]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/slab.c | |
3 | * Written by Mark Hemment, 1996/97. | |
4 | * (markhe@nextd.demon.co.uk) | |
5 | * | |
6 | * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli | |
7 | * | |
8 | * Major cleanup, different bufctl logic, per-cpu arrays | |
9 | * (c) 2000 Manfred Spraul | |
10 | * | |
11 | * Cleanup, make the head arrays unconditional, preparation for NUMA | |
12 | * (c) 2002 Manfred Spraul | |
13 | * | |
14 | * An implementation of the Slab Allocator as described in outline in; | |
15 | * UNIX Internals: The New Frontiers by Uresh Vahalia | |
16 | * Pub: Prentice Hall ISBN 0-13-101908-2 | |
17 | * or with a little more detail in; | |
18 | * The Slab Allocator: An Object-Caching Kernel Memory Allocator | |
19 | * Jeff Bonwick (Sun Microsystems). | |
20 | * Presented at: USENIX Summer 1994 Technical Conference | |
21 | * | |
22 | * The memory is organized in caches, one cache for each object type. | |
23 | * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct) | |
24 | * Each cache consists out of many slabs (they are small (usually one | |
25 | * page long) and always contiguous), and each slab contains multiple | |
26 | * initialized objects. | |
27 | * | |
28 | * This means, that your constructor is used only for newly allocated | |
183ff22b | 29 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
30 | * kmem_cache_free. |
31 | * | |
32 | * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM, | |
33 | * normal). If you need a special memory type, then must create a new | |
34 | * cache for that memory type. | |
35 | * | |
36 | * In order to reduce fragmentation, the slabs are sorted in 3 groups: | |
37 | * full slabs with 0 free objects | |
38 | * partial slabs | |
39 | * empty slabs with no allocated objects | |
40 | * | |
41 | * If partial slabs exist, then new allocations come from these slabs, | |
42 | * otherwise from empty slabs or new slabs are allocated. | |
43 | * | |
44 | * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache | |
45 | * during kmem_cache_destroy(). The caller must prevent concurrent allocs. | |
46 | * | |
47 | * Each cache has a short per-cpu head array, most allocs | |
48 | * and frees go into that array, and if that array overflows, then 1/2 | |
49 | * of the entries in the array are given back into the global cache. | |
50 | * The head array is strictly LIFO and should improve the cache hit rates. | |
51 | * On SMP, it additionally reduces the spinlock operations. | |
52 | * | |
a737b3e2 | 53 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
54 | * it's changed with a smp_call_function(). |
55 | * | |
56 | * SMP synchronization: | |
57 | * constructors and destructors are called without any locking. | |
343e0d7a | 58 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
59 | * are accessed without any locking. |
60 | * The per-cpu arrays are never accessed from the wrong cpu, no locking, | |
61 | * and local interrupts are disabled so slab code is preempt-safe. | |
62 | * The non-constant members are protected with a per-cache irq spinlock. | |
63 | * | |
64 | * Many thanks to Mark Hemment, who wrote another per-cpu slab patch | |
65 | * in 2000 - many ideas in the current implementation are derived from | |
66 | * his patch. | |
67 | * | |
68 | * Further notes from the original documentation: | |
69 | * | |
70 | * 11 April '97. Started multi-threading - markhe | |
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 | ||
1da177e4 LT |
89 | #include <linux/slab.h> |
90 | #include <linux/mm.h> | |
c9cf5528 | 91 | #include <linux/poison.h> |
1da177e4 LT |
92 | #include <linux/swap.h> |
93 | #include <linux/cache.h> | |
94 | #include <linux/interrupt.h> | |
95 | #include <linux/init.h> | |
96 | #include <linux/compiler.h> | |
101a5001 | 97 | #include <linux/cpuset.h> |
a0ec95a8 | 98 | #include <linux/proc_fs.h> |
1da177e4 LT |
99 | #include <linux/seq_file.h> |
100 | #include <linux/notifier.h> | |
101 | #include <linux/kallsyms.h> | |
102 | #include <linux/cpu.h> | |
103 | #include <linux/sysctl.h> | |
104 | #include <linux/module.h> | |
02af61bb | 105 | #include <linux/kmemtrace.h> |
1da177e4 | 106 | #include <linux/rcupdate.h> |
543537bd | 107 | #include <linux/string.h> |
138ae663 | 108 | #include <linux/uaccess.h> |
e498be7d | 109 | #include <linux/nodemask.h> |
d5cff635 | 110 | #include <linux/kmemleak.h> |
dc85da15 | 111 | #include <linux/mempolicy.h> |
fc0abb14 | 112 | #include <linux/mutex.h> |
8a8b6502 | 113 | #include <linux/fault-inject.h> |
e7eebaf6 | 114 | #include <linux/rtmutex.h> |
6a2d7a95 | 115 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 116 | #include <linux/debugobjects.h> |
c175eea4 | 117 | #include <linux/kmemcheck.h> |
1da177e4 | 118 | |
1da177e4 LT |
119 | #include <asm/cacheflush.h> |
120 | #include <asm/tlbflush.h> | |
121 | #include <asm/page.h> | |
122 | ||
123 | /* | |
50953fe9 | 124 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
125 | * 0 for faster, smaller code (especially in the critical paths). |
126 | * | |
127 | * STATS - 1 to collect stats for /proc/slabinfo. | |
128 | * 0 for faster, smaller code (especially in the critical paths). | |
129 | * | |
130 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
131 | */ | |
132 | ||
133 | #ifdef CONFIG_DEBUG_SLAB | |
134 | #define DEBUG 1 | |
135 | #define STATS 1 | |
136 | #define FORCED_DEBUG 1 | |
137 | #else | |
138 | #define DEBUG 0 | |
139 | #define STATS 0 | |
140 | #define FORCED_DEBUG 0 | |
141 | #endif | |
142 | ||
1da177e4 LT |
143 | /* Shouldn't this be in a header file somewhere? */ |
144 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 145 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 146 | |
1da177e4 LT |
147 | #ifndef ARCH_KMALLOC_MINALIGN |
148 | /* | |
149 | * Enforce a minimum alignment for the kmalloc caches. | |
150 | * Usually, the kmalloc caches are cache_line_size() aligned, except when | |
151 | * DEBUG and FORCED_DEBUG are enabled, then they are BYTES_PER_WORD aligned. | |
152 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed | |
b46b8f19 DW |
153 | * alignment larger than the alignment of a 64-bit integer. |
154 | * ARCH_KMALLOC_MINALIGN allows that. | |
155 | * Note that increasing this value may disable some debug features. | |
1da177e4 | 156 | */ |
b46b8f19 | 157 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
1da177e4 LT |
158 | #endif |
159 | ||
160 | #ifndef ARCH_SLAB_MINALIGN | |
161 | /* | |
162 | * Enforce a minimum alignment for all caches. | |
163 | * Intended for archs that get misalignment faults even for BYTES_PER_WORD | |
164 | * aligned buffers. Includes ARCH_KMALLOC_MINALIGN. | |
165 | * If possible: Do not enable this flag for CONFIG_DEBUG_SLAB, it disables | |
166 | * some debug features. | |
167 | */ | |
168 | #define ARCH_SLAB_MINALIGN 0 | |
169 | #endif | |
170 | ||
171 | #ifndef ARCH_KMALLOC_FLAGS | |
172 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
173 | #endif | |
174 | ||
175 | /* Legal flag mask for kmem_cache_create(). */ | |
176 | #if DEBUG | |
50953fe9 | 177 | # define CREATE_MASK (SLAB_RED_ZONE | \ |
1da177e4 | 178 | SLAB_POISON | SLAB_HWCACHE_ALIGN | \ |
ac2b898c | 179 | SLAB_CACHE_DMA | \ |
5af60839 | 180 | SLAB_STORE_USER | \ |
1da177e4 | 181 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
3ac7fe5a | 182 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ |
c175eea4 | 183 | SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) |
1da177e4 | 184 | #else |
ac2b898c | 185 | # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \ |
5af60839 | 186 | SLAB_CACHE_DMA | \ |
1da177e4 | 187 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
3ac7fe5a | 188 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ |
c175eea4 | 189 | SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) |
1da177e4 LT |
190 | #endif |
191 | ||
192 | /* | |
193 | * kmem_bufctl_t: | |
194 | * | |
195 | * Bufctl's are used for linking objs within a slab | |
196 | * linked offsets. | |
197 | * | |
198 | * This implementation relies on "struct page" for locating the cache & | |
199 | * slab an object belongs to. | |
200 | * This allows the bufctl structure to be small (one int), but limits | |
201 | * the number of objects a slab (not a cache) can contain when off-slab | |
202 | * bufctls are used. The limit is the size of the largest general cache | |
203 | * that does not use off-slab slabs. | |
204 | * For 32bit archs with 4 kB pages, is this 56. | |
205 | * This is not serious, as it is only for large objects, when it is unwise | |
206 | * to have too many per slab. | |
207 | * Note: This limit can be raised by introducing a general cache whose size | |
208 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. | |
209 | */ | |
210 | ||
fa5b08d5 | 211 | typedef unsigned int kmem_bufctl_t; |
1da177e4 LT |
212 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
213 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) | |
871751e2 AV |
214 | #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2) |
215 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3) | |
1da177e4 | 216 | |
1da177e4 LT |
217 | /* |
218 | * struct slab | |
219 | * | |
220 | * Manages the objs in a slab. Placed either at the beginning of mem allocated | |
221 | * for a slab, or allocated from an general cache. | |
222 | * Slabs are chained into three list: fully used, partial, fully free slabs. | |
223 | */ | |
224 | struct slab { | |
b28a02de PE |
225 | struct list_head list; |
226 | unsigned long colouroff; | |
227 | void *s_mem; /* including colour offset */ | |
228 | unsigned int inuse; /* num of objs active in slab */ | |
229 | kmem_bufctl_t free; | |
230 | unsigned short nodeid; | |
1da177e4 LT |
231 | }; |
232 | ||
233 | /* | |
234 | * struct slab_rcu | |
235 | * | |
236 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to | |
237 | * arrange for kmem_freepages to be called via RCU. This is useful if | |
238 | * we need to approach a kernel structure obliquely, from its address | |
239 | * obtained without the usual locking. We can lock the structure to | |
240 | * stabilize it and check it's still at the given address, only if we | |
241 | * can be sure that the memory has not been meanwhile reused for some | |
242 | * other kind of object (which our subsystem's lock might corrupt). | |
243 | * | |
244 | * rcu_read_lock before reading the address, then rcu_read_unlock after | |
245 | * taking the spinlock within the structure expected at that address. | |
246 | * | |
247 | * We assume struct slab_rcu can overlay struct slab when destroying. | |
248 | */ | |
249 | struct slab_rcu { | |
b28a02de | 250 | struct rcu_head head; |
343e0d7a | 251 | struct kmem_cache *cachep; |
b28a02de | 252 | void *addr; |
1da177e4 LT |
253 | }; |
254 | ||
255 | /* | |
256 | * struct array_cache | |
257 | * | |
1da177e4 LT |
258 | * Purpose: |
259 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
260 | * - reduce the number of linked list operations | |
261 | * - reduce spinlock operations | |
262 | * | |
263 | * The limit is stored in the per-cpu structure to reduce the data cache | |
264 | * footprint. | |
265 | * | |
266 | */ | |
267 | struct array_cache { | |
268 | unsigned int avail; | |
269 | unsigned int limit; | |
270 | unsigned int batchcount; | |
271 | unsigned int touched; | |
e498be7d | 272 | spinlock_t lock; |
bda5b655 | 273 | void *entry[]; /* |
a737b3e2 AM |
274 | * Must have this definition in here for the proper |
275 | * alignment of array_cache. Also simplifies accessing | |
276 | * the entries. | |
a737b3e2 | 277 | */ |
1da177e4 LT |
278 | }; |
279 | ||
a737b3e2 AM |
280 | /* |
281 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
282 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
283 | */ |
284 | #define BOOT_CPUCACHE_ENTRIES 1 | |
285 | struct arraycache_init { | |
286 | struct array_cache cache; | |
b28a02de | 287 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
288 | }; |
289 | ||
290 | /* | |
e498be7d | 291 | * The slab lists for all objects. |
1da177e4 LT |
292 | */ |
293 | struct kmem_list3 { | |
b28a02de PE |
294 | struct list_head slabs_partial; /* partial list first, better asm code */ |
295 | struct list_head slabs_full; | |
296 | struct list_head slabs_free; | |
297 | unsigned long free_objects; | |
b28a02de | 298 | unsigned int free_limit; |
2e1217cf | 299 | unsigned int colour_next; /* Per-node cache coloring */ |
b28a02de PE |
300 | spinlock_t list_lock; |
301 | struct array_cache *shared; /* shared per node */ | |
302 | struct array_cache **alien; /* on other nodes */ | |
35386e3b CL |
303 | unsigned long next_reap; /* updated without locking */ |
304 | int free_touched; /* updated without locking */ | |
1da177e4 LT |
305 | }; |
306 | ||
e498be7d CL |
307 | /* |
308 | * Need this for bootstrapping a per node allocator. | |
309 | */ | |
556a169d | 310 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
e498be7d CL |
311 | struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; |
312 | #define CACHE_CACHE 0 | |
556a169d PE |
313 | #define SIZE_AC MAX_NUMNODES |
314 | #define SIZE_L3 (2 * MAX_NUMNODES) | |
e498be7d | 315 | |
ed11d9eb CL |
316 | static int drain_freelist(struct kmem_cache *cache, |
317 | struct kmem_list3 *l3, int tofree); | |
318 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, | |
319 | int node); | |
83b519e8 | 320 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 321 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 322 | |
e498be7d | 323 | /* |
a737b3e2 AM |
324 | * This function must be completely optimized away if a constant is passed to |
325 | * it. Mostly the same as what is in linux/slab.h except it returns an index. | |
e498be7d | 326 | */ |
7243cc05 | 327 | static __always_inline int index_of(const size_t size) |
e498be7d | 328 | { |
5ec8a847 SR |
329 | extern void __bad_size(void); |
330 | ||
e498be7d CL |
331 | if (__builtin_constant_p(size)) { |
332 | int i = 0; | |
333 | ||
334 | #define CACHE(x) \ | |
335 | if (size <=x) \ | |
336 | return i; \ | |
337 | else \ | |
338 | i++; | |
1c61fc40 | 339 | #include <linux/kmalloc_sizes.h> |
e498be7d | 340 | #undef CACHE |
5ec8a847 | 341 | __bad_size(); |
7243cc05 | 342 | } else |
5ec8a847 | 343 | __bad_size(); |
e498be7d CL |
344 | return 0; |
345 | } | |
346 | ||
e0a42726 IM |
347 | static int slab_early_init = 1; |
348 | ||
e498be7d CL |
349 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) |
350 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) | |
1da177e4 | 351 | |
5295a74c | 352 | static void kmem_list3_init(struct kmem_list3 *parent) |
e498be7d CL |
353 | { |
354 | INIT_LIST_HEAD(&parent->slabs_full); | |
355 | INIT_LIST_HEAD(&parent->slabs_partial); | |
356 | INIT_LIST_HEAD(&parent->slabs_free); | |
357 | parent->shared = NULL; | |
358 | parent->alien = NULL; | |
2e1217cf | 359 | parent->colour_next = 0; |
e498be7d CL |
360 | spin_lock_init(&parent->list_lock); |
361 | parent->free_objects = 0; | |
362 | parent->free_touched = 0; | |
363 | } | |
364 | ||
a737b3e2 AM |
365 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
366 | do { \ | |
367 | INIT_LIST_HEAD(listp); \ | |
368 | list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ | |
e498be7d CL |
369 | } while (0) |
370 | ||
a737b3e2 AM |
371 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
372 | do { \ | |
e498be7d CL |
373 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
374 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
375 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
376 | } while (0) | |
1da177e4 | 377 | |
1da177e4 LT |
378 | #define CFLGS_OFF_SLAB (0x80000000UL) |
379 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
380 | ||
381 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
382 | /* |
383 | * Optimization question: fewer reaps means less probability for unnessary | |
384 | * cpucache drain/refill cycles. | |
1da177e4 | 385 | * |
dc6f3f27 | 386 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
387 | * which could lock up otherwise freeable slabs. |
388 | */ | |
389 | #define REAPTIMEOUT_CPUC (2*HZ) | |
390 | #define REAPTIMEOUT_LIST3 (4*HZ) | |
391 | ||
392 | #if STATS | |
393 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
394 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
395 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
396 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 397 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
398 | #define STATS_SET_HIGH(x) \ |
399 | do { \ | |
400 | if ((x)->num_active > (x)->high_mark) \ | |
401 | (x)->high_mark = (x)->num_active; \ | |
402 | } while (0) | |
1da177e4 LT |
403 | #define STATS_INC_ERR(x) ((x)->errors++) |
404 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 405 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 406 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
407 | #define STATS_SET_FREEABLE(x, i) \ |
408 | do { \ | |
409 | if ((x)->max_freeable < i) \ | |
410 | (x)->max_freeable = i; \ | |
411 | } while (0) | |
1da177e4 LT |
412 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
413 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
414 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
415 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
416 | #else | |
417 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
418 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
419 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
420 | #define STATS_INC_GROWN(x) do { } while (0) | |
ed11d9eb | 421 | #define STATS_ADD_REAPED(x,y) do { } while (0) |
1da177e4 LT |
422 | #define STATS_SET_HIGH(x) do { } while (0) |
423 | #define STATS_INC_ERR(x) do { } while (0) | |
424 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 425 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 426 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 427 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
428 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
429 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
430 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
431 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
432 | #endif | |
433 | ||
434 | #if DEBUG | |
1da177e4 | 435 | |
a737b3e2 AM |
436 | /* |
437 | * memory layout of objects: | |
1da177e4 | 438 | * 0 : objp |
3dafccf2 | 439 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
440 | * the end of an object is aligned with the end of the real |
441 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 442 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 443 | * redzone word. |
3dafccf2 MS |
444 | * cachep->obj_offset: The real object. |
445 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] | |
a737b3e2 AM |
446 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address |
447 | * [BYTES_PER_WORD long] | |
1da177e4 | 448 | */ |
343e0d7a | 449 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 450 | { |
3dafccf2 | 451 | return cachep->obj_offset; |
1da177e4 LT |
452 | } |
453 | ||
343e0d7a | 454 | static int obj_size(struct kmem_cache *cachep) |
1da177e4 | 455 | { |
3dafccf2 | 456 | return cachep->obj_size; |
1da177e4 LT |
457 | } |
458 | ||
b46b8f19 | 459 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
460 | { |
461 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
462 | return (unsigned long long*) (objp + obj_offset(cachep) - |
463 | sizeof(unsigned long long)); | |
1da177e4 LT |
464 | } |
465 | ||
b46b8f19 | 466 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
467 | { |
468 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
469 | if (cachep->flags & SLAB_STORE_USER) | |
b46b8f19 DW |
470 | return (unsigned long long *)(objp + cachep->buffer_size - |
471 | sizeof(unsigned long long) - | |
87a927c7 | 472 | REDZONE_ALIGN); |
b46b8f19 DW |
473 | return (unsigned long long *) (objp + cachep->buffer_size - |
474 | sizeof(unsigned long long)); | |
1da177e4 LT |
475 | } |
476 | ||
343e0d7a | 477 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
478 | { |
479 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3dafccf2 | 480 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
1da177e4 LT |
481 | } |
482 | ||
483 | #else | |
484 | ||
3dafccf2 MS |
485 | #define obj_offset(x) 0 |
486 | #define obj_size(cachep) (cachep->buffer_size) | |
b46b8f19 DW |
487 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
488 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
489 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
490 | ||
491 | #endif | |
492 | ||
36555751 EGM |
493 | #ifdef CONFIG_KMEMTRACE |
494 | size_t slab_buffer_size(struct kmem_cache *cachep) | |
495 | { | |
496 | return cachep->buffer_size; | |
497 | } | |
498 | EXPORT_SYMBOL(slab_buffer_size); | |
499 | #endif | |
500 | ||
1da177e4 LT |
501 | /* |
502 | * Do not go above this order unless 0 objects fit into the slab. | |
503 | */ | |
504 | #define BREAK_GFP_ORDER_HI 1 | |
505 | #define BREAK_GFP_ORDER_LO 0 | |
506 | static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; | |
507 | ||
a737b3e2 AM |
508 | /* |
509 | * Functions for storing/retrieving the cachep and or slab from the page | |
510 | * allocator. These are used to find the slab an obj belongs to. With kfree(), | |
511 | * these are used to find the cache which an obj belongs to. | |
1da177e4 | 512 | */ |
065d41cb PE |
513 | static inline void page_set_cache(struct page *page, struct kmem_cache *cache) |
514 | { | |
515 | page->lru.next = (struct list_head *)cache; | |
