629 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
630 | sizeof(void *) *nr); | |
631 | ||
632 | from->avail -= nr; | |
633 | to->avail += nr; | |
3ded175a CL |
634 | return nr; |
635 | } | |
636 | ||
765c4507 CL |
637 | #ifndef CONFIG_NUMA |
638 | ||
639 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 640 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 641 | |
c8522a3a JK |
642 | static inline struct alien_cache **alloc_alien_cache(int node, |
643 | int limit, gfp_t gfp) | |
765c4507 | 644 | { |
8888177e | 645 | return NULL; |
765c4507 CL |
646 | } |
647 | ||
c8522a3a | 648 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
649 | { |
650 | } | |
651 | ||
652 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
653 | { | |
654 | return 0; | |
655 | } | |
656 | ||
657 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
658 | gfp_t flags) | |
659 | { | |
660 | return NULL; | |
661 | } | |
662 | ||
8b98c169 | 663 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
664 | gfp_t flags, int nodeid) |
665 | { | |
666 | return NULL; | |
667 | } | |
668 | ||
4167e9b2 DR |
669 | static inline gfp_t gfp_exact_node(gfp_t flags) |
670 | { | |
444eb2a4 | 671 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
672 | } |
673 | ||
765c4507 CL |
674 | #else /* CONFIG_NUMA */ |
675 | ||
8b98c169 | 676 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 677 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 678 | |
c8522a3a JK |
679 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
680 | int batch, gfp_t gfp) | |
681 | { | |
5e804789 | 682 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
683 | struct alien_cache *alc = NULL; |
684 | ||
685 | alc = kmalloc_node(memsize, gfp, node); | |
686 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 687 | spin_lock_init(&alc->lock); |
c8522a3a JK |
688 | return alc; |
689 | } | |
690 | ||
691 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 692 | { |
c8522a3a | 693 | struct alien_cache **alc_ptr; |
5e804789 | 694 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
695 | int i; |
696 | ||
697 | if (limit > 1) | |
698 | limit = 12; | |
c8522a3a JK |
699 | alc_ptr = kzalloc_node(memsize, gfp, node); |
700 | if (!alc_ptr) | |
701 | return NULL; | |
702 | ||
703 | for_each_node(i) { | |
704 | if (i == node || !node_online(i)) | |
705 | continue; | |
706 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
707 | if (!alc_ptr[i]) { | |
708 | for (i--; i >= 0; i--) | |
709 | kfree(alc_ptr[i]); | |
710 | kfree(alc_ptr); | |
711 | return NULL; | |
e498be7d CL |
712 | } |
713 | } | |
c8522a3a | 714 | return alc_ptr; |
e498be7d CL |
715 | } |
716 | ||
c8522a3a | 717 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
718 | { |
719 | int i; | |
720 | ||
c8522a3a | 721 | if (!alc_ptr) |
e498be7d | 722 | return; |
e498be7d | 723 | for_each_node(i) |
c8522a3a JK |
724 | kfree(alc_ptr[i]); |
725 | kfree(alc_ptr); | |
e498be7d CL |
726 | } |
727 | ||
343e0d7a | 728 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
729 | struct array_cache *ac, int node, |
730 | struct list_head *list) | |
e498be7d | 731 | { |
18bf8541 | 732 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
733 | |
734 | if (ac->avail) { | |
ce8eb6c4 | 735 | spin_lock(&n->list_lock); |
e00946fe CL |
736 | /* |
737 | * Stuff objects into the remote nodes shared array first. | |
738 | * That way we could avoid the overhead of putting the objects | |
739 | * into the free lists and getting them back later. | |
740 | */ | |
ce8eb6c4 CL |
741 | if (n->shared) |
742 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 743 | |
833b706c | 744 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 745 | ac->avail = 0; |
ce8eb6c4 | 746 | spin_unlock(&n->list_lock); |
e498be7d CL |
747 | } |
748 | } | |
749 | ||
8fce4d8e CL |
750 | /* |
751 | * Called from cache_reap() to regularly drain alien caches round robin. | |
752 | */ | |
ce8eb6c4 | 753 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 754 | { |
909ea964 | 755 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 756 | |
ce8eb6c4 | 757 | if (n->alien) { |
c8522a3a JK |
758 | struct alien_cache *alc = n->alien[node]; |
759 | struct array_cache *ac; | |
760 | ||
761 | if (alc) { | |
762 | ac = &alc->ac; | |
49dfc304 | 763 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
764 | LIST_HEAD(list); |
765 | ||
766 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 767 | spin_unlock_irq(&alc->lock); |
833b706c | 768 | slabs_destroy(cachep, &list); |
c8522a3a | 769 | } |
8fce4d8e CL |
770 | } |
771 | } | |
772 | } | |
773 | ||
a737b3e2 | 774 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 775 | struct alien_cache **alien) |
e498be7d | 776 | { |
b28a02de | 777 | int i = 0; |
c8522a3a | 778 | struct alien_cache *alc; |
e498be7d CL |
779 | struct array_cache *ac; |
780 | unsigned long flags; | |
781 | ||
782 | for_each_online_node(i) { | |
c8522a3a JK |
783 | alc = alien[i]; |
784 | if (alc) { | |
833b706c JK |
785 | LIST_HEAD(list); |
786 | ||
c8522a3a | 787 | ac = &alc->ac; |
49dfc304 | 788 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 789 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 790 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 791 | slabs_destroy(cachep, &list); |
e498be7d CL |
792 | } |
793 | } | |
794 | } | |
729bd0b7 | 795 | |
25c4f304 JK |
796 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
797 | int node, int page_node) | |
729bd0b7 | 798 | { |
ce8eb6c4 | 799 | struct kmem_cache_node *n; |
c8522a3a JK |
800 | struct alien_cache *alien = NULL; |
801 | struct array_cache *ac; | |
97654dfa | 802 | LIST_HEAD(list); |
1ca4cb24 | 803 | |
18bf8541 | 804 | n = get_node(cachep, node); |
729bd0b7 | 805 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
806 | if (n->alien && n->alien[page_node]) { |
807 | alien = n->alien[page_node]; | |
c8522a3a | 808 | ac = &alien->ac; |
49dfc304 | 809 | spin_lock(&alien->lock); |
c8522a3a | 810 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 811 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 812 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 813 | } |
f68f8ddd | 814 | ac->entry[ac->avail++] = objp; |
49dfc304 | 815 | spin_unlock(&alien->lock); |
833b706c | 816 | slabs_destroy(cachep, &list); |
729bd0b7 | 817 | } else { |
25c4f304 | 818 | n = get_node(cachep, page_node); |
18bf8541 | 819 | spin_lock(&n->list_lock); |
25c4f304 | 820 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 821 | spin_unlock(&n->list_lock); |
97654dfa | 822 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
823 | } |
824 | return 1; | |
825 | } | |
25c4f304 JK |
826 | |
827 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
828 | { | |
829 | int page_node = page_to_nid(virt_to_page(objp)); | |
830 | int node = numa_mem_id(); | |
831 | /* | |
832 | * Make sure we are not freeing a object from another node to the array | |
833 | * cache on this cpu. | |
834 | */ | |
835 | if (likely(node == page_node)) | |
836 | return 0; | |
837 | ||
838 | return __cache_free_alien(cachep, objp, node, page_node); | |
839 | } | |
4167e9b2 DR |
840 | |
841 | /* | |
444eb2a4 MG |
842 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
843 | * warn about failures. | |
4167e9b2 DR |
844 | */ |
845 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
846 | { | |
444eb2a4 | 847 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 848 | } |
e498be7d CL |
849 | #endif |
850 | ||
ded0ecf6 JK |
851 | static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp) |
852 | { | |
853 | struct kmem_cache_node *n; | |
854 | ||
855 | /* | |
856 | * Set up the kmem_cache_node for cpu before we can | |
857 | * begin anything. Make sure some other cpu on this | |
858 | * node has not already allocated this | |
859 | */ | |
860 | n = get_node(cachep, node); | |
861 | if (n) { | |
862 | spin_lock_irq(&n->list_lock); | |
863 | n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount + | |
864 | cachep->num; | |
865 | spin_unlock_irq(&n->list_lock); | |
866 | ||
867 | return 0; | |
868 | } | |
869 | ||
870 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); | |
871 | if (!n) | |
872 | return -ENOMEM; | |
873 | ||
874 | kmem_cache_node_init(n); | |
875 | n->next_reap = jiffies + REAPTIMEOUT_NODE + | |
876 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
877 | ||
878 | n->free_limit = | |
879 | (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num; | |
880 | ||
881 | /* | |
882 | * The kmem_cache_nodes don't come and go as CPUs | |
883 | * come and go. slab_mutex is sufficient | |
884 | * protection here. | |
885 | */ | |
886 | cachep->node[node] = n; | |
887 | ||
888 | return 0; | |
889 | } | |
890 | ||
6731d4f1 | 891 | #if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP) |
8f9f8d9e | 892 | /* |
6a67368c | 893 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 894 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 895 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 896 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
897 | * already in use. |
898 | * | |
18004c5d | 899 | * Must hold slab_mutex. |
8f9f8d9e | 900 | */ |
6a67368c | 901 | static int init_cache_node_node(int node) |
8f9f8d9e | 902 | { |
ded0ecf6 | 903 | int ret; |
8f9f8d9e | 904 | struct kmem_cache *cachep; |
8f9f8d9e | 905 | |
18004c5d | 906 | list_for_each_entry(cachep, &slab_caches, list) { |
ded0ecf6 JK |
907 | ret = init_cache_node(cachep, node, GFP_KERNEL); |
908 | if (ret) | |
909 | return ret; | |
8f9f8d9e | 910 | } |
ded0ecf6 | 911 | |
8f9f8d9e DR |
912 | return 0; |
913 | } | |
6731d4f1 | 914 | #endif |
8f9f8d9e | 915 | |
c3d332b6 JK |
916 | static int setup_kmem_cache_node(struct kmem_cache *cachep, |
917 | int node, gfp_t gfp, bool force_change) | |
918 | { | |
919 | int ret = -ENOMEM; | |
920 | struct kmem_cache_node *n; | |
921 | struct array_cache *old_shared = NULL; | |
922 | struct array_cache *new_shared = NULL; | |
923 | struct alien_cache **new_alien = NULL; | |
924 | LIST_HEAD(list); | |
925 | ||
926 | if (use_alien_caches) { | |
927 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
928 | if (!new_alien) | |
929 | goto fail; | |
930 | } | |
931 | ||
932 | if (cachep->shared) { | |
933 | new_shared = alloc_arraycache(node, | |
934 | cachep->shared * cachep->batchcount, 0xbaadf00d, gfp); | |
935 | if (!new_shared) | |
936 | goto fail; | |
937 | } | |
938 | ||
939 | ret = init_cache_node(cachep, node, gfp); | |
940 | if (ret) | |
941 | goto fail; | |
942 | ||
943 | n = get_node(cachep, node); | |
944 | spin_lock_irq(&n->list_lock); | |
945 | if (n->shared && force_change) { | |
946 | free_block(cachep, n->shared->entry, | |
947 | n->shared->avail, node, &list); | |
948 | n->shared->avail = 0; | |
949 | } | |
950 | ||
951 | if (!n->shared || force_change) { | |
952 | old_shared = n->shared; | |
953 | n->shared = new_shared; | |
954 | new_shared = NULL; | |
955 | } | |
956 | ||
957 | if (!n->alien) { | |
958 | n->alien = new_alien; | |
959 | new_alien = NULL; | |
960 | } | |
961 | ||
962 | spin_unlock_irq(&n->list_lock); | |
963 | slabs_destroy(cachep, &list); | |
964 | ||
801faf0d JK |
965 | /* |
966 | * To protect lockless access to n->shared during irq disabled context. | |
967 | * If n->shared isn't NULL in irq disabled context, accessing to it is | |
968 | * guaranteed to be valid until irq is re-enabled, because it will be | |
969 | * freed after synchronize_sched(). | |
970 | */ | |
86d9f485 | 971 | if (old_shared && force_change) |
801faf0d JK |
972 | synchronize_sched(); |
973 | ||
c3d332b6 JK |
974 | fail: |
975 | kfree(old_shared); | |
976 | kfree(new_shared); | |
977 | free_alien_cache(new_alien); | |
978 | ||
979 | return ret; | |
980 | } | |
981 | ||
6731d4f1 SAS |
982 | #ifdef CONFIG_SMP |
983 | ||
0db0628d | 984 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
985 | { |
986 | struct kmem_cache *cachep; | |
ce8eb6c4 | 987 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 988 | int node = cpu_to_mem(cpu); |
a70f7302 | 989 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 990 | |
18004c5d | 991 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
992 | struct array_cache *nc; |
993 | struct array_cache *shared; | |
c8522a3a | 994 | struct alien_cache **alien; |
97654dfa | 995 | LIST_HEAD(list); |
fbf1e473 | 996 | |
18bf8541 | 997 | n = get_node(cachep, node); |
ce8eb6c4 | 998 | if (!n) |
bf0dea23 | 999 | continue; |
fbf1e473 | 1000 | |
ce8eb6c4 | 1001 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1002 | |
ce8eb6c4 CL |
1003 | /* Free limit for this kmem_cache_node */ |
1004 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1005 | |
1006 | /* cpu is dead; no one can alloc from it. */ | |
1007 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1008 | if (nc) { | |
97654dfa | 1009 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1010 | nc->avail = 0; |
1011 | } | |
fbf1e473 | 1012 | |
58463c1f | 1013 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1014 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1015 | goto free_slab; |
fbf1e473 AM |
1016 | } |
1017 | ||
ce8eb6c4 | 1018 | shared = n->shared; |
fbf1e473 AM |
1019 | if (shared) { |
1020 | free_block(cachep, shared->entry, | |
97654dfa | 1021 | shared->avail, node, &list); |
ce8eb6c4 | 1022 | n->shared = NULL; |
fbf1e473 AM |
1023 | } |
1024 | ||
ce8eb6c4 CL |
1025 | alien = n->alien; |
1026 | n->alien = NULL; | |
fbf1e473 | 1027 | |
ce8eb6c4 | 1028 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1029 | |
1030 | kfree(shared); | |
1031 | if (alien) { | |
1032 | drain_alien_cache(cachep, alien); | |
1033 | free_alien_cache(alien); | |
1034 | } | |
bf0dea23 JK |
1035 | |
1036 | free_slab: | |
97654dfa | 1037 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1038 | } |
1039 | /* | |
1040 | * In the previous loop, all the objects were freed to | |
1041 | * the respective cache's slabs, now we can go ahead and | |
1042 | * shrink each nodelist to its limit. | |
1043 | */ | |
18004c5d | 1044 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1045 | n = get_node(cachep, node); |
ce8eb6c4 | 1046 | if (!n) |
fbf1e473 | 1047 | continue; |
a5aa63a5 | 1048 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
1049 | } |
1050 | } | |
1051 | ||
0db0628d | 1052 | static int cpuup_prepare(long cpu) |
1da177e4 | 1053 | { |
343e0d7a | 1054 | struct kmem_cache *cachep; |
7d6e6d09 | 1055 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1056 | int err; |
1da177e4 | 1057 | |
fbf1e473 AM |
1058 | /* |
1059 | * We need to do this right in the beginning since | |
1060 | * alloc_arraycache's are going to use this list. | |
1061 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1062 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1063 | */ |
6a67368c | 1064 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1065 | if (err < 0) |
1066 | goto bad; | |
fbf1e473 AM |
1067 | |
1068 | /* | |
1069 | * Now we can go ahead with allocating the shared arrays and | |
1070 | * array caches | |
1071 | */ | |
18004c5d | 1072 | list_for_each_entry(cachep, &slab_caches, list) { |
c3d332b6 JK |
1073 | err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false); |
1074 | if (err) | |
1075 | goto bad; | |
fbf1e473 | 1076 | } |
ce79ddc8 | 1077 | |
fbf1e473 AM |
1078 | return 0; |
1079 | bad: | |
12d00f6a | 1080 | cpuup_canceled(cpu); |
fbf1e473 AM |
1081 | return -ENOMEM; |
1082 | } | |
1083 | ||
6731d4f1 | 1084 | int slab_prepare_cpu(unsigned int cpu) |
fbf1e473 | 1085 | { |
6731d4f1 | 1086 | int err; |
fbf1e473 | 1087 | |
6731d4f1 SAS |
1088 | mutex_lock(&slab_mutex); |
1089 | err = cpuup_prepare(cpu); | |
1090 | mutex_unlock(&slab_mutex); | |
1091 | return err; | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * This is called for a failed online attempt and for a successful | |
1096 | * offline. | |
1097 | * | |
1098 | * Even if all the cpus of a node are down, we don't free the | |
1099 | * kmem_list3 of any cache. This to avoid a race between cpu_down, and | |
1100 | * a kmalloc allocation from another cpu for memory from the node of | |
1101 | * the cpu going down. The list3 structure is usually allocated from | |
1102 | * kmem_cache_create() and gets destroyed at kmem_cache_destroy(). | |
1103 | */ | |
1104 | int slab_dead_cpu(unsigned int cpu) | |
1105 | { | |
1106 | mutex_lock(&slab_mutex); | |
1107 | cpuup_canceled(cpu); | |
1108 | mutex_unlock(&slab_mutex); | |
1109 | return 0; | |
1110 | } | |
8f5be20b | 1111 | #endif |
6731d4f1 SAS |
1112 | |
1113 | static int slab_online_cpu(unsigned int cpu) | |
1114 | { | |
1115 | start_cpu_timer(cpu); | |
1116 | return 0; | |
1da177e4 LT |
1117 | } |
1118 | ||
6731d4f1 SAS |
1119 | static int slab_offline_cpu(unsigned int cpu) |
1120 | { | |
1121 | /* | |
1122 | * Shutdown cache reaper. Note that the slab_mutex is held so | |
1123 | * that if cache_reap() is invoked it cannot do anything | |
1124 | * expensive but will only modify reap_work and reschedule the | |
