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 | |
29 | * slabs and you must pass objects with the same intializations to | |
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 | |
fc0abb14 | 71 | * The global cache-chain is protected by the mutex 'cache_chain_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 | ||
89 | #include <linux/config.h> | |
90 | #include <linux/slab.h> | |
91 | #include <linux/mm.h> | |
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> |
1da177e4 LT |
98 | #include <linux/seq_file.h> |
99 | #include <linux/notifier.h> | |
100 | #include <linux/kallsyms.h> | |
101 | #include <linux/cpu.h> | |
102 | #include <linux/sysctl.h> | |
103 | #include <linux/module.h> | |
104 | #include <linux/rcupdate.h> | |
543537bd | 105 | #include <linux/string.h> |
e498be7d | 106 | #include <linux/nodemask.h> |
dc85da15 | 107 | #include <linux/mempolicy.h> |
fc0abb14 | 108 | #include <linux/mutex.h> |
1da177e4 LT |
109 | |
110 | #include <asm/uaccess.h> | |
111 | #include <asm/cacheflush.h> | |
112 | #include <asm/tlbflush.h> | |
113 | #include <asm/page.h> | |
114 | ||
115 | /* | |
116 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_DEBUG_INITIAL, | |
117 | * SLAB_RED_ZONE & SLAB_POISON. | |
118 | * 0 for faster, smaller code (especially in the critical paths). | |
119 | * | |
120 | * STATS - 1 to collect stats for /proc/slabinfo. | |
121 | * 0 for faster, smaller code (especially in the critical paths). | |
122 | * | |
123 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
124 | */ | |
125 | ||
126 | #ifdef CONFIG_DEBUG_SLAB | |
127 | #define DEBUG 1 | |
128 | #define STATS 1 | |
129 | #define FORCED_DEBUG 1 | |
130 | #else | |
131 | #define DEBUG 0 | |
132 | #define STATS 0 | |
133 | #define FORCED_DEBUG 0 | |
134 | #endif | |
135 | ||
1da177e4 LT |
136 | /* Shouldn't this be in a header file somewhere? */ |
137 | #define BYTES_PER_WORD sizeof(void *) | |
138 | ||
139 | #ifndef cache_line_size | |
140 | #define cache_line_size() L1_CACHE_BYTES | |
141 | #endif | |
142 | ||
143 | #ifndef ARCH_KMALLOC_MINALIGN | |
144 | /* | |
145 | * Enforce a minimum alignment for the kmalloc caches. | |
146 | * Usually, the kmalloc caches are cache_line_size() aligned, except when | |
147 | * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned. | |
148 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed | |
149 | * alignment larger than BYTES_PER_WORD. ARCH_KMALLOC_MINALIGN allows that. | |
150 | * Note that this flag disables some debug features. | |
151 | */ | |
152 | #define ARCH_KMALLOC_MINALIGN 0 | |
153 | #endif | |
154 | ||
155 | #ifndef ARCH_SLAB_MINALIGN | |
156 | /* | |
157 | * Enforce a minimum alignment for all caches. | |
158 | * Intended for archs that get misalignment faults even for BYTES_PER_WORD | |
159 | * aligned buffers. Includes ARCH_KMALLOC_MINALIGN. | |
160 | * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables | |
161 | * some debug features. | |
162 | */ | |
163 | #define ARCH_SLAB_MINALIGN 0 | |
164 | #endif | |
165 | ||
166 | #ifndef ARCH_KMALLOC_FLAGS | |
167 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
168 | #endif | |
169 | ||
170 | /* Legal flag mask for kmem_cache_create(). */ | |
171 | #if DEBUG | |
172 | # define CREATE_MASK (SLAB_DEBUG_INITIAL | SLAB_RED_ZONE | \ | |
173 | SLAB_POISON | SLAB_HWCACHE_ALIGN | \ | |
ac2b898c | 174 | SLAB_CACHE_DMA | \ |
1da177e4 LT |
175 | SLAB_MUST_HWCACHE_ALIGN | SLAB_STORE_USER | \ |
176 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ | |
101a5001 | 177 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD) |
1da177e4 | 178 | #else |
ac2b898c | 179 | # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \ |
1da177e4 LT |
180 | SLAB_CACHE_DMA | SLAB_MUST_HWCACHE_ALIGN | \ |
181 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ | |
101a5001 | 182 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD) |
1da177e4 LT |
183 | #endif |
184 | ||
185 | /* | |
186 | * kmem_bufctl_t: | |
187 | * | |
188 | * Bufctl's are used for linking objs within a slab | |
189 | * linked offsets. | |
190 | * | |
191 | * This implementation relies on "struct page" for locating the cache & | |
192 | * slab an object belongs to. | |
193 | * This allows the bufctl structure to be small (one int), but limits | |
194 | * the number of objects a slab (not a cache) can contain when off-slab | |
195 | * bufctls are used. The limit is the size of the largest general cache | |
196 | * that does not use off-slab slabs. | |
197 | * For 32bit archs with 4 kB pages, is this 56. | |
198 | * This is not serious, as it is only for large objects, when it is unwise | |
199 | * to have too many per slab. | |
200 | * Note: This limit can be raised by introducing a general cache whose size | |
201 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. | |
202 | */ | |
203 | ||
fa5b08d5 | 204 | typedef unsigned int kmem_bufctl_t; |
1da177e4 LT |
205 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
206 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) | |
207 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-2) | |
208 | ||
209 | /* Max number of objs-per-slab for caches which use off-slab slabs. | |
210 | * Needed to avoid a possible looping condition in cache_grow(). | |
211 | */ | |
212 | static unsigned long offslab_limit; | |
213 | ||
214 | /* | |
215 | * struct slab | |
216 | * | |
217 | * Manages the objs in a slab. Placed either at the beginning of mem allocated | |
218 | * for a slab, or allocated from an general cache. | |
219 | * Slabs are chained into three list: fully used, partial, fully free slabs. | |
220 | */ | |
221 | struct slab { | |
b28a02de PE |
222 | struct list_head list; |
223 | unsigned long colouroff; | |
224 | void *s_mem; /* including colour offset */ | |
225 | unsigned int inuse; /* num of objs active in slab */ | |
226 | kmem_bufctl_t free; | |
227 | unsigned short nodeid; | |
1da177e4 LT |
228 | }; |
229 | ||
230 | /* | |
231 | * struct slab_rcu | |
232 | * | |
233 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to | |
234 | * arrange for kmem_freepages to be called via RCU. This is useful if | |
235 | * we need to approach a kernel structure obliquely, from its address | |
236 | * obtained without the usual locking. We can lock the structure to | |
237 | * stabilize it and check it's still at the given address, only if we | |
238 | * can be sure that the memory has not been meanwhile reused for some | |
239 | * other kind of object (which our subsystem's lock might corrupt). | |
240 | * | |
241 | * rcu_read_lock before reading the address, then rcu_read_unlock after | |
242 | * taking the spinlock within the structure expected at that address. | |
243 | * | |
244 | * We assume struct slab_rcu can overlay struct slab when destroying. | |
245 | */ | |
246 | struct slab_rcu { | |
b28a02de | 247 | struct rcu_head head; |
343e0d7a | 248 | struct kmem_cache *cachep; |
b28a02de | 249 | void *addr; |
1da177e4 LT |
250 | }; |
251 | ||
252 | /* | |
253 | * struct array_cache | |
254 | * | |
1da177e4 LT |
255 | * Purpose: |
256 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
257 | * - reduce the number of linked list operations | |
258 | * - reduce spinlock operations | |
259 | * | |
260 | * The limit is stored in the per-cpu structure to reduce the data cache | |
261 | * footprint. | |
262 | * | |
263 | */ | |
264 | struct array_cache { | |
265 | unsigned int avail; | |
266 | unsigned int limit; | |
267 | unsigned int batchcount; | |
268 | unsigned int touched; | |
e498be7d | 269 | spinlock_t lock; |
a737b3e2 AM |
270 | void *entry[0]; /* |
271 | * Must have this definition in here for the proper | |
272 | * alignment of array_cache. Also simplifies accessing | |
273 | * the entries. | |
274 | * [0] is for gcc 2.95. It should really be []. | |
275 | */ | |
1da177e4 LT |
276 | }; |
277 | ||
a737b3e2 AM |
278 | /* |
279 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
280 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
281 | */ |
282 | #define BOOT_CPUCACHE_ENTRIES 1 | |
283 | struct arraycache_init { | |
284 | struct array_cache cache; | |
b28a02de | 285 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
286 | }; |
287 | ||
288 | /* | |
e498be7d | 289 | * The slab lists for all objects. |
1da177e4 LT |
290 | */ |
291 | struct kmem_list3 { | |
b28a02de PE |
292 | struct list_head slabs_partial; /* partial list first, better asm code */ |
293 | struct list_head slabs_full; | |
294 | struct list_head slabs_free; | |
295 | unsigned long free_objects; | |
b28a02de | 296 | unsigned int free_limit; |
2e1217cf | 297 | unsigned int colour_next; /* Per-node cache coloring */ |
b28a02de PE |
298 | spinlock_t list_lock; |
299 | struct array_cache *shared; /* shared per node */ | |
300 | struct array_cache **alien; /* on other nodes */ | |
35386e3b CL |
301 | unsigned long next_reap; /* updated without locking */ |
302 | int free_touched; /* updated without locking */ | |
1da177e4 LT |
303 | }; |
304 | ||
e498be7d CL |
305 | /* |
306 | * Need this for bootstrapping a per node allocator. | |
307 | */ | |
308 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES + 1) | |
309 | struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; | |
310 | #define CACHE_CACHE 0 | |
311 | #define SIZE_AC 1 | |
312 | #define SIZE_L3 (1 + MAX_NUMNODES) | |
313 | ||
314 | /* | |
a737b3e2 AM |
315 | * This function must be completely optimized away if a constant is passed to |
316 | * it. Mostly the same as what is in linux/slab.h except it returns an index. | |
e498be7d | 317 | */ |
7243cc05 | 318 | static __always_inline int index_of(const size_t size) |
e498be7d | 319 | { |
5ec8a847 SR |
320 | extern void __bad_size(void); |
321 | ||
e498be7d CL |
322 | if (__builtin_constant_p(size)) { |
323 | int i = 0; | |
324 | ||
325 | #define CACHE(x) \ | |
326 | if (size <=x) \ | |
327 | return i; \ | |
328 | else \ | |
329 | i++; | |
330 | #include "linux/kmalloc_sizes.h" | |
331 | #undef CACHE | |
5ec8a847 | 332 | __bad_size(); |
7243cc05 | 333 | } else |
5ec8a847 | 334 | __bad_size(); |
e498be7d CL |
335 | return 0; |
336 | } | |
337 | ||
338 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) | |
339 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) | |
1da177e4 | 340 | |
5295a74c | 341 | static void kmem_list3_init(struct kmem_list3 *parent) |
e498be7d CL |
342 | { |
343 | INIT_LIST_HEAD(&parent->slabs_full); | |
344 | INIT_LIST_HEAD(&parent->slabs_partial); | |
345 | INIT_LIST_HEAD(&parent->slabs_free); | |
346 | parent->shared = NULL; | |
347 | parent->alien = NULL; | |
2e1217cf | 348 | parent->colour_next = 0; |
e498be7d CL |
349 | spin_lock_init(&parent->list_lock); |
350 | parent->free_objects = 0; | |
351 | parent->free_touched = 0; | |
352 | } | |
353 | ||
a737b3e2 AM |
354 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
355 | do { \ | |
356 | INIT_LIST_HEAD(listp); \ | |
357 | list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ | |
e498be7d CL |
358 | } while (0) |
359 | ||
a737b3e2 AM |
360 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
361 | do { \ | |
e498be7d CL |
362 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
363 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
364 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
365 | } while (0) | |
1da177e4 LT |
366 | |
367 | /* | |
343e0d7a | 368 | * struct kmem_cache |
1da177e4 LT |
369 | * |
370 | * manages a cache. | |
371 | */ | |
b28a02de | 372 | |
2109a2d1 | 373 | struct kmem_cache { |
1da177e4 | 374 | /* 1) per-cpu data, touched during every alloc/free */ |
b28a02de | 375 | struct array_cache *array[NR_CPUS]; |
b5d8ca7c | 376 | /* 2) Cache tunables. Protected by cache_chain_mutex */ |
b28a02de PE |
377 | unsigned int batchcount; |
378 | unsigned int limit; | |
379 | unsigned int shared; | |
b5d8ca7c | 380 | |
3dafccf2 | 381 | unsigned int buffer_size; |
b5d8ca7c | 382 | /* 3) touched by every alloc & free from the backend */ |
b28a02de | 383 | struct kmem_list3 *nodelists[MAX_NUMNODES]; |
b5d8ca7c | 384 | |
a737b3e2 AM |
385 | unsigned int flags; /* constant flags */ |
386 | unsigned int num; /* # of objs per slab */ | |
1da177e4 | 387 | |
b5d8ca7c | 388 | /* 4) cache_grow/shrink */ |
1da177e4 | 389 | /* order of pgs per slab (2^n) */ |
b28a02de | 390 | unsigned int gfporder; |
1da177e4 LT |
391 | |
392 | /* force GFP flags, e.g. GFP_DMA */ | |
b28a02de | 393 | gfp_t gfpflags; |
1da177e4 | 394 | |
a737b3e2 | 395 | size_t colour; /* cache colouring range */ |
b28a02de | 396 | unsigned int colour_off; /* colour offset */ |
343e0d7a | 397 | struct kmem_cache *slabp_cache; |
b28a02de | 398 | unsigned int slab_size; |
a737b3e2 | 399 | unsigned int dflags; /* dynamic flags */ |
1da177e4 LT |
400 | |
401 | /* constructor func */ | |
343e0d7a | 402 | void (*ctor) (void *, struct kmem_cache *, unsigned long); |
1da177e4 LT |
403 | |
404 | /* de-constructor func */ | |
343e0d7a | 405 | void (*dtor) (void *, struct kmem_cache *, unsigned long); |
1da177e4 | 406 | |
b5d8ca7c | 407 | /* 5) cache creation/removal */ |
b28a02de PE |
408 | const char *name; |
409 | struct list_head next; | |
1da177e4 | 410 | |
b5d8ca7c | 411 | /* 6) statistics */ |
1da177e4 | 412 | #if STATS |
b28a02de PE |
413 | unsigned long num_active; |
414 | unsigned long num_allocations; | |
415 | unsigned long high_mark; | |
416 | unsigned long grown; | |
417 | unsigned long reaped; | |
418 | unsigned long errors; | |
419 | unsigned long max_freeable; | |
420 | unsigned long node_allocs; | |
421 | unsigned long node_frees; | |
422 | atomic_t allochit; | |
423 | atomic_t allocmiss; | |
424 | atomic_t freehit; | |
425 | atomic_t freemiss; | |
1da177e4 LT |
426 | #endif |
427 | #if DEBUG | |
3dafccf2 MS |
428 | /* |
429 | * If debugging is enabled, then the allocator can add additional | |
430 | * fields and/or padding to every object. buffer_size contains the total | |
431 | * object size including these internal fields, the following two | |
432 | * variables contain the offset to the user object and its size. | |
433 | */ | |
434 | int obj_offset; | |
435 | int obj_size; | |
1da177e4 LT |
436 | #endif |
437 | }; | |
438 | ||
439 | #define CFLGS_OFF_SLAB (0x80000000UL) | |
440 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
441 | ||
442 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
443 | /* |
444 | * Optimization question: fewer reaps means less probability for unnessary | |
445 | * cpucache drain/refill cycles. | |
1da177e4 | 446 | * |
dc6f3f27 | 447 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
448 | * which could lock up otherwise freeable slabs. |
449 | */ | |
450 | #define REAPTIMEOUT_CPUC (2*HZ) | |
451 | #define REAPTIMEOUT_LIST3 (4*HZ) | |
452 | ||
453 | #if STATS | |
454 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
455 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
456 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
457 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
458 | #define STATS_INC_REAPED(x) ((x)->reaped++) | |
a737b3e2 AM |
459 | #define STATS_SET_HIGH(x) \ |
460 | do { \ | |
461 | if ((x)->num_active > (x)->high_mark) \ | |
462 | (x)->high_mark = (x)->num_active; \ | |
463 | } while (0) | |
1da177e4 LT |
464 | #define STATS_INC_ERR(x) ((x)->errors++) |
465 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 466 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
a737b3e2 AM |
467 | #define STATS_SET_FREEABLE(x, i) \ |
468 | do { \ | |
469 | if ((x)->max_freeable < i) \ | |
470 | (x)->max_freeable = i; \ | |
471 | } while (0) | |
1da177e4 LT |
472 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
473 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
474 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
475 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
476 | #else | |
477 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
478 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
479 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
480 | #define STATS_INC_GROWN(x) do { } while (0) | |
481 | #define STATS_INC_REAPED(x) do { } while (0) | |
482 | #define STATS_SET_HIGH(x) do { } while (0) | |
483 | #define STATS_INC_ERR(x) do { } while (0) | |
484 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 485 | #define STATS_INC_NODEFREES(x) do { } while (0) |
a737b3e2 | 486 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
487 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
488 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
489 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
490 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
491 | #endif | |
492 | ||
493 | #if DEBUG | |
a737b3e2 AM |
494 | /* |
495 | * Magic nums for obj red zoning. | |
1da177e4 LT |
496 | * Placed in the first word before and the first word after an obj. |
497 | */ | |
498 | #define RED_INACTIVE 0x5A2CF071UL /* when obj is inactive */ | |
499 | #define RED_ACTIVE 0x170FC2A5UL /* when obj is active */ | |
500 | ||
501 | /* ...and for poisoning */ | |
502 | #define POISON_INUSE 0x5a /* for use-uninitialised poisoning */ | |
503 | #define POISON_FREE 0x6b /* for use-after-free poisoning */ | |
504 | #define POISON_END 0xa5 /* end-byte of poisoning */ | |
505 | ||
a737b3e2 AM |
506 | /* |
507 | * memory layout of objects: | |
1da177e4 | 508 | * 0 : objp |
3dafccf2 | 509 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
510 | * the end of an object is aligned with the end of the real |
511 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 512 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 513 | * redzone word. |
3dafccf2 MS |
514 | * cachep->obj_offset: The real object. |
515 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] | |
a737b3e2 AM |
516 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address |
517 | * [BYTES_PER_WORD long] | |
1da177e4 | 518 | */ |
343e0d7a | 519 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 520 | { |
3dafccf2 | 521 | return cachep->obj_offset; |
1da177e4 LT |
522 | } |
523 | ||
343e0d7a | 524 | static int obj_size(struct kmem_cache *cachep) |
1da177e4 | 525 | { |
3dafccf2 | 526 | return cachep->obj_size; |
1da177e4 LT |
527 | } |
528 | ||
343e0d7a | 529 | static unsigned long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
530 | { |
531 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