516 | } | |
517 | ||
518 | static inline struct kmem_cache *page_get_cache(struct page *page) | |
519 | { | |
d85f3385 | 520 | page = compound_head(page); |
ddc2e812 | 521 | BUG_ON(!PageSlab(page)); |
065d41cb PE |
522 | return (struct kmem_cache *)page->lru.next; |
523 | } | |
524 | ||
525 | static inline void page_set_slab(struct page *page, struct slab *slab) | |
526 | { | |
527 | page->lru.prev = (struct list_head *)slab; | |
528 | } | |
529 | ||
530 | static inline struct slab *page_get_slab(struct page *page) | |
531 | { | |
ddc2e812 | 532 | BUG_ON(!PageSlab(page)); |
065d41cb PE |
533 | return (struct slab *)page->lru.prev; |
534 | } | |
1da177e4 | 535 | |
6ed5eb22 PE |
536 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
537 | { | |
b49af68f | 538 | struct page *page = virt_to_head_page(obj); |
6ed5eb22 PE |
539 | return page_get_cache(page); |
540 | } | |
541 | ||
542 | static inline struct slab *virt_to_slab(const void *obj) | |
543 | { | |
b49af68f | 544 | struct page *page = virt_to_head_page(obj); |
6ed5eb22 PE |
545 | return page_get_slab(page); |
546 | } | |
547 | ||
8fea4e96 PE |
548 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
549 | unsigned int idx) | |
550 | { | |
551 | return slab->s_mem + cache->buffer_size * idx; | |
552 | } | |
553 | ||
6a2d7a95 ED |
554 | /* |
555 | * We want to avoid an expensive divide : (offset / cache->buffer_size) | |
556 | * Using the fact that buffer_size is a constant for a particular cache, | |
557 | * we can replace (offset / cache->buffer_size) by | |
558 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) | |
559 | */ | |
560 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
561 | const struct slab *slab, void *obj) | |
8fea4e96 | 562 | { |
6a2d7a95 ED |
563 | u32 offset = (obj - slab->s_mem); |
564 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); | |
8fea4e96 PE |
565 | } |
566 | ||
a737b3e2 AM |
567 | /* |
568 | * These are the default caches for kmalloc. Custom caches can have other sizes. | |
569 | */ | |
1da177e4 LT |
570 | struct cache_sizes malloc_sizes[] = { |
571 | #define CACHE(x) { .cs_size = (x) }, | |
572 | #include <linux/kmalloc_sizes.h> | |
573 | CACHE(ULONG_MAX) | |
574 | #undef CACHE | |
575 | }; | |
576 | EXPORT_SYMBOL(malloc_sizes); | |
577 | ||
578 | /* Must match cache_sizes above. Out of line to keep cache footprint low. */ | |
579 | struct cache_names { | |
580 | char *name; | |
581 | char *name_dma; | |
582 | }; | |
583 | ||
584 | static struct cache_names __initdata cache_names[] = { | |
585 | #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, | |
586 | #include <linux/kmalloc_sizes.h> | |
b28a02de | 587 | {NULL,} |
1da177e4 LT |
588 | #undef CACHE |
589 | }; | |
590 | ||
591 | static struct arraycache_init initarray_cache __initdata = | |
b28a02de | 592 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 | 593 | static struct arraycache_init initarray_generic = |
b28a02de | 594 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
595 | |
596 | /* internal cache of cache description objs */ | |
343e0d7a | 597 | static struct kmem_cache cache_cache = { |
b28a02de PE |
598 | .batchcount = 1, |
599 | .limit = BOOT_CPUCACHE_ENTRIES, | |
600 | .shared = 1, | |
343e0d7a | 601 | .buffer_size = sizeof(struct kmem_cache), |
b28a02de | 602 | .name = "kmem_cache", |
1da177e4 LT |
603 | }; |
604 | ||
056c6241 RT |
605 | #define BAD_ALIEN_MAGIC 0x01020304ul |
606 | ||
f1aaee53 AV |
607 | #ifdef CONFIG_LOCKDEP |
608 | ||
609 | /* | |
610 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
611 | * for other slabs "off slab". | |
612 | * The locking for this is tricky in that it nests within the locks | |
613 | * of all other slabs in a few places; to deal with this special | |
614 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
615 | * |
616 | * We set lock class for alien array caches which are up during init. | |
617 | * The lock annotation will be lost if all cpus of a node goes down and | |
618 | * then comes back up during hotplug | |
f1aaee53 | 619 | */ |
056c6241 RT |
620 | static struct lock_class_key on_slab_l3_key; |
621 | static struct lock_class_key on_slab_alc_key; | |
622 | ||
623 | static inline void init_lock_keys(void) | |
f1aaee53 | 624 | |
f1aaee53 AV |
625 | { |
626 | int q; | |
056c6241 RT |
627 | struct cache_sizes *s = malloc_sizes; |
628 | ||
629 | while (s->cs_size != ULONG_MAX) { | |
630 | for_each_node(q) { | |
631 | struct array_cache **alc; | |
632 | int r; | |
633 | struct kmem_list3 *l3 = s->cs_cachep->nodelists[q]; | |
634 | if (!l3 || OFF_SLAB(s->cs_cachep)) | |
635 | continue; | |
636 | lockdep_set_class(&l3->list_lock, &on_slab_l3_key); | |
637 | alc = l3->alien; | |
638 | /* | |
639 | * FIXME: This check for BAD_ALIEN_MAGIC | |
640 | * should go away when common slab code is taught to | |
641 | * work even without alien caches. | |
642 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
643 | * for alloc_alien_cache, | |
644 | */ | |
645 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
646 | continue; | |
647 | for_each_node(r) { | |
648 | if (alc[r]) | |
649 | lockdep_set_class(&alc[r]->lock, | |
650 | &on_slab_alc_key); | |
651 | } | |
652 | } | |
653 | s++; | |
f1aaee53 AV |
654 | } |
655 | } | |
f1aaee53 | 656 | #else |
056c6241 | 657 | static inline void init_lock_keys(void) |
f1aaee53 AV |
658 | { |
659 | } | |
660 | #endif | |
661 | ||
8f5be20b | 662 | /* |
95402b38 | 663 | * Guard access to the cache-chain. |
8f5be20b | 664 | */ |
fc0abb14 | 665 | static DEFINE_MUTEX(cache_chain_mutex); |
1da177e4 LT |
666 | static struct list_head cache_chain; |
667 | ||
1da177e4 LT |
668 | /* |
669 | * chicken and egg problem: delay the per-cpu array allocation | |
670 | * until the general caches are up. | |
671 | */ | |
672 | static enum { | |
673 | NONE, | |
e498be7d CL |
674 | PARTIAL_AC, |
675 | PARTIAL_L3, | |
1da177e4 LT |
676 | FULL |
677 | } g_cpucache_up; | |
678 | ||
39d24e64 MK |
679 | /* |
680 | * used by boot code to determine if it can use slab based allocator | |
681 | */ | |
682 | int slab_is_available(void) | |
683 | { | |
684 | return g_cpucache_up == FULL; | |
685 | } | |
686 | ||
52bad64d | 687 | static DEFINE_PER_CPU(struct delayed_work, reap_work); |
1da177e4 | 688 | |
343e0d7a | 689 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
690 | { |
691 | return cachep->array[smp_processor_id()]; | |
692 | } | |
693 | ||
a737b3e2 AM |
694 | static inline struct kmem_cache *__find_general_cachep(size_t size, |
695 | gfp_t gfpflags) | |
1da177e4 LT |
696 | { |
697 | struct cache_sizes *csizep = malloc_sizes; | |
698 | ||
699 | #if DEBUG | |
700 | /* This happens if someone tries to call | |
b28a02de PE |
701 | * kmem_cache_create(), or __kmalloc(), before |
702 | * the generic caches are initialized. | |
703 | */ | |
c7e43c78 | 704 | BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); |
1da177e4 | 705 | #endif |
6cb8f913 CL |
706 | if (!size) |
707 | return ZERO_SIZE_PTR; | |
708 | ||
1da177e4 LT |
709 | while (size > csizep->cs_size) |
710 | csizep++; | |
711 | ||
712 | /* | |
0abf40c1 | 713 | * Really subtle: The last entry with cs->cs_size==ULONG_MAX |
1da177e4 LT |
714 | * has cs_{dma,}cachep==NULL. Thus no special case |
715 | * for large kmalloc calls required. | |
716 | */ | |
4b51d669 | 717 | #ifdef CONFIG_ZONE_DMA |
1da177e4 LT |
718 | if (unlikely(gfpflags & GFP_DMA)) |
719 | return csizep->cs_dmacachep; | |
4b51d669 | 720 | #endif |
1da177e4 LT |
721 | return csizep->cs_cachep; |
722 | } | |
723 | ||
b221385b | 724 | static struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) |
97e2bde4 MS |
725 | { |
726 | return __find_general_cachep(size, gfpflags); | |
727 | } | |
97e2bde4 | 728 | |
fbaccacf | 729 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
1da177e4 | 730 | { |
fbaccacf SR |
731 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
732 | } | |
1da177e4 | 733 | |
a737b3e2 AM |
734 | /* |
735 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
736 | */ | |
fbaccacf SR |
737 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
738 | size_t align, int flags, size_t *left_over, | |
739 | unsigned int *num) | |
740 | { | |
741 | int nr_objs; | |
742 | size_t mgmt_size; | |
743 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 744 | |
fbaccacf SR |
745 | /* |
746 | * The slab management structure can be either off the slab or | |
747 | * on it. For the latter case, the memory allocated for a | |
748 | * slab is used for: | |
749 | * | |
750 | * - The struct slab | |
751 | * - One kmem_bufctl_t for each object | |
752 | * - Padding to respect alignment of @align | |
753 | * - @buffer_size bytes for each object | |
754 | * | |
755 | * If the slab management structure is off the slab, then the | |
756 | * alignment will already be calculated into the size. Because | |
757 | * the slabs are all pages aligned, the objects will be at the | |
758 | * correct alignment when allocated. | |
759 | */ | |
760 | if (flags & CFLGS_OFF_SLAB) { | |
761 | mgmt_size = 0; | |
762 | nr_objs = slab_size / buffer_size; | |
763 | ||
764 | if (nr_objs > SLAB_LIMIT) | |
765 | nr_objs = SLAB_LIMIT; | |
766 | } else { | |
767 | /* | |
768 | * Ignore padding for the initial guess. The padding | |
769 | * is at most @align-1 bytes, and @buffer_size is at | |
770 | * least @align. In the worst case, this result will | |
771 | * be one greater than the number of objects that fit | |
772 | * into the memory allocation when taking the padding | |
773 | * into account. | |
774 | */ | |
775 | nr_objs = (slab_size - sizeof(struct slab)) / | |
776 | (buffer_size + sizeof(kmem_bufctl_t)); | |
777 | ||
778 | /* | |
779 | * This calculated number will be either the right | |
780 | * amount, or one greater than what we want. | |
781 | */ | |
782 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | |
783 | > slab_size) | |
784 | nr_objs--; | |
785 | ||
786 | if (nr_objs > SLAB_LIMIT) | |
787 | nr_objs = SLAB_LIMIT; | |
788 | ||
789 | mgmt_size = slab_mgmt_size(nr_objs, align); | |
790 | } | |
791 | *num = nr_objs; | |
792 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
793 | } |
794 | ||
d40cee24 | 795 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 796 | |
a737b3e2 AM |
797 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
798 | char *msg) | |
1da177e4 LT |
799 | { |
800 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 801 | function, cachep->name, msg); |
1da177e4 LT |
802 | dump_stack(); |
803 | } | |
804 | ||
3395ee05 PM |
805 | /* |
806 | * By default on NUMA we use alien caches to stage the freeing of | |
807 | * objects allocated from other nodes. This causes massive memory | |
808 | * inefficiencies when using fake NUMA setup to split memory into a | |
809 | * large number of small nodes, so it can be disabled on the command | |
810 | * line | |
811 | */ | |
812 | ||
813 | static int use_alien_caches __read_mostly = 1; | |
1807a1aa | 814 | static int numa_platform __read_mostly = 1; |
3395ee05 PM |
815 | static int __init noaliencache_setup(char *s) |
816 | { | |
817 | use_alien_caches = 0; | |
818 | return 1; | |
819 | } | |
820 | __setup("noaliencache", noaliencache_setup); | |
821 | ||
8fce4d8e CL |
822 | #ifdef CONFIG_NUMA |
823 | /* | |
824 | * Special reaping functions for NUMA systems called from cache_reap(). | |
825 | * These take care of doing round robin flushing of alien caches (containing | |
826 | * objects freed on different nodes from which they were allocated) and the | |
827 | * flushing of remote pcps by calling drain_node_pages. | |
828 | */ | |
829 | static DEFINE_PER_CPU(unsigned long, reap_node); | |
830 | ||
831 | static void init_reap_node(int cpu) | |
832 | { | |
833 | int node; | |
834 | ||
835 | node = next_node(cpu_to_node(cpu), node_online_map); | |
836 | if (node == MAX_NUMNODES) | |
442295c9 | 837 | node = first_node(node_online_map); |
8fce4d8e | 838 | |
7f6b8876 | 839 | per_cpu(reap_node, cpu) = node; |
8fce4d8e CL |
840 | } |
841 | ||
842 | static void next_reap_node(void) | |
843 | { | |
844 | int node = __get_cpu_var(reap_node); | |
845 | ||
8fce4d8e CL |
846 | node = next_node(node, node_online_map); |
847 | if (unlikely(node >= MAX_NUMNODES)) | |
848 | node = first_node(node_online_map); | |
849 | __get_cpu_var(reap_node) = node; | |
850 | } | |
851 | ||
852 | #else | |
853 | #define init_reap_node(cpu) do { } while (0) | |
854 | #define next_reap_node(void) do { } while (0) | |
855 | #endif | |
856 | ||
1da177e4 LT |
857 | /* |
858 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
859 | * via the workqueue/eventd. | |
860 | * Add the CPU number into the expiration time to minimize the possibility of | |
861 | * the CPUs getting into lockstep and contending for the global cache chain | |
862 | * lock. | |
863 | */ | |
897e679b | 864 | static void __cpuinit start_cpu_timer(int cpu) |
1da177e4 | 865 | { |
52bad64d | 866 | struct delayed_work *reap_work = &per_cpu(reap_work, cpu); |
1da177e4 LT |
867 | |
868 | /* | |
869 | * When this gets called from do_initcalls via cpucache_init(), | |
870 | * init_workqueues() has already run, so keventd will be setup | |
871 | * at that time. | |
872 | */ | |
52bad64d | 873 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 874 | init_reap_node(cpu); |
65f27f38 | 875 | INIT_DELAYED_WORK(reap_work, cache_reap); |
2b284214 AV |
876 | schedule_delayed_work_on(cpu, reap_work, |
877 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
878 | } |
879 | } | |
880 | ||
e498be7d | 881 | static struct array_cache *alloc_arraycache(int node, int entries, |
83b519e8 | 882 | int batchcount, gfp_t gfp) |
1da177e4 | 883 | { |
b28a02de | 884 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
885 | struct array_cache *nc = NULL; |
886 | ||
83b519e8 | 887 | nc = kmalloc_node(memsize, gfp, node); |
d5cff635 CM |
888 | /* |
889 | * The array_cache structures contain pointers to free object. | |
890 | * However, when such objects are allocated or transfered to another | |
891 | * cache the pointers are not cleared and they could be counted as | |
892 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
893 | * not scan such objects. | |
894 | */ | |
895 | kmemleak_no_scan(nc); | |
1da177e4 LT |
896 | if (nc) { |
897 | nc->avail = 0; | |
898 | nc->limit = entries; | |
899 | nc->batchcount = batchcount; | |
900 | nc->touched = 0; | |
e498be7d | 901 | spin_lock_init(&nc->lock); |
1da177e4 LT |
902 | } |
903 | return nc; | |
904 | } | |
905 | ||
3ded175a CL |
906 | /* |
907 | * Transfer objects in one arraycache to another. | |
908 | * Locking must be handled by the caller. | |
909 | * | |
910 | * Return the number of entries transferred. | |
911 | */ | |
912 | static int transfer_objects(struct array_cache *to, | |
913 | struct array_cache *from, unsigned int max) | |
914 | { | |
915 | /* Figure out how many entries to transfer */ | |
916 | int nr = min(min(from->avail, max), to->limit - to->avail); | |
917 | ||
918 | if (!nr) | |
919 | return 0; | |
920 | ||
921 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
922 | sizeof(void *) *nr); | |
923 | ||
924 | from->avail -= nr; | |
925 | to->avail += nr; | |
926 | to->touched = 1; | |
927 | return nr; | |
928 | } | |
929 | ||
765c4507 CL |
930 | #ifndef CONFIG_NUMA |
931 | ||
932 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
933 | #define reap_alien(cachep, l3) do { } while (0) | |
934 | ||
83b519e8 | 935 | static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
765c4507 CL |
936 | { |
937 | return (struct array_cache **)BAD_ALIEN_MAGIC; | |
938 | } | |
939 | ||
940 | static inline void free_alien_cache(struct array_cache **ac_ptr) | |
941 | { | |
942 | } | |
943 | ||
944 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
945 | { | |
946 | return 0; | |
947 | } | |
948 | ||
949 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
950 | gfp_t flags) | |
951 | { | |
952 | return NULL; | |
953 | } | |
954 | ||
8b98c169 | 955 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
956 | gfp_t flags, int nodeid) |
957 | { | |
958 | return NULL; | |
959 | } | |
960 | ||
961 | #else /* CONFIG_NUMA */ | |
962 | ||
8b98c169 | 963 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 964 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 965 | |
83b519e8 | 966 | static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
e498be7d CL |
967 | { |
968 | struct array_cache **ac_ptr; | |
8ef82866 | 969 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
970 | int i; |
971 | ||
972 | if (limit > 1) | |
973 | limit = 12; | |
83b519e8 | 974 | ac_ptr = kmalloc_node(memsize, gfp, node); |
e498be7d CL |
975 | if (ac_ptr) { |
976 | for_each_node(i) { | |
977 | if (i == node || !node_online(i)) { | |
978 | ac_ptr[i] = NULL; | |
979 | continue; | |
980 | } | |
83b519e8 | 981 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); |
e498be7d | 982 | if (!ac_ptr[i]) { |
cc550def | 983 | for (i--; i >= 0; i--) |
e498be7d CL |
984 | kfree(ac_ptr[i]); |
985 | kfree(ac_ptr); | |
986 | return NULL; | |
987 | } | |
988 | } | |
989 | } | |
990 | return ac_ptr; | |
991 | } | |
992 | ||
5295a74c | 993 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
994 | { |
995 | int i; | |
996 | ||
997 | if (!ac_ptr) | |
998 | return; | |
e498be7d | 999 | for_each_node(i) |
b28a02de | 1000 | kfree(ac_ptr[i]); |
e498be7d CL |
1001 | kfree(ac_ptr); |
1002 | } | |
1003 | ||
343e0d7a | 1004 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 1005 | struct array_cache *ac, int node) |
e498be7d CL |
1006 | { |
1007 | struct kmem_list3 *rl3 = cachep->nodelists[node]; | |
1008 | ||
1009 | if (ac->avail) { | |
1010 | spin_lock(&rl3->list_lock); | |
e00946fe CL |
1011 | /* |
1012 | * Stuff objects into the remote nodes shared array first. | |
1013 | * That way we could avoid the overhead of putting the objects | |
1014 | * into the free lists and getting them back later. | |
1015 | */ | |
693f7d36 JS |
1016 | if (rl3->shared) |
1017 | transfer_objects(rl3->shared, ac, ac->limit); | |
e00946fe | 1018 | |
ff69416e | 1019 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d CL |
1020 | ac->avail = 0; |
1021 | spin_unlock(&rl3->list_lock); | |
1022 | } | |
1023 | } | |
1024 | ||
8fce4d8e CL |
1025 | /* |
1026 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1027 | */ | |
1028 | static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) | |
1029 | { | |
1030 | int node = __get_cpu_var(reap_node); | |
1031 | ||
1032 | if (l3->alien) { | |
1033 | struct array_cache *ac = l3->alien[node]; | |
e00946fe CL |
1034 | |
1035 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { | |
8fce4d8e CL |
1036 | __drain_alien_cache(cachep, ac, node); |
1037 | spin_unlock_irq(&ac->lock); | |
1038 | } | |
1039 | } | |
1040 | } | |
1041 | ||
a737b3e2 AM |
1042 | static void drain_alien_cache(struct kmem_cache *cachep, |
1043 | struct array_cache **alien) | |
e498be7d | 1044 | { |
b28a02de | 1045 | int i = 0; |
e498be7d CL |
1046 | struct array_cache *ac; |
1047 | unsigned long flags; | |
1048 | ||
1049 | for_each_online_node(i) { | |
4484ebf1 | 1050 | ac = alien[i]; |
e498be7d CL |
1051 | if (ac) { |
1052 | spin_lock_irqsave(&ac->lock, flags); | |
1053 | __drain_alien_cache(cachep, ac, i); | |
1054 | spin_unlock_irqrestore(&ac->lock, flags); | |
1055 | } | |
1056 | } | |
1057 | } | |
729bd0b7 | 1058 | |
873623df | 1059 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 PE |
1060 | { |
1061 | struct slab *slabp = virt_to_slab(objp); | |
1062 | int nodeid = slabp->nodeid; | |
1063 | struct kmem_list3 *l3; | |
1064 | struct array_cache *alien = NULL; | |
1ca4cb24 PE |
1065 | int node; |
1066 | ||
1067 | node = numa_node_id(); | |
729bd0b7 PE |
1068 | |
1069 | /* | |
1070 | * Make sure we are not freeing a object from another node to the array | |
1071 | * cache on this cpu. | |
1072 | */ | |
62918a03 | 1073 | if (likely(slabp->nodeid == node)) |
729bd0b7 PE |
1074 | return 0; |
1075 | ||
1ca4cb24 | 1076 | l3 = cachep->nodelists[node]; |
729bd0b7 PE |
1077 | STATS_INC_NODEFREES(cachep); |
1078 | if (l3->alien && l3->alien[nodeid]) { | |
1079 | alien = l3->alien[nodeid]; | |
873623df | 1080 | spin_lock(&alien->lock); |
729bd0b7 PE |
1081 | if (unlikely(alien->avail == alien->limit)) { |
1082 | STATS_INC_ACOVERFLOW(cachep); | |
1083 | __drain_alien_cache(cachep, alien, nodeid); | |
1084 | } | |
1085 | alien->entry[alien->avail++] = objp; | |
1086 | spin_unlock(&alien->lock); | |
1087 | } else { | |
1088 | spin_lock(&(cachep->nodelists[nodeid])->list_lock); | |
1089 | free_block(cachep, &objp, 1, nodeid); | |
1090 | spin_unlock(&(cachep->nodelists[nodeid])->list_lock); | |
1091 | } | |
1092 | return 1; | |
1093 | } | |
e498be7d CL |
1094 | #endif |
1095 | ||
fbf1e473 AM |
1096 | static void __cpuinit cpuup_canceled(long cpu) |
1097 | { | |
1098 | struct kmem_cache *cachep; | |
1099 | struct kmem_list3 *l3 = NULL; | |
1100 | int node = cpu_to_node(cpu); | |
a70f7302 | 1101 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 AM |
1102 | |
1103 | list_for_each_entry(cachep, &cache_chain, next) { | |
1104 | struct array_cache *nc; | |
1105 | struct array_cache *shared; | |
1106 | struct array_cache **alien; | |
fbf1e473 | 1107 | |
fbf1e473 AM |
1108 | /* cpu is dead; no one can alloc from it. */ |
1109 | nc = cachep->array[cpu]; | |
1110 | cachep->array[cpu] = NULL; | |
1111 | l3 = cachep->nodelists[node]; | |
1112 | ||
1113 | if (!l3) | |
1114 | goto free_array_cache; | |
1115 | ||
