Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
072bb0aa MG |
160 | /* |
161 | * true if a page was allocated from pfmemalloc reserves for network-based | |
162 | * swap | |
163 | */ | |
164 | static bool pfmemalloc_active __read_mostly; | |
165 | ||
5bfe53a7 LJ |
166 | /* |
167 | * struct slab | |
168 | * | |
169 | * Manages the objs in a slab. Placed either at the beginning of mem allocated | |
170 | * for a slab, or allocated from an general cache. | |
171 | * Slabs are chained into three list: fully used, partial, fully free slabs. | |
172 | */ | |
173 | struct slab { | |
68126702 JK |
174 | struct { |
175 | struct list_head list; | |
176 | void *s_mem; /* including colour offset */ | |
106a74e1 | 177 | unsigned int active; /* num of objs active in slab */ |
5bfe53a7 LJ |
178 | }; |
179 | }; | |
180 | ||
1da177e4 LT |
181 | /* |
182 | * struct array_cache | |
183 | * | |
1da177e4 LT |
184 | * Purpose: |
185 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
186 | * - reduce the number of linked list operations | |
187 | * - reduce spinlock operations | |
188 | * | |
189 | * The limit is stored in the per-cpu structure to reduce the data cache | |
190 | * footprint. | |
191 | * | |
192 | */ | |
193 | struct array_cache { | |
194 | unsigned int avail; | |
195 | unsigned int limit; | |
196 | unsigned int batchcount; | |
197 | unsigned int touched; | |
e498be7d | 198 | spinlock_t lock; |
bda5b655 | 199 | void *entry[]; /* |
a737b3e2 AM |
200 | * Must have this definition in here for the proper |
201 | * alignment of array_cache. Also simplifies accessing | |
202 | * the entries. | |
072bb0aa MG |
203 | * |
204 | * Entries should not be directly dereferenced as | |
205 | * entries belonging to slabs marked pfmemalloc will | |
206 | * have the lower bits set SLAB_OBJ_PFMEMALLOC | |
a737b3e2 | 207 | */ |
1da177e4 LT |
208 | }; |
209 | ||
072bb0aa MG |
210 | #define SLAB_OBJ_PFMEMALLOC 1 |
211 | static inline bool is_obj_pfmemalloc(void *objp) | |
212 | { | |
213 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; | |
214 | } | |
215 | ||
216 | static inline void set_obj_pfmemalloc(void **objp) | |
217 | { | |
218 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); | |
219 | return; | |
220 | } | |
221 | ||
222 | static inline void clear_obj_pfmemalloc(void **objp) | |
223 | { | |
224 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); | |
225 | } | |
226 | ||
a737b3e2 AM |
227 | /* |
228 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
229 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
230 | */ |
231 | #define BOOT_CPUCACHE_ENTRIES 1 | |
232 | struct arraycache_init { | |
233 | struct array_cache cache; | |
b28a02de | 234 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
235 | }; |
236 | ||
e498be7d CL |
237 | /* |
238 | * Need this for bootstrapping a per node allocator. | |
239 | */ | |
556a169d | 240 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
ce8eb6c4 | 241 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 242 | #define CACHE_CACHE 0 |
556a169d | 243 | #define SIZE_AC MAX_NUMNODES |
ce8eb6c4 | 244 | #define SIZE_NODE (2 * MAX_NUMNODES) |
e498be7d | 245 | |
ed11d9eb | 246 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 247 | struct kmem_cache_node *n, int tofree); |
ed11d9eb CL |
248 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
249 | int node); | |
83b519e8 | 250 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 251 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 252 | |
e0a42726 IM |
253 | static int slab_early_init = 1; |
254 | ||
e3366016 | 255 | #define INDEX_AC kmalloc_index(sizeof(struct arraycache_init)) |
ce8eb6c4 | 256 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 257 | |
ce8eb6c4 | 258 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
259 | { |
260 | INIT_LIST_HEAD(&parent->slabs_full); | |
261 | INIT_LIST_HEAD(&parent->slabs_partial); | |
262 | INIT_LIST_HEAD(&parent->slabs_free); | |
263 | parent->shared = NULL; | |
264 | parent->alien = NULL; | |
2e1217cf | 265 | parent->colour_next = 0; |
e498be7d CL |
266 | spin_lock_init(&parent->list_lock); |
267 | parent->free_objects = 0; | |
268 | parent->free_touched = 0; | |
269 | } | |
270 | ||
a737b3e2 AM |
271 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
272 | do { \ | |
273 | INIT_LIST_HEAD(listp); \ | |
6a67368c | 274 | list_splice(&(cachep->node[nodeid]->slab), listp); \ |
e498be7d CL |
275 | } while (0) |
276 | ||
a737b3e2 AM |
277 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
278 | do { \ | |
e498be7d CL |
279 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
280 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
281 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
282 | } while (0) | |
1da177e4 | 283 | |
1da177e4 LT |
284 | #define CFLGS_OFF_SLAB (0x80000000UL) |
285 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
286 | ||
287 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
288 | /* |
289 | * Optimization question: fewer reaps means less probability for unnessary | |
290 | * cpucache drain/refill cycles. | |
1da177e4 | 291 | * |
dc6f3f27 | 292 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
293 | * which could lock up otherwise freeable slabs. |
294 | */ | |
295 | #define REAPTIMEOUT_CPUC (2*HZ) | |
296 | #define REAPTIMEOUT_LIST3 (4*HZ) | |
297 | ||
298 | #if STATS | |
299 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
300 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
301 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
302 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 303 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
304 | #define STATS_SET_HIGH(x) \ |
305 | do { \ | |
306 | if ((x)->num_active > (x)->high_mark) \ | |
307 | (x)->high_mark = (x)->num_active; \ | |
308 | } while (0) | |
1da177e4 LT |
309 | #define STATS_INC_ERR(x) ((x)->errors++) |
310 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 311 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 312 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
313 | #define STATS_SET_FREEABLE(x, i) \ |
314 | do { \ | |
315 | if ((x)->max_freeable < i) \ | |
316 | (x)->max_freeable = i; \ | |
317 | } while (0) | |
1da177e4 LT |
318 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
319 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
320 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
321 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
322 | #else | |
323 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
324 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
325 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
326 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 327 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
328 | #define STATS_SET_HIGH(x) do { } while (0) |
329 | #define STATS_INC_ERR(x) do { } while (0) | |
330 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 331 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 332 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 333 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
334 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
335 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
336 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
337 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
338 | #endif | |
339 | ||
340 | #if DEBUG | |
1da177e4 | 341 | |
a737b3e2 AM |
342 | /* |
343 | * memory layout of objects: | |
1da177e4 | 344 | * 0 : objp |
3dafccf2 | 345 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
346 | * the end of an object is aligned with the end of the real |
347 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 348 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 349 | * redzone word. |
3dafccf2 | 350 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
351 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
352 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 353 | * [BYTES_PER_WORD long] |
1da177e4 | 354 | */ |
343e0d7a | 355 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 356 | { |
3dafccf2 | 357 | return cachep->obj_offset; |
1da177e4 LT |
358 | } |
359 | ||
b46b8f19 | 360 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
361 | { |
362 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
363 | return (unsigned long long*) (objp + obj_offset(cachep) - |
364 | sizeof(unsigned long long)); | |
1da177e4 LT |
365 | } |
366 | ||
b46b8f19 | 367 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
368 | { |
369 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
370 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 371 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 372 | sizeof(unsigned long long) - |
87a927c7 | 373 | REDZONE_ALIGN); |
3b0efdfa | 374 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 375 | sizeof(unsigned long long)); |
1da177e4 LT |
376 | } |
377 | ||
343e0d7a | 378 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
379 | { |
380 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 381 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
382 | } |
383 | ||
384 | #else | |
385 | ||
3dafccf2 | 386 | #define obj_offset(x) 0 |
b46b8f19 DW |
387 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
388 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
389 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
390 | ||
391 | #endif | |
392 | ||
1da177e4 | 393 | /* |
3df1cccd DR |
394 | * Do not go above this order unless 0 objects fit into the slab or |
395 | * overridden on the command line. | |
1da177e4 | 396 | */ |
543585cc DR |
397 | #define SLAB_MAX_ORDER_HI 1 |
398 | #define SLAB_MAX_ORDER_LO 0 | |
399 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 400 | static bool slab_max_order_set __initdata; |
1da177e4 | 401 | |
6ed5eb22 PE |
402 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
403 | { | |
b49af68f | 404 | struct page *page = virt_to_head_page(obj); |
35026088 | 405 | return page->slab_cache; |
6ed5eb22 PE |
406 | } |
407 | ||
408 | static inline struct slab *virt_to_slab(const void *obj) | |
409 | { | |
b49af68f | 410 | struct page *page = virt_to_head_page(obj); |
35026088 CL |
411 | |
412 | VM_BUG_ON(!PageSlab(page)); | |
413 | return page->slab_page; | |
6ed5eb22 PE |
414 | } |
415 | ||
8fea4e96 PE |
416 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
417 | unsigned int idx) | |
418 | { | |
3b0efdfa | 419 | return slab->s_mem + cache->size * idx; |
8fea4e96 PE |
420 | } |
421 | ||
6a2d7a95 | 422 | /* |
3b0efdfa CL |
423 | * We want to avoid an expensive divide : (offset / cache->size) |
424 | * Using the fact that size is a constant for a particular cache, | |
425 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
426 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
427 | */ | |
428 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
429 | const struct slab *slab, void *obj) | |
8fea4e96 | 430 | { |
6a2d7a95 ED |
431 | u32 offset = (obj - slab->s_mem); |
432 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); | |
8fea4e96 PE |
433 | } |
434 | ||
1da177e4 | 435 | static struct arraycache_init initarray_generic = |
b28a02de | 436 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
437 | |
438 | /* internal cache of cache description objs */ | |
9b030cb8 | 439 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
440 | .batchcount = 1, |
441 | .limit = BOOT_CPUCACHE_ENTRIES, | |
442 | .shared = 1, | |
3b0efdfa | 443 | .size = sizeof(struct kmem_cache), |
b28a02de | 444 | .name = "kmem_cache", |
1da177e4 LT |
445 | }; |
446 | ||
056c6241 RT |
447 | #define BAD_ALIEN_MAGIC 0x01020304ul |
448 | ||
f1aaee53 AV |
449 | #ifdef CONFIG_LOCKDEP |
450 | ||
451 | /* | |
452 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
453 | * for other slabs "off slab". | |
454 | * The locking for this is tricky in that it nests within the locks | |
455 | * of all other slabs in a few places; to deal with this special | |
456 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
457 | * |
458 | * We set lock class for alien array caches which are up during init. | |
459 | * The lock annotation will be lost if all cpus of a node goes down and | |
460 | * then comes back up during hotplug | |
f1aaee53 | 461 | */ |
056c6241 RT |
462 | static struct lock_class_key on_slab_l3_key; |
463 | static struct lock_class_key on_slab_alc_key; | |
464 | ||
83835b3d PZ |
465 | static struct lock_class_key debugobj_l3_key; |
466 | static struct lock_class_key debugobj_alc_key; | |
467 | ||
468 | static void slab_set_lock_classes(struct kmem_cache *cachep, | |
469 | struct lock_class_key *l3_key, struct lock_class_key *alc_key, | |
470 | int q) | |
471 | { | |
472 | struct array_cache **alc; | |
ce8eb6c4 | 473 | struct kmem_cache_node *n; |
83835b3d PZ |
474 | int r; |
475 | ||
ce8eb6c4 CL |
476 | n = cachep->node[q]; |
477 | if (!n) | |
83835b3d PZ |
478 | return; |
479 | ||
ce8eb6c4 CL |
480 | lockdep_set_class(&n->list_lock, l3_key); |
481 | alc = n->alien; | |
83835b3d PZ |
482 | /* |
483 | * FIXME: This check for BAD_ALIEN_MAGIC | |
484 | * should go away when common slab code is taught to | |
485 | * work even without alien caches. | |
486 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
487 | * for alloc_alien_cache, | |
488 | */ | |
489 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
490 | return; | |
491 | for_each_node(r) { | |
492 | if (alc[r]) | |
493 | lockdep_set_class(&alc[r]->lock, alc_key); | |
494 | } | |
495 | } | |
496 | ||
497 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) | |
498 | { | |
499 | slab_set_lock_classes(cachep, &debugobj_l3_key, &debugobj_alc_key, node); | |
500 | } | |
501 | ||
502 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
503 | { | |
504 | int node; | |
505 | ||
506 | for_each_online_node(node) | |
507 | slab_set_debugobj_lock_classes_node(cachep, node); | |
508 | } | |
509 | ||
ce79ddc8 | 510 | static void init_node_lock_keys(int q) |
f1aaee53 | 511 | { |
e3366016 | 512 | int i; |
056c6241 | 513 | |
97d06609 | 514 | if (slab_state < UP) |
ce79ddc8 PE |
515 | return; |
516 | ||
0f8f8094 | 517 | for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) { |
ce8eb6c4 | 518 | struct kmem_cache_node *n; |
e3366016 CL |
519 | struct kmem_cache *cache = kmalloc_caches[i]; |
520 | ||
521 | if (!cache) | |
522 | continue; | |
ce79ddc8 | 523 | |
ce8eb6c4 CL |
524 | n = cache->node[q]; |
525 | if (!n || OFF_SLAB(cache)) | |
00afa758 | 526 | continue; |
83835b3d | 527 | |
e3366016 | 528 | slab_set_lock_classes(cache, &on_slab_l3_key, |
83835b3d | 529 | &on_slab_alc_key, q); |
f1aaee53 AV |
530 | } |
531 | } | |
ce79ddc8 | 532 | |
6ccfb5bc GC |
533 | static void on_slab_lock_classes_node(struct kmem_cache *cachep, int q) |
534 | { | |
6a67368c | 535 | if (!cachep->node[q]) |
6ccfb5bc GC |
536 | return; |
537 | ||
538 | slab_set_lock_classes(cachep, &on_slab_l3_key, | |
539 | &on_slab_alc_key, q); | |
540 | } | |
541 | ||
542 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) | |
543 | { | |
544 | int node; | |
545 | ||
546 | VM_BUG_ON(OFF_SLAB(cachep)); | |
547 | for_each_node(node) | |
548 | on_slab_lock_classes_node(cachep, node); | |
549 | } | |
550 | ||
ce79ddc8 PE |
551 | static inline void init_lock_keys(void) |
552 | { | |
553 | int node; | |
554 | ||
555 | for_each_node(node) | |
556 | init_node_lock_keys(node); | |
557 | } | |
f1aaee53 | 558 | #else |
ce79ddc8 PE |
559 | static void init_node_lock_keys(int q) |
560 | { | |
561 | } | |
562 | ||
056c6241 | 563 | static inline void init_lock_keys(void) |
f1aaee53 AV |
564 | { |
565 | } | |
83835b3d | 566 | |
6ccfb5bc GC |
567 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) |
568 | { | |
569 | } | |
570 | ||
571 | static inline void on_slab_lock_classes_node(struct kmem_cache *cachep, int node) | |
572 | { | |
573 | } | |
574 | ||
83835b3d PZ |
575 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) |
576 | { | |
577 | } | |
578 | ||
579 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
580 | { | |
581 | } | |
f1aaee53 AV |
582 | #endif |
583 | ||
1871e52c | 584 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 585 | |
343e0d7a | 586 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
587 | { |
588 | return cachep->array[smp_processor_id()]; | |
589 | } | |
590 | ||
fbaccacf | 591 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
1da177e4 | 592 | { |
16025177 | 593 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(unsigned int), align); |
fbaccacf | 594 | } |
1da177e4 | 595 | |
a737b3e2 AM |
596 | /* |
597 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
598 | */ | |
fbaccacf SR |
599 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
600 | size_t align, int flags, size_t *left_over, | |
601 | unsigned int *num) | |
602 | { | |
603 | int nr_objs; | |
604 | size_t mgmt_size; | |
605 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 606 | |
fbaccacf SR |
607 | /* |
608 | * The slab management structure can be either off the slab or | |
609 | * on it. For the latter case, the memory allocated for a | |
610 | * slab is used for: | |
611 | * | |
612 | * - The struct slab | |
16025177 | 613 | * - One unsigned int for each object |
fbaccacf SR |
614 | * - Padding to respect alignment of @align |
615 | * - @buffer_size bytes for each object | |
616 | * | |
617 | * If the slab management structure is off the slab, then the | |
618 | * alignment will already be calculated into the size. Because | |
619 | * the slabs are all pages aligned, the objects will be at the | |
620 | * correct alignment when allocated. | |
621 | */ | |
622 | if (flags & CFLGS_OFF_SLAB) { | |
623 | mgmt_size = 0; | |
624 | nr_objs = slab_size / buffer_size; | |
625 | ||
fbaccacf SR |
626 | } else { |
627 | /* | |
628 | * Ignore padding for the initial guess. The padding | |
629 | * is at most @align-1 bytes, and @buffer_size is at | |
630 | * least @align. In the worst case, this result will | |
631 | * be one greater than the number of objects that fit | |
632 | * into the memory allocation when taking the padding | |
633 | * into account. | |
634 | */ | |
635 | nr_objs = (slab_size - sizeof(struct slab)) / | |
16025177 | 636 | (buffer_size + sizeof(unsigned int)); |
fbaccacf SR |
637 | |
638 | /* | |
639 | * This calculated number will be either the right | |
640 | * amount, or one greater than what we want. | |
641 | */ | |
642 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | |
643 | > slab_size) | |
644 | nr_objs--; | |
645 | ||
fbaccacf SR |
646 | mgmt_size = slab_mgmt_size(nr_objs, align); |
647 | } | |
648 | *num = nr_objs; | |
649 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
650 | } |
651 | ||
f28510d3 | 652 | #if DEBUG |
d40cee24 | 653 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 654 | |
a737b3e2 AM |
655 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
656 | char *msg) | |
1da177e4 LT |
657 | { |
658 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 659 | function, cachep->name, msg); |
1da177e4 | 660 | dump_stack(); |
373d4d09 | 661 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 662 | } |
f28510d3 | 663 | #endif |
1da177e4 | 664 | |
3395ee05 PM |
665 | /* |
666 | * By default on NUMA we use alien caches to stage the freeing of | |
667 | * objects allocated from other nodes. This causes massive memory | |
668 | * inefficiencies when using fake NUMA setup to split memory into a | |
669 | * large number of small nodes, so it can be disabled on the command | |
670 | * line | |
671 | */ | |
672 | ||
673 | static int use_alien_caches __read_mostly = 1; | |
674 | static int __init noaliencache_setup(char *s) | |
675 | { | |
676 | use_alien_caches = 0; | |
677 | return 1; | |
678 | } | |
679 | __setup("noaliencache", noaliencache_setup); | |
680 | ||
3df1cccd DR |
681 | static int __init slab_max_order_setup(char *str) |
