]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/slab.c | |
4 | * Written by Mark Hemment, 1996/97. | |
5 | * (markhe@nextd.demon.co.uk) | |
6 | * | |
7 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
8 | * | |
9 | * Major cleanup, different bufctl logic, per-cpu arrays | |
10 | * (c) 2000 Manfred Spraul | |
11 | * | |
12 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
13 | * (c) 2002 Manfred Spraul | |
14 | * | |
15 | * An implementation of the Slab Allocator as described in outline in; | |
16 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
17 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
18 | * or with a little more detail in; | |
19 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
20 | * Jeff Bonwick (Sun Microsystems). | |
21 | * Presented at: USENIX Summer 1994 Technical Conference | |
22 | * | |
23 | * The memory is organized in caches, one cache for each object type. | |
24 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
25 | * Each cache consists out of many slabs (they are small (usually one | |
26 | * page long) and always contiguous), and each slab contains multiple | |
27 | * initialized objects. | |
28 | * | |
29 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 30 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
31 | * kmem_cache_free. |
32 | * | |
33 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
34 | * normal). If you need a special memory type, then must create a new | |
35 | * cache for that memory type. | |
36 | * | |
37 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
38 | * full slabs with 0 free objects | |
39 | * partial slabs | |
40 | * empty slabs with no allocated objects | |
41 | * | |
42 | * If partial slabs exist, then new allocations come from these slabs, | |
43 | * otherwise from empty slabs or new slabs are allocated. | |
44 | * | |
45 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
46 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
47 | * | |
48 | * Each cache has a short per-cpu head array, most allocs | |
49 | * and frees go into that array, and if that array overflows, then 1/2 | |
50 | * of the entries in the array are given back into the global cache. | |
51 | * The head array is strictly LIFO and should improve the cache hit rates. | |
52 | * On SMP, it additionally reduces the spinlock operations. | |
53 | * | |
a737b3e2 | 54 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
55 | * it's changed with a smp_call_function(). |
56 | * | |
57 | * SMP synchronization: | |
58 | * constructors and destructors are called without any locking. | |
343e0d7a | 59 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
60 | * are accessed without any locking. |
61 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
62 | * and local interrupts are disabled so slab code is preempt-safe. | |
63 | * The non-constant members are protected with a per-cache irq spinlock. | |
64 | * | |
65 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
66 | * in 2000 - many ideas in the current implementation are derived from | |
67 | * his patch. | |
68 | * | |
69 | * Further notes from the original documentation: | |
70 | * | |
71 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 72 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
73 | * The sem is only needed when accessing/extending the cache-chain, which |
74 | * can never happen inside an interrupt (kmem_cache_create(), | |
75 | * kmem_cache_shrink() and kmem_cache_reap()). | |
76 | * | |
77 | * At present, each engine can be growing a cache. This should be blocked. | |
78 | * | |
e498be7d CL |
79 | * 15 March 2005. NUMA slab allocator. |
80 | * Shai Fultheim <shai@scalex86.org>. | |
81 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
82 | * Alok N Kataria <alokk@calsoftinc.com> | |
83 | * Christoph Lameter <christoph@lameter.com> | |
84 | * | |
85 | * Modified the slab allocator to be node aware on NUMA systems. | |
86 | * Each node has its own list of partial, free and full slabs. | |
87 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 LT |
90 | #include <linux/slab.h> |
91 | #include <linux/mm.h> | |
c9cf5528 | 92 | #include <linux/poison.h> |
1da177e4 LT |
93 | #include <linux/swap.h> |
94 | #include <linux/cache.h> | |
95 | #include <linux/interrupt.h> | |
96 | #include <linux/init.h> | |
97 | #include <linux/compiler.h> | |
101a5001 | 98 | #include <linux/cpuset.h> |
a0ec95a8 | 99 | #include <linux/proc_fs.h> |
1da177e4 LT |
100 | #include <linux/seq_file.h> |
101 | #include <linux/notifier.h> | |
102 | #include <linux/kallsyms.h> | |
103 | #include <linux/cpu.h> | |
104 | #include <linux/sysctl.h> | |
105 | #include <linux/module.h> | |
106 | #include <linux/rcupdate.h> | |
543537bd | 107 | #include <linux/string.h> |
138ae663 | 108 | #include <linux/uaccess.h> |
e498be7d | 109 | #include <linux/nodemask.h> |
d5cff635 | 110 | #include <linux/kmemleak.h> |
dc85da15 | 111 | #include <linux/mempolicy.h> |
fc0abb14 | 112 | #include <linux/mutex.h> |
8a8b6502 | 113 | #include <linux/fault-inject.h> |
e7eebaf6 | 114 | #include <linux/rtmutex.h> |
6a2d7a95 | 115 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 116 | #include <linux/debugobjects.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
3f8c2452 | 119 | #include <linux/sched/task_stack.h> |
1da177e4 | 120 | |
381760ea MG |
121 | #include <net/sock.h> |
122 | ||
1da177e4 LT |
123 | #include <asm/cacheflush.h> |
124 | #include <asm/tlbflush.h> | |
125 | #include <asm/page.h> | |
126 | ||
4dee6b64 SR |
127 | #include <trace/events/kmem.h> |
128 | ||
072bb0aa MG |
129 | #include "internal.h" |
130 | ||
b9ce5ef4 GC |
131 | #include "slab.h" |
132 | ||
1da177e4 | 133 | /* |
50953fe9 | 134 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
135 | * 0 for faster, smaller code (especially in the critical paths). |
136 | * | |
137 | * STATS - 1 to collect stats for /proc/slabinfo. | |
138 | * 0 for faster, smaller code (especially in the critical paths). | |
139 | * | |
140 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
141 | */ | |
142 | ||
143 | #ifdef CONFIG_DEBUG_SLAB | |
144 | #define DEBUG 1 | |
145 | #define STATS 1 | |
146 | #define FORCED_DEBUG 1 | |
147 | #else | |
148 | #define DEBUG 0 | |
149 | #define STATS 0 | |
150 | #define FORCED_DEBUG 0 | |
151 | #endif | |
152 | ||
1da177e4 LT |
153 | /* Shouldn't this be in a header file somewhere? */ |
154 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 155 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 156 | |
1da177e4 LT |
157 | #ifndef ARCH_KMALLOC_FLAGS |
158 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
159 | #endif | |
160 | ||
f315e3fa JK |
161 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
162 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
163 | ||
164 | #if FREELIST_BYTE_INDEX | |
165 | typedef unsigned char freelist_idx_t; | |
166 | #else | |
167 | typedef unsigned short freelist_idx_t; | |
168 | #endif | |
169 | ||
30321c7b | 170 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 171 | |
1da177e4 LT |
172 | /* |
173 | * struct array_cache | |
174 | * | |
1da177e4 LT |
175 | * Purpose: |
176 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
177 | * - reduce the number of linked list operations | |
178 | * - reduce spinlock operations | |
179 | * | |
180 | * The limit is stored in the per-cpu structure to reduce the data cache | |
181 | * footprint. | |
182 | * | |
183 | */ | |
184 | struct array_cache { | |
185 | unsigned int avail; | |
186 | unsigned int limit; | |
187 | unsigned int batchcount; | |
188 | unsigned int touched; | |
bda5b655 | 189 | void *entry[]; /* |
a737b3e2 AM |
190 | * Must have this definition in here for the proper |
191 | * alignment of array_cache. Also simplifies accessing | |
192 | * the entries. | |
a737b3e2 | 193 | */ |
1da177e4 LT |
194 | }; |
195 | ||
c8522a3a JK |
196 | struct alien_cache { |
197 | spinlock_t lock; | |
198 | struct array_cache ac; | |
199 | }; | |
200 | ||
e498be7d CL |
201 | /* |
202 | * Need this for bootstrapping a per node allocator. | |
203 | */ | |
bf0dea23 | 204 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 205 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 206 | #define CACHE_CACHE 0 |
bf0dea23 | 207 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 208 | |
ed11d9eb | 209 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 210 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 211 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
212 | int node, struct list_head *list); |
213 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 214 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 215 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 216 | |
76b342bd JK |
217 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
218 | void **list); | |
219 | static inline void fixup_slab_list(struct kmem_cache *cachep, | |
220 | struct kmem_cache_node *n, struct page *page, | |
221 | void **list); | |
e0a42726 IM |
222 | static int slab_early_init = 1; |
223 | ||
ce8eb6c4 | 224 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 225 | |
ce8eb6c4 | 226 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
227 | { |
228 | INIT_LIST_HEAD(&parent->slabs_full); | |
229 | INIT_LIST_HEAD(&parent->slabs_partial); | |
230 | INIT_LIST_HEAD(&parent->slabs_free); | |
bf00bd34 | 231 | parent->total_slabs = 0; |
f728b0a5 | 232 | parent->free_slabs = 0; |
e498be7d CL |
233 | parent->shared = NULL; |
234 | parent->alien = NULL; | |
2e1217cf | 235 | parent->colour_next = 0; |
e498be7d CL |
236 | spin_lock_init(&parent->list_lock); |
237 | parent->free_objects = 0; | |
238 | parent->free_touched = 0; | |
239 | } | |
240 | ||
a737b3e2 AM |
241 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
242 | do { \ | |
243 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 244 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
245 | } while (0) |
246 | ||
a737b3e2 AM |
247 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
248 | do { \ | |
e498be7d CL |
249 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
250 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
251 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
252 | } while (0) | |
1da177e4 | 253 | |
4fd0b46e AD |
254 | #define CFLGS_OBJFREELIST_SLAB ((slab_flags_t __force)0x40000000U) |
255 | #define CFLGS_OFF_SLAB ((slab_flags_t __force)0x80000000U) | |
b03a017b | 256 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
257 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
258 | ||
259 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
260 | /* |
261 | * Optimization question: fewer reaps means less probability for unnessary | |
262 | * cpucache drain/refill cycles. | |
1da177e4 | 263 | * |
dc6f3f27 | 264 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
265 | * which could lock up otherwise freeable slabs. |
266 | */ | |
5f0985bb JZ |
267 | #define REAPTIMEOUT_AC (2*HZ) |
268 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
269 | |
270 | #if STATS | |
271 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
272 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
273 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
274 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 275 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
276 | #define STATS_SET_HIGH(x) \ |
277 | do { \ | |
278 | if ((x)->num_active > (x)->high_mark) \ | |
279 | (x)->high_mark = (x)->num_active; \ | |
280 | } while (0) | |
1da177e4 LT |
281 | #define STATS_INC_ERR(x) ((x)->errors++) |
282 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 283 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 284 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
285 | #define STATS_SET_FREEABLE(x, i) \ |
286 | do { \ | |
287 | if ((x)->max_freeable < i) \ | |
288 | (x)->max_freeable = i; \ | |
289 | } while (0) | |
1da177e4 LT |
290 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
291 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
292 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
293 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
294 | #else | |
295 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
296 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
297 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
298 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 299 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
300 | #define STATS_SET_HIGH(x) do { } while (0) |
301 | #define STATS_INC_ERR(x) do { } while (0) | |
302 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 303 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 304 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 305 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
306 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
307 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
308 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
309 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
310 | #endif | |
311 | ||
312 | #if DEBUG | |
1da177e4 | 313 | |
a737b3e2 AM |
314 | /* |
315 | * memory layout of objects: | |
1da177e4 | 316 | * 0 : objp |
3dafccf2 | 317 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
318 | * the end of an object is aligned with the end of the real |
319 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 320 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 321 | * redzone word. |
3dafccf2 | 322 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
323 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
324 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 325 | * [BYTES_PER_WORD long] |
1da177e4 | 326 | */ |
343e0d7a | 327 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 328 | { |
3dafccf2 | 329 | return cachep->obj_offset; |
1da177e4 LT |
330 | } |
331 | ||
b46b8f19 | 332 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
333 | { |
334 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
335 | return (unsigned long long*) (objp + obj_offset(cachep) - |
336 | sizeof(unsigned long long)); | |
1da177e4 LT |
337 | } |
338 | ||
b46b8f19 | 339 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
340 | { |
341 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
342 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 343 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 344 | sizeof(unsigned long long) - |
87a927c7 | 345 | REDZONE_ALIGN); |
3b0efdfa | 346 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 347 | sizeof(unsigned long long)); |
1da177e4 LT |
348 | } |
349 | ||
343e0d7a | 350 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
351 | { |
352 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 353 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
354 | } |
355 | ||
356 | #else | |
357 | ||
3dafccf2 | 358 | #define obj_offset(x) 0 |
b46b8f19 DW |
359 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
360 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
361 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
362 | ||
363 | #endif | |
364 | ||
1da177e4 | 365 | /* |
3df1cccd DR |
366 | * Do not go above this order unless 0 objects fit into the slab or |
367 | * overridden on the command line. | |
1da177e4 | 368 | */ |
543585cc DR |
369 | #define SLAB_MAX_ORDER_HI 1 |
370 | #define SLAB_MAX_ORDER_LO 0 | |
371 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 372 | static bool slab_max_order_set __initdata; |
1da177e4 | 373 | |
8456a648 | 374 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
375 | unsigned int idx) |
376 | { | |
8456a648 | 377 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
378 | } |
379 | ||
6fb92430 | 380 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 381 | /* internal cache of cache description objs */ |
9b030cb8 | 382 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
383 | .batchcount = 1, |
384 | .limit = BOOT_CPUCACHE_ENTRIES, | |
385 | .shared = 1, | |
3b0efdfa | 386 | .size = sizeof(struct kmem_cache), |
b28a02de | 387 | .name = "kmem_cache", |
1da177e4 LT |
388 | }; |
389 | ||
1871e52c | 390 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 391 | |
343e0d7a | 392 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 393 | { |
bf0dea23 | 394 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
395 | } |
396 | ||
a737b3e2 AM |
397 | /* |
398 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
399 | */ | |
70f75067 | 400 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
d50112ed | 401 | slab_flags_t flags, size_t *left_over) |
fbaccacf | 402 | { |
70f75067 | 403 | unsigned int num; |
fbaccacf | 404 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 405 | |
fbaccacf SR |
406 | /* |
407 | * The slab management structure can be either off the slab or | |
408 | * on it. For the latter case, the memory allocated for a | |
409 | * slab is used for: | |
410 | * | |
fbaccacf | 411 | * - @buffer_size bytes for each object |
2e6b3602 JK |
412 | * - One freelist_idx_t for each object |
413 | * | |
414 | * We don't need to consider alignment of freelist because | |
415 | * freelist will be at the end of slab page. The objects will be | |
416 | * at the correct alignment. | |
fbaccacf SR |
417 | * |
418 | * If the slab management structure is off the slab, then the | |
419 | * alignment will already be calculated into the size. Because | |
420 | * the slabs are all pages aligned, the objects will be at the | |
421 | * correct alignment when allocated. | |
422 | */ | |
b03a017b | 423 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 424 | num = slab_size / buffer_size; |
2e6b3602 | 425 | *left_over = slab_size % buffer_size; |
fbaccacf | 426 | } else { |
70f75067 | 427 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
428 | *left_over = slab_size % |
429 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 430 | } |
70f75067 JK |
431 | |
432 | return num; | |
1da177e4 LT |
433 | } |
434 | ||
f28510d3 | 435 | #if DEBUG |
d40cee24 | 436 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 437 | |
a737b3e2 AM |
438 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
439 | char *msg) | |
1da177e4 | 440 | { |
1170532b | 441 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 442 | function, cachep->name, msg); |
1da177e4 | 443 | dump_stack(); |
373d4d09 | 444 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 445 | } |
f28510d3 | 446 | #endif |
1da177e4 | 447 | |
3395ee05 PM |
448 | /* |
449 | * By default on NUMA we use alien caches to stage the freeing of | |
450 | * objects allocated from other nodes. This causes massive memory | |
451 | * inefficiencies when using fake NUMA setup to split memory into a | |
452 | * large number of small nodes, so it can be disabled on the command | |
453 | * line | |
454 | */ | |
455 | ||
456 | static int use_alien_caches __read_mostly = 1; | |
457 | static int __init noaliencache_setup(char *s) | |
458 | { | |
459 | use_alien_caches = 0; | |
460 | return 1; | |
461 | } | |
462 | __setup("noaliencache", noaliencache_setup); | |
463 | ||
3df1cccd DR |
464 | static int __init slab_max_order_setup(char *str) |
465 | { | |
466 | get_option(&str, &slab_max_order); | |
467 | slab_max_order = slab_max_order < 0 ? 0 : | |
468 | min(slab_max_order, MAX_ORDER - 1); | |
469 | slab_max_order_set = true; | |
470 | ||
471 | return 1; | |
472 | } | |
473 | __setup("slab_max_order=", slab_max_order_setup); | |
474 | ||
8fce4d8e CL |
475 | #ifdef CONFIG_NUMA |
476 | /* | |
477 | * Special reaping functions for NUMA systems called from cache_reap(). | |
478 | * These take care of doing round robin flushing of alien caches (containing | |
479 | * objects freed on different nodes from which they were allocated) and the | |
480 | * flushing of remote pcps by calling drain_node_pages. | |
481 | */ | |
1871e52c | 482 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
483 | |