1125 | * timer. | |
1126 | */ | |
1127 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); | |
1128 | /* Now the cache_reaper is guaranteed to be not running. */ | |
1129 | per_cpu(slab_reap_work, cpu).work.func = NULL; | |
1130 | return 0; | |
1131 | } | |
1da177e4 | 1132 | |
8f9f8d9e DR |
1133 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1134 | /* | |
1135 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1136 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1137 | * removed. | |
1138 | * | |
18004c5d | 1139 | * Must hold slab_mutex. |
8f9f8d9e | 1140 | */ |
6a67368c | 1141 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1142 | { |
1143 | struct kmem_cache *cachep; | |
1144 | int ret = 0; | |
1145 | ||
18004c5d | 1146 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1147 | struct kmem_cache_node *n; |
8f9f8d9e | 1148 | |
18bf8541 | 1149 | n = get_node(cachep, node); |
ce8eb6c4 | 1150 | if (!n) |
8f9f8d9e DR |
1151 | continue; |
1152 | ||
a5aa63a5 | 1153 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1154 | |
ce8eb6c4 CL |
1155 | if (!list_empty(&n->slabs_full) || |
1156 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1157 | ret = -EBUSY; |
1158 | break; | |
1159 | } | |
1160 | } | |
1161 | return ret; | |
1162 | } | |
1163 | ||
1164 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1165 | unsigned long action, void *arg) | |
1166 | { | |
1167 | struct memory_notify *mnb = arg; | |
1168 | int ret = 0; | |
1169 | int nid; | |
1170 | ||
1171 | nid = mnb->status_change_nid; | |
1172 | if (nid < 0) | |
1173 | goto out; | |
1174 | ||
1175 | switch (action) { | |
1176 | case MEM_GOING_ONLINE: | |
18004c5d | 1177 | mutex_lock(&slab_mutex); |
6a67368c | 1178 | ret = init_cache_node_node(nid); |
18004c5d | 1179 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1180 | break; |
1181 | case MEM_GOING_OFFLINE: | |
18004c5d | 1182 | mutex_lock(&slab_mutex); |
6a67368c | 1183 | ret = drain_cache_node_node(nid); |
18004c5d | 1184 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1185 | break; |
1186 | case MEM_ONLINE: | |
1187 | case MEM_OFFLINE: | |
1188 | case MEM_CANCEL_ONLINE: | |
1189 | case MEM_CANCEL_OFFLINE: | |
1190 | break; | |
1191 | } | |
1192 | out: | |
5fda1bd5 | 1193 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1194 | } |
1195 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1196 | ||
e498be7d | 1197 | /* |
ce8eb6c4 | 1198 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1199 | */ |
6744f087 | 1200 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1201 | int nodeid) |
e498be7d | 1202 | { |
6744f087 | 1203 | struct kmem_cache_node *ptr; |
e498be7d | 1204 | |
6744f087 | 1205 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1206 | BUG_ON(!ptr); |
1207 | ||
6744f087 | 1208 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1209 | /* |
1210 | * Do not assume that spinlocks can be initialized via memcpy: | |
1211 | */ | |
1212 | spin_lock_init(&ptr->list_lock); | |
1213 | ||
e498be7d | 1214 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1215 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1216 | } |
1217 | ||
556a169d | 1218 | /* |
ce8eb6c4 CL |
1219 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1220 | * size of kmem_cache_node. | |
556a169d | 1221 | */ |
ce8eb6c4 | 1222 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1223 | { |
1224 | int node; | |
1225 | ||
1226 | for_each_online_node(node) { | |
ce8eb6c4 | 1227 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1228 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1229 | REAPTIMEOUT_NODE + |
1230 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1231 | } |
1232 | } | |
1233 | ||
a737b3e2 AM |
1234 | /* |
1235 | * Initialisation. Called after the page allocator have been initialised and | |
1236 | * before smp_init(). | |
1da177e4 LT |
1237 | */ |
1238 | void __init kmem_cache_init(void) | |
1239 | { | |
e498be7d CL |
1240 | int i; |
1241 | ||
68126702 JK |
1242 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1243 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1244 | kmem_cache = &kmem_cache_boot; |
1245 | ||
8888177e | 1246 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1247 | use_alien_caches = 0; |
1248 | ||
3c583465 | 1249 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1250 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1251 | |
1da177e4 LT |
1252 | /* |
1253 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1254 | * page orders on machines with more than 32MB of memory if |
1255 | * not overridden on the command line. | |
1da177e4 | 1256 | */ |
3df1cccd | 1257 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1258 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1259 | |
1da177e4 LT |
1260 | /* Bootstrap is tricky, because several objects are allocated |
1261 | * from caches that do not exist yet: | |
9b030cb8 CL |
1262 | * 1) initialize the kmem_cache cache: it contains the struct |
1263 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1264 | * kmem_cache is statically allocated. | |
e498be7d | 1265 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1266 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1267 | * array at the end of the bootstrap. |
1da177e4 | 1268 | * 2) Create the first kmalloc cache. |
343e0d7a | 1269 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1270 | * An __init data area is used for the head array. |
1271 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1272 | * head arrays. | |
9b030cb8 | 1273 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1274 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1275 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1276 | * the other cache's with kmalloc allocated memory. |
1277 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1278 | */ |
1279 | ||
9b030cb8 | 1280 | /* 1) create the kmem_cache */ |
1da177e4 | 1281 | |
8da3430d | 1282 | /* |
b56efcf0 | 1283 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1284 | */ |
2f9baa9f | 1285 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1286 | offsetof(struct kmem_cache, node) + |
6744f087 | 1287 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1288 | SLAB_HWCACHE_ALIGN); |
1289 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1290 | slab_state = PARTIAL; |
1da177e4 | 1291 | |
a737b3e2 | 1292 | /* |
bf0dea23 JK |
1293 | * Initialize the caches that provide memory for the kmem_cache_node |
1294 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1295 | */ |
bf0dea23 | 1296 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
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; |
e12ba74d MG |
1415 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1416 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1417 | |
96db800f | 1418 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1419 | if (!page) { |
9a02d699 | 1420 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1421 | return NULL; |
8bdec192 | 1422 | } |
1da177e4 | 1423 | |
f3ccb2c4 VD |
1424 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1425 | __free_pages(page, cachep->gfporder); | |
1426 | return NULL; | |
1427 | } | |
1428 | ||
e1b6aa6f | 1429 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1430 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1431 | add_zone_page_state(page_zone(page), |
1432 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1433 | else | |
1434 | add_zone_page_state(page_zone(page), | |
1435 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
f68f8ddd | 1436 | |
a57a4988 | 1437 | __SetPageSlab(page); |
f68f8ddd JK |
1438 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1439 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1440 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1441 | |
b1eeab67 VN |
1442 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1443 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1444 | ||
1445 | if (cachep->ctor) | |
1446 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1447 | else | |
1448 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1449 | } | |
c175eea4 | 1450 | |
0c3aa83e | 1451 | return page; |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | /* | |
1455 | * Interface to system's page release. | |
1456 | */ | |
0c3aa83e | 1457 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1458 | { |
27ee57c9 VD |
1459 | int order = cachep->gfporder; |
1460 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1461 | |
27ee57c9 | 1462 | kmemcheck_free_shadow(page, order); |
c175eea4 | 1463 | |
972d1a7b CL |
1464 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1465 | sub_zone_page_state(page_zone(page), | |
1466 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1467 | else | |
1468 | sub_zone_page_state(page_zone(page), | |
1469 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1470 | |
a57a4988 | 1471 | BUG_ON(!PageSlab(page)); |
73293c2f | 1472 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1473 | __ClearPageSlab(page); |
8456a648 JK |
1474 | page_mapcount_reset(page); |
1475 | page->mapping = NULL; | |
1f458cbf | 1476 | |
1da177e4 LT |
1477 | if (current->reclaim_state) |
1478 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1479 | memcg_uncharge_slab(page, order, cachep); |
1480 | __free_pages(page, order); | |
1da177e4 LT |
1481 | } |
1482 | ||
1483 | static void kmem_rcu_free(struct rcu_head *head) | |
1484 | { | |
68126702 JK |
1485 | struct kmem_cache *cachep; |
1486 | struct page *page; | |
1da177e4 | 1487 | |
68126702 JK |
1488 | page = container_of(head, struct page, rcu_head); |
1489 | cachep = page->slab_cache; | |
1490 | ||
1491 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1492 | } |
1493 | ||
1494 | #if DEBUG | |
40b44137 JK |
1495 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1496 | { | |
1497 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1498 | (cachep->size % PAGE_SIZE) == 0) | |
1499 | return true; | |
1500 | ||
1501 | return false; | |
1502 | } | |
1da177e4 LT |
1503 | |
1504 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1505 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1506 | unsigned long caller) |
1da177e4 | 1507 | { |
8c138bc0 | 1508 | int size = cachep->object_size; |
1da177e4 | 1509 | |
3dafccf2 | 1510 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1511 | |
b28a02de | 1512 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1513 | return; |
1514 | ||
b28a02de PE |
1515 | *addr++ = 0x12345678; |
1516 | *addr++ = caller; | |
1517 | *addr++ = smp_processor_id(); | |
1518 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1519 | { |
1520 | unsigned long *sptr = &caller; | |
1521 | unsigned long svalue; | |
1522 | ||
1523 | while (!kstack_end(sptr)) { | |
1524 | svalue = *sptr++; | |
1525 | if (kernel_text_address(svalue)) { | |
b28a02de | 1526 | *addr++ = svalue; |
1da177e4 LT |
1527 | size -= sizeof(unsigned long); |
1528 | if (size <= sizeof(unsigned long)) | |
1529 | break; | |
1530 | } | |
1531 | } | |
1532 | ||
1533 | } | |
b28a02de | 1534 | *addr++ = 0x87654321; |
1da177e4 | 1535 | } |
40b44137 JK |
1536 | |
1537 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1538 | int map, unsigned long caller) | |
1539 | { | |
1540 | if (!is_debug_pagealloc_cache(cachep)) | |
1541 | return; | |
1542 | ||
1543 | if (caller) | |
1544 | store_stackinfo(cachep, objp, caller); | |
1545 | ||
1546 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1547 | } | |
1548 | ||
1549 | #else | |
1550 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1551 | int map, unsigned long caller) {} | |
1552 | ||
1da177e4 LT |
1553 | #endif |
1554 | ||
343e0d7a | 1555 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1556 | { |
8c138bc0 | 1557 | int size = cachep->object_size; |
3dafccf2 | 1558 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1559 | |
1560 | memset(addr, val, size); | |
b28a02de | 1561 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1562 | } |
1563 | ||
1564 | static void dump_line(char *data, int offset, int limit) | |
1565 | { | |
1566 | int i; | |
aa83aa40 DJ |
1567 | unsigned char error = 0; |
1568 | int bad_count = 0; | |
1569 | ||
1170532b | 1570 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1571 | for (i = 0; i < limit; i++) { |
1572 | if (data[offset + i] != POISON_FREE) { | |
1573 | error = data[offset + i]; | |
1574 | bad_count++; | |
1575 | } | |
aa83aa40 | 1576 | } |
fdde6abb SAS |
1577 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1578 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1579 | |
1580 | if (bad_count == 1) { | |
1581 | error ^= POISON_FREE; | |
1582 | if (!(error & (error - 1))) { | |
1170532b | 1583 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1584 | #ifdef CONFIG_X86 |
1170532b | 1585 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1586 | #else |
1170532b | 1587 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1588 | #endif |
1589 | } | |
1590 | } | |
1da177e4 LT |
1591 | } |
1592 | #endif | |
1593 | ||
1594 | #if DEBUG | |
1595 | ||
343e0d7a | 1596 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1597 | { |
1598 | int i, size; | |
1599 | char *realobj; | |
1600 | ||
1601 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1602 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1603 | *dbg_redzone1(cachep, objp), | |
1604 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1605 | } |
1606 | ||
1607 | if (cachep->flags & SLAB_STORE_USER) { | |
1170532b | 1608 | pr_err("Last user: [<%p>](%pSR)\n", |
071361d3 JP |
1609 | *dbg_userword(cachep, objp), |
1610 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1611 | } |
3dafccf2 | 1612 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1613 | size = cachep->object_size; |
b28a02de | 1614 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1615 | int limit; |
1616 | limit = 16; | |
b28a02de PE |
1617 | if (i + limit > size) |
1618 | limit = size - i; | |
1da177e4 LT |
1619 | dump_line(realobj, i, limit); |
1620 | } | |
1621 | } | |
1622 | ||
343e0d7a | 1623 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1624 | { |
1625 | char *realobj; | |
1626 | int size, i; | |
1627 | int lines = 0; | |
1628 | ||
40b44137 JK |
1629 | if (is_debug_pagealloc_cache(cachep)) |
1630 | return; | |
1631 | ||
3dafccf2 | 1632 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1633 | size = cachep->object_size; |
1da177e4 | 1634 | |
b28a02de | 1635 | for (i = 0; i < size; i++) { |
1da177e4 | 1636 | char exp = POISON_FREE; |
b28a02de | 1637 | if (i == size - 1) |
1da177e4 LT |
1638 | exp = POISON_END; |
1639 | if (realobj[i] != exp) { | |
1640 | int limit; | |
1641 | /* Mismatch ! */ | |
1642 | /* Print header */ | |
1643 | if (lines == 0) { | |
1170532b JP |
1644 | pr_err("Slab corruption (%s): %s start=%p, len=%d\n", |
1645 | print_tainted(), cachep->name, | |
1646 | realobj, size); | |
1da177e4 LT |
1647 | print_objinfo(cachep, objp, 0); |
1648 | } | |
1649 | /* Hexdump the affected line */ | |
b28a02de | 1650 | i = (i / 16) * 16; |
1da177e4 | 1651 | limit = 16; |
b28a02de PE |
1652 | if (i + limit > size) |
1653 | limit = size - i; | |
1da177e4 LT |
1654 | dump_line(realobj, i, limit); |
1655 | i += 16; | |
1656 | lines++; | |
1657 | /* Limit to 5 lines */ | |
1658 | if (lines > 5) | |
1659 | break; | |
1660 | } | |
1661 | } | |
1662 | if (lines != 0) { | |
1663 | /* Print some data about the neighboring objects, if they | |
1664 | * exist: | |
1665 | */ | |
8456a648 | 1666 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1667 | unsigned int objnr; |
1da177e4 | 1668 | |
8456a648 | 1669 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1670 | if (objnr) { |
8456a648 | 1671 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1672 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1673 | pr_err("Prev obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1674 | print_objinfo(cachep, objp, 2); |
1675 | } | |
b28a02de | 1676 | if (objnr + 1 < cachep->num) { |
8456a648 | 1677 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1678 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1679 | pr_err("Next obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1680 | print_objinfo(cachep, objp, 2); |
1681 | } | |
1682 | } | |
1683 | } | |
1684 | #endif | |
1685 | ||
12dd36fa | 1686 | #if DEBUG |
8456a648 JK |
1687 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1688 | struct page *page) | |
1da177e4 | 1689 | { |
1da177e4 | 1690 | int i; |
b03a017b JK |
1691 | |
1692 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1693 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1694 | POISON_FREE); | |
1695 | } | |
1696 | ||
1da177e4 | 1697 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1698 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1699 | |
1700 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1701 | check_poison_obj(cachep, objp); |
40b44137 | 1702 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1703 | } |
