]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
f315e3fa JK |
160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
162 | ||
163 | #if FREELIST_BYTE_INDEX | |
164 | typedef unsigned char freelist_idx_t; | |
165 | #else | |
166 | typedef unsigned short freelist_idx_t; | |
167 | #endif | |
168 | ||
30321c7b | 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 170 | |
1da177e4 LT |
171 | /* |
172 | * struct array_cache | |
173 | * | |
1da177e4 LT |
174 | * Purpose: |
175 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
176 | * - reduce the number of linked list operations | |
177 | * - reduce spinlock operations | |
178 | * | |
179 | * The limit is stored in the per-cpu structure to reduce the data cache | |
180 | * footprint. | |
181 | * | |
182 | */ | |
183 | struct array_cache { | |
184 | unsigned int avail; | |
185 | unsigned int limit; | |
186 | unsigned int batchcount; | |
187 | unsigned int touched; | |
bda5b655 | 188 | void *entry[]; /* |
a737b3e2 AM |
189 | * Must have this definition in here for the proper |
190 | * alignment of array_cache. Also simplifies accessing | |
191 | * the entries. | |
a737b3e2 | 192 | */ |
1da177e4 LT |
193 | }; |
194 | ||
c8522a3a JK |
195 | struct alien_cache { |
196 | spinlock_t lock; | |
197 | struct array_cache ac; | |
198 | }; | |
199 | ||
e498be7d CL |
200 | /* |
201 | * Need this for bootstrapping a per node allocator. | |
202 | */ | |
bf0dea23 | 203 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 204 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 205 | #define CACHE_CACHE 0 |
bf0dea23 | 206 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 207 | |
ed11d9eb | 208 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 209 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 210 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
211 | int node, struct list_head *list); |
212 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 213 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 214 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 215 | |
e0a42726 IM |
216 | static int slab_early_init = 1; |
217 | ||
ce8eb6c4 | 218 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 219 | |
ce8eb6c4 | 220 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
221 | { |
222 | INIT_LIST_HEAD(&parent->slabs_full); | |
223 | INIT_LIST_HEAD(&parent->slabs_partial); | |
224 | INIT_LIST_HEAD(&parent->slabs_free); | |
225 | parent->shared = NULL; | |
226 | parent->alien = NULL; | |
2e1217cf | 227 | parent->colour_next = 0; |
e498be7d CL |
228 | spin_lock_init(&parent->list_lock); |
229 | parent->free_objects = 0; | |
230 | parent->free_touched = 0; | |
231 | } | |
232 | ||
a737b3e2 AM |
233 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
234 | do { \ | |
235 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 236 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
237 | } while (0) |
238 | ||
a737b3e2 AM |
239 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
240 | do { \ | |
e498be7d CL |
241 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
242 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
243 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
244 | } while (0) | |
1da177e4 | 245 | |
b03a017b | 246 | #define CFLGS_OBJFREELIST_SLAB (0x40000000UL) |
1da177e4 | 247 | #define CFLGS_OFF_SLAB (0x80000000UL) |
b03a017b | 248 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
249 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
250 | ||
251 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
252 | /* |
253 | * Optimization question: fewer reaps means less probability for unnessary | |
254 | * cpucache drain/refill cycles. | |
1da177e4 | 255 | * |
dc6f3f27 | 256 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
257 | * which could lock up otherwise freeable slabs. |
258 | */ | |
5f0985bb JZ |
259 | #define REAPTIMEOUT_AC (2*HZ) |
260 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
261 | |
262 | #if STATS | |
263 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
264 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
265 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
266 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 267 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
268 | #define STATS_SET_HIGH(x) \ |
269 | do { \ | |
270 | if ((x)->num_active > (x)->high_mark) \ | |
271 | (x)->high_mark = (x)->num_active; \ | |
272 | } while (0) | |
1da177e4 LT |
273 | #define STATS_INC_ERR(x) ((x)->errors++) |
274 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 275 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 276 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
277 | #define STATS_SET_FREEABLE(x, i) \ |
278 | do { \ | |
279 | if ((x)->max_freeable < i) \ | |
280 | (x)->max_freeable = i; \ | |
281 | } while (0) | |
1da177e4 LT |
282 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
283 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
284 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
285 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
286 | #else | |
287 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
288 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
289 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
290 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 291 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
292 | #define STATS_SET_HIGH(x) do { } while (0) |
293 | #define STATS_INC_ERR(x) do { } while (0) | |
294 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 295 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 296 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 297 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
298 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
299 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
300 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
301 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
302 | #endif | |
303 | ||
304 | #if DEBUG | |
1da177e4 | 305 | |
a737b3e2 AM |
306 | /* |
307 | * memory layout of objects: | |
1da177e4 | 308 | * 0 : objp |
3dafccf2 | 309 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
310 | * the end of an object is aligned with the end of the real |
311 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 312 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 313 | * redzone word. |
3dafccf2 | 314 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
315 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
316 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 317 | * [BYTES_PER_WORD long] |
1da177e4 | 318 | */ |
343e0d7a | 319 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 320 | { |
3dafccf2 | 321 | return cachep->obj_offset; |
1da177e4 LT |
322 | } |
323 | ||
b46b8f19 | 324 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
325 | { |
326 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
327 | return (unsigned long long*) (objp + obj_offset(cachep) - |
328 | sizeof(unsigned long long)); | |
1da177e4 LT |
329 | } |
330 | ||
b46b8f19 | 331 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
332 | { |
333 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
334 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 335 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 336 | sizeof(unsigned long long) - |
87a927c7 | 337 | REDZONE_ALIGN); |
3b0efdfa | 338 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 339 | sizeof(unsigned long long)); |
1da177e4 LT |
340 | } |
341 | ||
343e0d7a | 342 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
343 | { |
344 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 345 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
346 | } |
347 | ||
348 | #else | |
349 | ||
3dafccf2 | 350 | #define obj_offset(x) 0 |
b46b8f19 DW |
351 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
352 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
353 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
354 | ||
355 | #endif | |
356 | ||
03787301 JK |
357 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
358 | ||
d31676df | 359 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 360 | { |
d31676df JK |
361 | return atomic_read(&cachep->store_user_clean) == 1; |
362 | } | |
03787301 | 363 | |
d31676df JK |
364 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
365 | { | |
366 | atomic_set(&cachep->store_user_clean, 1); | |
367 | } | |
03787301 | 368 | |
d31676df JK |
369 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
370 | { | |
371 | if (is_store_user_clean(cachep)) | |
372 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
373 | } |
374 | ||
375 | #else | |
d31676df | 376 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
377 | |
378 | #endif | |
379 | ||
1da177e4 | 380 | /* |
3df1cccd DR |
381 | * Do not go above this order unless 0 objects fit into the slab or |
382 | * overridden on the command line. | |
1da177e4 | 383 | */ |
543585cc DR |
384 | #define SLAB_MAX_ORDER_HI 1 |
385 | #define SLAB_MAX_ORDER_LO 0 | |
386 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 387 | static bool slab_max_order_set __initdata; |
1da177e4 | 388 | |
6ed5eb22 PE |
389 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
390 | { | |
b49af68f | 391 | struct page *page = virt_to_head_page(obj); |
35026088 | 392 | return page->slab_cache; |
6ed5eb22 PE |
393 | } |
394 | ||
8456a648 | 395 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
396 | unsigned int idx) |
397 | { | |
8456a648 | 398 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
399 | } |
400 | ||
6a2d7a95 | 401 | /* |
3b0efdfa CL |
402 | * We want to avoid an expensive divide : (offset / cache->size) |
403 | * Using the fact that size is a constant for a particular cache, | |
404 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
405 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
406 | */ | |
407 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 408 | const struct page *page, void *obj) |
8fea4e96 | 409 | { |
8456a648 | 410 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 411 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
412 | } |
413 | ||
6fb92430 | 414 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 415 | /* internal cache of cache description objs */ |
9b030cb8 | 416 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
417 | .batchcount = 1, |
418 | .limit = BOOT_CPUCACHE_ENTRIES, | |
419 | .shared = 1, | |
3b0efdfa | 420 | .size = sizeof(struct kmem_cache), |
b28a02de | 421 | .name = "kmem_cache", |
1da177e4 LT |
422 | }; |
423 | ||
1871e52c | 424 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 425 | |
343e0d7a | 426 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 427 | { |
bf0dea23 | 428 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
429 | } |
430 | ||
a737b3e2 AM |
431 | /* |
432 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
433 | */ | |
70f75067 JK |
434 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
435 | unsigned long flags, size_t *left_over) | |
fbaccacf | 436 | { |
70f75067 | 437 | unsigned int num; |
fbaccacf | 438 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 439 | |
fbaccacf SR |
440 | /* |
441 | * The slab management structure can be either off the slab or | |
442 | * on it. For the latter case, the memory allocated for a | |
443 | * slab is used for: | |
444 | * | |
fbaccacf | 445 | * - @buffer_size bytes for each object |
2e6b3602 JK |
446 | * - One freelist_idx_t for each object |
447 | * | |
448 | * We don't need to consider alignment of freelist because | |
449 | * freelist will be at the end of slab page. The objects will be | |
450 | * at the correct alignment. | |
fbaccacf SR |
451 | * |
452 | * If the slab management structure is off the slab, then the | |
453 | * alignment will already be calculated into the size. Because | |
454 | * the slabs are all pages aligned, the objects will be at the | |
455 | * correct alignment when allocated. | |
456 | */ | |
b03a017b | 457 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 458 | num = slab_size / buffer_size; |
2e6b3602 | 459 | *left_over = slab_size % buffer_size; |
fbaccacf | 460 | } else { |
70f75067 | 461 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
462 | *left_over = slab_size % |
463 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 464 | } |
70f75067 JK |
465 | |
466 | return num; | |
1da177e4 LT |
467 | } |
468 | ||
f28510d3 | 469 | #if DEBUG |
d40cee24 | 470 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 471 | |
a737b3e2 AM |
472 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
473 | char *msg) | |
1da177e4 | 474 | { |
1170532b | 475 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 476 | function, cachep->name, msg); |
1da177e4 | 477 | dump_stack(); |
373d4d09 | 478 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 479 | } |
f28510d3 | 480 | #endif |
1da177e4 | 481 | |
3395ee05 PM |
482 | /* |
483 | * By default on NUMA we use alien caches to stage the freeing of | |
484 | * objects allocated from other nodes. This causes massive memory | |
485 | * inefficiencies when using fake NUMA setup to split memory into a | |
486 | * large number of small nodes, so it can be disabled on the command | |
487 | * line | |
488 | */ | |
489 | ||
490 | static int use_alien_caches __read_mostly = 1; | |
491 | static int __init noaliencache_setup(char *s) | |
492 | { | |
493 | use_alien_caches = 0; | |
494 | return 1; | |
495 | } | |
496 | __setup("noaliencache", noaliencache_setup); | |
497 | ||
3df1cccd DR |
498 | static int __init slab_max_order_setup(char *str) |
499 | { | |
500 | get_option(&str, &slab_max_order); | |
501 | slab_max_order = slab_max_order < 0 ? 0 : | |
502 | min(slab_max_order, MAX_ORDER - 1); | |
503 | slab_max_order_set = true; | |
504 | ||
505 | return 1; | |
506 | } | |
507 | __setup("slab_max_order=", slab_max_order_setup); | |
508 | ||
8fce4d8e CL |
509 | #ifdef CONFIG_NUMA |
510 | /* | |
511 | * Special reaping functions for NUMA systems called from cache_reap(). | |
512 | * These take care of doing round robin flushing of alien caches (containing | |
513 | * objects freed on different nodes from which they were allocated) and the | |
514 | * flushing of remote pcps by calling drain_node_pages. | |
515 | */ | |
1871e52c | 516 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
517 | |
518 | static void init_reap_node(int cpu) | |
519 | { | |
520 | int node; | |
521 | ||
7d6e6d09 | 522 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 523 | if (node == MAX_NUMNODES) |
442295c9 | 524 | node = first_node(node_online_map); |
8fce4d8e | 525 | |
1871e52c | 526 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
527 | } |
528 | ||
529 | static void next_reap_node(void) | |
530 | { | |
909ea964 | 531 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 532 | |
8fce4d8e CL |
533 | node = next_node(node, node_online_map); |
534 | if (unlikely(node >= MAX_NUMNODES)) | |
535 | node = first_node(node_online_map); | |
909ea964 | 536 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
537 | } |
538 | ||
539 | #else | |
540 | #define init_reap_node(cpu) do { } while (0) | |
541 | #define next_reap_node(void) do { } while (0) | |
542 | #endif | |
543 | ||
1da177e4 LT |
544 | /* |
545 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
546 | * via the workqueue/eventd. | |
547 | * Add the CPU number into the expiration time to minimize the possibility of | |
548 | * the CPUs getting into lockstep and contending for the global cache chain | |
549 | * lock. | |
550 | */ | |
0db0628d | 551 | static void start_cpu_timer(int cpu) |
1da177e4 | 552 | { |
1871e52c | 553 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
554 | |
555 | /* | |
556 | * When this gets called from do_initcalls via cpucache_init(), | |
557 | * init_workqueues() has already run, so keventd will be setup | |
558 | * at that time. | |
559 | */ | |
52bad64d | 560 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 561 | init_reap_node(cpu); |
203b42f7 | 562 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
563 | schedule_delayed_work_on(cpu, reap_work, |
564 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
565 | } |
566 | } | |
567 | ||
1fe00d50 | 568 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 569 | { |
d5cff635 CM |
570 | /* |
571 | * The array_cache structures contain pointers to free object. | |
25985edc | 572 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
573 | * cache the pointers are not cleared and they could be counted as |
574 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
575 | * not scan such objects. | |
576 | */ | |
1fe00d50 JK |
577 | kmemleak_no_scan(ac); |
578 | if (ac) { | |
579 | ac->avail = 0; | |
580 | ac->limit = limit; | |
581 | ac->batchcount = batch; | |
582 | ac->touched = 0; | |
1da177e4 | 583 | } |
1fe00d50 JK |
584 | } |
585 | ||
586 | static struct array_cache *alloc_arraycache(int node, int entries, | |
587 | int batchcount, gfp_t gfp) | |
588 | { | |
5e804789 | 589 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
590 | struct array_cache *ac = NULL; |
591 | ||
592 | ac = kmalloc_node(memsize, gfp, node); | |
593 | init_arraycache(ac, entries, batchcount); | |
594 | return ac; | |
1da177e4 LT |
595 | } |
596 | ||
f68f8ddd JK |
597 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
598 | struct page *page, void *objp) | |
072bb0aa | 599 | { |
f68f8ddd JK |
600 | struct kmem_cache_node *n; |
601 | int page_node; | |
602 | LIST_HEAD(list); | |
072bb0aa | 603 | |
f68f8ddd JK |
604 | page_node = page_to_nid(page); |
605 | n = get_node(cachep, page_node); | |
381760ea | 606 | |
f68f8ddd JK |
607 | spin_lock(&n->list_lock); |
608 | free_block(cachep, &objp, 1, page_node, &list); | |
609 | spin_unlock(&n->list_lock); | |
381760ea | 610 | |
f68f8ddd | 611 | slabs_destroy(cachep, &list); |
072bb0aa MG |
612 | } |
613 | ||
3ded175a CL |
614 | /* |
615 | * Transfer objects in one arraycache to another. | |
616 | * Locking must be handled by the caller. | |
617 | * | |
618 | * Return the number of entries transferred. | |
619 | */ | |
620 | static int transfer_objects(struct array_cache *to, | |
621 | struct array_cache *from, unsigned int max) | |
622 | { | |
623 | /* Figure out how many entries to transfer */ | |
732eacc0 | 624 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
625 | |
626 | if (!nr) | |
627 | return 0; | |
628 | ||
629 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
630 | sizeof(void *) *nr); | |
631 | ||
632 | from->avail -= nr; | |
633 | to->avail += nr; | |
3ded175a CL |
634 | return nr; |
635 | } | |
636 | ||
765c4507 CL |
637 | #ifndef CONFIG_NUMA |
638 | ||
639 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 640 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 641 | |
c8522a3a JK |
642 | static inline struct alien_cache **alloc_alien_cache(int node, |
643 | int limit, gfp_t gfp) | |
765c4507 | 644 | { |
8888177e | 645 | return NULL; |
765c4507 CL |
646 | } |
647 | ||
c8522a3a | 648 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
649 | { |
650 | } | |
651 | ||
652 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
653 | { | |
654 | return 0; | |
655 | } | |
656 | ||
657 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
658 | gfp_t flags) | |
659 | { | |
660 | return NULL; | |
661 | } | |