484 | static void init_reap_node(int cpu) | |
485 | { | |
0edaf86c AM |
486 | per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu), |
487 | node_online_map); | |
8fce4d8e CL |
488 | } |
489 | ||
490 | static void next_reap_node(void) | |
491 | { | |
909ea964 | 492 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 493 | |
0edaf86c | 494 | node = next_node_in(node, node_online_map); |
909ea964 | 495 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
496 | } |
497 | ||
498 | #else | |
499 | #define init_reap_node(cpu) do { } while (0) | |
500 | #define next_reap_node(void) do { } while (0) | |
501 | #endif | |
502 | ||
1da177e4 LT |
503 | /* |
504 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
505 | * via the workqueue/eventd. | |
506 | * Add the CPU number into the expiration time to minimize the possibility of | |
507 | * the CPUs getting into lockstep and contending for the global cache chain | |
508 | * lock. | |
509 | */ | |
0db0628d | 510 | static void start_cpu_timer(int cpu) |
1da177e4 | 511 | { |
1871e52c | 512 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 | 513 | |
eac0337a | 514 | if (reap_work->work.func == NULL) { |
8fce4d8e | 515 | init_reap_node(cpu); |
203b42f7 | 516 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
517 | schedule_delayed_work_on(cpu, reap_work, |
518 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
519 | } |
520 | } | |
521 | ||
1fe00d50 | 522 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 523 | { |
1fe00d50 JK |
524 | if (ac) { |
525 | ac->avail = 0; | |
526 | ac->limit = limit; | |
527 | ac->batchcount = batch; | |
528 | ac->touched = 0; | |
1da177e4 | 529 | } |
1fe00d50 JK |
530 | } |
531 | ||
532 | static struct array_cache *alloc_arraycache(int node, int entries, | |
533 | int batchcount, gfp_t gfp) | |
534 | { | |
5e804789 | 535 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
536 | struct array_cache *ac = NULL; |
537 | ||
538 | ac = kmalloc_node(memsize, gfp, node); | |
92d1d07d QC |
539 | /* |
540 | * The array_cache structures contain pointers to free object. | |
541 | * However, when such objects are allocated or transferred to another | |
542 | * cache the pointers are not cleared and they could be counted as | |
543 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
544 | * not scan such objects. | |
545 | */ | |
546 | kmemleak_no_scan(ac); | |
1fe00d50 JK |
547 | init_arraycache(ac, entries, batchcount); |
548 | return ac; | |
1da177e4 LT |
549 | } |
550 | ||
f68f8ddd JK |
551 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
552 | struct page *page, void *objp) | |
072bb0aa | 553 | { |
f68f8ddd JK |
554 | struct kmem_cache_node *n; |
555 | int page_node; | |
556 | LIST_HEAD(list); | |
072bb0aa | 557 | |
f68f8ddd JK |
558 | page_node = page_to_nid(page); |
559 | n = get_node(cachep, page_node); | |
381760ea | 560 | |
f68f8ddd JK |
561 | spin_lock(&n->list_lock); |
562 | free_block(cachep, &objp, 1, page_node, &list); | |
563 | spin_unlock(&n->list_lock); | |
381760ea | 564 | |
f68f8ddd | 565 | slabs_destroy(cachep, &list); |
072bb0aa MG |
566 | } |
567 | ||
3ded175a CL |
568 | /* |
569 | * Transfer objects in one arraycache to another. | |
570 | * Locking must be handled by the caller. | |
571 | * | |
572 | * Return the number of entries transferred. | |
573 | */ | |
574 | static int transfer_objects(struct array_cache *to, | |
575 | struct array_cache *from, unsigned int max) | |
576 | { | |
577 | /* Figure out how many entries to transfer */ | |
732eacc0 | 578 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
579 | |
580 | if (!nr) | |
581 | return 0; | |
582 | ||
583 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
584 | sizeof(void *) *nr); | |
585 | ||
586 | from->avail -= nr; | |
587 | to->avail += nr; | |
3ded175a CL |
588 | return nr; |
589 | } | |
590 | ||
dabc3e29 KC |
591 | /* &alien->lock must be held by alien callers. */ |
592 | static __always_inline void __free_one(struct array_cache *ac, void *objp) | |
593 | { | |
594 | /* Avoid trivial double-free. */ | |
595 | if (IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) && | |
596 | WARN_ON_ONCE(ac->avail > 0 && ac->entry[ac->avail - 1] == objp)) | |
597 | return; | |
598 | ac->entry[ac->avail++] = objp; | |
599 | } | |
600 | ||
765c4507 CL |
601 | #ifndef CONFIG_NUMA |
602 | ||
603 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 604 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 605 | |
c8522a3a JK |
606 | static inline struct alien_cache **alloc_alien_cache(int node, |
607 | int limit, gfp_t gfp) | |
765c4507 | 608 | { |
8888177e | 609 | return NULL; |
765c4507 CL |
610 | } |
611 | ||
c8522a3a | 612 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
613 | { |
614 | } | |
615 | ||
616 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
617 | { | |
618 | return 0; | |
619 | } | |
620 | ||
621 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
622 | gfp_t flags) | |
623 | { | |
624 | return NULL; | |
625 | } | |
626 | ||
8b98c169 | 627 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
628 | gfp_t flags, int nodeid) |
629 | { | |
630 | return NULL; | |
631 | } | |
632 | ||
4167e9b2 DR |
633 | static inline gfp_t gfp_exact_node(gfp_t flags) |
634 | { | |
444eb2a4 | 635 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
636 | } |
637 | ||
765c4507 CL |
638 | #else /* CONFIG_NUMA */ |
639 | ||
8b98c169 | 640 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 641 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 642 | |
c8522a3a JK |
643 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
644 | int batch, gfp_t gfp) | |
645 | { | |
5e804789 | 646 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
647 | struct alien_cache *alc = NULL; |
648 | ||
649 | alc = kmalloc_node(memsize, gfp, node); | |
09c2e76e | 650 | if (alc) { |
92d1d07d | 651 | kmemleak_no_scan(alc); |
09c2e76e CL |
652 | init_arraycache(&alc->ac, entries, batch); |
653 | spin_lock_init(&alc->lock); | |
654 | } | |
c8522a3a JK |
655 | return alc; |
656 | } | |
657 | ||
658 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 659 | { |
c8522a3a | 660 | struct alien_cache **alc_ptr; |
e498be7d CL |
661 | int i; |
662 | ||
663 | if (limit > 1) | |
664 | limit = 12; | |
b9726c26 | 665 | alc_ptr = kcalloc_node(nr_node_ids, sizeof(void *), gfp, node); |
c8522a3a JK |
666 | if (!alc_ptr) |
667 | return NULL; | |
668 | ||
669 | for_each_node(i) { | |
670 | if (i == node || !node_online(i)) | |
671 | continue; | |
672 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
673 | if (!alc_ptr[i]) { | |
674 | for (i--; i >= 0; i--) | |
675 | kfree(alc_ptr[i]); | |
676 | kfree(alc_ptr); | |
677 | return NULL; | |
e498be7d CL |
678 | } |
679 | } | |
c8522a3a | 680 | return alc_ptr; |
e498be7d CL |
681 | } |
682 | ||
c8522a3a | 683 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
684 | { |
685 | int i; | |
686 | ||
c8522a3a | 687 | if (!alc_ptr) |
e498be7d | 688 | return; |
e498be7d | 689 | for_each_node(i) |
c8522a3a JK |
690 | kfree(alc_ptr[i]); |
691 | kfree(alc_ptr); | |
e498be7d CL |
692 | } |
693 | ||
343e0d7a | 694 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
695 | struct array_cache *ac, int node, |
696 | struct list_head *list) | |
e498be7d | 697 | { |
18bf8541 | 698 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
699 | |
700 | if (ac->avail) { | |
ce8eb6c4 | 701 | spin_lock(&n->list_lock); |
e00946fe CL |
702 | /* |
703 | * Stuff objects into the remote nodes shared array first. | |
704 | * That way we could avoid the overhead of putting the objects | |
705 | * into the free lists and getting them back later. | |
706 | */ | |
ce8eb6c4 CL |
707 | if (n->shared) |
708 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 709 | |
833b706c | 710 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 711 | ac->avail = 0; |
ce8eb6c4 | 712 | spin_unlock(&n->list_lock); |
e498be7d CL |
713 | } |
714 | } | |
715 | ||
8fce4d8e CL |
716 | /* |
717 | * Called from cache_reap() to regularly drain alien caches round robin. | |
718 | */ | |
ce8eb6c4 | 719 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 720 | { |
909ea964 | 721 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 722 | |
ce8eb6c4 | 723 | if (n->alien) { |
c8522a3a JK |
724 | struct alien_cache *alc = n->alien[node]; |
725 | struct array_cache *ac; | |
726 | ||
727 | if (alc) { | |
728 | ac = &alc->ac; | |
49dfc304 | 729 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
730 | LIST_HEAD(list); |
731 | ||
732 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 733 | spin_unlock_irq(&alc->lock); |
833b706c | 734 | slabs_destroy(cachep, &list); |
c8522a3a | 735 | } |
8fce4d8e CL |
736 | } |
737 | } | |
738 | } | |
739 | ||
a737b3e2 | 740 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 741 | struct alien_cache **alien) |
e498be7d | 742 | { |
b28a02de | 743 | int i = 0; |
c8522a3a | 744 | struct alien_cache *alc; |
e498be7d CL |
745 | struct array_cache *ac; |
746 | unsigned long flags; | |
747 | ||
748 | for_each_online_node(i) { | |
c8522a3a JK |
749 | alc = alien[i]; |
750 | if (alc) { | |
833b706c JK |
751 | LIST_HEAD(list); |
752 | ||
c8522a3a | 753 | ac = &alc->ac; |
49dfc304 | 754 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 755 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 756 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 757 | slabs_destroy(cachep, &list); |
e498be7d CL |
758 | } |
759 | } | |
760 | } | |
729bd0b7 | 761 | |
25c4f304 JK |
762 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
763 | int node, int page_node) | |
729bd0b7 | 764 | { |
ce8eb6c4 | 765 | struct kmem_cache_node *n; |
c8522a3a JK |
766 | struct alien_cache *alien = NULL; |
767 | struct array_cache *ac; | |
97654dfa | 768 | LIST_HEAD(list); |
1ca4cb24 | 769 | |
18bf8541 | 770 | n = get_node(cachep, node); |
729bd0b7 | 771 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
772 | if (n->alien && n->alien[page_node]) { |
773 | alien = n->alien[page_node]; | |
c8522a3a | 774 | ac = &alien->ac; |
49dfc304 | 775 | spin_lock(&alien->lock); |
c8522a3a | 776 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 777 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 778 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 779 | } |
dabc3e29 | 780 | __free_one(ac, objp); |
49dfc304 | 781 | spin_unlock(&alien->lock); |
833b706c | 782 | slabs_destroy(cachep, &list); |
729bd0b7 | 783 | } else { |
25c4f304 | 784 | n = get_node(cachep, page_node); |
18bf8541 | 785 | spin_lock(&n->list_lock); |
25c4f304 | 786 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 787 | spin_unlock(&n->list_lock); |
97654dfa | 788 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
789 | } |
790 | return 1; | |
791 | } | |
25c4f304 JK |
792 | |
793 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
794 | { | |
795 | int page_node = page_to_nid(virt_to_page(objp)); | |
796 | int node = numa_mem_id(); | |
797 | /* | |
798 | * Make sure we are not freeing a object from another node to the array | |
799 | * cache on this cpu. | |
800 | */ | |
801 | if (likely(node == page_node)) | |
802 | return 0; | |
803 | ||
804 | return __cache_free_alien(cachep, objp, node, page_node); | |
805 | } | |
4167e9b2 DR |
806 | |
807 | /* | |
444eb2a4 MG |
808 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
809 | * warn about failures. | |
4167e9b2 DR |
810 | */ |
811 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
812 | { | |
444eb2a4 | 813 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 814 | } |
e498be7d CL |
815 | #endif |
816 | ||
ded0ecf6 JK |
817 | static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp) |
818 | { | |
819 | struct kmem_cache_node *n; | |
820 | ||
821 | /* | |
822 | * Set up the kmem_cache_node for cpu before we can | |
823 | * begin anything. Make sure some other cpu on this | |
824 | * node has not already allocated this | |
825 | */ | |
826 | n = get_node(cachep, node); | |
827 | if (n) { | |
828 | spin_lock_irq(&n->list_lock); | |
829 | n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount + | |
830 | cachep->num; | |
831 | spin_unlock_irq(&n->list_lock); | |
832 | ||
833 | return 0; | |
834 | } | |
835 | ||
836 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); | |
837 | if (!n) | |
838 | return -ENOMEM; | |
839 | ||
840 | kmem_cache_node_init(n); | |
841 | n->next_reap = jiffies + REAPTIMEOUT_NODE + | |
842 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
843 | ||
844 | n->free_limit = | |
845 | (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num; | |
846 | ||
847 | /* | |
848 | * The kmem_cache_nodes don't come and go as CPUs | |
849 | * come and go. slab_mutex is sufficient | |
850 | * protection here. | |
851 | */ | |
852 | cachep->node[node] = n; | |
853 | ||
854 | return 0; | |
855 | } | |
856 | ||
6731d4f1 | 857 | #if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP) |
8f9f8d9e | 858 | /* |
6a67368c | 859 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 860 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 861 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 862 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
863 | * already in use. |
864 | * | |
18004c5d | 865 | * Must hold slab_mutex. |
8f9f8d9e | 866 | */ |
6a67368c | 867 | static int init_cache_node_node(int node) |
8f9f8d9e | 868 | { |
ded0ecf6 | 869 | int ret; |
8f9f8d9e | 870 | struct kmem_cache *cachep; |
8f9f8d9e | 871 | |
18004c5d | 872 | list_for_each_entry(cachep, &slab_caches, list) { |
ded0ecf6 JK |
873 | ret = init_cache_node(cachep, node, GFP_KERNEL); |
874 | if (ret) | |
875 | return ret; | |
8f9f8d9e | 876 | } |
ded0ecf6 | 877 | |
8f9f8d9e DR |
878 | return 0; |
879 | } | |
6731d4f1 | 880 | #endif |
8f9f8d9e | 881 | |
c3d332b6 JK |
882 | static int setup_kmem_cache_node(struct kmem_cache *cachep, |
883 | int node, gfp_t gfp, bool force_change) | |
884 | { | |
885 | int ret = -ENOMEM; | |
886 | struct kmem_cache_node *n; | |
887 | struct array_cache *old_shared = NULL; | |
888 | struct array_cache *new_shared = NULL; | |
889 | struct alien_cache **new_alien = NULL; | |
890 | LIST_HEAD(list); | |
891 | ||
892 | if (use_alien_caches) { | |
893 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
894 | if (!new_alien) | |
895 | goto fail; | |
896 | } | |
897 | ||
898 | if (cachep->shared) { | |
899 | new_shared = alloc_arraycache(node, | |
900 | cachep->shared * cachep->batchcount, 0xbaadf00d, gfp); | |
901 | if (!new_shared) | |
902 | goto fail; | |
903 | } | |
904 | ||
905 | ret = init_cache_node(cachep, node, gfp); | |
906 | if (ret) | |
907 | goto fail; | |
908 | ||
909 | n = get_node(cachep, node); | |
910 | spin_lock_irq(&n->list_lock); | |
911 | if (n->shared && force_change) { | |
912 | free_block(cachep, n->shared->entry, | |
913 | n->shared->avail, node, &list); | |
914 | n->shared->avail = 0; | |
915 | } | |
916 | ||
917 | if (!n->shared || force_change) { | |
918 | old_shared = n->shared; | |
919 | n->shared = new_shared; | |
920 | new_shared = NULL; | |
921 | } | |
922 | ||
923 | if (!n->alien) { | |
924 | n->alien = new_alien; | |
925 | new_alien = NULL; | |
926 | } | |
927 | ||
928 | spin_unlock_irq(&n->list_lock); | |
929 | slabs_destroy(cachep, &list); | |
930 | ||
801faf0d JK |
931 | /* |
932 | * To protect lockless access to n->shared during irq disabled context. | |
933 | * If n->shared isn't NULL in irq disabled context, accessing to it is | |
934 | * guaranteed to be valid until irq is re-enabled, because it will be | |
6564a25e | 935 | * freed after synchronize_rcu(). |
801faf0d | 936 | */ |
86d9f485 | 937 | if (old_shared && force_change) |
6564a25e | 938 | synchronize_rcu(); |
801faf0d | 939 | |
c3d332b6 JK |
940 | fail: |
941 | kfree(old_shared); | |
942 | kfree(new_shared); | |
943 | free_alien_cache(new_alien); | |
944 | ||
945 | return ret; | |
946 | } | |
947 | ||
6731d4f1 SAS |
948 | #ifdef CONFIG_SMP |
949 | ||
0db0628d | 950 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
951 | { |
952 | struct kmem_cache *cachep; | |
ce8eb6c4 | 953 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 954 | int node = cpu_to_mem(cpu); |
a70f7302 | 955 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 956 | |
18004c5d | 957 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
958 | struct array_cache *nc; |
959 | struct array_cache *shared; | |
c8522a3a | 960 | struct alien_cache **alien; |
97654dfa | 961 | LIST_HEAD(list); |
fbf1e473 | 962 | |
18bf8541 | 963 | n = get_node(cachep, node); |
ce8eb6c4 | 964 | if (!n) |
bf0dea23 | 965 | continue; |
fbf1e473 | 966 | |
ce8eb6c4 | 967 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 968 | |
ce8eb6c4 CL |
969 | /* Free limit for this kmem_cache_node */ |
970 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
971 | |
972 | /* cpu is dead; no one can alloc from it. */ | |
973 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
517f9f1e LR |
974 | free_block(cachep, nc->entry, nc->avail, node, &list); |
975 | nc->avail = 0; | |
fbf1e473 | 976 | |
58463c1f | 977 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 978 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 979 | goto free_slab; |
fbf1e473 AM |
980 | } |
981 | ||
ce8eb6c4 | 982 | shared = n->shared; |
fbf1e473 AM |
983 | if (shared) { |
984 | free_block(cachep, shared->entry, | |
97654dfa | 985 | shared->avail, node, &list); |
ce8eb6c4 | 986 | n->shared = NULL; |
fbf1e473 AM |
987 | } |
988 | ||
ce8eb6c4 CL |
989 | alien = n->alien; |
990 | n->alien = NULL; | |
fbf1e473 | 991 | |
ce8eb6c4 | 992 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
993 | |
994 | kfree(shared); | |
995 | if (alien) { | |
996 | drain_alien_cache(cachep, alien); | |
997 | free_alien_cache(alien); | |
998 | } | |
bf0dea23 JK |
999 | |
1000 | free_slab: | |
97654dfa | 1001 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1002 | } |
1003 | /* | |
1004 | * In the previous loop, all the objects were freed to | |
1005 | * the respective cache's slabs, now we can go ahead and | |
1006 | * shrink each nodelist to its limit. | |
1007 | */ | |
18004c5d | 1008 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1009 | n = get_node(cachep, node); |
ce8eb6c4 | 1010 | if (!n) |
fbf1e473 | 1011 | continue; |
a5aa63a5 | 1012 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
1013 | } |
1014 | } | |
1015 | ||
0db0628d | 1016 | static int cpuup_prepare(long cpu) |
1da177e4 | 1017 | { |
343e0d7a | 1018 | struct kmem_cache *cachep; |
7d6e6d09 | 1019 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1020 | int err; |
1da177e4 | 1021 | |
fbf1e473 AM |
1022 | /* |
1023 | * We need to do this right in the beginning since | |
1024 | * alloc_arraycache's are going to use this list. | |
1025 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1026 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1027 | */ |
6a67368c | 1028 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1029 | if (err < 0) |
1030 | goto bad; | |
fbf1e473 AM |
1031 | |
1032 | /* | |
1033 | * Now we can go ahead with allocating the shared arrays and | |
1034 | * array caches | |
1035 | */ | |
18004c5d | 1036 | list_for_each_entry(cachep, &slab_caches, list) { |
c3d332b6 JK |
1037 | err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false); |
1038 | if (err) | |
1039 | goto bad; | |
fbf1e473 | 1040 | } |
ce79ddc8 | 1041 | |
fbf1e473 AM |
1042 | return 0; |
1043 | bad: | |
12d00f6a | 1044 | cpuup_canceled(cpu); |
fbf1e473 AM |
1045 | return -ENOMEM; |
1046 | } | |
1047 | ||
6731d4f1 | 1048 | int slab_prepare_cpu(unsigned int cpu) |
fbf1e473 | 1049 | { |
6731d4f1 | 1050 | int err; |
fbf1e473 | 1051 | |
6731d4f1 SAS |
1052 | mutex_lock(&slab_mutex); |
1053 | err = cpuup_prepare(cpu); | |
1054 | mutex_unlock(&slab_mutex); | |
1055 | return err; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * This is called for a failed online attempt and for a successful | |
