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> |
8f9f8d9e | 118 | #include <linux/memory.h> |
1da177e4 | 119 | |
1da177e4 LT |
120 | #include <asm/cacheflush.h> |
121 | #include <asm/tlbflush.h> | |
122 | #include <asm/page.h> | |
123 | ||
124 | /* | |
50953fe9 | 125 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
126 | * 0 for faster, smaller code (especially in the critical paths). |
127 | * | |
128 | * STATS - 1 to collect stats for /proc/slabinfo. | |
129 | * 0 for faster, smaller code (especially in the critical paths). | |
130 | * | |
131 | * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible) | |
132 | */ | |
133 | ||
134 | #ifdef CONFIG_DEBUG_SLAB | |
135 | #define DEBUG 1 | |
136 | #define STATS 1 | |
137 | #define FORCED_DEBUG 1 | |
138 | #else | |
139 | #define DEBUG 0 | |
140 | #define STATS 0 | |
141 | #define FORCED_DEBUG 0 | |
142 | #endif | |
143 | ||
1da177e4 LT |
144 | /* Shouldn't this be in a header file somewhere? */ |
145 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 146 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 147 | |
1da177e4 LT |
148 | #ifndef ARCH_KMALLOC_FLAGS |
149 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
150 | #endif | |
151 | ||
152 | /* Legal flag mask for kmem_cache_create(). */ | |
153 | #if DEBUG | |
50953fe9 | 154 | # define CREATE_MASK (SLAB_RED_ZONE | \ |
1da177e4 | 155 | SLAB_POISON | SLAB_HWCACHE_ALIGN | \ |
ac2b898c | 156 | SLAB_CACHE_DMA | \ |
5af60839 | 157 | SLAB_STORE_USER | \ |
1da177e4 | 158 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
3ac7fe5a | 159 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ |
c175eea4 | 160 | SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) |
1da177e4 | 161 | #else |
ac2b898c | 162 | # define CREATE_MASK (SLAB_HWCACHE_ALIGN | \ |
5af60839 | 163 | SLAB_CACHE_DMA | \ |
1da177e4 | 164 | SLAB_RECLAIM_ACCOUNT | SLAB_PANIC | \ |
3ac7fe5a | 165 | SLAB_DESTROY_BY_RCU | SLAB_MEM_SPREAD | \ |
c175eea4 | 166 | SLAB_DEBUG_OBJECTS | SLAB_NOLEAKTRACE | SLAB_NOTRACK) |
1da177e4 LT |
167 | #endif |
168 | ||
169 | /* | |
170 | * kmem_bufctl_t: | |
171 | * | |
172 | * Bufctl's are used for linking objs within a slab | |
173 | * linked offsets. | |
174 | * | |
175 | * This implementation relies on "struct page" for locating the cache & | |
176 | * slab an object belongs to. | |
177 | * This allows the bufctl structure to be small (one int), but limits | |
178 | * the number of objects a slab (not a cache) can contain when off-slab | |
179 | * bufctls are used. The limit is the size of the largest general cache | |
180 | * that does not use off-slab slabs. | |
181 | * For 32bit archs with 4 kB pages, is this 56. | |
182 | * This is not serious, as it is only for large objects, when it is unwise | |
183 | * to have too many per slab. | |
184 | * Note: This limit can be raised by introducing a general cache whose size | |
185 | * is less than 512 (PAGE_SIZE<<3), but greater than 256. | |
186 | */ | |
187 | ||
fa5b08d5 | 188 | typedef unsigned int kmem_bufctl_t; |
1da177e4 LT |
189 | #define BUFCTL_END (((kmem_bufctl_t)(~0U))-0) |
190 | #define BUFCTL_FREE (((kmem_bufctl_t)(~0U))-1) | |
871751e2 AV |
191 | #define BUFCTL_ACTIVE (((kmem_bufctl_t)(~0U))-2) |
192 | #define SLAB_LIMIT (((kmem_bufctl_t)(~0U))-3) | |
1da177e4 | 193 | |
1da177e4 LT |
194 | /* |
195 | * struct slab | |
196 | * | |
197 | * Manages the objs in a slab. Placed either at the beginning of mem allocated | |
198 | * for a slab, or allocated from an general cache. | |
199 | * Slabs are chained into three list: fully used, partial, fully free slabs. | |
200 | */ | |
201 | struct slab { | |
b28a02de PE |
202 | struct list_head list; |
203 | unsigned long colouroff; | |
204 | void *s_mem; /* including colour offset */ | |
205 | unsigned int inuse; /* num of objs active in slab */ | |
206 | kmem_bufctl_t free; | |
207 | unsigned short nodeid; | |
1da177e4 LT |
208 | }; |
209 | ||
210 | /* | |
211 | * struct slab_rcu | |
212 | * | |
213 | * slab_destroy on a SLAB_DESTROY_BY_RCU cache uses this structure to | |
214 | * arrange for kmem_freepages to be called via RCU. This is useful if | |
215 | * we need to approach a kernel structure obliquely, from its address | |
216 | * obtained without the usual locking. We can lock the structure to | |
217 | * stabilize it and check it's still at the given address, only if we | |
218 | * can be sure that the memory has not been meanwhile reused for some | |
219 | * other kind of object (which our subsystem's lock might corrupt). | |
220 | * | |
221 | * rcu_read_lock before reading the address, then rcu_read_unlock after | |
222 | * taking the spinlock within the structure expected at that address. | |
223 | * | |
224 | * We assume struct slab_rcu can overlay struct slab when destroying. | |
225 | */ | |
226 | struct slab_rcu { | |
b28a02de | 227 | struct rcu_head head; |
343e0d7a | 228 | struct kmem_cache *cachep; |
b28a02de | 229 | void *addr; |
1da177e4 LT |
230 | }; |
231 | ||
232 | /* | |
233 | * struct array_cache | |
234 | * | |
1da177e4 LT |
235 | * Purpose: |
236 | * - LIFO ordering, to hand out cache-warm objects from _alloc | |
237 | * - reduce the number of linked list operations | |
238 | * - reduce spinlock operations | |
239 | * | |
240 | * The limit is stored in the per-cpu structure to reduce the data cache | |
241 | * footprint. | |
242 | * | |
243 | */ | |
244 | struct array_cache { | |
245 | unsigned int avail; | |
246 | unsigned int limit; | |
247 | unsigned int batchcount; | |
248 | unsigned int touched; | |
e498be7d | 249 | spinlock_t lock; |
bda5b655 | 250 | void *entry[]; /* |
a737b3e2 AM |
251 | * Must have this definition in here for the proper |
252 | * alignment of array_cache. Also simplifies accessing | |
253 | * the entries. | |
a737b3e2 | 254 | */ |
1da177e4 LT |
255 | }; |
256 | ||
a737b3e2 AM |
257 | /* |
258 | * bootstrap: The caches do not work without cpuarrays anymore, but the | |
259 | * cpuarrays are allocated from the generic caches... | |
1da177e4 LT |
260 | */ |
261 | #define BOOT_CPUCACHE_ENTRIES 1 | |
262 | struct arraycache_init { | |
263 | struct array_cache cache; | |
b28a02de | 264 | void *entries[BOOT_CPUCACHE_ENTRIES]; |
1da177e4 LT |
265 | }; |
266 | ||
267 | /* | |
e498be7d | 268 | * The slab lists for all objects. |
1da177e4 LT |
269 | */ |
270 | struct kmem_list3 { | |
b28a02de PE |
271 | struct list_head slabs_partial; /* partial list first, better asm code */ |
272 | struct list_head slabs_full; | |
273 | struct list_head slabs_free; | |
274 | unsigned long free_objects; | |
b28a02de | 275 | unsigned int free_limit; |
2e1217cf | 276 | unsigned int colour_next; /* Per-node cache coloring */ |
b28a02de PE |
277 | spinlock_t list_lock; |
278 | struct array_cache *shared; /* shared per node */ | |
279 | struct array_cache **alien; /* on other nodes */ | |
35386e3b CL |
280 | unsigned long next_reap; /* updated without locking */ |
281 | int free_touched; /* updated without locking */ | |
1da177e4 LT |
282 | }; |
283 | ||
e498be7d CL |
284 | /* |
285 | * Need this for bootstrapping a per node allocator. | |
286 | */ | |
556a169d | 287 | #define NUM_INIT_LISTS (3 * MAX_NUMNODES) |
e498be7d CL |
288 | struct kmem_list3 __initdata initkmem_list3[NUM_INIT_LISTS]; |
289 | #define CACHE_CACHE 0 | |
556a169d PE |
290 | #define SIZE_AC MAX_NUMNODES |
291 | #define SIZE_L3 (2 * MAX_NUMNODES) | |
e498be7d | 292 | |
ed11d9eb CL |
293 | static int drain_freelist(struct kmem_cache *cache, |
294 | struct kmem_list3 *l3, int tofree); | |
295 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, | |
296 | int node); | |
83b519e8 | 297 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 298 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 299 | |
e498be7d | 300 | /* |
a737b3e2 AM |
301 | * This function must be completely optimized away if a constant is passed to |
302 | * it. Mostly the same as what is in linux/slab.h except it returns an index. | |
e498be7d | 303 | */ |
7243cc05 | 304 | static __always_inline int index_of(const size_t size) |
e498be7d | 305 | { |
5ec8a847 SR |
306 | extern void __bad_size(void); |
307 | ||
e498be7d CL |
308 | if (__builtin_constant_p(size)) { |
309 | int i = 0; | |
310 | ||
311 | #define CACHE(x) \ | |
312 | if (size <=x) \ | |
313 | return i; \ | |
314 | else \ | |
315 | i++; | |
1c61fc40 | 316 | #include <linux/kmalloc_sizes.h> |
e498be7d | 317 | #undef CACHE |
5ec8a847 | 318 | __bad_size(); |
7243cc05 | 319 | } else |
5ec8a847 | 320 | __bad_size(); |
e498be7d CL |
321 | return 0; |
322 | } | |
323 | ||
e0a42726 IM |
324 | static int slab_early_init = 1; |
325 | ||
e498be7d CL |
326 | #define INDEX_AC index_of(sizeof(struct arraycache_init)) |
327 | #define INDEX_L3 index_of(sizeof(struct kmem_list3)) | |
1da177e4 | 328 | |
5295a74c | 329 | static void kmem_list3_init(struct kmem_list3 *parent) |
e498be7d CL |
330 | { |
331 | INIT_LIST_HEAD(&parent->slabs_full); | |
332 | INIT_LIST_HEAD(&parent->slabs_partial); | |
333 | INIT_LIST_HEAD(&parent->slabs_free); | |
334 | parent->shared = NULL; | |
335 | parent->alien = NULL; | |
2e1217cf | 336 | parent->colour_next = 0; |
e498be7d CL |
337 | spin_lock_init(&parent->list_lock); |
338 | parent->free_objects = 0; | |
339 | parent->free_touched = 0; | |
340 | } | |
341 | ||
a737b3e2 AM |
342 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
343 | do { \ | |
344 | INIT_LIST_HEAD(listp); \ | |
345 | list_splice(&(cachep->nodelists[nodeid]->slab), listp); \ | |
e498be7d CL |
346 | } while (0) |
347 | ||
a737b3e2 AM |
348 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
349 | do { \ | |
e498be7d CL |
350 | MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \ |
351 | MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \ | |
352 | MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \ | |
353 | } while (0) | |
1da177e4 | 354 | |
1da177e4 LT |
355 | #define CFLGS_OFF_SLAB (0x80000000UL) |
356 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) | |
357 | ||
358 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
359 | /* |
360 | * Optimization question: fewer reaps means less probability for unnessary | |
361 | * cpucache drain/refill cycles. | |
1da177e4 | 362 | * |
dc6f3f27 | 363 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
364 | * which could lock up otherwise freeable slabs. |
365 | */ | |
366 | #define REAPTIMEOUT_CPUC (2*HZ) | |
367 | #define REAPTIMEOUT_LIST3 (4*HZ) | |
368 | ||
369 | #if STATS | |
370 | #define STATS_INC_ACTIVE(x) ((x)->num_active++) | |
371 | #define STATS_DEC_ACTIVE(x) ((x)->num_active--) | |
372 | #define STATS_INC_ALLOCED(x) ((x)->num_allocations++) | |
373 | #define STATS_INC_GROWN(x) ((x)->grown++) | |
ed11d9eb | 374 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
375 | #define STATS_SET_HIGH(x) \ |
376 | do { \ | |
377 | if ((x)->num_active > (x)->high_mark) \ | |
378 | (x)->high_mark = (x)->num_active; \ | |
379 | } while (0) | |
1da177e4 LT |
380 | #define STATS_INC_ERR(x) ((x)->errors++) |
381 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 382 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 383 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
384 | #define STATS_SET_FREEABLE(x, i) \ |
385 | do { \ | |
386 | if ((x)->max_freeable < i) \ | |
387 | (x)->max_freeable = i; \ | |
388 | } while (0) | |
1da177e4 LT |
389 | #define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit) |
390 | #define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss) | |
391 | #define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit) | |
392 | #define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss) | |
393 | #else | |
394 | #define STATS_INC_ACTIVE(x) do { } while (0) | |
395 | #define STATS_DEC_ACTIVE(x) do { } while (0) | |
396 | #define STATS_INC_ALLOCED(x) do { } while (0) | |
397 | #define STATS_INC_GROWN(x) do { } while (0) | |
ed11d9eb | 398 | #define STATS_ADD_REAPED(x,y) do { } while (0) |
1da177e4 LT |
399 | #define STATS_SET_HIGH(x) do { } while (0) |
400 | #define STATS_INC_ERR(x) do { } while (0) | |
401 | #define STATS_INC_NODEALLOCS(x) do { } while (0) | |
e498be7d | 402 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 403 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 404 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
405 | #define STATS_INC_ALLOCHIT(x) do { } while (0) |
406 | #define STATS_INC_ALLOCMISS(x) do { } while (0) | |
407 | #define STATS_INC_FREEHIT(x) do { } while (0) | |
408 | #define STATS_INC_FREEMISS(x) do { } while (0) | |
409 | #endif | |
410 | ||
411 | #if DEBUG | |
1da177e4 | 412 | |
a737b3e2 AM |
413 | /* |
414 | * memory layout of objects: | |
1da177e4 | 415 | * 0 : objp |
3dafccf2 | 416 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
417 | * the end of an object is aligned with the end of the real |
418 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 419 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 420 | * redzone word. |
3dafccf2 MS |
421 | * cachep->obj_offset: The real object. |
422 | * cachep->buffer_size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] | |
a737b3e2 AM |
423 | * cachep->buffer_size - 1* BYTES_PER_WORD: last caller address |
424 | * [BYTES_PER_WORD long] | |
1da177e4 | 425 | */ |
343e0d7a | 426 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 427 | { |
3dafccf2 | 428 | return cachep->obj_offset; |
1da177e4 LT |
429 | } |
430 | ||
343e0d7a | 431 | static int obj_size(struct kmem_cache *cachep) |
1da177e4 | 432 | { |
3dafccf2 | 433 | return cachep->obj_size; |
1da177e4 LT |
434 | } |
435 | ||
b46b8f19 | 436 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
437 | { |
438 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
439 | return (unsigned long long*) (objp + obj_offset(cachep) - |
440 | sizeof(unsigned long long)); | |
1da177e4 LT |
441 | } |
442 | ||
b46b8f19 | 443 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
444 | { |
445 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
446 | if (cachep->flags & SLAB_STORE_USER) | |
b46b8f19 DW |
447 | return (unsigned long long *)(objp + cachep->buffer_size - |
448 | sizeof(unsigned long long) - | |
87a927c7 | 449 | REDZONE_ALIGN); |
b46b8f19 DW |
450 | return (unsigned long long *) (objp + cachep->buffer_size - |
451 | sizeof(unsigned long long)); | |
1da177e4 LT |
452 | } |
453 | ||
343e0d7a | 454 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
455 | { |
456 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3dafccf2 | 457 | return (void **)(objp + cachep->buffer_size - BYTES_PER_WORD); |
1da177e4 LT |
458 | } |
459 | ||
460 | #else | |
461 | ||
3dafccf2 MS |
462 | #define obj_offset(x) 0 |
463 | #define obj_size(cachep) (cachep->buffer_size) | |
b46b8f19 DW |
464 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
465 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
466 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
467 | ||
468 | #endif | |
469 | ||
0f24f128 | 470 | #ifdef CONFIG_TRACING |
36555751 EGM |
471 | size_t slab_buffer_size(struct kmem_cache *cachep) |
472 | { | |
473 | return cachep->buffer_size; | |
474 | } | |
475 | EXPORT_SYMBOL(slab_buffer_size); | |
476 | #endif | |
477 | ||
1da177e4 LT |
478 | /* |
479 | * Do not go above this order unless 0 objects fit into the slab. | |
480 | */ | |
481 | #define BREAK_GFP_ORDER_HI 1 | |
482 | #define BREAK_GFP_ORDER_LO 0 | |
483 | static int slab_break_gfp_order = BREAK_GFP_ORDER_LO; | |
484 | ||
a737b3e2 AM |
485 | /* |
486 | * Functions for storing/retrieving the cachep and or slab from the page | |
487 | * allocator. These are used to find the slab an obj belongs to. With kfree(), | |
488 | * these are used to find the cache which an obj belongs to. | |
1da177e4 | 489 | */ |
065d41cb PE |
490 | static inline void page_set_cache(struct page *page, struct kmem_cache *cache) |
491 | { | |
492 | page->lru.next = (struct list_head *)cache; | |
493 | } | |
494 | ||
495 | static inline struct kmem_cache *page_get_cache(struct page *page) | |
496 | { | |
d85f3385 | 497 | page = compound_head(page); |
ddc2e812 | 498 | BUG_ON(!PageSlab(page)); |
065d41cb PE |
499 | return (struct kmem_cache *)page->lru.next; |
500 | } | |
501 | ||
502 | static inline void page_set_slab(struct page *page, struct slab *slab) | |
503 | { | |
504 | page->lru.prev = (struct list_head *)slab; | |
505 | } | |
506 | ||
507 | static inline struct slab *page_get_slab(struct page *page) | |
508 | { | |
ddc2e812 | 509 | BUG_ON(!PageSlab(page)); |
065d41cb PE |
510 | return (struct slab *)page->lru.prev; |
511 | } | |
1da177e4 | 512 | |
6ed5eb22 PE |
513 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
514 | { | |
b49af68f | 515 | struct page *page = virt_to_head_page(obj); |
6ed5eb22 PE |
516 | return page_get_cache(page); |
517 | } | |
518 | ||
519 | static inline struct slab *virt_to_slab(const void *obj) | |
520 | { | |
b49af68f | 521 | struct page *page = virt_to_head_page(obj); |
6ed5eb22 PE |
522 | return page_get_slab(page); |
523 | } | |
524 | ||
8fea4e96 PE |
525 | static inline void *index_to_obj(struct kmem_cache *cache, struct slab *slab, |
526 | unsigned int idx) | |
527 | { | |
528 | return slab->s_mem + cache->buffer_size * idx; | |
529 | } | |
530 | ||
6a2d7a95 ED |
531 | /* |
532 | * We want to avoid an expensive divide : (offset / cache->buffer_size) | |
533 | * Using the fact that buffer_size is a constant for a particular cache, | |
534 | * we can replace (offset / cache->buffer_size) by | |
535 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) | |
536 | */ | |
537 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
538 | const struct slab *slab, void *obj) | |
8fea4e96 | 539 | { |
6a2d7a95 ED |
540 | u32 offset = (obj - slab->s_mem); |
541 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); | |
8fea4e96 PE |
542 | } |
543 | ||
a737b3e2 AM |
544 | /* |
545 | * These are the default caches for kmalloc. Custom caches can have other sizes. | |
546 | */ | |
1da177e4 LT |
547 | struct cache_sizes malloc_sizes[] = { |
548 | #define CACHE(x) { .cs_size = (x) }, | |
549 | #include <linux/kmalloc_sizes.h> | |
550 | CACHE(ULONG_MAX) | |
551 | #undef CACHE | |
552 | }; | |
553 | EXPORT_SYMBOL(malloc_sizes); | |
554 | ||
555 | /* Must match cache_sizes above. Out of line to keep cache footprint low. */ | |
556 | struct cache_names { | |
557 | char *name; | |
558 | char *name_dma; | |
559 | }; | |
560 | ||
561 | static struct cache_names __initdata cache_names[] = { | |
562 | #define CACHE(x) { .name = "size-" #x, .name_dma = "size-" #x "(DMA)" }, | |
563 | #include <linux/kmalloc_sizes.h> | |
b28a02de | 564 | {NULL,} |
1da177e4 LT |
565 | #undef CACHE |
566 | }; | |
567 | ||
568 | static struct arraycache_init initarray_cache __initdata = | |
b28a02de | 569 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 | 570 | static struct arraycache_init initarray_generic = |
b28a02de | 571 | { {0, BOOT_CPUCACHE_ENTRIES, 1, 0} }; |
1da177e4 LT |
572 | |
573 | /* internal cache of cache description objs */ | |
343e0d7a | 574 | static struct kmem_cache cache_cache = { |
b28a02de PE |
575 | .batchcount = 1, |
576 | .limit = BOOT_CPUCACHE_ENTRIES, | |
577 | .shared = 1, | |
343e0d7a | 578 | .buffer_size = sizeof(struct kmem_cache), |
b28a02de | 579 | .name = "kmem_cache", |
1da177e4 LT |
580 | }; |
581 | ||
056c6241 RT |
582 | #define BAD_ALIEN_MAGIC 0x01020304ul |
583 | ||
ce79ddc8 PE |
584 | /* |
585 | * chicken and egg problem: delay the per-cpu array allocation | |
586 | * until the general caches are up. | |
587 | */ | |
588 | static enum { | |
589 | NONE, | |
590 | PARTIAL_AC, | |
591 | PARTIAL_L3, | |
592 | EARLY, | |
593 | FULL | |
594 | } g_cpucache_up; | |
595 | ||
596 | /* | |
597 | * used by boot code to determine if it can use slab based allocator | |
598 | */ | |
599 | int slab_is_available(void) | |
600 | { | |
601 | return g_cpucache_up >= EARLY; | |
602 | } | |
603 | ||
f1aaee53 AV |
604 | #ifdef CONFIG_LOCKDEP |
605 | ||
606 | /* | |
607 | * Slab sometimes uses the kmalloc slabs to store the slab headers | |
608 | * for other slabs "off slab". | |
609 | * The locking for this is tricky in that it nests within the locks | |
610 | * of all other slabs in a few places; to deal with this special | |
611 | * locking we put on-slab caches into a separate lock-class. | |
056c6241 RT |
612 | * |
613 | * We set lock class for alien array caches which are up during init. | |
614 | * The lock annotation will be lost if all cpus of a node goes down and | |
615 | * then comes back up during hotplug | |
f1aaee53 | 616 | */ |
056c6241 RT |
617 | static struct lock_class_key on_slab_l3_key; |
618 | static struct lock_class_key on_slab_alc_key; | |
619 | ||
ce79ddc8 | 620 | static void init_node_lock_keys(int q) |
f1aaee53 | 621 | { |
056c6241 RT |
622 | struct cache_sizes *s = malloc_sizes; |
623 | ||
ce79ddc8 PE |
624 | if (g_cpucache_up != FULL) |
625 | return; | |
626 | ||
627 | for (s = malloc_sizes; s->cs_size != ULONG_MAX; s++) { | |
628 | struct array_cache **alc; | |
629 | struct kmem_list3 *l3; | |
630 | int r; | |
631 | ||
632 | l3 = s->cs_cachep->nodelists[q]; | |
633 | if (!l3 || OFF_SLAB(s->cs_cachep)) | |
00afa758 | 634 | continue; |
ce79ddc8 PE |
635 | lockdep_set_class(&l3->list_lock, &on_slab_l3_key); |
636 | alc = l3->alien; | |
637 | /* | |
638 | * FIXME: This check for BAD_ALIEN_MAGIC | |
639 | * should go away when common slab code is taught to | |
640 | * work even without alien caches. | |
641 | * Currently, non NUMA code returns BAD_ALIEN_MAGIC | |
642 | * for alloc_alien_cache, | |
643 | */ | |