1704 | if (cachep->flags & SLAB_RED_ZONE) { | |
1705 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1706 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1707 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1708 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1709 | } |
1da177e4 | 1710 | } |
12dd36fa | 1711 | } |
1da177e4 | 1712 | #else |
8456a648 JK |
1713 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1714 | struct page *page) | |
12dd36fa | 1715 | { |
12dd36fa | 1716 | } |
1da177e4 LT |
1717 | #endif |
1718 | ||
911851e6 RD |
1719 | /** |
1720 | * slab_destroy - destroy and release all objects in a slab | |
1721 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1722 | * @page: page pointer being destroyed |
911851e6 | 1723 | * |
8a7d9b43 WSH |
1724 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1725 | * Before calling the slab page must have been unlinked from the cache. The | |
1726 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1727 | */ |
8456a648 | 1728 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1729 | { |
7e007355 | 1730 | void *freelist; |
12dd36fa | 1731 | |
8456a648 JK |
1732 | freelist = page->freelist; |
1733 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1734 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1735 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1736 | else | |
0c3aa83e | 1737 | kmem_freepages(cachep, page); |
68126702 JK |
1738 | |
1739 | /* | |
8456a648 | 1740 | * From now on, we don't use freelist |
68126702 JK |
1741 | * although actual page can be freed in rcu context |
1742 | */ | |
1743 | if (OFF_SLAB(cachep)) | |
8456a648 | 1744 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1745 | } |
1746 | ||
97654dfa JK |
1747 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1748 | { | |
1749 | struct page *page, *n; | |
1750 | ||
1751 | list_for_each_entry_safe(page, n, list, lru) { | |
1752 | list_del(&page->lru); | |
1753 | slab_destroy(cachep, page); | |
1754 | } | |
1755 | } | |
1756 | ||
4d268eba | 1757 | /** |
a70773dd RD |
1758 | * calculate_slab_order - calculate size (page order) of slabs |
1759 | * @cachep: pointer to the cache that is being created | |
1760 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1761 | * @flags: slab allocation flags |
1762 | * | |
1763 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1764 | * |
1765 | * This could be made much more intelligent. For now, try to avoid using | |
1766 | * high order pages for slabs. When the gfp() functions are more friendly | |
1767 | * towards high-order requests, this should be changed. | |
1768 | */ | |
a737b3e2 | 1769 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1770 | size_t size, unsigned long flags) |
4d268eba PE |
1771 | { |
1772 | size_t left_over = 0; | |
9888e6fa | 1773 | int gfporder; |
4d268eba | 1774 | |
0aa817f0 | 1775 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1776 | unsigned int num; |
1777 | size_t remainder; | |
1778 | ||
70f75067 | 1779 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1780 | if (!num) |
1781 | continue; | |
9888e6fa | 1782 | |
f315e3fa JK |
1783 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1784 | if (num > SLAB_OBJ_MAX_NUM) | |
1785 | break; | |
1786 | ||
b1ab41c4 | 1787 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1788 | struct kmem_cache *freelist_cache; |
1789 | size_t freelist_size; | |
1790 | ||
1791 | freelist_size = num * sizeof(freelist_idx_t); | |
1792 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1793 | if (!freelist_cache) | |
1794 | continue; | |
1795 | ||
b1ab41c4 | 1796 | /* |
3217fd9b | 1797 | * Needed to avoid possible looping condition |
76b342bd | 1798 | * in cache_grow_begin() |
b1ab41c4 | 1799 | */ |
3217fd9b JK |
1800 | if (OFF_SLAB(freelist_cache)) |
1801 | continue; | |
b1ab41c4 | 1802 | |
3217fd9b JK |
1803 | /* check if off slab has enough benefit */ |
1804 | if (freelist_cache->size > cachep->size / 2) | |
1805 | continue; | |
b1ab41c4 | 1806 | } |
4d268eba | 1807 | |
9888e6fa | 1808 | /* Found something acceptable - save it away */ |
4d268eba | 1809 | cachep->num = num; |
9888e6fa | 1810 | cachep->gfporder = gfporder; |
4d268eba PE |
1811 | left_over = remainder; |
1812 | ||
f78bb8ad LT |
1813 | /* |
1814 | * A VFS-reclaimable slab tends to have most allocations | |
1815 | * as GFP_NOFS and we really don't want to have to be allocating | |
1816 | * higher-order pages when we are unable to shrink dcache. | |
1817 | */ | |
1818 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1819 | break; | |
1820 | ||
4d268eba PE |
1821 | /* |
1822 | * Large number of objects is good, but very large slabs are | |
1823 | * currently bad for the gfp()s. | |
1824 | */ | |
543585cc | 1825 | if (gfporder >= slab_max_order) |
4d268eba PE |
1826 | break; |
1827 | ||
9888e6fa LT |
1828 | /* |
1829 | * Acceptable internal fragmentation? | |
1830 | */ | |
a737b3e2 | 1831 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1832 | break; |
1833 | } | |
1834 | return left_over; | |
1835 | } | |
1836 | ||
bf0dea23 JK |
1837 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1838 | struct kmem_cache *cachep, int entries, int batchcount) | |
1839 | { | |
1840 | int cpu; | |
1841 | size_t size; | |
1842 | struct array_cache __percpu *cpu_cache; | |
1843 | ||
1844 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1845 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1846 | |
1847 | if (!cpu_cache) | |
1848 | return NULL; | |
1849 | ||
1850 | for_each_possible_cpu(cpu) { | |
1851 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1852 | entries, batchcount); | |
1853 | } | |
1854 | ||
1855 | return cpu_cache; | |
1856 | } | |
1857 | ||
bd721ea7 | 1858 | static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1859 | { |
97d06609 | 1860 | if (slab_state >= FULL) |
83b519e8 | 1861 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1862 | |
bf0dea23 JK |
1863 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1864 | if (!cachep->cpu_cache) | |
1865 | return 1; | |
1866 | ||
97d06609 | 1867 | if (slab_state == DOWN) { |
bf0dea23 JK |
1868 | /* Creation of first cache (kmem_cache). */ |
1869 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1870 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1871 | /* For kmem_cache_node */ |
1872 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1873 | } else { |
bf0dea23 | 1874 | int node; |
f30cf7d1 | 1875 | |
bf0dea23 JK |
1876 | for_each_online_node(node) { |
1877 | cachep->node[node] = kmalloc_node( | |
1878 | sizeof(struct kmem_cache_node), gfp, node); | |
1879 | BUG_ON(!cachep->node[node]); | |
1880 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1881 | } |
1882 | } | |
bf0dea23 | 1883 | |
6a67368c | 1884 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1885 | jiffies + REAPTIMEOUT_NODE + |
1886 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1887 | |
1888 | cpu_cache_get(cachep)->avail = 0; | |
1889 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1890 | cpu_cache_get(cachep)->batchcount = 1; | |
1891 | cpu_cache_get(cachep)->touched = 0; | |
1892 | cachep->batchcount = 1; | |
1893 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1894 | return 0; |
f30cf7d1 PE |
1895 | } |
1896 | ||
12220dea JK |
1897 | unsigned long kmem_cache_flags(unsigned long object_size, |
1898 | unsigned long flags, const char *name, | |
1899 | void (*ctor)(void *)) | |
1900 | { | |
1901 | return flags; | |
1902 | } | |
1903 | ||
1904 | struct kmem_cache * | |
1905 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
1906 | unsigned long flags, void (*ctor)(void *)) | |
1907 | { | |
1908 | struct kmem_cache *cachep; | |
1909 | ||
1910 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1911 | if (cachep) { | |
1912 | cachep->refcount++; | |
1913 | ||
1914 | /* | |
1915 | * Adjust the object sizes so that we clear | |
1916 | * the complete object on kzalloc. | |
1917 | */ | |
1918 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1919 | } | |
1920 | return cachep; | |
1921 | } | |
1922 | ||
b03a017b JK |
1923 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
1924 | size_t size, unsigned long flags) | |
1925 | { | |
1926 | size_t left; | |
1927 | ||
1928 | cachep->num = 0; | |
1929 | ||
1930 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
1931 | return false; | |
1932 | ||
1933 | left = calculate_slab_order(cachep, size, | |
1934 | flags | CFLGS_OBJFREELIST_SLAB); | |
1935 | if (!cachep->num) | |
1936 | return false; | |
1937 | ||
1938 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1939 | return false; | |
1940 | ||
1941 | cachep->colour = left / cachep->colour_off; | |
1942 | ||
1943 | return true; | |
1944 | } | |
1945 | ||
158e319b JK |
1946 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
1947 | size_t size, unsigned long flags) | |
1948 | { | |
1949 | size_t left; | |
1950 | ||
1951 | cachep->num = 0; | |
1952 | ||
1953 | /* | |
3217fd9b JK |
1954 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1955 | * to avoid recursive calls into kmemleak. | |
158e319b | 1956 | */ |
158e319b JK |
1957 | if (flags & SLAB_NOLEAKTRACE) |
1958 | return false; | |
1959 | ||
1960 | /* | |
1961 | * Size is large, assume best to place the slab management obj | |
1962 | * off-slab (should allow better packing of objs). | |
1963 | */ | |
1964 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1965 | if (!cachep->num) | |
1966 | return false; | |
1967 | ||
1968 | /* | |
1969 | * If the slab has been placed off-slab, and we have enough space then | |
1970 | * move it on-slab. This is at the expense of any extra colouring. | |
1971 | */ | |
1972 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1973 | return false; | |
1974 | ||
1975 | cachep->colour = left / cachep->colour_off; | |
1976 | ||
1977 | return true; | |
1978 | } | |
1979 | ||
1980 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
1981 | size_t size, unsigned long flags) | |
1982 | { | |
1983 | size_t left; | |
1984 | ||
1985 | cachep->num = 0; | |
1986 | ||
1987 | left = calculate_slab_order(cachep, size, flags); | |
1988 | if (!cachep->num) | |
1989 | return false; | |
1990 | ||
1991 | cachep->colour = left / cachep->colour_off; | |
1992 | ||
1993 | return true; | |
1994 | } | |
1995 | ||
1da177e4 | 1996 | /** |
039363f3 | 1997 | * __kmem_cache_create - Create a cache. |
a755b76a | 1998 | * @cachep: cache management descriptor |
1da177e4 | 1999 | * @flags: SLAB flags |
1da177e4 LT |
2000 | * |
2001 | * Returns a ptr to the cache on success, NULL on failure. | |
2002 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2003 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2004 | * |
1da177e4 LT |
2005 | * The flags are |
2006 | * | |
2007 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2008 | * to catch references to uninitialised memory. | |
2009 | * | |
2010 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2011 | * for buffer overruns. | |
2012 | * | |
1da177e4 LT |
2013 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2014 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2015 | * as davem. | |
2016 | */ | |
278b1bb1 | 2017 | int |
8a13a4cc | 2018 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2019 | { |
d4a5fca5 | 2020 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2021 | gfp_t gfp; |
278b1bb1 | 2022 | int err; |
8a13a4cc | 2023 | size_t size = cachep->size; |
1da177e4 | 2024 | |
1da177e4 | 2025 | #if DEBUG |
1da177e4 LT |
2026 | #if FORCED_DEBUG |
2027 | /* | |
2028 | * Enable redzoning and last user accounting, except for caches with | |
2029 | * large objects, if the increased size would increase the object size | |
2030 | * above the next power of two: caches with object sizes just above a | |
2031 | * power of two have a significant amount of internal fragmentation. | |
2032 | */ | |
87a927c7 DW |
2033 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2034 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2035 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2036 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2037 | flags |= SLAB_POISON; | |
2038 | #endif | |
1da177e4 | 2039 | #endif |
1da177e4 | 2040 | |
a737b3e2 AM |
2041 | /* |
2042 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2043 | * unaligned accesses for some archs when redzoning is used, and makes |
2044 | * sure any on-slab bufctl's are also correctly aligned. | |
2045 | */ | |
b28a02de PE |
2046 | if (size & (BYTES_PER_WORD - 1)) { |
2047 | size += (BYTES_PER_WORD - 1); | |
2048 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2049 | } |
2050 | ||
87a927c7 DW |
2051 | if (flags & SLAB_RED_ZONE) { |
2052 | ralign = REDZONE_ALIGN; | |
2053 | /* If redzoning, ensure that the second redzone is suitably | |
2054 | * aligned, by adjusting the object size accordingly. */ | |
2055 | size += REDZONE_ALIGN - 1; | |
2056 | size &= ~(REDZONE_ALIGN - 1); | |
2057 | } | |
ca5f9703 | 2058 | |
a44b56d3 | 2059 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2060 | if (ralign < cachep->align) { |
2061 | ralign = cachep->align; | |
1da177e4 | 2062 | } |
3ff84a7f PE |
2063 | /* disable debug if necessary */ |
2064 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2065 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2066 | /* |
ca5f9703 | 2067 | * 4) Store it. |
1da177e4 | 2068 | */ |
8a13a4cc | 2069 | cachep->align = ralign; |
158e319b JK |
2070 | cachep->colour_off = cache_line_size(); |
2071 | /* Offset must be a multiple of the alignment. */ | |
2072 | if (cachep->colour_off < cachep->align) | |
2073 | cachep->colour_off = cachep->align; | |
1da177e4 | 2074 | |
83b519e8 PE |
2075 | if (slab_is_available()) |
2076 | gfp = GFP_KERNEL; | |
2077 | else | |
2078 | gfp = GFP_NOWAIT; | |
2079 | ||
1da177e4 | 2080 | #if DEBUG |
1da177e4 | 2081 | |
ca5f9703 PE |
2082 | /* |
2083 | * Both debugging options require word-alignment which is calculated | |
2084 | * into align above. | |
2085 | */ | |
1da177e4 | 2086 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2087 | /* add space for red zone words */ |
3ff84a7f PE |
2088 | cachep->obj_offset += sizeof(unsigned long long); |
2089 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2090 | } |
2091 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2092 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2093 | * the real object. But if the second red zone needs to be |
2094 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2095 | */ |
87a927c7 DW |
2096 | if (flags & SLAB_RED_ZONE) |
2097 | size += REDZONE_ALIGN; | |
2098 | else | |
2099 | size += BYTES_PER_WORD; | |
1da177e4 | 2100 | } |
832a15d2 JK |
2101 | #endif |
2102 | ||
7ed2f9e6 AP |
2103 | kasan_cache_create(cachep, &size, &flags); |
2104 | ||
832a15d2 JK |
2105 | size = ALIGN(size, cachep->align); |
2106 | /* | |
2107 | * We should restrict the number of objects in a slab to implement | |
2108 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2109 | */ | |
2110 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2111 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2112 | ||
2113 | #if DEBUG | |
03a2d2a3 JK |
2114 | /* |
2115 | * To activate debug pagealloc, off-slab management is necessary | |
2116 | * requirement. In early phase of initialization, small sized slab | |
2117 | * doesn't get initialized so it would not be possible. So, we need | |
2118 | * to check size >= 256. It guarantees that all necessary small | |
2119 | * sized slab is initialized in current slab initialization sequence. | |
2120 | */ | |
40323278 | 2121 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2122 | size >= 256 && cachep->object_size > cache_line_size()) { |
2123 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2124 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2125 | ||
2126 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2127 | flags |= CFLGS_OFF_SLAB; | |
2128 | cachep->obj_offset += tmp_size - size; | |
2129 | size = tmp_size; | |
2130 | goto done; | |
2131 | } | |
2132 | } | |
1da177e4 | 2133 | } |
1da177e4 LT |
2134 | #endif |
2135 | ||
b03a017b JK |
2136 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2137 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2138 | goto done; | |
2139 | } | |
2140 | ||
158e319b | 2141 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2142 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2143 | goto done; |
832a15d2 | 2144 | } |
1da177e4 | 2145 | |
158e319b JK |
2146 | if (set_on_slab_cache(cachep, size, flags)) |
2147 | goto done; | |
1da177e4 | 2148 | |
158e319b | 2149 | return -E2BIG; |
1da177e4 | 2150 | |
158e319b JK |
2151 | done: |
2152 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2153 | cachep->flags = flags; |
a57a4988 | 2154 | cachep->allocflags = __GFP_COMP; |
a3187e43 | 2155 | if (flags & SLAB_CACHE_DMA) |
a618e89f | 2156 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2157 | cachep->size = size; |