1060 | * offline. | |
1061 | * | |
1062 | * Even if all the cpus of a node are down, we don't free the | |
1063 | * kmem_list3 of any cache. This to avoid a race between cpu_down, and | |
1064 | * a kmalloc allocation from another cpu for memory from the node of | |
1065 | * the cpu going down. The list3 structure is usually allocated from | |
1066 | * kmem_cache_create() and gets destroyed at kmem_cache_destroy(). | |
1067 | */ | |
1068 | int slab_dead_cpu(unsigned int cpu) | |
1069 | { | |
1070 | mutex_lock(&slab_mutex); | |
1071 | cpuup_canceled(cpu); | |
1072 | mutex_unlock(&slab_mutex); | |
1073 | return 0; | |
1074 | } | |
8f5be20b | 1075 | #endif |
6731d4f1 SAS |
1076 | |
1077 | static int slab_online_cpu(unsigned int cpu) | |
1078 | { | |
1079 | start_cpu_timer(cpu); | |
1080 | return 0; | |
1da177e4 LT |
1081 | } |
1082 | ||
6731d4f1 SAS |
1083 | static int slab_offline_cpu(unsigned int cpu) |
1084 | { | |
1085 | /* | |
1086 | * Shutdown cache reaper. Note that the slab_mutex is held so | |
1087 | * that if cache_reap() is invoked it cannot do anything | |
1088 | * expensive but will only modify reap_work and reschedule the | |
1089 | * timer. | |
1090 | */ | |
1091 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); | |
1092 | /* Now the cache_reaper is guaranteed to be not running. */ | |
1093 | per_cpu(slab_reap_work, cpu).work.func = NULL; | |
1094 | return 0; | |
1095 | } | |
1da177e4 | 1096 | |
8f9f8d9e DR |
1097 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1098 | /* | |
1099 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1100 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1101 | * removed. | |
1102 | * | |
18004c5d | 1103 | * Must hold slab_mutex. |
8f9f8d9e | 1104 | */ |
6a67368c | 1105 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1106 | { |
1107 | struct kmem_cache *cachep; | |
1108 | int ret = 0; | |
1109 | ||
18004c5d | 1110 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1111 | struct kmem_cache_node *n; |
8f9f8d9e | 1112 | |
18bf8541 | 1113 | n = get_node(cachep, node); |
ce8eb6c4 | 1114 | if (!n) |
8f9f8d9e DR |
1115 | continue; |
1116 | ||
a5aa63a5 | 1117 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1118 | |
ce8eb6c4 CL |
1119 | if (!list_empty(&n->slabs_full) || |
1120 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1121 | ret = -EBUSY; |
1122 | break; | |
1123 | } | |
1124 | } | |
1125 | return ret; | |
1126 | } | |
1127 | ||
1128 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1129 | unsigned long action, void *arg) | |
1130 | { | |
1131 | struct memory_notify *mnb = arg; | |
1132 | int ret = 0; | |
1133 | int nid; | |
1134 | ||
1135 | nid = mnb->status_change_nid; | |
1136 | if (nid < 0) | |
1137 | goto out; | |
1138 | ||
1139 | switch (action) { | |
1140 | case MEM_GOING_ONLINE: | |
18004c5d | 1141 | mutex_lock(&slab_mutex); |
6a67368c | 1142 | ret = init_cache_node_node(nid); |
18004c5d | 1143 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1144 | break; |
1145 | case MEM_GOING_OFFLINE: | |
18004c5d | 1146 | mutex_lock(&slab_mutex); |
6a67368c | 1147 | ret = drain_cache_node_node(nid); |
18004c5d | 1148 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1149 | break; |
1150 | case MEM_ONLINE: | |
1151 | case MEM_OFFLINE: | |
1152 | case MEM_CANCEL_ONLINE: | |
1153 | case MEM_CANCEL_OFFLINE: | |
1154 | break; | |
1155 | } | |
1156 | out: | |
5fda1bd5 | 1157 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1158 | } |
1159 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1160 | ||
e498be7d | 1161 | /* |
ce8eb6c4 | 1162 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1163 | */ |
6744f087 | 1164 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1165 | int nodeid) |
e498be7d | 1166 | { |
6744f087 | 1167 | struct kmem_cache_node *ptr; |
e498be7d | 1168 | |
6744f087 | 1169 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1170 | BUG_ON(!ptr); |
1171 | ||
6744f087 | 1172 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1173 | /* |
1174 | * Do not assume that spinlocks can be initialized via memcpy: | |
1175 | */ | |
1176 | spin_lock_init(&ptr->list_lock); | |
1177 | ||
e498be7d | 1178 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1179 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1180 | } |
1181 | ||
556a169d | 1182 | /* |
ce8eb6c4 CL |
1183 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1184 | * size of kmem_cache_node. | |
556a169d | 1185 | */ |
ce8eb6c4 | 1186 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1187 | { |
1188 | int node; | |
1189 | ||
1190 | for_each_online_node(node) { | |
ce8eb6c4 | 1191 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1192 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1193 | REAPTIMEOUT_NODE + |
1194 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1195 | } |
1196 | } | |
1197 | ||
a737b3e2 AM |
1198 | /* |
1199 | * Initialisation. Called after the page allocator have been initialised and | |
1200 | * before smp_init(). | |
1da177e4 LT |
1201 | */ |
1202 | void __init kmem_cache_init(void) | |
1203 | { | |
e498be7d CL |
1204 | int i; |
1205 | ||
9b030cb8 CL |
1206 | kmem_cache = &kmem_cache_boot; |
1207 | ||
8888177e | 1208 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1209 | use_alien_caches = 0; |
1210 | ||
3c583465 | 1211 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1212 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1213 | |
1da177e4 LT |
1214 | /* |
1215 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1216 | * page orders on machines with more than 32MB of memory if |
1217 | * not overridden on the command line. | |
1da177e4 | 1218 | */ |
ca79b0c2 | 1219 | if (!slab_max_order_set && totalram_pages() > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1220 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1221 | |
1da177e4 LT |
1222 | /* Bootstrap is tricky, because several objects are allocated |
1223 | * from caches that do not exist yet: | |
9b030cb8 CL |
1224 | * 1) initialize the kmem_cache cache: it contains the struct |
1225 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1226 | * kmem_cache is statically allocated. | |
e498be7d | 1227 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1228 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1229 | * array at the end of the bootstrap. |
1da177e4 | 1230 | * 2) Create the first kmalloc cache. |
343e0d7a | 1231 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1232 | * An __init data area is used for the head array. |
1233 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1234 | * head arrays. | |
9b030cb8 | 1235 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1236 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1237 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1238 | * the other cache's with kmalloc allocated memory. |
1239 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1240 | */ |
1241 | ||
9b030cb8 | 1242 | /* 1) create the kmem_cache */ |
1da177e4 | 1243 | |
8da3430d | 1244 | /* |
b56efcf0 | 1245 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1246 | */ |
2f9baa9f | 1247 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1248 | offsetof(struct kmem_cache, node) + |
6744f087 | 1249 | nr_node_ids * sizeof(struct kmem_cache_node *), |
8eb8284b | 1250 | SLAB_HWCACHE_ALIGN, 0, 0); |
2f9baa9f | 1251 | list_add(&kmem_cache->list, &slab_caches); |
c03914b7 | 1252 | memcg_link_cache(kmem_cache, NULL); |
bf0dea23 | 1253 | slab_state = PARTIAL; |
1da177e4 | 1254 | |
a737b3e2 | 1255 | /* |
bf0dea23 JK |
1256 | * Initialize the caches that provide memory for the kmem_cache_node |
1257 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1258 | */ |
cc252eae | 1259 | kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE] = create_kmalloc_cache( |
cb5d9fb3 | 1260 | kmalloc_info[INDEX_NODE].name[KMALLOC_NORMAL], |
dc0a7f75 PL |
1261 | kmalloc_info[INDEX_NODE].size, |
1262 | ARCH_KMALLOC_FLAGS, 0, | |
1263 | kmalloc_info[INDEX_NODE].size); | |
bf0dea23 | 1264 | slab_state = PARTIAL_NODE; |
34cc6990 | 1265 | setup_kmalloc_cache_index_table(); |
e498be7d | 1266 | |
e0a42726 IM |
1267 | slab_early_init = 0; |
1268 | ||
ce8eb6c4 | 1269 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1270 | { |
1ca4cb24 PE |
1271 | int nid; |
1272 | ||
9c09a95c | 1273 | for_each_online_node(nid) { |
ce8eb6c4 | 1274 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1275 | |
cc252eae | 1276 | init_list(kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE], |
ce8eb6c4 | 1277 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1278 | } |
1279 | } | |
1da177e4 | 1280 | |
f97d5f63 | 1281 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1282 | } |
1283 | ||
1284 | void __init kmem_cache_init_late(void) | |
1285 | { | |
1286 | struct kmem_cache *cachep; | |
1287 | ||
8429db5c | 1288 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1289 | mutex_lock(&slab_mutex); |
1290 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1291 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1292 | BUG(); | |
18004c5d | 1293 | mutex_unlock(&slab_mutex); |
056c6241 | 1294 | |
97d06609 CL |
1295 | /* Done! */ |
1296 | slab_state = FULL; | |
1297 | ||
8f9f8d9e DR |
1298 | #ifdef CONFIG_NUMA |
1299 | /* | |
1300 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1301 | * node. |
8f9f8d9e DR |
1302 | */ |
1303 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1304 | #endif | |
1305 | ||
a737b3e2 AM |
1306 | /* |
1307 | * The reap timers are started later, with a module init call: That part | |
1308 | * of the kernel is not yet operational. | |
1da177e4 LT |
1309 | */ |
1310 | } | |
1311 | ||
1312 | static int __init cpucache_init(void) | |
1313 | { | |
6731d4f1 | 1314 | int ret; |
1da177e4 | 1315 | |
a737b3e2 AM |
1316 | /* |
1317 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1318 | */ |
6731d4f1 SAS |
1319 | ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online", |
1320 | slab_online_cpu, slab_offline_cpu); | |
1321 | WARN_ON(ret < 0); | |
a164f896 | 1322 | |
1da177e4 LT |
1323 | return 0; |
1324 | } | |
1da177e4 LT |
1325 | __initcall(cpucache_init); |
1326 | ||
8bdec192 RA |
1327 | static noinline void |
1328 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1329 | { | |
9a02d699 | 1330 | #if DEBUG |
ce8eb6c4 | 1331 | struct kmem_cache_node *n; |
8bdec192 RA |
1332 | unsigned long flags; |
1333 | int node; | |
9a02d699 DR |
1334 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1335 | DEFAULT_RATELIMIT_BURST); | |
1336 | ||
1337 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1338 | return; | |
8bdec192 | 1339 | |
5b3810e5 VB |
1340 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1341 | nodeid, gfpflags, &gfpflags); | |
1342 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1343 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1344 | |
18bf8541 | 1345 | for_each_kmem_cache_node(cachep, node, n) { |
bf00bd34 | 1346 | unsigned long total_slabs, free_slabs, free_objs; |
8bdec192 | 1347 | |
ce8eb6c4 | 1348 | spin_lock_irqsave(&n->list_lock, flags); |
bf00bd34 DR |
1349 | total_slabs = n->total_slabs; |
1350 | free_slabs = n->free_slabs; | |
1351 | free_objs = n->free_objects; | |
ce8eb6c4 | 1352 | spin_unlock_irqrestore(&n->list_lock, flags); |
8bdec192 | 1353 | |
bf00bd34 DR |
1354 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n", |
1355 | node, total_slabs - free_slabs, total_slabs, | |
1356 | (total_slabs * cachep->num) - free_objs, | |
1357 | total_slabs * cachep->num); | |
8bdec192 | 1358 | } |
9a02d699 | 1359 | #endif |
8bdec192 RA |
1360 | } |
1361 | ||
1da177e4 | 1362 | /* |
8a7d9b43 WSH |
1363 | * Interface to system's page allocator. No need to hold the |
1364 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1365 | * |
1366 | * If we requested dmaable memory, we will get it. Even if we | |
1367 | * did not request dmaable memory, we might get it, but that | |
1368 | * would be relatively rare and ignorable. | |
1369 | */ | |
0c3aa83e JK |
1370 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1371 | int nodeid) | |
1da177e4 LT |
1372 | { |
1373 | struct page *page; | |
765c4507 | 1374 | |
a618e89f | 1375 | flags |= cachep->allocflags; |
e1b6aa6f | 1376 | |
75f296d9 | 1377 | page = __alloc_pages_node(nodeid, flags, cachep->gfporder); |
8bdec192 | 1378 | if (!page) { |
9a02d699 | 1379 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1380 | return NULL; |
8bdec192 | 1381 | } |
1da177e4 | 1382 | |
6cea1d56 | 1383 | if (charge_slab_page(page, flags, cachep->gfporder, cachep)) { |
f3ccb2c4 VD |
1384 | __free_pages(page, cachep->gfporder); |
1385 | return NULL; | |
1386 | } | |
1387 | ||
a57a4988 | 1388 | __SetPageSlab(page); |
f68f8ddd JK |
1389 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1390 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1391 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1392 | |
0c3aa83e | 1393 | return page; |
1da177e4 LT |
1394 | } |
1395 | ||
1396 | /* | |
1397 | * Interface to system's page release. | |
1398 | */ | |
0c3aa83e | 1399 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1400 | { |
27ee57c9 | 1401 | int order = cachep->gfporder; |
73293c2f | 1402 | |
a57a4988 | 1403 | BUG_ON(!PageSlab(page)); |
73293c2f | 1404 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1405 | __ClearPageSlab(page); |
8456a648 JK |
1406 | page_mapcount_reset(page); |
1407 | page->mapping = NULL; | |
1f458cbf | 1408 | |
1da177e4 | 1409 | if (current->reclaim_state) |
6cea1d56 RG |
1410 | current->reclaim_state->reclaimed_slab += 1 << order; |
1411 | uncharge_slab_page(page, order, cachep); | |
27ee57c9 | 1412 | __free_pages(page, order); |
1da177e4 LT |
1413 | } |
1414 | ||
1415 | static void kmem_rcu_free(struct rcu_head *head) | |
1416 | { | |
68126702 JK |
1417 | struct kmem_cache *cachep; |
1418 | struct page *page; | |
1da177e4 | 1419 | |
68126702 JK |
1420 | page = container_of(head, struct page, rcu_head); |
1421 | cachep = page->slab_cache; | |
1422 | ||
1423 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1424 | } |
1425 | ||
1426 | #if DEBUG | |
40b44137 JK |
1427 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1428 | { | |
8e57f8ac | 1429 | if (debug_pagealloc_enabled_static() && OFF_SLAB(cachep) && |
40b44137 JK |
1430 | (cachep->size % PAGE_SIZE) == 0) |
1431 | return true; | |
1432 | ||
1433 | return false; | |
1434 | } | |
1da177e4 LT |
1435 | |
1436 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
80552f0f | 1437 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, int map) |
40b44137 JK |
1438 | { |
1439 | if (!is_debug_pagealloc_cache(cachep)) | |
1440 | return; | |
1441 | ||
40b44137 JK |
1442 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); |
1443 | } | |
1444 | ||
1445 | #else | |
1446 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
80552f0f | 1447 | int map) {} |
40b44137 | 1448 | |
1da177e4 LT |
1449 | #endif |
1450 | ||
343e0d7a | 1451 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1452 | { |
8c138bc0 | 1453 | int size = cachep->object_size; |
3dafccf2 | 1454 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1455 | |
1456 | memset(addr, val, size); | |
b28a02de | 1457 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1458 | } |
1459 | ||
1460 | static void dump_line(char *data, int offset, int limit) | |
1461 | { | |
1462 | int i; | |
aa83aa40 DJ |
1463 | unsigned char error = 0; |
1464 | int bad_count = 0; | |
1465 | ||
1170532b | 1466 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1467 | for (i = 0; i < limit; i++) { |
1468 | if (data[offset + i] != POISON_FREE) { | |
1469 | error = data[offset + i]; | |
1470 | bad_count++; | |
1471 | } | |
aa83aa40 | 1472 | } |
fdde6abb SAS |
1473 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1474 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1475 | |
1476 | if (bad_count == 1) { | |
1477 | error ^= POISON_FREE; | |
1478 | if (!(error & (error - 1))) { | |
1170532b | 1479 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1480 | #ifdef CONFIG_X86 |
1170532b | 1481 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1482 | #else |
1170532b | 1483 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1484 | #endif |
1485 | } | |
1486 | } | |
1da177e4 LT |
1487 | } |
1488 | #endif | |
1489 | ||
1490 | #if DEBUG | |
1491 | ||
343e0d7a | 1492 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1493 | { |
1494 | int i, size; | |
1495 | char *realobj; | |
1496 | ||
1497 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1498 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1499 | *dbg_redzone1(cachep, objp), | |
1500 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1501 | } |
1502 | ||
85c3e4a5 GU |
1503 | if (cachep->flags & SLAB_STORE_USER) |
1504 | pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp)); | |
3dafccf2 | 1505 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1506 | size = cachep->object_size; |
b28a02de | 1507 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1508 | int limit; |
1509 | limit = 16; | |
b28a02de PE |
1510 | if (i + limit > size) |
1511 | limit = size - i; | |
1da177e4 LT |
1512 | dump_line(realobj, i, limit); |
1513 | } | |
1514 | } | |
1515 | ||
343e0d7a | 1516 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1517 | { |
1518 | char *realobj; | |
1519 | int size, i; | |
1520 | int lines = 0; | |
1521 | ||
40b44137 JK |
1522 | if (is_debug_pagealloc_cache(cachep)) |
1523 | return; | |
1524 | ||
3dafccf2 | 1525 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1526 | size = cachep->object_size; |
1da177e4 | 1527 | |
b28a02de | 1528 | for (i = 0; i < size; i++) { |
1da177e4 | 1529 | char exp = POISON_FREE; |
b28a02de | 1530 | if (i == size - 1) |
1da177e4 LT |
1531 | exp = POISON_END; |
1532 | if (realobj[i] != exp) { | |
1533 | int limit; | |
1534 | /* Mismatch ! */ | |
1535 | /* Print header */ | |
1536 | if (lines == 0) { | |
85c3e4a5 | 1537 | pr_err("Slab corruption (%s): %s start=%px, len=%d\n", |
1170532b JP |
1538 | print_tainted(), cachep->name, |
1539 | realobj, size); | |
1da177e4 LT |
1540 | print_objinfo(cachep, objp, 0); |
1541 | } | |
1542 | /* Hexdump the affected line */ | |
b28a02de | 1543 | i = (i / 16) * 16; |
1da177e4 | 1544 | limit = 16; |
b28a02de PE |
1545 | if (i + limit > size) |
1546 | limit = size - i; | |
1da177e4 LT |
1547 | dump_line(realobj, i, limit); |
1548 | i += 16; | |
1549 | lines++; | |
1550 | /* Limit to 5 lines */ | |
1551 | if (lines > 5) | |
1552 | break; | |
1553 | } | |
1554 | } | |
1555 | if (lines != 0) { | |
1556 | /* Print some data about the neighboring objects, if they | |
1557 | * exist: | |
1558 | */ | |
8456a648 | 1559 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1560 | unsigned int objnr; |
1da177e4 | 1561 | |
8456a648 | 1562 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1563 | if (objnr) { |
8456a648 | 1564 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1565 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1566 | pr_err("Prev obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1567 | print_objinfo(cachep, objp, 2); |
1568 | } | |
b28a02de | 1569 | if (objnr + 1 < cachep->num) { |
8456a648 | 1570 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1571 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1572 | pr_err("Next obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1573 | print_objinfo(cachep, objp, 2); |
1574 | } | |
1575 | } | |
1576 | } | |
1577 | #endif | |
1578 | ||
12dd36fa | 1579 | #if DEBUG |
8456a648 JK |
1580 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1581 | struct page *page) | |
1da177e4 | 1582 | { |
1da177e4 | 1583 | int i; |
b03a017b JK |
1584 | |
1585 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1586 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1587 | POISON_FREE); | |
1588 | } | |
1589 | ||