644 | if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC) | |
00afa758 | 645 | continue; |
ce79ddc8 PE |
646 | for_each_node(r) { |
647 | if (alc[r]) | |
648 | lockdep_set_class(&alc[r]->lock, | |
649 | &on_slab_alc_key); | |
056c6241 | 650 | } |
f1aaee53 AV |
651 | } |
652 | } | |
ce79ddc8 PE |
653 | |
654 | static inline void init_lock_keys(void) | |
655 | { | |
656 | int node; | |
657 | ||
658 | for_each_node(node) | |
659 | init_node_lock_keys(node); | |
660 | } | |
f1aaee53 | 661 | #else |
ce79ddc8 PE |
662 | static void init_node_lock_keys(int q) |
663 | { | |
664 | } | |
665 | ||
056c6241 | 666 | static inline void init_lock_keys(void) |
f1aaee53 AV |
667 | { |
668 | } | |
669 | #endif | |
670 | ||
8f5be20b | 671 | /* |
95402b38 | 672 | * Guard access to the cache-chain. |
8f5be20b | 673 | */ |
fc0abb14 | 674 | static DEFINE_MUTEX(cache_chain_mutex); |
1da177e4 LT |
675 | static struct list_head cache_chain; |
676 | ||
1871e52c | 677 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 678 | |
343e0d7a | 679 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 LT |
680 | { |
681 | return cachep->array[smp_processor_id()]; | |
682 | } | |
683 | ||
a737b3e2 AM |
684 | static inline struct kmem_cache *__find_general_cachep(size_t size, |
685 | gfp_t gfpflags) | |
1da177e4 LT |
686 | { |
687 | struct cache_sizes *csizep = malloc_sizes; | |
688 | ||
689 | #if DEBUG | |
690 | /* This happens if someone tries to call | |
b28a02de PE |
691 | * kmem_cache_create(), or __kmalloc(), before |
692 | * the generic caches are initialized. | |
693 | */ | |
c7e43c78 | 694 | BUG_ON(malloc_sizes[INDEX_AC].cs_cachep == NULL); |
1da177e4 | 695 | #endif |
6cb8f913 CL |
696 | if (!size) |
697 | return ZERO_SIZE_PTR; | |
698 | ||
1da177e4 LT |
699 | while (size > csizep->cs_size) |
700 | csizep++; | |
701 | ||
702 | /* | |
0abf40c1 | 703 | * Really subtle: The last entry with cs->cs_size==ULONG_MAX |
1da177e4 LT |
704 | * has cs_{dma,}cachep==NULL. Thus no special case |
705 | * for large kmalloc calls required. | |
706 | */ | |
4b51d669 | 707 | #ifdef CONFIG_ZONE_DMA |
1da177e4 LT |
708 | if (unlikely(gfpflags & GFP_DMA)) |
709 | return csizep->cs_dmacachep; | |
4b51d669 | 710 | #endif |
1da177e4 LT |
711 | return csizep->cs_cachep; |
712 | } | |
713 | ||
b221385b | 714 | static struct kmem_cache *kmem_find_general_cachep(size_t size, gfp_t gfpflags) |
97e2bde4 MS |
715 | { |
716 | return __find_general_cachep(size, gfpflags); | |
717 | } | |
97e2bde4 | 718 | |
fbaccacf | 719 | static size_t slab_mgmt_size(size_t nr_objs, size_t align) |
1da177e4 | 720 | { |
fbaccacf SR |
721 | return ALIGN(sizeof(struct slab)+nr_objs*sizeof(kmem_bufctl_t), align); |
722 | } | |
1da177e4 | 723 | |
a737b3e2 AM |
724 | /* |
725 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
726 | */ | |
fbaccacf SR |
727 | static void cache_estimate(unsigned long gfporder, size_t buffer_size, |
728 | size_t align, int flags, size_t *left_over, | |
729 | unsigned int *num) | |
730 | { | |
731 | int nr_objs; | |
732 | size_t mgmt_size; | |
733 | size_t slab_size = PAGE_SIZE << gfporder; | |
1da177e4 | 734 | |
fbaccacf SR |
735 | /* |
736 | * The slab management structure can be either off the slab or | |
737 | * on it. For the latter case, the memory allocated for a | |
738 | * slab is used for: | |
739 | * | |
740 | * - The struct slab | |
741 | * - One kmem_bufctl_t for each object | |
742 | * - Padding to respect alignment of @align | |
743 | * - @buffer_size bytes for each object | |
744 | * | |
745 | * If the slab management structure is off the slab, then the | |
746 | * alignment will already be calculated into the size. Because | |
747 | * the slabs are all pages aligned, the objects will be at the | |
748 | * correct alignment when allocated. | |
749 | */ | |
750 | if (flags & CFLGS_OFF_SLAB) { | |
751 | mgmt_size = 0; | |
752 | nr_objs = slab_size / buffer_size; | |
753 | ||
754 | if (nr_objs > SLAB_LIMIT) | |
755 | nr_objs = SLAB_LIMIT; | |
756 | } else { | |
757 | /* | |
758 | * Ignore padding for the initial guess. The padding | |
759 | * is at most @align-1 bytes, and @buffer_size is at | |
760 | * least @align. In the worst case, this result will | |
761 | * be one greater than the number of objects that fit | |
762 | * into the memory allocation when taking the padding | |
763 | * into account. | |
764 | */ | |
765 | nr_objs = (slab_size - sizeof(struct slab)) / | |
766 | (buffer_size + sizeof(kmem_bufctl_t)); | |
767 | ||
768 | /* | |
769 | * This calculated number will be either the right | |
770 | * amount, or one greater than what we want. | |
771 | */ | |
772 | if (slab_mgmt_size(nr_objs, align) + nr_objs*buffer_size | |
773 | > slab_size) | |
774 | nr_objs--; | |
775 | ||
776 | if (nr_objs > SLAB_LIMIT) | |
777 | nr_objs = SLAB_LIMIT; | |
778 | ||
779 | mgmt_size = slab_mgmt_size(nr_objs, align); | |
780 | } | |
781 | *num = nr_objs; | |
782 | *left_over = slab_size - nr_objs*buffer_size - mgmt_size; | |
1da177e4 LT |
783 | } |
784 | ||
d40cee24 | 785 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 786 | |
a737b3e2 AM |
787 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
788 | char *msg) | |
1da177e4 LT |
789 | { |
790 | printk(KERN_ERR "slab error in %s(): cache `%s': %s\n", | |
b28a02de | 791 | function, cachep->name, msg); |
1da177e4 LT |
792 | dump_stack(); |
793 | } | |
794 | ||
3395ee05 PM |
795 | /* |
796 | * By default on NUMA we use alien caches to stage the freeing of | |
797 | * objects allocated from other nodes. This causes massive memory | |
798 | * inefficiencies when using fake NUMA setup to split memory into a | |
799 | * large number of small nodes, so it can be disabled on the command | |
800 | * line | |
801 | */ | |
802 | ||
803 | static int use_alien_caches __read_mostly = 1; | |
804 | static int __init noaliencache_setup(char *s) | |
805 | { | |
806 | use_alien_caches = 0; | |
807 | return 1; | |
808 | } | |
809 | __setup("noaliencache", noaliencache_setup); | |
810 | ||
8fce4d8e CL |
811 | #ifdef CONFIG_NUMA |
812 | /* | |
813 | * Special reaping functions for NUMA systems called from cache_reap(). | |
814 | * These take care of doing round robin flushing of alien caches (containing | |
815 | * objects freed on different nodes from which they were allocated) and the | |
816 | * flushing of remote pcps by calling drain_node_pages. | |
817 | */ | |
1871e52c | 818 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
819 | |
820 | static void init_reap_node(int cpu) | |
821 | { | |
822 | int node; | |
823 | ||
7d6e6d09 | 824 | node = next_node(cpu_to_mem(cpu), node_online_map); |
8fce4d8e | 825 | if (node == MAX_NUMNODES) |
442295c9 | 826 | node = first_node(node_online_map); |
8fce4d8e | 827 | |
1871e52c | 828 | per_cpu(slab_reap_node, cpu) = node; |
8fce4d8e CL |
829 | } |
830 | ||
831 | static void next_reap_node(void) | |
832 | { | |
1871e52c | 833 | int node = __get_cpu_var(slab_reap_node); |
8fce4d8e | 834 | |
8fce4d8e CL |
835 | node = next_node(node, node_online_map); |
836 | if (unlikely(node >= MAX_NUMNODES)) | |
837 | node = first_node(node_online_map); | |
1871e52c | 838 | __get_cpu_var(slab_reap_node) = node; |
8fce4d8e CL |
839 | } |
840 | ||
841 | #else | |
842 | #define init_reap_node(cpu) do { } while (0) | |
843 | #define next_reap_node(void) do { } while (0) | |
844 | #endif | |
845 | ||
1da177e4 LT |
846 | /* |
847 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
848 | * via the workqueue/eventd. | |
849 | * Add the CPU number into the expiration time to minimize the possibility of | |
850 | * the CPUs getting into lockstep and contending for the global cache chain | |
851 | * lock. | |
852 | */ | |
897e679b | 853 | static void __cpuinit start_cpu_timer(int cpu) |
1da177e4 | 854 | { |
1871e52c | 855 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 LT |
856 | |
857 | /* | |
858 | * When this gets called from do_initcalls via cpucache_init(), | |
859 | * init_workqueues() has already run, so keventd will be setup | |
860 | * at that time. | |
861 | */ | |
52bad64d | 862 | if (keventd_up() && reap_work->work.func == NULL) { |
8fce4d8e | 863 | init_reap_node(cpu); |
65f27f38 | 864 | INIT_DELAYED_WORK(reap_work, cache_reap); |
2b284214 AV |
865 | schedule_delayed_work_on(cpu, reap_work, |
866 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
867 | } |
868 | } | |
869 | ||
e498be7d | 870 | static struct array_cache *alloc_arraycache(int node, int entries, |
83b519e8 | 871 | int batchcount, gfp_t gfp) |
1da177e4 | 872 | { |
b28a02de | 873 | int memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1da177e4 LT |
874 | struct array_cache *nc = NULL; |
875 | ||
83b519e8 | 876 | nc = kmalloc_node(memsize, gfp, node); |
d5cff635 CM |
877 | /* |
878 | * The array_cache structures contain pointers to free object. | |
879 | * However, when such objects are allocated or transfered to another | |
880 | * cache the pointers are not cleared and they could be counted as | |
881 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
882 | * not scan such objects. | |
883 | */ | |
884 | kmemleak_no_scan(nc); | |
1da177e4 LT |
885 | if (nc) { |
886 | nc->avail = 0; | |
887 | nc->limit = entries; | |
888 | nc->batchcount = batchcount; | |
889 | nc->touched = 0; | |
e498be7d | 890 | spin_lock_init(&nc->lock); |
1da177e4 LT |
891 | } |
892 | return nc; | |
893 | } | |
894 | ||
3ded175a CL |
895 | /* |
896 | * Transfer objects in one arraycache to another. | |
897 | * Locking must be handled by the caller. | |
898 | * | |
899 | * Return the number of entries transferred. | |
900 | */ | |
901 | static int transfer_objects(struct array_cache *to, | |
902 | struct array_cache *from, unsigned int max) | |
903 | { | |
904 | /* Figure out how many entries to transfer */ | |
905 | int nr = min(min(from->avail, max), to->limit - to->avail); | |
906 | ||
907 | if (!nr) | |
908 | return 0; | |
909 | ||
910 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
911 | sizeof(void *) *nr); | |
912 | ||
913 | from->avail -= nr; | |
914 | to->avail += nr; | |
3ded175a CL |
915 | return nr; |
916 | } | |
917 | ||
765c4507 CL |
918 | #ifndef CONFIG_NUMA |
919 | ||
920 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
921 | #define reap_alien(cachep, l3) do { } while (0) | |
922 | ||
83b519e8 | 923 | static inline struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
765c4507 CL |
924 | { |
925 | return (struct array_cache **)BAD_ALIEN_MAGIC; | |
926 | } | |
927 | ||
928 | static inline void free_alien_cache(struct array_cache **ac_ptr) | |
929 | { | |
930 | } | |
931 | ||
932 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
933 | { | |
934 | return 0; | |
935 | } | |
936 | ||
937 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
938 | gfp_t flags) | |
939 | { | |
940 | return NULL; | |
941 | } | |
942 | ||
8b98c169 | 943 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
944 | gfp_t flags, int nodeid) |
945 | { | |
946 | return NULL; | |
947 | } | |
948 | ||
949 | #else /* CONFIG_NUMA */ | |
950 | ||
8b98c169 | 951 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 952 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 953 | |
83b519e8 | 954 | static struct array_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) |
e498be7d CL |
955 | { |
956 | struct array_cache **ac_ptr; | |
8ef82866 | 957 | int memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
958 | int i; |
959 | ||
960 | if (limit > 1) | |
961 | limit = 12; | |
f3186a9c | 962 | ac_ptr = kzalloc_node(memsize, gfp, node); |
e498be7d CL |
963 | if (ac_ptr) { |
964 | for_each_node(i) { | |
f3186a9c | 965 | if (i == node || !node_online(i)) |
e498be7d | 966 | continue; |
83b519e8 | 967 | ac_ptr[i] = alloc_arraycache(node, limit, 0xbaadf00d, gfp); |
e498be7d | 968 | if (!ac_ptr[i]) { |
cc550def | 969 | for (i--; i >= 0; i--) |
e498be7d CL |
970 | kfree(ac_ptr[i]); |
971 | kfree(ac_ptr); | |
972 | return NULL; | |
973 | } | |
974 | } | |
975 | } | |
976 | return ac_ptr; | |
977 | } | |
978 | ||
5295a74c | 979 | static void free_alien_cache(struct array_cache **ac_ptr) |
e498be7d CL |
980 | { |
981 | int i; | |
982 | ||
983 | if (!ac_ptr) | |
984 | return; | |
e498be7d | 985 | for_each_node(i) |
b28a02de | 986 | kfree(ac_ptr[i]); |
e498be7d CL |
987 | kfree(ac_ptr); |
988 | } | |
989 | ||
343e0d7a | 990 | static void __drain_alien_cache(struct kmem_cache *cachep, |
5295a74c | 991 | struct array_cache *ac, int node) |
e498be7d CL |
992 | { |
993 | struct kmem_list3 *rl3 = cachep->nodelists[node]; | |
994 | ||
995 | if (ac->avail) { | |
996 | spin_lock(&rl3->list_lock); | |
e00946fe CL |
997 | /* |
998 | * Stuff objects into the remote nodes shared array first. | |
999 | * That way we could avoid the overhead of putting the objects | |
1000 | * into the free lists and getting them back later. | |
1001 | */ | |
693f7d36 | 1002 | if (rl3->shared) |
1003 | transfer_objects(rl3->shared, ac, ac->limit); | |
e00946fe | 1004 | |
ff69416e | 1005 | free_block(cachep, ac->entry, ac->avail, node); |
e498be7d CL |
1006 | ac->avail = 0; |
1007 | spin_unlock(&rl3->list_lock); | |
1008 | } | |
1009 | } | |
1010 | ||
8fce4d8e CL |
1011 | /* |
1012 | * Called from cache_reap() to regularly drain alien caches round robin. | |
1013 | */ | |
1014 | static void reap_alien(struct kmem_cache *cachep, struct kmem_list3 *l3) | |
1015 | { | |
1871e52c | 1016 | int node = __get_cpu_var(slab_reap_node); |
8fce4d8e CL |
1017 | |
1018 | if (l3->alien) { | |
1019 | struct array_cache *ac = l3->alien[node]; | |
e00946fe CL |
1020 | |
1021 | if (ac && ac->avail && spin_trylock_irq(&ac->lock)) { | |
8fce4d8e CL |
1022 | __drain_alien_cache(cachep, ac, node); |
1023 | spin_unlock_irq(&ac->lock); | |
1024 | } | |
1025 | } | |
1026 | } | |
1027 | ||
a737b3e2 AM |
1028 | static void drain_alien_cache(struct kmem_cache *cachep, |
1029 | struct array_cache **alien) | |
e498be7d | 1030 | { |
b28a02de | 1031 | int i = 0; |
e498be7d CL |
1032 | struct array_cache *ac; |
1033 | unsigned long flags; | |
1034 | ||
1035 | for_each_online_node(i) { | |
4484ebf1 | 1036 | ac = alien[i]; |
e498be7d CL |
1037 | if (ac) { |
1038 | spin_lock_irqsave(&ac->lock, flags); | |
1039 | __drain_alien_cache(cachep, ac, i); | |
1040 | spin_unlock_irqrestore(&ac->lock, flags); | |
1041 | } | |
1042 | } | |
1043 | } | |
729bd0b7 | 1044 | |
873623df | 1045 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) |
729bd0b7 PE |
1046 | { |
1047 | struct slab *slabp = virt_to_slab(objp); | |
1048 | int nodeid = slabp->nodeid; | |
1049 | struct kmem_list3 *l3; | |
1050 | struct array_cache *alien = NULL; | |
1ca4cb24 PE |
1051 | int node; |
1052 | ||
7d6e6d09 | 1053 | node = numa_mem_id(); |
729bd0b7 PE |
1054 | |
1055 | /* | |
1056 | * Make sure we are not freeing a object from another node to the array | |
1057 | * cache on this cpu. | |
1058 | */ | |
62918a03 | 1059 | if (likely(slabp->nodeid == node)) |
729bd0b7 PE |
1060 | return 0; |
1061 | ||
1ca4cb24 | 1062 | l3 = cachep->nodelists[node]; |
729bd0b7 PE |
1063 | STATS_INC_NODEFREES(cachep); |
1064 | if (l3->alien && l3->alien[nodeid]) { | |
1065 | alien = l3->alien[nodeid]; | |
873623df | 1066 | spin_lock(&alien->lock); |
729bd0b7 PE |
1067 | if (unlikely(alien->avail == alien->limit)) { |
1068 | STATS_INC_ACOVERFLOW(cachep); | |
1069 | __drain_alien_cache(cachep, alien, nodeid); | |
1070 | } | |
1071 | alien->entry[alien->avail++] = objp; | |
1072 | spin_unlock(&alien->lock); | |
1073 | } else { | |
1074 | spin_lock(&(cachep->nodelists[nodeid])->list_lock); | |
1075 | free_block(cachep, &objp, 1, nodeid); | |
1076 | spin_unlock(&(cachep->nodelists[nodeid])->list_lock); | |
1077 | } | |
1078 | return 1; | |
1079 | } | |
e498be7d CL |
1080 | #endif |
1081 | ||
8f9f8d9e DR |
1082 | /* |
1083 | * Allocates and initializes nodelists for a node on each slab cache, used for | |
1084 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_list3 | |
1085 | * will be allocated off-node since memory is not yet online for the new node. | |
1086 | * When hotplugging memory or a cpu, existing nodelists are not replaced if | |
1087 | * already in use. | |
1088 | * | |
1089 | * Must hold cache_chain_mutex. | |
1090 | */ | |
1091 | static int init_cache_nodelists_node(int node) | |
1092 | { | |
1093 | struct kmem_cache *cachep; | |
1094 | struct kmem_list3 *l3; | |
1095 | const int memsize = sizeof(struct kmem_list3); | |
1096 | ||
1097 | list_for_each_entry(cachep, &cache_chain, next) { | |
1098 | /* | |
1099 | * Set up the size64 kmemlist for cpu before we can | |
1100 | * begin anything. Make sure some other cpu on this | |
1101 | * node has not already allocated this | |
1102 | */ | |
1103 | if (!cachep->nodelists[node]) { | |
1104 | l3 = kmalloc_node(memsize, GFP_KERNEL, node); | |
1105 | if (!l3) | |
1106 | return -ENOMEM; | |
1107 | kmem_list3_init(l3); | |
1108 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
1109 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1110 | ||
1111 | /* | |
1112 | * The l3s don't come and go as CPUs come and | |
1113 | * go. cache_chain_mutex is sufficient | |
1114 | * protection here. | |
1115 | */ | |
1116 | cachep->nodelists[node] = l3; | |
1117 | } | |
1118 | ||
1119 | spin_lock_irq(&cachep->nodelists[node]->list_lock); | |
1120 | cachep->nodelists[node]->free_limit = | |
1121 | (1 + nr_cpus_node(node)) * | |
1122 | cachep->batchcount + cachep->num; | |
1123 | spin_unlock_irq(&cachep->nodelists[node]->list_lock); | |
1124 | } | |
1125 | return 0; | |
1126 | } | |
1127 | ||
fbf1e473 AM |
1128 | static void __cpuinit cpuup_canceled(long cpu) |
1129 | { | |
1130 | struct kmem_cache *cachep; | |
1131 | struct kmem_list3 *l3 = NULL; | |
7d6e6d09 | 1132 | int node = cpu_to_mem(cpu); |
a70f7302 | 1133 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 AM |
1134 | |
1135 | list_for_each_entry(cachep, &cache_chain, next) { | |
1136 | struct array_cache *nc; | |
1137 | struct array_cache *shared; | |
1138 | struct array_cache **alien; | |
fbf1e473 | 1139 | |
fbf1e473 AM |
1140 | /* cpu is dead; no one can alloc from it. */ |
1141 | nc = cachep->array[cpu]; | |
1142 | cachep->array[cpu] = NULL; | |
1143 | l3 = cachep->nodelists[node]; | |
1144 | ||
1145 | if (!l3) | |
1146 | goto free_array_cache; | |
1147 | ||
1148 | spin_lock_irq(&l3->list_lock); | |
1149 | ||
1150 | /* Free limit for this kmem_list3 */ | |
1151 | l3->free_limit -= cachep->batchcount; | |
1152 | if (nc) | |
1153 | free_block(cachep, nc->entry, nc->avail, node); | |
1154 | ||
58463c1f | 1155 | if (!cpumask_empty(mask)) { |
fbf1e473 AM |
1156 | spin_unlock_irq(&l3->list_lock); |
1157 | goto free_array_cache; | |
1158 | } | |
1159 | ||
1160 | shared = l3->shared; | |
1161 | if (shared) { | |
1162 | free_block(cachep, shared->entry, | |
1163 | shared->avail, node); | |
1164 | l3->shared = NULL; | |
1165 | } | |
1166 | ||
1167 | alien = l3->alien; | |
1168 | l3->alien = NULL; | |
1169 | ||
1170 | spin_unlock_irq(&l3->list_lock); | |
1171 | ||
1172 | kfree(shared); | |
1173 | if (alien) { | |
1174 | drain_alien_cache(cachep, alien); | |
1175 | free_alien_cache(alien); | |
1176 | } | |
1177 | free_array_cache: | |
1178 | kfree(nc); | |
1179 | } | |
1180 | /* | |
1181 | * In the previous loop, all the objects were freed to | |
1182 | * the respective cache's slabs, now we can go ahead and | |
1183 | * shrink each nodelist to its limit. | |
1184 | */ | |
1185 | list_for_each_entry(cachep, &cache_chain, next) { | |
1186 | l3 = cachep->nodelists[node]; | |
1187 | if (!l3) | |
1188 | continue; | |
1189 | drain_freelist(cachep, l3, l3->free_objects); | |
1190 | } | |
1191 | } | |
1192 | ||
1193 | static int __cpuinit cpuup_prepare(long cpu) | |
1da177e4 | 1194 | { |
343e0d7a | 1195 | struct kmem_cache *cachep; |
e498be7d | 1196 | struct kmem_list3 *l3 = NULL; |
7d6e6d09 | 1197 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1198 | int err; |
1da177e4 | 1199 | |
fbf1e473 AM |
1200 | /* |
1201 | * We need to do this right in the beginning since | |
1202 | * alloc_arraycache's are going to use this list. | |
1203 | * kmalloc_node allows us to add the slab to the right | |
1204 | * kmem_list3 and not this cpu's kmem_list3 | |
1205 | */ | |
8f9f8d9e DR |
1206 | err = init_cache_nodelists_node(node); |
1207 | if (err < 0) | |
1208 | goto bad; | |
fbf1e473 AM |
1209 | |
1210 | /* | |
1211 | * Now we can go ahead with allocating the shared arrays and | |
1212 | * array caches | |
1213 | */ | |
1214 | list_for_each_entry(cachep, &cache_chain, next) { | |
1215 | struct array_cache *nc; | |
1216 | struct array_cache *shared = NULL; | |
1217 | struct array_cache **alien = NULL; | |
1218 | ||
1219 | nc = alloc_arraycache(node, cachep->limit, | |
83b519e8 | 1220 | cachep->batchcount, GFP_KERNEL); |
fbf1e473 AM |
1221 | if (!nc) |
1222 | goto bad; | |
1223 | if (cachep->shared) { | |
1224 | shared = alloc_arraycache(node, | |
1225 | cachep->shared * cachep->batchcount, | |
83b519e8 | 1226 | 0xbaadf00d, GFP_KERNEL); |
12d00f6a AM |
1227 | if (!shared) { |
1228 | kfree(nc); | |
1da177e4 | 1229 | goto bad; |
12d00f6a | 1230 | } |
fbf1e473 AM |
1231 | } |
1232 | if (use_alien_caches) { | |
83b519e8 | 1233 | alien = alloc_alien_cache(node, cachep->limit, GFP_KERNEL); |
12d00f6a AM |
1234 | if (!alien) { |
1235 | kfree(shared); | |
1236 | kfree(nc); | |
fbf1e473 | 1237 | goto bad; |
12d00f6a | 1238 | } |
fbf1e473 AM |
1239 | } |
1240 | cachep->array[cpu] = nc; | |
1241 | l3 = cachep->nodelists[node]; | |
1242 | BUG_ON(!l3); | |
1243 | ||
1244 | spin_lock_irq(&l3->list_lock); | |
1245 | if (!l3->shared) { | |
1246 | /* | |
