Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
18004c5d | 71 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
72 | * The sem is only needed when accessing/extending the cache-chain, which |
73 | * can never happen inside an interrupt (kmem_cache_create(), | |
74 | * kmem_cache_shrink() and kmem_cache_reap()). | |
75 | * | |
76 | * At present, each engine can be growing a cache. This should be blocked. | |
77 | * | |
e498be7d CL |
78 | * 15 March 2005. NUMA slab allocator. |
79 | * Shai Fultheim <shai@scalex86.org>. | |
80 | * Shobhit Dayal <shobhit@calsoftinc.com> | |
81 | * Alok N Kataria <alokk@calsoftinc.com> | |
82 | * Christoph Lameter <christoph@lameter.com> | |
83 | * | |
84 | * Modified the slab allocator to be node aware on NUMA systems. | |
85 | * Each node has its own list of partial, free and full slabs. | |
86 | * All object allocations for a node occur from node specific slab lists. | |
1da177e4 LT |
87 | */ |
88 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
105 | #include <linux/rcupdate.h> | |
543537bd | 106 | #include <linux/string.h> |
138ae663 | 107 | #include <linux/uaccess.h> |
e498be7d | 108 | #include <linux/nodemask.h> |
d5cff635 | 109 | #include <linux/kmemleak.h> |
dc85da15 | 110 | #include <linux/mempolicy.h> |
fc0abb14 | 111 | #include <linux/mutex.h> |
8a8b6502 | 112 | #include <linux/fault-inject.h> |
e7eebaf6 | 113 | #include <linux/rtmutex.h> |
6a2d7a95 | 114 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 115 | #include <linux/debugobjects.h> |
c175eea4 | 116 | #include <linux/kmemcheck.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
1da177e4 | 119 | |
381760ea MG |
120 | #include <net/sock.h> |
121 | ||
1da177e4 LT |
122 | #include <asm/cacheflush.h> |
123 | #include <asm/tlbflush.h> | |
124 | #include <asm/page.h> | |
125 | ||
4dee6b64 SR |
126 | #include <trace/events/kmem.h> |
127 | ||
072bb0aa MG |
128 | #include "internal.h" |
129 | ||
b9ce5ef4 GC |
130 | #include "slab.h" |
131 | ||
1da177e4 | 132 | /* |
50953fe9 | 133 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
134 | * 0 for faster, smaller code (especially in the critical paths). |
135 | * | |
136 | * STATS - 1 to collect stats for /proc/slabinfo. | |
137 | * 0 for faster, smaller code (especially in the critical paths). | |
138 | * | |
139 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
140 | */ | |
141 | ||
142 | #ifdef CONFIG_DEBUG_SLAB | |
143 | #define DEBUG 1 | |
144 | #define STATS 1 | |
145 | #define FORCED_DEBUG 1 | |
146 | #else | |
147 | #define DEBUG 0 | |
148 | #define STATS 0 | |
149 | #define FORCED_DEBUG 0 | |
150 | #endif | |
151 | ||
1da177e4 LT |
152 | /* Shouldn't this be in a header file somewhere? */ |
153 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 154 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 155 | |
1da177e4 LT |
156 | #ifndef ARCH_KMALLOC_FLAGS |
157 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
158 | #endif | |
159 | ||
f315e3fa JK |
160 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
161 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
162 | ||
163 | #if FREELIST_BYTE_INDEX | |
164 | typedef unsigned char freelist_idx_t; | |
165 | #else | |
166 | typedef unsigned short freelist_idx_t; | |
167 | #endif | |
168 | ||
30321c7b | 169 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 170 | |
072bb0aa MG |
171 | /* |
172 | * true if a page was allocated from pfmemalloc reserves for network-based | |
173 | * swap | |
174 | */ | |
175 | static bool pfmemalloc_active __read_mostly; | |
176 | ||
1da177e4 LT |
177 | /* |
178 | * struct array_cache | |
179 | * | |
1da177e4 LT |
180 | * Purpose: |
181 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
182 | * - reduce the number of linked list operations | |
183 | * - reduce spinlock operations | |
184 | * | |
185 | * The limit is stored in the per-cpu structure to reduce the data cache | |
186 | * footprint. | |
187 | * | |
188 | */ | |
189 | struct array_cache { | |
190 | unsigned int avail; | |
191 | unsigned int limit; | |
192 | unsigned int batchcount; | |
193 | unsigned int touched; | |
e498be7d | 194 | spinlock_t lock; |
bda5b655 | 195 | void *entry[]; /* |
a737b3e2 AM |
196 | * Must have this definition in here for the proper |
197 | * alignment of array_cache. Also simplifies accessing | |
198 | * the entries. | |
072bb0aa MG |
199 | * |
200 | * Entries should not be directly dereferenced as | |
201 | * entries belonging to slabs marked pfmemalloc will | |
202 | * have the lower bits set SLAB_OBJ_PFMEMALLOC | |
a737b3e2 | 203 | */ |
1da177e4 LT |
204 | }; |
205 | ||
072bb0aa MG |
206 | #define SLAB_OBJ_PFMEMALLOC 1 |
207 | static inline bool is_obj_pfmemalloc(void *objp) | |
208 | { | |
209 | return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC; | |
210 | } | |
211 | ||
212 | static inline void set_obj_pfmemalloc(void **objp) | |
213 | { | |
214 | *objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC); | |
215 | return; | |
216 | } | |
217 | ||
218 | static inline void clear_obj_pfmemalloc(void **objp) | |
219 | { | |
220 | *objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC); | |
221 | } | |
222 | ||
a737b3e2 AM |
223 | /* |
224 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
225 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
226 | */ |
227 | #define BOOT_CPUCACHE_ENTRIES 1 | |
228 | struct arraycache_init { | |
229 | struct array_cache cache; | |
b28a02de | 230 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
231 | }; |
232 | ||
e498be7d CL |
233 | /* |
234 | * Need this for bootstrapping a per node allocator. | |
235 | */ | |
556a169d | 236 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
ce8eb6c4 | 237 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 238 | #define CACHE_CACHE 0 |
556a169d | 239 | #define SIZE_AC MAX_NUMNODES |
ce8eb6c4 | 240 | #define SIZE_NODE (2 * MAX_NUMNODES) |
e498be7d | 241 | |
ed11d9eb | 242 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 243 | struct kmem_cache_node *n, int tofree); |
ed11d9eb CL |
244 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
245 | int node); | |
83b519e8 | 246 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 247 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 248 | |
e0a42726 IM |
249 | static int slab_early_init = 1; |
250 | ||
e3366016 | 251 | #define INDEX_AC kmalloc_index(sizeof(struct arraycache_init)) |
ce8eb6c4 | 252 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 253 | |
ce8eb6c4 | 254 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
255 | { |
256 | INIT_LIST_HEAD(&parent->slabs_full); | |
257 | INIT_LIST_HEAD(&parent->slabs_partial); | |
258 | INIT_LIST_HEAD(&parent->slabs_free); | |
259 | parent->shared = NULL; | |
260 | parent->alien = NULL; | |
2e1217cf | 261 | parent->colour_next = 0; |
e498be7d CL |
262 | spin_lock_init(&parent->list_lock); |
263 | parent->free_objects = 0; | |
264 | parent->free_touched = 0; | |
265 | } | |
266 | ||
a737b3e2 AM |
267 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
268 | do { \ | |
269 | INIT_LIST_HEAD(listp); \ | |
6a67368c | 270 | list_splice(&(cachep->node[nodeid]->slab), listp); \ |
e498be7d CL |
271 | } while (0) |
272 | ||
a737b3e2 AM |
273 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
274 | do { \ | |
e498be7d CL |
275 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
276 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
277 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
278 | } while (0) | |
1da177e4 | 279 | |
1da177e4 LT |
280 | #define CFLGS_OFF_SLAB (0x80000000UL) |
281 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
282 | ||
283 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
284 | /* |
285 | * Optimization question: fewer reaps means less probability for unnessary | |
286 | * cpucache drain/refill cycles. | |
1da177e4 | 287 | * |
dc6f3f27 | 288 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
289 | * which could lock up otherwise freeable slabs. |
290 | */ | |
5f0985bb JZ |
291 | #define REAPTIMEOUT_AC (2*HZ) |
292 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
293 | |
294 | #if STATS | |
295 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
296 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
297 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
298 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 299 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
300 | #define STATS_SET_HIGH(x) \ |
301 | do { \ | |
302 | if ((x)->num_active > (x)->high_mark) \ | |
303 | (x)->high_mark = (x)->num_active; \ | |
304 | } while (0) | |
1da177e4 LT |
305 | #define STATS_INC_ERR(x) ((x)->errors++) |
306 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 307 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 308 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
309 | #define STATS_SET_FREEABLE(x, i) \ |
310 | do { \ | |
311 | if ((x)->max_freeable < i) \ | |
312 | (x)->max_freeable = i; \ | |
313 | } while (0) | |
1da177e4 LT |
314 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
315 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
316 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
317 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
318 | #else | |
319 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
320 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
321 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
322 | #define STATS_INC_GROWN(x) do { } while (0) | |
4e60c86b | 323 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
324 | #define STATS_SET_HIGH(x) do { } while (0) |
325 | #define STATS_INC_ERR(x) do { } while (0) | |
326 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 327 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 328 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 329 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
330 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
331 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
332 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
333 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
334 | #endif | |
335 | ||
336 | #if DEBUG | |
1da177e4 | 337 | |
a737b3e2 AM |
338 | /* |
339 | * memory layout of objects: | |
1da177e4 | 340 | * 0 : objp |
3dafccf2 | 341 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
342 | * the end of an object is aligned with the end of the real |
343 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 344 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 345 | * redzone word. |
3dafccf2 | 346 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
347 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
348 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 349 | * [BYTES_PER_WORD long] |
1da177e4 | 350 | */ |
343e0d7a | 351 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 352 | { |
3dafccf2 | 353 | return cachep->obj_offset; |
1da177e4 LT |
354 | } |
355 | ||
b46b8f19 | 356 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
357 | { |
358 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
359 | return (unsigned long long*) (objp + obj_offset(cachep) - |
360 | sizeof(unsigned long long)); | |
1da177e4 LT |
361 | } |
362 | ||
b46b8f19 | 363 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
364 | { |
365 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
366 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 367 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 368 | sizeof(unsigned long long) - |
87a927c7 | 369 | REDZONE_ALIGN); |
3b0efdfa | 370 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 371 | sizeof(unsigned long long)); |
1da177e4 LT |
372 | } |
373 | ||
343e0d7a | 374 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
375 | { |
376 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 377 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
378 | } |
379 | ||
380 | #else | |
381 | ||
3dafccf2 | 382 | #define obj_offset(x) 0 |
b46b8f19 DW |
383 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
384 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
385 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
386 | ||
387 | #endif | |
388 | ||
03787301 JK |
389 | #define OBJECT_FREE (0) |
390 | #define OBJECT_ACTIVE (1) | |
391 | ||
392 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
393 | ||
394 | static void set_obj_status(struct page *page, int idx, int val) | |
395 | { | |
396 | int freelist_size; | |
397 | char *status; | |
398 | struct kmem_cache *cachep = page->slab_cache; | |
399 | ||
400 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
401 | status = (char *)page->freelist + freelist_size; | |
402 | status[idx] = val; | |
403 | } | |
404 | ||
405 | static inline unsigned int get_obj_status(struct page *page, int idx) | |
406 | { | |
407 | int freelist_size; | |
408 | char *status; | |
409 | struct kmem_cache *cachep = page->slab_cache; | |
410 | ||
411 | freelist_size = cachep->num * sizeof(freelist_idx_t); | |
412 | status = (char *)page->freelist + freelist_size; | |
413 | ||
414 | return status[idx]; | |
415 | } | |
416 | ||
417 | #else | |
418 | static inline void set_obj_status(struct page *page, int idx, int val) {} | |
419 | ||
420 | #endif | |
421 | ||
1da177e4 | 422 | /* |
3df1cccd DR |
423 | * Do not go above this order unless 0 objects fit into the slab or |
424 | * overridden on the command line. | |
1da177e4 | 425 | */ |
543585cc DR |
426 | #define SLAB_MAX_ORDER_HI 1 |
427 | #define SLAB_MAX_ORDER_LO 0 | |
428 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 429 | static bool slab_max_order_set __initdata; |
1da177e4 | 430 | |
6ed5eb22 PE |
431 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
432 | { | |
b49af68f | 433 | struct page *page = virt_to_head_page(obj); |
35026088 | 434 | return page->slab_cache; |
6ed5eb22 PE |
435 | } |
436 | ||
8456a648 | 437 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
438 | unsigned int idx) |
439 | { | |
8456a648 | 440 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
441 | } |
442 | ||
6a2d7a95 | 443 | /* |
3b0efdfa CL |
444 | * We want to avoid an expensive divide : (offset / cache->size) |
445 | * Using the fact that size is a constant for a particular cache, | |
446 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
447 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
448 | */ | |
449 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 450 | const struct page *page, void *obj) |
8fea4e96 | 451 | { |
8456a648 | 452 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 453 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
454 | } |
455 | ||
1da177e4 | 456 | static struct arraycache_init initarray_generic = |
b28a02de | 457 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
458 | |
459 | /* internal cache of cache description objs */ | |
9b030cb8 | 460 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
461 | .batchcount = 1, |
462 | .limit = BOOT_CPUCACHE_ENTRIES, | |
463 | .shared = 1, | |
3b0efdfa | 464 | .size = sizeof(struct kmem_cache), |
b28a02de | 465 | .name = "kmem_cache", |
1da177e4 LT |
466 | }; |
467 | ||
056c6241 RT |
468 | #define BAD_ALIEN_MAGIC 0x01020304ul |
469 | ||
f1aaee53 AV |
470 | #ifdef CONFIG_LOCKDEP |
471 | ||
472 | /* | |
473 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
474 | * for other slabs "off slab". | |
475 | * The locking for this is tricky in that it nests within the locks | |
476 | * of all other slabs in a few places; to deal with this special | |
477 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
478 | * |
479 | * We set lock class for alien array caches which are up during init. | |
480 | * The lock annotation will be lost if all cpus of a node goes down and | |
481 | * then comes back up during hotplug | |
f1aaee53 | 482 | */ |
056c6241 RT |
483 | static struct lock_class_key on_slab_l3_key; |
484 | static struct lock_class_key on_slab_alc_key; | |
485 | ||
83835b3d PZ |
486 | static struct lock_class_key debugobj_l3_key; |
487 | static struct lock_class_key debugobj_alc_key; | |
488 | ||
489 | static void slab_set_lock_classes(struct kmem_cache *cachep, | |
490 | struct lock_class_key *l3_key, struct lock_class_key *alc_key, | |
491 | int q) | |
492 | { | |
493 | struct array_cache **alc; | |
ce8eb6c4 | 494 | struct kmem_cache_node *n; |
83835b3d PZ |
495 | int r; |
496 | ||
ce8eb6c4 CL |
497 | n = cachep->node[q]; |
498 | if (!n) | |
83835b3d PZ |
499 | return; |
500 | ||
ce8eb6c4 CL |
501 | lockdep_set_class(&n->list_lock, l3_key); |
502 | alc = n->alien; | |
83835b3d PZ |
503 | /* |
504 | * FIXME: This check for BAD_ALIEN_MAGIC | |
505 | * should go away when common slab code is taught to | |
506 | * work even without alien caches. | |
507 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
508 | * for alloc_alien_cache, | |
509 | */ | |
510 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
511 | return; | |
512 | for_each_node(r) { | |
513 | if (alc[r]) | |
514 | lockdep_set_class(&alc[r]->lock, alc_key); | |
515 | } | |
516 | } | |
517 | ||
518 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) | |
519 | { | |
520 | slab_set_lock_classes(cachep, &debugobj_l3_key, &debugobj_alc_key, node); | |
521 | } | |
522 | ||
523 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
524 | { | |
525 | int node; | |
526 | ||
527 | for_each_online_node(node) | |
528 | slab_set_debugobj_lock_classes_node(cachep, node); | |
529 | } | |
530 | ||
ce79ddc8 | 531 | static void init_node_lock_keys(int q) |
f1aaee53 | 532 | { |
e3366016 | 533 | int i; |
056c6241 | 534 | |
97d06609 | 535 | if (slab_state < UP) |
ce79ddc8 PE |
536 | return; |
537 | ||
0f8f8094 | 538 | for (i = 1; i <= KMALLOC_SHIFT_HIGH; i++) { |
ce8eb6c4 | 539 | struct kmem_cache_node *n; |
e3366016 CL |
540 | struct kmem_cache *cache = kmalloc_caches[i]; |
541 | ||
542 | if (!cache) | |
543 | continue; | |
ce79ddc8 | 544 | |
ce8eb6c4 CL |
545 | n = cache->node[q]; |
546 | if (!n || OFF_SLAB(cache)) | |
00afa758 | 547 | continue; |
83835b3d | 548 | |
e3366016 | 549 | slab_set_lock_classes(cache, &on_slab_l3_key, |
83835b3d | 550 | &on_slab_alc_key, q); |
f1aaee53 AV |
551 | } |
552 | } | |
ce79ddc8 | 553 | |
6ccfb5bc GC |
554 | static void on_slab_lock_classes_node(struct kmem_cache *cachep, int q) |
555 | { | |
6a67368c | 556 | if (!cachep->node[q]) |
6ccfb5bc GC |
557 | return; |
558 | ||
559 | slab_set_lock_classes(cachep, &on_slab_l3_key, | |
560 | &on_slab_alc_key, q); | |
561 | } | |
562 | ||
563 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) | |
564 | { | |
565 | int node; | |
566 | ||
567 | VM_BUG_ON(OFF_SLAB(cachep)); | |
568 | for_each_node(node) | |
569 | on_slab_lock_classes_node(cachep, node); | |
570 | } | |
571 | ||
ce79ddc8 PE |
572 | static inline void init_lock_keys(void) |
573 | { | |
574 | int node; | |
575 | ||
576 | for_each_node(node) | |
577 | init_node_lock_keys(node); | |
578 | } | |
f1aaee53 | 579 | #else |
ce79ddc8 PE |
580 | static void init_node_lock_keys(int q) |
581 | { | |
582 | } | |
583 | ||
056c6241 | 584 | static inline void init_lock_keys(void) |
f1aaee53 AV |
585 | { |
586 | } | |
83835b3d | 587 | |
6ccfb5bc GC |
588 | static inline void on_slab_lock_classes(struct kmem_cache *cachep) |
589 | { | |
590 | } | |
591 | ||
592 | static inline void on_slab_lock_classes_node(struct kmem_cache *cachep, int node) | |
593 | { | |
594 | } | |
595 | ||
83835b3d PZ |
596 | static void slab_set_debugobj_lock_classes_node(struct kmem_cache *cachep, int node) |
597 | { | |
598 | } | |
599 | ||
600 | static void slab_set_debugobj_lock_classes(struct kmem_cache *cachep) | |
601 | { | |
602 | } | |
f1aaee53 AV |
603 | #endif |
604 | ||
1871e52c | 605 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 606 | |
343e0d7a | 607 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
608 | { |
609 | return cachep->array[smp_processor_id()]; | |
610 | } | |
611 | ||
03787301 JK |
612 | static size_t calculate_freelist_size(int nr_objs, size_t align) |
613 | { | |
614 | size_t freelist_size; | |
615 | ||
616 | freelist_size = nr_objs * sizeof(freelist_idx_t); | |
617 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) | |
618 | freelist_size += nr_objs * sizeof(char); | |
619 | ||
620 | if (align) | |
621 | freelist_size = ALIGN(freelist_size, align); | |
622 | ||
623 | return freelist_size; | |
624 | } | |
625 | ||
9cef2e2b JK |
626 | static int calculate_nr_objs(size_t slab_size, size_t buffer_size, |
627 | size_t idx_size, size_t align) | |
1da177e4 | 628 | { |
9cef2e2b | 629 | int nr_objs; |
03787301 | 630 | size_t remained_size; |