1da177e4 | 1590 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1591 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1592 | |
1593 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1594 | check_poison_obj(cachep, objp); |
80552f0f | 1595 | slab_kernel_map(cachep, objp, 1); |
1da177e4 LT |
1596 | } |
1597 | if (cachep->flags & SLAB_RED_ZONE) { | |
1598 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1599 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1600 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1601 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1602 | } |
1da177e4 | 1603 | } |
12dd36fa | 1604 | } |
1da177e4 | 1605 | #else |
8456a648 JK |
1606 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1607 | struct page *page) | |
12dd36fa | 1608 | { |
12dd36fa | 1609 | } |
1da177e4 LT |
1610 | #endif |
1611 | ||
911851e6 RD |
1612 | /** |
1613 | * slab_destroy - destroy and release all objects in a slab | |
1614 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1615 | * @page: page pointer being destroyed |
911851e6 | 1616 | * |
8a7d9b43 WSH |
1617 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1618 | * Before calling the slab page must have been unlinked from the cache. The | |
1619 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1620 | */ |
8456a648 | 1621 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1622 | { |
7e007355 | 1623 | void *freelist; |
12dd36fa | 1624 | |
8456a648 JK |
1625 | freelist = page->freelist; |
1626 | slab_destroy_debugcheck(cachep, page); | |
5f0d5a3a | 1627 | if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU)) |
bc4f610d KS |
1628 | call_rcu(&page->rcu_head, kmem_rcu_free); |
1629 | else | |
0c3aa83e | 1630 | kmem_freepages(cachep, page); |
68126702 JK |
1631 | |
1632 | /* | |
8456a648 | 1633 | * From now on, we don't use freelist |
68126702 JK |
1634 | * although actual page can be freed in rcu context |
1635 | */ | |
1636 | if (OFF_SLAB(cachep)) | |
8456a648 | 1637 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1638 | } |
1639 | ||
97654dfa JK |
1640 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1641 | { | |
1642 | struct page *page, *n; | |
1643 | ||
16cb0ec7 TH |
1644 | list_for_each_entry_safe(page, n, list, slab_list) { |
1645 | list_del(&page->slab_list); | |
97654dfa JK |
1646 | slab_destroy(cachep, page); |
1647 | } | |
1648 | } | |
1649 | ||
4d268eba | 1650 | /** |
a70773dd RD |
1651 | * calculate_slab_order - calculate size (page order) of slabs |
1652 | * @cachep: pointer to the cache that is being created | |
1653 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1654 | * @flags: slab allocation flags |
1655 | * | |
1656 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1657 | * |
1658 | * This could be made much more intelligent. For now, try to avoid using | |
1659 | * high order pages for slabs. When the gfp() functions are more friendly | |
1660 | * towards high-order requests, this should be changed. | |
a862f68a MR |
1661 | * |
1662 | * Return: number of left-over bytes in a slab | |
4d268eba | 1663 | */ |
a737b3e2 | 1664 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
d50112ed | 1665 | size_t size, slab_flags_t flags) |
4d268eba PE |
1666 | { |
1667 | size_t left_over = 0; | |
9888e6fa | 1668 | int gfporder; |
4d268eba | 1669 | |
0aa817f0 | 1670 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1671 | unsigned int num; |
1672 | size_t remainder; | |
1673 | ||
70f75067 | 1674 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1675 | if (!num) |
1676 | continue; | |
9888e6fa | 1677 | |
f315e3fa JK |
1678 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1679 | if (num > SLAB_OBJ_MAX_NUM) | |
1680 | break; | |
1681 | ||
b1ab41c4 | 1682 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1683 | struct kmem_cache *freelist_cache; |
1684 | size_t freelist_size; | |
1685 | ||
1686 | freelist_size = num * sizeof(freelist_idx_t); | |
1687 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1688 | if (!freelist_cache) | |
1689 | continue; | |
1690 | ||
b1ab41c4 | 1691 | /* |
3217fd9b | 1692 | * Needed to avoid possible looping condition |
76b342bd | 1693 | * in cache_grow_begin() |
b1ab41c4 | 1694 | */ |
3217fd9b JK |
1695 | if (OFF_SLAB(freelist_cache)) |
1696 | continue; | |
b1ab41c4 | 1697 | |
3217fd9b JK |
1698 | /* check if off slab has enough benefit */ |
1699 | if (freelist_cache->size > cachep->size / 2) | |
1700 | continue; | |
b1ab41c4 | 1701 | } |
4d268eba | 1702 | |
9888e6fa | 1703 | /* Found something acceptable - save it away */ |
4d268eba | 1704 | cachep->num = num; |
9888e6fa | 1705 | cachep->gfporder = gfporder; |
4d268eba PE |
1706 | left_over = remainder; |
1707 | ||
f78bb8ad LT |
1708 | /* |
1709 | * A VFS-reclaimable slab tends to have most allocations | |
1710 | * as GFP_NOFS and we really don't want to have to be allocating | |
1711 | * higher-order pages when we are unable to shrink dcache. | |
1712 | */ | |
1713 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1714 | break; | |
1715 | ||
4d268eba PE |
1716 | /* |
1717 | * Large number of objects is good, but very large slabs are | |
1718 | * currently bad for the gfp()s. | |
1719 | */ | |
543585cc | 1720 | if (gfporder >= slab_max_order) |
4d268eba PE |
1721 | break; |
1722 | ||
9888e6fa LT |
1723 | /* |
1724 | * Acceptable internal fragmentation? | |
1725 | */ | |
a737b3e2 | 1726 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1727 | break; |
1728 | } | |
1729 | return left_over; | |
1730 | } | |
1731 | ||
bf0dea23 JK |
1732 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1733 | struct kmem_cache *cachep, int entries, int batchcount) | |
1734 | { | |
1735 | int cpu; | |
1736 | size_t size; | |
1737 | struct array_cache __percpu *cpu_cache; | |
1738 | ||
1739 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1740 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1741 | |
1742 | if (!cpu_cache) | |
1743 | return NULL; | |
1744 | ||
1745 | for_each_possible_cpu(cpu) { | |
1746 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1747 | entries, batchcount); | |
1748 | } | |
1749 | ||
1750 | return cpu_cache; | |
1751 | } | |
1752 | ||
bd721ea7 | 1753 | static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1754 | { |
97d06609 | 1755 | if (slab_state >= FULL) |
83b519e8 | 1756 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1757 | |
bf0dea23 JK |
1758 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1759 | if (!cachep->cpu_cache) | |
1760 | return 1; | |
1761 | ||
97d06609 | 1762 | if (slab_state == DOWN) { |
bf0dea23 JK |
1763 | /* Creation of first cache (kmem_cache). */ |
1764 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1765 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1766 | /* For kmem_cache_node */ |
1767 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1768 | } else { |
bf0dea23 | 1769 | int node; |
f30cf7d1 | 1770 | |
bf0dea23 JK |
1771 | for_each_online_node(node) { |
1772 | cachep->node[node] = kmalloc_node( | |
1773 | sizeof(struct kmem_cache_node), gfp, node); | |
1774 | BUG_ON(!cachep->node[node]); | |
1775 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1776 | } |
1777 | } | |
bf0dea23 | 1778 | |
6a67368c | 1779 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1780 | jiffies + REAPTIMEOUT_NODE + |
1781 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1782 | |
1783 | cpu_cache_get(cachep)->avail = 0; | |
1784 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1785 | cpu_cache_get(cachep)->batchcount = 1; | |
1786 | cpu_cache_get(cachep)->touched = 0; | |
1787 | cachep->batchcount = 1; | |
1788 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1789 | return 0; |
f30cf7d1 PE |
1790 | } |
1791 | ||
0293d1fd | 1792 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1793 | slab_flags_t flags, const char *name, |
12220dea JK |
1794 | void (*ctor)(void *)) |
1795 | { | |
1796 | return flags; | |
1797 | } | |
1798 | ||
1799 | struct kmem_cache * | |
f4957d5b | 1800 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 1801 | slab_flags_t flags, void (*ctor)(void *)) |
12220dea JK |
1802 | { |
1803 | struct kmem_cache *cachep; | |
1804 | ||
1805 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1806 | if (cachep) { | |
1807 | cachep->refcount++; | |
1808 | ||
1809 | /* | |
1810 | * Adjust the object sizes so that we clear | |
1811 | * the complete object on kzalloc. | |
1812 | */ | |
1813 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1814 | } | |
1815 | return cachep; | |
1816 | } | |
1817 | ||
b03a017b | 1818 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1819 | size_t size, slab_flags_t flags) |
b03a017b JK |
1820 | { |
1821 | size_t left; | |
1822 | ||
1823 | cachep->num = 0; | |
1824 | ||
6471384a AP |
1825 | /* |
1826 | * If slab auto-initialization on free is enabled, store the freelist | |
1827 | * off-slab, so that its contents don't end up in one of the allocated | |
1828 | * objects. | |
1829 | */ | |
1830 | if (unlikely(slab_want_init_on_free(cachep))) | |
1831 | return false; | |
1832 | ||
5f0d5a3a | 1833 | if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU) |
b03a017b JK |
1834 | return false; |
1835 | ||
1836 | left = calculate_slab_order(cachep, size, | |
1837 | flags | CFLGS_OBJFREELIST_SLAB); | |
1838 | if (!cachep->num) | |
1839 | return false; | |
1840 | ||
1841 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1842 | return false; | |
1843 | ||
1844 | cachep->colour = left / cachep->colour_off; | |
1845 | ||
1846 | return true; | |
1847 | } | |
1848 | ||
158e319b | 1849 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1850 | size_t size, slab_flags_t flags) |
158e319b JK |
1851 | { |
1852 | size_t left; | |
1853 | ||
1854 | cachep->num = 0; | |
1855 | ||
1856 | /* | |
3217fd9b JK |
1857 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1858 | * to avoid recursive calls into kmemleak. | |
158e319b | 1859 | */ |
158e319b JK |
1860 | if (flags & SLAB_NOLEAKTRACE) |
1861 | return false; | |
1862 | ||
1863 | /* | |
1864 | * Size is large, assume best to place the slab management obj | |
1865 | * off-slab (should allow better packing of objs). | |
1866 | */ | |
1867 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1868 | if (!cachep->num) | |
1869 | return false; | |
1870 | ||
1871 | /* | |
1872 | * If the slab has been placed off-slab, and we have enough space then | |
1873 | * move it on-slab. This is at the expense of any extra colouring. | |
1874 | */ | |
1875 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1876 | return false; | |
1877 | ||
1878 | cachep->colour = left / cachep->colour_off; | |
1879 | ||
1880 | return true; | |
1881 | } | |
1882 | ||
1883 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
d50112ed | 1884 | size_t size, slab_flags_t flags) |
158e319b JK |
1885 | { |
1886 | size_t left; | |
1887 | ||
1888 | cachep->num = 0; | |
1889 | ||
1890 | left = calculate_slab_order(cachep, size, flags); | |
1891 | if (!cachep->num) | |
1892 | return false; | |
1893 | ||
1894 | cachep->colour = left / cachep->colour_off; | |
1895 | ||
1896 | return true; | |
1897 | } | |
1898 | ||
1da177e4 | 1899 | /** |
039363f3 | 1900 | * __kmem_cache_create - Create a cache. |
a755b76a | 1901 | * @cachep: cache management descriptor |
1da177e4 | 1902 | * @flags: SLAB flags |
1da177e4 LT |
1903 | * |
1904 | * Returns a ptr to the cache on success, NULL on failure. | |
1905 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1906 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1907 | * |
1da177e4 LT |
1908 | * The flags are |
1909 | * | |
1910 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1911 | * to catch references to uninitialised memory. | |
1912 | * | |
1913 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1914 | * for buffer overruns. | |
1915 | * | |
1da177e4 LT |
1916 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1917 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1918 | * as davem. | |
a862f68a MR |
1919 | * |
1920 | * Return: a pointer to the created cache or %NULL in case of error | |
1da177e4 | 1921 | */ |
d50112ed | 1922 | int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags) |
1da177e4 | 1923 | { |
d4a5fca5 | 1924 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 1925 | gfp_t gfp; |
278b1bb1 | 1926 | int err; |
be4a7988 | 1927 | unsigned int size = cachep->size; |
1da177e4 | 1928 | |
1da177e4 | 1929 | #if DEBUG |
1da177e4 LT |
1930 | #if FORCED_DEBUG |
1931 | /* | |
1932 | * Enable redzoning and last user accounting, except for caches with | |
1933 | * large objects, if the increased size would increase the object size | |
1934 | * above the next power of two: caches with object sizes just above a | |
1935 | * power of two have a significant amount of internal fragmentation. | |
1936 | */ | |
87a927c7 DW |
1937 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
1938 | 2 * sizeof(unsigned long long))) | |
b28a02de | 1939 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
5f0d5a3a | 1940 | if (!(flags & SLAB_TYPESAFE_BY_RCU)) |
1da177e4 LT |
1941 | flags |= SLAB_POISON; |
1942 | #endif | |
1da177e4 | 1943 | #endif |
1da177e4 | 1944 | |
a737b3e2 AM |
1945 | /* |
1946 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
1947 | * unaligned accesses for some archs when redzoning is used, and makes |
1948 | * sure any on-slab bufctl's are also correctly aligned. | |
1949 | */ | |
e0771950 | 1950 | size = ALIGN(size, BYTES_PER_WORD); |
1da177e4 | 1951 | |
87a927c7 DW |
1952 | if (flags & SLAB_RED_ZONE) { |
1953 | ralign = REDZONE_ALIGN; | |
1954 | /* If redzoning, ensure that the second redzone is suitably | |
1955 | * aligned, by adjusting the object size accordingly. */ | |
e0771950 | 1956 | size = ALIGN(size, REDZONE_ALIGN); |
87a927c7 | 1957 | } |
ca5f9703 | 1958 | |
a44b56d3 | 1959 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
1960 | if (ralign < cachep->align) { |
1961 | ralign = cachep->align; | |
1da177e4 | 1962 | } |
3ff84a7f PE |
1963 | /* disable debug if necessary */ |
1964 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 1965 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 1966 | /* |
ca5f9703 | 1967 | * 4) Store it. |
1da177e4 | 1968 | */ |
8a13a4cc | 1969 | cachep->align = ralign; |
158e319b JK |
1970 | cachep->colour_off = cache_line_size(); |
1971 | /* Offset must be a multiple of the alignment. */ | |
1972 | if (cachep->colour_off < cachep->align) | |
1973 | cachep->colour_off = cachep->align; | |
1da177e4 | 1974 | |
83b519e8 PE |
1975 | if (slab_is_available()) |
1976 | gfp = GFP_KERNEL; | |
1977 | else | |
1978 | gfp = GFP_NOWAIT; | |
1979 | ||
1da177e4 | 1980 | #if DEBUG |
1da177e4 | 1981 | |
ca5f9703 PE |
1982 | /* |
1983 | * Both debugging options require word-alignment which is calculated | |
1984 | * into align above. | |
1985 | */ | |
1da177e4 | 1986 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 1987 | /* add space for red zone words */ |
3ff84a7f PE |
1988 | cachep->obj_offset += sizeof(unsigned long long); |
1989 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
1990 | } |
1991 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 1992 | /* user store requires one word storage behind the end of |
87a927c7 DW |
1993 | * the real object. But if the second red zone needs to be |
1994 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 1995 | */ |
87a927c7 DW |
1996 | if (flags & SLAB_RED_ZONE) |
1997 | size += REDZONE_ALIGN; | |
1998 | else | |
1999 | size += BYTES_PER_WORD; | |
1da177e4 | 2000 | } |
832a15d2 JK |
2001 | #endif |
2002 | ||
7ed2f9e6 AP |
2003 | kasan_cache_create(cachep, &size, &flags); |
2004 | ||
832a15d2 JK |
2005 | size = ALIGN(size, cachep->align); |
2006 | /* | |
2007 | * We should restrict the number of objects in a slab to implement | |
2008 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2009 | */ | |
2010 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2011 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2012 | ||
2013 | #if DEBUG | |
03a2d2a3 JK |
2014 | /* |
2015 | * To activate debug pagealloc, off-slab management is necessary | |
2016 | * requirement. In early phase of initialization, small sized slab | |
2017 | * doesn't get initialized so it would not be possible. So, we need | |
2018 | * to check size >= 256. It guarantees that all necessary small | |
2019 | * sized slab is initialized in current slab initialization sequence. | |
2020 | */ | |
8e57f8ac | 2021 | if (debug_pagealloc_enabled_static() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2022 | size >= 256 && cachep->object_size > cache_line_size()) { |
2023 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2024 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2025 | ||
2026 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2027 | flags |= CFLGS_OFF_SLAB; | |
2028 | cachep->obj_offset += tmp_size - size; | |
2029 | size = tmp_size; | |
2030 | goto done; | |
2031 | } | |
2032 | } | |
1da177e4 | 2033 | } |
1da177e4 LT |
2034 | #endif |
2035 | ||
b03a017b JK |
2036 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2037 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2038 | goto done; | |
2039 | } | |
2040 | ||
158e319b | 2041 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2042 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2043 | goto done; |
832a15d2 | 2044 | } |
1da177e4 | 2045 | |
158e319b JK |
2046 | if (set_on_slab_cache(cachep, size, flags)) |
2047 | goto done; | |
1da177e4 | 2048 | |
158e319b | 2049 | return -E2BIG; |
1da177e4 | 2050 | |
158e319b JK |
2051 | done: |
2052 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2053 | cachep->flags = flags; |
a57a4988 | 2054 | cachep->allocflags = __GFP_COMP; |
a3187e43 | 2055 | if (flags & SLAB_CACHE_DMA) |
a618e89f | 2056 | cachep->allocflags |= GFP_DMA; |
6d6ea1e9 NB |
2057 | if (flags & SLAB_CACHE_DMA32) |
2058 | cachep->allocflags |= GFP_DMA32; | |
a3ba0744 DR |
2059 | if (flags & SLAB_RECLAIM_ACCOUNT) |
2060 | cachep->allocflags |= __GFP_RECLAIMABLE; | |
3b0efdfa | 2061 | cachep->size = size; |
6a2d7a95 | 2062 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2063 | |
40b44137 JK |
2064 | #if DEBUG |
2065 | /* | |
2066 | * If we're going to use the generic kernel_map_pages() | |
2067 | * poisoning, then it's going to smash the contents of | |
2068 | * the redzone and userword anyhow, so switch them off. | |
2069 | */ | |
2070 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2071 | (cachep->flags & SLAB_POISON) && | |
2072 | is_debug_pagealloc_cache(cachep)) | |
2073 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2074 | #endif | |
2075 | ||
2076 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2077 | cachep->freelist_cache = |
2078 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2079 | } |
1da177e4 | 2080 | |
278b1bb1 CL |
2081 | err = setup_cpu_cache(cachep, gfp); |
2082 | if (err) { | |
52b4b950 | 2083 | __kmem_cache_release(cachep); |
278b1bb1 | 2084 | return err; |
2ed3a4ef | 2085 | } |
1da177e4 | 2086 | |
278b1bb1 | 2087 | return 0; |
1da177e4 | 2088 | } |
1da177e4 LT |
2089 | |
2090 | #if DEBUG | |
2091 | static void check_irq_off(void) | |
2092 | { | |
2093 | BUG_ON(!irqs_disabled()); | |
2094 | } | |
2095 | ||
2096 | static void check_irq_on(void) | |
2097 | { | |
2098 | BUG_ON(irqs_disabled()); | |
2099 | } | |
2100 | ||
18726ca8 JK |
2101 | static void check_mutex_acquired(void) |
2102 | { | |
2103 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2104 | } | |
2105 | ||
343e0d7a | 2106 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2107 | { |
2108 | #ifdef CONFIG_SMP | |
2109 | check_irq_off(); | |
18bf8541 | 2110 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2111 | #endif |
2112 | } | |
e498be7d | 2113 | |
343e0d7a | 2114 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2115 | { |
2116 | #ifdef CONFIG_SMP | |
2117 | check_irq_off(); | |
18bf8541 | 2118 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2119 | #endif |