3dafccf2 | 532 | return (unsigned long*) (objp+obj_offset(cachep)-BYTES_PER_WORD); |
1da177e4 LT |
533 | } |
534 | ||
343e0d7a | 535 | static unsigned long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
536 | { |
537 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
538 | if (cachep->flags & SLAB_STORE_USER) | |
3dafccf2 | 539 | return (unsigned long *)(objp + cachep->buffer_size - |
b28a02de | 540 | 2 * BYTES_PER_WORD); |
3dafccf2 | 541 | return (unsigned long *)(objp + cachep->buffer_size - BYTES_PER_WORD); |
1da177e4 LT |
542 | } |
543 | ||
343e0d7a | 544 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
545 | { |
546 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3dafccf2 | 547 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
1da177e4 LT |
548 | } |
549 | ||
550 | #else | |
551 | ||
3dafccf2 MS |
552 | #define obj_offset(x) 0 |
553 | #define obj_size(cachep) (cachep->buffer_size) | |
1da177e4 LT |
554 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long *)NULL;}) |
555 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long *)NULL;}) | |
556 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) | |
557 | ||
558 | #endif | |
559 | ||
560 | /* | |
a737b3e2 AM |
561 | * Maximum size of an obj (in 2^order pages) and absolute limit for the gfp |
562 | * order. | |
1da177e4 LT |
563 | */ |
564 | #if defined(CONFIG_LARGE_ALLOCS) | |
565 | #define MAX_OBJ_ORDER 13 /* up to 32Mb */ | |
566 | #define MAX_GFP_ORDER 13 /* up to 32Mb */ | |
567 | #elif defined(CONFIG_MMU) | |
568 | #define MAX_OBJ_ORDER 5 /* 32 pages */ | |
569 | #define MAX_GFP_ORDER 5 /* 32 pages */ | |
570 | #else | |
571 | #define MAX_OBJ_ORDER 8 /* up to 1Mb */ | |
572 | #define MAX_GFP_ORDER 8 /* up to 1Mb */ | |
573 | #endif | |
574 | ||
575 | /* | |
576 | * Do not go above this order unless 0 objects fit into the slab. | |
577 | */ | |
578 | #define BREAK_GFP_ORDER_HI 1 | |
579 | #define BREAK_GFP_ORDER_LO 0 | |
580 | static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; | |
581 | ||
a737b3e2 AM |
582 | /* |
583 | * Functions for storing/retrieving the cachep and or slab from the page | |
584 | * allocator. These are used to find the slab an obj belongs to. With kfree(), | |
585 | * these are used to find the cache which an obj belongs to. | |
1da177e4 | 586 | */ |
065d41cb PE |
587 | static inline void page_set_cache(struct page *page, struct kmem_cache *cache) |
588 | { | |
589 | page->lru.next = (struct list_head *)cache; | |
590 | } | |
591 | ||
592 | static inline struct kmem_cache *page_get_cache(struct page *page) | |
593 | { | |
84097518 NP |
594 | if (unlikely(PageCompound(page))) |
595 | page = (struct page *)page_private(page); | |
065d41cb PE |
596 | return (struct kmem_cache *)page->lru.next; |
597 | } | |
598 | ||
599 | static inline void page_set_slab(struct page *page, struct slab *slab) | |
600 | { | |
601 | page->lru.prev = (struct list_head *)slab; | |
602 | } | |
603 | ||
604 | static inline struct slab *page_get_slab(struct page *page) | |
605 | { | |
84097518 NP |
606 | if (unlikely(PageCompound(page))) |
607 | page = (struct page *)page_private(page); | |
065d41cb PE |
608 | return (struct slab *)page->lru.prev; |
609 | } | |
1da177e4 | 610 | |
6ed5eb22 PE |
611 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
612 | { | |
613 | struct page *page = virt_to_page(obj); | |
614 | return page_get_cache(page); | |
615 | } | |
616 | ||
617 | static inline struct slab *virt_to_slab(const void *obj) | |
618 | { | |
619 | struct page *page = virt_to_page(obj); | |
620 | return page_get_slab(page); | |
621 | } | |
622 | ||
8fea4e96 PE |
623 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
624 | unsigned int idx) | |
625 | { | |
626 | return slab->s_mem + cache->buffer_size * idx; | |
627 | } | |
628 | ||
629 | static inline unsigned int obj_to_index(struct kmem_cache *cache, | |
630 | struct slab *slab, void *obj) | |
631 | { | |
632 | return (unsigned)(obj - slab->s_mem) / cache->buffer_size; | |
633 | } | |
634 | ||
a737b3e2 AM |
635 | /* |
636 | * These are the default caches for kmalloc. Custom caches can have other sizes. | |
637 | */ | |
1da177e4 LT |
638 | struct cache_sizes malloc_sizes[] = { |
639 | #define CACHE(x) { .cs_size = (x) }, | |
640 | #include <linux/kmalloc_sizes.h> | |
641 | CACHE(ULONG_MAX) | |
642 | #undef CACHE | |
643 | }; | |
644 | EXPORT_SYMBOL(malloc_sizes); | |
645 | ||
646 | /* Must match cache_sizes above. Out of line to keep cache footprint low. */ | |
647 | struct cache_names { | |
648 | char *name; | |
649 | char *name_dma; | |
650 | }; | |
651 | ||
652 | static struct cache_names __initdata cache_names[] = { | |
653 | #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, | |
654 | #include <linux/kmalloc_sizes.h> | |
b28a02de | 655 | {NULL,} |
1da177e4 LT |
656 | #undef CACHE |
657 | }; | |
658 | ||
659 | static struct arraycache_init initarray_cache __initdata = | |
b28a02de | 660 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 | 661 | static struct arraycache_init initarray_generic = |
b28a02de | 662 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
663 | |
664 | /* internal cache of cache description objs */ | |
343e0d7a | 665 | static struct kmem_cache cache_cache = { |
b28a02de PE |
666 | .batchcount = 1, |
667 | .limit = BOOT_CPUCACHE_ENTRIES, | |
668 | .shared = 1, | |
343e0d7a | 669 | .buffer_size = sizeof(struct kmem_cache), |
b28a02de | 670 | .name = "kmem_cache", |
1da177e4 | 671 | #if DEBUG |
343e0d7a | 672 | .obj_size = sizeof(struct kmem_cache), |
1da177e4 LT |
673 | #endif |
674 | }; | |
675 | ||
676 | /* Guard access to the cache-chain. */ | |
fc0abb14 | 677 | static DEFINE_MUTEX(cache_chain_mutex); |
1da177e4 LT |
678 | static struct list_head cache_chain; |
679 | ||
680 | /* | |
a737b3e2 AM |
681 | * vm_enough_memory() looks at this to determine how many slab-allocated pages |
682 | * are possibly freeable under pressure | |
1da177e4 LT |
683 | * |
684 | * SLAB_RECLAIM_ACCOUNT turns this on per-slab | |
685 | */ | |
686 | atomic_t slab_reclaim_pages; | |
1da177e4 LT |
687 | |
688 | /* | |
689 | * chicken and egg problem: delay the per-cpu array allocation | |
690 | * until the general caches are up. | |
691 | */ | |
692 | static enum { | |
693 | NONE, | |
e498be7d CL |
694 | PARTIAL_AC, |
695 | PARTIAL_L3, | |
1da177e4 LT |
696 | FULL |
697 | } g_cpucache_up; | |
698 | ||
699 | static DEFINE_PER_CPU(struct work_struct, reap_work); | |
700 | ||
a737b3e2 AM |
701 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
702 | int node); | |
343e0d7a | 703 | static void enable_cpucache(struct kmem_cache *cachep); |
b28a02de | 704 | static void cache_reap(void *unused); |
343e0d7a | 705 | static int __node_shrink(struct kmem_cache *cachep, int node); |
1da177e4 | 706 | |
343e0d7a | 707 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
708 | { |
709 | return cachep->array[smp_processor_id()]; | |
710 | } | |
711 | ||
a737b3e2 AM |
712 | static inline struct kmem_cache *__find_general_cachep(size_t size, |
713 | gfp_t gfpflags) | |
1da177e4 LT |
714 | { |
715 | struct cache_sizes *csizep = malloc_sizes; | |
716 | ||
717 | #if DEBUG | |
718 | /* This happens if someone tries to call | |
b28a02de PE |
719 | * kmem_cache_create(), or __kmalloc(), before |
720 | * the generic caches are initialized. | |
721 | */ | |
c7e43c78 | 722 | BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); |
1da177e4 LT |
723 | #endif |
724 | while (size > csizep->cs_size) | |
725 | csizep++; | |
726 | ||
727 | /* | |
0abf40c1 | 728 | * Really subtle: The last entry with cs->cs_size==ULONG_MAX |
1da177e4 LT |
729 | * has cs_{dma,}cachep==NULL. Thus no special case |
730 | * for large kmalloc calls required. | |
731 | */ | |
732 | if (unlikely(gfpflags & GFP_DMA)) | |
733 | return csizep->cs_dmacachep; | |
734 | return csizep->cs_cachep; | |
735 | } | |
736 | ||
343e0d7a | 737 | struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) |
97e2bde4 MS |
738 | { |
739 | return __find_general_cachep(size, gfpflags); | |
740 | } | |
741 | EXPORT_SYMBOL(kmem_find_general_cachep); | |
742 | ||
fbaccacf | 743 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
1da177e4 | 744 | { |
fbaccacf SR |
745 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
746 | } | |
1da177e4 | 747 | |
a737b3e2 AM |
748 | /* |
749 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
750 | */ | |
fbaccacf SR |
751 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
752 | size_t align, int flags, size_t *left_over, | |
753 | unsigned int *num) | |
754 | { | |
755 | int nr_objs; | |
756 | size_t mgmt_size; | |
757 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 758 | |
fbaccacf SR |
759 | /* |
760 | * The slab management structure can be either off the slab or | |
761 | * on it. For the latter case, the memory allocated for a | |
762 | * slab is used for: | |
763 | * | |
764 | * - The struct slab | |
765 | * - One kmem_bufctl_t for each object | |
766 | * - Padding to respect alignment of @align | |
767 | * - @buffer_size bytes for each object | |
768 | * | |
769 | * If the slab management structure is off the slab, then the | |
770 | * alignment will already be calculated into the size. Because | |
771 | * the slabs are all pages aligned, the objects will be at the | |
772 | * correct alignment when allocated. | |
773 | */ | |
774 | if (flags & CFLGS_OFF_SLAB) { | |
775 | mgmt_size = 0; | |
776 | nr_objs = slab_size / buffer_size; | |
777 | ||
778 | if (nr_objs > SLAB_LIMIT) | |
779 | nr_objs = SLAB_LIMIT; | |
780 | } else { | |
781 | /* | |
782 | * Ignore padding for the initial guess. The padding | |
783 | * is at most @align-1 bytes, and @buffer_size is at | |
784 | * least @align. In the worst case, this result will | |
785 | * be one greater than the number of objects that fit | |
786 | * into the memory allocation when taking the padding | |
787 | * into account. | |
788 | */ | |
789 | nr_objs = (slab_size - sizeof(struct slab)) / | |
790 | (buffer_size + sizeof(kmem_bufctl_t)); | |
791 | ||
792 | /* | |
793 | * This calculated number will be either the right | |
794 | * amount, or one greater than what we want. | |
795 | */ | |
796 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | |
797 | > slab_size) | |
798 | nr_objs--; | |
799 | ||
800 | if (nr_objs > SLAB_LIMIT) | |
801 | nr_objs = SLAB_LIMIT; | |
802 | ||
803 | mgmt_size = slab_mgmt_size(nr_objs, align); | |
804 | } | |
805 | *num = nr_objs; | |
806 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
807 | } |
808 | ||
809 | #define slab_error(cachep, msg) __slab_error(__FUNCTION__, cachep, msg) | |
810 | ||
a737b3e2 AM |
811 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
812 | char *msg) | |
1da177e4 LT |
813 | { |
814 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 815 | function, cachep->name, msg); |
1da177e4 LT |
816 | dump_stack(); |
817 | } | |
818 | ||
8fce4d8e CL |
819 | #ifdef CONFIG_NUMA |
820 | /* | |
821 | * Special reaping functions for NUMA systems called from cache_reap(). | |
822 | * These take care of doing round robin flushing of alien caches (containing | |
823 | * objects freed on different nodes from which they were allocated) and the | |
824 | * flushing of remote pcps by calling drain_node_pages. | |
825 | */ | |
826 | static DEFINE_PER_CPU(unsigned long, reap_node); | |
827 | ||
828 | static void init_reap_node(int cpu) | |
829 | { | |
830 | int node; | |
831 | ||
832 | node = next_node(cpu_to_node(cpu), node_online_map); | |
833 | if (node == MAX_NUMNODES) | |
442295c9 | 834 | node = first_node(node_online_map); |
8fce4d8e CL |
835 | |
836 | __get_cpu_var(reap_node) = node; | |
837 | } | |
838 | ||
839 | static void next_reap_node(void) | |
840 | { | |
841 | int node = __get_cpu_var(reap_node); | |
842 | ||
843 | /* | |
844 | * Also drain per cpu pages on remote zones | |
845 | */ | |
846 | if (node != numa_node_id()) | |
847 | drain_node_pages(node); | |
848 | ||
849 | node = next_node(node, node_online_map); | |
850 | if (unlikely(node >= MAX_NUMNODES)) | |
851 | node = first_node(node_online_map); | |
852 | __get_cpu_var(reap_node) = node; | |
853 | } | |
854 | ||
855 | #else | |
856 | #define init_reap_node(cpu) do { } while (0) | |
857 | #define next_reap_node(void) do { } while (0) | |
858 | #endif | |
859 | ||
1da177e4 LT |
860 | /* |
861 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
862 | * via the workqueue/eventd. | |
863 | * Add the CPU number into the expiration time to minimize the possibility of | |
864 | * the CPUs getting into lockstep and contending for the global cache chain | |
865 | * lock. | |
866 | */ | |
867 | static void __devinit start_cpu_timer(int cpu) | |
868 | { | |
869 | struct work_struct *reap_work = &per_cpu(reap_work, cpu); | |
870 | ||
871 | /* | |
872 | * When this gets called from do_initcalls via cpucache_init(), | |
873 | * init_workqueues() has already run, so keventd will be setup | |
874 | * at that time. | |
875 | */ | |
876 | if (keventd_up() && reap_work->func == NULL) { | |
8fce4d8e | 877 | init_reap_node(cpu); |
1da177e4 LT |
878 | INIT_WORK(reap_work, cache_reap, NULL); |
879 | schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu); | |
880 | } | |
881 | } | |
882 | ||
e498be7d | 883 | static struct array_cache *alloc_arraycache(int node, int entries, |
b28a02de | 884 | int batchcount) |
1da177e4 | 885 | { |
b28a02de | 886 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
887 | struct array_cache *nc = NULL; |
888 | ||
e498be7d | 889 | nc = kmalloc_node(memsize, GFP_KERNEL, node); |
1da177e4 LT |
890 | if (nc) { |
891 | nc->avail = 0; | |
892 | nc->limit = entries; | |
893 | nc->batchcount = batchcount; | |
894 | nc->touched = 0; | |
e498be7d | 895 | spin_lock_init(&nc->lock); |
1da177e4 LT |
896 | } |
897 | return nc; | |
898 | } | |
899 | ||
e498be7d | 900 | #ifdef CONFIG_NUMA |
343e0d7a | 901 | static void *__cache_alloc_node(struct kmem_cache *, gfp_t, int); |
dc85da15 | 902 | |
5295a74c | 903 | static struct array_cache **alloc_alien_cache(int node, int limit) |
e498be7d CL |
904 | { |
905 | struct array_cache **ac_ptr; | |
b28a02de | 906 | int memsize = sizeof(void *) * MAX_NUMNODES; |
e498be7d CL |
907 | int i; |
908 | ||
909 | if (limit > 1) | |
910 | limit = 12; | |
911 | ac_ptr = kmalloc_node(memsize, GFP_KERNEL, node); | |
912 | if (ac_ptr) { | |
913 | for_each_node(i) { | |
914 | if (i == node || !node_online(i)) { | |
915 | ac_ptr[i] = NULL; | |
916 | continue; | |
917 | } | |
918 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d); | |
919 | if (!ac_ptr[i]) { | |
b28a02de | 920 | for (i--; i <= 0; i--) |
e498be7d CL |
921 | kfree(ac_ptr[i]); |
922 | kfree(ac_ptr); | |
923 | return NULL; | |
924 | } | |
925 | } | |
926 | } | |
927 | return ac_ptr; | |
928 | } | |
929 | ||
5295a74c | 930 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
931 | { |
932 | int i; | |
933 | ||
934 | if (!ac_ptr) | |
935 | return; | |
e498be7d | 936 | for_each_node(i) |
b28a02de | 937 | kfree(ac_ptr[i]); |
e498be7d CL |
938 | kfree(ac_ptr); |
939 | } | |
940 | ||
343e0d7a | 941 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 942 | struct array_cache *ac, int node) |
e498be7d CL |
943 | { |
944 | struct kmem_list3 *rl3 = cachep->nodelists[node]; | |
945 | ||
946 | if (ac->avail) { | |
947 | spin_lock(&rl3->list_lock); | |
ff69416e | 948 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d CL |
949 | ac->avail = 0; |
950 | spin_unlock(&rl3->list_lock); | |
951 | } | |
952 | } | |
953 | ||
8fce4d8e CL |
954 | /* |
955 | * Called from cache_reap() to regularly drain alien caches round robin. | |
956 | */ | |
957 | static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) | |
958 | { | |
959 | int node = __get_cpu_var(reap_node); | |
960 | ||
961 | if (l3->alien) { | |
962 | struct array_cache *ac = l3->alien[node]; | |
963 | if (ac && ac->avail) { | |
964 | spin_lock_irq(&ac->lock); | |
965 | __drain_alien_cache(cachep, ac, node); | |
966 | spin_unlock_irq(&ac->lock); | |
967 | } | |
968 | } | |
969 | } | |
970 | ||
a737b3e2 AM |
971 | static void drain_alien_cache(struct kmem_cache *cachep, |
972 | struct array_cache **alien) | |
e498be7d | 973 | { |
b28a02de | 974 | int i = 0; |
e498be7d CL |
975 | struct array_cache *ac; |
976 | unsigned long flags; | |
977 | ||
978 | for_each_online_node(i) { | |
4484ebf1 | 979 | ac = alien[i]; |
e498be7d CL |
980 | if (ac) { |
981 | spin_lock_irqsave(&ac->lock, flags); | |
982 | __drain_alien_cache(cachep, ac, i); | |
983 | spin_unlock_irqrestore(&ac->lock, flags); | |
984 | } | |
985 | } | |
986 | } | |
987 | #else | |
7a21ef6f | 988 | |
4484ebf1 | 989 | #define drain_alien_cache(cachep, alien) do { } while (0) |
8fce4d8e | 990 | #define reap_alien(cachep, l3) do { } while (0) |
4484ebf1 | 991 | |
7a21ef6f LT |
992 | static inline struct array_cache **alloc_alien_cache(int node, int limit) |
993 | { | |
994 | return (struct array_cache **) 0x01020304ul; | |
995 | } | |
996 | ||
4484ebf1 RT |
997 | static inline void free_alien_cache(struct array_cache **ac_ptr) |
998 | { | |
999 | } | |
7a21ef6f | 1000 | |
e498be7d CL |
1001 | #endif |
1002 | ||
1da177e4 | 1003 | static int __devinit cpuup_callback(struct notifier_block *nfb, |
b28a02de | 1004 | unsigned long action, void *hcpu) |
1da177e4 LT |
1005 | { |
1006 | long cpu = (long)hcpu; | |
343e0d7a | 1007 | struct kmem_cache *cachep; |
e498be7d CL |
1008 | struct kmem_list3 *l3 = NULL; |
1009 | int node = cpu_to_node(cpu); | |
1010 | int memsize = sizeof(struct kmem_list3); | |
1da177e4 LT |
1011 | |
1012 | switch (action) { | |
1013 | case CPU_UP_PREPARE: | |
fc0abb14 | 1014 | mutex_lock(&cache_chain_mutex); |
a737b3e2 AM |
1015 | /* |
1016 | * We need to do this right in the beginning since | |
e498be7d CL |
1017 | * alloc_arraycache's are going to use this list. |
1018 | * kmalloc_node allows us to add the slab to the right | |
1019 | * kmem_list3 and not this cpu's kmem_list3 | |
1020 | */ | |
1021 | ||
1da177e4 | 1022 | list_for_each_entry(cachep, &cache_chain, next) { |
a737b3e2 AM |
1023 | /* |
1024 | * Set up the size64 kmemlist for cpu before we can | |
e498be7d CL |
1025 | * begin anything. Make sure some other cpu on this |
1026 | * node has not already allocated this | |
1027 | */ | |
1028 | if (!cachep->nodelists[node]) { | |
a737b3e2 AM |
1029 | l3 = kmalloc_node(memsize, GFP_KERNEL, node); |
1030 | if (!l3) | |
e498be7d CL |