662 | ||
8b98c169 | 663 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
664 | gfp_t flags, int nodeid) |
665 | { | |
666 | return NULL; | |
667 | } | |
668 | ||
4167e9b2 DR |
669 | static inline gfp_t gfp_exact_node(gfp_t flags) |
670 | { | |
444eb2a4 | 671 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
672 | } |
673 | ||
765c4507 CL |
674 | #else /* CONFIG_NUMA */ |
675 | ||
8b98c169 | 676 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 677 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 678 | |
c8522a3a JK |
679 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
680 | int batch, gfp_t gfp) | |
681 | { | |
5e804789 | 682 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
683 | struct alien_cache *alc = NULL; |
684 | ||
685 | alc = kmalloc_node(memsize, gfp, node); | |
686 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 687 | spin_lock_init(&alc->lock); |
c8522a3a JK |
688 | return alc; |
689 | } | |
690 | ||
691 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 692 | { |
c8522a3a | 693 | struct alien_cache **alc_ptr; |
5e804789 | 694 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
695 | int i; |
696 | ||
697 | if (limit > 1) | |
698 | limit = 12; | |
c8522a3a JK |
699 | alc_ptr = kzalloc_node(memsize, gfp, node); |
700 | if (!alc_ptr) | |
701 | return NULL; | |
702 | ||
703 | for_each_node(i) { | |
704 | if (i == node || !node_online(i)) | |
705 | continue; | |
706 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
707 | if (!alc_ptr[i]) { | |
708 | for (i--; i >= 0; i--) | |
709 | kfree(alc_ptr[i]); | |
710 | kfree(alc_ptr); | |
711 | return NULL; | |
e498be7d CL |
712 | } |
713 | } | |
c8522a3a | 714 | return alc_ptr; |
e498be7d CL |
715 | } |
716 | ||
c8522a3a | 717 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
718 | { |
719 | int i; | |
720 | ||
c8522a3a | 721 | if (!alc_ptr) |
e498be7d | 722 | return; |
e498be7d | 723 | for_each_node(i) |
c8522a3a JK |
724 | kfree(alc_ptr[i]); |
725 | kfree(alc_ptr); | |
e498be7d CL |
726 | } |
727 | ||
343e0d7a | 728 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
729 | struct array_cache *ac, int node, |
730 | struct list_head *list) | |
e498be7d | 731 | { |
18bf8541 | 732 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
733 | |
734 | if (ac->avail) { | |
ce8eb6c4 | 735 | spin_lock(&n->list_lock); |
e00946fe CL |
736 | /* |
737 | * Stuff objects into the remote nodes shared array first. | |
738 | * That way we could avoid the overhead of putting the objects | |
739 | * into the free lists and getting them back later. | |
740 | */ | |
ce8eb6c4 CL |
741 | if (n->shared) |
742 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 743 | |
833b706c | 744 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 745 | ac->avail = 0; |
ce8eb6c4 | 746 | spin_unlock(&n->list_lock); |
e498be7d CL |
747 | } |
748 | } | |
749 | ||
8fce4d8e CL |
750 | /* |
751 | * Called from cache_reap() to regularly drain alien caches round robin. | |
752 | */ | |
ce8eb6c4 | 753 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 754 | { |
909ea964 | 755 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 756 | |
ce8eb6c4 | 757 | if (n->alien) { |
c8522a3a JK |
758 | struct alien_cache *alc = n->alien[node]; |
759 | struct array_cache *ac; | |
760 | ||
761 | if (alc) { | |
762 | ac = &alc->ac; | |
49dfc304 | 763 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
764 | LIST_HEAD(list); |
765 | ||
766 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 767 | spin_unlock_irq(&alc->lock); |
833b706c | 768 | slabs_destroy(cachep, &list); |
c8522a3a | 769 | } |
8fce4d8e CL |
770 | } |
771 | } | |
772 | } | |
773 | ||
a737b3e2 | 774 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 775 | struct alien_cache **alien) |
e498be7d | 776 | { |
b28a02de | 777 | int i = 0; |
c8522a3a | 778 | struct alien_cache *alc; |
e498be7d CL |
779 | struct array_cache *ac; |
780 | unsigned long flags; | |
781 | ||
782 | for_each_online_node(i) { | |
c8522a3a JK |
783 | alc = alien[i]; |
784 | if (alc) { | |
833b706c JK |
785 | LIST_HEAD(list); |
786 | ||
c8522a3a | 787 | ac = &alc->ac; |
49dfc304 | 788 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 789 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 790 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 791 | slabs_destroy(cachep, &list); |
e498be7d CL |
792 | } |
793 | } | |
794 | } | |
729bd0b7 | 795 | |
25c4f304 JK |
796 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
797 | int node, int page_node) | |
729bd0b7 | 798 | { |
ce8eb6c4 | 799 | struct kmem_cache_node *n; |
c8522a3a JK |
800 | struct alien_cache *alien = NULL; |
801 | struct array_cache *ac; | |
97654dfa | 802 | LIST_HEAD(list); |
1ca4cb24 | 803 | |
18bf8541 | 804 | n = get_node(cachep, node); |
729bd0b7 | 805 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
806 | if (n->alien && n->alien[page_node]) { |
807 | alien = n->alien[page_node]; | |
c8522a3a | 808 | ac = &alien->ac; |
49dfc304 | 809 | spin_lock(&alien->lock); |
c8522a3a | 810 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 811 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 812 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 813 | } |
f68f8ddd | 814 | ac->entry[ac->avail++] = objp; |
49dfc304 | 815 | spin_unlock(&alien->lock); |
833b706c | 816 | slabs_destroy(cachep, &list); |
729bd0b7 | 817 | } else { |
25c4f304 | 818 | n = get_node(cachep, page_node); |
18bf8541 | 819 | spin_lock(&n->list_lock); |
25c4f304 | 820 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 821 | spin_unlock(&n->list_lock); |
97654dfa | 822 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
823 | } |
824 | return 1; | |
825 | } | |
25c4f304 JK |
826 | |
827 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
828 | { | |
829 | int page_node = page_to_nid(virt_to_page(objp)); | |
830 | int node = numa_mem_id(); | |
831 | /* | |
832 | * Make sure we are not freeing a object from another node to the array | |
833 | * cache on this cpu. | |
834 | */ | |
835 | if (likely(node == page_node)) | |
836 | return 0; | |
837 | ||
838 | return __cache_free_alien(cachep, objp, node, page_node); | |
839 | } | |
4167e9b2 DR |
840 | |
841 | /* | |
444eb2a4 MG |
842 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
843 | * warn about failures. | |
4167e9b2 DR |
844 | */ |
845 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
846 | { | |
444eb2a4 | 847 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 848 | } |
e498be7d CL |
849 | #endif |
850 | ||
8f9f8d9e | 851 | /* |
6a67368c | 852 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 853 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 854 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 855 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
856 | * already in use. |
857 | * | |
18004c5d | 858 | * Must hold slab_mutex. |
8f9f8d9e | 859 | */ |
6a67368c | 860 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
861 | { |
862 | struct kmem_cache *cachep; | |
ce8eb6c4 | 863 | struct kmem_cache_node *n; |
5e804789 | 864 | const size_t memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 865 | |
18004c5d | 866 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 867 | /* |
5f0985bb | 868 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
869 | * begin anything. Make sure some other cpu on this |
870 | * node has not already allocated this | |
871 | */ | |
18bf8541 CL |
872 | n = get_node(cachep, node); |
873 | if (!n) { | |
ce8eb6c4 CL |
874 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
875 | if (!n) | |
8f9f8d9e | 876 | return -ENOMEM; |
ce8eb6c4 | 877 | kmem_cache_node_init(n); |
5f0985bb JZ |
878 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
879 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
880 | |
881 | /* | |
5f0985bb JZ |
882 | * The kmem_cache_nodes don't come and go as CPUs |
883 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
884 | * protection here. |
885 | */ | |
ce8eb6c4 | 886 | cachep->node[node] = n; |
8f9f8d9e DR |
887 | } |
888 | ||
18bf8541 CL |
889 | spin_lock_irq(&n->list_lock); |
890 | n->free_limit = | |
8f9f8d9e DR |
891 | (1 + nr_cpus_node(node)) * |
892 | cachep->batchcount + cachep->num; | |
18bf8541 | 893 | spin_unlock_irq(&n->list_lock); |
8f9f8d9e DR |
894 | } |
895 | return 0; | |
896 | } | |
897 | ||
0fa8103b WL |
898 | static inline int slabs_tofree(struct kmem_cache *cachep, |
899 | struct kmem_cache_node *n) | |
900 | { | |
901 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
902 | } | |
903 | ||
0db0628d | 904 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
905 | { |
906 | struct kmem_cache *cachep; | |
ce8eb6c4 | 907 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 908 | int node = cpu_to_mem(cpu); |
a70f7302 | 909 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 910 | |
18004c5d | 911 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
912 | struct array_cache *nc; |
913 | struct array_cache *shared; | |
c8522a3a | 914 | struct alien_cache **alien; |
97654dfa | 915 | LIST_HEAD(list); |
fbf1e473 | 916 | |
18bf8541 | 917 | n = get_node(cachep, node); |
ce8eb6c4 | 918 | if (!n) |
bf0dea23 | 919 | continue; |
fbf1e473 | 920 | |
ce8eb6c4 | 921 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 922 | |
ce8eb6c4 CL |
923 | /* Free limit for this kmem_cache_node */ |
924 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
925 | |
926 | /* cpu is dead; no one can alloc from it. */ | |
927 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
928 | if (nc) { | |
97654dfa | 929 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
930 | nc->avail = 0; |
931 | } | |
fbf1e473 | 932 | |
58463c1f | 933 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 934 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 935 | goto free_slab; |
fbf1e473 AM |
936 | } |
937 | ||
ce8eb6c4 | 938 | shared = n->shared; |
fbf1e473 AM |
939 | if (shared) { |
940 | free_block(cachep, shared->entry, | |
97654dfa | 941 | shared->avail, node, &list); |
ce8eb6c4 | 942 | n->shared = NULL; |
fbf1e473 AM |
943 | } |
944 | ||
ce8eb6c4 CL |
945 | alien = n->alien; |
946 | n->alien = NULL; | |
fbf1e473 | 947 | |
ce8eb6c4 | 948 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
949 | |
950 | kfree(shared); | |
951 | if (alien) { | |
952 | drain_alien_cache(cachep, alien); | |
953 | free_alien_cache(alien); | |
954 | } | |
bf0dea23 JK |
955 | |
956 | free_slab: | |
97654dfa | 957 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
958 | } |
959 | /* | |
960 | * In the previous loop, all the objects were freed to | |
961 | * the respective cache's slabs, now we can go ahead and | |
962 | * shrink each nodelist to its limit. | |
963 | */ | |
18004c5d | 964 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 965 | n = get_node(cachep, node); |
ce8eb6c4 | 966 | if (!n) |
fbf1e473 | 967 | continue; |
0fa8103b | 968 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
969 | } |
970 | } | |
971 | ||
0db0628d | 972 | static int cpuup_prepare(long cpu) |
1da177e4 | 973 | { |
343e0d7a | 974 | struct kmem_cache *cachep; |
ce8eb6c4 | 975 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 976 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 977 | int err; |
1da177e4 | 978 | |
fbf1e473 AM |
979 | /* |
980 | * We need to do this right in the beginning since | |
981 | * alloc_arraycache's are going to use this list. | |
982 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 983 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 984 | */ |
6a67368c | 985 | err = init_cache_node_node(node); |
8f9f8d9e DR |
986 | if (err < 0) |
987 | goto bad; | |
fbf1e473 AM |
988 | |
989 | /* | |
990 | * Now we can go ahead with allocating the shared arrays and | |
991 | * array caches | |
992 | */ | |
18004c5d | 993 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 | 994 | struct array_cache *shared = NULL; |
c8522a3a | 995 | struct alien_cache **alien = NULL; |
fbf1e473 | 996 | |
fbf1e473 AM |
997 | if (cachep->shared) { |
998 | shared = alloc_arraycache(node, | |
999 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1000 | 0xbaadf00d, GFP_KERNEL); |
bf0dea23 | 1001 | if (!shared) |
1da177e4 | 1002 | goto bad; |
fbf1e473 AM |
1003 | } |
1004 | if (use_alien_caches) { | |
83b519e8 | 1005 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1006 | if (!alien) { |
1007 | kfree(shared); | |
fbf1e473 | 1008 | goto bad; |
12d00f6a | 1009 | } |
fbf1e473 | 1010 | } |
18bf8541 | 1011 | n = get_node(cachep, node); |
ce8eb6c4 | 1012 | BUG_ON(!n); |
fbf1e473 | 1013 | |
ce8eb6c4 CL |
1014 | spin_lock_irq(&n->list_lock); |
1015 | if (!n->shared) { | |
fbf1e473 AM |
1016 | /* |
1017 | * We are serialised from CPU_DEAD or | |
1018 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1019 | */ | |
ce8eb6c4 | 1020 | n->shared = shared; |
fbf1e473 AM |
1021 | shared = NULL; |
1022 | } | |
4484ebf1 | 1023 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1024 | if (!n->alien) { |
1025 | n->alien = alien; | |
fbf1e473 | 1026 | alien = NULL; |
1da177e4 | 1027 | } |
fbf1e473 | 1028 | #endif |
ce8eb6c4 | 1029 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1030 | kfree(shared); |
1031 | free_alien_cache(alien); | |
1032 | } | |
ce79ddc8 | 1033 | |
fbf1e473 AM |
1034 | return 0; |
1035 | bad: | |
12d00f6a | 1036 | cpuup_canceled(cpu); |
fbf1e473 AM |
1037 | return -ENOMEM; |
1038 | } | |
1039 | ||
0db0628d | 1040 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1041 | unsigned long action, void *hcpu) |
1042 | { | |
1043 | long cpu = (long)hcpu; | |
1044 | int err = 0; | |
1045 | ||
1046 | switch (action) { | |
fbf1e473 AM |
1047 | case CPU_UP_PREPARE: |
1048 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1049 | mutex_lock(&slab_mutex); |
fbf1e473 | 1050 | err = cpuup_prepare(cpu); |
18004c5d | 1051 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1052 | break; |
1053 | case CPU_ONLINE: | |
8bb78442 | 1054 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1055 | start_cpu_timer(cpu); |
1056 | break; | |
1057 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1058 | case CPU_DOWN_PREPARE: |
8bb78442 | 1059 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1060 | /* |
18004c5d | 1061 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1062 | * held so that if cache_reap() is invoked it cannot do |
1063 | * anything expensive but will only modify reap_work | |
1064 | * and reschedule the timer. | |
1065 | */ | |
afe2c511 | 1066 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1067 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1068 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1069 | break; |
1070 | case CPU_DOWN_FAILED: | |
8bb78442 | 1071 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1072 | start_cpu_timer(cpu); |
1073 | break; | |
1da177e4 | 1074 | case CPU_DEAD: |
8bb78442 | 1075 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1076 | /* |
1077 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1078 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1079 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1080 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1081 | * structure is usually allocated from kmem_cache_create() and |
1082 | * gets destroyed at kmem_cache_destroy(). | |
1083 | */ | |
183ff22b | 1084 | /* fall through */ |
8f5be20b | 1085 | #endif |
1da177e4 | 1086 | case CPU_UP_CANCELED: |
8bb78442 | 1087 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1088 | mutex_lock(&slab_mutex); |
fbf1e473 | 1089 | cpuup_canceled(cpu); |
18004c5d | 1090 | mutex_unlock(&slab_mutex); |
1da177e4 | 1091 | break; |
1da177e4 | 1092 | } |
eac40680 | 1093 | return notifier_from_errno(err); |
1da177e4 LT |
1094 | } |
1095 | ||
0db0628d | 1096 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1097 | &cpuup_callback, NULL, 0 |
1098 | }; | |
1da177e4 | 1099 | |
8f9f8d9e DR |
1100 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1101 | /* | |
1102 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1103 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1104 | * removed. | |
1105 | * | |
18004c5d | 1106 | * Must hold slab_mutex. |
8f9f8d9e | 1107 | */ |
6a67368c | 1108 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1109 | { |
1110 | struct kmem_cache *cachep; | |
1111 | int ret = 0; | |
1112 | ||
18004c5d | 1113 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1114 | struct kmem_cache_node *n; |
8f9f8d9e | 1115 | |
18bf8541 | 1116 | n = get_node(cachep, node); |
ce8eb6c4 | 1117 | if (!n) |
8f9f8d9e DR |
1118 | continue; |
1119 | ||
0fa8103b | 1120 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1121 | |
ce8eb6c4 CL |
1122 | if (!list_empty(&n->slabs_full) || |
1123 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1124 | ret = -EBUSY; |
1125 | break; | |
1126 | } | |
1127 | } | |
1128 | return ret; | |
1129 | } | |
1130 | ||
1131 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1132 | unsigned long action, void *arg) | |
1133 | { | |
1134 | struct memory_notify *mnb = arg; | |
1135 | int ret = 0; | |
1136 | int nid; | |
1137 | ||
1138 | nid = mnb->status_change_nid; | |
1139 | if (nid < 0) | |
1140 | goto out; | |
1141 | ||
1142 | switch (action) { | |
1143 | case MEM_GOING_ONLINE: | |
18004c5d | 1144 | mutex_lock(&slab_mutex); |
6a67368c | 1145 | ret = init_cache_node_node(nid); |
18004c5d | 1146 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1147 | break; |
1148 | case MEM_GOING_OFFLINE: | |
18004c5d | 1149 | mutex_lock(&slab_mutex); |
6a67368c | 1150 | ret = drain_cache_node_node(nid); |
18004c5d | 1151 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1152 | break; |
1153 | case MEM_ONLINE: | |
1154 | case MEM_OFFLINE: | |
1155 | case MEM_CANCEL_ONLINE: | |
1156 | case MEM_CANCEL_OFFLINE: | |
1157 | break; | |
1158 | } | |
1159 | out: | |
5fda1bd5 | 1160 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1161 | } |
1162 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1163 | ||
e498be7d | 1164 | /* |
ce8eb6c4 | 1165 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1166 | */ |
6744f087 | 1167 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1168 | int nodeid) |
e498be7d | 1169 | { |
6744f087 | 1170 | struct kmem_cache_node *ptr; |
e498be7d | 1171 | |
6744f087 | 1172 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1173 | BUG_ON(!ptr); |
1174 | ||
6744f087 | 1175 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1176 | /* |
1177 | * Do not assume that spinlocks can be initialized via memcpy: | |
1178 | */ | |
1179 | spin_lock_init(&ptr->list_lock); | |
1180 | ||
e498be7d | 1181 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1182 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1183 | } |
1184 | ||
556a169d | 1185 | /* |
ce8eb6c4 CL |
1186 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1187 | * size of kmem_cache_node. | |
556a169d | 1188 | */ |
ce8eb6c4 | 1189 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1190 | { |
1191 | int node; | |
1192 | ||
1193 | for_each_online_node(node) { | |
ce8eb6c4 | 1194 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1195 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1196 | REAPTIMEOUT_NODE + |
1197 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1198 | } |
1199 | } | |
1200 | ||
a737b3e2 AM |
1201 | /* |
1202 | * Initialisation. Called after the page allocator have been initialised and | |