2120 | } | |
2121 | ||
1da177e4 LT |
2122 | #else |
2123 | #define check_irq_off() do { } while(0) | |
2124 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2125 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2126 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2127 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2128 | #endif |
2129 | ||
18726ca8 JK |
2130 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2131 | int node, bool free_all, struct list_head *list) | |
2132 | { | |
2133 | int tofree; | |
2134 | ||
2135 | if (!ac || !ac->avail) | |
2136 | return; | |
2137 | ||
2138 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2139 | if (tofree > ac->avail) | |
2140 | tofree = (ac->avail + 1) / 2; | |
2141 | ||
2142 | free_block(cachep, ac->entry, tofree, node, list); | |
2143 | ac->avail -= tofree; | |
2144 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2145 | } | |
aab2207c | 2146 | |
1da177e4 LT |
2147 | static void do_drain(void *arg) |
2148 | { | |
a737b3e2 | 2149 | struct kmem_cache *cachep = arg; |
1da177e4 | 2150 | struct array_cache *ac; |
7d6e6d09 | 2151 | int node = numa_mem_id(); |
18bf8541 | 2152 | struct kmem_cache_node *n; |
97654dfa | 2153 | LIST_HEAD(list); |
1da177e4 LT |
2154 | |
2155 | check_irq_off(); | |
9a2dba4b | 2156 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2157 | n = get_node(cachep, node); |
2158 | spin_lock(&n->list_lock); | |
97654dfa | 2159 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2160 | spin_unlock(&n->list_lock); |
97654dfa | 2161 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2162 | ac->avail = 0; |
2163 | } | |
2164 | ||
343e0d7a | 2165 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2166 | { |
ce8eb6c4 | 2167 | struct kmem_cache_node *n; |
e498be7d | 2168 | int node; |
18726ca8 | 2169 | LIST_HEAD(list); |
e498be7d | 2170 | |
15c8b6c1 | 2171 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2172 | check_irq_on(); |
18bf8541 CL |
2173 | for_each_kmem_cache_node(cachep, node, n) |
2174 | if (n->alien) | |
ce8eb6c4 | 2175 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2176 | |
18726ca8 JK |
2177 | for_each_kmem_cache_node(cachep, node, n) { |
2178 | spin_lock_irq(&n->list_lock); | |
2179 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2180 | spin_unlock_irq(&n->list_lock); | |
2181 | ||
2182 | slabs_destroy(cachep, &list); | |
2183 | } | |
1da177e4 LT |
2184 | } |
2185 | ||
ed11d9eb CL |
2186 | /* |
2187 | * Remove slabs from the list of free slabs. | |
2188 | * Specify the number of slabs to drain in tofree. | |
2189 | * | |
2190 | * Returns the actual number of slabs released. | |
2191 | */ | |
2192 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2193 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2194 | { |
ed11d9eb CL |
2195 | struct list_head *p; |
2196 | int nr_freed; | |
8456a648 | 2197 | struct page *page; |
1da177e4 | 2198 | |
ed11d9eb | 2199 | nr_freed = 0; |
ce8eb6c4 | 2200 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2201 | |
ce8eb6c4 CL |
2202 | spin_lock_irq(&n->list_lock); |
2203 | p = n->slabs_free.prev; | |
2204 | if (p == &n->slabs_free) { | |
2205 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2206 | goto out; |
2207 | } | |
1da177e4 | 2208 | |
16cb0ec7 TH |
2209 | page = list_entry(p, struct page, slab_list); |
2210 | list_del(&page->slab_list); | |
f728b0a5 | 2211 | n->free_slabs--; |
bf00bd34 | 2212 | n->total_slabs--; |
ed11d9eb CL |
2213 | /* |
2214 | * Safe to drop the lock. The slab is no longer linked | |
2215 | * to the cache. | |
2216 | */ | |
ce8eb6c4 CL |
2217 | n->free_objects -= cache->num; |
2218 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2219 | slab_destroy(cache, page); |
ed11d9eb | 2220 | nr_freed++; |
1da177e4 | 2221 | } |
ed11d9eb CL |
2222 | out: |
2223 | return nr_freed; | |
1da177e4 LT |
2224 | } |
2225 | ||
f9e13c0a SB |
2226 | bool __kmem_cache_empty(struct kmem_cache *s) |
2227 | { | |
2228 | int node; | |
2229 | struct kmem_cache_node *n; | |
2230 | ||
2231 | for_each_kmem_cache_node(s, node, n) | |
2232 | if (!list_empty(&n->slabs_full) || | |
2233 | !list_empty(&n->slabs_partial)) | |
2234 | return false; | |
2235 | return true; | |
2236 | } | |
2237 | ||
c9fc5864 | 2238 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2239 | { |
18bf8541 CL |
2240 | int ret = 0; |
2241 | int node; | |
ce8eb6c4 | 2242 | struct kmem_cache_node *n; |
e498be7d CL |
2243 | |
2244 | drain_cpu_caches(cachep); | |
2245 | ||
2246 | check_irq_on(); | |
18bf8541 | 2247 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2248 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2249 | |
ce8eb6c4 CL |
2250 | ret += !list_empty(&n->slabs_full) || |
2251 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2252 | } |
2253 | return (ret ? 1 : 0); | |
2254 | } | |
2255 | ||
c9fc5864 TH |
2256 | #ifdef CONFIG_MEMCG |
2257 | void __kmemcg_cache_deactivate(struct kmem_cache *cachep) | |
2258 | { | |
2259 | __kmem_cache_shrink(cachep); | |
2260 | } | |
43486694 RG |
2261 | |
2262 | void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) | |
2263 | { | |
2264 | } | |
c9fc5864 TH |
2265 | #endif |
2266 | ||
945cf2b6 | 2267 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 | 2268 | { |
c9fc5864 | 2269 | return __kmem_cache_shrink(cachep); |
52b4b950 DS |
2270 | } |
2271 | ||
2272 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2273 | { |
12c3667f | 2274 | int i; |
ce8eb6c4 | 2275 | struct kmem_cache_node *n; |
1da177e4 | 2276 | |
c7ce4f60 TG |
2277 | cache_random_seq_destroy(cachep); |
2278 | ||
bf0dea23 | 2279 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2280 | |
ce8eb6c4 | 2281 | /* NUMA: free the node structures */ |
18bf8541 CL |
2282 | for_each_kmem_cache_node(cachep, i, n) { |
2283 | kfree(n->shared); | |
2284 | free_alien_cache(n->alien); | |
2285 | kfree(n); | |
2286 | cachep->node[i] = NULL; | |
12c3667f | 2287 | } |
1da177e4 | 2288 | } |
1da177e4 | 2289 | |
e5ac9c5a RT |
2290 | /* |
2291 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2292 | * |
2293 | * For a slab cache when the slab descriptor is off-slab, the | |
2294 | * slab descriptor can't come from the same cache which is being created, | |
2295 | * Because if it is the case, that means we defer the creation of | |
2296 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2297 | * And we eventually call down to __kmem_cache_create(), which | |
2298 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2299 | * This is a "chicken-and-egg" problem. | |
2300 | * | |
2301 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2302 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2303 | */ |
7e007355 | 2304 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2305 | struct page *page, int colour_off, |
2306 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2307 | { |
7e007355 | 2308 | void *freelist; |
0c3aa83e | 2309 | void *addr = page_address(page); |
b28a02de | 2310 | |
51dedad0 | 2311 | page->s_mem = addr + colour_off; |
2e6b3602 JK |
2312 | page->active = 0; |
2313 | ||
b03a017b JK |
2314 | if (OBJFREELIST_SLAB(cachep)) |
2315 | freelist = NULL; | |
2316 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2317 | /* Slab management obj is off-slab. */ |
8456a648 | 2318 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2319 | local_flags, nodeid); |
8456a648 | 2320 | if (!freelist) |
1da177e4 LT |
2321 | return NULL; |
2322 | } else { | |
2e6b3602 JK |
2323 | /* We will use last bytes at the slab for freelist */ |
2324 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2325 | cachep->freelist_size; | |
1da177e4 | 2326 | } |
2e6b3602 | 2327 | |
8456a648 | 2328 | return freelist; |
1da177e4 LT |
2329 | } |
2330 | ||
7cc68973 | 2331 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2332 | { |
a41adfaa | 2333 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2334 | } |
2335 | ||
2336 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2337 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2338 | { |
a41adfaa | 2339 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2340 | } |
2341 | ||
10b2e9e8 | 2342 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2343 | { |
10b2e9e8 | 2344 | #if DEBUG |
1da177e4 LT |
2345 | int i; |
2346 | ||
2347 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2348 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2349 | |
1da177e4 LT |
2350 | if (cachep->flags & SLAB_STORE_USER) |
2351 | *dbg_userword(cachep, objp) = NULL; | |
2352 | ||
2353 | if (cachep->flags & SLAB_RED_ZONE) { | |
2354 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2355 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2356 | } | |
2357 | /* | |
a737b3e2 AM |
2358 | * Constructors are not allowed to allocate memory from the same |
2359 | * cache which they are a constructor for. Otherwise, deadlock. | |
2360 | * They must also be threaded. | |
1da177e4 | 2361 | */ |
7ed2f9e6 AP |
2362 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2363 | kasan_unpoison_object_data(cachep, | |
2364 | objp + obj_offset(cachep)); | |
51cc5068 | 2365 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2366 | kasan_poison_object_data( |
2367 | cachep, objp + obj_offset(cachep)); | |
2368 | } | |
1da177e4 LT |
2369 | |
2370 | if (cachep->flags & SLAB_RED_ZONE) { | |
2371 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2372 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2373 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2374 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2375 | } |
40b44137 JK |
2376 | /* need to poison the objs? */ |
2377 | if (cachep->flags & SLAB_POISON) { | |
2378 | poison_obj(cachep, objp, POISON_FREE); | |
80552f0f | 2379 | slab_kernel_map(cachep, objp, 0); |
40b44137 | 2380 | } |
10b2e9e8 | 2381 | } |
1da177e4 | 2382 | #endif |
10b2e9e8 JK |
2383 | } |
2384 | ||
c7ce4f60 TG |
2385 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
2386 | /* Hold information during a freelist initialization */ | |
2387 | union freelist_init_state { | |
2388 | struct { | |
2389 | unsigned int pos; | |
7c00fce9 | 2390 | unsigned int *list; |
c7ce4f60 | 2391 | unsigned int count; |
c7ce4f60 TG |
2392 | }; |
2393 | struct rnd_state rnd_state; | |
2394 | }; | |
2395 | ||
2396 | /* | |
2397 | * Initialize the state based on the randomization methode available. | |
2398 | * return true if the pre-computed list is available, false otherwize. | |
2399 | */ | |
2400 | static bool freelist_state_initialize(union freelist_init_state *state, | |
2401 | struct kmem_cache *cachep, | |
2402 | unsigned int count) | |
2403 | { | |
2404 | bool ret; | |
2405 | unsigned int rand; | |
2406 | ||
2407 | /* Use best entropy available to define a random shift */ | |
7c00fce9 | 2408 | rand = get_random_int(); |
c7ce4f60 TG |
2409 | |
2410 | /* Use a random state if the pre-computed list is not available */ | |
2411 | if (!cachep->random_seq) { | |
2412 | prandom_seed_state(&state->rnd_state, rand); | |
2413 | ret = false; | |
2414 | } else { | |
2415 | state->list = cachep->random_seq; | |
2416 | state->count = count; | |
c4e490cf | 2417 | state->pos = rand % count; |
c7ce4f60 TG |
2418 | ret = true; |
2419 | } | |
2420 | return ret; | |
2421 | } | |
2422 | ||
2423 | /* Get the next entry on the list and randomize it using a random shift */ | |
2424 | static freelist_idx_t next_random_slot(union freelist_init_state *state) | |
2425 | { | |
c4e490cf JS |
2426 | if (state->pos >= state->count) |
2427 | state->pos = 0; | |
2428 | return state->list[state->pos++]; | |
c7ce4f60 TG |
2429 | } |
2430 | ||
7c00fce9 TG |
2431 | /* Swap two freelist entries */ |
2432 | static void swap_free_obj(struct page *page, unsigned int a, unsigned int b) | |
2433 | { | |
2434 | swap(((freelist_idx_t *)page->freelist)[a], | |
2435 | ((freelist_idx_t *)page->freelist)[b]); | |
2436 | } | |
2437 | ||
c7ce4f60 TG |
2438 | /* |
2439 | * Shuffle the freelist initialization state based on pre-computed lists. | |
2440 | * return true if the list was successfully shuffled, false otherwise. | |
2441 | */ | |
2442 | static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page) | |
2443 | { | |
7c00fce9 | 2444 | unsigned int objfreelist = 0, i, rand, count = cachep->num; |
c7ce4f60 TG |
2445 | union freelist_init_state state; |
2446 | bool precomputed; | |
2447 | ||
2448 | if (count < 2) | |
2449 | return false; | |
2450 | ||
2451 | precomputed = freelist_state_initialize(&state, cachep, count); | |
2452 | ||
2453 | /* Take a random entry as the objfreelist */ | |
2454 | if (OBJFREELIST_SLAB(cachep)) { | |
2455 | if (!precomputed) | |
2456 | objfreelist = count - 1; | |
2457 | else | |
2458 | objfreelist = next_random_slot(&state); | |
2459 | page->freelist = index_to_obj(cachep, page, objfreelist) + | |
2460 | obj_offset(cachep); | |
2461 | count--; | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * On early boot, generate the list dynamically. | |
2466 | * Later use a pre-computed list for speed. | |
2467 | */ | |
2468 | if (!precomputed) { | |
7c00fce9 TG |
2469 | for (i = 0; i < count; i++) |
2470 | set_free_obj(page, i, i); | |
2471 | ||
2472 | /* Fisher-Yates shuffle */ | |
2473 | for (i = count - 1; i > 0; i--) { | |
2474 | rand = prandom_u32_state(&state.rnd_state); | |
2475 | rand %= (i + 1); | |
2476 | swap_free_obj(page, i, rand); | |
2477 | } | |
c7ce4f60 TG |
2478 | } else { |
2479 | for (i = 0; i < count; i++) | |
2480 | set_free_obj(page, i, next_random_slot(&state)); | |
2481 | } | |
2482 | ||
2483 | if (OBJFREELIST_SLAB(cachep)) | |
2484 | set_free_obj(page, cachep->num - 1, objfreelist); | |
2485 | ||
2486 | return true; | |
2487 | } | |
2488 | #else | |
2489 | static inline bool shuffle_freelist(struct kmem_cache *cachep, | |
2490 | struct page *page) | |
2491 | { | |
2492 | return false; | |
2493 | } | |
2494 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
2495 | ||
10b2e9e8 JK |
2496 | static void cache_init_objs(struct kmem_cache *cachep, |
2497 | struct page *page) | |
2498 | { | |
2499 | int i; | |
7ed2f9e6 | 2500 | void *objp; |
c7ce4f60 | 2501 | bool shuffled; |
10b2e9e8 JK |
2502 | |
2503 | cache_init_objs_debug(cachep, page); | |
2504 | ||
c7ce4f60 TG |
2505 | /* Try to randomize the freelist if enabled */ |
2506 | shuffled = shuffle_freelist(cachep, page); | |
2507 | ||
2508 | if (!shuffled && OBJFREELIST_SLAB(cachep)) { | |
b03a017b JK |
2509 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + |
2510 | obj_offset(cachep); | |
2511 | } | |
2512 | ||
10b2e9e8 | 2513 | for (i = 0; i < cachep->num; i++) { |
b3cbd9bf | 2514 | objp = index_to_obj(cachep, page, i); |
4d176711 | 2515 | objp = kasan_init_slab_obj(cachep, objp); |
b3cbd9bf | 2516 | |
10b2e9e8 | 2517 | /* constructor could break poison info */ |
7ed2f9e6 | 2518 | if (DEBUG == 0 && cachep->ctor) { |
7ed2f9e6 AP |
2519 | kasan_unpoison_object_data(cachep, objp); |
2520 | cachep->ctor(objp); | |
2521 | kasan_poison_object_data(cachep, objp); | |
2522 | } | |
10b2e9e8 | 2523 | |
c7ce4f60 TG |
2524 | if (!shuffled) |
2525 | set_free_obj(page, i, i); | |
1da177e4 | 2526 | } |
1da177e4 LT |
2527 | } |
2528 | ||
260b61dd | 2529 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2530 | { |
b1cb0982 | 2531 | void *objp; |
78d382d7 | 2532 | |
e5c58dfd | 2533 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2534 | page->active++; |
78d382d7 MD |
2535 | |
2536 | return objp; | |
2537 | } | |
2538 | ||
260b61dd JK |
2539 | static void slab_put_obj(struct kmem_cache *cachep, |
2540 | struct page *page, void *objp) | |
78d382d7 | 2541 | { |
8456a648 | 2542 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2543 | #if DEBUG |
16025177 | 2544 | unsigned int i; |
b1cb0982 | 2545 | |
b1cb0982 | 2546 | /* Verify double free bug */ |
8456a648 | 2547 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2548 | if (get_free_obj(page, i) == objnr) { |
85c3e4a5 | 2549 | pr_err("slab: double free detected in cache '%s', objp %px\n", |
756a025f | 2550 | cachep->name, objp); |
b1cb0982 JK |
2551 | BUG(); |
2552 | } | |
78d382d7 MD |
2553 | } |
2554 | #endif | |
8456a648 | 2555 | page->active--; |
b03a017b JK |
2556 | if (!page->freelist) |
2557 | page->freelist = objp + obj_offset(cachep); | |
2558 | ||
e5c58dfd | 2559 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2560 | } |
2561 | ||
4776874f PE |
2562 | /* |
2563 | * Map pages beginning at addr to the given cache and slab. This is required | |
2564 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2565 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2566 | */ |
8456a648 | 2567 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2568 | void *freelist) |
1da177e4 | 2569 | { |
a57a4988 | 2570 | page->slab_cache = cache; |
8456a648 | 2571 | page->freelist = freelist; |
1da177e4 LT |
2572 | } |
2573 | ||
2574 | /* | |
2575 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2576 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2577 | */ | |
76b342bd JK |
2578 | static struct page *cache_grow_begin(struct kmem_cache *cachep, |
2579 | gfp_t flags, int nodeid) | |
1da177e4 | 2580 | { |
7e007355 | 2581 | void *freelist; |
b28a02de PE |
2582 | size_t offset; |
2583 | gfp_t local_flags; | |
511e3a05 | 2584 | int page_node; |
ce8eb6c4 | 2585 | struct kmem_cache_node *n; |
511e3a05 | 2586 | struct page *page; |
1da177e4 | 2587 | |
a737b3e2 AM |
2588 | /* |
2589 | * Be lazy and only check for valid flags here, keeping it out of the | |
2590 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2591 | */ |
c871ac4e | 2592 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
bacdcb34 | 2593 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
2594 | flags &= ~GFP_SLAB_BUG_MASK; |
2595 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
2596 | invalid_mask, &invalid_mask, flags, &flags); | |
2597 | dump_stack(); | |
c871ac4e | 2598 | } |
128227e7 | 2599 | WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO)); |
6cb06229 | 2600 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2601 | |
1da177e4 | 2602 | check_irq_off(); |
d0164adc | 2603 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2604 | local_irq_enable(); |
2605 | ||
a737b3e2 AM |
2606 | /* |
2607 | * Get mem for the objs. Attempt to allocate a physical page from | |
2608 | * 'nodeid'. | |
e498be7d | 2609 | */ |
511e3a05 | 2610 | page = kmem_getpages(cachep, local_flags, nodeid); |
0c3aa83e | 2611 | if (!page) |
1da177e4 LT |
2612 | goto failed; |
2613 | ||
511e3a05 JK |
2614 | page_node = page_to_nid(page); |
2615 | n = get_node(cachep, page_node); | |
03d1d43a JK |
2616 | |
2617 | /* Get colour for the slab, and cal the next value. */ | |
2618 | n->colour_next++; | |
2619 | if (n->colour_next >= cachep->colour) | |
2620 | n->colour_next = 0; | |
2621 | ||
2622 | offset = n->colour_next; | |
2623 | if (offset >= cachep->colour) | |
2624 | offset = 0; | |
2625 | ||
2626 | offset *= cachep->colour_off; | |
2627 | ||
51dedad0 AK |
2628 | /* |
2629 | * Call kasan_poison_slab() before calling alloc_slabmgmt(), so | |
2630 | * page_address() in the latter returns a non-tagged pointer, | |
2631 | * as it should be for slab pages. | |
2632 | */ | |
2633 | kasan_poison_slab(page); | |
2634 | ||
1da177e4 | 2635 | /* Get slab management. */ |
8456a648 | 2636 | freelist = alloc_slabmgmt(cachep, page, offset, |
511e3a05 | 2637 | local_flags & ~GFP_CONSTRAINT_MASK, page_node); |
b03a017b | 2638 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2639 | goto opps1; |
2640 | ||
8456a648 | 2641 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2642 | |
8456a648 | 2643 | cache_init_objs(cachep, page); |
1da177e4 | 2644 | |
d0164adc | 2645 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2646 | local_irq_disable(); |
1da177e4 | 2647 | |
76b342bd JK |
2648 | return page; |
2649 | ||
a737b3e2 | 2650 | opps1: |
0c3aa83e | 2651 | kmem_freepages(cachep, page); |