9cef2e2b | 631 | size_t freelist_size; |
03787301 | 632 | int extra_space = 0; |
9cef2e2b | 633 | |
03787301 JK |
634 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
635 | extra_space = sizeof(char); | |
9cef2e2b JK |
636 | /* |
637 | * Ignore padding for the initial guess. The padding | |
638 | * is at most @align-1 bytes, and @buffer_size is at | |
639 | * least @align. In the worst case, this result will | |
640 | * be one greater than the number of objects that fit | |
641 | * into the memory allocation when taking the padding | |
642 | * into account. | |
643 | */ | |
03787301 | 644 | nr_objs = slab_size / (buffer_size + idx_size + extra_space); |
9cef2e2b JK |
645 | |
646 | /* | |
647 | * This calculated number will be either the right | |
648 | * amount, or one greater than what we want. | |
649 | */ | |
03787301 JK |
650 | remained_size = slab_size - nr_objs * buffer_size; |
651 | freelist_size = calculate_freelist_size(nr_objs, align); | |
652 | if (remained_size < freelist_size) | |
9cef2e2b JK |
653 | nr_objs--; |
654 | ||
655 | return nr_objs; | |
fbaccacf | 656 | } |
1da177e4 | 657 | |
a737b3e2 AM |
658 | /* |
659 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
660 | */ | |
fbaccacf SR |
661 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
662 | size_t align, int flags, size_t *left_over, | |
663 | unsigned int *num) | |
664 | { | |
665 | int nr_objs; | |
666 | size_t mgmt_size; | |
667 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 668 | |
fbaccacf SR |
669 | /* |
670 | * The slab management structure can be either off the slab or | |
671 | * on it. For the latter case, the memory allocated for a | |
672 | * slab is used for: | |
673 | * | |
16025177 | 674 | * - One unsigned int for each object |
fbaccacf SR |
675 | * - Padding to respect alignment of @align |
676 | * - @buffer_size bytes for each object | |
677 | * | |
678 | * If the slab management structure is off the slab, then the | |
679 | * alignment will already be calculated into the size. Because | |
680 | * the slabs are all pages aligned, the objects will be at the | |
681 | * correct alignment when allocated. | |
682 | */ | |
683 | if (flags & CFLGS_OFF_SLAB) { | |
684 | mgmt_size = 0; | |
685 | nr_objs = slab_size / buffer_size; | |
686 | ||
fbaccacf | 687 | } else { |
9cef2e2b | 688 | nr_objs = calculate_nr_objs(slab_size, buffer_size, |
a41adfaa | 689 | sizeof(freelist_idx_t), align); |
03787301 | 690 | mgmt_size = calculate_freelist_size(nr_objs, align); |
fbaccacf SR |
691 | } |
692 | *num = nr_objs; | |
693 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
694 | } |
695 | ||
f28510d3 | 696 | #if DEBUG |
d40cee24 | 697 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 698 | |
a737b3e2 AM |
699 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
700 | char *msg) | |
1da177e4 LT |
701 | { |
702 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 703 | function, cachep->name, msg); |
1da177e4 | 704 | dump_stack(); |
373d4d09 | 705 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 706 | } |
f28510d3 | 707 | #endif |
1da177e4 | 708 | |
3395ee05 PM |
709 | /* |
710 | * By default on NUMA we use alien caches to stage the freeing of | |
711 | * objects allocated from other nodes. This causes massive memory | |
712 | * inefficiencies when using fake NUMA setup to split memory into a | |
713 | * large number of small nodes, so it can be disabled on the command | |
714 | * line | |
715 | */ | |
716 | ||
717 | static int use_alien_caches __read_mostly = 1; | |
718 | static int __init noaliencache_setup(char *s) | |
719 | { | |
720 | use_alien_caches = 0; | |
721 | return 1; | |
722 | } | |
723 | __setup("noaliencache", noaliencache_setup); | |
724 | ||
3df1cccd DR |
725 | static int __init slab_max_order_setup(char *str) |
726 | { | |
727 | get_option(&str, &slab_max_order); | |
728 | slab_max_order = slab_max_order < 0 ? 0 : | |
729 | min(slab_max_order, MAX_ORDER - 1); | |
730 | slab_max_order_set = true; | |
731 | ||
732 | return 1; | |
733 | } | |
734 | __setup("slab_max_order=", slab_max_order_setup); | |
735 | ||
8fce4d8e CL |
736 | #ifdef CONFIG_NUMA |
737 | /* | |
738 | * Special reaping functions for NUMA systems called from cache_reap(). | |
739 | * These take care of doing round robin flushing of alien caches (containing | |
740 | * objects freed on different nodes from which they were allocated) and the | |
741 | * flushing of remote pcps by calling drain_node_pages. | |
742 | */ | |
1871e52c | 743 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
744 | |
745 | static void init_reap_node(int cpu) | |
746 | { | |
747 | int node; | |
748 | ||
7d6e6d09 | 749 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 750 | if (node == MAX_NUMNODES) |
442295c9 | 751 | node = first_node(node_online_map); |
8fce4d8e | 752 | |
1871e52c | 753 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
754 | } |
755 | ||
756 | static void next_reap_node(void) | |
757 | { | |
909ea964 | 758 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 759 | |
8fce4d8e CL |
760 | node = next_node(node, node_online_map); |
761 | if (unlikely(node >= MAX_NUMNODES)) | |
762 | node = first_node(node_online_map); | |
909ea964 | 763 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
764 | } |
765 | ||
766 | #else | |
767 | #define init_reap_node(cpu) do { } while (0) | |
768 | #define next_reap_node(void) do { } while (0) | |
769 | #endif | |
770 | ||
1da177e4 LT |
771 | /* |
772 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
773 | * via the workqueue/eventd. | |
774 | * Add the CPU number into the expiration time to minimize the possibility of | |
775 | * the CPUs getting into lockstep and contending for the global cache chain | |
776 | * lock. | |
777 | */ | |
0db0628d | 778 | static void start_cpu_timer(int cpu) |
1da177e4 | 779 | { |
1871e52c | 780 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
781 | |
782 | /* | |
783 | * When this gets called from do_initcalls via cpucache_init(), | |
784 | * init_workqueues() has already run, so keventd will be setup | |
785 | * at that time. | |
786 | */ | |
52bad64d | 787 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 788 | init_reap_node(cpu); |
203b42f7 | 789 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
790 | schedule_delayed_work_on(cpu, reap_work, |
791 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
792 | } |
793 | } | |
794 | ||
e498be7d | 795 | static struct array_cache *alloc_arraycache(int node, int entries, |
83b519e8 | 796 | int batchcount, gfp_t gfp) |
1da177e4 | 797 | { |
b28a02de | 798 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
799 | struct array_cache *nc = NULL; |
800 | ||
83b519e8 | 801 | nc = kmalloc_node(memsize, gfp, node); |
d5cff635 CM |
802 | /* |
803 | * The array_cache structures contain pointers to free object. | |
25985edc | 804 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
805 | * cache the pointers are not cleared and they could be counted as |
806 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
807 | * not scan such objects. | |
808 | */ | |
809 | kmemleak_no_scan(nc); | |
1da177e4 LT |
810 | if (nc) { |
811 | nc->avail = 0; | |
812 | nc->limit = entries; | |
813 | nc->batchcount = batchcount; | |
814 | nc->touched = 0; | |
e498be7d | 815 | spin_lock_init(&nc->lock); |
1da177e4 LT |
816 | } |
817 | return nc; | |
818 | } | |
819 | ||
8456a648 | 820 | static inline bool is_slab_pfmemalloc(struct page *page) |
072bb0aa | 821 | { |
072bb0aa MG |
822 | return PageSlabPfmemalloc(page); |
823 | } | |
824 | ||
825 | /* Clears pfmemalloc_active if no slabs have pfmalloc set */ | |
826 | static void recheck_pfmemalloc_active(struct kmem_cache *cachep, | |
827 | struct array_cache *ac) | |
828 | { | |
ce8eb6c4 | 829 | struct kmem_cache_node *n = cachep->node[numa_mem_id()]; |
8456a648 | 830 | struct page *page; |
072bb0aa MG |
831 | unsigned long flags; |
832 | ||
833 | if (!pfmemalloc_active) | |
834 | return; | |
835 | ||
ce8eb6c4 | 836 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 JK |
837 | list_for_each_entry(page, &n->slabs_full, lru) |
838 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
839 | goto out; |
840 | ||
8456a648 JK |
841 | list_for_each_entry(page, &n->slabs_partial, lru) |
842 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
843 | goto out; |
844 | ||
8456a648 JK |
845 | list_for_each_entry(page, &n->slabs_free, lru) |
846 | if (is_slab_pfmemalloc(page)) | |
072bb0aa MG |
847 | goto out; |
848 | ||
849 | pfmemalloc_active = false; | |
850 | out: | |
ce8eb6c4 | 851 | spin_unlock_irqrestore(&n->list_lock, flags); |
072bb0aa MG |
852 | } |
853 | ||
381760ea | 854 | static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac, |
072bb0aa MG |
855 | gfp_t flags, bool force_refill) |
856 | { | |
857 | int i; | |
858 | void *objp = ac->entry[--ac->avail]; | |
859 | ||
860 | /* Ensure the caller is allowed to use objects from PFMEMALLOC slab */ | |
861 | if (unlikely(is_obj_pfmemalloc(objp))) { | |
ce8eb6c4 | 862 | struct kmem_cache_node *n; |
072bb0aa MG |
863 | |
864 | if (gfp_pfmemalloc_allowed(flags)) { | |
865 | clear_obj_pfmemalloc(&objp); | |
866 | return objp; | |
867 | } | |
868 | ||
869 | /* The caller cannot use PFMEMALLOC objects, find another one */ | |
d014dc2e | 870 | for (i = 0; i < ac->avail; i++) { |
072bb0aa MG |
871 | /* If a !PFMEMALLOC object is found, swap them */ |
872 | if (!is_obj_pfmemalloc(ac->entry[i])) { | |
873 | objp = ac->entry[i]; | |
874 | ac->entry[i] = ac->entry[ac->avail]; | |
875 | ac->entry[ac->avail] = objp; | |
876 | return objp; | |
877 | } | |
878 | } | |
879 | ||
880 | /* | |
881 | * If there are empty slabs on the slabs_free list and we are | |
882 | * being forced to refill the cache, mark this one !pfmemalloc. | |
883 | */ | |
ce8eb6c4 CL |
884 | n = cachep->node[numa_mem_id()]; |
885 | if (!list_empty(&n->slabs_free) && force_refill) { | |
8456a648 | 886 | struct page *page = virt_to_head_page(objp); |
7ecccf9d | 887 | ClearPageSlabPfmemalloc(page); |
072bb0aa MG |
888 | clear_obj_pfmemalloc(&objp); |
889 | recheck_pfmemalloc_active(cachep, ac); | |
890 | return objp; | |
891 | } | |
892 | ||
893 | /* No !PFMEMALLOC objects available */ | |
894 | ac->avail++; | |
895 | objp = NULL; | |
896 | } | |
897 | ||
898 | return objp; | |
899 | } | |
900 | ||
381760ea MG |
901 | static inline void *ac_get_obj(struct kmem_cache *cachep, |
902 | struct array_cache *ac, gfp_t flags, bool force_refill) | |
903 | { | |
904 | void *objp; | |
905 | ||
906 | if (unlikely(sk_memalloc_socks())) | |
907 | objp = __ac_get_obj(cachep, ac, flags, force_refill); | |
908 | else | |
909 | objp = ac->entry[--ac->avail]; | |
910 | ||
911 | return objp; | |
912 | } | |
913 | ||
914 | static void *__ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
072bb0aa MG |
915 | void *objp) |
916 | { | |
917 | if (unlikely(pfmemalloc_active)) { | |
918 | /* Some pfmemalloc slabs exist, check if this is one */ | |
30c29bea | 919 | struct page *page = virt_to_head_page(objp); |
072bb0aa MG |
920 | if (PageSlabPfmemalloc(page)) |
921 | set_obj_pfmemalloc(&objp); | |
922 | } | |
923 | ||
381760ea MG |
924 | return objp; |
925 | } | |
926 | ||
927 | static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac, | |
928 | void *objp) | |
929 | { | |
930 | if (unlikely(sk_memalloc_socks())) | |
931 | objp = __ac_put_obj(cachep, ac, objp); | |
932 | ||
072bb0aa MG |
933 | ac->entry[ac->avail++] = objp; |
934 | } | |
935 | ||
3ded175a CL |
936 | /* |
937 | * Transfer objects in one arraycache to another. | |
938 | * Locking must be handled by the caller. | |
939 | * | |
940 | * Return the number of entries transferred. | |
941 | */ | |
942 | static int transfer_objects(struct array_cache *to, | |
943 | struct array_cache *from, unsigned int max) | |
944 | { | |
945 | /* Figure out how many entries to transfer */ | |
732eacc0 | 946 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
947 | |
948 | if (!nr) | |
949 | return 0; | |
950 | ||
951 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
952 | sizeof(void *) *nr); | |
953 | ||
954 | from->avail -= nr; | |
955 | to->avail += nr; | |
3ded175a CL |
956 | return nr; |
957 | } | |
958 | ||
765c4507 CL |
959 | #ifndef CONFIG_NUMA |
960 | ||
961 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 962 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 963 | |
83b519e8 | 964 | static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
765c4507 CL |
965 | { |
966 | return (struct array_cache **)BAD_ALIEN_MAGIC; | |
967 | } | |
968 | ||
969 | static inline void free_alien_cache(struct array_cache **ac_ptr) | |
970 | { | |
971 | } | |
972 | ||
973 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
974 | { | |
975 | return 0; | |
976 | } | |
977 | ||
978 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
979 | gfp_t flags) | |
980 | { | |
981 | return NULL; | |
982 | } | |
983 | ||
8b98c169 | 984 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
985 | gfp_t flags, int nodeid) |
986 | { | |
987 | return NULL; | |
988 | } | |
989 | ||
990 | #else /* CONFIG_NUMA */ | |
991 | ||
8b98c169 | 992 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 993 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 994 | |
83b519e8 | 995 | static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
e498be7d CL |
996 | { |
997 | struct array_cache **ac_ptr; | |
8ef82866 | 998 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
999 | int i; |
1000 | ||
1001 | if (limit > 1) | |
1002 | limit = 12; | |
f3186a9c | 1003 | ac_ptr = kzalloc_node(memsize, gfp, node); |
e498be7d CL |
1004 | if (ac_ptr) { |
1005 | for_each_node(i) { | |
f3186a9c | 1006 | if (i == node || !node_online(i)) |
e498be7d | 1007 | continue; |
83b519e8 | 1008 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); |
e498be7d | 1009 | if (!ac_ptr[i]) { |
cc550def | 1010 | for (i--; i >= 0; i--) |
e498be7d CL |
1011 | kfree(ac_ptr[i]); |
1012 | kfree(ac_ptr); | |
1013 | return NULL; | |
1014 | } | |
1015 | } | |
1016 | } | |
1017 | return ac_ptr; | |
1018 | } | |
1019 | ||
5295a74c | 1020 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
1021 | { |
1022 | int i; | |
1023 | ||
1024 | if (!ac_ptr) | |
1025 | return; | |
e498be7d | 1026 | for_each_node(i) |
b28a02de | 1027 | kfree(ac_ptr[i]); |
e498be7d CL |
1028 | kfree(ac_ptr); |
1029 | } | |
1030 | ||
343e0d7a | 1031 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 1032 | struct array_cache *ac, int node) |
e498be7d | 1033 | { |
ce8eb6c4 | 1034 | struct kmem_cache_node *n = cachep->node[node]; |
e498be7d CL |
1035 | |
1036 | if (ac->avail) { | |
ce8eb6c4 | 1037 | spin_lock(&n->list_lock); |
e00946fe CL |
1038 | /* |
1039 | * Stuff objects into the remote nodes shared array first. | |
1040 | * That way we could avoid the overhead of putting the objects | |
1041 | * into the free lists and getting them back later. | |
1042 | */ | |
ce8eb6c4 CL |
1043 | if (n->shared) |
1044 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 1045 | |
ff69416e | 1046 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d | 1047 | ac->avail = 0; |
ce8eb6c4 | 1048 | spin_unlock(&n->list_lock); |
e498be7d CL |
1049 | } |
1050 | } | |
1051 | ||
8fce4d8e CL |
1052 | /* |
1053 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1054 | */ | |
ce8eb6c4 | 1055 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 1056 | { |
909ea964 | 1057 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 1058 | |
ce8eb6c4 CL |
1059 | if (n->alien) { |
1060 | struct array_cache *ac = n->alien[node]; | |
e00946fe CL |
1061 | |
1062 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { | |
8fce4d8e CL |
1063 | __drain_alien_cache(cachep, ac, node); |
1064 | spin_unlock_irq(&ac->lock); | |
1065 | } | |
1066 | } | |
1067 | } | |
1068 | ||
a737b3e2 AM |
1069 | static void drain_alien_cache(struct kmem_cache *cachep, |
1070 | struct array_cache **alien) | |
e498be7d | 1071 | { |
b28a02de | 1072 | int i = 0; |
e498be7d CL |
1073 | struct array_cache *ac; |
1074 | unsigned long flags; | |
1075 | ||
1076 | for_each_online_node(i) { | |
4484ebf1 | 1077 | ac = alien[i]; |
e498be7d CL |
1078 | if (ac) { |
1079 | spin_lock_irqsave(&ac->lock, flags); | |
1080 | __drain_alien_cache(cachep, ac, i); | |
1081 | spin_unlock_irqrestore(&ac->lock, flags); | |
1082 | } | |
1083 | } | |
1084 | } | |
729bd0b7 | 1085 | |
873623df | 1086 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 | 1087 | { |
1ea991b0 | 1088 | int nodeid = page_to_nid(virt_to_page(objp)); |
ce8eb6c4 | 1089 | struct kmem_cache_node *n; |
729bd0b7 | 1090 | struct array_cache *alien = NULL; |
1ca4cb24 PE |
1091 | int node; |
1092 | ||
7d6e6d09 | 1093 | node = numa_mem_id(); |
729bd0b7 PE |
1094 | |
1095 | /* | |
1096 | * Make sure we are not freeing a object from another node to the array | |
1097 | * cache on this cpu. | |
1098 | */ | |
1ea991b0 | 1099 | if (likely(nodeid == node)) |
729bd0b7 PE |
1100 | return 0; |
1101 | ||
ce8eb6c4 | 1102 | n = cachep->node[node]; |
729bd0b7 | 1103 | STATS_INC_NODEFREES(cachep); |
ce8eb6c4 CL |
1104 | if (n->alien && n->alien[nodeid]) { |
1105 | alien = n->alien[nodeid]; | |
873623df | 1106 | spin_lock(&alien->lock); |
729bd0b7 PE |
1107 | if (unlikely(alien->avail == alien->limit)) { |
1108 | STATS_INC_ACOVERFLOW(cachep); | |
1109 | __drain_alien_cache(cachep, alien, nodeid); | |
1110 | } | |
072bb0aa | 1111 | ac_put_obj(cachep, alien, objp); |
729bd0b7 PE |
1112 | spin_unlock(&alien->lock); |
1113 | } else { | |
6a67368c | 1114 | spin_lock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 | 1115 | free_block(cachep, &objp, 1, nodeid); |
6a67368c | 1116 | spin_unlock(&(cachep->node[nodeid])->list_lock); |
729bd0b7 PE |
1117 | } |
1118 | return 1; | |
1119 | } | |
e498be7d CL |
1120 | #endif |
1121 | ||
8f9f8d9e | 1122 | /* |
6a67368c | 1123 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 1124 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 1125 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 1126 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
1127 | * already in use. |
1128 | * | |
18004c5d | 1129 | * Must hold slab_mutex. |
8f9f8d9e | 1130 | */ |
6a67368c | 1131 | static int init_cache_node_node(int node) |
8f9f8d9e DR |
1132 | { |
1133 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1134 | struct kmem_cache_node *n; |
6744f087 | 1135 | const int memsize = sizeof(struct kmem_cache_node); |
8f9f8d9e | 1136 | |
18004c5d | 1137 | list_for_each_entry(cachep, &slab_caches, list) { |
8f9f8d9e | 1138 | /* |
5f0985bb | 1139 | * Set up the kmem_cache_node for cpu before we can |
8f9f8d9e DR |
1140 | * begin anything. Make sure some other cpu on this |
1141 | * node has not already allocated this | |
1142 | */ | |
6a67368c | 1143 | if (!cachep->node[node]) { |
ce8eb6c4 CL |
1144 | n = kmalloc_node(memsize, GFP_KERNEL, node); |
1145 | if (!n) | |
8f9f8d9e | 1146 | return -ENOMEM; |
ce8eb6c4 | 1147 | kmem_cache_node_init(n); |
5f0985bb JZ |
1148 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
1149 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
8f9f8d9e DR |
1150 | |
1151 | /* | |
5f0985bb JZ |
1152 | * The kmem_cache_nodes don't come and go as CPUs |
1153 | * come and go. slab_mutex is sufficient | |
8f9f8d9e DR |
1154 | * protection here. |
1155 | */ | |
ce8eb6c4 | 1156 | cachep->node[node] = n; |
8f9f8d9e DR |
1157 | } |
1158 | ||
6a67368c CL |
1159 | spin_lock_irq(&cachep->node[node]->list_lock); |
1160 | cachep->node[node]->free_limit = | |
8f9f8d9e DR |
1161 | (1 + nr_cpus_node(node)) * |
1162 | cachep->batchcount + cachep->num; | |
6a67368c | 1163 | spin_unlock_irq(&cachep->node[node]->list_lock); |
8f9f8d9e DR |
1164 | } |
1165 | return 0; | |
1166 | } | |
1167 | ||
0fa8103b WL |
1168 | static inline int slabs_tofree(struct kmem_cache *cachep, |
1169 | struct kmem_cache_node *n) | |
1170 | { | |
1171 | return (n->free_objects + cachep->num - 1) / cachep->num; | |
1172 | } | |
1173 | ||
0db0628d | 1174 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
1175 | { |
1176 | struct kmem_cache *cachep; | |
ce8eb6c4 | 1177 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1178 | int node = cpu_to_mem(cpu); |
a70f7302 | 1179 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 1180 | |
18004c5d | 1181 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1182 | struct array_cache *nc; |
1183 | struct array_cache *shared; | |
1184 | struct array_cache **alien; | |
fbf1e473 | 1185 | |
fbf1e473 AM |
1186 | /* cpu is dead; no one can alloc from it. */ |
1187 | nc = cachep->array[cpu]; | |
1188 | cachep->array[cpu] = NULL; | |