1247 | * We are serialised from CPU_DEAD or | |
1248 | * CPU_UP_CANCELLED by the cpucontrol lock | |
1249 | */ | |
1250 | l3->shared = shared; | |
1251 | shared = NULL; | |
1252 | } | |
4484ebf1 | 1253 | #ifdef CONFIG_NUMA |
fbf1e473 AM |
1254 | if (!l3->alien) { |
1255 | l3->alien = alien; | |
1256 | alien = NULL; | |
1da177e4 | 1257 | } |
fbf1e473 AM |
1258 | #endif |
1259 | spin_unlock_irq(&l3->list_lock); | |
1260 | kfree(shared); | |
1261 | free_alien_cache(alien); | |
1262 | } | |
ce79ddc8 PE |
1263 | init_node_lock_keys(node); |
1264 | ||
fbf1e473 AM |
1265 | return 0; |
1266 | bad: | |
12d00f6a | 1267 | cpuup_canceled(cpu); |
fbf1e473 AM |
1268 | return -ENOMEM; |
1269 | } | |
1270 | ||
1271 | static int __cpuinit cpuup_callback(struct notifier_block *nfb, | |
1272 | unsigned long action, void *hcpu) | |
1273 | { | |
1274 | long cpu = (long)hcpu; | |
1275 | int err = 0; | |
1276 | ||
1277 | switch (action) { | |
fbf1e473 AM |
1278 | case CPU_UP_PREPARE: |
1279 | case CPU_UP_PREPARE_FROZEN: | |
95402b38 | 1280 | mutex_lock(&cache_chain_mutex); |
fbf1e473 | 1281 | err = cpuup_prepare(cpu); |
95402b38 | 1282 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
1283 | break; |
1284 | case CPU_ONLINE: | |
8bb78442 | 1285 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
1286 | start_cpu_timer(cpu); |
1287 | break; | |
1288 | #ifdef CONFIG_HOTPLUG_CPU | |
5830c590 | 1289 | case CPU_DOWN_PREPARE: |
8bb78442 | 1290 | case CPU_DOWN_PREPARE_FROZEN: |
5830c590 CL |
1291 | /* |
1292 | * Shutdown cache reaper. Note that the cache_chain_mutex is | |
1293 | * held so that if cache_reap() is invoked it cannot do | |
1294 | * anything expensive but will only modify reap_work | |
1295 | * and reschedule the timer. | |
1296 | */ | |
1871e52c | 1297 | cancel_rearming_delayed_work(&per_cpu(slab_reap_work, cpu)); |
5830c590 | 1298 | /* Now the cache_reaper is guaranteed to be not running. */ |
1871e52c | 1299 | per_cpu(slab_reap_work, cpu).work.func = NULL; |
5830c590 CL |
1300 | break; |
1301 | case CPU_DOWN_FAILED: | |
8bb78442 | 1302 | case CPU_DOWN_FAILED_FROZEN: |
5830c590 CL |
1303 | start_cpu_timer(cpu); |
1304 | break; | |
1da177e4 | 1305 | case CPU_DEAD: |
8bb78442 | 1306 | case CPU_DEAD_FROZEN: |
4484ebf1 RT |
1307 | /* |
1308 | * Even if all the cpus of a node are down, we don't free the | |
1309 | * kmem_list3 of any cache. This to avoid a race between | |
1310 | * cpu_down, and a kmalloc allocation from another cpu for | |
1311 | * memory from the node of the cpu going down. The list3 | |
1312 | * structure is usually allocated from kmem_cache_create() and | |
1313 | * gets destroyed at kmem_cache_destroy(). | |
1314 | */ | |
183ff22b | 1315 | /* fall through */ |
8f5be20b | 1316 | #endif |
1da177e4 | 1317 | case CPU_UP_CANCELED: |
8bb78442 | 1318 | case CPU_UP_CANCELED_FROZEN: |
95402b38 | 1319 | mutex_lock(&cache_chain_mutex); |
fbf1e473 | 1320 | cpuup_canceled(cpu); |
fc0abb14 | 1321 | mutex_unlock(&cache_chain_mutex); |
1da177e4 | 1322 | break; |
1da177e4 | 1323 | } |
eac40680 | 1324 | return notifier_from_errno(err); |
1da177e4 LT |
1325 | } |
1326 | ||
74b85f37 CS |
1327 | static struct notifier_block __cpuinitdata cpucache_notifier = { |
1328 | &cpuup_callback, NULL, 0 | |
1329 | }; | |
1da177e4 | 1330 | |
8f9f8d9e DR |
1331 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1332 | /* | |
1333 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1334 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1335 | * removed. | |
1336 | * | |
1337 | * Must hold cache_chain_mutex. | |
1338 | */ | |
1339 | static int __meminit drain_cache_nodelists_node(int node) | |
1340 | { | |
1341 | struct kmem_cache *cachep; | |
1342 | int ret = 0; | |
1343 | ||
1344 | list_for_each_entry(cachep, &cache_chain, next) { | |
1345 | struct kmem_list3 *l3; | |
1346 | ||
1347 | l3 = cachep->nodelists[node]; | |
1348 | if (!l3) | |
1349 | continue; | |
1350 | ||
1351 | drain_freelist(cachep, l3, l3->free_objects); | |
1352 | ||
1353 | if (!list_empty(&l3->slabs_full) || | |
1354 | !list_empty(&l3->slabs_partial)) { | |
1355 | ret = -EBUSY; | |
1356 | break; | |
1357 | } | |
1358 | } | |
1359 | return ret; | |
1360 | } | |
1361 | ||
1362 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1363 | unsigned long action, void *arg) | |
1364 | { | |
1365 | struct memory_notify *mnb = arg; | |
1366 | int ret = 0; | |
1367 | int nid; | |
1368 | ||
1369 | nid = mnb->status_change_nid; | |
1370 | if (nid < 0) | |
1371 | goto out; | |
1372 | ||
1373 | switch (action) { | |
1374 | case MEM_GOING_ONLINE: | |
1375 | mutex_lock(&cache_chain_mutex); | |
1376 | ret = init_cache_nodelists_node(nid); | |
1377 | mutex_unlock(&cache_chain_mutex); | |
1378 | break; | |
1379 | case MEM_GOING_OFFLINE: | |
1380 | mutex_lock(&cache_chain_mutex); | |
1381 | ret = drain_cache_nodelists_node(nid); | |
1382 | mutex_unlock(&cache_chain_mutex); | |
1383 | break; | |
1384 | case MEM_ONLINE: | |
1385 | case MEM_OFFLINE: | |
1386 | case MEM_CANCEL_ONLINE: | |
1387 | case MEM_CANCEL_OFFLINE: | |
1388 | break; | |
1389 | } | |
1390 | out: | |
1391 | return ret ? notifier_from_errno(ret) : NOTIFY_OK; | |
1392 | } | |
1393 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1394 | ||
e498be7d CL |
1395 | /* |
1396 | * swap the static kmem_list3 with kmalloced memory | |
1397 | */ | |
8f9f8d9e DR |
1398 | static void __init init_list(struct kmem_cache *cachep, struct kmem_list3 *list, |
1399 | int nodeid) | |
e498be7d CL |
1400 | { |
1401 | struct kmem_list3 *ptr; | |
1402 | ||
83b519e8 | 1403 | ptr = kmalloc_node(sizeof(struct kmem_list3), GFP_NOWAIT, nodeid); |
e498be7d CL |
1404 | BUG_ON(!ptr); |
1405 | ||
e498be7d | 1406 | memcpy(ptr, list, sizeof(struct kmem_list3)); |
2b2d5493 IM |
1407 | /* |
1408 | * Do not assume that spinlocks can be initialized via memcpy: | |
1409 | */ | |
1410 | spin_lock_init(&ptr->list_lock); | |
1411 | ||
e498be7d CL |
1412 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
1413 | cachep->nodelists[nodeid] = ptr; | |
e498be7d CL |
1414 | } |
1415 | ||
556a169d PE |
1416 | /* |
1417 | * For setting up all the kmem_list3s for cache whose buffer_size is same as | |
1418 | * size of kmem_list3. | |
1419 | */ | |
1420 | static void __init set_up_list3s(struct kmem_cache *cachep, int index) | |
1421 | { | |
1422 | int node; | |
1423 | ||
1424 | for_each_online_node(node) { | |
1425 | cachep->nodelists[node] = &initkmem_list3[index + node]; | |
1426 | cachep->nodelists[node]->next_reap = jiffies + | |
1427 | REAPTIMEOUT_LIST3 + | |
1428 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
1429 | } | |
1430 | } | |
1431 | ||
a737b3e2 AM |
1432 | /* |
1433 | * Initialisation. Called after the page allocator have been initialised and | |
1434 | * before smp_init(). | |
1da177e4 LT |
1435 | */ |
1436 | void __init kmem_cache_init(void) | |
1437 | { | |
1438 | size_t left_over; | |
1439 | struct cache_sizes *sizes; | |
1440 | struct cache_names *names; | |
e498be7d | 1441 | int i; |
07ed76b2 | 1442 | int order; |
1ca4cb24 | 1443 | int node; |
e498be7d | 1444 | |
b6e68bc1 | 1445 | if (num_possible_nodes() == 1) |
62918a03 SS |
1446 | use_alien_caches = 0; |
1447 | ||
e498be7d CL |
1448 | for (i = 0; i < NUM_INIT_LISTS; i++) { |
1449 | kmem_list3_init(&initkmem_list3[i]); | |
1450 | if (i < MAX_NUMNODES) | |
1451 | cache_cache.nodelists[i] = NULL; | |
1452 | } | |
556a169d | 1453 | set_up_list3s(&cache_cache, CACHE_CACHE); |
1da177e4 LT |
1454 | |
1455 | /* | |
1456 | * Fragmentation resistance on low memory - only use bigger | |
1457 | * page orders on machines with more than 32MB of memory. | |
1458 | */ | |
4481374c | 1459 | if (totalram_pages > (32 << 20) >> PAGE_SHIFT) |
1da177e4 LT |
1460 | slab_break_gfp_order = BREAK_GFP_ORDER_HI; |
1461 | ||
1da177e4 LT |
1462 | /* Bootstrap is tricky, because several objects are allocated |
1463 | * from caches that do not exist yet: | |
a737b3e2 AM |
1464 | * 1) initialize the cache_cache cache: it contains the struct |
1465 | * kmem_cache structures of all caches, except cache_cache itself: | |
1466 | * cache_cache is statically allocated. | |
e498be7d CL |
1467 | * Initially an __init data area is used for the head array and the |
1468 | * kmem_list3 structures, it's replaced with a kmalloc allocated | |
1469 | * array at the end of the bootstrap. | |
1da177e4 | 1470 | * 2) Create the first kmalloc cache. |
343e0d7a | 1471 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1472 | * An __init data area is used for the head array. |
1473 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1474 | * head arrays. | |
1da177e4 LT |
1475 | * 4) Replace the __init data head arrays for cache_cache and the first |
1476 | * kmalloc cache with kmalloc allocated arrays. | |
e498be7d CL |
1477 | * 5) Replace the __init data for kmem_list3 for cache_cache and |
1478 | * the other cache's with kmalloc allocated memory. | |
1479 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1480 | */ |
1481 | ||
7d6e6d09 | 1482 | node = numa_mem_id(); |
1ca4cb24 | 1483 | |
1da177e4 | 1484 | /* 1) create the cache_cache */ |
1da177e4 LT |
1485 | INIT_LIST_HEAD(&cache_chain); |
1486 | list_add(&cache_cache.next, &cache_chain); | |
1487 | cache_cache.colour_off = cache_line_size(); | |
1488 | cache_cache.array[smp_processor_id()] = &initarray_cache.cache; | |
ec1f5eee | 1489 | cache_cache.nodelists[node] = &initkmem_list3[CACHE_CACHE + node]; |
1da177e4 | 1490 | |
8da3430d ED |
1491 | /* |
1492 | * struct kmem_cache size depends on nr_node_ids, which | |
1493 | * can be less than MAX_NUMNODES. | |
1494 | */ | |
1495 | cache_cache.buffer_size = offsetof(struct kmem_cache, nodelists) + | |
1496 | nr_node_ids * sizeof(struct kmem_list3 *); | |
1497 | #if DEBUG | |
1498 | cache_cache.obj_size = cache_cache.buffer_size; | |
1499 | #endif | |
a737b3e2 AM |
1500 | cache_cache.buffer_size = ALIGN(cache_cache.buffer_size, |
1501 | cache_line_size()); | |
6a2d7a95 ED |
1502 | cache_cache.reciprocal_buffer_size = |
1503 | reciprocal_value(cache_cache.buffer_size); | |
1da177e4 | 1504 | |
07ed76b2 JS |
1505 | for (order = 0; order < MAX_ORDER; order++) { |
1506 | cache_estimate(order, cache_cache.buffer_size, | |
1507 | cache_line_size(), 0, &left_over, &cache_cache.num); | |
1508 | if (cache_cache.num) | |
1509 | break; | |
1510 | } | |
40094fa6 | 1511 | BUG_ON(!cache_cache.num); |
07ed76b2 | 1512 | cache_cache.gfporder = order; |
b28a02de | 1513 | cache_cache.colour = left_over / cache_cache.colour_off; |
b28a02de PE |
1514 | cache_cache.slab_size = ALIGN(cache_cache.num * sizeof(kmem_bufctl_t) + |
1515 | sizeof(struct slab), cache_line_size()); | |
1da177e4 LT |
1516 | |
1517 | /* 2+3) create the kmalloc caches */ | |
1518 | sizes = malloc_sizes; | |
1519 | names = cache_names; | |
1520 | ||
a737b3e2 AM |
1521 | /* |
1522 | * Initialize the caches that provide memory for the array cache and the | |
1523 | * kmem_list3 structures first. Without this, further allocations will | |
1524 | * bug. | |
e498be7d CL |
1525 | */ |
1526 | ||
1527 | sizes[INDEX_AC].cs_cachep = kmem_cache_create(names[INDEX_AC].name, | |
a737b3e2 AM |
1528 | sizes[INDEX_AC].cs_size, |
1529 | ARCH_KMALLOC_MINALIGN, | |
1530 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1531 | NULL); |
e498be7d | 1532 | |
a737b3e2 | 1533 | if (INDEX_AC != INDEX_L3) { |
e498be7d | 1534 | sizes[INDEX_L3].cs_cachep = |
a737b3e2 AM |
1535 | kmem_cache_create(names[INDEX_L3].name, |
1536 | sizes[INDEX_L3].cs_size, | |
1537 | ARCH_KMALLOC_MINALIGN, | |
1538 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1539 | NULL); |
a737b3e2 | 1540 | } |
e498be7d | 1541 | |
e0a42726 IM |
1542 | slab_early_init = 0; |
1543 | ||
1da177e4 | 1544 | while (sizes->cs_size != ULONG_MAX) { |
e498be7d CL |
1545 | /* |
1546 | * For performance, all the general caches are L1 aligned. | |
1da177e4 LT |
1547 | * This should be particularly beneficial on SMP boxes, as it |
1548 | * eliminates "false sharing". | |
1549 | * Note for systems short on memory removing the alignment will | |
e498be7d CL |
1550 | * allow tighter packing of the smaller caches. |
1551 | */ | |
a737b3e2 | 1552 | if (!sizes->cs_cachep) { |
e498be7d | 1553 | sizes->cs_cachep = kmem_cache_create(names->name, |
a737b3e2 AM |
1554 | sizes->cs_size, |
1555 | ARCH_KMALLOC_MINALIGN, | |
1556 | ARCH_KMALLOC_FLAGS|SLAB_PANIC, | |
20c2df83 | 1557 | NULL); |
a737b3e2 | 1558 | } |
4b51d669 CL |
1559 | #ifdef CONFIG_ZONE_DMA |
1560 | sizes->cs_dmacachep = kmem_cache_create( | |
1561 | names->name_dma, | |
a737b3e2 AM |
1562 | sizes->cs_size, |
1563 | ARCH_KMALLOC_MINALIGN, | |
1564 | ARCH_KMALLOC_FLAGS|SLAB_CACHE_DMA| | |
1565 | SLAB_PANIC, | |
20c2df83 | 1566 | NULL); |
4b51d669 | 1567 | #endif |
1da177e4 LT |
1568 | sizes++; |
1569 | names++; | |
1570 | } | |
1571 | /* 4) Replace the bootstrap head arrays */ | |
1572 | { | |
2b2d5493 | 1573 | struct array_cache *ptr; |
e498be7d | 1574 | |
83b519e8 | 1575 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1576 | |
9a2dba4b PE |
1577 | BUG_ON(cpu_cache_get(&cache_cache) != &initarray_cache.cache); |
1578 | memcpy(ptr, cpu_cache_get(&cache_cache), | |
b28a02de | 1579 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1580 | /* |
1581 | * Do not assume that spinlocks can be initialized via memcpy: | |
1582 | */ | |
1583 | spin_lock_init(&ptr->lock); | |
1584 | ||
1da177e4 | 1585 | cache_cache.array[smp_processor_id()] = ptr; |
e498be7d | 1586 | |
83b519e8 | 1587 | ptr = kmalloc(sizeof(struct arraycache_init), GFP_NOWAIT); |
e498be7d | 1588 | |
9a2dba4b | 1589 | BUG_ON(cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep) |
b28a02de | 1590 | != &initarray_generic.cache); |
9a2dba4b | 1591 | memcpy(ptr, cpu_cache_get(malloc_sizes[INDEX_AC].cs_cachep), |
b28a02de | 1592 | sizeof(struct arraycache_init)); |
2b2d5493 IM |
1593 | /* |
1594 | * Do not assume that spinlocks can be initialized via memcpy: | |
1595 | */ | |
1596 | spin_lock_init(&ptr->lock); | |
1597 | ||
e498be7d | 1598 | malloc_sizes[INDEX_AC].cs_cachep->array[smp_processor_id()] = |
b28a02de | 1599 | ptr; |
1da177e4 | 1600 | } |
e498be7d CL |
1601 | /* 5) Replace the bootstrap kmem_list3's */ |
1602 | { | |
1ca4cb24 PE |
1603 | int nid; |
1604 | ||
9c09a95c | 1605 | for_each_online_node(nid) { |
ec1f5eee | 1606 | init_list(&cache_cache, &initkmem_list3[CACHE_CACHE + nid], nid); |
556a169d | 1607 | |
e498be7d | 1608 | init_list(malloc_sizes[INDEX_AC].cs_cachep, |
1ca4cb24 | 1609 | &initkmem_list3[SIZE_AC + nid], nid); |
e498be7d CL |
1610 | |
1611 | if (INDEX_AC != INDEX_L3) { | |
1612 | init_list(malloc_sizes[INDEX_L3].cs_cachep, | |
1ca4cb24 | 1613 | &initkmem_list3[SIZE_L3 + nid], nid); |
e498be7d CL |
1614 | } |
1615 | } | |
1616 | } | |
1da177e4 | 1617 | |
8429db5c | 1618 | g_cpucache_up = EARLY; |
8429db5c PE |
1619 | } |
1620 | ||
1621 | void __init kmem_cache_init_late(void) | |
1622 | { | |
1623 | struct kmem_cache *cachep; | |
1624 | ||
8429db5c PE |
1625 | /* 6) resize the head arrays to their final sizes */ |
1626 | mutex_lock(&cache_chain_mutex); | |
1627 | list_for_each_entry(cachep, &cache_chain, next) | |
1628 | if (enable_cpucache(cachep, GFP_NOWAIT)) | |
1629 | BUG(); | |
1630 | mutex_unlock(&cache_chain_mutex); | |
056c6241 | 1631 | |
1da177e4 LT |
1632 | /* Done! */ |
1633 | g_cpucache_up = FULL; | |
1634 | ||
ec5a36f9 PE |
1635 | /* Annotate slab for lockdep -- annotate the malloc caches */ |
1636 | init_lock_keys(); | |
1637 | ||
a737b3e2 AM |
1638 | /* |
1639 | * Register a cpu startup notifier callback that initializes | |
1640 | * cpu_cache_get for all new cpus | |
1da177e4 LT |
1641 | */ |
1642 | register_cpu_notifier(&cpucache_notifier); | |
1da177e4 | 1643 | |
8f9f8d9e DR |
1644 | #ifdef CONFIG_NUMA |
1645 | /* | |
1646 | * Register a memory hotplug callback that initializes and frees | |
1647 | * nodelists. | |
1648 | */ | |
1649 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1650 | #endif | |
1651 | ||
a737b3e2 AM |
1652 | /* |
1653 | * The reap timers are started later, with a module init call: That part | |
1654 | * of the kernel is not yet operational. | |
1da177e4 LT |
1655 | */ |
1656 | } | |
1657 | ||
1658 | static int __init cpucache_init(void) | |
1659 | { | |
1660 | int cpu; | |
1661 | ||
a737b3e2 AM |
1662 | /* |
1663 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1664 | */ |
e498be7d | 1665 | for_each_online_cpu(cpu) |
a737b3e2 | 1666 | start_cpu_timer(cpu); |
1da177e4 LT |
1667 | return 0; |
1668 | } | |
1da177e4 LT |
1669 | __initcall(cpucache_init); |
1670 | ||
1671 | /* | |
1672 | * Interface to system's page allocator. No need to hold the cache-lock. | |
1673 | * | |
1674 | * If we requested dmaable memory, we will get it. Even if we | |
1675 | * did not request dmaable memory, we might get it, but that | |
1676 | * would be relatively rare and ignorable. | |
1677 | */ | |
343e0d7a | 1678 | static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
1da177e4 LT |
1679 | { |
1680 | struct page *page; | |
e1b6aa6f | 1681 | int nr_pages; |
1da177e4 LT |
1682 | int i; |
1683 | ||
d6fef9da | 1684 | #ifndef CONFIG_MMU |
e1b6aa6f CH |
1685 | /* |
1686 | * Nommu uses slab's for process anonymous memory allocations, and thus | |
1687 | * requires __GFP_COMP to properly refcount higher order allocations | |
d6fef9da | 1688 | */ |
e1b6aa6f | 1689 | flags |= __GFP_COMP; |
d6fef9da | 1690 | #endif |
765c4507 | 1691 | |
3c517a61 | 1692 | flags |= cachep->gfpflags; |
e12ba74d MG |
1693 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1694 | flags |= __GFP_RECLAIMABLE; | |
e1b6aa6f | 1695 | |
517d0869 | 1696 | page = alloc_pages_exact_node(nodeid, flags | __GFP_NOTRACK, cachep->gfporder); |
1da177e4 LT |
1697 | if (!page) |
1698 | return NULL; | |
1da177e4 | 1699 | |
e1b6aa6f | 1700 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1701 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
972d1a7b CL |
1702 | add_zone_page_state(page_zone(page), |
1703 | NR_SLAB_RECLAIMABLE, nr_pages); | |
1704 | else | |
1705 | add_zone_page_state(page_zone(page), | |
1706 | NR_SLAB_UNRECLAIMABLE, nr_pages); | |
e1b6aa6f CH |
1707 | for (i = 0; i < nr_pages; i++) |
1708 | __SetPageSlab(page + i); | |
c175eea4 | 1709 | |
b1eeab67 VN |
1710 | if (kmemcheck_enabled && !(cachep->flags & SLAB_NOTRACK)) { |
1711 | kmemcheck_alloc_shadow(page, cachep->gfporder, flags, nodeid); | |
1712 | ||
1713 | if (cachep->ctor) | |
1714 | kmemcheck_mark_uninitialized_pages(page, nr_pages); | |
1715 | else | |
1716 | kmemcheck_mark_unallocated_pages(page, nr_pages); | |
1717 | } | |
c175eea4 | 1718 | |
e1b6aa6f | 1719 | return page_address(page); |
1da177e4 LT |
1720 | } |
1721 | ||
1722 | /* | |
1723 | * Interface to system's page release. | |
1724 | */ | |
343e0d7a | 1725 | static void kmem_freepages(struct kmem_cache *cachep, void *addr) |
1da177e4 | 1726 | { |
b28a02de | 1727 | unsigned long i = (1 << cachep->gfporder); |
1da177e4 LT |
1728 | struct page *page = virt_to_page(addr); |
1729 | const unsigned long nr_freed = i; | |
1730 | ||
b1eeab67 | 1731 | kmemcheck_free_shadow(page, cachep->gfporder); |
c175eea4 | 1732 | |
972d1a7b CL |
1733 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
1734 | sub_zone_page_state(page_zone(page), | |
1735 | NR_SLAB_RECLAIMABLE, nr_freed); | |
1736 | else | |
1737 | sub_zone_page_state(page_zone(page), | |
1738 | NR_SLAB_UNRECLAIMABLE, nr_freed); | |
1da177e4 | 1739 | while (i--) { |
f205b2fe NP |
1740 | BUG_ON(!PageSlab(page)); |
1741 | __ClearPageSlab(page); | |
1da177e4 LT |
1742 | page++; |
1743 | } | |
1da177e4 LT |
1744 | if (current->reclaim_state) |
1745 | current->reclaim_state->reclaimed_slab += nr_freed; | |
1746 | free_pages((unsigned long)addr, cachep->gfporder); | |
1da177e4 LT |
1747 | } |
1748 | ||
1749 | static void kmem_rcu_free(struct rcu_head *head) | |
1750 | { | |
b28a02de | 1751 | struct slab_rcu *slab_rcu = (struct slab_rcu *)head; |
343e0d7a | 1752 | struct kmem_cache *cachep = slab_rcu->cachep; |
1da177e4 LT |
1753 | |
1754 | kmem_freepages(cachep, slab_rcu->addr); | |
1755 | if (OFF_SLAB(cachep)) | |
1756 | kmem_cache_free(cachep->slabp_cache, slab_rcu); | |
1757 | } | |
1758 | ||
1759 | #if DEBUG | |
1760 | ||
1761 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1762 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1763 | unsigned long caller) |
1da177e4 | 1764 | { |
3dafccf2 | 1765 | int size = obj_size(cachep); |
1da177e4 | 1766 | |
3dafccf2 | 1767 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1768 | |
b28a02de | 1769 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1770 | return; |
1771 | ||
b28a02de PE |
1772 | *addr++ = 0x12345678; |
1773 | *addr++ = caller; | |
1774 | *addr++ = smp_processor_id(); | |
1775 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1776 | { |
1777 | unsigned long *sptr = &caller; | |
1778 | unsigned long svalue; | |
1779 | ||
1780 | while (!kstack_end(sptr)) { | |
1781 | svalue = *sptr++; | |
1782 | if (kernel_text_address(svalue)) { | |
b28a02de | 1783 | *addr++ = svalue; |
1da177e4 LT |
1784 | size -= sizeof(unsigned long); |
1785 | if (size <= sizeof(unsigned long)) | |
1786 | break; | |
1787 | } | |
1788 | } | |
1789 | ||
1790 | } | |
b28a02de | 1791 | *addr++ = 0x87654321; |
1da177e4 LT |
1792 | } |
1793 | #endif | |
1794 | ||
343e0d7a | 1795 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1796 | { |
3dafccf2 MS |
1797 | int size = obj_size(cachep); |
1798 | addr = &((char *)addr)[obj_offset(cachep)]; | |
1da177e4 LT |
1799 | |
1800 | memset(addr, val, size); | |
b28a02de | 1801 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1802 | } |