6a2d7a95 | 2158 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2159 | |
40b44137 JK |
2160 | #if DEBUG |
2161 | /* | |
2162 | * If we're going to use the generic kernel_map_pages() | |
2163 | * poisoning, then it's going to smash the contents of | |
2164 | * the redzone and userword anyhow, so switch them off. | |
2165 | */ | |
2166 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2167 | (cachep->flags & SLAB_POISON) && | |
2168 | is_debug_pagealloc_cache(cachep)) | |
2169 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2170 | #endif | |
2171 | ||
2172 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2173 | cachep->freelist_cache = |
2174 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2175 | } |
1da177e4 | 2176 | |
278b1bb1 CL |
2177 | err = setup_cpu_cache(cachep, gfp); |
2178 | if (err) { | |
52b4b950 | 2179 | __kmem_cache_release(cachep); |
278b1bb1 | 2180 | return err; |
2ed3a4ef | 2181 | } |
1da177e4 | 2182 | |
278b1bb1 | 2183 | return 0; |
1da177e4 | 2184 | } |
1da177e4 LT |
2185 | |
2186 | #if DEBUG | |
2187 | static void check_irq_off(void) | |
2188 | { | |
2189 | BUG_ON(!irqs_disabled()); | |
2190 | } | |
2191 | ||
2192 | static void check_irq_on(void) | |
2193 | { | |
2194 | BUG_ON(irqs_disabled()); | |
2195 | } | |
2196 | ||
18726ca8 JK |
2197 | static void check_mutex_acquired(void) |
2198 | { | |
2199 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2200 | } | |
2201 | ||
343e0d7a | 2202 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2203 | { |
2204 | #ifdef CONFIG_SMP | |
2205 | check_irq_off(); | |
18bf8541 | 2206 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2207 | #endif |
2208 | } | |
e498be7d | 2209 | |
343e0d7a | 2210 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2211 | { |
2212 | #ifdef CONFIG_SMP | |
2213 | check_irq_off(); | |
18bf8541 | 2214 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2215 | #endif |
2216 | } | |
2217 | ||
1da177e4 LT |
2218 | #else |
2219 | #define check_irq_off() do { } while(0) | |
2220 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2221 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2222 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2223 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2224 | #endif |
2225 | ||
18726ca8 JK |
2226 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2227 | int node, bool free_all, struct list_head *list) | |
2228 | { | |
2229 | int tofree; | |
2230 | ||
2231 | if (!ac || !ac->avail) | |
2232 | return; | |
2233 | ||
2234 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2235 | if (tofree > ac->avail) | |
2236 | tofree = (ac->avail + 1) / 2; | |
2237 | ||
2238 | free_block(cachep, ac->entry, tofree, node, list); | |
2239 | ac->avail -= tofree; | |
2240 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2241 | } | |
aab2207c | 2242 | |
1da177e4 LT |
2243 | static void do_drain(void *arg) |
2244 | { | |
a737b3e2 | 2245 | struct kmem_cache *cachep = arg; |
1da177e4 | 2246 | struct array_cache *ac; |
7d6e6d09 | 2247 | int node = numa_mem_id(); |
18bf8541 | 2248 | struct kmem_cache_node *n; |
97654dfa | 2249 | LIST_HEAD(list); |
1da177e4 LT |
2250 | |
2251 | check_irq_off(); | |
9a2dba4b | 2252 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2253 | n = get_node(cachep, node); |
2254 | spin_lock(&n->list_lock); | |
97654dfa | 2255 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2256 | spin_unlock(&n->list_lock); |
97654dfa | 2257 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2258 | ac->avail = 0; |
2259 | } | |
2260 | ||
343e0d7a | 2261 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2262 | { |
ce8eb6c4 | 2263 | struct kmem_cache_node *n; |
e498be7d | 2264 | int node; |
18726ca8 | 2265 | LIST_HEAD(list); |
e498be7d | 2266 | |
15c8b6c1 | 2267 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2268 | check_irq_on(); |
18bf8541 CL |
2269 | for_each_kmem_cache_node(cachep, node, n) |
2270 | if (n->alien) | |
ce8eb6c4 | 2271 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2272 | |
18726ca8 JK |
2273 | for_each_kmem_cache_node(cachep, node, n) { |
2274 | spin_lock_irq(&n->list_lock); | |
2275 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2276 | spin_unlock_irq(&n->list_lock); | |
2277 | ||
2278 | slabs_destroy(cachep, &list); | |
2279 | } | |
1da177e4 LT |
2280 | } |
2281 | ||
ed11d9eb CL |
2282 | /* |
2283 | * Remove slabs from the list of free slabs. | |
2284 | * Specify the number of slabs to drain in tofree. | |
2285 | * | |
2286 | * Returns the actual number of slabs released. | |
2287 | */ | |
2288 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2289 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2290 | { |
ed11d9eb CL |
2291 | struct list_head *p; |
2292 | int nr_freed; | |
8456a648 | 2293 | struct page *page; |
1da177e4 | 2294 | |
ed11d9eb | 2295 | nr_freed = 0; |
ce8eb6c4 | 2296 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2297 | |
ce8eb6c4 CL |
2298 | spin_lock_irq(&n->list_lock); |
2299 | p = n->slabs_free.prev; | |
2300 | if (p == &n->slabs_free) { | |
2301 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2302 | goto out; |
2303 | } | |
1da177e4 | 2304 | |
8456a648 | 2305 | page = list_entry(p, struct page, lru); |
8456a648 | 2306 | list_del(&page->lru); |
f728b0a5 | 2307 | n->free_slabs--; |
bf00bd34 | 2308 | n->total_slabs--; |
ed11d9eb CL |
2309 | /* |
2310 | * Safe to drop the lock. The slab is no longer linked | |
2311 | * to the cache. | |
2312 | */ | |
ce8eb6c4 CL |
2313 | n->free_objects -= cache->num; |
2314 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2315 | slab_destroy(cache, page); |
ed11d9eb | 2316 | nr_freed++; |
1da177e4 | 2317 | } |
ed11d9eb CL |
2318 | out: |
2319 | return nr_freed; | |
1da177e4 LT |
2320 | } |
2321 | ||
89e364db | 2322 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2323 | { |
18bf8541 CL |
2324 | int ret = 0; |
2325 | int node; | |
ce8eb6c4 | 2326 | struct kmem_cache_node *n; |
e498be7d CL |
2327 | |
2328 | drain_cpu_caches(cachep); | |
2329 | ||
2330 | check_irq_on(); | |
18bf8541 | 2331 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2332 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2333 | |
ce8eb6c4 CL |
2334 | ret += !list_empty(&n->slabs_full) || |
2335 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2336 | } |
2337 | return (ret ? 1 : 0); | |
2338 | } | |
2339 | ||
945cf2b6 | 2340 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 | 2341 | { |
89e364db | 2342 | return __kmem_cache_shrink(cachep); |
52b4b950 DS |
2343 | } |
2344 | ||
2345 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2346 | { |
12c3667f | 2347 | int i; |
ce8eb6c4 | 2348 | struct kmem_cache_node *n; |
1da177e4 | 2349 | |
c7ce4f60 TG |
2350 | cache_random_seq_destroy(cachep); |
2351 | ||
bf0dea23 | 2352 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2353 | |
ce8eb6c4 | 2354 | /* NUMA: free the node structures */ |
18bf8541 CL |
2355 | for_each_kmem_cache_node(cachep, i, n) { |
2356 | kfree(n->shared); | |
2357 | free_alien_cache(n->alien); | |
2358 | kfree(n); | |
2359 | cachep->node[i] = NULL; | |
12c3667f | 2360 | } |
1da177e4 | 2361 | } |
1da177e4 | 2362 | |
e5ac9c5a RT |
2363 | /* |
2364 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2365 | * |
2366 | * For a slab cache when the slab descriptor is off-slab, the | |
2367 | * slab descriptor can't come from the same cache which is being created, | |
2368 | * Because if it is the case, that means we defer the creation of | |
2369 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2370 | * And we eventually call down to __kmem_cache_create(), which | |
2371 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2372 | * This is a "chicken-and-egg" problem. | |
2373 | * | |
2374 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2375 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2376 | */ |
7e007355 | 2377 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2378 | struct page *page, int colour_off, |
2379 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2380 | { |
7e007355 | 2381 | void *freelist; |
0c3aa83e | 2382 | void *addr = page_address(page); |
b28a02de | 2383 | |
2e6b3602 JK |
2384 | page->s_mem = addr + colour_off; |
2385 | page->active = 0; | |
2386 | ||
b03a017b JK |
2387 | if (OBJFREELIST_SLAB(cachep)) |
2388 | freelist = NULL; | |
2389 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2390 | /* Slab management obj is off-slab. */ |
8456a648 | 2391 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2392 | local_flags, nodeid); |
8456a648 | 2393 | if (!freelist) |
1da177e4 LT |
2394 | return NULL; |
2395 | } else { | |
2e6b3602 JK |
2396 | /* We will use last bytes at the slab for freelist */ |
2397 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2398 | cachep->freelist_size; | |
1da177e4 | 2399 | } |
2e6b3602 | 2400 | |
8456a648 | 2401 | return freelist; |
1da177e4 LT |
2402 | } |
2403 | ||
7cc68973 | 2404 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2405 | { |
a41adfaa | 2406 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2407 | } |
2408 | ||
2409 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2410 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2411 | { |
a41adfaa | 2412 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2413 | } |
2414 | ||
10b2e9e8 | 2415 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2416 | { |
10b2e9e8 | 2417 | #if DEBUG |
1da177e4 LT |
2418 | int i; |
2419 | ||
2420 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2421 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2422 | |
1da177e4 LT |
2423 | if (cachep->flags & SLAB_STORE_USER) |
2424 | *dbg_userword(cachep, objp) = NULL; | |
2425 | ||
2426 | if (cachep->flags & SLAB_RED_ZONE) { | |
2427 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2428 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2429 | } | |
2430 | /* | |
a737b3e2 AM |
2431 | * Constructors are not allowed to allocate memory from the same |
2432 | * cache which they are a constructor for. Otherwise, deadlock. | |
2433 | * They must also be threaded. | |
1da177e4 | 2434 | */ |
7ed2f9e6 AP |
2435 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2436 | kasan_unpoison_object_data(cachep, | |
2437 | objp + obj_offset(cachep)); | |
51cc5068 | 2438 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2439 | kasan_poison_object_data( |
2440 | cachep, objp + obj_offset(cachep)); | |
2441 | } | |
1da177e4 LT |
2442 | |
2443 | if (cachep->flags & SLAB_RED_ZONE) { | |
2444 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2445 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2446 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2447 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2448 | } |
40b44137 JK |
2449 | /* need to poison the objs? */ |
2450 | if (cachep->flags & SLAB_POISON) { | |
2451 | poison_obj(cachep, objp, POISON_FREE); | |
2452 | slab_kernel_map(cachep, objp, 0, 0); | |
2453 | } | |
10b2e9e8 | 2454 | } |
1da177e4 | 2455 | #endif |
10b2e9e8 JK |
2456 | } |
2457 | ||
c7ce4f60 TG |
2458 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
2459 | /* Hold information during a freelist initialization */ | |
2460 | union freelist_init_state { | |
2461 | struct { | |
2462 | unsigned int pos; | |
7c00fce9 | 2463 | unsigned int *list; |
c7ce4f60 TG |
2464 | unsigned int count; |
2465 | unsigned int rand; | |
2466 | }; | |
2467 | struct rnd_state rnd_state; | |
2468 | }; | |
2469 | ||
2470 | /* | |
2471 | * Initialize the state based on the randomization methode available. | |
2472 | * return true if the pre-computed list is available, false otherwize. | |
2473 | */ | |
2474 | static bool freelist_state_initialize(union freelist_init_state *state, | |
2475 | struct kmem_cache *cachep, | |
2476 | unsigned int count) | |
2477 | { | |
2478 | bool ret; | |
2479 | unsigned int rand; | |
2480 | ||
2481 | /* Use best entropy available to define a random shift */ | |
7c00fce9 | 2482 | rand = get_random_int(); |
c7ce4f60 TG |
2483 | |
2484 | /* Use a random state if the pre-computed list is not available */ | |
2485 | if (!cachep->random_seq) { | |
2486 | prandom_seed_state(&state->rnd_state, rand); | |
2487 | ret = false; | |
2488 | } else { | |
2489 | state->list = cachep->random_seq; | |
2490 | state->count = count; | |
2491 | state->pos = 0; | |
2492 | state->rand = rand; | |
2493 | ret = true; | |
2494 | } | |
2495 | return ret; | |
2496 | } | |
2497 | ||
2498 | /* Get the next entry on the list and randomize it using a random shift */ | |
2499 | static freelist_idx_t next_random_slot(union freelist_init_state *state) | |
2500 | { | |
2501 | return (state->list[state->pos++] + state->rand) % state->count; | |
2502 | } | |
2503 | ||
7c00fce9 TG |
2504 | /* Swap two freelist entries */ |
2505 | static void swap_free_obj(struct page *page, unsigned int a, unsigned int b) | |
2506 | { | |
2507 | swap(((freelist_idx_t *)page->freelist)[a], | |
2508 | ((freelist_idx_t *)page->freelist)[b]); | |
2509 | } | |
2510 | ||
c7ce4f60 TG |
2511 | /* |
2512 | * Shuffle the freelist initialization state based on pre-computed lists. | |
2513 | * return true if the list was successfully shuffled, false otherwise. | |
2514 | */ | |
2515 | static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page) | |
2516 | { | |
7c00fce9 | 2517 | unsigned int objfreelist = 0, i, rand, count = cachep->num; |
c7ce4f60 TG |
2518 | union freelist_init_state state; |
2519 | bool precomputed; | |
2520 | ||
2521 | if (count < 2) | |
2522 | return false; | |
2523 | ||
2524 | precomputed = freelist_state_initialize(&state, cachep, count); | |
2525 | ||
2526 | /* Take a random entry as the objfreelist */ | |
2527 | if (OBJFREELIST_SLAB(cachep)) { | |
2528 | if (!precomputed) | |
2529 | objfreelist = count - 1; | |
2530 | else | |
2531 | objfreelist = next_random_slot(&state); | |
2532 | page->freelist = index_to_obj(cachep, page, objfreelist) + | |
2533 | obj_offset(cachep); | |
2534 | count--; | |
2535 | } | |
2536 | ||
2537 | /* | |
2538 | * On early boot, generate the list dynamically. | |
2539 | * Later use a pre-computed list for speed. | |
2540 | */ | |
2541 | if (!precomputed) { | |
7c00fce9 TG |
2542 | for (i = 0; i < count; i++) |
2543 | set_free_obj(page, i, i); | |
2544 | ||
2545 | /* Fisher-Yates shuffle */ | |
2546 | for (i = count - 1; i > 0; i--) { | |
2547 | rand = prandom_u32_state(&state.rnd_state); | |
2548 | rand %= (i + 1); | |
2549 | swap_free_obj(page, i, rand); | |
2550 | } | |
c7ce4f60 TG |
2551 | } else { |
2552 | for (i = 0; i < count; i++) | |
2553 | set_free_obj(page, i, next_random_slot(&state)); | |
2554 | } | |
2555 | ||
2556 | if (OBJFREELIST_SLAB(cachep)) | |
2557 | set_free_obj(page, cachep->num - 1, objfreelist); | |
2558 | ||
2559 | return true; | |
2560 | } | |
2561 | #else | |
2562 | static inline bool shuffle_freelist(struct kmem_cache *cachep, | |
2563 | struct page *page) | |
2564 | { | |
2565 | return false; | |
2566 | } | |
2567 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
2568 | ||
10b2e9e8 JK |
2569 | static void cache_init_objs(struct kmem_cache *cachep, |
2570 | struct page *page) | |
2571 | { | |
2572 | int i; | |
7ed2f9e6 | 2573 | void *objp; |
c7ce4f60 | 2574 | bool shuffled; |
10b2e9e8 JK |
2575 | |
2576 | cache_init_objs_debug(cachep, page); | |
2577 | ||
c7ce4f60 TG |
2578 | /* Try to randomize the freelist if enabled */ |
2579 | shuffled = shuffle_freelist(cachep, page); | |
2580 | ||
2581 | if (!shuffled && OBJFREELIST_SLAB(cachep)) { | |
b03a017b JK |
2582 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + |
2583 | obj_offset(cachep); | |
2584 | } | |
2585 | ||
10b2e9e8 | 2586 | for (i = 0; i < cachep->num; i++) { |
b3cbd9bf AR |
2587 | objp = index_to_obj(cachep, page, i); |
2588 | kasan_init_slab_obj(cachep, objp); | |
2589 | ||
10b2e9e8 | 2590 | /* constructor could break poison info */ |
7ed2f9e6 | 2591 | if (DEBUG == 0 && cachep->ctor) { |
7ed2f9e6 AP |
2592 | kasan_unpoison_object_data(cachep, objp); |
2593 | cachep->ctor(objp); | |
2594 | kasan_poison_object_data(cachep, objp); | |
2595 | } | |
10b2e9e8 | 2596 | |
c7ce4f60 TG |
2597 | if (!shuffled) |
2598 | set_free_obj(page, i, i); | |
1da177e4 | 2599 | } |
1da177e4 LT |
2600 | } |
2601 | ||
260b61dd | 2602 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2603 | { |
b1cb0982 | 2604 | void *objp; |
78d382d7 | 2605 | |
e5c58dfd | 2606 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2607 | page->active++; |
78d382d7 | 2608 | |
d31676df JK |
2609 | #if DEBUG |
2610 | if (cachep->flags & SLAB_STORE_USER) | |
2611 | set_store_user_dirty(cachep); | |
2612 | #endif | |
2613 | ||
78d382d7 MD |
2614 | return objp; |
2615 | } | |
2616 | ||
260b61dd JK |
2617 | static void slab_put_obj(struct kmem_cache *cachep, |
2618 | struct page *page, void *objp) | |
78d382d7 | 2619 | { |
8456a648 | 2620 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2621 | #if DEBUG |
16025177 | 2622 | unsigned int i; |