1031 | goto bad; |
1032 | kmem_list3_init(l3); | |
1033 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
b28a02de | 1034 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
e498be7d | 1035 | |
4484ebf1 RT |
1036 | /* |
1037 | * The l3s don't come and go as CPUs come and | |
1038 | * go. cache_chain_mutex is sufficient | |
1039 | * protection here. | |
1040 | */ | |
e498be7d CL |
1041 | cachep->nodelists[node] = l3; |
1042 | } | |
1da177e4 | 1043 | |
e498be7d CL |
1044 | spin_lock_irq(&cachep->nodelists[node]->list_lock); |
1045 | cachep->nodelists[node]->free_limit = | |
a737b3e2 AM |
1046 | (1 + nr_cpus_node(node)) * |
1047 | cachep->batchcount + cachep->num; | |
e498be7d CL |
1048 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); |
1049 | } | |
1050 | ||
a737b3e2 AM |
1051 | /* |
1052 | * Now we can go ahead with allocating the shared arrays and | |
1053 | * array caches | |
1054 | */ | |
e498be7d | 1055 | list_for_each_entry(cachep, &cache_chain, next) { |
cd105df4 | 1056 | struct array_cache *nc; |
4484ebf1 RT |
1057 | struct array_cache *shared; |
1058 | struct array_cache **alien; | |
cd105df4 | 1059 | |
e498be7d | 1060 | nc = alloc_arraycache(node, cachep->limit, |
4484ebf1 | 1061 | cachep->batchcount); |
1da177e4 LT |
1062 | if (!nc) |
1063 | goto bad; | |
4484ebf1 RT |
1064 | shared = alloc_arraycache(node, |
1065 | cachep->shared * cachep->batchcount, | |
1066 | 0xbaadf00d); | |
1067 | if (!shared) | |
1068 | goto bad; | |
7a21ef6f | 1069 | |
4484ebf1 RT |
1070 | alien = alloc_alien_cache(node, cachep->limit); |
1071 | if (!alien) | |
1072 | goto bad; | |
1da177e4 | 1073 | cachep->array[cpu] = nc; |
e498be7d CL |
1074 | l3 = cachep->nodelists[node]; |
1075 | BUG_ON(!l3); | |
e498be7d | 1076 | |
4484ebf1 RT |
1077 | spin_lock_irq(&l3->list_lock); |
1078 | if (!l3->shared) { | |
1079 | /* | |
1080 | * We are serialised from CPU_DEAD or | |
1081 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1082 | */ | |
1083 | l3->shared = shared; | |
1084 | shared = NULL; | |
e498be7d | 1085 | } |
4484ebf1 RT |
1086 | #ifdef CONFIG_NUMA |
1087 | if (!l3->alien) { | |
1088 | l3->alien = alien; | |
1089 | alien = NULL; | |
1090 | } | |
1091 | #endif | |
1092 | spin_unlock_irq(&l3->list_lock); | |
4484ebf1 RT |
1093 | kfree(shared); |
1094 | free_alien_cache(alien); | |
1da177e4 | 1095 | } |
fc0abb14 | 1096 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1097 | break; |
1098 | case CPU_ONLINE: | |
1099 | start_cpu_timer(cpu); | |
1100 | break; | |
1101 | #ifdef CONFIG_HOTPLUG_CPU | |
1102 | case CPU_DEAD: | |
4484ebf1 RT |
1103 | /* |
1104 | * Even if all the cpus of a node are down, we don't free the | |
1105 | * kmem_list3 of any cache. This to avoid a race between | |
1106 | * cpu_down, and a kmalloc allocation from another cpu for | |
1107 | * memory from the node of the cpu going down. The list3 | |
1108 | * structure is usually allocated from kmem_cache_create() and | |
1109 | * gets destroyed at kmem_cache_destroy(). | |
1110 | */ | |
1da177e4 LT |
1111 | /* fall thru */ |
1112 | case CPU_UP_CANCELED: | |
fc0abb14 | 1113 | mutex_lock(&cache_chain_mutex); |
1da177e4 LT |
1114 | list_for_each_entry(cachep, &cache_chain, next) { |
1115 | struct array_cache *nc; | |
4484ebf1 RT |
1116 | struct array_cache *shared; |
1117 | struct array_cache **alien; | |
e498be7d | 1118 | cpumask_t mask; |
1da177e4 | 1119 | |
e498be7d | 1120 | mask = node_to_cpumask(node); |
1da177e4 LT |
1121 | /* cpu is dead; no one can alloc from it. */ |
1122 | nc = cachep->array[cpu]; | |
1123 | cachep->array[cpu] = NULL; | |
e498be7d CL |
1124 | l3 = cachep->nodelists[node]; |
1125 | ||
1126 | if (!l3) | |
4484ebf1 | 1127 | goto free_array_cache; |
e498be7d | 1128 | |
ca3b9b91 | 1129 | spin_lock_irq(&l3->list_lock); |
e498be7d CL |
1130 | |
1131 | /* Free limit for this kmem_list3 */ | |
1132 | l3->free_limit -= cachep->batchcount; | |
1133 | if (nc) | |
ff69416e | 1134 | free_block(cachep, nc->entry, nc->avail, node); |
e498be7d CL |
1135 | |
1136 | if (!cpus_empty(mask)) { | |
ca3b9b91 | 1137 | spin_unlock_irq(&l3->list_lock); |
4484ebf1 | 1138 | goto free_array_cache; |
b28a02de | 1139 | } |
e498be7d | 1140 | |
4484ebf1 RT |
1141 | shared = l3->shared; |
1142 | if (shared) { | |
e498be7d | 1143 | free_block(cachep, l3->shared->entry, |
b28a02de | 1144 | l3->shared->avail, node); |
e498be7d CL |
1145 | l3->shared = NULL; |
1146 | } | |
e498be7d | 1147 | |
4484ebf1 RT |
1148 | alien = l3->alien; |
1149 | l3->alien = NULL; | |
1150 | ||
1151 | spin_unlock_irq(&l3->list_lock); | |
1152 | ||
1153 | kfree(shared); | |
1154 | if (alien) { | |
1155 | drain_alien_cache(cachep, alien); | |
1156 | free_alien_cache(alien); | |
e498be7d | 1157 | } |
4484ebf1 | 1158 | free_array_cache: |
1da177e4 LT |
1159 | kfree(nc); |
1160 | } | |
4484ebf1 RT |
1161 | /* |
1162 | * In the previous loop, all the objects were freed to | |
1163 | * the respective cache's slabs, now we can go ahead and | |
1164 | * shrink each nodelist to its limit. | |
1165 | */ | |
1166 | list_for_each_entry(cachep, &cache_chain, next) { | |
1167 | l3 = cachep->nodelists[node]; | |
1168 | if (!l3) | |
1169 | continue; | |
1170 | spin_lock_irq(&l3->list_lock); | |
1171 | /* free slabs belonging to this node */ | |
1172 | __node_shrink(cachep, node); | |
1173 | spin_unlock_irq(&l3->list_lock); | |
1174 | } | |
fc0abb14 | 1175 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1176 | break; |
1177 | #endif | |
1178 | } | |
1179 | return NOTIFY_OK; | |
a737b3e2 | 1180 | bad: |
fc0abb14 | 1181 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1182 | return NOTIFY_BAD; |
1183 | } | |
1184 | ||
1185 | static struct notifier_block cpucache_notifier = { &cpuup_callback, NULL, 0 }; | |
1186 | ||
e498be7d CL |
1187 | /* |
1188 | * swap the static kmem_list3 with kmalloced memory | |
1189 | */ | |
a737b3e2 AM |
1190 | static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, |
1191 | int nodeid) | |
e498be7d CL |
1192 | { |
1193 | struct kmem_list3 *ptr; | |
1194 | ||
1195 | BUG_ON(cachep->nodelists[nodeid] != list); | |
1196 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, nodeid); | |
1197 | BUG_ON(!ptr); | |
1198 | ||
1199 | local_irq_disable(); | |
1200 | memcpy(ptr, list, sizeof(struct kmem_list3)); | |
1201 | MAKE_ALL_LISTS(cachep, ptr, nodeid); | |
1202 | cachep->nodelists[nodeid] = ptr; | |
1203 | local_irq_enable(); | |
1204 | } | |
1205 | ||
a737b3e2 AM |
1206 | /* |
1207 | * Initialisation. Called after the page allocator have been initialised and | |
1208 | * before smp_init(). | |
1da177e4 LT |
1209 | */ |
1210 | void __init kmem_cache_init(void) | |
1211 | { | |
1212 | size_t left_over; | |
1213 | struct cache_sizes *sizes; | |
1214 | struct cache_names *names; | |
e498be7d | 1215 | int i; |
07ed76b2 | 1216 | int order; |
e498be7d CL |
1217 | |
1218 | for (i = 0; i < NUM_INIT_LISTS; i++) { | |
1219 | kmem_list3_init(&initkmem_list3[i]); | |
1220 | if (i < MAX_NUMNODES) | |
1221 | cache_cache.nodelists[i] = NULL; | |
1222 | } | |
1da177e4 LT |
1223 | |
1224 | /* | |
1225 | * Fragmentation resistance on low memory - only use bigger | |
1226 | * page orders on machines with more than 32MB of memory. | |
1227 | */ | |
1228 | if (num_physpages > (32 << 20) >> PAGE_SHIFT) | |
1229 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; | |
1230 | ||
1da177e4 LT |
1231 | /* Bootstrap is tricky, because several objects are allocated |
1232 | * from caches that do not exist yet: | |
a737b3e2 AM |
1233 | * 1) initialize the cache_cache cache: it contains the struct |
1234 | * kmem_cache structures of all caches, except cache_cache itself: | |
1235 | * cache_cache is statically allocated. | |
e498be7d CL |
1236 | * Initially an __init data area is used for the head array and the |
1237 | * kmem_list3 structures, it's replaced with a kmalloc allocated | |
1238 | * array at the end of the bootstrap. | |
1da177e4 | 1239 | * 2) Create the first kmalloc cache. |
343e0d7a | 1240 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1241 | * An __init data area is used for the head array. |
1242 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1243 | * head arrays. | |
1da177e4 LT |
1244 | * 4) Replace the __init data head arrays for cache_cache and the first |
1245 | * kmalloc cache with kmalloc allocated arrays. | |
e498be7d CL |
1246 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
1247 | * the other cache's with kmalloc allocated memory. | |
1248 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1249 | */ |
1250 | ||
1251 | /* 1) create the cache_cache */ | |
1da177e4 LT |
1252 | INIT_LIST_HEAD(&cache_chain); |
1253 | list_add(&cache_cache.next, &cache_chain); | |
1254 | cache_cache.colour_off = cache_line_size(); | |
1255 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; | |
e498be7d | 1256 | cache_cache.nodelists[numa_node_id()] = &initkmem_list3[CACHE_CACHE]; |
1da177e4 | 1257 | |
a737b3e2 AM |
1258 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, |
1259 | cache_line_size()); | |
1da177e4 | 1260 | |
07ed76b2 JS |
1261 | for (order = 0; order < MAX_ORDER; order++) { |
1262 | cache_estimate(order, cache_cache.buffer_size, | |
1263 | cache_line_size(), 0, &left_over, &cache_cache.num); | |
1264 | if (cache_cache.num) | |
1265 | break; | |
1266 | } | |
1da177e4 LT |
1267 | if (!cache_cache.num) |
1268 | BUG(); | |
07ed76b2 | 1269 | cache_cache.gfporder = order; |
b28a02de | 1270 | cache_cache.colour = left_over / cache_cache.colour_off; |
b28a02de PE |
1271 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
1272 | sizeof(struct slab), cache_line_size()); | |
1da177e4 LT |
1273 | |
1274 | /* 2+3) create the kmalloc caches */ | |
1275 | sizes = malloc_sizes; | |
1276 | names = cache_names; | |
1277 | ||
a737b3e2 AM |
1278 | /* |
1279 | * Initialize the caches that provide memory for the array cache and the | |
1280 | * kmem_list3 structures first. Without this, further allocations will | |
1281 | * bug. | |
e498be7d CL |
1282 | */ |
1283 | ||
1284 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, | |
a737b3e2 AM |
1285 | sizes[INDEX_AC].cs_size, |
1286 | ARCH_KMALLOC_MINALIGN, | |
1287 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
1288 | NULL, NULL); | |
e498be7d | 1289 | |
a737b3e2 | 1290 | if (INDEX_AC != INDEX_L3) { |
e498be7d | 1291 | sizes[INDEX_L3].cs_cachep = |
a737b3e2 AM |
1292 | kmem_cache_create(names[INDEX_L3].name, |
1293 | sizes[INDEX_L3].cs_size, | |
1294 | ARCH_KMALLOC_MINALIGN, | |
1295 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
1296 | NULL, NULL); | |
1297 | } | |
e498be7d | 1298 | |
1da177e4 | 1299 | while (sizes->cs_size != ULONG_MAX) { |
e498be7d CL |
1300 | /* |
1301 | * For performance, all the general caches are L1 aligned. | |
1da177e4 LT |
1302 | * This should be particularly beneficial on SMP boxes, as it |
1303 | * eliminates "false sharing". | |
1304 | * Note for systems short on memory removing the alignment will | |
e498be7d CL |
1305 | * allow tighter packing of the smaller caches. |
1306 | */ | |
a737b3e2 | 1307 | if (!sizes->cs_cachep) { |
e498be7d | 1308 | sizes->cs_cachep = kmem_cache_create(names->name, |
a737b3e2 AM |
1309 | sizes->cs_size, |
1310 | ARCH_KMALLOC_MINALIGN, | |
1311 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
1312 | NULL, NULL); | |
1313 | } | |
1da177e4 LT |
1314 | |
1315 | /* Inc off-slab bufctl limit until the ceiling is hit. */ | |
1316 | if (!(OFF_SLAB(sizes->cs_cachep))) { | |
b28a02de | 1317 | offslab_limit = sizes->cs_size - sizeof(struct slab); |
1da177e4 LT |
1318 | offslab_limit /= sizeof(kmem_bufctl_t); |
1319 | } | |
1320 | ||
1321 | sizes->cs_dmacachep = kmem_cache_create(names->name_dma, | |
a737b3e2 AM |
1322 | sizes->cs_size, |
1323 | ARCH_KMALLOC_MINALIGN, | |
1324 | ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA| | |
1325 | SLAB_PANIC, | |
1326 | NULL, NULL); | |
1da177e4 LT |
1327 | sizes++; |
1328 | names++; | |
1329 | } | |
1330 | /* 4) Replace the bootstrap head arrays */ | |
1331 | { | |
b28a02de | 1332 | void *ptr; |
e498be7d | 1333 | |
1da177e4 | 1334 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
e498be7d | 1335 | |
1da177e4 | 1336 | local_irq_disable(); |
9a2dba4b PE |
1337 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
1338 | memcpy(ptr, cpu_cache_get(&cache_cache), | |
b28a02de | 1339 | sizeof(struct arraycache_init)); |
1da177e4 LT |
1340 | cache_cache.array[smp_processor_id()] = ptr; |
1341 | local_irq_enable(); | |
e498be7d | 1342 | |
1da177e4 | 1343 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); |
e498be7d | 1344 | |
1da177e4 | 1345 | local_irq_disable(); |
9a2dba4b | 1346 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
b28a02de | 1347 | != &initarray_generic.cache); |
9a2dba4b | 1348 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
b28a02de | 1349 | sizeof(struct arraycache_init)); |
e498be7d | 1350 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
b28a02de | 1351 | ptr; |
1da177e4 LT |
1352 | local_irq_enable(); |
1353 | } | |
e498be7d CL |
1354 | /* 5) Replace the bootstrap kmem_list3's */ |
1355 | { | |
1356 | int node; | |
1357 | /* Replace the static kmem_list3 structures for the boot cpu */ | |
1358 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE], | |
b28a02de | 1359 | numa_node_id()); |
e498be7d CL |
1360 | |
1361 | for_each_online_node(node) { | |
1362 | init_list(malloc_sizes[INDEX_AC].cs_cachep, | |
b28a02de | 1363 | &initkmem_list3[SIZE_AC + node], node); |
e498be7d CL |
1364 | |
1365 | if (INDEX_AC != INDEX_L3) { | |
1366 | init_list(malloc_sizes[INDEX_L3].cs_cachep, | |
b28a02de PE |
1367 | &initkmem_list3[SIZE_L3 + node], |
1368 | node); | |
e498be7d CL |
1369 | } |
1370 | } | |
1371 | } | |
1da177e4 | 1372 | |
e498be7d | 1373 | /* 6) resize the head arrays to their final sizes */ |
1da177e4 | 1374 | { |
343e0d7a | 1375 | struct kmem_cache *cachep; |
fc0abb14 | 1376 | mutex_lock(&cache_chain_mutex); |
1da177e4 | 1377 | list_for_each_entry(cachep, &cache_chain, next) |
a737b3e2 | 1378 | enable_cpucache(cachep); |
fc0abb14 | 1379 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1380 | } |
1381 | ||
1382 | /* Done! */ | |
1383 | g_cpucache_up = FULL; | |
1384 | ||
a737b3e2 AM |
1385 | /* |
1386 | * Register a cpu startup notifier callback that initializes | |
1387 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1388 | */ |
1389 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1390 | |
a737b3e2 AM |
1391 | /* |
1392 | * The reap timers are started later, with a module init call: That part | |
1393 | * of the kernel is not yet operational. | |
1da177e4 LT |
1394 | */ |
1395 | } | |
1396 | ||
1397 | static int __init cpucache_init(void) | |
1398 | { | |
1399 | int cpu; | |
1400 | ||
a737b3e2 AM |
1401 | /* |
1402 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1403 | */ |
e498be7d | 1404 | for_each_online_cpu(cpu) |
a737b3e2 | 1405 | start_cpu_timer(cpu); |
1da177e4 LT |
1406 | return 0; |
1407 | } | |
1da177e4 LT |
1408 | __initcall(cpucache_init); |
1409 | ||
1410 | /* | |
1411 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1412 | * | |
1413 | * If we requested dmaable memory, we will get it. Even if we | |
1414 | * did not request dmaable memory, we might get it, but that | |
1415 | * would be relatively rare and ignorable. | |
1416 | */ | |
343e0d7a | 1417 | static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1da177e4 LT |
1418 | { |
1419 | struct page *page; | |
1420 | void *addr; | |
1421 | int i; | |
1422 | ||
1423 | flags |= cachep->gfpflags; | |
50c85a19 | 1424 | page = alloc_pages_node(nodeid, flags, cachep->gfporder); |
1da177e4 LT |
1425 | if (!page) |
1426 | return NULL; | |
1427 | addr = page_address(page); | |
1428 | ||
1429 | i = (1 << cachep->gfporder); | |
1430 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) | |
1431 | atomic_add(i, &slab_reclaim_pages); | |
1432 | add_page_state(nr_slab, i); | |
1433 | while (i--) { | |
f205b2fe | 1434 | __SetPageSlab(page); |
1da177e4 LT |
1435 | page++; |
1436 | } | |
1437 | return addr; | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Interface to system's page release. | |
1442 | */ | |
343e0d7a | 1443 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
1da177e4 | 1444 | { |
b28a02de | 1445 | unsigned long i = (1 << cachep->gfporder); |
1da177e4 LT |
1446 | struct page *page = virt_to_page(addr); |
1447 | const unsigned long nr_freed = i; | |
1448 | ||
1449 | while (i--) { | |
f205b2fe NP |
1450 | BUG_ON(!PageSlab(page)); |
1451 | __ClearPageSlab(page); | |
1da177e4 LT |
1452 | page++; |
1453 | } | |
1454 | sub_page_state(nr_slab, nr_freed); | |
1455 | if (current->reclaim_state) | |
1456 | current->reclaim_state->reclaimed_slab += nr_freed; | |
1457 | free_pages((unsigned long)addr, cachep->gfporder); | |
b28a02de PE |
1458 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1459 | atomic_sub(1 << cachep->gfporder, &slab_reclaim_pages); | |
1da177e4 LT |
1460 | } |
1461 | ||
1462 | static void kmem_rcu_free(struct rcu_head *head) | |
1463 | { | |
b28a02de | 1464 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
343e0d7a | 1465 | struct kmem_cache *cachep = slab_rcu->cachep; |
1da177e4 LT |
1466 | |
1467 | kmem_freepages(cachep, slab_rcu->addr); | |
1468 | if (OFF_SLAB(cachep)) | |
1469 | kmem_cache_free(cachep->slabp_cache, slab_rcu); | |
1470 | } | |
1471 | ||
1472 | #if DEBUG | |
1473 | ||
1474 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1475 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1476 | unsigned long caller) |
1da177e4 | 1477 | { |
3dafccf2 | 1478 | int size = obj_size(cachep); |
1da177e4 | 1479 | |
3dafccf2 | 1480 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1481 | |
b28a02de | 1482 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1483 | return; |
1484 | ||
b28a02de PE |
1485 | *addr++ = 0x12345678; |
1486 | *addr++ = caller; | |
1487 | *addr++ = smp_processor_id(); | |
1488 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1489 | { |
1490 | unsigned long *sptr = &caller; | |
1491 | unsigned long svalue; | |
1492 | ||
1493 | while (!kstack_end(sptr)) { | |
1494 | svalue = *sptr++; | |
1495 | if (kernel_text_address(svalue)) { | |
b28a02de | 1496 | *addr++ = svalue; |
1da177e4 LT |
1497 | size -= sizeof(unsigned long); |
1498 | if (size <= sizeof(unsigned long)) | |
1499 | break; | |
1500 | } | |
1501 | } | |
1502 | ||
1503 | } | |