1203 | * before smp_init(). | |
1da177e4 LT |
1204 | */ |
1205 | void __init kmem_cache_init(void) | |
1206 | { | |
e498be7d CL |
1207 | int i; |
1208 | ||
68126702 JK |
1209 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1210 | sizeof(struct rcu_head)); | |
9b030cb8 CL |
1211 | kmem_cache = &kmem_cache_boot; |
1212 | ||
8888177e | 1213 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1214 | use_alien_caches = 0; |
1215 | ||
3c583465 | 1216 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1217 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1218 | |
1da177e4 LT |
1219 | /* |
1220 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1221 | * page orders on machines with more than 32MB of memory if |
1222 | * not overridden on the command line. | |
1da177e4 | 1223 | */ |
3df1cccd | 1224 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1225 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1226 | |
1da177e4 LT |
1227 | /* Bootstrap is tricky, because several objects are allocated |
1228 | * from caches that do not exist yet: | |
9b030cb8 CL |
1229 | * 1) initialize the kmem_cache cache: it contains the struct |
1230 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1231 | * kmem_cache is statically allocated. | |
e498be7d | 1232 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1233 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1234 | * array at the end of the bootstrap. |
1da177e4 | 1235 | * 2) Create the first kmalloc cache. |
343e0d7a | 1236 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1237 | * An __init data area is used for the head array. |
1238 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1239 | * head arrays. | |
9b030cb8 | 1240 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1241 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1242 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1243 | * the other cache's with kmalloc allocated memory. |
1244 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1245 | */ |
1246 | ||
9b030cb8 | 1247 | /* 1) create the kmem_cache */ |
1da177e4 | 1248 | |
8da3430d | 1249 | /* |
b56efcf0 | 1250 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1251 | */ |
2f9baa9f | 1252 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1253 | offsetof(struct kmem_cache, node) + |
6744f087 | 1254 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1255 | SLAB_HWCACHE_ALIGN); |
1256 | list_add(&kmem_cache->list, &slab_caches); | |
bf0dea23 | 1257 | slab_state = PARTIAL; |
1da177e4 | 1258 | |
a737b3e2 | 1259 | /* |
bf0dea23 JK |
1260 | * Initialize the caches that provide memory for the kmem_cache_node |
1261 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1262 | */ |
bf0dea23 | 1263 | kmalloc_caches[INDEX_NODE] = create_kmalloc_cache("kmalloc-node", |
ce8eb6c4 | 1264 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); |
bf0dea23 | 1265 | slab_state = PARTIAL_NODE; |
34cc6990 | 1266 | setup_kmalloc_cache_index_table(); |
e498be7d | 1267 | |
e0a42726 IM |
1268 | slab_early_init = 0; |
1269 | ||
ce8eb6c4 | 1270 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1271 | { |
1ca4cb24 PE |
1272 | int nid; |
1273 | ||
9c09a95c | 1274 | for_each_online_node(nid) { |
ce8eb6c4 | 1275 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1276 | |
bf0dea23 | 1277 | init_list(kmalloc_caches[INDEX_NODE], |
ce8eb6c4 | 1278 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1279 | } |
1280 | } | |
1da177e4 | 1281 | |
f97d5f63 | 1282 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1283 | } |
1284 | ||
1285 | void __init kmem_cache_init_late(void) | |
1286 | { | |
1287 | struct kmem_cache *cachep; | |
1288 | ||
97d06609 | 1289 | slab_state = UP; |
52cef189 | 1290 | |
8429db5c | 1291 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1292 | mutex_lock(&slab_mutex); |
1293 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1294 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1295 | BUG(); | |
18004c5d | 1296 | mutex_unlock(&slab_mutex); |
056c6241 | 1297 | |
97d06609 CL |
1298 | /* Done! */ |
1299 | slab_state = FULL; | |
1300 | ||
a737b3e2 AM |
1301 | /* |
1302 | * Register a cpu startup notifier callback that initializes | |
1303 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1304 | */ |
1305 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1306 | |
8f9f8d9e DR |
1307 | #ifdef CONFIG_NUMA |
1308 | /* | |
1309 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1310 | * node. |
8f9f8d9e DR |
1311 | */ |
1312 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1313 | #endif | |
1314 | ||
a737b3e2 AM |
1315 | /* |
1316 | * The reap timers are started later, with a module init call: That part | |
1317 | * of the kernel is not yet operational. | |
1da177e4 LT |
1318 | */ |
1319 | } | |
1320 | ||
1321 | static int __init cpucache_init(void) | |
1322 | { | |
1323 | int cpu; | |
1324 | ||
a737b3e2 AM |
1325 | /* |
1326 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1327 | */ |
e498be7d | 1328 | for_each_online_cpu(cpu) |
a737b3e2 | 1329 | start_cpu_timer(cpu); |
a164f896 GC |
1330 | |
1331 | /* Done! */ | |
97d06609 | 1332 | slab_state = FULL; |
1da177e4 LT |
1333 | return 0; |
1334 | } | |
1da177e4 LT |
1335 | __initcall(cpucache_init); |
1336 | ||
8bdec192 RA |
1337 | static noinline void |
1338 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1339 | { | |
9a02d699 | 1340 | #if DEBUG |
ce8eb6c4 | 1341 | struct kmem_cache_node *n; |
8456a648 | 1342 | struct page *page; |
8bdec192 RA |
1343 | unsigned long flags; |
1344 | int node; | |
9a02d699 DR |
1345 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1346 | DEFAULT_RATELIMIT_BURST); | |
1347 | ||
1348 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1349 | return; | |
8bdec192 | 1350 | |
5b3810e5 VB |
1351 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1352 | nodeid, gfpflags, &gfpflags); | |
1353 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1354 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1355 | |
18bf8541 | 1356 | for_each_kmem_cache_node(cachep, node, n) { |
8bdec192 RA |
1357 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; |
1358 | unsigned long active_slabs = 0, num_slabs = 0; | |
1359 | ||
ce8eb6c4 | 1360 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1361 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1362 | active_objs += cachep->num; |
1363 | active_slabs++; | |
1364 | } | |
8456a648 JK |
1365 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1366 | active_objs += page->active; | |
8bdec192 RA |
1367 | active_slabs++; |
1368 | } | |
8456a648 | 1369 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1370 | num_slabs++; |
1371 | ||
ce8eb6c4 CL |
1372 | free_objects += n->free_objects; |
1373 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1374 | |
1375 | num_slabs += active_slabs; | |
1376 | num_objs = num_slabs * cachep->num; | |
5b3810e5 | 1377 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", |
8bdec192 RA |
1378 | node, active_slabs, num_slabs, active_objs, num_objs, |
1379 | free_objects); | |
1380 | } | |
9a02d699 | 1381 | #endif |
8bdec192 RA |
1382 | } |
1383 | ||
1da177e4 | 1384 | /* |
8a7d9b43 WSH |
1385 | * Interface to system's page allocator. No need to hold the |
1386 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1387 | * |
1388 | * If we requested dmaable memory, we will get it. Even if we | |
1389 | * did not request dmaable memory, we might get it, but that | |
1390 | * would be relatively rare and ignorable. | |
1391 | */ | |
0c3aa83e JK |
1392 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1393 | int nodeid) | |
1da177e4 LT |
1394 | { |
1395 | struct page *page; | |
e1b6aa6f | 1396 | int nr_pages; |
765c4507 | 1397 | |
a618e89f | 1398 | flags |= cachep->allocflags; |
e12ba74d MG |
1399 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1400 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1401 | |
96db800f | 1402 | page = __alloc_pages_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1403 | if (!page) { |
9a02d699 | 1404 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1405 | return NULL; |
8bdec192 | 1406 | } |
1da177e4 | 1407 | |
f3ccb2c4 VD |
1408 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1409 | __free_pages(page, cachep->gfporder); | |
1410 | return NULL; | |
1411 | } | |
1412 | ||
e1b6aa6f | 1413 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1414 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1415 | add_zone_page_state(page_zone(page), |
1416 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1417 | else | |
1418 | add_zone_page_state(page_zone(page), | |
1419 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
f68f8ddd | 1420 | |
a57a4988 | 1421 | __SetPageSlab(page); |
f68f8ddd JK |
1422 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1423 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1424 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1425 | |
b1eeab67 VN |
1426 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1427 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1428 | ||
1429 | if (cachep->ctor) | |
1430 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1431 | else | |
1432 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1433 | } | |
c175eea4 | 1434 | |
0c3aa83e | 1435 | return page; |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | /* | |
1439 | * Interface to system's page release. | |
1440 | */ | |
0c3aa83e | 1441 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1442 | { |
27ee57c9 VD |
1443 | int order = cachep->gfporder; |
1444 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1445 | |
27ee57c9 | 1446 | kmemcheck_free_shadow(page, order); |
c175eea4 | 1447 | |
972d1a7b CL |
1448 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1449 | sub_zone_page_state(page_zone(page), | |
1450 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1451 | else | |
1452 | sub_zone_page_state(page_zone(page), | |
1453 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1454 | |
a57a4988 | 1455 | BUG_ON(!PageSlab(page)); |
73293c2f | 1456 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1457 | __ClearPageSlab(page); |
8456a648 JK |
1458 | page_mapcount_reset(page); |
1459 | page->mapping = NULL; | |
1f458cbf | 1460 | |
1da177e4 LT |
1461 | if (current->reclaim_state) |
1462 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1463 | memcg_uncharge_slab(page, order, cachep); |
1464 | __free_pages(page, order); | |
1da177e4 LT |
1465 | } |
1466 | ||
1467 | static void kmem_rcu_free(struct rcu_head *head) | |
1468 | { | |
68126702 JK |
1469 | struct kmem_cache *cachep; |
1470 | struct page *page; | |
1da177e4 | 1471 | |
68126702 JK |
1472 | page = container_of(head, struct page, rcu_head); |
1473 | cachep = page->slab_cache; | |
1474 | ||
1475 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1476 | } |
1477 | ||
1478 | #if DEBUG | |
40b44137 JK |
1479 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1480 | { | |
1481 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1482 | (cachep->size % PAGE_SIZE) == 0) | |
1483 | return true; | |
1484 | ||
1485 | return false; | |
1486 | } | |
1da177e4 LT |
1487 | |
1488 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1489 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1490 | unsigned long caller) |
1da177e4 | 1491 | { |
8c138bc0 | 1492 | int size = cachep->object_size; |
1da177e4 | 1493 | |
3dafccf2 | 1494 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1495 | |
b28a02de | 1496 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1497 | return; |
1498 | ||
b28a02de PE |
1499 | *addr++ = 0x12345678; |
1500 | *addr++ = caller; | |
1501 | *addr++ = smp_processor_id(); | |
1502 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1503 | { |
1504 | unsigned long *sptr = &caller; | |
1505 | unsigned long svalue; | |
1506 | ||
1507 | while (!kstack_end(sptr)) { | |
1508 | svalue = *sptr++; | |
1509 | if (kernel_text_address(svalue)) { | |
b28a02de | 1510 | *addr++ = svalue; |
1da177e4 LT |
1511 | size -= sizeof(unsigned long); |
1512 | if (size <= sizeof(unsigned long)) | |
1513 | break; | |
1514 | } | |
1515 | } | |
1516 | ||
1517 | } | |
b28a02de | 1518 | *addr++ = 0x87654321; |
1da177e4 | 1519 | } |
40b44137 JK |
1520 | |
1521 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1522 | int map, unsigned long caller) | |
1523 | { | |
1524 | if (!is_debug_pagealloc_cache(cachep)) | |
1525 | return; | |
1526 | ||
1527 | if (caller) | |
1528 | store_stackinfo(cachep, objp, caller); | |
1529 | ||
1530 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1531 | } | |
1532 | ||
1533 | #else | |
1534 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1535 | int map, unsigned long caller) {} | |
1536 | ||
1da177e4 LT |
1537 | #endif |
1538 | ||
343e0d7a | 1539 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1540 | { |
8c138bc0 | 1541 | int size = cachep->object_size; |
3dafccf2 | 1542 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1543 | |
1544 | memset(addr, val, size); | |
b28a02de | 1545 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1546 | } |
1547 | ||
1548 | static void dump_line(char *data, int offset, int limit) | |
1549 | { | |
1550 | int i; | |
aa83aa40 DJ |
1551 | unsigned char error = 0; |
1552 | int bad_count = 0; | |
1553 | ||
1170532b | 1554 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1555 | for (i = 0; i < limit; i++) { |
1556 | if (data[offset + i] != POISON_FREE) { | |
1557 | error = data[offset + i]; | |
1558 | bad_count++; | |
1559 | } | |
aa83aa40 | 1560 | } |
fdde6abb SAS |
1561 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1562 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1563 | |
1564 | if (bad_count == 1) { | |
1565 | error ^= POISON_FREE; | |
1566 | if (!(error & (error - 1))) { | |
1170532b | 1567 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1568 | #ifdef CONFIG_X86 |
1170532b | 1569 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1570 | #else |
1170532b | 1571 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1572 | #endif |
1573 | } | |
1574 | } | |
1da177e4 LT |
1575 | } |
1576 | #endif | |
1577 | ||
1578 | #if DEBUG | |
1579 | ||
343e0d7a | 1580 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1581 | { |
1582 | int i, size; | |
1583 | char *realobj; | |
1584 | ||
1585 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1586 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1587 | *dbg_redzone1(cachep, objp), | |
1588 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1589 | } |
1590 | ||
1591 | if (cachep->flags & SLAB_STORE_USER) { | |
1170532b | 1592 | pr_err("Last user: [<%p>](%pSR)\n", |
071361d3 JP |
1593 | *dbg_userword(cachep, objp), |
1594 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1595 | } |
3dafccf2 | 1596 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1597 | size = cachep->object_size; |
b28a02de | 1598 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1599 | int limit; |
1600 | limit = 16; | |
b28a02de PE |
1601 | if (i + limit > size) |
1602 | limit = size - i; | |
1da177e4 LT |
1603 | dump_line(realobj, i, limit); |
1604 | } | |
1605 | } | |
1606 | ||
343e0d7a | 1607 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1608 | { |
1609 | char *realobj; | |
1610 | int size, i; | |
1611 | int lines = 0; | |
1612 | ||
40b44137 JK |
1613 | if (is_debug_pagealloc_cache(cachep)) |
1614 | return; | |
1615 | ||
3dafccf2 | 1616 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1617 | size = cachep->object_size; |
1da177e4 | 1618 | |
b28a02de | 1619 | for (i = 0; i < size; i++) { |
1da177e4 | 1620 | char exp = POISON_FREE; |
b28a02de | 1621 | if (i == size - 1) |
1da177e4 LT |
1622 | exp = POISON_END; |
1623 | if (realobj[i] != exp) { | |
1624 | int limit; | |
1625 | /* Mismatch ! */ | |
1626 | /* Print header */ | |
1627 | if (lines == 0) { | |
1170532b JP |
1628 | pr_err("Slab corruption (%s): %s start=%p, len=%d\n", |
1629 | print_tainted(), cachep->name, | |
1630 | realobj, size); | |
1da177e4 LT |
1631 | print_objinfo(cachep, objp, 0); |
1632 | } | |
1633 | /* Hexdump the affected line */ | |
b28a02de | 1634 | i = (i / 16) * 16; |
1da177e4 | 1635 | limit = 16; |
b28a02de PE |
1636 | if (i + limit > size) |
1637 | limit = size - i; | |
1da177e4 LT |
1638 | dump_line(realobj, i, limit); |
1639 | i += 16; | |
1640 | lines++; | |
1641 | /* Limit to 5 lines */ | |
1642 | if (lines > 5) | |
1643 | break; | |
1644 | } | |
1645 | } | |
1646 | if (lines != 0) { | |
1647 | /* Print some data about the neighboring objects, if they | |
1648 | * exist: | |
1649 | */ | |
8456a648 | 1650 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1651 | unsigned int objnr; |
1da177e4 | 1652 | |
8456a648 | 1653 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1654 | if (objnr) { |
8456a648 | 1655 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1656 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1657 | pr_err("Prev obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1658 | print_objinfo(cachep, objp, 2); |
1659 | } | |
b28a02de | 1660 | if (objnr + 1 < cachep->num) { |
8456a648 | 1661 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1662 | realobj = (char *)objp + obj_offset(cachep); |
1170532b | 1663 | pr_err("Next obj: start=%p, len=%d\n", realobj, size); |
1da177e4 LT |
1664 | print_objinfo(cachep, objp, 2); |
1665 | } | |
1666 | } | |
1667 | } | |
1668 | #endif | |
1669 | ||
12dd36fa | 1670 | #if DEBUG |
8456a648 JK |
1671 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1672 | struct page *page) | |
1da177e4 | 1673 | { |
1da177e4 | 1674 | int i; |
b03a017b JK |
1675 | |
1676 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1677 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1678 | POISON_FREE); | |
1679 | } | |
1680 | ||
1da177e4 | 1681 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1682 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1683 | |
1684 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1685 | check_poison_obj(cachep, objp); |
40b44137 | 1686 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1687 | } |
1688 | if (cachep->flags & SLAB_RED_ZONE) { | |
1689 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1690 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1691 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1692 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1693 | } |
1da177e4 | 1694 | } |
12dd36fa | 1695 | } |
1da177e4 | 1696 | #else |
8456a648 JK |
1697 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1698 | struct page *page) | |
12dd36fa | 1699 | { |
12dd36fa | 1700 | } |
1da177e4 LT |
1701 | #endif |
1702 | ||
911851e6 RD |
1703 | /** |
1704 | * slab_destroy - destroy and release all objects in a slab | |
1705 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1706 | * @page: page pointer being destroyed |
911851e6 | 1707 | * |
8a7d9b43 WSH |
1708 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1709 | * Before calling the slab page must have been unlinked from the cache. The | |
1710 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1711 | */ |
8456a648 | 1712 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1713 | { |
7e007355 | 1714 | void *freelist; |
12dd36fa | 1715 | |
8456a648 JK |
1716 | freelist = page->freelist; |