b1cb0982 | 2623 | |
b1cb0982 | 2624 | /* Verify double free bug */ |
8456a648 | 2625 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2626 | if (get_free_obj(page, i) == objnr) { |
1170532b | 2627 | pr_err("slab: double free detected in cache '%s', objp %p\n", |
756a025f | 2628 | cachep->name, objp); |
b1cb0982 JK |
2629 | BUG(); |
2630 | } | |
78d382d7 MD |
2631 | } |
2632 | #endif | |
8456a648 | 2633 | page->active--; |
b03a017b JK |
2634 | if (!page->freelist) |
2635 | page->freelist = objp + obj_offset(cachep); | |
2636 | ||
e5c58dfd | 2637 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2638 | } |
2639 | ||
4776874f PE |
2640 | /* |
2641 | * Map pages beginning at addr to the given cache and slab. This is required | |
2642 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2643 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2644 | */ |
8456a648 | 2645 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2646 | void *freelist) |
1da177e4 | 2647 | { |
a57a4988 | 2648 | page->slab_cache = cache; |
8456a648 | 2649 | page->freelist = freelist; |
1da177e4 LT |
2650 | } |
2651 | ||
2652 | /* | |
2653 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2654 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2655 | */ | |
76b342bd JK |
2656 | static struct page *cache_grow_begin(struct kmem_cache *cachep, |
2657 | gfp_t flags, int nodeid) | |
1da177e4 | 2658 | { |
7e007355 | 2659 | void *freelist; |
b28a02de PE |
2660 | size_t offset; |
2661 | gfp_t local_flags; | |
511e3a05 | 2662 | int page_node; |
ce8eb6c4 | 2663 | struct kmem_cache_node *n; |
511e3a05 | 2664 | struct page *page; |
1da177e4 | 2665 | |
a737b3e2 AM |
2666 | /* |
2667 | * Be lazy and only check for valid flags here, keeping it out of the | |
2668 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2669 | */ |
c871ac4e | 2670 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
bacdcb34 | 2671 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
2672 | flags &= ~GFP_SLAB_BUG_MASK; |
2673 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
2674 | invalid_mask, &invalid_mask, flags, &flags); | |
2675 | dump_stack(); | |
c871ac4e | 2676 | } |
6cb06229 | 2677 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2678 | |
1da177e4 | 2679 | check_irq_off(); |
d0164adc | 2680 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2681 | local_irq_enable(); |
2682 | ||
a737b3e2 AM |
2683 | /* |
2684 | * Get mem for the objs. Attempt to allocate a physical page from | |
2685 | * 'nodeid'. | |
e498be7d | 2686 | */ |
511e3a05 | 2687 | page = kmem_getpages(cachep, local_flags, nodeid); |
0c3aa83e | 2688 | if (!page) |
1da177e4 LT |
2689 | goto failed; |
2690 | ||
511e3a05 JK |
2691 | page_node = page_to_nid(page); |
2692 | n = get_node(cachep, page_node); | |
03d1d43a JK |
2693 | |
2694 | /* Get colour for the slab, and cal the next value. */ | |
2695 | n->colour_next++; | |
2696 | if (n->colour_next >= cachep->colour) | |
2697 | n->colour_next = 0; | |
2698 | ||
2699 | offset = n->colour_next; | |
2700 | if (offset >= cachep->colour) | |
2701 | offset = 0; | |
2702 | ||
2703 | offset *= cachep->colour_off; | |
2704 | ||
1da177e4 | 2705 | /* Get slab management. */ |
8456a648 | 2706 | freelist = alloc_slabmgmt(cachep, page, offset, |
511e3a05 | 2707 | local_flags & ~GFP_CONSTRAINT_MASK, page_node); |
b03a017b | 2708 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2709 | goto opps1; |
2710 | ||
8456a648 | 2711 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2712 | |
7ed2f9e6 | 2713 | kasan_poison_slab(page); |
8456a648 | 2714 | cache_init_objs(cachep, page); |
1da177e4 | 2715 | |
d0164adc | 2716 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2717 | local_irq_disable(); |
1da177e4 | 2718 | |
76b342bd JK |
2719 | return page; |
2720 | ||
a737b3e2 | 2721 | opps1: |
0c3aa83e | 2722 | kmem_freepages(cachep, page); |
a737b3e2 | 2723 | failed: |
d0164adc | 2724 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2725 | local_irq_disable(); |
76b342bd JK |
2726 | return NULL; |
2727 | } | |
2728 | ||
2729 | static void cache_grow_end(struct kmem_cache *cachep, struct page *page) | |
2730 | { | |
2731 | struct kmem_cache_node *n; | |
2732 | void *list = NULL; | |
2733 | ||
2734 | check_irq_off(); | |
2735 | ||
2736 | if (!page) | |
2737 | return; | |
2738 | ||
2739 | INIT_LIST_HEAD(&page->lru); | |
2740 | n = get_node(cachep, page_to_nid(page)); | |
2741 | ||
2742 | spin_lock(&n->list_lock); | |
bf00bd34 | 2743 | n->total_slabs++; |
f728b0a5 | 2744 | if (!page->active) { |
76b342bd | 2745 | list_add_tail(&page->lru, &(n->slabs_free)); |
f728b0a5 | 2746 | n->free_slabs++; |
bf00bd34 | 2747 | } else |
76b342bd | 2748 | fixup_slab_list(cachep, n, page, &list); |
07a63c41 | 2749 | |
76b342bd JK |
2750 | STATS_INC_GROWN(cachep); |
2751 | n->free_objects += cachep->num - page->active; | |
2752 | spin_unlock(&n->list_lock); | |
2753 | ||
2754 | fixup_objfreelist_debug(cachep, &list); | |
1da177e4 LT |
2755 | } |
2756 | ||
2757 | #if DEBUG | |
2758 | ||
2759 | /* | |
2760 | * Perform extra freeing checks: | |
2761 | * - detect bad pointers. | |
2762 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2763 | */ |
2764 | static void kfree_debugcheck(const void *objp) | |
2765 | { | |
1da177e4 | 2766 | if (!virt_addr_valid(objp)) { |
1170532b | 2767 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2768 | (unsigned long)objp); |
2769 | BUG(); | |
1da177e4 | 2770 | } |
1da177e4 LT |
2771 | } |
2772 | ||
58ce1fd5 PE |
2773 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2774 | { | |
b46b8f19 | 2775 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2776 | |
2777 | redzone1 = *dbg_redzone1(cache, obj); | |
2778 | redzone2 = *dbg_redzone2(cache, obj); | |
2779 | ||
2780 | /* | |
2781 | * Redzone is ok. | |
2782 | */ | |
2783 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2784 | return; | |
2785 | ||
2786 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2787 | slab_error(cache, "double free detected"); | |
2788 | else | |
2789 | slab_error(cache, "memory outside object was overwritten"); | |
2790 | ||
1170532b JP |
2791 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2792 | obj, redzone1, redzone2); | |
58ce1fd5 PE |
2793 | } |
2794 | ||
343e0d7a | 2795 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2796 | unsigned long caller) |
1da177e4 | 2797 | { |
1da177e4 | 2798 | unsigned int objnr; |
8456a648 | 2799 | struct page *page; |
1da177e4 | 2800 | |
80cbd911 MW |
2801 | BUG_ON(virt_to_cache(objp) != cachep); |
2802 | ||
3dafccf2 | 2803 | objp -= obj_offset(cachep); |
1da177e4 | 2804 | kfree_debugcheck(objp); |
b49af68f | 2805 | page = virt_to_head_page(objp); |
1da177e4 | 2806 | |
1da177e4 | 2807 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2808 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2809 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2810 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2811 | } | |
d31676df JK |
2812 | if (cachep->flags & SLAB_STORE_USER) { |
2813 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2814 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2815 | } |
1da177e4 | 2816 | |
8456a648 | 2817 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2818 | |
2819 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2820 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2821 | |
1da177e4 | 2822 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2823 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2824 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2825 | } |
2826 | return objp; | |
2827 | } | |
2828 | ||
1da177e4 LT |
2829 | #else |
2830 | #define kfree_debugcheck(x) do { } while(0) | |
2831 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2832 | #endif |
2833 | ||
b03a017b JK |
2834 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2835 | void **list) | |
2836 | { | |
2837 | #if DEBUG | |
2838 | void *next = *list; | |
2839 | void *objp; | |
2840 | ||
2841 | while (next) { | |
2842 | objp = next - obj_offset(cachep); | |
2843 | next = *(void **)next; | |
2844 | poison_obj(cachep, objp, POISON_FREE); | |
2845 | } | |
2846 | #endif | |
2847 | } | |
2848 | ||
d8410234 | 2849 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2850 | struct kmem_cache_node *n, struct page *page, |
2851 | void **list) | |
d8410234 JK |
2852 | { |
2853 | /* move slabp to correct slabp list: */ | |
2854 | list_del(&page->lru); | |
b03a017b | 2855 | if (page->active == cachep->num) { |
d8410234 | 2856 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2857 | if (OBJFREELIST_SLAB(cachep)) { |
2858 | #if DEBUG | |
2859 | /* Poisoning will be done without holding the lock */ | |
2860 | if (cachep->flags & SLAB_POISON) { | |
2861 | void **objp = page->freelist; | |
2862 | ||
2863 | *objp = *list; | |
2864 | *list = objp; | |
2865 | } | |
2866 | #endif | |
2867 | page->freelist = NULL; | |
2868 | } | |
2869 | } else | |
d8410234 JK |
2870 | list_add(&page->lru, &n->slabs_partial); |
2871 | } | |
2872 | ||
f68f8ddd JK |
2873 | /* Try to find non-pfmemalloc slab if needed */ |
2874 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
bf00bd34 | 2875 | struct page *page, bool pfmemalloc) |
f68f8ddd JK |
2876 | { |
2877 | if (!page) | |
2878 | return NULL; | |
2879 | ||
2880 | if (pfmemalloc) | |
2881 | return page; | |
2882 | ||
2883 | if (!PageSlabPfmemalloc(page)) | |
2884 | return page; | |
2885 | ||
2886 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2887 | if (n->free_objects > n->free_limit) { | |
2888 | ClearPageSlabPfmemalloc(page); | |
2889 | return page; | |
2890 | } | |
2891 | ||
2892 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2893 | list_del(&page->lru); | |
bf00bd34 | 2894 | if (!page->active) { |
f68f8ddd | 2895 | list_add_tail(&page->lru, &n->slabs_free); |
bf00bd34 | 2896 | n->free_slabs++; |
f728b0a5 | 2897 | } else |
f68f8ddd JK |
2898 | list_add_tail(&page->lru, &n->slabs_partial); |
2899 | ||
2900 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2901 | if (!PageSlabPfmemalloc(page)) | |
2902 | return page; | |
2903 | } | |
2904 | ||
f728b0a5 | 2905 | n->free_touched = 1; |
f68f8ddd | 2906 | list_for_each_entry(page, &n->slabs_free, lru) { |
f728b0a5 | 2907 | if (!PageSlabPfmemalloc(page)) { |
bf00bd34 | 2908 | n->free_slabs--; |
f68f8ddd | 2909 | return page; |
f728b0a5 | 2910 | } |
f68f8ddd JK |
2911 | } |
2912 | ||
2913 | return NULL; | |
2914 | } | |
2915 | ||
2916 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2917 | { |
2918 | struct page *page; | |
2919 | ||
f728b0a5 | 2920 | assert_spin_locked(&n->list_lock); |
bf00bd34 | 2921 | page = list_first_entry_or_null(&n->slabs_partial, struct page, lru); |
7aa0d227 GT |
2922 | if (!page) { |
2923 | n->free_touched = 1; | |
bf00bd34 DR |
2924 | page = list_first_entry_or_null(&n->slabs_free, struct page, |
2925 | lru); | |
f728b0a5 | 2926 | if (page) |
bf00bd34 | 2927 | n->free_slabs--; |
7aa0d227 GT |
2928 | } |
2929 | ||
f68f8ddd | 2930 | if (sk_memalloc_socks()) |
bf00bd34 | 2931 | page = get_valid_first_slab(n, page, pfmemalloc); |
f68f8ddd | 2932 | |
7aa0d227 GT |
2933 | return page; |
2934 | } | |
2935 | ||
f68f8ddd JK |
2936 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2937 | struct kmem_cache_node *n, gfp_t flags) | |
2938 | { | |
2939 | struct page *page; | |
2940 | void *obj; | |
2941 | void *list = NULL; | |
2942 | ||
2943 | if (!gfp_pfmemalloc_allowed(flags)) | |
2944 | return NULL; | |
2945 | ||
2946 | spin_lock(&n->list_lock); | |
2947 | page = get_first_slab(n, true); | |
2948 | if (!page) { | |
2949 | spin_unlock(&n->list_lock); | |
2950 | return NULL; | |
2951 | } | |
2952 | ||
2953 | obj = slab_get_obj(cachep, page); | |
2954 | n->free_objects--; | |
2955 | ||
2956 | fixup_slab_list(cachep, n, page, &list); | |
2957 | ||
2958 | spin_unlock(&n->list_lock); | |
2959 | fixup_objfreelist_debug(cachep, &list); | |
2960 | ||
2961 | return obj; | |
2962 | } | |
2963 | ||
213b4695 JK |
2964 | /* |
2965 | * Slab list should be fixed up by fixup_slab_list() for existing slab | |
2966 | * or cache_grow_end() for new slab | |
2967 | */ | |
2968 | static __always_inline int alloc_block(struct kmem_cache *cachep, | |
2969 | struct array_cache *ac, struct page *page, int batchcount) | |
2970 | { | |
2971 | /* | |
2972 | * There must be at least one object available for | |
2973 | * allocation. | |
2974 | */ | |
2975 | BUG_ON(page->active >= cachep->num); | |
2976 | ||
2977 | while (page->active < cachep->num && batchcount--) { | |
2978 | STATS_INC_ALLOCED(cachep); | |
2979 | STATS_INC_ACTIVE(cachep); | |
2980 | STATS_SET_HIGH(cachep); | |
2981 | ||
2982 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); | |
2983 | } | |
2984 | ||
2985 | return batchcount; | |
2986 | } | |
2987 | ||
f68f8ddd | 2988 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2989 | { |
2990 | int batchcount; | |
ce8eb6c4 | 2991 | struct kmem_cache_node *n; |
801faf0d | 2992 | struct array_cache *ac, *shared; |
1ca4cb24 | 2993 | int node; |
b03a017b | 2994 | void *list = NULL; |
76b342bd | 2995 | struct page *page; |
1ca4cb24 | 2996 | |
1da177e4 | 2997 | check_irq_off(); |
7d6e6d09 | 2998 | node = numa_mem_id(); |
f68f8ddd | 2999 | |
9a2dba4b | 3000 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
3001 | batchcount = ac->batchcount; |
3002 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
3003 | /* |
3004 | * If there was little recent activity on this cache, then | |
3005 | * perform only a partial refill. Otherwise we could generate | |
3006 | * refill bouncing. | |
1da177e4 LT |
3007 | */ |
3008 | batchcount = BATCHREFILL_LIMIT; | |
3009 | } | |
18bf8541 | 3010 | n = get_node(cachep, node); |
e498be7d | 3011 | |
ce8eb6c4 | 3012 | BUG_ON(ac->avail > 0 || !n); |
801faf0d JK |
3013 | shared = READ_ONCE(n->shared); |
3014 | if (!n->free_objects && (!shared || !shared->avail)) | |
3015 | goto direct_grow; | |
3016 | ||
ce8eb6c4 | 3017 | spin_lock(&n->list_lock); |
801faf0d | 3018 | shared = READ_ONCE(n->shared); |
1da177e4 | 3019 | |
3ded175a | 3020 | /* See if we can refill from the shared array */ |
801faf0d JK |
3021 | if (shared && transfer_objects(ac, shared, batchcount)) { |
3022 | shared->touched = 1; | |
3ded175a | 3023 | goto alloc_done; |
44b57f1c | 3024 | } |
3ded175a | 3025 | |
1da177e4 | 3026 | while (batchcount > 0) { |
1da177e4 | 3027 | /* Get slab alloc is to come from. */ |
f68f8ddd | 3028 | page = get_first_slab(n, false); |
7aa0d227 GT |
3029 | if (!page) |
3030 | goto must_grow; | |
1da177e4 | 3031 | |
1da177e4 | 3032 | check_spinlock_acquired(cachep); |
714b8171 | 3033 | |
213b4695 | 3034 | batchcount = alloc_block(cachep, ac, page, batchcount); |
b03a017b | 3035 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
3036 | } |
3037 | ||
a737b3e2 | 3038 | must_grow: |
ce8eb6c4 | 3039 | n->free_objects -= ac->avail; |
a737b3e2 | 3040 | alloc_done: |
ce8eb6c4 | 3041 | spin_unlock(&n->list_lock); |
b03a017b | 3042 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 | 3043 | |
801faf0d | 3044 | direct_grow: |
1da177e4 | 3045 | if (unlikely(!ac->avail)) { |
f68f8ddd JK |
3046 | /* Check if we can use obj in pfmemalloc slab */ |
3047 | if (sk_memalloc_socks()) { | |
3048 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
3049 | ||
3050 | if (obj) | |
3051 | return obj; | |
3052 | } | |
3053 | ||
76b342bd | 3054 | page = cache_grow_begin(cachep, gfp_exact_node(flags), node); |
e498be7d | 3055 | |
76b342bd JK |
3056 | /* |
3057 | * cache_grow_begin() can reenable interrupts, | |
3058 | * then ac could change. | |
3059 | */ | |
9a2dba4b | 3060 | ac = cpu_cache_get(cachep); |
213b4695 JK |
3061 | if (!ac->avail && page) |
3062 | alloc_block(cachep, ac, page, batchcount); | |
3063 | cache_grow_end(cachep, page); | |
072bb0aa | 3064 | |
213b4695 | 3065 | if (!ac->avail) |
1da177e4 | 3066 | return NULL; |
1da177e4 LT |
3067 | } |
3068 | ac->touched = 1; | |
072bb0aa | 3069 | |
f68f8ddd | 3070 | return ac->entry[--ac->avail]; |
1da177e4 LT |
3071 | } |
3072 | ||
a737b3e2 AM |
3073 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3074 | gfp_t flags) | |
1da177e4 | 3075 | { |
d0164adc | 3076 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
3077 | } |
3078 | ||
3079 | #if DEBUG | |
a737b3e2 | 3080 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3081 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3082 | { |
b28a02de | 3083 | if (!objp) |
1da177e4 | 3084 | return objp; |
b28a02de | 3085 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3086 | check_poison_obj(cachep, objp); |
40b44137 | 3087 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
3088 | poison_obj(cachep, objp, POISON_INUSE); |
3089 | } | |
3090 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3091 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3092 | |
3093 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3094 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3095 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 3096 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