1116 | spin_lock_irq(&l3->list_lock); | |
1117 | ||
1118 | /* Free limit for this kmem_list3 */ | |
1119 | l3->free_limit -= cachep->batchcount; | |
1120 | if (nc) | |
1121 | free_block(cachep, nc->entry, nc->avail, node); | |
1122 | ||
c5f59f08 | 1123 | if (!cpus_empty(*mask)) { |
fbf1e473 AM |
1124 | spin_unlock_irq(&l3->list_lock); |
1125 | goto free_array_cache; | |
1126 | } | |
1127 | ||
1128 | shared = l3->shared; | |
1129 | if (shared) { | |
1130 | free_block(cachep, shared->entry, | |
1131 | shared->avail, node); | |
1132 | l3->shared = NULL; | |
1133 | } | |
1134 | ||
1135 | alien = l3->alien; | |
1136 | l3->alien = NULL; | |
1137 | ||
1138 | spin_unlock_irq(&l3->list_lock); | |
1139 | ||
1140 | kfree(shared); | |
1141 | if (alien) { | |
1142 | drain_alien_cache(cachep, alien); | |
1143 | free_alien_cache(alien); | |
1144 | } | |
1145 | free_array_cache: | |
1146 | kfree(nc); | |
1147 | } | |
1148 | /* | |
1149 | * In the previous loop, all the objects were freed to | |
1150 | * the respective cache's slabs, now we can go ahead and | |
1151 | * shrink each nodelist to its limit. | |
1152 | */ | |
1153 | list_for_each_entry(cachep, &cache_chain, next) { | |
1154 | l3 = cachep->nodelists[node]; | |
1155 | if (!l3) | |
1156 | continue; | |
1157 | drain_freelist(cachep, l3, l3->free_objects); | |
1158 | } | |
1159 | } | |
1160 | ||
1161 | static int __cpuinit cpuup_prepare(long cpu) | |
1da177e4 | 1162 | { |
343e0d7a | 1163 | struct kmem_cache *cachep; |
e498be7d CL |
1164 | struct kmem_list3 *l3 = NULL; |
1165 | int node = cpu_to_node(cpu); | |
ea02e3dd | 1166 | const int memsize = sizeof(struct kmem_list3); |
1da177e4 | 1167 | |
fbf1e473 AM |
1168 | /* |
1169 | * We need to do this right in the beginning since | |
1170 | * alloc_arraycache's are going to use this list. | |
1171 | * kmalloc_node allows us to add the slab to the right | |
1172 | * kmem_list3 and not this cpu's kmem_list3 | |
1173 | */ | |
1174 | ||
1175 | list_for_each_entry(cachep, &cache_chain, next) { | |
a737b3e2 | 1176 | /* |
fbf1e473 AM |
1177 | * Set up the size64 kmemlist for cpu before we can |
1178 | * begin anything. Make sure some other cpu on this | |
1179 | * node has not already allocated this | |
e498be7d | 1180 | */ |
fbf1e473 AM |
1181 | if (!cachep->nodelists[node]) { |
1182 | l3 = kmalloc_node(memsize, GFP_KERNEL, node); | |
1183 | if (!l3) | |
1184 | goto bad; | |
1185 | kmem_list3_init(l3); | |
1186 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
1187 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
e498be7d | 1188 | |
a737b3e2 | 1189 | /* |
fbf1e473 AM |
1190 | * The l3s don't come and go as CPUs come and |
1191 | * go. cache_chain_mutex is sufficient | |
1192 | * protection here. | |
e498be7d | 1193 | */ |
fbf1e473 | 1194 | cachep->nodelists[node] = l3; |
e498be7d CL |
1195 | } |
1196 | ||
fbf1e473 AM |
1197 | spin_lock_irq(&cachep->nodelists[node]->list_lock); |
1198 | cachep->nodelists[node]->free_limit = | |
1199 | (1 + nr_cpus_node(node)) * | |
1200 | cachep->batchcount + cachep->num; | |
1201 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Now we can go ahead with allocating the shared arrays and | |
1206 | * array caches | |
1207 | */ | |
1208 | list_for_each_entry(cachep, &cache_chain, next) { | |
1209 | struct array_cache *nc; | |
1210 | struct array_cache *shared = NULL; | |
1211 | struct array_cache **alien = NULL; | |
1212 | ||
1213 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1214 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1215 | if (!nc) |
1216 | goto bad; | |
1217 | if (cachep->shared) { | |
1218 | shared = alloc_arraycache(node, | |
1219 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1220 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1221 | if (!shared) { |
1222 | kfree(nc); | |
1da177e4 | 1223 | goto bad; |
12d00f6a | 1224 | } |
fbf1e473 AM |
1225 | } |
1226 | if (use_alien_caches) { | |
83b519e8 | 1227 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1228 | if (!alien) { |
1229 | kfree(shared); | |
1230 | kfree(nc); | |
fbf1e473 | 1231 | goto bad; |
12d00f6a | 1232 | } |
fbf1e473 AM |
1233 | } |
1234 | cachep->array[cpu] = nc; | |
1235 | l3 = cachep->nodelists[node]; | |
1236 | BUG_ON(!l3); | |
1237 | ||
1238 | spin_lock_irq(&l3->list_lock); | |
1239 | if (!l3->shared) { | |
1240 | /* | |
1241 | * We are serialised from CPU_DEAD or | |
1242 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1243 | */ | |
1244 | l3->shared = shared; | |
1245 | shared = NULL; | |
1246 | } | |
4484ebf1 | 1247 | #ifdef CONFIG_NUMA |
fbf1e473 AM |
1248 | if (!l3->alien) { |
1249 | l3->alien = alien; | |
1250 | alien = NULL; | |
1da177e4 | 1251 | } |
fbf1e473 AM |
1252 | #endif |
1253 | spin_unlock_irq(&l3->list_lock); | |
1254 | kfree(shared); | |
1255 | free_alien_cache(alien); | |
1256 | } | |
1257 | return 0; | |
1258 | bad: | |
12d00f6a | 1259 | cpuup_canceled(cpu); |
fbf1e473 AM |
1260 | return -ENOMEM; |
1261 | } | |
1262 | ||
1263 | static int __cpuinit cpuup_callback(struct notifier_block *nfb, | |
1264 | unsigned long action, void *hcpu) | |
1265 | { | |
1266 | long cpu = (long)hcpu; | |
1267 | int err = 0; | |
1268 | ||
1269 | switch (action) { | |
fbf1e473 AM |
1270 | case CPU_UP_PREPARE: |
1271 | case CPU_UP_PREPARE_FROZEN: | |
95402b38 | 1272 | mutex_lock(&cache_chain_mutex); |
fbf1e473 | 1273 | err = cpuup_prepare(cpu); |
95402b38 | 1274 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1275 | break; |
1276 | case CPU_ONLINE: | |
8bb78442 | 1277 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1278 | start_cpu_timer(cpu); |
1279 | break; | |
1280 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1281 | case CPU_DOWN_PREPARE: |
8bb78442 | 1282 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 CL |
1283 | /* |
1284 | * Shutdown cache reaper. Note that the cache_chain_mutex is | |
1285 | * held so that if cache_reap() is invoked it cannot do | |
1286 | * anything expensive but will only modify reap_work | |
1287 | * and reschedule the timer. | |
1288 | */ | |
1289 | cancel_rearming_delayed_work(&per_cpu(reap_work, cpu)); | |
1290 | /* Now the cache_reaper is guaranteed to be not running. */ | |
1291 | per_cpu(reap_work, cpu).work.func = NULL; | |
1292 | break; | |
1293 | case CPU_DOWN_FAILED: | |
8bb78442 | 1294 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1295 | start_cpu_timer(cpu); |
1296 | break; | |
1da177e4 | 1297 | case CPU_DEAD: |
8bb78442 | 1298 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1299 | /* |
1300 | * Even if all the cpus of a node are down, we don't free the | |
1301 | * kmem_list3 of any cache. This to avoid a race between | |
1302 | * cpu_down, and a kmalloc allocation from another cpu for | |
1303 | * memory from the node of the cpu going down. The list3 | |
1304 | * structure is usually allocated from kmem_cache_create() and | |
1305 | * gets destroyed at kmem_cache_destroy(). | |
1306 | */ | |
183ff22b | 1307 | /* fall through */ |
8f5be20b | 1308 | #endif |
1da177e4 | 1309 | case CPU_UP_CANCELED: |
8bb78442 | 1310 | case CPU_UP_CANCELED_FROZEN: |
95402b38 | 1311 | mutex_lock(&cache_chain_mutex); |
fbf1e473 | 1312 | cpuup_canceled(cpu); |
fc0abb14 | 1313 | mutex_unlock(&cache_chain_mutex); |
1da177e4 | 1314 | break; |
1da177e4 | 1315 | } |
fbf1e473 | 1316 | return err ? NOTIFY_BAD : NOTIFY_OK; |
1da177e4 LT |
1317 | } |
1318 | ||
74b85f37 CS |
1319 | static struct notifier_block __cpuinitdata cpucache_notifier = { |
1320 | &cpuup_callback, NULL, 0 | |
1321 | }; | |
1da177e4 | 1322 | |
e498be7d CL |
1323 | /* |
1324 | * swap the static kmem_list3 with kmalloced memory | |
1325 | */ | |
a737b3e2 AM |
1326 | static void init_list(struct kmem_cache *cachep, struct kmem_list3 *list, |
1327 | int nodeid) | |
e498be7d CL |
1328 | { |
1329 | struct kmem_list3 *ptr; | |
1330 | ||
83b519e8 | 1331 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_NOWAIT, nodeid); |
e498be7d CL |
1332 | BUG_ON(!ptr); |
1333 | ||
e498be7d | 1334 | memcpy(ptr, list, sizeof(struct kmem_list3)); |
2b2d5493 IM |
1335 | /* |
1336 | * Do not assume that spinlocks can be initialized via memcpy: | |
1337 | */ | |
1338 | spin_lock_init(&ptr->list_lock); | |
1339 | ||
e498be7d CL |
1340 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
1341 | cachep->nodelists[nodeid] = ptr; | |
e498be7d CL |
1342 | } |
1343 | ||
556a169d PE |
1344 | /* |
1345 | * For setting up all the kmem_list3s for cache whose buffer_size is same as | |
1346 | * size of kmem_list3. | |
1347 | */ | |
1348 | static void __init set_up_list3s(struct kmem_cache *cachep, int index) | |
1349 | { | |
1350 | int node; | |
1351 | ||
1352 | for_each_online_node(node) { | |
1353 | cachep->nodelists[node] = &initkmem_list3[index + node]; | |
1354 | cachep->nodelists[node]->next_reap = jiffies + | |
1355 | REAPTIMEOUT_LIST3 + | |
1356 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1357 | } | |
1358 | } | |
1359 | ||
a737b3e2 AM |
1360 | /* |
1361 | * Initialisation. Called after the page allocator have been initialised and | |
1362 | * before smp_init(). | |
1da177e4 LT |
1363 | */ |
1364 | void __init kmem_cache_init(void) | |
1365 | { | |
1366 | size_t left_over; | |
1367 | struct cache_sizes *sizes; | |
1368 | struct cache_names *names; | |
e498be7d | 1369 | int i; |
07ed76b2 | 1370 | int order; |
1ca4cb24 | 1371 | int node; |
e498be7d | 1372 | |
1807a1aa | 1373 | if (num_possible_nodes() == 1) { |
62918a03 | 1374 | use_alien_caches = 0; |
1807a1aa SS |
1375 | numa_platform = 0; |
1376 | } | |
62918a03 | 1377 | |
e498be7d CL |
1378 | for (i = 0; i < NUM_INIT_LISTS; i++) { |
1379 | kmem_list3_init(&initkmem_list3[i]); | |
1380 | if (i < MAX_NUMNODES) | |
1381 | cache_cache.nodelists[i] = NULL; | |
1382 | } | |
556a169d | 1383 | set_up_list3s(&cache_cache, CACHE_CACHE); |
1da177e4 LT |
1384 | |
1385 | /* | |
1386 | * Fragmentation resistance on low memory - only use bigger | |
1387 | * page orders on machines with more than 32MB of memory. | |
1388 | */ | |
1389 | if (num_physpages > (32 << 20) >> PAGE_SHIFT) | |
1390 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; | |
1391 | ||
1da177e4 LT |
1392 | /* Bootstrap is tricky, because several objects are allocated |
1393 | * from caches that do not exist yet: | |
a737b3e2 AM |
1394 | * 1) initialize the cache_cache cache: it contains the struct |
1395 | * kmem_cache structures of all caches, except cache_cache itself: | |
1396 | * cache_cache is statically allocated. | |
e498be7d CL |
1397 | * Initially an __init data area is used for the head array and the |
1398 | * kmem_list3 structures, it's replaced with a kmalloc allocated | |
1399 | * array at the end of the bootstrap. | |
1da177e4 | 1400 | * 2) Create the first kmalloc cache. |
343e0d7a | 1401 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1402 | * An __init data area is used for the head array. |
1403 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1404 | * head arrays. | |
1da177e4 LT |
1405 | * 4) Replace the __init data head arrays for cache_cache and the first |
1406 | * kmalloc cache with kmalloc allocated arrays. | |
e498be7d CL |
1407 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
1408 | * the other cache's with kmalloc allocated memory. | |
1409 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1410 | */ |
1411 | ||
1ca4cb24 PE |
1412 | node = numa_node_id(); |
1413 | ||
1da177e4 | 1414 | /* 1) create the cache_cache */ |
1da177e4 LT |
1415 | INIT_LIST_HEAD(&cache_chain); |
1416 | list_add(&cache_cache.next, &cache_chain); | |
1417 | cache_cache.colour_off = cache_line_size(); | |
1418 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; | |
ec1f5eee | 1419 | cache_cache.nodelists[node] = &initkmem_list3[CACHE_CACHE + node]; |
1da177e4 | 1420 | |
8da3430d ED |
1421 | /* |
1422 | * struct kmem_cache size depends on nr_node_ids, which | |
1423 | * can be less than MAX_NUMNODES. | |
1424 | */ | |
1425 | cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) + | |
1426 | nr_node_ids * sizeof(struct kmem_list3 *); | |
1427 | #if DEBUG | |
1428 | cache_cache.obj_size = cache_cache.buffer_size; | |
1429 | #endif | |
a737b3e2 AM |
1430 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, |
1431 | cache_line_size()); | |
6a2d7a95 ED |
1432 | cache_cache.reciprocal_buffer_size = |
1433 | reciprocal_value(cache_cache.buffer_size); | |
1da177e4 | 1434 | |
07ed76b2 JS |
1435 | for (order = 0; order < MAX_ORDER; order++) { |
1436 | cache_estimate(order, cache_cache.buffer_size, | |
1437 | cache_line_size(), 0, &left_over, &cache_cache.num); | |
1438 | if (cache_cache.num) | |
1439 | break; | |
1440 | } | |
40094fa6 | 1441 | BUG_ON(!cache_cache.num); |
07ed76b2 | 1442 | cache_cache.gfporder = order; |
b28a02de | 1443 | cache_cache.colour = left_over / cache_cache.colour_off; |
b28a02de PE |
1444 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
1445 | sizeof(struct slab), cache_line_size()); | |
1da177e4 LT |
1446 | |
1447 | /* 2+3) create the kmalloc caches */ | |
1448 | sizes = malloc_sizes; | |
1449 | names = cache_names; | |
1450 | ||
a737b3e2 AM |
1451 | /* |
1452 | * Initialize the caches that provide memory for the array cache and the | |
1453 | * kmem_list3 structures first. Without this, further allocations will | |
1454 | * bug. | |
e498be7d CL |
1455 | */ |
1456 | ||
1457 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, | |
a737b3e2 AM |
1458 | sizes[INDEX_AC].cs_size, |
1459 | ARCH_KMALLOC_MINALIGN, | |
1460 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1461 | NULL); |
e498be7d | 1462 | |
a737b3e2 | 1463 | if (INDEX_AC != INDEX_L3) { |
e498be7d | 1464 | sizes[INDEX_L3].cs_cachep = |
a737b3e2 AM |
1465 | kmem_cache_create(names[INDEX_L3].name, |
1466 | sizes[INDEX_L3].cs_size, | |
1467 | ARCH_KMALLOC_MINALIGN, | |
1468 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1469 | NULL); |
a737b3e2 | 1470 | } |
e498be7d | 1471 | |
e0a42726 IM |
1472 | slab_early_init = 0; |
1473 | ||
1da177e4 | 1474 | while (sizes->cs_size != ULONG_MAX) { |
e498be7d CL |
1475 | /* |
1476 | * For performance, all the general caches are L1 aligned. | |
1da177e4 LT |
1477 | * This should be particularly beneficial on SMP boxes, as it |
1478 | * eliminates "false sharing". | |
1479 | * Note for systems short on memory removing the alignment will | |
e498be7d CL |
1480 | * allow tighter packing of the smaller caches. |
1481 | */ | |
a737b3e2 | 1482 | if (!sizes->cs_cachep) { |
e498be7d | 1483 | sizes->cs_cachep = kmem_cache_create(names->name, |
a737b3e2 AM |
1484 | sizes->cs_size, |
1485 | ARCH_KMALLOC_MINALIGN, | |
1486 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1487 | NULL); |
a737b3e2 | 1488 | } |
4b51d669 CL |
1489 | #ifdef CONFIG_ZONE_DMA |
1490 | sizes->cs_dmacachep = kmem_cache_create( | |
1491 | names->name_dma, | |
a737b3e2 AM |
1492 | sizes->cs_size, |
1493 | ARCH_KMALLOC_MINALIGN, | |
1494 | ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA| | |
1495 | SLAB_PANIC, | |
20c2df83 | 1496 | NULL); |
4b51d669 | 1497 | #endif |
1da177e4 LT |
1498 | sizes++; |
1499 | names++; | |
1500 | } | |
1501 | /* 4) Replace the bootstrap head arrays */ | |
1502 | { | |
2b2d5493 | 1503 | struct array_cache *ptr; |
e498be7d | 1504 | |
83b519e8 | 1505 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1506 | |
9a2dba4b PE |
1507 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
1508 | memcpy(ptr, cpu_cache_get(&cache_cache), | |
b28a02de | 1509 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1510 | /* |
1511 | * Do not assume that spinlocks can be initialized via memcpy: | |
1512 | */ | |
1513 | spin_lock_init(&ptr->lock); | |
1514 | ||
1da177e4 | 1515 | cache_cache.array[smp_processor_id()] = ptr; |
e498be7d | 1516 | |
83b519e8 | 1517 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1518 | |
9a2dba4b | 1519 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
b28a02de | 1520 | != &initarray_generic.cache); |
9a2dba4b | 1521 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
b28a02de | 1522 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1523 | /* |
1524 | * Do not assume that spinlocks can be initialized via memcpy: | |
1525 | */ | |
1526 | spin_lock_init(&ptr->lock); | |
1527 | ||
e498be7d | 1528 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
b28a02de | 1529 | ptr; |
1da177e4 | 1530 | } |
e498be7d CL |
1531 | /* 5) Replace the bootstrap kmem_list3's */ |
1532 | { | |
1ca4cb24 PE |
1533 | int nid; |
1534 | ||
9c09a95c | 1535 | for_each_online_node(nid) { |
ec1f5eee | 1536 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE + nid], nid); |
556a169d | 1537 | |
e498be7d | 1538 | init_list(malloc_sizes[INDEX_AC].cs_cachep, |
1ca4cb24 | 1539 | &initkmem_list3[SIZE_AC + nid], nid); |
e498be7d CL |
1540 | |
1541 | if (INDEX_AC != INDEX_L3) { | |
1542 | init_list(malloc_sizes[INDEX_L3].cs_cachep, | |
1ca4cb24 | 1543 | &initkmem_list3[SIZE_L3 + nid], nid); |
e498be7d CL |
1544 | } |
1545 | } | |
1546 | } | |
1da177e4 | 1547 | |
e498be7d | 1548 | /* 6) resize the head arrays to their final sizes */ |
1da177e4 | 1549 | { |
343e0d7a | 1550 | struct kmem_cache *cachep; |
fc0abb14 | 1551 | mutex_lock(&cache_chain_mutex); |
1da177e4 | 1552 | list_for_each_entry(cachep, &cache_chain, next) |
83b519e8 | 1553 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
2ed3a4ef | 1554 | BUG(); |
fc0abb14 | 1555 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1556 | } |
1557 | ||
056c6241 RT |
1558 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1559 | init_lock_keys(); | |
1560 | ||
1561 | ||
1da177e4 LT |
1562 | /* Done! */ |
1563 | g_cpucache_up = FULL; | |
1564 | ||
a737b3e2 AM |
1565 | /* |
1566 | * Register a cpu startup notifier callback that initializes | |
1567 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1568 | */ |
1569 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1570 | |
a737b3e2 AM |
1571 | /* |
1572 | * The reap timers are started later, with a module init call: That part | |
1573 | * of the kernel is not yet operational. | |
1da177e4 LT |
1574 | */ |
1575 | } | |
1576 | ||
1577 | static int __init cpucache_init(void) | |
1578 | { | |
1579 | int cpu; | |
1580 | ||
a737b3e2 AM |
1581 | /* |
1582 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1583 | */ |
e498be7d | 1584 | for_each_online_cpu(cpu) |
a737b3e2 | 1585 | start_cpu_timer(cpu); |
1da177e4 LT |
1586 | return 0; |
1587 | } | |
1da177e4 LT |
1588 | __initcall(cpucache_init); |
1589 | ||
1590 | /* | |
1591 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1592 | * | |
1593 | * If we requested dmaable memory, we will get it. Even if we | |
1594 | * did not request dmaable memory, we might get it, but that | |
1595 | * would be relatively rare and ignorable. | |
1596 | */ | |
343e0d7a | 1597 | static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1da177e4 LT |
1598 | { |
1599 | struct page *page; | |
e1b6aa6f | 1600 | int nr_pages; |
1da177e4 LT |
1601 | int i; |
1602 | ||
d6fef9da | 1603 | #ifndef CONFIG_MMU |
e1b6aa6f CH |
1604 | /* |
1605 | * Nommu uses slab's for process anonymous memory allocations, and thus | |
1606 | * requires __GFP_COMP to properly refcount higher order allocations | |
d6fef9da | 1607 | */ |
e1b6aa6f | 1608 | flags |= __GFP_COMP; |
d6fef9da | 1609 | #endif |
765c4507 | 1610 | |
3c517a61 | 1611 | flags |= cachep->gfpflags; |
e12ba74d MG |
1612 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1613 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f CH |
1614 | |
1615 | page = alloc_pages_node(nodeid, flags, cachep->gfporder); | |
1da177e4 LT |
1616 | if (!page) |
1617 | return NULL; | |
1da177e4 | 1618 | |
e1b6aa6f | 1619 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1620 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1621 | add_zone_page_state(page_zone(page), |
1622 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1623 | else | |
1624 | add_zone_page_state(page_zone(page), | |
1625 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
e1b6aa6f CH |
1626 | for (i = 0; i < nr_pages; i++) |
1627 | __SetPageSlab(page + i); | |
c175eea4 PE |
1628 | |
1629 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) | |
1630 | kmemcheck_alloc_shadow(cachep, flags, nodeid, page, cachep->gfporder); | |
1631 | ||
e1b6aa6f | 1632 | return page_address(page); |
1da177e4 LT |
1633 | } |
1634 | ||
1635 | /* | |
1636 | * Interface to system's page release. | |
1637 | */ | |
343e0d7a | 1638 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
1da177e4 | 1639 | { |
b28a02de | 1640 | unsigned long i = (1 << cachep->gfporder); |
1da177e4 LT |
1641 | struct page *page = virt_to_page(addr); |
1642 | const unsigned long nr_freed = i; | |
1643 | ||
c175eea4 PE |
1644 | if (kmemcheck_page_is_tracked(page)) |
1645 | kmemcheck_free_shadow(cachep, page, cachep->gfporder); | |
1646 | ||
972d1a7b CL |
1647 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1648 | sub_zone_page_state(page_zone(page), | |
1649 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1650 | else | |
1651 | sub_zone_page_state(page_zone(page), | |
1652 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
1da177e4 | 1653 | while (i--) { |
f205b2fe NP |
1654 | BUG_ON(!PageSlab(page)); |
1655 | __ClearPageSlab(page); | |
1da177e4 LT |
1656 | page++; |
1657 | } | |
1da177e4 LT |
1658 | if (current->reclaim_state) |
1659 | current->reclaim_state->reclaimed_slab += nr_freed; | |
1660 | free_pages((unsigned long)addr, cachep->gfporder); | |
1da177e4 LT |
1661 | } |
1662 | ||
1663 | static void kmem_rcu_free(struct rcu_head *head) | |