1717 | slab_destroy_debugcheck(cachep, page); | |
bc4f610d KS |
1718 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) |
1719 | call_rcu(&page->rcu_head, kmem_rcu_free); | |
1720 | else | |
0c3aa83e | 1721 | kmem_freepages(cachep, page); |
68126702 JK |
1722 | |
1723 | /* | |
8456a648 | 1724 | * From now on, we don't use freelist |
68126702 JK |
1725 | * although actual page can be freed in rcu context |
1726 | */ | |
1727 | if (OFF_SLAB(cachep)) | |
8456a648 | 1728 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1729 | } |
1730 | ||
97654dfa JK |
1731 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1732 | { | |
1733 | struct page *page, *n; | |
1734 | ||
1735 | list_for_each_entry_safe(page, n, list, lru) { | |
1736 | list_del(&page->lru); | |
1737 | slab_destroy(cachep, page); | |
1738 | } | |
1739 | } | |
1740 | ||
4d268eba | 1741 | /** |
a70773dd RD |
1742 | * calculate_slab_order - calculate size (page order) of slabs |
1743 | * @cachep: pointer to the cache that is being created | |
1744 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1745 | * @flags: slab allocation flags |
1746 | * | |
1747 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1748 | * |
1749 | * This could be made much more intelligent. For now, try to avoid using | |
1750 | * high order pages for slabs. When the gfp() functions are more friendly | |
1751 | * towards high-order requests, this should be changed. | |
1752 | */ | |
a737b3e2 | 1753 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
2e6b3602 | 1754 | size_t size, unsigned long flags) |
4d268eba PE |
1755 | { |
1756 | size_t left_over = 0; | |
9888e6fa | 1757 | int gfporder; |
4d268eba | 1758 | |
0aa817f0 | 1759 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1760 | unsigned int num; |
1761 | size_t remainder; | |
1762 | ||
70f75067 | 1763 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1764 | if (!num) |
1765 | continue; | |
9888e6fa | 1766 | |
f315e3fa JK |
1767 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1768 | if (num > SLAB_OBJ_MAX_NUM) | |
1769 | break; | |
1770 | ||
b1ab41c4 | 1771 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1772 | struct kmem_cache *freelist_cache; |
1773 | size_t freelist_size; | |
1774 | ||
1775 | freelist_size = num * sizeof(freelist_idx_t); | |
1776 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1777 | if (!freelist_cache) | |
1778 | continue; | |
1779 | ||
b1ab41c4 | 1780 | /* |
3217fd9b JK |
1781 | * Needed to avoid possible looping condition |
1782 | * in cache_grow() | |
b1ab41c4 | 1783 | */ |
3217fd9b JK |
1784 | if (OFF_SLAB(freelist_cache)) |
1785 | continue; | |
b1ab41c4 | 1786 | |
3217fd9b JK |
1787 | /* check if off slab has enough benefit */ |
1788 | if (freelist_cache->size > cachep->size / 2) | |
1789 | continue; | |
b1ab41c4 | 1790 | } |
4d268eba | 1791 | |
9888e6fa | 1792 | /* Found something acceptable - save it away */ |
4d268eba | 1793 | cachep->num = num; |
9888e6fa | 1794 | cachep->gfporder = gfporder; |
4d268eba PE |
1795 | left_over = remainder; |
1796 | ||
f78bb8ad LT |
1797 | /* |
1798 | * A VFS-reclaimable slab tends to have most allocations | |
1799 | * as GFP_NOFS and we really don't want to have to be allocating | |
1800 | * higher-order pages when we are unable to shrink dcache. | |
1801 | */ | |
1802 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1803 | break; | |
1804 | ||
4d268eba PE |
1805 | /* |
1806 | * Large number of objects is good, but very large slabs are | |
1807 | * currently bad for the gfp()s. | |
1808 | */ | |
543585cc | 1809 | if (gfporder >= slab_max_order) |
4d268eba PE |
1810 | break; |
1811 | ||
9888e6fa LT |
1812 | /* |
1813 | * Acceptable internal fragmentation? | |
1814 | */ | |
a737b3e2 | 1815 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1816 | break; |
1817 | } | |
1818 | return left_over; | |
1819 | } | |
1820 | ||
bf0dea23 JK |
1821 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1822 | struct kmem_cache *cachep, int entries, int batchcount) | |
1823 | { | |
1824 | int cpu; | |
1825 | size_t size; | |
1826 | struct array_cache __percpu *cpu_cache; | |
1827 | ||
1828 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1829 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1830 | |
1831 | if (!cpu_cache) | |
1832 | return NULL; | |
1833 | ||
1834 | for_each_possible_cpu(cpu) { | |
1835 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1836 | entries, batchcount); | |
1837 | } | |
1838 | ||
1839 | return cpu_cache; | |
1840 | } | |
1841 | ||
83b519e8 | 1842 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1843 | { |
97d06609 | 1844 | if (slab_state >= FULL) |
83b519e8 | 1845 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1846 | |
bf0dea23 JK |
1847 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1848 | if (!cachep->cpu_cache) | |
1849 | return 1; | |
1850 | ||
97d06609 | 1851 | if (slab_state == DOWN) { |
bf0dea23 JK |
1852 | /* Creation of first cache (kmem_cache). */ |
1853 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1854 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1855 | /* For kmem_cache_node */ |
1856 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1857 | } else { |
bf0dea23 | 1858 | int node; |
f30cf7d1 | 1859 | |
bf0dea23 JK |
1860 | for_each_online_node(node) { |
1861 | cachep->node[node] = kmalloc_node( | |
1862 | sizeof(struct kmem_cache_node), gfp, node); | |
1863 | BUG_ON(!cachep->node[node]); | |
1864 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1865 | } |
1866 | } | |
bf0dea23 | 1867 | |
6a67368c | 1868 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1869 | jiffies + REAPTIMEOUT_NODE + |
1870 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1871 | |
1872 | cpu_cache_get(cachep)->avail = 0; | |
1873 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1874 | cpu_cache_get(cachep)->batchcount = 1; | |
1875 | cpu_cache_get(cachep)->touched = 0; | |
1876 | cachep->batchcount = 1; | |
1877 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1878 | return 0; |
f30cf7d1 PE |
1879 | } |
1880 | ||
12220dea JK |
1881 | unsigned long kmem_cache_flags(unsigned long object_size, |
1882 | unsigned long flags, const char *name, | |
1883 | void (*ctor)(void *)) | |
1884 | { | |
1885 | return flags; | |
1886 | } | |
1887 | ||
1888 | struct kmem_cache * | |
1889 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
1890 | unsigned long flags, void (*ctor)(void *)) | |
1891 | { | |
1892 | struct kmem_cache *cachep; | |
1893 | ||
1894 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1895 | if (cachep) { | |
1896 | cachep->refcount++; | |
1897 | ||
1898 | /* | |
1899 | * Adjust the object sizes so that we clear | |
1900 | * the complete object on kzalloc. | |
1901 | */ | |
1902 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1903 | } | |
1904 | return cachep; | |
1905 | } | |
1906 | ||
b03a017b JK |
1907 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
1908 | size_t size, unsigned long flags) | |
1909 | { | |
1910 | size_t left; | |
1911 | ||
1912 | cachep->num = 0; | |
1913 | ||
1914 | if (cachep->ctor || flags & SLAB_DESTROY_BY_RCU) | |
1915 | return false; | |
1916 | ||
1917 | left = calculate_slab_order(cachep, size, | |
1918 | flags | CFLGS_OBJFREELIST_SLAB); | |
1919 | if (!cachep->num) | |
1920 | return false; | |
1921 | ||
1922 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1923 | return false; | |
1924 | ||
1925 | cachep->colour = left / cachep->colour_off; | |
1926 | ||
1927 | return true; | |
1928 | } | |
1929 | ||
158e319b JK |
1930 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
1931 | size_t size, unsigned long flags) | |
1932 | { | |
1933 | size_t left; | |
1934 | ||
1935 | cachep->num = 0; | |
1936 | ||
1937 | /* | |
3217fd9b JK |
1938 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1939 | * to avoid recursive calls into kmemleak. | |
158e319b | 1940 | */ |
158e319b JK |
1941 | if (flags & SLAB_NOLEAKTRACE) |
1942 | return false; | |
1943 | ||
1944 | /* | |
1945 | * Size is large, assume best to place the slab management obj | |
1946 | * off-slab (should allow better packing of objs). | |
1947 | */ | |
1948 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1949 | if (!cachep->num) | |
1950 | return false; | |
1951 | ||
1952 | /* | |
1953 | * If the slab has been placed off-slab, and we have enough space then | |
1954 | * move it on-slab. This is at the expense of any extra colouring. | |
1955 | */ | |
1956 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1957 | return false; | |
1958 | ||
1959 | cachep->colour = left / cachep->colour_off; | |
1960 | ||
1961 | return true; | |
1962 | } | |
1963 | ||
1964 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
1965 | size_t size, unsigned long flags) | |
1966 | { | |
1967 | size_t left; | |
1968 | ||
1969 | cachep->num = 0; | |
1970 | ||
1971 | left = calculate_slab_order(cachep, size, flags); | |
1972 | if (!cachep->num) | |
1973 | return false; | |
1974 | ||
1975 | cachep->colour = left / cachep->colour_off; | |
1976 | ||
1977 | return true; | |
1978 | } | |
1979 | ||
1da177e4 | 1980 | /** |
039363f3 | 1981 | * __kmem_cache_create - Create a cache. |
a755b76a | 1982 | * @cachep: cache management descriptor |
1da177e4 | 1983 | * @flags: SLAB flags |
1da177e4 LT |
1984 | * |
1985 | * Returns a ptr to the cache on success, NULL on failure. | |
1986 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1987 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1988 | * |
1da177e4 LT |
1989 | * The flags are |
1990 | * | |
1991 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1992 | * to catch references to uninitialised memory. | |
1993 | * | |
1994 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1995 | * for buffer overruns. | |
1996 | * | |
1da177e4 LT |
1997 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1998 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1999 | * as davem. | |
2000 | */ | |
278b1bb1 | 2001 | int |
8a13a4cc | 2002 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2003 | { |
d4a5fca5 | 2004 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 2005 | gfp_t gfp; |
278b1bb1 | 2006 | int err; |
8a13a4cc | 2007 | size_t size = cachep->size; |
1da177e4 | 2008 | |
1da177e4 | 2009 | #if DEBUG |
1da177e4 LT |
2010 | #if FORCED_DEBUG |
2011 | /* | |
2012 | * Enable redzoning and last user accounting, except for caches with | |
2013 | * large objects, if the increased size would increase the object size | |
2014 | * above the next power of two: caches with object sizes just above a | |
2015 | * power of two have a significant amount of internal fragmentation. | |
2016 | */ | |
87a927c7 DW |
2017 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2018 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2019 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2020 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2021 | flags |= SLAB_POISON; | |
2022 | #endif | |
1da177e4 | 2023 | #endif |
1da177e4 | 2024 | |
a737b3e2 AM |
2025 | /* |
2026 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2027 | * unaligned accesses for some archs when redzoning is used, and makes |
2028 | * sure any on-slab bufctl's are also correctly aligned. | |
2029 | */ | |
b28a02de PE |
2030 | if (size & (BYTES_PER_WORD - 1)) { |
2031 | size += (BYTES_PER_WORD - 1); | |
2032 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2033 | } |
2034 | ||
87a927c7 DW |
2035 | if (flags & SLAB_RED_ZONE) { |
2036 | ralign = REDZONE_ALIGN; | |
2037 | /* If redzoning, ensure that the second redzone is suitably | |
2038 | * aligned, by adjusting the object size accordingly. */ | |
2039 | size += REDZONE_ALIGN - 1; | |
2040 | size &= ~(REDZONE_ALIGN - 1); | |
2041 | } | |
ca5f9703 | 2042 | |
a44b56d3 | 2043 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2044 | if (ralign < cachep->align) { |
2045 | ralign = cachep->align; | |
1da177e4 | 2046 | } |
3ff84a7f PE |
2047 | /* disable debug if necessary */ |
2048 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2049 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2050 | /* |
ca5f9703 | 2051 | * 4) Store it. |
1da177e4 | 2052 | */ |
8a13a4cc | 2053 | cachep->align = ralign; |
158e319b JK |
2054 | cachep->colour_off = cache_line_size(); |
2055 | /* Offset must be a multiple of the alignment. */ | |
2056 | if (cachep->colour_off < cachep->align) | |
2057 | cachep->colour_off = cachep->align; | |
1da177e4 | 2058 | |
83b519e8 PE |
2059 | if (slab_is_available()) |
2060 | gfp = GFP_KERNEL; | |
2061 | else | |
2062 | gfp = GFP_NOWAIT; | |
2063 | ||
1da177e4 | 2064 | #if DEBUG |
1da177e4 | 2065 | |
ca5f9703 PE |
2066 | /* |
2067 | * Both debugging options require word-alignment which is calculated | |
2068 | * into align above. | |
2069 | */ | |
1da177e4 | 2070 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2071 | /* add space for red zone words */ |
3ff84a7f PE |
2072 | cachep->obj_offset += sizeof(unsigned long long); |
2073 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2074 | } |
2075 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2076 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2077 | * the real object. But if the second red zone needs to be |
2078 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2079 | */ |
87a927c7 DW |
2080 | if (flags & SLAB_RED_ZONE) |
2081 | size += REDZONE_ALIGN; | |
2082 | else | |
2083 | size += BYTES_PER_WORD; | |
1da177e4 | 2084 | } |
832a15d2 JK |
2085 | #endif |
2086 | ||
7ed2f9e6 AP |
2087 | kasan_cache_create(cachep, &size, &flags); |
2088 | ||
832a15d2 JK |
2089 | size = ALIGN(size, cachep->align); |
2090 | /* | |
2091 | * We should restrict the number of objects in a slab to implement | |
2092 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2093 | */ | |
2094 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2095 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2096 | ||
2097 | #if DEBUG | |
03a2d2a3 JK |
2098 | /* |
2099 | * To activate debug pagealloc, off-slab management is necessary | |
2100 | * requirement. In early phase of initialization, small sized slab | |
2101 | * doesn't get initialized so it would not be possible. So, we need | |
2102 | * to check size >= 256. It guarantees that all necessary small | |
2103 | * sized slab is initialized in current slab initialization sequence. | |
2104 | */ | |
40323278 | 2105 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2106 | size >= 256 && cachep->object_size > cache_line_size()) { |
2107 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2108 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2109 | ||
2110 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2111 | flags |= CFLGS_OFF_SLAB; | |
2112 | cachep->obj_offset += tmp_size - size; | |
2113 | size = tmp_size; | |
2114 | goto done; | |
2115 | } | |
2116 | } | |
1da177e4 | 2117 | } |
1da177e4 LT |
2118 | #endif |
2119 | ||
b03a017b JK |
2120 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2121 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2122 | goto done; | |
2123 | } | |
2124 | ||
158e319b | 2125 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2126 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2127 | goto done; |
832a15d2 | 2128 | } |
1da177e4 | 2129 | |
158e319b JK |
2130 | if (set_on_slab_cache(cachep, size, flags)) |
2131 | goto done; | |
1da177e4 | 2132 | |
158e319b | 2133 | return -E2BIG; |
1da177e4 | 2134 | |
158e319b JK |
2135 | done: |
2136 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2137 | cachep->flags = flags; |
a57a4988 | 2138 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2139 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2140 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2141 | cachep->size = size; |
6a2d7a95 | 2142 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2143 | |
40b44137 JK |
2144 | #if DEBUG |
2145 | /* | |
2146 | * If we're going to use the generic kernel_map_pages() | |
2147 | * poisoning, then it's going to smash the contents of | |
2148 | * the redzone and userword anyhow, so switch them off. | |
2149 | */ | |
2150 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2151 | (cachep->flags & SLAB_POISON) && | |
2152 | is_debug_pagealloc_cache(cachep)) | |
2153 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2154 | #endif | |
2155 | ||
2156 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2157 | cachep->freelist_cache = |
2158 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2159 | } |
1da177e4 | 2160 | |
278b1bb1 CL |
2161 | err = setup_cpu_cache(cachep, gfp); |
2162 | if (err) { | |
52b4b950 | 2163 | __kmem_cache_release(cachep); |
278b1bb1 | 2164 | return err; |
2ed3a4ef | 2165 | } |
1da177e4 | 2166 | |
278b1bb1 | 2167 | return 0; |
1da177e4 | 2168 | } |
1da177e4 LT |
2169 | |
2170 | #if DEBUG | |
2171 | static void check_irq_off(void) | |
2172 | { | |
2173 | BUG_ON(!irqs_disabled()); | |
2174 | } | |
2175 | ||
2176 | static void check_irq_on(void) | |
2177 | { | |
2178 | BUG_ON(irqs_disabled()); | |
2179 | } | |
2180 | ||
18726ca8 JK |
2181 | static void check_mutex_acquired(void) |
2182 | { | |
2183 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2184 | } | |
2185 | ||
343e0d7a | 2186 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2187 | { |
2188 | #ifdef CONFIG_SMP | |
2189 | check_irq_off(); | |
18bf8541 | 2190 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2191 | #endif |
2192 | } | |
e498be7d | 2193 | |
343e0d7a | 2194 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2195 | { |
2196 | #ifdef CONFIG_SMP | |
2197 | check_irq_off(); | |
18bf8541 | 2198 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2199 | #endif |
2200 | } | |
2201 | ||
1da177e4 LT |
2202 | #else |
2203 | #define check_irq_off() do { } while(0) | |
2204 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2205 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2206 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2207 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2208 | #endif |
2209 | ||
18726ca8 JK |
2210 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2211 | int node, bool free_all, struct list_head *list) | |
2212 | { | |
2213 | int tofree; | |
2214 | ||
2215 | if (!ac || !ac->avail) | |
2216 | return; | |
2217 | ||
2218 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2219 | if (tofree > ac->avail) | |
2220 | tofree = (ac->avail + 1) / 2; | |
2221 | ||
2222 | free_block(cachep, ac->entry, tofree, node, list); | |
2223 | ac->avail -= tofree; | |
2224 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2225 | } | |
aab2207c | 2226 | |
1da177e4 LT |
2227 | static void do_drain(void *arg) |
2228 | { | |
a737b3e2 | 2229 | struct kmem_cache *cachep = arg; |
1da177e4 | 2230 | struct array_cache *ac; |
7d6e6d09 | 2231 | int node = numa_mem_id(); |
18bf8541 | 2232 | struct kmem_cache_node *n; |
97654dfa | 2233 | LIST_HEAD(list); |
1da177e4 LT |
2234 | |
2235 | check_irq_off(); | |
9a2dba4b | 2236 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2237 | n = get_node(cachep, node); |
2238 | spin_lock(&n->list_lock); | |
97654dfa | 2239 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2240 | spin_unlock(&n->list_lock); |
97654dfa | 2241 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2242 | ac->avail = 0; |
2243 | } | |
2244 | ||
343e0d7a | 2245 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2246 | { |