b28a02de | 1504 | *addr++ = 0x87654321; |
1da177e4 LT |
1505 | } |
1506 | #endif | |
1507 | ||
343e0d7a | 1508 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1509 | { |
3dafccf2 MS |
1510 | int size = obj_size(cachep); |
1511 | addr = &((char *)addr)[obj_offset(cachep)]; | |
1da177e4 LT |
1512 | |
1513 | memset(addr, val, size); | |
b28a02de | 1514 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1515 | } |
1516 | ||
1517 | static void dump_line(char *data, int offset, int limit) | |
1518 | { | |
1519 | int i; | |
1520 | printk(KERN_ERR "%03x:", offset); | |
a737b3e2 | 1521 | for (i = 0; i < limit; i++) |
b28a02de | 1522 | printk(" %02x", (unsigned char)data[offset + i]); |
1da177e4 LT |
1523 | printk("\n"); |
1524 | } | |
1525 | #endif | |
1526 | ||
1527 | #if DEBUG | |
1528 | ||
343e0d7a | 1529 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1530 | { |
1531 | int i, size; | |
1532 | char *realobj; | |
1533 | ||
1534 | if (cachep->flags & SLAB_RED_ZONE) { | |
1535 | printk(KERN_ERR "Redzone: 0x%lx/0x%lx.\n", | |
a737b3e2 AM |
1536 | *dbg_redzone1(cachep, objp), |
1537 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1538 | } |
1539 | ||
1540 | if (cachep->flags & SLAB_STORE_USER) { | |
1541 | printk(KERN_ERR "Last user: [<%p>]", | |
a737b3e2 | 1542 | *dbg_userword(cachep, objp)); |
1da177e4 | 1543 | print_symbol("(%s)", |
a737b3e2 | 1544 | (unsigned long)*dbg_userword(cachep, objp)); |
1da177e4 LT |
1545 | printk("\n"); |
1546 | } | |
3dafccf2 MS |
1547 | realobj = (char *)objp + obj_offset(cachep); |
1548 | size = obj_size(cachep); | |
b28a02de | 1549 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1550 | int limit; |
1551 | limit = 16; | |
b28a02de PE |
1552 | if (i + limit > size) |
1553 | limit = size - i; | |
1da177e4 LT |
1554 | dump_line(realobj, i, limit); |
1555 | } | |
1556 | } | |
1557 | ||
343e0d7a | 1558 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1559 | { |
1560 | char *realobj; | |
1561 | int size, i; | |
1562 | int lines = 0; | |
1563 | ||
3dafccf2 MS |
1564 | realobj = (char *)objp + obj_offset(cachep); |
1565 | size = obj_size(cachep); | |
1da177e4 | 1566 | |
b28a02de | 1567 | for (i = 0; i < size; i++) { |
1da177e4 | 1568 | char exp = POISON_FREE; |
b28a02de | 1569 | if (i == size - 1) |
1da177e4 LT |
1570 | exp = POISON_END; |
1571 | if (realobj[i] != exp) { | |
1572 | int limit; | |
1573 | /* Mismatch ! */ | |
1574 | /* Print header */ | |
1575 | if (lines == 0) { | |
b28a02de | 1576 | printk(KERN_ERR |
a737b3e2 AM |
1577 | "Slab corruption: start=%p, len=%d\n", |
1578 | realobj, size); | |
1da177e4 LT |
1579 | print_objinfo(cachep, objp, 0); |
1580 | } | |
1581 | /* Hexdump the affected line */ | |
b28a02de | 1582 | i = (i / 16) * 16; |
1da177e4 | 1583 | limit = 16; |
b28a02de PE |
1584 | if (i + limit > size) |
1585 | limit = size - i; | |
1da177e4 LT |
1586 | dump_line(realobj, i, limit); |
1587 | i += 16; | |
1588 | lines++; | |
1589 | /* Limit to 5 lines */ | |
1590 | if (lines > 5) | |
1591 | break; | |
1592 | } | |
1593 | } | |
1594 | if (lines != 0) { | |
1595 | /* Print some data about the neighboring objects, if they | |
1596 | * exist: | |
1597 | */ | |
6ed5eb22 | 1598 | struct slab *slabp = virt_to_slab(objp); |
8fea4e96 | 1599 | unsigned int objnr; |
1da177e4 | 1600 | |
8fea4e96 | 1601 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 | 1602 | if (objnr) { |
8fea4e96 | 1603 | objp = index_to_obj(cachep, slabp, objnr - 1); |
3dafccf2 | 1604 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1605 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1606 | realobj, size); |
1da177e4 LT |
1607 | print_objinfo(cachep, objp, 2); |
1608 | } | |
b28a02de | 1609 | if (objnr + 1 < cachep->num) { |
8fea4e96 | 1610 | objp = index_to_obj(cachep, slabp, objnr + 1); |
3dafccf2 | 1611 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1612 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1613 | realobj, size); |
1da177e4 LT |
1614 | print_objinfo(cachep, objp, 2); |
1615 | } | |
1616 | } | |
1617 | } | |
1618 | #endif | |
1619 | ||
12dd36fa MD |
1620 | #if DEBUG |
1621 | /** | |
911851e6 RD |
1622 | * slab_destroy_objs - destroy a slab and its objects |
1623 | * @cachep: cache pointer being destroyed | |
1624 | * @slabp: slab pointer being destroyed | |
1625 | * | |
1626 | * Call the registered destructor for each object in a slab that is being | |
1627 | * destroyed. | |
1da177e4 | 1628 | */ |
343e0d7a | 1629 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 | 1630 | { |
1da177e4 LT |
1631 | int i; |
1632 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 1633 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
1634 | |
1635 | if (cachep->flags & SLAB_POISON) { | |
1636 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 AM |
1637 | if (cachep->buffer_size % PAGE_SIZE == 0 && |
1638 | OFF_SLAB(cachep)) | |
b28a02de | 1639 | kernel_map_pages(virt_to_page(objp), |
a737b3e2 | 1640 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
1641 | else |
1642 | check_poison_obj(cachep, objp); | |
1643 | #else | |
1644 | check_poison_obj(cachep, objp); | |
1645 | #endif | |
1646 | } | |
1647 | if (cachep->flags & SLAB_RED_ZONE) { | |
1648 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1649 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1650 | "was overwritten"); |
1da177e4 LT |
1651 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1652 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1653 | "was overwritten"); |
1da177e4 LT |
1654 | } |
1655 | if (cachep->dtor && !(cachep->flags & SLAB_POISON)) | |
3dafccf2 | 1656 | (cachep->dtor) (objp + obj_offset(cachep), cachep, 0); |
1da177e4 | 1657 | } |
12dd36fa | 1658 | } |
1da177e4 | 1659 | #else |
343e0d7a | 1660 | static void slab_destroy_objs(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 1661 | { |
1da177e4 LT |
1662 | if (cachep->dtor) { |
1663 | int i; | |
1664 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 1665 | void *objp = index_to_obj(cachep, slabp, i); |
b28a02de | 1666 | (cachep->dtor) (objp, cachep, 0); |
1da177e4 LT |
1667 | } |
1668 | } | |
12dd36fa | 1669 | } |
1da177e4 LT |
1670 | #endif |
1671 | ||
911851e6 RD |
1672 | /** |
1673 | * slab_destroy - destroy and release all objects in a slab | |
1674 | * @cachep: cache pointer being destroyed | |
1675 | * @slabp: slab pointer being destroyed | |
1676 | * | |
12dd36fa | 1677 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
1678 | * Before calling the slab must have been unlinked from the cache. The |
1679 | * cache-lock is not held/needed. | |
12dd36fa | 1680 | */ |
343e0d7a | 1681 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa MD |
1682 | { |
1683 | void *addr = slabp->s_mem - slabp->colouroff; | |
1684 | ||
1685 | slab_destroy_objs(cachep, slabp); | |
1da177e4 LT |
1686 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
1687 | struct slab_rcu *slab_rcu; | |
1688 | ||
b28a02de | 1689 | slab_rcu = (struct slab_rcu *)slabp; |
1da177e4 LT |
1690 | slab_rcu->cachep = cachep; |
1691 | slab_rcu->addr = addr; | |
1692 | call_rcu(&slab_rcu->head, kmem_rcu_free); | |
1693 | } else { | |
1694 | kmem_freepages(cachep, addr); | |
1695 | if (OFF_SLAB(cachep)) | |
1696 | kmem_cache_free(cachep->slabp_cache, slabp); | |
1697 | } | |
1698 | } | |
1699 | ||
a737b3e2 AM |
1700 | /* |
1701 | * For setting up all the kmem_list3s for cache whose buffer_size is same as | |
1702 | * size of kmem_list3. | |
1703 | */ | |
343e0d7a | 1704 | static void set_up_list3s(struct kmem_cache *cachep, int index) |
e498be7d CL |
1705 | { |
1706 | int node; | |
1707 | ||
1708 | for_each_online_node(node) { | |
b28a02de | 1709 | cachep->nodelists[node] = &initkmem_list3[index + node]; |
e498be7d | 1710 | cachep->nodelists[node]->next_reap = jiffies + |
b28a02de PE |
1711 | REAPTIMEOUT_LIST3 + |
1712 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
e498be7d CL |
1713 | } |
1714 | } | |
1715 | ||
4d268eba | 1716 | /** |
a70773dd RD |
1717 | * calculate_slab_order - calculate size (page order) of slabs |
1718 | * @cachep: pointer to the cache that is being created | |
1719 | * @size: size of objects to be created in this cache. | |
1720 | * @align: required alignment for the objects. | |
1721 | * @flags: slab allocation flags | |
1722 | * | |
1723 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1724 | * |
1725 | * This could be made much more intelligent. For now, try to avoid using | |
1726 | * high order pages for slabs. When the gfp() functions are more friendly | |
1727 | * towards high-order requests, this should be changed. | |
1728 | */ | |
a737b3e2 | 1729 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1730 | size_t size, size_t align, unsigned long flags) |
4d268eba PE |
1731 | { |
1732 | size_t left_over = 0; | |
9888e6fa | 1733 | int gfporder; |
4d268eba | 1734 | |
a737b3e2 | 1735 | for (gfporder = 0; gfporder <= MAX_GFP_ORDER; gfporder++) { |
4d268eba PE |
1736 | unsigned int num; |
1737 | size_t remainder; | |
1738 | ||
9888e6fa | 1739 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1740 | if (!num) |
1741 | continue; | |
9888e6fa | 1742 | |
4d268eba | 1743 | /* More than offslab_limit objects will cause problems */ |
9888e6fa | 1744 | if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) |
4d268eba PE |
1745 | break; |
1746 | ||
9888e6fa | 1747 | /* Found something acceptable - save it away */ |
4d268eba | 1748 | cachep->num = num; |
9888e6fa | 1749 | cachep->gfporder = gfporder; |
4d268eba PE |
1750 | left_over = remainder; |
1751 | ||
f78bb8ad LT |
1752 | /* |
1753 | * A VFS-reclaimable slab tends to have most allocations | |
1754 | * as GFP_NOFS and we really don't want to have to be allocating | |
1755 | * higher-order pages when we are unable to shrink dcache. | |
1756 | */ | |
1757 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1758 | break; | |
1759 | ||
4d268eba PE |
1760 | /* |
1761 | * Large number of objects is good, but very large slabs are | |
1762 | * currently bad for the gfp()s. | |
1763 | */ | |
9888e6fa | 1764 | if (gfporder >= slab_break_gfp_order) |
4d268eba PE |
1765 | break; |
1766 | ||
9888e6fa LT |
1767 | /* |
1768 | * Acceptable internal fragmentation? | |
1769 | */ | |
a737b3e2 | 1770 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1771 | break; |
1772 | } | |
1773 | return left_over; | |
1774 | } | |
1775 | ||
f30cf7d1 PE |
1776 | static void setup_cpu_cache(struct kmem_cache *cachep) |
1777 | { | |
1778 | if (g_cpucache_up == FULL) { | |
1779 | enable_cpucache(cachep); | |
1780 | return; | |
1781 | } | |
1782 | if (g_cpucache_up == NONE) { | |
1783 | /* | |
1784 | * Note: the first kmem_cache_create must create the cache | |
1785 | * that's used by kmalloc(24), otherwise the creation of | |
1786 | * further caches will BUG(). | |
1787 | */ | |
1788 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
1789 | ||
1790 | /* | |
1791 | * If the cache that's used by kmalloc(sizeof(kmem_list3)) is | |
1792 | * the first cache, then we need to set up all its list3s, | |
1793 | * otherwise the creation of further caches will BUG(). | |
1794 | */ | |
1795 | set_up_list3s(cachep, SIZE_AC); | |
1796 | if (INDEX_AC == INDEX_L3) | |
1797 | g_cpucache_up = PARTIAL_L3; | |
1798 | else | |
1799 | g_cpucache_up = PARTIAL_AC; | |
1800 | } else { | |
1801 | cachep->array[smp_processor_id()] = | |
1802 | kmalloc(sizeof(struct arraycache_init), GFP_KERNEL); | |
1803 | ||
1804 | if (g_cpucache_up == PARTIAL_AC) { | |
1805 | set_up_list3s(cachep, SIZE_L3); | |
1806 | g_cpucache_up = PARTIAL_L3; | |
1807 | } else { | |
1808 | int node; | |
1809 | for_each_online_node(node) { | |
1810 | cachep->nodelists[node] = | |
1811 | kmalloc_node(sizeof(struct kmem_list3), | |
1812 | GFP_KERNEL, node); | |
1813 | BUG_ON(!cachep->nodelists[node]); | |
1814 | kmem_list3_init(cachep->nodelists[node]); | |
1815 | } | |
1816 | } | |
1817 | } | |
1818 | cachep->nodelists[numa_node_id()]->next_reap = | |
1819 | jiffies + REAPTIMEOUT_LIST3 + | |
1820 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1821 | ||
1822 | cpu_cache_get(cachep)->avail = 0; | |
1823 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1824 | cpu_cache_get(cachep)->batchcount = 1; | |
1825 | cpu_cache_get(cachep)->touched = 0; | |
1826 | cachep->batchcount = 1; | |
1827 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
1828 | } | |
1829 | ||
1da177e4 LT |
1830 | /** |
1831 | * kmem_cache_create - Create a cache. | |
1832 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
1833 | * @size: The size of objects to be created in this cache. | |
1834 | * @align: The required alignment for the objects. | |
1835 | * @flags: SLAB flags | |
1836 | * @ctor: A constructor for the objects. | |
1837 | * @dtor: A destructor for the objects. | |
1838 | * | |
1839 | * Returns a ptr to the cache on success, NULL on failure. | |
1840 | * Cannot be called within a int, but can be interrupted. | |
1841 | * The @ctor is run when new pages are allocated by the cache | |
1842 | * and the @dtor is run before the pages are handed back. | |
1843 | * | |
1844 | * @name must be valid until the cache is destroyed. This implies that | |
a737b3e2 AM |
1845 | * the module calling this has to destroy the cache before getting unloaded. |
1846 | * | |
1da177e4 LT |
1847 | * The flags are |
1848 | * | |
1849 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1850 | * to catch references to uninitialised memory. | |
1851 | * | |
1852 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1853 | * for buffer overruns. | |
1854 | * | |
1da177e4 LT |
1855 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1856 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1857 | * as davem. | |
1858 | */ | |
343e0d7a | 1859 | struct kmem_cache * |
1da177e4 | 1860 | kmem_cache_create (const char *name, size_t size, size_t align, |
a737b3e2 AM |
1861 | unsigned long flags, |
1862 | void (*ctor)(void*, struct kmem_cache *, unsigned long), | |
343e0d7a | 1863 | void (*dtor)(void*, struct kmem_cache *, unsigned long)) |
1da177e4 LT |
1864 | { |
1865 | size_t left_over, slab_size, ralign; | |
343e0d7a | 1866 | struct kmem_cache *cachep = NULL; |
4f12bb4f | 1867 | struct list_head *p; |
1da177e4 LT |
1868 | |
1869 | /* | |
1870 | * Sanity checks... these are all serious usage bugs. | |
1871 | */ | |
a737b3e2 | 1872 | if (!name || in_interrupt() || (size < BYTES_PER_WORD) || |
b28a02de | 1873 | (size > (1 << MAX_OBJ_ORDER) * PAGE_SIZE) || (dtor && !ctor)) { |
a737b3e2 AM |
1874 | printk(KERN_ERR "%s: Early error in slab %s\n", __FUNCTION__, |
1875 | name); | |
b28a02de PE |
1876 | BUG(); |
1877 | } | |
1da177e4 | 1878 | |
f0188f47 RT |
1879 | /* |
1880 | * Prevent CPUs from coming and going. | |
1881 | * lock_cpu_hotplug() nests outside cache_chain_mutex | |
1882 | */ | |
1883 | lock_cpu_hotplug(); | |
1884 | ||
fc0abb14 | 1885 | mutex_lock(&cache_chain_mutex); |
4f12bb4f AM |
1886 | |
1887 | list_for_each(p, &cache_chain) { | |
343e0d7a | 1888 | struct kmem_cache *pc = list_entry(p, struct kmem_cache, next); |
4f12bb4f AM |
1889 | mm_segment_t old_fs = get_fs(); |
1890 | char tmp; | |
1891 | int res; | |
1892 | ||
1893 | /* | |
1894 | * This happens when the module gets unloaded and doesn't | |
1895 | * destroy its slab cache and no-one else reuses the vmalloc | |
1896 | * area of the module. Print a warning. | |
1897 | */ | |
1898 | set_fs(KERNEL_DS); | |
1899 | res = __get_user(tmp, pc->name); | |
1900 | set_fs(old_fs); | |
1901 | if (res) { | |
1902 | printk("SLAB: cache with size %d has lost its name\n", | |
3dafccf2 | 1903 | pc->buffer_size); |
4f12bb4f AM |
1904 | continue; |
1905 | } | |
1906 | ||
b28a02de | 1907 | if (!strcmp(pc->name, name)) { |
4f12bb4f AM |
1908 | printk("kmem_cache_create: duplicate cache %s\n", name); |
1909 | dump_stack(); | |
1910 | goto oops; | |
1911 | } | |
1912 | } | |
1913 | ||
1da177e4 LT |
1914 | #if DEBUG |
1915 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
1916 | if ((flags & SLAB_DEBUG_INITIAL) && !ctor) { | |
1917 | /* No constructor, but inital state check requested */ | |
1918 | printk(KERN_ERR "%s: No con, but init state check " | |
b28a02de | 1919 | "requested - %s\n", __FUNCTION__, name); |
1da177e4 LT |
1920 | flags &= ~SLAB_DEBUG_INITIAL; |
1921 | } | |
1da177e4 LT |
1922 | #if FORCED_DEBUG |
1923 | /* | |
1924 | * Enable redzoning and last user accounting, except for caches with | |
1925 | * large objects, if the increased size would increase the object size | |
1926 | * above the next power of two: caches with object sizes just above a | |
1927 | * power of two have a significant amount of internal fragmentation. | |
1928 | */ | |
a737b3e2 | 1929 | if (size < 4096 || fls(size - 1) == fls(size-1 + 3 * BYTES_PER_WORD)) |
b28a02de | 1930 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
1931 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
1932 | flags |= SLAB_POISON; | |
1933 | #endif | |
1934 | if (flags & SLAB_DESTROY_BY_RCU) | |
1935 | BUG_ON(flags & SLAB_POISON); | |
1936 | #endif | |
1937 | if (flags & SLAB_DESTROY_BY_RCU) | |
1938 | BUG_ON(dtor); | |
1939 | ||
1940 | /* | |
a737b3e2 AM |
1941 | * Always checks flags, a caller might be expecting debug support which |
1942 | * isn't available. | |
1da177e4 LT |
1943 | */ |
1944 | if (flags & ~CREATE_MASK) | |
1945 | BUG(); | |
1946 | ||
a737b3e2 AM |
1947 | /* |
1948 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
1949 | * unaligned accesses for some archs when redzoning is used, and makes |
1950 | * sure any on-slab bufctl's are also correctly aligned. | |
1951 | */ | |
b28a02de PE |
1952 | if (size & (BYTES_PER_WORD - 1)) { |
1953 | size += (BYTES_PER_WORD - 1); | |
1954 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
1955 | } |
1956 | ||
a737b3e2 AM |
1957 | /* calculate the final buffer alignment: */ |
1958 | ||
1da177e4 LT |
1959 | /* 1) arch recommendation: can be overridden for debug */ |
1960 | if (flags & SLAB_HWCACHE_ALIGN) { | |
a737b3e2 AM |
1961 | /* |
1962 | * Default alignment: as specified by the arch code. Except if | |
1963 | * an object is really small, then squeeze multiple objects into | |
1964 | * one cacheline. | |
1da177e4 LT |
1965 | */ |
1966 | ralign = cache_line_size(); | |
b28a02de | 1967 | while (size <= ralign / 2) |
1da177e4 LT |
1968 | ralign /= 2; |
1969 | } else { | |
1970 | ralign = BYTES_PER_WORD; | |
1971 | } | |