1664 | { | |
b28a02de | 1665 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
343e0d7a | 1666 | struct kmem_cache *cachep = slab_rcu->cachep; |
1da177e4 LT |
1667 | |
1668 | kmem_freepages(cachep, slab_rcu->addr); | |
1669 | if (OFF_SLAB(cachep)) | |
1670 | kmem_cache_free(cachep->slabp_cache, slab_rcu); | |
1671 | } | |
1672 | ||
1673 | #if DEBUG | |
1674 | ||
1675 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1676 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1677 | unsigned long caller) |
1da177e4 | 1678 | { |
3dafccf2 | 1679 | int size = obj_size(cachep); |
1da177e4 | 1680 | |
3dafccf2 | 1681 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1682 | |
b28a02de | 1683 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1684 | return; |
1685 | ||
b28a02de PE |
1686 | *addr++ = 0x12345678; |
1687 | *addr++ = caller; | |
1688 | *addr++ = smp_processor_id(); | |
1689 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1690 | { |
1691 | unsigned long *sptr = &caller; | |
1692 | unsigned long svalue; | |
1693 | ||
1694 | while (!kstack_end(sptr)) { | |
1695 | svalue = *sptr++; | |
1696 | if (kernel_text_address(svalue)) { | |
b28a02de | 1697 | *addr++ = svalue; |
1da177e4 LT |
1698 | size -= sizeof(unsigned long); |
1699 | if (size <= sizeof(unsigned long)) | |
1700 | break; | |
1701 | } | |
1702 | } | |
1703 | ||
1704 | } | |
b28a02de | 1705 | *addr++ = 0x87654321; |
1da177e4 LT |
1706 | } |
1707 | #endif | |
1708 | ||
343e0d7a | 1709 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1710 | { |
3dafccf2 MS |
1711 | int size = obj_size(cachep); |
1712 | addr = &((char *)addr)[obj_offset(cachep)]; | |
1da177e4 LT |
1713 | |
1714 | memset(addr, val, size); | |
b28a02de | 1715 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1716 | } |
1717 | ||
1718 | static void dump_line(char *data, int offset, int limit) | |
1719 | { | |
1720 | int i; | |
aa83aa40 DJ |
1721 | unsigned char error = 0; |
1722 | int bad_count = 0; | |
1723 | ||
1da177e4 | 1724 | printk(KERN_ERR "%03x:", offset); |
aa83aa40 DJ |
1725 | for (i = 0; i < limit; i++) { |
1726 | if (data[offset + i] != POISON_FREE) { | |
1727 | error = data[offset + i]; | |
1728 | bad_count++; | |
1729 | } | |
b28a02de | 1730 | printk(" %02x", (unsigned char)data[offset + i]); |
aa83aa40 | 1731 | } |
1da177e4 | 1732 | printk("\n"); |
aa83aa40 DJ |
1733 | |
1734 | if (bad_count == 1) { | |
1735 | error ^= POISON_FREE; | |
1736 | if (!(error & (error - 1))) { | |
1737 | printk(KERN_ERR "Single bit error detected. Probably " | |
1738 | "bad RAM.\n"); | |
1739 | #ifdef CONFIG_X86 | |
1740 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1741 | "test tool.\n"); | |
1742 | #else | |
1743 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1744 | #endif | |
1745 | } | |
1746 | } | |
1da177e4 LT |
1747 | } |
1748 | #endif | |
1749 | ||
1750 | #if DEBUG | |
1751 | ||
343e0d7a | 1752 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1753 | { |
1754 | int i, size; | |
1755 | char *realobj; | |
1756 | ||
1757 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1758 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1759 | *dbg_redzone1(cachep, objp), |
1760 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1761 | } |
1762 | ||
1763 | if (cachep->flags & SLAB_STORE_USER) { | |
1764 | printk(KERN_ERR "Last user: [<%p>]", | |
a737b3e2 | 1765 | *dbg_userword(cachep, objp)); |
1da177e4 | 1766 | print_symbol("(%s)", |
a737b3e2 | 1767 | (unsigned long)*dbg_userword(cachep, objp)); |
1da177e4 LT |
1768 | printk("\n"); |
1769 | } | |
3dafccf2 MS |
1770 | realobj = (char *)objp + obj_offset(cachep); |
1771 | size = obj_size(cachep); | |
b28a02de | 1772 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1773 | int limit; |
1774 | limit = 16; | |
b28a02de PE |
1775 | if (i + limit > size) |
1776 | limit = size - i; | |
1da177e4 LT |
1777 | dump_line(realobj, i, limit); |
1778 | } | |
1779 | } | |
1780 | ||
343e0d7a | 1781 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1782 | { |
1783 | char *realobj; | |
1784 | int size, i; | |
1785 | int lines = 0; | |
1786 | ||
3dafccf2 MS |
1787 | realobj = (char *)objp + obj_offset(cachep); |
1788 | size = obj_size(cachep); | |
1da177e4 | 1789 | |
b28a02de | 1790 | for (i = 0; i < size; i++) { |
1da177e4 | 1791 | char exp = POISON_FREE; |
b28a02de | 1792 | if (i == size - 1) |
1da177e4 LT |
1793 | exp = POISON_END; |
1794 | if (realobj[i] != exp) { | |
1795 | int limit; | |
1796 | /* Mismatch ! */ | |
1797 | /* Print header */ | |
1798 | if (lines == 0) { | |
b28a02de | 1799 | printk(KERN_ERR |
e94a40c5 DH |
1800 | "Slab corruption: %s start=%p, len=%d\n", |
1801 | cachep->name, realobj, size); | |
1da177e4 LT |
1802 | print_objinfo(cachep, objp, 0); |
1803 | } | |
1804 | /* Hexdump the affected line */ | |
b28a02de | 1805 | i = (i / 16) * 16; |
1da177e4 | 1806 | limit = 16; |
b28a02de PE |
1807 | if (i + limit > size) |
1808 | limit = size - i; | |
1da177e4 LT |
1809 | dump_line(realobj, i, limit); |
1810 | i += 16; | |
1811 | lines++; | |
1812 | /* Limit to 5 lines */ | |
1813 | if (lines > 5) | |
1814 | break; | |
1815 | } | |
1816 | } | |
1817 | if (lines != 0) { | |
1818 | /* Print some data about the neighboring objects, if they | |
1819 | * exist: | |
1820 | */ | |
6ed5eb22 | 1821 | struct slab *slabp = virt_to_slab(objp); |
8fea4e96 | 1822 | unsigned int objnr; |
1da177e4 | 1823 | |
8fea4e96 | 1824 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 | 1825 | if (objnr) { |
8fea4e96 | 1826 | objp = index_to_obj(cachep, slabp, objnr - 1); |
3dafccf2 | 1827 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1828 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1829 | realobj, size); |
1da177e4 LT |
1830 | print_objinfo(cachep, objp, 2); |
1831 | } | |
b28a02de | 1832 | if (objnr + 1 < cachep->num) { |
8fea4e96 | 1833 | objp = index_to_obj(cachep, slabp, objnr + 1); |
3dafccf2 | 1834 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1835 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1836 | realobj, size); |
1da177e4 LT |
1837 | print_objinfo(cachep, objp, 2); |
1838 | } | |
1839 | } | |
1840 | } | |
1841 | #endif | |
1842 | ||
12dd36fa | 1843 | #if DEBUG |
e79aec29 | 1844 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 | 1845 | { |
1da177e4 LT |
1846 | int i; |
1847 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 1848 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
1849 | |
1850 | if (cachep->flags & SLAB_POISON) { | |
1851 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 AM |
1852 | if (cachep->buffer_size % PAGE_SIZE == 0 && |
1853 | OFF_SLAB(cachep)) | |
b28a02de | 1854 | kernel_map_pages(virt_to_page(objp), |
a737b3e2 | 1855 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
1856 | else |
1857 | check_poison_obj(cachep, objp); | |
1858 | #else | |
1859 | check_poison_obj(cachep, objp); | |
1860 | #endif | |
1861 | } | |
1862 | if (cachep->flags & SLAB_RED_ZONE) { | |
1863 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1864 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1865 | "was overwritten"); |
1da177e4 LT |
1866 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1867 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1868 | "was overwritten"); |
1da177e4 | 1869 | } |
1da177e4 | 1870 | } |
12dd36fa | 1871 | } |
1da177e4 | 1872 | #else |
e79aec29 | 1873 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 1874 | { |
12dd36fa | 1875 | } |
1da177e4 LT |
1876 | #endif |
1877 | ||
911851e6 RD |
1878 | /** |
1879 | * slab_destroy - destroy and release all objects in a slab | |
1880 | * @cachep: cache pointer being destroyed | |
1881 | * @slabp: slab pointer being destroyed | |
1882 | * | |
12dd36fa | 1883 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
1884 | * Before calling the slab must have been unlinked from the cache. The |
1885 | * cache-lock is not held/needed. | |
12dd36fa | 1886 | */ |
343e0d7a | 1887 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa MD |
1888 | { |
1889 | void *addr = slabp->s_mem - slabp->colouroff; | |
1890 | ||
e79aec29 | 1891 | slab_destroy_debugcheck(cachep, slabp); |
1da177e4 LT |
1892 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
1893 | struct slab_rcu *slab_rcu; | |
1894 | ||
b28a02de | 1895 | slab_rcu = (struct slab_rcu *)slabp; |
1da177e4 LT |
1896 | slab_rcu->cachep = cachep; |
1897 | slab_rcu->addr = addr; | |
1898 | call_rcu(&slab_rcu->head, kmem_rcu_free); | |
1899 | } else { | |
1900 | kmem_freepages(cachep, addr); | |
873623df IM |
1901 | if (OFF_SLAB(cachep)) |
1902 | kmem_cache_free(cachep->slabp_cache, slabp); | |
1da177e4 LT |
1903 | } |
1904 | } | |
1905 | ||
117f6eb1 CL |
1906 | static void __kmem_cache_destroy(struct kmem_cache *cachep) |
1907 | { | |
1908 | int i; | |
1909 | struct kmem_list3 *l3; | |
1910 | ||
1911 | for_each_online_cpu(i) | |
1912 | kfree(cachep->array[i]); | |
1913 | ||
1914 | /* NUMA: free the list3 structures */ | |
1915 | for_each_online_node(i) { | |
1916 | l3 = cachep->nodelists[i]; | |
1917 | if (l3) { | |
1918 | kfree(l3->shared); | |
1919 | free_alien_cache(l3->alien); | |
1920 | kfree(l3); | |
1921 | } | |
1922 | } | |
1923 | kmem_cache_free(&cache_cache, cachep); | |
1924 | } | |
1925 | ||
1926 | ||
4d268eba | 1927 | /** |
a70773dd RD |
1928 | * calculate_slab_order - calculate size (page order) of slabs |
1929 | * @cachep: pointer to the cache that is being created | |
1930 | * @size: size of objects to be created in this cache. | |
1931 | * @align: required alignment for the objects. | |
1932 | * @flags: slab allocation flags | |
1933 | * | |
1934 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1935 | * |
1936 | * This could be made much more intelligent. For now, try to avoid using | |
1937 | * high order pages for slabs. When the gfp() functions are more friendly | |
1938 | * towards high-order requests, this should be changed. | |
1939 | */ | |
a737b3e2 | 1940 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 1941 | size_t size, size_t align, unsigned long flags) |
4d268eba | 1942 | { |
b1ab41c4 | 1943 | unsigned long offslab_limit; |
4d268eba | 1944 | size_t left_over = 0; |
9888e6fa | 1945 | int gfporder; |
4d268eba | 1946 | |
0aa817f0 | 1947 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1948 | unsigned int num; |
1949 | size_t remainder; | |
1950 | ||
9888e6fa | 1951 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
1952 | if (!num) |
1953 | continue; | |
9888e6fa | 1954 | |
b1ab41c4 IM |
1955 | if (flags & CFLGS_OFF_SLAB) { |
1956 | /* | |
1957 | * Max number of objs-per-slab for caches which | |
1958 | * use off-slab slabs. Needed to avoid a possible | |
1959 | * looping condition in cache_grow(). | |
1960 | */ | |
1961 | offslab_limit = size - sizeof(struct slab); | |
1962 | offslab_limit /= sizeof(kmem_bufctl_t); | |
1963 | ||
1964 | if (num > offslab_limit) | |
1965 | break; | |
1966 | } | |
4d268eba | 1967 | |
9888e6fa | 1968 | /* Found something acceptable - save it away */ |
4d268eba | 1969 | cachep->num = num; |
9888e6fa | 1970 | cachep->gfporder = gfporder; |
4d268eba PE |
1971 | left_over = remainder; |
1972 | ||
f78bb8ad LT |
1973 | /* |
1974 | * A VFS-reclaimable slab tends to have most allocations | |
1975 | * as GFP_NOFS and we really don't want to have to be allocating | |
1976 | * higher-order pages when we are unable to shrink dcache. | |
1977 | */ | |
1978 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1979 | break; | |
1980 | ||
4d268eba PE |
1981 | /* |
1982 | * Large number of objects is good, but very large slabs are | |
1983 | * currently bad for the gfp()s. | |
1984 | */ | |
9888e6fa | 1985 | if (gfporder >= slab_break_gfp_order) |
4d268eba PE |
1986 | break; |
1987 | ||
9888e6fa LT |
1988 | /* |
1989 | * Acceptable internal fragmentation? | |
1990 | */ | |
a737b3e2 | 1991 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1992 | break; |
1993 | } | |
1994 | return left_over; | |
1995 | } | |
1996 | ||
83b519e8 | 1997 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1998 | { |
2ed3a4ef | 1999 | if (g_cpucache_up == FULL) |
83b519e8 | 2000 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2001 | |
f30cf7d1 PE |
2002 | if (g_cpucache_up == NONE) { |
2003 | /* | |
2004 | * Note: the first kmem_cache_create must create the cache | |
2005 | * that's used by kmalloc(24), otherwise the creation of | |
2006 | * further caches will BUG(). | |
2007 | */ | |
2008 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2009 | ||
2010 | /* | |
2011 | * If the cache that's used by kmalloc(sizeof(kmem_list3)) is | |
2012 | * the first cache, then we need to set up all its list3s, | |
2013 | * otherwise the creation of further caches will BUG(). | |
2014 | */ | |
2015 | set_up_list3s(cachep, SIZE_AC); | |
2016 | if (INDEX_AC == INDEX_L3) | |
2017 | g_cpucache_up = PARTIAL_L3; | |
2018 | else | |
2019 | g_cpucache_up = PARTIAL_AC; | |
2020 | } else { | |
2021 | cachep->array[smp_processor_id()] = | |
83b519e8 | 2022 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 PE |
2023 | |
2024 | if (g_cpucache_up == PARTIAL_AC) { | |
2025 | set_up_list3s(cachep, SIZE_L3); | |
2026 | g_cpucache_up = PARTIAL_L3; | |
2027 | } else { | |
2028 | int node; | |
556a169d | 2029 | for_each_online_node(node) { |
f30cf7d1 PE |
2030 | cachep->nodelists[node] = |
2031 | kmalloc_node(sizeof(struct kmem_list3), | |
2032 | GFP_KERNEL, node); | |
2033 | BUG_ON(!cachep->nodelists[node]); | |
2034 | kmem_list3_init(cachep->nodelists[node]); | |
2035 | } | |
2036 | } | |
2037 | } | |
2038 | cachep->nodelists[numa_node_id()]->next_reap = | |
2039 | jiffies + REAPTIMEOUT_LIST3 + | |
2040 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
2041 | ||
2042 | cpu_cache_get(cachep)->avail = 0; | |
2043 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2044 | cpu_cache_get(cachep)->batchcount = 1; | |
2045 | cpu_cache_get(cachep)->touched = 0; | |
2046 | cachep->batchcount = 1; | |
2047 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2048 | return 0; |
f30cf7d1 PE |
2049 | } |
2050 | ||
1da177e4 LT |
2051 | /** |
2052 | * kmem_cache_create - Create a cache. | |
2053 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
2054 | * @size: The size of objects to be created in this cache. | |
2055 | * @align: The required alignment for the objects. | |
2056 | * @flags: SLAB flags | |
2057 | * @ctor: A constructor for the objects. | |
1da177e4 LT |
2058 | * |
2059 | * Returns a ptr to the cache on success, NULL on failure. | |
2060 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2061 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 LT |
2062 | * |
2063 | * @name must be valid until the cache is destroyed. This implies that | |
a737b3e2 | 2064 | * the module calling this has to destroy the cache before getting unloaded. |
249da166 CM |
2065 | * Note that kmem_cache_name() is not guaranteed to return the same pointer, |
2066 | * therefore applications must manage it themselves. | |
a737b3e2 | 2067 | * |
1da177e4 LT |
2068 | * The flags are |
2069 | * | |
2070 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2071 | * to catch references to uninitialised memory. | |
2072 | * | |
2073 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2074 | * for buffer overruns. | |
2075 | * | |
1da177e4 LT |
2076 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2077 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2078 | * as davem. | |
2079 | */ | |
343e0d7a | 2080 | struct kmem_cache * |
1da177e4 | 2081 | kmem_cache_create (const char *name, size_t size, size_t align, |
51cc5068 | 2082 | unsigned long flags, void (*ctor)(void *)) |
1da177e4 LT |
2083 | { |
2084 | size_t left_over, slab_size, ralign; | |
7a7c381d | 2085 | struct kmem_cache *cachep = NULL, *pc; |
83b519e8 | 2086 | gfp_t gfp; |
1da177e4 LT |
2087 | |
2088 | /* | |
2089 | * Sanity checks... these are all serious usage bugs. | |
2090 | */ | |
a737b3e2 | 2091 | if (!name || in_interrupt() || (size < BYTES_PER_WORD) || |
20c2df83 | 2092 | size > KMALLOC_MAX_SIZE) { |
d40cee24 | 2093 | printk(KERN_ERR "%s: Early error in slab %s\n", __func__, |
a737b3e2 | 2094 | name); |
b28a02de PE |
2095 | BUG(); |
2096 | } | |
1da177e4 | 2097 | |
f0188f47 | 2098 | /* |
8f5be20b | 2099 | * We use cache_chain_mutex to ensure a consistent view of |
174596a0 | 2100 | * cpu_online_mask as well. Please see cpuup_callback |
f0188f47 | 2101 | */ |
83b519e8 PE |
2102 | if (slab_is_available()) { |
2103 | get_online_cpus(); | |
2104 | mutex_lock(&cache_chain_mutex); | |
2105 | } | |
4f12bb4f | 2106 | |
7a7c381d | 2107 | list_for_each_entry(pc, &cache_chain, next) { |
4f12bb4f AM |
2108 | char tmp; |
2109 | int res; | |
2110 | ||
2111 | /* | |
2112 | * This happens when the module gets unloaded and doesn't | |
2113 | * destroy its slab cache and no-one else reuses the vmalloc | |
2114 | * area of the module. Print a warning. | |
2115 | */ | |
138ae663 | 2116 | res = probe_kernel_address(pc->name, tmp); |
4f12bb4f | 2117 | if (res) { |
b4169525 | 2118 | printk(KERN_ERR |
2119 | "SLAB: cache with size %d has lost its name\n", | |
3dafccf2 | 2120 | pc->buffer_size); |
4f12bb4f AM |
2121 | continue; |
2122 | } | |
2123 | ||
b28a02de | 2124 | if (!strcmp(pc->name, name)) { |
b4169525 | 2125 | printk(KERN_ERR |
2126 | "kmem_cache_create: duplicate cache %s\n", name); | |
4f12bb4f AM |
2127 | dump_stack(); |
2128 | goto oops; | |
2129 | } | |
2130 | } | |
2131 | ||
1da177e4 LT |
2132 | #if DEBUG |
2133 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
1da177e4 LT |
2134 | #if FORCED_DEBUG |
2135 | /* | |
2136 | * Enable redzoning and last user accounting, except for caches with | |
2137 | * large objects, if the increased size would increase the object size | |
2138 | * above the next power of two: caches with object sizes just above a | |
2139 | * power of two have a significant amount of internal fragmentation. | |
2140 | */ | |
87a927c7 DW |
2141 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2142 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2143 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2144 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2145 | flags |= SLAB_POISON; | |
2146 | #endif | |
2147 | if (flags & SLAB_DESTROY_BY_RCU) | |
2148 | BUG_ON(flags & SLAB_POISON); | |
2149 | #endif | |
1da177e4 | 2150 | /* |
a737b3e2 AM |
2151 | * Always checks flags, a caller might be expecting debug support which |
2152 | * isn't available. | |
1da177e4 | 2153 | */ |
40094fa6 | 2154 | BUG_ON(flags & ~CREATE_MASK); |
1da177e4 | 2155 | |
a737b3e2 AM |
2156 | /* |
2157 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2158 | * unaligned accesses for some archs when redzoning is used, and makes |
2159 | * sure any on-slab bufctl's are also correctly aligned. | |
2160 | */ | |
b28a02de PE |
2161 | if (size & (BYTES_PER_WORD - 1)) { |
2162 | size += (BYTES_PER_WORD - 1); | |
2163 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2164 | } |
2165 | ||
a737b3e2 AM |
2166 | /* calculate the final buffer alignment: */ |
2167 | ||
1da177e4 LT |
2168 | /* 1) arch recommendation: can be overridden for debug */ |
2169 | if (flags & SLAB_HWCACHE_ALIGN) { | |
a737b3e2 AM |
2170 | /* |
2171 | * Default alignment: as specified by the arch code. Except if | |
2172 | * an object is really small, then squeeze multiple objects into | |
2173 | * one cacheline. | |
1da177e4 LT |
2174 | */ |
2175 | ralign = cache_line_size(); | |
b28a02de | 2176 | while (size <= ralign / 2) |
1da177e4 LT |
2177 | ralign /= 2; |
2178 | } else { | |
2179 | ralign = BYTES_PER_WORD; | |
2180 | } | |
ca5f9703 PE |
2181 | |
2182 | /* | |
87a927c7 DW |
2183 | * Redzoning and user store require word alignment or possibly larger. |
2184 | * Note this will be overridden by architecture or caller mandated | |
2185 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2186 | */ |
87a927c7 DW |
2187 | if (flags & SLAB_STORE_USER) |
2188 | ralign = BYTES_PER_WORD; | |
2189 | ||
2190 | if (flags & SLAB_RED_ZONE) { | |
2191 | ralign = REDZONE_ALIGN; | |
2192 | /* If redzoning, ensure that the second redzone is suitably | |
2193 | * aligned, by adjusting the object size accordingly. */ | |
2194 | size += REDZONE_ALIGN - 1; | |
2195 | size &= ~(REDZONE_ALIGN - 1); | |
2196 | } | |
ca5f9703 | 2197 | |
a44b56d3 | 2198 | /* 2) arch mandated alignment */ |
1da177e4 LT |
2199 | if (ralign < ARCH_SLAB_MINALIGN) { |
2200 | ralign = ARCH_SLAB_MINALIGN; | |
1da177e4 | 2201 | } |
a44b56d3 | 2202 | /* 3) caller mandated alignment */ |
1da177e4 LT |
2203 | if (ralign < align) { |
2204 | ralign = align; | |
1da177e4 | 2205 | } |
a44b56d3 | 2206 | /* disable debug if necessary */ |
b46b8f19 | 2207 | if (ralign > __alignof__(unsigned long long)) |
a44b56d3 | 2208 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2209 | /* |
ca5f9703 | 2210 | * 4) Store it. |
1da177e4 LT |
2211 | */ |
2212 | align = ralign; | |
2213 | ||
83b519e8 PE |
2214 | if (slab_is_available()) |
2215 | gfp = GFP_KERNEL; | |
2216 | else | |
2217 | gfp = GFP_NOWAIT; | |
2218 | ||
1da177e4 | 2219 | /* Get cache's description obj. */ |
83b519e8 | 2220 | cachep = kmem_cache_zalloc(&cache_cache, gfp); |
1da177e4 | 2221 | if (!cachep) |
4f12bb4f | 2222 | goto oops; |
1da177e4 LT |
2223 | |
2224 | #if DEBUG | |
3dafccf2 | 2225 | cachep->obj_size = size; |
1da177e4 | 2226 | |
ca5f9703 PE |
2227 | /* |
2228 | * Both debugging options require word-alignment which is calculated | |
2229 | * into align above. | |