ce8eb6c4 | 2247 | struct kmem_cache_node *n; |
e498be7d | 2248 | int node; |
18726ca8 | 2249 | LIST_HEAD(list); |
e498be7d | 2250 | |
15c8b6c1 | 2251 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2252 | check_irq_on(); |
18bf8541 CL |
2253 | for_each_kmem_cache_node(cachep, node, n) |
2254 | if (n->alien) | |
ce8eb6c4 | 2255 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2256 | |
18726ca8 JK |
2257 | for_each_kmem_cache_node(cachep, node, n) { |
2258 | spin_lock_irq(&n->list_lock); | |
2259 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2260 | spin_unlock_irq(&n->list_lock); | |
2261 | ||
2262 | slabs_destroy(cachep, &list); | |
2263 | } | |
1da177e4 LT |
2264 | } |
2265 | ||
ed11d9eb CL |
2266 | /* |
2267 | * Remove slabs from the list of free slabs. | |
2268 | * Specify the number of slabs to drain in tofree. | |
2269 | * | |
2270 | * Returns the actual number of slabs released. | |
2271 | */ | |
2272 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2273 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2274 | { |
ed11d9eb CL |
2275 | struct list_head *p; |
2276 | int nr_freed; | |
8456a648 | 2277 | struct page *page; |
1da177e4 | 2278 | |
ed11d9eb | 2279 | nr_freed = 0; |
ce8eb6c4 | 2280 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2281 | |
ce8eb6c4 CL |
2282 | spin_lock_irq(&n->list_lock); |
2283 | p = n->slabs_free.prev; | |
2284 | if (p == &n->slabs_free) { | |
2285 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2286 | goto out; |
2287 | } | |
1da177e4 | 2288 | |
8456a648 | 2289 | page = list_entry(p, struct page, lru); |
8456a648 | 2290 | list_del(&page->lru); |
ed11d9eb CL |
2291 | /* |
2292 | * Safe to drop the lock. The slab is no longer linked | |
2293 | * to the cache. | |
2294 | */ | |
ce8eb6c4 CL |
2295 | n->free_objects -= cache->num; |
2296 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2297 | slab_destroy(cache, page); |
ed11d9eb | 2298 | nr_freed++; |
1da177e4 | 2299 | } |
ed11d9eb CL |
2300 | out: |
2301 | return nr_freed; | |
1da177e4 LT |
2302 | } |
2303 | ||
d6e0b7fa | 2304 | int __kmem_cache_shrink(struct kmem_cache *cachep, bool deactivate) |
e498be7d | 2305 | { |
18bf8541 CL |
2306 | int ret = 0; |
2307 | int node; | |
ce8eb6c4 | 2308 | struct kmem_cache_node *n; |
e498be7d CL |
2309 | |
2310 | drain_cpu_caches(cachep); | |
2311 | ||
2312 | check_irq_on(); | |
18bf8541 | 2313 | for_each_kmem_cache_node(cachep, node, n) { |
0fa8103b | 2314 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2315 | |
ce8eb6c4 CL |
2316 | ret += !list_empty(&n->slabs_full) || |
2317 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2318 | } |
2319 | return (ret ? 1 : 0); | |
2320 | } | |
2321 | ||
945cf2b6 | 2322 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 DS |
2323 | { |
2324 | return __kmem_cache_shrink(cachep, false); | |
2325 | } | |
2326 | ||
2327 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2328 | { |
12c3667f | 2329 | int i; |
ce8eb6c4 | 2330 | struct kmem_cache_node *n; |
1da177e4 | 2331 | |
bf0dea23 | 2332 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2333 | |
ce8eb6c4 | 2334 | /* NUMA: free the node structures */ |
18bf8541 CL |
2335 | for_each_kmem_cache_node(cachep, i, n) { |
2336 | kfree(n->shared); | |
2337 | free_alien_cache(n->alien); | |
2338 | kfree(n); | |
2339 | cachep->node[i] = NULL; | |
12c3667f | 2340 | } |
1da177e4 | 2341 | } |
1da177e4 | 2342 | |
e5ac9c5a RT |
2343 | /* |
2344 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2345 | * |
2346 | * For a slab cache when the slab descriptor is off-slab, the | |
2347 | * slab descriptor can't come from the same cache which is being created, | |
2348 | * Because if it is the case, that means we defer the creation of | |
2349 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2350 | * And we eventually call down to __kmem_cache_create(), which | |
2351 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2352 | * This is a "chicken-and-egg" problem. | |
2353 | * | |
2354 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2355 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2356 | */ |
7e007355 | 2357 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2358 | struct page *page, int colour_off, |
2359 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2360 | { |
7e007355 | 2361 | void *freelist; |
0c3aa83e | 2362 | void *addr = page_address(page); |
b28a02de | 2363 | |
2e6b3602 JK |
2364 | page->s_mem = addr + colour_off; |
2365 | page->active = 0; | |
2366 | ||
b03a017b JK |
2367 | if (OBJFREELIST_SLAB(cachep)) |
2368 | freelist = NULL; | |
2369 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2370 | /* Slab management obj is off-slab. */ |
8456a648 | 2371 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2372 | local_flags, nodeid); |
8456a648 | 2373 | if (!freelist) |
1da177e4 LT |
2374 | return NULL; |
2375 | } else { | |
2e6b3602 JK |
2376 | /* We will use last bytes at the slab for freelist */ |
2377 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2378 | cachep->freelist_size; | |
1da177e4 | 2379 | } |
2e6b3602 | 2380 | |
8456a648 | 2381 | return freelist; |
1da177e4 LT |
2382 | } |
2383 | ||
7cc68973 | 2384 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2385 | { |
a41adfaa | 2386 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2387 | } |
2388 | ||
2389 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2390 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2391 | { |
a41adfaa | 2392 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2393 | } |
2394 | ||
10b2e9e8 | 2395 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2396 | { |
10b2e9e8 | 2397 | #if DEBUG |
1da177e4 LT |
2398 | int i; |
2399 | ||
2400 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2401 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2402 | |
1da177e4 LT |
2403 | if (cachep->flags & SLAB_STORE_USER) |
2404 | *dbg_userword(cachep, objp) = NULL; | |
2405 | ||
2406 | if (cachep->flags & SLAB_RED_ZONE) { | |
2407 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2408 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2409 | } | |
2410 | /* | |
a737b3e2 AM |
2411 | * Constructors are not allowed to allocate memory from the same |
2412 | * cache which they are a constructor for. Otherwise, deadlock. | |
2413 | * They must also be threaded. | |
1da177e4 | 2414 | */ |
7ed2f9e6 AP |
2415 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2416 | kasan_unpoison_object_data(cachep, | |
2417 | objp + obj_offset(cachep)); | |
51cc5068 | 2418 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2419 | kasan_poison_object_data( |
2420 | cachep, objp + obj_offset(cachep)); | |
2421 | } | |
1da177e4 LT |
2422 | |
2423 | if (cachep->flags & SLAB_RED_ZONE) { | |
2424 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2425 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2426 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2427 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2428 | } |
40b44137 JK |
2429 | /* need to poison the objs? */ |
2430 | if (cachep->flags & SLAB_POISON) { | |
2431 | poison_obj(cachep, objp, POISON_FREE); | |
2432 | slab_kernel_map(cachep, objp, 0, 0); | |
2433 | } | |
10b2e9e8 | 2434 | } |
1da177e4 | 2435 | #endif |
10b2e9e8 JK |
2436 | } |
2437 | ||
2438 | static void cache_init_objs(struct kmem_cache *cachep, | |
2439 | struct page *page) | |
2440 | { | |
2441 | int i; | |
7ed2f9e6 | 2442 | void *objp; |
10b2e9e8 JK |
2443 | |
2444 | cache_init_objs_debug(cachep, page); | |
2445 | ||
b03a017b JK |
2446 | if (OBJFREELIST_SLAB(cachep)) { |
2447 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + | |
2448 | obj_offset(cachep); | |
2449 | } | |
2450 | ||
10b2e9e8 JK |
2451 | for (i = 0; i < cachep->num; i++) { |
2452 | /* constructor could break poison info */ | |
7ed2f9e6 AP |
2453 | if (DEBUG == 0 && cachep->ctor) { |
2454 | objp = index_to_obj(cachep, page, i); | |
2455 | kasan_unpoison_object_data(cachep, objp); | |
2456 | cachep->ctor(objp); | |
2457 | kasan_poison_object_data(cachep, objp); | |
2458 | } | |
10b2e9e8 | 2459 | |
e5c58dfd | 2460 | set_free_obj(page, i, i); |
1da177e4 | 2461 | } |
1da177e4 LT |
2462 | } |
2463 | ||
343e0d7a | 2464 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2465 | { |
4b51d669 CL |
2466 | if (CONFIG_ZONE_DMA_FLAG) { |
2467 | if (flags & GFP_DMA) | |
a618e89f | 2468 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2469 | else |
a618e89f | 2470 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2471 | } |
1da177e4 LT |
2472 | } |
2473 | ||
260b61dd | 2474 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2475 | { |
b1cb0982 | 2476 | void *objp; |
78d382d7 | 2477 | |
e5c58dfd | 2478 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2479 | page->active++; |
78d382d7 | 2480 | |
d31676df JK |
2481 | #if DEBUG |
2482 | if (cachep->flags & SLAB_STORE_USER) | |
2483 | set_store_user_dirty(cachep); | |
2484 | #endif | |
2485 | ||
78d382d7 MD |
2486 | return objp; |
2487 | } | |
2488 | ||
260b61dd JK |
2489 | static void slab_put_obj(struct kmem_cache *cachep, |
2490 | struct page *page, void *objp) | |
78d382d7 | 2491 | { |
8456a648 | 2492 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2493 | #if DEBUG |
16025177 | 2494 | unsigned int i; |
b1cb0982 | 2495 | |
b1cb0982 | 2496 | /* Verify double free bug */ |
8456a648 | 2497 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2498 | if (get_free_obj(page, i) == objnr) { |
1170532b | 2499 | pr_err("slab: double free detected in cache '%s', objp %p\n", |
756a025f | 2500 | cachep->name, objp); |
b1cb0982 JK |
2501 | BUG(); |
2502 | } | |
78d382d7 MD |
2503 | } |
2504 | #endif | |
8456a648 | 2505 | page->active--; |
b03a017b JK |
2506 | if (!page->freelist) |
2507 | page->freelist = objp + obj_offset(cachep); | |
2508 | ||
e5c58dfd | 2509 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2510 | } |
2511 | ||
4776874f PE |
2512 | /* |
2513 | * Map pages beginning at addr to the given cache and slab. This is required | |
2514 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2515 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2516 | */ |
8456a648 | 2517 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2518 | void *freelist) |
1da177e4 | 2519 | { |
a57a4988 | 2520 | page->slab_cache = cache; |
8456a648 | 2521 | page->freelist = freelist; |
1da177e4 LT |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2526 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2527 | */ | |
3c517a61 | 2528 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2529 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2530 | { |
7e007355 | 2531 | void *freelist; |
b28a02de PE |
2532 | size_t offset; |
2533 | gfp_t local_flags; | |
ce8eb6c4 | 2534 | struct kmem_cache_node *n; |
1da177e4 | 2535 | |
a737b3e2 AM |
2536 | /* |
2537 | * Be lazy and only check for valid flags here, keeping it out of the | |
2538 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2539 | */ |
c871ac4e AM |
2540 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
2541 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
2542 | BUG(); | |
2543 | } | |
6cb06229 | 2544 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2545 | |
ce8eb6c4 | 2546 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2547 | check_irq_off(); |
18bf8541 | 2548 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 2549 | spin_lock(&n->list_lock); |
1da177e4 LT |
2550 | |
2551 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2552 | offset = n->colour_next; |
2553 | n->colour_next++; | |
2554 | if (n->colour_next >= cachep->colour) | |
2555 | n->colour_next = 0; | |
2556 | spin_unlock(&n->list_lock); | |
1da177e4 | 2557 | |
2e1217cf | 2558 | offset *= cachep->colour_off; |
1da177e4 | 2559 | |
d0164adc | 2560 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2561 | local_irq_enable(); |
2562 | ||
2563 | /* | |
2564 | * The test for missing atomic flag is performed here, rather than | |
2565 | * the more obvious place, simply to reduce the critical path length | |
2566 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2567 | * will eventually be caught here (where it matters). | |
2568 | */ | |
2569 | kmem_flagcheck(cachep, flags); | |
2570 | ||
a737b3e2 AM |
2571 | /* |
2572 | * Get mem for the objs. Attempt to allocate a physical page from | |
2573 | * 'nodeid'. | |
e498be7d | 2574 | */ |
0c3aa83e JK |
2575 | if (!page) |
2576 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2577 | if (!page) | |
1da177e4 LT |
2578 | goto failed; |
2579 | ||
2580 | /* Get slab management. */ | |
8456a648 | 2581 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2582 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
b03a017b | 2583 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2584 | goto opps1; |
2585 | ||
8456a648 | 2586 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2587 | |
7ed2f9e6 | 2588 | kasan_poison_slab(page); |
8456a648 | 2589 | cache_init_objs(cachep, page); |
1da177e4 | 2590 | |
d0164adc | 2591 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2592 | local_irq_disable(); |
2593 | check_irq_off(); | |
ce8eb6c4 | 2594 | spin_lock(&n->list_lock); |
1da177e4 LT |
2595 | |
2596 | /* Make slab active. */ | |
8456a648 | 2597 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2598 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2599 | n->free_objects += cachep->num; |
2600 | spin_unlock(&n->list_lock); | |
1da177e4 | 2601 | return 1; |
a737b3e2 | 2602 | opps1: |
0c3aa83e | 2603 | kmem_freepages(cachep, page); |
a737b3e2 | 2604 | failed: |
d0164adc | 2605 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2606 | local_irq_disable(); |
2607 | return 0; | |
2608 | } | |
2609 | ||
2610 | #if DEBUG | |
2611 | ||
2612 | /* | |
2613 | * Perform extra freeing checks: | |
2614 | * - detect bad pointers. | |
2615 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2616 | */ |
2617 | static void kfree_debugcheck(const void *objp) | |
2618 | { | |
1da177e4 | 2619 | if (!virt_addr_valid(objp)) { |
1170532b | 2620 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2621 | (unsigned long)objp); |
2622 | BUG(); | |
1da177e4 | 2623 | } |
1da177e4 LT |
2624 | } |
2625 | ||
58ce1fd5 PE |
2626 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2627 | { | |
b46b8f19 | 2628 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2629 | |
2630 | redzone1 = *dbg_redzone1(cache, obj); | |
2631 | redzone2 = *dbg_redzone2(cache, obj); | |
2632 | ||
2633 | /* | |
2634 | * Redzone is ok. | |
2635 | */ | |
2636 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2637 | return; | |
2638 | ||
2639 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2640 | slab_error(cache, "double free detected"); | |
2641 | else | |
2642 | slab_error(cache, "memory outside object was overwritten"); | |
2643 | ||
1170532b JP |
2644 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2645 | obj, redzone1, redzone2); | |
58ce1fd5 PE |
2646 | } |
2647 | ||
343e0d7a | 2648 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2649 | unsigned long caller) |
1da177e4 | 2650 | { |
1da177e4 | 2651 | unsigned int objnr; |
8456a648 | 2652 | struct page *page; |
1da177e4 | 2653 | |
80cbd911 MW |
2654 | BUG_ON(virt_to_cache(objp) != cachep); |
2655 | ||
3dafccf2 | 2656 | objp -= obj_offset(cachep); |
1da177e4 | 2657 | kfree_debugcheck(objp); |
b49af68f | 2658 | page = virt_to_head_page(objp); |
1da177e4 | 2659 | |
1da177e4 | 2660 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2661 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2662 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2663 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2664 | } | |
d31676df JK |
2665 | if (cachep->flags & SLAB_STORE_USER) { |
2666 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2667 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2668 | } |
1da177e4 | 2669 | |
8456a648 | 2670 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2671 | |
2672 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2673 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2674 | |
1da177e4 | 2675 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2676 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2677 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2678 | } |
2679 | return objp; | |
2680 | } | |
2681 | ||
1da177e4 LT |
2682 | #else |
2683 | #define kfree_debugcheck(x) do { } while(0) | |
2684 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2685 | #endif |
2686 | ||
b03a017b JK |
2687 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2688 | void **list) | |
2689 | { | |
2690 | #if DEBUG | |
2691 | void *next = *list; | |
2692 | void *objp; | |
2693 | ||
2694 | while (next) { | |
2695 | objp = next - obj_offset(cachep); | |
2696 | next = *(void **)next; | |
2697 | poison_obj(cachep, objp, POISON_FREE); | |
2698 | } | |
2699 | #endif | |
2700 | } | |
2701 | ||
d8410234 | 2702 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2703 | struct kmem_cache_node *n, struct page *page, |
2704 | void **list) | |
d8410234 JK |
2705 | { |
2706 | /* move slabp to correct slabp list: */ | |
2707 | list_del(&page->lru); | |
b03a017b | 2708 | if (page->active == cachep->num) { |
d8410234 | 2709 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2710 | if (OBJFREELIST_SLAB(cachep)) { |
2711 | #if DEBUG | |
2712 | /* Poisoning will be done without holding the lock */ | |
2713 | if (cachep->flags & SLAB_POISON) { | |
2714 | void **objp = page->freelist; | |
2715 | ||
2716 | *objp = *list; | |
2717 | *list = objp; | |
2718 | } | |
2719 | #endif | |
2720 | page->freelist = NULL; | |
2721 | } | |
2722 | } else | |
d8410234 JK |
2723 | list_add(&page->lru, &n->slabs_partial); |
2724 | } | |
2725 | ||
f68f8ddd JK |
2726 | /* Try to find non-pfmemalloc slab if needed */ |
2727 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
2728 | struct page *page, bool pfmemalloc) | |
2729 | { | |
2730 | if (!page) | |
2731 | return NULL; | |
2732 | ||
2733 | if (pfmemalloc) | |
2734 | return page; | |
2735 | ||
2736 | if (!PageSlabPfmemalloc(page)) | |
2737 | return page; | |
2738 | ||
2739 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2740 | if (n->free_objects > n->free_limit) { | |
2741 | ClearPageSlabPfmemalloc(page); | |
2742 | return page; | |
2743 | } | |
2744 | ||
2745 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2746 | list_del(&page->lru); | |
2747 | if (!page->active) | |
2748 | list_add_tail(&page->lru, &n->slabs_free); | |
2749 | else | |
2750 | list_add_tail(&page->lru, &n->slabs_partial); | |
2751 | ||
2752 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2753 | if (!PageSlabPfmemalloc(page)) | |
2754 | return page; | |
2755 | } | |
2756 | ||
2757 | list_for_each_entry(page, &n->slabs_free, lru) { | |
2758 | if (!PageSlabPfmemalloc(page)) | |
2759 | return page; | |
2760 | } | |
2761 | ||
2762 | return NULL; | |
2763 | } | |
2764 | ||
2765 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2766 | { |
2767 | struct page *page; | |
2768 | ||
2769 | page = list_first_entry_or_null(&n->slabs_partial, | |
2770 | struct page, lru); | |
2771 | if (!page) { | |
2772 | n->free_touched = 1; | |
2773 | page = list_first_entry_or_null(&n->slabs_free, | |
2774 | struct page, lru); | |
2775 | } | |
2776 | ||
f68f8ddd JK |
2777 | if (sk_memalloc_socks()) |