1972 | /* 2) arch mandated alignment: disables debug if necessary */ | |
1973 | if (ralign < ARCH_SLAB_MINALIGN) { | |
1974 | ralign = ARCH_SLAB_MINALIGN; | |
1975 | if (ralign > BYTES_PER_WORD) | |
b28a02de | 1976 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
1da177e4 LT |
1977 | } |
1978 | /* 3) caller mandated alignment: disables debug if necessary */ | |
1979 | if (ralign < align) { | |
1980 | ralign = align; | |
1981 | if (ralign > BYTES_PER_WORD) | |
b28a02de | 1982 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
1da177e4 | 1983 | } |
a737b3e2 AM |
1984 | /* |
1985 | * 4) Store it. Note that the debug code below can reduce | |
1da177e4 LT |
1986 | * the alignment to BYTES_PER_WORD. |
1987 | */ | |
1988 | align = ralign; | |
1989 | ||
1990 | /* Get cache's description obj. */ | |
343e0d7a | 1991 | cachep = kmem_cache_alloc(&cache_cache, SLAB_KERNEL); |
1da177e4 | 1992 | if (!cachep) |
4f12bb4f | 1993 | goto oops; |
343e0d7a | 1994 | memset(cachep, 0, sizeof(struct kmem_cache)); |
1da177e4 LT |
1995 | |
1996 | #if DEBUG | |
3dafccf2 | 1997 | cachep->obj_size = size; |
1da177e4 LT |
1998 | |
1999 | if (flags & SLAB_RED_ZONE) { | |
2000 | /* redzoning only works with word aligned caches */ | |
2001 | align = BYTES_PER_WORD; | |
2002 | ||
2003 | /* add space for red zone words */ | |
3dafccf2 | 2004 | cachep->obj_offset += BYTES_PER_WORD; |
b28a02de | 2005 | size += 2 * BYTES_PER_WORD; |
1da177e4 LT |
2006 | } |
2007 | if (flags & SLAB_STORE_USER) { | |
2008 | /* user store requires word alignment and | |
2009 | * one word storage behind the end of the real | |
2010 | * object. | |
2011 | */ | |
2012 | align = BYTES_PER_WORD; | |
2013 | size += BYTES_PER_WORD; | |
2014 | } | |
2015 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
b28a02de | 2016 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
3dafccf2 MS |
2017 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
2018 | cachep->obj_offset += PAGE_SIZE - size; | |
1da177e4 LT |
2019 | size = PAGE_SIZE; |
2020 | } | |
2021 | #endif | |
2022 | #endif | |
2023 | ||
2024 | /* Determine if the slab management is 'on' or 'off' slab. */ | |
b28a02de | 2025 | if (size >= (PAGE_SIZE >> 3)) |
1da177e4 LT |
2026 | /* |
2027 | * Size is large, assume best to place the slab management obj | |
2028 | * off-slab (should allow better packing of objs). | |
2029 | */ | |
2030 | flags |= CFLGS_OFF_SLAB; | |
2031 | ||
2032 | size = ALIGN(size, align); | |
2033 | ||
f78bb8ad | 2034 | left_over = calculate_slab_order(cachep, size, align, flags); |
1da177e4 LT |
2035 | |
2036 | if (!cachep->num) { | |
2037 | printk("kmem_cache_create: couldn't create cache %s.\n", name); | |
2038 | kmem_cache_free(&cache_cache, cachep); | |
2039 | cachep = NULL; | |
4f12bb4f | 2040 | goto oops; |
1da177e4 | 2041 | } |
b28a02de PE |
2042 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
2043 | + sizeof(struct slab), align); | |
1da177e4 LT |
2044 | |
2045 | /* | |
2046 | * If the slab has been placed off-slab, and we have enough space then | |
2047 | * move it on-slab. This is at the expense of any extra colouring. | |
2048 | */ | |
2049 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { | |
2050 | flags &= ~CFLGS_OFF_SLAB; | |
2051 | left_over -= slab_size; | |
2052 | } | |
2053 | ||
2054 | if (flags & CFLGS_OFF_SLAB) { | |
2055 | /* really off slab. No need for manual alignment */ | |
b28a02de PE |
2056 | slab_size = |
2057 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); | |
1da177e4 LT |
2058 | } |
2059 | ||
2060 | cachep->colour_off = cache_line_size(); | |
2061 | /* Offset must be a multiple of the alignment. */ | |
2062 | if (cachep->colour_off < align) | |
2063 | cachep->colour_off = align; | |
b28a02de | 2064 | cachep->colour = left_over / cachep->colour_off; |
1da177e4 LT |
2065 | cachep->slab_size = slab_size; |
2066 | cachep->flags = flags; | |
2067 | cachep->gfpflags = 0; | |
2068 | if (flags & SLAB_CACHE_DMA) | |
2069 | cachep->gfpflags |= GFP_DMA; | |
3dafccf2 | 2070 | cachep->buffer_size = size; |
1da177e4 LT |
2071 | |
2072 | if (flags & CFLGS_OFF_SLAB) | |
b2d55073 | 2073 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
1da177e4 LT |
2074 | cachep->ctor = ctor; |
2075 | cachep->dtor = dtor; | |
2076 | cachep->name = name; | |
2077 | ||
1da177e4 | 2078 | |
f30cf7d1 | 2079 | setup_cpu_cache(cachep); |
1da177e4 | 2080 | |
1da177e4 LT |
2081 | /* cache setup completed, link it into the list */ |
2082 | list_add(&cachep->next, &cache_chain); | |
a737b3e2 | 2083 | oops: |
1da177e4 LT |
2084 | if (!cachep && (flags & SLAB_PANIC)) |
2085 | panic("kmem_cache_create(): failed to create slab `%s'\n", | |
b28a02de | 2086 | name); |
fc0abb14 | 2087 | mutex_unlock(&cache_chain_mutex); |
f0188f47 | 2088 | unlock_cpu_hotplug(); |
1da177e4 LT |
2089 | return cachep; |
2090 | } | |
2091 | EXPORT_SYMBOL(kmem_cache_create); | |
2092 | ||
2093 | #if DEBUG | |
2094 | static void check_irq_off(void) | |
2095 | { | |
2096 | BUG_ON(!irqs_disabled()); | |
2097 | } | |
2098 | ||
2099 | static void check_irq_on(void) | |
2100 | { | |
2101 | BUG_ON(irqs_disabled()); | |
2102 | } | |
2103 | ||
343e0d7a | 2104 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2105 | { |
2106 | #ifdef CONFIG_SMP | |
2107 | check_irq_off(); | |
e498be7d | 2108 | assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); |
1da177e4 LT |
2109 | #endif |
2110 | } | |
e498be7d | 2111 | |
343e0d7a | 2112 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2113 | { |
2114 | #ifdef CONFIG_SMP | |
2115 | check_irq_off(); | |
2116 | assert_spin_locked(&cachep->nodelists[node]->list_lock); | |
2117 | #endif | |
2118 | } | |
2119 | ||
1da177e4 LT |
2120 | #else |
2121 | #define check_irq_off() do { } while(0) | |
2122 | #define check_irq_on() do { } while(0) | |
2123 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2124 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2125 | #endif |
2126 | ||
aab2207c CL |
2127 | static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, |
2128 | struct array_cache *ac, | |
2129 | int force, int node); | |
2130 | ||
1da177e4 LT |
2131 | static void do_drain(void *arg) |
2132 | { | |
a737b3e2 | 2133 | struct kmem_cache *cachep = arg; |
1da177e4 | 2134 | struct array_cache *ac; |
ff69416e | 2135 | int node = numa_node_id(); |
1da177e4 LT |
2136 | |
2137 | check_irq_off(); | |
9a2dba4b | 2138 | ac = cpu_cache_get(cachep); |
ff69416e CL |
2139 | spin_lock(&cachep->nodelists[node]->list_lock); |
2140 | free_block(cachep, ac->entry, ac->avail, node); | |
2141 | spin_unlock(&cachep->nodelists[node]->list_lock); | |
1da177e4 LT |
2142 | ac->avail = 0; |
2143 | } | |
2144 | ||
343e0d7a | 2145 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2146 | { |
e498be7d CL |
2147 | struct kmem_list3 *l3; |
2148 | int node; | |
2149 | ||
a07fa394 | 2150 | on_each_cpu(do_drain, cachep, 1, 1); |
1da177e4 | 2151 | check_irq_on(); |
b28a02de | 2152 | for_each_online_node(node) { |
e498be7d CL |
2153 | l3 = cachep->nodelists[node]; |
2154 | if (l3) { | |
aab2207c | 2155 | drain_array(cachep, l3, l3->shared, 1, node); |
e498be7d | 2156 | if (l3->alien) |
4484ebf1 | 2157 | drain_alien_cache(cachep, l3->alien); |
e498be7d CL |
2158 | } |
2159 | } | |
1da177e4 LT |
2160 | } |
2161 | ||
343e0d7a | 2162 | static int __node_shrink(struct kmem_cache *cachep, int node) |
1da177e4 LT |
2163 | { |
2164 | struct slab *slabp; | |
e498be7d | 2165 | struct kmem_list3 *l3 = cachep->nodelists[node]; |
1da177e4 LT |
2166 | int ret; |
2167 | ||
e498be7d | 2168 | for (;;) { |
1da177e4 LT |
2169 | struct list_head *p; |
2170 | ||
e498be7d CL |
2171 | p = l3->slabs_free.prev; |
2172 | if (p == &l3->slabs_free) | |
1da177e4 LT |
2173 | break; |
2174 | ||
e498be7d | 2175 | slabp = list_entry(l3->slabs_free.prev, struct slab, list); |
1da177e4 LT |
2176 | #if DEBUG |
2177 | if (slabp->inuse) | |
2178 | BUG(); | |
2179 | #endif | |
2180 | list_del(&slabp->list); | |
2181 | ||
e498be7d CL |
2182 | l3->free_objects -= cachep->num; |
2183 | spin_unlock_irq(&l3->list_lock); | |
1da177e4 | 2184 | slab_destroy(cachep, slabp); |
e498be7d | 2185 | spin_lock_irq(&l3->list_lock); |
1da177e4 | 2186 | } |
b28a02de | 2187 | ret = !list_empty(&l3->slabs_full) || !list_empty(&l3->slabs_partial); |
1da177e4 LT |
2188 | return ret; |
2189 | } | |
2190 | ||
343e0d7a | 2191 | static int __cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2192 | { |
2193 | int ret = 0, i = 0; | |
2194 | struct kmem_list3 *l3; | |
2195 | ||
2196 | drain_cpu_caches(cachep); | |
2197 | ||
2198 | check_irq_on(); | |
2199 | for_each_online_node(i) { | |
2200 | l3 = cachep->nodelists[i]; | |
2201 | if (l3) { | |
2202 | spin_lock_irq(&l3->list_lock); | |
2203 | ret += __node_shrink(cachep, i); | |
2204 | spin_unlock_irq(&l3->list_lock); | |
2205 | } | |
2206 | } | |
2207 | return (ret ? 1 : 0); | |
2208 | } | |
2209 | ||
1da177e4 LT |
2210 | /** |
2211 | * kmem_cache_shrink - Shrink a cache. | |
2212 | * @cachep: The cache to shrink. | |
2213 | * | |
2214 | * Releases as many slabs as possible for a cache. | |
2215 | * To help debugging, a zero exit status indicates all slabs were released. | |
2216 | */ | |
343e0d7a | 2217 | int kmem_cache_shrink(struct kmem_cache *cachep) |
1da177e4 LT |
2218 | { |
2219 | if (!cachep || in_interrupt()) | |
2220 | BUG(); | |
2221 | ||
2222 | return __cache_shrink(cachep); | |
2223 | } | |
2224 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2225 | ||
2226 | /** | |
2227 | * kmem_cache_destroy - delete a cache | |
2228 | * @cachep: the cache to destroy | |
2229 | * | |
343e0d7a | 2230 | * Remove a struct kmem_cache object from the slab cache. |
1da177e4 LT |
2231 | * Returns 0 on success. |
2232 | * | |
2233 | * It is expected this function will be called by a module when it is | |
2234 | * unloaded. This will remove the cache completely, and avoid a duplicate | |
2235 | * cache being allocated each time a module is loaded and unloaded, if the | |
2236 | * module doesn't have persistent in-kernel storage across loads and unloads. | |
2237 | * | |
2238 | * The cache must be empty before calling this function. | |
2239 | * | |
2240 | * The caller must guarantee that noone will allocate memory from the cache | |
2241 | * during the kmem_cache_destroy(). | |
2242 | */ | |
343e0d7a | 2243 | int kmem_cache_destroy(struct kmem_cache *cachep) |
1da177e4 LT |
2244 | { |
2245 | int i; | |
e498be7d | 2246 | struct kmem_list3 *l3; |
1da177e4 LT |
2247 | |
2248 | if (!cachep || in_interrupt()) | |
2249 | BUG(); | |
2250 | ||
2251 | /* Don't let CPUs to come and go */ | |
2252 | lock_cpu_hotplug(); | |
2253 | ||
2254 | /* Find the cache in the chain of caches. */ | |
fc0abb14 | 2255 | mutex_lock(&cache_chain_mutex); |
1da177e4 LT |
2256 | /* |
2257 | * the chain is never empty, cache_cache is never destroyed | |
2258 | */ | |
2259 | list_del(&cachep->next); | |
fc0abb14 | 2260 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
2261 | |
2262 | if (__cache_shrink(cachep)) { | |
2263 | slab_error(cachep, "Can't free all objects"); | |
fc0abb14 | 2264 | mutex_lock(&cache_chain_mutex); |
b28a02de | 2265 | list_add(&cachep->next, &cache_chain); |
fc0abb14 | 2266 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
2267 | unlock_cpu_hotplug(); |
2268 | return 1; | |
2269 | } | |
2270 | ||
2271 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
fbd568a3 | 2272 | synchronize_rcu(); |
1da177e4 | 2273 | |
e498be7d | 2274 | for_each_online_cpu(i) |
b28a02de | 2275 | kfree(cachep->array[i]); |
1da177e4 LT |
2276 | |
2277 | /* NUMA: free the list3 structures */ | |
e498be7d | 2278 | for_each_online_node(i) { |
a737b3e2 AM |
2279 | l3 = cachep->nodelists[i]; |
2280 | if (l3) { | |
e498be7d CL |
2281 | kfree(l3->shared); |
2282 | free_alien_cache(l3->alien); | |
2283 | kfree(l3); | |
2284 | } | |
2285 | } | |
1da177e4 | 2286 | kmem_cache_free(&cache_cache, cachep); |
1da177e4 | 2287 | unlock_cpu_hotplug(); |
1da177e4 LT |
2288 | return 0; |
2289 | } | |
2290 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2291 | ||
2292 | /* Get the memory for a slab management obj. */ | |
343e0d7a | 2293 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
b28a02de | 2294 | int colour_off, gfp_t local_flags) |
1da177e4 LT |
2295 | { |
2296 | struct slab *slabp; | |
b28a02de | 2297 | |
1da177e4 LT |
2298 | if (OFF_SLAB(cachep)) { |
2299 | /* Slab management obj is off-slab. */ | |
2300 | slabp = kmem_cache_alloc(cachep->slabp_cache, local_flags); | |
2301 | if (!slabp) | |
2302 | return NULL; | |
2303 | } else { | |
b28a02de | 2304 | slabp = objp + colour_off; |
1da177e4 LT |
2305 | colour_off += cachep->slab_size; |
2306 | } | |
2307 | slabp->inuse = 0; | |
2308 | slabp->colouroff = colour_off; | |
b28a02de | 2309 | slabp->s_mem = objp + colour_off; |
1da177e4 LT |
2310 | return slabp; |
2311 | } | |
2312 | ||
2313 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) | |
2314 | { | |
b28a02de | 2315 | return (kmem_bufctl_t *) (slabp + 1); |
1da177e4 LT |
2316 | } |
2317 | ||
343e0d7a | 2318 | static void cache_init_objs(struct kmem_cache *cachep, |
b28a02de | 2319 | struct slab *slabp, unsigned long ctor_flags) |
1da177e4 LT |
2320 | { |
2321 | int i; | |
2322 | ||
2323 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2324 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2325 | #if DEBUG |
2326 | /* need to poison the objs? */ | |
2327 | if (cachep->flags & SLAB_POISON) | |
2328 | poison_obj(cachep, objp, POISON_FREE); | |
2329 | if (cachep->flags & SLAB_STORE_USER) | |
2330 | *dbg_userword(cachep, objp) = NULL; | |
2331 | ||
2332 | if (cachep->flags & SLAB_RED_ZONE) { | |
2333 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2334 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2335 | } | |
2336 | /* | |
a737b3e2 AM |
2337 | * Constructors are not allowed to allocate memory from the same |
2338 | * cache which they are a constructor for. Otherwise, deadlock. | |
2339 | * They must also be threaded. | |
1da177e4 LT |
2340 | */ |
2341 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
3dafccf2 | 2342 | cachep->ctor(objp + obj_offset(cachep), cachep, |
b28a02de | 2343 | ctor_flags); |
1da177e4 LT |
2344 | |
2345 | if (cachep->flags & SLAB_RED_ZONE) { | |
2346 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2347 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2348 | " end of an object"); |
1da177e4 LT |
2349 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2350 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2351 | " start of an object"); |
1da177e4 | 2352 | } |
a737b3e2 AM |
2353 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && |
2354 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) | |
b28a02de | 2355 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2356 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2357 | #else |
2358 | if (cachep->ctor) | |
2359 | cachep->ctor(objp, cachep, ctor_flags); | |
2360 | #endif | |
b28a02de | 2361 | slab_bufctl(slabp)[i] = i + 1; |
1da177e4 | 2362 | } |
b28a02de | 2363 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
1da177e4 LT |
2364 | slabp->free = 0; |
2365 | } | |
2366 | ||
343e0d7a | 2367 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2368 | { |
a737b3e2 AM |
2369 | if (flags & SLAB_DMA) |
2370 | BUG_ON(!(cachep->gfpflags & GFP_DMA)); | |
2371 | else | |
2372 | BUG_ON(cachep->gfpflags & GFP_DMA); | |
1da177e4 LT |
2373 | } |
2374 | ||
a737b3e2 AM |
2375 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
2376 | int nodeid) | |
78d382d7 | 2377 | { |
8fea4e96 | 2378 | void *objp = index_to_obj(cachep, slabp, slabp->free); |
78d382d7 MD |
2379 | kmem_bufctl_t next; |
2380 | ||
2381 | slabp->inuse++; | |
2382 | next = slab_bufctl(slabp)[slabp->free]; | |
2383 | #if DEBUG | |
2384 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | |
2385 | WARN_ON(slabp->nodeid != nodeid); | |
2386 | #endif | |
2387 | slabp->free = next; | |
2388 | ||
2389 | return objp; | |
2390 | } | |
2391 | ||
a737b3e2 AM |
2392 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
2393 | void *objp, int nodeid) | |
78d382d7 | 2394 | { |
8fea4e96 | 2395 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
78d382d7 MD |
2396 | |
2397 | #if DEBUG | |
2398 | /* Verify that the slab belongs to the intended node */ | |
2399 | WARN_ON(slabp->nodeid != nodeid); | |
2400 | ||
2401 | if (slab_bufctl(slabp)[objnr] != BUFCTL_FREE) { | |
2402 | printk(KERN_ERR "slab: double free detected in cache " | |
a737b3e2 | 2403 | "'%s', objp %p\n", cachep->name, objp); |
78d382d7 MD |
2404 | BUG(); |
2405 | } | |
2406 | #endif | |
2407 | slab_bufctl(slabp)[objnr] = slabp->free; | |
2408 | slabp->free = objnr; | |
2409 | slabp->inuse--; | |
2410 | } | |
2411 | ||
a737b3e2 AM |
2412 | static void set_slab_attr(struct kmem_cache *cachep, struct slab *slabp, |
2413 | void *objp) | |
1da177e4 LT |
2414 | { |
2415 | int i; | |
2416 | struct page *page; | |
2417 | ||
2418 | /* Nasty!!!!!! I hope this is OK. */ | |
1da177e4 | 2419 | page = virt_to_page(objp); |
84097518 NP |
2420 | |
2421 | i = 1; | |
2422 | if (likely(!PageCompound(page))) | |
2423 | i <<= cachep->gfporder; | |
1da177e4 | 2424 | do { |
065d41cb PE |
2425 | page_set_cache(page, cachep); |
2426 | page_set_slab(page, slabp); | |
1da177e4 LT |
2427 | page++; |
2428 | } while (--i); | |
2429 | } | |
2430 | ||
2431 | /* | |
2432 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2433 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2434 | */ | |
343e0d7a | 2435 | static int cache_grow(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1da177e4 | 2436 | { |
b28a02de PE |
2437 | struct slab *slabp; |
2438 | void *objp; | |
2439 | size_t offset; | |
2440 | gfp_t local_flags; | |
2441 | unsigned long ctor_flags; | |
e498be7d | 2442 | struct kmem_list3 *l3; |
1da177e4 | 2443 | |
a737b3e2 AM |
2444 | /* |
2445 | * Be lazy and only check for valid flags here, keeping it out of the | |
2446 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2447 | */ |
b28a02de | 2448 | if (flags & ~(SLAB_DMA | SLAB_LEVEL_MASK | SLAB_NO_GROW)) |
1da177e4 LT |
2449 | BUG(); |
2450 | if (flags & SLAB_NO_GROW) | |
2451 | return 0; | |
2452 | ||
2453 | ctor_flags = SLAB_CTOR_CONSTRUCTOR; | |
2454 | local_flags = (flags & SLAB_LEVEL_MASK); | |
2455 | if (!(local_flags & __GFP_WAIT)) | |
2456 | /* | |
2457 | * Not allowed to sleep. Need to tell a constructor about | |