1803 | ||
1804 | static void dump_line(char *data, int offset, int limit) | |
1805 | { | |
1806 | int i; | |
aa83aa40 DJ |
1807 | unsigned char error = 0; |
1808 | int bad_count = 0; | |
1809 | ||
1da177e4 | 1810 | printk(KERN_ERR "%03x:", offset); |
aa83aa40 DJ |
1811 | for (i = 0; i < limit; i++) { |
1812 | if (data[offset + i] != POISON_FREE) { | |
1813 | error = data[offset + i]; | |
1814 | bad_count++; | |
1815 | } | |
b28a02de | 1816 | printk(" %02x", (unsigned char)data[offset + i]); |
aa83aa40 | 1817 | } |
1da177e4 | 1818 | printk("\n"); |
aa83aa40 DJ |
1819 | |
1820 | if (bad_count == 1) { | |
1821 | error ^= POISON_FREE; | |
1822 | if (!(error & (error - 1))) { | |
1823 | printk(KERN_ERR "Single bit error detected. Probably " | |
1824 | "bad RAM.\n"); | |
1825 | #ifdef CONFIG_X86 | |
1826 | printk(KERN_ERR "Run memtest86+ or a similar memory " | |
1827 | "test tool.\n"); | |
1828 | #else | |
1829 | printk(KERN_ERR "Run a memory test tool.\n"); | |
1830 | #endif | |
1831 | } | |
1832 | } | |
1da177e4 LT |
1833 | } |
1834 | #endif | |
1835 | ||
1836 | #if DEBUG | |
1837 | ||
343e0d7a | 1838 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1839 | { |
1840 | int i, size; | |
1841 | char *realobj; | |
1842 | ||
1843 | if (cachep->flags & SLAB_RED_ZONE) { | |
b46b8f19 | 1844 | printk(KERN_ERR "Redzone: 0x%llx/0x%llx.\n", |
a737b3e2 AM |
1845 | *dbg_redzone1(cachep, objp), |
1846 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1847 | } |
1848 | ||
1849 | if (cachep->flags & SLAB_STORE_USER) { | |
1850 | printk(KERN_ERR "Last user: [<%p>]", | |
a737b3e2 | 1851 | *dbg_userword(cachep, objp)); |
1da177e4 | 1852 | print_symbol("(%s)", |
a737b3e2 | 1853 | (unsigned long)*dbg_userword(cachep, objp)); |
1da177e4 LT |
1854 | printk("\n"); |
1855 | } | |
3dafccf2 MS |
1856 | realobj = (char *)objp + obj_offset(cachep); |
1857 | size = obj_size(cachep); | |
b28a02de | 1858 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1859 | int limit; |
1860 | limit = 16; | |
b28a02de PE |
1861 | if (i + limit > size) |
1862 | limit = size - i; | |
1da177e4 LT |
1863 | dump_line(realobj, i, limit); |
1864 | } | |
1865 | } | |
1866 | ||
343e0d7a | 1867 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1868 | { |
1869 | char *realobj; | |
1870 | int size, i; | |
1871 | int lines = 0; | |
1872 | ||
3dafccf2 MS |
1873 | realobj = (char *)objp + obj_offset(cachep); |
1874 | size = obj_size(cachep); | |
1da177e4 | 1875 | |
b28a02de | 1876 | for (i = 0; i < size; i++) { |
1da177e4 | 1877 | char exp = POISON_FREE; |
b28a02de | 1878 | if (i == size - 1) |
1da177e4 LT |
1879 | exp = POISON_END; |
1880 | if (realobj[i] != exp) { | |
1881 | int limit; | |
1882 | /* Mismatch ! */ | |
1883 | /* Print header */ | |
1884 | if (lines == 0) { | |
b28a02de | 1885 | printk(KERN_ERR |
e94a40c5 DH |
1886 | "Slab corruption: %s start=%p, len=%d\n", |
1887 | cachep->name, realobj, size); | |
1da177e4 LT |
1888 | print_objinfo(cachep, objp, 0); |
1889 | } | |
1890 | /* Hexdump the affected line */ | |
b28a02de | 1891 | i = (i / 16) * 16; |
1da177e4 | 1892 | limit = 16; |
b28a02de PE |
1893 | if (i + limit > size) |
1894 | limit = size - i; | |
1da177e4 LT |
1895 | dump_line(realobj, i, limit); |
1896 | i += 16; | |
1897 | lines++; | |
1898 | /* Limit to 5 lines */ | |
1899 | if (lines > 5) | |
1900 | break; | |
1901 | } | |
1902 | } | |
1903 | if (lines != 0) { | |
1904 | /* Print some data about the neighboring objects, if they | |
1905 | * exist: | |
1906 | */ | |
6ed5eb22 | 1907 | struct slab *slabp = virt_to_slab(objp); |
8fea4e96 | 1908 | unsigned int objnr; |
1da177e4 | 1909 | |
8fea4e96 | 1910 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 | 1911 | if (objnr) { |
8fea4e96 | 1912 | objp = index_to_obj(cachep, slabp, objnr - 1); |
3dafccf2 | 1913 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1914 | printk(KERN_ERR "Prev obj: start=%p, len=%d\n", |
b28a02de | 1915 | realobj, size); |
1da177e4 LT |
1916 | print_objinfo(cachep, objp, 2); |
1917 | } | |
b28a02de | 1918 | if (objnr + 1 < cachep->num) { |
8fea4e96 | 1919 | objp = index_to_obj(cachep, slabp, objnr + 1); |
3dafccf2 | 1920 | realobj = (char *)objp + obj_offset(cachep); |
1da177e4 | 1921 | printk(KERN_ERR "Next obj: start=%p, len=%d\n", |
b28a02de | 1922 | realobj, size); |
1da177e4 LT |
1923 | print_objinfo(cachep, objp, 2); |
1924 | } | |
1925 | } | |
1926 | } | |
1927 | #endif | |
1928 | ||
12dd36fa | 1929 | #if DEBUG |
e79aec29 | 1930 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 | 1931 | { |
1da177e4 LT |
1932 | int i; |
1933 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 1934 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
1935 | |
1936 | if (cachep->flags & SLAB_POISON) { | |
1937 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 AM |
1938 | if (cachep->buffer_size % PAGE_SIZE == 0 && |
1939 | OFF_SLAB(cachep)) | |
b28a02de | 1940 | kernel_map_pages(virt_to_page(objp), |
a737b3e2 | 1941 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
1942 | else |
1943 | check_poison_obj(cachep, objp); | |
1944 | #else | |
1945 | check_poison_obj(cachep, objp); | |
1946 | #endif | |
1947 | } | |
1948 | if (cachep->flags & SLAB_RED_ZONE) { | |
1949 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
1950 | slab_error(cachep, "start of a freed object " | |
b28a02de | 1951 | "was overwritten"); |
1da177e4 LT |
1952 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
1953 | slab_error(cachep, "end of a freed object " | |
b28a02de | 1954 | "was overwritten"); |
1da177e4 | 1955 | } |
1da177e4 | 1956 | } |
12dd36fa | 1957 | } |
1da177e4 | 1958 | #else |
e79aec29 | 1959 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa | 1960 | { |
12dd36fa | 1961 | } |
1da177e4 LT |
1962 | #endif |
1963 | ||
911851e6 RD |
1964 | /** |
1965 | * slab_destroy - destroy and release all objects in a slab | |
1966 | * @cachep: cache pointer being destroyed | |
1967 | * @slabp: slab pointer being destroyed | |
1968 | * | |
12dd36fa | 1969 | * Destroy all the objs in a slab, and release the mem back to the system. |
a737b3e2 AM |
1970 | * Before calling the slab must have been unlinked from the cache. The |
1971 | * cache-lock is not held/needed. | |
12dd36fa | 1972 | */ |
343e0d7a | 1973 | static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp) |
12dd36fa MD |
1974 | { |
1975 | void *addr = slabp->s_mem - slabp->colouroff; | |
1976 | ||
e79aec29 | 1977 | slab_destroy_debugcheck(cachep, slabp); |
1da177e4 LT |
1978 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) { |
1979 | struct slab_rcu *slab_rcu; | |
1980 | ||
b28a02de | 1981 | slab_rcu = (struct slab_rcu *)slabp; |
1da177e4 LT |
1982 | slab_rcu->cachep = cachep; |
1983 | slab_rcu->addr = addr; | |
1984 | call_rcu(&slab_rcu->head, kmem_rcu_free); | |
1985 | } else { | |
1986 | kmem_freepages(cachep, addr); | |
873623df IM |
1987 | if (OFF_SLAB(cachep)) |
1988 | kmem_cache_free(cachep->slabp_cache, slabp); | |
1da177e4 LT |
1989 | } |
1990 | } | |
1991 | ||
117f6eb1 CL |
1992 | static void __kmem_cache_destroy(struct kmem_cache *cachep) |
1993 | { | |
1994 | int i; | |
1995 | struct kmem_list3 *l3; | |
1996 | ||
1997 | for_each_online_cpu(i) | |
1998 | kfree(cachep->array[i]); | |
1999 | ||
2000 | /* NUMA: free the list3 structures */ | |
2001 | for_each_online_node(i) { | |
2002 | l3 = cachep->nodelists[i]; | |
2003 | if (l3) { | |
2004 | kfree(l3->shared); | |
2005 | free_alien_cache(l3->alien); | |
2006 | kfree(l3); | |
2007 | } | |
2008 | } | |
2009 | kmem_cache_free(&cache_cache, cachep); | |
2010 | } | |
2011 | ||
2012 | ||
4d268eba | 2013 | /** |
a70773dd RD |
2014 | * calculate_slab_order - calculate size (page order) of slabs |
2015 | * @cachep: pointer to the cache that is being created | |
2016 | * @size: size of objects to be created in this cache. | |
2017 | * @align: required alignment for the objects. | |
2018 | * @flags: slab allocation flags | |
2019 | * | |
2020 | * Also calculates the number of objects per slab. | |
4d268eba PE |
2021 | * |
2022 | * This could be made much more intelligent. For now, try to avoid using | |
2023 | * high order pages for slabs. When the gfp() functions are more friendly | |
2024 | * towards high-order requests, this should be changed. | |
2025 | */ | |
a737b3e2 | 2026 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
ee13d785 | 2027 | size_t size, size_t align, unsigned long flags) |
4d268eba | 2028 | { |
b1ab41c4 | 2029 | unsigned long offslab_limit; |
4d268eba | 2030 | size_t left_over = 0; |
9888e6fa | 2031 | int gfporder; |
4d268eba | 2032 | |
0aa817f0 | 2033 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
2034 | unsigned int num; |
2035 | size_t remainder; | |
2036 | ||
9888e6fa | 2037 | cache_estimate(gfporder, size, align, flags, &remainder, &num); |
4d268eba PE |
2038 | if (!num) |
2039 | continue; | |
9888e6fa | 2040 | |
b1ab41c4 IM |
2041 | if (flags & CFLGS_OFF_SLAB) { |
2042 | /* | |
2043 | * Max number of objs-per-slab for caches which | |
2044 | * use off-slab slabs. Needed to avoid a possible | |
2045 | * looping condition in cache_grow(). | |
2046 | */ | |
2047 | offslab_limit = size - sizeof(struct slab); | |
2048 | offslab_limit /= sizeof(kmem_bufctl_t); | |
2049 | ||
2050 | if (num > offslab_limit) | |
2051 | break; | |
2052 | } | |
4d268eba | 2053 | |
9888e6fa | 2054 | /* Found something acceptable - save it away */ |
4d268eba | 2055 | cachep->num = num; |
9888e6fa | 2056 | cachep->gfporder = gfporder; |
4d268eba PE |
2057 | left_over = remainder; |
2058 | ||
f78bb8ad LT |
2059 | /* |
2060 | * A VFS-reclaimable slab tends to have most allocations | |
2061 | * as GFP_NOFS and we really don't want to have to be allocating | |
2062 | * higher-order pages when we are unable to shrink dcache. | |
2063 | */ | |
2064 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
2065 | break; | |
2066 | ||
4d268eba PE |
2067 | /* |
2068 | * Large number of objects is good, but very large slabs are | |
2069 | * currently bad for the gfp()s. | |
2070 | */ | |
9888e6fa | 2071 | if (gfporder >= slab_break_gfp_order) |
4d268eba PE |
2072 | break; |
2073 | ||
9888e6fa LT |
2074 | /* |
2075 | * Acceptable internal fragmentation? | |
2076 | */ | |
a737b3e2 | 2077 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
2078 | break; |
2079 | } | |
2080 | return left_over; | |
2081 | } | |
2082 | ||
83b519e8 | 2083 | static int __init_refok setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 2084 | { |
2ed3a4ef | 2085 | if (g_cpucache_up == FULL) |
83b519e8 | 2086 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 2087 | |
f30cf7d1 PE |
2088 | if (g_cpucache_up == NONE) { |
2089 | /* | |
2090 | * Note: the first kmem_cache_create must create the cache | |
2091 | * that's used by kmalloc(24), otherwise the creation of | |
2092 | * further caches will BUG(). | |
2093 | */ | |
2094 | cachep->array[smp_processor_id()] = &initarray_generic.cache; | |
2095 | ||
2096 | /* | |
2097 | * If the cache that's used by kmalloc(sizeof(kmem_list3)) is | |
2098 | * the first cache, then we need to set up all its list3s, | |
2099 | * otherwise the creation of further caches will BUG(). | |
2100 | */ | |
2101 | set_up_list3s(cachep, SIZE_AC); | |
2102 | if (INDEX_AC == INDEX_L3) | |
2103 | g_cpucache_up = PARTIAL_L3; | |
2104 | else | |
2105 | g_cpucache_up = PARTIAL_AC; | |
2106 | } else { | |
2107 | cachep->array[smp_processor_id()] = | |
83b519e8 | 2108 | kmalloc(sizeof(struct arraycache_init), gfp); |
f30cf7d1 PE |
2109 | |
2110 | if (g_cpucache_up == PARTIAL_AC) { | |
2111 | set_up_list3s(cachep, SIZE_L3); | |
2112 | g_cpucache_up = PARTIAL_L3; | |
2113 | } else { | |
2114 | int node; | |
556a169d | 2115 | for_each_online_node(node) { |
f30cf7d1 PE |
2116 | cachep->nodelists[node] = |
2117 | kmalloc_node(sizeof(struct kmem_list3), | |
eb91f1d0 | 2118 | gfp, node); |
f30cf7d1 PE |
2119 | BUG_ON(!cachep->nodelists[node]); |
2120 | kmem_list3_init(cachep->nodelists[node]); | |
2121 | } | |
2122 | } | |
2123 | } | |
7d6e6d09 | 2124 | cachep->nodelists[numa_mem_id()]->next_reap = |
f30cf7d1 PE |
2125 | jiffies + REAPTIMEOUT_LIST3 + |
2126 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; | |
2127 | ||
2128 | cpu_cache_get(cachep)->avail = 0; | |
2129 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
2130 | cpu_cache_get(cachep)->batchcount = 1; | |
2131 | cpu_cache_get(cachep)->touched = 0; | |
2132 | cachep->batchcount = 1; | |
2133 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 2134 | return 0; |
f30cf7d1 PE |
2135 | } |
2136 | ||
1da177e4 LT |
2137 | /** |
2138 | * kmem_cache_create - Create a cache. | |
2139 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
2140 | * @size: The size of objects to be created in this cache. | |
2141 | * @align: The required alignment for the objects. | |
2142 | * @flags: SLAB flags | |
2143 | * @ctor: A constructor for the objects. | |
1da177e4 LT |
2144 | * |
2145 | * Returns a ptr to the cache on success, NULL on failure. | |
2146 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 2147 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 LT |
2148 | * |
2149 | * @name must be valid until the cache is destroyed. This implies that | |
a737b3e2 | 2150 | * the module calling this has to destroy the cache before getting unloaded. |
249da166 CM |
2151 | * Note that kmem_cache_name() is not guaranteed to return the same pointer, |
2152 | * therefore applications must manage it themselves. | |
a737b3e2 | 2153 | * |
1da177e4 LT |
2154 | * The flags are |
2155 | * | |
2156 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
2157 | * to catch references to uninitialised memory. | |
2158 | * | |
2159 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
2160 | * for buffer overruns. | |
2161 | * | |
1da177e4 LT |
2162 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
2163 | * cacheline. This can be beneficial if you're counting cycles as closely | |
2164 | * as davem. | |
2165 | */ | |
343e0d7a | 2166 | struct kmem_cache * |
1da177e4 | 2167 | kmem_cache_create (const char *name, size_t size, size_t align, |
51cc5068 | 2168 | unsigned long flags, void (*ctor)(void *)) |
1da177e4 LT |
2169 | { |
2170 | size_t left_over, slab_size, ralign; | |
7a7c381d | 2171 | struct kmem_cache *cachep = NULL, *pc; |
83b519e8 | 2172 | gfp_t gfp; |
1da177e4 LT |
2173 | |
2174 | /* | |
2175 | * Sanity checks... these are all serious usage bugs. | |
2176 | */ | |
a737b3e2 | 2177 | if (!name || in_interrupt() || (size < BYTES_PER_WORD) || |
20c2df83 | 2178 | size > KMALLOC_MAX_SIZE) { |
d40cee24 | 2179 | printk(KERN_ERR "%s: Early error in slab %s\n", __func__, |
a737b3e2 | 2180 | name); |
b28a02de PE |
2181 | BUG(); |
2182 | } | |
1da177e4 | 2183 | |
f0188f47 | 2184 | /* |
8f5be20b | 2185 | * We use cache_chain_mutex to ensure a consistent view of |
174596a0 | 2186 | * cpu_online_mask as well. Please see cpuup_callback |
f0188f47 | 2187 | */ |
83b519e8 PE |
2188 | if (slab_is_available()) { |
2189 | get_online_cpus(); | |
2190 | mutex_lock(&cache_chain_mutex); | |
2191 | } | |
4f12bb4f | 2192 | |
7a7c381d | 2193 | list_for_each_entry(pc, &cache_chain, next) { |
4f12bb4f AM |
2194 | char tmp; |
2195 | int res; | |
2196 | ||
2197 | /* | |
2198 | * This happens when the module gets unloaded and doesn't | |
2199 | * destroy its slab cache and no-one else reuses the vmalloc | |
2200 | * area of the module. Print a warning. | |
2201 | */ | |
138ae663 | 2202 | res = probe_kernel_address(pc->name, tmp); |
4f12bb4f | 2203 | if (res) { |
b4169525 | 2204 | printk(KERN_ERR |
2205 | "SLAB: cache with size %d has lost its name\n", | |
3dafccf2 | 2206 | pc->buffer_size); |
4f12bb4f AM |
2207 | continue; |
2208 | } | |
2209 | ||
b28a02de | 2210 | if (!strcmp(pc->name, name)) { |
b4169525 | 2211 | printk(KERN_ERR |
2212 | "kmem_cache_create: duplicate cache %s\n", name); | |
4f12bb4f AM |
2213 | dump_stack(); |
2214 | goto oops; | |
2215 | } | |
2216 | } | |
2217 | ||
1da177e4 LT |
2218 | #if DEBUG |
2219 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
1da177e4 LT |
2220 | #if FORCED_DEBUG |
2221 | /* | |
2222 | * Enable redzoning and last user accounting, except for caches with | |
2223 | * large objects, if the increased size would increase the object size | |
2224 | * above the next power of two: caches with object sizes just above a | |
2225 | * power of two have a significant amount of internal fragmentation. | |
2226 | */ | |
87a927c7 DW |
2227 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2228 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2229 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
1da177e4 LT |
2230 | if (!(flags & SLAB_DESTROY_BY_RCU)) |
2231 | flags |= SLAB_POISON; | |
2232 | #endif | |
2233 | if (flags & SLAB_DESTROY_BY_RCU) | |
2234 | BUG_ON(flags & SLAB_POISON); | |
2235 | #endif | |
1da177e4 | 2236 | /* |
a737b3e2 AM |
2237 | * Always checks flags, a caller might be expecting debug support which |
2238 | * isn't available. | |
1da177e4 | 2239 | */ |
40094fa6 | 2240 | BUG_ON(flags & ~CREATE_MASK); |
1da177e4 | 2241 | |
a737b3e2 AM |
2242 | /* |
2243 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2244 | * unaligned accesses for some archs when redzoning is used, and makes |
2245 | * sure any on-slab bufctl's are also correctly aligned. | |
2246 | */ | |
b28a02de PE |
2247 | if (size & (BYTES_PER_WORD - 1)) { |
2248 | size += (BYTES_PER_WORD - 1); | |
2249 | size &= ~(BYTES_PER_WORD - 1); | |
1da177e4 LT |
2250 | } |
2251 | ||
a737b3e2 AM |
2252 | /* calculate the final buffer alignment: */ |
2253 | ||
1da177e4 LT |
2254 | /* 1) arch recommendation: can be overridden for debug */ |
2255 | if (flags & SLAB_HWCACHE_ALIGN) { | |
a737b3e2 AM |
2256 | /* |
2257 | * Default alignment: as specified by the arch code. Except if | |
2258 | * an object is really small, then squeeze multiple objects into | |
2259 | * one cacheline. | |
1da177e4 LT |
2260 | */ |
2261 | ralign = cache_line_size(); | |
b28a02de | 2262 | while (size <= ralign / 2) |
1da177e4 LT |
2263 | ralign /= 2; |
2264 | } else { | |
2265 | ralign = BYTES_PER_WORD; | |
2266 | } | |
ca5f9703 PE |
2267 | |
2268 | /* | |
87a927c7 DW |
2269 | * Redzoning and user store require word alignment or possibly larger. |
2270 | * Note this will be overridden by architecture or caller mandated | |
2271 | * alignment if either is greater than BYTES_PER_WORD. | |
ca5f9703 | 2272 | */ |
87a927c7 DW |
2273 | if (flags & SLAB_STORE_USER) |
2274 | ralign = BYTES_PER_WORD; | |
2275 | ||
2276 | if (flags & SLAB_RED_ZONE) { | |
2277 | ralign = REDZONE_ALIGN; | |
2278 | /* If redzoning, ensure that the second redzone is suitably | |
2279 | * aligned, by adjusting the object size accordingly. */ | |
2280 | size += REDZONE_ALIGN - 1; | |
2281 | size &= ~(REDZONE_ALIGN - 1); | |
2282 | } | |
ca5f9703 | 2283 | |
a44b56d3 | 2284 | /* 2) arch mandated alignment */ |
1da177e4 LT |
2285 | if (ralign < ARCH_SLAB_MINALIGN) { |
2286 | ralign = ARCH_SLAB_MINALIGN; | |
1da177e4 | 2287 | } |
a44b56d3 | 2288 | /* 3) caller mandated alignment */ |
1da177e4 LT |
2289 | if (ralign < align) { |
2290 | ralign = align; | |
1da177e4 | 2291 | } |
5c5e3b33 SL |
2292 | /* disable debug if not aligning with REDZONE_ALIGN */ |
2293 | if (ralign & (__alignof__(unsigned long long) - 1)) | |
a44b56d3 | 2294 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2295 | /* |
ca5f9703 | 2296 | * 4) Store it. |
1da177e4 LT |
2297 | */ |
2298 | align = ralign; | |
2299 | ||
83b519e8 PE |
2300 | if (slab_is_available()) |
2301 | gfp = GFP_KERNEL; | |
2302 | else | |
2303 | gfp = GFP_NOWAIT; | |
2304 | ||
1da177e4 | 2305 | /* Get cache's description obj. */ |
83b519e8 | 2306 | cachep = kmem_cache_zalloc(&cache_cache, gfp); |
1da177e4 | 2307 | if (!cachep) |
4f12bb4f | 2308 | goto oops; |
1da177e4 LT |
2309 | |
2310 | #if DEBUG | |
3dafccf2 | 2311 | cachep->obj_size = size; |
1da177e4 | 2312 | |
ca5f9703 PE |
2313 | /* |
2314 | * Both debugging options require word-alignment which is calculated | |
2315 | * into align above. | |
2316 | */ | |
1da177e4 | 2317 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2318 | /* add space for red zone words */ |
5c5e3b33 SL |
2319 | cachep->obj_offset += align; |
2320 | size += align + sizeof(unsigned long long); | |
1da177e4 LT |
2321 | } |
2322 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2323 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2324 | * the real object. But if the second red zone needs to be |
2325 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2326 | */ |
87a927c7 DW |
2327 | if (flags & SLAB_RED_ZONE) |
2328 | size += REDZONE_ALIGN; | |
2329 | else | |
2330 | size += BYTES_PER_WORD; | |
1da177e4 LT |
2331 | } |
2332 | #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC) | |
b28a02de | 2333 | if (size >= malloc_sizes[INDEX_L3 + 1].cs_size |
3dafccf2 MS |
2334 | && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) { |
2335 | cachep->obj_offset += PAGE_SIZE - size; | |
1da177e4 LT |
2336 | size = PAGE_SIZE; |
2337 | } | |
2338 | #endif | |
2339 | #endif | |
2340 | ||
e0a42726 IM |
2341 | /* |
2342 | * Determine if the slab management is 'on' or 'off' slab. | |
2343 | * (bootstrapping cannot cope with offslab caches so don't do | |
e7cb55b9 CM |
2344 | * it too early on. Always use on-slab management when |
2345 | * SLAB_NOLEAKTRACE to avoid recursive calls into kmemleak) | |
e0a42726 | 2346 | */ |
e7cb55b9 CM |
2347 | if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init && |
2348 | !(flags & SLAB_NOLEAKTRACE)) | |
1da177e4 LT |
2349 | /* |