2778 | return get_valid_first_slab(n, page, pfmemalloc); | |
2779 | ||
7aa0d227 GT |
2780 | return page; |
2781 | } | |
2782 | ||
f68f8ddd JK |
2783 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2784 | struct kmem_cache_node *n, gfp_t flags) | |
2785 | { | |
2786 | struct page *page; | |
2787 | void *obj; | |
2788 | void *list = NULL; | |
2789 | ||
2790 | if (!gfp_pfmemalloc_allowed(flags)) | |
2791 | return NULL; | |
2792 | ||
2793 | spin_lock(&n->list_lock); | |
2794 | page = get_first_slab(n, true); | |
2795 | if (!page) { | |
2796 | spin_unlock(&n->list_lock); | |
2797 | return NULL; | |
2798 | } | |
2799 | ||
2800 | obj = slab_get_obj(cachep, page); | |
2801 | n->free_objects--; | |
2802 | ||
2803 | fixup_slab_list(cachep, n, page, &list); | |
2804 | ||
2805 | spin_unlock(&n->list_lock); | |
2806 | fixup_objfreelist_debug(cachep, &list); | |
2807 | ||
2808 | return obj; | |
2809 | } | |
2810 | ||
2811 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) | |
1da177e4 LT |
2812 | { |
2813 | int batchcount; | |
ce8eb6c4 | 2814 | struct kmem_cache_node *n; |
1da177e4 | 2815 | struct array_cache *ac; |
1ca4cb24 | 2816 | int node; |
b03a017b | 2817 | void *list = NULL; |
1ca4cb24 | 2818 | |
1da177e4 | 2819 | check_irq_off(); |
7d6e6d09 | 2820 | node = numa_mem_id(); |
f68f8ddd | 2821 | |
072bb0aa | 2822 | retry: |
9a2dba4b | 2823 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2824 | batchcount = ac->batchcount; |
2825 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2826 | /* |
2827 | * If there was little recent activity on this cache, then | |
2828 | * perform only a partial refill. Otherwise we could generate | |
2829 | * refill bouncing. | |
1da177e4 LT |
2830 | */ |
2831 | batchcount = BATCHREFILL_LIMIT; | |
2832 | } | |
18bf8541 | 2833 | n = get_node(cachep, node); |
e498be7d | 2834 | |
ce8eb6c4 CL |
2835 | BUG_ON(ac->avail > 0 || !n); |
2836 | spin_lock(&n->list_lock); | |
1da177e4 | 2837 | |
3ded175a | 2838 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2839 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2840 | n->shared->touched = 1; | |
3ded175a | 2841 | goto alloc_done; |
44b57f1c | 2842 | } |
3ded175a | 2843 | |
1da177e4 | 2844 | while (batchcount > 0) { |
8456a648 | 2845 | struct page *page; |
1da177e4 | 2846 | /* Get slab alloc is to come from. */ |
f68f8ddd | 2847 | page = get_first_slab(n, false); |
7aa0d227 GT |
2848 | if (!page) |
2849 | goto must_grow; | |
1da177e4 | 2850 | |
1da177e4 | 2851 | check_spinlock_acquired(cachep); |
714b8171 PE |
2852 | |
2853 | /* | |
2854 | * The slab was either on partial or free list so | |
2855 | * there must be at least one object available for | |
2856 | * allocation. | |
2857 | */ | |
8456a648 | 2858 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2859 | |
8456a648 | 2860 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2861 | STATS_INC_ALLOCED(cachep); |
2862 | STATS_INC_ACTIVE(cachep); | |
2863 | STATS_SET_HIGH(cachep); | |
2864 | ||
f68f8ddd | 2865 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); |
1da177e4 | 2866 | } |
1da177e4 | 2867 | |
b03a017b | 2868 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2869 | } |
2870 | ||
a737b3e2 | 2871 | must_grow: |
ce8eb6c4 | 2872 | n->free_objects -= ac->avail; |
a737b3e2 | 2873 | alloc_done: |
ce8eb6c4 | 2874 | spin_unlock(&n->list_lock); |
b03a017b | 2875 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 LT |
2876 | |
2877 | if (unlikely(!ac->avail)) { | |
2878 | int x; | |
f68f8ddd JK |
2879 | |
2880 | /* Check if we can use obj in pfmemalloc slab */ | |
2881 | if (sk_memalloc_socks()) { | |
2882 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
2883 | ||
2884 | if (obj) | |
2885 | return obj; | |
2886 | } | |
2887 | ||
4167e9b2 | 2888 | x = cache_grow(cachep, gfp_exact_node(flags), node, NULL); |
e498be7d | 2889 | |
a737b3e2 | 2890 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2891 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2892 | node = numa_mem_id(); |
072bb0aa MG |
2893 | |
2894 | /* no objects in sight? abort */ | |
f68f8ddd | 2895 | if (!x && ac->avail == 0) |
1da177e4 LT |
2896 | return NULL; |
2897 | ||
a737b3e2 | 2898 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2899 | goto retry; |
2900 | } | |
2901 | ac->touched = 1; | |
072bb0aa | 2902 | |
f68f8ddd | 2903 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2904 | } |
2905 | ||
a737b3e2 AM |
2906 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2907 | gfp_t flags) | |
1da177e4 | 2908 | { |
d0164adc | 2909 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
2910 | #if DEBUG |
2911 | kmem_flagcheck(cachep, flags); | |
2912 | #endif | |
2913 | } | |
2914 | ||
2915 | #if DEBUG | |
a737b3e2 | 2916 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2917 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2918 | { |
b28a02de | 2919 | if (!objp) |
1da177e4 | 2920 | return objp; |
b28a02de | 2921 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2922 | check_poison_obj(cachep, objp); |
40b44137 | 2923 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
2924 | poison_obj(cachep, objp, POISON_INUSE); |
2925 | } | |
2926 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2927 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2928 | |
2929 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2930 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2931 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 2932 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
1170532b JP |
2933 | pr_err("%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
2934 | objp, *dbg_redzone1(cachep, objp), | |
2935 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2936 | } |
2937 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2938 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2939 | } | |
03787301 | 2940 | |
3dafccf2 | 2941 | objp += obj_offset(cachep); |
4f104934 | 2942 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2943 | cachep->ctor(objp); |
7ea466f2 TH |
2944 | if (ARCH_SLAB_MINALIGN && |
2945 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
1170532b | 2946 | pr_err("0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 2947 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 2948 | } |
1da177e4 LT |
2949 | return objp; |
2950 | } | |
2951 | #else | |
2952 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2953 | #endif | |
2954 | ||
343e0d7a | 2955 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2956 | { |
b28a02de | 2957 | void *objp; |
1da177e4 LT |
2958 | struct array_cache *ac; |
2959 | ||
5c382300 | 2960 | check_irq_off(); |
8a8b6502 | 2961 | |
9a2dba4b | 2962 | ac = cpu_cache_get(cachep); |
1da177e4 | 2963 | if (likely(ac->avail)) { |
1da177e4 | 2964 | ac->touched = 1; |
f68f8ddd | 2965 | objp = ac->entry[--ac->avail]; |
072bb0aa | 2966 | |
f68f8ddd JK |
2967 | STATS_INC_ALLOCHIT(cachep); |
2968 | goto out; | |
1da177e4 | 2969 | } |
072bb0aa MG |
2970 | |
2971 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 2972 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
2973 | /* |
2974 | * the 'ac' may be updated by cache_alloc_refill(), | |
2975 | * and kmemleak_erase() requires its correct value. | |
2976 | */ | |
2977 | ac = cpu_cache_get(cachep); | |
2978 | ||
2979 | out: | |
d5cff635 CM |
2980 | /* |
2981 | * To avoid a false negative, if an object that is in one of the | |
2982 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
2983 | * treat the array pointers as a reference to the object. | |
2984 | */ | |
f3d8b53a O |
2985 | if (objp) |
2986 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
2987 | return objp; |
2988 | } | |
2989 | ||
e498be7d | 2990 | #ifdef CONFIG_NUMA |
c61afb18 | 2991 | /* |
2ad654bc | 2992 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
2993 | * |
2994 | * If we are in_interrupt, then process context, including cpusets and | |
2995 | * mempolicy, may not apply and should not be used for allocation policy. | |
2996 | */ | |
2997 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
2998 | { | |
2999 | int nid_alloc, nid_here; | |
3000 | ||
765c4507 | 3001 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3002 | return NULL; |
7d6e6d09 | 3003 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3004 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3005 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3006 | else if (current->mempolicy) |
2a389610 | 3007 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3008 | if (nid_alloc != nid_here) |
8b98c169 | 3009 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3010 | return NULL; |
3011 | } | |
3012 | ||
765c4507 CL |
3013 | /* |
3014 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3015 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3016 | * available node for available objects. If that fails then we |
3c517a61 CL |
3017 | * perform an allocation without specifying a node. This allows the page |
3018 | * allocator to do its reclaim / fallback magic. We then insert the | |
3019 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3020 | */ |
8c8cc2c1 | 3021 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3022 | { |
8c8cc2c1 PE |
3023 | struct zonelist *zonelist; |
3024 | gfp_t local_flags; | |
dd1a239f | 3025 | struct zoneref *z; |
54a6eb5c MG |
3026 | struct zone *zone; |
3027 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3028 | void *obj = NULL; |
3c517a61 | 3029 | int nid; |
cc9a6c87 | 3030 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3031 | |
3032 | if (flags & __GFP_THISNODE) | |
3033 | return NULL; | |
3034 | ||
6cb06229 | 3035 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3036 | |
cc9a6c87 | 3037 | retry_cpuset: |
d26914d1 | 3038 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3039 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3040 | |
3c517a61 CL |
3041 | retry: |
3042 | /* | |
3043 | * Look through allowed nodes for objects available | |
3044 | * from existing per node queues. | |
3045 | */ | |
54a6eb5c MG |
3046 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3047 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3048 | |
061d7074 | 3049 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3050 | get_node(cache, nid) && |
3051 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3052 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3053 | gfp_exact_node(flags), nid); |
481c5346 CL |
3054 | if (obj) |
3055 | break; | |
3056 | } | |
3c517a61 CL |
3057 | } |
3058 | ||
cfce6604 | 3059 | if (!obj) { |
3c517a61 CL |
3060 | /* |
3061 | * This allocation will be performed within the constraints | |
3062 | * of the current cpuset / memory policy requirements. | |
3063 | * We may trigger various forms of reclaim on the allowed | |
3064 | * set and go into memory reserves if necessary. | |
3065 | */ | |
0c3aa83e JK |
3066 | struct page *page; |
3067 | ||
d0164adc | 3068 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 CL |
3069 | local_irq_enable(); |
3070 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3071 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
d0164adc | 3072 | if (gfpflags_allow_blocking(local_flags)) |
dd47ea75 | 3073 | local_irq_disable(); |
0c3aa83e | 3074 | if (page) { |
3c517a61 CL |
3075 | /* |
3076 | * Insert into the appropriate per node queues | |
3077 | */ | |
0c3aa83e JK |
3078 | nid = page_to_nid(page); |
3079 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 | 3080 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3081 | gfp_exact_node(flags), nid); |
3c517a61 CL |
3082 | if (!obj) |
3083 | /* | |
3084 | * Another processor may allocate the | |
3085 | * objects in the slab since we are | |
3086 | * not holding any locks. | |
3087 | */ | |
3088 | goto retry; | |
3089 | } else { | |
b6a60451 | 3090 | /* cache_grow already freed obj */ |
3c517a61 CL |
3091 | obj = NULL; |
3092 | } | |
3093 | } | |
aedb0eb1 | 3094 | } |
cc9a6c87 | 3095 | |
d26914d1 | 3096 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3097 | goto retry_cpuset; |
765c4507 CL |
3098 | return obj; |
3099 | } | |
3100 | ||
e498be7d CL |
3101 | /* |
3102 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3103 | */ |
8b98c169 | 3104 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3105 | int nodeid) |
e498be7d | 3106 | { |
8456a648 | 3107 | struct page *page; |
ce8eb6c4 | 3108 | struct kmem_cache_node *n; |
b28a02de | 3109 | void *obj; |
b03a017b | 3110 | void *list = NULL; |
b28a02de PE |
3111 | int x; |
3112 | ||
7c3fbbdd | 3113 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3114 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3115 | BUG_ON(!n); |
b28a02de | 3116 | |
a737b3e2 | 3117 | retry: |
ca3b9b91 | 3118 | check_irq_off(); |
ce8eb6c4 | 3119 | spin_lock(&n->list_lock); |
f68f8ddd | 3120 | page = get_first_slab(n, false); |
7aa0d227 GT |
3121 | if (!page) |
3122 | goto must_grow; | |
b28a02de | 3123 | |
b28a02de | 3124 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3125 | |
3126 | STATS_INC_NODEALLOCS(cachep); | |
3127 | STATS_INC_ACTIVE(cachep); | |
3128 | STATS_SET_HIGH(cachep); | |
3129 | ||
8456a648 | 3130 | BUG_ON(page->active == cachep->num); |
b28a02de | 3131 | |
260b61dd | 3132 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3133 | n->free_objects--; |
b28a02de | 3134 | |
b03a017b | 3135 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3136 | |
ce8eb6c4 | 3137 | spin_unlock(&n->list_lock); |
b03a017b | 3138 | fixup_objfreelist_debug(cachep, &list); |
b28a02de | 3139 | goto done; |
e498be7d | 3140 | |
a737b3e2 | 3141 | must_grow: |
ce8eb6c4 | 3142 | spin_unlock(&n->list_lock); |
4167e9b2 | 3143 | x = cache_grow(cachep, gfp_exact_node(flags), nodeid, NULL); |
765c4507 CL |
3144 | if (x) |
3145 | goto retry; | |
1da177e4 | 3146 | |
8c8cc2c1 | 3147 | return fallback_alloc(cachep, flags); |
e498be7d | 3148 | |
a737b3e2 | 3149 | done: |
b28a02de | 3150 | return obj; |
e498be7d | 3151 | } |
8c8cc2c1 | 3152 | |
8c8cc2c1 | 3153 | static __always_inline void * |
48356303 | 3154 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3155 | unsigned long caller) |
8c8cc2c1 PE |
3156 | { |
3157 | unsigned long save_flags; | |
3158 | void *ptr; | |
7d6e6d09 | 3159 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3160 | |
dcce284a | 3161 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3162 | cachep = slab_pre_alloc_hook(cachep, flags); |
3163 | if (unlikely(!cachep)) | |
824ebef1 AM |
3164 | return NULL; |
3165 | ||
8c8cc2c1 PE |
3166 | cache_alloc_debugcheck_before(cachep, flags); |
3167 | local_irq_save(save_flags); | |
3168 | ||
eacbbae3 | 3169 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3170 | nodeid = slab_node; |
8c8cc2c1 | 3171 | |
18bf8541 | 3172 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3173 | /* Node not bootstrapped yet */ |
3174 | ptr = fallback_alloc(cachep, flags); | |
3175 | goto out; | |
3176 | } | |
3177 | ||
7d6e6d09 | 3178 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3179 | /* |
3180 | * Use the locally cached objects if possible. | |
3181 | * However ____cache_alloc does not allow fallback | |
3182 | * to other nodes. It may fail while we still have | |
3183 | * objects on other nodes available. | |
3184 | */ | |
3185 | ptr = ____cache_alloc(cachep, flags); | |
3186 | if (ptr) | |
3187 | goto out; | |
3188 | } | |
3189 | /* ___cache_alloc_node can fall back to other nodes */ | |
3190 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3191 | out: | |
3192 | local_irq_restore(save_flags); | |
3193 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3194 | ||
d5e3ed66 JDB |
3195 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3196 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3197 | |
d5e3ed66 | 3198 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3199 | return ptr; |
3200 | } | |
3201 | ||
3202 | static __always_inline void * | |
3203 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3204 | { | |
3205 | void *objp; | |
3206 | ||
2ad654bc | 3207 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3208 | objp = alternate_node_alloc(cache, flags); |
3209 | if (objp) | |
3210 | goto out; | |
3211 | } | |
3212 | objp = ____cache_alloc(cache, flags); | |
3213 | ||
3214 | /* | |
3215 | * We may just have run out of memory on the local node. | |
3216 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3217 | */ | |
7d6e6d09 LS |
3218 | if (!objp) |
3219 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3220 | |
3221 | out: | |
3222 | return objp; | |
3223 | } | |
3224 | #else | |
3225 | ||
3226 | static __always_inline void * | |
3227 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3228 | { | |
3229 | return ____cache_alloc(cachep, flags); | |
3230 | } | |
3231 | ||
3232 | #endif /* CONFIG_NUMA */ | |
3233 | ||
3234 | static __always_inline void * | |
48356303 | 3235 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3236 | { |
3237 | unsigned long save_flags; | |
3238 | void *objp; | |
3239 | ||
dcce284a | 3240 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3241 | cachep = slab_pre_alloc_hook(cachep, flags); |
3242 | if (unlikely(!cachep)) | |
824ebef1 AM |
3243 | return NULL; |
3244 | ||
8c8cc2c1 PE |
3245 | cache_alloc_debugcheck_before(cachep, flags); |
3246 | local_irq_save(save_flags); | |
3247 | objp = __do_cache_alloc(cachep, flags); | |
3248 | local_irq_restore(save_flags); | |
3249 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3250 | prefetchw(objp); | |
3251 | ||
d5e3ed66 JDB |
3252 | if (unlikely(flags & __GFP_ZERO) && objp) |
3253 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3254 | |
d5e3ed66 | 3255 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3256 | return objp; |
3257 | } | |
e498be7d CL |
3258 | |
3259 | /* | |
5f0985bb | 3260 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3261 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3262 | */ |
97654dfa JK |
3263 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3264 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3265 | { |
3266 | int i; | |
25c063fb | 3267 | struct kmem_cache_node *n = get_node(cachep, node); |
1da177e4 LT |
3268 | |
3269 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3270 | void *objp; |
8456a648 | 3271 | struct page *page; |
1da177e4 | 3272 | |
072bb0aa MG |
3273 | objp = objpp[i]; |
3274 | ||
8456a648 | 3275 | page = virt_to_head_page(objp); |
8456a648 | 3276 | list_del(&page->lru); |
ff69416e | 3277 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3278 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3279 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3280 | n->free_objects++; |
1da177e4 LT |
3281 | |
3282 | /* fixup slab chains */ | |
8456a648 | 3283 | if (page->active == 0) { |
ce8eb6c4 CL |
3284 | if (n->free_objects > n->free_limit) { |
3285 | n->free_objects -= cachep->num; | |
97654dfa | 3286 | list_add_tail(&page->lru, list); |
1da177e4 | 3287 | } else { |
8456a648 | 3288 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3289 | } |
3290 | } else { | |
3291 | /* Unconditionally move a slab to the end of the | |
3292 | * partial list on free - maximum time for the | |
3293 | * other objects to be freed, too. | |
3294 | */ | |
8456a648 | 3295 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3296 | } |
3297 | } | |
3298 | } | |
3299 | ||