682 | { | |
683 | get_option(&str, &slab_max_order); | |
684 | slab_max_order = slab_max_order < 0 ? 0 : | |
685 | min(slab_max_order, MAX_ORDER - 1); | |
686 | slab_max_order_set = true; | |
687 | ||
688 | return 1; | |
689 | } | |
690 | __setup("slab_max_order=", slab_max_order_setup); | |
691 | ||
8fce4d8e CL |
692 | #ifdef CONFIG_NUMA |
693 | /* | |
694 | * Special reaping functions for NUMA systems called from cache_reap(). | |
695 | * These take care of doing round robin flushing of alien caches (containing | |
696 | * objects freed on different nodes from which they were allocated) and the | |
697 | * flushing of remote pcps by calling drain_node_pages. | |
698 | */ | |
1871e52c | 699 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
700 | |
701 | static void init_reap_node(int cpu) | |
702 | { | |
703 | int node; | |
704 | ||
7d6e6d09 | 705 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 706 | if (node == MAX_NUMNODES) |
442295c9 | 707 | node = first_node(node_online_map); |
8fce4d8e | 708 | |
1871e52c | 709 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
710 | } |
711 | ||
712 | static void next_reap_node(void) | |
713 | { | |
909ea964 | 714 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 715 | |
8fce4d8e CL |
716 | node = next_node(node, node_online_map); |
717 | if (unlikely(node >= MAX_NUMNODES)) | |
718 | node = first_node(node_online_map); | |
909ea964 | 719 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
720 | } |
721 | ||
722 | #else | |
723 | #define init_reap_node(cpu) do { } while (0) | |
724 | #define next_reap_node(void) do { } while (0) | |
725 | #endif | |
726 | ||
1da177e4 LT |
727 | /* |
728 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
729 | * via the workqueue/eventd. | |
730 | * Add the CPU number into the expiration time to minimize the possibility of | |
731 | * the CPUs getting into lockstep and contending for the global cache chain | |
732 | * lock. | |
733 | */ | |
0db0628d | 734 | static void start_cpu_timer(int cpu) |
1da177e4 | 735 | { |
1871e52c | 736 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
737 | |
738 | /* | |
739 | * When this gets called from do_initcalls via cpucache_init(), | |
740 | * init_workqueues() has already run, so keventd will be setup | |
741 | * at that time. | |
742 | */ | |
52bad64d | 743 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 744 | init_reap_node(cpu); |
203b42f7 | 745 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
746 | schedule_delayed_work_on(cpu, reap_work, |
747 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
748 | } |
749 | } | |
750 | ||
e498be7d | 751 | static struct array_cache *alloc_arraycache(int node, int entries, |
83b519e8 | 752 | int batchcount, gfp_t gfp) |
1da177e4 | 753 | { |
b28a02de | 754 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
755 | struct array_cache *nc = NULL; |
756 | ||
83b519e8 | 757 | nc = kmalloc_node(memsize, gfp, node); |
d5cff635 CM |
758 | /* |
759 | * The array_cache structures contain pointers to free object. | |
25985edc | 760 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
761 | * cache the pointers are not cleared and they could be counted as |
762 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
763 | * not scan such objects. | |
764 | */ | |
765 | kmemleak_no_scan(nc); | |
1da177e4 LT |
766 | if (nc) { |
767 | nc->avail = 0; | |
768 | nc->limit = entries; | |
769 | nc->batchcount = batchcount; | |
770 | nc->touched = 0; | |
e498be7d | 771 | spin_lock_init(&nc->lock); |
1da177e4 LT |
772 | } |
773 | return nc; | |
774 | } | |
775 | ||
072bb0aa MG |
776 | static inline bool is_slab_pfmemalloc(struct slab *slabp) |
777 | { | |
778 | struct page *page = virt_to_page(slabp->s_mem); | |
779 | ||
780 | return PageSlabPfmemalloc(page); | |
781 | } | |
782 | ||
783 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
784 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
785 | struct array_cache *ac) | |
786 | { | |
ce8eb6c4 | 787 | struct kmem_cache_node *n = cachep->node[numa_mem_id()]; |
072bb0aa MG |
788 | struct slab *slabp; |
789 | unsigned long flags; | |
790 | ||
791 | if (!pfmemalloc_active) | |
792 | return; | |
793 | ||
ce8eb6c4 CL |
794 | spin_lock_irqsave(&n->list_lock, flags); |
795 | list_for_each_entry(slabp, &n->slabs_full, list) | |
072bb0aa MG |
796 | if (is_slab_pfmemalloc(slabp)) |
797 | goto out; | |
798 | ||
ce8eb6c4 | 799 | list_for_each_entry(slabp, &n->slabs_partial, list) |
072bb0aa MG |
800 | if (is_slab_pfmemalloc(slabp)) |
801 | goto out; | |
802 | ||
ce8eb6c4 | 803 | list_for_each_entry(slabp, &n->slabs_free, list) |
072bb0aa MG |
804 | if (is_slab_pfmemalloc(slabp)) |
805 | goto out; | |
806 | ||
807 | pfmemalloc_active = false; | |
808 | out: | |
ce8eb6c4 | 809 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
810 | } |
811 | ||
381760ea | 812 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
813 | gfp_t flags, bool force_refill) |
814 | { | |
815 | int i; | |
816 | void *objp = ac->entry[--ac->avail]; | |
817 | ||
818 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
819 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 820 | struct kmem_cache_node *n; |
072bb0aa MG |
821 | |
822 | if (gfp_pfmemalloc_allowed(flags)) { | |
823 | clear_obj_pfmemalloc(&objp); | |
824 | return objp; | |
825 | } | |
826 | ||
827 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 828 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
829 | /* If a !PFMEMALLOC object is found, swap them */ |
830 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
831 | objp = ac->entry[i]; | |
832 | ac->entry[i] = ac->entry[ac->avail]; | |
833 | ac->entry[ac->avail] = objp; | |
834 | return objp; | |
835 | } | |
836 | } | |
837 | ||
838 | /* | |
839 | * If there are empty slabs on the slabs_free list and we are | |
840 | * being forced to refill the cache, mark this one !pfmemalloc. | |
841 | */ | |
ce8eb6c4 CL |
842 | n = cachep->node[numa_mem_id()]; |
843 | if (!list_empty(&n->slabs_free) && force_refill) { | |
072bb0aa | 844 | struct slab *slabp = virt_to_slab(objp); |
30c29bea | 845 | ClearPageSlabPfmemalloc(virt_to_head_page(slabp->s_mem)); |
072bb0aa MG |
846 | clear_obj_pfmemalloc(&objp); |
847 | recheck_pfmemalloc_active(cachep, ac); | |
848 | return objp; | |
849 | } | |
850 | ||
851 | /* No !PFMEMALLOC objects available */ | |
852 | ac->avail++; | |
853 | objp = NULL; | |
854 | } | |
855 | ||
856 | return objp; | |
857 | } | |
858 | ||
381760ea MG |
859 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
860 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
861 | { | |
862 | void *objp; | |
863 | ||
864 | if (unlikely(sk_memalloc_socks())) | |
865 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
866 | else | |
867 | objp = ac->entry[--ac->avail]; | |
868 | ||
869 | return objp; | |
870 | } | |
871 | ||
872 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
873 | void *objp) |
874 | { | |
875 | if (unlikely(pfmemalloc_active)) { | |
876 | /* Some pfmemalloc slabs exist, check if this is one */ | |
73293c2f JK |
877 | struct slab *slabp = virt_to_slab(objp); |
878 | struct page *page = virt_to_head_page(slabp->s_mem); | |
072bb0aa MG |
879 | if (PageSlabPfmemalloc(page)) |
880 | set_obj_pfmemalloc(&objp); | |
881 | } | |
882 | ||
381760ea MG |
883 | return objp; |
884 | } | |
885 | ||
886 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
887 | void *objp) | |
888 | { | |
889 | if (unlikely(sk_memalloc_socks())) | |
890 | objp = __ac_put_obj(cachep, ac, objp); | |
891 | ||
072bb0aa MG |
892 | ac->entry[ac->avail++] = objp; |
893 | } | |
894 | ||
3ded175a CL |
895 | /* |
896 | * Transfer objects in one arraycache to another. | |
897 | * Locking must be handled by the caller. | |
898 | * | |
899 | * Return the number of entries transferred. | |
900 | */ | |
901 | static int transfer_objects(struct array_cache *to, | |
902 | struct array_cache *from, unsigned int max) | |
903 | { | |
904 | /* Figure out how many entries to transfer */ | |
732eacc0 | 905 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
906 | |
907 | if (!nr) | |
908 | return 0; | |
909 | ||
910 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
911 | sizeof(void *) *nr); | |
912 | ||
913 | from->avail -= nr; | |
914 | to->avail += nr; | |
3ded175a CL |
915 | return nr; |
916 | } | |
917 | ||
765c4507 CL |
918 | #ifndef CONFIG_NUMA |
919 | ||
920 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 921 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 922 | |
83b519e8 | 923 | static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
765c4507 CL |
924 | { |
925 | return (struct array_cache **)BAD_ALIEN_MAGIC; | |
926 | } | |
927 | ||
928 | static inline void free_alien_cache(struct array_cache **ac_ptr) | |
929 | { | |
930 | } | |
931 | ||
932 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
933 | { | |
934 | return 0; | |
935 | } | |
936 | ||
937 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
938 | gfp_t flags) | |
939 | { | |
940 | return NULL; | |
941 | } | |
942 | ||
8b98c169 | 943 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
944 | gfp_t flags, int nodeid) |
945 | { | |
946 | return NULL; | |
947 | } | |
948 | ||
949 | #else /* CONFIG_NUMA */ | |
950 | ||
8b98c169 | 951 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 952 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 953 | |
83b519e8 | 954 | static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
e498be7d CL |
955 | { |
956 | struct array_cache **ac_ptr; | |
8ef82866 | 957 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
958 | int i; |
959 | ||
960 | if (limit > 1) | |
961 | limit = 12; | |
f3186a9c | 962 | ac_ptr = kzalloc_node(memsize, gfp, node); |
e498be7d CL |
963 | if (ac_ptr) { |
964 | for_each_node(i) { | |
f3186a9c | 965 | if (i == node || !node_online(i)) |
e498be7d | 966 | continue; |
83b519e8 | 967 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); |
e498be7d | 968 | if (!ac_ptr[i]) { |
cc550def | 969 | for (i--; i >= 0; i--) |
e498be7d CL |
970 | kfree(ac_ptr[i]); |
971 | kfree(ac_ptr); | |
972 | return NULL; | |
973 | } | |
974 | } | |
975 | } | |
976 | return ac_ptr; | |
977 | } | |
978 | ||
5295a74c | 979 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
980 | { |
981 | int i; | |
982 | ||
983 | if (!ac_ptr) | |
984 | return; | |
e498be7d | 985 | for_each_node(i) |
b28a02de | 986 | kfree(ac_ptr[i]); |
e498be7d CL |
987 | kfree(ac_ptr); |
988 | } | |
989 | ||
343e0d7a | 990 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 991 | struct array_cache *ac, int node) |
e498be7d | 992 | { |
ce8eb6c4 | 993 | struct kmem_cache_node *n = cachep->node[node]; |
e498be7d CL |
994 | |
995 | if (ac->avail) { | |
ce8eb6c4 | 996 | spin_lock(&n->list_lock); |
e00946fe CL |
997 | /* |
998 | * Stuff objects into the remote nodes shared array first. | |
999 | * That way we could avoid the overhead of putting the objects | |
1000 | * into the free lists and getting them back later. | |
1001 | */ | |
ce8eb6c4 CL |
1002 | if (n->shared) |
1003 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 1004 | |
ff69416e | 1005 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d | 1006 | ac->avail = 0; |
ce8eb6c4 | 1007 | spin_unlock(&n->list_lock); |
e498be7d CL |
1008 | } |
1009 | } | |
1010 | ||
8fce4d8e CL |
1011 | /* |
1012 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1013 | */ | |
ce8eb6c4 | 1014 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 1015 | { |
909ea964 | 1016 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 1017 | |
ce8eb6c4 CL |
1018 | if (n->alien) { |
1019 | struct array_cache *ac = n->alien[node]; | |
e00946fe CL |
1020 | |
1021 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { | |
8fce4d8e CL |
1022 | __drain_alien_cache(cachep, ac, node); |
1023 | spin_unlock_irq(&ac->lock); | |
1024 | } | |
1025 | } | |
1026 | } | |
1027 | ||
a737b3e2 AM |
1028 | static void drain_alien_cache(struct kmem_cache *cachep, |
1029 | struct array_cache **alien) | |
e498be7d | 1030 | { |
b28a02de | 1031 | int i = 0; |
e498be7d CL |
1032 | struct array_cache *ac; |
1033 | unsigned long flags; | |
1034 | ||
1035 | for_each_online_node(i) { | |
4484ebf1 | 1036 | ac = alien[i]; |
e498be7d CL |
1037 | if (ac) { |
1038 | spin_lock_irqsave(&ac->lock, flags); | |
1039 | __drain_alien_cache(cachep, ac, i); | |
1040 | spin_unlock_irqrestore(&ac->lock, flags); | |
1041 | } | |
1042 | } | |
1043 | } | |
729bd0b7 | 1044 | |
873623df | 1045 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 1046 | { |
1ea991b0 | 1047 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 1048 | struct kmem_cache_node *n; |
729bd0b7 | 1049 | struct array_cache *alien = NULL; |
1ca4cb24 PE |
1050 | int node; |
1051 | ||
7d6e6d09 | 1052 | node = numa_mem_id(); |
729bd0b7 PE |
1053 | |
1054 | /* | |
1055 | * Make sure we are not freeing a object from another node to the array | |
1056 | * cache on this cpu. | |
1057 | */ | |
1ea991b0 | 1058 | if (likely(nodeid == node)) |
729bd0b7 PE |
1059 | return 0; |
1060 | ||
ce8eb6c4 | 1061 | n = cachep->node[node]; |
729bd0b7 | 1062 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1063 | if (n->alien && n->alien[nodeid]) { |
1064 | alien = n->alien[nodeid]; | |
873623df | 1065 | spin_lock(&alien->lock); |
729bd0b7 PE |
1066 | if (unlikely(alien->avail == alien->limit)) { |
1067 | STATS_INC_ACOVERFLOW(cachep); | |
1068 | __drain_alien_cache(cachep, alien, nodeid); | |
1069 | } | |
072bb0aa | 1070 | ac_put_obj(cachep, alien, objp); |
729bd0b7 PE |
1071 | spin_unlock(&alien->lock); |
1072 | } else { | |
6a67368c | 1073 | spin_lock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 | 1074 | free_block(cachep, &objp, 1, nodeid); |
6a67368c | 1075 | spin_unlock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 PE |
1076 | } |
1077 | return 1; | |
1078 | } | |
e498be7d CL |
1079 | #endif |
1080 | ||
8f9f8d9e | 1081 | /* |
6a67368c | 1082 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1083 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1084 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1085 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1086 | * already in use. |
1087 | * | |
18004c5d | 1088 | * Must hold slab_mutex. |
8f9f8d9e | 1089 | */ |
6a67368c | 1090 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1091 | { |
1092 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1093 | struct kmem_cache_node *n; |
6744f087 | 1094 | const int memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1095 | |
18004c5d | 1096 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e DR |
1097 | /* |
1098 | * Set up the size64 kmemlist for cpu before we can | |
1099 | * begin anything. Make sure some other cpu on this | |
1100 | * node has not already allocated this | |
1101 | */ | |
6a67368c | 1102 | if (!cachep->node[node]) { |
ce8eb6c4 CL |
1103 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1104 | if (!n) | |
8f9f8d9e | 1105 | return -ENOMEM; |
ce8eb6c4 CL |
1106 | kmem_cache_node_init(n); |
1107 | n->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
8f9f8d9e DR |
1108 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
1109 | ||
1110 | /* | |
1111 | * The l3s don't come and go as CPUs come and | |
18004c5d | 1112 | * go. slab_mutex is sufficient |
8f9f8d9e DR |
1113 | * protection here. |
1114 | */ | |
ce8eb6c4 | 1115 | cachep->node[node] = n; |
8f9f8d9e DR |
1116 | } |
1117 | ||
6a67368c CL |
1118 | spin_lock_irq(&cachep->node[node]->list_lock); |
1119 | cachep->node[node]->free_limit = | |
8f9f8d9e DR |
1120 | (1 + nr_cpus_node(node)) * |
1121 | cachep->batchcount + cachep->num; | |
6a67368c | 1122 | spin_unlock_irq(&cachep->node[node]->list_lock); |
8f9f8d9e DR |
1123 | } |
1124 | return 0; | |
1125 | } | |
1126 | ||
0fa8103b WL |
1127 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1128 | struct kmem_cache_node *n) | |
1129 | { | |
1130 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1131 | } | |
1132 | ||
0db0628d | 1133 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1134 | { |
1135 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1136 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1137 | int node = cpu_to_mem(cpu); |
a70f7302 | 1138 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1139 | |
18004c5d | 1140 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1141 | struct array_cache *nc; |
1142 | struct array_cache *shared; | |
1143 | struct array_cache **alien; | |
fbf1e473 | 1144 | |
fbf1e473 AM |
1145 | /* cpu is dead; no one can alloc from it. */ |
1146 | nc = cachep->array[cpu]; | |
1147 | cachep->array[cpu] = NULL; | |
ce8eb6c4 | 1148 | n = cachep->node[node]; |
fbf1e473 | 1149 | |
ce8eb6c4 | 1150 | if (!n) |
fbf1e473 AM |
1151 | goto free_array_cache; |
1152 | ||
ce8eb6c4 | 1153 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1154 | |
ce8eb6c4 CL |
1155 | /* Free limit for this kmem_cache_node */ |
1156 | n->free_limit -= cachep->batchcount; | |
fbf1e473 AM |
1157 | if (nc) |
1158 | free_block(cachep, nc->entry, nc->avail, node); | |
1159 | ||
58463c1f | 1160 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1161 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1162 | goto free_array_cache; |
1163 | } | |
1164 | ||
ce8eb6c4 | 1165 | shared = n->shared; |
fbf1e473 AM |
1166 | if (shared) { |
1167 | free_block(cachep, shared->entry, | |
1168 | shared->avail, node); | |
ce8eb6c4 | 1169 | n->shared = NULL; |
fbf1e473 AM |
1170 | } |
1171 | ||
ce8eb6c4 CL |
1172 | alien = n->alien; |
1173 | n->alien = NULL; | |
fbf1e473 | 1174 | |
ce8eb6c4 | 1175 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1176 | |
1177 | kfree(shared); | |
1178 | if (alien) { | |
1179 | drain_alien_cache(cachep, alien); | |
1180 | free_alien_cache(alien); | |
1181 | } | |
1182 | free_array_cache: | |
1183 | kfree(nc); | |
1184 | } | |
1185 | /* | |
1186 | * In the previous loop, all the objects were freed to | |
1187 | * the respective cache's slabs, now we can go ahead and | |
1188 | * shrink each nodelist to its limit. | |
1189 | */ | |
18004c5d | 1190 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 CL |
1191 | n = cachep->node[node]; |
1192 | if (!n) | |
fbf1e473 | 1193 | continue; |
0fa8103b | 1194 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1195 | } |
1196 | } | |
1197 | ||
0db0628d | 1198 | static int cpuup_prepare(long cpu) |
1da177e4 | 1199 | { |
343e0d7a | 1200 | struct kmem_cache *cachep; |
ce8eb6c4 | 1201 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1202 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1203 | int err; |
1da177e4 | 1204 | |
fbf1e473 AM |
1205 | /* |
1206 | * We need to do this right in the beginning since | |
1207 | * alloc_arraycache's are going to use this list. | |
1208 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1209 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1210 | */ |
6a67368c | 1211 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1212 | if (err < 0) |
1213 | goto bad; | |
fbf1e473 AM |
1214 | |
1215 | /* | |
1216 | * Now we can go ahead with allocating the shared arrays and | |
1217 | * array caches | |
1218 | */ | |
18004c5d | 1219 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1220 | struct array_cache *nc; |
1221 | struct array_cache *shared = NULL; | |
1222 | struct array_cache **alien = NULL; | |
1223 | ||
1224 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1225 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1226 | if (!nc) |
1227 | goto bad; | |
1228 | if (cachep->shared) { | |
1229 | shared = alloc_arraycache(node, | |
1230 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1231 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1232 | if (!shared) { |
1233 | kfree(nc); | |
1da177e4 | 1234 | goto bad; |
12d00f6a | 1235 | } |
fbf1e473 AM |