2350 | * Size is large, assume best to place the slab management obj | |
2351 | * off-slab (should allow better packing of objs). | |
2352 | */ | |
2353 | flags |= CFLGS_OFF_SLAB; | |
2354 | ||
2355 | size = ALIGN(size, align); | |
2356 | ||
f78bb8ad | 2357 | left_over = calculate_slab_order(cachep, size, align, flags); |
1da177e4 LT |
2358 | |
2359 | if (!cachep->num) { | |
b4169525 | 2360 | printk(KERN_ERR |
2361 | "kmem_cache_create: couldn't create cache %s.\n", name); | |
1da177e4 LT |
2362 | kmem_cache_free(&cache_cache, cachep); |
2363 | cachep = NULL; | |
4f12bb4f | 2364 | goto oops; |
1da177e4 | 2365 | } |
b28a02de PE |
2366 | slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t) |
2367 | + sizeof(struct slab), align); | |
1da177e4 LT |
2368 | |
2369 | /* | |
2370 | * If the slab has been placed off-slab, and we have enough space then | |
2371 | * move it on-slab. This is at the expense of any extra colouring. | |
2372 | */ | |
2373 | if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) { | |
2374 | flags &= ~CFLGS_OFF_SLAB; | |
2375 | left_over -= slab_size; | |
2376 | } | |
2377 | ||
2378 | if (flags & CFLGS_OFF_SLAB) { | |
2379 | /* really off slab. No need for manual alignment */ | |
b28a02de PE |
2380 | slab_size = |
2381 | cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab); | |
67461365 RL |
2382 | |
2383 | #ifdef CONFIG_PAGE_POISONING | |
2384 | /* If we're going to use the generic kernel_map_pages() | |
2385 | * poisoning, then it's going to smash the contents of | |
2386 | * the redzone and userword anyhow, so switch them off. | |
2387 | */ | |
2388 | if (size % PAGE_SIZE == 0 && flags & SLAB_POISON) | |
2389 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2390 | #endif | |
1da177e4 LT |
2391 | } |
2392 | ||
2393 | cachep->colour_off = cache_line_size(); | |
2394 | /* Offset must be a multiple of the alignment. */ | |
2395 | if (cachep->colour_off < align) | |
2396 | cachep->colour_off = align; | |
b28a02de | 2397 | cachep->colour = left_over / cachep->colour_off; |
1da177e4 LT |
2398 | cachep->slab_size = slab_size; |
2399 | cachep->flags = flags; | |
2400 | cachep->gfpflags = 0; | |
4b51d669 | 2401 | if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA)) |
1da177e4 | 2402 | cachep->gfpflags |= GFP_DMA; |
3dafccf2 | 2403 | cachep->buffer_size = size; |
6a2d7a95 | 2404 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2405 | |
e5ac9c5a | 2406 | if (flags & CFLGS_OFF_SLAB) { |
b2d55073 | 2407 | cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u); |
e5ac9c5a RT |
2408 | /* |
2409 | * This is a possibility for one of the malloc_sizes caches. | |
2410 | * But since we go off slab only for object size greater than | |
2411 | * PAGE_SIZE/8, and malloc_sizes gets created in ascending order, | |
2412 | * this should not happen at all. | |
2413 | * But leave a BUG_ON for some lucky dude. | |
2414 | */ | |
6cb8f913 | 2415 | BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache)); |
e5ac9c5a | 2416 | } |
1da177e4 | 2417 | cachep->ctor = ctor; |
1da177e4 LT |
2418 | cachep->name = name; |
2419 | ||
83b519e8 | 2420 | if (setup_cpu_cache(cachep, gfp)) { |
2ed3a4ef CL |
2421 | __kmem_cache_destroy(cachep); |
2422 | cachep = NULL; | |
2423 | goto oops; | |
2424 | } | |
1da177e4 | 2425 | |
1da177e4 LT |
2426 | /* cache setup completed, link it into the list */ |
2427 | list_add(&cachep->next, &cache_chain); | |
a737b3e2 | 2428 | oops: |
1da177e4 LT |
2429 | if (!cachep && (flags & SLAB_PANIC)) |
2430 | panic("kmem_cache_create(): failed to create slab `%s'\n", | |
b28a02de | 2431 | name); |
83b519e8 PE |
2432 | if (slab_is_available()) { |
2433 | mutex_unlock(&cache_chain_mutex); | |
2434 | put_online_cpus(); | |
2435 | } | |
1da177e4 LT |
2436 | return cachep; |
2437 | } | |
2438 | EXPORT_SYMBOL(kmem_cache_create); | |
2439 | ||
2440 | #if DEBUG | |
2441 | static void check_irq_off(void) | |
2442 | { | |
2443 | BUG_ON(!irqs_disabled()); | |
2444 | } | |
2445 | ||
2446 | static void check_irq_on(void) | |
2447 | { | |
2448 | BUG_ON(irqs_disabled()); | |
2449 | } | |
2450 | ||
343e0d7a | 2451 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2452 | { |
2453 | #ifdef CONFIG_SMP | |
2454 | check_irq_off(); | |
7d6e6d09 | 2455 | assert_spin_locked(&cachep->nodelists[numa_mem_id()]->list_lock); |
1da177e4 LT |
2456 | #endif |
2457 | } | |
e498be7d | 2458 | |
343e0d7a | 2459 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2460 | { |
2461 | #ifdef CONFIG_SMP | |
2462 | check_irq_off(); | |
2463 | assert_spin_locked(&cachep->nodelists[node]->list_lock); | |
2464 | #endif | |
2465 | } | |
2466 | ||
1da177e4 LT |
2467 | #else |
2468 | #define check_irq_off() do { } while(0) | |
2469 | #define check_irq_on() do { } while(0) | |
2470 | #define check_spinlock_acquired(x) do { } while(0) | |
e498be7d | 2471 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2472 | #endif |
2473 | ||
aab2207c CL |
2474 | static void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, |
2475 | struct array_cache *ac, | |
2476 | int force, int node); | |
2477 | ||
1da177e4 LT |
2478 | static void do_drain(void *arg) |
2479 | { | |
a737b3e2 | 2480 | struct kmem_cache *cachep = arg; |
1da177e4 | 2481 | struct array_cache *ac; |
7d6e6d09 | 2482 | int node = numa_mem_id(); |
1da177e4 LT |
2483 | |
2484 | check_irq_off(); | |
9a2dba4b | 2485 | ac = cpu_cache_get(cachep); |
ff69416e CL |
2486 | spin_lock(&cachep->nodelists[node]->list_lock); |
2487 | free_block(cachep, ac->entry, ac->avail, node); | |
2488 | spin_unlock(&cachep->nodelists[node]->list_lock); | |
1da177e4 LT |
2489 | ac->avail = 0; |
2490 | } | |
2491 | ||
343e0d7a | 2492 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2493 | { |
e498be7d CL |
2494 | struct kmem_list3 *l3; |
2495 | int node; | |
2496 | ||
15c8b6c1 | 2497 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2498 | check_irq_on(); |
b28a02de | 2499 | for_each_online_node(node) { |
e498be7d | 2500 | l3 = cachep->nodelists[node]; |
a4523a8b RD |
2501 | if (l3 && l3->alien) |
2502 | drain_alien_cache(cachep, l3->alien); | |
2503 | } | |
2504 | ||
2505 | for_each_online_node(node) { | |
2506 | l3 = cachep->nodelists[node]; | |
2507 | if (l3) | |
aab2207c | 2508 | drain_array(cachep, l3, l3->shared, 1, node); |
e498be7d | 2509 | } |
1da177e4 LT |
2510 | } |
2511 | ||
ed11d9eb CL |
2512 | /* |
2513 | * Remove slabs from the list of free slabs. | |
2514 | * Specify the number of slabs to drain in tofree. | |
2515 | * | |
2516 | * Returns the actual number of slabs released. | |
2517 | */ | |
2518 | static int drain_freelist(struct kmem_cache *cache, | |
2519 | struct kmem_list3 *l3, int tofree) | |
1da177e4 | 2520 | { |
ed11d9eb CL |
2521 | struct list_head *p; |
2522 | int nr_freed; | |
1da177e4 | 2523 | struct slab *slabp; |
1da177e4 | 2524 | |
ed11d9eb CL |
2525 | nr_freed = 0; |
2526 | while (nr_freed < tofree && !list_empty(&l3->slabs_free)) { | |
1da177e4 | 2527 | |
ed11d9eb | 2528 | spin_lock_irq(&l3->list_lock); |
e498be7d | 2529 | p = l3->slabs_free.prev; |
ed11d9eb CL |
2530 | if (p == &l3->slabs_free) { |
2531 | spin_unlock_irq(&l3->list_lock); | |
2532 | goto out; | |
2533 | } | |
1da177e4 | 2534 | |
ed11d9eb | 2535 | slabp = list_entry(p, struct slab, list); |
1da177e4 | 2536 | #if DEBUG |
40094fa6 | 2537 | BUG_ON(slabp->inuse); |
1da177e4 LT |
2538 | #endif |
2539 | list_del(&slabp->list); | |
ed11d9eb CL |
2540 | /* |
2541 | * Safe to drop the lock. The slab is no longer linked | |
2542 | * to the cache. | |
2543 | */ | |
2544 | l3->free_objects -= cache->num; | |
e498be7d | 2545 | spin_unlock_irq(&l3->list_lock); |
ed11d9eb CL |
2546 | slab_destroy(cache, slabp); |
2547 | nr_freed++; | |
1da177e4 | 2548 | } |
ed11d9eb CL |
2549 | out: |
2550 | return nr_freed; | |
1da177e4 LT |
2551 | } |
2552 | ||
8f5be20b | 2553 | /* Called with cache_chain_mutex held to protect against cpu hotplug */ |
343e0d7a | 2554 | static int __cache_shrink(struct kmem_cache *cachep) |
e498be7d CL |
2555 | { |
2556 | int ret = 0, i = 0; | |
2557 | struct kmem_list3 *l3; | |
2558 | ||
2559 | drain_cpu_caches(cachep); | |
2560 | ||
2561 | check_irq_on(); | |
2562 | for_each_online_node(i) { | |
2563 | l3 = cachep->nodelists[i]; | |
ed11d9eb CL |
2564 | if (!l3) |
2565 | continue; | |
2566 | ||
2567 | drain_freelist(cachep, l3, l3->free_objects); | |
2568 | ||
2569 | ret += !list_empty(&l3->slabs_full) || | |
2570 | !list_empty(&l3->slabs_partial); | |
e498be7d CL |
2571 | } |
2572 | return (ret ? 1 : 0); | |
2573 | } | |
2574 | ||
1da177e4 LT |
2575 | /** |
2576 | * kmem_cache_shrink - Shrink a cache. | |
2577 | * @cachep: The cache to shrink. | |
2578 | * | |
2579 | * Releases as many slabs as possible for a cache. | |
2580 | * To help debugging, a zero exit status indicates all slabs were released. | |
2581 | */ | |
343e0d7a | 2582 | int kmem_cache_shrink(struct kmem_cache *cachep) |
1da177e4 | 2583 | { |
8f5be20b | 2584 | int ret; |
40094fa6 | 2585 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2586 | |
95402b38 | 2587 | get_online_cpus(); |
8f5be20b RT |
2588 | mutex_lock(&cache_chain_mutex); |
2589 | ret = __cache_shrink(cachep); | |
2590 | mutex_unlock(&cache_chain_mutex); | |
95402b38 | 2591 | put_online_cpus(); |
8f5be20b | 2592 | return ret; |
1da177e4 LT |
2593 | } |
2594 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2595 | ||
2596 | /** | |
2597 | * kmem_cache_destroy - delete a cache | |
2598 | * @cachep: the cache to destroy | |
2599 | * | |
72fd4a35 | 2600 | * Remove a &struct kmem_cache object from the slab cache. |
1da177e4 LT |
2601 | * |
2602 | * It is expected this function will be called by a module when it is | |
2603 | * unloaded. This will remove the cache completely, and avoid a duplicate | |
2604 | * cache being allocated each time a module is loaded and unloaded, if the | |
2605 | * module doesn't have persistent in-kernel storage across loads and unloads. | |
2606 | * | |
2607 | * The cache must be empty before calling this function. | |
2608 | * | |
2609 | * The caller must guarantee that noone will allocate memory from the cache | |
2610 | * during the kmem_cache_destroy(). | |
2611 | */ | |
133d205a | 2612 | void kmem_cache_destroy(struct kmem_cache *cachep) |
1da177e4 | 2613 | { |
40094fa6 | 2614 | BUG_ON(!cachep || in_interrupt()); |
1da177e4 | 2615 | |
1da177e4 | 2616 | /* Find the cache in the chain of caches. */ |
95402b38 | 2617 | get_online_cpus(); |
fc0abb14 | 2618 | mutex_lock(&cache_chain_mutex); |
1da177e4 LT |
2619 | /* |
2620 | * the chain is never empty, cache_cache is never destroyed | |
2621 | */ | |
2622 | list_del(&cachep->next); | |
1da177e4 LT |
2623 | if (__cache_shrink(cachep)) { |
2624 | slab_error(cachep, "Can't free all objects"); | |
b28a02de | 2625 | list_add(&cachep->next, &cache_chain); |
fc0abb14 | 2626 | mutex_unlock(&cache_chain_mutex); |
95402b38 | 2627 | put_online_cpus(); |
133d205a | 2628 | return; |
1da177e4 LT |
2629 | } |
2630 | ||
2631 | if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) | |
7ed9f7e5 | 2632 | rcu_barrier(); |
1da177e4 | 2633 | |
117f6eb1 | 2634 | __kmem_cache_destroy(cachep); |
8f5be20b | 2635 | mutex_unlock(&cache_chain_mutex); |
95402b38 | 2636 | put_online_cpus(); |
1da177e4 LT |
2637 | } |
2638 | EXPORT_SYMBOL(kmem_cache_destroy); | |
2639 | ||
e5ac9c5a RT |
2640 | /* |
2641 | * Get the memory for a slab management obj. | |
2642 | * For a slab cache when the slab descriptor is off-slab, slab descriptors | |
2643 | * always come from malloc_sizes caches. The slab descriptor cannot | |
2644 | * come from the same cache which is getting created because, | |
2645 | * when we are searching for an appropriate cache for these | |
2646 | * descriptors in kmem_cache_create, we search through the malloc_sizes array. | |
2647 | * If we are creating a malloc_sizes cache here it would not be visible to | |
2648 | * kmem_find_general_cachep till the initialization is complete. | |
2649 | * Hence we cannot have slabp_cache same as the original cache. | |
2650 | */ | |
343e0d7a | 2651 | static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp, |
5b74ada7 RT |
2652 | int colour_off, gfp_t local_flags, |
2653 | int nodeid) | |
1da177e4 LT |
2654 | { |
2655 | struct slab *slabp; | |
b28a02de | 2656 | |
1da177e4 LT |
2657 | if (OFF_SLAB(cachep)) { |
2658 | /* Slab management obj is off-slab. */ | |
5b74ada7 | 2659 | slabp = kmem_cache_alloc_node(cachep->slabp_cache, |
8759ec50 | 2660 | local_flags, nodeid); |
d5cff635 CM |
2661 | /* |
2662 | * If the first object in the slab is leaked (it's allocated | |
2663 | * but no one has a reference to it), we want to make sure | |
2664 | * kmemleak does not treat the ->s_mem pointer as a reference | |
2665 | * to the object. Otherwise we will not report the leak. | |
2666 | */ | |
c017b4be CM |
2667 | kmemleak_scan_area(&slabp->list, sizeof(struct list_head), |
2668 | local_flags); | |
1da177e4 LT |
2669 | if (!slabp) |
2670 | return NULL; | |
2671 | } else { | |
b28a02de | 2672 | slabp = objp + colour_off; |
1da177e4 LT |
2673 | colour_off += cachep->slab_size; |
2674 | } | |
2675 | slabp->inuse = 0; | |
2676 | slabp->colouroff = colour_off; | |
b28a02de | 2677 | slabp->s_mem = objp + colour_off; |
5b74ada7 | 2678 | slabp->nodeid = nodeid; |
e51bfd0a | 2679 | slabp->free = 0; |
1da177e4 LT |
2680 | return slabp; |
2681 | } | |
2682 | ||
2683 | static inline kmem_bufctl_t *slab_bufctl(struct slab *slabp) | |
2684 | { | |
b28a02de | 2685 | return (kmem_bufctl_t *) (slabp + 1); |
1da177e4 LT |
2686 | } |
2687 | ||
343e0d7a | 2688 | static void cache_init_objs(struct kmem_cache *cachep, |
a35afb83 | 2689 | struct slab *slabp) |
1da177e4 LT |
2690 | { |
2691 | int i; | |
2692 | ||
2693 | for (i = 0; i < cachep->num; i++) { | |
8fea4e96 | 2694 | void *objp = index_to_obj(cachep, slabp, i); |
1da177e4 LT |
2695 | #if DEBUG |
2696 | /* need to poison the objs? */ | |
2697 | if (cachep->flags & SLAB_POISON) | |
2698 | poison_obj(cachep, objp, POISON_FREE); | |
2699 | if (cachep->flags & SLAB_STORE_USER) | |
2700 | *dbg_userword(cachep, objp) = NULL; | |
2701 | ||
2702 | if (cachep->flags & SLAB_RED_ZONE) { | |
2703 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2704 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2705 | } | |
2706 | /* | |
a737b3e2 AM |
2707 | * Constructors are not allowed to allocate memory from the same |
2708 | * cache which they are a constructor for. Otherwise, deadlock. | |
2709 | * They must also be threaded. | |
1da177e4 LT |
2710 | */ |
2711 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) | |
51cc5068 | 2712 | cachep->ctor(objp + obj_offset(cachep)); |
1da177e4 LT |
2713 | |
2714 | if (cachep->flags & SLAB_RED_ZONE) { | |
2715 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
2716 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2717 | " end of an object"); |
1da177e4 LT |
2718 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
2719 | slab_error(cachep, "constructor overwrote the" | |
b28a02de | 2720 | " start of an object"); |
1da177e4 | 2721 | } |
a737b3e2 AM |
2722 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && |
2723 | OFF_SLAB(cachep) && cachep->flags & SLAB_POISON) | |
b28a02de | 2724 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2725 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2726 | #else |
2727 | if (cachep->ctor) | |
51cc5068 | 2728 | cachep->ctor(objp); |
1da177e4 | 2729 | #endif |
b28a02de | 2730 | slab_bufctl(slabp)[i] = i + 1; |
1da177e4 | 2731 | } |
b28a02de | 2732 | slab_bufctl(slabp)[i - 1] = BUFCTL_END; |
1da177e4 LT |
2733 | } |
2734 | ||
343e0d7a | 2735 | static void kmem_flagcheck(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 2736 | { |
4b51d669 CL |
2737 | if (CONFIG_ZONE_DMA_FLAG) { |
2738 | if (flags & GFP_DMA) | |
2739 | BUG_ON(!(cachep->gfpflags & GFP_DMA)); | |
2740 | else | |
2741 | BUG_ON(cachep->gfpflags & GFP_DMA); | |
2742 | } | |
1da177e4 LT |
2743 | } |
2744 | ||
a737b3e2 AM |
2745 | static void *slab_get_obj(struct kmem_cache *cachep, struct slab *slabp, |
2746 | int nodeid) | |
78d382d7 | 2747 | { |
8fea4e96 | 2748 | void *objp = index_to_obj(cachep, slabp, slabp->free); |
78d382d7 MD |
2749 | kmem_bufctl_t next; |
2750 | ||
2751 | slabp->inuse++; | |
2752 | next = slab_bufctl(slabp)[slabp->free]; | |
2753 | #if DEBUG | |
2754 | slab_bufctl(slabp)[slabp->free] = BUFCTL_FREE; | |
2755 | WARN_ON(slabp->nodeid != nodeid); | |
2756 | #endif | |
2757 | slabp->free = next; | |
2758 | ||
2759 | return objp; | |
2760 | } | |
2761 | ||
a737b3e2 AM |
2762 | static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp, |
2763 | void *objp, int nodeid) | |
78d382d7 | 2764 | { |
8fea4e96 | 2765 | unsigned int objnr = obj_to_index(cachep, slabp, objp); |
78d382d7 MD |
2766 | |
2767 | #if DEBUG | |
2768 | /* Verify that the slab belongs to the intended node */ | |
2769 | WARN_ON(slabp->nodeid != nodeid); | |
2770 | ||
871751e2 | 2771 | if (slab_bufctl(slabp)[objnr] + 1 <= SLAB_LIMIT + 1) { |
78d382d7 | 2772 | printk(KERN_ERR "slab: double free detected in cache " |
a737b3e2 | 2773 | "'%s', objp %p\n", cachep->name, objp); |
78d382d7 MD |
2774 | BUG(); |
2775 | } | |
2776 | #endif | |
2777 | slab_bufctl(slabp)[objnr] = slabp->free; | |
2778 | slabp->free = objnr; | |
2779 | slabp->inuse--; | |
2780 | } | |
2781 | ||
4776874f PE |
2782 | /* |
2783 | * Map pages beginning at addr to the given cache and slab. This is required | |
2784 | * for the slab allocator to be able to lookup the cache and slab of a | |
2785 | * virtual address for kfree, ksize, kmem_ptr_validate, and slab debugging. | |
2786 | */ | |
2787 | static void slab_map_pages(struct kmem_cache *cache, struct slab *slab, | |
2788 | void *addr) | |
1da177e4 | 2789 | { |
4776874f | 2790 | int nr_pages; |
1da177e4 LT |
2791 | struct page *page; |
2792 | ||
4776874f | 2793 | page = virt_to_page(addr); |
84097518 | 2794 | |
4776874f | 2795 | nr_pages = 1; |
84097518 | 2796 | if (likely(!PageCompound(page))) |
4776874f PE |
2797 | nr_pages <<= cache->gfporder; |
2798 | ||
1da177e4 | 2799 | do { |
4776874f PE |
2800 | page_set_cache(page, cache); |
2801 | page_set_slab(page, slab); | |
1da177e4 | 2802 | page++; |
4776874f | 2803 | } while (--nr_pages); |
1da177e4 LT |
2804 | } |
2805 | ||
2806 | /* | |
2807 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2808 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2809 | */ | |
3c517a61 CL |
2810 | static int cache_grow(struct kmem_cache *cachep, |
2811 | gfp_t flags, int nodeid, void *objp) | |
1da177e4 | 2812 | { |
b28a02de | 2813 | struct slab *slabp; |
b28a02de PE |
2814 | size_t offset; |
2815 | gfp_t local_flags; | |
e498be7d | 2816 | struct kmem_list3 *l3; |
1da177e4 | 2817 | |
a737b3e2 AM |
2818 | /* |
2819 | * Be lazy and only check for valid flags here, keeping it out of the | |
2820 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2821 | */ |
6cb06229 CL |
2822 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
2823 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); | |
1da177e4 | 2824 | |
2e1217cf | 2825 | /* Take the l3 list lock to change the colour_next on this node */ |
1da177e4 | 2826 | check_irq_off(); |
2e1217cf RT |
2827 | l3 = cachep->nodelists[nodeid]; |
2828 | spin_lock(&l3->list_lock); | |
1da177e4 LT |
2829 | |
2830 | /* Get colour for the slab, and cal the next value. */ | |
2e1217cf RT |
2831 | offset = l3->colour_next; |
2832 | l3->colour_next++; | |
2833 | if (l3->colour_next >= cachep->colour) | |
2834 | l3->colour_next = 0; | |
2835 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2836 | |
2e1217cf | 2837 | offset *= cachep->colour_off; |
1da177e4 LT |
2838 | |
2839 | if (local_flags & __GFP_WAIT) | |
2840 | local_irq_enable(); | |
2841 | ||
2842 | /* | |
2843 | * The test for missing atomic flag is performed here, rather than | |
2844 | * the more obvious place, simply to reduce the critical path length | |
2845 | * in kmem_cache_alloc(). If a caller is seriously mis-behaving they | |
2846 | * will eventually be caught here (where it matters). | |
2847 | */ | |
2848 | kmem_flagcheck(cachep, flags); | |
2849 | ||
a737b3e2 AM |
2850 | /* |
2851 | * Get mem for the objs. Attempt to allocate a physical page from | |
2852 | * 'nodeid'. | |
e498be7d | 2853 | */ |
3c517a61 | 2854 | if (!objp) |
b8c1c5da | 2855 | objp = kmem_getpages(cachep, local_flags, nodeid); |
a737b3e2 | 2856 | if (!objp) |
1da177e4 LT |
2857 | goto failed; |
2858 | ||
2859 | /* Get slab management. */ | |
3c517a61 | 2860 | slabp = alloc_slabmgmt(cachep, objp, offset, |
6cb06229 | 2861 | local_flags & ~GFP_CONSTRAINT_MASK, nodeid); |
a737b3e2 | 2862 | if (!slabp) |
1da177e4 LT |
2863 | goto opps1; |
2864 | ||
4776874f | 2865 | slab_map_pages(cachep, slabp, objp); |
1da177e4 | 2866 | |
a35afb83 | 2867 | cache_init_objs(cachep, slabp); |
1da177e4 LT |
2868 | |
2869 | if (local_flags & __GFP_WAIT) | |
2870 | local_irq_disable(); | |
2871 | check_irq_off(); | |
e498be7d | 2872 | spin_lock(&l3->list_lock); |
1da177e4 LT |
2873 | |
2874 | /* Make slab active. */ | |
e498be7d | 2875 | list_add_tail(&slabp->list, &(l3->slabs_free)); |
1da177e4 | 2876 | STATS_INC_GROWN(cachep); |
e498be7d CL |
2877 | l3->free_objects += cachep->num; |
2878 | spin_unlock(&l3->list_lock); | |
1da177e4 | 2879 | return 1; |
a737b3e2 | 2880 | opps1: |
1da177e4 | 2881 | kmem_freepages(cachep, objp); |
a737b3e2 | 2882 | failed: |
1da177e4 LT |
2883 | if (local_flags & __GFP_WAIT) |
2884 | local_irq_disable(); | |
2885 | return 0; | |
2886 | } | |
2887 | ||
2888 | #if DEBUG | |
2889 | ||
2890 | /* | |
2891 | * Perform extra freeing checks: | |