1170532b JP |
3097 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
3098 | objp, *dbg_redzone1(cachep, objp), | |
3099 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3100 | } |
3101 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3102 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3103 | } | |
03787301 | 3104 | |
3dafccf2 | 3105 | objp += obj_offset(cachep); |
4f104934 | 3106 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3107 | cachep->ctor(objp); |
7ea466f2 TH |
3108 | if (ARCH_SLAB_MINALIGN && |
3109 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
1170532b | 3110 | pr_err("0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3111 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3112 | } |
1da177e4 LT |
3113 | return objp; |
3114 | } | |
3115 | #else | |
3116 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3117 | #endif | |
3118 | ||
343e0d7a | 3119 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3120 | { |
b28a02de | 3121 | void *objp; |
1da177e4 LT |
3122 | struct array_cache *ac; |
3123 | ||
5c382300 | 3124 | check_irq_off(); |
8a8b6502 | 3125 | |
9a2dba4b | 3126 | ac = cpu_cache_get(cachep); |
1da177e4 | 3127 | if (likely(ac->avail)) { |
1da177e4 | 3128 | ac->touched = 1; |
f68f8ddd | 3129 | objp = ac->entry[--ac->avail]; |
072bb0aa | 3130 | |
f68f8ddd JK |
3131 | STATS_INC_ALLOCHIT(cachep); |
3132 | goto out; | |
1da177e4 | 3133 | } |
072bb0aa MG |
3134 | |
3135 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 3136 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
3137 | /* |
3138 | * the 'ac' may be updated by cache_alloc_refill(), | |
3139 | * and kmemleak_erase() requires its correct value. | |
3140 | */ | |
3141 | ac = cpu_cache_get(cachep); | |
3142 | ||
3143 | out: | |
d5cff635 CM |
3144 | /* |
3145 | * To avoid a false negative, if an object that is in one of the | |
3146 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3147 | * treat the array pointers as a reference to the object. | |
3148 | */ | |
f3d8b53a O |
3149 | if (objp) |
3150 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3151 | return objp; |
3152 | } | |
3153 | ||
e498be7d | 3154 | #ifdef CONFIG_NUMA |
c61afb18 | 3155 | /* |
2ad654bc | 3156 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3157 | * |
3158 | * If we are in_interrupt, then process context, including cpusets and | |
3159 | * mempolicy, may not apply and should not be used for allocation policy. | |
3160 | */ | |
3161 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3162 | { | |
3163 | int nid_alloc, nid_here; | |
3164 | ||
765c4507 | 3165 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3166 | return NULL; |
7d6e6d09 | 3167 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3168 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3169 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3170 | else if (current->mempolicy) |
2a389610 | 3171 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3172 | if (nid_alloc != nid_here) |
8b98c169 | 3173 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3174 | return NULL; |
3175 | } | |
3176 | ||
765c4507 CL |
3177 | /* |
3178 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3179 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3180 | * available node for available objects. If that fails then we |
3c517a61 CL |
3181 | * perform an allocation without specifying a node. This allows the page |
3182 | * allocator to do its reclaim / fallback magic. We then insert the | |
3183 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3184 | */ |
8c8cc2c1 | 3185 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3186 | { |
8c8cc2c1 | 3187 | struct zonelist *zonelist; |
dd1a239f | 3188 | struct zoneref *z; |
54a6eb5c MG |
3189 | struct zone *zone; |
3190 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3191 | void *obj = NULL; |
76b342bd | 3192 | struct page *page; |
3c517a61 | 3193 | int nid; |
cc9a6c87 | 3194 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3195 | |
3196 | if (flags & __GFP_THISNODE) | |
3197 | return NULL; | |
3198 | ||
cc9a6c87 | 3199 | retry_cpuset: |
d26914d1 | 3200 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3201 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3202 | |
3c517a61 CL |
3203 | retry: |
3204 | /* | |
3205 | * Look through allowed nodes for objects available | |
3206 | * from existing per node queues. | |
3207 | */ | |
54a6eb5c MG |
3208 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3209 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3210 | |
061d7074 | 3211 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3212 | get_node(cache, nid) && |
3213 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3214 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3215 | gfp_exact_node(flags), nid); |
481c5346 CL |
3216 | if (obj) |
3217 | break; | |
3218 | } | |
3c517a61 CL |
3219 | } |
3220 | ||
cfce6604 | 3221 | if (!obj) { |
3c517a61 CL |
3222 | /* |
3223 | * This allocation will be performed within the constraints | |
3224 | * of the current cpuset / memory policy requirements. | |
3225 | * We may trigger various forms of reclaim on the allowed | |
3226 | * set and go into memory reserves if necessary. | |
3227 | */ | |
76b342bd JK |
3228 | page = cache_grow_begin(cache, flags, numa_mem_id()); |
3229 | cache_grow_end(cache, page); | |
3230 | if (page) { | |
3231 | nid = page_to_nid(page); | |
511e3a05 JK |
3232 | obj = ____cache_alloc_node(cache, |
3233 | gfp_exact_node(flags), nid); | |
0c3aa83e | 3234 | |
3c517a61 | 3235 | /* |
511e3a05 JK |
3236 | * Another processor may allocate the objects in |
3237 | * the slab since we are not holding any locks. | |
3c517a61 | 3238 | */ |
511e3a05 JK |
3239 | if (!obj) |
3240 | goto retry; | |
3c517a61 | 3241 | } |
aedb0eb1 | 3242 | } |
cc9a6c87 | 3243 | |
d26914d1 | 3244 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3245 | goto retry_cpuset; |
765c4507 CL |
3246 | return obj; |
3247 | } | |
3248 | ||
e498be7d CL |
3249 | /* |
3250 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3251 | */ |
8b98c169 | 3252 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3253 | int nodeid) |
e498be7d | 3254 | { |
8456a648 | 3255 | struct page *page; |
ce8eb6c4 | 3256 | struct kmem_cache_node *n; |
213b4695 | 3257 | void *obj = NULL; |
b03a017b | 3258 | void *list = NULL; |
b28a02de | 3259 | |
7c3fbbdd | 3260 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3261 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3262 | BUG_ON(!n); |
b28a02de | 3263 | |
ca3b9b91 | 3264 | check_irq_off(); |
ce8eb6c4 | 3265 | spin_lock(&n->list_lock); |
f68f8ddd | 3266 | page = get_first_slab(n, false); |
7aa0d227 GT |
3267 | if (!page) |
3268 | goto must_grow; | |
b28a02de | 3269 | |
b28a02de | 3270 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3271 | |
3272 | STATS_INC_NODEALLOCS(cachep); | |
3273 | STATS_INC_ACTIVE(cachep); | |
3274 | STATS_SET_HIGH(cachep); | |
3275 | ||
8456a648 | 3276 | BUG_ON(page->active == cachep->num); |
b28a02de | 3277 | |
260b61dd | 3278 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3279 | n->free_objects--; |
b28a02de | 3280 | |
b03a017b | 3281 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3282 | |
ce8eb6c4 | 3283 | spin_unlock(&n->list_lock); |
b03a017b | 3284 | fixup_objfreelist_debug(cachep, &list); |
213b4695 | 3285 | return obj; |
e498be7d | 3286 | |
a737b3e2 | 3287 | must_grow: |
ce8eb6c4 | 3288 | spin_unlock(&n->list_lock); |
76b342bd | 3289 | page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid); |
213b4695 JK |
3290 | if (page) { |
3291 | /* This slab isn't counted yet so don't update free_objects */ | |
3292 | obj = slab_get_obj(cachep, page); | |
3293 | } | |
76b342bd | 3294 | cache_grow_end(cachep, page); |
1da177e4 | 3295 | |
213b4695 | 3296 | return obj ? obj : fallback_alloc(cachep, flags); |
e498be7d | 3297 | } |
8c8cc2c1 | 3298 | |
8c8cc2c1 | 3299 | static __always_inline void * |
48356303 | 3300 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3301 | unsigned long caller) |
8c8cc2c1 PE |
3302 | { |
3303 | unsigned long save_flags; | |
3304 | void *ptr; | |
7d6e6d09 | 3305 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3306 | |
dcce284a | 3307 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3308 | cachep = slab_pre_alloc_hook(cachep, flags); |
3309 | if (unlikely(!cachep)) | |
824ebef1 AM |
3310 | return NULL; |
3311 | ||
8c8cc2c1 PE |
3312 | cache_alloc_debugcheck_before(cachep, flags); |
3313 | local_irq_save(save_flags); | |
3314 | ||
eacbbae3 | 3315 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3316 | nodeid = slab_node; |
8c8cc2c1 | 3317 | |
18bf8541 | 3318 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3319 | /* Node not bootstrapped yet */ |
3320 | ptr = fallback_alloc(cachep, flags); | |
3321 | goto out; | |
3322 | } | |
3323 | ||
7d6e6d09 | 3324 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3325 | /* |
3326 | * Use the locally cached objects if possible. | |
3327 | * However ____cache_alloc does not allow fallback | |
3328 | * to other nodes. It may fail while we still have | |
3329 | * objects on other nodes available. | |
3330 | */ | |
3331 | ptr = ____cache_alloc(cachep, flags); | |
3332 | if (ptr) | |
3333 | goto out; | |
3334 | } | |
3335 | /* ___cache_alloc_node can fall back to other nodes */ | |
3336 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3337 | out: | |
3338 | local_irq_restore(save_flags); | |
3339 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3340 | ||
d5e3ed66 JDB |
3341 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3342 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3343 | |
d5e3ed66 | 3344 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3345 | return ptr; |
3346 | } | |
3347 | ||
3348 | static __always_inline void * | |
3349 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3350 | { | |
3351 | void *objp; | |
3352 | ||
2ad654bc | 3353 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3354 | objp = alternate_node_alloc(cache, flags); |
3355 | if (objp) | |
3356 | goto out; | |
3357 | } | |
3358 | objp = ____cache_alloc(cache, flags); | |
3359 | ||
3360 | /* | |
3361 | * We may just have run out of memory on the local node. | |
3362 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3363 | */ | |
7d6e6d09 LS |
3364 | if (!objp) |
3365 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3366 | |
3367 | out: | |
3368 | return objp; | |
3369 | } | |
3370 | #else | |
3371 | ||
3372 | static __always_inline void * | |
3373 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3374 | { | |
3375 | return ____cache_alloc(cachep, flags); | |
3376 | } | |
3377 | ||
3378 | #endif /* CONFIG_NUMA */ | |
3379 | ||
3380 | static __always_inline void * | |
48356303 | 3381 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3382 | { |
3383 | unsigned long save_flags; | |
3384 | void *objp; | |
3385 | ||
dcce284a | 3386 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3387 | cachep = slab_pre_alloc_hook(cachep, flags); |
3388 | if (unlikely(!cachep)) | |
824ebef1 AM |
3389 | return NULL; |
3390 | ||
8c8cc2c1 PE |
3391 | cache_alloc_debugcheck_before(cachep, flags); |
3392 | local_irq_save(save_flags); | |
3393 | objp = __do_cache_alloc(cachep, flags); | |
3394 | local_irq_restore(save_flags); | |
3395 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3396 | prefetchw(objp); | |
3397 | ||
d5e3ed66 JDB |
3398 | if (unlikely(flags & __GFP_ZERO) && objp) |
3399 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3400 | |
d5e3ed66 | 3401 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3402 | return objp; |
3403 | } | |
e498be7d CL |
3404 | |
3405 | /* | |
5f0985bb | 3406 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3407 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3408 | */ |
97654dfa JK |
3409 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3410 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3411 | { |
3412 | int i; | |
25c063fb | 3413 | struct kmem_cache_node *n = get_node(cachep, node); |
6052b788 JK |
3414 | struct page *page; |
3415 | ||
3416 | n->free_objects += nr_objects; | |
1da177e4 LT |
3417 | |
3418 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3419 | void *objp; |
8456a648 | 3420 | struct page *page; |
1da177e4 | 3421 | |
072bb0aa MG |
3422 | objp = objpp[i]; |
3423 | ||
8456a648 | 3424 | page = virt_to_head_page(objp); |
8456a648 | 3425 | list_del(&page->lru); |
ff69416e | 3426 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3427 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3428 | STATS_DEC_ACTIVE(cachep); |
1da177e4 LT |
3429 | |
3430 | /* fixup slab chains */ | |
f728b0a5 | 3431 | if (page->active == 0) { |
6052b788 | 3432 | list_add(&page->lru, &n->slabs_free); |
f728b0a5 | 3433 | n->free_slabs++; |
f728b0a5 | 3434 | } else { |
1da177e4 LT |
3435 | /* Unconditionally move a slab to the end of the |
3436 | * partial list on free - maximum time for the | |
3437 | * other objects to be freed, too. | |
3438 | */ | |
8456a648 | 3439 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3440 | } |
3441 | } | |
6052b788 JK |
3442 | |
3443 | while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) { | |
3444 | n->free_objects -= cachep->num; | |
3445 | ||
3446 | page = list_last_entry(&n->slabs_free, struct page, lru); | |
de24baec | 3447 | list_move(&page->lru, list); |
f728b0a5 | 3448 | n->free_slabs--; |
bf00bd34 | 3449 | n->total_slabs--; |
6052b788 | 3450 | } |
1da177e4 LT |
3451 | } |
3452 | ||
343e0d7a | 3453 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3454 | { |
3455 | int batchcount; | |
ce8eb6c4 | 3456 | struct kmem_cache_node *n; |
7d6e6d09 | 3457 | int node = numa_mem_id(); |
97654dfa | 3458 | LIST_HEAD(list); |
1da177e4 LT |
3459 | |
3460 | batchcount = ac->batchcount; | |
260b61dd | 3461 | |
1da177e4 | 3462 | check_irq_off(); |
18bf8541 | 3463 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3464 | spin_lock(&n->list_lock); |
3465 | if (n->shared) { | |
3466 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3467 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3468 | if (max) { |
3469 | if (batchcount > max) | |
3470 | batchcount = max; | |
e498be7d | 3471 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3472 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3473 | shared_array->avail += batchcount; |
3474 | goto free_done; | |
3475 | } | |
3476 | } | |
3477 | ||
97654dfa | 3478 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3479 | free_done: |
1da177e4 LT |
3480 | #if STATS |
3481 | { | |
3482 | int i = 0; | |
73c0219d | 3483 | struct page *page; |
1da177e4 | 3484 | |
73c0219d | 3485 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3486 | BUG_ON(page->active); |
1da177e4 LT |
3487 | |
3488 | i++; | |
1da177e4 LT |
3489 | } |
3490 | STATS_SET_FREEABLE(cachep, i); | |
3491 | } | |
3492 | #endif | |
ce8eb6c4 | 3493 | spin_unlock(&n->list_lock); |
97654dfa | 3494 | slabs_destroy(cachep, &list); |
1da177e4 | 3495 | ac->avail -= batchcount; |
a737b3e2 | 3496 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3497 | } |
3498 | ||
3499 | /* | |
a737b3e2 AM |
3500 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3501 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3502 | */ |
a947eb95 | 3503 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3504 | unsigned long caller) |
1da177e4 | 3505 | { |
55834c59 AP |
3506 | /* Put the object into the quarantine, don't touch it for now. */ |
3507 | if (kasan_slab_free(cachep, objp)) | |
3508 | return; | |
3509 | ||
3510 | ___cache_free(cachep, objp, caller); | |
3511 | } | |
1da177e4 | 3512 | |
55834c59 AP |
3513 | void ___cache_free(struct kmem_cache *cachep, void *objp, |
3514 | unsigned long caller) | |
3515 | { | |
3516 | struct array_cache *ac = cpu_cache_get(cachep); | |
7ed2f9e6 | 3517 | |
1da177e4 | 3518 | check_irq_off(); |
d5cff635 | 3519 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3520 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3521 | |
8c138bc0 | 3522 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3523 | |
1807a1aa SS |
3524 | /* |
3525 | * Skip calling cache_free_alien() when the platform is not numa. | |
3526 | * This will avoid cache misses that happen while accessing slabp (which | |
3527 | * is per page memory reference) to get nodeid. Instead use a global | |
3528 | * variable to skip the call, which is mostly likely to be present in | |
3529 | * the cache. | |
3530 | */ | |
b6e68bc1 | 3531 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3532 | return; |
3533 | ||
3d880194 | 3534 | if (ac->avail < ac->limit) { |
1da177e4 | 3535 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3536 | } else { |
3537 | STATS_INC_FREEMISS(cachep); | |
3538 | cache_flusharray(cachep, ac); | |
1da177e4 | 3539 | } |