1236 | } |
1237 | if (use_alien_caches) { | |
83b519e8 | 1238 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1239 | if (!alien) { |
1240 | kfree(shared); | |
1241 | kfree(nc); | |
fbf1e473 | 1242 | goto bad; |
12d00f6a | 1243 | } |
fbf1e473 AM |
1244 | } |
1245 | cachep->array[cpu] = nc; | |
ce8eb6c4 CL |
1246 | n = cachep->node[node]; |
1247 | BUG_ON(!n); | |
fbf1e473 | 1248 | |
ce8eb6c4 CL |
1249 | spin_lock_irq(&n->list_lock); |
1250 | if (!n->shared) { | |
fbf1e473 AM |
1251 | /* |
1252 | * We are serialised from CPU_DEAD or | |
1253 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1254 | */ | |
ce8eb6c4 | 1255 | n->shared = shared; |
fbf1e473 AM |
1256 | shared = NULL; |
1257 | } | |
4484ebf1 | 1258 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1259 | if (!n->alien) { |
1260 | n->alien = alien; | |
fbf1e473 | 1261 | alien = NULL; |
1da177e4 | 1262 | } |
fbf1e473 | 1263 | #endif |
ce8eb6c4 | 1264 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1265 | kfree(shared); |
1266 | free_alien_cache(alien); | |
83835b3d PZ |
1267 | if (cachep->flags & SLAB_DEBUG_OBJECTS) |
1268 | slab_set_debugobj_lock_classes_node(cachep, node); | |
6ccfb5bc GC |
1269 | else if (!OFF_SLAB(cachep) && |
1270 | !(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
1271 | on_slab_lock_classes_node(cachep, node); | |
fbf1e473 | 1272 | } |
ce79ddc8 PE |
1273 | init_node_lock_keys(node); |
1274 | ||
fbf1e473 AM |
1275 | return 0; |
1276 | bad: | |
12d00f6a | 1277 | cpuup_canceled(cpu); |
fbf1e473 AM |
1278 | return -ENOMEM; |
1279 | } | |
1280 | ||
0db0628d | 1281 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1282 | unsigned long action, void *hcpu) |
1283 | { | |
1284 | long cpu = (long)hcpu; | |
1285 | int err = 0; | |
1286 | ||
1287 | switch (action) { | |
fbf1e473 AM |
1288 | case CPU_UP_PREPARE: |
1289 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1290 | mutex_lock(&slab_mutex); |
fbf1e473 | 1291 | err = cpuup_prepare(cpu); |
18004c5d | 1292 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1293 | break; |
1294 | case CPU_ONLINE: | |
8bb78442 | 1295 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1296 | start_cpu_timer(cpu); |
1297 | break; | |
1298 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1299 | case CPU_DOWN_PREPARE: |
8bb78442 | 1300 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1301 | /* |
18004c5d | 1302 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1303 | * held so that if cache_reap() is invoked it cannot do |
1304 | * anything expensive but will only modify reap_work | |
1305 | * and reschedule the timer. | |
1306 | */ | |
afe2c511 | 1307 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1308 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1309 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1310 | break; |
1311 | case CPU_DOWN_FAILED: | |
8bb78442 | 1312 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1313 | start_cpu_timer(cpu); |
1314 | break; | |
1da177e4 | 1315 | case CPU_DEAD: |
8bb78442 | 1316 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1317 | /* |
1318 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1319 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1320 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1321 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1322 | * structure is usually allocated from kmem_cache_create() and |
1323 | * gets destroyed at kmem_cache_destroy(). | |
1324 | */ | |
183ff22b | 1325 | /* fall through */ |
8f5be20b | 1326 | #endif |
1da177e4 | 1327 | case CPU_UP_CANCELED: |
8bb78442 | 1328 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1329 | mutex_lock(&slab_mutex); |
fbf1e473 | 1330 | cpuup_canceled(cpu); |
18004c5d | 1331 | mutex_unlock(&slab_mutex); |
1da177e4 | 1332 | break; |
1da177e4 | 1333 | } |
eac40680 | 1334 | return notifier_from_errno(err); |
1da177e4 LT |
1335 | } |
1336 | ||
0db0628d | 1337 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1338 | &cpuup_callback, NULL, 0 |
1339 | }; | |
1da177e4 | 1340 | |
8f9f8d9e DR |
1341 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1342 | /* | |
1343 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1344 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1345 | * removed. | |
1346 | * | |
18004c5d | 1347 | * Must hold slab_mutex. |
8f9f8d9e | 1348 | */ |
6a67368c | 1349 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1350 | { |
1351 | struct kmem_cache *cachep; | |
1352 | int ret = 0; | |
1353 | ||
18004c5d | 1354 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1355 | struct kmem_cache_node *n; |
8f9f8d9e | 1356 | |
ce8eb6c4 CL |
1357 | n = cachep->node[node]; |
1358 | if (!n) | |
8f9f8d9e DR |
1359 | continue; |
1360 | ||
0fa8103b | 1361 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1362 | |
ce8eb6c4 CL |
1363 | if (!list_empty(&n->slabs_full) || |
1364 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1365 | ret = -EBUSY; |
1366 | break; | |
1367 | } | |
1368 | } | |
1369 | return ret; | |
1370 | } | |
1371 | ||
1372 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1373 | unsigned long action, void *arg) | |
1374 | { | |
1375 | struct memory_notify *mnb = arg; | |
1376 | int ret = 0; | |
1377 | int nid; | |
1378 | ||
1379 | nid = mnb->status_change_nid; | |
1380 | if (nid < 0) | |
1381 | goto out; | |
1382 | ||
1383 | switch (action) { | |
1384 | case MEM_GOING_ONLINE: | |
18004c5d | 1385 | mutex_lock(&slab_mutex); |
6a67368c | 1386 | ret = init_cache_node_node(nid); |
18004c5d | 1387 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1388 | break; |
1389 | case MEM_GOING_OFFLINE: | |
18004c5d | 1390 | mutex_lock(&slab_mutex); |
6a67368c | 1391 | ret = drain_cache_node_node(nid); |
18004c5d | 1392 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1393 | break; |
1394 | case MEM_ONLINE: | |
1395 | case MEM_OFFLINE: | |
1396 | case MEM_CANCEL_ONLINE: | |
1397 | case MEM_CANCEL_OFFLINE: | |
1398 | break; | |
1399 | } | |
1400 | out: | |
5fda1bd5 | 1401 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1402 | } |
1403 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1404 | ||
e498be7d | 1405 | /* |
ce8eb6c4 | 1406 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1407 | */ |
6744f087 | 1408 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1409 | int nodeid) |
e498be7d | 1410 | { |
6744f087 | 1411 | struct kmem_cache_node *ptr; |
e498be7d | 1412 | |
6744f087 | 1413 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1414 | BUG_ON(!ptr); |
1415 | ||
6744f087 | 1416 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1417 | /* |
1418 | * Do not assume that spinlocks can be initialized via memcpy: | |
1419 | */ | |
1420 | spin_lock_init(&ptr->list_lock); | |
1421 | ||
e498be7d | 1422 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1423 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1424 | } |
1425 | ||
556a169d | 1426 | /* |
ce8eb6c4 CL |
1427 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1428 | * size of kmem_cache_node. | |
556a169d | 1429 | */ |
ce8eb6c4 | 1430 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1431 | { |
1432 | int node; | |
1433 | ||
1434 | for_each_online_node(node) { | |
ce8eb6c4 | 1435 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1436 | cachep->node[node]->next_reap = jiffies + |
556a169d PE |
1437 | REAPTIMEOUT_LIST3 + |
1438 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1439 | } | |
1440 | } | |
1441 | ||
3c583465 CL |
1442 | /* |
1443 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1444 | * the node pointer. |
3c583465 | 1445 | */ |
6a67368c | 1446 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1447 | { |
6a67368c | 1448 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1449 | } |
1450 | ||
a737b3e2 AM |
1451 | /* |
1452 | * Initialisation. Called after the page allocator have been initialised and | |
1453 | * before smp_init(). | |
1da177e4 LT |
1454 | */ |
1455 | void __init kmem_cache_init(void) | |
1456 | { | |
e498be7d CL |
1457 | int i; |
1458 | ||
68126702 JK |
1459 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1460 | sizeof(struct rcu_head)); | |
9b030cb8 | 1461 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1462 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1463 | |
b6e68bc1 | 1464 | if (num_possible_nodes() == 1) |
62918a03 SS |
1465 | use_alien_caches = 0; |
1466 | ||
3c583465 | 1467 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1468 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1469 | |
ce8eb6c4 | 1470 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1471 | |
1472 | /* | |
1473 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1474 | * page orders on machines with more than 32MB of memory if |
1475 | * not overridden on the command line. | |
1da177e4 | 1476 | */ |
3df1cccd | 1477 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1478 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1479 | |
1da177e4 LT |
1480 | /* Bootstrap is tricky, because several objects are allocated |
1481 | * from caches that do not exist yet: | |
9b030cb8 CL |
1482 | * 1) initialize the kmem_cache cache: it contains the struct |
1483 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1484 | * kmem_cache is statically allocated. | |
e498be7d | 1485 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1486 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1487 | * array at the end of the bootstrap. |
1da177e4 | 1488 | * 2) Create the first kmalloc cache. |
343e0d7a | 1489 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1490 | * An __init data area is used for the head array. |
1491 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1492 | * head arrays. | |
9b030cb8 | 1493 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1494 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1495 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1496 | * the other cache's with kmalloc allocated memory. |
1497 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1498 | */ |
1499 | ||
9b030cb8 | 1500 | /* 1) create the kmem_cache */ |
1da177e4 | 1501 | |
8da3430d | 1502 | /* |
b56efcf0 | 1503 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1504 | */ |
2f9baa9f CL |
1505 | create_boot_cache(kmem_cache, "kmem_cache", |
1506 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1507 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1508 | SLAB_HWCACHE_ALIGN); |
1509 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1510 | |
1511 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1512 | |
a737b3e2 AM |
1513 | /* |
1514 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1515 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1516 | * bug. |
e498be7d CL |
1517 | */ |
1518 | ||
e3366016 CL |
1519 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1520 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1521 | |
ce8eb6c4 CL |
1522 | if (INDEX_AC != INDEX_NODE) |
1523 | kmalloc_caches[INDEX_NODE] = | |
1524 | create_kmalloc_cache("kmalloc-node", | |
1525 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1526 | |
e0a42726 IM |
1527 | slab_early_init = 0; |
1528 | ||
1da177e4 LT |
1529 | /* 4) Replace the bootstrap head arrays */ |
1530 | { | |
2b2d5493 | 1531 | struct array_cache *ptr; |
e498be7d | 1532 | |
83b519e8 | 1533 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1534 | |
9b030cb8 | 1535 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1536 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1537 | /* |
1538 | * Do not assume that spinlocks can be initialized via memcpy: | |
1539 | */ | |
1540 | spin_lock_init(&ptr->lock); | |
1541 | ||
9b030cb8 | 1542 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1543 | |
83b519e8 | 1544 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1545 | |
e3366016 | 1546 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1547 | != &initarray_generic.cache); |
e3366016 | 1548 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1549 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1550 | /* |
1551 | * Do not assume that spinlocks can be initialized via memcpy: | |
1552 | */ | |
1553 | spin_lock_init(&ptr->lock); | |
1554 | ||
e3366016 | 1555 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1556 | } |
ce8eb6c4 | 1557 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1558 | { |
1ca4cb24 PE |
1559 | int nid; |
1560 | ||
9c09a95c | 1561 | for_each_online_node(nid) { |
ce8eb6c4 | 1562 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1563 | |
e3366016 | 1564 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1565 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1566 | |
ce8eb6c4 CL |
1567 | if (INDEX_AC != INDEX_NODE) { |
1568 | init_list(kmalloc_caches[INDEX_NODE], | |
1569 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1570 | } |
1571 | } | |
1572 | } | |
1da177e4 | 1573 | |
f97d5f63 | 1574 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1575 | } |
1576 | ||
1577 | void __init kmem_cache_init_late(void) | |
1578 | { | |
1579 | struct kmem_cache *cachep; | |
1580 | ||
97d06609 | 1581 | slab_state = UP; |
52cef189 | 1582 | |
8429db5c | 1583 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1584 | mutex_lock(&slab_mutex); |
1585 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1586 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1587 | BUG(); | |
18004c5d | 1588 | mutex_unlock(&slab_mutex); |
056c6241 | 1589 | |
947ca185 MW |
1590 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1591 | init_lock_keys(); | |
1592 | ||
97d06609 CL |
1593 | /* Done! */ |
1594 | slab_state = FULL; | |
1595 | ||
a737b3e2 AM |
1596 | /* |
1597 | * Register a cpu startup notifier callback that initializes | |
1598 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1599 | */ |
1600 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1601 | |
8f9f8d9e DR |
1602 | #ifdef CONFIG_NUMA |
1603 | /* | |
1604 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1605 | * node. |
8f9f8d9e DR |
1606 | */ |
1607 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1608 | #endif | |
1609 | ||
a737b3e2 AM |
1610 | /* |
1611 | * The reap timers are started later, with a module init call: That part | |
1612 | * of the kernel is not yet operational. | |
1da177e4 LT |
1613 | */ |
1614 | } | |
1615 | ||
1616 | static int __init cpucache_init(void) | |
1617 | { | |
1618 | int cpu; | |
1619 | ||
a737b3e2 AM |
1620 | /* |
1621 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1622 | */ |
e498be7d | 1623 | for_each_online_cpu(cpu) |
a737b3e2 | 1624 | start_cpu_timer(cpu); |
a164f896 GC |
1625 | |
1626 | /* Done! */ | |
97d06609 | 1627 | slab_state = FULL; |
1da177e4 LT |
1628 | return 0; |
1629 | } | |
1da177e4 LT |
1630 | __initcall(cpucache_init); |
1631 | ||
8bdec192 RA |
1632 | static noinline void |
1633 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1634 | { | |
ce8eb6c4 | 1635 | struct kmem_cache_node *n; |
8bdec192 RA |
1636 | struct slab *slabp; |
1637 | unsigned long flags; | |
1638 | int node; | |
1639 | ||
1640 | printk(KERN_WARNING | |
1641 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1642 | nodeid, gfpflags); | |
1643 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1644 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 RA |
1645 | |
1646 | for_each_online_node(node) { | |
1647 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; | |
1648 | unsigned long active_slabs = 0, num_slabs = 0; | |
1649 | ||
ce8eb6c4 CL |
1650 | n = cachep->node[node]; |
1651 | if (!n) | |
8bdec192 RA |
1652 | continue; |
1653 | ||
ce8eb6c4 CL |
1654 | spin_lock_irqsave(&n->list_lock, flags); |
1655 | list_for_each_entry(slabp, &n->slabs_full, list) { | |
8bdec192 RA |
1656 | active_objs += cachep->num; |
1657 | active_slabs++; | |
1658 | } | |
ce8eb6c4 | 1659 | list_for_each_entry(slabp, &n->slabs_partial, list) { |
106a74e1 | 1660 | active_objs += slabp->active; |
8bdec192 RA |
1661 | active_slabs++; |
1662 | } | |
ce8eb6c4 | 1663 | list_for_each_entry(slabp, &n->slabs_free, list) |
8bdec192 RA |
1664 | num_slabs++; |
1665 | ||
ce8eb6c4 CL |
1666 | free_objects += n->free_objects; |
1667 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1668 | |
1669 | num_slabs += active_slabs; | |
1670 | num_objs = num_slabs * cachep->num; | |
1671 | printk(KERN_WARNING | |
1672 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1673 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1674 | free_objects); | |
1675 | } | |
1676 | } | |
1677 | ||
1da177e4 LT |
1678 | /* |
1679 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1680 | * | |
1681 | * If we requested dmaable memory, we will get it. Even if we | |
1682 | * did not request dmaable memory, we might get it, but that | |
1683 | * would be relatively rare and ignorable. | |
1684 | */ | |
0c3aa83e JK |
1685 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1686 | int nodeid) | |
1da177e4 LT |
1687 | { |
1688 | struct page *page; | |
e1b6aa6f | 1689 | int nr_pages; |
765c4507 | 1690 | |
a618e89f | 1691 | flags |= cachep->allocflags; |
e12ba74d MG |
1692 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1693 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1694 | |
517d0869 | 1695 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 RA |
1696 | if (!page) { |
1697 | if (!(flags & __GFP_NOWARN) && printk_ratelimit()) | |
1698 | slab_out_of_memory(cachep, flags, nodeid); | |
1da177e4 | 1699 | return NULL; |
8bdec192 | 1700 | } |
1da177e4 | 1701 | |
b37f1dd0 | 1702 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1703 | if (unlikely(page->pfmemalloc)) |
1704 | pfmemalloc_active = true; | |
1705 | ||
e1b6aa6f | 1706 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1707 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1708 | add_zone_page_state(page_zone(page), |
1709 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1710 | else | |
1711 | add_zone_page_state(page_zone(page), | |
1712 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 JK |
1713 | __SetPageSlab(page); |
1714 | if (page->pfmemalloc) | |
1715 | SetPageSlabPfmemalloc(page); | |
1f458cbf | 1716 | memcg_bind_pages(cachep, cachep->gfporder); |
072bb0aa | 1717 | |
b1eeab67 VN |
1718 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1719 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1720 | ||
1721 | if (cachep->ctor) | |
1722 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1723 | else | |
1724 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1725 | } | |
c175eea4 | 1726 | |
0c3aa83e | 1727 | return page; |
1da177e4 LT |
1728 | } |
1729 | ||
1730 | /* | |
1731 | * Interface to system's page release. | |
1732 | */ | |
0c3aa83e | 1733 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1734 | { |
a57a4988 | 1735 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1736 | |
b1eeab67 | 1737 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1738 | |
972d1a7b CL |
1739 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1740 | sub_zone_page_state(page_zone(page), | |
1741 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1742 | else | |
1743 | sub_zone_page_state(page_zone(page), | |
1744 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1745 | |
a57a4988 | 1746 | BUG_ON(!PageSlab(page)); |
73293c2f | 1747 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1748 | __ClearPageSlab(page); |
1f458cbf GC |
1749 | |
1750 | memcg_release_pages(cachep, cachep->gfporder); | |
1da177e4 LT |
1751 | if (current->reclaim_state) |
1752 | current->reclaim_state->reclaimed_slab += nr_freed; | |
0c3aa83e | 1753 | __free_memcg_kmem_pages(page, cachep->gfporder); |
1da177e4 LT |
1754 | } |
1755 | ||
1756 | static void kmem_rcu_free(struct rcu_head *head) | |
1757 | { | |
68126702 JK |
1758 | struct kmem_cache *cachep; |
1759 | struct page *page; | |
1da177e4 | 1760 | |
68126702 JK |