ce8eb6c4 | 1189 | n = cachep->node[node]; |
fbf1e473 | 1190 | |
ce8eb6c4 | 1191 | if (!n) |
fbf1e473 AM |
1192 | goto free_array_cache; |
1193 | ||
ce8eb6c4 | 1194 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 1195 | |
ce8eb6c4 CL |
1196 | /* Free limit for this kmem_cache_node */ |
1197 | n->free_limit -= cachep->batchcount; | |
fbf1e473 AM |
1198 | if (nc) |
1199 | free_block(cachep, nc->entry, nc->avail, node); | |
1200 | ||
58463c1f | 1201 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1202 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1203 | goto free_array_cache; |
1204 | } | |
1205 | ||
ce8eb6c4 | 1206 | shared = n->shared; |
fbf1e473 AM |
1207 | if (shared) { |
1208 | free_block(cachep, shared->entry, | |
1209 | shared->avail, node); | |
ce8eb6c4 | 1210 | n->shared = NULL; |
fbf1e473 AM |
1211 | } |
1212 | ||
ce8eb6c4 CL |
1213 | alien = n->alien; |
1214 | n->alien = NULL; | |
fbf1e473 | 1215 | |
ce8eb6c4 | 1216 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1217 | |
1218 | kfree(shared); | |
1219 | if (alien) { | |
1220 | drain_alien_cache(cachep, alien); | |
1221 | free_alien_cache(alien); | |
1222 | } | |
1223 | free_array_cache: | |
1224 | kfree(nc); | |
1225 | } | |
1226 | /* | |
1227 | * In the previous loop, all the objects were freed to | |
1228 | * the respective cache's slabs, now we can go ahead and | |
1229 | * shrink each nodelist to its limit. | |
1230 | */ | |
18004c5d | 1231 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 CL |
1232 | n = cachep->node[node]; |
1233 | if (!n) | |
fbf1e473 | 1234 | continue; |
0fa8103b | 1235 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
fbf1e473 AM |
1236 | } |
1237 | } | |
1238 | ||
0db0628d | 1239 | static int cpuup_prepare(long cpu) |
1da177e4 | 1240 | { |
343e0d7a | 1241 | struct kmem_cache *cachep; |
ce8eb6c4 | 1242 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 1243 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1244 | int err; |
1da177e4 | 1245 | |
fbf1e473 AM |
1246 | /* |
1247 | * We need to do this right in the beginning since | |
1248 | * alloc_arraycache's are going to use this list. | |
1249 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1250 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1251 | */ |
6a67368c | 1252 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1253 | if (err < 0) |
1254 | goto bad; | |
fbf1e473 AM |
1255 | |
1256 | /* | |
1257 | * Now we can go ahead with allocating the shared arrays and | |
1258 | * array caches | |
1259 | */ | |
18004c5d | 1260 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
1261 | struct array_cache *nc; |
1262 | struct array_cache *shared = NULL; | |
1263 | struct array_cache **alien = NULL; | |
1264 | ||
1265 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1266 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1267 | if (!nc) |
1268 | goto bad; | |
1269 | if (cachep->shared) { | |
1270 | shared = alloc_arraycache(node, | |
1271 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1272 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1273 | if (!shared) { |
1274 | kfree(nc); | |
1da177e4 | 1275 | goto bad; |
12d00f6a | 1276 | } |
fbf1e473 AM |
1277 | } |
1278 | if (use_alien_caches) { | |
83b519e8 | 1279 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1280 | if (!alien) { |
1281 | kfree(shared); | |
1282 | kfree(nc); | |
fbf1e473 | 1283 | goto bad; |
12d00f6a | 1284 | } |
fbf1e473 AM |
1285 | } |
1286 | cachep->array[cpu] = nc; | |
ce8eb6c4 CL |
1287 | n = cachep->node[node]; |
1288 | BUG_ON(!n); | |
fbf1e473 | 1289 | |
ce8eb6c4 CL |
1290 | spin_lock_irq(&n->list_lock); |
1291 | if (!n->shared) { | |
fbf1e473 AM |
1292 | /* |
1293 | * We are serialised from CPU_DEAD or | |
1294 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1295 | */ | |
ce8eb6c4 | 1296 | n->shared = shared; |
fbf1e473 AM |
1297 | shared = NULL; |
1298 | } | |
4484ebf1 | 1299 | #ifdef CONFIG_NUMA |
ce8eb6c4 CL |
1300 | if (!n->alien) { |
1301 | n->alien = alien; | |
fbf1e473 | 1302 | alien = NULL; |
1da177e4 | 1303 | } |
fbf1e473 | 1304 | #endif |
ce8eb6c4 | 1305 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1306 | kfree(shared); |
1307 | free_alien_cache(alien); | |
83835b3d PZ |
1308 | if (cachep->flags & SLAB_DEBUG_OBJECTS) |
1309 | slab_set_debugobj_lock_classes_node(cachep, node); | |
6ccfb5bc GC |
1310 | else if (!OFF_SLAB(cachep) && |
1311 | !(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
1312 | on_slab_lock_classes_node(cachep, node); | |
fbf1e473 | 1313 | } |
ce79ddc8 PE |
1314 | init_node_lock_keys(node); |
1315 | ||
fbf1e473 AM |
1316 | return 0; |
1317 | bad: | |
12d00f6a | 1318 | cpuup_canceled(cpu); |
fbf1e473 AM |
1319 | return -ENOMEM; |
1320 | } | |
1321 | ||
0db0628d | 1322 | static int cpuup_callback(struct notifier_block *nfb, |
fbf1e473 AM |
1323 | unsigned long action, void *hcpu) |
1324 | { | |
1325 | long cpu = (long)hcpu; | |
1326 | int err = 0; | |
1327 | ||
1328 | switch (action) { | |
fbf1e473 AM |
1329 | case CPU_UP_PREPARE: |
1330 | case CPU_UP_PREPARE_FROZEN: | |
18004c5d | 1331 | mutex_lock(&slab_mutex); |
fbf1e473 | 1332 | err = cpuup_prepare(cpu); |
18004c5d | 1333 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
1334 | break; |
1335 | case CPU_ONLINE: | |
8bb78442 | 1336 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1337 | start_cpu_timer(cpu); |
1338 | break; | |
1339 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1340 | case CPU_DOWN_PREPARE: |
8bb78442 | 1341 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 | 1342 | /* |
18004c5d | 1343 | * Shutdown cache reaper. Note that the slab_mutex is |
5830c590 CL |
1344 | * held so that if cache_reap() is invoked it cannot do |
1345 | * anything expensive but will only modify reap_work | |
1346 | * and reschedule the timer. | |
1347 | */ | |
afe2c511 | 1348 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1349 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1350 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1351 | break; |
1352 | case CPU_DOWN_FAILED: | |
8bb78442 | 1353 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1354 | start_cpu_timer(cpu); |
1355 | break; | |
1da177e4 | 1356 | case CPU_DEAD: |
8bb78442 | 1357 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1358 | /* |
1359 | * Even if all the cpus of a node are down, we don't free the | |
ce8eb6c4 | 1360 | * kmem_cache_node of any cache. This to avoid a race between |
4484ebf1 | 1361 | * cpu_down, and a kmalloc allocation from another cpu for |
ce8eb6c4 | 1362 | * memory from the node of the cpu going down. The node |
4484ebf1 RT |
1363 | * structure is usually allocated from kmem_cache_create() and |
1364 | * gets destroyed at kmem_cache_destroy(). | |
1365 | */ | |
183ff22b | 1366 | /* fall through */ |
8f5be20b | 1367 | #endif |
1da177e4 | 1368 | case CPU_UP_CANCELED: |
8bb78442 | 1369 | case CPU_UP_CANCELED_FROZEN: |
18004c5d | 1370 | mutex_lock(&slab_mutex); |
fbf1e473 | 1371 | cpuup_canceled(cpu); |
18004c5d | 1372 | mutex_unlock(&slab_mutex); |
1da177e4 | 1373 | break; |
1da177e4 | 1374 | } |
eac40680 | 1375 | return notifier_from_errno(err); |
1da177e4 LT |
1376 | } |
1377 | ||
0db0628d | 1378 | static struct notifier_block cpucache_notifier = { |
74b85f37 CS |
1379 | &cpuup_callback, NULL, 0 |
1380 | }; | |
1da177e4 | 1381 | |
8f9f8d9e DR |
1382 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1383 | /* | |
1384 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1385 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1386 | * removed. | |
1387 | * | |
18004c5d | 1388 | * Must hold slab_mutex. |
8f9f8d9e | 1389 | */ |
6a67368c | 1390 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1391 | { |
1392 | struct kmem_cache *cachep; | |
1393 | int ret = 0; | |
1394 | ||
18004c5d | 1395 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1396 | struct kmem_cache_node *n; |
8f9f8d9e | 1397 | |
ce8eb6c4 CL |
1398 | n = cachep->node[node]; |
1399 | if (!n) | |
8f9f8d9e DR |
1400 | continue; |
1401 | ||
0fa8103b | 1402 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
8f9f8d9e | 1403 | |
ce8eb6c4 CL |
1404 | if (!list_empty(&n->slabs_full) || |
1405 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1406 | ret = -EBUSY; |
1407 | break; | |
1408 | } | |
1409 | } | |
1410 | return ret; | |
1411 | } | |
1412 | ||
1413 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1414 | unsigned long action, void *arg) | |
1415 | { | |
1416 | struct memory_notify *mnb = arg; | |
1417 | int ret = 0; | |
1418 | int nid; | |
1419 | ||
1420 | nid = mnb->status_change_nid; | |
1421 | if (nid < 0) | |
1422 | goto out; | |
1423 | ||
1424 | switch (action) { | |
1425 | case MEM_GOING_ONLINE: | |
18004c5d | 1426 | mutex_lock(&slab_mutex); |
6a67368c | 1427 | ret = init_cache_node_node(nid); |
18004c5d | 1428 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1429 | break; |
1430 | case MEM_GOING_OFFLINE: | |
18004c5d | 1431 | mutex_lock(&slab_mutex); |
6a67368c | 1432 | ret = drain_cache_node_node(nid); |
18004c5d | 1433 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1434 | break; |
1435 | case MEM_ONLINE: | |
1436 | case MEM_OFFLINE: | |
1437 | case MEM_CANCEL_ONLINE: | |
1438 | case MEM_CANCEL_OFFLINE: | |
1439 | break; | |
1440 | } | |
1441 | out: | |
5fda1bd5 | 1442 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1443 | } |
1444 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1445 | ||
e498be7d | 1446 | /* |
ce8eb6c4 | 1447 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1448 | */ |
6744f087 | 1449 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1450 | int nodeid) |
e498be7d | 1451 | { |
6744f087 | 1452 | struct kmem_cache_node *ptr; |
e498be7d | 1453 | |
6744f087 | 1454 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1455 | BUG_ON(!ptr); |
1456 | ||
6744f087 | 1457 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1458 | /* |
1459 | * Do not assume that spinlocks can be initialized via memcpy: | |
1460 | */ | |
1461 | spin_lock_init(&ptr->list_lock); | |
1462 | ||
e498be7d | 1463 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1464 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1465 | } |
1466 | ||
556a169d | 1467 | /* |
ce8eb6c4 CL |
1468 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1469 | * size of kmem_cache_node. | |
556a169d | 1470 | */ |
ce8eb6c4 | 1471 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1472 | { |
1473 | int node; | |
1474 | ||
1475 | for_each_online_node(node) { | |
ce8eb6c4 | 1476 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1477 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1478 | REAPTIMEOUT_NODE + |
1479 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1480 | } |
1481 | } | |
1482 | ||
3c583465 CL |
1483 | /* |
1484 | * The memory after the last cpu cache pointer is used for the | |
6a67368c | 1485 | * the node pointer. |
3c583465 | 1486 | */ |
6a67368c | 1487 | static void setup_node_pointer(struct kmem_cache *cachep) |
3c583465 | 1488 | { |
6a67368c | 1489 | cachep->node = (struct kmem_cache_node **)&cachep->array[nr_cpu_ids]; |
3c583465 CL |
1490 | } |
1491 | ||
a737b3e2 AM |
1492 | /* |
1493 | * Initialisation. Called after the page allocator have been initialised and | |
1494 | * before smp_init(). | |
1da177e4 LT |
1495 | */ |
1496 | void __init kmem_cache_init(void) | |
1497 | { | |
e498be7d CL |
1498 | int i; |
1499 | ||
68126702 JK |
1500 | BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) < |
1501 | sizeof(struct rcu_head)); | |
9b030cb8 | 1502 | kmem_cache = &kmem_cache_boot; |
6a67368c | 1503 | setup_node_pointer(kmem_cache); |
9b030cb8 | 1504 | |
b6e68bc1 | 1505 | if (num_possible_nodes() == 1) |
62918a03 SS |
1506 | use_alien_caches = 0; |
1507 | ||
3c583465 | 1508 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1509 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1510 | |
ce8eb6c4 | 1511 | set_up_node(kmem_cache, CACHE_CACHE); |
1da177e4 LT |
1512 | |
1513 | /* | |
1514 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1515 | * page orders on machines with more than 32MB of memory if |
1516 | * not overridden on the command line. | |
1da177e4 | 1517 | */ |
3df1cccd | 1518 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1519 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1520 | |
1da177e4 LT |
1521 | /* Bootstrap is tricky, because several objects are allocated |
1522 | * from caches that do not exist yet: | |
9b030cb8 CL |
1523 | * 1) initialize the kmem_cache cache: it contains the struct |
1524 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1525 | * kmem_cache is statically allocated. | |
e498be7d | 1526 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1527 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1528 | * array at the end of the bootstrap. |
1da177e4 | 1529 | * 2) Create the first kmalloc cache. |
343e0d7a | 1530 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1531 | * An __init data area is used for the head array. |
1532 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1533 | * head arrays. | |
9b030cb8 | 1534 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1535 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1536 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1537 | * the other cache's with kmalloc allocated memory. |
1538 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1539 | */ |
1540 | ||
9b030cb8 | 1541 | /* 1) create the kmem_cache */ |
1da177e4 | 1542 | |
8da3430d | 1543 | /* |
b56efcf0 | 1544 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1545 | */ |
2f9baa9f CL |
1546 | create_boot_cache(kmem_cache, "kmem_cache", |
1547 | offsetof(struct kmem_cache, array[nr_cpu_ids]) + | |
6744f087 | 1548 | nr_node_ids * sizeof(struct kmem_cache_node *), |
2f9baa9f CL |
1549 | SLAB_HWCACHE_ALIGN); |
1550 | list_add(&kmem_cache->list, &slab_caches); | |
1da177e4 LT |
1551 | |
1552 | /* 2+3) create the kmalloc caches */ | |
1da177e4 | 1553 | |
a737b3e2 AM |
1554 | /* |
1555 | * Initialize the caches that provide memory for the array cache and the | |
ce8eb6c4 | 1556 | * kmem_cache_node structures first. Without this, further allocations will |
a737b3e2 | 1557 | * bug. |
e498be7d CL |
1558 | */ |
1559 | ||
e3366016 CL |
1560 | kmalloc_caches[INDEX_AC] = create_kmalloc_cache("kmalloc-ac", |
1561 | kmalloc_size(INDEX_AC), ARCH_KMALLOC_FLAGS); | |
45530c44 | 1562 | |
ce8eb6c4 CL |
1563 | if (INDEX_AC != INDEX_NODE) |
1564 | kmalloc_caches[INDEX_NODE] = | |
1565 | create_kmalloc_cache("kmalloc-node", | |
1566 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS); | |
e498be7d | 1567 | |
e0a42726 IM |
1568 | slab_early_init = 0; |
1569 | ||
1da177e4 LT |
1570 | /* 4) Replace the bootstrap head arrays */ |
1571 | { | |
2b2d5493 | 1572 | struct array_cache *ptr; |
e498be7d | 1573 | |
83b519e8 | 1574 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1575 | |
9b030cb8 | 1576 | memcpy(ptr, cpu_cache_get(kmem_cache), |
b28a02de | 1577 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1578 | /* |
1579 | * Do not assume that spinlocks can be initialized via memcpy: | |
1580 | */ | |
1581 | spin_lock_init(&ptr->lock); | |
1582 | ||
9b030cb8 | 1583 | kmem_cache->array[smp_processor_id()] = ptr; |
e498be7d | 1584 | |
83b519e8 | 1585 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1586 | |
e3366016 | 1587 | BUG_ON(cpu_cache_get(kmalloc_caches[INDEX_AC]) |
b28a02de | 1588 | != &initarray_generic.cache); |
e3366016 | 1589 | memcpy(ptr, cpu_cache_get(kmalloc_caches[INDEX_AC]), |
b28a02de | 1590 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1591 | /* |
1592 | * Do not assume that spinlocks can be initialized via memcpy: | |
1593 | */ | |
1594 | spin_lock_init(&ptr->lock); | |
1595 | ||
e3366016 | 1596 | kmalloc_caches[INDEX_AC]->array[smp_processor_id()] = ptr; |
1da177e4 | 1597 | } |
ce8eb6c4 | 1598 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1599 | { |
1ca4cb24 PE |
1600 | int nid; |
1601 | ||
9c09a95c | 1602 | for_each_online_node(nid) { |
ce8eb6c4 | 1603 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1604 | |
e3366016 | 1605 | init_list(kmalloc_caches[INDEX_AC], |
ce8eb6c4 | 1606 | &init_kmem_cache_node[SIZE_AC + nid], nid); |
e498be7d | 1607 | |
ce8eb6c4 CL |
1608 | if (INDEX_AC != INDEX_NODE) { |
1609 | init_list(kmalloc_caches[INDEX_NODE], | |
1610 | &init_kmem_cache_node[SIZE_NODE + nid], nid); | |
e498be7d CL |
1611 | } |
1612 | } | |
1613 | } | |
1da177e4 | 1614 | |
f97d5f63 | 1615 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1616 | } |
1617 | ||
1618 | void __init kmem_cache_init_late(void) | |
1619 | { | |
1620 | struct kmem_cache *cachep; | |
1621 | ||
97d06609 | 1622 | slab_state = UP; |
52cef189 | 1623 | |
8429db5c | 1624 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1625 | mutex_lock(&slab_mutex); |
1626 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1627 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1628 | BUG(); | |
18004c5d | 1629 | mutex_unlock(&slab_mutex); |
056c6241 | 1630 | |
947ca185 MW |
1631 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1632 | init_lock_keys(); | |
1633 | ||
97d06609 CL |
1634 | /* Done! */ |
1635 | slab_state = FULL; | |
1636 | ||
a737b3e2 AM |
1637 | /* |
1638 | * Register a cpu startup notifier callback that initializes | |
1639 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1640 | */ |
1641 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1642 | |
8f9f8d9e DR |
1643 | #ifdef CONFIG_NUMA |
1644 | /* | |
1645 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1646 | * node. |
8f9f8d9e DR |
1647 | */ |
1648 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1649 | #endif | |
1650 | ||
a737b3e2 AM |
1651 | /* |
1652 | * The reap timers are started later, with a module init call: That part | |
1653 | * of the kernel is not yet operational. | |
1da177e4 LT |
1654 | */ |
1655 | } | |
1656 | ||
1657 | static int __init cpucache_init(void) | |
1658 | { | |
1659 | int cpu; | |
1660 | ||
a737b3e2 AM |
1661 | /* |
1662 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1663 | */ |
e498be7d | 1664 | for_each_online_cpu(cpu) |
a737b3e2 | 1665 | start_cpu_timer(cpu); |
a164f896 GC |
1666 | |
1667 | /* Done! */ | |
97d06609 | 1668 | slab_state = FULL; |
1da177e4 LT |
1669 | return 0; |
1670 | } | |
1da177e4 LT |
1671 | __initcall(cpucache_init); |
1672 | ||
8bdec192 RA |
1673 | static noinline void |
1674 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1675 | { | |
9a02d699 | 1676 | #if DEBUG |
ce8eb6c4 | 1677 | struct kmem_cache_node *n; |
8456a648 | 1678 | struct page *page; |
8bdec192 RA |
1679 | unsigned long flags; |
1680 | int node; | |
9a02d699 DR |
1681 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1682 | DEFAULT_RATELIMIT_BURST); | |
1683 | ||
1684 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1685 | return; | |
8bdec192 RA |
1686 | |
1687 | printk(KERN_WARNING | |
1688 | "SLAB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
1689 | nodeid, gfpflags); | |
1690 | printk(KERN_WARNING " cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1691 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 RA |
1692 | |
1693 | for_each_online_node(node) { | |
1694 | unsigned long active_objs = 0, num_objs = 0, free_objects = 0; | |
1695 | unsigned long active_slabs = 0, num_slabs = 0; | |
1696 | ||
ce8eb6c4 CL |
1697 | n = cachep->node[node]; |
1698 | if (!n) | |
8bdec192 RA |
1699 | continue; |
1700 | ||
ce8eb6c4 | 1701 | spin_lock_irqsave(&n->list_lock, flags); |
8456a648 | 1702 | list_for_each_entry(page, &n->slabs_full, lru) { |
8bdec192 RA |
1703 | active_objs += cachep->num; |
1704 | active_slabs++; | |
1705 | } | |
8456a648 JK |
1706 | list_for_each_entry(page, &n->slabs_partial, lru) { |
1707 | active_objs += page->active; | |
8bdec192 RA |
1708 | active_slabs++; |
1709 | } | |
8456a648 | 1710 | list_for_each_entry(page, &n->slabs_free, lru) |
8bdec192 RA |
1711 | num_slabs++; |
1712 | ||
ce8eb6c4 CL |
1713 | free_objects += n->free_objects; |
1714 | spin_unlock_irqrestore(&n->list_lock, flags); | |
8bdec192 RA |
1715 | |
1716 | num_slabs += active_slabs; | |
1717 | num_objs = num_slabs * cachep->num; | |
1718 | printk(KERN_WARNING | |
1719 | " node %d: slabs: %ld/%ld, objs: %ld/%ld, free: %ld\n", | |
1720 | node, active_slabs, num_slabs, active_objs, num_objs, | |
1721 | free_objects); | |
1722 | } | |
9a02d699 | 1723 | #endif |
8bdec192 RA |
1724 | } |
1725 | ||
1da177e4 LT |
1726 | /* |
1727 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1728 | * | |
1729 | * If we requested dmaable memory, we will get it. Even if we | |