2458 | * this - it might need to know... | |
2459 | */ | |
2460 | ctor_flags |= SLAB_CTOR_ATOMIC; | |
2461 | ||
2e1217cf | 2462 | /* Take the l3 list lock to change the colour_next on this node */ |
1da177e4 | 2463 | check_irq_off(); |
2e1217cf RT |
2464 | l3 = cachep->nodelists[nodeid]; |
2465 | spin_lock(&l3->list_lock); | |
1da177e4 LT |
2466 | |
2467 | /* Get colour for the slab, and cal the next value. */ | |
2e1217cf RT |
2468 | offset = l3->colour_next; |
2469 | l3->colour_next++; | |
2470 | if (l3->colour_next >= cachep->colour) | |
2471 | l3->colour_next = 0; | |
2472 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2473 | |
2e1217cf | 2474 | offset *= cachep->colour_off; |
1da177e4 LT |
2475 | |
2476 | if (local_flags & __GFP_WAIT) | |
2477 | local_irq_enable(); | |
2478 | ||
2479 | /* | |
2480 | * The test for missing atomic flag is performed here, rather than | |
2481 | * the more obvious place, simply to reduce the critical path length | |
2482 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2483 | * will eventually be caught here (where it matters). | |
2484 | */ | |
2485 | kmem_flagcheck(cachep, flags); | |
2486 | ||
a737b3e2 AM |
2487 | /* |
2488 | * Get mem for the objs. Attempt to allocate a physical page from | |
2489 | * 'nodeid'. | |
e498be7d | 2490 | */ |
a737b3e2 AM |
2491 | objp = kmem_getpages(cachep, flags, nodeid); |
2492 | if (!objp) | |
1da177e4 LT |
2493 | goto failed; |
2494 | ||
2495 | /* Get slab management. */ | |
a737b3e2 AM |
2496 | slabp = alloc_slabmgmt(cachep, objp, offset, local_flags); |
2497 | if (!slabp) | |
1da177e4 LT |
2498 | goto opps1; |
2499 | ||
e498be7d | 2500 | slabp->nodeid = nodeid; |
1da177e4 LT |
2501 | set_slab_attr(cachep, slabp, objp); |
2502 | ||
2503 | cache_init_objs(cachep, slabp, ctor_flags); | |
2504 | ||
2505 | if (local_flags & __GFP_WAIT) | |
2506 | local_irq_disable(); | |
2507 | check_irq_off(); | |
e498be7d | 2508 | spin_lock(&l3->list_lock); |
1da177e4 LT |
2509 | |
2510 | /* Make slab active. */ | |
e498be7d | 2511 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
1da177e4 | 2512 | STATS_INC_GROWN(cachep); |
e498be7d CL |
2513 | l3->free_objects += cachep->num; |
2514 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2515 | return 1; |
a737b3e2 | 2516 | opps1: |
1da177e4 | 2517 | kmem_freepages(cachep, objp); |
a737b3e2 | 2518 | failed: |
1da177e4 LT |
2519 | if (local_flags & __GFP_WAIT) |
2520 | local_irq_disable(); | |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | #if DEBUG | |
2525 | ||
2526 | /* | |
2527 | * Perform extra freeing checks: | |
2528 | * - detect bad pointers. | |
2529 | * - POISON/RED_ZONE checking | |
2530 | * - destructor calls, for caches with POISON+dtor | |
2531 | */ | |
2532 | static void kfree_debugcheck(const void *objp) | |
2533 | { | |
2534 | struct page *page; | |
2535 | ||
2536 | if (!virt_addr_valid(objp)) { | |
2537 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2538 | (unsigned long)objp); |
2539 | BUG(); | |
1da177e4 LT |
2540 | } |
2541 | page = virt_to_page(objp); | |
2542 | if (!PageSlab(page)) { | |
b28a02de PE |
2543 | printk(KERN_ERR "kfree_debugcheck: bad ptr %lxh.\n", |
2544 | (unsigned long)objp); | |
1da177e4 LT |
2545 | BUG(); |
2546 | } | |
2547 | } | |
2548 | ||
343e0d7a | 2549 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
b28a02de | 2550 | void *caller) |
1da177e4 LT |
2551 | { |
2552 | struct page *page; | |
2553 | unsigned int objnr; | |
2554 | struct slab *slabp; | |
2555 | ||
3dafccf2 | 2556 | objp -= obj_offset(cachep); |
1da177e4 LT |
2557 | kfree_debugcheck(objp); |
2558 | page = virt_to_page(objp); | |
2559 | ||
065d41cb | 2560 | if (page_get_cache(page) != cachep) { |
a737b3e2 AM |
2561 | printk(KERN_ERR "mismatch in kmem_cache_free: expected " |
2562 | "cache %p, got %p\n", | |
b28a02de | 2563 | page_get_cache(page), cachep); |
1da177e4 | 2564 | printk(KERN_ERR "%p is %s.\n", cachep, cachep->name); |
b28a02de PE |
2565 | printk(KERN_ERR "%p is %s.\n", page_get_cache(page), |
2566 | page_get_cache(page)->name); | |
1da177e4 LT |
2567 | WARN_ON(1); |
2568 | } | |
065d41cb | 2569 | slabp = page_get_slab(page); |
1da177e4 LT |
2570 | |
2571 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2572 | if (*dbg_redzone1(cachep, objp) != RED_ACTIVE || |
2573 | *dbg_redzone2(cachep, objp) != RED_ACTIVE) { | |
2574 | slab_error(cachep, "double free, or memory outside" | |
2575 | " object was overwritten"); | |
2576 | printk(KERN_ERR "%p: redzone 1:0x%lx, " | |
2577 | "redzone 2:0x%lx.\n", | |
b28a02de PE |
2578 | objp, *dbg_redzone1(cachep, objp), |
2579 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2580 | } |
2581 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2582 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2583 | } | |
2584 | if (cachep->flags & SLAB_STORE_USER) | |
2585 | *dbg_userword(cachep, objp) = caller; | |
2586 | ||
8fea4e96 | 2587 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 LT |
2588 | |
2589 | BUG_ON(objnr >= cachep->num); | |
8fea4e96 | 2590 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
1da177e4 LT |
2591 | |
2592 | if (cachep->flags & SLAB_DEBUG_INITIAL) { | |
a737b3e2 AM |
2593 | /* |
2594 | * Need to call the slab's constructor so the caller can | |
2595 | * perform a verify of its state (debugging). Called without | |
2596 | * the cache-lock held. | |
1da177e4 | 2597 | */ |
3dafccf2 | 2598 | cachep->ctor(objp + obj_offset(cachep), |
b28a02de | 2599 | cachep, SLAB_CTOR_CONSTRUCTOR | SLAB_CTOR_VERIFY); |
1da177e4 LT |
2600 | } |
2601 | if (cachep->flags & SLAB_POISON && cachep->dtor) { | |
2602 | /* we want to cache poison the object, | |
2603 | * call the destruction callback | |
2604 | */ | |
3dafccf2 | 2605 | cachep->dtor(objp + obj_offset(cachep), cachep, 0); |
1da177e4 LT |
2606 | } |
2607 | if (cachep->flags & SLAB_POISON) { | |
2608 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 | 2609 | if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
1da177e4 | 2610 | store_stackinfo(cachep, objp, (unsigned long)caller); |
b28a02de | 2611 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2612 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2613 | } else { |
2614 | poison_obj(cachep, objp, POISON_FREE); | |
2615 | } | |
2616 | #else | |
2617 | poison_obj(cachep, objp, POISON_FREE); | |
2618 | #endif | |
2619 | } | |
2620 | return objp; | |
2621 | } | |
2622 | ||
343e0d7a | 2623 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 LT |
2624 | { |
2625 | kmem_bufctl_t i; | |
2626 | int entries = 0; | |
b28a02de | 2627 | |
1da177e4 LT |
2628 | /* Check slab's freelist to see if this obj is there. */ |
2629 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { | |
2630 | entries++; | |
2631 | if (entries > cachep->num || i >= cachep->num) | |
2632 | goto bad; | |
2633 | } | |
2634 | if (entries != cachep->num - slabp->inuse) { | |
a737b3e2 AM |
2635 | bad: |
2636 | printk(KERN_ERR "slab: Internal list corruption detected in " | |
2637 | "cache '%s'(%d), slabp %p(%d). Hexdump:\n", | |
2638 | cachep->name, cachep->num, slabp, slabp->inuse); | |
b28a02de | 2639 | for (i = 0; |
264132bc | 2640 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
b28a02de | 2641 | i++) { |
a737b3e2 | 2642 | if (i % 16 == 0) |
1da177e4 | 2643 | printk("\n%03x:", i); |
b28a02de | 2644 | printk(" %02x", ((unsigned char *)slabp)[i]); |
1da177e4 LT |
2645 | } |
2646 | printk("\n"); | |
2647 | BUG(); | |
2648 | } | |
2649 | } | |
2650 | #else | |
2651 | #define kfree_debugcheck(x) do { } while(0) | |
2652 | #define cache_free_debugcheck(x,objp,z) (objp) | |
2653 | #define check_slabp(x,y) do { } while(0) | |
2654 | #endif | |
2655 | ||
343e0d7a | 2656 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2657 | { |
2658 | int batchcount; | |
2659 | struct kmem_list3 *l3; | |
2660 | struct array_cache *ac; | |
2661 | ||
2662 | check_irq_off(); | |
9a2dba4b | 2663 | ac = cpu_cache_get(cachep); |
a737b3e2 | 2664 | retry: |
1da177e4 LT |
2665 | batchcount = ac->batchcount; |
2666 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2667 | /* |
2668 | * If there was little recent activity on this cache, then | |
2669 | * perform only a partial refill. Otherwise we could generate | |
2670 | * refill bouncing. | |
1da177e4 LT |
2671 | */ |
2672 | batchcount = BATCHREFILL_LIMIT; | |
2673 | } | |
e498be7d CL |
2674 | l3 = cachep->nodelists[numa_node_id()]; |
2675 | ||
2676 | BUG_ON(ac->avail > 0 || !l3); | |
2677 | spin_lock(&l3->list_lock); | |
1da177e4 | 2678 | |
1da177e4 LT |
2679 | if (l3->shared) { |
2680 | struct array_cache *shared_array = l3->shared; | |
2681 | if (shared_array->avail) { | |
2682 | if (batchcount > shared_array->avail) | |
2683 | batchcount = shared_array->avail; | |
2684 | shared_array->avail -= batchcount; | |
2685 | ac->avail = batchcount; | |
e498be7d | 2686 | memcpy(ac->entry, |
b28a02de PE |
2687 | &(shared_array->entry[shared_array->avail]), |
2688 | sizeof(void *) * batchcount); | |
1da177e4 LT |
2689 | shared_array->touched = 1; |
2690 | goto alloc_done; | |
2691 | } | |
2692 | } | |
2693 | while (batchcount > 0) { | |
2694 | struct list_head *entry; | |
2695 | struct slab *slabp; | |
2696 | /* Get slab alloc is to come from. */ | |
2697 | entry = l3->slabs_partial.next; | |
2698 | if (entry == &l3->slabs_partial) { | |
2699 | l3->free_touched = 1; | |
2700 | entry = l3->slabs_free.next; | |
2701 | if (entry == &l3->slabs_free) | |
2702 | goto must_grow; | |
2703 | } | |
2704 | ||
2705 | slabp = list_entry(entry, struct slab, list); | |
2706 | check_slabp(cachep, slabp); | |
2707 | check_spinlock_acquired(cachep); | |
2708 | while (slabp->inuse < cachep->num && batchcount--) { | |
1da177e4 LT |
2709 | STATS_INC_ALLOCED(cachep); |
2710 | STATS_INC_ACTIVE(cachep); | |
2711 | STATS_SET_HIGH(cachep); | |
2712 | ||
78d382d7 MD |
2713 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
2714 | numa_node_id()); | |
1da177e4 LT |
2715 | } |
2716 | check_slabp(cachep, slabp); | |
2717 | ||
2718 | /* move slabp to correct slabp list: */ | |
2719 | list_del(&slabp->list); | |
2720 | if (slabp->free == BUFCTL_END) | |
2721 | list_add(&slabp->list, &l3->slabs_full); | |
2722 | else | |
2723 | list_add(&slabp->list, &l3->slabs_partial); | |
2724 | } | |
2725 | ||
a737b3e2 | 2726 | must_grow: |
1da177e4 | 2727 | l3->free_objects -= ac->avail; |
a737b3e2 | 2728 | alloc_done: |
e498be7d | 2729 | spin_unlock(&l3->list_lock); |
1da177e4 LT |
2730 | |
2731 | if (unlikely(!ac->avail)) { | |
2732 | int x; | |
e498be7d CL |
2733 | x = cache_grow(cachep, flags, numa_node_id()); |
2734 | ||
a737b3e2 | 2735 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2736 | ac = cpu_cache_get(cachep); |
a737b3e2 | 2737 | if (!x && ac->avail == 0) /* no objects in sight? abort */ |
1da177e4 LT |
2738 | return NULL; |
2739 | ||
a737b3e2 | 2740 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2741 | goto retry; |
2742 | } | |
2743 | ac->touched = 1; | |
e498be7d | 2744 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2745 | } |
2746 | ||
a737b3e2 AM |
2747 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
2748 | gfp_t flags) | |
1da177e4 LT |
2749 | { |
2750 | might_sleep_if(flags & __GFP_WAIT); | |
2751 | #if DEBUG | |
2752 | kmem_flagcheck(cachep, flags); | |
2753 | #endif | |
2754 | } | |
2755 | ||
2756 | #if DEBUG | |
a737b3e2 AM |
2757 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
2758 | gfp_t flags, void *objp, void *caller) | |
1da177e4 | 2759 | { |
b28a02de | 2760 | if (!objp) |
1da177e4 | 2761 | return objp; |
b28a02de | 2762 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2763 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3dafccf2 | 2764 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 2765 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2766 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
2767 | else |
2768 | check_poison_obj(cachep, objp); | |
2769 | #else | |
2770 | check_poison_obj(cachep, objp); | |
2771 | #endif | |
2772 | poison_obj(cachep, objp, POISON_INUSE); | |
2773 | } | |
2774 | if (cachep->flags & SLAB_STORE_USER) | |
2775 | *dbg_userword(cachep, objp) = caller; | |
2776 | ||
2777 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
2778 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
2779 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
2780 | slab_error(cachep, "double free, or memory outside" | |
2781 | " object was overwritten"); | |
b28a02de | 2782 | printk(KERN_ERR |
a737b3e2 AM |
2783 | "%p: redzone 1:0x%lx, redzone 2:0x%lx\n", |
2784 | objp, *dbg_redzone1(cachep, objp), | |
2785 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
2786 | } |
2787 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
2788 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
2789 | } | |
3dafccf2 | 2790 | objp += obj_offset(cachep); |
1da177e4 | 2791 | if (cachep->ctor && cachep->flags & SLAB_POISON) { |
b28a02de | 2792 | unsigned long ctor_flags = SLAB_CTOR_CONSTRUCTOR; |
1da177e4 LT |
2793 | |
2794 | if (!(flags & __GFP_WAIT)) | |
2795 | ctor_flags |= SLAB_CTOR_ATOMIC; | |
2796 | ||
2797 | cachep->ctor(objp, cachep, ctor_flags); | |
b28a02de | 2798 | } |
1da177e4 LT |
2799 | return objp; |
2800 | } | |
2801 | #else | |
2802 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
2803 | #endif | |
2804 | ||
343e0d7a | 2805 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2806 | { |
b28a02de | 2807 | void *objp; |
1da177e4 LT |
2808 | struct array_cache *ac; |
2809 | ||
dc85da15 | 2810 | #ifdef CONFIG_NUMA |
86c562a9 | 2811 | if (unlikely(current->mempolicy && !in_interrupt())) { |
dc85da15 CL |
2812 | int nid = slab_node(current->mempolicy); |
2813 | ||
101a5001 PJ |
2814 | if (nid != numa_node_id()) |
2815 | return __cache_alloc_node(cachep, flags, nid); | |
2816 | } | |
2817 | if (unlikely(cpuset_do_slab_mem_spread() && | |
2818 | (cachep->flags & SLAB_MEM_SPREAD) && | |
2819 | !in_interrupt())) { | |
2820 | int nid = cpuset_mem_spread_node(); | |
2821 | ||
dc85da15 CL |
2822 | if (nid != numa_node_id()) |
2823 | return __cache_alloc_node(cachep, flags, nid); | |
2824 | } | |
2825 | #endif | |
2826 | ||
5c382300 | 2827 | check_irq_off(); |
9a2dba4b | 2828 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2829 | if (likely(ac->avail)) { |
2830 | STATS_INC_ALLOCHIT(cachep); | |
2831 | ac->touched = 1; | |
e498be7d | 2832 | objp = ac->entry[--ac->avail]; |
1da177e4 LT |
2833 | } else { |
2834 | STATS_INC_ALLOCMISS(cachep); | |
2835 | objp = cache_alloc_refill(cachep, flags); | |
2836 | } | |
5c382300 AK |
2837 | return objp; |
2838 | } | |
2839 | ||
a737b3e2 AM |
2840 | static __always_inline void *__cache_alloc(struct kmem_cache *cachep, |
2841 | gfp_t flags, void *caller) | |
5c382300 AK |
2842 | { |
2843 | unsigned long save_flags; | |
b28a02de | 2844 | void *objp; |
5c382300 AK |
2845 | |
2846 | cache_alloc_debugcheck_before(cachep, flags); | |
2847 | ||
2848 | local_irq_save(save_flags); | |
2849 | objp = ____cache_alloc(cachep, flags); | |
1da177e4 | 2850 | local_irq_restore(save_flags); |
34342e86 | 2851 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, |
7fd6b141 | 2852 | caller); |
34342e86 | 2853 | prefetchw(objp); |
1da177e4 LT |
2854 | return objp; |
2855 | } | |
2856 | ||
e498be7d CL |
2857 | #ifdef CONFIG_NUMA |
2858 | /* | |
2859 | * A interface to enable slab creation on nodeid | |
1da177e4 | 2860 | */ |
a737b3e2 AM |
2861 | static void *__cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
2862 | int nodeid) | |
e498be7d CL |
2863 | { |
2864 | struct list_head *entry; | |
b28a02de PE |
2865 | struct slab *slabp; |
2866 | struct kmem_list3 *l3; | |
2867 | void *obj; | |
b28a02de PE |
2868 | int x; |
2869 | ||
2870 | l3 = cachep->nodelists[nodeid]; | |
2871 | BUG_ON(!l3); | |
2872 | ||
a737b3e2 | 2873 | retry: |
ca3b9b91 | 2874 | check_irq_off(); |
b28a02de PE |
2875 | spin_lock(&l3->list_lock); |
2876 | entry = l3->slabs_partial.next; | |
2877 | if (entry == &l3->slabs_partial) { | |
2878 | l3->free_touched = 1; | |
2879 | entry = l3->slabs_free.next; | |
2880 | if (entry == &l3->slabs_free) | |
2881 | goto must_grow; | |
2882 | } | |
2883 | ||
2884 | slabp = list_entry(entry, struct slab, list); | |
2885 | check_spinlock_acquired_node(cachep, nodeid); | |
2886 | check_slabp(cachep, slabp); | |
2887 | ||
2888 | STATS_INC_NODEALLOCS(cachep); | |
2889 | STATS_INC_ACTIVE(cachep); | |
2890 | STATS_SET_HIGH(cachep); | |
2891 | ||
2892 | BUG_ON(slabp->inuse == cachep->num); | |
2893 | ||
78d382d7 | 2894 | obj = slab_get_obj(cachep, slabp, nodeid); |
b28a02de PE |
2895 | check_slabp(cachep, slabp); |
2896 | l3->free_objects--; | |
2897 | /* move slabp to correct slabp list: */ | |
2898 | list_del(&slabp->list); | |
2899 | ||
a737b3e2 | 2900 | if (slabp->free == BUFCTL_END) |
b28a02de | 2901 | list_add(&slabp->list, &l3->slabs_full); |
a737b3e2 | 2902 | else |
b28a02de | 2903 | list_add(&slabp->list, &l3->slabs_partial); |
e498be7d | 2904 | |
b28a02de PE |
2905 | spin_unlock(&l3->list_lock); |
2906 | goto done; | |
e498be7d | 2907 | |
a737b3e2 | 2908 | must_grow: |
b28a02de PE |
2909 | spin_unlock(&l3->list_lock); |
2910 | x = cache_grow(cachep, flags, nodeid); | |
1da177e4 | 2911 | |
b28a02de PE |
2912 | if (!x) |
2913 | return NULL; | |
e498be7d | 2914 | |
b28a02de | 2915 | goto retry; |
a737b3e2 | 2916 | done: |
b28a02de | 2917 | return obj; |
e498be7d CL |
2918 | } |
2919 | #endif | |
2920 | ||
2921 | /* | |
2922 | * Caller needs to acquire correct kmem_list's list_lock | |
2923 | */ | |
343e0d7a | 2924 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 2925 | int node) |
1da177e4 LT |
2926 | { |
2927 | int i; | |
e498be7d | 2928 | struct kmem_list3 *l3; |
1da177e4 LT |
2929 | |
2930 | for (i = 0; i < nr_objects; i++) { | |
2931 | void *objp = objpp[i]; | |
2932 | struct slab *slabp; | |
1da177e4 | 2933 | |
6ed5eb22 | 2934 | slabp = virt_to_slab(objp); |
ff69416e | 2935 | l3 = cachep->nodelists[node]; |
1da177e4 | 2936 | list_del(&slabp->list); |
ff69416e | 2937 | check_spinlock_acquired_node(cachep, node); |
1da177e4 | 2938 | check_slabp(cachep, slabp); |
78d382d7 | 2939 | slab_put_obj(cachep, slabp, objp, node); |
1da177e4 | 2940 | STATS_DEC_ACTIVE(cachep); |
e498be7d | 2941 | l3->free_objects++; |
1da177e4 LT |
2942 | check_slabp(cachep, slabp); |
2943 | ||
2944 | /* fixup slab chains */ | |
2945 | if (slabp->inuse == 0) { | |