1761 | page = container_of(head, struct page, rcu_head); |
1762 | cachep = page->slab_cache; | |
1763 | ||
1764 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1765 | } |
1766 | ||
1767 | #if DEBUG | |
1768 | ||
1769 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1770 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1771 | unsigned long caller) |
1da177e4 | 1772 | { |
8c138bc0 | 1773 | int size = cachep->object_size; |
1da177e4 | 1774 | |
3dafccf2 | 1775 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1776 | |
b28a02de | 1777 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1778 | return; |
1779 | ||
b28a02de PE |
1780 | *addr++ = 0x12345678; |
1781 | *addr++ = caller; | |
1782 | *addr++ = smp_processor_id(); | |
1783 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1784 | { |
1785 | unsigned long *sptr = &caller; | |
1786 | unsigned long svalue; | |
1787 | ||
1788 | while (!kstack_end(sptr)) { | |
1789 | svalue = *sptr++; | |
1790 | if (kernel_text_address(svalue)) { | |
b28a02de | 1791 | *addr++ = svalue; |
1da177e4 LT |
1792 | size -= sizeof(unsigned long); |
1793 | if (size <= sizeof(unsigned long)) | |
1794 | break; | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | } | |
b28a02de | 1799 | *addr++ = 0x87654321; |
1da177e4 LT |
1800 | } |
1801 | #endif | |
1802 | ||
343e0d7a | 1803 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1804 | { |
8c138bc0 | 1805 | int size = cachep->object_size; |
3dafccf2 | 1806 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1807 | |
1808 | memset(addr, val, size); | |
b28a02de | 1809 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1810 | } |
1811 | ||
1812 | static void dump_line(char *data, int offset, int limit) | |
1813 | { | |
1814 | int i; | |
aa83aa40 DJ |
1815 | unsigned char error = 0; |
1816 | int bad_count = 0; | |
1817 | ||
fdde6abb | 1818 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1819 | for (i = 0; i < limit; i++) { |
1820 | if (data[offset + i] != POISON_FREE) { | |
1821 | error = data[offset + i]; | |
1822 | bad_count++; | |
1823 | } | |
aa83aa40 | 1824 | } |
fdde6abb SAS |
1825 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1826 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1827 | |
1828 | if (bad_count == 1) { | |
1829 | error ^= POISON_FREE; | |
1830 | if (!(error & (error - 1))) { | |
1831 | printk(KERN_ERR "Single bit error detected. Probably " | |
1832 | "bad RAM.\n"); | |
1833 | #ifdef CONFIG_X86 | |
1834 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1835 | "test tool.\n"); | |
1836 | #else | |
1837 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1838 | #endif | |
1839 | } | |
1840 | } | |
1da177e4 LT |
1841 | } |
1842 | #endif | |
1843 | ||
1844 | #if DEBUG | |
1845 | ||
343e0d7a | 1846 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1847 | { |
1848 | int i, size; | |
1849 | char *realobj; | |
1850 | ||
1851 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1852 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1853 | *dbg_redzone1(cachep, objp), |
1854 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1855 | } |
1856 | ||
1857 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1858 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1859 | *dbg_userword(cachep, objp), | |
1860 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1861 | } |
3dafccf2 | 1862 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1863 | size = cachep->object_size; |
b28a02de | 1864 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1865 | int limit; |
1866 | limit = 16; | |
b28a02de PE |
1867 | if (i + limit > size) |
1868 | limit = size - i; | |
1da177e4 LT |
1869 | dump_line(realobj, i, limit); |
1870 | } | |
1871 | } | |
1872 | ||
343e0d7a | 1873 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1874 | { |
1875 | char *realobj; | |
1876 | int size, i; | |
1877 | int lines = 0; | |
1878 | ||
3dafccf2 | 1879 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1880 | size = cachep->object_size; |
1da177e4 | 1881 | |
b28a02de | 1882 | for (i = 0; i < size; i++) { |
1da177e4 | 1883 | char exp = POISON_FREE; |
b28a02de | 1884 | if (i == size - 1) |
1da177e4 LT |
1885 | exp = POISON_END; |
1886 | if (realobj[i] != exp) { | |
1887 | int limit; | |
1888 | /* Mismatch ! */ | |
1889 | /* Print header */ | |
1890 | if (lines == 0) { | |
b28a02de | 1891 | printk(KERN_ERR |
face37f5 DJ |
1892 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1893 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1894 | print_objinfo(cachep, objp, 0); |
1895 | } | |
1896 | /* Hexdump the affected line */ | |
b28a02de | 1897 | i = (i / 16) * 16; |
1da177e4 | 1898 | limit = 16; |
b28a02de PE |
1899 | if (i + limit > size) |
1900 | limit = size - i; | |
1da177e4 LT |
1901 | dump_line(realobj, i, limit); |
1902 | i += 16; | |
1903 | lines++; | |
1904 | /* Limit to 5 lines */ | |
1905 | if (lines > 5) | |
1906 | break; | |
1907 | } | |
1908 | } | |
1909 | if (lines != 0) { | |
1910 | /* Print some data about the neighboring objects, if they | |
1911 | * exist: | |
1912 | */ | |
6ed5eb22 | 1913 | struct slab *slabp = virt_to_slab(objp); |
8fea4e96 | 1914 | unsigned int objnr; |
1da177e4 | 1915 | |
8fea4e96 | 1916 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 | 1917 | if (objnr) { |
8fea4e96 | 1918 | objp = index_to_obj(cachep, slabp, objnr - 1); |
3dafccf2 | 1919 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1920 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1921 | realobj, size); |
1da177e4 LT |
1922 | print_objinfo(cachep, objp, 2); |
1923 | } | |
b28a02de | 1924 | if (objnr + 1 < cachep->num) { |
8fea4e96 | 1925 | objp = index_to_obj(cachep, slabp, objnr + 1); |
3dafccf2 | 1926 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1927 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1928 | realobj, size); |
1da177e4 LT |
1929 | print_objinfo(cachep, objp, 2); |
1930 | } | |
1931 | } | |
1932 | } | |
1933 | #endif | |
1934 | ||
12dd36fa | 1935 | #if DEBUG |
e79aec29 | 1936 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 | 1937 | { |
1da177e4 LT |
1938 | int i; |
1939 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 1940 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
1941 | |
1942 | if (cachep->flags & SLAB_POISON) { | |
1943 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 1944 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 1945 | OFF_SLAB(cachep)) |
b28a02de | 1946 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 1947 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
1948 | else |
1949 | check_poison_obj(cachep, objp); | |
1950 | #else | |
1951 | check_poison_obj(cachep, objp); | |
1952 | #endif | |
1953 | } | |
1954 | if (cachep->flags & SLAB_RED_ZONE) { | |
1955 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1956 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1957 | "was overwritten"); |
1da177e4 LT |
1958 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1959 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1960 | "was overwritten"); |
1da177e4 | 1961 | } |
1da177e4 | 1962 | } |
12dd36fa | 1963 | } |
1da177e4 | 1964 | #else |
e79aec29 | 1965 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 1966 | { |
12dd36fa | 1967 | } |
1da177e4 LT |
1968 | #endif |
1969 | ||
911851e6 RD |
1970 | /** |
1971 | * slab_destroy - destroy and release all objects in a slab | |
1972 | * @cachep: cache pointer being destroyed | |
1973 | * @slabp: slab pointer being destroyed | |
1974 | * | |
12dd36fa | 1975 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
1976 | * Before calling the slab must have been unlinked from the cache. The |
1977 | * cache-lock is not held/needed. | |
12dd36fa | 1978 | */ |
343e0d7a | 1979 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 1980 | { |
0c3aa83e | 1981 | struct page *page = virt_to_head_page(slabp->s_mem); |
12dd36fa | 1982 | |
e79aec29 | 1983 | slab_destroy_debugcheck(cachep, slabp); |
1da177e4 | 1984 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
1985 | struct rcu_head *head; |
1986 | ||
1987 | /* | |
1988 | * RCU free overloads the RCU head over the LRU. | |
1989 | * slab_page has been overloeaded over the LRU, | |
1990 | * however it is not used from now on so that | |
1991 | * we can use it safely. | |
1992 | */ | |
1993 | head = (void *)&page->rcu_head; | |
1994 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 1995 | |
1da177e4 | 1996 | } else { |
0c3aa83e | 1997 | kmem_freepages(cachep, page); |
1da177e4 | 1998 | } |
68126702 JK |
1999 | |
2000 | /* | |
2001 | * From now on, we don't use slab management | |
2002 | * although actual page can be freed in rcu context | |
2003 | */ | |
2004 | if (OFF_SLAB(cachep)) | |
2005 | kmem_cache_free(cachep->slabp_cache, slabp); | |
1da177e4 LT |
2006 | } |
2007 | ||
4d268eba | 2008 | /** |
a70773dd RD |
2009 | * calculate_slab_order - calculate size (page order) of slabs |
2010 | * @cachep: pointer to the cache that is being created | |
2011 | * @size: size of objects to be created in this cache. | |
2012 | * @align: required alignment for the objects. | |
2013 | * @flags: slab allocation flags | |
2014 | * | |
2015 | * Also calculates the number of objects per slab. | |
4d268eba PE |
2016 | * |
2017 | * This could be made much more intelligent. For now, try to avoid using | |
2018 | * high order pages for slabs. When the gfp() functions are more friendly | |
2019 | * towards high-order requests, this should be changed. | |
2020 | */ | |
a737b3e2 | 2021 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 2022 | size_t size, size_t align, unsigned long flags) |
4d268eba | 2023 | { |
b1ab41c4 | 2024 | unsigned long offslab_limit; |
4d268eba | 2025 | size_t left_over = 0; |
9888e6fa | 2026 | int gfporder; |
4d268eba | 2027 | |
0aa817f0 | 2028 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
2029 | unsigned int num; |
2030 | size_t remainder; | |
2031 | ||
9888e6fa | 2032 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
2033 | if (!num) |
2034 | continue; | |
9888e6fa | 2035 | |
b1ab41c4 IM |
2036 | if (flags & CFLGS_OFF_SLAB) { |
2037 | /* | |
2038 | * Max number of objs-per-slab for caches which | |
2039 | * use off-slab slabs. Needed to avoid a possible | |
2040 | * looping condition in cache_grow(). | |
2041 | */ | |
2042 | offslab_limit = size - sizeof(struct slab); | |
16025177 | 2043 | offslab_limit /= sizeof(unsigned int); |
b1ab41c4 IM |
2044 | |
2045 | if (num > offslab_limit) | |
2046 | break; | |
2047 | } | |
4d268eba | 2048 | |
9888e6fa | 2049 | /* Found something acceptable - save it away */ |
4d268eba | 2050 | cachep->num = num; |
9888e6fa | 2051 | cachep->gfporder = gfporder; |
4d268eba PE |
2052 | left_over = remainder; |
2053 | ||
f78bb8ad LT |
2054 | /* |
2055 | * A VFS-reclaimable slab tends to have most allocations | |
2056 | * as GFP_NOFS and we really don't want to have to be allocating | |
2057 | * higher-order pages when we are unable to shrink dcache. | |
2058 | */ | |
2059 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2060 | break; | |
2061 | ||
4d268eba PE |
2062 | /* |
2063 | * Large number of objects is good, but very large slabs are | |
2064 | * currently bad for the gfp()s. | |
2065 | */ | |
543585cc | 2066 | if (gfporder >= slab_max_order) |
4d268eba PE |
2067 | break; |
2068 | ||
9888e6fa LT |
2069 | /* |
2070 | * Acceptable internal fragmentation? | |
2071 | */ | |
a737b3e2 | 2072 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2073 | break; |
2074 | } | |
2075 | return left_over; | |
2076 | } | |
2077 | ||
83b519e8 | 2078 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2079 | { |
97d06609 | 2080 | if (slab_state >= FULL) |
83b519e8 | 2081 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2082 | |
97d06609 | 2083 | if (slab_state == DOWN) { |
f30cf7d1 | 2084 | /* |
2f9baa9f | 2085 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2086 | * The setup_node is taken care |
2f9baa9f CL |
2087 | * of by the caller of __kmem_cache_create |
2088 | */ | |
2089 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2090 | slab_state = PARTIAL; | |
2091 | } else if (slab_state == PARTIAL) { | |
2092 | /* | |
2093 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2094 | * that's used by kmalloc(24), otherwise the creation of |
2095 | * further caches will BUG(). | |
2096 | */ | |
2097 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2098 | ||
2099 | /* | |
ce8eb6c4 CL |
2100 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2101 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2102 | * otherwise the creation of further caches will BUG(). |
2103 | */ | |
ce8eb6c4 CL |
2104 | set_up_node(cachep, SIZE_AC); |
2105 | if (INDEX_AC == INDEX_NODE) | |
2106 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2107 | else |
97d06609 | 2108 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2109 | } else { |
2f9baa9f | 2110 | /* Remaining boot caches */ |
f30cf7d1 | 2111 | cachep->array[smp_processor_id()] = |
83b519e8 | 2112 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2113 | |
97d06609 | 2114 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2115 | set_up_node(cachep, SIZE_NODE); |
2116 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2117 | } else { |
2118 | int node; | |
556a169d | 2119 | for_each_online_node(node) { |
6a67368c | 2120 | cachep->node[node] = |
6744f087 | 2121 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2122 | gfp, node); |
6a67368c | 2123 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2124 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2125 | } |
2126 | } | |
2127 | } | |
6a67368c | 2128 | cachep->node[numa_mem_id()]->next_reap = |
f30cf7d1 PE |
2129 | jiffies + REAPTIMEOUT_LIST3 + |
2130 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
2131 | ||
2132 | cpu_cache_get(cachep)->avail = 0; | |
2133 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2134 | cpu_cache_get(cachep)->batchcount = 1; | |
2135 | cpu_cache_get(cachep)->touched = 0; | |
2136 | cachep->batchcount = 1; | |
2137 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2138 | return 0; |
f30cf7d1 PE |
2139 | } |
2140 | ||
1da177e4 | 2141 | /** |
039363f3 | 2142 | * __kmem_cache_create - Create a cache. |
a755b76a | 2143 | * @cachep: cache management descriptor |
1da177e4 | 2144 | * @flags: SLAB flags |
1da177e4 LT |
2145 | * |
2146 | * Returns a ptr to the cache on success, NULL on failure. | |
2147 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2148 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2149 | * |
1da177e4 LT |
2150 | * The flags are |
2151 | * | |
2152 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2153 | * to catch references to uninitialised memory. | |
2154 | * | |
2155 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2156 | * for buffer overruns. | |
2157 | * | |
1da177e4 LT |
2158 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2159 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2160 | * as davem. | |
2161 | */ | |
278b1bb1 | 2162 | int |
8a13a4cc | 2163 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 LT |
2164 | { |
2165 | size_t left_over, slab_size, ralign; | |
83b519e8 | 2166 | gfp_t gfp; |
278b1bb1 | 2167 | int err; |
8a13a4cc | 2168 | size_t size = cachep->size; |
1da177e4 | 2169 | |
1da177e4 | 2170 | #if DEBUG |
1da177e4 LT |
2171 | #if FORCED_DEBUG |
2172 | /* | |
2173 | * Enable redzoning and last user accounting, except for caches with | |
2174 | * large objects, if the increased size would increase the object size | |
2175 | * above the next power of two: caches with object sizes just above a | |
2176 | * power of two have a significant amount of internal fragmentation. | |
2177 | */ | |
87a927c7 DW |
2178 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2179 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2180 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2181 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2182 | flags |= SLAB_POISON; | |
2183 | #endif | |
2184 | if (flags & SLAB_DESTROY_BY_RCU) | |
2185 | BUG_ON(flags & SLAB_POISON); | |
2186 | #endif | |
1da177e4 | 2187 | |
a737b3e2 AM |
2188 | /* |
2189 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2190 | * unaligned accesses for some archs when redzoning is used, and makes |
2191 | * sure any on-slab bufctl's are also correctly aligned. | |
2192 | */ | |
b28a02de PE |
2193 | if (size & (BYTES_PER_WORD - 1)) { |
2194 | size += (BYTES_PER_WORD - 1); | |
2195 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2196 | } |
2197 | ||
ca5f9703 | 2198 | /* |
87a927c7 DW |
2199 | * Redzoning and user store require word alignment or possibly larger. |
2200 | * Note this will be overridden by architecture or caller mandated | |
2201 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2202 | */ |
87a927c7 DW |
2203 | if (flags & SLAB_STORE_USER) |
2204 | ralign = BYTES_PER_WORD; | |
2205 | ||
2206 | if (flags & SLAB_RED_ZONE) { | |
2207 | ralign = REDZONE_ALIGN; | |
2208 | /* If redzoning, ensure that the second redzone is suitably | |
2209 | * aligned, by adjusting the object size accordingly. */ | |
2210 | size += REDZONE_ALIGN - 1; | |
2211 | size &= ~(REDZONE_ALIGN - 1); | |
2212 | } | |
ca5f9703 | 2213 | |
a44b56d3 | 2214 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2215 | if (ralign < cachep->align) { |
2216 | ralign = cachep->align; | |
1da177e4 | 2217 | } |
3ff84a7f PE |
2218 | /* disable debug if necessary */ |
2219 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2220 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2221 | /* |
ca5f9703 | 2222 | * 4) Store it. |
1da177e4 | 2223 | */ |
8a13a4cc | 2224 | cachep->align = ralign; |
1da177e4 | 2225 | |
83b519e8 PE |
2226 | if (slab_is_available()) |
2227 | gfp = GFP_KERNEL; | |
2228 | else | |
2229 | gfp = GFP_NOWAIT; | |
2230 | ||
6a67368c | 2231 | setup_node_pointer(cachep); |
1da177e4 | 2232 | #if DEBUG |
1da177e4 | 2233 | |
ca5f9703 PE |
2234 | /* |
2235 | * Both debugging options require word-alignment which is calculated | |
2236 | * into align above. | |
2237 | */ | |
1da177e4 | 2238 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2239 | /* add space for red zone words */ |
3ff84a7f PE |
2240 | cachep->obj_offset += sizeof(unsigned long long); |
2241 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2242 | } |
2243 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2244 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2245 | * the real object. But if the second red zone needs to be |
2246 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2247 | */ |
87a927c7 DW |
2248 | if (flags & SLAB_RED_ZONE) |
2249 | size += REDZONE_ALIGN; | |
2250 | else | |
2251 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2252 | } |
2253 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2254 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2255 | && cachep->object_size > cache_line_size() |
2256 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2257 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2258 | size = PAGE_SIZE; |
2259 | } | |
2260 | #endif | |
2261 | #endif | |
2262 | ||
e0a42726 IM |
2263 | /* |
2264 | * Determine if the slab management is 'on' or 'off' slab. | |
2265 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2266 | * it too early on. Always use on-slab management when |
2267 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2268 | */ |
e7cb55b9 CM |
2269 | if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init && |
2270 | !(flags & SLAB_NOLEAKTRACE)) | |
1da177e4 LT |
2271 | /* |
2272 | * Size is large, assume best to place the slab management obj | |
2273 | * off-slab (should allow better packing of objs). | |
2274 | */ | |
2275 | flags |= CFLGS_OFF_SLAB; | |
2276 | ||
8a13a4cc | 2277 | size = ALIGN(size, cachep->align); |
1da177e4 | 2278 | |