1730 | * did not request dmaable memory, we might get it, but that | |
1731 | * would be relatively rare and ignorable. | |
1732 | */ | |
0c3aa83e JK |
1733 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1734 | int nodeid) | |
1da177e4 LT |
1735 | { |
1736 | struct page *page; | |
e1b6aa6f | 1737 | int nr_pages; |
765c4507 | 1738 | |
a618e89f | 1739 | flags |= cachep->allocflags; |
e12ba74d MG |
1740 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1741 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1742 | |
5dfb4175 VD |
1743 | if (memcg_charge_slab(cachep, flags, cachep->gfporder)) |
1744 | return NULL; | |
1745 | ||
517d0869 | 1746 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
8bdec192 | 1747 | if (!page) { |
5dfb4175 | 1748 | memcg_uncharge_slab(cachep, cachep->gfporder); |
9a02d699 | 1749 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1750 | return NULL; |
8bdec192 | 1751 | } |
1da177e4 | 1752 | |
b37f1dd0 | 1753 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
072bb0aa MG |
1754 | if (unlikely(page->pfmemalloc)) |
1755 | pfmemalloc_active = true; | |
1756 | ||
e1b6aa6f | 1757 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1758 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1759 | add_zone_page_state(page_zone(page), |
1760 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1761 | else | |
1762 | add_zone_page_state(page_zone(page), | |
1763 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
a57a4988 JK |
1764 | __SetPageSlab(page); |
1765 | if (page->pfmemalloc) | |
1766 | SetPageSlabPfmemalloc(page); | |
072bb0aa | 1767 | |
b1eeab67 VN |
1768 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1769 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1770 | ||
1771 | if (cachep->ctor) | |
1772 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1773 | else | |
1774 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1775 | } | |
c175eea4 | 1776 | |
0c3aa83e | 1777 | return page; |
1da177e4 LT |
1778 | } |
1779 | ||
1780 | /* | |
1781 | * Interface to system's page release. | |
1782 | */ | |
0c3aa83e | 1783 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1784 | { |
a57a4988 | 1785 | const unsigned long nr_freed = (1 << cachep->gfporder); |
1da177e4 | 1786 | |
b1eeab67 | 1787 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1788 | |
972d1a7b CL |
1789 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1790 | sub_zone_page_state(page_zone(page), | |
1791 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1792 | else | |
1793 | sub_zone_page_state(page_zone(page), | |
1794 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
73293c2f | 1795 | |
a57a4988 | 1796 | BUG_ON(!PageSlab(page)); |
73293c2f | 1797 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1798 | __ClearPageSlab(page); |
8456a648 JK |
1799 | page_mapcount_reset(page); |
1800 | page->mapping = NULL; | |
1f458cbf | 1801 | |
1da177e4 LT |
1802 | if (current->reclaim_state) |
1803 | current->reclaim_state->reclaimed_slab += nr_freed; | |
5dfb4175 VD |
1804 | __free_pages(page, cachep->gfporder); |
1805 | memcg_uncharge_slab(cachep, cachep->gfporder); | |
1da177e4 LT |
1806 | } |
1807 | ||
1808 | static void kmem_rcu_free(struct rcu_head *head) | |
1809 | { | |
68126702 JK |
1810 | struct kmem_cache *cachep; |
1811 | struct page *page; | |
1da177e4 | 1812 | |
68126702 JK |
1813 | page = container_of(head, struct page, rcu_head); |
1814 | cachep = page->slab_cache; | |
1815 | ||
1816 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1817 | } |
1818 | ||
1819 | #if DEBUG | |
1820 | ||
1821 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1822 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1823 | unsigned long caller) |
1da177e4 | 1824 | { |
8c138bc0 | 1825 | int size = cachep->object_size; |
1da177e4 | 1826 | |
3dafccf2 | 1827 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1828 | |
b28a02de | 1829 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1830 | return; |
1831 | ||
b28a02de PE |
1832 | *addr++ = 0x12345678; |
1833 | *addr++ = caller; | |
1834 | *addr++ = smp_processor_id(); | |
1835 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1836 | { |
1837 | unsigned long *sptr = &caller; | |
1838 | unsigned long svalue; | |
1839 | ||
1840 | while (!kstack_end(sptr)) { | |
1841 | svalue = *sptr++; | |
1842 | if (kernel_text_address(svalue)) { | |
b28a02de | 1843 | *addr++ = svalue; |
1da177e4 LT |
1844 | size -= sizeof(unsigned long); |
1845 | if (size <= sizeof(unsigned long)) | |
1846 | break; | |
1847 | } | |
1848 | } | |
1849 | ||
1850 | } | |
b28a02de | 1851 | *addr++ = 0x87654321; |
1da177e4 LT |
1852 | } |
1853 | #endif | |
1854 | ||
343e0d7a | 1855 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1856 | { |
8c138bc0 | 1857 | int size = cachep->object_size; |
3dafccf2 | 1858 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1859 | |
1860 | memset(addr, val, size); | |
b28a02de | 1861 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1862 | } |
1863 | ||
1864 | static void dump_line(char *data, int offset, int limit) | |
1865 | { | |
1866 | int i; | |
aa83aa40 DJ |
1867 | unsigned char error = 0; |
1868 | int bad_count = 0; | |
1869 | ||
fdde6abb | 1870 | printk(KERN_ERR "%03x: ", offset); |
aa83aa40 DJ |
1871 | for (i = 0; i < limit; i++) { |
1872 | if (data[offset + i] != POISON_FREE) { | |
1873 | error = data[offset + i]; | |
1874 | bad_count++; | |
1875 | } | |
aa83aa40 | 1876 | } |
fdde6abb SAS |
1877 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1878 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1879 | |
1880 | if (bad_count == 1) { | |
1881 | error ^= POISON_FREE; | |
1882 | if (!(error & (error - 1))) { | |
1883 | printk(KERN_ERR "Single bit error detected. Probably " | |
1884 | "bad RAM.\n"); | |
1885 | #ifdef CONFIG_X86 | |
1886 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1887 | "test tool.\n"); | |
1888 | #else | |
1889 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1890 | #endif | |
1891 | } | |
1892 | } | |
1da177e4 LT |
1893 | } |
1894 | #endif | |
1895 | ||
1896 | #if DEBUG | |
1897 | ||
343e0d7a | 1898 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1899 | { |
1900 | int i, size; | |
1901 | char *realobj; | |
1902 | ||
1903 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1904 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1905 | *dbg_redzone1(cachep, objp), |
1906 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1907 | } |
1908 | ||
1909 | if (cachep->flags & SLAB_STORE_USER) { | |
071361d3 JP |
1910 | printk(KERN_ERR "Last user: [<%p>](%pSR)\n", |
1911 | *dbg_userword(cachep, objp), | |
1912 | *dbg_userword(cachep, objp)); | |
1da177e4 | 1913 | } |
3dafccf2 | 1914 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1915 | size = cachep->object_size; |
b28a02de | 1916 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1917 | int limit; |
1918 | limit = 16; | |
b28a02de PE |
1919 | if (i + limit > size) |
1920 | limit = size - i; | |
1da177e4 LT |
1921 | dump_line(realobj, i, limit); |
1922 | } | |
1923 | } | |
1924 | ||
343e0d7a | 1925 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1926 | { |
1927 | char *realobj; | |
1928 | int size, i; | |
1929 | int lines = 0; | |
1930 | ||
3dafccf2 | 1931 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1932 | size = cachep->object_size; |
1da177e4 | 1933 | |
b28a02de | 1934 | for (i = 0; i < size; i++) { |
1da177e4 | 1935 | char exp = POISON_FREE; |
b28a02de | 1936 | if (i == size - 1) |
1da177e4 LT |
1937 | exp = POISON_END; |
1938 | if (realobj[i] != exp) { | |
1939 | int limit; | |
1940 | /* Mismatch ! */ | |
1941 | /* Print header */ | |
1942 | if (lines == 0) { | |
b28a02de | 1943 | printk(KERN_ERR |
face37f5 DJ |
1944 | "Slab corruption (%s): %s start=%p, len=%d\n", |
1945 | print_tainted(), cachep->name, realobj, size); | |
1da177e4 LT |
1946 | print_objinfo(cachep, objp, 0); |
1947 | } | |
1948 | /* Hexdump the affected line */ | |
b28a02de | 1949 | i = (i / 16) * 16; |
1da177e4 | 1950 | limit = 16; |
b28a02de PE |
1951 | if (i + limit > size) |
1952 | limit = size - i; | |
1da177e4 LT |
1953 | dump_line(realobj, i, limit); |
1954 | i += 16; | |
1955 | lines++; | |
1956 | /* Limit to 5 lines */ | |
1957 | if (lines > 5) | |
1958 | break; | |
1959 | } | |
1960 | } | |
1961 | if (lines != 0) { | |
1962 | /* Print some data about the neighboring objects, if they | |
1963 | * exist: | |
1964 | */ | |
8456a648 | 1965 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1966 | unsigned int objnr; |
1da177e4 | 1967 | |
8456a648 | 1968 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1969 | if (objnr) { |
8456a648 | 1970 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1971 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1972 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1973 | realobj, size); |
1da177e4 LT |
1974 | print_objinfo(cachep, objp, 2); |
1975 | } | |
b28a02de | 1976 | if (objnr + 1 < cachep->num) { |
8456a648 | 1977 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1978 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1979 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1980 | realobj, size); |
1da177e4 LT |
1981 | print_objinfo(cachep, objp, 2); |
1982 | } | |
1983 | } | |
1984 | } | |
1985 | #endif | |
1986 | ||
12dd36fa | 1987 | #if DEBUG |
8456a648 JK |
1988 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1989 | struct page *page) | |
1da177e4 | 1990 | { |
1da177e4 LT |
1991 | int i; |
1992 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 1993 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1994 | |
1995 | if (cachep->flags & SLAB_POISON) { | |
1996 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 1997 | if (cachep->size % PAGE_SIZE == 0 && |
a737b3e2 | 1998 | OFF_SLAB(cachep)) |
b28a02de | 1999 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2000 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
2001 | else |
2002 | check_poison_obj(cachep, objp); | |
2003 | #else | |
2004 | check_poison_obj(cachep, objp); | |
2005 | #endif | |
2006 | } | |
2007 | if (cachep->flags & SLAB_RED_ZONE) { | |
2008 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
2009 | slab_error(cachep, "start of a freed object " | |
b28a02de | 2010 | "was overwritten"); |
1da177e4 LT |
2011 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
2012 | slab_error(cachep, "end of a freed object " | |
b28a02de | 2013 | "was overwritten"); |
1da177e4 | 2014 | } |
1da177e4 | 2015 | } |
12dd36fa | 2016 | } |
1da177e4 | 2017 | #else |
8456a648 JK |
2018 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
2019 | struct page *page) | |
12dd36fa | 2020 | { |
12dd36fa | 2021 | } |
1da177e4 LT |
2022 | #endif |
2023 | ||
911851e6 RD |
2024 | /** |
2025 | * slab_destroy - destroy and release all objects in a slab | |
2026 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 2027 | * @page: page pointer being destroyed |
911851e6 | 2028 | * |
12dd36fa | 2029 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
2030 | * Before calling the slab must have been unlinked from the cache. The |
2031 | * cache-lock is not held/needed. | |
12dd36fa | 2032 | */ |
8456a648 | 2033 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 2034 | { |
7e007355 | 2035 | void *freelist; |
12dd36fa | 2036 | |
8456a648 JK |
2037 | freelist = page->freelist; |
2038 | slab_destroy_debugcheck(cachep, page); | |
1da177e4 | 2039 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
68126702 JK |
2040 | struct rcu_head *head; |
2041 | ||
2042 | /* | |
2043 | * RCU free overloads the RCU head over the LRU. | |
2044 | * slab_page has been overloeaded over the LRU, | |
2045 | * however it is not used from now on so that | |
2046 | * we can use it safely. | |
2047 | */ | |
2048 | head = (void *)&page->rcu_head; | |
2049 | call_rcu(head, kmem_rcu_free); | |
1da177e4 | 2050 | |
1da177e4 | 2051 | } else { |
0c3aa83e | 2052 | kmem_freepages(cachep, page); |
1da177e4 | 2053 | } |
68126702 JK |
2054 | |
2055 | /* | |
8456a648 | 2056 | * From now on, we don't use freelist |
68126702 JK |
2057 | * although actual page can be freed in rcu context |
2058 | */ | |
2059 | if (OFF_SLAB(cachep)) | |
8456a648 | 2060 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
2061 | } |
2062 | ||
4d268eba | 2063 | /** |
a70773dd RD |
2064 | * calculate_slab_order - calculate size (page order) of slabs |
2065 | * @cachep: pointer to the cache that is being created | |
2066 | * @size: size of objects to be created in this cache. | |
2067 | * @align: required alignment for the objects. | |
2068 | * @flags: slab allocation flags | |
2069 | * | |
2070 | * Also calculates the number of objects per slab. | |
4d268eba PE |
2071 | * |
2072 | * This could be made much more intelligent. For now, try to avoid using | |
2073 | * high order pages for slabs. When the gfp() functions are more friendly | |
2074 | * towards high-order requests, this should be changed. | |
2075 | */ | |
a737b3e2 | 2076 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 2077 | size_t size, size_t align, unsigned long flags) |
4d268eba | 2078 | { |
b1ab41c4 | 2079 | unsigned long offslab_limit; |
4d268eba | 2080 | size_t left_over = 0; |
9888e6fa | 2081 | int gfporder; |
4d268eba | 2082 | |
0aa817f0 | 2083 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
2084 | unsigned int num; |
2085 | size_t remainder; | |
2086 | ||
9888e6fa | 2087 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
2088 | if (!num) |
2089 | continue; | |
9888e6fa | 2090 | |
f315e3fa JK |
2091 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
2092 | if (num > SLAB_OBJ_MAX_NUM) | |
2093 | break; | |
2094 | ||
b1ab41c4 | 2095 | if (flags & CFLGS_OFF_SLAB) { |
03787301 | 2096 | size_t freelist_size_per_obj = sizeof(freelist_idx_t); |
b1ab41c4 IM |
2097 | /* |
2098 | * Max number of objs-per-slab for caches which | |
2099 | * use off-slab slabs. Needed to avoid a possible | |
2100 | * looping condition in cache_grow(). | |
2101 | */ | |
03787301 JK |
2102 | if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK)) |
2103 | freelist_size_per_obj += sizeof(char); | |
8456a648 | 2104 | offslab_limit = size; |
03787301 | 2105 | offslab_limit /= freelist_size_per_obj; |
b1ab41c4 IM |
2106 | |
2107 | if (num > offslab_limit) | |
2108 | break; | |
2109 | } | |
4d268eba | 2110 | |
9888e6fa | 2111 | /* Found something acceptable - save it away */ |
4d268eba | 2112 | cachep->num = num; |
9888e6fa | 2113 | cachep->gfporder = gfporder; |
4d268eba PE |
2114 | left_over = remainder; |
2115 | ||
f78bb8ad LT |
2116 | /* |
2117 | * A VFS-reclaimable slab tends to have most allocations | |
2118 | * as GFP_NOFS and we really don't want to have to be allocating | |
2119 | * higher-order pages when we are unable to shrink dcache. | |
2120 | */ | |
2121 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2122 | break; | |
2123 | ||
4d268eba PE |
2124 | /* |
2125 | * Large number of objects is good, but very large slabs are | |
2126 | * currently bad for the gfp()s. | |
2127 | */ | |
543585cc | 2128 | if (gfporder >= slab_max_order) |
4d268eba PE |
2129 | break; |
2130 | ||
9888e6fa LT |
2131 | /* |
2132 | * Acceptable internal fragmentation? | |
2133 | */ | |
a737b3e2 | 2134 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2135 | break; |
2136 | } | |
2137 | return left_over; | |
2138 | } | |
2139 | ||
83b519e8 | 2140 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2141 | { |
97d06609 | 2142 | if (slab_state >= FULL) |
83b519e8 | 2143 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2144 | |
97d06609 | 2145 | if (slab_state == DOWN) { |
f30cf7d1 | 2146 | /* |
2f9baa9f | 2147 | * Note: Creation of first cache (kmem_cache). |
ce8eb6c4 | 2148 | * The setup_node is taken care |
2f9baa9f CL |
2149 | * of by the caller of __kmem_cache_create |
2150 | */ | |
2151 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2152 | slab_state = PARTIAL; | |
2153 | } else if (slab_state == PARTIAL) { | |
2154 | /* | |
2155 | * Note: the second kmem_cache_create must create the cache | |
f30cf7d1 PE |
2156 | * that's used by kmalloc(24), otherwise the creation of |
2157 | * further caches will BUG(). | |
2158 | */ | |
2159 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2160 | ||
2161 | /* | |
ce8eb6c4 CL |
2162 | * If the cache that's used by kmalloc(sizeof(kmem_cache_node)) is |
2163 | * the second cache, then we need to set up all its node/, | |
f30cf7d1 PE |
2164 | * otherwise the creation of further caches will BUG(). |
2165 | */ | |
ce8eb6c4 CL |
2166 | set_up_node(cachep, SIZE_AC); |
2167 | if (INDEX_AC == INDEX_NODE) | |
2168 | slab_state = PARTIAL_NODE; | |
f30cf7d1 | 2169 | else |
97d06609 | 2170 | slab_state = PARTIAL_ARRAYCACHE; |
f30cf7d1 | 2171 | } else { |
2f9baa9f | 2172 | /* Remaining boot caches */ |
f30cf7d1 | 2173 | cachep->array[smp_processor_id()] = |
83b519e8 | 2174 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 | 2175 | |
97d06609 | 2176 | if (slab_state == PARTIAL_ARRAYCACHE) { |
ce8eb6c4 CL |
2177 | set_up_node(cachep, SIZE_NODE); |
2178 | slab_state = PARTIAL_NODE; | |
f30cf7d1 PE |
2179 | } else { |
2180 | int node; | |
556a169d | 2181 | for_each_online_node(node) { |
6a67368c | 2182 | cachep->node[node] = |
6744f087 | 2183 | kmalloc_node(sizeof(struct kmem_cache_node), |
eb91f1d0 | 2184 | gfp, node); |
6a67368c | 2185 | BUG_ON(!cachep->node[node]); |
ce8eb6c4 | 2186 | kmem_cache_node_init(cachep->node[node]); |
f30cf7d1 PE |
2187 | } |
2188 | } | |
2189 | } | |
6a67368c | 2190 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
2191 | jiffies + REAPTIMEOUT_NODE + |
2192 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
2193 | |
2194 | cpu_cache_get(cachep)->avail = 0; | |
2195 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2196 | cpu_cache_get(cachep)->batchcount = 1; | |
2197 | cpu_cache_get(cachep)->touched = 0; | |
2198 | cachep->batchcount = 1; | |
2199 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2200 | return 0; |
f30cf7d1 PE |
2201 | } |
2202 | ||
1da177e4 | 2203 | /** |
039363f3 | 2204 | * __kmem_cache_create - Create a cache. |
a755b76a | 2205 | * @cachep: cache management descriptor |
1da177e4 | 2206 | * @flags: SLAB flags |
1da177e4 LT |
2207 | * |
2208 | * Returns a ptr to the cache on success, NULL on failure. | |
2209 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2210 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 2211 | * |
1da177e4 LT |
2212 | * The flags are |
2213 | * | |
2214 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2215 | * to catch references to uninitialised memory. | |
2216 | * | |
2217 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2218 | * for buffer overruns. | |
2219 | * | |
1da177e4 LT |
2220 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2221 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2222 | * as davem. | |
2223 | */ | |
278b1bb1 | 2224 | int |
8a13a4cc | 2225 | __kmem_cache_create (struct kmem_cache *cachep, unsigned long flags) |
1da177e4 | 2226 | { |
8456a648 | 2227 | size_t left_over, freelist_size, ralign; |
83b519e8 | 2228 | gfp_t gfp; |
278b1bb1 | 2229 | int err; |
8a13a4cc | 2230 | size_t size = cachep->size; |
1da177e4 | 2231 | |
1da177e4 | 2232 | #if DEBUG |
1da177e4 LT |
2233 | #if FORCED_DEBUG |
2234 | /* | |
2235 | * Enable redzoning and last user accounting, except for caches with | |
2236 | * large objects, if the increased size would increase the object size | |
2237 | * above the next power of two: caches with object sizes just above a | |
2238 | * power of two have a significant amount of internal fragmentation. | |
2239 | */ | |
87a927c7 DW |
2240 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2241 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2242 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2243 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2244 | flags |= SLAB_POISON; | |
2245 | #endif | |
2246 | if (flags & SLAB_DESTROY_BY_RCU) | |
2247 | BUG_ON(flags & SLAB_POISON); | |
2248 | #endif | |
1da177e4 | 2249 | |
a737b3e2 AM |
2250 | /* |
2251 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2252 | * unaligned accesses for some archs when redzoning is used, and makes |
2253 | * sure any on-slab bufctl's are also correctly aligned. | |
2254 | */ | |
b28a02de PE |
2255 | if (size & (BYTES_PER_WORD - 1)) { |
2256 | size += (BYTES_PER_WORD - 1); | |
2257 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2258 | } |
2259 | ||
ca5f9703 | 2260 | /* |
87a927c7 DW |
2261 | * Redzoning and user store require word alignment or possibly larger. |