42c8c99c | 3540 | |
f68f8ddd JK |
3541 | if (sk_memalloc_socks()) { |
3542 | struct page *page = virt_to_head_page(objp); | |
3543 | ||
3544 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3545 | cache_free_pfmemalloc(cachep, page, objp); | |
3546 | return; | |
3547 | } | |
3548 | } | |
3549 | ||
3550 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3551 | } |
3552 | ||
3553 | /** | |
3554 | * kmem_cache_alloc - Allocate an object | |
3555 | * @cachep: The cache to allocate from. | |
3556 | * @flags: See kmalloc(). | |
3557 | * | |
3558 | * Allocate an object from this cache. The flags are only relevant | |
3559 | * if the cache has no available objects. | |
3560 | */ | |
343e0d7a | 3561 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3562 | { |
48356303 | 3563 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3564 | |
505f5dcb | 3565 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3566 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3567 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3568 | |
3569 | return ret; | |
1da177e4 LT |
3570 | } |
3571 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3572 | ||
7b0501dd JDB |
3573 | static __always_inline void |
3574 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3575 | size_t size, void **p, unsigned long caller) | |
3576 | { | |
3577 | size_t i; | |
3578 | ||
3579 | for (i = 0; i < size; i++) | |
3580 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3581 | } | |
3582 | ||
865762a8 | 3583 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3584 | void **p) |
484748f0 | 3585 | { |
2a777eac JDB |
3586 | size_t i; |
3587 | ||
3588 | s = slab_pre_alloc_hook(s, flags); | |
3589 | if (!s) | |
3590 | return 0; | |
3591 | ||
3592 | cache_alloc_debugcheck_before(s, flags); | |
3593 | ||
3594 | local_irq_disable(); | |
3595 | for (i = 0; i < size; i++) { | |
3596 | void *objp = __do_cache_alloc(s, flags); | |
3597 | ||
2a777eac JDB |
3598 | if (unlikely(!objp)) |
3599 | goto error; | |
3600 | p[i] = objp; | |
3601 | } | |
3602 | local_irq_enable(); | |
3603 | ||
7b0501dd JDB |
3604 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3605 | ||
2a777eac JDB |
3606 | /* Clear memory outside IRQ disabled section */ |
3607 | if (unlikely(flags & __GFP_ZERO)) | |
3608 | for (i = 0; i < size; i++) | |
3609 | memset(p[i], 0, s->object_size); | |
3610 | ||
3611 | slab_post_alloc_hook(s, flags, size, p); | |
3612 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3613 | return size; | |
3614 | error: | |
3615 | local_irq_enable(); | |
7b0501dd | 3616 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3617 | slab_post_alloc_hook(s, flags, i, p); |
3618 | __kmem_cache_free_bulk(s, i, p); | |
3619 | return 0; | |
484748f0 CL |
3620 | } |
3621 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3622 | ||
0f24f128 | 3623 | #ifdef CONFIG_TRACING |
85beb586 | 3624 | void * |
4052147c | 3625 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3626 | { |
85beb586 SR |
3627 | void *ret; |
3628 | ||
48356303 | 3629 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3630 | |
505f5dcb | 3631 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3632 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3633 | size, cachep->size, flags); |
85beb586 | 3634 | return ret; |
36555751 | 3635 | } |
85beb586 | 3636 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3637 | #endif |
3638 | ||
1da177e4 | 3639 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3640 | /** |
3641 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3642 | * @cachep: The cache to allocate from. | |
3643 | * @flags: See kmalloc(). | |
3644 | * @nodeid: node number of the target node. | |
3645 | * | |
3646 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3647 | * node, which can improve the performance for cpu bound structures. | |
3648 | * | |
3649 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3650 | */ | |
8b98c169 CH |
3651 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3652 | { | |
48356303 | 3653 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3654 | |
505f5dcb | 3655 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3656 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3657 | cachep->object_size, cachep->size, |
ca2b84cb | 3658 | flags, nodeid); |
36555751 EGM |
3659 | |
3660 | return ret; | |
8b98c169 | 3661 | } |
1da177e4 LT |
3662 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3663 | ||
0f24f128 | 3664 | #ifdef CONFIG_TRACING |
4052147c | 3665 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3666 | gfp_t flags, |
4052147c EG |
3667 | int nodeid, |
3668 | size_t size) | |
36555751 | 3669 | { |
85beb586 SR |
3670 | void *ret; |
3671 | ||
592f4145 | 3672 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3673 | |
3674 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3675 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3676 | size, cachep->size, |
85beb586 SR |
3677 | flags, nodeid); |
3678 | return ret; | |
36555751 | 3679 | } |
85beb586 | 3680 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3681 | #endif |
3682 | ||
8b98c169 | 3683 | static __always_inline void * |
7c0cb9c6 | 3684 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3685 | { |
343e0d7a | 3686 | struct kmem_cache *cachep; |
7ed2f9e6 | 3687 | void *ret; |
97e2bde4 | 3688 | |
2c59dd65 | 3689 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3690 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3691 | return cachep; | |
7ed2f9e6 | 3692 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3693 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3694 | |
3695 | return ret; | |
97e2bde4 | 3696 | } |
8b98c169 | 3697 | |
8b98c169 CH |
3698 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3699 | { | |
7c0cb9c6 | 3700 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3701 | } |
dbe5e69d | 3702 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3703 | |
3704 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3705 | int node, unsigned long caller) |
8b98c169 | 3706 | { |
7c0cb9c6 | 3707 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3708 | } |
3709 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3710 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3711 | |
3712 | /** | |
800590f5 | 3713 | * __do_kmalloc - allocate memory |
1da177e4 | 3714 | * @size: how many bytes of memory are required. |
800590f5 | 3715 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3716 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3717 | */ |
7fd6b141 | 3718 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3719 | unsigned long caller) |
1da177e4 | 3720 | { |
343e0d7a | 3721 | struct kmem_cache *cachep; |
36555751 | 3722 | void *ret; |
1da177e4 | 3723 | |
2c59dd65 | 3724 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3725 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3726 | return cachep; | |
48356303 | 3727 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3728 | |
505f5dcb | 3729 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3730 | trace_kmalloc(caller, ret, |
3b0efdfa | 3731 | size, cachep->size, flags); |
36555751 EGM |
3732 | |
3733 | return ret; | |
7fd6b141 PE |
3734 | } |
3735 | ||
7fd6b141 PE |
3736 | void *__kmalloc(size_t size, gfp_t flags) |
3737 | { | |
7c0cb9c6 | 3738 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3739 | } |
3740 | EXPORT_SYMBOL(__kmalloc); | |
3741 | ||
ce71e27c | 3742 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3743 | { |
7c0cb9c6 | 3744 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3745 | } |
3746 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3747 | |
1da177e4 LT |
3748 | /** |
3749 | * kmem_cache_free - Deallocate an object | |
3750 | * @cachep: The cache the allocation was from. | |
3751 | * @objp: The previously allocated object. | |
3752 | * | |
3753 | * Free an object which was previously allocated from this | |
3754 | * cache. | |
3755 | */ | |
343e0d7a | 3756 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3757 | { |
3758 | unsigned long flags; | |
b9ce5ef4 GC |
3759 | cachep = cache_from_obj(cachep, objp); |
3760 | if (!cachep) | |
3761 | return; | |
1da177e4 LT |
3762 | |
3763 | local_irq_save(flags); | |
d97d476b | 3764 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3765 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3766 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3767 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3768 | local_irq_restore(flags); |
36555751 | 3769 | |
ca2b84cb | 3770 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3771 | } |
3772 | EXPORT_SYMBOL(kmem_cache_free); | |
3773 | ||
e6cdb58d JDB |
3774 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3775 | { | |
3776 | struct kmem_cache *s; | |
3777 | size_t i; | |
3778 | ||
3779 | local_irq_disable(); | |
3780 | for (i = 0; i < size; i++) { | |
3781 | void *objp = p[i]; | |
3782 | ||
ca257195 JDB |
3783 | if (!orig_s) /* called via kfree_bulk */ |
3784 | s = virt_to_cache(objp); | |
3785 | else | |
3786 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3787 | |
3788 | debug_check_no_locks_freed(objp, s->object_size); | |
3789 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3790 | debug_check_no_obj_freed(objp, s->object_size); | |
3791 | ||
3792 | __cache_free(s, objp, _RET_IP_); | |
3793 | } | |
3794 | local_irq_enable(); | |
3795 | ||
3796 | /* FIXME: add tracing */ | |
3797 | } | |
3798 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3799 | ||
1da177e4 LT |
3800 | /** |
3801 | * kfree - free previously allocated memory | |
3802 | * @objp: pointer returned by kmalloc. | |
3803 | * | |
80e93eff PE |
3804 | * If @objp is NULL, no operation is performed. |
3805 | * | |
1da177e4 LT |
3806 | * Don't free memory not originally allocated by kmalloc() |
3807 | * or you will run into trouble. | |
3808 | */ | |
3809 | void kfree(const void *objp) | |
3810 | { | |
343e0d7a | 3811 | struct kmem_cache *c; |
1da177e4 LT |
3812 | unsigned long flags; |
3813 | ||
2121db74 PE |
3814 | trace_kfree(_RET_IP_, objp); |
3815 | ||
6cb8f913 | 3816 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3817 | return; |
3818 | local_irq_save(flags); | |
3819 | kfree_debugcheck(objp); | |
6ed5eb22 | 3820 | c = virt_to_cache(objp); |
8c138bc0 CL |
3821 | debug_check_no_locks_freed(objp, c->object_size); |
3822 | ||
3823 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3824 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3825 | local_irq_restore(flags); |
3826 | } | |
3827 | EXPORT_SYMBOL(kfree); | |
3828 | ||
e498be7d | 3829 | /* |
ce8eb6c4 | 3830 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3831 | */ |
c3d332b6 | 3832 | static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d | 3833 | { |
c3d332b6 | 3834 | int ret; |
e498be7d | 3835 | int node; |
ce8eb6c4 | 3836 | struct kmem_cache_node *n; |
e498be7d | 3837 | |
9c09a95c | 3838 | for_each_online_node(node) { |
c3d332b6 JK |
3839 | ret = setup_kmem_cache_node(cachep, node, gfp, true); |
3840 | if (ret) | |
e498be7d CL |
3841 | goto fail; |
3842 | ||
e498be7d | 3843 | } |
c3d332b6 | 3844 | |
cafeb02e | 3845 | return 0; |
0718dc2a | 3846 | |
a737b3e2 | 3847 | fail: |
3b0efdfa | 3848 | if (!cachep->list.next) { |
0718dc2a CL |
3849 | /* Cache is not active yet. Roll back what we did */ |
3850 | node--; | |
3851 | while (node >= 0) { | |
18bf8541 CL |
3852 | n = get_node(cachep, node); |
3853 | if (n) { | |
ce8eb6c4 CL |
3854 | kfree(n->shared); |
3855 | free_alien_cache(n->alien); | |
3856 | kfree(n); | |
6a67368c | 3857 | cachep->node[node] = NULL; |
0718dc2a CL |
3858 | } |
3859 | node--; | |
3860 | } | |
3861 | } | |
cafeb02e | 3862 | return -ENOMEM; |
e498be7d CL |
3863 | } |
3864 | ||
18004c5d | 3865 | /* Always called with the slab_mutex held */ |
943a451a | 3866 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3867 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3868 | { |
bf0dea23 JK |
3869 | struct array_cache __percpu *cpu_cache, *prev; |
3870 | int cpu; | |
1da177e4 | 3871 | |
bf0dea23 JK |
3872 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3873 | if (!cpu_cache) | |
d2e7b7d0 SS |
3874 | return -ENOMEM; |
3875 | ||
bf0dea23 JK |
3876 | prev = cachep->cpu_cache; |
3877 | cachep->cpu_cache = cpu_cache; | |
3878 | kick_all_cpus_sync(); | |
e498be7d | 3879 | |
1da177e4 | 3880 | check_irq_on(); |
1da177e4 LT |
3881 | cachep->batchcount = batchcount; |
3882 | cachep->limit = limit; | |
e498be7d | 3883 | cachep->shared = shared; |
1da177e4 | 3884 | |
bf0dea23 | 3885 | if (!prev) |
c3d332b6 | 3886 | goto setup_node; |
bf0dea23 JK |
3887 | |
3888 | for_each_online_cpu(cpu) { | |
97654dfa | 3889 | LIST_HEAD(list); |
18bf8541 CL |
3890 | int node; |
3891 | struct kmem_cache_node *n; | |
bf0dea23 | 3892 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3893 | |
bf0dea23 | 3894 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3895 | n = get_node(cachep, node); |
3896 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3897 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3898 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3899 | slabs_destroy(cachep, &list); |
1da177e4 | 3900 | } |
bf0dea23 JK |
3901 | free_percpu(prev); |
3902 | ||
c3d332b6 JK |
3903 | setup_node: |
3904 | return setup_kmem_cache_nodes(cachep, gfp); | |
1da177e4 LT |
3905 | } |
3906 | ||
943a451a GC |
3907 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3908 | int batchcount, int shared, gfp_t gfp) | |
3909 | { | |
3910 | int ret; | |
426589f5 | 3911 | struct kmem_cache *c; |
943a451a GC |
3912 | |
3913 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3914 | ||
3915 | if (slab_state < FULL) | |
3916 | return ret; | |
3917 | ||
3918 | if ((ret < 0) || !is_root_cache(cachep)) | |
3919 | return ret; | |
3920 | ||
426589f5 VD |
3921 | lockdep_assert_held(&slab_mutex); |
3922 | for_each_memcg_cache(c, cachep) { | |
3923 | /* return value determined by the root cache only */ | |
3924 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3925 | } |
3926 | ||
3927 | return ret; | |
3928 | } | |
3929 | ||
18004c5d | 3930 | /* Called with slab_mutex held always */ |
83b519e8 | 3931 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3932 | { |
3933 | int err; | |
943a451a GC |
3934 | int limit = 0; |
3935 | int shared = 0; | |
3936 | int batchcount = 0; | |
3937 | ||
7c00fce9 | 3938 | err = cache_random_seq_create(cachep, cachep->num, gfp); |
c7ce4f60 TG |
3939 | if (err) |
3940 | goto end; | |
3941 | ||
943a451a GC |
3942 | if (!is_root_cache(cachep)) { |
3943 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3944 | limit = root->limit; | |
3945 | shared = root->shared; | |
3946 | batchcount = root->batchcount; | |
3947 | } | |
1da177e4 | 3948 | |
943a451a GC |
3949 | if (limit && shared && batchcount) |
3950 | goto skip_setup; | |
a737b3e2 AM |
3951 | /* |
3952 | * The head array serves three purposes: | |
1da177e4 LT |
3953 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3954 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3955 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3956 | * bufctl chains: array operations are cheaper. |
3957 | * The numbers are guessed, we should auto-tune as described by | |
3958 | * Bonwick. | |
3959 | */ | |
3b0efdfa | 3960 | if (cachep->size > 131072) |
1da177e4 | 3961 | limit = 1; |
3b0efdfa | 3962 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3963 | limit = 8; |
3b0efdfa | 3964 | else if (cachep->size > 1024) |
1da177e4 | 3965 | limit = 24; |
3b0efdfa | 3966 | else if (cachep->size > 256) |
1da177e4 LT |
3967 | limit = 54; |
3968 | else | |
3969 | limit = 120; | |
3970 | ||
a737b3e2 AM |
3971 | /* |
3972 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3973 | * allocation behaviour: Most allocs on one cpu, most free operations |
3974 | * on another cpu. For these cases, an efficient object passing between | |
3975 | * cpus is necessary. This is provided by a shared array. The array | |
3976 | * replaces Bonwick's magazine layer. | |
3977 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3978 | * to a larger limit. Thus disabled by default. | |
3979 | */ | |
3980 | shared = 0; | |
3b0efdfa | 3981 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3982 | shared = 8; |
1da177e4 LT |
3983 | |
3984 | #if DEBUG | |
a737b3e2 AM |
3985 | /* |
3986 | * With debugging enabled, large batchcount lead to excessively long | |
3987 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3988 | */ |
3989 | if (limit > 32) | |
3990 | limit = 32; | |
3991 | #endif | |
943a451a GC |
3992 | batchcount = (limit + 1) / 2; |
3993 | skip_setup: | |
3994 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
c7ce4f60 | 3995 | end: |
1da177e4 | 3996 | if (err) |
1170532b | 3997 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3998 | cachep->name, -err); |
2ed3a4ef | 3999 | return err; |
1da177e4 LT |
4000 | } |
4001 | ||
1b55253a | 4002 | /* |
ce8eb6c4 CL |