e498be7d CL |
2946 | if (l3->free_objects > l3->free_limit) { |
2947 | l3->free_objects -= cachep->num; | |
1da177e4 LT |
2948 | slab_destroy(cachep, slabp); |
2949 | } else { | |
e498be7d | 2950 | list_add(&slabp->list, &l3->slabs_free); |
1da177e4 LT |
2951 | } |
2952 | } else { | |
2953 | /* Unconditionally move a slab to the end of the | |
2954 | * partial list on free - maximum time for the | |
2955 | * other objects to be freed, too. | |
2956 | */ | |
e498be7d | 2957 | list_add_tail(&slabp->list, &l3->slabs_partial); |
1da177e4 LT |
2958 | } |
2959 | } | |
2960 | } | |
2961 | ||
343e0d7a | 2962 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
2963 | { |
2964 | int batchcount; | |
e498be7d | 2965 | struct kmem_list3 *l3; |
ff69416e | 2966 | int node = numa_node_id(); |
1da177e4 LT |
2967 | |
2968 | batchcount = ac->batchcount; | |
2969 | #if DEBUG | |
2970 | BUG_ON(!batchcount || batchcount > ac->avail); | |
2971 | #endif | |
2972 | check_irq_off(); | |
ff69416e | 2973 | l3 = cachep->nodelists[node]; |
e498be7d CL |
2974 | spin_lock(&l3->list_lock); |
2975 | if (l3->shared) { | |
2976 | struct array_cache *shared_array = l3->shared; | |
b28a02de | 2977 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
2978 | if (max) { |
2979 | if (batchcount > max) | |
2980 | batchcount = max; | |
e498be7d | 2981 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 2982 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
2983 | shared_array->avail += batchcount; |
2984 | goto free_done; | |
2985 | } | |
2986 | } | |
2987 | ||
ff69416e | 2988 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 2989 | free_done: |
1da177e4 LT |
2990 | #if STATS |
2991 | { | |
2992 | int i = 0; | |
2993 | struct list_head *p; | |
2994 | ||
e498be7d CL |
2995 | p = l3->slabs_free.next; |
2996 | while (p != &(l3->slabs_free)) { | |
1da177e4 LT |
2997 | struct slab *slabp; |
2998 | ||
2999 | slabp = list_entry(p, struct slab, list); | |
3000 | BUG_ON(slabp->inuse); | |
3001 | ||
3002 | i++; | |
3003 | p = p->next; | |
3004 | } | |
3005 | STATS_SET_FREEABLE(cachep, i); | |
3006 | } | |
3007 | #endif | |
e498be7d | 3008 | spin_unlock(&l3->list_lock); |
1da177e4 | 3009 | ac->avail -= batchcount; |
a737b3e2 | 3010 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3011 | } |
3012 | ||
3013 | /* | |
a737b3e2 AM |
3014 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3015 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3016 | */ |
343e0d7a | 3017 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 | 3018 | { |
9a2dba4b | 3019 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3020 | |
3021 | check_irq_off(); | |
3022 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); | |
3023 | ||
e498be7d CL |
3024 | /* Make sure we are not freeing a object from another |
3025 | * node to the array cache on this cpu. | |
3026 | */ | |
3027 | #ifdef CONFIG_NUMA | |
3028 | { | |
3029 | struct slab *slabp; | |
6ed5eb22 | 3030 | slabp = virt_to_slab(objp); |
e498be7d CL |
3031 | if (unlikely(slabp->nodeid != numa_node_id())) { |
3032 | struct array_cache *alien = NULL; | |
3033 | int nodeid = slabp->nodeid; | |
a737b3e2 | 3034 | struct kmem_list3 *l3; |
e498be7d | 3035 | |
a737b3e2 | 3036 | l3 = cachep->nodelists[numa_node_id()]; |
e498be7d CL |
3037 | STATS_INC_NODEFREES(cachep); |
3038 | if (l3->alien && l3->alien[nodeid]) { | |
3039 | alien = l3->alien[nodeid]; | |
3040 | spin_lock(&alien->lock); | |
3041 | if (unlikely(alien->avail == alien->limit)) | |
3042 | __drain_alien_cache(cachep, | |
b28a02de | 3043 | alien, nodeid); |
e498be7d CL |
3044 | alien->entry[alien->avail++] = objp; |
3045 | spin_unlock(&alien->lock); | |
3046 | } else { | |
3047 | spin_lock(&(cachep->nodelists[nodeid])-> | |
b28a02de | 3048 | list_lock); |
ff69416e | 3049 | free_block(cachep, &objp, 1, nodeid); |
e498be7d | 3050 | spin_unlock(&(cachep->nodelists[nodeid])-> |
b28a02de | 3051 | list_lock); |
e498be7d CL |
3052 | } |
3053 | return; | |
3054 | } | |
3055 | } | |
3056 | #endif | |
1da177e4 LT |
3057 | if (likely(ac->avail < ac->limit)) { |
3058 | STATS_INC_FREEHIT(cachep); | |
e498be7d | 3059 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3060 | return; |
3061 | } else { | |
3062 | STATS_INC_FREEMISS(cachep); | |
3063 | cache_flusharray(cachep, ac); | |
e498be7d | 3064 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3065 | } |
3066 | } | |
3067 | ||
3068 | /** | |
3069 | * kmem_cache_alloc - Allocate an object | |
3070 | * @cachep: The cache to allocate from. | |
3071 | * @flags: See kmalloc(). | |
3072 | * | |
3073 | * Allocate an object from this cache. The flags are only relevant | |
3074 | * if the cache has no available objects. | |
3075 | */ | |
343e0d7a | 3076 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3077 | { |
7fd6b141 | 3078 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); |
1da177e4 LT |
3079 | } |
3080 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3081 | ||
3082 | /** | |
3083 | * kmem_ptr_validate - check if an untrusted pointer might | |
3084 | * be a slab entry. | |
3085 | * @cachep: the cache we're checking against | |
3086 | * @ptr: pointer to validate | |
3087 | * | |
3088 | * This verifies that the untrusted pointer looks sane: | |
3089 | * it is _not_ a guarantee that the pointer is actually | |
3090 | * part of the slab cache in question, but it at least | |
3091 | * validates that the pointer can be dereferenced and | |
3092 | * looks half-way sane. | |
3093 | * | |
3094 | * Currently only used for dentry validation. | |
3095 | */ | |
343e0d7a | 3096 | int fastcall kmem_ptr_validate(struct kmem_cache *cachep, void *ptr) |
1da177e4 | 3097 | { |
b28a02de | 3098 | unsigned long addr = (unsigned long)ptr; |
1da177e4 | 3099 | unsigned long min_addr = PAGE_OFFSET; |
b28a02de | 3100 | unsigned long align_mask = BYTES_PER_WORD - 1; |
3dafccf2 | 3101 | unsigned long size = cachep->buffer_size; |
1da177e4 LT |
3102 | struct page *page; |
3103 | ||
3104 | if (unlikely(addr < min_addr)) | |
3105 | goto out; | |
3106 | if (unlikely(addr > (unsigned long)high_memory - size)) | |
3107 | goto out; | |
3108 | if (unlikely(addr & align_mask)) | |
3109 | goto out; | |
3110 | if (unlikely(!kern_addr_valid(addr))) | |
3111 | goto out; | |
3112 | if (unlikely(!kern_addr_valid(addr + size - 1))) | |
3113 | goto out; | |
3114 | page = virt_to_page(ptr); | |
3115 | if (unlikely(!PageSlab(page))) | |
3116 | goto out; | |
065d41cb | 3117 | if (unlikely(page_get_cache(page) != cachep)) |
1da177e4 LT |
3118 | goto out; |
3119 | return 1; | |
a737b3e2 | 3120 | out: |
1da177e4 LT |
3121 | return 0; |
3122 | } | |
3123 | ||
3124 | #ifdef CONFIG_NUMA | |
3125 | /** | |
3126 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3127 | * @cachep: The cache to allocate from. | |
3128 | * @flags: See kmalloc(). | |
3129 | * @nodeid: node number of the target node. | |
3130 | * | |
3131 | * Identical to kmem_cache_alloc, except that this function is slow | |
3132 | * and can sleep. And it will allocate memory on the given node, which | |
3133 | * can improve the performance for cpu bound structures. | |
e498be7d CL |
3134 | * New and improved: it will now make sure that the object gets |
3135 | * put on the correct node list so that there is no false sharing. | |
1da177e4 | 3136 | */ |
343e0d7a | 3137 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1da177e4 | 3138 | { |
e498be7d CL |
3139 | unsigned long save_flags; |
3140 | void *ptr; | |
1da177e4 | 3141 | |
e498be7d CL |
3142 | cache_alloc_debugcheck_before(cachep, flags); |
3143 | local_irq_save(save_flags); | |
18f820f6 CL |
3144 | |
3145 | if (nodeid == -1 || nodeid == numa_node_id() || | |
a737b3e2 | 3146 | !cachep->nodelists[nodeid]) |
5c382300 AK |
3147 | ptr = ____cache_alloc(cachep, flags); |
3148 | else | |
3149 | ptr = __cache_alloc_node(cachep, flags, nodeid); | |
e498be7d | 3150 | local_irq_restore(save_flags); |
18f820f6 CL |
3151 | |
3152 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, | |
3153 | __builtin_return_address(0)); | |
1da177e4 | 3154 | |
e498be7d | 3155 | return ptr; |
1da177e4 LT |
3156 | } |
3157 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
3158 | ||
dd0fc66f | 3159 | void *kmalloc_node(size_t size, gfp_t flags, int node) |
97e2bde4 | 3160 | { |
343e0d7a | 3161 | struct kmem_cache *cachep; |
97e2bde4 MS |
3162 | |
3163 | cachep = kmem_find_general_cachep(size, flags); | |
3164 | if (unlikely(cachep == NULL)) | |
3165 | return NULL; | |
3166 | return kmem_cache_alloc_node(cachep, flags, node); | |
3167 | } | |
3168 | EXPORT_SYMBOL(kmalloc_node); | |
1da177e4 LT |
3169 | #endif |
3170 | ||
3171 | /** | |
3172 | * kmalloc - allocate memory | |
3173 | * @size: how many bytes of memory are required. | |
3174 | * @flags: the type of memory to allocate. | |
911851e6 | 3175 | * @caller: function caller for debug tracking of the caller |
1da177e4 LT |
3176 | * |
3177 | * kmalloc is the normal method of allocating memory | |
3178 | * in the kernel. | |
3179 | * | |
3180 | * The @flags argument may be one of: | |
3181 | * | |
3182 | * %GFP_USER - Allocate memory on behalf of user. May sleep. | |
3183 | * | |
3184 | * %GFP_KERNEL - Allocate normal kernel ram. May sleep. | |
3185 | * | |
3186 | * %GFP_ATOMIC - Allocation will not sleep. Use inside interrupt handlers. | |
3187 | * | |
3188 | * Additionally, the %GFP_DMA flag may be set to indicate the memory | |
3189 | * must be suitable for DMA. This can mean different things on different | |
3190 | * platforms. For example, on i386, it means that the memory must come | |
3191 | * from the first 16MB. | |
3192 | */ | |
7fd6b141 PE |
3193 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
3194 | void *caller) | |
1da177e4 | 3195 | { |
343e0d7a | 3196 | struct kmem_cache *cachep; |
1da177e4 | 3197 | |
97e2bde4 MS |
3198 | /* If you want to save a few bytes .text space: replace |
3199 | * __ with kmem_. | |
3200 | * Then kmalloc uses the uninlined functions instead of the inline | |
3201 | * functions. | |
3202 | */ | |
3203 | cachep = __find_general_cachep(size, flags); | |
dbdb9045 AM |
3204 | if (unlikely(cachep == NULL)) |
3205 | return NULL; | |
7fd6b141 PE |
3206 | return __cache_alloc(cachep, flags, caller); |
3207 | } | |
3208 | ||
3209 | #ifndef CONFIG_DEBUG_SLAB | |
3210 | ||
3211 | void *__kmalloc(size_t size, gfp_t flags) | |
3212 | { | |
3213 | return __do_kmalloc(size, flags, NULL); | |
1da177e4 LT |
3214 | } |
3215 | EXPORT_SYMBOL(__kmalloc); | |
3216 | ||
7fd6b141 PE |
3217 | #else |
3218 | ||
3219 | void *__kmalloc_track_caller(size_t size, gfp_t flags, void *caller) | |
3220 | { | |
3221 | return __do_kmalloc(size, flags, caller); | |
3222 | } | |
3223 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
3224 | ||
3225 | #endif | |
3226 | ||
1da177e4 LT |
3227 | #ifdef CONFIG_SMP |
3228 | /** | |
3229 | * __alloc_percpu - allocate one copy of the object for every present | |
3230 | * cpu in the system, zeroing them. | |
3231 | * Objects should be dereferenced using the per_cpu_ptr macro only. | |
3232 | * | |
3233 | * @size: how many bytes of memory are required. | |
1da177e4 | 3234 | */ |
f9f75005 | 3235 | void *__alloc_percpu(size_t size) |
1da177e4 LT |
3236 | { |
3237 | int i; | |
b28a02de | 3238 | struct percpu_data *pdata = kmalloc(sizeof(*pdata), GFP_KERNEL); |
1da177e4 LT |
3239 | |
3240 | if (!pdata) | |
3241 | return NULL; | |
3242 | ||
e498be7d CL |
3243 | /* |
3244 | * Cannot use for_each_online_cpu since a cpu may come online | |
3245 | * and we have no way of figuring out how to fix the array | |
3246 | * that we have allocated then.... | |
3247 | */ | |
3248 | for_each_cpu(i) { | |
3249 | int node = cpu_to_node(i); | |
3250 | ||
3251 | if (node_online(node)) | |
3252 | pdata->ptrs[i] = kmalloc_node(size, GFP_KERNEL, node); | |
3253 | else | |
3254 | pdata->ptrs[i] = kmalloc(size, GFP_KERNEL); | |
1da177e4 LT |
3255 | |
3256 | if (!pdata->ptrs[i]) | |
3257 | goto unwind_oom; | |
3258 | memset(pdata->ptrs[i], 0, size); | |
3259 | } | |
3260 | ||
3261 | /* Catch derefs w/o wrappers */ | |
b28a02de | 3262 | return (void *)(~(unsigned long)pdata); |
1da177e4 | 3263 | |
a737b3e2 | 3264 | unwind_oom: |
1da177e4 LT |
3265 | while (--i >= 0) { |
3266 | if (!cpu_possible(i)) | |
3267 | continue; | |
3268 | kfree(pdata->ptrs[i]); | |
3269 | } | |
3270 | kfree(pdata); | |
3271 | return NULL; | |
3272 | } | |
3273 | EXPORT_SYMBOL(__alloc_percpu); | |
3274 | #endif | |
3275 | ||
3276 | /** | |
3277 | * kmem_cache_free - Deallocate an object | |
3278 | * @cachep: The cache the allocation was from. | |
3279 | * @objp: The previously allocated object. | |
3280 | * | |
3281 | * Free an object which was previously allocated from this | |
3282 | * cache. | |
3283 | */ | |
343e0d7a | 3284 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3285 | { |
3286 | unsigned long flags; | |
3287 | ||
3288 | local_irq_save(flags); | |
3289 | __cache_free(cachep, objp); | |
3290 | local_irq_restore(flags); | |
3291 | } | |
3292 | EXPORT_SYMBOL(kmem_cache_free); | |
3293 | ||
1da177e4 LT |
3294 | /** |
3295 | * kfree - free previously allocated memory | |
3296 | * @objp: pointer returned by kmalloc. | |
3297 | * | |
80e93eff PE |
3298 | * If @objp is NULL, no operation is performed. |
3299 | * | |
1da177e4 LT |
3300 | * Don't free memory not originally allocated by kmalloc() |
3301 | * or you will run into trouble. | |
3302 | */ | |
3303 | void kfree(const void *objp) | |
3304 | { | |
343e0d7a | 3305 | struct kmem_cache *c; |
1da177e4 LT |
3306 | unsigned long flags; |
3307 | ||
3308 | if (unlikely(!objp)) | |
3309 | return; | |
3310 | local_irq_save(flags); | |
3311 | kfree_debugcheck(objp); | |
6ed5eb22 | 3312 | c = virt_to_cache(objp); |
3dafccf2 | 3313 | mutex_debug_check_no_locks_freed(objp, obj_size(c)); |
b28a02de | 3314 | __cache_free(c, (void *)objp); |
1da177e4 LT |
3315 | local_irq_restore(flags); |
3316 | } | |
3317 | EXPORT_SYMBOL(kfree); | |
3318 | ||
3319 | #ifdef CONFIG_SMP | |
3320 | /** | |
3321 | * free_percpu - free previously allocated percpu memory | |
3322 | * @objp: pointer returned by alloc_percpu. | |
3323 | * | |
3324 | * Don't free memory not originally allocated by alloc_percpu() | |
3325 | * The complemented objp is to check for that. | |
3326 | */ | |
b28a02de | 3327 | void free_percpu(const void *objp) |
1da177e4 LT |
3328 | { |
3329 | int i; | |
b28a02de | 3330 | struct percpu_data *p = (struct percpu_data *)(~(unsigned long)objp); |
1da177e4 | 3331 | |
e498be7d CL |
3332 | /* |
3333 | * We allocate for all cpus so we cannot use for online cpu here. | |
3334 | */ | |
3335 | for_each_cpu(i) | |
b28a02de | 3336 | kfree(p->ptrs[i]); |
1da177e4 LT |
3337 | kfree(p); |
3338 | } | |
3339 | EXPORT_SYMBOL(free_percpu); | |
3340 | #endif | |
3341 | ||
343e0d7a | 3342 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
1da177e4 | 3343 | { |
3dafccf2 | 3344 | return obj_size(cachep); |
1da177e4 LT |
3345 | } |
3346 | EXPORT_SYMBOL(kmem_cache_size); | |
3347 | ||
343e0d7a | 3348 | const char *kmem_cache_name(struct kmem_cache *cachep) |
1944972d ACM |
3349 | { |
3350 | return cachep->name; | |
3351 | } | |
3352 | EXPORT_SYMBOL_GPL(kmem_cache_name); | |
3353 | ||
e498be7d CL |
3354 | /* |
3355 | * This initializes kmem_list3 for all nodes. | |
3356 | */ | |
343e0d7a | 3357 | static int alloc_kmemlist(struct kmem_cache *cachep) |
e498be7d CL |
3358 | { |
3359 | int node; | |
3360 | struct kmem_list3 *l3; | |
3361 | int err = 0; | |
3362 | ||
3363 | for_each_online_node(node) { | |
3364 | struct array_cache *nc = NULL, *new; | |
3365 | struct array_cache **new_alien = NULL; | |
3366 | #ifdef CONFIG_NUMA | |
a737b3e2 AM |
3367 | new_alien = alloc_alien_cache(node, cachep->limit); |
3368 | if (!new_alien) | |
e498be7d CL |
3369 | goto fail; |
3370 | #endif | |
a737b3e2 AM |
3371 | new = alloc_arraycache(node, cachep->shared*cachep->batchcount, |
3372 | 0xbaadf00d); | |
3373 | if (!new) | |
e498be7d | 3374 | goto fail; |
a737b3e2 AM |
3375 | l3 = cachep->nodelists[node]; |
3376 | if (l3) { | |
e498be7d CL |
3377 | spin_lock_irq(&l3->list_lock); |
3378 | ||
a737b3e2 AM |
3379 | nc = cachep->nodelists[node]->shared; |
3380 | if (nc) | |
b28a02de | 3381 | free_block(cachep, nc->entry, nc->avail, node); |
e498be7d CL |
3382 | |
3383 | l3->shared = new; | |
3384 | if (!cachep->nodelists[node]->alien) { | |
3385 | l3->alien = new_alien; | |
3386 | new_alien = NULL; | |
3387 | } | |
b28a02de | 3388 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3389 | cachep->batchcount + cachep->num; |
e498be7d CL |
3390 | spin_unlock_irq(&l3->list_lock); |
3391 | kfree(nc); | |
3392 | free_alien_cache(new_alien); | |
3393 | continue; | |
3394 | } | |
a737b3e2 AM |
3395 | l3 = kmalloc_node(sizeof(struct kmem_list3), GFP_KERNEL, node); |
3396 | if (!l3) | |
e498be7d CL |
3397 | goto fail; |
3398 | ||
3399 | kmem_list3_init(l3); | |
3400 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
a737b3e2 | 3401 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
e498be7d CL |
3402 | l3->shared = new; |
3403 | l3->alien = new_alien; | |
b28a02de | 3404 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3405 | cachep->batchcount + cachep->num; |
e498be7d CL |
3406 | cachep->nodelists[node] = l3; |
3407 | } | |
3408 | return err; | |
a737b3e2 | 3409 | fail: |
e498be7d CL |
3410 | err = -ENOMEM; |
3411 | return err; | |
3412 | } | |
3413 | ||
1da177e4 | 3414 | struct ccupdate_struct { |
343e0d7a | 3415 | struct kmem_cache *cachep; |
1da177e4 LT |
3416 | struct array_cache *new[NR_CPUS]; |
3417 | }; | |
3418 | ||
3419 | static void do_ccupdate_local(void *info) | |
3420 | { | |
a737b3e2 | 3421 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3422 | struct array_cache *old; |
3423 | ||
3424 | check_irq_off(); | |
9a2dba4b | 3425 | old = cpu_cache_get(new->cachep); |
e498be7d | 3426 | |
1da177e4 LT |
3427 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3428 | new->new[smp_processor_id()] = old; | |
3429 | } | |
3430 | ||
b5d8ca7c | 3431 | /* Always called with the cache_chain_mutex held */ |
a737b3e2 AM |
3432 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3433 | int batchcount, int shared) | |
1da177e4 LT |
3434 | { |
3435 | struct ccupdate_struct new; | |
e498be7d | 3436 | int i, err; |
1da177e4 | 3437 | |
b28a02de | 3438 | memset(&new.new, 0, sizeof(new.new)); |
e498be7d | 3439 | for_each_online_cpu(i) { |