a737b3e2 | 2652 | failed: |
d0164adc | 2653 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2654 | local_irq_disable(); |
76b342bd JK |
2655 | return NULL; |
2656 | } | |
2657 | ||
2658 | static void cache_grow_end(struct kmem_cache *cachep, struct page *page) | |
2659 | { | |
2660 | struct kmem_cache_node *n; | |
2661 | void *list = NULL; | |
2662 | ||
2663 | check_irq_off(); | |
2664 | ||
2665 | if (!page) | |
2666 | return; | |
2667 | ||
16cb0ec7 | 2668 | INIT_LIST_HEAD(&page->slab_list); |
76b342bd JK |
2669 | n = get_node(cachep, page_to_nid(page)); |
2670 | ||
2671 | spin_lock(&n->list_lock); | |
bf00bd34 | 2672 | n->total_slabs++; |
f728b0a5 | 2673 | if (!page->active) { |
16cb0ec7 | 2674 | list_add_tail(&page->slab_list, &n->slabs_free); |
f728b0a5 | 2675 | n->free_slabs++; |
bf00bd34 | 2676 | } else |
76b342bd | 2677 | fixup_slab_list(cachep, n, page, &list); |
07a63c41 | 2678 | |
76b342bd JK |
2679 | STATS_INC_GROWN(cachep); |
2680 | n->free_objects += cachep->num - page->active; | |
2681 | spin_unlock(&n->list_lock); | |
2682 | ||
2683 | fixup_objfreelist_debug(cachep, &list); | |
1da177e4 LT |
2684 | } |
2685 | ||
2686 | #if DEBUG | |
2687 | ||
2688 | /* | |
2689 | * Perform extra freeing checks: | |
2690 | * - detect bad pointers. | |
2691 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2692 | */ |
2693 | static void kfree_debugcheck(const void *objp) | |
2694 | { | |
1da177e4 | 2695 | if (!virt_addr_valid(objp)) { |
1170532b | 2696 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2697 | (unsigned long)objp); |
2698 | BUG(); | |
1da177e4 | 2699 | } |
1da177e4 LT |
2700 | } |
2701 | ||
58ce1fd5 PE |
2702 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2703 | { | |
b46b8f19 | 2704 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2705 | |
2706 | redzone1 = *dbg_redzone1(cache, obj); | |
2707 | redzone2 = *dbg_redzone2(cache, obj); | |
2708 | ||
2709 | /* | |
2710 | * Redzone is ok. | |
2711 | */ | |
2712 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2713 | return; | |
2714 | ||
2715 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2716 | slab_error(cache, "double free detected"); | |
2717 | else | |
2718 | slab_error(cache, "memory outside object was overwritten"); | |
2719 | ||
85c3e4a5 | 2720 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b | 2721 | obj, redzone1, redzone2); |
58ce1fd5 PE |
2722 | } |
2723 | ||
343e0d7a | 2724 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2725 | unsigned long caller) |
1da177e4 | 2726 | { |
1da177e4 | 2727 | unsigned int objnr; |
8456a648 | 2728 | struct page *page; |
1da177e4 | 2729 | |
80cbd911 MW |
2730 | BUG_ON(virt_to_cache(objp) != cachep); |
2731 | ||
3dafccf2 | 2732 | objp -= obj_offset(cachep); |
1da177e4 | 2733 | kfree_debugcheck(objp); |
b49af68f | 2734 | page = virt_to_head_page(objp); |
1da177e4 | 2735 | |
1da177e4 | 2736 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2737 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2738 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2739 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2740 | } | |
7878c231 | 2741 | if (cachep->flags & SLAB_STORE_USER) |
7c0cb9c6 | 2742 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2743 | |
8456a648 | 2744 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2745 | |
2746 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2747 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2748 | |
1da177e4 | 2749 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2750 | poison_obj(cachep, objp, POISON_FREE); |
80552f0f | 2751 | slab_kernel_map(cachep, objp, 0); |
1da177e4 LT |
2752 | } |
2753 | return objp; | |
2754 | } | |
2755 | ||
1da177e4 LT |
2756 | #else |
2757 | #define kfree_debugcheck(x) do { } while(0) | |
2758 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2759 | #endif |
2760 | ||
b03a017b JK |
2761 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2762 | void **list) | |
2763 | { | |
2764 | #if DEBUG | |
2765 | void *next = *list; | |
2766 | void *objp; | |
2767 | ||
2768 | while (next) { | |
2769 | objp = next - obj_offset(cachep); | |
2770 | next = *(void **)next; | |
2771 | poison_obj(cachep, objp, POISON_FREE); | |
2772 | } | |
2773 | #endif | |
2774 | } | |
2775 | ||
d8410234 | 2776 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2777 | struct kmem_cache_node *n, struct page *page, |
2778 | void **list) | |
d8410234 JK |
2779 | { |
2780 | /* move slabp to correct slabp list: */ | |
16cb0ec7 | 2781 | list_del(&page->slab_list); |
b03a017b | 2782 | if (page->active == cachep->num) { |
16cb0ec7 | 2783 | list_add(&page->slab_list, &n->slabs_full); |
b03a017b JK |
2784 | if (OBJFREELIST_SLAB(cachep)) { |
2785 | #if DEBUG | |
2786 | /* Poisoning will be done without holding the lock */ | |
2787 | if (cachep->flags & SLAB_POISON) { | |
2788 | void **objp = page->freelist; | |
2789 | ||
2790 | *objp = *list; | |
2791 | *list = objp; | |
2792 | } | |
2793 | #endif | |
2794 | page->freelist = NULL; | |
2795 | } | |
2796 | } else | |
16cb0ec7 | 2797 | list_add(&page->slab_list, &n->slabs_partial); |
d8410234 JK |
2798 | } |
2799 | ||
f68f8ddd JK |
2800 | /* Try to find non-pfmemalloc slab if needed */ |
2801 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
bf00bd34 | 2802 | struct page *page, bool pfmemalloc) |
f68f8ddd JK |
2803 | { |
2804 | if (!page) | |
2805 | return NULL; | |
2806 | ||
2807 | if (pfmemalloc) | |
2808 | return page; | |
2809 | ||
2810 | if (!PageSlabPfmemalloc(page)) | |
2811 | return page; | |
2812 | ||
2813 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2814 | if (n->free_objects > n->free_limit) { | |
2815 | ClearPageSlabPfmemalloc(page); | |
2816 | return page; | |
2817 | } | |
2818 | ||
2819 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
16cb0ec7 | 2820 | list_del(&page->slab_list); |
bf00bd34 | 2821 | if (!page->active) { |
16cb0ec7 | 2822 | list_add_tail(&page->slab_list, &n->slabs_free); |
bf00bd34 | 2823 | n->free_slabs++; |
f728b0a5 | 2824 | } else |
16cb0ec7 | 2825 | list_add_tail(&page->slab_list, &n->slabs_partial); |
f68f8ddd | 2826 | |
16cb0ec7 | 2827 | list_for_each_entry(page, &n->slabs_partial, slab_list) { |
f68f8ddd JK |
2828 | if (!PageSlabPfmemalloc(page)) |
2829 | return page; | |
2830 | } | |
2831 | ||
f728b0a5 | 2832 | n->free_touched = 1; |
16cb0ec7 | 2833 | list_for_each_entry(page, &n->slabs_free, slab_list) { |
f728b0a5 | 2834 | if (!PageSlabPfmemalloc(page)) { |
bf00bd34 | 2835 | n->free_slabs--; |
f68f8ddd | 2836 | return page; |
f728b0a5 | 2837 | } |
f68f8ddd JK |
2838 | } |
2839 | ||
2840 | return NULL; | |
2841 | } | |
2842 | ||
2843 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2844 | { |
2845 | struct page *page; | |
2846 | ||
f728b0a5 | 2847 | assert_spin_locked(&n->list_lock); |
16cb0ec7 TH |
2848 | page = list_first_entry_or_null(&n->slabs_partial, struct page, |
2849 | slab_list); | |
7aa0d227 GT |
2850 | if (!page) { |
2851 | n->free_touched = 1; | |
bf00bd34 | 2852 | page = list_first_entry_or_null(&n->slabs_free, struct page, |
16cb0ec7 | 2853 | slab_list); |
f728b0a5 | 2854 | if (page) |
bf00bd34 | 2855 | n->free_slabs--; |
7aa0d227 GT |
2856 | } |
2857 | ||
f68f8ddd | 2858 | if (sk_memalloc_socks()) |
bf00bd34 | 2859 | page = get_valid_first_slab(n, page, pfmemalloc); |
f68f8ddd | 2860 | |
7aa0d227 GT |
2861 | return page; |
2862 | } | |
2863 | ||
f68f8ddd JK |
2864 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2865 | struct kmem_cache_node *n, gfp_t flags) | |
2866 | { | |
2867 | struct page *page; | |
2868 | void *obj; | |
2869 | void *list = NULL; | |
2870 | ||
2871 | if (!gfp_pfmemalloc_allowed(flags)) | |
2872 | return NULL; | |
2873 | ||
2874 | spin_lock(&n->list_lock); | |
2875 | page = get_first_slab(n, true); | |
2876 | if (!page) { | |
2877 | spin_unlock(&n->list_lock); | |
2878 | return NULL; | |
2879 | } | |
2880 | ||
2881 | obj = slab_get_obj(cachep, page); | |
2882 | n->free_objects--; | |
2883 | ||
2884 | fixup_slab_list(cachep, n, page, &list); | |
2885 | ||
2886 | spin_unlock(&n->list_lock); | |
2887 | fixup_objfreelist_debug(cachep, &list); | |
2888 | ||
2889 | return obj; | |
2890 | } | |
2891 | ||
213b4695 JK |
2892 | /* |
2893 | * Slab list should be fixed up by fixup_slab_list() for existing slab | |
2894 | * or cache_grow_end() for new slab | |
2895 | */ | |
2896 | static __always_inline int alloc_block(struct kmem_cache *cachep, | |
2897 | struct array_cache *ac, struct page *page, int batchcount) | |
2898 | { | |
2899 | /* | |
2900 | * There must be at least one object available for | |
2901 | * allocation. | |
2902 | */ | |
2903 | BUG_ON(page->active >= cachep->num); | |
2904 | ||
2905 | while (page->active < cachep->num && batchcount--) { | |
2906 | STATS_INC_ALLOCED(cachep); | |
2907 | STATS_INC_ACTIVE(cachep); | |
2908 | STATS_SET_HIGH(cachep); | |
2909 | ||
2910 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); | |
2911 | } | |
2912 | ||
2913 | return batchcount; | |
2914 | } | |
2915 | ||
f68f8ddd | 2916 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2917 | { |
2918 | int batchcount; | |
ce8eb6c4 | 2919 | struct kmem_cache_node *n; |
801faf0d | 2920 | struct array_cache *ac, *shared; |
1ca4cb24 | 2921 | int node; |
b03a017b | 2922 | void *list = NULL; |
76b342bd | 2923 | struct page *page; |
1ca4cb24 | 2924 | |
1da177e4 | 2925 | check_irq_off(); |
7d6e6d09 | 2926 | node = numa_mem_id(); |
f68f8ddd | 2927 | |
9a2dba4b | 2928 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2929 | batchcount = ac->batchcount; |
2930 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2931 | /* |
2932 | * If there was little recent activity on this cache, then | |
2933 | * perform only a partial refill. Otherwise we could generate | |
2934 | * refill bouncing. | |
1da177e4 LT |
2935 | */ |
2936 | batchcount = BATCHREFILL_LIMIT; | |
2937 | } | |
18bf8541 | 2938 | n = get_node(cachep, node); |
e498be7d | 2939 | |
ce8eb6c4 | 2940 | BUG_ON(ac->avail > 0 || !n); |
801faf0d JK |
2941 | shared = READ_ONCE(n->shared); |
2942 | if (!n->free_objects && (!shared || !shared->avail)) | |
2943 | goto direct_grow; | |
2944 | ||
ce8eb6c4 | 2945 | spin_lock(&n->list_lock); |
801faf0d | 2946 | shared = READ_ONCE(n->shared); |
1da177e4 | 2947 | |
3ded175a | 2948 | /* See if we can refill from the shared array */ |
801faf0d JK |
2949 | if (shared && transfer_objects(ac, shared, batchcount)) { |
2950 | shared->touched = 1; | |
3ded175a | 2951 | goto alloc_done; |
44b57f1c | 2952 | } |
3ded175a | 2953 | |
1da177e4 | 2954 | while (batchcount > 0) { |
1da177e4 | 2955 | /* Get slab alloc is to come from. */ |
f68f8ddd | 2956 | page = get_first_slab(n, false); |
7aa0d227 GT |
2957 | if (!page) |
2958 | goto must_grow; | |
1da177e4 | 2959 | |
1da177e4 | 2960 | check_spinlock_acquired(cachep); |
714b8171 | 2961 | |
213b4695 | 2962 | batchcount = alloc_block(cachep, ac, page, batchcount); |
b03a017b | 2963 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
2964 | } |
2965 | ||
a737b3e2 | 2966 | must_grow: |
ce8eb6c4 | 2967 | n->free_objects -= ac->avail; |
a737b3e2 | 2968 | alloc_done: |
ce8eb6c4 | 2969 | spin_unlock(&n->list_lock); |
b03a017b | 2970 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 | 2971 | |
801faf0d | 2972 | direct_grow: |
1da177e4 | 2973 | if (unlikely(!ac->avail)) { |
f68f8ddd JK |
2974 | /* Check if we can use obj in pfmemalloc slab */ |
2975 | if (sk_memalloc_socks()) { | |
2976 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
2977 | ||
2978 | if (obj) | |
2979 | return obj; | |
2980 | } | |
2981 | ||
76b342bd | 2982 | page = cache_grow_begin(cachep, gfp_exact_node(flags), node); |
e498be7d | 2983 | |
76b342bd JK |
2984 | /* |
2985 | * cache_grow_begin() can reenable interrupts, | |
2986 | * then ac could change. | |
2987 | */ | |
9a2dba4b | 2988 | ac = cpu_cache_get(cachep); |
213b4695 JK |
2989 | if (!ac->avail && page) |
2990 | alloc_block(cachep, ac, page, batchcount); | |
2991 | cache_grow_end(cachep, page); | |
072bb0aa | 2992 | |
213b4695 | 2993 | if (!ac->avail) |
1da177e4 | 2994 | return NULL; |
1da177e4 LT |
2995 | } |
2996 | ac->touched = 1; | |
072bb0aa | 2997 | |
f68f8ddd | 2998 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2999 | } |
3000 | ||
a737b3e2 AM |
3001 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3002 | gfp_t flags) | |
1da177e4 | 3003 | { |
d0164adc | 3004 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
3005 | } |
3006 | ||
3007 | #if DEBUG | |
a737b3e2 | 3008 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3009 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3010 | { |
128227e7 | 3011 | WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO)); |
b28a02de | 3012 | if (!objp) |
1da177e4 | 3013 | return objp; |
b28a02de | 3014 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3015 | check_poison_obj(cachep, objp); |
80552f0f | 3016 | slab_kernel_map(cachep, objp, 1); |
1da177e4 LT |
3017 | poison_obj(cachep, objp, POISON_INUSE); |
3018 | } | |
3019 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3020 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3021 | |
3022 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3023 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3024 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 3025 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
85c3e4a5 | 3026 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b JP |
3027 | objp, *dbg_redzone1(cachep, objp), |
3028 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3029 | } |
3030 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3031 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3032 | } | |
03787301 | 3033 | |
3dafccf2 | 3034 | objp += obj_offset(cachep); |
4f104934 | 3035 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3036 | cachep->ctor(objp); |
7ea466f2 TH |
3037 | if (ARCH_SLAB_MINALIGN && |
3038 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
85c3e4a5 | 3039 | pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3040 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3041 | } |
1da177e4 LT |
3042 | return objp; |
3043 | } | |
3044 | #else | |
3045 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3046 | #endif | |
3047 | ||
343e0d7a | 3048 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3049 | { |
b28a02de | 3050 | void *objp; |
1da177e4 LT |
3051 | struct array_cache *ac; |
3052 | ||
5c382300 | 3053 | check_irq_off(); |
8a8b6502 | 3054 | |
9a2dba4b | 3055 | ac = cpu_cache_get(cachep); |
1da177e4 | 3056 | if (likely(ac->avail)) { |
1da177e4 | 3057 | ac->touched = 1; |
f68f8ddd | 3058 | objp = ac->entry[--ac->avail]; |
072bb0aa | 3059 | |
f68f8ddd JK |
3060 | STATS_INC_ALLOCHIT(cachep); |
3061 | goto out; | |
1da177e4 | 3062 | } |
072bb0aa MG |
3063 | |
3064 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 3065 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
3066 | /* |
3067 | * the 'ac' may be updated by cache_alloc_refill(), | |
3068 | * and kmemleak_erase() requires its correct value. | |
3069 | */ | |
3070 | ac = cpu_cache_get(cachep); | |
3071 | ||
3072 | out: | |
d5cff635 CM |
3073 | /* |
3074 | * To avoid a false negative, if an object that is in one of the | |
3075 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3076 | * treat the array pointers as a reference to the object. | |
3077 | */ | |
f3d8b53a O |
3078 | if (objp) |
3079 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3080 | return objp; |
3081 | } | |
3082 | ||
e498be7d | 3083 | #ifdef CONFIG_NUMA |
c61afb18 | 3084 | /* |
2ad654bc | 3085 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3086 | * |
3087 | * If we are in_interrupt, then process context, including cpusets and | |
3088 | * mempolicy, may not apply and should not be used for allocation policy. | |
3089 | */ | |
3090 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3091 | { | |
3092 | int nid_alloc, nid_here; | |
3093 | ||
765c4507 | 3094 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3095 | return NULL; |
7d6e6d09 | 3096 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3097 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3098 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3099 | else if (current->mempolicy) |
2a389610 | 3100 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3101 | if (nid_alloc != nid_here) |
8b98c169 | 3102 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3103 | return NULL; |
3104 | } | |
3105 | ||
765c4507 CL |
3106 | /* |
3107 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3108 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3109 | * available node for available objects. If that fails then we |
3c517a61 CL |
3110 | * perform an allocation without specifying a node. This allows the page |
3111 | * allocator to do its reclaim / fallback magic. We then insert the | |
3112 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3113 | */ |
8c8cc2c1 | 3114 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3115 | { |
8c8cc2c1 | 3116 | struct zonelist *zonelist; |
dd1a239f | 3117 | struct zoneref *z; |
54a6eb5c | 3118 | struct zone *zone; |
97a225e6 | 3119 | enum zone_type highest_zoneidx = gfp_zone(flags); |
765c4507 | 3120 | void *obj = NULL; |
76b342bd | 3121 | struct page *page; |
3c517a61 | 3122 | int nid; |
cc9a6c87 | 3123 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3124 | |
3125 | if (flags & __GFP_THISNODE) | |
3126 | return NULL; | |
3127 | ||
cc9a6c87 | 3128 | retry_cpuset: |
d26914d1 | 3129 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3130 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3131 | |
3c517a61 CL |
3132 | retry: |
3133 | /* | |
3134 | * Look through allowed nodes for objects available | |
3135 | * from existing per node queues. | |
3136 | */ | |
97a225e6 | 3137 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
54a6eb5c | 3138 | nid = zone_to_nid(zone); |
aedb0eb1 | 3139 | |
061d7074 | 3140 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3141 | get_node(cache, nid) && |
3142 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3143 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3144 | gfp_exact_node(flags), nid); |
481c5346 CL |
3145 | if (obj) |
3146 | break; | |
3147 | } | |
3c517a61 CL |
3148 | } |
3149 | ||
cfce6604 | 3150 | if (!obj) { |
3c517a61 CL |
3151 | /* |
3152 | * This allocation will be performed within the constraints | |
3153 | * of the current cpuset / memory policy requirements. | |
3154 | * We may trigger various forms of reclaim on the allowed | |
3155 | * set and go into memory reserves if necessary. | |
3156 | */ | |
76b342bd JK |
3157 | page = cache_grow_begin(cache, flags, numa_mem_id()); |
3158 | cache_grow_end(cache, page); | |
3159 | if (page) { | |
3160 | nid = page_to_nid(page); | |
511e3a05 JK |
3161 | obj = ____cache_alloc_node(cache, |
3162 | gfp_exact_node(flags), nid); | |
0c3aa83e | 3163 | |
3c517a61 | 3164 | /* |
511e3a05 JK |
3165 | * Another processor may allocate the objects in |
3166 | * the slab since we are not holding any locks. | |
3c517a61 | 3167 | */ |
511e3a05 JK |
3168 | if (!obj) |
3169 | goto retry; | |
3c517a61 | 3170 | } |
aedb0eb1 | 3171 | } |
cc9a6c87 | 3172 | |
d26914d1 | 3173 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3174 | goto retry_cpuset; |
765c4507 CL |
3175 | return obj; |
3176 | } | |
3177 | ||
e498be7d CL |
3178 | /* |
3179 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3180 | */ |
8b98c169 | 3181 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3182 | int nodeid) |
e498be7d | 3183 | { |
8456a648 | 3184 | struct page *page; |
ce8eb6c4 | 3185 | struct kmem_cache_node *n; |
213b4695 | 3186 | void *obj = NULL; |
b03a017b | 3187 | void *list = NULL; |
b28a02de | 3188 | |
7c3fbbdd | 3189 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3190 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3191 | BUG_ON(!n); |