2230 | */ | |
1da177e4 | 2231 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2232 | /* add space for red zone words */ |
b46b8f19 DW |
2233 | cachep->obj_offset += sizeof(unsigned long long); |
2234 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2235 | } |
2236 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2237 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2238 | * the real object. But if the second red zone needs to be |
2239 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2240 | */ |
87a927c7 DW |
2241 | if (flags & SLAB_RED_ZONE) |
2242 | size += REDZONE_ALIGN; | |
2243 | else | |
2244 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2245 | } |
2246 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
b28a02de | 2247 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
3dafccf2 MS |
2248 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
2249 | cachep->obj_offset += PAGE_SIZE - size; | |
1da177e4 LT |
2250 | size = PAGE_SIZE; |
2251 | } | |
2252 | #endif | |
2253 | #endif | |
2254 | ||
e0a42726 IM |
2255 | /* |
2256 | * Determine if the slab management is 'on' or 'off' slab. | |
2257 | * (bootstrapping cannot cope with offslab caches so don't do | |
2258 | * it too early on.) | |
2259 | */ | |
2260 | if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init) | |
1da177e4 LT |
2261 | /* |
2262 | * Size is large, assume best to place the slab management obj | |
2263 | * off-slab (should allow better packing of objs). | |
2264 | */ | |
2265 | flags |= CFLGS_OFF_SLAB; | |
2266 | ||
2267 | size = ALIGN(size, align); | |
2268 | ||
f78bb8ad | 2269 | left_over = calculate_slab_order(cachep, size, align, flags); |
1da177e4 LT |
2270 | |
2271 | if (!cachep->num) { | |
b4169525 | 2272 | printk(KERN_ERR |
2273 | "kmem_cache_create: couldn't create cache %s.\n", name); | |
1da177e4 LT |
2274 | kmem_cache_free(&cache_cache, cachep); |
2275 | cachep = NULL; | |
4f12bb4f | 2276 | goto oops; |
1da177e4 | 2277 | } |
b28a02de PE |
2278 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
2279 | + sizeof(struct slab), align); | |
1da177e4 LT |
2280 | |
2281 | /* | |
2282 | * If the slab has been placed off-slab, and we have enough space then | |
2283 | * move it on-slab. This is at the expense of any extra colouring. | |
2284 | */ | |
2285 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { | |
2286 | flags &= ~CFLGS_OFF_SLAB; | |
2287 | left_over -= slab_size; | |
2288 | } | |
2289 | ||
2290 | if (flags & CFLGS_OFF_SLAB) { | |
2291 | /* really off slab. No need for manual alignment */ | |
b28a02de PE |
2292 | slab_size = |
2293 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); | |
1da177e4 LT |
2294 | } |
2295 | ||
2296 | cachep->colour_off = cache_line_size(); | |
2297 | /* Offset must be a multiple of the alignment. */ | |
2298 | if (cachep->colour_off < align) | |
2299 | cachep->colour_off = align; | |
b28a02de | 2300 | cachep->colour = left_over / cachep->colour_off; |
1da177e4 LT |
2301 | cachep->slab_size = slab_size; |
2302 | cachep->flags = flags; | |
2303 | cachep->gfpflags = 0; | |
4b51d669 | 2304 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
1da177e4 | 2305 | cachep->gfpflags |= GFP_DMA; |
3dafccf2 | 2306 | cachep->buffer_size = size; |
6a2d7a95 | 2307 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2308 | |
e5ac9c5a | 2309 | if (flags & CFLGS_OFF_SLAB) { |
b2d55073 | 2310 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
e5ac9c5a RT |
2311 | /* |
2312 | * This is a possibility for one of the malloc_sizes caches. | |
2313 | * But since we go off slab only for object size greater than | |
2314 | * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, | |
2315 | * this should not happen at all. | |
2316 | * But leave a BUG_ON for some lucky dude. | |
2317 | */ | |
6cb8f913 | 2318 | BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache)); |
e5ac9c5a | 2319 | } |
1da177e4 | 2320 | cachep->ctor = ctor; |
1da177e4 LT |
2321 | cachep->name = name; |
2322 | ||
83b519e8 | 2323 | if (setup_cpu_cache(cachep, gfp)) { |
2ed3a4ef CL |
2324 | __kmem_cache_destroy(cachep); |
2325 | cachep = NULL; | |
2326 | goto oops; | |
2327 | } | |
1da177e4 | 2328 | |
1da177e4 LT |
2329 | /* cache setup completed, link it into the list */ |
2330 | list_add(&cachep->next, &cache_chain); | |
a737b3e2 | 2331 | oops: |
1da177e4 LT |
2332 | if (!cachep && (flags & SLAB_PANIC)) |
2333 | panic("kmem_cache_create(): failed to create slab `%s'\n", | |
b28a02de | 2334 | name); |
83b519e8 PE |
2335 | if (slab_is_available()) { |
2336 | mutex_unlock(&cache_chain_mutex); | |
2337 | put_online_cpus(); | |
2338 | } | |
1da177e4 LT |
2339 | return cachep; |
2340 | } | |
2341 | EXPORT_SYMBOL(kmem_cache_create); | |
2342 | ||
2343 | #if DEBUG | |
2344 | static void check_irq_off(void) | |
2345 | { | |
2346 | BUG_ON(!irqs_disabled()); | |
2347 | } | |
2348 | ||
2349 | static void check_irq_on(void) | |
2350 | { | |
2351 | BUG_ON(irqs_disabled()); | |
2352 | } | |
2353 | ||
343e0d7a | 2354 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2355 | { |
2356 | #ifdef CONFIG_SMP | |
2357 | check_irq_off(); | |
e498be7d | 2358 | assert_spin_locked(&cachep->nodelists[numa_node_id()]->list_lock); |
1da177e4 LT |
2359 | #endif |
2360 | } | |
e498be7d | 2361 | |
343e0d7a | 2362 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2363 | { |
2364 | #ifdef CONFIG_SMP | |
2365 | check_irq_off(); | |
2366 | assert_spin_locked(&cachep->nodelists[node]->list_lock); | |
2367 | #endif | |
2368 | } | |
2369 | ||
1da177e4 LT |
2370 | #else |
2371 | #define check_irq_off() do { } while(0) | |
2372 | #define check_irq_on() do { } while(0) | |
2373 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2374 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2375 | #endif |
2376 | ||
aab2207c CL |
2377 | static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, |
2378 | struct array_cache *ac, | |
2379 | int force, int node); | |
2380 | ||
1da177e4 LT |
2381 | static void do_drain(void *arg) |
2382 | { | |
a737b3e2 | 2383 | struct kmem_cache *cachep = arg; |
1da177e4 | 2384 | struct array_cache *ac; |
ff69416e | 2385 | int node = numa_node_id(); |
1da177e4 LT |
2386 | |
2387 | check_irq_off(); | |
9a2dba4b | 2388 | ac = cpu_cache_get(cachep); |
ff69416e CL |
2389 | spin_lock(&cachep->nodelists[node]->list_lock); |
2390 | free_block(cachep, ac->entry, ac->avail, node); | |
2391 | spin_unlock(&cachep->nodelists[node]->list_lock); | |
1da177e4 LT |
2392 | ac->avail = 0; |
2393 | } | |
2394 | ||
343e0d7a | 2395 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2396 | { |
e498be7d CL |
2397 | struct kmem_list3 *l3; |
2398 | int node; | |
2399 | ||
15c8b6c1 | 2400 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2401 | check_irq_on(); |
b28a02de | 2402 | for_each_online_node(node) { |
e498be7d | 2403 | l3 = cachep->nodelists[node]; |
a4523a8b RD |
2404 | if (l3 && l3->alien) |
2405 | drain_alien_cache(cachep, l3->alien); | |
2406 | } | |
2407 | ||
2408 | for_each_online_node(node) { | |
2409 | l3 = cachep->nodelists[node]; | |
2410 | if (l3) | |
aab2207c | 2411 | drain_array(cachep, l3, l3->shared, 1, node); |
e498be7d | 2412 | } |
1da177e4 LT |
2413 | } |
2414 | ||
ed11d9eb CL |
2415 | /* |
2416 | * Remove slabs from the list of free slabs. | |
2417 | * Specify the number of slabs to drain in tofree. | |
2418 | * | |
2419 | * Returns the actual number of slabs released. | |
2420 | */ | |
2421 | static int drain_freelist(struct kmem_cache *cache, | |
2422 | struct kmem_list3 *l3, int tofree) | |
1da177e4 | 2423 | { |
ed11d9eb CL |
2424 | struct list_head *p; |
2425 | int nr_freed; | |
1da177e4 | 2426 | struct slab *slabp; |
1da177e4 | 2427 | |
ed11d9eb CL |
2428 | nr_freed = 0; |
2429 | while (nr_freed < tofree && !list_empty(&l3->slabs_free)) { | |
1da177e4 | 2430 | |
ed11d9eb | 2431 | spin_lock_irq(&l3->list_lock); |
e498be7d | 2432 | p = l3->slabs_free.prev; |
ed11d9eb CL |
2433 | if (p == &l3->slabs_free) { |
2434 | spin_unlock_irq(&l3->list_lock); | |
2435 | goto out; | |
2436 | } | |
1da177e4 | 2437 | |
ed11d9eb | 2438 | slabp = list_entry(p, struct slab, list); |
1da177e4 | 2439 | #if DEBUG |
40094fa6 | 2440 | BUG_ON(slabp->inuse); |
1da177e4 LT |
2441 | #endif |
2442 | list_del(&slabp->list); | |
ed11d9eb CL |
2443 | /* |
2444 | * Safe to drop the lock. The slab is no longer linked | |
2445 | * to the cache. | |
2446 | */ | |
2447 | l3->free_objects -= cache->num; | |
e498be7d | 2448 | spin_unlock_irq(&l3->list_lock); |
ed11d9eb CL |
2449 | slab_destroy(cache, slabp); |
2450 | nr_freed++; | |
1da177e4 | 2451 | } |
ed11d9eb CL |
2452 | out: |
2453 | return nr_freed; | |
1da177e4 LT |
2454 | } |
2455 | ||
8f5be20b | 2456 | /* Called with cache_chain_mutex held to protect against cpu hotplug */ |
343e0d7a | 2457 | static int __cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2458 | { |
2459 | int ret = 0, i = 0; | |
2460 | struct kmem_list3 *l3; | |
2461 | ||
2462 | drain_cpu_caches(cachep); | |
2463 | ||
2464 | check_irq_on(); | |
2465 | for_each_online_node(i) { | |
2466 | l3 = cachep->nodelists[i]; | |
ed11d9eb CL |
2467 | if (!l3) |
2468 | continue; | |
2469 | ||
2470 | drain_freelist(cachep, l3, l3->free_objects); | |
2471 | ||
2472 | ret += !list_empty(&l3->slabs_full) || | |
2473 | !list_empty(&l3->slabs_partial); | |
e498be7d CL |
2474 | } |
2475 | return (ret ? 1 : 0); | |
2476 | } | |
2477 | ||
1da177e4 LT |
2478 | /** |
2479 | * kmem_cache_shrink - Shrink a cache. | |
2480 | * @cachep: The cache to shrink. | |
2481 | * | |
2482 | * Releases as many slabs as possible for a cache. | |
2483 | * To help debugging, a zero exit status indicates all slabs were released. | |
2484 | */ | |
343e0d7a | 2485 | int kmem_cache_shrink(struct kmem_cache *cachep) |
1da177e4 | 2486 | { |
8f5be20b | 2487 | int ret; |
40094fa6 | 2488 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2489 | |
95402b38 | 2490 | get_online_cpus(); |
8f5be20b RT |
2491 | mutex_lock(&cache_chain_mutex); |
2492 | ret = __cache_shrink(cachep); | |
2493 | mutex_unlock(&cache_chain_mutex); | |
95402b38 | 2494 | put_online_cpus(); |
8f5be20b | 2495 | return ret; |
1da177e4 LT |
2496 | } |
2497 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2498 | ||
2499 | /** | |
2500 | * kmem_cache_destroy - delete a cache | |
2501 | * @cachep: the cache to destroy | |
2502 | * | |
72fd4a35 | 2503 | * Remove a &struct kmem_cache object from the slab cache. |
1da177e4 LT |
2504 | * |
2505 | * It is expected this function will be called by a module when it is | |
2506 | * unloaded. This will remove the cache completely, and avoid a duplicate | |
2507 | * cache being allocated each time a module is loaded and unloaded, if the | |
2508 | * module doesn't have persistent in-kernel storage across loads and unloads. | |
2509 | * | |
2510 | * The cache must be empty before calling this function. | |
2511 | * | |
2512 | * The caller must guarantee that noone will allocate memory from the cache | |
2513 | * during the kmem_cache_destroy(). | |
2514 | */ | |
133d205a | 2515 | void kmem_cache_destroy(struct kmem_cache *cachep) |
1da177e4 | 2516 | { |
40094fa6 | 2517 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2518 | |
1da177e4 | 2519 | /* Find the cache in the chain of caches. */ |
95402b38 | 2520 | get_online_cpus(); |
fc0abb14 | 2521 | mutex_lock(&cache_chain_mutex); |
1da177e4 LT |
2522 | /* |
2523 | * the chain is never empty, cache_cache is never destroyed | |
2524 | */ | |
2525 | list_del(&cachep->next); | |
1da177e4 LT |
2526 | if (__cache_shrink(cachep)) { |
2527 | slab_error(cachep, "Can't free all objects"); | |
b28a02de | 2528 | list_add(&cachep->next, &cache_chain); |
fc0abb14 | 2529 | mutex_unlock(&cache_chain_mutex); |
95402b38 | 2530 | put_online_cpus(); |
133d205a | 2531 | return; |
1da177e4 LT |
2532 | } |
2533 | ||
2534 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
fbd568a3 | 2535 | synchronize_rcu(); |
1da177e4 | 2536 | |
117f6eb1 | 2537 | __kmem_cache_destroy(cachep); |
8f5be20b | 2538 | mutex_unlock(&cache_chain_mutex); |
95402b38 | 2539 | put_online_cpus(); |
1da177e4 LT |
2540 | } |
2541 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2542 | ||
e5ac9c5a RT |
2543 | /* |
2544 | * Get the memory for a slab management obj. | |
2545 | * For a slab cache when the slab descriptor is off-slab, slab descriptors | |
2546 | * always come from malloc_sizes caches. The slab descriptor cannot | |
2547 | * come from the same cache which is getting created because, | |
2548 | * when we are searching for an appropriate cache for these | |
2549 | * descriptors in kmem_cache_create, we search through the malloc_sizes array. | |
2550 | * If we are creating a malloc_sizes cache here it would not be visible to | |
2551 | * kmem_find_general_cachep till the initialization is complete. | |
2552 | * Hence we cannot have slabp_cache same as the original cache. | |
2553 | */ | |
343e0d7a | 2554 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
5b74ada7 RT |
2555 | int colour_off, gfp_t local_flags, |
2556 | int nodeid) | |
1da177e4 LT |
2557 | { |
2558 | struct slab *slabp; | |
b28a02de | 2559 | |
1da177e4 LT |
2560 | if (OFF_SLAB(cachep)) { |
2561 | /* Slab management obj is off-slab. */ | |
5b74ada7 | 2562 | slabp = kmem_cache_alloc_node(cachep->slabp_cache, |
8759ec50 | 2563 | local_flags, nodeid); |
d5cff635 CM |
2564 | /* |
2565 | * If the first object in the slab is leaked (it's allocated | |
2566 | * but no one has a reference to it), we want to make sure | |
2567 | * kmemleak does not treat the ->s_mem pointer as a reference | |
2568 | * to the object. Otherwise we will not report the leak. | |
2569 | */ | |
2570 | kmemleak_scan_area(slabp, offsetof(struct slab, list), | |
2571 | sizeof(struct list_head), local_flags); | |
1da177e4 LT |
2572 | if (!slabp) |
2573 | return NULL; | |
2574 | } else { | |
b28a02de | 2575 | slabp = objp + colour_off; |
1da177e4 LT |
2576 | colour_off += cachep->slab_size; |
2577 | } | |
2578 | slabp->inuse = 0; | |
2579 | slabp->colouroff = colour_off; | |
b28a02de | 2580 | slabp->s_mem = objp + colour_off; |
5b74ada7 | 2581 | slabp->nodeid = nodeid; |
e51bfd0a | 2582 | slabp->free = 0; |
1da177e4 LT |
2583 | return slabp; |
2584 | } | |
2585 | ||
2586 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) | |
2587 | { | |
b28a02de | 2588 | return (kmem_bufctl_t *) (slabp + 1); |
1da177e4 LT |
2589 | } |
2590 | ||
343e0d7a | 2591 | static void cache_init_objs(struct kmem_cache *cachep, |
a35afb83 | 2592 | struct slab *slabp) |
1da177e4 LT |
2593 | { |
2594 | int i; | |
2595 | ||
2596 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2597 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2598 | #if DEBUG |
2599 | /* need to poison the objs? */ | |
2600 | if (cachep->flags & SLAB_POISON) | |
2601 | poison_obj(cachep, objp, POISON_FREE); | |
2602 | if (cachep->flags & SLAB_STORE_USER) | |
2603 | *dbg_userword(cachep, objp) = NULL; | |
2604 | ||
2605 | if (cachep->flags & SLAB_RED_ZONE) { | |
2606 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2607 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2608 | } | |
2609 | /* | |
a737b3e2 AM |
2610 | * Constructors are not allowed to allocate memory from the same |
2611 | * cache which they are a constructor for. Otherwise, deadlock. | |
2612 | * They must also be threaded. | |
1da177e4 LT |
2613 | */ |
2614 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2615 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2616 | |
2617 | if (cachep->flags & SLAB_RED_ZONE) { | |
2618 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2619 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2620 | " end of an object"); |
1da177e4 LT |
2621 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2622 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2623 | " start of an object"); |
1da177e4 | 2624 | } |
a737b3e2 AM |
2625 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && |
2626 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) | |
b28a02de | 2627 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2628 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2629 | #else |
2630 | if (cachep->ctor) | |
51cc5068 | 2631 | cachep->ctor(objp); |
1da177e4 | 2632 | #endif |
b28a02de | 2633 | slab_bufctl(slabp)[i] = i + 1; |
1da177e4 | 2634 | } |
b28a02de | 2635 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
1da177e4 LT |
2636 | } |
2637 | ||
343e0d7a | 2638 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2639 | { |
4b51d669 CL |
2640 | if (CONFIG_ZONE_DMA_FLAG) { |
2641 | if (flags & GFP_DMA) | |
2642 | BUG_ON(!(cachep->gfpflags & GFP_DMA)); | |
2643 | else | |
2644 | BUG_ON(cachep->gfpflags & GFP_DMA); | |
2645 | } | |
1da177e4 LT |
2646 | } |
2647 | ||
a737b3e2 AM |
2648 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
2649 | int nodeid) | |
78d382d7 | 2650 | { |
8fea4e96 | 2651 | void *objp = index_to_obj(cachep, slabp, slabp->free); |
78d382d7 MD |
2652 | kmem_bufctl_t next; |
2653 | ||
2654 | slabp->inuse++; | |
2655 | next = slab_bufctl(slabp)[slabp->free]; | |
2656 | #if DEBUG | |
2657 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | |
2658 | WARN_ON(slabp->nodeid != nodeid); | |
2659 | #endif | |
2660 | slabp->free = next; | |
2661 | ||
2662 | return objp; | |
2663 | } | |
2664 | ||
a737b3e2 AM |
2665 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
2666 | void *objp, int nodeid) | |
78d382d7 | 2667 | { |
8fea4e96 | 2668 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
78d382d7 MD |
2669 | |
2670 | #if DEBUG | |
2671 | /* Verify that the slab belongs to the intended node */ | |
2672 | WARN_ON(slabp->nodeid != nodeid); | |
2673 | ||
871751e2 | 2674 | if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) { |
78d382d7 | 2675 | printk(KERN_ERR "slab: double free detected in cache " |
a737b3e2 | 2676 | "'%s', objp %p\n", cachep->name, objp); |
78d382d7 MD |
2677 | BUG(); |
2678 | } | |
2679 | #endif | |
2680 | slab_bufctl(slabp)[objnr] = slabp->free; | |
2681 | slabp->free = objnr; | |
2682 | slabp->inuse--; | |
2683 | } | |
2684 | ||
4776874f PE |
2685 | /* |
2686 | * Map pages beginning at addr to the given cache and slab. This is required | |
2687 | * for the slab allocator to be able to lookup the cache and slab of a | |
2688 | * virtual address for kfree, ksize, kmem_ptr_validate, and slab debugging. | |
2689 | */ | |
2690 | static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, | |
2691 | void *addr) | |
1da177e4 | 2692 | { |
4776874f | 2693 | int nr_pages; |
1da177e4 LT |
2694 | struct page *page; |
2695 | ||
4776874f | 2696 | page = virt_to_page(addr); |
84097518 | 2697 | |
4776874f | 2698 | nr_pages = 1; |
84097518 | 2699 | if (likely(!PageCompound(page))) |
4776874f PE |
2700 | nr_pages <<= cache->gfporder; |
2701 | ||
1da177e4 | 2702 | do { |
4776874f PE |
2703 | page_set_cache(page, cache); |
2704 | page_set_slab(page, slab); | |
1da177e4 | 2705 | page++; |
4776874f | 2706 | } while (--nr_pages); |
1da177e4 LT |
2707 | } |
2708 | ||
2709 | /* | |
2710 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2711 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2712 | */ | |
3c517a61 CL |
2713 | static int cache_grow(struct kmem_cache *cachep, |
2714 | gfp_t flags, int nodeid, void *objp) | |
1da177e4 | 2715 | { |
b28a02de | 2716 | struct slab *slabp; |
b28a02de PE |
2717 | size_t offset; |
2718 | gfp_t local_flags; | |
e498be7d | 2719 | struct kmem_list3 *l3; |
1da177e4 | 2720 | |
a737b3e2 AM |
2721 | /* |
2722 | * Be lazy and only check for valid flags here, keeping it out of the | |
2723 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2724 | */ |
6cb06229 CL |
2725 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2726 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2727 | |
2e1217cf | 2728 | /* Take the l3 list lock to change the colour_next on this node */ |
1da177e4 | 2729 | check_irq_off(); |
2e1217cf RT |
2730 | l3 = cachep->nodelists[nodeid]; |
2731 | spin_lock(&l3->list_lock); | |
1da177e4 LT |
2732 | |
2733 | /* Get colour for the slab, and cal the next value. */ | |
2e1217cf RT |
2734 | offset = l3->colour_next; |
2735 | l3->colour_next++; | |
2736 | if (l3->colour_next >= cachep->colour) | |
2737 | l3->colour_next = 0; | |
2738 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2739 | |
2e1217cf | 2740 | offset *= cachep->colour_off; |
1da177e4 LT |
2741 | |
2742 | if (local_flags & __GFP_WAIT) | |
2743 | local_irq_enable(); | |
2744 | ||
2745 | /* | |
2746 | * The test for missing atomic flag is performed here, rather than | |
2747 | * the more obvious place, simply to reduce the critical path length | |
2748 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2749 | * will eventually be caught here (where it matters). | |
2750 | */ | |
2751 | kmem_flagcheck(cachep, flags); | |
2752 | ||
a737b3e2 AM |
2753 | /* |
2754 | * Get mem for the objs. Attempt to allocate a physical page from | |
2755 | * 'nodeid'. | |
e498be7d | 2756 | */ |
3c517a61 | 2757 | if (!objp) |
b8c1c5da | 2758 | objp = kmem_getpages(cachep, local_flags, nodeid); |
a737b3e2 | 2759 | if (!objp) |
1da177e4 LT |
2760 | goto failed; |
2761 | ||
2762 | /* Get slab management. */ | |
3c517a61 | 2763 | slabp = alloc_slabmgmt(cachep, objp, offset, |
6cb06229 | 2764 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
a737b3e2 | 2765 | if (!slabp) |
1da177e4 LT |
2766 | goto opps1; |
2767 | ||
4776874f | 2768 | slab_map_pages(cachep, slabp, objp); |