2262 | * Note this will be overridden by architecture or caller mandated | |
2263 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2264 | */ |
87a927c7 DW |
2265 | if (flags & SLAB_STORE_USER) |
2266 | ralign = BYTES_PER_WORD; | |
2267 | ||
2268 | if (flags & SLAB_RED_ZONE) { | |
2269 | ralign = REDZONE_ALIGN; | |
2270 | /* If redzoning, ensure that the second redzone is suitably | |
2271 | * aligned, by adjusting the object size accordingly. */ | |
2272 | size += REDZONE_ALIGN - 1; | |
2273 | size &= ~(REDZONE_ALIGN - 1); | |
2274 | } | |
ca5f9703 | 2275 | |
a44b56d3 | 2276 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2277 | if (ralign < cachep->align) { |
2278 | ralign = cachep->align; | |
1da177e4 | 2279 | } |
3ff84a7f PE |
2280 | /* disable debug if necessary */ |
2281 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2282 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2283 | /* |
ca5f9703 | 2284 | * 4) Store it. |
1da177e4 | 2285 | */ |
8a13a4cc | 2286 | cachep->align = ralign; |
1da177e4 | 2287 | |
83b519e8 PE |
2288 | if (slab_is_available()) |
2289 | gfp = GFP_KERNEL; | |
2290 | else | |
2291 | gfp = GFP_NOWAIT; | |
2292 | ||
6a67368c | 2293 | setup_node_pointer(cachep); |
1da177e4 | 2294 | #if DEBUG |
1da177e4 | 2295 | |
ca5f9703 PE |
2296 | /* |
2297 | * Both debugging options require word-alignment which is calculated | |
2298 | * into align above. | |
2299 | */ | |
1da177e4 | 2300 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2301 | /* add space for red zone words */ |
3ff84a7f PE |
2302 | cachep->obj_offset += sizeof(unsigned long long); |
2303 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2304 | } |
2305 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2306 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2307 | * the real object. But if the second red zone needs to be |
2308 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2309 | */ |
87a927c7 DW |
2310 | if (flags & SLAB_RED_ZONE) |
2311 | size += REDZONE_ALIGN; | |
2312 | else | |
2313 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2314 | } |
2315 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
ce8eb6c4 | 2316 | if (size >= kmalloc_size(INDEX_NODE + 1) |
608da7e3 TH |
2317 | && cachep->object_size > cache_line_size() |
2318 | && ALIGN(size, cachep->align) < PAGE_SIZE) { | |
2319 | cachep->obj_offset += PAGE_SIZE - ALIGN(size, cachep->align); | |
1da177e4 LT |
2320 | size = PAGE_SIZE; |
2321 | } | |
2322 | #endif | |
2323 | #endif | |
2324 | ||
e0a42726 IM |
2325 | /* |
2326 | * Determine if the slab management is 'on' or 'off' slab. | |
2327 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2328 | * it too early on. Always use on-slab management when |
2329 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2330 | */ |
8fc9cf42 | 2331 | if ((size >= (PAGE_SIZE >> 5)) && !slab_early_init && |
e7cb55b9 | 2332 | !(flags & SLAB_NOLEAKTRACE)) |
1da177e4 LT |
2333 | /* |
2334 | * Size is large, assume best to place the slab management obj | |
2335 | * off-slab (should allow better packing of objs). | |
2336 | */ | |
2337 | flags |= CFLGS_OFF_SLAB; | |
2338 | ||
8a13a4cc | 2339 | size = ALIGN(size, cachep->align); |
f315e3fa JK |
2340 | /* |
2341 | * We should restrict the number of objects in a slab to implement | |
2342 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2343 | */ | |
2344 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2345 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
1da177e4 | 2346 | |
8a13a4cc | 2347 | left_over = calculate_slab_order(cachep, size, cachep->align, flags); |
1da177e4 | 2348 | |
8a13a4cc | 2349 | if (!cachep->num) |
278b1bb1 | 2350 | return -E2BIG; |
1da177e4 | 2351 | |
03787301 | 2352 | freelist_size = calculate_freelist_size(cachep->num, cachep->align); |
1da177e4 LT |
2353 | |
2354 | /* | |
2355 | * If the slab has been placed off-slab, and we have enough space then | |
2356 | * move it on-slab. This is at the expense of any extra colouring. | |
2357 | */ | |
8456a648 | 2358 | if (flags & CFLGS_OFF_SLAB && left_over >= freelist_size) { |
1da177e4 | 2359 | flags &= ~CFLGS_OFF_SLAB; |
8456a648 | 2360 | left_over -= freelist_size; |
1da177e4 LT |
2361 | } |
2362 | ||
2363 | if (flags & CFLGS_OFF_SLAB) { | |
2364 | /* really off slab. No need for manual alignment */ | |
03787301 | 2365 | freelist_size = calculate_freelist_size(cachep->num, 0); |
67461365 RL |
2366 | |
2367 | #ifdef CONFIG_PAGE_POISONING | |
2368 | /* If we're going to use the generic kernel_map_pages() | |
2369 | * poisoning, then it's going to smash the contents of | |
2370 | * the redzone and userword anyhow, so switch them off. | |
2371 | */ | |
2372 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2373 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2374 | #endif | |
1da177e4 LT |
2375 | } |
2376 | ||
2377 | cachep->colour_off = cache_line_size(); | |
2378 | /* Offset must be a multiple of the alignment. */ | |
8a13a4cc CL |
2379 | if (cachep->colour_off < cachep->align) |
2380 | cachep->colour_off = cachep->align; | |
b28a02de | 2381 | cachep->colour = left_over / cachep->colour_off; |
8456a648 | 2382 | cachep->freelist_size = freelist_size; |
1da177e4 | 2383 | cachep->flags = flags; |
a57a4988 | 2384 | cachep->allocflags = __GFP_COMP; |
4b51d669 | 2385 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
a618e89f | 2386 | cachep->allocflags |= GFP_DMA; |
3b0efdfa | 2387 | cachep->size = size; |
6a2d7a95 | 2388 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2389 | |
e5ac9c5a | 2390 | if (flags & CFLGS_OFF_SLAB) { |
8456a648 | 2391 | cachep->freelist_cache = kmalloc_slab(freelist_size, 0u); |
e5ac9c5a | 2392 | /* |
5f0985bb | 2393 | * This is a possibility for one of the kmalloc_{dma,}_caches. |
e5ac9c5a | 2394 | * But since we go off slab only for object size greater than |
5f0985bb JZ |
2395 | * PAGE_SIZE/8, and kmalloc_{dma,}_caches get created |
2396 | * in ascending order,this should not happen at all. | |
e5ac9c5a RT |
2397 | * But leave a BUG_ON for some lucky dude. |
2398 | */ | |
8456a648 | 2399 | BUG_ON(ZERO_OR_NULL_PTR(cachep->freelist_cache)); |
e5ac9c5a | 2400 | } |
1da177e4 | 2401 | |
278b1bb1 CL |
2402 | err = setup_cpu_cache(cachep, gfp); |
2403 | if (err) { | |
12c3667f | 2404 | __kmem_cache_shutdown(cachep); |
278b1bb1 | 2405 | return err; |
2ed3a4ef | 2406 | } |
1da177e4 | 2407 | |
83835b3d PZ |
2408 | if (flags & SLAB_DEBUG_OBJECTS) { |
2409 | /* | |
2410 | * Would deadlock through slab_destroy()->call_rcu()-> | |
2411 | * debug_object_activate()->kmem_cache_alloc(). | |
2412 | */ | |
2413 | WARN_ON_ONCE(flags & SLAB_DESTROY_BY_RCU); | |
2414 | ||
2415 | slab_set_debugobj_lock_classes(cachep); | |
6ccfb5bc GC |
2416 | } else if (!OFF_SLAB(cachep) && !(flags & SLAB_DESTROY_BY_RCU)) |
2417 | on_slab_lock_classes(cachep); | |
83835b3d | 2418 | |
278b1bb1 | 2419 | return 0; |
1da177e4 | 2420 | } |
1da177e4 LT |
2421 | |
2422 | #if DEBUG | |
2423 | static void check_irq_off(void) | |
2424 | { | |
2425 | BUG_ON(!irqs_disabled()); | |
2426 | } | |
2427 | ||
2428 | static void check_irq_on(void) | |
2429 | { | |
2430 | BUG_ON(irqs_disabled()); | |
2431 | } | |
2432 | ||
343e0d7a | 2433 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2434 | { |
2435 | #ifdef CONFIG_SMP | |
2436 | check_irq_off(); | |
6a67368c | 2437 | assert_spin_locked(&cachep->node[numa_mem_id()]->list_lock); |
1da177e4 LT |
2438 | #endif |
2439 | } | |
e498be7d | 2440 | |
343e0d7a | 2441 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2442 | { |
2443 | #ifdef CONFIG_SMP | |
2444 | check_irq_off(); | |
6a67368c | 2445 | assert_spin_locked(&cachep->node[node]->list_lock); |
e498be7d CL |
2446 | #endif |
2447 | } | |
2448 | ||
1da177e4 LT |
2449 | #else |
2450 | #define check_irq_off() do { } while(0) | |
2451 | #define check_irq_on() do { } while(0) | |
2452 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2453 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2454 | #endif |
2455 | ||
ce8eb6c4 | 2456 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
aab2207c CL |
2457 | struct array_cache *ac, |
2458 | int force, int node); | |
2459 | ||
1da177e4 LT |
2460 | static void do_drain(void *arg) |
2461 | { | |
a737b3e2 | 2462 | struct kmem_cache *cachep = arg; |
1da177e4 | 2463 | struct array_cache *ac; |
7d6e6d09 | 2464 | int node = numa_mem_id(); |
1da177e4 LT |
2465 | |
2466 | check_irq_off(); | |
9a2dba4b | 2467 | ac = cpu_cache_get(cachep); |
6a67368c | 2468 | spin_lock(&cachep->node[node]->list_lock); |
ff69416e | 2469 | free_block(cachep, ac->entry, ac->avail, node); |
6a67368c | 2470 | spin_unlock(&cachep->node[node]->list_lock); |
1da177e4 LT |
2471 | ac->avail = 0; |
2472 | } | |
2473 | ||
343e0d7a | 2474 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2475 | { |
ce8eb6c4 | 2476 | struct kmem_cache_node *n; |
e498be7d CL |
2477 | int node; |
2478 | ||
15c8b6c1 | 2479 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2480 | check_irq_on(); |
b28a02de | 2481 | for_each_online_node(node) { |
ce8eb6c4 CL |
2482 | n = cachep->node[node]; |
2483 | if (n && n->alien) | |
2484 | drain_alien_cache(cachep, n->alien); | |
a4523a8b RD |
2485 | } |
2486 | ||
2487 | for_each_online_node(node) { | |
ce8eb6c4 CL |
2488 | n = cachep->node[node]; |
2489 | if (n) | |
2490 | drain_array(cachep, n, n->shared, 1, node); | |
e498be7d | 2491 | } |
1da177e4 LT |
2492 | } |
2493 | ||
ed11d9eb CL |
2494 | /* |
2495 | * Remove slabs from the list of free slabs. | |
2496 | * Specify the number of slabs to drain in tofree. | |
2497 | * | |
2498 | * Returns the actual number of slabs released. | |
2499 | */ | |
2500 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2501 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2502 | { |
ed11d9eb CL |
2503 | struct list_head *p; |
2504 | int nr_freed; | |
8456a648 | 2505 | struct page *page; |
1da177e4 | 2506 | |
ed11d9eb | 2507 | nr_freed = 0; |
ce8eb6c4 | 2508 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2509 | |
ce8eb6c4 CL |
2510 | spin_lock_irq(&n->list_lock); |
2511 | p = n->slabs_free.prev; | |
2512 | if (p == &n->slabs_free) { | |
2513 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2514 | goto out; |
2515 | } | |
1da177e4 | 2516 | |
8456a648 | 2517 | page = list_entry(p, struct page, lru); |
1da177e4 | 2518 | #if DEBUG |
8456a648 | 2519 | BUG_ON(page->active); |
1da177e4 | 2520 | #endif |
8456a648 | 2521 | list_del(&page->lru); |
ed11d9eb CL |
2522 | /* |
2523 | * Safe to drop the lock. The slab is no longer linked | |
2524 | * to the cache. | |
2525 | */ | |
ce8eb6c4 CL |
2526 | n->free_objects -= cache->num; |
2527 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2528 | slab_destroy(cache, page); |
ed11d9eb | 2529 | nr_freed++; |
1da177e4 | 2530 | } |
ed11d9eb CL |
2531 | out: |
2532 | return nr_freed; | |
1da177e4 LT |
2533 | } |
2534 | ||
03afc0e2 | 2535 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2536 | { |
2537 | int ret = 0, i = 0; | |
ce8eb6c4 | 2538 | struct kmem_cache_node *n; |
e498be7d CL |
2539 | |
2540 | drain_cpu_caches(cachep); | |
2541 | ||
2542 | check_irq_on(); | |
2543 | for_each_online_node(i) { | |
ce8eb6c4 CL |
2544 | n = cachep->node[i]; |
2545 | if (!n) | |
ed11d9eb CL |
2546 | continue; |
2547 | ||
0fa8103b | 2548 | drain_freelist(cachep, n, slabs_tofree(cachep, n)); |
ed11d9eb | 2549 | |
ce8eb6c4 CL |
2550 | ret += !list_empty(&n->slabs_full) || |
2551 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2552 | } |
2553 | return (ret ? 1 : 0); | |
2554 | } | |
2555 | ||
945cf2b6 | 2556 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
1da177e4 | 2557 | { |
12c3667f | 2558 | int i; |
ce8eb6c4 | 2559 | struct kmem_cache_node *n; |
03afc0e2 | 2560 | int rc = __kmem_cache_shrink(cachep); |
1da177e4 | 2561 | |
12c3667f CL |
2562 | if (rc) |
2563 | return rc; | |
1da177e4 | 2564 | |
12c3667f CL |
2565 | for_each_online_cpu(i) |
2566 | kfree(cachep->array[i]); | |
1da177e4 | 2567 | |
ce8eb6c4 | 2568 | /* NUMA: free the node structures */ |
12c3667f | 2569 | for_each_online_node(i) { |
ce8eb6c4 CL |
2570 | n = cachep->node[i]; |
2571 | if (n) { | |
2572 | kfree(n->shared); | |
2573 | free_alien_cache(n->alien); | |
2574 | kfree(n); | |
12c3667f CL |
2575 | } |
2576 | } | |
2577 | return 0; | |
1da177e4 | 2578 | } |
1da177e4 | 2579 | |
e5ac9c5a RT |
2580 | /* |
2581 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2582 | * |
2583 | * For a slab cache when the slab descriptor is off-slab, the | |
2584 | * slab descriptor can't come from the same cache which is being created, | |
2585 | * Because if it is the case, that means we defer the creation of | |
2586 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2587 | * And we eventually call down to __kmem_cache_create(), which | |
2588 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2589 | * This is a "chicken-and-egg" problem. | |
2590 | * | |
2591 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2592 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2593 | */ |
7e007355 | 2594 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2595 | struct page *page, int colour_off, |
2596 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2597 | { |
7e007355 | 2598 | void *freelist; |
0c3aa83e | 2599 | void *addr = page_address(page); |
b28a02de | 2600 | |
1da177e4 LT |
2601 | if (OFF_SLAB(cachep)) { |
2602 | /* Slab management obj is off-slab. */ | |
8456a648 | 2603 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2604 | local_flags, nodeid); |
8456a648 | 2605 | if (!freelist) |
1da177e4 LT |
2606 | return NULL; |
2607 | } else { | |
8456a648 JK |
2608 | freelist = addr + colour_off; |
2609 | colour_off += cachep->freelist_size; | |
1da177e4 | 2610 | } |
8456a648 JK |
2611 | page->active = 0; |
2612 | page->s_mem = addr + colour_off; | |
2613 | return freelist; | |
1da177e4 LT |
2614 | } |
2615 | ||
7cc68973 | 2616 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2617 | { |
a41adfaa | 2618 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2619 | } |
2620 | ||
2621 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2622 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2623 | { |
a41adfaa | 2624 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2625 | } |
2626 | ||
343e0d7a | 2627 | static void cache_init_objs(struct kmem_cache *cachep, |
8456a648 | 2628 | struct page *page) |
1da177e4 LT |
2629 | { |
2630 | int i; | |
2631 | ||
2632 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2633 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
2634 | #if DEBUG |
2635 | /* need to poison the objs? */ | |
2636 | if (cachep->flags & SLAB_POISON) | |
2637 | poison_obj(cachep, objp, POISON_FREE); | |
2638 | if (cachep->flags & SLAB_STORE_USER) | |
2639 | *dbg_userword(cachep, objp) = NULL; | |
2640 | ||
2641 | if (cachep->flags & SLAB_RED_ZONE) { | |
2642 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2643 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2644 | } | |
2645 | /* | |
a737b3e2 AM |
2646 | * Constructors are not allowed to allocate memory from the same |
2647 | * cache which they are a constructor for. Otherwise, deadlock. | |
2648 | * They must also be threaded. | |
1da177e4 LT |
2649 | */ |
2650 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2651 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2652 | |
2653 | if (cachep->flags & SLAB_RED_ZONE) { | |
2654 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2655 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2656 | " end of an object"); |
1da177e4 LT |
2657 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2658 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2659 | " start of an object"); |
1da177e4 | 2660 | } |
3b0efdfa | 2661 | if ((cachep->size % PAGE_SIZE) == 0 && |
a737b3e2 | 2662 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) |
b28a02de | 2663 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2664 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2665 | #else |
2666 | if (cachep->ctor) | |
51cc5068 | 2667 | cachep->ctor(objp); |
1da177e4 | 2668 | #endif |
03787301 | 2669 | set_obj_status(page, i, OBJECT_FREE); |
e5c58dfd | 2670 | set_free_obj(page, i, i); |
1da177e4 | 2671 | } |
1da177e4 LT |
2672 | } |
2673 | ||
343e0d7a | 2674 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2675 | { |
4b51d669 CL |
2676 | if (CONFIG_ZONE_DMA_FLAG) { |
2677 | if (flags & GFP_DMA) | |
a618e89f | 2678 | BUG_ON(!(cachep->allocflags & GFP_DMA)); |
4b51d669 | 2679 | else |
a618e89f | 2680 | BUG_ON(cachep->allocflags & GFP_DMA); |
4b51d669 | 2681 | } |
1da177e4 LT |
2682 | } |
2683 | ||
8456a648 | 2684 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2685 | int nodeid) |
78d382d7 | 2686 | { |
b1cb0982 | 2687 | void *objp; |
78d382d7 | 2688 | |
e5c58dfd | 2689 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2690 | page->active++; |
78d382d7 | 2691 | #if DEBUG |
1ea991b0 | 2692 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2693 | #endif |
78d382d7 MD |
2694 | |
2695 | return objp; | |
2696 | } | |
2697 | ||
8456a648 | 2698 | static void slab_put_obj(struct kmem_cache *cachep, struct page *page, |
a737b3e2 | 2699 | void *objp, int nodeid) |
78d382d7 | 2700 | { |
8456a648 | 2701 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2702 | #if DEBUG |
16025177 | 2703 | unsigned int i; |
b1cb0982 | 2704 | |
78d382d7 | 2705 | /* Verify that the slab belongs to the intended node */ |
1ea991b0 | 2706 | WARN_ON(page_to_nid(virt_to_page(objp)) != nodeid); |
78d382d7 | 2707 | |
b1cb0982 | 2708 | /* Verify double free bug */ |
8456a648 | 2709 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2710 | if (get_free_obj(page, i) == objnr) { |
b1cb0982 JK |
2711 | printk(KERN_ERR "slab: double free detected in cache " |
2712 | "'%s', objp %p\n", cachep->name, objp); | |
2713 | BUG(); | |
2714 | } | |
78d382d7 MD |
2715 | } |
2716 | #endif | |
8456a648 | 2717 | page->active--; |
e5c58dfd | 2718 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2719 | } |
2720 | ||
4776874f PE |
2721 | /* |
2722 | * Map pages beginning at addr to the given cache and slab. This is required | |
2723 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2724 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2725 | */ |
8456a648 | 2726 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2727 | void *freelist) |
1da177e4 | 2728 | { |
a57a4988 | 2729 | page->slab_cache = cache; |
8456a648 | 2730 | page->freelist = freelist; |
1da177e4 LT |
2731 | } |
2732 | ||
2733 | /* | |
2734 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2735 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2736 | */ | |
3c517a61 | 2737 | static int cache_grow(struct kmem_cache *cachep, |
0c3aa83e | 2738 | gfp_t flags, int nodeid, struct page *page) |
1da177e4 | 2739 | { |
7e007355 | 2740 | void *freelist; |
b28a02de PE |
2741 | size_t offset; |
2742 | gfp_t local_flags; | |
ce8eb6c4 | 2743 | struct kmem_cache_node *n; |
1da177e4 | 2744 | |
a737b3e2 AM |
2745 | /* |
2746 | * Be lazy and only check for valid flags here, keeping it out of the | |
2747 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2748 | */ |
6cb06229 CL |
2749 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2750 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2751 | |
ce8eb6c4 | 2752 | /* Take the node list lock to change the colour_next on this node */ |
1da177e4 | 2753 | check_irq_off(); |
ce8eb6c4 CL |
2754 | n = cachep->node[nodeid]; |
2755 | spin_lock(&n->list_lock); | |
1da177e4 LT |
2756 | |
2757 | /* Get colour for the slab, and cal the next value. */ | |
ce8eb6c4 CL |
2758 | offset = n->colour_next; |
2759 | n->colour_next++; | |
2760 | if (n->colour_next >= cachep->colour) | |
2761 | n->colour_next = 0; | |
2762 | spin_unlock(&n->list_lock); | |
1da177e4 | 2763 | |
2e1217cf | 2764 | offset *= cachep->colour_off; |
1da177e4 LT |
2765 | |
2766 | if (local_flags & __GFP_WAIT) | |
2767 | local_irq_enable(); | |
2768 | ||
2769 | /* | |
2770 | * The test for missing atomic flag is performed here, rather than | |
2771 | * the more obvious place, simply to reduce the critical path length | |
2772 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2773 | * will eventually be caught here (where it matters). | |
2774 | */ | |
2775 | kmem_flagcheck(cachep, flags); | |
2776 | ||
a737b3e2 AM |
2777 | /* |
2778 | * Get mem for the objs. Attempt to allocate a physical page from | |
2779 | * 'nodeid'. | |
e498be7d | 2780 | */ |
0c3aa83e JK |
2781 | if (!page) |
2782 | page = kmem_getpages(cachep, local_flags, nodeid); | |
2783 | if (!page) | |
1da177e4 LT |
2784 | goto failed; |
2785 | ||