a737b3e2 AM |
3440 | new.new[i] = alloc_arraycache(cpu_to_node(i), limit, |
3441 | batchcount); | |
e498be7d | 3442 | if (!new.new[i]) { |
b28a02de PE |
3443 | for (i--; i >= 0; i--) |
3444 | kfree(new.new[i]); | |
e498be7d | 3445 | return -ENOMEM; |
1da177e4 LT |
3446 | } |
3447 | } | |
3448 | new.cachep = cachep; | |
3449 | ||
a07fa394 | 3450 | on_each_cpu(do_ccupdate_local, (void *)&new, 1, 1); |
e498be7d | 3451 | |
1da177e4 | 3452 | check_irq_on(); |
1da177e4 LT |
3453 | cachep->batchcount = batchcount; |
3454 | cachep->limit = limit; | |
e498be7d | 3455 | cachep->shared = shared; |
1da177e4 | 3456 | |
e498be7d | 3457 | for_each_online_cpu(i) { |
1da177e4 LT |
3458 | struct array_cache *ccold = new.new[i]; |
3459 | if (!ccold) | |
3460 | continue; | |
e498be7d | 3461 | spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
ff69416e | 3462 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); |
e498be7d | 3463 | spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
1da177e4 LT |
3464 | kfree(ccold); |
3465 | } | |
1da177e4 | 3466 | |
e498be7d CL |
3467 | err = alloc_kmemlist(cachep); |
3468 | if (err) { | |
3469 | printk(KERN_ERR "alloc_kmemlist failed for %s, error %d.\n", | |
b28a02de | 3470 | cachep->name, -err); |
e498be7d | 3471 | BUG(); |
1da177e4 | 3472 | } |
1da177e4 LT |
3473 | return 0; |
3474 | } | |
3475 | ||
b5d8ca7c | 3476 | /* Called with cache_chain_mutex held always */ |
343e0d7a | 3477 | static void enable_cpucache(struct kmem_cache *cachep) |
1da177e4 LT |
3478 | { |
3479 | int err; | |
3480 | int limit, shared; | |
3481 | ||
a737b3e2 AM |
3482 | /* |
3483 | * The head array serves three purposes: | |
1da177e4 LT |
3484 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3485 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3486 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3487 | * bufctl chains: array operations are cheaper. |
3488 | * The numbers are guessed, we should auto-tune as described by | |
3489 | * Bonwick. | |
3490 | */ | |
3dafccf2 | 3491 | if (cachep->buffer_size > 131072) |
1da177e4 | 3492 | limit = 1; |
3dafccf2 | 3493 | else if (cachep->buffer_size > PAGE_SIZE) |
1da177e4 | 3494 | limit = 8; |
3dafccf2 | 3495 | else if (cachep->buffer_size > 1024) |
1da177e4 | 3496 | limit = 24; |
3dafccf2 | 3497 | else if (cachep->buffer_size > 256) |
1da177e4 LT |
3498 | limit = 54; |
3499 | else | |
3500 | limit = 120; | |
3501 | ||
a737b3e2 AM |
3502 | /* |
3503 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3504 | * allocation behaviour: Most allocs on one cpu, most free operations |
3505 | * on another cpu. For these cases, an efficient object passing between | |
3506 | * cpus is necessary. This is provided by a shared array. The array | |
3507 | * replaces Bonwick's magazine layer. | |
3508 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3509 | * to a larger limit. Thus disabled by default. | |
3510 | */ | |
3511 | shared = 0; | |
3512 | #ifdef CONFIG_SMP | |
3dafccf2 | 3513 | if (cachep->buffer_size <= PAGE_SIZE) |
1da177e4 LT |
3514 | shared = 8; |
3515 | #endif | |
3516 | ||
3517 | #if DEBUG | |
a737b3e2 AM |
3518 | /* |
3519 | * With debugging enabled, large batchcount lead to excessively long | |
3520 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3521 | */ |
3522 | if (limit > 32) | |
3523 | limit = 32; | |
3524 | #endif | |
b28a02de | 3525 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared); |
1da177e4 LT |
3526 | if (err) |
3527 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3528 | cachep->name, -err); |
1da177e4 LT |
3529 | } |
3530 | ||
1b55253a CL |
3531 | /* |
3532 | * Drain an array if it contains any elements taking the l3 lock only if | |
b18e7e65 CL |
3533 | * necessary. Note that the l3 listlock also protects the array_cache |
3534 | * if drain_array() is used on the shared array. | |
1b55253a CL |
3535 | */ |
3536 | void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, | |
3537 | struct array_cache *ac, int force, int node) | |
1da177e4 LT |
3538 | { |
3539 | int tofree; | |
3540 | ||
1b55253a CL |
3541 | if (!ac || !ac->avail) |
3542 | return; | |
1da177e4 LT |
3543 | if (ac->touched && !force) { |
3544 | ac->touched = 0; | |
b18e7e65 | 3545 | } else { |
1b55253a | 3546 | spin_lock_irq(&l3->list_lock); |
b18e7e65 CL |
3547 | if (ac->avail) { |
3548 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3549 | if (tofree > ac->avail) | |
3550 | tofree = (ac->avail + 1) / 2; | |
3551 | free_block(cachep, ac->entry, tofree, node); | |
3552 | ac->avail -= tofree; | |
3553 | memmove(ac->entry, &(ac->entry[tofree]), | |
3554 | sizeof(void *) * ac->avail); | |
3555 | } | |
1b55253a | 3556 | spin_unlock_irq(&l3->list_lock); |
1da177e4 LT |
3557 | } |
3558 | } | |
3559 | ||
3560 | /** | |
3561 | * cache_reap - Reclaim memory from caches. | |
1e5d5331 | 3562 | * @unused: unused parameter |
1da177e4 LT |
3563 | * |
3564 | * Called from workqueue/eventd every few seconds. | |
3565 | * Purpose: | |
3566 | * - clear the per-cpu caches for this CPU. | |
3567 | * - return freeable pages to the main free memory pool. | |
3568 | * | |
a737b3e2 AM |
3569 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
3570 | * again on the next iteration. | |
1da177e4 LT |
3571 | */ |
3572 | static void cache_reap(void *unused) | |
3573 | { | |
3574 | struct list_head *walk; | |
e498be7d | 3575 | struct kmem_list3 *l3; |
aab2207c | 3576 | int node = numa_node_id(); |
1da177e4 | 3577 | |
fc0abb14 | 3578 | if (!mutex_trylock(&cache_chain_mutex)) { |
1da177e4 | 3579 | /* Give up. Setup the next iteration. */ |
b28a02de PE |
3580 | schedule_delayed_work(&__get_cpu_var(reap_work), |
3581 | REAPTIMEOUT_CPUC); | |
1da177e4 LT |
3582 | return; |
3583 | } | |
3584 | ||
3585 | list_for_each(walk, &cache_chain) { | |
343e0d7a | 3586 | struct kmem_cache *searchp; |
b28a02de | 3587 | struct list_head *p; |
1da177e4 LT |
3588 | int tofree; |
3589 | struct slab *slabp; | |
3590 | ||
343e0d7a | 3591 | searchp = list_entry(walk, struct kmem_cache, next); |
1da177e4 LT |
3592 | check_irq_on(); |
3593 | ||
35386e3b CL |
3594 | /* |
3595 | * We only take the l3 lock if absolutely necessary and we | |
3596 | * have established with reasonable certainty that | |
3597 | * we can do some work if the lock was obtained. | |
3598 | */ | |
aab2207c | 3599 | l3 = searchp->nodelists[node]; |
35386e3b | 3600 | |
8fce4d8e | 3601 | reap_alien(searchp, l3); |
1da177e4 | 3602 | |
aab2207c | 3603 | drain_array(searchp, l3, cpu_cache_get(searchp), 0, node); |
1da177e4 | 3604 | |
35386e3b CL |
3605 | /* |
3606 | * These are racy checks but it does not matter | |
3607 | * if we skip one check or scan twice. | |
3608 | */ | |
e498be7d | 3609 | if (time_after(l3->next_reap, jiffies)) |
35386e3b | 3610 | goto next; |
1da177e4 | 3611 | |
e498be7d | 3612 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
1da177e4 | 3613 | |
aab2207c | 3614 | drain_array(searchp, l3, l3->shared, 0, node); |
1da177e4 | 3615 | |
e498be7d CL |
3616 | if (l3->free_touched) { |
3617 | l3->free_touched = 0; | |
35386e3b | 3618 | goto next; |
1da177e4 LT |
3619 | } |
3620 | ||
a737b3e2 AM |
3621 | tofree = (l3->free_limit + 5 * searchp->num - 1) / |
3622 | (5 * searchp->num); | |
1da177e4 | 3623 | do { |
35386e3b CL |
3624 | /* |
3625 | * Do not lock if there are no free blocks. | |
3626 | */ | |
3627 | if (list_empty(&l3->slabs_free)) | |
3628 | break; | |
3629 | ||
3630 | spin_lock_irq(&l3->list_lock); | |
e498be7d | 3631 | p = l3->slabs_free.next; |
35386e3b CL |
3632 | if (p == &(l3->slabs_free)) { |
3633 | spin_unlock_irq(&l3->list_lock); | |
1da177e4 | 3634 | break; |
35386e3b | 3635 | } |
1da177e4 LT |
3636 | |
3637 | slabp = list_entry(p, struct slab, list); | |
3638 | BUG_ON(slabp->inuse); | |
3639 | list_del(&slabp->list); | |
3640 | STATS_INC_REAPED(searchp); | |
3641 | ||
a737b3e2 AM |
3642 | /* |
3643 | * Safe to drop the lock. The slab is no longer linked | |
3644 | * to the cache. searchp cannot disappear, we hold | |
1da177e4 LT |
3645 | * cache_chain_lock |
3646 | */ | |
e498be7d CL |
3647 | l3->free_objects -= searchp->num; |
3648 | spin_unlock_irq(&l3->list_lock); | |
1da177e4 | 3649 | slab_destroy(searchp, slabp); |
b28a02de | 3650 | } while (--tofree > 0); |
35386e3b | 3651 | next: |
1da177e4 LT |
3652 | cond_resched(); |
3653 | } | |
3654 | check_irq_on(); | |
fc0abb14 | 3655 | mutex_unlock(&cache_chain_mutex); |
8fce4d8e | 3656 | next_reap_node(); |
a737b3e2 | 3657 | /* Set up the next iteration */ |
cd61ef62 | 3658 | schedule_delayed_work(&__get_cpu_var(reap_work), REAPTIMEOUT_CPUC); |
1da177e4 LT |
3659 | } |
3660 | ||
3661 | #ifdef CONFIG_PROC_FS | |
3662 | ||
85289f98 | 3663 | static void print_slabinfo_header(struct seq_file *m) |
1da177e4 | 3664 | { |
85289f98 PE |
3665 | /* |
3666 | * Output format version, so at least we can change it | |
3667 | * without _too_ many complaints. | |
3668 | */ | |
1da177e4 | 3669 | #if STATS |
85289f98 | 3670 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
1da177e4 | 3671 | #else |
85289f98 | 3672 | seq_puts(m, "slabinfo - version: 2.1\n"); |
1da177e4 | 3673 | #endif |
85289f98 PE |
3674 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
3675 | "<objperslab> <pagesperslab>"); | |
3676 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
3677 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1da177e4 | 3678 | #if STATS |
85289f98 PE |
3679 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
3680 | "<error> <maxfreeable> <nodeallocs> <remotefrees>"); | |
3681 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
1da177e4 | 3682 | #endif |
85289f98 PE |
3683 | seq_putc(m, '\n'); |
3684 | } | |
3685 | ||
3686 | static void *s_start(struct seq_file *m, loff_t *pos) | |
3687 | { | |
3688 | loff_t n = *pos; | |
3689 | struct list_head *p; | |
3690 | ||
fc0abb14 | 3691 | mutex_lock(&cache_chain_mutex); |
85289f98 PE |
3692 | if (!n) |
3693 | print_slabinfo_header(m); | |
1da177e4 LT |
3694 | p = cache_chain.next; |
3695 | while (n--) { | |
3696 | p = p->next; | |
3697 | if (p == &cache_chain) | |
3698 | return NULL; | |
3699 | } | |
343e0d7a | 3700 | return list_entry(p, struct kmem_cache, next); |
1da177e4 LT |
3701 | } |
3702 | ||
3703 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
3704 | { | |
343e0d7a | 3705 | struct kmem_cache *cachep = p; |
1da177e4 | 3706 | ++*pos; |
a737b3e2 AM |
3707 | return cachep->next.next == &cache_chain ? |
3708 | NULL : list_entry(cachep->next.next, struct kmem_cache, next); | |
1da177e4 LT |
3709 | } |
3710 | ||
3711 | static void s_stop(struct seq_file *m, void *p) | |
3712 | { | |
fc0abb14 | 3713 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
3714 | } |
3715 | ||
3716 | static int s_show(struct seq_file *m, void *p) | |
3717 | { | |
343e0d7a | 3718 | struct kmem_cache *cachep = p; |
1da177e4 | 3719 | struct list_head *q; |
b28a02de PE |
3720 | struct slab *slabp; |
3721 | unsigned long active_objs; | |
3722 | unsigned long num_objs; | |
3723 | unsigned long active_slabs = 0; | |
3724 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 3725 | const char *name; |
1da177e4 | 3726 | char *error = NULL; |
e498be7d CL |
3727 | int node; |
3728 | struct kmem_list3 *l3; | |
1da177e4 | 3729 | |
1da177e4 LT |
3730 | active_objs = 0; |
3731 | num_slabs = 0; | |
e498be7d CL |
3732 | for_each_online_node(node) { |
3733 | l3 = cachep->nodelists[node]; | |
3734 | if (!l3) | |
3735 | continue; | |
3736 | ||
ca3b9b91 RT |
3737 | check_irq_on(); |
3738 | spin_lock_irq(&l3->list_lock); | |
e498be7d | 3739 | |
b28a02de | 3740 | list_for_each(q, &l3->slabs_full) { |
e498be7d CL |
3741 | slabp = list_entry(q, struct slab, list); |
3742 | if (slabp->inuse != cachep->num && !error) | |
3743 | error = "slabs_full accounting error"; | |
3744 | active_objs += cachep->num; | |
3745 | active_slabs++; | |
3746 | } | |
b28a02de | 3747 | list_for_each(q, &l3->slabs_partial) { |
e498be7d CL |
3748 | slabp = list_entry(q, struct slab, list); |
3749 | if (slabp->inuse == cachep->num && !error) | |
3750 | error = "slabs_partial inuse accounting error"; | |
3751 | if (!slabp->inuse && !error) | |
3752 | error = "slabs_partial/inuse accounting error"; | |
3753 | active_objs += slabp->inuse; | |
3754 | active_slabs++; | |
3755 | } | |
b28a02de | 3756 | list_for_each(q, &l3->slabs_free) { |
e498be7d CL |
3757 | slabp = list_entry(q, struct slab, list); |
3758 | if (slabp->inuse && !error) | |
3759 | error = "slabs_free/inuse accounting error"; | |
3760 | num_slabs++; | |
3761 | } | |
3762 | free_objects += l3->free_objects; | |
4484ebf1 RT |
3763 | if (l3->shared) |
3764 | shared_avail += l3->shared->avail; | |
e498be7d | 3765 | |
ca3b9b91 | 3766 | spin_unlock_irq(&l3->list_lock); |
1da177e4 | 3767 | } |
b28a02de PE |
3768 | num_slabs += active_slabs; |
3769 | num_objs = num_slabs * cachep->num; | |
e498be7d | 3770 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
3771 | error = "free_objects accounting error"; |
3772 | ||
b28a02de | 3773 | name = cachep->name; |
1da177e4 LT |
3774 | if (error) |
3775 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
3776 | ||
3777 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
3dafccf2 | 3778 | name, active_objs, num_objs, cachep->buffer_size, |
b28a02de | 3779 | cachep->num, (1 << cachep->gfporder)); |
1da177e4 | 3780 | seq_printf(m, " : tunables %4u %4u %4u", |
b28a02de | 3781 | cachep->limit, cachep->batchcount, cachep->shared); |
e498be7d | 3782 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
b28a02de | 3783 | active_slabs, num_slabs, shared_avail); |
1da177e4 | 3784 | #if STATS |
b28a02de | 3785 | { /* list3 stats */ |
1da177e4 LT |
3786 | unsigned long high = cachep->high_mark; |
3787 | unsigned long allocs = cachep->num_allocations; | |
3788 | unsigned long grown = cachep->grown; | |
3789 | unsigned long reaped = cachep->reaped; | |
3790 | unsigned long errors = cachep->errors; | |
3791 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 3792 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 3793 | unsigned long node_frees = cachep->node_frees; |
1da177e4 | 3794 | |
e498be7d | 3795 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ |
a737b3e2 AM |
3796 | %4lu %4lu %4lu %4lu", allocs, high, grown, |
3797 | reaped, errors, max_freeable, node_allocs, | |
3798 | node_frees); | |
1da177e4 LT |
3799 | } |
3800 | /* cpu stats */ | |
3801 | { | |
3802 | unsigned long allochit = atomic_read(&cachep->allochit); | |
3803 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
3804 | unsigned long freehit = atomic_read(&cachep->freehit); | |
3805 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
3806 | ||
3807 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 3808 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
3809 | } |
3810 | #endif | |
3811 | seq_putc(m, '\n'); | |
1da177e4 LT |
3812 | return 0; |
3813 | } | |
3814 | ||
3815 | /* | |
3816 | * slabinfo_op - iterator that generates /proc/slabinfo | |
3817 | * | |
3818 | * Output layout: | |
3819 | * cache-name | |
3820 | * num-active-objs | |
3821 | * total-objs | |
3822 | * object size | |
3823 | * num-active-slabs | |
3824 | * total-slabs | |
3825 | * num-pages-per-slab | |
3826 | * + further values on SMP and with statistics enabled | |
3827 | */ | |
3828 | ||
3829 | struct seq_operations slabinfo_op = { | |
b28a02de PE |
3830 | .start = s_start, |
3831 | .next = s_next, | |
3832 | .stop = s_stop, | |
3833 | .show = s_show, | |
1da177e4 LT |
3834 | }; |
3835 | ||
3836 | #define MAX_SLABINFO_WRITE 128 | |
3837 | /** | |
3838 | * slabinfo_write - Tuning for the slab allocator | |
3839 | * @file: unused | |
3840 | * @buffer: user buffer | |
3841 | * @count: data length | |
3842 | * @ppos: unused | |
3843 | */ | |
b28a02de PE |
3844 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
3845 | size_t count, loff_t *ppos) | |
1da177e4 | 3846 | { |
b28a02de | 3847 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 LT |
3848 | int limit, batchcount, shared, res; |
3849 | struct list_head *p; | |
b28a02de | 3850 | |
1da177e4 LT |
3851 | if (count > MAX_SLABINFO_WRITE) |
3852 | return -EINVAL; | |
3853 | if (copy_from_user(&kbuf, buffer, count)) | |
3854 | return -EFAULT; | |
b28a02de | 3855 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
3856 | |
3857 | tmp = strchr(kbuf, ' '); | |
3858 | if (!tmp) | |
3859 | return -EINVAL; | |
3860 | *tmp = '\0'; | |
3861 | tmp++; | |
3862 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
3863 | return -EINVAL; | |
3864 | ||
3865 | /* Find the cache in the chain of caches. */ | |
fc0abb14 | 3866 | mutex_lock(&cache_chain_mutex); |
1da177e4 | 3867 | res = -EINVAL; |
b28a02de | 3868 | list_for_each(p, &cache_chain) { |
a737b3e2 | 3869 | struct kmem_cache *cachep; |
1da177e4 | 3870 | |
a737b3e2 | 3871 | cachep = list_entry(p, struct kmem_cache, next); |
1da177e4 | 3872 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
3873 | if (limit < 1 || batchcount < 1 || |
3874 | batchcount > limit || shared < 0) { | |
e498be7d | 3875 | res = 0; |
1da177e4 | 3876 | } else { |
e498be7d | 3877 | res = do_tune_cpucache(cachep, limit, |
b28a02de | 3878 | batchcount, shared); |
1da177e4 LT |
3879 | } |
3880 | break; | |
3881 | } | |
3882 | } | |
fc0abb14 | 3883 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
3884 | if (res >= 0) |
3885 | res = count; | |
3886 | return res; | |
3887 | } | |
3888 | #endif | |
3889 | ||
00e145b6 MS |
3890 | /** |
3891 | * ksize - get the actual amount of memory allocated for a given object | |
3892 | * @objp: Pointer to the object | |
3893 | * | |
3894 | * kmalloc may internally round up allocations and return more memory | |
3895 | * than requested. ksize() can be used to determine the actual amount of | |
3896 | * memory allocated. The caller may use this additional memory, even though | |
3897 | * a smaller amount of memory was initially specified with the kmalloc call. | |
3898 | * The caller must guarantee that objp points to a valid object previously | |
3899 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
3900 | * must not be freed during the duration of the call. | |
3901 | */ | |
1da177e4 LT |
3902 | unsigned int ksize(const void *objp) |
3903 | { | |
00e145b6 MS |
3904 | if (unlikely(objp == NULL)) |
3905 | return 0; | |
1da177e4 | 3906 | |
6ed5eb22 | 3907 | return obj_size(virt_to_cache(objp)); |
1da177e4 | 3908 | } |