343e0d7a | 3300 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3301 | { |
3302 | int batchcount; | |
ce8eb6c4 | 3303 | struct kmem_cache_node *n; |
7d6e6d09 | 3304 | int node = numa_mem_id(); |
97654dfa | 3305 | LIST_HEAD(list); |
1da177e4 LT |
3306 | |
3307 | batchcount = ac->batchcount; | |
260b61dd | 3308 | |
1da177e4 | 3309 | check_irq_off(); |
18bf8541 | 3310 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3311 | spin_lock(&n->list_lock); |
3312 | if (n->shared) { | |
3313 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3314 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3315 | if (max) { |
3316 | if (batchcount > max) | |
3317 | batchcount = max; | |
e498be7d | 3318 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3319 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3320 | shared_array->avail += batchcount; |
3321 | goto free_done; | |
3322 | } | |
3323 | } | |
3324 | ||
97654dfa | 3325 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3326 | free_done: |
1da177e4 LT |
3327 | #if STATS |
3328 | { | |
3329 | int i = 0; | |
73c0219d | 3330 | struct page *page; |
1da177e4 | 3331 | |
73c0219d | 3332 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3333 | BUG_ON(page->active); |
1da177e4 LT |
3334 | |
3335 | i++; | |
1da177e4 LT |
3336 | } |
3337 | STATS_SET_FREEABLE(cachep, i); | |
3338 | } | |
3339 | #endif | |
ce8eb6c4 | 3340 | spin_unlock(&n->list_lock); |
97654dfa | 3341 | slabs_destroy(cachep, &list); |
1da177e4 | 3342 | ac->avail -= batchcount; |
a737b3e2 | 3343 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3344 | } |
3345 | ||
3346 | /* | |
a737b3e2 AM |
3347 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3348 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3349 | */ |
a947eb95 | 3350 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3351 | unsigned long caller) |
1da177e4 | 3352 | { |
9a2dba4b | 3353 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 | 3354 | |
7ed2f9e6 AP |
3355 | kasan_slab_free(cachep, objp); |
3356 | ||
1da177e4 | 3357 | check_irq_off(); |
d5cff635 | 3358 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3359 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3360 | |
8c138bc0 | 3361 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3362 | |
1807a1aa SS |
3363 | /* |
3364 | * Skip calling cache_free_alien() when the platform is not numa. | |
3365 | * This will avoid cache misses that happen while accessing slabp (which | |
3366 | * is per page memory reference) to get nodeid. Instead use a global | |
3367 | * variable to skip the call, which is mostly likely to be present in | |
3368 | * the cache. | |
3369 | */ | |
b6e68bc1 | 3370 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3371 | return; |
3372 | ||
3d880194 | 3373 | if (ac->avail < ac->limit) { |
1da177e4 | 3374 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3375 | } else { |
3376 | STATS_INC_FREEMISS(cachep); | |
3377 | cache_flusharray(cachep, ac); | |
1da177e4 | 3378 | } |
42c8c99c | 3379 | |
f68f8ddd JK |
3380 | if (sk_memalloc_socks()) { |
3381 | struct page *page = virt_to_head_page(objp); | |
3382 | ||
3383 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3384 | cache_free_pfmemalloc(cachep, page, objp); | |
3385 | return; | |
3386 | } | |
3387 | } | |
3388 | ||
3389 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3390 | } |
3391 | ||
3392 | /** | |
3393 | * kmem_cache_alloc - Allocate an object | |
3394 | * @cachep: The cache to allocate from. | |
3395 | * @flags: See kmalloc(). | |
3396 | * | |
3397 | * Allocate an object from this cache. The flags are only relevant | |
3398 | * if the cache has no available objects. | |
3399 | */ | |
343e0d7a | 3400 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3401 | { |
48356303 | 3402 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3403 | |
505f5dcb | 3404 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3405 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3406 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3407 | |
3408 | return ret; | |
1da177e4 LT |
3409 | } |
3410 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3411 | ||
7b0501dd JDB |
3412 | static __always_inline void |
3413 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3414 | size_t size, void **p, unsigned long caller) | |
3415 | { | |
3416 | size_t i; | |
3417 | ||
3418 | for (i = 0; i < size; i++) | |
3419 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3420 | } | |
3421 | ||
865762a8 | 3422 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3423 | void **p) |
484748f0 | 3424 | { |
2a777eac JDB |
3425 | size_t i; |
3426 | ||
3427 | s = slab_pre_alloc_hook(s, flags); | |
3428 | if (!s) | |
3429 | return 0; | |
3430 | ||
3431 | cache_alloc_debugcheck_before(s, flags); | |
3432 | ||
3433 | local_irq_disable(); | |
3434 | for (i = 0; i < size; i++) { | |
3435 | void *objp = __do_cache_alloc(s, flags); | |
3436 | ||
2a777eac JDB |
3437 | if (unlikely(!objp)) |
3438 | goto error; | |
3439 | p[i] = objp; | |
3440 | } | |
3441 | local_irq_enable(); | |
3442 | ||
7b0501dd JDB |
3443 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3444 | ||
2a777eac JDB |
3445 | /* Clear memory outside IRQ disabled section */ |
3446 | if (unlikely(flags & __GFP_ZERO)) | |
3447 | for (i = 0; i < size; i++) | |
3448 | memset(p[i], 0, s->object_size); | |
3449 | ||
3450 | slab_post_alloc_hook(s, flags, size, p); | |
3451 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3452 | return size; | |
3453 | error: | |
3454 | local_irq_enable(); | |
7b0501dd | 3455 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3456 | slab_post_alloc_hook(s, flags, i, p); |
3457 | __kmem_cache_free_bulk(s, i, p); | |
3458 | return 0; | |
484748f0 CL |
3459 | } |
3460 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3461 | ||
0f24f128 | 3462 | #ifdef CONFIG_TRACING |
85beb586 | 3463 | void * |
4052147c | 3464 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3465 | { |
85beb586 SR |
3466 | void *ret; |
3467 | ||
48356303 | 3468 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3469 | |
505f5dcb | 3470 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3471 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3472 | size, cachep->size, flags); |
85beb586 | 3473 | return ret; |
36555751 | 3474 | } |
85beb586 | 3475 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3476 | #endif |
3477 | ||
1da177e4 | 3478 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3479 | /** |
3480 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3481 | * @cachep: The cache to allocate from. | |
3482 | * @flags: See kmalloc(). | |
3483 | * @nodeid: node number of the target node. | |
3484 | * | |
3485 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3486 | * node, which can improve the performance for cpu bound structures. | |
3487 | * | |
3488 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3489 | */ | |
8b98c169 CH |
3490 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3491 | { | |
48356303 | 3492 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3493 | |
505f5dcb | 3494 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3495 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3496 | cachep->object_size, cachep->size, |
ca2b84cb | 3497 | flags, nodeid); |
36555751 EGM |
3498 | |
3499 | return ret; | |
8b98c169 | 3500 | } |
1da177e4 LT |
3501 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3502 | ||
0f24f128 | 3503 | #ifdef CONFIG_TRACING |
4052147c | 3504 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3505 | gfp_t flags, |
4052147c EG |
3506 | int nodeid, |
3507 | size_t size) | |
36555751 | 3508 | { |
85beb586 SR |
3509 | void *ret; |
3510 | ||
592f4145 | 3511 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3512 | |
3513 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3514 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3515 | size, cachep->size, |
85beb586 SR |
3516 | flags, nodeid); |
3517 | return ret; | |
36555751 | 3518 | } |
85beb586 | 3519 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3520 | #endif |
3521 | ||
8b98c169 | 3522 | static __always_inline void * |
7c0cb9c6 | 3523 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3524 | { |
343e0d7a | 3525 | struct kmem_cache *cachep; |
7ed2f9e6 | 3526 | void *ret; |
97e2bde4 | 3527 | |
2c59dd65 | 3528 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3529 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3530 | return cachep; | |
7ed2f9e6 | 3531 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3532 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3533 | |
3534 | return ret; | |
97e2bde4 | 3535 | } |
8b98c169 | 3536 | |
8b98c169 CH |
3537 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3538 | { | |
7c0cb9c6 | 3539 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3540 | } |
dbe5e69d | 3541 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3542 | |
3543 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3544 | int node, unsigned long caller) |
8b98c169 | 3545 | { |
7c0cb9c6 | 3546 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3547 | } |
3548 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3549 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3550 | |
3551 | /** | |
800590f5 | 3552 | * __do_kmalloc - allocate memory |
1da177e4 | 3553 | * @size: how many bytes of memory are required. |
800590f5 | 3554 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3555 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3556 | */ |
7fd6b141 | 3557 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3558 | unsigned long caller) |
1da177e4 | 3559 | { |
343e0d7a | 3560 | struct kmem_cache *cachep; |
36555751 | 3561 | void *ret; |
1da177e4 | 3562 | |
2c59dd65 | 3563 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3564 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3565 | return cachep; | |
48356303 | 3566 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3567 | |
505f5dcb | 3568 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3569 | trace_kmalloc(caller, ret, |
3b0efdfa | 3570 | size, cachep->size, flags); |
36555751 EGM |
3571 | |
3572 | return ret; | |
7fd6b141 PE |
3573 | } |
3574 | ||
7fd6b141 PE |
3575 | void *__kmalloc(size_t size, gfp_t flags) |
3576 | { | |
7c0cb9c6 | 3577 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3578 | } |
3579 | EXPORT_SYMBOL(__kmalloc); | |
3580 | ||
ce71e27c | 3581 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3582 | { |
7c0cb9c6 | 3583 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3584 | } |
3585 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3586 | |
1da177e4 LT |
3587 | /** |
3588 | * kmem_cache_free - Deallocate an object | |
3589 | * @cachep: The cache the allocation was from. | |
3590 | * @objp: The previously allocated object. | |
3591 | * | |
3592 | * Free an object which was previously allocated from this | |
3593 | * cache. | |
3594 | */ | |
343e0d7a | 3595 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3596 | { |
3597 | unsigned long flags; | |
b9ce5ef4 GC |
3598 | cachep = cache_from_obj(cachep, objp); |
3599 | if (!cachep) | |
3600 | return; | |
1da177e4 LT |
3601 | |
3602 | local_irq_save(flags); | |
d97d476b | 3603 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3604 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3605 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3606 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3607 | local_irq_restore(flags); |
36555751 | 3608 | |
ca2b84cb | 3609 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3610 | } |
3611 | EXPORT_SYMBOL(kmem_cache_free); | |
3612 | ||
e6cdb58d JDB |
3613 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3614 | { | |
3615 | struct kmem_cache *s; | |
3616 | size_t i; | |
3617 | ||
3618 | local_irq_disable(); | |
3619 | for (i = 0; i < size; i++) { | |
3620 | void *objp = p[i]; | |
3621 | ||
ca257195 JDB |
3622 | if (!orig_s) /* called via kfree_bulk */ |
3623 | s = virt_to_cache(objp); | |
3624 | else | |
3625 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3626 | |
3627 | debug_check_no_locks_freed(objp, s->object_size); | |
3628 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3629 | debug_check_no_obj_freed(objp, s->object_size); | |
3630 | ||
3631 | __cache_free(s, objp, _RET_IP_); | |
3632 | } | |
3633 | local_irq_enable(); | |
3634 | ||
3635 | /* FIXME: add tracing */ | |
3636 | } | |
3637 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3638 | ||
1da177e4 LT |
3639 | /** |
3640 | * kfree - free previously allocated memory | |
3641 | * @objp: pointer returned by kmalloc. | |
3642 | * | |
80e93eff PE |
3643 | * If @objp is NULL, no operation is performed. |
3644 | * | |
1da177e4 LT |
3645 | * Don't free memory not originally allocated by kmalloc() |
3646 | * or you will run into trouble. | |
3647 | */ | |
3648 | void kfree(const void *objp) | |
3649 | { | |
343e0d7a | 3650 | struct kmem_cache *c; |
1da177e4 LT |
3651 | unsigned long flags; |
3652 | ||
2121db74 PE |
3653 | trace_kfree(_RET_IP_, objp); |
3654 | ||
6cb8f913 | 3655 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3656 | return; |
3657 | local_irq_save(flags); | |
3658 | kfree_debugcheck(objp); | |
6ed5eb22 | 3659 | c = virt_to_cache(objp); |
8c138bc0 CL |
3660 | debug_check_no_locks_freed(objp, c->object_size); |
3661 | ||
3662 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3663 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3664 | local_irq_restore(flags); |
3665 | } | |
3666 | EXPORT_SYMBOL(kfree); | |
3667 | ||
e498be7d | 3668 | /* |
ce8eb6c4 | 3669 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3670 | */ |
5f0985bb | 3671 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3672 | { |
3673 | int node; | |
ce8eb6c4 | 3674 | struct kmem_cache_node *n; |
cafeb02e | 3675 | struct array_cache *new_shared; |
c8522a3a | 3676 | struct alien_cache **new_alien = NULL; |
e498be7d | 3677 | |
9c09a95c | 3678 | for_each_online_node(node) { |
cafeb02e | 3679 | |
b455def2 L |
3680 | if (use_alien_caches) { |
3681 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
3682 | if (!new_alien) | |
3683 | goto fail; | |
3684 | } | |
cafeb02e | 3685 | |
63109846 ED |
3686 | new_shared = NULL; |
3687 | if (cachep->shared) { | |
3688 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3689 | cachep->shared*cachep->batchcount, |
83b519e8 | 3690 | 0xbaadf00d, gfp); |
63109846 ED |
3691 | if (!new_shared) { |
3692 | free_alien_cache(new_alien); | |
3693 | goto fail; | |
3694 | } | |
0718dc2a | 3695 | } |
cafeb02e | 3696 | |
18bf8541 | 3697 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3698 | if (n) { |
3699 | struct array_cache *shared = n->shared; | |
97654dfa | 3700 | LIST_HEAD(list); |
cafeb02e | 3701 | |
ce8eb6c4 | 3702 | spin_lock_irq(&n->list_lock); |
e498be7d | 3703 | |
cafeb02e | 3704 | if (shared) |
0718dc2a | 3705 | free_block(cachep, shared->entry, |
97654dfa | 3706 | shared->avail, node, &list); |
e498be7d | 3707 | |
ce8eb6c4 CL |
3708 | n->shared = new_shared; |
3709 | if (!n->alien) { | |
3710 | n->alien = new_alien; | |
e498be7d CL |
3711 | new_alien = NULL; |
3712 | } | |
ce8eb6c4 | 3713 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3714 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3715 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3716 | slabs_destroy(cachep, &list); |
cafeb02e | 3717 | kfree(shared); |
e498be7d CL |
3718 | free_alien_cache(new_alien); |
3719 | continue; | |
3720 | } | |
ce8eb6c4 CL |
3721 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3722 | if (!n) { | |
0718dc2a CL |
3723 | free_alien_cache(new_alien); |
3724 | kfree(new_shared); | |
e498be7d | 3725 | goto fail; |
0718dc2a | 3726 | } |
e498be7d | 3727 | |
ce8eb6c4 | 3728 | kmem_cache_node_init(n); |
5f0985bb JZ |
3729 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3730 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3731 | n->shared = new_shared; |
3732 | n->alien = new_alien; | |
3733 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3734 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3735 | cachep->node[node] = n; |
e498be7d | 3736 | } |
cafeb02e | 3737 | return 0; |
0718dc2a | 3738 | |
a737b3e2 | 3739 | fail: |
3b0efdfa | 3740 | if (!cachep->list.next) { |
0718dc2a CL |
3741 | /* Cache is not active yet. Roll back what we did */ |
3742 | node--; | |
3743 | while (node >= 0) { | |
18bf8541 CL |
3744 | n = get_node(cachep, node); |
3745 | if (n) { | |
ce8eb6c4 CL |
3746 | kfree(n->shared); |
3747 | free_alien_cache(n->alien); | |
3748 | kfree(n); | |
6a67368c | 3749 | cachep->node[node] = NULL; |
0718dc2a CL |
3750 | } |
3751 | node--; | |
3752 | } | |
3753 | } | |
cafeb02e | 3754 | return -ENOMEM; |
e498be7d CL |
3755 | } |
3756 | ||
18004c5d | 3757 | /* Always called with the slab_mutex held */ |
943a451a | 3758 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3759 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3760 | { |
bf0dea23 JK |
3761 | struct array_cache __percpu *cpu_cache, *prev; |
3762 | int cpu; | |
1da177e4 | 3763 | |
bf0dea23 JK |
3764 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3765 | if (!cpu_cache) | |
d2e7b7d0 SS |
3766 | return -ENOMEM; |
3767 | ||
bf0dea23 JK |
3768 | prev = cachep->cpu_cache; |
3769 | cachep->cpu_cache = cpu_cache; | |
3770 | kick_all_cpus_sync(); | |
e498be7d | 3771 | |
1da177e4 | 3772 | check_irq_on(); |
1da177e4 LT |
3773 | cachep->batchcount = batchcount; |
3774 | cachep->limit = limit; | |
e498be7d | 3775 | cachep->shared = shared; |
1da177e4 | 3776 | |
bf0dea23 JK |
3777 | if (!prev) |
3778 | goto alloc_node; | |
3779 | ||
3780 | for_each_online_cpu(cpu) { | |
97654dfa | 3781 | LIST_HEAD(list); |
18bf8541 CL |
3782 | int node; |
3783 | struct kmem_cache_node *n; | |
bf0dea23 | 3784 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3785 | |
bf0dea23 | 3786 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3787 | n = get_node(cachep, node); |
3788 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3789 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3790 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3791 | slabs_destroy(cachep, &list); |
1da177e4 | 3792 | } |
bf0dea23 JK |
3793 | free_percpu(prev); |
3794 | ||
3795 | alloc_node: | |
5f0985bb | 3796 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3797 | } |
3798 | ||
943a451a GC |
3799 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3800 | int batchcount, int shared, gfp_t gfp) | |
3801 | { | |
3802 | int ret; | |
426589f5 | 3803 | struct kmem_cache *c; |
943a451a GC |
3804 | |
3805 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3806 | ||
3807 | if (slab_state < FULL) | |
3808 | return ret; | |
3809 | ||
3810 | if ((ret < 0) || !is_root_cache(cachep)) | |
3811 | return ret; | |
3812 | ||
426589f5 VD |
3813 | lockdep_assert_held(&slab_mutex); |
3814 | for_each_memcg_cache(c, cachep) { | |
3815 | /* return value determined by the root cache only */ | |
3816 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3817 | } |
3818 | ||
3819 | return ret; | |
3820 | } | |
3821 | ||
18004c5d | 3822 | /* Called with slab_mutex held always */ |
83b519e8 | 3823 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3824 | { |
3825 | int err; | |
943a451a GC |
3826 | int limit = 0; |
3827 | int shared = 0; | |
3828 | int batchcount = 0; | |
3829 | ||