8a13a4cc | 2279 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2280 | |
8a13a4cc | 2281 | if (!cachep->num) |
278b1bb1 | 2282 | return -E2BIG; |
1da177e4 | 2283 | |
16025177 | 2284 | slab_size = ALIGN(cachep->num * sizeof(unsigned int) |
8a13a4cc | 2285 | + sizeof(struct slab), cachep->align); |
1da177e4 LT |
2286 | |
2287 | /* | |
2288 | * If the slab has been placed off-slab, and we have enough space then | |
2289 | * move it on-slab. This is at the expense of any extra colouring. | |
2290 | */ | |
2291 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { | |
2292 | flags &= ~CFLGS_OFF_SLAB; | |
2293 | left_over -= slab_size; | |
2294 | } | |
2295 | ||
2296 | if (flags & CFLGS_OFF_SLAB) { | |
2297 | /* really off slab. No need for manual alignment */ | |
b28a02de | 2298 | slab_size = |
16025177 | 2299 | cachep->num * sizeof(unsigned int) + sizeof(struct slab); |
67461365 RL |
2300 | |
2301 | #ifdef CONFIG_PAGE_POISONING | |
2302 | /* If we're going to use the generic kernel_map_pages() | |
2303 | * poisoning, then it's going to smash the contents of | |
2304 | * the redzone and userword anyhow, so switch them off. | |
2305 | */ | |
2306 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2307 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2308 | #endif | |
1da177e4 LT |
2309 | } |
2310 | ||
2311 | cachep->colour_off = cache_line_size(); | |
2312 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2313 | if (cachep->colour_off < cachep->align) |
2314 | cachep->colour_off = cachep->align; | |
b28a02de | 2315 | cachep->colour = left_over / cachep->colour_off; |
1da177e4 LT |
2316 | cachep->slab_size = slab_size; |
2317 | cachep->flags = flags; | |
a57a4988 | 2318 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2319 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2320 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2321 | cachep->size = size; |
6a2d7a95 | 2322 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2323 | |
e5ac9c5a | 2324 | if (flags & CFLGS_OFF_SLAB) { |
2c59dd65 | 2325 | cachep->slabp_cache = kmalloc_slab(slab_size, 0u); |
e5ac9c5a RT |
2326 | /* |
2327 | * This is a possibility for one of the malloc_sizes caches. | |
2328 | * But since we go off slab only for object size greater than | |
2329 | * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, | |
2330 | * this should not happen at all. | |
2331 | * But leave a BUG_ON for some lucky dude. | |
2332 | */ | |
6cb8f913 | 2333 | BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache)); |
e5ac9c5a | 2334 | } |
1da177e4 | 2335 | |
278b1bb1 CL |
2336 | err = setup_cpu_cache(cachep, gfp); |
2337 | if (err) { | |
12c3667f | 2338 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2339 | return err; |
2ed3a4ef | 2340 | } |
1da177e4 | 2341 | |
83835b3d PZ |
2342 | if (flags & SLAB_DEBUG_OBJECTS) { |
2343 | /* | |
2344 | * Would deadlock through slab_destroy()->call_rcu()-> | |
2345 | * debug_object_activate()->kmem_cache_alloc(). | |
2346 | */ | |
2347 | WARN_ON_ONCE(flags & SLAB_DESTROY_BY_RCU); | |
2348 | ||
2349 | slab_set_debugobj_lock_classes(cachep); | |
6ccfb5bc GC |
2350 | } else if (!OFF_SLAB(cachep) && !(flags & SLAB_DESTROY_BY_RCU)) |
2351 | on_slab_lock_classes(cachep); | |
83835b3d | 2352 | |
278b1bb1 | 2353 | return 0; |
1da177e4 | 2354 | } |
1da177e4 LT |
2355 | |
2356 | #if DEBUG | |
2357 | static void check_irq_off(void) | |
2358 | { | |
2359 | BUG_ON(!irqs_disabled()); | |
2360 | } | |
2361 | ||
2362 | static void check_irq_on(void) | |
2363 | { | |
2364 | BUG_ON(irqs_disabled()); | |
2365 | } | |
2366 | ||
343e0d7a | 2367 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2368 | { |
2369 | #ifdef CONFIG_SMP | |
2370 | check_irq_off(); | |
6a67368c | 2371 | assert_spin_locked(&cachep->node[numa_mem_id()]->list_lock); |
1da177e4 LT |
2372 | #endif |
2373 | } | |
e498be7d | 2374 | |
343e0d7a | 2375 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2376 | { |
2377 | #ifdef CONFIG_SMP | |
2378 | check_irq_off(); | |
6a67368c | 2379 | assert_spin_locked(&cachep->node[node]->list_lock); |
e498be7d CL |
2380 | #endif |
2381 | } | |
2382 | ||
1da177e4 LT |
2383 | #else |
2384 | #define check_irq_off() do { } while(0) | |
2385 | #define check_irq_on() do { } while(0) | |
2386 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2387 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2388 | #endif |
2389 | ||
ce8eb6c4 | 2390 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2391 | struct array_cache *ac, |
2392 | int force, int node); | |
2393 | ||
1da177e4 LT |
2394 | static void do_drain(void *arg) |
2395 | { | |
a737b3e2 | 2396 | struct kmem_cache *cachep = arg; |
1da177e4 | 2397 | struct array_cache *ac; |
7d6e6d09 | 2398 | int node = numa_mem_id(); |
1da177e4 LT |
2399 | |
2400 | check_irq_off(); | |
9a2dba4b | 2401 | ac = cpu_cache_get(cachep); |
6a67368c | 2402 | spin_lock(&cachep->node[node]->list_lock); |
ff69416e | 2403 | free_block(cachep, ac->entry, ac->avail, node); |
6a67368c | 2404 | spin_unlock(&cachep->node[node]->list_lock); |
1da177e4 LT |
2405 | ac->avail = 0; |
2406 | } | |
2407 | ||
343e0d7a | 2408 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2409 | { |
ce8eb6c4 | 2410 | struct kmem_cache_node *n; |
e498be7d CL |
2411 | int node; |
2412 | ||
15c8b6c1 | 2413 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2414 | check_irq_on(); |
b28a02de | 2415 | for_each_online_node(node) { |
ce8eb6c4 CL |
2416 | n = cachep->node[node]; |
2417 | if (n && n->alien) | |
2418 | drain_alien_cache(cachep, n->alien); | |
a4523a8b RD |
2419 | } |
2420 | ||
2421 | for_each_online_node(node) { | |
ce8eb6c4 CL |
2422 | n = cachep->node[node]; |
2423 | if (n) | |
2424 | drain_array(cachep, n, n->shared, 1, node); | |
e498be7d | 2425 | } |
1da177e4 LT |
2426 | } |
2427 | ||
ed11d9eb CL |
2428 | /* |
2429 | * Remove slabs from the list of free slabs. | |
2430 | * Specify the number of slabs to drain in tofree. | |
2431 | * | |
2432 | * Returns the actual number of slabs released. | |
2433 | */ | |
2434 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2435 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2436 | { |
ed11d9eb CL |
2437 | struct list_head *p; |
2438 | int nr_freed; | |
1da177e4 | 2439 | struct slab *slabp; |
1da177e4 | 2440 | |
ed11d9eb | 2441 | nr_freed = 0; |
ce8eb6c4 | 2442 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2443 | |
ce8eb6c4 CL |
2444 | spin_lock_irq(&n->list_lock); |
2445 | p = n->slabs_free.prev; | |
2446 | if (p == &n->slabs_free) { | |
2447 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2448 | goto out; |
2449 | } | |
1da177e4 | 2450 | |
ed11d9eb | 2451 | slabp = list_entry(p, struct slab, list); |
1da177e4 | 2452 | #if DEBUG |
106a74e1 | 2453 | BUG_ON(slabp->active); |
1da177e4 LT |
2454 | #endif |
2455 | list_del(&slabp->list); | |
ed11d9eb CL |
2456 | /* |
2457 | * Safe to drop the lock. The slab is no longer linked | |
2458 | * to the cache. | |
2459 | */ | |
ce8eb6c4 CL |
2460 | n->free_objects -= cache->num; |
2461 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2462 | slab_destroy(cache, slabp); |
2463 | nr_freed++; | |
1da177e4 | 2464 | } |
ed11d9eb CL |
2465 | out: |
2466 | return nr_freed; | |
1da177e4 LT |
2467 | } |
2468 | ||
18004c5d | 2469 | /* Called with slab_mutex held to protect against cpu hotplug */ |
343e0d7a | 2470 | static int __cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2471 | { |
2472 | int ret = 0, i = 0; | |
ce8eb6c4 | 2473 | struct kmem_cache_node *n; |
e498be7d CL |
2474 | |
2475 | drain_cpu_caches(cachep); | |
2476 | ||
2477 | check_irq_on(); | |
2478 | for_each_online_node(i) { | |
ce8eb6c4 CL |
2479 | n = cachep->node[i]; |
2480 | if (!n) | |
ed11d9eb CL |
2481 | continue; |
2482 | ||
0fa8103b | 2483 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2484 | |
ce8eb6c4 CL |
2485 | ret += !list_empty(&n->slabs_full) || |
2486 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2487 | } |
2488 | return (ret ? 1 : 0); | |
2489 | } | |
2490 | ||
1da177e4 LT |
2491 | /** |
2492 | * kmem_cache_shrink - Shrink a cache. | |
2493 | * @cachep: The cache to shrink. | |
2494 | * | |
2495 | * Releases as many slabs as possible for a cache. | |
2496 | * To help debugging, a zero exit status indicates all slabs were released. | |
2497 | */ | |
343e0d7a | 2498 | int kmem_cache_shrink(struct kmem_cache *cachep) |
1da177e4 | 2499 | { |
8f5be20b | 2500 | int ret; |
40094fa6 | 2501 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2502 | |
95402b38 | 2503 | get_online_cpus(); |
18004c5d | 2504 | mutex_lock(&slab_mutex); |
8f5be20b | 2505 | ret = __cache_shrink(cachep); |
18004c5d | 2506 | mutex_unlock(&slab_mutex); |
95402b38 | 2507 | put_online_cpus(); |
8f5be20b | 2508 | return ret; |
1da177e4 LT |
2509 | } |
2510 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2511 | ||
945cf2b6 | 2512 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2513 | { |
12c3667f | 2514 | int i; |
ce8eb6c4 | 2515 | struct kmem_cache_node *n; |
12c3667f | 2516 | int rc = __cache_shrink(cachep); |
1da177e4 | 2517 | |
12c3667f CL |
2518 | if (rc) |
2519 | return rc; | |
1da177e4 | 2520 | |
12c3667f CL |
2521 | for_each_online_cpu(i) |
2522 | kfree(cachep->array[i]); | |
1da177e4 | 2523 | |
ce8eb6c4 | 2524 | /* NUMA: free the node structures */ |
12c3667f | 2525 | for_each_online_node(i) { |
ce8eb6c4 CL |
2526 | n = cachep->node[i]; |
2527 | if (n) { | |
2528 | kfree(n->shared); | |
2529 | free_alien_cache(n->alien); | |
2530 | kfree(n); | |
12c3667f CL |
2531 | } |
2532 | } | |
2533 | return 0; | |
1da177e4 | 2534 | } |
1da177e4 | 2535 | |
e5ac9c5a RT |
2536 | /* |
2537 | * Get the memory for a slab management obj. | |
2538 | * For a slab cache when the slab descriptor is off-slab, slab descriptors | |
2539 | * always come from malloc_sizes caches. The slab descriptor cannot | |
2540 | * come from the same cache which is getting created because, | |
2541 | * when we are searching for an appropriate cache for these | |
2542 | * descriptors in kmem_cache_create, we search through the malloc_sizes array. | |
2543 | * If we are creating a malloc_sizes cache here it would not be visible to | |
2544 | * kmem_find_general_cachep till the initialization is complete. | |
2545 | * Hence we cannot have slabp_cache same as the original cache. | |
2546 | */ | |
0c3aa83e JK |
2547 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, |
2548 | struct page *page, int colour_off, | |
2549 | gfp_t local_flags, int nodeid) | |
1da177e4 LT |
2550 | { |
2551 | struct slab *slabp; | |
0c3aa83e | 2552 | void *addr = page_address(page); |
b28a02de | 2553 | |
1da177e4 LT |
2554 | if (OFF_SLAB(cachep)) { |
2555 | /* Slab management obj is off-slab. */ | |
5b74ada7 | 2556 | slabp = kmem_cache_alloc_node(cachep->slabp_cache, |
8759ec50 | 2557 | local_flags, nodeid); |
d5cff635 CM |
2558 | /* |
2559 | * If the first object in the slab is leaked (it's allocated | |
2560 | * but no one has a reference to it), we want to make sure | |
2561 | * kmemleak does not treat the ->s_mem pointer as a reference | |
2562 | * to the object. Otherwise we will not report the leak. | |
2563 | */ | |
c017b4be CM |
2564 | kmemleak_scan_area(&slabp->list, sizeof(struct list_head), |
2565 | local_flags); | |
1da177e4 LT |
2566 | if (!slabp) |
2567 | return NULL; | |
2568 | } else { | |
0c3aa83e | 2569 | slabp = addr + colour_off; |
1da177e4 LT |
2570 | colour_off += cachep->slab_size; |
2571 | } | |
106a74e1 | 2572 | slabp->active = 0; |
0c3aa83e | 2573 | slabp->s_mem = addr + colour_off; |
1da177e4 LT |
2574 | return slabp; |
2575 | } | |
2576 | ||
16025177 | 2577 | static inline unsigned int *slab_bufctl(struct slab *slabp) |
1da177e4 | 2578 | { |
16025177 | 2579 | return (unsigned int *) (slabp + 1); |
1da177e4 LT |
2580 | } |
2581 | ||
343e0d7a | 2582 | static void cache_init_objs(struct kmem_cache *cachep, |
a35afb83 | 2583 | struct slab *slabp) |
1da177e4 LT |
2584 | { |
2585 | int i; | |
2586 | ||
2587 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2588 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2589 | #if DEBUG |
2590 | /* need to poison the objs? */ | |
2591 | if (cachep->flags & SLAB_POISON) | |
2592 | poison_obj(cachep, objp, POISON_FREE); | |
2593 | if (cachep->flags & SLAB_STORE_USER) | |
2594 | *dbg_userword(cachep, objp) = NULL; | |
2595 | ||
2596 | if (cachep->flags & SLAB_RED_ZONE) { | |
2597 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2598 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2599 | } | |
2600 | /* | |
a737b3e2 AM |
2601 | * Constructors are not allowed to allocate memory from the same |
2602 | * cache which they are a constructor for. Otherwise, deadlock. | |
2603 | * They must also be threaded. | |
1da177e4 LT |
2604 | */ |
2605 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2606 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2607 | |
2608 | if (cachep->flags & SLAB_RED_ZONE) { | |
2609 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2610 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2611 | " end of an object"); |
1da177e4 LT |
2612 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2613 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2614 | " start of an object"); |
1da177e4 | 2615 | } |
3b0efdfa | 2616 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2617 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2618 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2619 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2620 | #else |
2621 | if (cachep->ctor) | |
51cc5068 | 2622 | cachep->ctor(objp); |
1da177e4 | 2623 | #endif |
b1cb0982 | 2624 | slab_bufctl(slabp)[i] = i; |
1da177e4 | 2625 | } |
1da177e4 LT |
2626 | } |
2627 | ||
343e0d7a | 2628 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2629 | { |
4b51d669 CL |
2630 | if (CONFIG_ZONE_DMA_FLAG) { |
2631 | if (flags & GFP_DMA) | |
a618e89f | 2632 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2633 | else |
a618e89f | 2634 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2635 | } |
1da177e4 LT |
2636 | } |
2637 | ||
a737b3e2 AM |
2638 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
2639 | int nodeid) | |
78d382d7 | 2640 | { |
b1cb0982 | 2641 | void *objp; |
78d382d7 | 2642 | |
106a74e1 JK |
2643 | objp = index_to_obj(cachep, slabp, slab_bufctl(slabp)[slabp->active]); |
2644 | slabp->active++; | |
78d382d7 | 2645 | #if DEBUG |
1ea991b0 | 2646 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2647 | #endif |
78d382d7 MD |
2648 | |
2649 | return objp; | |
2650 | } | |
2651 | ||
a737b3e2 AM |
2652 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
2653 | void *objp, int nodeid) | |
78d382d7 | 2654 | { |
8fea4e96 | 2655 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
78d382d7 | 2656 | #if DEBUG |
16025177 | 2657 | unsigned int i; |
b1cb0982 | 2658 | |
78d382d7 | 2659 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2660 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2661 | |
b1cb0982 | 2662 | /* Verify double free bug */ |
106a74e1 | 2663 | for (i = slabp->active; i < cachep->num; i++) { |
b1cb0982 JK |
2664 | if (slab_bufctl(slabp)[i] == objnr) { |
2665 | printk(KERN_ERR "slab: double free detected in cache " | |
2666 | "'%s', objp %p\n", cachep->name, objp); | |
2667 | BUG(); | |
2668 | } | |
78d382d7 MD |
2669 | } |
2670 | #endif | |
106a74e1 JK |
2671 | slabp->active--; |
2672 | slab_bufctl(slabp)[slabp->active] = objnr; | |
78d382d7 MD |
2673 | } |
2674 | ||
4776874f PE |
2675 | /* |
2676 | * Map pages beginning at addr to the given cache and slab. This is required | |
2677 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2678 | * virtual address for kfree, ksize, and slab debugging. |
4776874f PE |
2679 | */ |
2680 | static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, | |
0c3aa83e | 2681 | struct page *page) |
1da177e4 | 2682 | { |
a57a4988 JK |
2683 | page->slab_cache = cache; |
2684 | page->slab_page = slab; | |
1da177e4 LT |
2685 | } |
2686 | ||
2687 | /* | |
2688 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2689 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2690 | */ | |
3c517a61 | 2691 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2692 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2693 | { |
b28a02de | 2694 | struct slab *slabp; |
b28a02de PE |
2695 | size_t offset; |
2696 | gfp_t local_flags; | |
ce8eb6c4 | 2697 | struct kmem_cache_node *n; |
1da177e4 | 2698 | |
a737b3e2 AM |
2699 | /* |
2700 | * Be lazy and only check for valid flags here, keeping it out of the | |
2701 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2702 | */ |
6cb06229 CL |
2703 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2704 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2705 | |
ce8eb6c4 | 2706 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2707 | check_irq_off(); |
ce8eb6c4 CL |
2708 | n = cachep->node[nodeid]; |
2709 | spin_lock(&n->list_lock); | |
1da177e4 LT |
2710 | |
2711 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2712 | offset = n->colour_next; |
2713 | n->colour_next++; | |
2714 | if (n->colour_next >= cachep->colour) | |
2715 | n->colour_next = 0; | |
2716 | spin_unlock(&n->list_lock); | |
1da177e4 | 2717 | |
2e1217cf | 2718 | offset *= cachep->colour_off; |
1da177e4 LT |
2719 | |
2720 | if (local_flags & __GFP_WAIT) | |
2721 | local_irq_enable(); | |
2722 | ||
2723 | /* | |
2724 | * The test for missing atomic flag is performed here, rather than | |
2725 | * the more obvious place, simply to reduce the critical path length | |
2726 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2727 | * will eventually be caught here (where it matters). | |
2728 | */ | |
2729 | kmem_flagcheck(cachep, flags); | |
2730 | ||
a737b3e2 AM |
2731 | /* |
2732 | * Get mem for the objs. Attempt to allocate a physical page from | |
2733 | * 'nodeid'. | |
e498be7d | 2734 | */ |
0c3aa83e JK |
2735 | if (!page) |
2736 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2737 | if (!page) | |
1da177e4 LT |
2738 | goto failed; |
2739 | ||
2740 | /* Get slab management. */ | |
0c3aa83e | 2741 | slabp = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2742 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
a737b3e2 | 2743 | if (!slabp) |
1da177e4 LT |
2744 | goto opps1; |
2745 | ||
0c3aa83e | 2746 | slab_map_pages(cachep, slabp, page); |
1da177e4 | 2747 | |
a35afb83 | 2748 | cache_init_objs(cachep, slabp); |
1da177e4 LT |
2749 | |
2750 | if (local_flags & __GFP_WAIT) | |
2751 | local_irq_disable(); | |
2752 | check_irq_off(); | |
ce8eb6c4 | 2753 | spin_lock(&n->list_lock); |
1da177e4 LT |
2754 | |
2755 | /* Make slab active. */ | |
ce8eb6c4 | 2756 | list_add_tail(&slabp->list, &(n->slabs_free)); |
1da177e4 | 2757 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2758 | n->free_objects += cachep->num; |
2759 | spin_unlock(&n->list_lock); | |
1da177e4 | 2760 | return 1; |
a737b3e2 | 2761 | opps1: |
0c3aa83e | 2762 | kmem_freepages(cachep, page); |
a737b3e2 | 2763 | failed: |
1da177e4 LT |
2764 | if (local_flags & __GFP_WAIT) |
2765 | local_irq_disable(); | |
2766 | return 0; | |
2767 | } | |
2768 | ||
2769 | #if DEBUG | |
2770 | ||
2771 | /* | |
2772 | * Perform extra freeing checks: | |
2773 | * - detect bad pointers. | |
2774 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2775 | */ |
2776 | static void kfree_debugcheck(const void *objp) | |
2777 | { | |
1da177e4 LT |
2778 | if (!virt_addr_valid(objp)) { |
2779 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2780 | (unsigned long)objp); |
2781 | BUG(); | |
1da177e4 | 2782 | } |
1da177e4 LT |
2783 | } |
2784 | ||
58ce1fd5 PE |
2785 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2786 | { | |
b46b8f19 | 2787 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2788 | |
2789 | redzone1 = *dbg_redzone1(cache, obj); | |