b28a02de | 3192 | |
ca3b9b91 | 3193 | check_irq_off(); |
ce8eb6c4 | 3194 | spin_lock(&n->list_lock); |
f68f8ddd | 3195 | page = get_first_slab(n, false); |
7aa0d227 GT |
3196 | if (!page) |
3197 | goto must_grow; | |
b28a02de | 3198 | |
b28a02de | 3199 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3200 | |
3201 | STATS_INC_NODEALLOCS(cachep); | |
3202 | STATS_INC_ACTIVE(cachep); | |
3203 | STATS_SET_HIGH(cachep); | |
3204 | ||
8456a648 | 3205 | BUG_ON(page->active == cachep->num); |
b28a02de | 3206 | |
260b61dd | 3207 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3208 | n->free_objects--; |
b28a02de | 3209 | |
b03a017b | 3210 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3211 | |
ce8eb6c4 | 3212 | spin_unlock(&n->list_lock); |
b03a017b | 3213 | fixup_objfreelist_debug(cachep, &list); |
213b4695 | 3214 | return obj; |
e498be7d | 3215 | |
a737b3e2 | 3216 | must_grow: |
ce8eb6c4 | 3217 | spin_unlock(&n->list_lock); |
76b342bd | 3218 | page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid); |
213b4695 JK |
3219 | if (page) { |
3220 | /* This slab isn't counted yet so don't update free_objects */ | |
3221 | obj = slab_get_obj(cachep, page); | |
3222 | } | |
76b342bd | 3223 | cache_grow_end(cachep, page); |
1da177e4 | 3224 | |
213b4695 | 3225 | return obj ? obj : fallback_alloc(cachep, flags); |
e498be7d | 3226 | } |
8c8cc2c1 | 3227 | |
8c8cc2c1 | 3228 | static __always_inline void * |
48356303 | 3229 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3230 | unsigned long caller) |
8c8cc2c1 PE |
3231 | { |
3232 | unsigned long save_flags; | |
3233 | void *ptr; | |
7d6e6d09 | 3234 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3235 | |
dcce284a | 3236 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3237 | cachep = slab_pre_alloc_hook(cachep, flags); |
3238 | if (unlikely(!cachep)) | |
824ebef1 AM |
3239 | return NULL; |
3240 | ||
8c8cc2c1 PE |
3241 | cache_alloc_debugcheck_before(cachep, flags); |
3242 | local_irq_save(save_flags); | |
3243 | ||
eacbbae3 | 3244 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3245 | nodeid = slab_node; |
8c8cc2c1 | 3246 | |
18bf8541 | 3247 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3248 | /* Node not bootstrapped yet */ |
3249 | ptr = fallback_alloc(cachep, flags); | |
3250 | goto out; | |
3251 | } | |
3252 | ||
7d6e6d09 | 3253 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3254 | /* |
3255 | * Use the locally cached objects if possible. | |
3256 | * However ____cache_alloc does not allow fallback | |
3257 | * to other nodes. It may fail while we still have | |
3258 | * objects on other nodes available. | |
3259 | */ | |
3260 | ptr = ____cache_alloc(cachep, flags); | |
3261 | if (ptr) | |
3262 | goto out; | |
3263 | } | |
3264 | /* ___cache_alloc_node can fall back to other nodes */ | |
3265 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3266 | out: | |
3267 | local_irq_restore(save_flags); | |
3268 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3269 | ||
6471384a | 3270 | if (unlikely(slab_want_init_on_alloc(flags, cachep)) && ptr) |
d5e3ed66 | 3271 | memset(ptr, 0, cachep->object_size); |
d07dbea4 | 3272 | |
d5e3ed66 | 3273 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3274 | return ptr; |
3275 | } | |
3276 | ||
3277 | static __always_inline void * | |
3278 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3279 | { | |
3280 | void *objp; | |
3281 | ||
2ad654bc | 3282 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3283 | objp = alternate_node_alloc(cache, flags); |
3284 | if (objp) | |
3285 | goto out; | |
3286 | } | |
3287 | objp = ____cache_alloc(cache, flags); | |
3288 | ||
3289 | /* | |
3290 | * We may just have run out of memory on the local node. | |
3291 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3292 | */ | |
7d6e6d09 LS |
3293 | if (!objp) |
3294 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3295 | |
3296 | out: | |
3297 | return objp; | |
3298 | } | |
3299 | #else | |
3300 | ||
3301 | static __always_inline void * | |
3302 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3303 | { | |
3304 | return ____cache_alloc(cachep, flags); | |
3305 | } | |
3306 | ||
3307 | #endif /* CONFIG_NUMA */ | |
3308 | ||
3309 | static __always_inline void * | |
48356303 | 3310 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3311 | { |
3312 | unsigned long save_flags; | |
3313 | void *objp; | |
3314 | ||
dcce284a | 3315 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3316 | cachep = slab_pre_alloc_hook(cachep, flags); |
3317 | if (unlikely(!cachep)) | |
824ebef1 AM |
3318 | return NULL; |
3319 | ||
8c8cc2c1 PE |
3320 | cache_alloc_debugcheck_before(cachep, flags); |
3321 | local_irq_save(save_flags); | |
3322 | objp = __do_cache_alloc(cachep, flags); | |
3323 | local_irq_restore(save_flags); | |
3324 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3325 | prefetchw(objp); | |
3326 | ||
6471384a | 3327 | if (unlikely(slab_want_init_on_alloc(flags, cachep)) && objp) |
d5e3ed66 | 3328 | memset(objp, 0, cachep->object_size); |
d07dbea4 | 3329 | |
d5e3ed66 | 3330 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3331 | return objp; |
3332 | } | |
e498be7d CL |
3333 | |
3334 | /* | |
5f0985bb | 3335 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3336 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3337 | */ |
97654dfa JK |
3338 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3339 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3340 | { |
3341 | int i; | |
25c063fb | 3342 | struct kmem_cache_node *n = get_node(cachep, node); |
6052b788 JK |
3343 | struct page *page; |
3344 | ||
3345 | n->free_objects += nr_objects; | |
1da177e4 LT |
3346 | |
3347 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3348 | void *objp; |
8456a648 | 3349 | struct page *page; |
1da177e4 | 3350 | |
072bb0aa MG |
3351 | objp = objpp[i]; |
3352 | ||
8456a648 | 3353 | page = virt_to_head_page(objp); |
16cb0ec7 | 3354 | list_del(&page->slab_list); |
ff69416e | 3355 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3356 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3357 | STATS_DEC_ACTIVE(cachep); |
1da177e4 LT |
3358 | |
3359 | /* fixup slab chains */ | |
f728b0a5 | 3360 | if (page->active == 0) { |
16cb0ec7 | 3361 | list_add(&page->slab_list, &n->slabs_free); |
f728b0a5 | 3362 | n->free_slabs++; |
f728b0a5 | 3363 | } else { |
1da177e4 LT |
3364 | /* Unconditionally move a slab to the end of the |
3365 | * partial list on free - maximum time for the | |
3366 | * other objects to be freed, too. | |
3367 | */ | |
16cb0ec7 | 3368 | list_add_tail(&page->slab_list, &n->slabs_partial); |
1da177e4 LT |
3369 | } |
3370 | } | |
6052b788 JK |
3371 | |
3372 | while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) { | |
3373 | n->free_objects -= cachep->num; | |
3374 | ||
16cb0ec7 TH |
3375 | page = list_last_entry(&n->slabs_free, struct page, slab_list); |
3376 | list_move(&page->slab_list, list); | |
f728b0a5 | 3377 | n->free_slabs--; |
bf00bd34 | 3378 | n->total_slabs--; |
6052b788 | 3379 | } |
1da177e4 LT |
3380 | } |
3381 | ||
343e0d7a | 3382 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3383 | { |
3384 | int batchcount; | |
ce8eb6c4 | 3385 | struct kmem_cache_node *n; |
7d6e6d09 | 3386 | int node = numa_mem_id(); |
97654dfa | 3387 | LIST_HEAD(list); |
1da177e4 LT |
3388 | |
3389 | batchcount = ac->batchcount; | |
260b61dd | 3390 | |
1da177e4 | 3391 | check_irq_off(); |
18bf8541 | 3392 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3393 | spin_lock(&n->list_lock); |
3394 | if (n->shared) { | |
3395 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3396 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3397 | if (max) { |
3398 | if (batchcount > max) | |
3399 | batchcount = max; | |
e498be7d | 3400 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3401 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3402 | shared_array->avail += batchcount; |
3403 | goto free_done; | |
3404 | } | |
3405 | } | |
3406 | ||
97654dfa | 3407 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3408 | free_done: |
1da177e4 LT |
3409 | #if STATS |
3410 | { | |
3411 | int i = 0; | |
73c0219d | 3412 | struct page *page; |
1da177e4 | 3413 | |
16cb0ec7 | 3414 | list_for_each_entry(page, &n->slabs_free, slab_list) { |
8456a648 | 3415 | BUG_ON(page->active); |
1da177e4 LT |
3416 | |
3417 | i++; | |
1da177e4 LT |
3418 | } |
3419 | STATS_SET_FREEABLE(cachep, i); | |
3420 | } | |
3421 | #endif | |
ce8eb6c4 | 3422 | spin_unlock(&n->list_lock); |
97654dfa | 3423 | slabs_destroy(cachep, &list); |
1da177e4 | 3424 | ac->avail -= batchcount; |
a737b3e2 | 3425 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3426 | } |
3427 | ||
3428 | /* | |
a737b3e2 AM |
3429 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3430 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3431 | */ |
ee3ce779 DV |
3432 | static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp, |
3433 | unsigned long caller) | |
1da177e4 | 3434 | { |
55834c59 | 3435 | /* Put the object into the quarantine, don't touch it for now. */ |
ee3ce779 | 3436 | if (kasan_slab_free(cachep, objp, _RET_IP_)) |
55834c59 AP |
3437 | return; |
3438 | ||
3439 | ___cache_free(cachep, objp, caller); | |
3440 | } | |
1da177e4 | 3441 | |
55834c59 AP |
3442 | void ___cache_free(struct kmem_cache *cachep, void *objp, |
3443 | unsigned long caller) | |
3444 | { | |
3445 | struct array_cache *ac = cpu_cache_get(cachep); | |
7ed2f9e6 | 3446 | |
1da177e4 | 3447 | check_irq_off(); |
6471384a AP |
3448 | if (unlikely(slab_want_init_on_free(cachep))) |
3449 | memset(objp, 0, cachep->object_size); | |
d5cff635 | 3450 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3451 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3452 | |
1807a1aa SS |
3453 | /* |
3454 | * Skip calling cache_free_alien() when the platform is not numa. | |
3455 | * This will avoid cache misses that happen while accessing slabp (which | |
3456 | * is per page memory reference) to get nodeid. Instead use a global | |
3457 | * variable to skip the call, which is mostly likely to be present in | |
3458 | * the cache. | |
3459 | */ | |
b6e68bc1 | 3460 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3461 | return; |
3462 | ||
3d880194 | 3463 | if (ac->avail < ac->limit) { |
1da177e4 | 3464 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3465 | } else { |
3466 | STATS_INC_FREEMISS(cachep); | |
3467 | cache_flusharray(cachep, ac); | |
1da177e4 | 3468 | } |
42c8c99c | 3469 | |
f68f8ddd JK |
3470 | if (sk_memalloc_socks()) { |
3471 | struct page *page = virt_to_head_page(objp); | |
3472 | ||
3473 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3474 | cache_free_pfmemalloc(cachep, page, objp); | |
3475 | return; | |
3476 | } | |
3477 | } | |
3478 | ||
dabc3e29 | 3479 | __free_one(ac, objp); |
1da177e4 LT |
3480 | } |
3481 | ||
3482 | /** | |
3483 | * kmem_cache_alloc - Allocate an object | |
3484 | * @cachep: The cache to allocate from. | |
3485 | * @flags: See kmalloc(). | |
3486 | * | |
3487 | * Allocate an object from this cache. The flags are only relevant | |
3488 | * if the cache has no available objects. | |
a862f68a MR |
3489 | * |
3490 | * Return: pointer to the new object or %NULL in case of error | |
1da177e4 | 3491 | */ |
343e0d7a | 3492 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3493 | { |
48356303 | 3494 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3495 | |
ca2b84cb | 3496 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3497 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3498 | |
3499 | return ret; | |
1da177e4 LT |
3500 | } |
3501 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3502 | ||
7b0501dd JDB |
3503 | static __always_inline void |
3504 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3505 | size_t size, void **p, unsigned long caller) | |
3506 | { | |
3507 | size_t i; | |
3508 | ||
3509 | for (i = 0; i < size; i++) | |
3510 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3511 | } | |
3512 | ||
865762a8 | 3513 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3514 | void **p) |
484748f0 | 3515 | { |
2a777eac JDB |
3516 | size_t i; |
3517 | ||
3518 | s = slab_pre_alloc_hook(s, flags); | |
3519 | if (!s) | |
3520 | return 0; | |
3521 | ||
3522 | cache_alloc_debugcheck_before(s, flags); | |
3523 | ||
3524 | local_irq_disable(); | |
3525 | for (i = 0; i < size; i++) { | |
3526 | void *objp = __do_cache_alloc(s, flags); | |
3527 | ||
2a777eac JDB |
3528 | if (unlikely(!objp)) |
3529 | goto error; | |
3530 | p[i] = objp; | |
3531 | } | |
3532 | local_irq_enable(); | |
3533 | ||
7b0501dd JDB |
3534 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3535 | ||
2a777eac | 3536 | /* Clear memory outside IRQ disabled section */ |
6471384a | 3537 | if (unlikely(slab_want_init_on_alloc(flags, s))) |
2a777eac JDB |
3538 | for (i = 0; i < size; i++) |
3539 | memset(p[i], 0, s->object_size); | |
3540 | ||
3541 | slab_post_alloc_hook(s, flags, size, p); | |
3542 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3543 | return size; | |
3544 | error: | |
3545 | local_irq_enable(); | |
7b0501dd | 3546 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3547 | slab_post_alloc_hook(s, flags, i, p); |
3548 | __kmem_cache_free_bulk(s, i, p); | |
3549 | return 0; | |
484748f0 CL |
3550 | } |
3551 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3552 | ||
0f24f128 | 3553 | #ifdef CONFIG_TRACING |
85beb586 | 3554 | void * |
4052147c | 3555 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3556 | { |
85beb586 SR |
3557 | void *ret; |
3558 | ||
48356303 | 3559 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3560 | |
0116523c | 3561 | ret = kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3562 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3563 | size, cachep->size, flags); |
85beb586 | 3564 | return ret; |
36555751 | 3565 | } |
85beb586 | 3566 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3567 | #endif |
3568 | ||
1da177e4 | 3569 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3570 | /** |
3571 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3572 | * @cachep: The cache to allocate from. | |
3573 | * @flags: See kmalloc(). | |
3574 | * @nodeid: node number of the target node. | |
3575 | * | |
3576 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3577 | * node, which can improve the performance for cpu bound structures. | |
3578 | * | |
3579 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
a862f68a MR |
3580 | * |
3581 | * Return: pointer to the new object or %NULL in case of error | |
d0d04b78 | 3582 | */ |
8b98c169 CH |
3583 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3584 | { | |
48356303 | 3585 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3586 | |
ca2b84cb | 3587 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3588 | cachep->object_size, cachep->size, |
ca2b84cb | 3589 | flags, nodeid); |
36555751 EGM |
3590 | |
3591 | return ret; | |
8b98c169 | 3592 | } |
1da177e4 LT |
3593 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3594 | ||
0f24f128 | 3595 | #ifdef CONFIG_TRACING |
4052147c | 3596 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3597 | gfp_t flags, |
4052147c EG |
3598 | int nodeid, |
3599 | size_t size) | |
36555751 | 3600 | { |
85beb586 SR |
3601 | void *ret; |
3602 | ||
592f4145 | 3603 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb | 3604 | |
0116523c | 3605 | ret = kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3606 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3607 | size, cachep->size, |
85beb586 SR |
3608 | flags, nodeid); |
3609 | return ret; | |
36555751 | 3610 | } |
85beb586 | 3611 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3612 | #endif |
3613 | ||
8b98c169 | 3614 | static __always_inline void * |
7c0cb9c6 | 3615 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3616 | { |
343e0d7a | 3617 | struct kmem_cache *cachep; |
7ed2f9e6 | 3618 | void *ret; |
97e2bde4 | 3619 | |
61448479 DV |
3620 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3621 | return NULL; | |
2c59dd65 | 3622 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3623 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3624 | return cachep; | |
7ed2f9e6 | 3625 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
0116523c | 3626 | ret = kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3627 | |
3628 | return ret; | |
97e2bde4 | 3629 | } |
8b98c169 | 3630 | |
8b98c169 CH |
3631 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3632 | { | |
7c0cb9c6 | 3633 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3634 | } |
dbe5e69d | 3635 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3636 | |
3637 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3638 | int node, unsigned long caller) |
8b98c169 | 3639 | { |
7c0cb9c6 | 3640 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3641 | } |
3642 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3643 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3644 | |
3645 | /** | |
800590f5 | 3646 | * __do_kmalloc - allocate memory |
1da177e4 | 3647 | * @size: how many bytes of memory are required. |
800590f5 | 3648 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3649 | * @caller: function caller for debug tracking of the caller |
a862f68a MR |
3650 | * |
3651 | * Return: pointer to the allocated memory or %NULL in case of error | |
1da177e4 | 3652 | */ |
7fd6b141 | 3653 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3654 | unsigned long caller) |
1da177e4 | 3655 | { |
343e0d7a | 3656 | struct kmem_cache *cachep; |
36555751 | 3657 | void *ret; |
1da177e4 | 3658 | |
61448479 DV |
3659 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3660 | return NULL; | |
2c59dd65 | 3661 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3662 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3663 | return cachep; | |
48356303 | 3664 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3665 | |
0116523c | 3666 | ret = kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3667 | trace_kmalloc(caller, ret, |
3b0efdfa | 3668 | size, cachep->size, flags); |
36555751 EGM |
3669 | |
3670 | return ret; | |
7fd6b141 PE |
3671 | } |
3672 | ||
7fd6b141 PE |
3673 | void *__kmalloc(size_t size, gfp_t flags) |
3674 | { | |
7c0cb9c6 | 3675 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3676 | } |
3677 | EXPORT_SYMBOL(__kmalloc); | |
3678 | ||
ce71e27c | 3679 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3680 | { |
7c0cb9c6 | 3681 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3682 | } |
3683 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3684 | |
1da177e4 LT |
3685 | /** |
3686 | * kmem_cache_free - Deallocate an object | |
3687 | * @cachep: The cache the allocation was from. | |
3688 | * @objp: The previously allocated object. | |
3689 | * | |
3690 | * Free an object which was previously allocated from this | |
3691 | * cache. | |
3692 | */ | |
343e0d7a | 3693 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3694 | { |
3695 | unsigned long flags; | |
b9ce5ef4 GC |
3696 | cachep = cache_from_obj(cachep, objp); |
3697 | if (!cachep) | |
3698 | return; | |
1da177e4 LT |
3699 | |
3700 | local_irq_save(flags); | |
d97d476b | 3701 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3702 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3703 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3704 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3705 | local_irq_restore(flags); |
36555751 | 3706 | |
ca2b84cb | 3707 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3708 | } |
3709 | EXPORT_SYMBOL(kmem_cache_free); | |
3710 | ||
e6cdb58d JDB |
3711 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3712 | { | |
3713 | struct kmem_cache *s; | |
3714 | size_t i; | |
3715 | ||
3716 | local_irq_disable(); | |
3717 | for (i = 0; i < size; i++) { | |