2786 | /* Get slab management. */ | |
8456a648 | 2787 | freelist = alloc_slabmgmt(cachep, page, offset, |
6cb06229 | 2788 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
8456a648 | 2789 | if (!freelist) |
1da177e4 LT |
2790 | goto opps1; |
2791 | ||
8456a648 | 2792 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2793 | |
8456a648 | 2794 | cache_init_objs(cachep, page); |
1da177e4 LT |
2795 | |
2796 | if (local_flags & __GFP_WAIT) | |
2797 | local_irq_disable(); | |
2798 | check_irq_off(); | |
ce8eb6c4 | 2799 | spin_lock(&n->list_lock); |
1da177e4 LT |
2800 | |
2801 | /* Make slab active. */ | |
8456a648 | 2802 | list_add_tail(&page->lru, &(n->slabs_free)); |
1da177e4 | 2803 | STATS_INC_GROWN(cachep); |
ce8eb6c4 CL |
2804 | n->free_objects += cachep->num; |
2805 | spin_unlock(&n->list_lock); | |
1da177e4 | 2806 | return 1; |
a737b3e2 | 2807 | opps1: |
0c3aa83e | 2808 | kmem_freepages(cachep, page); |
a737b3e2 | 2809 | failed: |
1da177e4 LT |
2810 | if (local_flags & __GFP_WAIT) |
2811 | local_irq_disable(); | |
2812 | return 0; | |
2813 | } | |
2814 | ||
2815 | #if DEBUG | |
2816 | ||
2817 | /* | |
2818 | * Perform extra freeing checks: | |
2819 | * - detect bad pointers. | |
2820 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2821 | */ |
2822 | static void kfree_debugcheck(const void *objp) | |
2823 | { | |
1da177e4 LT |
2824 | if (!virt_addr_valid(objp)) { |
2825 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2826 | (unsigned long)objp); |
2827 | BUG(); | |
1da177e4 | 2828 | } |
1da177e4 LT |
2829 | } |
2830 | ||
58ce1fd5 PE |
2831 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2832 | { | |
b46b8f19 | 2833 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2834 | |
2835 | redzone1 = *dbg_redzone1(cache, obj); | |
2836 | redzone2 = *dbg_redzone2(cache, obj); | |
2837 | ||
2838 | /* | |
2839 | * Redzone is ok. | |
2840 | */ | |
2841 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2842 | return; | |
2843 | ||
2844 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2845 | slab_error(cache, "double free detected"); | |
2846 | else | |
2847 | slab_error(cache, "memory outside object was overwritten"); | |
2848 | ||
b46b8f19 | 2849 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2850 | obj, redzone1, redzone2); |
2851 | } | |
2852 | ||
343e0d7a | 2853 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2854 | unsigned long caller) |
1da177e4 | 2855 | { |
1da177e4 | 2856 | unsigned int objnr; |
8456a648 | 2857 | struct page *page; |
1da177e4 | 2858 | |
80cbd911 MW |
2859 | BUG_ON(virt_to_cache(objp) != cachep); |
2860 | ||
3dafccf2 | 2861 | objp -= obj_offset(cachep); |
1da177e4 | 2862 | kfree_debugcheck(objp); |
b49af68f | 2863 | page = virt_to_head_page(objp); |
1da177e4 | 2864 | |
1da177e4 | 2865 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2866 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2867 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2868 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2869 | } | |
2870 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 2871 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 | 2872 | |
8456a648 | 2873 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2874 | |
2875 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2876 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2877 | |
03787301 | 2878 | set_obj_status(page, objnr, OBJECT_FREE); |
1da177e4 LT |
2879 | if (cachep->flags & SLAB_POISON) { |
2880 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3b0efdfa | 2881 | if ((cachep->size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
7c0cb9c6 | 2882 | store_stackinfo(cachep, objp, caller); |
b28a02de | 2883 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 2884 | cachep->size / PAGE_SIZE, 0); |
1da177e4 LT |
2885 | } else { |
2886 | poison_obj(cachep, objp, POISON_FREE); | |
2887 | } | |
2888 | #else | |
2889 | poison_obj(cachep, objp, POISON_FREE); | |
2890 | #endif | |
2891 | } | |
2892 | return objp; | |
2893 | } | |
2894 | ||
1da177e4 LT |
2895 | #else |
2896 | #define kfree_debugcheck(x) do { } while(0) | |
2897 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2898 | #endif |
2899 | ||
072bb0aa MG |
2900 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags, |
2901 | bool force_refill) | |
1da177e4 LT |
2902 | { |
2903 | int batchcount; | |
ce8eb6c4 | 2904 | struct kmem_cache_node *n; |
1da177e4 | 2905 | struct array_cache *ac; |
1ca4cb24 PE |
2906 | int node; |
2907 | ||
1da177e4 | 2908 | check_irq_off(); |
7d6e6d09 | 2909 | node = numa_mem_id(); |
072bb0aa MG |
2910 | if (unlikely(force_refill)) |
2911 | goto force_grow; | |
2912 | retry: | |
9a2dba4b | 2913 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2914 | batchcount = ac->batchcount; |
2915 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2916 | /* |
2917 | * If there was little recent activity on this cache, then | |
2918 | * perform only a partial refill. Otherwise we could generate | |
2919 | * refill bouncing. | |
1da177e4 LT |
2920 | */ |
2921 | batchcount = BATCHREFILL_LIMIT; | |
2922 | } | |
ce8eb6c4 | 2923 | n = cachep->node[node]; |
e498be7d | 2924 | |
ce8eb6c4 CL |
2925 | BUG_ON(ac->avail > 0 || !n); |
2926 | spin_lock(&n->list_lock); | |
1da177e4 | 2927 | |
3ded175a | 2928 | /* See if we can refill from the shared array */ |
ce8eb6c4 CL |
2929 | if (n->shared && transfer_objects(ac, n->shared, batchcount)) { |
2930 | n->shared->touched = 1; | |
3ded175a | 2931 | goto alloc_done; |
44b57f1c | 2932 | } |
3ded175a | 2933 | |
1da177e4 LT |
2934 | while (batchcount > 0) { |
2935 | struct list_head *entry; | |
8456a648 | 2936 | struct page *page; |
1da177e4 | 2937 | /* Get slab alloc is to come from. */ |
ce8eb6c4 CL |
2938 | entry = n->slabs_partial.next; |
2939 | if (entry == &n->slabs_partial) { | |
2940 | n->free_touched = 1; | |
2941 | entry = n->slabs_free.next; | |
2942 | if (entry == &n->slabs_free) | |
1da177e4 LT |
2943 | goto must_grow; |
2944 | } | |
2945 | ||
8456a648 | 2946 | page = list_entry(entry, struct page, lru); |
1da177e4 | 2947 | check_spinlock_acquired(cachep); |
714b8171 PE |
2948 | |
2949 | /* | |
2950 | * The slab was either on partial or free list so | |
2951 | * there must be at least one object available for | |
2952 | * allocation. | |
2953 | */ | |
8456a648 | 2954 | BUG_ON(page->active >= cachep->num); |
714b8171 | 2955 | |
8456a648 | 2956 | while (page->active < cachep->num && batchcount--) { |
1da177e4 LT |
2957 | STATS_INC_ALLOCED(cachep); |
2958 | STATS_INC_ACTIVE(cachep); | |
2959 | STATS_SET_HIGH(cachep); | |
2960 | ||
8456a648 | 2961 | ac_put_obj(cachep, ac, slab_get_obj(cachep, page, |
072bb0aa | 2962 | node)); |
1da177e4 | 2963 | } |
1da177e4 LT |
2964 | |
2965 | /* move slabp to correct slabp list: */ | |
8456a648 JK |
2966 | list_del(&page->lru); |
2967 | if (page->active == cachep->num) | |
34bf6ef9 | 2968 | list_add(&page->lru, &n->slabs_full); |
1da177e4 | 2969 | else |
34bf6ef9 | 2970 | list_add(&page->lru, &n->slabs_partial); |
1da177e4 LT |
2971 | } |
2972 | ||
a737b3e2 | 2973 | must_grow: |
ce8eb6c4 | 2974 | n->free_objects -= ac->avail; |
a737b3e2 | 2975 | alloc_done: |
ce8eb6c4 | 2976 | spin_unlock(&n->list_lock); |
1da177e4 LT |
2977 | |
2978 | if (unlikely(!ac->avail)) { | |
2979 | int x; | |
072bb0aa | 2980 | force_grow: |
3c517a61 | 2981 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 2982 | |
a737b3e2 | 2983 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2984 | ac = cpu_cache_get(cachep); |
51cd8e6f | 2985 | node = numa_mem_id(); |
072bb0aa MG |
2986 | |
2987 | /* no objects in sight? abort */ | |
2988 | if (!x && (ac->avail == 0 || force_refill)) | |
1da177e4 LT |
2989 | return NULL; |
2990 | ||
a737b3e2 | 2991 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2992 | goto retry; |
2993 | } | |
2994 | ac->touched = 1; | |
072bb0aa MG |
2995 | |
2996 | return ac_get_obj(cachep, ac, flags, force_refill); | |
1da177e4 LT |
2997 | } |
2998 | ||
a737b3e2 AM |
2999 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3000 | gfp_t flags) | |
1da177e4 LT |
3001 | { |
3002 | might_sleep_if(flags & __GFP_WAIT); | |
3003 | #if DEBUG | |
3004 | kmem_flagcheck(cachep, flags); | |
3005 | #endif | |
3006 | } | |
3007 | ||
3008 | #if DEBUG | |
a737b3e2 | 3009 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3010 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3011 | { |
03787301 JK |
3012 | struct page *page; |
3013 | ||
b28a02de | 3014 | if (!objp) |
1da177e4 | 3015 | return objp; |
b28a02de | 3016 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3017 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3b0efdfa | 3018 | if ((cachep->size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 3019 | kernel_map_pages(virt_to_page(objp), |
3b0efdfa | 3020 | cachep->size / PAGE_SIZE, 1); |
1da177e4 LT |
3021 | else |
3022 | check_poison_obj(cachep, objp); | |
3023 | #else | |
3024 | check_poison_obj(cachep, objp); | |
3025 | #endif | |
3026 | poison_obj(cachep, objp, POISON_INUSE); | |
3027 | } | |
3028 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3029 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3030 | |
3031 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3032 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3033 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
3034 | slab_error(cachep, "double free, or memory outside" | |
3035 | " object was overwritten"); | |
b28a02de | 3036 | printk(KERN_ERR |
b46b8f19 | 3037 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
3038 | objp, *dbg_redzone1(cachep, objp), |
3039 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3040 | } |
3041 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3042 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3043 | } | |
03787301 JK |
3044 | |
3045 | page = virt_to_head_page(objp); | |
3046 | set_obj_status(page, obj_to_index(cachep, page, objp), OBJECT_ACTIVE); | |
3dafccf2 | 3047 | objp += obj_offset(cachep); |
4f104934 | 3048 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3049 | cachep->ctor(objp); |
7ea466f2 TH |
3050 | if (ARCH_SLAB_MINALIGN && |
3051 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
a44b56d3 | 3052 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3053 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3054 | } |
1da177e4 LT |
3055 | return objp; |
3056 | } | |
3057 | #else | |
3058 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3059 | #endif | |
3060 | ||
773ff60e | 3061 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 | 3062 | { |
9b030cb8 | 3063 | if (cachep == kmem_cache) |
773ff60e | 3064 | return false; |
8a8b6502 | 3065 | |
8c138bc0 | 3066 | return should_failslab(cachep->object_size, flags, cachep->flags); |
8a8b6502 AM |
3067 | } |
3068 | ||
343e0d7a | 3069 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3070 | { |
b28a02de | 3071 | void *objp; |
1da177e4 | 3072 | struct array_cache *ac; |
072bb0aa | 3073 | bool force_refill = false; |
1da177e4 | 3074 | |
5c382300 | 3075 | check_irq_off(); |
8a8b6502 | 3076 | |
9a2dba4b | 3077 | ac = cpu_cache_get(cachep); |
1da177e4 | 3078 | if (likely(ac->avail)) { |
1da177e4 | 3079 | ac->touched = 1; |
072bb0aa MG |
3080 | objp = ac_get_obj(cachep, ac, flags, false); |
3081 | ||
ddbf2e83 | 3082 | /* |
072bb0aa MG |
3083 | * Allow for the possibility all avail objects are not allowed |
3084 | * by the current flags | |
ddbf2e83 | 3085 | */ |
072bb0aa MG |
3086 | if (objp) { |
3087 | STATS_INC_ALLOCHIT(cachep); | |
3088 | goto out; | |
3089 | } | |
3090 | force_refill = true; | |
1da177e4 | 3091 | } |
072bb0aa MG |
3092 | |
3093 | STATS_INC_ALLOCMISS(cachep); | |
3094 | objp = cache_alloc_refill(cachep, flags, force_refill); | |
3095 | /* | |
3096 | * the 'ac' may be updated by cache_alloc_refill(), | |
3097 | * and kmemleak_erase() requires its correct value. | |
3098 | */ | |
3099 | ac = cpu_cache_get(cachep); | |
3100 | ||
3101 | out: | |
d5cff635 CM |
3102 | /* |
3103 | * To avoid a false negative, if an object that is in one of the | |
3104 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3105 | * treat the array pointers as a reference to the object. | |
3106 | */ | |
f3d8b53a O |
3107 | if (objp) |
3108 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3109 | return objp; |
3110 | } | |
3111 | ||
e498be7d | 3112 | #ifdef CONFIG_NUMA |
c61afb18 | 3113 | /* |
f0432d15 | 3114 | * Try allocating on another node if PF_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3115 | * |
3116 | * If we are in_interrupt, then process context, including cpusets and | |
3117 | * mempolicy, may not apply and should not be used for allocation policy. | |
3118 | */ | |
3119 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3120 | { | |
3121 | int nid_alloc, nid_here; | |
3122 | ||
765c4507 | 3123 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3124 | return NULL; |
7d6e6d09 | 3125 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3126 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3127 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3128 | else if (current->mempolicy) |
2a389610 | 3129 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3130 | if (nid_alloc != nid_here) |
8b98c169 | 3131 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3132 | return NULL; |
3133 | } | |
3134 | ||
765c4507 CL |
3135 | /* |
3136 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3137 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3138 | * available node for available objects. If that fails then we |
3c517a61 CL |
3139 | * perform an allocation without specifying a node. This allows the page |
3140 | * allocator to do its reclaim / fallback magic. We then insert the | |
3141 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3142 | */ |
8c8cc2c1 | 3143 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3144 | { |
8c8cc2c1 PE |
3145 | struct zonelist *zonelist; |
3146 | gfp_t local_flags; | |
dd1a239f | 3147 | struct zoneref *z; |
54a6eb5c MG |
3148 | struct zone *zone; |
3149 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3150 | void *obj = NULL; |
3c517a61 | 3151 | int nid; |
cc9a6c87 | 3152 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3153 | |
3154 | if (flags & __GFP_THISNODE) | |
3155 | return NULL; | |
3156 | ||
6cb06229 | 3157 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3158 | |
cc9a6c87 | 3159 | retry_cpuset: |
d26914d1 | 3160 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3161 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3162 | |
3c517a61 CL |
3163 | retry: |
3164 | /* | |
3165 | * Look through allowed nodes for objects available | |
3166 | * from existing per node queues. | |
3167 | */ | |
54a6eb5c MG |
3168 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3169 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3170 | |
54a6eb5c | 3171 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
6a67368c CL |
3172 | cache->node[nid] && |
3173 | cache->node[nid]->free_objects) { | |
3c517a61 CL |
3174 | obj = ____cache_alloc_node(cache, |
3175 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3176 | if (obj) |
3177 | break; | |
3178 | } | |
3c517a61 CL |
3179 | } |
3180 | ||
cfce6604 | 3181 | if (!obj) { |
3c517a61 CL |
3182 | /* |
3183 | * This allocation will be performed within the constraints | |
3184 | * of the current cpuset / memory policy requirements. | |
3185 | * We may trigger various forms of reclaim on the allowed | |
3186 | * set and go into memory reserves if necessary. | |
3187 | */ | |
0c3aa83e JK |
3188 | struct page *page; |
3189 | ||
dd47ea75 CL |
3190 | if (local_flags & __GFP_WAIT) |
3191 | local_irq_enable(); | |
3192 | kmem_flagcheck(cache, flags); | |
0c3aa83e | 3193 | page = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3194 | if (local_flags & __GFP_WAIT) |
3195 | local_irq_disable(); | |
0c3aa83e | 3196 | if (page) { |
3c517a61 CL |
3197 | /* |
3198 | * Insert into the appropriate per node queues | |
3199 | */ | |
0c3aa83e JK |
3200 | nid = page_to_nid(page); |
3201 | if (cache_grow(cache, flags, nid, page)) { | |
3c517a61 CL |
3202 | obj = ____cache_alloc_node(cache, |
3203 | flags | GFP_THISNODE, nid); | |
3204 | if (!obj) | |
3205 | /* | |
3206 | * Another processor may allocate the | |
3207 | * objects in the slab since we are | |
3208 | * not holding any locks. | |
3209 | */ | |
3210 | goto retry; | |
3211 | } else { | |
b6a60451 | 3212 | /* cache_grow already freed obj */ |
3c517a61 CL |
3213 | obj = NULL; |
3214 | } | |
3215 | } | |
aedb0eb1 | 3216 | } |
cc9a6c87 | 3217 | |
d26914d1 | 3218 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3219 | goto retry_cpuset; |
765c4507 CL |
3220 | return obj; |
3221 | } | |
3222 | ||
e498be7d CL |
3223 | /* |
3224 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3225 | */ |
8b98c169 | 3226 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3227 | int nodeid) |
e498be7d CL |
3228 | { |
3229 | struct list_head *entry; | |
8456a648 | 3230 | struct page *page; |
ce8eb6c4 | 3231 | struct kmem_cache_node *n; |
b28a02de | 3232 | void *obj; |
b28a02de PE |
3233 | int x; |
3234 | ||
14e50c6a | 3235 | VM_BUG_ON(nodeid > num_online_nodes()); |
ce8eb6c4 CL |
3236 | n = cachep->node[nodeid]; |
3237 | BUG_ON(!n); | |
b28a02de | 3238 | |
a737b3e2 | 3239 | retry: |
ca3b9b91 | 3240 | check_irq_off(); |
ce8eb6c4 CL |
3241 | spin_lock(&n->list_lock); |
3242 | entry = n->slabs_partial.next; | |
3243 | if (entry == &n->slabs_partial) { | |
3244 | n->free_touched = 1; | |
3245 | entry = n->slabs_free.next; | |
3246 | if (entry == &n->slabs_free) | |
b28a02de PE |
3247 | goto must_grow; |
3248 | } | |
3249 | ||
8456a648 | 3250 | page = list_entry(entry, struct page, lru); |
b28a02de | 3251 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3252 | |
3253 | STATS_INC_NODEALLOCS(cachep); | |
3254 | STATS_INC_ACTIVE(cachep); | |
3255 | STATS_SET_HIGH(cachep); | |
3256 | ||
8456a648 | 3257 | BUG_ON(page->active == cachep->num); |
b28a02de | 3258 | |
8456a648 | 3259 | obj = slab_get_obj(cachep, page, nodeid); |
ce8eb6c4 | 3260 | n->free_objects--; |
b28a02de | 3261 | /* move slabp to correct slabp list: */ |
8456a648 | 3262 | list_del(&page->lru); |
b28a02de | 3263 | |
8456a648 JK |
3264 | if (page->active == cachep->num) |
3265 | list_add(&page->lru, &n->slabs_full); | |
a737b3e2 | 3266 | else |
8456a648 | 3267 | list_add(&page->lru, &n->slabs_partial); |
e498be7d | 3268 | |
ce8eb6c4 | 3269 | spin_unlock(&n->list_lock); |
b28a02de | 3270 | goto done; |
e498be7d | 3271 | |
a737b3e2 | 3272 | must_grow: |
ce8eb6c4 | 3273 | spin_unlock(&n->list_lock); |
3c517a61 | 3274 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3275 | if (x) |
3276 | goto retry; | |
1da177e4 | 3277 | |
8c8cc2c1 | 3278 | return fallback_alloc(cachep, flags); |
e498be7d | 3279 | |
a737b3e2 | 3280 | done: |
b28a02de | 3281 | return obj; |
e498be7d | 3282 | } |
8c8cc2c1 | 3283 | |
8c8cc2c1 | 3284 | static __always_inline void * |
48356303 | 3285 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3286 | unsigned long caller) |
8c8cc2c1 PE |
3287 | { |
3288 | unsigned long save_flags; | |
3289 | void *ptr; | |
7d6e6d09 | 3290 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3291 | |
dcce284a | 3292 | flags &= gfp_allowed_mask; |
7e85ee0c | 3293 | |
cf40bd16 NP |
3294 | lockdep_trace_alloc(flags); |
3295 | ||
773ff60e | 3296 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3297 | return NULL; |
3298 | ||
d79923fa GC |
3299 | cachep = memcg_kmem_get_cache(cachep, flags); |
3300 | ||
8c8cc2c1 PE |
3301 | cache_alloc_debugcheck_before(cachep, flags); |
3302 | local_irq_save(save_flags); | |
3303 | ||
eacbbae3 | 3304 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3305 | nodeid = slab_node; |
8c8cc2c1 | 3306 | |
6a67368c | 3307 | if (unlikely(!cachep->node[nodeid])) { |
8c8cc2c1 PE |
3308 | /* Node not bootstrapped yet */ |
3309 | ptr = fallback_alloc(cachep, flags); | |
3310 | goto out; | |
3311 | } | |
3312 | ||
7d6e6d09 | 3313 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3314 | /* |
3315 | * Use the locally cached objects if possible. | |
3316 | * However ____cache_alloc does not allow fallback | |
3317 | * to other nodes. It may fail while we still have | |
3318 | * objects on other nodes available. | |
3319 | */ | |
3320 | ptr = ____cache_alloc(cachep, flags); | |
3321 | if (ptr) | |
3322 | goto out; | |
3323 | } | |
3324 | /* ___cache_alloc_node can fall back to other nodes */ | |
3325 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3326 | out: | |
3327 | local_irq_restore(save_flags); | |