3830 | if (!is_root_cache(cachep)) { | |
3831 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3832 | limit = root->limit; | |
3833 | shared = root->shared; | |
3834 | batchcount = root->batchcount; | |
3835 | } | |
1da177e4 | 3836 | |
943a451a GC |
3837 | if (limit && shared && batchcount) |
3838 | goto skip_setup; | |
a737b3e2 AM |
3839 | /* |
3840 | * The head array serves three purposes: | |
1da177e4 LT |
3841 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3842 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3843 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3844 | * bufctl chains: array operations are cheaper. |
3845 | * The numbers are guessed, we should auto-tune as described by | |
3846 | * Bonwick. | |
3847 | */ | |
3b0efdfa | 3848 | if (cachep->size > 131072) |
1da177e4 | 3849 | limit = 1; |
3b0efdfa | 3850 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3851 | limit = 8; |
3b0efdfa | 3852 | else if (cachep->size > 1024) |
1da177e4 | 3853 | limit = 24; |
3b0efdfa | 3854 | else if (cachep->size > 256) |
1da177e4 LT |
3855 | limit = 54; |
3856 | else | |
3857 | limit = 120; | |
3858 | ||
a737b3e2 AM |
3859 | /* |
3860 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3861 | * allocation behaviour: Most allocs on one cpu, most free operations |
3862 | * on another cpu. For these cases, an efficient object passing between | |
3863 | * cpus is necessary. This is provided by a shared array. The array | |
3864 | * replaces Bonwick's magazine layer. | |
3865 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3866 | * to a larger limit. Thus disabled by default. | |
3867 | */ | |
3868 | shared = 0; | |
3b0efdfa | 3869 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3870 | shared = 8; |
1da177e4 LT |
3871 | |
3872 | #if DEBUG | |
a737b3e2 AM |
3873 | /* |
3874 | * With debugging enabled, large batchcount lead to excessively long | |
3875 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3876 | */ |
3877 | if (limit > 32) | |
3878 | limit = 32; | |
3879 | #endif | |
943a451a GC |
3880 | batchcount = (limit + 1) / 2; |
3881 | skip_setup: | |
3882 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 | 3883 | if (err) |
1170532b | 3884 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3885 | cachep->name, -err); |
2ed3a4ef | 3886 | return err; |
1da177e4 LT |
3887 | } |
3888 | ||
1b55253a | 3889 | /* |
ce8eb6c4 CL |
3890 | * Drain an array if it contains any elements taking the node lock only if |
3891 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3892 | * if drain_array() is used on the shared array. |
1b55253a | 3893 | */ |
ce8eb6c4 | 3894 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 3895 | struct array_cache *ac, int node) |
1da177e4 | 3896 | { |
97654dfa | 3897 | LIST_HEAD(list); |
18726ca8 JK |
3898 | |
3899 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
3900 | check_mutex_acquired(); | |
1da177e4 | 3901 | |
1b55253a CL |
3902 | if (!ac || !ac->avail) |
3903 | return; | |
18726ca8 JK |
3904 | |
3905 | if (ac->touched) { | |
1da177e4 | 3906 | ac->touched = 0; |
18726ca8 | 3907 | return; |
1da177e4 | 3908 | } |
18726ca8 JK |
3909 | |
3910 | spin_lock_irq(&n->list_lock); | |
3911 | drain_array_locked(cachep, ac, node, false, &list); | |
3912 | spin_unlock_irq(&n->list_lock); | |
3913 | ||
3914 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
3915 | } |
3916 | ||
3917 | /** | |
3918 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3919 | * @w: work descriptor |
1da177e4 LT |
3920 | * |
3921 | * Called from workqueue/eventd every few seconds. | |
3922 | * Purpose: | |
3923 | * - clear the per-cpu caches for this CPU. | |
3924 | * - return freeable pages to the main free memory pool. | |
3925 | * | |
a737b3e2 AM |
3926 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3927 | * again on the next iteration. | |
1da177e4 | 3928 | */ |
7c5cae36 | 3929 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3930 | { |
7a7c381d | 3931 | struct kmem_cache *searchp; |
ce8eb6c4 | 3932 | struct kmem_cache_node *n; |
7d6e6d09 | 3933 | int node = numa_mem_id(); |
bf6aede7 | 3934 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3935 | |
18004c5d | 3936 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3937 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3938 | goto out; |
1da177e4 | 3939 | |
18004c5d | 3940 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3941 | check_irq_on(); |
3942 | ||
35386e3b | 3943 | /* |
ce8eb6c4 | 3944 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
3945 | * have established with reasonable certainty that |
3946 | * we can do some work if the lock was obtained. | |
3947 | */ | |
18bf8541 | 3948 | n = get_node(searchp, node); |
35386e3b | 3949 | |
ce8eb6c4 | 3950 | reap_alien(searchp, n); |
1da177e4 | 3951 | |
18726ca8 | 3952 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 3953 | |
35386e3b CL |
3954 | /* |
3955 | * These are racy checks but it does not matter | |
3956 | * if we skip one check or scan twice. | |
3957 | */ | |
ce8eb6c4 | 3958 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 3959 | goto next; |
1da177e4 | 3960 | |
5f0985bb | 3961 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 3962 | |
18726ca8 | 3963 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 3964 | |
ce8eb6c4 CL |
3965 | if (n->free_touched) |
3966 | n->free_touched = 0; | |
ed11d9eb CL |
3967 | else { |
3968 | int freed; | |
1da177e4 | 3969 | |
ce8eb6c4 | 3970 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
3971 | 5 * searchp->num - 1) / (5 * searchp->num)); |
3972 | STATS_ADD_REAPED(searchp, freed); | |
3973 | } | |
35386e3b | 3974 | next: |
1da177e4 LT |
3975 | cond_resched(); |
3976 | } | |
3977 | check_irq_on(); | |
18004c5d | 3978 | mutex_unlock(&slab_mutex); |
8fce4d8e | 3979 | next_reap_node(); |
7c5cae36 | 3980 | out: |
a737b3e2 | 3981 | /* Set up the next iteration */ |
5f0985bb | 3982 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
3983 | } |
3984 | ||
158a9624 | 3985 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 3986 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 3987 | { |
8456a648 | 3988 | struct page *page; |
b28a02de PE |
3989 | unsigned long active_objs; |
3990 | unsigned long num_objs; | |
3991 | unsigned long active_slabs = 0; | |
3992 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3993 | const char *name; |
1da177e4 | 3994 | char *error = NULL; |
e498be7d | 3995 | int node; |
ce8eb6c4 | 3996 | struct kmem_cache_node *n; |
1da177e4 | 3997 | |
1da177e4 LT |
3998 | active_objs = 0; |
3999 | num_slabs = 0; | |
18bf8541 | 4000 | for_each_kmem_cache_node(cachep, node, n) { |
e498be7d | 4001 | |
ca3b9b91 | 4002 | check_irq_on(); |
ce8eb6c4 | 4003 | spin_lock_irq(&n->list_lock); |
e498be7d | 4004 | |
8456a648 JK |
4005 | list_for_each_entry(page, &n->slabs_full, lru) { |
4006 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4007 | error = "slabs_full accounting error"; |
4008 | active_objs += cachep->num; | |
4009 | active_slabs++; | |
4010 | } | |
8456a648 JK |
4011 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4012 | if (page->active == cachep->num && !error) | |
106a74e1 | 4013 | error = "slabs_partial accounting error"; |
8456a648 | 4014 | if (!page->active && !error) |
106a74e1 | 4015 | error = "slabs_partial accounting error"; |
8456a648 | 4016 | active_objs += page->active; |
e498be7d CL |
4017 | active_slabs++; |
4018 | } | |
8456a648 JK |
4019 | list_for_each_entry(page, &n->slabs_free, lru) { |
4020 | if (page->active && !error) | |
106a74e1 | 4021 | error = "slabs_free accounting error"; |
e498be7d CL |
4022 | num_slabs++; |
4023 | } | |
ce8eb6c4 CL |
4024 | free_objects += n->free_objects; |
4025 | if (n->shared) | |
4026 | shared_avail += n->shared->avail; | |
e498be7d | 4027 | |
ce8eb6c4 | 4028 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4029 | } |
b28a02de PE |
4030 | num_slabs += active_slabs; |
4031 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4032 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4033 | error = "free_objects accounting error"; |
4034 | ||
b28a02de | 4035 | name = cachep->name; |
1da177e4 | 4036 | if (error) |
1170532b | 4037 | pr_err("slab: cache %s error: %s\n", name, error); |
1da177e4 | 4038 | |
0d7561c6 GC |
4039 | sinfo->active_objs = active_objs; |
4040 | sinfo->num_objs = num_objs; | |
4041 | sinfo->active_slabs = active_slabs; | |
4042 | sinfo->num_slabs = num_slabs; | |
4043 | sinfo->shared_avail = shared_avail; | |
4044 | sinfo->limit = cachep->limit; | |
4045 | sinfo->batchcount = cachep->batchcount; | |
4046 | sinfo->shared = cachep->shared; | |
4047 | sinfo->objects_per_slab = cachep->num; | |
4048 | sinfo->cache_order = cachep->gfporder; | |
4049 | } | |
4050 | ||
4051 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4052 | { | |
1da177e4 | 4053 | #if STATS |
ce8eb6c4 | 4054 | { /* node stats */ |
1da177e4 LT |
4055 | unsigned long high = cachep->high_mark; |
4056 | unsigned long allocs = cachep->num_allocations; | |
4057 | unsigned long grown = cachep->grown; | |
4058 | unsigned long reaped = cachep->reaped; | |
4059 | unsigned long errors = cachep->errors; | |
4060 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4061 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4062 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4063 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4064 | |
756a025f | 4065 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4066 | allocs, high, grown, |
4067 | reaped, errors, max_freeable, node_allocs, | |
4068 | node_frees, overflows); | |
1da177e4 LT |
4069 | } |
4070 | /* cpu stats */ | |
4071 | { | |
4072 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4073 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4074 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4075 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4076 | ||
4077 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4078 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4079 | } |
4080 | #endif | |
1da177e4 LT |
4081 | } |
4082 | ||
1da177e4 LT |
4083 | #define MAX_SLABINFO_WRITE 128 |
4084 | /** | |
4085 | * slabinfo_write - Tuning for the slab allocator | |
4086 | * @file: unused | |
4087 | * @buffer: user buffer | |
4088 | * @count: data length | |
4089 | * @ppos: unused | |
4090 | */ | |
b7454ad3 | 4091 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4092 | size_t count, loff_t *ppos) |
1da177e4 | 4093 | { |
b28a02de | 4094 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4095 | int limit, batchcount, shared, res; |
7a7c381d | 4096 | struct kmem_cache *cachep; |
b28a02de | 4097 | |
1da177e4 LT |
4098 | if (count > MAX_SLABINFO_WRITE) |
4099 | return -EINVAL; | |
4100 | if (copy_from_user(&kbuf, buffer, count)) | |
4101 | return -EFAULT; | |
b28a02de | 4102 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4103 | |
4104 | tmp = strchr(kbuf, ' '); | |
4105 | if (!tmp) | |
4106 | return -EINVAL; | |
4107 | *tmp = '\0'; | |
4108 | tmp++; | |
4109 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4110 | return -EINVAL; | |
4111 | ||
4112 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4113 | mutex_lock(&slab_mutex); |
1da177e4 | 4114 | res = -EINVAL; |
18004c5d | 4115 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4116 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4117 | if (limit < 1 || batchcount < 1 || |
4118 | batchcount > limit || shared < 0) { | |
e498be7d | 4119 | res = 0; |
1da177e4 | 4120 | } else { |
e498be7d | 4121 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4122 | batchcount, shared, |
4123 | GFP_KERNEL); | |
1da177e4 LT |
4124 | } |
4125 | break; | |
4126 | } | |
4127 | } | |
18004c5d | 4128 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4129 | if (res >= 0) |
4130 | res = count; | |
4131 | return res; | |
4132 | } | |
871751e2 AV |
4133 | |
4134 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4135 | ||
871751e2 AV |
4136 | static inline int add_caller(unsigned long *n, unsigned long v) |
4137 | { | |
4138 | unsigned long *p; | |
4139 | int l; | |
4140 | if (!v) | |
4141 | return 1; | |
4142 | l = n[1]; | |
4143 | p = n + 2; | |
4144 | while (l) { | |
4145 | int i = l/2; | |
4146 | unsigned long *q = p + 2 * i; | |
4147 | if (*q == v) { | |
4148 | q[1]++; | |
4149 | return 1; | |
4150 | } | |
4151 | if (*q > v) { | |
4152 | l = i; | |
4153 | } else { | |
4154 | p = q + 2; | |
4155 | l -= i + 1; | |
4156 | } | |
4157 | } | |
4158 | if (++n[1] == n[0]) | |
4159 | return 0; | |
4160 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4161 | p[0] = v; | |
4162 | p[1] = 1; | |
4163 | return 1; | |
4164 | } | |
4165 | ||
8456a648 JK |
4166 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4167 | struct page *page) | |
871751e2 AV |
4168 | { |
4169 | void *p; | |
d31676df JK |
4170 | int i, j; |
4171 | unsigned long v; | |
b1cb0982 | 4172 | |
871751e2 AV |
4173 | if (n[0] == n[1]) |
4174 | return; | |
8456a648 | 4175 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4176 | bool active = true; |
4177 | ||
4178 | for (j = page->active; j < c->num; j++) { | |
4179 | if (get_free_obj(page, j) == i) { | |
4180 | active = false; | |
4181 | break; | |
4182 | } | |
4183 | } | |
4184 | ||
4185 | if (!active) | |
871751e2 | 4186 | continue; |
b1cb0982 | 4187 | |
d31676df JK |
4188 | /* |
4189 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4190 | * mapping is established when actual object allocation and | |
4191 | * we could mistakenly access the unmapped object in the cpu | |
4192 | * cache. | |
4193 | */ | |
4194 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4195 | continue; | |
4196 | ||
4197 | if (!add_caller(n, v)) | |
871751e2 AV |
4198 | return; |
4199 | } | |
4200 | } | |
4201 | ||
4202 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4203 | { | |
4204 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4205 | unsigned long offset, size; |
9281acea | 4206 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4207 | |
a5c43dae | 4208 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4209 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4210 | if (modname[0]) |
871751e2 AV |
4211 | seq_printf(m, " [%s]", modname); |
4212 | return; | |
4213 | } | |
4214 | #endif | |
4215 | seq_printf(m, "%p", (void *)address); | |
4216 | } | |
4217 | ||
4218 | static int leaks_show(struct seq_file *m, void *p) | |
4219 | { | |
0672aa7c | 4220 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4221 | struct page *page; |
ce8eb6c4 | 4222 | struct kmem_cache_node *n; |
871751e2 | 4223 | const char *name; |
db845067 | 4224 | unsigned long *x = m->private; |
871751e2 AV |
4225 | int node; |
4226 | int i; | |
4227 | ||
4228 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4229 | return 0; | |
4230 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4231 | return 0; | |
4232 | ||
d31676df JK |
4233 | /* |
4234 | * Set store_user_clean and start to grab stored user information | |
4235 | * for all objects on this cache. If some alloc/free requests comes | |
4236 | * during the processing, information would be wrong so restart | |
4237 | * whole processing. | |
4238 | */ | |
4239 | do { | |
4240 | set_store_user_clean(cachep); | |
4241 | drain_cpu_caches(cachep); | |
4242 | ||
4243 | x[1] = 0; | |
871751e2 | 4244 | |
d31676df | 4245 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4246 | |
d31676df JK |
4247 | check_irq_on(); |
4248 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4249 | |
d31676df JK |
4250 | list_for_each_entry(page, &n->slabs_full, lru) |
4251 | handle_slab(x, cachep, page); | |
4252 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4253 | handle_slab(x, cachep, page); | |
4254 | spin_unlock_irq(&n->list_lock); | |
4255 | } | |
4256 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4257 | |
871751e2 | 4258 | name = cachep->name; |
db845067 | 4259 | if (x[0] == x[1]) { |
871751e2 | 4260 | /* Increase the buffer size */ |
18004c5d | 4261 | mutex_unlock(&slab_mutex); |
db845067 | 4262 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4263 | if (!m->private) { |
4264 | /* Too bad, we are really out */ | |
db845067 | 4265 | m->private = x; |
18004c5d | 4266 | mutex_lock(&slab_mutex); |
871751e2 AV |
4267 | return -ENOMEM; |
4268 | } | |
db845067 CL |
4269 | *(unsigned long *)m->private = x[0] * 2; |
4270 | kfree(x); | |
18004c5d | 4271 | mutex_lock(&slab_mutex); |
871751e2 AV |
4272 | /* Now make sure this entry will be retried */ |
4273 | m->count = m->size; | |
4274 | return 0; | |
4275 | } | |
db845067 CL |
4276 | for (i = 0; i < x[1]; i++) { |
4277 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4278 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4279 | seq_putc(m, '\n'); |
4280 | } | |
d2e7b7d0 | 4281 | |
871751e2 AV |
4282 | return 0; |
4283 | } | |
4284 | ||
a0ec95a8 | 4285 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4286 | .start = slab_start, |
276a2439 WL |
4287 | .next = slab_next, |
4288 | .stop = slab_stop, | |
871751e2 AV |
4289 | .show = leaks_show, |
4290 | }; | |
a0ec95a8 AD |
4291 | |
4292 | static int slabstats_open(struct inode *inode, struct file *file) | |
4293 | { | |
b208ce32 RJ |
4294 | unsigned long *n; |
4295 | ||
4296 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4297 | if (!n) | |
4298 | return -ENOMEM; | |
4299 | ||
4300 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4301 | ||
4302 | return 0; | |
a0ec95a8 AD |
4303 | } |
4304 | ||
4305 | static const struct file_operations proc_slabstats_operations = { | |
4306 | .open = slabstats_open, | |
4307 | .read = seq_read, | |
4308 | .llseek = seq_lseek, | |
4309 | .release = seq_release_private, | |
4310 | }; | |
4311 | #endif | |
4312 | ||
4313 | static int __init slab_proc_init(void) | |
4314 | { | |
4315 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4316 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4317 | #endif |
a0ec95a8 AD |
4318 | return 0; |
4319 | } | |
4320 | module_init(slab_proc_init); | |
1da177e4 LT |
4321 | #endif |
4322 | ||
00e145b6 MS |
4323 | /** |
4324 | * ksize - get the actual amount of memory allocated for a given object | |
4325 | * @objp: Pointer to the object | |
4326 | * | |
4327 | * kmalloc may internally round up allocations and return more memory | |
4328 | * than requested. ksize() can be used to determine the actual amount of | |
4329 | * memory allocated. The caller may use this additional memory, even though | |
4330 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4331 | * The caller must guarantee that objp points to a valid object previously | |
4332 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4333 | * must not be freed during the duration of the call. | |
4334 | */ | |
fd76bab2 | 4335 | size_t ksize(const void *objp) |
1da177e4 | 4336 | { |
7ed2f9e6 AP |
4337 | size_t size; |
4338 | ||
ef8b4520 CL |
4339 | BUG_ON(!objp); |
4340 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4341 | return 0; |
1da177e4 | 4342 | |
7ed2f9e6 AP |
4343 | size = virt_to_cache(objp)->object_size; |
4344 | /* We assume that ksize callers could use the whole allocated area, | |
4345 | * so we need to unpoison this area. | |
4346 | */ | |
505f5dcb | 4347 | kasan_krealloc(objp, size, GFP_NOWAIT); |
7ed2f9e6 AP |
4348 | |
4349 | return size; | |
1da177e4 | 4350 | } |
b1aabecd | 4351 | EXPORT_SYMBOL(ksize); |