1da177e4 | 2769 | |
a35afb83 | 2770 | cache_init_objs(cachep, slabp); |
1da177e4 LT |
2771 | |
2772 | if (local_flags & __GFP_WAIT) | |
2773 | local_irq_disable(); | |
2774 | check_irq_off(); | |
e498be7d | 2775 | spin_lock(&l3->list_lock); |
1da177e4 LT |
2776 | |
2777 | /* Make slab active. */ | |
e498be7d | 2778 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
1da177e4 | 2779 | STATS_INC_GROWN(cachep); |
e498be7d CL |
2780 | l3->free_objects += cachep->num; |
2781 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2782 | return 1; |
a737b3e2 | 2783 | opps1: |
1da177e4 | 2784 | kmem_freepages(cachep, objp); |
a737b3e2 | 2785 | failed: |
1da177e4 LT |
2786 | if (local_flags & __GFP_WAIT) |
2787 | local_irq_disable(); | |
2788 | return 0; | |
2789 | } | |
2790 | ||
2791 | #if DEBUG | |
2792 | ||
2793 | /* | |
2794 | * Perform extra freeing checks: | |
2795 | * - detect bad pointers. | |
2796 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2797 | */ |
2798 | static void kfree_debugcheck(const void *objp) | |
2799 | { | |
1da177e4 LT |
2800 | if (!virt_addr_valid(objp)) { |
2801 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2802 | (unsigned long)objp); |
2803 | BUG(); | |
1da177e4 | 2804 | } |
1da177e4 LT |
2805 | } |
2806 | ||
58ce1fd5 PE |
2807 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2808 | { | |
b46b8f19 | 2809 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2810 | |
2811 | redzone1 = *dbg_redzone1(cache, obj); | |
2812 | redzone2 = *dbg_redzone2(cache, obj); | |
2813 | ||
2814 | /* | |
2815 | * Redzone is ok. | |
2816 | */ | |
2817 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2818 | return; | |
2819 | ||
2820 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2821 | slab_error(cache, "double free detected"); | |
2822 | else | |
2823 | slab_error(cache, "memory outside object was overwritten"); | |
2824 | ||
b46b8f19 | 2825 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2826 | obj, redzone1, redzone2); |
2827 | } | |
2828 | ||
343e0d7a | 2829 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
b28a02de | 2830 | void *caller) |
1da177e4 LT |
2831 | { |
2832 | struct page *page; | |
2833 | unsigned int objnr; | |
2834 | struct slab *slabp; | |
2835 | ||
80cbd911 MW |
2836 | BUG_ON(virt_to_cache(objp) != cachep); |
2837 | ||
3dafccf2 | 2838 | objp -= obj_offset(cachep); |
1da177e4 | 2839 | kfree_debugcheck(objp); |
b49af68f | 2840 | page = virt_to_head_page(objp); |
1da177e4 | 2841 | |
065d41cb | 2842 | slabp = page_get_slab(page); |
1da177e4 LT |
2843 | |
2844 | if (cachep->flags & SLAB_RED_ZONE) { | |
58ce1fd5 | 2845 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2846 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2847 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2848 | } | |
2849 | if (cachep->flags & SLAB_STORE_USER) | |
2850 | *dbg_userword(cachep, objp) = caller; | |
2851 | ||
8fea4e96 | 2852 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 LT |
2853 | |
2854 | BUG_ON(objnr >= cachep->num); | |
8fea4e96 | 2855 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
1da177e4 | 2856 | |
871751e2 AV |
2857 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
2858 | slab_bufctl(slabp)[objnr] = BUFCTL_FREE; | |
2859 | #endif | |
1da177e4 LT |
2860 | if (cachep->flags & SLAB_POISON) { |
2861 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 | 2862 | if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
1da177e4 | 2863 | store_stackinfo(cachep, objp, (unsigned long)caller); |
b28a02de | 2864 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2865 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2866 | } else { |
2867 | poison_obj(cachep, objp, POISON_FREE); | |
2868 | } | |
2869 | #else | |
2870 | poison_obj(cachep, objp, POISON_FREE); | |
2871 | #endif | |
2872 | } | |
2873 | return objp; | |
2874 | } | |
2875 | ||
343e0d7a | 2876 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 LT |
2877 | { |
2878 | kmem_bufctl_t i; | |
2879 | int entries = 0; | |
b28a02de | 2880 | |
1da177e4 LT |
2881 | /* Check slab's freelist to see if this obj is there. */ |
2882 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { | |
2883 | entries++; | |
2884 | if (entries > cachep->num || i >= cachep->num) | |
2885 | goto bad; | |
2886 | } | |
2887 | if (entries != cachep->num - slabp->inuse) { | |
a737b3e2 AM |
2888 | bad: |
2889 | printk(KERN_ERR "slab: Internal list corruption detected in " | |
2890 | "cache '%s'(%d), slabp %p(%d). Hexdump:\n", | |
2891 | cachep->name, cachep->num, slabp, slabp->inuse); | |
b28a02de | 2892 | for (i = 0; |
264132bc | 2893 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
b28a02de | 2894 | i++) { |
a737b3e2 | 2895 | if (i % 16 == 0) |
1da177e4 | 2896 | printk("\n%03x:", i); |
b28a02de | 2897 | printk(" %02x", ((unsigned char *)slabp)[i]); |
1da177e4 LT |
2898 | } |
2899 | printk("\n"); | |
2900 | BUG(); | |
2901 | } | |
2902 | } | |
2903 | #else | |
2904 | #define kfree_debugcheck(x) do { } while(0) | |
2905 | #define cache_free_debugcheck(x,objp,z) (objp) | |
2906 | #define check_slabp(x,y) do { } while(0) | |
2907 | #endif | |
2908 | ||
343e0d7a | 2909 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2910 | { |
2911 | int batchcount; | |
2912 | struct kmem_list3 *l3; | |
2913 | struct array_cache *ac; | |
1ca4cb24 PE |
2914 | int node; |
2915 | ||
6d2144d3 | 2916 | retry: |
1da177e4 | 2917 | check_irq_off(); |
6d2144d3 | 2918 | node = numa_node_id(); |
9a2dba4b | 2919 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2920 | batchcount = ac->batchcount; |
2921 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2922 | /* |
2923 | * If there was little recent activity on this cache, then | |
2924 | * perform only a partial refill. Otherwise we could generate | |
2925 | * refill bouncing. | |
1da177e4 LT |
2926 | */ |
2927 | batchcount = BATCHREFILL_LIMIT; | |
2928 | } | |
1ca4cb24 | 2929 | l3 = cachep->nodelists[node]; |
e498be7d CL |
2930 | |
2931 | BUG_ON(ac->avail > 0 || !l3); | |
2932 | spin_lock(&l3->list_lock); | |
1da177e4 | 2933 | |
3ded175a CL |
2934 | /* See if we can refill from the shared array */ |
2935 | if (l3->shared && transfer_objects(ac, l3->shared, batchcount)) | |
2936 | goto alloc_done; | |
2937 | ||
1da177e4 LT |
2938 | while (batchcount > 0) { |
2939 | struct list_head *entry; | |
2940 | struct slab *slabp; | |
2941 | /* Get slab alloc is to come from. */ | |
2942 | entry = l3->slabs_partial.next; | |
2943 | if (entry == &l3->slabs_partial) { | |
2944 | l3->free_touched = 1; | |
2945 | entry = l3->slabs_free.next; | |
2946 | if (entry == &l3->slabs_free) | |
2947 | goto must_grow; | |
2948 | } | |
2949 | ||
2950 | slabp = list_entry(entry, struct slab, list); | |
2951 | check_slabp(cachep, slabp); | |
2952 | check_spinlock_acquired(cachep); | |
714b8171 PE |
2953 | |
2954 | /* | |
2955 | * The slab was either on partial or free list so | |
2956 | * there must be at least one object available for | |
2957 | * allocation. | |
2958 | */ | |
249b9f33 | 2959 | BUG_ON(slabp->inuse >= cachep->num); |
714b8171 | 2960 | |
1da177e4 | 2961 | while (slabp->inuse < cachep->num && batchcount--) { |
1da177e4 LT |
2962 | STATS_INC_ALLOCED(cachep); |
2963 | STATS_INC_ACTIVE(cachep); | |
2964 | STATS_SET_HIGH(cachep); | |
2965 | ||
78d382d7 | 2966 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
1ca4cb24 | 2967 | node); |
1da177e4 LT |
2968 | } |
2969 | check_slabp(cachep, slabp); | |
2970 | ||
2971 | /* move slabp to correct slabp list: */ | |
2972 | list_del(&slabp->list); | |
2973 | if (slabp->free == BUFCTL_END) | |
2974 | list_add(&slabp->list, &l3->slabs_full); | |
2975 | else | |
2976 | list_add(&slabp->list, &l3->slabs_partial); | |
2977 | } | |
2978 | ||
a737b3e2 | 2979 | must_grow: |
1da177e4 | 2980 | l3->free_objects -= ac->avail; |
a737b3e2 | 2981 | alloc_done: |
e498be7d | 2982 | spin_unlock(&l3->list_lock); |
1da177e4 LT |
2983 | |
2984 | if (unlikely(!ac->avail)) { | |
2985 | int x; | |
3c517a61 | 2986 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 2987 | |
a737b3e2 | 2988 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 2989 | ac = cpu_cache_get(cachep); |
a737b3e2 | 2990 | if (!x && ac->avail == 0) /* no objects in sight? abort */ |
1da177e4 LT |
2991 | return NULL; |
2992 | ||
a737b3e2 | 2993 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
2994 | goto retry; |
2995 | } | |
2996 | ac->touched = 1; | |
e498be7d | 2997 | return ac->entry[--ac->avail]; |
1da177e4 LT |
2998 | } |
2999 | ||
a737b3e2 AM |
3000 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3001 | gfp_t flags) | |
1da177e4 LT |
3002 | { |
3003 | might_sleep_if(flags & __GFP_WAIT); | |
3004 | #if DEBUG | |
3005 | kmem_flagcheck(cachep, flags); | |
3006 | #endif | |
3007 | } | |
3008 | ||
3009 | #if DEBUG | |
a737b3e2 AM |
3010 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
3011 | gfp_t flags, void *objp, void *caller) | |
1da177e4 | 3012 | { |
b28a02de | 3013 | if (!objp) |
1da177e4 | 3014 | return objp; |
b28a02de | 3015 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3016 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3dafccf2 | 3017 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 3018 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 3019 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
3020 | else |
3021 | check_poison_obj(cachep, objp); | |
3022 | #else | |
3023 | check_poison_obj(cachep, objp); | |
3024 | #endif | |
3025 | poison_obj(cachep, objp, POISON_INUSE); | |
3026 | } | |
3027 | if (cachep->flags & SLAB_STORE_USER) | |
3028 | *dbg_userword(cachep, objp) = caller; | |
3029 | ||
3030 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3031 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3032 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
3033 | slab_error(cachep, "double free, or memory outside" | |
3034 | " object was overwritten"); | |
b28a02de | 3035 | printk(KERN_ERR |
b46b8f19 | 3036 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
3037 | objp, *dbg_redzone1(cachep, objp), |
3038 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3039 | } |
3040 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3041 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3042 | } | |
871751e2 AV |
3043 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
3044 | { | |
3045 | struct slab *slabp; | |
3046 | unsigned objnr; | |
3047 | ||
b49af68f | 3048 | slabp = page_get_slab(virt_to_head_page(objp)); |
871751e2 AV |
3049 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
3050 | slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE; | |
3051 | } | |
3052 | #endif | |
3dafccf2 | 3053 | objp += obj_offset(cachep); |
4f104934 | 3054 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3055 | cachep->ctor(objp); |
a44b56d3 KH |
3056 | #if ARCH_SLAB_MINALIGN |
3057 | if ((u32)objp & (ARCH_SLAB_MINALIGN-1)) { | |
3058 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", | |
3059 | objp, ARCH_SLAB_MINALIGN); | |
3060 | } | |
3061 | #endif | |
1da177e4 LT |
3062 | return objp; |
3063 | } | |
3064 | #else | |
3065 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3066 | #endif | |
3067 | ||
773ff60e | 3068 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 AM |
3069 | { |
3070 | if (cachep == &cache_cache) | |
773ff60e | 3071 | return false; |
8a8b6502 | 3072 | |
773ff60e | 3073 | return should_failslab(obj_size(cachep), flags); |
8a8b6502 AM |
3074 | } |
3075 | ||
343e0d7a | 3076 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3077 | { |
b28a02de | 3078 | void *objp; |
1da177e4 LT |
3079 | struct array_cache *ac; |
3080 | ||
5c382300 | 3081 | check_irq_off(); |
8a8b6502 | 3082 | |
9a2dba4b | 3083 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
3084 | if (likely(ac->avail)) { |
3085 | STATS_INC_ALLOCHIT(cachep); | |
3086 | ac->touched = 1; | |
e498be7d | 3087 | objp = ac->entry[--ac->avail]; |
1da177e4 LT |
3088 | } else { |
3089 | STATS_INC_ALLOCMISS(cachep); | |
3090 | objp = cache_alloc_refill(cachep, flags); | |
3091 | } | |
d5cff635 CM |
3092 | /* |
3093 | * To avoid a false negative, if an object that is in one of the | |
3094 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3095 | * treat the array pointers as a reference to the object. | |
3096 | */ | |
3097 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3098 | return objp; |
3099 | } | |
3100 | ||
e498be7d | 3101 | #ifdef CONFIG_NUMA |
c61afb18 | 3102 | /* |
b2455396 | 3103 | * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. |
c61afb18 PJ |
3104 | * |
3105 | * If we are in_interrupt, then process context, including cpusets and | |
3106 | * mempolicy, may not apply and should not be used for allocation policy. | |
3107 | */ | |
3108 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3109 | { | |
3110 | int nid_alloc, nid_here; | |
3111 | ||
765c4507 | 3112 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 PJ |
3113 | return NULL; |
3114 | nid_alloc = nid_here = numa_node_id(); | |
3115 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) | |
3116 | nid_alloc = cpuset_mem_spread_node(); | |
3117 | else if (current->mempolicy) | |
3118 | nid_alloc = slab_node(current->mempolicy); | |
3119 | if (nid_alloc != nid_here) | |
8b98c169 | 3120 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3121 | return NULL; |
3122 | } | |
3123 | ||
765c4507 CL |
3124 | /* |
3125 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 CL |
3126 | * certain node and fall back is permitted. First we scan all the |
3127 | * available nodelists for available objects. If that fails then we | |
3128 | * perform an allocation without specifying a node. This allows the page | |
3129 | * allocator to do its reclaim / fallback magic. We then insert the | |
3130 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3131 | */ |
8c8cc2c1 | 3132 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3133 | { |
8c8cc2c1 PE |
3134 | struct zonelist *zonelist; |
3135 | gfp_t local_flags; | |
dd1a239f | 3136 | struct zoneref *z; |
54a6eb5c MG |
3137 | struct zone *zone; |
3138 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3139 | void *obj = NULL; |
3c517a61 | 3140 | int nid; |
8c8cc2c1 PE |
3141 | |
3142 | if (flags & __GFP_THISNODE) | |
3143 | return NULL; | |
3144 | ||
0e88460d | 3145 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
6cb06229 | 3146 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3147 | |
3c517a61 CL |
3148 | retry: |
3149 | /* | |
3150 | * Look through allowed nodes for objects available | |
3151 | * from existing per node queues. | |
3152 | */ | |
54a6eb5c MG |
3153 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3154 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3155 | |
54a6eb5c | 3156 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
3c517a61 | 3157 | cache->nodelists[nid] && |
481c5346 | 3158 | cache->nodelists[nid]->free_objects) { |
3c517a61 CL |
3159 | obj = ____cache_alloc_node(cache, |
3160 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3161 | if (obj) |
3162 | break; | |
3163 | } | |
3c517a61 CL |
3164 | } |
3165 | ||
cfce6604 | 3166 | if (!obj) { |
3c517a61 CL |
3167 | /* |
3168 | * This allocation will be performed within the constraints | |
3169 | * of the current cpuset / memory policy requirements. | |
3170 | * We may trigger various forms of reclaim on the allowed | |
3171 | * set and go into memory reserves if necessary. | |
3172 | */ | |
dd47ea75 CL |
3173 | if (local_flags & __GFP_WAIT) |
3174 | local_irq_enable(); | |
3175 | kmem_flagcheck(cache, flags); | |
9ac33b2b | 3176 | obj = kmem_getpages(cache, local_flags, -1); |
dd47ea75 CL |
3177 | if (local_flags & __GFP_WAIT) |
3178 | local_irq_disable(); | |
3c517a61 CL |
3179 | if (obj) { |
3180 | /* | |
3181 | * Insert into the appropriate per node queues | |
3182 | */ | |
3183 | nid = page_to_nid(virt_to_page(obj)); | |
3184 | if (cache_grow(cache, flags, nid, obj)) { | |
3185 | obj = ____cache_alloc_node(cache, | |
3186 | flags | GFP_THISNODE, nid); | |
3187 | if (!obj) | |
3188 | /* | |
3189 | * Another processor may allocate the | |
3190 | * objects in the slab since we are | |
3191 | * not holding any locks. | |
3192 | */ | |
3193 | goto retry; | |
3194 | } else { | |
b6a60451 | 3195 | /* cache_grow already freed obj */ |
3c517a61 CL |
3196 | obj = NULL; |
3197 | } | |
3198 | } | |
aedb0eb1 | 3199 | } |
765c4507 CL |
3200 | return obj; |
3201 | } | |
3202 | ||
e498be7d CL |
3203 | /* |
3204 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3205 | */ |
8b98c169 | 3206 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3207 | int nodeid) |
e498be7d CL |
3208 | { |
3209 | struct list_head *entry; | |
b28a02de PE |
3210 | struct slab *slabp; |
3211 | struct kmem_list3 *l3; | |
3212 | void *obj; | |
b28a02de PE |
3213 | int x; |
3214 | ||
3215 | l3 = cachep->nodelists[nodeid]; | |
3216 | BUG_ON(!l3); | |
3217 | ||
a737b3e2 | 3218 | retry: |
ca3b9b91 | 3219 | check_irq_off(); |
b28a02de PE |
3220 | spin_lock(&l3->list_lock); |
3221 | entry = l3->slabs_partial.next; | |
3222 | if (entry == &l3->slabs_partial) { | |
3223 | l3->free_touched = 1; | |
3224 | entry = l3->slabs_free.next; | |
3225 | if (entry == &l3->slabs_free) | |
3226 | goto must_grow; | |
3227 | } | |
3228 | ||
3229 | slabp = list_entry(entry, struct slab, list); | |
3230 | check_spinlock_acquired_node(cachep, nodeid); | |
3231 | check_slabp(cachep, slabp); | |
3232 | ||
3233 | STATS_INC_NODEALLOCS(cachep); | |
3234 | STATS_INC_ACTIVE(cachep); | |
3235 | STATS_SET_HIGH(cachep); | |
3236 | ||
3237 | BUG_ON(slabp->inuse == cachep->num); | |
3238 | ||
78d382d7 | 3239 | obj = slab_get_obj(cachep, slabp, nodeid); |
b28a02de PE |
3240 | check_slabp(cachep, slabp); |
3241 | l3->free_objects--; | |
3242 | /* move slabp to correct slabp list: */ | |
3243 | list_del(&slabp->list); | |
3244 | ||
a737b3e2 | 3245 | if (slabp->free == BUFCTL_END) |
b28a02de | 3246 | list_add(&slabp->list, &l3->slabs_full); |
a737b3e2 | 3247 | else |
b28a02de | 3248 | list_add(&slabp->list, &l3->slabs_partial); |
e498be7d | 3249 | |
b28a02de PE |
3250 | spin_unlock(&l3->list_lock); |
3251 | goto done; | |
e498be7d | 3252 | |
a737b3e2 | 3253 | must_grow: |
b28a02de | 3254 | spin_unlock(&l3->list_lock); |
3c517a61 | 3255 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3256 | if (x) |
3257 | goto retry; | |
1da177e4 | 3258 | |
8c8cc2c1 | 3259 | return fallback_alloc(cachep, flags); |
e498be7d | 3260 | |
a737b3e2 | 3261 | done: |
b28a02de | 3262 | return obj; |
e498be7d | 3263 | } |
8c8cc2c1 PE |
3264 | |
3265 | /** | |
3266 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3267 | * @cachep: The cache to allocate from. | |
3268 | * @flags: See kmalloc(). | |
3269 | * @nodeid: node number of the target node. | |
3270 | * @caller: return address of caller, used for debug information | |
3271 | * | |
3272 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3273 | * node, which can improve the performance for cpu bound structures. | |
3274 | * | |
3275 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3276 | */ | |
3277 | static __always_inline void * | |
3278 | __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, | |
3279 | void *caller) | |
3280 | { | |
3281 | unsigned long save_flags; | |
3282 | void *ptr; | |
3283 | ||
cf40bd16 NP |
3284 | lockdep_trace_alloc(flags); |
3285 | ||
773ff60e | 3286 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3287 | return NULL; |
3288 | ||
8c8cc2c1 PE |
3289 | cache_alloc_debugcheck_before(cachep, flags); |
3290 | local_irq_save(save_flags); | |
3291 | ||
3292 | if (unlikely(nodeid == -1)) | |
3293 | nodeid = numa_node_id(); | |
3294 | ||
3295 | if (unlikely(!cachep->nodelists[nodeid])) { | |
3296 | /* Node not bootstrapped yet */ | |
3297 | ptr = fallback_alloc(cachep, flags); | |
3298 | goto out; | |
3299 | } | |
3300 | ||
3301 | if (nodeid == numa_node_id()) { | |
3302 | /* | |
3303 | * Use the locally cached objects if possible. | |
3304 | * However ____cache_alloc does not allow fallback | |
3305 | * to other nodes. It may fail while we still have | |
3306 | * objects on other nodes available. | |
3307 | */ | |
3308 | ptr = ____cache_alloc(cachep, flags); | |
3309 | if (ptr) | |
3310 | goto out; | |
3311 | } | |
3312 | /* ___cache_alloc_node can fall back to other nodes */ | |
3313 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3314 | out: | |
3315 | local_irq_restore(save_flags); | |
3316 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
d5cff635 CM |
3317 | kmemleak_alloc_recursive(ptr, obj_size(cachep), 1, cachep->flags, |
3318 | flags); | |
8c8cc2c1 | 3319 | |
c175eea4 PE |
3320 | if (likely(ptr)) |
3321 | kmemcheck_slab_alloc(cachep, flags, ptr, obj_size(cachep)); | |
3322 | ||
d07dbea4 CL |
3323 | if (unlikely((flags & __GFP_ZERO) && ptr)) |
3324 | memset(ptr, 0, obj_size(cachep)); | |
3325 | ||
8c8cc2c1 PE |
3326 | return ptr; |
3327 | } | |
3328 | ||
3329 | static __always_inline void * | |
3330 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3331 | { | |
3332 | void *objp; | |
3333 | ||
3334 | if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { | |
3335 | objp = alternate_node_alloc(cache, flags); | |
3336 | if (objp) | |
3337 | goto out; | |
3338 | } | |
3339 | objp = ____cache_alloc(cache, flags); | |
3340 | ||
3341 | /* | |
3342 | * We may just have run out of memory on the local node. | |
3343 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3344 | */ | |
3345 | if (!objp) | |