2892 | * - detect bad pointers. | |
2893 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2894 | */ |
2895 | static void kfree_debugcheck(const void *objp) | |
2896 | { | |
1da177e4 LT |
2897 | if (!virt_addr_valid(objp)) { |
2898 | printk(KERN_ERR "kfree_debugcheck: out of range ptr %lxh.\n", | |
b28a02de PE |
2899 | (unsigned long)objp); |
2900 | BUG(); | |
1da177e4 | 2901 | } |
1da177e4 LT |
2902 | } |
2903 | ||
58ce1fd5 PE |
2904 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2905 | { | |
b46b8f19 | 2906 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2907 | |
2908 | redzone1 = *dbg_redzone1(cache, obj); | |
2909 | redzone2 = *dbg_redzone2(cache, obj); | |
2910 | ||
2911 | /* | |
2912 | * Redzone is ok. | |
2913 | */ | |
2914 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2915 | return; | |
2916 | ||
2917 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2918 | slab_error(cache, "double free detected"); | |
2919 | else | |
2920 | slab_error(cache, "memory outside object was overwritten"); | |
2921 | ||
b46b8f19 | 2922 | printk(KERN_ERR "%p: redzone 1:0x%llx, redzone 2:0x%llx.\n", |
58ce1fd5 PE |
2923 | obj, redzone1, redzone2); |
2924 | } | |
2925 | ||
343e0d7a | 2926 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
b28a02de | 2927 | void *caller) |
1da177e4 LT |
2928 | { |
2929 | struct page *page; | |
2930 | unsigned int objnr; | |
2931 | struct slab *slabp; | |
2932 | ||
80cbd911 MW |
2933 | BUG_ON(virt_to_cache(objp) != cachep); |
2934 | ||
3dafccf2 | 2935 | objp -= obj_offset(cachep); |
1da177e4 | 2936 | kfree_debugcheck(objp); |
b49af68f | 2937 | page = virt_to_head_page(objp); |
1da177e4 | 2938 | |
065d41cb | 2939 | slabp = page_get_slab(page); |
1da177e4 LT |
2940 | |
2941 | if (cachep->flags & SLAB_RED_ZONE) { | |
58ce1fd5 | 2942 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2943 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2944 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2945 | } | |
2946 | if (cachep->flags & SLAB_STORE_USER) | |
2947 | *dbg_userword(cachep, objp) = caller; | |
2948 | ||
8fea4e96 | 2949 | objnr = obj_to_index(cachep, slabp, objp); |
1da177e4 LT |
2950 | |
2951 | BUG_ON(objnr >= cachep->num); | |
8fea4e96 | 2952 | BUG_ON(objp != index_to_obj(cachep, slabp, objnr)); |
1da177e4 | 2953 | |
871751e2 AV |
2954 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
2955 | slab_bufctl(slabp)[objnr] = BUFCTL_FREE; | |
2956 | #endif | |
1da177e4 LT |
2957 | if (cachep->flags & SLAB_POISON) { |
2958 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
a737b3e2 | 2959 | if ((cachep->buffer_size % PAGE_SIZE)==0 && OFF_SLAB(cachep)) { |
1da177e4 | 2960 | store_stackinfo(cachep, objp, (unsigned long)caller); |
b28a02de | 2961 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 2962 | cachep->buffer_size / PAGE_SIZE, 0); |
1da177e4 LT |
2963 | } else { |
2964 | poison_obj(cachep, objp, POISON_FREE); | |
2965 | } | |
2966 | #else | |
2967 | poison_obj(cachep, objp, POISON_FREE); | |
2968 | #endif | |
2969 | } | |
2970 | return objp; | |
2971 | } | |
2972 | ||
343e0d7a | 2973 | static void check_slabp(struct kmem_cache *cachep, struct slab *slabp) |
1da177e4 LT |
2974 | { |
2975 | kmem_bufctl_t i; | |
2976 | int entries = 0; | |
b28a02de | 2977 | |
1da177e4 LT |
2978 | /* Check slab's freelist to see if this obj is there. */ |
2979 | for (i = slabp->free; i != BUFCTL_END; i = slab_bufctl(slabp)[i]) { | |
2980 | entries++; | |
2981 | if (entries > cachep->num || i >= cachep->num) | |
2982 | goto bad; | |
2983 | } | |
2984 | if (entries != cachep->num - slabp->inuse) { | |
a737b3e2 AM |
2985 | bad: |
2986 | printk(KERN_ERR "slab: Internal list corruption detected in " | |
2987 | "cache '%s'(%d), slabp %p(%d). Hexdump:\n", | |
2988 | cachep->name, cachep->num, slabp, slabp->inuse); | |
b28a02de | 2989 | for (i = 0; |
264132bc | 2990 | i < sizeof(*slabp) + cachep->num * sizeof(kmem_bufctl_t); |
b28a02de | 2991 | i++) { |
a737b3e2 | 2992 | if (i % 16 == 0) |
1da177e4 | 2993 | printk("\n%03x:", i); |
b28a02de | 2994 | printk(" %02x", ((unsigned char *)slabp)[i]); |
1da177e4 LT |
2995 | } |
2996 | printk("\n"); | |
2997 | BUG(); | |
2998 | } | |
2999 | } | |
3000 | #else | |
3001 | #define kfree_debugcheck(x) do { } while(0) | |
3002 | #define cache_free_debugcheck(x,objp,z) (objp) | |
3003 | #define check_slabp(x,y) do { } while(0) | |
3004 | #endif | |
3005 | ||
343e0d7a | 3006 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
3007 | { |
3008 | int batchcount; | |
3009 | struct kmem_list3 *l3; | |
3010 | struct array_cache *ac; | |
1ca4cb24 PE |
3011 | int node; |
3012 | ||
6d2144d3 | 3013 | retry: |
1da177e4 | 3014 | check_irq_off(); |
7d6e6d09 | 3015 | node = numa_mem_id(); |
9a2dba4b | 3016 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
3017 | batchcount = ac->batchcount; |
3018 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
3019 | /* |
3020 | * If there was little recent activity on this cache, then | |
3021 | * perform only a partial refill. Otherwise we could generate | |
3022 | * refill bouncing. | |
1da177e4 LT |
3023 | */ |
3024 | batchcount = BATCHREFILL_LIMIT; | |
3025 | } | |
1ca4cb24 | 3026 | l3 = cachep->nodelists[node]; |
e498be7d CL |
3027 | |
3028 | BUG_ON(ac->avail > 0 || !l3); | |
3029 | spin_lock(&l3->list_lock); | |
1da177e4 | 3030 | |
3ded175a | 3031 | /* See if we can refill from the shared array */ |
44b57f1c NP |
3032 | if (l3->shared && transfer_objects(ac, l3->shared, batchcount)) { |
3033 | l3->shared->touched = 1; | |
3ded175a | 3034 | goto alloc_done; |
44b57f1c | 3035 | } |
3ded175a | 3036 | |
1da177e4 LT |
3037 | while (batchcount > 0) { |
3038 | struct list_head *entry; | |
3039 | struct slab *slabp; | |
3040 | /* Get slab alloc is to come from. */ | |
3041 | entry = l3->slabs_partial.next; | |
3042 | if (entry == &l3->slabs_partial) { | |
3043 | l3->free_touched = 1; | |
3044 | entry = l3->slabs_free.next; | |
3045 | if (entry == &l3->slabs_free) | |
3046 | goto must_grow; | |
3047 | } | |
3048 | ||
3049 | slabp = list_entry(entry, struct slab, list); | |
3050 | check_slabp(cachep, slabp); | |
3051 | check_spinlock_acquired(cachep); | |
714b8171 PE |
3052 | |
3053 | /* | |
3054 | * The slab was either on partial or free list so | |
3055 | * there must be at least one object available for | |
3056 | * allocation. | |
3057 | */ | |
249b9f33 | 3058 | BUG_ON(slabp->inuse >= cachep->num); |
714b8171 | 3059 | |
1da177e4 | 3060 | while (slabp->inuse < cachep->num && batchcount--) { |
1da177e4 LT |
3061 | STATS_INC_ALLOCED(cachep); |
3062 | STATS_INC_ACTIVE(cachep); | |
3063 | STATS_SET_HIGH(cachep); | |
3064 | ||
78d382d7 | 3065 | ac->entry[ac->avail++] = slab_get_obj(cachep, slabp, |
1ca4cb24 | 3066 | node); |
1da177e4 LT |
3067 | } |
3068 | check_slabp(cachep, slabp); | |
3069 | ||
3070 | /* move slabp to correct slabp list: */ | |
3071 | list_del(&slabp->list); | |
3072 | if (slabp->free == BUFCTL_END) | |
3073 | list_add(&slabp->list, &l3->slabs_full); | |
3074 | else | |
3075 | list_add(&slabp->list, &l3->slabs_partial); | |
3076 | } | |
3077 | ||
a737b3e2 | 3078 | must_grow: |
1da177e4 | 3079 | l3->free_objects -= ac->avail; |
a737b3e2 | 3080 | alloc_done: |
e498be7d | 3081 | spin_unlock(&l3->list_lock); |
1da177e4 LT |
3082 | |
3083 | if (unlikely(!ac->avail)) { | |
3084 | int x; | |
3c517a61 | 3085 | x = cache_grow(cachep, flags | GFP_THISNODE, node, NULL); |
e498be7d | 3086 | |
a737b3e2 | 3087 | /* cache_grow can reenable interrupts, then ac could change. */ |
9a2dba4b | 3088 | ac = cpu_cache_get(cachep); |
a737b3e2 | 3089 | if (!x && ac->avail == 0) /* no objects in sight? abort */ |
1da177e4 LT |
3090 | return NULL; |
3091 | ||
a737b3e2 | 3092 | if (!ac->avail) /* objects refilled by interrupt? */ |
1da177e4 LT |
3093 | goto retry; |
3094 | } | |
3095 | ac->touched = 1; | |
e498be7d | 3096 | return ac->entry[--ac->avail]; |
1da177e4 LT |
3097 | } |
3098 | ||
a737b3e2 AM |
3099 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3100 | gfp_t flags) | |
1da177e4 LT |
3101 | { |
3102 | might_sleep_if(flags & __GFP_WAIT); | |
3103 | #if DEBUG | |
3104 | kmem_flagcheck(cachep, flags); | |
3105 | #endif | |
3106 | } | |
3107 | ||
3108 | #if DEBUG | |
a737b3e2 AM |
3109 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
3110 | gfp_t flags, void *objp, void *caller) | |
1da177e4 | 3111 | { |
b28a02de | 3112 | if (!objp) |
1da177e4 | 3113 | return objp; |
b28a02de | 3114 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3115 | #ifdef CONFIG_DEBUG_PAGEALLOC |
3dafccf2 | 3116 | if ((cachep->buffer_size % PAGE_SIZE) == 0 && OFF_SLAB(cachep)) |
b28a02de | 3117 | kernel_map_pages(virt_to_page(objp), |
3dafccf2 | 3118 | cachep->buffer_size / PAGE_SIZE, 1); |
1da177e4 LT |
3119 | else |
3120 | check_poison_obj(cachep, objp); | |
3121 | #else | |
3122 | check_poison_obj(cachep, objp); | |
3123 | #endif | |
3124 | poison_obj(cachep, objp, POISON_INUSE); | |
3125 | } | |
3126 | if (cachep->flags & SLAB_STORE_USER) | |
3127 | *dbg_userword(cachep, objp) = caller; | |
3128 | ||
3129 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3130 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3131 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
3132 | slab_error(cachep, "double free, or memory outside" | |
3133 | " object was overwritten"); | |
b28a02de | 3134 | printk(KERN_ERR |
b46b8f19 | 3135 | "%p: redzone 1:0x%llx, redzone 2:0x%llx\n", |
a737b3e2 AM |
3136 | objp, *dbg_redzone1(cachep, objp), |
3137 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3138 | } |
3139 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3140 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3141 | } | |
871751e2 AV |
3142 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
3143 | { | |
3144 | struct slab *slabp; | |
3145 | unsigned objnr; | |
3146 | ||
b49af68f | 3147 | slabp = page_get_slab(virt_to_head_page(objp)); |
871751e2 AV |
3148 | objnr = (unsigned)(objp - slabp->s_mem) / cachep->buffer_size; |
3149 | slab_bufctl(slabp)[objnr] = BUFCTL_ACTIVE; | |
3150 | } | |
3151 | #endif | |
3dafccf2 | 3152 | objp += obj_offset(cachep); |
4f104934 | 3153 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3154 | cachep->ctor(objp); |
a44b56d3 KH |
3155 | #if ARCH_SLAB_MINALIGN |
3156 | if ((u32)objp & (ARCH_SLAB_MINALIGN-1)) { | |
3157 | printk(KERN_ERR "0x%p: not aligned to ARCH_SLAB_MINALIGN=%d\n", | |
3158 | objp, ARCH_SLAB_MINALIGN); | |
3159 | } | |
3160 | #endif | |
1da177e4 LT |
3161 | return objp; |
3162 | } | |
3163 | #else | |
3164 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3165 | #endif | |
3166 | ||
773ff60e | 3167 | static bool slab_should_failslab(struct kmem_cache *cachep, gfp_t flags) |
8a8b6502 AM |
3168 | { |
3169 | if (cachep == &cache_cache) | |
773ff60e | 3170 | return false; |
8a8b6502 | 3171 | |
4c13dd3b | 3172 | return should_failslab(obj_size(cachep), flags, cachep->flags); |
8a8b6502 AM |
3173 | } |
3174 | ||
343e0d7a | 3175 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3176 | { |
b28a02de | 3177 | void *objp; |
1da177e4 LT |
3178 | struct array_cache *ac; |
3179 | ||
5c382300 | 3180 | check_irq_off(); |
8a8b6502 | 3181 | |
9a2dba4b | 3182 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
3183 | if (likely(ac->avail)) { |
3184 | STATS_INC_ALLOCHIT(cachep); | |
3185 | ac->touched = 1; | |
e498be7d | 3186 | objp = ac->entry[--ac->avail]; |
1da177e4 LT |
3187 | } else { |
3188 | STATS_INC_ALLOCMISS(cachep); | |
3189 | objp = cache_alloc_refill(cachep, flags); | |
ddbf2e83 O |
3190 | /* |
3191 | * the 'ac' may be updated by cache_alloc_refill(), | |
3192 | * and kmemleak_erase() requires its correct value. | |
3193 | */ | |
3194 | ac = cpu_cache_get(cachep); | |
1da177e4 | 3195 | } |
d5cff635 CM |
3196 | /* |
3197 | * To avoid a false negative, if an object that is in one of the | |
3198 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3199 | * treat the array pointers as a reference to the object. | |
3200 | */ | |
f3d8b53a O |
3201 | if (objp) |
3202 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3203 | return objp; |
3204 | } | |
3205 | ||
e498be7d | 3206 | #ifdef CONFIG_NUMA |
c61afb18 | 3207 | /* |
b2455396 | 3208 | * Try allocating on another node if PF_SPREAD_SLAB|PF_MEMPOLICY. |
c61afb18 PJ |
3209 | * |
3210 | * If we are in_interrupt, then process context, including cpusets and | |
3211 | * mempolicy, may not apply and should not be used for allocation policy. | |
3212 | */ | |
3213 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3214 | { | |
3215 | int nid_alloc, nid_here; | |
3216 | ||
765c4507 | 3217 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3218 | return NULL; |
7d6e6d09 | 3219 | nid_alloc = nid_here = numa_mem_id(); |
c0ff7453 | 3220 | get_mems_allowed(); |
c61afb18 | 3221 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3222 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 PJ |
3223 | else if (current->mempolicy) |
3224 | nid_alloc = slab_node(current->mempolicy); | |
c0ff7453 | 3225 | put_mems_allowed(); |
c61afb18 | 3226 | if (nid_alloc != nid_here) |
8b98c169 | 3227 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3228 | return NULL; |
3229 | } | |
3230 | ||
765c4507 CL |
3231 | /* |
3232 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 CL |
3233 | * certain node and fall back is permitted. First we scan all the |
3234 | * available nodelists for available objects. If that fails then we | |
3235 | * perform an allocation without specifying a node. This allows the page | |
3236 | * allocator to do its reclaim / fallback magic. We then insert the | |
3237 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3238 | */ |
8c8cc2c1 | 3239 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3240 | { |
8c8cc2c1 PE |
3241 | struct zonelist *zonelist; |
3242 | gfp_t local_flags; | |
dd1a239f | 3243 | struct zoneref *z; |
54a6eb5c MG |
3244 | struct zone *zone; |
3245 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3246 | void *obj = NULL; |
3c517a61 | 3247 | int nid; |
8c8cc2c1 PE |
3248 | |
3249 | if (flags & __GFP_THISNODE) | |
3250 | return NULL; | |
3251 | ||
c0ff7453 | 3252 | get_mems_allowed(); |
0e88460d | 3253 | zonelist = node_zonelist(slab_node(current->mempolicy), flags); |
6cb06229 | 3254 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
765c4507 | 3255 | |
3c517a61 CL |
3256 | retry: |
3257 | /* | |
3258 | * Look through allowed nodes for objects available | |
3259 | * from existing per node queues. | |
3260 | */ | |
54a6eb5c MG |
3261 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3262 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3263 | |
54a6eb5c | 3264 | if (cpuset_zone_allowed_hardwall(zone, flags) && |
3c517a61 | 3265 | cache->nodelists[nid] && |
481c5346 | 3266 | cache->nodelists[nid]->free_objects) { |
3c517a61 CL |
3267 | obj = ____cache_alloc_node(cache, |
3268 | flags | GFP_THISNODE, nid); | |
481c5346 CL |
3269 | if (obj) |
3270 | break; | |
3271 | } | |
3c517a61 CL |
3272 | } |
3273 | ||
cfce6604 | 3274 | if (!obj) { |
3c517a61 CL |
3275 | /* |
3276 | * This allocation will be performed within the constraints | |
3277 | * of the current cpuset / memory policy requirements. | |
3278 | * We may trigger various forms of reclaim on the allowed | |
3279 | * set and go into memory reserves if necessary. | |
3280 | */ | |
dd47ea75 CL |
3281 | if (local_flags & __GFP_WAIT) |
3282 | local_irq_enable(); | |
3283 | kmem_flagcheck(cache, flags); | |
7d6e6d09 | 3284 | obj = kmem_getpages(cache, local_flags, numa_mem_id()); |
dd47ea75 CL |
3285 | if (local_flags & __GFP_WAIT) |
3286 | local_irq_disable(); | |
3c517a61 CL |
3287 | if (obj) { |
3288 | /* | |
3289 | * Insert into the appropriate per node queues | |
3290 | */ | |
3291 | nid = page_to_nid(virt_to_page(obj)); | |
3292 | if (cache_grow(cache, flags, nid, obj)) { | |
3293 | obj = ____cache_alloc_node(cache, | |
3294 | flags | GFP_THISNODE, nid); | |
3295 | if (!obj) | |
3296 | /* | |
3297 | * Another processor may allocate the | |
3298 | * objects in the slab since we are | |
3299 | * not holding any locks. | |
3300 | */ | |
3301 | goto retry; | |
3302 | } else { | |
b6a60451 | 3303 | /* cache_grow already freed obj */ |
3c517a61 CL |
3304 | obj = NULL; |
3305 | } | |
3306 | } | |
aedb0eb1 | 3307 | } |
c0ff7453 | 3308 | put_mems_allowed(); |
765c4507 CL |
3309 | return obj; |
3310 | } | |
3311 | ||
e498be7d CL |
3312 | /* |
3313 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3314 | */ |
8b98c169 | 3315 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3316 | int nodeid) |
e498be7d CL |
3317 | { |
3318 | struct list_head *entry; | |
b28a02de PE |
3319 | struct slab *slabp; |
3320 | struct kmem_list3 *l3; | |
3321 | void *obj; | |
b28a02de PE |
3322 | int x; |
3323 | ||
3324 | l3 = cachep->nodelists[nodeid]; | |
3325 | BUG_ON(!l3); | |
3326 | ||
a737b3e2 | 3327 | retry: |
ca3b9b91 | 3328 | check_irq_off(); |
b28a02de PE |
3329 | spin_lock(&l3->list_lock); |
3330 | entry = l3->slabs_partial.next; | |
3331 | if (entry == &l3->slabs_partial) { | |
3332 | l3->free_touched = 1; | |
3333 | entry = l3->slabs_free.next; | |
3334 | if (entry == &l3->slabs_free) | |
3335 | goto must_grow; | |
3336 | } | |
3337 | ||
3338 | slabp = list_entry(entry, struct slab, list); | |
3339 | check_spinlock_acquired_node(cachep, nodeid); | |
3340 | check_slabp(cachep, slabp); | |
3341 | ||
3342 | STATS_INC_NODEALLOCS(cachep); | |
3343 | STATS_INC_ACTIVE(cachep); | |
3344 | STATS_SET_HIGH(cachep); | |
3345 | ||
3346 | BUG_ON(slabp->inuse == cachep->num); | |
3347 | ||
78d382d7 | 3348 | obj = slab_get_obj(cachep, slabp, nodeid); |
b28a02de PE |
3349 | check_slabp(cachep, slabp); |
3350 | l3->free_objects--; | |
3351 | /* move slabp to correct slabp list: */ | |
3352 | list_del(&slabp->list); | |
3353 | ||
a737b3e2 | 3354 | if (slabp->free == BUFCTL_END) |
b28a02de | 3355 | list_add(&slabp->list, &l3->slabs_full); |
a737b3e2 | 3356 | else |
b28a02de | 3357 | list_add(&slabp->list, &l3->slabs_partial); |
e498be7d | 3358 | |
b28a02de PE |
3359 | spin_unlock(&l3->list_lock); |
3360 | goto done; | |
e498be7d | 3361 | |
a737b3e2 | 3362 | must_grow: |
b28a02de | 3363 | spin_unlock(&l3->list_lock); |
3c517a61 | 3364 | x = cache_grow(cachep, flags | GFP_THISNODE, nodeid, NULL); |
765c4507 CL |
3365 | if (x) |
3366 | goto retry; | |
1da177e4 | 3367 | |
8c8cc2c1 | 3368 | return fallback_alloc(cachep, flags); |
e498be7d | 3369 | |
a737b3e2 | 3370 | done: |
b28a02de | 3371 | return obj; |
e498be7d | 3372 | } |
8c8cc2c1 PE |
3373 | |
3374 | /** | |
3375 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3376 | * @cachep: The cache to allocate from. | |
3377 | * @flags: See kmalloc(). | |
3378 | * @nodeid: node number of the target node. | |
3379 | * @caller: return address of caller, used for debug information | |
3380 | * | |
3381 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3382 | * node, which can improve the performance for cpu bound structures. | |
3383 | * | |
3384 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3385 | */ | |
3386 | static __always_inline void * | |
3387 | __cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, | |
3388 | void *caller) | |
3389 | { | |
3390 | unsigned long save_flags; | |
3391 | void *ptr; | |
7d6e6d09 | 3392 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3393 | |
dcce284a | 3394 | flags &= gfp_allowed_mask; |
7e85ee0c | 3395 | |
cf40bd16 NP |
3396 | lockdep_trace_alloc(flags); |
3397 | ||
773ff60e | 3398 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3399 | return NULL; |
3400 | ||
8c8cc2c1 PE |
3401 | cache_alloc_debugcheck_before(cachep, flags); |
3402 | local_irq_save(save_flags); | |
3403 | ||
8e15b79c | 3404 | if (nodeid == -1) |
7d6e6d09 | 3405 | nodeid = slab_node; |
8c8cc2c1 PE |
3406 | |
3407 | if (unlikely(!cachep->nodelists[nodeid])) { | |
3408 | /* Node not bootstrapped yet */ | |
3409 | ptr = fallback_alloc(cachep, flags); | |
3410 | goto out; | |
3411 | } | |
3412 | ||
7d6e6d09 | 3413 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3414 | /* |
3415 | * Use the locally cached objects if possible. | |
3416 | * However ____cache_alloc does not allow fallback | |
3417 | * to other nodes. It may fail while we still have | |
3418 | * objects on other nodes available. | |
3419 | */ | |
3420 | ptr = ____cache_alloc(cachep, flags); | |
3421 | if (ptr) | |
3422 | goto out; | |
3423 | } | |
3424 | /* ___cache_alloc_node can fall back to other nodes */ | |
3425 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3426 | out: | |
3427 | local_irq_restore(save_flags); | |
3428 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
d5cff635 CM |
3429 | kmemleak_alloc_recursive(ptr, obj_size(cachep), 1, cachep->flags, |
3430 | flags); | |
8c8cc2c1 | 3431 | |
c175eea4 PE |
3432 | if (likely(ptr)) |
3433 | kmemcheck_slab_alloc(cachep, flags, ptr, obj_size(cachep)); | |
3434 | ||
d07dbea4 CL |
3435 | if (unlikely((flags & __GFP_ZERO) && ptr)) |
3436 | memset(ptr, 0, obj_size(cachep)); | |
3437 | ||
8c8cc2c1 PE |
3438 | return ptr; |
3439 | } | |
3440 | ||
3441 | static __always_inline void * | |
3442 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3443 | { | |
3444 | void *objp; | |
3445 | ||
3446 | if (unlikely(current->flags & (PF_SPREAD_SLAB | PF_MEMPOLICY))) { | |
3447 | objp = alternate_node_alloc(cache, flags); | |
3448 | if (objp) | |
3449 | goto out; | |
3450 | } | |
3451 | objp = ____cache_alloc(cache, flags); | |
3452 | ||
3453 | /* | |
3454 | * We may just have run out of memory on the local node. | |