3718 | void *objp = p[i]; | |
3719 | ||
ca257195 JDB |
3720 | if (!orig_s) /* called via kfree_bulk */ |
3721 | s = virt_to_cache(objp); | |
3722 | else | |
3723 | s = cache_from_obj(orig_s, objp); | |
a64b5378 KC |
3724 | if (!s) |
3725 | continue; | |
e6cdb58d JDB |
3726 | |
3727 | debug_check_no_locks_freed(objp, s->object_size); | |
3728 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3729 | debug_check_no_obj_freed(objp, s->object_size); | |
3730 | ||
3731 | __cache_free(s, objp, _RET_IP_); | |
3732 | } | |
3733 | local_irq_enable(); | |
3734 | ||
3735 | /* FIXME: add tracing */ | |
3736 | } | |
3737 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3738 | ||
1da177e4 LT |
3739 | /** |
3740 | * kfree - free previously allocated memory | |
3741 | * @objp: pointer returned by kmalloc. | |
3742 | * | |
80e93eff PE |
3743 | * If @objp is NULL, no operation is performed. |
3744 | * | |
1da177e4 LT |
3745 | * Don't free memory not originally allocated by kmalloc() |
3746 | * or you will run into trouble. | |
3747 | */ | |
3748 | void kfree(const void *objp) | |
3749 | { | |
343e0d7a | 3750 | struct kmem_cache *c; |
1da177e4 LT |
3751 | unsigned long flags; |
3752 | ||
2121db74 PE |
3753 | trace_kfree(_RET_IP_, objp); |
3754 | ||
6cb8f913 | 3755 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3756 | return; |
3757 | local_irq_save(flags); | |
3758 | kfree_debugcheck(objp); | |
6ed5eb22 | 3759 | c = virt_to_cache(objp); |
a64b5378 KC |
3760 | if (!c) { |
3761 | local_irq_restore(flags); | |
3762 | return; | |
3763 | } | |
8c138bc0 CL |
3764 | debug_check_no_locks_freed(objp, c->object_size); |
3765 | ||
3766 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3767 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3768 | local_irq_restore(flags); |
3769 | } | |
3770 | EXPORT_SYMBOL(kfree); | |
3771 | ||
e498be7d | 3772 | /* |
ce8eb6c4 | 3773 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3774 | */ |
c3d332b6 | 3775 | static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d | 3776 | { |
c3d332b6 | 3777 | int ret; |
e498be7d | 3778 | int node; |
ce8eb6c4 | 3779 | struct kmem_cache_node *n; |
e498be7d | 3780 | |
9c09a95c | 3781 | for_each_online_node(node) { |
c3d332b6 JK |
3782 | ret = setup_kmem_cache_node(cachep, node, gfp, true); |
3783 | if (ret) | |
e498be7d CL |
3784 | goto fail; |
3785 | ||
e498be7d | 3786 | } |
c3d332b6 | 3787 | |
cafeb02e | 3788 | return 0; |
0718dc2a | 3789 | |
a737b3e2 | 3790 | fail: |
3b0efdfa | 3791 | if (!cachep->list.next) { |
0718dc2a CL |
3792 | /* Cache is not active yet. Roll back what we did */ |
3793 | node--; | |
3794 | while (node >= 0) { | |
18bf8541 CL |
3795 | n = get_node(cachep, node); |
3796 | if (n) { | |
ce8eb6c4 CL |
3797 | kfree(n->shared); |
3798 | free_alien_cache(n->alien); | |
3799 | kfree(n); | |
6a67368c | 3800 | cachep->node[node] = NULL; |
0718dc2a CL |
3801 | } |
3802 | node--; | |
3803 | } | |
3804 | } | |
cafeb02e | 3805 | return -ENOMEM; |
e498be7d CL |
3806 | } |
3807 | ||
18004c5d | 3808 | /* Always called with the slab_mutex held */ |
943a451a | 3809 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3810 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3811 | { |
bf0dea23 JK |
3812 | struct array_cache __percpu *cpu_cache, *prev; |
3813 | int cpu; | |
1da177e4 | 3814 | |
bf0dea23 JK |
3815 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3816 | if (!cpu_cache) | |
d2e7b7d0 SS |
3817 | return -ENOMEM; |
3818 | ||
bf0dea23 JK |
3819 | prev = cachep->cpu_cache; |
3820 | cachep->cpu_cache = cpu_cache; | |
a87c75fb GT |
3821 | /* |
3822 | * Without a previous cpu_cache there's no need to synchronize remote | |
3823 | * cpus, so skip the IPIs. | |
3824 | */ | |
3825 | if (prev) | |
3826 | kick_all_cpus_sync(); | |
e498be7d | 3827 | |
1da177e4 | 3828 | check_irq_on(); |
1da177e4 LT |
3829 | cachep->batchcount = batchcount; |
3830 | cachep->limit = limit; | |
e498be7d | 3831 | cachep->shared = shared; |
1da177e4 | 3832 | |
bf0dea23 | 3833 | if (!prev) |
c3d332b6 | 3834 | goto setup_node; |
bf0dea23 JK |
3835 | |
3836 | for_each_online_cpu(cpu) { | |
97654dfa | 3837 | LIST_HEAD(list); |
18bf8541 CL |
3838 | int node; |
3839 | struct kmem_cache_node *n; | |
bf0dea23 | 3840 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3841 | |
bf0dea23 | 3842 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3843 | n = get_node(cachep, node); |
3844 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3845 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3846 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3847 | slabs_destroy(cachep, &list); |
1da177e4 | 3848 | } |
bf0dea23 JK |
3849 | free_percpu(prev); |
3850 | ||
c3d332b6 JK |
3851 | setup_node: |
3852 | return setup_kmem_cache_nodes(cachep, gfp); | |
1da177e4 LT |
3853 | } |
3854 | ||
943a451a GC |
3855 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3856 | int batchcount, int shared, gfp_t gfp) | |
3857 | { | |
3858 | int ret; | |
426589f5 | 3859 | struct kmem_cache *c; |
943a451a GC |
3860 | |
3861 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3862 | ||
3863 | if (slab_state < FULL) | |
3864 | return ret; | |
3865 | ||
3866 | if ((ret < 0) || !is_root_cache(cachep)) | |
3867 | return ret; | |
3868 | ||
426589f5 VD |
3869 | lockdep_assert_held(&slab_mutex); |
3870 | for_each_memcg_cache(c, cachep) { | |
3871 | /* return value determined by the root cache only */ | |
3872 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3873 | } |
3874 | ||
3875 | return ret; | |
3876 | } | |
3877 | ||
18004c5d | 3878 | /* Called with slab_mutex held always */ |
83b519e8 | 3879 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3880 | { |
3881 | int err; | |
943a451a GC |
3882 | int limit = 0; |
3883 | int shared = 0; | |
3884 | int batchcount = 0; | |
3885 | ||
7c00fce9 | 3886 | err = cache_random_seq_create(cachep, cachep->num, gfp); |
c7ce4f60 TG |
3887 | if (err) |
3888 | goto end; | |
3889 | ||
943a451a GC |
3890 | if (!is_root_cache(cachep)) { |
3891 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3892 | limit = root->limit; | |
3893 | shared = root->shared; | |
3894 | batchcount = root->batchcount; | |
3895 | } | |
1da177e4 | 3896 | |
943a451a GC |
3897 | if (limit && shared && batchcount) |
3898 | goto skip_setup; | |
a737b3e2 AM |
3899 | /* |
3900 | * The head array serves three purposes: | |
1da177e4 LT |
3901 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3902 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3903 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3904 | * bufctl chains: array operations are cheaper. |
3905 | * The numbers are guessed, we should auto-tune as described by | |
3906 | * Bonwick. | |
3907 | */ | |
3b0efdfa | 3908 | if (cachep->size > 131072) |
1da177e4 | 3909 | limit = 1; |
3b0efdfa | 3910 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3911 | limit = 8; |
3b0efdfa | 3912 | else if (cachep->size > 1024) |
1da177e4 | 3913 | limit = 24; |
3b0efdfa | 3914 | else if (cachep->size > 256) |
1da177e4 LT |
3915 | limit = 54; |
3916 | else | |
3917 | limit = 120; | |
3918 | ||
a737b3e2 AM |
3919 | /* |
3920 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3921 | * allocation behaviour: Most allocs on one cpu, most free operations |
3922 | * on another cpu. For these cases, an efficient object passing between | |
3923 | * cpus is necessary. This is provided by a shared array. The array | |
3924 | * replaces Bonwick's magazine layer. | |
3925 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3926 | * to a larger limit. Thus disabled by default. | |
3927 | */ | |
3928 | shared = 0; | |
3b0efdfa | 3929 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3930 | shared = 8; |
1da177e4 LT |
3931 | |
3932 | #if DEBUG | |
a737b3e2 AM |
3933 | /* |
3934 | * With debugging enabled, large batchcount lead to excessively long | |
3935 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3936 | */ |
3937 | if (limit > 32) | |
3938 | limit = 32; | |
3939 | #endif | |
943a451a GC |
3940 | batchcount = (limit + 1) / 2; |
3941 | skip_setup: | |
3942 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
c7ce4f60 | 3943 | end: |
1da177e4 | 3944 | if (err) |
1170532b | 3945 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3946 | cachep->name, -err); |
2ed3a4ef | 3947 | return err; |
1da177e4 LT |
3948 | } |
3949 | ||
1b55253a | 3950 | /* |
ce8eb6c4 CL |
3951 | * Drain an array if it contains any elements taking the node lock only if |
3952 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3953 | * if drain_array() is used on the shared array. |
1b55253a | 3954 | */ |
ce8eb6c4 | 3955 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 3956 | struct array_cache *ac, int node) |
1da177e4 | 3957 | { |
97654dfa | 3958 | LIST_HEAD(list); |
18726ca8 JK |
3959 | |
3960 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
3961 | check_mutex_acquired(); | |
1da177e4 | 3962 | |
1b55253a CL |
3963 | if (!ac || !ac->avail) |
3964 | return; | |
18726ca8 JK |
3965 | |
3966 | if (ac->touched) { | |
1da177e4 | 3967 | ac->touched = 0; |
18726ca8 | 3968 | return; |
1da177e4 | 3969 | } |
18726ca8 JK |
3970 | |
3971 | spin_lock_irq(&n->list_lock); | |
3972 | drain_array_locked(cachep, ac, node, false, &list); | |
3973 | spin_unlock_irq(&n->list_lock); | |
3974 | ||
3975 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
3976 | } |
3977 | ||
3978 | /** | |
3979 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3980 | * @w: work descriptor |
1da177e4 LT |
3981 | * |
3982 | * Called from workqueue/eventd every few seconds. | |
3983 | * Purpose: | |
3984 | * - clear the per-cpu caches for this CPU. | |
3985 | * - return freeable pages to the main free memory pool. | |
3986 | * | |
a737b3e2 AM |
3987 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3988 | * again on the next iteration. | |
1da177e4 | 3989 | */ |
7c5cae36 | 3990 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3991 | { |
7a7c381d | 3992 | struct kmem_cache *searchp; |
ce8eb6c4 | 3993 | struct kmem_cache_node *n; |
7d6e6d09 | 3994 | int node = numa_mem_id(); |
bf6aede7 | 3995 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3996 | |
18004c5d | 3997 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3998 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3999 | goto out; |
1da177e4 | 4000 | |
18004c5d | 4001 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4002 | check_irq_on(); |
4003 | ||
35386e3b | 4004 | /* |
ce8eb6c4 | 4005 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4006 | * have established with reasonable certainty that |
4007 | * we can do some work if the lock was obtained. | |
4008 | */ | |
18bf8541 | 4009 | n = get_node(searchp, node); |
35386e3b | 4010 | |
ce8eb6c4 | 4011 | reap_alien(searchp, n); |
1da177e4 | 4012 | |
18726ca8 | 4013 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 4014 | |
35386e3b CL |
4015 | /* |
4016 | * These are racy checks but it does not matter | |
4017 | * if we skip one check or scan twice. | |
4018 | */ | |
ce8eb6c4 | 4019 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4020 | goto next; |
1da177e4 | 4021 | |
5f0985bb | 4022 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4023 | |
18726ca8 | 4024 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 4025 | |
ce8eb6c4 CL |
4026 | if (n->free_touched) |
4027 | n->free_touched = 0; | |
ed11d9eb CL |
4028 | else { |
4029 | int freed; | |
1da177e4 | 4030 | |
ce8eb6c4 | 4031 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4032 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4033 | STATS_ADD_REAPED(searchp, freed); | |
4034 | } | |
35386e3b | 4035 | next: |
1da177e4 LT |
4036 | cond_resched(); |
4037 | } | |
4038 | check_irq_on(); | |
18004c5d | 4039 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4040 | next_reap_node(); |
7c5cae36 | 4041 | out: |
a737b3e2 | 4042 | /* Set up the next iteration */ |
a9f2a846 VB |
4043 | schedule_delayed_work_on(smp_processor_id(), work, |
4044 | round_jiffies_relative(REAPTIMEOUT_AC)); | |
1da177e4 LT |
4045 | } |
4046 | ||
0d7561c6 | 4047 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4048 | { |
f728b0a5 | 4049 | unsigned long active_objs, num_objs, active_slabs; |
bf00bd34 DR |
4050 | unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0; |
4051 | unsigned long free_slabs = 0; | |
e498be7d | 4052 | int node; |
ce8eb6c4 | 4053 | struct kmem_cache_node *n; |
1da177e4 | 4054 | |
18bf8541 | 4055 | for_each_kmem_cache_node(cachep, node, n) { |
ca3b9b91 | 4056 | check_irq_on(); |
ce8eb6c4 | 4057 | spin_lock_irq(&n->list_lock); |
e498be7d | 4058 | |
bf00bd34 DR |
4059 | total_slabs += n->total_slabs; |
4060 | free_slabs += n->free_slabs; | |
f728b0a5 | 4061 | free_objs += n->free_objects; |
07a63c41 | 4062 | |
ce8eb6c4 CL |
4063 | if (n->shared) |
4064 | shared_avail += n->shared->avail; | |
e498be7d | 4065 | |
ce8eb6c4 | 4066 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4067 | } |
bf00bd34 DR |
4068 | num_objs = total_slabs * cachep->num; |
4069 | active_slabs = total_slabs - free_slabs; | |
f728b0a5 | 4070 | active_objs = num_objs - free_objs; |
1da177e4 | 4071 | |
0d7561c6 GC |
4072 | sinfo->active_objs = active_objs; |
4073 | sinfo->num_objs = num_objs; | |
4074 | sinfo->active_slabs = active_slabs; | |
bf00bd34 | 4075 | sinfo->num_slabs = total_slabs; |
0d7561c6 GC |
4076 | sinfo->shared_avail = shared_avail; |
4077 | sinfo->limit = cachep->limit; | |
4078 | sinfo->batchcount = cachep->batchcount; | |
4079 | sinfo->shared = cachep->shared; | |
4080 | sinfo->objects_per_slab = cachep->num; | |
4081 | sinfo->cache_order = cachep->gfporder; | |
4082 | } | |
4083 | ||
4084 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4085 | { | |
1da177e4 | 4086 | #if STATS |
ce8eb6c4 | 4087 | { /* node stats */ |
1da177e4 LT |
4088 | unsigned long high = cachep->high_mark; |
4089 | unsigned long allocs = cachep->num_allocations; | |
4090 | unsigned long grown = cachep->grown; | |
4091 | unsigned long reaped = cachep->reaped; | |
4092 | unsigned long errors = cachep->errors; | |
4093 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4094 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4095 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4096 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4097 | |
756a025f | 4098 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4099 | allocs, high, grown, |
4100 | reaped, errors, max_freeable, node_allocs, | |
4101 | node_frees, overflows); | |
1da177e4 LT |
4102 | } |
4103 | /* cpu stats */ | |
4104 | { | |
4105 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4106 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4107 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4108 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4109 | ||
4110 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4111 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4112 | } |
4113 | #endif | |
1da177e4 LT |
4114 | } |
4115 | ||
1da177e4 LT |
4116 | #define MAX_SLABINFO_WRITE 128 |
4117 | /** | |
4118 | * slabinfo_write - Tuning for the slab allocator | |
4119 | * @file: unused | |
4120 | * @buffer: user buffer | |
4121 | * @count: data length | |
4122 | * @ppos: unused | |
a862f68a MR |
4123 | * |
4124 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 4125 | */ |
b7454ad3 | 4126 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4127 | size_t count, loff_t *ppos) |
1da177e4 | 4128 | { |
b28a02de | 4129 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4130 | int limit, batchcount, shared, res; |
7a7c381d | 4131 | struct kmem_cache *cachep; |
b28a02de | 4132 | |
1da177e4 LT |
4133 | if (count > MAX_SLABINFO_WRITE) |
4134 | return -EINVAL; | |
4135 | if (copy_from_user(&kbuf, buffer, count)) | |
4136 | return -EFAULT; | |
b28a02de | 4137 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4138 | |
4139 | tmp = strchr(kbuf, ' '); | |
4140 | if (!tmp) | |
4141 | return -EINVAL; | |
4142 | *tmp = '\0'; | |
4143 | tmp++; | |
4144 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4145 | return -EINVAL; | |
4146 | ||
4147 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4148 | mutex_lock(&slab_mutex); |
1da177e4 | 4149 | res = -EINVAL; |
18004c5d | 4150 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4151 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4152 | if (limit < 1 || batchcount < 1 || |
4153 | batchcount > limit || shared < 0) { | |
e498be7d | 4154 | res = 0; |
1da177e4 | 4155 | } else { |
e498be7d | 4156 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4157 | batchcount, shared, |
4158 | GFP_KERNEL); | |
1da177e4 LT |
4159 | } |
4160 | break; | |
4161 | } | |
4162 | } | |
18004c5d | 4163 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4164 | if (res >= 0) |
4165 | res = count; | |
4166 | return res; | |
4167 | } | |
871751e2 | 4168 | |
04385fc5 KC |
4169 | #ifdef CONFIG_HARDENED_USERCOPY |
4170 | /* | |
afcc90f8 KC |
4171 | * Rejects incorrectly sized objects and objects that are to be copied |
4172 | * to/from userspace but do not fall entirely within the containing slab | |
4173 | * cache's usercopy region. | |
04385fc5 KC |
4174 | * |
4175 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4176 | * to indicate an error. | |
4177 | */ | |
f4e6e289 KC |
4178 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4179 | bool to_user) | |
04385fc5 KC |
4180 | { |
4181 | struct kmem_cache *cachep; | |
4182 | unsigned int objnr; | |
4183 | unsigned long offset; | |
4184 | ||
219667c2 AK |
4185 | ptr = kasan_reset_tag(ptr); |
4186 | ||
04385fc5 KC |
4187 | /* Find and validate object. */ |
4188 | cachep = page->slab_cache; | |
4189 | objnr = obj_to_index(cachep, page, (void *)ptr); | |
4190 | BUG_ON(objnr >= cachep->num); | |
4191 | ||
4192 | /* Find offset within object. */ | |
4193 | offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep); | |
4194 | ||
afcc90f8 KC |
4195 | /* Allow address range falling entirely within usercopy region. */ |
4196 | if (offset >= cachep->useroffset && | |
4197 | offset - cachep->useroffset <= cachep->usersize && | |
4198 | n <= cachep->useroffset - offset + cachep->usersize) | |
f4e6e289 | 4199 | return; |
04385fc5 | 4200 | |
afcc90f8 KC |
4201 | /* |
4202 | * If the copy is still within the allocated object, produce | |
4203 | * a warning instead of rejecting the copy. This is intended | |
4204 | * to be a temporary method to find any missing usercopy | |
4205 | * whitelists. | |
4206 | */ | |
2d891fbc KC |
4207 | if (usercopy_fallback && |
4208 | offset <= cachep->object_size && | |
afcc90f8 KC |
4209 | n <= cachep->object_size - offset) { |
4210 | usercopy_warn("SLAB object", cachep->name, to_user, offset, n); | |
4211 | return; | |
4212 | } | |
04385fc5 | 4213 | |
f4e6e289 | 4214 | usercopy_abort("SLAB object", cachep->name, to_user, offset, n); |
04385fc5 KC |
4215 | } |
4216 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4217 | ||
00e145b6 | 4218 | /** |
10d1f8cb | 4219 | * __ksize -- Uninstrumented ksize. |
87bf4f71 | 4220 | * @objp: pointer to the object |
00e145b6 | 4221 | * |
10d1f8cb ME |
4222 | * Unlike ksize(), __ksize() is uninstrumented, and does not provide the same |
4223 | * safety checks as ksize() with KASAN instrumentation enabled. | |
87bf4f71 RD |
4224 | * |
4225 | * Return: size of the actual memory used by @objp in bytes | |
00e145b6 | 4226 | */ |
10d1f8cb | 4227 | size_t __ksize(const void *objp) |
1da177e4 | 4228 | { |
a64b5378 | 4229 | struct kmem_cache *c; |
7ed2f9e6 AP |
4230 | size_t size; |
4231 | ||
ef8b4520 CL |
4232 | BUG_ON(!objp); |
4233 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4234 | return 0; |
1da177e4 | 4235 | |
a64b5378 KC |
4236 | c = virt_to_cache(objp); |
4237 | size = c ? c->object_size : 0; | |
7ed2f9e6 AP |
4238 | |
4239 | return size; | |
1da177e4 | 4240 | } |
10d1f8cb | 4241 | EXPORT_SYMBOL(__ksize); |