3346 | objp = ____cache_alloc_node(cache, flags, numa_node_id()); | |
3347 | ||
3348 | out: | |
3349 | return objp; | |
3350 | } | |
3351 | #else | |
3352 | ||
3353 | static __always_inline void * | |
3354 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3355 | { | |
3356 | return ____cache_alloc(cachep, flags); | |
3357 | } | |
3358 | ||
3359 | #endif /* CONFIG_NUMA */ | |
3360 | ||
3361 | static __always_inline void * | |
3362 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) | |
3363 | { | |
3364 | unsigned long save_flags; | |
3365 | void *objp; | |
3366 | ||
cf40bd16 NP |
3367 | lockdep_trace_alloc(flags); |
3368 | ||
773ff60e | 3369 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3370 | return NULL; |
3371 | ||
8c8cc2c1 PE |
3372 | cache_alloc_debugcheck_before(cachep, flags); |
3373 | local_irq_save(save_flags); | |
3374 | objp = __do_cache_alloc(cachep, flags); | |
3375 | local_irq_restore(save_flags); | |
3376 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
d5cff635 CM |
3377 | kmemleak_alloc_recursive(objp, obj_size(cachep), 1, cachep->flags, |
3378 | flags); | |
8c8cc2c1 PE |
3379 | prefetchw(objp); |
3380 | ||
c175eea4 PE |
3381 | if (likely(objp)) |
3382 | kmemcheck_slab_alloc(cachep, flags, objp, obj_size(cachep)); | |
3383 | ||
d07dbea4 CL |
3384 | if (unlikely((flags & __GFP_ZERO) && objp)) |
3385 | memset(objp, 0, obj_size(cachep)); | |
3386 | ||
8c8cc2c1 PE |
3387 | return objp; |
3388 | } | |
e498be7d CL |
3389 | |
3390 | /* | |
3391 | * Caller needs to acquire correct kmem_list's list_lock | |
3392 | */ | |
343e0d7a | 3393 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 3394 | int node) |
1da177e4 LT |
3395 | { |
3396 | int i; | |
e498be7d | 3397 | struct kmem_list3 *l3; |
1da177e4 LT |
3398 | |
3399 | for (i = 0; i < nr_objects; i++) { | |
3400 | void *objp = objpp[i]; | |
3401 | struct slab *slabp; | |
1da177e4 | 3402 | |
6ed5eb22 | 3403 | slabp = virt_to_slab(objp); |
ff69416e | 3404 | l3 = cachep->nodelists[node]; |
1da177e4 | 3405 | list_del(&slabp->list); |
ff69416e | 3406 | check_spinlock_acquired_node(cachep, node); |
1da177e4 | 3407 | check_slabp(cachep, slabp); |
78d382d7 | 3408 | slab_put_obj(cachep, slabp, objp, node); |
1da177e4 | 3409 | STATS_DEC_ACTIVE(cachep); |
e498be7d | 3410 | l3->free_objects++; |
1da177e4 LT |
3411 | check_slabp(cachep, slabp); |
3412 | ||
3413 | /* fixup slab chains */ | |
3414 | if (slabp->inuse == 0) { | |
e498be7d CL |
3415 | if (l3->free_objects > l3->free_limit) { |
3416 | l3->free_objects -= cachep->num; | |
e5ac9c5a RT |
3417 | /* No need to drop any previously held |
3418 | * lock here, even if we have a off-slab slab | |
3419 | * descriptor it is guaranteed to come from | |
3420 | * a different cache, refer to comments before | |
3421 | * alloc_slabmgmt. | |
3422 | */ | |
1da177e4 LT |
3423 | slab_destroy(cachep, slabp); |
3424 | } else { | |
e498be7d | 3425 | list_add(&slabp->list, &l3->slabs_free); |
1da177e4 LT |
3426 | } |
3427 | } else { | |
3428 | /* Unconditionally move a slab to the end of the | |
3429 | * partial list on free - maximum time for the | |
3430 | * other objects to be freed, too. | |
3431 | */ | |
e498be7d | 3432 | list_add_tail(&slabp->list, &l3->slabs_partial); |
1da177e4 LT |
3433 | } |
3434 | } | |
3435 | } | |
3436 | ||
343e0d7a | 3437 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3438 | { |
3439 | int batchcount; | |
e498be7d | 3440 | struct kmem_list3 *l3; |
ff69416e | 3441 | int node = numa_node_id(); |
1da177e4 LT |
3442 | |
3443 | batchcount = ac->batchcount; | |
3444 | #if DEBUG | |
3445 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3446 | #endif | |
3447 | check_irq_off(); | |
ff69416e | 3448 | l3 = cachep->nodelists[node]; |
873623df | 3449 | spin_lock(&l3->list_lock); |
e498be7d CL |
3450 | if (l3->shared) { |
3451 | struct array_cache *shared_array = l3->shared; | |
b28a02de | 3452 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3453 | if (max) { |
3454 | if (batchcount > max) | |
3455 | batchcount = max; | |
e498be7d | 3456 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3457 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3458 | shared_array->avail += batchcount; |
3459 | goto free_done; | |
3460 | } | |
3461 | } | |
3462 | ||
ff69416e | 3463 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 3464 | free_done: |
1da177e4 LT |
3465 | #if STATS |
3466 | { | |
3467 | int i = 0; | |
3468 | struct list_head *p; | |
3469 | ||
e498be7d CL |
3470 | p = l3->slabs_free.next; |
3471 | while (p != &(l3->slabs_free)) { | |
1da177e4 LT |
3472 | struct slab *slabp; |
3473 | ||
3474 | slabp = list_entry(p, struct slab, list); | |
3475 | BUG_ON(slabp->inuse); | |
3476 | ||
3477 | i++; | |
3478 | p = p->next; | |
3479 | } | |
3480 | STATS_SET_FREEABLE(cachep, i); | |
3481 | } | |
3482 | #endif | |
e498be7d | 3483 | spin_unlock(&l3->list_lock); |
1da177e4 | 3484 | ac->avail -= batchcount; |
a737b3e2 | 3485 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3486 | } |
3487 | ||
3488 | /* | |
a737b3e2 AM |
3489 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3490 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3491 | */ |
873623df | 3492 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 | 3493 | { |
9a2dba4b | 3494 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3495 | |
3496 | check_irq_off(); | |
d5cff635 | 3497 | kmemleak_free_recursive(objp, cachep->flags); |
1da177e4 LT |
3498 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
3499 | ||
c175eea4 PE |
3500 | kmemcheck_slab_free(cachep, objp, obj_size(cachep)); |
3501 | ||
1807a1aa SS |
3502 | /* |
3503 | * Skip calling cache_free_alien() when the platform is not numa. | |
3504 | * This will avoid cache misses that happen while accessing slabp (which | |
3505 | * is per page memory reference) to get nodeid. Instead use a global | |
3506 | * variable to skip the call, which is mostly likely to be present in | |
3507 | * the cache. | |
3508 | */ | |
3509 | if (numa_platform && cache_free_alien(cachep, objp)) | |
729bd0b7 PE |
3510 | return; |
3511 | ||
1da177e4 LT |
3512 | if (likely(ac->avail < ac->limit)) { |
3513 | STATS_INC_FREEHIT(cachep); | |
e498be7d | 3514 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3515 | return; |
3516 | } else { | |
3517 | STATS_INC_FREEMISS(cachep); | |
3518 | cache_flusharray(cachep, ac); | |
e498be7d | 3519 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3520 | } |
3521 | } | |
3522 | ||
3523 | /** | |
3524 | * kmem_cache_alloc - Allocate an object | |
3525 | * @cachep: The cache to allocate from. | |
3526 | * @flags: See kmalloc(). | |
3527 | * | |
3528 | * Allocate an object from this cache. The flags are only relevant | |
3529 | * if the cache has no available objects. | |
3530 | */ | |
343e0d7a | 3531 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3532 | { |
36555751 EGM |
3533 | void *ret = __cache_alloc(cachep, flags, __builtin_return_address(0)); |
3534 | ||
ca2b84cb EGM |
3535 | trace_kmem_cache_alloc(_RET_IP_, ret, |
3536 | obj_size(cachep), cachep->buffer_size, flags); | |
36555751 EGM |
3537 | |
3538 | return ret; | |
1da177e4 LT |
3539 | } |
3540 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3541 | ||
36555751 EGM |
3542 | #ifdef CONFIG_KMEMTRACE |
3543 | void *kmem_cache_alloc_notrace(struct kmem_cache *cachep, gfp_t flags) | |
3544 | { | |
3545 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); | |
3546 | } | |
3547 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
3548 | #endif | |
3549 | ||
1da177e4 | 3550 | /** |
7682486b | 3551 | * kmem_ptr_validate - check if an untrusted pointer might be a slab entry. |
1da177e4 LT |
3552 | * @cachep: the cache we're checking against |
3553 | * @ptr: pointer to validate | |
3554 | * | |
7682486b | 3555 | * This verifies that the untrusted pointer looks sane; |
1da177e4 LT |
3556 | * it is _not_ a guarantee that the pointer is actually |
3557 | * part of the slab cache in question, but it at least | |
3558 | * validates that the pointer can be dereferenced and | |
3559 | * looks half-way sane. | |
3560 | * | |
3561 | * Currently only used for dentry validation. | |
3562 | */ | |
b7f869a2 | 3563 | int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) |
1da177e4 | 3564 | { |
b28a02de | 3565 | unsigned long addr = (unsigned long)ptr; |
1da177e4 | 3566 | unsigned long min_addr = PAGE_OFFSET; |
b28a02de | 3567 | unsigned long align_mask = BYTES_PER_WORD - 1; |
3dafccf2 | 3568 | unsigned long size = cachep->buffer_size; |
1da177e4 LT |
3569 | struct page *page; |
3570 | ||
3571 | if (unlikely(addr < min_addr)) | |
3572 | goto out; | |
3573 | if (unlikely(addr > (unsigned long)high_memory - size)) | |
3574 | goto out; | |
3575 | if (unlikely(addr & align_mask)) | |
3576 | goto out; | |
3577 | if (unlikely(!kern_addr_valid(addr))) | |
3578 | goto out; | |
3579 | if (unlikely(!kern_addr_valid(addr + size - 1))) | |
3580 | goto out; | |
3581 | page = virt_to_page(ptr); | |
3582 | if (unlikely(!PageSlab(page))) | |
3583 | goto out; | |
065d41cb | 3584 | if (unlikely(page_get_cache(page) != cachep)) |
1da177e4 LT |
3585 | goto out; |
3586 | return 1; | |
a737b3e2 | 3587 | out: |
1da177e4 LT |
3588 | return 0; |
3589 | } | |
3590 | ||
3591 | #ifdef CONFIG_NUMA | |
8b98c169 CH |
3592 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3593 | { | |
36555751 EGM |
3594 | void *ret = __cache_alloc_node(cachep, flags, nodeid, |
3595 | __builtin_return_address(0)); | |
3596 | ||
ca2b84cb EGM |
3597 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3598 | obj_size(cachep), cachep->buffer_size, | |
3599 | flags, nodeid); | |
36555751 EGM |
3600 | |
3601 | return ret; | |
8b98c169 | 3602 | } |
1da177e4 LT |
3603 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3604 | ||
36555751 EGM |
3605 | #ifdef CONFIG_KMEMTRACE |
3606 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *cachep, | |
3607 | gfp_t flags, | |
3608 | int nodeid) | |
3609 | { | |
3610 | return __cache_alloc_node(cachep, flags, nodeid, | |
3611 | __builtin_return_address(0)); | |
3612 | } | |
3613 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
3614 | #endif | |
3615 | ||
8b98c169 CH |
3616 | static __always_inline void * |
3617 | __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller) | |
97e2bde4 | 3618 | { |
343e0d7a | 3619 | struct kmem_cache *cachep; |
36555751 | 3620 | void *ret; |
97e2bde4 MS |
3621 | |
3622 | cachep = kmem_find_general_cachep(size, flags); | |
6cb8f913 CL |
3623 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3624 | return cachep; | |
36555751 EGM |
3625 | ret = kmem_cache_alloc_node_notrace(cachep, flags, node); |
3626 | ||
ca2b84cb EGM |
3627 | trace_kmalloc_node((unsigned long) caller, ret, |
3628 | size, cachep->buffer_size, flags, node); | |
36555751 EGM |
3629 | |
3630 | return ret; | |
97e2bde4 | 3631 | } |
8b98c169 | 3632 | |
36555751 | 3633 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_KMEMTRACE) |
8b98c169 CH |
3634 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3635 | { | |
3636 | return __do_kmalloc_node(size, flags, node, | |
3637 | __builtin_return_address(0)); | |
3638 | } | |
dbe5e69d | 3639 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3640 | |
3641 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3642 | int node, unsigned long caller) |
8b98c169 | 3643 | { |
ce71e27c | 3644 | return __do_kmalloc_node(size, flags, node, (void *)caller); |
8b98c169 CH |
3645 | } |
3646 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3647 | #else | |
3648 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3649 | { | |
3650 | return __do_kmalloc_node(size, flags, node, NULL); | |
3651 | } | |
3652 | EXPORT_SYMBOL(__kmalloc_node); | |
3653 | #endif /* CONFIG_DEBUG_SLAB */ | |
3654 | #endif /* CONFIG_NUMA */ | |
1da177e4 LT |
3655 | |
3656 | /** | |
800590f5 | 3657 | * __do_kmalloc - allocate memory |
1da177e4 | 3658 | * @size: how many bytes of memory are required. |
800590f5 | 3659 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3660 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3661 | */ |
7fd6b141 PE |
3662 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
3663 | void *caller) | |
1da177e4 | 3664 | { |
343e0d7a | 3665 | struct kmem_cache *cachep; |
36555751 | 3666 | void *ret; |
1da177e4 | 3667 | |
97e2bde4 MS |
3668 | /* If you want to save a few bytes .text space: replace |
3669 | * __ with kmem_. | |
3670 | * Then kmalloc uses the uninlined functions instead of the inline | |
3671 | * functions. | |
3672 | */ | |
3673 | cachep = __find_general_cachep(size, flags); | |
a5c96d8a LT |
3674 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3675 | return cachep; | |
36555751 EGM |
3676 | ret = __cache_alloc(cachep, flags, caller); |
3677 | ||
ca2b84cb EGM |
3678 | trace_kmalloc((unsigned long) caller, ret, |
3679 | size, cachep->buffer_size, flags); | |
36555751 EGM |
3680 | |
3681 | return ret; | |
7fd6b141 PE |
3682 | } |
3683 | ||
7fd6b141 | 3684 | |
36555751 | 3685 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_KMEMTRACE) |
7fd6b141 PE |
3686 | void *__kmalloc(size_t size, gfp_t flags) |
3687 | { | |
871751e2 | 3688 | return __do_kmalloc(size, flags, __builtin_return_address(0)); |
1da177e4 LT |
3689 | } |
3690 | EXPORT_SYMBOL(__kmalloc); | |
3691 | ||
ce71e27c | 3692 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3693 | { |
ce71e27c | 3694 | return __do_kmalloc(size, flags, (void *)caller); |
7fd6b141 PE |
3695 | } |
3696 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3697 | |
3698 | #else | |
3699 | void *__kmalloc(size_t size, gfp_t flags) | |
3700 | { | |
3701 | return __do_kmalloc(size, flags, NULL); | |
3702 | } | |
3703 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3704 | #endif |
3705 | ||
1da177e4 LT |
3706 | /** |
3707 | * kmem_cache_free - Deallocate an object | |
3708 | * @cachep: The cache the allocation was from. | |
3709 | * @objp: The previously allocated object. | |
3710 | * | |
3711 | * Free an object which was previously allocated from this | |
3712 | * cache. | |
3713 | */ | |
343e0d7a | 3714 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3715 | { |
3716 | unsigned long flags; | |
3717 | ||
3718 | local_irq_save(flags); | |
898552c9 | 3719 | debug_check_no_locks_freed(objp, obj_size(cachep)); |
3ac7fe5a TG |
3720 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
3721 | debug_check_no_obj_freed(objp, obj_size(cachep)); | |
873623df | 3722 | __cache_free(cachep, objp); |
1da177e4 | 3723 | local_irq_restore(flags); |
36555751 | 3724 | |
ca2b84cb | 3725 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3726 | } |
3727 | EXPORT_SYMBOL(kmem_cache_free); | |
3728 | ||
1da177e4 LT |
3729 | /** |
3730 | * kfree - free previously allocated memory | |
3731 | * @objp: pointer returned by kmalloc. | |
3732 | * | |
80e93eff PE |
3733 | * If @objp is NULL, no operation is performed. |
3734 | * | |
1da177e4 LT |
3735 | * Don't free memory not originally allocated by kmalloc() |
3736 | * or you will run into trouble. | |
3737 | */ | |
3738 | void kfree(const void *objp) | |
3739 | { | |
343e0d7a | 3740 | struct kmem_cache *c; |
1da177e4 LT |
3741 | unsigned long flags; |
3742 | ||
2121db74 PE |
3743 | trace_kfree(_RET_IP_, objp); |
3744 | ||
6cb8f913 | 3745 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3746 | return; |
3747 | local_irq_save(flags); | |
3748 | kfree_debugcheck(objp); | |
6ed5eb22 | 3749 | c = virt_to_cache(objp); |
f9b8404c | 3750 | debug_check_no_locks_freed(objp, obj_size(c)); |
3ac7fe5a | 3751 | debug_check_no_obj_freed(objp, obj_size(c)); |
873623df | 3752 | __cache_free(c, (void *)objp); |
1da177e4 LT |
3753 | local_irq_restore(flags); |
3754 | } | |
3755 | EXPORT_SYMBOL(kfree); | |
3756 | ||
343e0d7a | 3757 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
1da177e4 | 3758 | { |
3dafccf2 | 3759 | return obj_size(cachep); |
1da177e4 LT |
3760 | } |
3761 | EXPORT_SYMBOL(kmem_cache_size); | |
3762 | ||
343e0d7a | 3763 | const char *kmem_cache_name(struct kmem_cache *cachep) |
1944972d ACM |
3764 | { |
3765 | return cachep->name; | |
3766 | } | |
3767 | EXPORT_SYMBOL_GPL(kmem_cache_name); | |
3768 | ||
e498be7d | 3769 | /* |
183ff22b | 3770 | * This initializes kmem_list3 or resizes various caches for all nodes. |
e498be7d | 3771 | */ |
83b519e8 | 3772 | static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3773 | { |
3774 | int node; | |
3775 | struct kmem_list3 *l3; | |
cafeb02e | 3776 | struct array_cache *new_shared; |
3395ee05 | 3777 | struct array_cache **new_alien = NULL; |
e498be7d | 3778 | |
9c09a95c | 3779 | for_each_online_node(node) { |
cafeb02e | 3780 | |
3395ee05 | 3781 | if (use_alien_caches) { |
83b519e8 | 3782 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3783 | if (!new_alien) |
3784 | goto fail; | |
3785 | } | |
cafeb02e | 3786 | |
63109846 ED |
3787 | new_shared = NULL; |
3788 | if (cachep->shared) { | |
3789 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3790 | cachep->shared*cachep->batchcount, |
83b519e8 | 3791 | 0xbaadf00d, gfp); |
63109846 ED |
3792 | if (!new_shared) { |
3793 | free_alien_cache(new_alien); | |
3794 | goto fail; | |
3795 | } | |
0718dc2a | 3796 | } |
cafeb02e | 3797 | |
a737b3e2 AM |
3798 | l3 = cachep->nodelists[node]; |
3799 | if (l3) { | |
cafeb02e CL |
3800 | struct array_cache *shared = l3->shared; |
3801 | ||
e498be7d CL |
3802 | spin_lock_irq(&l3->list_lock); |
3803 | ||
cafeb02e | 3804 | if (shared) |
0718dc2a CL |
3805 | free_block(cachep, shared->entry, |
3806 | shared->avail, node); | |
e498be7d | 3807 | |
cafeb02e CL |
3808 | l3->shared = new_shared; |
3809 | if (!l3->alien) { | |
e498be7d CL |
3810 | l3->alien = new_alien; |
3811 | new_alien = NULL; | |
3812 | } | |
b28a02de | 3813 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3814 | cachep->batchcount + cachep->num; |
e498be7d | 3815 | spin_unlock_irq(&l3->list_lock); |
cafeb02e | 3816 | kfree(shared); |
e498be7d CL |
3817 | free_alien_cache(new_alien); |
3818 | continue; | |
3819 | } | |
83b519e8 | 3820 | l3 = kmalloc_node(sizeof(struct kmem_list3), gfp, node); |
0718dc2a CL |
3821 | if (!l3) { |
3822 | free_alien_cache(new_alien); | |
3823 | kfree(new_shared); | |
e498be7d | 3824 | goto fail; |
0718dc2a | 3825 | } |
e498be7d CL |
3826 | |
3827 | kmem_list3_init(l3); | |
3828 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
a737b3e2 | 3829 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
cafeb02e | 3830 | l3->shared = new_shared; |
e498be7d | 3831 | l3->alien = new_alien; |
b28a02de | 3832 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3833 | cachep->batchcount + cachep->num; |
e498be7d CL |
3834 | cachep->nodelists[node] = l3; |
3835 | } | |
cafeb02e | 3836 | return 0; |
0718dc2a | 3837 | |
a737b3e2 | 3838 | fail: |
0718dc2a CL |
3839 | if (!cachep->next.next) { |
3840 | /* Cache is not active yet. Roll back what we did */ | |
3841 | node--; | |
3842 | while (node >= 0) { | |
3843 | if (cachep->nodelists[node]) { | |
3844 | l3 = cachep->nodelists[node]; | |
3845 | ||
3846 | kfree(l3->shared); | |
3847 | free_alien_cache(l3->alien); | |
3848 | kfree(l3); | |
3849 | cachep->nodelists[node] = NULL; | |
3850 | } | |
3851 | node--; | |
3852 | } | |
3853 | } | |
cafeb02e | 3854 | return -ENOMEM; |
e498be7d CL |
3855 | } |
3856 | ||
1da177e4 | 3857 | struct ccupdate_struct { |
343e0d7a | 3858 | struct kmem_cache *cachep; |
1da177e4 LT |
3859 | struct array_cache *new[NR_CPUS]; |
3860 | }; | |
3861 | ||
3862 | static void do_ccupdate_local(void *info) | |
3863 | { | |
a737b3e2 | 3864 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3865 | struct array_cache *old; |
3866 | ||
3867 | check_irq_off(); | |
9a2dba4b | 3868 | old = cpu_cache_get(new->cachep); |
e498be7d | 3869 | |
1da177e4 LT |
3870 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3871 | new->new[smp_processor_id()] = old; | |
3872 | } | |
3873 | ||
b5d8ca7c | 3874 | /* Always called with the cache_chain_mutex held */ |
a737b3e2 | 3875 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3876 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3877 | { |
d2e7b7d0 | 3878 | struct ccupdate_struct *new; |
2ed3a4ef | 3879 | int i; |
1da177e4 | 3880 | |
83b519e8 | 3881 | new = kzalloc(sizeof(*new), gfp); |
d2e7b7d0 SS |
3882 | if (!new) |
3883 | return -ENOMEM; | |
3884 | ||
e498be7d | 3885 | for_each_online_cpu(i) { |
d2e7b7d0 | 3886 | new->new[i] = alloc_arraycache(cpu_to_node(i), limit, |
83b519e8 | 3887 | batchcount, gfp); |
d2e7b7d0 | 3888 | if (!new->new[i]) { |
b28a02de | 3889 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3890 | kfree(new->new[i]); |
3891 | kfree(new); | |
e498be7d | 3892 | return -ENOMEM; |
1da177e4 LT |
3893 | } |
3894 | } | |
d2e7b7d0 | 3895 | new->cachep = cachep; |
1da177e4 | 3896 | |
15c8b6c1 | 3897 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 3898 | |
1da177e4 | 3899 | check_irq_on(); |
1da177e4 LT |
3900 | cachep->batchcount = batchcount; |
3901 | cachep->limit = limit; | |
e498be7d | 3902 | cachep->shared = shared; |
1da177e4 | 3903 | |
e498be7d | 3904 | for_each_online_cpu(i) { |
d2e7b7d0 | 3905 | struct array_cache *ccold = new->new[i]; |
1da177e4 LT |
3906 | if (!ccold) |
3907 | continue; | |
e498be7d | 3908 | spin_lock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
ff69416e | 3909 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_node(i)); |
e498be7d | 3910 | spin_unlock_irq(&cachep->nodelists[cpu_to_node(i)]->list_lock); |
1da177e4 LT |
3911 | kfree(ccold); |