3455 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3456 | */ | |
7d6e6d09 LS |
3457 | if (!objp) |
3458 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3459 | |
3460 | out: | |
3461 | return objp; | |
3462 | } | |
3463 | #else | |
3464 | ||
3465 | static __always_inline void * | |
3466 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3467 | { | |
3468 | return ____cache_alloc(cachep, flags); | |
3469 | } | |
3470 | ||
3471 | #endif /* CONFIG_NUMA */ | |
3472 | ||
3473 | static __always_inline void * | |
3474 | __cache_alloc(struct kmem_cache *cachep, gfp_t flags, void *caller) | |
3475 | { | |
3476 | unsigned long save_flags; | |
3477 | void *objp; | |
3478 | ||
dcce284a | 3479 | flags &= gfp_allowed_mask; |
7e85ee0c | 3480 | |
cf40bd16 NP |
3481 | lockdep_trace_alloc(flags); |
3482 | ||
773ff60e | 3483 | if (slab_should_failslab(cachep, flags)) |
824ebef1 AM |
3484 | return NULL; |
3485 | ||
8c8cc2c1 PE |
3486 | cache_alloc_debugcheck_before(cachep, flags); |
3487 | local_irq_save(save_flags); | |
3488 | objp = __do_cache_alloc(cachep, flags); | |
3489 | local_irq_restore(save_flags); | |
3490 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
d5cff635 CM |
3491 | kmemleak_alloc_recursive(objp, obj_size(cachep), 1, cachep->flags, |
3492 | flags); | |
8c8cc2c1 PE |
3493 | prefetchw(objp); |
3494 | ||
c175eea4 PE |
3495 | if (likely(objp)) |
3496 | kmemcheck_slab_alloc(cachep, flags, objp, obj_size(cachep)); | |
3497 | ||
d07dbea4 CL |
3498 | if (unlikely((flags & __GFP_ZERO) && objp)) |
3499 | memset(objp, 0, obj_size(cachep)); | |
3500 | ||
8c8cc2c1 PE |
3501 | return objp; |
3502 | } | |
e498be7d CL |
3503 | |
3504 | /* | |
3505 | * Caller needs to acquire correct kmem_list's list_lock | |
3506 | */ | |
343e0d7a | 3507 | static void free_block(struct kmem_cache *cachep, void **objpp, int nr_objects, |
b28a02de | 3508 | int node) |
1da177e4 LT |
3509 | { |
3510 | int i; | |
e498be7d | 3511 | struct kmem_list3 *l3; |
1da177e4 LT |
3512 | |
3513 | for (i = 0; i < nr_objects; i++) { | |
3514 | void *objp = objpp[i]; | |
3515 | struct slab *slabp; | |
1da177e4 | 3516 | |
6ed5eb22 | 3517 | slabp = virt_to_slab(objp); |
ff69416e | 3518 | l3 = cachep->nodelists[node]; |
1da177e4 | 3519 | list_del(&slabp->list); |
ff69416e | 3520 | check_spinlock_acquired_node(cachep, node); |
1da177e4 | 3521 | check_slabp(cachep, slabp); |
78d382d7 | 3522 | slab_put_obj(cachep, slabp, objp, node); |
1da177e4 | 3523 | STATS_DEC_ACTIVE(cachep); |
e498be7d | 3524 | l3->free_objects++; |
1da177e4 LT |
3525 | check_slabp(cachep, slabp); |
3526 | ||
3527 | /* fixup slab chains */ | |
3528 | if (slabp->inuse == 0) { | |
e498be7d CL |
3529 | if (l3->free_objects > l3->free_limit) { |
3530 | l3->free_objects -= cachep->num; | |
e5ac9c5a RT |
3531 | /* No need to drop any previously held |
3532 | * lock here, even if we have a off-slab slab | |
3533 | * descriptor it is guaranteed to come from | |
3534 | * a different cache, refer to comments before | |
3535 | * alloc_slabmgmt. | |
3536 | */ | |
1da177e4 LT |
3537 | slab_destroy(cachep, slabp); |
3538 | } else { | |
e498be7d | 3539 | list_add(&slabp->list, &l3->slabs_free); |
1da177e4 LT |
3540 | } |
3541 | } else { | |
3542 | /* Unconditionally move a slab to the end of the | |
3543 | * partial list on free - maximum time for the | |
3544 | * other objects to be freed, too. | |
3545 | */ | |
e498be7d | 3546 | list_add_tail(&slabp->list, &l3->slabs_partial); |
1da177e4 LT |
3547 | } |
3548 | } | |
3549 | } | |
3550 | ||
343e0d7a | 3551 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3552 | { |
3553 | int batchcount; | |
e498be7d | 3554 | struct kmem_list3 *l3; |
7d6e6d09 | 3555 | int node = numa_mem_id(); |
1da177e4 LT |
3556 | |
3557 | batchcount = ac->batchcount; | |
3558 | #if DEBUG | |
3559 | BUG_ON(!batchcount || batchcount > ac->avail); | |
3560 | #endif | |
3561 | check_irq_off(); | |
ff69416e | 3562 | l3 = cachep->nodelists[node]; |
873623df | 3563 | spin_lock(&l3->list_lock); |
e498be7d CL |
3564 | if (l3->shared) { |
3565 | struct array_cache *shared_array = l3->shared; | |
b28a02de | 3566 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3567 | if (max) { |
3568 | if (batchcount > max) | |
3569 | batchcount = max; | |
e498be7d | 3570 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3571 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3572 | shared_array->avail += batchcount; |
3573 | goto free_done; | |
3574 | } | |
3575 | } | |
3576 | ||
ff69416e | 3577 | free_block(cachep, ac->entry, batchcount, node); |
a737b3e2 | 3578 | free_done: |
1da177e4 LT |
3579 | #if STATS |
3580 | { | |
3581 | int i = 0; | |
3582 | struct list_head *p; | |
3583 | ||
e498be7d CL |
3584 | p = l3->slabs_free.next; |
3585 | while (p != &(l3->slabs_free)) { | |
1da177e4 LT |
3586 | struct slab *slabp; |
3587 | ||
3588 | slabp = list_entry(p, struct slab, list); | |
3589 | BUG_ON(slabp->inuse); | |
3590 | ||
3591 | i++; | |
3592 | p = p->next; | |
3593 | } | |
3594 | STATS_SET_FREEABLE(cachep, i); | |
3595 | } | |
3596 | #endif | |
e498be7d | 3597 | spin_unlock(&l3->list_lock); |
1da177e4 | 3598 | ac->avail -= batchcount; |
a737b3e2 | 3599 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3600 | } |
3601 | ||
3602 | /* | |
a737b3e2 AM |
3603 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3604 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3605 | */ |
873623df | 3606 | static inline void __cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 | 3607 | { |
9a2dba4b | 3608 | struct array_cache *ac = cpu_cache_get(cachep); |
1da177e4 LT |
3609 | |
3610 | check_irq_off(); | |
d5cff635 | 3611 | kmemleak_free_recursive(objp, cachep->flags); |
1da177e4 LT |
3612 | objp = cache_free_debugcheck(cachep, objp, __builtin_return_address(0)); |
3613 | ||
c175eea4 PE |
3614 | kmemcheck_slab_free(cachep, objp, obj_size(cachep)); |
3615 | ||
1807a1aa SS |
3616 | /* |
3617 | * Skip calling cache_free_alien() when the platform is not numa. | |
3618 | * This will avoid cache misses that happen while accessing slabp (which | |
3619 | * is per page memory reference) to get nodeid. Instead use a global | |
3620 | * variable to skip the call, which is mostly likely to be present in | |
3621 | * the cache. | |
3622 | */ | |
b6e68bc1 | 3623 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3624 | return; |
3625 | ||
1da177e4 LT |
3626 | if (likely(ac->avail < ac->limit)) { |
3627 | STATS_INC_FREEHIT(cachep); | |
e498be7d | 3628 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3629 | return; |
3630 | } else { | |
3631 | STATS_INC_FREEMISS(cachep); | |
3632 | cache_flusharray(cachep, ac); | |
e498be7d | 3633 | ac->entry[ac->avail++] = objp; |
1da177e4 LT |
3634 | } |
3635 | } | |
3636 | ||
3637 | /** | |
3638 | * kmem_cache_alloc - Allocate an object | |
3639 | * @cachep: The cache to allocate from. | |
3640 | * @flags: See kmalloc(). | |
3641 | * | |
3642 | * Allocate an object from this cache. The flags are only relevant | |
3643 | * if the cache has no available objects. | |
3644 | */ | |
343e0d7a | 3645 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3646 | { |
36555751 EGM |
3647 | void *ret = __cache_alloc(cachep, flags, __builtin_return_address(0)); |
3648 | ||
ca2b84cb EGM |
3649 | trace_kmem_cache_alloc(_RET_IP_, ret, |
3650 | obj_size(cachep), cachep->buffer_size, flags); | |
36555751 EGM |
3651 | |
3652 | return ret; | |
1da177e4 LT |
3653 | } |
3654 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3655 | ||
0f24f128 | 3656 | #ifdef CONFIG_TRACING |
36555751 EGM |
3657 | void *kmem_cache_alloc_notrace(struct kmem_cache *cachep, gfp_t flags) |
3658 | { | |
3659 | return __cache_alloc(cachep, flags, __builtin_return_address(0)); | |
3660 | } | |
3661 | EXPORT_SYMBOL(kmem_cache_alloc_notrace); | |
3662 | #endif | |
3663 | ||
1da177e4 | 3664 | /** |
7682486b | 3665 | * kmem_ptr_validate - check if an untrusted pointer might be a slab entry. |
1da177e4 LT |
3666 | * @cachep: the cache we're checking against |
3667 | * @ptr: pointer to validate | |
3668 | * | |
7682486b | 3669 | * This verifies that the untrusted pointer looks sane; |
1da177e4 LT |
3670 | * it is _not_ a guarantee that the pointer is actually |
3671 | * part of the slab cache in question, but it at least | |
3672 | * validates that the pointer can be dereferenced and | |
3673 | * looks half-way sane. | |
3674 | * | |
3675 | * Currently only used for dentry validation. | |
3676 | */ | |
b7f869a2 | 3677 | int kmem_ptr_validate(struct kmem_cache *cachep, const void *ptr) |
1da177e4 | 3678 | { |
3dafccf2 | 3679 | unsigned long size = cachep->buffer_size; |
1da177e4 LT |
3680 | struct page *page; |
3681 | ||
fc1c1833 | 3682 | if (unlikely(!kern_ptr_validate(ptr, size))) |
1da177e4 LT |
3683 | goto out; |
3684 | page = virt_to_page(ptr); | |
3685 | if (unlikely(!PageSlab(page))) | |
3686 | goto out; | |
065d41cb | 3687 | if (unlikely(page_get_cache(page) != cachep)) |
1da177e4 LT |
3688 | goto out; |
3689 | return 1; | |
a737b3e2 | 3690 | out: |
1da177e4 LT |
3691 | return 0; |
3692 | } | |
3693 | ||
3694 | #ifdef CONFIG_NUMA | |
8b98c169 CH |
3695 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3696 | { | |
36555751 EGM |
3697 | void *ret = __cache_alloc_node(cachep, flags, nodeid, |
3698 | __builtin_return_address(0)); | |
3699 | ||
ca2b84cb EGM |
3700 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3701 | obj_size(cachep), cachep->buffer_size, | |
3702 | flags, nodeid); | |
36555751 EGM |
3703 | |
3704 | return ret; | |
8b98c169 | 3705 | } |
1da177e4 LT |
3706 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3707 | ||
0f24f128 | 3708 | #ifdef CONFIG_TRACING |
36555751 EGM |
3709 | void *kmem_cache_alloc_node_notrace(struct kmem_cache *cachep, |
3710 | gfp_t flags, | |
3711 | int nodeid) | |
3712 | { | |
3713 | return __cache_alloc_node(cachep, flags, nodeid, | |
3714 | __builtin_return_address(0)); | |
3715 | } | |
3716 | EXPORT_SYMBOL(kmem_cache_alloc_node_notrace); | |
3717 | #endif | |
3718 | ||
8b98c169 CH |
3719 | static __always_inline void * |
3720 | __do_kmalloc_node(size_t size, gfp_t flags, int node, void *caller) | |
97e2bde4 | 3721 | { |
343e0d7a | 3722 | struct kmem_cache *cachep; |
36555751 | 3723 | void *ret; |
97e2bde4 MS |
3724 | |
3725 | cachep = kmem_find_general_cachep(size, flags); | |
6cb8f913 CL |
3726 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3727 | return cachep; | |
36555751 EGM |
3728 | ret = kmem_cache_alloc_node_notrace(cachep, flags, node); |
3729 | ||
ca2b84cb EGM |
3730 | trace_kmalloc_node((unsigned long) caller, ret, |
3731 | size, cachep->buffer_size, flags, node); | |
36555751 EGM |
3732 | |
3733 | return ret; | |
97e2bde4 | 3734 | } |
8b98c169 | 3735 | |
0bb38a5c | 3736 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
8b98c169 CH |
3737 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3738 | { | |
3739 | return __do_kmalloc_node(size, flags, node, | |
3740 | __builtin_return_address(0)); | |
3741 | } | |
dbe5e69d | 3742 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3743 | |
3744 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3745 | int node, unsigned long caller) |
8b98c169 | 3746 | { |
ce71e27c | 3747 | return __do_kmalloc_node(size, flags, node, (void *)caller); |
8b98c169 CH |
3748 | } |
3749 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
3750 | #else | |
3751 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
3752 | { | |
3753 | return __do_kmalloc_node(size, flags, node, NULL); | |
3754 | } | |
3755 | EXPORT_SYMBOL(__kmalloc_node); | |
0bb38a5c | 3756 | #endif /* CONFIG_DEBUG_SLAB || CONFIG_TRACING */ |
8b98c169 | 3757 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3758 | |
3759 | /** | |
800590f5 | 3760 | * __do_kmalloc - allocate memory |
1da177e4 | 3761 | * @size: how many bytes of memory are required. |
800590f5 | 3762 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3763 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3764 | */ |
7fd6b141 PE |
3765 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
3766 | void *caller) | |
1da177e4 | 3767 | { |
343e0d7a | 3768 | struct kmem_cache *cachep; |
36555751 | 3769 | void *ret; |
1da177e4 | 3770 | |
97e2bde4 MS |
3771 | /* If you want to save a few bytes .text space: replace |
3772 | * __ with kmem_. | |
3773 | * Then kmalloc uses the uninlined functions instead of the inline | |
3774 | * functions. | |
3775 | */ | |
3776 | cachep = __find_general_cachep(size, flags); | |
a5c96d8a LT |
3777 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3778 | return cachep; | |
36555751 EGM |
3779 | ret = __cache_alloc(cachep, flags, caller); |
3780 | ||
ca2b84cb EGM |
3781 | trace_kmalloc((unsigned long) caller, ret, |
3782 | size, cachep->buffer_size, flags); | |
36555751 EGM |
3783 | |
3784 | return ret; | |
7fd6b141 PE |
3785 | } |
3786 | ||
7fd6b141 | 3787 | |
0bb38a5c | 3788 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_TRACING) |
7fd6b141 PE |
3789 | void *__kmalloc(size_t size, gfp_t flags) |
3790 | { | |
871751e2 | 3791 | return __do_kmalloc(size, flags, __builtin_return_address(0)); |
1da177e4 LT |
3792 | } |
3793 | EXPORT_SYMBOL(__kmalloc); | |
3794 | ||
ce71e27c | 3795 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3796 | { |
ce71e27c | 3797 | return __do_kmalloc(size, flags, (void *)caller); |
7fd6b141 PE |
3798 | } |
3799 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea CH |
3800 | |
3801 | #else | |
3802 | void *__kmalloc(size_t size, gfp_t flags) | |
3803 | { | |
3804 | return __do_kmalloc(size, flags, NULL); | |
3805 | } | |
3806 | EXPORT_SYMBOL(__kmalloc); | |
7fd6b141 PE |
3807 | #endif |
3808 | ||
1da177e4 LT |
3809 | /** |
3810 | * kmem_cache_free - Deallocate an object | |
3811 | * @cachep: The cache the allocation was from. | |
3812 | * @objp: The previously allocated object. | |
3813 | * | |
3814 | * Free an object which was previously allocated from this | |
3815 | * cache. | |
3816 | */ | |
343e0d7a | 3817 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3818 | { |
3819 | unsigned long flags; | |
3820 | ||
3821 | local_irq_save(flags); | |
898552c9 | 3822 | debug_check_no_locks_freed(objp, obj_size(cachep)); |
3ac7fe5a TG |
3823 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
3824 | debug_check_no_obj_freed(objp, obj_size(cachep)); | |
873623df | 3825 | __cache_free(cachep, objp); |
1da177e4 | 3826 | local_irq_restore(flags); |
36555751 | 3827 | |
ca2b84cb | 3828 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3829 | } |
3830 | EXPORT_SYMBOL(kmem_cache_free); | |
3831 | ||
1da177e4 LT |
3832 | /** |
3833 | * kfree - free previously allocated memory | |
3834 | * @objp: pointer returned by kmalloc. | |
3835 | * | |
80e93eff PE |
3836 | * If @objp is NULL, no operation is performed. |
3837 | * | |
1da177e4 LT |
3838 | * Don't free memory not originally allocated by kmalloc() |
3839 | * or you will run into trouble. | |
3840 | */ | |
3841 | void kfree(const void *objp) | |
3842 | { | |
343e0d7a | 3843 | struct kmem_cache *c; |
1da177e4 LT |
3844 | unsigned long flags; |
3845 | ||
2121db74 PE |
3846 | trace_kfree(_RET_IP_, objp); |
3847 | ||
6cb8f913 | 3848 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3849 | return; |
3850 | local_irq_save(flags); | |
3851 | kfree_debugcheck(objp); | |
6ed5eb22 | 3852 | c = virt_to_cache(objp); |
f9b8404c | 3853 | debug_check_no_locks_freed(objp, obj_size(c)); |
3ac7fe5a | 3854 | debug_check_no_obj_freed(objp, obj_size(c)); |
873623df | 3855 | __cache_free(c, (void *)objp); |
1da177e4 LT |
3856 | local_irq_restore(flags); |
3857 | } | |
3858 | EXPORT_SYMBOL(kfree); | |
3859 | ||
343e0d7a | 3860 | unsigned int kmem_cache_size(struct kmem_cache *cachep) |
1da177e4 | 3861 | { |
3dafccf2 | 3862 | return obj_size(cachep); |
1da177e4 LT |
3863 | } |
3864 | EXPORT_SYMBOL(kmem_cache_size); | |
3865 | ||
343e0d7a | 3866 | const char *kmem_cache_name(struct kmem_cache *cachep) |
1944972d ACM |
3867 | { |
3868 | return cachep->name; | |
3869 | } | |
3870 | EXPORT_SYMBOL_GPL(kmem_cache_name); | |
3871 | ||
e498be7d | 3872 | /* |
183ff22b | 3873 | * This initializes kmem_list3 or resizes various caches for all nodes. |
e498be7d | 3874 | */ |
83b519e8 | 3875 | static int alloc_kmemlist(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d CL |
3876 | { |
3877 | int node; | |
3878 | struct kmem_list3 *l3; | |
cafeb02e | 3879 | struct array_cache *new_shared; |
3395ee05 | 3880 | struct array_cache **new_alien = NULL; |
e498be7d | 3881 | |
9c09a95c | 3882 | for_each_online_node(node) { |
cafeb02e | 3883 | |
3395ee05 | 3884 | if (use_alien_caches) { |
83b519e8 | 3885 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); |
3395ee05 PM |
3886 | if (!new_alien) |
3887 | goto fail; | |
3888 | } | |
cafeb02e | 3889 | |
63109846 ED |
3890 | new_shared = NULL; |
3891 | if (cachep->shared) { | |
3892 | new_shared = alloc_arraycache(node, | |
0718dc2a | 3893 | cachep->shared*cachep->batchcount, |
83b519e8 | 3894 | 0xbaadf00d, gfp); |
63109846 ED |
3895 | if (!new_shared) { |
3896 | free_alien_cache(new_alien); | |
3897 | goto fail; | |
3898 | } | |
0718dc2a | 3899 | } |
cafeb02e | 3900 | |
a737b3e2 AM |
3901 | l3 = cachep->nodelists[node]; |
3902 | if (l3) { | |
cafeb02e CL |
3903 | struct array_cache *shared = l3->shared; |
3904 | ||
e498be7d CL |
3905 | spin_lock_irq(&l3->list_lock); |
3906 | ||
cafeb02e | 3907 | if (shared) |
0718dc2a CL |
3908 | free_block(cachep, shared->entry, |
3909 | shared->avail, node); | |
e498be7d | 3910 | |
cafeb02e CL |
3911 | l3->shared = new_shared; |
3912 | if (!l3->alien) { | |
e498be7d CL |
3913 | l3->alien = new_alien; |
3914 | new_alien = NULL; | |
3915 | } | |
b28a02de | 3916 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3917 | cachep->batchcount + cachep->num; |
e498be7d | 3918 | spin_unlock_irq(&l3->list_lock); |
cafeb02e | 3919 | kfree(shared); |
e498be7d CL |
3920 | free_alien_cache(new_alien); |
3921 | continue; | |
3922 | } | |
83b519e8 | 3923 | l3 = kmalloc_node(sizeof(struct kmem_list3), gfp, node); |
0718dc2a CL |
3924 | if (!l3) { |
3925 | free_alien_cache(new_alien); | |
3926 | kfree(new_shared); | |
e498be7d | 3927 | goto fail; |
0718dc2a | 3928 | } |
e498be7d CL |
3929 | |
3930 | kmem_list3_init(l3); | |
3931 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3 + | |
a737b3e2 | 3932 | ((unsigned long)cachep) % REAPTIMEOUT_LIST3; |
cafeb02e | 3933 | l3->shared = new_shared; |
e498be7d | 3934 | l3->alien = new_alien; |
b28a02de | 3935 | l3->free_limit = (1 + nr_cpus_node(node)) * |
a737b3e2 | 3936 | cachep->batchcount + cachep->num; |
e498be7d CL |
3937 | cachep->nodelists[node] = l3; |
3938 | } | |
cafeb02e | 3939 | return 0; |
0718dc2a | 3940 | |
a737b3e2 | 3941 | fail: |
0718dc2a CL |
3942 | if (!cachep->next.next) { |
3943 | /* Cache is not active yet. Roll back what we did */ | |
3944 | node--; | |
3945 | while (node >= 0) { | |
3946 | if (cachep->nodelists[node]) { | |
3947 | l3 = cachep->nodelists[node]; | |
3948 | ||
3949 | kfree(l3->shared); | |
3950 | free_alien_cache(l3->alien); | |
3951 | kfree(l3); | |
3952 | cachep->nodelists[node] = NULL; | |
3953 | } | |
3954 | node--; | |
3955 | } | |
3956 | } | |
cafeb02e | 3957 | return -ENOMEM; |
e498be7d CL |
3958 | } |
3959 | ||
1da177e4 | 3960 | struct ccupdate_struct { |
343e0d7a | 3961 | struct kmem_cache *cachep; |
1da177e4 LT |
3962 | struct array_cache *new[NR_CPUS]; |
3963 | }; | |
3964 | ||
3965 | static void do_ccupdate_local(void *info) | |
3966 | { | |
a737b3e2 | 3967 | struct ccupdate_struct *new = info; |
1da177e4 LT |
3968 | struct array_cache *old; |
3969 | ||
3970 | check_irq_off(); | |
9a2dba4b | 3971 | old = cpu_cache_get(new->cachep); |
e498be7d | 3972 | |
1da177e4 LT |
3973 | new->cachep->array[smp_processor_id()] = new->new[smp_processor_id()]; |
3974 | new->new[smp_processor_id()] = old; | |
3975 | } | |
3976 | ||
b5d8ca7c | 3977 | /* Always called with the cache_chain_mutex held */ |
a737b3e2 | 3978 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3979 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3980 | { |
d2e7b7d0 | 3981 | struct ccupdate_struct *new; |
2ed3a4ef | 3982 | int i; |
1da177e4 | 3983 | |
83b519e8 | 3984 | new = kzalloc(sizeof(*new), gfp); |
d2e7b7d0 SS |
3985 | if (!new) |
3986 | return -ENOMEM; | |
3987 | ||
e498be7d | 3988 | for_each_online_cpu(i) { |
7d6e6d09 | 3989 | new->new[i] = alloc_arraycache(cpu_to_mem(i), limit, |
83b519e8 | 3990 | batchcount, gfp); |
d2e7b7d0 | 3991 | if (!new->new[i]) { |
b28a02de | 3992 | for (i--; i >= 0; i--) |
d2e7b7d0 SS |
3993 | kfree(new->new[i]); |
3994 | kfree(new); | |
e498be7d | 3995 | return -ENOMEM; |
1da177e4 LT |
3996 | } |
3997 | } | |
d2e7b7d0 | 3998 | new->cachep = cachep; |
1da177e4 | 3999 | |
15c8b6c1 | 4000 | on_each_cpu(do_ccupdate_local, (void *)new, 1); |
e498be7d | 4001 | |
1da177e4 | 4002 | check_irq_on(); |
1da177e4 LT |
4003 | cachep->batchcount = batchcount; |
4004 | cachep->limit = limit; | |
e498be7d | 4005 | cachep->shared = shared; |
1da177e4 | 4006 | |
e498be7d | 4007 | for_each_online_cpu(i) { |
d2e7b7d0 | 4008 | struct array_cache *ccold = new->new[i]; |
1da177e4 LT |
4009 | if (!ccold) |
4010 | continue; | |
7d6e6d09 LS |
4011 | spin_lock_irq(&cachep->nodelists[cpu_to_mem(i)]->list_lock); |
4012 | free_block(cachep, ccold->entry, ccold->avail, cpu_to_mem(i)); | |
4013 | spin_unlock_irq(&cachep->nodelists[cpu_to_mem(i)]->list_lock); | |
1da177e4 LT |
4014 | kfree(ccold); |
4015 | } | |
d2e7b7d0 | 4016 | kfree(new); |
83b519e8 | 4017 | return alloc_kmemlist(cachep, gfp); |
1da177e4 LT |
4018 | } |
4019 | ||