2790 | redzone2 = *dbg_redzone2(cache, obj); | |
2791 | ||
2792 | /* | |
2793 | * Redzone is ok. | |
2794 | */ | |
2795 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2796 | return; | |
2797 | ||
2798 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2799 | slab_error(cache, "double free detected"); | |
2800 | else | |
2801 | slab_error(cache, "memory outside object was overwritten"); | |
2802 | ||
b46b8f19 | 2803 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2804 | obj, redzone1, redzone2); |
2805 | } | |
2806 | ||
343e0d7a | 2807 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2808 | unsigned long caller) |
1da177e4 | 2809 | { |
1da177e4 LT |
2810 | unsigned int objnr; |
2811 | struct slab *slabp; | |
2812 | ||
80cbd911 MW |
2813 | BUG_ON(virt_to_cache(objp) != cachep); |
2814 | ||
3dafccf2 | 2815 | objp -= obj_offset(cachep); |
1da177e4 | 2816 | kfree_debugcheck(objp); |
56f295ef | 2817 | slabp = virt_to_slab(objp); |
1da177e4 LT |
2818 | |
2819 | if (cachep->flags & SLAB_RED_ZONE) { | |
58ce1fd5 | 2820 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2821 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2822 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2823 | } | |
2824 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2825 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2826 | |
8fea4e96 | 2827 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 LT |
2828 | |
2829 | BUG_ON(objnr >= cachep->num); | |
8fea4e96 | 2830 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
1da177e4 | 2831 | |
1da177e4 LT |
2832 | if (cachep->flags & SLAB_POISON) { |
2833 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2834 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2835 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2836 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2837 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2838 | } else { |
2839 | poison_obj(cachep, objp, POISON_FREE); | |
2840 | } | |
2841 | #else | |
2842 | poison_obj(cachep, objp, POISON_FREE); | |
2843 | #endif | |
2844 | } | |
2845 | return objp; | |
2846 | } | |
2847 | ||
1da177e4 LT |
2848 | #else |
2849 | #define kfree_debugcheck(x) do { } while(0) | |
2850 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2851 | #endif |
2852 | ||
072bb0aa MG |
2853 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2854 | bool force_refill) | |
1da177e4 LT |
2855 | { |
2856 | int batchcount; | |
ce8eb6c4 | 2857 | struct kmem_cache_node *n; |
1da177e4 | 2858 | struct array_cache *ac; |
1ca4cb24 PE |
2859 | int node; |
2860 | ||
1da177e4 | 2861 | check_irq_off(); |
7d6e6d09 | 2862 | node = numa_mem_id(); |
072bb0aa MG |
2863 | if (unlikely(force_refill)) |
2864 | goto force_grow; | |
2865 | retry: | |
9a2dba4b | 2866 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2867 | batchcount = ac->batchcount; |
2868 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2869 | /* |
2870 | * If there was little recent activity on this cache, then | |
2871 | * perform only a partial refill. Otherwise we could generate | |
2872 | * refill bouncing. | |
1da177e4 LT |
2873 | */ |
2874 | batchcount = BATCHREFILL_LIMIT; | |
2875 | } | |
ce8eb6c4 | 2876 | n = cachep->node[node]; |
e498be7d | 2877 | |
ce8eb6c4 CL |
2878 | BUG_ON(ac->avail > 0 || !n); |
2879 | spin_lock(&n->list_lock); | |
1da177e4 | 2880 | |
3ded175a | 2881 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2882 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2883 | n->shared->touched = 1; | |
3ded175a | 2884 | goto alloc_done; |
44b57f1c | 2885 | } |
3ded175a | 2886 | |
1da177e4 LT |
2887 | while (batchcount > 0) { |
2888 | struct list_head *entry; | |
2889 | struct slab *slabp; | |
2890 | /* Get slab alloc is to come from. */ | |
ce8eb6c4 CL |
2891 | entry = n->slabs_partial.next; |
2892 | if (entry == &n->slabs_partial) { | |
2893 | n->free_touched = 1; | |
2894 | entry = n->slabs_free.next; | |
2895 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2896 | goto must_grow; |
2897 | } | |
2898 | ||
2899 | slabp = list_entry(entry, struct slab, list); | |
1da177e4 | 2900 | check_spinlock_acquired(cachep); |
714b8171 PE |
2901 | |
2902 | /* | |
2903 | * The slab was either on partial or free list so | |
2904 | * there must be at least one object available for | |
2905 | * allocation. | |
2906 | */ | |
106a74e1 | 2907 | BUG_ON(slabp->active >= cachep->num); |
714b8171 | 2908 | |
106a74e1 | 2909 | while (slabp->active < cachep->num && batchcount--) { |
1da177e4 LT |
2910 | STATS_INC_ALLOCED(cachep); |
2911 | STATS_INC_ACTIVE(cachep); | |
2912 | STATS_SET_HIGH(cachep); | |
2913 | ||
072bb0aa MG |
2914 | ac_put_obj(cachep, ac, slab_get_obj(cachep, slabp, |
2915 | node)); | |
1da177e4 | 2916 | } |
1da177e4 LT |
2917 | |
2918 | /* move slabp to correct slabp list: */ | |
2919 | list_del(&slabp->list); | |
106a74e1 | 2920 | if (slabp->active == cachep->num) |
ce8eb6c4 | 2921 | list_add(&slabp->list, &n->slabs_full); |
1da177e4 | 2922 | else |
ce8eb6c4 | 2923 | list_add(&slabp->list, &n->slabs_partial); |
1da177e4 LT |
2924 | } |
2925 | ||
a737b3e2 | 2926 | must_grow: |
ce8eb6c4 | 2927 | n->free_objects -= ac->avail; |
a737b3e2 | 2928 | alloc_done: |
ce8eb6c4 | 2929 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2930 | |
2931 | if (unlikely(!ac->avail)) { | |
2932 | int x; | |
072bb0aa | 2933 | force_grow: |
3c517a61 | 2934 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 2935 | |
a737b3e2 | 2936 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2937 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2938 | node = numa_mem_id(); |
072bb0aa MG |
2939 | |
2940 | /* no objects in sight? abort */ | |
2941 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2942 | return NULL; |
2943 | ||
a737b3e2 | 2944 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2945 | goto retry; |
2946 | } | |
2947 | ac->touched = 1; | |
072bb0aa MG |
2948 | |
2949 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2950 | } |
2951 | ||
a737b3e2 AM |
2952 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2953 | gfp_t flags) | |
1da177e4 LT |
2954 | { |
2955 | might_sleep_if(flags & __GFP_WAIT); | |
2956 | #if DEBUG | |
2957 | kmem_flagcheck(cachep, flags); | |
2958 | #endif | |
2959 | } | |
2960 | ||
2961 | #if DEBUG | |
a737b3e2 | 2962 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 2963 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 2964 | { |
b28a02de | 2965 | if (!objp) |
1da177e4 | 2966 | return objp; |
b28a02de | 2967 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2968 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 2969 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 2970 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2971 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2972 | else |
2973 | check_poison_obj(cachep, objp); | |
2974 | #else | |
2975 | check_poison_obj(cachep, objp); | |
2976 | #endif | |
2977 | poison_obj(cachep, objp, POISON_INUSE); | |
2978 | } | |
2979 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2980 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
2981 | |
2982 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2983 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2984 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2985 | slab_error(cachep, "double free, or memory outside" | |
2986 | " object was overwritten"); | |
b28a02de | 2987 | printk(KERN_ERR |
b46b8f19 | 2988 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
2989 | objp, *dbg_redzone1(cachep, objp), |
2990 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2991 | } |
2992 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2993 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2994 | } | |
3dafccf2 | 2995 | objp += obj_offset(cachep); |
4f104934 | 2996 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 2997 | cachep->ctor(objp); |
7ea466f2 TH |
2998 | if (ARCH_SLAB_MINALIGN && |
2999 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 3000 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3001 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3002 | } |
1da177e4 LT |
3003 | return objp; |
3004 | } | |
3005 | #else | |
3006 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3007 | #endif | |
3008 | ||
773ff60e | 3009 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 3010 | { |
9b030cb8 | 3011 | if (cachep == kmem_cache) |
773ff60e | 3012 | return false; |
8a8b6502 | 3013 | |
8c138bc0 | 3014 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
3015 | } |
3016 | ||
343e0d7a | 3017 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3018 | { |
b28a02de | 3019 | void *objp; |
1da177e4 | 3020 | struct array_cache *ac; |
072bb0aa | 3021 | bool force_refill = false; |
1da177e4 | 3022 | |
5c382300 | 3023 | check_irq_off(); |
8a8b6502 | 3024 | |
9a2dba4b | 3025 | ac = cpu_cache_get(cachep); |
1da177e4 | 3026 | if (likely(ac->avail)) { |
1da177e4 | 3027 | ac->touched = 1; |
072bb0aa MG |
3028 | objp = ac_get_obj(cachep, ac, flags, false); |
3029 | ||
ddbf2e83 | 3030 | /* |
072bb0aa MG |
3031 | * Allow for the possibility all avail objects are not allowed |
3032 | * by the current flags | |
ddbf2e83 | 3033 | */ |
072bb0aa MG |
3034 | if (objp) { |
3035 | STATS_INC_ALLOCHIT(cachep); | |
3036 | goto out; | |
3037 | } | |
3038 | force_refill = true; | |
1da177e4 | 3039 | } |
072bb0aa MG |
3040 | |
3041 | STATS_INC_ALLOCMISS(cachep); | |
3042 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
3043 | /* | |
3044 | * the 'ac' may be updated by cache_alloc_refill(), | |
3045 | * and kmemleak_erase() requires its correct value. | |
3046 | */ | |
3047 | ac = cpu_cache_get(cachep); | |
3048 | ||
3049 | out: | |
d5cff635 CM |
3050 | /* |
3051 | * To avoid a false negative, if an object that is in one of the | |
3052 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3053 | * treat the array pointers as a reference to the object. | |
3054 | */ | |
f3d8b53a O |
3055 | if (objp) |
3056 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3057 | return objp; |
3058 | } | |
3059 | ||
e498be7d | 3060 | #ifdef CONFIG_NUMA |
c61afb18 | 3061 | /* |
b2455396 | 3062 | * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. |
c61afb18 PJ |
3063 | * |
3064 | * If we are in_interrupt, then process context, including cpusets and | |
3065 | * mempolicy, may not apply and should not be used for allocation policy. | |
3066 | */ | |
3067 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3068 | { | |
3069 | int nid_alloc, nid_here; | |
3070 | ||
765c4507 | 3071 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3072 | return NULL; |
7d6e6d09 | 3073 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3074 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3075 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3076 | else if (current->mempolicy) |
e7b691b0 | 3077 | nid_alloc = slab_node(); |
c61afb18 | 3078 | if (nid_alloc != nid_here) |
8b98c169 | 3079 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3080 | return NULL; |
3081 | } | |
3082 | ||
765c4507 CL |
3083 | /* |
3084 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3085 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3086 | * available node for available objects. If that fails then we |
3c517a61 CL |
3087 | * perform an allocation without specifying a node. This allows the page |
3088 | * allocator to do its reclaim / fallback magic. We then insert the | |
3089 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3090 | */ |
8c8cc2c1 | 3091 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3092 | { |
8c8cc2c1 PE |
3093 | struct zonelist *zonelist; |
3094 | gfp_t local_flags; | |
dd1a239f | 3095 | struct zoneref *z; |
54a6eb5c MG |
3096 | struct zone *zone; |
3097 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3098 | void *obj = NULL; |
3c517a61 | 3099 | int nid; |
cc9a6c87 | 3100 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3101 | |
3102 | if (flags & __GFP_THISNODE) | |
3103 | return NULL; | |
3104 | ||
6cb06229 | 3105 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3106 | |
cc9a6c87 MG |
3107 | retry_cpuset: |
3108 | cpuset_mems_cookie = get_mems_allowed(); | |
e7b691b0 | 3109 | zonelist = node_zonelist(slab_node(), flags); |
cc9a6c87 | 3110 | |
3c517a61 CL |
3111 | retry: |
3112 | /* | |
3113 | * Look through allowed nodes for objects available | |
3114 | * from existing per node queues. | |
3115 | */ | |
54a6eb5c MG |
3116 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3117 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3118 | |
54a6eb5c | 3119 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
6a67368c CL |
3120 | cache->node[nid] && |
3121 | cache->node[nid]->free_objects) { | |
3c517a61 CL |
3122 | obj = ____cache_alloc_node(cache, |
3123 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3124 | if (obj) |
3125 | break; | |
3126 | } | |
3c517a61 CL |
3127 | } |
3128 | ||
cfce6604 | 3129 | if (!obj) { |
3c517a61 CL |
3130 | /* |
3131 | * This allocation will be performed within the constraints | |
3132 | * of the current cpuset / memory policy requirements. | |
3133 | * We may trigger various forms of reclaim on the allowed | |
3134 | * set and go into memory reserves if necessary. | |
3135 | */ | |
0c3aa83e JK |
3136 | struct page *page; |
3137 | ||
dd47ea75 CL |
3138 | if (local_flags & __GFP_WAIT) |
3139 | local_irq_enable(); | |
3140 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3141 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3142 | if (local_flags & __GFP_WAIT) |
3143 | local_irq_disable(); | |
0c3aa83e | 3144 | if (page) { |
3c517a61 CL |
3145 | /* |
3146 | * Insert into the appropriate per node queues | |
3147 | */ | |
0c3aa83e JK |
3148 | nid = page_to_nid(page); |
3149 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3150 | obj = ____cache_alloc_node(cache, |
3151 | flags | GFP_THISNODE, nid); | |
3152 | if (!obj) | |
3153 | /* | |
3154 | * Another processor may allocate the | |
3155 | * objects in the slab since we are | |
3156 | * not holding any locks. | |
3157 | */ | |
3158 | goto retry; | |
3159 | } else { | |
b6a60451 | 3160 | /* cache_grow already freed obj */ |
3c517a61 CL |
3161 | obj = NULL; |
3162 | } | |
3163 | } | |
aedb0eb1 | 3164 | } |
cc9a6c87 MG |
3165 | |
3166 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !obj)) | |
3167 | goto retry_cpuset; | |
765c4507 CL |
3168 | return obj; |
3169 | } | |
3170 | ||
e498be7d CL |
3171 | /* |
3172 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3173 | */ |
8b98c169 | 3174 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3175 | int nodeid) |
e498be7d CL |
3176 | { |
3177 | struct list_head *entry; | |
b28a02de | 3178 | struct slab *slabp; |
ce8eb6c4 | 3179 | struct kmem_cache_node *n; |
b28a02de | 3180 | void *obj; |
b28a02de PE |
3181 | int x; |
3182 | ||
14e50c6a | 3183 | VM_BUG_ON(nodeid > num_online_nodes()); |
ce8eb6c4 CL |
3184 | n = cachep->node[nodeid]; |
3185 | BUG_ON(!n); | |
b28a02de | 3186 | |
a737b3e2 | 3187 | retry: |
ca3b9b91 | 3188 | check_irq_off(); |
ce8eb6c4 CL |
3189 | spin_lock(&n->list_lock); |
3190 | entry = n->slabs_partial.next; | |
3191 | if (entry == &n->slabs_partial) { | |
3192 | n->free_touched = 1; | |
3193 | entry = n->slabs_free.next; | |
3194 | if (entry == &n->slabs_free) | |
b28a02de PE |
3195 | goto must_grow; |
3196 | } | |
3197 | ||
3198 | slabp = list_entry(entry, struct slab, list); | |
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 | ||
106a74e1 | 3205 | BUG_ON(slabp->active == cachep->num); |
b28a02de | 3206 | |
78d382d7 | 3207 | obj = slab_get_obj(cachep, slabp, nodeid); |
ce8eb6c4 | 3208 | n->free_objects--; |
b28a02de PE |
3209 | /* move slabp to correct slabp list: */ |
3210 | list_del(&slabp->list); | |
3211 | ||
106a74e1 | 3212 | if (slabp->active == cachep->num) |
ce8eb6c4 | 3213 | list_add(&slabp->list, &n->slabs_full); |
a737b3e2 | 3214 | else |
ce8eb6c4 | 3215 | list_add(&slabp->list, &n->slabs_partial); |
e498be7d | 3216 | |
ce8eb6c4 | 3217 | spin_unlock(&n->list_lock); |
b28a02de | 3218 | goto done; |
e498be7d | 3219 | |
a737b3e2 | 3220 | must_grow: |
ce8eb6c4 | 3221 | spin_unlock(&n->list_lock); |
3c517a61 | 3222 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3223 | if (x) |
3224 | goto retry; | |
1da177e4 | 3225 | |
8c8cc2c1 | 3226 | return fallback_alloc(cachep, flags); |
e498be7d | 3227 | |
a737b3e2 | 3228 | done: |
b28a02de | 3229 | return obj; |
e498be7d | 3230 | } |
8c8cc2c1 | 3231 | |
8c8cc2c1 | 3232 | static __always_inline void * |
48356303 | 3233 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3234 | unsigned long caller) |
8c8cc2c1 PE |
3235 | { |
3236 | unsigned long save_flags; | |
3237 | void *ptr; | |
7d6e6d09 | 3238 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3239 | |
dcce284a | 3240 | flags &= gfp_allowed_mask; |
7e85ee0c | 3241 | |
cf40bd16 NP |
3242 | lockdep_trace_alloc(flags); |
3243 | ||
773ff60e | 3244 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3245 | return NULL; |
3246 | ||
d79923fa GC |
3247 | cachep = memcg_kmem_get_cache(cachep, flags); |
3248 | ||
8c8cc2c1 PE |
3249 | cache_alloc_debugcheck_before(cachep, flags); |
3250 | local_irq_save(save_flags); | |
3251 | ||
eacbbae3 | 3252 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3253 | nodeid = slab_node; |
8c8cc2c1 | 3254 | |
6a67368c | 3255 | if (unlikely(!cachep->node[nodeid])) { |
8c8cc2c1 PE |
3256 | /* Node not bootstrapped yet */ |
3257 | ptr = fallback_alloc(cachep, flags); | |
3258 | goto out; | |
3259 | } | |
3260 | ||
7d6e6d09 | 3261 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3262 | /* |
3263 | * Use the locally cached objects if possible. | |
3264 | * However ____cache_alloc does not allow fallback | |
3265 | * to other nodes. It may fail while we still have | |
3266 | * objects on other nodes available. | |
3267 | */ | |
3268 | ptr = ____cache_alloc(cachep, flags); | |
3269 | if (ptr) | |
3270 | goto out; | |
3271 | } | |
3272 | /* ___cache_alloc_node can fall back to other nodes */ | |
3273 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3274 | out: | |
3275 | local_irq_restore(save_flags); | |
3276 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3277 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3278 | flags); |
8c8cc2c1 | 3279 | |
c175eea4 | 3280 | if (likely(ptr)) |
8c138bc0 | 3281 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
c175eea4 | 3282 | |
d07dbea4 | 3283 | if (unlikely((flags & __GFP_ZERO) && ptr)) |
8c138bc0 | 3284 | memset(ptr, 0, cachep->object_size); |
d07dbea4 | 3285 | |
8c8cc2c1 PE |
3286 | return ptr; |
3287 | } | |
3288 | ||
3289 | static __always_inline void * | |
3290 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3291 | { | |
3292 | void *objp; | |
3293 | ||
3294 | if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { | |
3295 | objp = alternate_node_alloc(cache, flags); | |
3296 | if (objp) | |
3297 | goto out; | |
3298 | } | |
3299 | objp = ____cache_alloc(cache, flags); | |
3300 | ||
3301 | /* | |
3302 | * We may just have run out of memory on the local node. | |
3303 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3304 | */ | |
7d6e6d09 LS |
3305 | if (!objp) |
3306 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3307 | |
3308 | out: | |
3309 | return objp; | |
3310 | } | |
3311 | #else | |
3312 | ||
3313 | static __always_inline void * | |
3314 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3315 | { | |
3316 | return ____cache_alloc(cachep, flags); | |
3317 | } | |
3318 | ||
3319 | #endif /* CONFIG_NUMA */ | |
3320 | ||
3321 | static __always_inline void * | |
48356303 | 3322 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3323 | { |
3324 | unsigned long save_flags; | |
3325 | void *objp; | |
3326 | ||
dcce284a | 3327 | flags &= gfp_allowed_mask; |