3328 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
8c138bc0 | 3329 | kmemleak_alloc_recursive(ptr, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3330 | flags); |
8c8cc2c1 | 3331 | |
5087c822 | 3332 | if (likely(ptr)) { |
8c138bc0 | 3333 | kmemcheck_slab_alloc(cachep, flags, ptr, cachep->object_size); |
5087c822 JP |
3334 | if (unlikely(flags & __GFP_ZERO)) |
3335 | memset(ptr, 0, cachep->object_size); | |
3336 | } | |
d07dbea4 | 3337 | |
8c8cc2c1 PE |
3338 | return ptr; |
3339 | } | |
3340 | ||
3341 | static __always_inline void * | |
3342 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3343 | { | |
3344 | void *objp; | |
3345 | ||
f0432d15 | 3346 | if (current->mempolicy || unlikely(current->flags & PF_SPREAD_SLAB)) { |
8c8cc2c1 PE |
3347 | objp = alternate_node_alloc(cache, flags); |
3348 | if (objp) | |
3349 | goto out; | |
3350 | } | |
3351 | objp = ____cache_alloc(cache, flags); | |
3352 | ||
3353 | /* | |
3354 | * We may just have run out of memory on the local node. | |
3355 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3356 | */ | |
7d6e6d09 LS |
3357 | if (!objp) |
3358 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3359 | |
3360 | out: | |
3361 | return objp; | |
3362 | } | |
3363 | #else | |
3364 | ||
3365 | static __always_inline void * | |
3366 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3367 | { | |
3368 | return ____cache_alloc(cachep, flags); | |
3369 | } | |
3370 | ||
3371 | #endif /* CONFIG_NUMA */ | |
3372 | ||
3373 | static __always_inline void * | |
48356303 | 3374 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3375 | { |
3376 | unsigned long save_flags; | |
3377 | void *objp; | |
3378 | ||
dcce284a | 3379 | flags &= gfp_allowed_mask; |
7e85ee0c | 3380 | |
cf40bd16 NP |
3381 | lockdep_trace_alloc(flags); |
3382 | ||
773ff60e | 3383 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3384 | return NULL; |
3385 | ||
d79923fa GC |
3386 | cachep = memcg_kmem_get_cache(cachep, flags); |
3387 | ||
8c8cc2c1 PE |
3388 | cache_alloc_debugcheck_before(cachep, flags); |
3389 | local_irq_save(save_flags); | |
3390 | objp = __do_cache_alloc(cachep, flags); | |
3391 | local_irq_restore(save_flags); | |
3392 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
8c138bc0 | 3393 | kmemleak_alloc_recursive(objp, cachep->object_size, 1, cachep->flags, |
d5cff635 | 3394 | flags); |
8c8cc2c1 PE |
3395 | prefetchw(objp); |
3396 | ||
5087c822 | 3397 | if (likely(objp)) { |
8c138bc0 | 3398 | kmemcheck_slab_alloc(cachep, flags, objp, cachep->object_size); |
5087c822 JP |
3399 | if (unlikely(flags & __GFP_ZERO)) |
3400 | memset(objp, 0, cachep->object_size); | |
3401 | } | |
d07dbea4 | 3402 | |
8c8cc2c1 PE |
3403 | return objp; |
3404 | } | |
e498be7d CL |
3405 | |
3406 | /* | |
5f0985bb | 3407 | * Caller needs to acquire correct kmem_cache_node's list_lock |
e498be7d | 3408 | */ |
343e0d7a | 3409 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 3410 | int node) |
1da177e4 LT |
3411 | { |
3412 | int i; | |
ce8eb6c4 | 3413 | struct kmem_cache_node *n; |
1da177e4 LT |
3414 | |
3415 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3416 | void *objp; |
8456a648 | 3417 | struct page *page; |
1da177e4 | 3418 | |
072bb0aa MG |
3419 | clear_obj_pfmemalloc(&objpp[i]); |
3420 | objp = objpp[i]; | |
3421 | ||
8456a648 | 3422 | page = virt_to_head_page(objp); |
ce8eb6c4 | 3423 | n = cachep->node[node]; |
8456a648 | 3424 | list_del(&page->lru); |
ff69416e | 3425 | check_spinlock_acquired_node(cachep, node); |
8456a648 | 3426 | slab_put_obj(cachep, page, objp, node); |
1da177e4 | 3427 | STATS_DEC_ACTIVE(cachep); |
ce8eb6c4 | 3428 | n->free_objects++; |
1da177e4 LT |
3429 | |
3430 | /* fixup slab chains */ | |
8456a648 | 3431 | if (page->active == 0) { |
ce8eb6c4 CL |
3432 | if (n->free_objects > n->free_limit) { |
3433 | n->free_objects -= cachep->num; | |
e5ac9c5a RT |
3434 | /* No need to drop any previously held |
3435 | * lock here, even if we have a off-slab slab | |
3436 | * descriptor it is guaranteed to come from | |
3437 | * a different cache, refer to comments before | |
3438 | * alloc_slabmgmt. | |
3439 | */ | |
8456a648 | 3440 | slab_destroy(cachep, page); |
1da177e4 | 3441 | } else { |
8456a648 | 3442 | list_add(&page->lru, &n->slabs_free); |
1da177e4 LT |
3443 | } |
3444 | } else { | |
3445 | /* Unconditionally move a slab to the end of the | |
3446 | * partial list on free - maximum time for the | |
3447 | * other objects to be freed, too. | |
3448 | */ | |
8456a648 | 3449 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3450 | } |
3451 | } | |
3452 | } | |
3453 | ||
343e0d7a | 3454 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3455 | { |
3456 | int batchcount; | |
ce8eb6c4 | 3457 | struct kmem_cache_node *n; |
7d6e6d09 | 3458 | int node = numa_mem_id(); |
1da177e4 LT |
3459 | |
3460 | batchcount = ac->batchcount; | |
3461 | #if DEBUG | |
3462 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3463 | #endif | |
3464 | check_irq_off(); | |
ce8eb6c4 CL |
3465 | n = cachep->node[node]; |
3466 | spin_lock(&n->list_lock); | |
3467 | if (n->shared) { | |
3468 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3469 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3470 | if (max) { |
3471 | if (batchcount > max) | |
3472 | batchcount = max; | |
e498be7d | 3473 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3474 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3475 | shared_array->avail += batchcount; |
3476 | goto free_done; | |
3477 | } | |
3478 | } | |
3479 | ||
ff69416e | 3480 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 3481 | free_done: |
1da177e4 LT |
3482 | #if STATS |
3483 | { | |
3484 | int i = 0; | |
3485 | struct list_head *p; | |
3486 | ||
ce8eb6c4 CL |
3487 | p = n->slabs_free.next; |
3488 | while (p != &(n->slabs_free)) { | |
8456a648 | 3489 | struct page *page; |
1da177e4 | 3490 | |
8456a648 JK |
3491 | page = list_entry(p, struct page, lru); |
3492 | BUG_ON(page->active); | |
1da177e4 LT |
3493 | |
3494 | i++; | |
3495 | p = p->next; | |
3496 | } | |
3497 | STATS_SET_FREEABLE(cachep, i); | |
3498 | } | |
3499 | #endif | |
ce8eb6c4 | 3500 | spin_unlock(&n->list_lock); |
1da177e4 | 3501 | ac->avail -= batchcount; |
a737b3e2 | 3502 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3503 | } |
3504 | ||
3505 | /* | |
a737b3e2 AM |
3506 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3507 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3508 | */ |
a947eb95 | 3509 | static inline void __cache_free(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 3510 | unsigned long caller) |
1da177e4 | 3511 | { |
9a2dba4b | 3512 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3513 | |
3514 | check_irq_off(); | |
d5cff635 | 3515 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3516 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3517 | |
8c138bc0 | 3518 | kmemcheck_slab_free(cachep, objp, cachep->object_size); |
c175eea4 | 3519 | |
1807a1aa SS |
3520 | /* |
3521 | * Skip calling cache_free_alien() when the platform is not numa. | |
3522 | * This will avoid cache misses that happen while accessing slabp (which | |
3523 | * is per page memory reference) to get nodeid. Instead use a global | |
3524 | * variable to skip the call, which is mostly likely to be present in | |
3525 | * the cache. | |
3526 | */ | |
b6e68bc1 | 3527 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3528 | return; |
3529 | ||
1da177e4 LT |
3530 | if (likely(ac->avail < ac->limit)) { |
3531 | STATS_INC_FREEHIT(cachep); | |
1da177e4 LT |
3532 | } else { |
3533 | STATS_INC_FREEMISS(cachep); | |
3534 | cache_flusharray(cachep, ac); | |
1da177e4 | 3535 | } |
42c8c99c | 3536 | |
072bb0aa | 3537 | ac_put_obj(cachep, ac, objp); |
1da177e4 LT |
3538 | } |
3539 | ||
3540 | /** | |
3541 | * kmem_cache_alloc - Allocate an object | |
3542 | * @cachep: The cache to allocate from. | |
3543 | * @flags: See kmalloc(). | |
3544 | * | |
3545 | * Allocate an object from this cache. The flags are only relevant | |
3546 | * if the cache has no available objects. | |
3547 | */ | |
343e0d7a | 3548 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3549 | { |
48356303 | 3550 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3551 | |
ca2b84cb | 3552 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3553 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3554 | |
3555 | return ret; | |
1da177e4 LT |
3556 | } |
3557 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3558 | ||
0f24f128 | 3559 | #ifdef CONFIG_TRACING |
85beb586 | 3560 | void * |
4052147c | 3561 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3562 | { |
85beb586 SR |
3563 | void *ret; |
3564 | ||
48356303 | 3565 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 SR |
3566 | |
3567 | trace_kmalloc(_RET_IP_, ret, | |
ff4fcd01 | 3568 | size, cachep->size, flags); |
85beb586 | 3569 | return ret; |
36555751 | 3570 | } |
85beb586 | 3571 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3572 | #endif |
3573 | ||
1da177e4 | 3574 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3575 | /** |
3576 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3577 | * @cachep: The cache to allocate from. | |
3578 | * @flags: See kmalloc(). | |
3579 | * @nodeid: node number of the target node. | |
3580 | * | |
3581 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3582 | * node, which can improve the performance for cpu bound structures. | |
3583 | * | |
3584 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3585 | */ | |
8b98c169 CH |
3586 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3587 | { | |
48356303 | 3588 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3589 | |
ca2b84cb | 3590 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3591 | cachep->object_size, cachep->size, |
ca2b84cb | 3592 | flags, nodeid); |
36555751 EGM |
3593 | |
3594 | return ret; | |
8b98c169 | 3595 | } |
1da177e4 LT |
3596 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3597 | ||
0f24f128 | 3598 | #ifdef CONFIG_TRACING |
4052147c | 3599 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3600 | gfp_t flags, |
4052147c EG |
3601 | int nodeid, |
3602 | size_t size) | |
36555751 | 3603 | { |
85beb586 SR |
3604 | void *ret; |
3605 | ||
592f4145 | 3606 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
7c0cb9c6 | 3607 | |
85beb586 | 3608 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3609 | size, cachep->size, |
85beb586 SR |
3610 | flags, nodeid); |
3611 | return ret; | |
36555751 | 3612 | } |
85beb586 | 3613 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3614 | #endif |
3615 | ||
8b98c169 | 3616 | static __always_inline void * |
7c0cb9c6 | 3617 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3618 | { |
343e0d7a | 3619 | struct kmem_cache *cachep; |
97e2bde4 | 3620 | |
2c59dd65 | 3621 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3622 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3623 | return cachep; | |
4052147c | 3624 | return kmem_cache_alloc_node_trace(cachep, flags, node, size); |
97e2bde4 | 3625 | } |
8b98c169 | 3626 | |
0bb38a5c | 3627 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3628 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3629 | { | |
7c0cb9c6 | 3630 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3631 | } |
dbe5e69d | 3632 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3633 | |
3634 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3635 | int node, unsigned long caller) |
8b98c169 | 3636 | { |
7c0cb9c6 | 3637 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3638 | } |
3639 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3640 | #else | |
3641 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3642 | { | |
7c0cb9c6 | 3643 | return __do_kmalloc_node(size, flags, node, 0); |
8b98c169 CH |
3644 | } |
3645 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3646 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3647 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3648 | |
3649 | /** | |
800590f5 | 3650 | * __do_kmalloc - allocate memory |
1da177e4 | 3651 | * @size: how many bytes of memory are required. |
800590f5 | 3652 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3653 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3654 | */ |
7fd6b141 | 3655 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3656 | unsigned long caller) |
1da177e4 | 3657 | { |
343e0d7a | 3658 | struct kmem_cache *cachep; |
36555751 | 3659 | void *ret; |
1da177e4 | 3660 | |
2c59dd65 | 3661 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3662 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3663 | return cachep; | |
48356303 | 3664 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3665 | |
7c0cb9c6 | 3666 | trace_kmalloc(caller, ret, |
3b0efdfa | 3667 | size, cachep->size, flags); |
36555751 EGM |
3668 | |
3669 | return ret; | |
7fd6b141 PE |
3670 | } |
3671 | ||
7fd6b141 | 3672 | |
0bb38a5c | 3673 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3674 | void *__kmalloc(size_t size, gfp_t flags) |
3675 | { | |
7c0cb9c6 | 3676 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3677 | } |
3678 | EXPORT_SYMBOL(__kmalloc); | |
3679 | ||
ce71e27c | 3680 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3681 | { |
7c0cb9c6 | 3682 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3683 | } |
3684 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3685 | |
3686 | #else | |
3687 | void *__kmalloc(size_t size, gfp_t flags) | |
3688 | { | |
7c0cb9c6 | 3689 | return __do_kmalloc(size, flags, 0); |
1d2c8eea CH |
3690 | } |
3691 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3692 | #endif |
3693 | ||
1da177e4 LT |
3694 | /** |
3695 | * kmem_cache_free - Deallocate an object | |
3696 | * @cachep: The cache the allocation was from. | |
3697 | * @objp: The previously allocated object. | |
3698 | * | |
3699 | * Free an object which was previously allocated from this | |
3700 | * cache. | |
3701 | */ | |
343e0d7a | 3702 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3703 | { |
3704 | unsigned long flags; | |
b9ce5ef4 GC |
3705 | cachep = cache_from_obj(cachep, objp); |
3706 | if (!cachep) | |
3707 | return; | |
1da177e4 LT |
3708 | |
3709 | local_irq_save(flags); | |
d97d476b | 3710 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3711 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3712 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3713 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3714 | local_irq_restore(flags); |
36555751 | 3715 | |
ca2b84cb | 3716 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3717 | } |
3718 | EXPORT_SYMBOL(kmem_cache_free); | |
3719 | ||
1da177e4 LT |
3720 | /** |
3721 | * kfree - free previously allocated memory | |
3722 | * @objp: pointer returned by kmalloc. | |
3723 | * | |
80e93eff PE |
3724 | * If @objp is NULL, no operation is performed. |
3725 | * | |
1da177e4 LT |
3726 | * Don't free memory not originally allocated by kmalloc() |
3727 | * or you will run into trouble. | |
3728 | */ | |
3729 | void kfree(const void *objp) | |
3730 | { | |
343e0d7a | 3731 | struct kmem_cache *c; |
1da177e4 LT |
3732 | unsigned long flags; |
3733 | ||
2121db74 PE |
3734 | trace_kfree(_RET_IP_, objp); |
3735 | ||
6cb8f913 | 3736 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3737 | return; |
3738 | local_irq_save(flags); | |
3739 | kfree_debugcheck(objp); | |
6ed5eb22 | 3740 | c = virt_to_cache(objp); |
8c138bc0 CL |
3741 | debug_check_no_locks_freed(objp, c->object_size); |
3742 | ||
3743 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3744 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3745 | local_irq_restore(flags); |
3746 | } | |
3747 | EXPORT_SYMBOL(kfree); | |
3748 | ||
e498be7d | 3749 | /* |
ce8eb6c4 | 3750 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3751 | */ |
5f0985bb | 3752 | static int alloc_kmem_cache_node(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3753 | { |
3754 | int node; | |
ce8eb6c4 | 3755 | struct kmem_cache_node *n; |
cafeb02e | 3756 | struct array_cache *new_shared; |
3395ee05 | 3757 | struct array_cache **new_alien = NULL; |
e498be7d | 3758 | |
9c09a95c | 3759 | for_each_online_node(node) { |
cafeb02e | 3760 | |
3395ee05 | 3761 | if (use_alien_caches) { |
83b519e8 | 3762 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3763 | if (!new_alien) |
3764 | goto fail; | |
3765 | } | |
cafeb02e | 3766 | |
63109846 ED |
3767 | new_shared = NULL; |
3768 | if (cachep->shared) { | |
3769 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3770 | cachep->shared*cachep->batchcount, |
83b519e8 | 3771 | 0xbaadf00d, gfp); |
63109846 ED |
3772 | if (!new_shared) { |
3773 | free_alien_cache(new_alien); | |
3774 | goto fail; | |
3775 | } | |
0718dc2a | 3776 | } |
cafeb02e | 3777 | |
ce8eb6c4 CL |
3778 | n = cachep->node[node]; |
3779 | if (n) { | |
3780 | struct array_cache *shared = n->shared; | |
cafeb02e | 3781 | |
ce8eb6c4 | 3782 | spin_lock_irq(&n->list_lock); |
e498be7d | 3783 | |
cafeb02e | 3784 | if (shared) |
0718dc2a CL |
3785 | free_block(cachep, shared->entry, |
3786 | shared->avail, node); | |
e498be7d | 3787 | |
ce8eb6c4 CL |
3788 | n->shared = new_shared; |
3789 | if (!n->alien) { | |
3790 | n->alien = new_alien; | |
e498be7d CL |
3791 | new_alien = NULL; |
3792 | } | |
ce8eb6c4 | 3793 | n->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3794 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3795 | spin_unlock_irq(&n->list_lock); |
cafeb02e | 3796 | kfree(shared); |
e498be7d CL |
3797 | free_alien_cache(new_alien); |
3798 | continue; | |
3799 | } | |
ce8eb6c4 CL |
3800 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); |
3801 | if (!n) { | |
0718dc2a CL |
3802 | free_alien_cache(new_alien); |
3803 | kfree(new_shared); | |
e498be7d | 3804 | goto fail; |
0718dc2a | 3805 | } |
e498be7d | 3806 | |
ce8eb6c4 | 3807 | kmem_cache_node_init(n); |
5f0985bb JZ |
3808 | n->next_reap = jiffies + REAPTIMEOUT_NODE + |
3809 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
ce8eb6c4 CL |
3810 | n->shared = new_shared; |
3811 | n->alien = new_alien; | |
3812 | n->free_limit = (1 + nr_cpus_node(node)) * | |
a737b3e2 | 3813 | cachep->batchcount + cachep->num; |
ce8eb6c4 | 3814 | cachep->node[node] = n; |
e498be7d | 3815 | } |
cafeb02e | 3816 | return 0; |
0718dc2a | 3817 | |
a737b3e2 | 3818 | fail: |
3b0efdfa | 3819 | if (!cachep->list.next) { |
0718dc2a CL |
3820 | /* Cache is not active yet. Roll back what we did */ |
3821 | node--; | |
3822 | while (node >= 0) { | |
6a67368c | 3823 | if (cachep->node[node]) { |
ce8eb6c4 | 3824 | n = cachep->node[node]; |
0718dc2a | 3825 | |
ce8eb6c4 CL |
3826 | kfree(n->shared); |
3827 | free_alien_cache(n->alien); | |
3828 | kfree(n); | |
6a67368c | 3829 | cachep->node[node] = NULL; |
0718dc2a CL |
3830 | } |
3831 | node--; | |
3832 | } | |
3833 | } | |
cafeb02e | 3834 | return -ENOMEM; |
e498be7d CL |
3835 | } |
3836 | ||
1da177e4 | 3837 | struct ccupdate_struct { |
343e0d7a | 3838 | struct kmem_cache *cachep; |
acfe7d74 | 3839 | struct array_cache *new[0]; |
1da177e4 LT |
3840 | }; |
3841 | ||
3842 | static void do_ccupdate_local(void *info) | |
3843 | { | |
a737b3e2 | 3844 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3845 | struct array_cache *old; |
3846 | ||
3847 | check_irq_off(); | |
9a2dba4b | 3848 | old = cpu_cache_get(new->cachep); |
e498be7d | 3849 | |
1da177e4 LT |
3850 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3851 | new->new[smp_processor_id()] = old; | |
3852 | } | |
3853 | ||
18004c5d | 3854 | /* Always called with the slab_mutex held */ |
943a451a | 3855 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3856 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3857 | { |
d2e7b7d0 | 3858 | struct ccupdate_struct *new; |
2ed3a4ef | 3859 | int i; |
1da177e4 | 3860 | |
acfe7d74 ED |
3861 | new = kzalloc(sizeof(*new) + nr_cpu_ids * sizeof(struct array_cache *), |
3862 | gfp); | |
d2e7b7d0 SS |
3863 | if (!new) |
3864 | return -ENOMEM; | |
3865 | ||
e498be7d | 3866 | for_each_online_cpu(i) { |
7d6e6d09 | 3867 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3868 | batchcount, gfp); |
d2e7b7d0 | 3869 | if (!new->new[i]) { |
b28a02de | 3870 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3871 | kfree(new->new[i]); |
3872 | kfree(new); | |
e498be7d | 3873 | return -ENOMEM; |
1da177e4 LT |
3874 | } |
3875 | } | |
d2e7b7d0 | 3876 | new->cachep = cachep; |
1da177e4 | 3877 | |
15c8b6c1 | 3878 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3879 | |
1da177e4 | 3880 | check_irq_on(); |
1da177e4 LT |
3881 | cachep->batchcount = batchcount; |
3882 | cachep->limit = limit; | |