4003 | * Drain an array if it contains any elements taking the node lock only if |
4004 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 4005 | * if drain_array() is used on the shared array. |
1b55253a | 4006 | */ |
ce8eb6c4 | 4007 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 4008 | struct array_cache *ac, int node) |
1da177e4 | 4009 | { |
97654dfa | 4010 | LIST_HEAD(list); |
18726ca8 JK |
4011 | |
4012 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
4013 | check_mutex_acquired(); | |
1da177e4 | 4014 | |
1b55253a CL |
4015 | if (!ac || !ac->avail) |
4016 | return; | |
18726ca8 JK |
4017 | |
4018 | if (ac->touched) { | |
1da177e4 | 4019 | ac->touched = 0; |
18726ca8 | 4020 | return; |
1da177e4 | 4021 | } |
18726ca8 JK |
4022 | |
4023 | spin_lock_irq(&n->list_lock); | |
4024 | drain_array_locked(cachep, ac, node, false, &list); | |
4025 | spin_unlock_irq(&n->list_lock); | |
4026 | ||
4027 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
4028 | } |
4029 | ||
4030 | /** | |
4031 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4032 | * @w: work descriptor |
1da177e4 LT |
4033 | * |
4034 | * Called from workqueue/eventd every few seconds. | |
4035 | * Purpose: | |
4036 | * - clear the per-cpu caches for this CPU. | |
4037 | * - return freeable pages to the main free memory pool. | |
4038 | * | |
a737b3e2 AM |
4039 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4040 | * again on the next iteration. | |
1da177e4 | 4041 | */ |
7c5cae36 | 4042 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4043 | { |
7a7c381d | 4044 | struct kmem_cache *searchp; |
ce8eb6c4 | 4045 | struct kmem_cache_node *n; |
7d6e6d09 | 4046 | int node = numa_mem_id(); |
bf6aede7 | 4047 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4048 | |
18004c5d | 4049 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4050 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4051 | goto out; |
1da177e4 | 4052 | |
18004c5d | 4053 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4054 | check_irq_on(); |
4055 | ||
35386e3b | 4056 | /* |
ce8eb6c4 | 4057 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4058 | * have established with reasonable certainty that |
4059 | * we can do some work if the lock was obtained. | |
4060 | */ | |
18bf8541 | 4061 | n = get_node(searchp, node); |
35386e3b | 4062 | |
ce8eb6c4 | 4063 | reap_alien(searchp, n); |
1da177e4 | 4064 | |
18726ca8 | 4065 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 4066 | |
35386e3b CL |
4067 | /* |
4068 | * These are racy checks but it does not matter | |
4069 | * if we skip one check or scan twice. | |
4070 | */ | |
ce8eb6c4 | 4071 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4072 | goto next; |
1da177e4 | 4073 | |
5f0985bb | 4074 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4075 | |
18726ca8 | 4076 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 4077 | |
ce8eb6c4 CL |
4078 | if (n->free_touched) |
4079 | n->free_touched = 0; | |
ed11d9eb CL |
4080 | else { |
4081 | int freed; | |
1da177e4 | 4082 | |
ce8eb6c4 | 4083 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4084 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4085 | STATS_ADD_REAPED(searchp, freed); | |
4086 | } | |
35386e3b | 4087 | next: |
1da177e4 LT |
4088 | cond_resched(); |
4089 | } | |
4090 | check_irq_on(); | |
18004c5d | 4091 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4092 | next_reap_node(); |
7c5cae36 | 4093 | out: |
a737b3e2 | 4094 | /* Set up the next iteration */ |
5f0985bb | 4095 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
4096 | } |
4097 | ||
158a9624 | 4098 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4099 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4100 | { |
f728b0a5 | 4101 | unsigned long active_objs, num_objs, active_slabs; |
bf00bd34 DR |
4102 | unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0; |
4103 | unsigned long free_slabs = 0; | |
e498be7d | 4104 | int node; |
ce8eb6c4 | 4105 | struct kmem_cache_node *n; |
1da177e4 | 4106 | |
18bf8541 | 4107 | for_each_kmem_cache_node(cachep, node, n) { |
ca3b9b91 | 4108 | check_irq_on(); |
ce8eb6c4 | 4109 | spin_lock_irq(&n->list_lock); |
e498be7d | 4110 | |
bf00bd34 DR |
4111 | total_slabs += n->total_slabs; |
4112 | free_slabs += n->free_slabs; | |
f728b0a5 | 4113 | free_objs += n->free_objects; |
07a63c41 | 4114 | |
ce8eb6c4 CL |
4115 | if (n->shared) |
4116 | shared_avail += n->shared->avail; | |
e498be7d | 4117 | |
ce8eb6c4 | 4118 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4119 | } |
bf00bd34 DR |
4120 | num_objs = total_slabs * cachep->num; |
4121 | active_slabs = total_slabs - free_slabs; | |
f728b0a5 | 4122 | active_objs = num_objs - free_objs; |
1da177e4 | 4123 | |
0d7561c6 GC |
4124 | sinfo->active_objs = active_objs; |
4125 | sinfo->num_objs = num_objs; | |
4126 | sinfo->active_slabs = active_slabs; | |
bf00bd34 | 4127 | sinfo->num_slabs = total_slabs; |
0d7561c6 GC |
4128 | sinfo->shared_avail = shared_avail; |
4129 | sinfo->limit = cachep->limit; | |
4130 | sinfo->batchcount = cachep->batchcount; | |
4131 | sinfo->shared = cachep->shared; | |
4132 | sinfo->objects_per_slab = cachep->num; | |
4133 | sinfo->cache_order = cachep->gfporder; | |
4134 | } | |
4135 | ||
4136 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4137 | { | |
1da177e4 | 4138 | #if STATS |
ce8eb6c4 | 4139 | { /* node stats */ |
1da177e4 LT |
4140 | unsigned long high = cachep->high_mark; |
4141 | unsigned long allocs = cachep->num_allocations; | |
4142 | unsigned long grown = cachep->grown; | |
4143 | unsigned long reaped = cachep->reaped; | |
4144 | unsigned long errors = cachep->errors; | |
4145 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4146 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4147 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4148 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4149 | |
756a025f | 4150 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4151 | allocs, high, grown, |
4152 | reaped, errors, max_freeable, node_allocs, | |
4153 | node_frees, overflows); | |
1da177e4 LT |
4154 | } |
4155 | /* cpu stats */ | |
4156 | { | |
4157 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4158 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4159 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4160 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4161 | ||
4162 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4163 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4164 | } |
4165 | #endif | |
1da177e4 LT |
4166 | } |
4167 | ||
1da177e4 LT |
4168 | #define MAX_SLABINFO_WRITE 128 |
4169 | /** | |
4170 | * slabinfo_write - Tuning for the slab allocator | |
4171 | * @file: unused | |
4172 | * @buffer: user buffer | |
4173 | * @count: data length | |
4174 | * @ppos: unused | |
4175 | */ | |
b7454ad3 | 4176 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4177 | size_t count, loff_t *ppos) |
1da177e4 | 4178 | { |
b28a02de | 4179 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4180 | int limit, batchcount, shared, res; |
7a7c381d | 4181 | struct kmem_cache *cachep; |
b28a02de | 4182 | |
1da177e4 LT |
4183 | if (count > MAX_SLABINFO_WRITE) |
4184 | return -EINVAL; | |
4185 | if (copy_from_user(&kbuf, buffer, count)) | |
4186 | return -EFAULT; | |
b28a02de | 4187 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4188 | |
4189 | tmp = strchr(kbuf, ' '); | |
4190 | if (!tmp) | |
4191 | return -EINVAL; | |
4192 | *tmp = '\0'; | |
4193 | tmp++; | |
4194 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4195 | return -EINVAL; | |
4196 | ||
4197 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4198 | mutex_lock(&slab_mutex); |
1da177e4 | 4199 | res = -EINVAL; |
18004c5d | 4200 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4201 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4202 | if (limit < 1 || batchcount < 1 || |
4203 | batchcount > limit || shared < 0) { | |
e498be7d | 4204 | res = 0; |
1da177e4 | 4205 | } else { |
e498be7d | 4206 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4207 | batchcount, shared, |
4208 | GFP_KERNEL); | |
1da177e4 LT |
4209 | } |
4210 | break; | |
4211 | } | |
4212 | } | |
18004c5d | 4213 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4214 | if (res >= 0) |
4215 | res = count; | |
4216 | return res; | |
4217 | } | |
871751e2 AV |
4218 | |
4219 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4220 | ||
871751e2 AV |
4221 | static inline int add_caller(unsigned long *n, unsigned long v) |
4222 | { | |
4223 | unsigned long *p; | |
4224 | int l; | |
4225 | if (!v) | |
4226 | return 1; | |
4227 | l = n[1]; | |
4228 | p = n + 2; | |
4229 | while (l) { | |
4230 | int i = l/2; | |
4231 | unsigned long *q = p + 2 * i; | |
4232 | if (*q == v) { | |
4233 | q[1]++; | |
4234 | return 1; | |
4235 | } | |
4236 | if (*q > v) { | |
4237 | l = i; | |
4238 | } else { | |
4239 | p = q + 2; | |
4240 | l -= i + 1; | |
4241 | } | |
4242 | } | |
4243 | if (++n[1] == n[0]) | |
4244 | return 0; | |
4245 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4246 | p[0] = v; | |
4247 | p[1] = 1; | |
4248 | return 1; | |
4249 | } | |
4250 | ||
8456a648 JK |
4251 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4252 | struct page *page) | |
871751e2 AV |
4253 | { |
4254 | void *p; | |
d31676df JK |
4255 | int i, j; |
4256 | unsigned long v; | |
b1cb0982 | 4257 | |
871751e2 AV |
4258 | if (n[0] == n[1]) |
4259 | return; | |
8456a648 | 4260 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4261 | bool active = true; |
4262 | ||
4263 | for (j = page->active; j < c->num; j++) { | |
4264 | if (get_free_obj(page, j) == i) { | |
4265 | active = false; | |
4266 | break; | |
4267 | } | |
4268 | } | |
4269 | ||
4270 | if (!active) | |
871751e2 | 4271 | continue; |
b1cb0982 | 4272 | |
d31676df JK |
4273 | /* |
4274 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4275 | * mapping is established when actual object allocation and | |
4276 | * we could mistakenly access the unmapped object in the cpu | |
4277 | * cache. | |
4278 | */ | |
4279 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4280 | continue; | |
4281 | ||
4282 | if (!add_caller(n, v)) | |
871751e2 AV |
4283 | return; |
4284 | } | |
4285 | } | |
4286 | ||
4287 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4288 | { | |
4289 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4290 | unsigned long offset, size; |
9281acea | 4291 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4292 | |
a5c43dae | 4293 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4294 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4295 | if (modname[0]) |
871751e2 AV |
4296 | seq_printf(m, " [%s]", modname); |
4297 | return; | |
4298 | } | |
4299 | #endif | |
4300 | seq_printf(m, "%p", (void *)address); | |
4301 | } | |
4302 | ||
4303 | static int leaks_show(struct seq_file *m, void *p) | |
4304 | { | |
0672aa7c | 4305 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4306 | struct page *page; |
ce8eb6c4 | 4307 | struct kmem_cache_node *n; |
871751e2 | 4308 | const char *name; |
db845067 | 4309 | unsigned long *x = m->private; |
871751e2 AV |
4310 | int node; |
4311 | int i; | |
4312 | ||
4313 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4314 | return 0; | |
4315 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4316 | return 0; | |
4317 | ||
d31676df JK |
4318 | /* |
4319 | * Set store_user_clean and start to grab stored user information | |
4320 | * for all objects on this cache. If some alloc/free requests comes | |
4321 | * during the processing, information would be wrong so restart | |
4322 | * whole processing. | |
4323 | */ | |
4324 | do { | |
4325 | set_store_user_clean(cachep); | |
4326 | drain_cpu_caches(cachep); | |
4327 | ||
4328 | x[1] = 0; | |
871751e2 | 4329 | |
d31676df | 4330 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4331 | |
d31676df JK |
4332 | check_irq_on(); |
4333 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4334 | |
d31676df JK |
4335 | list_for_each_entry(page, &n->slabs_full, lru) |
4336 | handle_slab(x, cachep, page); | |
4337 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4338 | handle_slab(x, cachep, page); | |
4339 | spin_unlock_irq(&n->list_lock); | |
4340 | } | |
4341 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4342 | |
871751e2 | 4343 | name = cachep->name; |
db845067 | 4344 | if (x[0] == x[1]) { |
871751e2 | 4345 | /* Increase the buffer size */ |
18004c5d | 4346 | mutex_unlock(&slab_mutex); |
db845067 | 4347 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4348 | if (!m->private) { |
4349 | /* Too bad, we are really out */ | |
db845067 | 4350 | m->private = x; |
18004c5d | 4351 | mutex_lock(&slab_mutex); |
871751e2 AV |
4352 | return -ENOMEM; |
4353 | } | |
db845067 CL |
4354 | *(unsigned long *)m->private = x[0] * 2; |
4355 | kfree(x); | |
18004c5d | 4356 | mutex_lock(&slab_mutex); |
871751e2 AV |
4357 | /* Now make sure this entry will be retried */ |
4358 | m->count = m->size; | |
4359 | return 0; | |
4360 | } | |
db845067 CL |
4361 | for (i = 0; i < x[1]; i++) { |
4362 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4363 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4364 | seq_putc(m, '\n'); |
4365 | } | |
d2e7b7d0 | 4366 | |
871751e2 AV |
4367 | return 0; |
4368 | } | |
4369 | ||
a0ec95a8 | 4370 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4371 | .start = slab_start, |
276a2439 WL |
4372 | .next = slab_next, |
4373 | .stop = slab_stop, | |
871751e2 AV |
4374 | .show = leaks_show, |
4375 | }; | |
a0ec95a8 AD |
4376 | |
4377 | static int slabstats_open(struct inode *inode, struct file *file) | |
4378 | { | |
b208ce32 RJ |
4379 | unsigned long *n; |
4380 | ||
4381 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4382 | if (!n) | |
4383 | return -ENOMEM; | |
4384 | ||
4385 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4386 | ||
4387 | return 0; | |
a0ec95a8 AD |
4388 | } |
4389 | ||
4390 | static const struct file_operations proc_slabstats_operations = { | |
4391 | .open = slabstats_open, | |
4392 | .read = seq_read, | |
4393 | .llseek = seq_lseek, | |
4394 | .release = seq_release_private, | |
4395 | }; | |
4396 | #endif | |
4397 | ||
4398 | static int __init slab_proc_init(void) | |
4399 | { | |
4400 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4401 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4402 | #endif |
a0ec95a8 AD |
4403 | return 0; |
4404 | } | |
4405 | module_init(slab_proc_init); | |
1da177e4 LT |
4406 | #endif |
4407 | ||
04385fc5 KC |
4408 | #ifdef CONFIG_HARDENED_USERCOPY |
4409 | /* | |
4410 | * Rejects objects that are incorrectly sized. | |
4411 | * | |
4412 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4413 | * to indicate an error. | |
4414 | */ | |
4415 | const char *__check_heap_object(const void *ptr, unsigned long n, | |
4416 | struct page *page) | |
4417 | { | |
4418 | struct kmem_cache *cachep; | |
4419 | unsigned int objnr; | |
4420 | unsigned long offset; | |
4421 | ||
4422 | /* Find and validate object. */ | |
4423 | cachep = page->slab_cache; | |
4424 | objnr = obj_to_index(cachep, page, (void *)ptr); | |
4425 | BUG_ON(objnr >= cachep->num); | |
4426 | ||
4427 | /* Find offset within object. */ | |
4428 | offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep); | |
4429 | ||
4430 | /* Allow address range falling entirely within object size. */ | |
4431 | if (offset <= cachep->object_size && n <= cachep->object_size - offset) | |
4432 | return NULL; | |
4433 | ||
4434 | return cachep->name; | |
4435 | } | |
4436 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4437 | ||
00e145b6 MS |
4438 | /** |
4439 | * ksize - get the actual amount of memory allocated for a given object | |
4440 | * @objp: Pointer to the object | |
4441 | * | |
4442 | * kmalloc may internally round up allocations and return more memory | |
4443 | * than requested. ksize() can be used to determine the actual amount of | |
4444 | * memory allocated. The caller may use this additional memory, even though | |
4445 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4446 | * The caller must guarantee that objp points to a valid object previously | |
4447 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4448 | * must not be freed during the duration of the call. | |
4449 | */ | |
fd76bab2 | 4450 | size_t ksize(const void *objp) |
1da177e4 | 4451 | { |
7ed2f9e6 AP |
4452 | size_t size; |
4453 | ||
ef8b4520 CL |
4454 | BUG_ON(!objp); |
4455 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4456 | return 0; |
1da177e4 | 4457 | |
7ed2f9e6 AP |
4458 | size = virt_to_cache(objp)->object_size; |
4459 | /* We assume that ksize callers could use the whole allocated area, | |
4460 | * so we need to unpoison this area. | |
4461 | */ | |
4ebb31a4 | 4462 | kasan_unpoison_shadow(objp, size); |
7ed2f9e6 AP |
4463 | |
4464 | return size; | |
1da177e4 | 4465 | } |
b1aabecd | 4466 | EXPORT_SYMBOL(ksize); |