7e85ee0c | 3328 | |
cf40bd16 NP |
3329 | lockdep_trace_alloc(flags); |
3330 | ||
773ff60e | 3331 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3332 | return NULL; |
3333 | ||
d79923fa GC |
3334 | cachep = memcg_kmem_get_cache(cachep, flags); |
3335 | ||
8c8cc2c1 PE |
3336 | cache_alloc_debugcheck_before(cachep, flags); |
3337 | local_irq_save(save_flags); | |
3338 | objp = __do_cache_alloc(cachep, flags); | |
3339 | local_irq_restore(save_flags); | |
3340 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3341 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3342 | flags); |
8c8cc2c1 PE |
3343 | prefetchw(objp); |
3344 | ||
c175eea4 | 3345 | if (likely(objp)) |
8c138bc0 | 3346 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
c175eea4 | 3347 | |
d07dbea4 | 3348 | if (unlikely((flags & __GFP_ZERO) && objp)) |
8c138bc0 | 3349 | memset(objp, 0, cachep->object_size); |
d07dbea4 | 3350 | |
8c8cc2c1 PE |
3351 | return objp; |
3352 | } | |
e498be7d CL |
3353 | |
3354 | /* | |
3355 | * Caller needs to acquire correct kmem_list's list_lock | |
3356 | */ | |
343e0d7a | 3357 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 3358 | int node) |
1da177e4 LT |
3359 | { |
3360 | int i; | |
ce8eb6c4 | 3361 | struct kmem_cache_node *n; |
1da177e4 LT |
3362 | |
3363 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3364 | void *objp; |
1da177e4 | 3365 | struct slab *slabp; |
1da177e4 | 3366 | |
072bb0aa MG |
3367 | clear_obj_pfmemalloc(&objpp[i]); |
3368 | objp = objpp[i]; | |
3369 | ||
6ed5eb22 | 3370 | slabp = virt_to_slab(objp); |
ce8eb6c4 | 3371 | n = cachep->node[node]; |
1da177e4 | 3372 | list_del(&slabp->list); |
ff69416e | 3373 | check_spinlock_acquired_node(cachep, node); |
78d382d7 | 3374 | slab_put_obj(cachep, slabp, objp, node); |
1da177e4 | 3375 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3376 | n->free_objects++; |
1da177e4 LT |
3377 | |
3378 | /* fixup slab chains */ | |
106a74e1 | 3379 | if (slabp->active == 0) { |
ce8eb6c4 CL |
3380 | if (n->free_objects > n->free_limit) { |
3381 | n->free_objects -= cachep->num; | |
e5ac9c5a RT |
3382 | /* No need to drop any previously held |
3383 | * lock here, even if we have a off-slab slab | |
3384 | * descriptor it is guaranteed to come from | |
3385 | * a different cache, refer to comments before | |
3386 | * alloc_slabmgmt. | |
3387 | */ | |
1da177e4 LT |
3388 | slab_destroy(cachep, slabp); |
3389 | } else { | |
ce8eb6c4 | 3390 | list_add(&slabp->list, &n->slabs_free); |
1da177e4 LT |
3391 | } |
3392 | } else { | |
3393 | /* Unconditionally move a slab to the end of the | |
3394 | * partial list on free - maximum time for the | |
3395 | * other objects to be freed, too. | |
3396 | */ | |
ce8eb6c4 | 3397 | list_add_tail(&slabp->list, &n->slabs_partial); |
1da177e4 LT |
3398 | } |
3399 | } | |
3400 | } | |
3401 | ||
343e0d7a | 3402 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3403 | { |
3404 | int batchcount; | |
ce8eb6c4 | 3405 | struct kmem_cache_node *n; |
7d6e6d09 | 3406 | int node = numa_mem_id(); |
1da177e4 LT |
3407 | |
3408 | batchcount = ac->batchcount; | |
3409 | #if DEBUG | |
3410 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3411 | #endif | |
3412 | check_irq_off(); | |
ce8eb6c4 CL |
3413 | n = cachep->node[node]; |
3414 | spin_lock(&n->list_lock); | |
3415 | if (n->shared) { | |
3416 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3417 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3418 | if (max) { |
3419 | if (batchcount > max) | |
3420 | batchcount = max; | |
e498be7d | 3421 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3422 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3423 | shared_array->avail += batchcount; |
3424 | goto free_done; | |
3425 | } | |
3426 | } | |
3427 | ||
ff69416e | 3428 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 3429 | free_done: |
1da177e4 LT |
3430 | #if STATS |
3431 | { | |
3432 | int i = 0; | |
3433 | struct list_head *p; | |
3434 | ||
ce8eb6c4 CL |
3435 | p = n->slabs_free.next; |
3436 | while (p != &(n->slabs_free)) { | |
1da177e4 LT |
3437 | struct slab *slabp; |
3438 | ||
3439 | slabp = list_entry(p, struct slab, list); | |
106a74e1 | 3440 | BUG_ON(slabp->active); |
1da177e4 LT |
3441 | |
3442 | i++; | |
3443 | p = p->next; | |
3444 | } | |
3445 | STATS_SET_FREEABLE(cachep, i); | |
3446 | } | |
3447 | #endif | |
ce8eb6c4 | 3448 | spin_unlock(&n->list_lock); |
1da177e4 | 3449 | ac->avail -= batchcount; |
a737b3e2 | 3450 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3451 | } |
3452 | ||
3453 | /* | |
a737b3e2 AM |
3454 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3455 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3456 | */ |
a947eb95 | 3457 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3458 | unsigned long caller) |
1da177e4 | 3459 | { |
9a2dba4b | 3460 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3461 | |
3462 | check_irq_off(); | |
d5cff635 | 3463 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3464 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3465 | |
8c138bc0 | 3466 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3467 | |
1807a1aa SS |
3468 | /* |
3469 | * Skip calling cache_free_alien() when the platform is not numa. | |
3470 | * This will avoid cache misses that happen while accessing slabp (which | |
3471 | * is per page memory reference) to get nodeid. Instead use a global | |
3472 | * variable to skip the call, which is mostly likely to be present in | |
3473 | * the cache. | |
3474 | */ | |
b6e68bc1 | 3475 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3476 | return; |
3477 | ||
1da177e4 LT |
3478 | if (likely(ac->avail < ac->limit)) { |
3479 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3480 | } else { |
3481 | STATS_INC_FREEMISS(cachep); | |
3482 | cache_flusharray(cachep, ac); | |
1da177e4 | 3483 | } |
42c8c99c | 3484 | |
072bb0aa | 3485 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3486 | } |
3487 | ||
3488 | /** | |
3489 | * kmem_cache_alloc - Allocate an object | |
3490 | * @cachep: The cache to allocate from. | |
3491 | * @flags: See kmalloc(). | |
3492 | * | |
3493 | * Allocate an object from this cache. The flags are only relevant | |
3494 | * if the cache has no available objects. | |
3495 | */ | |
343e0d7a | 3496 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3497 | { |
48356303 | 3498 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3499 | |
ca2b84cb | 3500 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3501 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3502 | |
3503 | return ret; | |
1da177e4 LT |
3504 | } |
3505 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3506 | ||
0f24f128 | 3507 | #ifdef CONFIG_TRACING |
85beb586 | 3508 | void * |
4052147c | 3509 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3510 | { |
85beb586 SR |
3511 | void *ret; |
3512 | ||
48356303 | 3513 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3514 | |
3515 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3516 | size, cachep->size, flags); |
85beb586 | 3517 | return ret; |
36555751 | 3518 | } |
85beb586 | 3519 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3520 | #endif |
3521 | ||
1da177e4 | 3522 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3523 | /** |
3524 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3525 | * @cachep: The cache to allocate from. | |
3526 | * @flags: See kmalloc(). | |
3527 | * @nodeid: node number of the target node. | |
3528 | * | |
3529 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3530 | * node, which can improve the performance for cpu bound structures. | |
3531 | * | |
3532 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3533 | */ | |
8b98c169 CH |
3534 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3535 | { | |
48356303 | 3536 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3537 | |
ca2b84cb | 3538 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3539 | cachep->object_size, cachep->size, |
ca2b84cb | 3540 | flags, nodeid); |
36555751 EGM |
3541 | |
3542 | return ret; | |
8b98c169 | 3543 | } |
1da177e4 LT |
3544 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3545 | ||
0f24f128 | 3546 | #ifdef CONFIG_TRACING |
4052147c | 3547 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3548 | gfp_t flags, |
4052147c EG |
3549 | int nodeid, |
3550 | size_t size) | |
36555751 | 3551 | { |
85beb586 SR |
3552 | void *ret; |
3553 | ||
592f4145 | 3554 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3555 | |
85beb586 | 3556 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3557 | size, cachep->size, |
85beb586 SR |
3558 | flags, nodeid); |
3559 | return ret; | |
36555751 | 3560 | } |
85beb586 | 3561 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3562 | #endif |
3563 | ||
8b98c169 | 3564 | static __always_inline void * |
7c0cb9c6 | 3565 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3566 | { |
343e0d7a | 3567 | struct kmem_cache *cachep; |
97e2bde4 | 3568 | |
2c59dd65 | 3569 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3570 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3571 | return cachep; | |
4052147c | 3572 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3573 | } |
8b98c169 | 3574 | |
0bb38a5c | 3575 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3576 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3577 | { | |
7c0cb9c6 | 3578 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3579 | } |
dbe5e69d | 3580 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3581 | |
3582 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3583 | int node, unsigned long caller) |
8b98c169 | 3584 | { |
7c0cb9c6 | 3585 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3586 | } |
3587 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3588 | #else | |
3589 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3590 | { | |
7c0cb9c6 | 3591 | return __do_kmalloc_node(size, flags, node, 0); |
8b98c169 CH |
3592 | } |
3593 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3594 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3595 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3596 | |
3597 | /** | |
800590f5 | 3598 | * __do_kmalloc - allocate memory |
1da177e4 | 3599 | * @size: how many bytes of memory are required. |
800590f5 | 3600 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3601 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3602 | */ |
7fd6b141 | 3603 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3604 | unsigned long caller) |
1da177e4 | 3605 | { |
343e0d7a | 3606 | struct kmem_cache *cachep; |
36555751 | 3607 | void *ret; |
1da177e4 | 3608 | |
97e2bde4 MS |
3609 | /* If you want to save a few bytes .text space: replace |
3610 | * __ with kmem_. | |
3611 | * Then kmalloc uses the uninlined functions instead of the inline | |
3612 | * functions. | |
3613 | */ | |
2c59dd65 | 3614 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3615 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3616 | return cachep; | |
48356303 | 3617 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3618 | |
7c0cb9c6 | 3619 | trace_kmalloc(caller, ret, |
3b0efdfa | 3620 | size, cachep->size, flags); |
36555751 EGM |
3621 | |
3622 | return ret; | |
7fd6b141 PE |
3623 | } |
3624 | ||
7fd6b141 | 3625 | |
0bb38a5c | 3626 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3627 | void *__kmalloc(size_t size, gfp_t flags) |
3628 | { | |
7c0cb9c6 | 3629 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3630 | } |
3631 | EXPORT_SYMBOL(__kmalloc); | |
3632 | ||
ce71e27c | 3633 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3634 | { |
7c0cb9c6 | 3635 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3636 | } |
3637 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3638 | |
3639 | #else | |
3640 | void *__kmalloc(size_t size, gfp_t flags) | |
3641 | { | |
7c0cb9c6 | 3642 | return __do_kmalloc(size, flags, 0); |
1d2c8eea CH |
3643 | } |
3644 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3645 | #endif |
3646 | ||
1da177e4 LT |
3647 | /** |
3648 | * kmem_cache_free - Deallocate an object | |
3649 | * @cachep: The cache the allocation was from. | |
3650 | * @objp: The previously allocated object. | |
3651 | * | |
3652 | * Free an object which was previously allocated from this | |
3653 | * cache. | |
3654 | */ | |
343e0d7a | 3655 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3656 | { |
3657 | unsigned long flags; | |
b9ce5ef4 GC |
3658 | cachep = cache_from_obj(cachep, objp); |
3659 | if (!cachep) | |
3660 | return; | |
1da177e4 LT |
3661 | |
3662 | local_irq_save(flags); | |
d97d476b | 3663 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3664 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3665 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3666 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3667 | local_irq_restore(flags); |
36555751 | 3668 | |
ca2b84cb | 3669 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3670 | } |
3671 | EXPORT_SYMBOL(kmem_cache_free); | |
3672 | ||
1da177e4 LT |
3673 | /** |
3674 | * kfree - free previously allocated memory | |
3675 | * @objp: pointer returned by kmalloc. | |
3676 | * | |
80e93eff PE |
3677 | * If @objp is NULL, no operation is performed. |
3678 | * | |
1da177e4 LT |
3679 | * Don't free memory not originally allocated by kmalloc() |
3680 | * or you will run into trouble. | |
3681 | */ | |
3682 | void kfree(const void *objp) | |
3683 | { | |
343e0d7a | 3684 | struct kmem_cache *c; |
1da177e4 LT |
3685 | unsigned long flags; |
3686 | ||
2121db74 PE |
3687 | trace_kfree(_RET_IP_, objp); |
3688 | ||
6cb8f913 | 3689 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3690 | return; |
3691 | local_irq_save(flags); | |
3692 | kfree_debugcheck(objp); | |
6ed5eb22 | 3693 | c = virt_to_cache(objp); |
8c138bc0 CL |
3694 | debug_check_no_locks_freed(objp, c->object_size); |
3695 | ||
3696 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3697 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3698 | local_irq_restore(flags); |
3699 | } | |
3700 | EXPORT_SYMBOL(kfree); | |
3701 | ||
e498be7d | 3702 | /* |
ce8eb6c4 | 3703 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3704 | */ |
83b519e8 | 3705 | static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3706 | { |
3707 | int node; | |
ce8eb6c4 | 3708 | struct kmem_cache_node *n; |
cafeb02e | 3709 | struct array_cache *new_shared; |
3395ee05 | 3710 | struct array_cache **new_alien = NULL; |
e498be7d | 3711 | |
9c09a95c | 3712 | for_each_online_node(node) { |
cafeb02e | 3713 | |
3395ee05 | 3714 | if (use_alien_caches) { |
83b519e8 | 3715 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3716 | if (!new_alien) |
3717 | goto fail; | |
3718 | } | |
cafeb02e | 3719 | |
63109846 ED |
3720 | new_shared = NULL; |
3721 | if (cachep->shared) { | |
3722 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3723 | cachep->shared*cachep->batchcount, |
83b519e8 | 3724 | 0xbaadf00d, gfp); |
63109846 ED |
3725 | if (!new_shared) { |
3726 | free_alien_cache(new_alien); | |
3727 | goto fail; | |
3728 | } | |
0718dc2a | 3729 | } |
cafeb02e | 3730 | |
ce8eb6c4 CL |
3731 | n = cachep->node[node]; |
3732 | if (n) { | |
3733 | struct array_cache *shared = n->shared; | |
cafeb02e | 3734 | |
ce8eb6c4 | 3735 | spin_lock_irq(&n->list_lock); |
e498be7d | 3736 | |
cafeb02e | 3737 | if (shared) |
0718dc2a CL |
3738 | free_block(cachep, shared->entry, |
3739 | shared->avail, node); | |
e498be7d | 3740 | |
ce8eb6c4 CL |
3741 | n->shared = new_shared; |
3742 | if (!n->alien) { | |
3743 | n->alien = new_alien; | |
e498be7d CL |
3744 | new_alien = NULL; |
3745 | } | |
ce8eb6c4 | 3746 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3747 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3748 | spin_unlock_irq(&n->list_lock); |
cafeb02e | 3749 | kfree(shared); |
e498be7d CL |
3750 | free_alien_cache(new_alien); |
3751 | continue; | |
3752 | } | |
ce8eb6c4 CL |
3753 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3754 | if (!n) { | |
0718dc2a CL |
3755 | free_alien_cache(new_alien); |
3756 | kfree(new_shared); | |
e498be7d | 3757 | goto fail; |
0718dc2a | 3758 | } |
e498be7d | 3759 | |
ce8eb6c4 CL |
3760 | kmem_cache_node_init(n); |
3761 | n->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
a737b3e2 | 3762 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
ce8eb6c4 CL |
3763 | n->shared = new_shared; |
3764 | n->alien = new_alien; | |
3765 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3766 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3767 | cachep->node[node] = n; |
e498be7d | 3768 | } |
cafeb02e | 3769 | return 0; |
0718dc2a | 3770 | |
a737b3e2 | 3771 | fail: |
3b0efdfa | 3772 | if (!cachep->list.next) { |
0718dc2a CL |
3773 | /* Cache is not active yet. Roll back what we did */ |
3774 | node--; | |
3775 | while (node >= 0) { | |
6a67368c | 3776 | if (cachep->node[node]) { |
ce8eb6c4 | 3777 | n = cachep->node[node]; |
0718dc2a | 3778 | |
ce8eb6c4 CL |
3779 | kfree(n->shared); |
3780 | free_alien_cache(n->alien); | |
3781 | kfree(n); | |
6a67368c | 3782 | cachep->node[node] = NULL; |
0718dc2a CL |
3783 | } |
3784 | node--; | |
3785 | } | |
3786 | } | |
cafeb02e | 3787 | return -ENOMEM; |
e498be7d CL |
3788 | } |
3789 | ||
1da177e4 | 3790 | struct ccupdate_struct { |
343e0d7a | 3791 | struct kmem_cache *cachep; |
acfe7d74 | 3792 | struct array_cache *new[0]; |
1da177e4 LT |
3793 | }; |
3794 | ||
3795 | static void do_ccupdate_local(void *info) | |
3796 | { | |
a737b3e2 | 3797 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3798 | struct array_cache *old; |
3799 | ||
3800 | check_irq_off(); | |
9a2dba4b | 3801 | old = cpu_cache_get(new->cachep); |
e498be7d | 3802 | |
1da177e4 LT |
3803 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3804 | new->new[smp_processor_id()] = old; | |
3805 | } | |
3806 | ||
18004c5d | 3807 | /* Always called with the slab_mutex held */ |
943a451a | 3808 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3809 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3810 | { |
d2e7b7d0 | 3811 | struct ccupdate_struct *new; |
2ed3a4ef | 3812 | int i; |
1da177e4 | 3813 | |
acfe7d74 ED |
3814 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3815 | gfp); | |
d2e7b7d0 SS |
3816 | if (!new) |
3817 | return -ENOMEM; | |
3818 | ||
e498be7d | 3819 | for_each_online_cpu(i) { |
7d6e6d09 | 3820 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3821 | batchcount, gfp); |
d2e7b7d0 | 3822 | if (!new->new[i]) { |
b28a02de | 3823 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3824 | kfree(new->new[i]); |
3825 | kfree(new); | |
e498be7d | 3826 | return -ENOMEM; |
1da177e4 LT |
3827 | } |
3828 | } | |
d2e7b7d0 | 3829 | new->cachep = cachep; |
1da177e4 | 3830 | |
15c8b6c1 | 3831 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3832 | |
1da177e4 | 3833 | check_irq_on(); |
1da177e4 LT |
3834 | cachep->batchcount = batchcount; |
3835 | cachep->limit = limit; | |
e498be7d | 3836 | cachep->shared = shared; |
1da177e4 | 3837 | |
e498be7d | 3838 | for_each_online_cpu(i) { |
d2e7b7d0 | 3839 | struct array_cache *ccold = new->new[i]; |
1da177e4 LT |
3840 | if (!ccold) |
3841 | continue; | |
6a67368c | 3842 | spin_lock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
7d6e6d09 | 3843 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_mem(i)); |
6a67368c | 3844 | spin_unlock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
1da177e4 LT |
3845 | kfree(ccold); |
3846 | } | |
d2e7b7d0 | 3847 | kfree(new); |
83b519e8 | 3848 | return alloc_kmemlist(cachep, gfp); |
1da177e4 LT |
3849 | } |
3850 | ||
943a451a GC |
3851 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3852 | int batchcount, int shared, gfp_t gfp) | |
3853 | { | |
3854 | int ret; | |
3855 | struct kmem_cache *c = NULL; | |
3856 | int i = 0; | |
3857 | ||