e498be7d | 3883 | cachep->shared = shared; |
1da177e4 | 3884 | |
e498be7d | 3885 | for_each_online_cpu(i) { |
d2e7b7d0 | 3886 | struct array_cache *ccold = new->new[i]; |
1da177e4 LT |
3887 | if (!ccold) |
3888 | continue; | |
6a67368c | 3889 | spin_lock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
7d6e6d09 | 3890 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_mem(i)); |
6a67368c | 3891 | spin_unlock_irq(&cachep->node[cpu_to_mem(i)]->list_lock); |
1da177e4 LT |
3892 | kfree(ccold); |
3893 | } | |
d2e7b7d0 | 3894 | kfree(new); |
5f0985bb | 3895 | return alloc_kmem_cache_node(cachep, gfp); |
1da177e4 LT |
3896 | } |
3897 | ||
943a451a GC |
3898 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3899 | int batchcount, int shared, gfp_t gfp) | |
3900 | { | |
3901 | int ret; | |
3902 | struct kmem_cache *c = NULL; | |
3903 | int i = 0; | |
3904 | ||
3905 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3906 | ||
3907 | if (slab_state < FULL) | |
3908 | return ret; | |
3909 | ||
3910 | if ((ret < 0) || !is_root_cache(cachep)) | |
3911 | return ret; | |
3912 | ||
ebe945c2 | 3913 | VM_BUG_ON(!mutex_is_locked(&slab_mutex)); |
943a451a | 3914 | for_each_memcg_cache_index(i) { |
2ade4de8 | 3915 | c = cache_from_memcg_idx(cachep, i); |
943a451a GC |
3916 | if (c) |
3917 | /* return value determined by the parent cache only */ | |
3918 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
3919 | } | |
3920 | ||
3921 | return ret; | |
3922 | } | |
3923 | ||
18004c5d | 3924 | /* Called with slab_mutex held always */ |
83b519e8 | 3925 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3926 | { |
3927 | int err; | |
943a451a GC |
3928 | int limit = 0; |
3929 | int shared = 0; | |
3930 | int batchcount = 0; | |
3931 | ||
3932 | if (!is_root_cache(cachep)) { | |
3933 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3934 | limit = root->limit; | |
3935 | shared = root->shared; | |
3936 | batchcount = root->batchcount; | |
3937 | } | |
1da177e4 | 3938 | |
943a451a GC |
3939 | if (limit && shared && batchcount) |
3940 | goto skip_setup; | |
a737b3e2 AM |
3941 | /* |
3942 | * The head array serves three purposes: | |
1da177e4 LT |
3943 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3944 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3945 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3946 | * bufctl chains: array operations are cheaper. |
3947 | * The numbers are guessed, we should auto-tune as described by | |
3948 | * Bonwick. | |
3949 | */ | |
3b0efdfa | 3950 | if (cachep->size > 131072) |
1da177e4 | 3951 | limit = 1; |
3b0efdfa | 3952 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3953 | limit = 8; |
3b0efdfa | 3954 | else if (cachep->size > 1024) |
1da177e4 | 3955 | limit = 24; |
3b0efdfa | 3956 | else if (cachep->size > 256) |
1da177e4 LT |
3957 | limit = 54; |
3958 | else | |
3959 | limit = 120; | |
3960 | ||
a737b3e2 AM |
3961 | /* |
3962 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3963 | * allocation behaviour: Most allocs on one cpu, most free operations |
3964 | * on another cpu. For these cases, an efficient object passing between | |
3965 | * cpus is necessary. This is provided by a shared array. The array | |
3966 | * replaces Bonwick's magazine layer. | |
3967 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3968 | * to a larger limit. Thus disabled by default. | |
3969 | */ | |
3970 | shared = 0; | |
3b0efdfa | 3971 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3972 | shared = 8; |
1da177e4 LT |
3973 | |
3974 | #if DEBUG | |
a737b3e2 AM |
3975 | /* |
3976 | * With debugging enabled, large batchcount lead to excessively long | |
3977 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3978 | */ |
3979 | if (limit > 32) | |
3980 | limit = 32; | |
3981 | #endif | |
943a451a GC |
3982 | batchcount = (limit + 1) / 2; |
3983 | skip_setup: | |
3984 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
1da177e4 LT |
3985 | if (err) |
3986 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3987 | cachep->name, -err); |
2ed3a4ef | 3988 | return err; |
1da177e4 LT |
3989 | } |
3990 | ||
1b55253a | 3991 | /* |
ce8eb6c4 CL |
3992 | * Drain an array if it contains any elements taking the node lock only if |
3993 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 3994 | * if drain_array() is used on the shared array. |
1b55253a | 3995 | */ |
ce8eb6c4 | 3996 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
1b55253a | 3997 | struct array_cache *ac, int force, int node) |
1da177e4 LT |
3998 | { |
3999 | int tofree; | |
4000 | ||
1b55253a CL |
4001 | if (!ac || !ac->avail) |
4002 | return; | |
1da177e4 LT |
4003 | if (ac->touched && !force) { |
4004 | ac->touched = 0; | |
b18e7e65 | 4005 | } else { |
ce8eb6c4 | 4006 | spin_lock_irq(&n->list_lock); |
b18e7e65 CL |
4007 | if (ac->avail) { |
4008 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
4009 | if (tofree > ac->avail) | |
4010 | tofree = (ac->avail + 1) / 2; | |
4011 | free_block(cachep, ac->entry, tofree, node); | |
4012 | ac->avail -= tofree; | |
4013 | memmove(ac->entry, &(ac->entry[tofree]), | |
4014 | sizeof(void *) * ac->avail); | |
4015 | } | |
ce8eb6c4 | 4016 | spin_unlock_irq(&n->list_lock); |
1da177e4 LT |
4017 | } |
4018 | } | |
4019 | ||
4020 | /** | |
4021 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4022 | * @w: work descriptor |
1da177e4 LT |
4023 | * |
4024 | * Called from workqueue/eventd every few seconds. | |
4025 | * Purpose: | |
4026 | * - clear the per-cpu caches for this CPU. | |
4027 | * - return freeable pages to the main free memory pool. | |
4028 | * | |
a737b3e2 AM |
4029 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4030 | * again on the next iteration. | |
1da177e4 | 4031 | */ |
7c5cae36 | 4032 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4033 | { |
7a7c381d | 4034 | struct kmem_cache *searchp; |
ce8eb6c4 | 4035 | struct kmem_cache_node *n; |
7d6e6d09 | 4036 | int node = numa_mem_id(); |
bf6aede7 | 4037 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4038 | |
18004c5d | 4039 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4040 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4041 | goto out; |
1da177e4 | 4042 | |
18004c5d | 4043 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4044 | check_irq_on(); |
4045 | ||
35386e3b | 4046 | /* |
ce8eb6c4 | 4047 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4048 | * have established with reasonable certainty that |
4049 | * we can do some work if the lock was obtained. | |
4050 | */ | |
ce8eb6c4 | 4051 | n = searchp->node[node]; |
35386e3b | 4052 | |
ce8eb6c4 | 4053 | reap_alien(searchp, n); |
1da177e4 | 4054 | |
ce8eb6c4 | 4055 | drain_array(searchp, n, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4056 | |
35386e3b CL |
4057 | /* |
4058 | * These are racy checks but it does not matter | |
4059 | * if we skip one check or scan twice. | |
4060 | */ | |
ce8eb6c4 | 4061 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4062 | goto next; |
1da177e4 | 4063 | |
5f0985bb | 4064 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4065 | |
ce8eb6c4 | 4066 | drain_array(searchp, n, n->shared, 0, node); |
1da177e4 | 4067 | |
ce8eb6c4 CL |
4068 | if (n->free_touched) |
4069 | n->free_touched = 0; | |
ed11d9eb CL |
4070 | else { |
4071 | int freed; | |
1da177e4 | 4072 | |
ce8eb6c4 | 4073 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4074 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4075 | STATS_ADD_REAPED(searchp, freed); | |
4076 | } | |
35386e3b | 4077 | next: |
1da177e4 LT |
4078 | cond_resched(); |
4079 | } | |
4080 | check_irq_on(); | |
18004c5d | 4081 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4082 | next_reap_node(); |
7c5cae36 | 4083 | out: |
a737b3e2 | 4084 | /* Set up the next iteration */ |
5f0985bb | 4085 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_AC)); |
1da177e4 LT |
4086 | } |
4087 | ||
158a9624 | 4088 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 4089 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4090 | { |
8456a648 | 4091 | struct page *page; |
b28a02de PE |
4092 | unsigned long active_objs; |
4093 | unsigned long num_objs; | |
4094 | unsigned long active_slabs = 0; | |
4095 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4096 | const char *name; |
1da177e4 | 4097 | char *error = NULL; |
e498be7d | 4098 | int node; |
ce8eb6c4 | 4099 | struct kmem_cache_node *n; |
1da177e4 | 4100 | |
1da177e4 LT |
4101 | active_objs = 0; |
4102 | num_slabs = 0; | |
e498be7d | 4103 | for_each_online_node(node) { |
ce8eb6c4 CL |
4104 | n = cachep->node[node]; |
4105 | if (!n) | |
e498be7d CL |
4106 | continue; |
4107 | ||
ca3b9b91 | 4108 | check_irq_on(); |
ce8eb6c4 | 4109 | spin_lock_irq(&n->list_lock); |
e498be7d | 4110 | |
8456a648 JK |
4111 | list_for_each_entry(page, &n->slabs_full, lru) { |
4112 | if (page->active != cachep->num && !error) | |
e498be7d CL |
4113 | error = "slabs_full accounting error"; |
4114 | active_objs += cachep->num; | |
4115 | active_slabs++; | |
4116 | } | |
8456a648 JK |
4117 | list_for_each_entry(page, &n->slabs_partial, lru) { |
4118 | if (page->active == cachep->num && !error) | |
106a74e1 | 4119 | error = "slabs_partial accounting error"; |
8456a648 | 4120 | if (!page->active && !error) |
106a74e1 | 4121 | error = "slabs_partial accounting error"; |
8456a648 | 4122 | active_objs += page->active; |
e498be7d CL |
4123 | active_slabs++; |
4124 | } | |
8456a648 JK |
4125 | list_for_each_entry(page, &n->slabs_free, lru) { |
4126 | if (page->active && !error) | |
106a74e1 | 4127 | error = "slabs_free accounting error"; |
e498be7d CL |
4128 | num_slabs++; |
4129 | } | |
ce8eb6c4 CL |
4130 | free_objects += n->free_objects; |
4131 | if (n->shared) | |
4132 | shared_avail += n->shared->avail; | |
e498be7d | 4133 | |
ce8eb6c4 | 4134 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4135 | } |
b28a02de PE |
4136 | num_slabs += active_slabs; |
4137 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4138 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4139 | error = "free_objects accounting error"; |
4140 | ||
b28a02de | 4141 | name = cachep->name; |
1da177e4 LT |
4142 | if (error) |
4143 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4144 | ||
0d7561c6 GC |
4145 | sinfo->active_objs = active_objs; |
4146 | sinfo->num_objs = num_objs; | |
4147 | sinfo->active_slabs = active_slabs; | |
4148 | sinfo->num_slabs = num_slabs; | |
4149 | sinfo->shared_avail = shared_avail; | |
4150 | sinfo->limit = cachep->limit; | |
4151 | sinfo->batchcount = cachep->batchcount; | |
4152 | sinfo->shared = cachep->shared; | |
4153 | sinfo->objects_per_slab = cachep->num; | |
4154 | sinfo->cache_order = cachep->gfporder; | |
4155 | } | |
4156 | ||
4157 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4158 | { | |
1da177e4 | 4159 | #if STATS |
ce8eb6c4 | 4160 | { /* node stats */ |
1da177e4 LT |
4161 | unsigned long high = cachep->high_mark; |
4162 | unsigned long allocs = cachep->num_allocations; | |
4163 | unsigned long grown = cachep->grown; | |
4164 | unsigned long reaped = cachep->reaped; | |
4165 | unsigned long errors = cachep->errors; | |
4166 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4167 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4168 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4169 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4170 | |
e92dd4fd JP |
4171 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4172 | "%4lu %4lu %4lu %4lu %4lu", | |
4173 | allocs, high, grown, | |
4174 | reaped, errors, max_freeable, node_allocs, | |
4175 | node_frees, overflows); | |
1da177e4 LT |
4176 | } |
4177 | /* cpu stats */ | |
4178 | { | |
4179 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4180 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4181 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4182 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4183 | ||
4184 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4185 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4186 | } |
4187 | #endif | |
1da177e4 LT |
4188 | } |
4189 | ||
1da177e4 LT |
4190 | #define MAX_SLABINFO_WRITE 128 |
4191 | /** | |
4192 | * slabinfo_write - Tuning for the slab allocator | |
4193 | * @file: unused | |
4194 | * @buffer: user buffer | |
4195 | * @count: data length | |
4196 | * @ppos: unused | |
4197 | */ | |
b7454ad3 | 4198 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4199 | size_t count, loff_t *ppos) |
1da177e4 | 4200 | { |
b28a02de | 4201 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4202 | int limit, batchcount, shared, res; |
7a7c381d | 4203 | struct kmem_cache *cachep; |
b28a02de | 4204 | |
1da177e4 LT |
4205 | if (count > MAX_SLABINFO_WRITE) |
4206 | return -EINVAL; | |
4207 | if (copy_from_user(&kbuf, buffer, count)) | |
4208 | return -EFAULT; | |
b28a02de | 4209 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4210 | |
4211 | tmp = strchr(kbuf, ' '); | |
4212 | if (!tmp) | |
4213 | return -EINVAL; | |
4214 | *tmp = '\0'; | |
4215 | tmp++; | |
4216 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4217 | return -EINVAL; | |
4218 | ||
4219 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4220 | mutex_lock(&slab_mutex); |
1da177e4 | 4221 | res = -EINVAL; |
18004c5d | 4222 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4223 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4224 | if (limit < 1 || batchcount < 1 || |
4225 | batchcount > limit || shared < 0) { | |
e498be7d | 4226 | res = 0; |
1da177e4 | 4227 | } else { |
e498be7d | 4228 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4229 | batchcount, shared, |
4230 | GFP_KERNEL); | |
1da177e4 LT |
4231 | } |
4232 | break; | |
4233 | } | |
4234 | } | |
18004c5d | 4235 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4236 | if (res >= 0) |
4237 | res = count; | |
4238 | return res; | |
4239 | } | |
871751e2 AV |
4240 | |
4241 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4242 | ||
4243 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4244 | { | |
18004c5d CL |
4245 | mutex_lock(&slab_mutex); |
4246 | return seq_list_start(&slab_caches, *pos); | |
871751e2 AV |
4247 | } |
4248 | ||
4249 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4250 | { | |
4251 | unsigned long *p; | |
4252 | int l; | |
4253 | if (!v) | |
4254 | return 1; | |
4255 | l = n[1]; | |
4256 | p = n + 2; | |
4257 | while (l) { | |
4258 | int i = l/2; | |
4259 | unsigned long *q = p + 2 * i; | |
4260 | if (*q == v) { | |
4261 | q[1]++; | |
4262 | return 1; | |
4263 | } | |
4264 | if (*q > v) { | |
4265 | l = i; | |
4266 | } else { | |
4267 | p = q + 2; | |
4268 | l -= i + 1; | |
4269 | } | |
4270 | } | |
4271 | if (++n[1] == n[0]) | |
4272 | return 0; | |
4273 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4274 | p[0] = v; | |
4275 | p[1] = 1; | |
4276 | return 1; | |
4277 | } | |
4278 | ||
8456a648 JK |
4279 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4280 | struct page *page) | |
871751e2 AV |
4281 | { |
4282 | void *p; | |
03787301 | 4283 | int i; |
b1cb0982 | 4284 | |
871751e2 AV |
4285 | if (n[0] == n[1]) |
4286 | return; | |
8456a648 | 4287 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
03787301 | 4288 | if (get_obj_status(page, i) != OBJECT_ACTIVE) |
871751e2 | 4289 | continue; |
b1cb0982 | 4290 | |
871751e2 AV |
4291 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) |
4292 | return; | |
4293 | } | |
4294 | } | |
4295 | ||
4296 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4297 | { | |
4298 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4299 | unsigned long offset, size; |
9281acea | 4300 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4301 | |
a5c43dae | 4302 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4303 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4304 | if (modname[0]) |
871751e2 AV |
4305 | seq_printf(m, " [%s]", modname); |
4306 | return; | |
4307 | } | |
4308 | #endif | |
4309 | seq_printf(m, "%p", (void *)address); | |
4310 | } | |
4311 | ||
4312 | static int leaks_show(struct seq_file *m, void *p) | |
4313 | { | |
0672aa7c | 4314 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4315 | struct page *page; |
ce8eb6c4 | 4316 | struct kmem_cache_node *n; |
871751e2 | 4317 | const char *name; |
db845067 | 4318 | unsigned long *x = m->private; |
871751e2 AV |
4319 | int node; |
4320 | int i; | |
4321 | ||
4322 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4323 | return 0; | |
4324 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4325 | return 0; | |
4326 | ||
4327 | /* OK, we can do it */ | |
4328 | ||
db845067 | 4329 | x[1] = 0; |
871751e2 AV |
4330 | |
4331 | for_each_online_node(node) { | |
ce8eb6c4 CL |
4332 | n = cachep->node[node]; |
4333 | if (!n) | |
871751e2 AV |
4334 | continue; |
4335 | ||
4336 | check_irq_on(); | |
ce8eb6c4 | 4337 | spin_lock_irq(&n->list_lock); |
871751e2 | 4338 | |
8456a648 JK |
4339 | list_for_each_entry(page, &n->slabs_full, lru) |
4340 | handle_slab(x, cachep, page); | |
4341 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4342 | handle_slab(x, cachep, page); | |
ce8eb6c4 | 4343 | spin_unlock_irq(&n->list_lock); |
871751e2 AV |
4344 | } |
4345 | name = cachep->name; | |
db845067 | 4346 | if (x[0] == x[1]) { |
871751e2 | 4347 | /* Increase the buffer size */ |
18004c5d | 4348 | mutex_unlock(&slab_mutex); |
db845067 | 4349 | m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL); |
871751e2 AV |
4350 | if (!m->private) { |
4351 | /* Too bad, we are really out */ | |
db845067 | 4352 | m->private = x; |
18004c5d | 4353 | mutex_lock(&slab_mutex); |
871751e2 AV |
4354 | return -ENOMEM; |
4355 | } | |
db845067 CL |
4356 | *(unsigned long *)m->private = x[0] * 2; |
4357 | kfree(x); | |
18004c5d | 4358 | mutex_lock(&slab_mutex); |
871751e2 AV |
4359 | /* Now make sure this entry will be retried */ |
4360 | m->count = m->size; | |
4361 | return 0; | |
4362 | } | |
db845067 CL |
4363 | for (i = 0; i < x[1]; i++) { |
4364 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4365 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4366 | seq_putc(m, '\n'); |
4367 | } | |
d2e7b7d0 | 4368 | |
871751e2 AV |
4369 | return 0; |
4370 | } | |
4371 | ||
a0ec95a8 | 4372 | static const struct seq_operations slabstats_op = { |
871751e2 | 4373 | .start = leaks_start, |
276a2439 WL |
4374 | .next = slab_next, |
4375 | .stop = slab_stop, | |
871751e2 AV |
4376 | .show = leaks_show, |
4377 | }; | |
a0ec95a8 AD |
4378 | |
4379 | static int slabstats_open(struct inode *inode, struct file *file) | |
4380 | { | |
4381 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4382 | int ret = -ENOMEM; | |
4383 | if (n) { | |
4384 | ret = seq_open(file, &slabstats_op); | |
4385 | if (!ret) { | |
4386 | struct seq_file *m = file->private_data; | |
4387 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4388 | m->private = n; | |
4389 | n = NULL; | |
4390 | } | |
4391 | kfree(n); | |
4392 | } | |
4393 | return ret; | |
4394 | } | |
4395 | ||
4396 | static const struct file_operations proc_slabstats_operations = { | |
4397 | .open = slabstats_open, | |
4398 | .read = seq_read, | |
4399 | .llseek = seq_lseek, | |
4400 | .release = seq_release_private, | |
4401 | }; | |
4402 | #endif | |
4403 | ||
4404 | static int __init slab_proc_init(void) | |
4405 | { | |
4406 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4407 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4408 | #endif |
a0ec95a8 AD |
4409 | return 0; |
4410 | } | |
4411 | module_init(slab_proc_init); | |
1da177e4 LT |
4412 | #endif |
4413 | ||
00e145b6 MS |
4414 | /** |
4415 | * ksize - get the actual amount of memory allocated for a given object | |
4416 | * @objp: Pointer to the object | |
4417 | * | |
4418 | * kmalloc may internally round up allocations and return more memory | |
4419 | * than requested. ksize() can be used to determine the actual amount of | |
4420 | * memory allocated. The caller may use this additional memory, even though | |
4421 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4422 | * The caller must guarantee that objp points to a valid object previously | |
4423 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4424 | * must not be freed during the duration of the call. | |
4425 | */ | |
fd76bab2 | 4426 | size_t ksize(const void *objp) |
1da177e4 | 4427 | { |
ef8b4520 CL |
4428 | BUG_ON(!objp); |
4429 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4430 | return 0; |
1da177e4 | 4431 | |
8c138bc0 | 4432 | return virt_to_cache(objp)->object_size; |
1da177e4 | 4433 | } |
b1aabecd | 4434 | EXPORT_SYMBOL(ksize); |