3912 | } | |
d2e7b7d0 | 3913 | kfree(new); |
83b519e8 | 3914 | return alloc_kmemlist(cachep, gfp); |
1da177e4 LT |
3915 | } |
3916 | ||
b5d8ca7c | 3917 | /* Called with cache_chain_mutex held always */ |
83b519e8 | 3918 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3919 | { |
3920 | int err; | |
3921 | int limit, shared; | |
3922 | ||
a737b3e2 AM |
3923 | /* |
3924 | * The head array serves three purposes: | |
1da177e4 LT |
3925 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3926 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3927 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3928 | * bufctl chains: array operations are cheaper. |
3929 | * The numbers are guessed, we should auto-tune as described by | |
3930 | * Bonwick. | |
3931 | */ | |
3dafccf2 | 3932 | if (cachep->buffer_size > 131072) |
1da177e4 | 3933 | limit = 1; |
3dafccf2 | 3934 | else if (cachep->buffer_size > PAGE_SIZE) |
1da177e4 | 3935 | limit = 8; |
3dafccf2 | 3936 | else if (cachep->buffer_size > 1024) |
1da177e4 | 3937 | limit = 24; |
3dafccf2 | 3938 | else if (cachep->buffer_size > 256) |
1da177e4 LT |
3939 | limit = 54; |
3940 | else | |
3941 | limit = 120; | |
3942 | ||
a737b3e2 AM |
3943 | /* |
3944 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3945 | * allocation behaviour: Most allocs on one cpu, most free operations |
3946 | * on another cpu. For these cases, an efficient object passing between | |
3947 | * cpus is necessary. This is provided by a shared array. The array | |
3948 | * replaces Bonwick's magazine layer. | |
3949 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3950 | * to a larger limit. Thus disabled by default. | |
3951 | */ | |
3952 | shared = 0; | |
364fbb29 | 3953 | if (cachep->buffer_size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3954 | shared = 8; |
1da177e4 LT |
3955 | |
3956 | #if DEBUG | |
a737b3e2 AM |
3957 | /* |
3958 | * With debugging enabled, large batchcount lead to excessively long | |
3959 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3960 | */ |
3961 | if (limit > 32) | |
3962 | limit = 32; | |
3963 | #endif | |
83b519e8 | 3964 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared, gfp); |
1da177e4 LT |
3965 | if (err) |
3966 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 3967 | cachep->name, -err); |
2ed3a4ef | 3968 | return err; |
1da177e4 LT |
3969 | } |
3970 | ||
1b55253a CL |
3971 | /* |
3972 | * Drain an array if it contains any elements taking the l3 lock only if | |
b18e7e65 CL |
3973 | * necessary. Note that the l3 listlock also protects the array_cache |
3974 | * if drain_array() is used on the shared array. | |
1b55253a CL |
3975 | */ |
3976 | void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, | |
3977 | struct array_cache *ac, int force, int node) | |
1da177e4 LT |
3978 | { |
3979 | int tofree; | |
3980 | ||
1b55253a CL |
3981 | if (!ac || !ac->avail) |
3982 | return; | |
1da177e4 LT |
3983 | if (ac->touched && !force) { |
3984 | ac->touched = 0; | |
b18e7e65 | 3985 | } else { |
1b55253a | 3986 | spin_lock_irq(&l3->list_lock); |
b18e7e65 CL |
3987 | if (ac->avail) { |
3988 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
3989 | if (tofree > ac->avail) | |
3990 | tofree = (ac->avail + 1) / 2; | |
3991 | free_block(cachep, ac->entry, tofree, node); | |
3992 | ac->avail -= tofree; | |
3993 | memmove(ac->entry, &(ac->entry[tofree]), | |
3994 | sizeof(void *) * ac->avail); | |
3995 | } | |
1b55253a | 3996 | spin_unlock_irq(&l3->list_lock); |
1da177e4 LT |
3997 | } |
3998 | } | |
3999 | ||
4000 | /** | |
4001 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4002 | * @w: work descriptor |
1da177e4 LT |
4003 | * |
4004 | * Called from workqueue/eventd every few seconds. | |
4005 | * Purpose: | |
4006 | * - clear the per-cpu caches for this CPU. | |
4007 | * - return freeable pages to the main free memory pool. | |
4008 | * | |
a737b3e2 AM |
4009 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4010 | * again on the next iteration. | |
1da177e4 | 4011 | */ |
7c5cae36 | 4012 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4013 | { |
7a7c381d | 4014 | struct kmem_cache *searchp; |
e498be7d | 4015 | struct kmem_list3 *l3; |
aab2207c | 4016 | int node = numa_node_id(); |
bf6aede7 | 4017 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4018 | |
7c5cae36 | 4019 | if (!mutex_trylock(&cache_chain_mutex)) |
1da177e4 | 4020 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4021 | goto out; |
1da177e4 | 4022 | |
7a7c381d | 4023 | list_for_each_entry(searchp, &cache_chain, next) { |
1da177e4 LT |
4024 | check_irq_on(); |
4025 | ||
35386e3b CL |
4026 | /* |
4027 | * We only take the l3 lock if absolutely necessary and we | |
4028 | * have established with reasonable certainty that | |
4029 | * we can do some work if the lock was obtained. | |
4030 | */ | |
aab2207c | 4031 | l3 = searchp->nodelists[node]; |
35386e3b | 4032 | |
8fce4d8e | 4033 | reap_alien(searchp, l3); |
1da177e4 | 4034 | |
aab2207c | 4035 | drain_array(searchp, l3, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4036 | |
35386e3b CL |
4037 | /* |
4038 | * These are racy checks but it does not matter | |
4039 | * if we skip one check or scan twice. | |
4040 | */ | |
e498be7d | 4041 | if (time_after(l3->next_reap, jiffies)) |
35386e3b | 4042 | goto next; |
1da177e4 | 4043 | |
e498be7d | 4044 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
1da177e4 | 4045 | |
aab2207c | 4046 | drain_array(searchp, l3, l3->shared, 0, node); |
1da177e4 | 4047 | |
ed11d9eb | 4048 | if (l3->free_touched) |
e498be7d | 4049 | l3->free_touched = 0; |
ed11d9eb CL |
4050 | else { |
4051 | int freed; | |
1da177e4 | 4052 | |
ed11d9eb CL |
4053 | freed = drain_freelist(searchp, l3, (l3->free_limit + |
4054 | 5 * searchp->num - 1) / (5 * searchp->num)); | |
4055 | STATS_ADD_REAPED(searchp, freed); | |
4056 | } | |
35386e3b | 4057 | next: |
1da177e4 LT |
4058 | cond_resched(); |
4059 | } | |
4060 | check_irq_on(); | |
fc0abb14 | 4061 | mutex_unlock(&cache_chain_mutex); |
8fce4d8e | 4062 | next_reap_node(); |
7c5cae36 | 4063 | out: |
a737b3e2 | 4064 | /* Set up the next iteration */ |
7c5cae36 | 4065 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); |
1da177e4 LT |
4066 | } |
4067 | ||
158a9624 | 4068 | #ifdef CONFIG_SLABINFO |
1da177e4 | 4069 | |
85289f98 | 4070 | static void print_slabinfo_header(struct seq_file *m) |
1da177e4 | 4071 | { |
85289f98 PE |
4072 | /* |
4073 | * Output format version, so at least we can change it | |
4074 | * without _too_ many complaints. | |
4075 | */ | |
1da177e4 | 4076 | #if STATS |
85289f98 | 4077 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
1da177e4 | 4078 | #else |
85289f98 | 4079 | seq_puts(m, "slabinfo - version: 2.1\n"); |
1da177e4 | 4080 | #endif |
85289f98 PE |
4081 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
4082 | "<objperslab> <pagesperslab>"); | |
4083 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4084 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1da177e4 | 4085 | #if STATS |
85289f98 | 4086 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
fb7faf33 | 4087 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
85289f98 | 4088 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1da177e4 | 4089 | #endif |
85289f98 PE |
4090 | seq_putc(m, '\n'); |
4091 | } | |
4092 | ||
4093 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4094 | { | |
4095 | loff_t n = *pos; | |
85289f98 | 4096 | |
fc0abb14 | 4097 | mutex_lock(&cache_chain_mutex); |
85289f98 PE |
4098 | if (!n) |
4099 | print_slabinfo_header(m); | |
b92151ba PE |
4100 | |
4101 | return seq_list_start(&cache_chain, *pos); | |
1da177e4 LT |
4102 | } |
4103 | ||
4104 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4105 | { | |
b92151ba | 4106 | return seq_list_next(p, &cache_chain, pos); |
1da177e4 LT |
4107 | } |
4108 | ||
4109 | static void s_stop(struct seq_file *m, void *p) | |
4110 | { | |
fc0abb14 | 4111 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
4112 | } |
4113 | ||
4114 | static int s_show(struct seq_file *m, void *p) | |
4115 | { | |
b92151ba | 4116 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); |
b28a02de PE |
4117 | struct slab *slabp; |
4118 | unsigned long active_objs; | |
4119 | unsigned long num_objs; | |
4120 | unsigned long active_slabs = 0; | |
4121 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4122 | const char *name; |
1da177e4 | 4123 | char *error = NULL; |
e498be7d CL |
4124 | int node; |
4125 | struct kmem_list3 *l3; | |
1da177e4 | 4126 | |
1da177e4 LT |
4127 | active_objs = 0; |
4128 | num_slabs = 0; | |
e498be7d CL |
4129 | for_each_online_node(node) { |
4130 | l3 = cachep->nodelists[node]; | |
4131 | if (!l3) | |
4132 | continue; | |
4133 | ||
ca3b9b91 RT |
4134 | check_irq_on(); |
4135 | spin_lock_irq(&l3->list_lock); | |
e498be7d | 4136 | |
7a7c381d | 4137 | list_for_each_entry(slabp, &l3->slabs_full, list) { |
e498be7d CL |
4138 | if (slabp->inuse != cachep->num && !error) |
4139 | error = "slabs_full accounting error"; | |
4140 | active_objs += cachep->num; | |
4141 | active_slabs++; | |
4142 | } | |
7a7c381d | 4143 | list_for_each_entry(slabp, &l3->slabs_partial, list) { |
e498be7d CL |
4144 | if (slabp->inuse == cachep->num && !error) |
4145 | error = "slabs_partial inuse accounting error"; | |
4146 | if (!slabp->inuse && !error) | |
4147 | error = "slabs_partial/inuse accounting error"; | |
4148 | active_objs += slabp->inuse; | |
4149 | active_slabs++; | |
4150 | } | |
7a7c381d | 4151 | list_for_each_entry(slabp, &l3->slabs_free, list) { |
e498be7d CL |
4152 | if (slabp->inuse && !error) |
4153 | error = "slabs_free/inuse accounting error"; | |
4154 | num_slabs++; | |
4155 | } | |
4156 | free_objects += l3->free_objects; | |
4484ebf1 RT |
4157 | if (l3->shared) |
4158 | shared_avail += l3->shared->avail; | |
e498be7d | 4159 | |
ca3b9b91 | 4160 | spin_unlock_irq(&l3->list_lock); |
1da177e4 | 4161 | } |
b28a02de PE |
4162 | num_slabs += active_slabs; |
4163 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4164 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4165 | error = "free_objects accounting error"; |
4166 | ||
b28a02de | 4167 | name = cachep->name; |
1da177e4 LT |
4168 | if (error) |
4169 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4170 | ||
4171 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
3dafccf2 | 4172 | name, active_objs, num_objs, cachep->buffer_size, |
b28a02de | 4173 | cachep->num, (1 << cachep->gfporder)); |
1da177e4 | 4174 | seq_printf(m, " : tunables %4u %4u %4u", |
b28a02de | 4175 | cachep->limit, cachep->batchcount, cachep->shared); |
e498be7d | 4176 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
b28a02de | 4177 | active_slabs, num_slabs, shared_avail); |
1da177e4 | 4178 | #if STATS |
b28a02de | 4179 | { /* list3 stats */ |
1da177e4 LT |
4180 | unsigned long high = cachep->high_mark; |
4181 | unsigned long allocs = cachep->num_allocations; | |
4182 | unsigned long grown = cachep->grown; | |
4183 | unsigned long reaped = cachep->reaped; | |
4184 | unsigned long errors = cachep->errors; | |
4185 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4186 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4187 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4188 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4189 | |
e498be7d | 4190 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu \ |
fb7faf33 | 4191 | %4lu %4lu %4lu %4lu %4lu", allocs, high, grown, |
a737b3e2 | 4192 | reaped, errors, max_freeable, node_allocs, |
fb7faf33 | 4193 | node_frees, overflows); |
1da177e4 LT |
4194 | } |
4195 | /* cpu stats */ | |
4196 | { | |
4197 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4198 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4199 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4200 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4201 | ||
4202 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4203 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4204 | } |
4205 | #endif | |
4206 | seq_putc(m, '\n'); | |
1da177e4 LT |
4207 | return 0; |
4208 | } | |
4209 | ||
4210 | /* | |
4211 | * slabinfo_op - iterator that generates /proc/slabinfo | |
4212 | * | |
4213 | * Output layout: | |
4214 | * cache-name | |
4215 | * num-active-objs | |
4216 | * total-objs | |
4217 | * object size | |
4218 | * num-active-slabs | |
4219 | * total-slabs | |
4220 | * num-pages-per-slab | |
4221 | * + further values on SMP and with statistics enabled | |
4222 | */ | |
4223 | ||
7b3c3a50 | 4224 | static const struct seq_operations slabinfo_op = { |
b28a02de PE |
4225 | .start = s_start, |
4226 | .next = s_next, | |
4227 | .stop = s_stop, | |
4228 | .show = s_show, | |
1da177e4 LT |
4229 | }; |
4230 | ||
4231 | #define MAX_SLABINFO_WRITE 128 | |
4232 | /** | |
4233 | * slabinfo_write - Tuning for the slab allocator | |
4234 | * @file: unused | |
4235 | * @buffer: user buffer | |
4236 | * @count: data length | |
4237 | * @ppos: unused | |
4238 | */ | |
b28a02de PE |
4239 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
4240 | size_t count, loff_t *ppos) | |
1da177e4 | 4241 | { |
b28a02de | 4242 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4243 | int limit, batchcount, shared, res; |
7a7c381d | 4244 | struct kmem_cache *cachep; |
b28a02de | 4245 | |
1da177e4 LT |
4246 | if (count > MAX_SLABINFO_WRITE) |
4247 | return -EINVAL; | |
4248 | if (copy_from_user(&kbuf, buffer, count)) | |
4249 | return -EFAULT; | |
b28a02de | 4250 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4251 | |
4252 | tmp = strchr(kbuf, ' '); | |
4253 | if (!tmp) | |
4254 | return -EINVAL; | |
4255 | *tmp = '\0'; | |
4256 | tmp++; | |
4257 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4258 | return -EINVAL; | |
4259 | ||
4260 | /* Find the cache in the chain of caches. */ | |
fc0abb14 | 4261 | mutex_lock(&cache_chain_mutex); |
1da177e4 | 4262 | res = -EINVAL; |
7a7c381d | 4263 | list_for_each_entry(cachep, &cache_chain, next) { |
1da177e4 | 4264 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4265 | if (limit < 1 || batchcount < 1 || |
4266 | batchcount > limit || shared < 0) { | |
e498be7d | 4267 | res = 0; |
1da177e4 | 4268 | } else { |
e498be7d | 4269 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4270 | batchcount, shared, |
4271 | GFP_KERNEL); | |
1da177e4 LT |
4272 | } |
4273 | break; | |
4274 | } | |
4275 | } | |
fc0abb14 | 4276 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
4277 | if (res >= 0) |
4278 | res = count; | |
4279 | return res; | |
4280 | } | |
871751e2 | 4281 | |
7b3c3a50 AD |
4282 | static int slabinfo_open(struct inode *inode, struct file *file) |
4283 | { | |
4284 | return seq_open(file, &slabinfo_op); | |
4285 | } | |
4286 | ||
4287 | static const struct file_operations proc_slabinfo_operations = { | |
4288 | .open = slabinfo_open, | |
4289 | .read = seq_read, | |
4290 | .write = slabinfo_write, | |
4291 | .llseek = seq_lseek, | |
4292 | .release = seq_release, | |
4293 | }; | |
4294 | ||
871751e2 AV |
4295 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
4296 | ||
4297 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4298 | { | |
871751e2 | 4299 | mutex_lock(&cache_chain_mutex); |
b92151ba | 4300 | return seq_list_start(&cache_chain, *pos); |
871751e2 AV |
4301 | } |
4302 | ||
4303 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4304 | { | |
4305 | unsigned long *p; | |
4306 | int l; | |
4307 | if (!v) | |
4308 | return 1; | |
4309 | l = n[1]; | |
4310 | p = n + 2; | |
4311 | while (l) { | |
4312 | int i = l/2; | |
4313 | unsigned long *q = p + 2 * i; | |
4314 | if (*q == v) { | |
4315 | q[1]++; | |
4316 | return 1; | |
4317 | } | |
4318 | if (*q > v) { | |
4319 | l = i; | |
4320 | } else { | |
4321 | p = q + 2; | |
4322 | l -= i + 1; | |
4323 | } | |
4324 | } | |
4325 | if (++n[1] == n[0]) | |
4326 | return 0; | |
4327 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4328 | p[0] = v; | |
4329 | p[1] = 1; | |
4330 | return 1; | |
4331 | } | |
4332 | ||
4333 | static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) | |
4334 | { | |
4335 | void *p; | |
4336 | int i; | |
4337 | if (n[0] == n[1]) | |
4338 | return; | |
4339 | for (i = 0, p = s->s_mem; i < c->num; i++, p += c->buffer_size) { | |
4340 | if (slab_bufctl(s)[i] != BUFCTL_ACTIVE) | |
4341 | continue; | |
4342 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) | |
4343 | return; | |
4344 | } | |
4345 | } | |
4346 | ||
4347 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4348 | { | |
4349 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4350 | unsigned long offset, size; |
9281acea | 4351 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4352 | |
a5c43dae | 4353 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4354 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4355 | if (modname[0]) |
871751e2 AV |
4356 | seq_printf(m, " [%s]", modname); |
4357 | return; | |
4358 | } | |
4359 | #endif | |
4360 | seq_printf(m, "%p", (void *)address); | |
4361 | } | |
4362 | ||
4363 | static int leaks_show(struct seq_file *m, void *p) | |
4364 | { | |
b92151ba | 4365 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); |
871751e2 AV |
4366 | struct slab *slabp; |
4367 | struct kmem_list3 *l3; | |
4368 | const char *name; | |
4369 | unsigned long *n = m->private; | |
4370 | int node; | |
4371 | int i; | |
4372 | ||
4373 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4374 | return 0; | |
4375 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4376 | return 0; | |
4377 | ||
4378 | /* OK, we can do it */ | |
4379 | ||
4380 | n[1] = 0; | |
4381 | ||
4382 | for_each_online_node(node) { | |
4383 | l3 = cachep->nodelists[node]; | |
4384 | if (!l3) | |
4385 | continue; | |
4386 | ||
4387 | check_irq_on(); | |
4388 | spin_lock_irq(&l3->list_lock); | |
4389 | ||
7a7c381d | 4390 | list_for_each_entry(slabp, &l3->slabs_full, list) |
871751e2 | 4391 | handle_slab(n, cachep, slabp); |
7a7c381d | 4392 | list_for_each_entry(slabp, &l3->slabs_partial, list) |
871751e2 | 4393 | handle_slab(n, cachep, slabp); |
871751e2 AV |
4394 | spin_unlock_irq(&l3->list_lock); |
4395 | } | |
4396 | name = cachep->name; | |
4397 | if (n[0] == n[1]) { | |
4398 | /* Increase the buffer size */ | |
4399 | mutex_unlock(&cache_chain_mutex); | |
4400 | m->private = kzalloc(n[0] * 4 * sizeof(unsigned long), GFP_KERNEL); | |
4401 | if (!m->private) { | |
4402 | /* Too bad, we are really out */ | |
4403 | m->private = n; | |
4404 | mutex_lock(&cache_chain_mutex); | |
4405 | return -ENOMEM; | |
4406 | } | |
4407 | *(unsigned long *)m->private = n[0] * 2; | |
4408 | kfree(n); | |
4409 | mutex_lock(&cache_chain_mutex); | |
4410 | /* Now make sure this entry will be retried */ | |
4411 | m->count = m->size; | |
4412 | return 0; | |
4413 | } | |
4414 | for (i = 0; i < n[1]; i++) { | |
4415 | seq_printf(m, "%s: %lu ", name, n[2*i+3]); | |
4416 | show_symbol(m, n[2*i+2]); | |
4417 | seq_putc(m, '\n'); | |
4418 | } | |
d2e7b7d0 | 4419 | |
871751e2 AV |
4420 | return 0; |
4421 | } | |
4422 | ||
a0ec95a8 | 4423 | static const struct seq_operations slabstats_op = { |
871751e2 AV |
4424 | .start = leaks_start, |
4425 | .next = s_next, | |
4426 | .stop = s_stop, | |
4427 | .show = leaks_show, | |
4428 | }; | |
a0ec95a8 AD |
4429 | |
4430 | static int slabstats_open(struct inode *inode, struct file *file) | |
4431 | { | |
4432 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4433 | int ret = -ENOMEM; | |
4434 | if (n) { | |
4435 | ret = seq_open(file, &slabstats_op); | |
4436 | if (!ret) { | |
4437 | struct seq_file *m = file->private_data; | |
4438 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4439 | m->private = n; | |
4440 | n = NULL; | |
4441 | } | |
4442 | kfree(n); | |
4443 | } | |
4444 | return ret; | |
4445 | } | |
4446 | ||
4447 | static const struct file_operations proc_slabstats_operations = { | |
4448 | .open = slabstats_open, | |
4449 | .read = seq_read, | |
4450 | .llseek = seq_lseek, | |
4451 | .release = seq_release_private, | |
4452 | }; | |
4453 | #endif | |
4454 | ||
4455 | static int __init slab_proc_init(void) | |
4456 | { | |
7b3c3a50 | 4457 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); |
a0ec95a8 AD |
4458 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
4459 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4460 | #endif |
a0ec95a8 AD |
4461 | return 0; |
4462 | } | |
4463 | module_init(slab_proc_init); | |
1da177e4 LT |
4464 | #endif |
4465 | ||
00e145b6 MS |
4466 | /** |
4467 | * ksize - get the actual amount of memory allocated for a given object | |
4468 | * @objp: Pointer to the object | |
4469 | * | |
4470 | * kmalloc may internally round up allocations and return more memory | |
4471 | * than requested. ksize() can be used to determine the actual amount of | |
4472 | * memory allocated. The caller may use this additional memory, even though | |
4473 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4474 | * The caller must guarantee that objp points to a valid object previously | |
4475 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4476 | * must not be freed during the duration of the call. | |
4477 | */ | |
fd76bab2 | 4478 | size_t ksize(const void *objp) |
1da177e4 | 4479 | { |
ef8b4520 CL |
4480 | BUG_ON(!objp); |
4481 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4482 | return 0; |
1da177e4 | 4483 | |
6ed5eb22 | 4484 | return obj_size(virt_to_cache(objp)); |
1da177e4 | 4485 | } |
b1aabecd | 4486 | EXPORT_SYMBOL(ksize); |