b5d8ca7c | 4020 | /* Called with cache_chain_mutex held always */ |
83b519e8 | 4021 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
4022 | { |
4023 | int err; | |
4024 | int limit, shared; | |
4025 | ||
a737b3e2 AM |
4026 | /* |
4027 | * The head array serves three purposes: | |
1da177e4 LT |
4028 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
4029 | * - reduce the number of spinlock operations. | |
a737b3e2 | 4030 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
4031 | * bufctl chains: array operations are cheaper. |
4032 | * The numbers are guessed, we should auto-tune as described by | |
4033 | * Bonwick. | |
4034 | */ | |
3dafccf2 | 4035 | if (cachep->buffer_size > 131072) |
1da177e4 | 4036 | limit = 1; |
3dafccf2 | 4037 | else if (cachep->buffer_size > PAGE_SIZE) |
1da177e4 | 4038 | limit = 8; |
3dafccf2 | 4039 | else if (cachep->buffer_size > 1024) |
1da177e4 | 4040 | limit = 24; |
3dafccf2 | 4041 | else if (cachep->buffer_size > 256) |
1da177e4 LT |
4042 | limit = 54; |
4043 | else | |
4044 | limit = 120; | |
4045 | ||
a737b3e2 AM |
4046 | /* |
4047 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
4048 | * allocation behaviour: Most allocs on one cpu, most free operations |
4049 | * on another cpu. For these cases, an efficient object passing between | |
4050 | * cpus is necessary. This is provided by a shared array. The array | |
4051 | * replaces Bonwick's magazine layer. | |
4052 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
4053 | * to a larger limit. Thus disabled by default. | |
4054 | */ | |
4055 | shared = 0; | |
364fbb29 | 4056 | if (cachep->buffer_size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 4057 | shared = 8; |
1da177e4 LT |
4058 | |
4059 | #if DEBUG | |
a737b3e2 AM |
4060 | /* |
4061 | * With debugging enabled, large batchcount lead to excessively long | |
4062 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
4063 | */ |
4064 | if (limit > 32) | |
4065 | limit = 32; | |
4066 | #endif | |
83b519e8 | 4067 | err = do_tune_cpucache(cachep, limit, (limit + 1) / 2, shared, gfp); |
1da177e4 LT |
4068 | if (err) |
4069 | printk(KERN_ERR "enable_cpucache failed for %s, error %d.\n", | |
b28a02de | 4070 | cachep->name, -err); |
2ed3a4ef | 4071 | return err; |
1da177e4 LT |
4072 | } |
4073 | ||
1b55253a CL |
4074 | /* |
4075 | * Drain an array if it contains any elements taking the l3 lock only if | |
b18e7e65 CL |
4076 | * necessary. Note that the l3 listlock also protects the array_cache |
4077 | * if drain_array() is used on the shared array. | |
1b55253a CL |
4078 | */ |
4079 | void drain_array(struct kmem_cache *cachep, struct kmem_list3 *l3, | |
4080 | struct array_cache *ac, int force, int node) | |
1da177e4 LT |
4081 | { |
4082 | int tofree; | |
4083 | ||
1b55253a CL |
4084 | if (!ac || !ac->avail) |
4085 | return; | |
1da177e4 LT |
4086 | if (ac->touched && !force) { |
4087 | ac->touched = 0; | |
b18e7e65 | 4088 | } else { |
1b55253a | 4089 | spin_lock_irq(&l3->list_lock); |
b18e7e65 CL |
4090 | if (ac->avail) { |
4091 | tofree = force ? ac->avail : (ac->limit + 4) / 5; | |
4092 | if (tofree > ac->avail) | |
4093 | tofree = (ac->avail + 1) / 2; | |
4094 | free_block(cachep, ac->entry, tofree, node); | |
4095 | ac->avail -= tofree; | |
4096 | memmove(ac->entry, &(ac->entry[tofree]), | |
4097 | sizeof(void *) * ac->avail); | |
4098 | } | |
1b55253a | 4099 | spin_unlock_irq(&l3->list_lock); |
1da177e4 LT |
4100 | } |
4101 | } | |
4102 | ||
4103 | /** | |
4104 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4105 | * @w: work descriptor |
1da177e4 LT |
4106 | * |
4107 | * Called from workqueue/eventd every few seconds. | |
4108 | * Purpose: | |
4109 | * - clear the per-cpu caches for this CPU. | |
4110 | * - return freeable pages to the main free memory pool. | |
4111 | * | |
a737b3e2 AM |
4112 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4113 | * again on the next iteration. | |
1da177e4 | 4114 | */ |
7c5cae36 | 4115 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4116 | { |
7a7c381d | 4117 | struct kmem_cache *searchp; |
e498be7d | 4118 | struct kmem_list3 *l3; |
7d6e6d09 | 4119 | int node = numa_mem_id(); |
bf6aede7 | 4120 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4121 | |
7c5cae36 | 4122 | if (!mutex_trylock(&cache_chain_mutex)) |
1da177e4 | 4123 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4124 | goto out; |
1da177e4 | 4125 | |
7a7c381d | 4126 | list_for_each_entry(searchp, &cache_chain, next) { |
1da177e4 LT |
4127 | check_irq_on(); |
4128 | ||
35386e3b CL |
4129 | /* |
4130 | * We only take the l3 lock if absolutely necessary and we | |
4131 | * have established with reasonable certainty that | |
4132 | * we can do some work if the lock was obtained. | |
4133 | */ | |
aab2207c | 4134 | l3 = searchp->nodelists[node]; |
35386e3b | 4135 | |
8fce4d8e | 4136 | reap_alien(searchp, l3); |
1da177e4 | 4137 | |
aab2207c | 4138 | drain_array(searchp, l3, cpu_cache_get(searchp), 0, node); |
1da177e4 | 4139 | |
35386e3b CL |
4140 | /* |
4141 | * These are racy checks but it does not matter | |
4142 | * if we skip one check or scan twice. | |
4143 | */ | |
e498be7d | 4144 | if (time_after(l3->next_reap, jiffies)) |
35386e3b | 4145 | goto next; |
1da177e4 | 4146 | |
e498be7d | 4147 | l3->next_reap = jiffies + REAPTIMEOUT_LIST3; |
1da177e4 | 4148 | |
aab2207c | 4149 | drain_array(searchp, l3, l3->shared, 0, node); |
1da177e4 | 4150 | |
ed11d9eb | 4151 | if (l3->free_touched) |
e498be7d | 4152 | l3->free_touched = 0; |
ed11d9eb CL |
4153 | else { |
4154 | int freed; | |
1da177e4 | 4155 | |
ed11d9eb CL |
4156 | freed = drain_freelist(searchp, l3, (l3->free_limit + |
4157 | 5 * searchp->num - 1) / (5 * searchp->num)); | |
4158 | STATS_ADD_REAPED(searchp, freed); | |
4159 | } | |
35386e3b | 4160 | next: |
1da177e4 LT |
4161 | cond_resched(); |
4162 | } | |
4163 | check_irq_on(); | |
fc0abb14 | 4164 | mutex_unlock(&cache_chain_mutex); |
8fce4d8e | 4165 | next_reap_node(); |
7c5cae36 | 4166 | out: |
a737b3e2 | 4167 | /* Set up the next iteration */ |
7c5cae36 | 4168 | schedule_delayed_work(work, round_jiffies_relative(REAPTIMEOUT_CPUC)); |
1da177e4 LT |
4169 | } |
4170 | ||
158a9624 | 4171 | #ifdef CONFIG_SLABINFO |
1da177e4 | 4172 | |
85289f98 | 4173 | static void print_slabinfo_header(struct seq_file *m) |
1da177e4 | 4174 | { |
85289f98 PE |
4175 | /* |
4176 | * Output format version, so at least we can change it | |
4177 | * without _too_ many complaints. | |
4178 | */ | |
1da177e4 | 4179 | #if STATS |
85289f98 | 4180 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); |
1da177e4 | 4181 | #else |
85289f98 | 4182 | seq_puts(m, "slabinfo - version: 2.1\n"); |
1da177e4 | 4183 | #endif |
85289f98 PE |
4184 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " |
4185 | "<objperslab> <pagesperslab>"); | |
4186 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
4187 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1da177e4 | 4188 | #if STATS |
85289f98 | 4189 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " |
fb7faf33 | 4190 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
85289f98 | 4191 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1da177e4 | 4192 | #endif |
85289f98 PE |
4193 | seq_putc(m, '\n'); |
4194 | } | |
4195 | ||
4196 | static void *s_start(struct seq_file *m, loff_t *pos) | |
4197 | { | |
4198 | loff_t n = *pos; | |
85289f98 | 4199 | |
fc0abb14 | 4200 | mutex_lock(&cache_chain_mutex); |
85289f98 PE |
4201 | if (!n) |
4202 | print_slabinfo_header(m); | |
b92151ba PE |
4203 | |
4204 | return seq_list_start(&cache_chain, *pos); | |
1da177e4 LT |
4205 | } |
4206 | ||
4207 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
4208 | { | |
b92151ba | 4209 | return seq_list_next(p, &cache_chain, pos); |
1da177e4 LT |
4210 | } |
4211 | ||
4212 | static void s_stop(struct seq_file *m, void *p) | |
4213 | { | |
fc0abb14 | 4214 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
4215 | } |
4216 | ||
4217 | static int s_show(struct seq_file *m, void *p) | |
4218 | { | |
b92151ba | 4219 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); |
b28a02de PE |
4220 | struct slab *slabp; |
4221 | unsigned long active_objs; | |
4222 | unsigned long num_objs; | |
4223 | unsigned long active_slabs = 0; | |
4224 | unsigned long num_slabs, free_objects = 0, shared_avail = 0; | |
e498be7d | 4225 | const char *name; |
1da177e4 | 4226 | char *error = NULL; |
e498be7d CL |
4227 | int node; |
4228 | struct kmem_list3 *l3; | |
1da177e4 | 4229 | |
1da177e4 LT |
4230 | active_objs = 0; |
4231 | num_slabs = 0; | |
e498be7d CL |
4232 | for_each_online_node(node) { |
4233 | l3 = cachep->nodelists[node]; | |
4234 | if (!l3) | |
4235 | continue; | |
4236 | ||
ca3b9b91 RT |
4237 | check_irq_on(); |
4238 | spin_lock_irq(&l3->list_lock); | |
e498be7d | 4239 | |
7a7c381d | 4240 | list_for_each_entry(slabp, &l3->slabs_full, list) { |
e498be7d CL |
4241 | if (slabp->inuse != cachep->num && !error) |
4242 | error = "slabs_full accounting error"; | |
4243 | active_objs += cachep->num; | |
4244 | active_slabs++; | |
4245 | } | |
7a7c381d | 4246 | list_for_each_entry(slabp, &l3->slabs_partial, list) { |
e498be7d CL |
4247 | if (slabp->inuse == cachep->num && !error) |
4248 | error = "slabs_partial inuse accounting error"; | |
4249 | if (!slabp->inuse && !error) | |
4250 | error = "slabs_partial/inuse accounting error"; | |
4251 | active_objs += slabp->inuse; | |
4252 | active_slabs++; | |
4253 | } | |
7a7c381d | 4254 | list_for_each_entry(slabp, &l3->slabs_free, list) { |
e498be7d CL |
4255 | if (slabp->inuse && !error) |
4256 | error = "slabs_free/inuse accounting error"; | |
4257 | num_slabs++; | |
4258 | } | |
4259 | free_objects += l3->free_objects; | |
4484ebf1 RT |
4260 | if (l3->shared) |
4261 | shared_avail += l3->shared->avail; | |
e498be7d | 4262 | |
ca3b9b91 | 4263 | spin_unlock_irq(&l3->list_lock); |
1da177e4 | 4264 | } |
b28a02de PE |
4265 | num_slabs += active_slabs; |
4266 | num_objs = num_slabs * cachep->num; | |
e498be7d | 4267 | if (num_objs - active_objs != free_objects && !error) |
1da177e4 LT |
4268 | error = "free_objects accounting error"; |
4269 | ||
b28a02de | 4270 | name = cachep->name; |
1da177e4 LT |
4271 | if (error) |
4272 | printk(KERN_ERR "slab: cache %s error: %s\n", name, error); | |
4273 | ||
4274 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
3dafccf2 | 4275 | name, active_objs, num_objs, cachep->buffer_size, |
b28a02de | 4276 | cachep->num, (1 << cachep->gfporder)); |
1da177e4 | 4277 | seq_printf(m, " : tunables %4u %4u %4u", |
b28a02de | 4278 | cachep->limit, cachep->batchcount, cachep->shared); |
e498be7d | 4279 | seq_printf(m, " : slabdata %6lu %6lu %6lu", |
b28a02de | 4280 | active_slabs, num_slabs, shared_avail); |
1da177e4 | 4281 | #if STATS |
b28a02de | 4282 | { /* list3 stats */ |
1da177e4 LT |
4283 | unsigned long high = cachep->high_mark; |
4284 | unsigned long allocs = cachep->num_allocations; | |
4285 | unsigned long grown = cachep->grown; | |
4286 | unsigned long reaped = cachep->reaped; | |
4287 | unsigned long errors = cachep->errors; | |
4288 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4289 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4290 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4291 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4292 | |
e92dd4fd JP |
4293 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu " |
4294 | "%4lu %4lu %4lu %4lu %4lu", | |
4295 | allocs, high, grown, | |
4296 | reaped, errors, max_freeable, node_allocs, | |
4297 | node_frees, overflows); | |
1da177e4 LT |
4298 | } |
4299 | /* cpu stats */ | |
4300 | { | |
4301 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4302 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4303 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4304 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4305 | ||
4306 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4307 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4308 | } |
4309 | #endif | |
4310 | seq_putc(m, '\n'); | |
1da177e4 LT |
4311 | return 0; |
4312 | } | |
4313 | ||
4314 | /* | |
4315 | * slabinfo_op - iterator that generates /proc/slabinfo | |
4316 | * | |
4317 | * Output layout: | |
4318 | * cache-name | |
4319 | * num-active-objs | |
4320 | * total-objs | |
4321 | * object size | |
4322 | * num-active-slabs | |
4323 | * total-slabs | |
4324 | * num-pages-per-slab | |
4325 | * + further values on SMP and with statistics enabled | |
4326 | */ | |
4327 | ||
7b3c3a50 | 4328 | static const struct seq_operations slabinfo_op = { |
b28a02de PE |
4329 | .start = s_start, |
4330 | .next = s_next, | |
4331 | .stop = s_stop, | |
4332 | .show = s_show, | |
1da177e4 LT |
4333 | }; |
4334 | ||
4335 | #define MAX_SLABINFO_WRITE 128 | |
4336 | /** | |
4337 | * slabinfo_write - Tuning for the slab allocator | |
4338 | * @file: unused | |
4339 | * @buffer: user buffer | |
4340 | * @count: data length | |
4341 | * @ppos: unused | |
4342 | */ | |
b28a02de PE |
4343 | ssize_t slabinfo_write(struct file *file, const char __user * buffer, |
4344 | size_t count, loff_t *ppos) | |
1da177e4 | 4345 | { |
b28a02de | 4346 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4347 | int limit, batchcount, shared, res; |
7a7c381d | 4348 | struct kmem_cache *cachep; |
b28a02de | 4349 | |
1da177e4 LT |
4350 | if (count > MAX_SLABINFO_WRITE) |
4351 | return -EINVAL; | |
4352 | if (copy_from_user(&kbuf, buffer, count)) | |
4353 | return -EFAULT; | |
b28a02de | 4354 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4355 | |
4356 | tmp = strchr(kbuf, ' '); | |
4357 | if (!tmp) | |
4358 | return -EINVAL; | |
4359 | *tmp = '\0'; | |
4360 | tmp++; | |
4361 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4362 | return -EINVAL; | |
4363 | ||
4364 | /* Find the cache in the chain of caches. */ | |
fc0abb14 | 4365 | mutex_lock(&cache_chain_mutex); |
1da177e4 | 4366 | res = -EINVAL; |
7a7c381d | 4367 | list_for_each_entry(cachep, &cache_chain, next) { |
1da177e4 | 4368 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4369 | if (limit < 1 || batchcount < 1 || |
4370 | batchcount > limit || shared < 0) { | |
e498be7d | 4371 | res = 0; |
1da177e4 | 4372 | } else { |
e498be7d | 4373 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4374 | batchcount, shared, |
4375 | GFP_KERNEL); | |
1da177e4 LT |
4376 | } |
4377 | break; | |
4378 | } | |
4379 | } | |
fc0abb14 | 4380 | mutex_unlock(&cache_chain_mutex); |
1da177e4 LT |
4381 | if (res >= 0) |
4382 | res = count; | |
4383 | return res; | |
4384 | } | |
871751e2 | 4385 | |
7b3c3a50 AD |
4386 | static int slabinfo_open(struct inode *inode, struct file *file) |
4387 | { | |
4388 | return seq_open(file, &slabinfo_op); | |
4389 | } | |
4390 | ||
4391 | static const struct file_operations proc_slabinfo_operations = { | |
4392 | .open = slabinfo_open, | |
4393 | .read = seq_read, | |
4394 | .write = slabinfo_write, | |
4395 | .llseek = seq_lseek, | |
4396 | .release = seq_release, | |
4397 | }; | |
4398 | ||
871751e2 AV |
4399 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
4400 | ||
4401 | static void *leaks_start(struct seq_file *m, loff_t *pos) | |
4402 | { | |
871751e2 | 4403 | mutex_lock(&cache_chain_mutex); |
b92151ba | 4404 | return seq_list_start(&cache_chain, *pos); |
871751e2 AV |
4405 | } |
4406 | ||
4407 | static inline int add_caller(unsigned long *n, unsigned long v) | |
4408 | { | |
4409 | unsigned long *p; | |
4410 | int l; | |
4411 | if (!v) | |
4412 | return 1; | |
4413 | l = n[1]; | |
4414 | p = n + 2; | |
4415 | while (l) { | |
4416 | int i = l/2; | |
4417 | unsigned long *q = p + 2 * i; | |
4418 | if (*q == v) { | |
4419 | q[1]++; | |
4420 | return 1; | |
4421 | } | |
4422 | if (*q > v) { | |
4423 | l = i; | |
4424 | } else { | |
4425 | p = q + 2; | |
4426 | l -= i + 1; | |
4427 | } | |
4428 | } | |
4429 | if (++n[1] == n[0]) | |
4430 | return 0; | |
4431 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4432 | p[0] = v; | |
4433 | p[1] = 1; | |
4434 | return 1; | |
4435 | } | |
4436 | ||
4437 | static void handle_slab(unsigned long *n, struct kmem_cache *c, struct slab *s) | |
4438 | { | |
4439 | void *p; | |
4440 | int i; | |
4441 | if (n[0] == n[1]) | |
4442 | return; | |
4443 | for (i = 0, p = s->s_mem; i < c->num; i++, p += c->buffer_size) { | |
4444 | if (slab_bufctl(s)[i] != BUFCTL_ACTIVE) | |
4445 | continue; | |
4446 | if (!add_caller(n, (unsigned long)*dbg_userword(c, p))) | |
4447 | return; | |
4448 | } | |
4449 | } | |
4450 | ||
4451 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4452 | { | |
4453 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4454 | unsigned long offset, size; |
9281acea | 4455 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4456 | |
a5c43dae | 4457 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4458 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4459 | if (modname[0]) |
871751e2 AV |
4460 | seq_printf(m, " [%s]", modname); |
4461 | return; | |
4462 | } | |
4463 | #endif | |
4464 | seq_printf(m, "%p", (void *)address); | |
4465 | } | |
4466 | ||
4467 | static int leaks_show(struct seq_file *m, void *p) | |
4468 | { | |
b92151ba | 4469 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, next); |
871751e2 AV |
4470 | struct slab *slabp; |
4471 | struct kmem_list3 *l3; | |
4472 | const char *name; | |
4473 | unsigned long *n = m->private; | |
4474 | int node; | |
4475 | int i; | |
4476 | ||
4477 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4478 | return 0; | |
4479 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4480 | return 0; | |
4481 | ||
4482 | /* OK, we can do it */ | |
4483 | ||
4484 | n[1] = 0; | |
4485 | ||
4486 | for_each_online_node(node) { | |
4487 | l3 = cachep->nodelists[node]; | |
4488 | if (!l3) | |
4489 | continue; | |
4490 | ||
4491 | check_irq_on(); | |
4492 | spin_lock_irq(&l3->list_lock); | |
4493 | ||
7a7c381d | 4494 | list_for_each_entry(slabp, &l3->slabs_full, list) |
871751e2 | 4495 | handle_slab(n, cachep, slabp); |
7a7c381d | 4496 | list_for_each_entry(slabp, &l3->slabs_partial, list) |
871751e2 | 4497 | handle_slab(n, cachep, slabp); |
871751e2 AV |
4498 | spin_unlock_irq(&l3->list_lock); |
4499 | } | |
4500 | name = cachep->name; | |
4501 | if (n[0] == n[1]) { | |
4502 | /* Increase the buffer size */ | |
4503 | mutex_unlock(&cache_chain_mutex); | |
4504 | m->private = kzalloc(n[0] * 4 * sizeof(unsigned long), GFP_KERNEL); | |
4505 | if (!m->private) { | |
4506 | /* Too bad, we are really out */ | |
4507 | m->private = n; | |
4508 | mutex_lock(&cache_chain_mutex); | |
4509 | return -ENOMEM; | |
4510 | } | |
4511 | *(unsigned long *)m->private = n[0] * 2; | |
4512 | kfree(n); | |
4513 | mutex_lock(&cache_chain_mutex); | |
4514 | /* Now make sure this entry will be retried */ | |
4515 | m->count = m->size; | |
4516 | return 0; | |
4517 | } | |
4518 | for (i = 0; i < n[1]; i++) { | |
4519 | seq_printf(m, "%s: %lu ", name, n[2*i+3]); | |
4520 | show_symbol(m, n[2*i+2]); | |
4521 | seq_putc(m, '\n'); | |
4522 | } | |
d2e7b7d0 | 4523 | |
871751e2 AV |
4524 | return 0; |
4525 | } | |
4526 | ||
a0ec95a8 | 4527 | static const struct seq_operations slabstats_op = { |
871751e2 AV |
4528 | .start = leaks_start, |
4529 | .next = s_next, | |
4530 | .stop = s_stop, | |
4531 | .show = leaks_show, | |
4532 | }; | |
a0ec95a8 AD |
4533 | |
4534 | static int slabstats_open(struct inode *inode, struct file *file) | |
4535 | { | |
4536 | unsigned long *n = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
4537 | int ret = -ENOMEM; | |
4538 | if (n) { | |
4539 | ret = seq_open(file, &slabstats_op); | |
4540 | if (!ret) { | |
4541 | struct seq_file *m = file->private_data; | |
4542 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4543 | m->private = n; | |
4544 | n = NULL; | |
4545 | } | |
4546 | kfree(n); | |
4547 | } | |
4548 | return ret; | |
4549 | } | |
4550 | ||
4551 | static const struct file_operations proc_slabstats_operations = { | |
4552 | .open = slabstats_open, | |
4553 | .read = seq_read, | |
4554 | .llseek = seq_lseek, | |
4555 | .release = seq_release_private, | |
4556 | }; | |
4557 | #endif | |
4558 | ||
4559 | static int __init slab_proc_init(void) | |
4560 | { | |
7b3c3a50 | 4561 | proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations); |
a0ec95a8 AD |
4562 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
4563 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4564 | #endif |
a0ec95a8 AD |
4565 | return 0; |
4566 | } | |
4567 | module_init(slab_proc_init); | |
1da177e4 LT |
4568 | #endif |
4569 | ||
00e145b6 MS |
4570 | /** |
4571 | * ksize - get the actual amount of memory allocated for a given object | |
4572 | * @objp: Pointer to the object | |
4573 | * | |
4574 | * kmalloc may internally round up allocations and return more memory | |
4575 | * than requested. ksize() can be used to determine the actual amount of | |
4576 | * memory allocated. The caller may use this additional memory, even though | |
4577 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4578 | * The caller must guarantee that objp points to a valid object previously | |
4579 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4580 | * must not be freed during the duration of the call. | |
4581 | */ | |
fd76bab2 | 4582 | size_t ksize(const void *objp) |
1da177e4 | 4583 | { |
ef8b4520 CL |
4584 | BUG_ON(!objp); |
4585 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4586 | return 0; |
1da177e4 | 4587 | |
6ed5eb22 | 4588 | return obj_size(virt_to_cache(objp)); |
1da177e4 | 4589 | } |
b1aabecd | 4590 | EXPORT_SYMBOL(ksize); |