3858 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3859 | ||
3860 | if (slab_state < FULL) | |
3861 | return ret; | |
3862 | ||
3863 | if ((ret < 0) || !is_root_cache(cachep)) | |
3864 | return ret; | |
3865 | ||
ebe945c2 | 3866 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a GC |
3867 | for_each_memcg_cache_index(i) { |
3868 | c = cache_from_memcg(cachep, i); | |
3869 | if (c) | |
3870 | /* return value determined by the parent cache only */ | |
3871 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3872 | } | |
3873 | ||
3874 | return ret; | |
3875 | } | |
3876 | ||
18004c5d | 3877 | /* Called with slab_mutex held always */ |
83b519e8 | 3878 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3879 | { |
3880 | int err; | |
943a451a GC |
3881 | int limit = 0; |
3882 | int shared = 0; | |
3883 | int batchcount = 0; | |
3884 | ||
3885 | if (!is_root_cache(cachep)) { | |
3886 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3887 | limit = root->limit; | |
3888 | shared = root->shared; | |
3889 | batchcount = root->batchcount; | |
3890 | } | |
1da177e4 | 3891 | |
943a451a GC |
3892 | if (limit && shared && batchcount) |
3893 | goto skip_setup; | |
a737b3e2 AM |
3894 | /* |
3895 | * The head array serves three purposes: | |
1da177e4 LT |
3896 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3897 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3898 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3899 | * bufctl chains: array operations are cheaper. |
3900 | * The numbers are guessed, we should auto-tune as described by | |
3901 | * Bonwick. | |
3902 | */ | |
3b0efdfa | 3903 | if (cachep->size > 131072) |
1da177e4 | 3904 | limit = 1; |
3b0efdfa | 3905 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3906 | limit = 8; |
3b0efdfa | 3907 | else if (cachep->size > 1024) |
1da177e4 | 3908 | limit = 24; |
3b0efdfa | 3909 | else if (cachep->size > 256) |
1da177e4 LT |
3910 | limit = 54; |
3911 | else | |
3912 | limit = 120; | |
3913 | ||
a737b3e2 AM |
3914 | /* |
3915 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3916 | * allocation behaviour: Most allocs on one cpu, most free operations |
3917 | * on another cpu. For these cases, an efficient object passing between | |
3918 | * cpus is necessary. This is provided by a shared array. The array | |
3919 | * replaces Bonwick's magazine layer. | |
3920 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3921 | * to a larger limit. Thus disabled by default. | |
3922 | */ | |
3923 | shared = 0; | |
3b0efdfa | 3924 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3925 | shared = 8; |
1da177e4 LT |
3926 | |
3927 | #if DEBUG | |
a737b3e2 AM |
3928 | /* |
3929 | * With debugging enabled, large batchcount lead to excessively long | |
3930 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3931 | */ |
3932 | if (limit > 32) | |
3933 | limit = 32; | |
3934 | #endif | |
943a451a GC |
3935 | batchcount = (limit + 1) / 2; |
3936 | skip_setup: | |
3937 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3938 | if (err) |
3939 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3940 | cachep->name, -err); |
2ed3a4ef | 3941 | return err; |
1da177e4 LT |
3942 | } |
3943 | ||
1b55253a | 3944 | /* |
ce8eb6c4 CL |
3945 | * Drain an array if it contains any elements taking the node lock only if |
3946 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3947 | * if drain_array() is used on the shared array. |
1b55253a | 3948 | */ |
ce8eb6c4 | 3949 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3950 | struct array_cache *ac, int force, int node) |
1da177e4 LT |
3951 | { |
3952 | int tofree; | |
3953 | ||
1b55253a CL |
3954 | if (!ac || !ac->avail) |
3955 | return; | |
1da177e4 LT |
3956 | if (ac->touched && !force) { |
3957 | ac->touched = 0; | |
b18e7e65 | 3958 | } else { |
ce8eb6c4 | 3959 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
3960 | if (ac->avail) { |
3961 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3962 | if (tofree > ac->avail) | |
3963 | tofree = (ac->avail + 1) / 2; | |
3964 | free_block(cachep, ac->entry, tofree, node); | |
3965 | ac->avail -= tofree; | |
3966 | memmove(ac->entry, &(ac->entry[tofree]), | |
3967 | sizeof(void *) * ac->avail); | |
3968 | } | |
ce8eb6c4 | 3969 | spin_unlock_irq(&n->list_lock); |
1da177e4 LT |
3970 | } |
3971 | } | |
3972 | ||
3973 | /** | |
3974 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 3975 | * @w: work descriptor |
1da177e4 LT |
3976 | * |
3977 | * Called from workqueue/eventd every few seconds. | |
3978 | * Purpose: | |
3979 | * - clear the per-cpu caches for this CPU. | |
3980 | * - return freeable pages to the main free memory pool. | |
3981 | * | |
a737b3e2 AM |
3982 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3983 | * again on the next iteration. | |
1da177e4 | 3984 | */ |
7c5cae36 | 3985 | static void cache_reap(struct work_struct *w) |
1da177e4 | 3986 | { |
7a7c381d | 3987 | struct kmem_cache *searchp; |
ce8eb6c4 | 3988 | struct kmem_cache_node *n; |
7d6e6d09 | 3989 | int node = numa_mem_id(); |
bf6aede7 | 3990 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 3991 | |
18004c5d | 3992 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 3993 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 3994 | goto out; |
1da177e4 | 3995 | |
18004c5d | 3996 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
3997 | check_irq_on(); |
3998 | ||
35386e3b | 3999 | /* |
ce8eb6c4 | 4000 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4001 | * have established with reasonable certainty that |
4002 | * we can do some work if the lock was obtained. | |
4003 | */ | |
ce8eb6c4 | 4004 | n = searchp->node[node]; |
35386e3b | 4005 | |
ce8eb6c4 | 4006 | reap_alien(searchp, n); |
1da177e4 | 4007 | |
ce8eb6c4 | 4008 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4009 | |
35386e3b CL |
4010 | /* |
4011 | * These are racy checks but it does not matter | |
4012 | * if we skip one check or scan twice. | |
4013 | */ | |
ce8eb6c4 | 4014 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4015 | goto next; |
1da177e4 | 4016 | |
ce8eb6c4 | 4017 | n->next_reap = jiffies + REAPTIMEOUT_LIST3; |
1da177e4 | 4018 | |
ce8eb6c4 | 4019 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 4020 | |
ce8eb6c4 CL |
4021 | if (n->free_touched) |
4022 | n->free_touched = 0; | |
ed11d9eb CL |
4023 | else { |
4024 | int freed; | |
1da177e4 | 4025 | |
ce8eb6c4 | 4026 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4027 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4028 | STATS_ADD_REAPED(searchp, freed); | |
4029 | } | |
35386e3b | 4030 | next: |
1da177e4 LT |
4031 | cond_resched(); |
4032 | } | |
4033 | check_irq_on(); | |
18004c5d | 4034 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4035 | next_reap_node(); |
7c5cae36 | 4036 | out: |
a737b3e2 | 4037 | /* Set up the next iteration */ |
7c5cae36 | 4038 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); |
1da177e4 LT |
4039 | } |
4040 | ||
158a9624 | 4041 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4042 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4043 | { |
b28a02de PE |
4044 | struct slab *slabp; |
4045 | unsigned long active_objs; | |
4046 | unsigned long num_objs; | |
4047 | unsigned long active_slabs = 0; | |
4048 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4049 | const char *name; |
1da177e4 | 4050 | char *error = NULL; |
e498be7d | 4051 | int node; |
ce8eb6c4 | 4052 | struct kmem_cache_node *n; |
1da177e4 | 4053 | |
1da177e4 LT |
4054 | active_objs = 0; |
4055 | num_slabs = 0; | |
e498be7d | 4056 | for_each_online_node(node) { |
ce8eb6c4 CL |
4057 | n = cachep->node[node]; |
4058 | if (!n) | |
e498be7d CL |
4059 | continue; |
4060 | ||
ca3b9b91 | 4061 | check_irq_on(); |
ce8eb6c4 | 4062 | spin_lock_irq(&n->list_lock); |
e498be7d | 4063 | |
ce8eb6c4 | 4064 | list_for_each_entry(slabp, &n->slabs_full, list) { |
106a74e1 | 4065 | if (slabp->active != cachep->num && !error) |
e498be7d CL |
4066 | error = "slabs_full accounting error"; |
4067 | active_objs += cachep->num; | |
4068 | active_slabs++; | |
4069 | } | |
ce8eb6c4 | 4070 | list_for_each_entry(slabp, &n->slabs_partial, list) { |
106a74e1 JK |
4071 | if (slabp->active == cachep->num && !error) |
4072 | error = "slabs_partial accounting error"; | |
4073 | if (!slabp->active && !error) | |
4074 | error = "slabs_partial accounting error"; | |
4075 | active_objs += slabp->active; | |
e498be7d CL |
4076 | active_slabs++; |
4077 | } | |
ce8eb6c4 | 4078 | list_for_each_entry(slabp, &n->slabs_free, list) { |
106a74e1 JK |
4079 | if (slabp->active && !error) |
4080 | error = "slabs_free accounting error"; | |
e498be7d CL |
4081 | num_slabs++; |
4082 | } | |
ce8eb6c4 CL |
4083 | free_objects += n->free_objects; |
4084 | if (n->shared) | |
4085 | shared_avail += n->shared->avail; | |
e498be7d | 4086 | |
ce8eb6c4 | 4087 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4088 | } |
b28a02de PE |
4089 | num_slabs += active_slabs; |
4090 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4091 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4092 | error = "free_objects accounting error"; |
4093 | ||
b28a02de | 4094 | name = cachep->name; |
1da177e4 LT |
4095 | if (error) |
4096 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4097 | ||
0d7561c6 GC |
4098 | sinfo->active_objs = active_objs; |
4099 | sinfo->num_objs = num_objs; | |
4100 | sinfo->active_slabs = active_slabs; | |
4101 | sinfo->num_slabs = num_slabs; | |
4102 | sinfo->shared_avail = shared_avail; | |
4103 | sinfo->limit = cachep->limit; | |
4104 | sinfo->batchcount = cachep->batchcount; | |
4105 | sinfo->shared = cachep->shared; | |
4106 | sinfo->objects_per_slab = cachep->num; | |
4107 | sinfo->cache_order = cachep->gfporder; | |
4108 | } | |
4109 | ||
4110 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4111 | { | |
1da177e4 | 4112 | #if STATS |
ce8eb6c4 | 4113 | { /* node stats */ |
1da177e4 LT |
4114 | unsigned long high = cachep->high_mark; |
4115 | unsigned long allocs = cachep->num_allocations; | |
4116 | unsigned long grown = cachep->grown; | |
4117 | unsigned long reaped = cachep->reaped; | |
4118 | unsigned long errors = cachep->errors; | |
4119 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4120 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4121 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4122 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4123 | |
e92dd4fd JP |
4124 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4125 | "%4lu %4lu %4lu %4lu %4lu", | |
4126 | allocs, high, grown, | |
4127 | reaped, errors, max_freeable, node_allocs, | |
4128 | node_frees, overflows); | |
1da177e4 LT |
4129 | } |
4130 | /* cpu stats */ | |
4131 | { | |
4132 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4133 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4134 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4135 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4136 | ||
4137 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4138 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4139 | } |
4140 | #endif | |
1da177e4 LT |
4141 | } |
4142 | ||
1da177e4 LT |
4143 | #define MAX_SLABINFO_WRITE 128 |
4144 | /** | |
4145 | * slabinfo_write - Tuning for the slab allocator | |
4146 | * @file: unused | |
4147 | * @buffer: user buffer | |
4148 | * @count: data length | |
4149 | * @ppos: unused | |
4150 | */ | |
b7454ad3 | 4151 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4152 | size_t count, loff_t *ppos) |
1da177e4 | 4153 | { |
b28a02de | 4154 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4155 | int limit, batchcount, shared, res; |
7a7c381d | 4156 | struct kmem_cache *cachep; |
b28a02de | 4157 | |
1da177e4 LT |
4158 | if (count > MAX_SLABINFO_WRITE) |
4159 | return -EINVAL; | |
4160 | if (copy_from_user(&kbuf, buffer, count)) | |
4161 | return -EFAULT; | |
b28a02de | 4162 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4163 | |
4164 | tmp = strchr(kbuf, ' '); | |
4165 | if (!tmp) | |
4166 | return -EINVAL; | |
4167 | *tmp = '\0'; | |
4168 | tmp++; | |
4169 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4170 | return -EINVAL; | |
4171 | ||
4172 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4173 | mutex_lock(&slab_mutex); |
1da177e4 | 4174 | res = -EINVAL; |
18004c5d | 4175 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4176 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4177 | if (limit < 1 || batchcount < 1 || |
4178 | batchcount > limit || shared < 0) { | |
e498be7d | 4179 | res = 0; |
1da177e4 | 4180 | } else { |
e498be7d | 4181 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4182 | batchcount, shared, |
4183 | GFP_KERNEL); | |
1da177e4 LT |
4184 | } |
4185 | break; | |
4186 | } | |
4187 | } | |
18004c5d | 4188 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4189 | if (res >= 0) |
4190 | res = count; | |
4191 | return res; | |
4192 | } | |
871751e2 AV |
4193 | |
4194 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4195 | ||
4196 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4197 | { | |
18004c5d CL |
4198 | mutex_lock(&slab_mutex); |
4199 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4200 | } |
4201 | ||
4202 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4203 | { | |
4204 | unsigned long *p; | |
4205 | int l; | |
4206 | if (!v) | |
4207 | return 1; | |
4208 | l = n[1]; | |
4209 | p = n + 2; | |
4210 | while (l) { | |
4211 | int i = l/2; | |
4212 | unsigned long *q = p + 2 * i; | |
4213 | if (*q == v) { | |
4214 | q[1]++; | |
4215 | return 1; | |
4216 | } | |
4217 | if (*q > v) { | |
4218 | l = i; | |
4219 | } else { | |
4220 | p = q + 2; | |
4221 | l -= i + 1; | |
4222 | } | |
4223 | } | |
4224 | if (++n[1] == n[0]) | |
4225 | return 0; | |
4226 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4227 | p[0] = v; | |
4228 | p[1] = 1; | |
4229 | return 1; | |
4230 | } | |
4231 | ||
4232 | static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) | |
4233 | { | |
4234 | void *p; | |
b1cb0982 JK |
4235 | int i, j; |
4236 | ||
871751e2 AV |
4237 | if (n[0] == n[1]) |
4238 | return; | |
3b0efdfa | 4239 | for (i = 0, p = s->s_mem; i < c->num; i++, p += c->size) { |
b1cb0982 JK |
4240 | bool active = true; |
4241 | ||
106a74e1 | 4242 | for (j = s->active; j < c->num; j++) { |
b1cb0982 JK |
4243 | /* Skip freed item */ |
4244 | if (slab_bufctl(s)[j] == i) { | |
4245 | active = false; | |
4246 | break; | |
4247 | } | |
4248 | } | |
4249 | if (!active) | |
871751e2 | 4250 | continue; |
b1cb0982 | 4251 | |
871751e2 AV |
4252 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4253 | return; | |
4254 | } | |
4255 | } | |
4256 | ||
4257 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4258 | { | |
4259 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4260 | unsigned long offset, size; |
9281acea | 4261 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4262 | |
a5c43dae | 4263 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4264 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4265 | if (modname[0]) |
871751e2 AV |
4266 | seq_printf(m, " [%s]", modname); |
4267 | return; | |
4268 | } | |
4269 | #endif | |
4270 | seq_printf(m, "%p", (void *)address); | |
4271 | } | |
4272 | ||
4273 | static int leaks_show(struct seq_file *m, void *p) | |
4274 | { | |
0672aa7c | 4275 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
871751e2 | 4276 | struct slab *slabp; |
ce8eb6c4 | 4277 | struct kmem_cache_node *n; |
871751e2 | 4278 | const char *name; |
db845067 | 4279 | unsigned long *x = m->private; |
871751e2 AV |
4280 | int node; |
4281 | int i; | |
4282 | ||
4283 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4284 | return 0; | |
4285 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4286 | return 0; | |
4287 | ||
4288 | /* OK, we can do it */ | |
4289 | ||
db845067 | 4290 | x[1] = 0; |
871751e2 AV |
4291 | |
4292 | for_each_online_node(node) { | |
ce8eb6c4 CL |
4293 | n = cachep->node[node]; |
4294 | if (!n) | |
871751e2 AV |
4295 | continue; |
4296 | ||
4297 | check_irq_on(); | |
ce8eb6c4 | 4298 | spin_lock_irq(&n->list_lock); |
871751e2 | 4299 | |
ce8eb6c4 | 4300 | list_for_each_entry(slabp, &n->slabs_full, list) |
db845067 | 4301 | handle_slab(x, cachep, slabp); |
ce8eb6c4 | 4302 | list_for_each_entry(slabp, &n->slabs_partial, list) |
db845067 | 4303 | handle_slab(x, cachep, slabp); |
ce8eb6c4 | 4304 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4305 | } |
4306 | name = cachep->name; | |
db845067 | 4307 | if (x[0] == x[1]) { |
871751e2 | 4308 | /* Increase the buffer size */ |
18004c5d | 4309 | mutex_unlock(&slab_mutex); |
db845067 | 4310 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4311 | if (!m->private) { |
4312 | /* Too bad, we are really out */ | |
db845067 | 4313 | m->private = x; |
18004c5d | 4314 | mutex_lock(&slab_mutex); |
871751e2 AV |
4315 | return -ENOMEM; |
4316 | } | |
db845067 CL |
4317 | *(unsigned long *)m->private = x[0] * 2; |
4318 | kfree(x); | |
18004c5d | 4319 | mutex_lock(&slab_mutex); |
871751e2 AV |
4320 | /* Now make sure this entry will be retried */ |
4321 | m->count = m->size; | |
4322 | return 0; | |
4323 | } | |
db845067 CL |
4324 | for (i = 0; i < x[1]; i++) { |
4325 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4326 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4327 | seq_putc(m, '\n'); |
4328 | } | |
d2e7b7d0 | 4329 | |
871751e2 AV |
4330 | return 0; |
4331 | } | |
4332 | ||
a0ec95a8 | 4333 | static const struct seq_operations slabstats_op = { |
871751e2 | 4334 | .start = leaks_start, |
276a2439 WL |
4335 | .next = slab_next, |
4336 | .stop = slab_stop, | |
871751e2 AV |
4337 | .show = leaks_show, |
4338 | }; | |
a0ec95a8 AD |
4339 | |
4340 | static int slabstats_open(struct inode *inode, struct file *file) | |
4341 | { | |
4342 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4343 | int ret = -ENOMEM; | |
4344 | if (n) { | |
4345 | ret = seq_open(file, &slabstats_op); | |
4346 | if (!ret) { | |
4347 | struct seq_file *m = file->private_data; | |
4348 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4349 | m->private = n; | |
4350 | n = NULL; | |
4351 | } | |
4352 | kfree(n); | |
4353 | } | |
4354 | return ret; | |
4355 | } | |
4356 | ||
4357 | static const struct file_operations proc_slabstats_operations = { | |
4358 | .open = slabstats_open, | |
4359 | .read = seq_read, | |
4360 | .llseek = seq_lseek, | |
4361 | .release = seq_release_private, | |
4362 | }; | |
4363 | #endif | |
4364 | ||
4365 | static int __init slab_proc_init(void) | |
4366 | { | |
4367 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4368 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4369 | #endif |
a0ec95a8 AD |
4370 | return 0; |
4371 | } | |
4372 | module_init(slab_proc_init); | |
1da177e4 LT |
4373 | #endif |
4374 | ||
00e145b6 MS |
4375 | /** |
4376 | * ksize - get the actual amount of memory allocated for a given object | |
4377 | * @objp: Pointer to the object | |
4378 | * | |
4379 | * kmalloc may internally round up allocations and return more memory | |
4380 | * than requested. ksize() can be used to determine the actual amount of | |
4381 | * memory allocated. The caller may use this additional memory, even though | |
4382 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4383 | * The caller must guarantee that objp points to a valid object previously | |
4384 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4385 | * must not be freed during the duration of the call. | |
4386 | */ | |
fd76bab2 | 4387 | size_t ksize(const void *objp) |
1da177e4 | 4388 | { |
ef8b4520 CL |
4389 | BUG_ON(!objp); |
4390 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4391 | return 0; |
1da177e4 | 4392 | |
8c138bc0 | 4393 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4394 | } |
b1aabecd | 4395 | EXPORT_SYMBOL(ksize); |